summaryrefslogtreecommitdiff
path: root/drivers/gpu/drm/xe
diff options
context:
space:
mode:
Diffstat (limited to 'drivers/gpu/drm/xe')
-rw-r--r--drivers/gpu/drm/xe/.gitignore4
-rw-r--r--drivers/gpu/drm/xe/.kunitconfig13
-rw-r--r--drivers/gpu/drm/xe/Kconfig96
-rw-r--r--drivers/gpu/drm/xe/Kconfig.debug107
-rw-r--r--drivers/gpu/drm/xe/Kconfig.profile54
-rw-r--r--drivers/gpu/drm/xe/Makefile305
-rw-r--r--drivers/gpu/drm/xe/abi/gsc_command_header_abi.h46
-rw-r--r--drivers/gpu/drm/xe/abi/gsc_mkhi_commands_abi.h39
-rw-r--r--drivers/gpu/drm/xe/abi/gsc_pxp_commands_abi.h59
-rw-r--r--drivers/gpu/drm/xe/abi/guc_actions_abi.h219
-rw-r--r--drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h249
-rw-r--r--drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h127
-rw-r--r--drivers/gpu/drm/xe/abi/guc_communication_mmio_abi.h49
-rw-r--r--drivers/gpu/drm/xe/abi/guc_errors_abi.h37
-rw-r--r--drivers/gpu/drm/xe/abi/guc_klvs_abi.h322
-rw-r--r--drivers/gpu/drm/xe/abi/guc_messages_abi.h234
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_lmem.h1
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_mman.h17
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h65
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object_frontbuffer.h12
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gt/intel_rps.h11
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_active.h22
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_active_types.h13
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_config.h19
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_debugfs.h14
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_drv.h233
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_fixed.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_gem.h9
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_gem_stolen.h79
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_gpu_error.h17
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_irq.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_reg.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_reg_defs.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_trace.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_utils.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_vgpu.h44
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h34
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/i915_vma_types.h74
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_clock_gating.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_gt_types.h11
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_mchbar_regs.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_pci_config.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_pcode.h42
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_runtime_pm.h16
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_step.h20
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_uc_fw.h11
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h175
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/intel_wakeref.h8
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/pxp/intel_pxp.h28
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/soc/intel_dram.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/soc/intel_pch.h6
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband.h132
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband_reg.h6
-rw-r--r--drivers/gpu/drm/xe/display/ext/i915_irq.c77
-rw-r--r--drivers/gpu/drm/xe/display/ext/i915_utils.c26
-rw-r--r--drivers/gpu/drm/xe/display/intel_fb_bo.c74
-rw-r--r--drivers/gpu/drm/xe/display/intel_fb_bo.h24
-rw-r--r--drivers/gpu/drm/xe/display/intel_fbdev_fb.c104
-rw-r--r--drivers/gpu/drm/xe/display/intel_fbdev_fb.h21
-rw-r--r--drivers/gpu/drm/xe/display/xe_display_misc.c16
-rw-r--r--drivers/gpu/drm/xe/display/xe_display_rps.c17
-rw-r--r--drivers/gpu/drm/xe/display/xe_dsb_buffer.c71
-rw-r--r--drivers/gpu/drm/xe/display/xe_fb_pin.c384
-rw-r--r--drivers/gpu/drm/xe/display/xe_hdcp_gsc.c34
-rw-r--r--drivers/gpu/drm/xe/display/xe_plane_initial.c291
-rw-r--r--drivers/gpu/drm/xe/instructions/xe_gfxpipe_commands.h160
-rw-r--r--drivers/gpu/drm/xe/instructions/xe_gsc_commands.h36
-rw-r--r--drivers/gpu/drm/xe/instructions/xe_instr_defs.h33
-rw-r--r--drivers/gpu/drm/xe/instructions/xe_mi_commands.h61
-rw-r--r--drivers/gpu/drm/xe/regs/xe_engine_regs.h184
-rw-r--r--drivers/gpu/drm/xe/regs/xe_gpu_commands.h70
-rw-r--r--drivers/gpu/drm/xe/regs/xe_gsc_regs.h41
-rw-r--r--drivers/gpu/drm/xe/regs/xe_gt_regs.h478
-rw-r--r--drivers/gpu/drm/xe/regs/xe_guc_regs.h143
-rw-r--r--drivers/gpu/drm/xe/regs/xe_lrc_layout.h17
-rw-r--r--drivers/gpu/drm/xe/regs/xe_mchbar_regs.h44
-rw-r--r--drivers/gpu/drm/xe/regs/xe_reg_defs.h120
-rw-r--r--drivers/gpu/drm/xe/regs/xe_regs.h68
-rw-r--r--drivers/gpu/drm/xe/regs/xe_sriov_regs.h17
-rw-r--r--drivers/gpu/drm/xe/tests/Makefile10
-rw-r--r--drivers/gpu/drm/xe/tests/xe_bo.c353
-rw-r--r--drivers/gpu/drm/xe/tests/xe_bo_test.c26
-rw-r--r--drivers/gpu/drm/xe/tests/xe_bo_test.h14
-rw-r--r--drivers/gpu/drm/xe/tests/xe_dma_buf.c278
-rw-r--r--drivers/gpu/drm/xe/tests/xe_dma_buf_test.c25
-rw-r--r--drivers/gpu/drm/xe/tests/xe_dma_buf_test.h13
-rw-r--r--drivers/gpu/drm/xe/tests/xe_lmtt_test.c73
-rw-r--r--drivers/gpu/drm/xe/tests/xe_migrate.c444
-rw-r--r--drivers/gpu/drm/xe/tests/xe_migrate_test.c25
-rw-r--r--drivers/gpu/drm/xe/tests/xe_migrate_test.h13
-rw-r--r--drivers/gpu/drm/xe/tests/xe_mocs.c130
-rw-r--r--drivers/gpu/drm/xe/tests/xe_mocs_test.c24
-rw-r--r--drivers/gpu/drm/xe/tests/xe_mocs_test.h13
-rw-r--r--drivers/gpu/drm/xe/tests/xe_pci.c166
-rw-r--r--drivers/gpu/drm/xe/tests/xe_pci_test.c71
-rw-r--r--drivers/gpu/drm/xe/tests/xe_pci_test.h36
-rw-r--r--drivers/gpu/drm/xe/tests/xe_rtp_test.c319
-rw-r--r--drivers/gpu/drm/xe/tests/xe_test.h67
-rw-r--r--drivers/gpu/drm/xe/tests/xe_wa_test.c170
-rw-r--r--drivers/gpu/drm/xe/xe_assert.h174
-rw-r--r--drivers/gpu/drm/xe/xe_bb.c110
-rw-r--r--drivers/gpu/drm/xe/xe_bb.h25
-rw-r--r--drivers/gpu/drm/xe/xe_bb_types.h20
-rw-r--r--drivers/gpu/drm/xe/xe_bo.c2269
-rw-r--r--drivers/gpu/drm/xe/xe_bo.h355
-rw-r--r--drivers/gpu/drm/xe/xe_bo_doc.h179
-rw-r--r--drivers/gpu/drm/xe/xe_bo_evict.c228
-rw-r--r--drivers/gpu/drm/xe/xe_bo_evict.h15
-rw-r--r--drivers/gpu/drm/xe/xe_bo_types.h96
-rw-r--r--drivers/gpu/drm/xe/xe_debugfs.c148
-rw-r--r--drivers/gpu/drm/xe/xe_debugfs.h13
-rw-r--r--drivers/gpu/drm/xe/xe_devcoredump.c196
-rw-r--r--drivers/gpu/drm/xe/xe_devcoredump.h20
-rw-r--r--drivers/gpu/drm/xe/xe_devcoredump_types.h55
-rw-r--r--drivers/gpu/drm/xe/xe_device.c700
-rw-r--r--drivers/gpu/drm/xe/xe_device.h173
-rw-r--r--drivers/gpu/drm/xe/xe_device_sysfs.c89
-rw-r--r--drivers/gpu/drm/xe/xe_device_sysfs.h13
-rw-r--r--drivers/gpu/drm/xe/xe_device_types.h545
-rw-r--r--drivers/gpu/drm/xe/xe_display.c422
-rw-r--r--drivers/gpu/drm/xe/xe_display.h72
-rw-r--r--drivers/gpu/drm/xe/xe_dma_buf.c322
-rw-r--r--drivers/gpu/drm/xe/xe_dma_buf.h15
-rw-r--r--drivers/gpu/drm/xe/xe_drm_client.c204
-rw-r--r--drivers/gpu/drm/xe/xe_drm_client.h70
-rw-r--r--drivers/gpu/drm/xe/xe_drv.h23
-rw-r--r--drivers/gpu/drm/xe/xe_exec.c350
-rw-r--r--drivers/gpu/drm/xe/xe_exec.h14
-rw-r--r--drivers/gpu/drm/xe/xe_exec_queue.c956
-rw-r--r--drivers/gpu/drm/xe/xe_exec_queue.h69
-rw-r--r--drivers/gpu/drm/xe/xe_exec_queue_types.h222
-rw-r--r--drivers/gpu/drm/xe/xe_execlist.c474
-rw-r--r--drivers/gpu/drm/xe/xe_execlist.h21
-rw-r--r--drivers/gpu/drm/xe/xe_execlist_types.h49
-rw-r--r--drivers/gpu/drm/xe/xe_force_wake.c199
-rw-r--r--drivers/gpu/drm/xe/xe_force_wake.h38
-rw-r--r--drivers/gpu/drm/xe/xe_force_wake_types.h86
-rw-r--r--drivers/gpu/drm/xe/xe_gen_wa_oob.c165
-rw-r--r--drivers/gpu/drm/xe/xe_ggtt.c428
-rw-r--r--drivers/gpu/drm/xe/xe_ggtt.h33
-rw-r--r--drivers/gpu/drm/xe/xe_ggtt_types.h39
-rw-r--r--drivers/gpu/drm/xe/xe_gpu_scheduler.c101
-rw-r--r--drivers/gpu/drm/xe/xe_gpu_scheduler.h73
-rw-r--r--drivers/gpu/drm/xe/xe_gpu_scheduler_types.h57
-rw-r--r--drivers/gpu/drm/xe/xe_gsc.c438
-rw-r--r--drivers/gpu/drm/xe/xe_gsc.h20
-rw-r--r--drivers/gpu/drm/xe/xe_gsc_submit.c184
-rw-r--r--drivers/gpu/drm/xe/xe_gsc_submit.h30
-rw-r--r--drivers/gpu/drm/xe/xe_gsc_types.h39
-rw-r--r--drivers/gpu/drm/xe/xe_gt.c778
-rw-r--r--drivers/gpu/drm/xe/xe_gt.h72
-rw-r--r--drivers/gpu/drm/xe/xe_gt_ccs_mode.c191
-rw-r--r--drivers/gpu/drm/xe/xe_gt_ccs_mode.h24
-rw-r--r--drivers/gpu/drm/xe/xe_gt_clock.c85
-rw-r--r--drivers/gpu/drm/xe/xe_gt_clock.h15
-rw-r--r--drivers/gpu/drm/xe/xe_gt_debugfs.c249
-rw-r--r--drivers/gpu/drm/xe/xe_gt_debugfs.h13
-rw-r--r--drivers/gpu/drm/xe/xe_gt_freq.c219
-rw-r--r--drivers/gpu/drm/xe/xe_gt_freq.h13
-rw-r--r--drivers/gpu/drm/xe/xe_gt_idle.c192
-rw-r--r--drivers/gpu/drm/xe/xe_gt_idle.h17
-rw-r--r--drivers/gpu/drm/xe/xe_gt_idle_types.h38
-rw-r--r--drivers/gpu/drm/xe/xe_gt_mcr.c685
-rw-r--r--drivers/gpu/drm/xe/xe_gt_mcr.h29
-rw-r--r--drivers/gpu/drm/xe/xe_gt_pagefault.c646
-rw-r--r--drivers/gpu/drm/xe/xe_gt_pagefault.h19
-rw-r--r--drivers/gpu/drm/xe/xe_gt_printk.h46
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sysfs.c61
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sysfs.h19
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sysfs_types.h26
-rw-r--r--drivers/gpu/drm/xe/xe_gt_throttle_sysfs.c251
-rw-r--r--drivers/gpu/drm/xe/xe_gt_throttle_sysfs.h16
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c406
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h26
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation_types.h28
-rw-r--r--drivers/gpu/drm/xe/xe_gt_topology.c169
-rw-r--r--drivers/gpu/drm/xe/xe_gt_topology.h25
-rw-r--r--drivers/gpu/drm/xe/xe_gt_types.h363
-rw-r--r--drivers/gpu/drm/xe/xe_guc.c911
-rw-r--r--drivers/gpu/drm/xe/xe_guc.h72
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ads.c672
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ads.h17
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ads_types.h25
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct.c1320
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct.h59
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct_types.h115
-rw-r--r--drivers/gpu/drm/xe/xe_guc_debugfs.c74
-rw-r--r--drivers/gpu/drm/xe/xe_guc_debugfs.h14
-rw-r--r--drivers/gpu/drm/xe/xe_guc_exec_queue_types.h54
-rw-r--r--drivers/gpu/drm/xe/xe_guc_fwif.h361
-rw-r--r--drivers/gpu/drm/xe/xe_guc_hwconfig.c104
-rw-r--r--drivers/gpu/drm/xe/xe_guc_hwconfig.h17
-rw-r--r--drivers/gpu/drm/xe/xe_guc_log.c97
-rw-r--r--drivers/gpu/drm/xe/xe_guc_log.h48
-rw-r--r--drivers/gpu/drm/xe/xe_guc_log_types.h23
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc.c1000
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc.h31
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc_types.h34
-rw-r--r--drivers/gpu/drm/xe/xe_guc_submit.c1990
-rw-r--r--drivers/gpu/drm/xe/xe_guc_submit.h38
-rw-r--r--drivers/gpu/drm/xe/xe_guc_submit_types.h155
-rw-r--r--drivers/gpu/drm/xe/xe_guc_types.h81
-rw-r--r--drivers/gpu/drm/xe/xe_heci_gsc.c234
-rw-r--r--drivers/gpu/drm/xe/xe_heci_gsc.h35
-rw-r--r--drivers/gpu/drm/xe/xe_huc.c307
-rw-r--r--drivers/gpu/drm/xe/xe_huc.h26
-rw-r--r--drivers/gpu/drm/xe/xe_huc_debugfs.c70
-rw-r--r--drivers/gpu/drm/xe/xe_huc_debugfs.h14
-rw-r--r--drivers/gpu/drm/xe/xe_huc_types.h24
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine.c883
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine.h70
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.c675
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.h36
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine_types.h225
-rw-r--r--drivers/gpu/drm/xe/xe_hw_fence.c230
-rw-r--r--drivers/gpu/drm/xe/xe_hw_fence.h30
-rw-r--r--drivers/gpu/drm/xe/xe_hw_fence_types.h72
-rw-r--r--drivers/gpu/drm/xe/xe_hwmon.c776
-rw-r--r--drivers/gpu/drm/xe/xe_hwmon.h19
-rw-r--r--drivers/gpu/drm/xe/xe_irq.c666
-rw-r--r--drivers/gpu/drm/xe/xe_irq.h19
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt.c506
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt.h27
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt_2l.c150
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt_ml.c161
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt_types.h63
-rw-r--r--drivers/gpu/drm/xe/xe_lrc.c1272
-rw-r--r--drivers/gpu/drm/xe/xe_lrc.h58
-rw-r--r--drivers/gpu/drm/xe/xe_lrc_types.h46
-rw-r--r--drivers/gpu/drm/xe/xe_macros.h18
-rw-r--r--drivers/gpu/drm/xe/xe_map.h93
-rw-r--r--drivers/gpu/drm/xe/xe_migrate.c1410
-rw-r--r--drivers/gpu/drm/xe/xe_migrate.h110
-rw-r--r--drivers/gpu/drm/xe/xe_migrate_doc.h88
-rw-r--r--drivers/gpu/drm/xe/xe_mmio.c524
-rw-r--r--drivers/gpu/drm/xe/xe_mmio.h107
-rw-r--r--drivers/gpu/drm/xe/xe_mocs.c580
-rw-r--r--drivers/gpu/drm/xe/xe_mocs.h17
-rw-r--r--drivers/gpu/drm/xe/xe_module.c101
-rw-r--r--drivers/gpu/drm/xe/xe_module.h26
-rw-r--r--drivers/gpu/drm/xe/xe_pat.c459
-rw-r--r--drivers/gpu/drm/xe/xe_pat.h61
-rw-r--r--drivers/gpu/drm/xe/xe_pci.c951
-rw-r--r--drivers/gpu/drm/xe/xe_pci.h12
-rw-r--r--drivers/gpu/drm/xe/xe_pci_types.h46
-rw-r--r--drivers/gpu/drm/xe/xe_pcode.c296
-rw-r--r--drivers/gpu/drm/xe/xe_pcode.h30
-rw-r--r--drivers/gpu/drm/xe/xe_pcode_api.h49
-rw-r--r--drivers/gpu/drm/xe/xe_platform_types.h37
-rw-r--r--drivers/gpu/drm/xe/xe_pm.c405
-rw-r--r--drivers/gpu/drm/xe/xe_pm.h35
-rw-r--r--drivers/gpu/drm/xe/xe_preempt_fence.c158
-rw-r--r--drivers/gpu/drm/xe/xe_preempt_fence.h61
-rw-r--r--drivers/gpu/drm/xe/xe_preempt_fence_types.h32
-rw-r--r--drivers/gpu/drm/xe/xe_pt.c1653
-rw-r--r--drivers/gpu/drm/xe/xe_pt.h48
-rw-r--r--drivers/gpu/drm/xe/xe_pt_types.h77
-rw-r--r--drivers/gpu/drm/xe/xe_pt_walk.c160
-rw-r--r--drivers/gpu/drm/xe/xe_pt_walk.h161
-rw-r--r--drivers/gpu/drm/xe/xe_query.c552
-rw-r--r--drivers/gpu/drm/xe/xe_query.h14
-rw-r--r--drivers/gpu/drm/xe/xe_range_fence.c156
-rw-r--r--drivers/gpu/drm/xe/xe_range_fence.h75
-rw-r--r--drivers/gpu/drm/xe/xe_reg_sr.c284
-rw-r--r--drivers/gpu/drm/xe/xe_reg_sr.h28
-rw-r--r--drivers/gpu/drm/xe/xe_reg_sr_types.h37
-rw-r--r--drivers/gpu/drm/xe/xe_reg_whitelist.c146
-rw-r--r--drivers/gpu/drm/xe/xe_reg_whitelist.h23
-rw-r--r--drivers/gpu/drm/xe/xe_res_cursor.h240
-rw-r--r--drivers/gpu/drm/xe/xe_ring_ops.c482
-rw-r--r--drivers/gpu/drm/xe/xe_ring_ops.h17
-rw-r--r--drivers/gpu/drm/xe/xe_ring_ops_types.h22
-rw-r--r--drivers/gpu/drm/xe/xe_rtp.c325
-rw-r--r--drivers/gpu/drm/xe/xe_rtp.h430
-rw-r--r--drivers/gpu/drm/xe/xe_rtp_helpers.h81
-rw-r--r--drivers/gpu/drm/xe/xe_rtp_types.h124
-rw-r--r--drivers/gpu/drm/xe/xe_sa.c106
-rw-r--r--drivers/gpu/drm/xe/xe_sa.h40
-rw-r--r--drivers/gpu/drm/xe/xe_sa_types.h19
-rw-r--r--drivers/gpu/drm/xe/xe_sched_job.c280
-rw-r--r--drivers/gpu/drm/xe/xe_sched_job.h80
-rw-r--r--drivers/gpu/drm/xe/xe_sched_job_types.h46
-rw-r--r--drivers/gpu/drm/xe/xe_sriov.c55
-rw-r--r--drivers/gpu/drm/xe/xe_sriov.h42
-rw-r--r--drivers/gpu/drm/xe/xe_sriov_printk.h46
-rw-r--r--drivers/gpu/drm/xe/xe_sriov_types.h28
-rw-r--r--drivers/gpu/drm/xe/xe_step.c264
-rw-r--r--drivers/gpu/drm/xe/xe_step.h23
-rw-r--r--drivers/gpu/drm/xe/xe_step_types.h50
-rw-r--r--drivers/gpu/drm/xe/xe_sync.c344
-rw-r--r--drivers/gpu/drm/xe/xe_sync.h36
-rw-r--r--drivers/gpu/drm/xe/xe_sync_types.h28
-rw-r--r--drivers/gpu/drm/xe/xe_tile.c185
-rw-r--r--drivers/gpu/drm/xe/xe_tile.h18
-rw-r--r--drivers/gpu/drm/xe/xe_tile_sysfs.c57
-rw-r--r--drivers/gpu/drm/xe/xe_tile_sysfs.h19
-rw-r--r--drivers/gpu/drm/xe/xe_tile_sysfs_types.h27
-rw-r--r--drivers/gpu/drm/xe/xe_trace.c9
-rw-r--r--drivers/gpu/drm/xe/xe_trace.h608
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c334
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h21
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_sys_mgr.c118
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_sys_mgr.h13
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_vram_mgr.c480
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_vram_mgr.h44
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h52
-rw-r--r--drivers/gpu/drm/xe/xe_tuning.c121
-rw-r--r--drivers/gpu/drm/xe/xe_tuning.h16
-rw-r--r--drivers/gpu/drm/xe/xe_uc.c258
-rw-r--r--drivers/gpu/drm/xe/xe_uc.h24
-rw-r--r--drivers/gpu/drm/xe/xe_uc_debugfs.c26
-rw-r--r--drivers/gpu/drm/xe/xe_uc_debugfs.h14
-rw-r--r--drivers/gpu/drm/xe/xe_uc_fw.c882
-rw-r--r--drivers/gpu/drm/xe/xe_uc_fw.h184
-rw-r--r--drivers/gpu/drm/xe/xe_uc_fw_abi.h321
-rw-r--r--drivers/gpu/drm/xe/xe_uc_fw_types.h146
-rw-r--r--drivers/gpu/drm/xe/xe_uc_types.h28
-rw-r--r--drivers/gpu/drm/xe/xe_vm.c3209
-rw-r--r--drivers/gpu/drm/xe/xe_vm.h263
-rw-r--r--drivers/gpu/drm/xe/xe_vm_doc.h555
-rw-r--r--drivers/gpu/drm/xe/xe_vm_types.h373
-rw-r--r--drivers/gpu/drm/xe/xe_wa.c895
-rw-r--r--drivers/gpu/drm/xe/xe_wa.h32
-rw-r--r--drivers/gpu/drm/xe/xe_wa_oob.rules24
-rw-r--r--drivers/gpu/drm/xe/xe_wait_user_fence.c179
-rw-r--r--drivers/gpu/drm/xe/xe_wait_user_fence.h15
-rw-r--r--drivers/gpu/drm/xe/xe_wopcm.c270
-rw-r--r--drivers/gpu/drm/xe/xe_wopcm.h16
-rw-r--r--drivers/gpu/drm/xe/xe_wopcm_types.h26
330 files changed, 59579 insertions, 0 deletions
diff --git a/drivers/gpu/drm/xe/.gitignore b/drivers/gpu/drm/xe/.gitignore
new file mode 100644
index 000000000000..8778bf132674
--- /dev/null
+++ b/drivers/gpu/drm/xe/.gitignore
@@ -0,0 +1,4 @@
+# SPDX-License-Identifier: GPL-2.0-only
+*.hdrtest
+/generated
+/xe_gen_wa_oob
diff --git a/drivers/gpu/drm/xe/.kunitconfig b/drivers/gpu/drm/xe/.kunitconfig
new file mode 100644
index 000000000000..9590eac91af3
--- /dev/null
+++ b/drivers/gpu/drm/xe/.kunitconfig
@@ -0,0 +1,13 @@
+# xe dependencies
+CONFIG_KUNIT=y
+CONFIG_PCI=y
+CONFIG_PCI_IOV=y
+CONFIG_DEBUG_FS=y
+CONFIG_DRM=y
+CONFIG_DRM_FBDEV_EMULATION=y
+CONFIG_DRM_KMS_HELPER=y
+CONFIG_DRM_XE=y
+CONFIG_DRM_XE_DISPLAY=n
+CONFIG_EXPERT=y
+CONFIG_FB=y
+CONFIG_DRM_XE_KUNIT_TEST=y
diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig
new file mode 100644
index 000000000000..1cced50d8d8c
--- /dev/null
+++ b/drivers/gpu/drm/xe/Kconfig
@@ -0,0 +1,96 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config DRM_XE
+ tristate "Intel Xe Graphics"
+ depends on DRM && PCI && MMU && (m || (y && KUNIT=y)) && 64BIT
+ select INTERVAL_TREE
+ # we need shmfs for the swappable backing store, and in particular
+ # the shmem_readpage() which depends upon tmpfs
+ select SHMEM
+ select TMPFS
+ select DRM_BUDDY
+ select DRM_EXEC
+ select DRM_KMS_HELPER
+ select DRM_PANEL
+ select DRM_SUBALLOC_HELPER
+ select DRM_DISPLAY_DP_HELPER
+ select DRM_DISPLAY_HDCP_HELPER
+ select DRM_DISPLAY_HDMI_HELPER
+ select DRM_DISPLAY_HELPER
+ select DRM_MIPI_DSI
+ select RELAY
+ select IRQ_WORK
+ # xe depends on ACPI_VIDEO when ACPI is enabled
+ # but for select to work, need to select ACPI_VIDEO's dependencies, ick
+ select BACKLIGHT_CLASS_DEVICE if ACPI
+ select INPUT if ACPI
+ select ACPI_VIDEO if X86 && ACPI
+ select ACPI_BUTTON if ACPI
+ select ACPI_WMI if X86 && ACPI
+ select SYNC_FILE
+ select IOSF_MBI
+ select CRC32
+ select SND_HDA_I915 if SND_HDA_CORE
+ select CEC_CORE if CEC_NOTIFIER
+ select VMAP_PFN
+ select DRM_TTM
+ select DRM_TTM_HELPER
+ select DRM_EXEC
+ select DRM_GPUVM
+ select DRM_SCHED
+ select MMU_NOTIFIER
+ select WANT_DEV_COREDUMP
+ select AUXILIARY_BUS
+ help
+ Experimental driver for Intel Xe series GPUs
+
+ If "M" is selected, the module will be called xe.
+
+config DRM_XE_DISPLAY
+ bool "Enable display support"
+ depends on DRM_XE && EXPERT && DRM_XE=m
+ select FB_IOMEM_HELPERS
+ select I2C
+ select I2C_ALGOBIT
+ default y
+ help
+ Disable this option only if you want to compile out display support.
+
+config DRM_XE_FORCE_PROBE
+ string "Force probe xe for selected Intel hardware IDs"
+ depends on DRM_XE
+ help
+ This is the default value for the xe.force_probe module
+ parameter. Using the module parameter overrides this option.
+
+ Force probe the xe for Intel graphics devices that are
+ recognized but not properly supported by this kernel version. It is
+ recommended to upgrade to a kernel version with proper support as soon
+ as it is available.
+
+ It can also be used to block the probe of recognized and fully
+ supported devices.
+
+ Use "" to disable force probe. If in doubt, use this.
+
+ Use "<pci-id>[,<pci-id>,...]" to force probe the xe for listed
+ devices. For example, "4500" or "4500,4571".
+
+ Use "*" to force probe the driver for all known devices.
+
+ Use "!" right before the ID to block the probe of the device. For
+ example, "4500,!4571" forces the probe of 4500 and blocks the probe of
+ 4571.
+
+ Use "!*" to block the probe of the driver for all known devices.
+
+menu "drm/Xe Debugging"
+depends on DRM_XE
+depends on EXPERT
+source "drivers/gpu/drm/xe/Kconfig.debug"
+endmenu
+
+menu "drm/xe Profile Guided Optimisation"
+ visible if EXPERT
+ depends on DRM_XE
+ source "drivers/gpu/drm/xe/Kconfig.profile"
+endmenu
diff --git a/drivers/gpu/drm/xe/Kconfig.debug b/drivers/gpu/drm/xe/Kconfig.debug
new file mode 100644
index 000000000000..549065f57a78
--- /dev/null
+++ b/drivers/gpu/drm/xe/Kconfig.debug
@@ -0,0 +1,107 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config DRM_XE_WERROR
+ bool "Force GCC to throw an error instead of a warning when compiling"
+ # As this may inadvertently break the build, only allow the user
+ # to shoot oneself in the foot iff they aim really hard
+ depends on EXPERT
+ # We use the dependency on !COMPILE_TEST to not be enabled in
+ # allmodconfig or allyesconfig configurations
+ depends on !COMPILE_TEST
+ default n
+ help
+ Add -Werror to the build flags for (and only for) xe.ko.
+ Do not enable this unless you are writing code for the xe.ko module.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_DEBUG
+ bool "Enable additional driver debugging"
+ depends on DRM_XE
+ depends on EXPERT
+ depends on !COMPILE_TEST
+ default n
+ help
+ Choose this option to turn on extra driver debugging that may affect
+ performance but will catch some internal issues.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_DEBUG_VM
+ bool "Enable extra VM debugging info"
+ default n
+ help
+ Enable extra VM debugging info
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_DEBUG_SRIOV
+ bool "Enable extra SR-IOV debugging"
+ default n
+ help
+ Enable extra SR-IOV debugging info.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_DEBUG_MEM
+ bool "Enable passing SYS/VRAM addresses to user space"
+ default n
+ help
+ Pass object location trough uapi. Intended for extended
+ testing and development only.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_SIMPLE_ERROR_CAPTURE
+ bool "Enable simple error capture to dmesg on job timeout"
+ default n
+ help
+ Choose this option when debugging an unexpected job timeout
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_KUNIT_TEST
+ tristate "KUnit tests for the drm xe driver" if !KUNIT_ALL_TESTS
+ depends on DRM_XE && KUNIT && DEBUG_FS
+ default KUNIT_ALL_TESTS
+ select DRM_EXPORT_FOR_TESTS if m
+ select DRM_KUNIT_TEST_HELPERS
+ help
+ Choose this option to allow the driver to perform selftests under
+ the kunit framework
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_LARGE_GUC_BUFFER
+ bool "Enable larger guc log buffer"
+ default n
+ help
+ Choose this option when debugging guc issues.
+ Buffer should be large enough for complex issues.
+
+ Recommended for driver developers only.
+
+ If in doubt, say "N".
+
+config DRM_XE_USERPTR_INVAL_INJECT
+ bool "Inject userptr invalidation -EINVAL errors"
+ default n
+ help
+ Choose this option when debugging error paths that
+ are hit during checks for userptr invalidations.
+
+ Recomended for driver developers only.
+ If in doubt, say "N".
diff --git a/drivers/gpu/drm/xe/Kconfig.profile b/drivers/gpu/drm/xe/Kconfig.profile
new file mode 100644
index 000000000000..ba17a25e8db3
--- /dev/null
+++ b/drivers/gpu/drm/xe/Kconfig.profile
@@ -0,0 +1,54 @@
+config DRM_XE_JOB_TIMEOUT_MAX
+ int "Default max job timeout (ms)"
+ default 10000 # milliseconds
+ help
+ Configures the default max job timeout after which job will
+ be forcefully taken away from scheduler.
+config DRM_XE_JOB_TIMEOUT_MIN
+ int "Default min job timeout (ms)"
+ default 1 # milliseconds
+ help
+ Configures the default min job timeout after which job will
+ be forcefully taken away from scheduler.
+config DRM_XE_TIMESLICE_MAX
+ int "Default max timeslice duration (us)"
+ default 10000000 # microseconds
+ help
+ Configures the default max timeslice duration between multiple
+ contexts by guc scheduling.
+config DRM_XE_TIMESLICE_MIN
+ int "Default min timeslice duration (us)"
+ default 1 # microseconds
+ help
+ Configures the default min timeslice duration between multiple
+ contexts by guc scheduling.
+config DRM_XE_PREEMPT_TIMEOUT
+ int "Preempt timeout (us, jiffy granularity)"
+ default 640000 # microseconds
+ help
+ How long to wait (in microseconds) for a preemption event to occur
+ when submitting a new context. If the current context does not hit
+ an arbitration point and yield to HW before the timer expires, the
+ HW will be reset to allow the more important context to execute.
+config DRM_XE_PREEMPT_TIMEOUT_MAX
+ int "Default max preempt timeout (us)"
+ default 10000000 # microseconds
+ help
+ Configures the default max preempt timeout after which context
+ will be forcefully taken away and higher priority context will
+ run.
+config DRM_XE_PREEMPT_TIMEOUT_MIN
+ int "Default min preempt timeout (us)"
+ default 1 # microseconds
+ help
+ Configures the default min preempt timeout after which context
+ will be forcefully taken away and higher priority context will
+ run.
+config DRM_XE_ENABLE_SCHEDTIMEOUT_LIMIT
+ bool "Default configuration of limitation on scheduler timeout"
+ default y
+ help
+ Configures the enablement of limitation on scheduler timeout
+ to apply to applicable user. For elevated user, all above MIN
+ and MAX values will apply when this configuration is enable to
+ apply limitation. By default limitation is applied.
diff --git a/drivers/gpu/drm/xe/Makefile b/drivers/gpu/drm/xe/Makefile
new file mode 100644
index 000000000000..53bd2a8ba1ae
--- /dev/null
+++ b/drivers/gpu/drm/xe/Makefile
@@ -0,0 +1,305 @@
+# SPDX-License-Identifier: GPL-2.0
+#
+# Makefile for the drm device driver. This driver provides support for the
+# Direct Rendering Infrastructure (DRI) in XFree86 4.1.0 and higher.
+
+# Unconditionally enable W=1 warnings locally
+# --- begin copy-paste W=1 warnings from scripts/Makefile.extrawarn
+subdir-ccflags-y += -Wextra -Wunused -Wno-unused-parameter
+subdir-ccflags-y += -Wmissing-declarations
+subdir-ccflags-y += $(call cc-option, -Wrestrict)
+subdir-ccflags-y += -Wmissing-format-attribute
+subdir-ccflags-y += -Wmissing-prototypes
+subdir-ccflags-y += -Wold-style-definition
+subdir-ccflags-y += -Wmissing-include-dirs
+subdir-ccflags-y += $(call cc-option, -Wunused-but-set-variable)
+subdir-ccflags-y += $(call cc-option, -Wunused-const-variable)
+subdir-ccflags-y += $(call cc-option, -Wpacked-not-aligned)
+subdir-ccflags-y += $(call cc-option, -Wformat-overflow)
+subdir-ccflags-y += $(call cc-option, -Wformat-truncation)
+subdir-ccflags-y += $(call cc-option, -Wstringop-overflow)
+subdir-ccflags-y += $(call cc-option, -Wstringop-truncation)
+# The following turn off the warnings enabled by -Wextra
+ifeq ($(findstring 2, $(KBUILD_EXTRA_WARN)),)
+subdir-ccflags-y += -Wno-missing-field-initializers
+subdir-ccflags-y += -Wno-type-limits
+subdir-ccflags-y += -Wno-shift-negative-value
+endif
+ifeq ($(findstring 3, $(KBUILD_EXTRA_WARN)),)
+subdir-ccflags-y += -Wno-sign-compare
+endif
+# --- end copy-paste
+
+# Enable -Werror in CI and development
+subdir-ccflags-$(CONFIG_DRM_XE_WERROR) += -Werror
+
+subdir-ccflags-y += -I$(obj) -I$(srctree)/$(src)
+
+# generated sources
+hostprogs := xe_gen_wa_oob
+
+generated_oob := $(obj)/generated/xe_wa_oob.c $(obj)/generated/xe_wa_oob.h
+
+quiet_cmd_wa_oob = GEN $(notdir $(generated_oob))
+ cmd_wa_oob = mkdir -p $(@D); $^ $(generated_oob)
+
+$(generated_oob) &: $(obj)/xe_gen_wa_oob $(srctree)/$(src)/xe_wa_oob.rules
+ $(call cmd,wa_oob)
+
+uses_generated_oob := \
+ $(obj)/xe_gsc.o \
+ $(obj)/xe_guc.o \
+ $(obj)/xe_migrate.o \
+ $(obj)/xe_ring_ops.o \
+ $(obj)/xe_vm.o \
+ $(obj)/xe_wa.o \
+ $(obj)/xe_ttm_stolen_mgr.o
+
+$(uses_generated_oob): $(generated_oob)
+
+# Please keep these build lists sorted!
+
+# core driver code
+
+xe-y += xe_bb.o \
+ xe_bo.o \
+ xe_bo_evict.o \
+ xe_debugfs.o \
+ xe_devcoredump.o \
+ xe_device.o \
+ xe_device_sysfs.o \
+ xe_dma_buf.o \
+ xe_drm_client.o \
+ xe_exec.o \
+ xe_execlist.o \
+ xe_exec_queue.o \
+ xe_force_wake.o \
+ xe_ggtt.o \
+ xe_gpu_scheduler.o \
+ xe_gsc.o \
+ xe_gsc_submit.o \
+ xe_gt.o \
+ xe_gt_ccs_mode.o \
+ xe_gt_clock.o \
+ xe_gt_debugfs.o \
+ xe_gt_freq.o \
+ xe_gt_idle.o \
+ xe_gt_mcr.o \
+ xe_gt_pagefault.o \
+ xe_gt_sysfs.o \
+ xe_gt_throttle_sysfs.o \
+ xe_gt_tlb_invalidation.o \
+ xe_gt_topology.o \
+ xe_guc.o \
+ xe_guc_ads.o \
+ xe_guc_ct.o \
+ xe_guc_debugfs.o \
+ xe_guc_hwconfig.o \
+ xe_guc_log.o \
+ xe_guc_pc.o \
+ xe_guc_submit.o \
+ xe_heci_gsc.o \
+ xe_hw_engine.o \
+ xe_hw_engine_class_sysfs.o \
+ xe_hw_fence.o \
+ xe_huc.o \
+ xe_huc_debugfs.o \
+ xe_irq.o \
+ xe_lrc.o \
+ xe_migrate.o \
+ xe_mmio.o \
+ xe_mocs.o \
+ xe_module.o \
+ xe_pat.o \
+ xe_pci.o \
+ xe_pcode.o \
+ xe_pm.o \
+ xe_preempt_fence.o \
+ xe_pt.o \
+ xe_pt_walk.o \
+ xe_query.o \
+ xe_range_fence.o \
+ xe_reg_sr.o \
+ xe_reg_whitelist.o \
+ xe_rtp.o \
+ xe_ring_ops.o \
+ xe_sa.o \
+ xe_sched_job.o \
+ xe_step.o \
+ xe_sync.o \
+ xe_tile.o \
+ xe_tile_sysfs.o \
+ xe_trace.o \
+ xe_ttm_sys_mgr.o \
+ xe_ttm_stolen_mgr.o \
+ xe_ttm_vram_mgr.o \
+ xe_tuning.o \
+ xe_uc.o \
+ xe_uc_debugfs.o \
+ xe_uc_fw.o \
+ xe_vm.o \
+ xe_wait_user_fence.o \
+ xe_wa.o \
+ xe_wopcm.o
+
+# graphics hardware monitoring (HWMON) support
+xe-$(CONFIG_HWMON) += xe_hwmon.o
+
+# graphics virtualization (SR-IOV) support
+xe-y += xe_sriov.o
+
+xe-$(CONFIG_PCI_IOV) += \
+ xe_lmtt.o \
+ xe_lmtt_2l.o \
+ xe_lmtt_ml.o
+
+# i915 Display compat #defines and #includes
+subdir-ccflags-$(CONFIG_DRM_XE_DISPLAY) += \
+ -I$(srctree)/$(src)/display/ext \
+ -I$(srctree)/$(src)/compat-i915-headers \
+ -I$(srctree)/drivers/gpu/drm/xe/display/ \
+ -I$(srctree)/drivers/gpu/drm/i915/display/ \
+ -Ddrm_i915_gem_object=xe_bo \
+ -Ddrm_i915_private=xe_device
+
+CFLAGS_i915-display/intel_fbdev.o = $(call cc-disable-warning, override-init)
+CFLAGS_i915-display/intel_display_device.o = $(call cc-disable-warning, override-init)
+
+# Rule to build SOC code shared with i915
+$(obj)/i915-soc/%.o: $(srctree)/drivers/gpu/drm/i915/soc/%.c FORCE
+ $(call cmd,force_checksrc)
+ $(call if_changed_rule,cc_o_c)
+
+# Rule to build display code shared with i915
+$(obj)/i915-display/%.o: $(srctree)/drivers/gpu/drm/i915/display/%.c FORCE
+ $(call cmd,force_checksrc)
+ $(call if_changed_rule,cc_o_c)
+
+# Display code specific to xe
+xe-$(CONFIG_DRM_XE_DISPLAY) += \
+ xe_display.o \
+ display/xe_fb_pin.o \
+ display/xe_hdcp_gsc.o \
+ display/xe_plane_initial.o \
+ display/xe_display_rps.o \
+ display/xe_display_misc.o \
+ display/xe_dsb_buffer.o \
+ display/intel_fbdev_fb.o \
+ display/intel_fb_bo.o \
+ display/ext/i915_irq.o \
+ display/ext/i915_utils.o
+
+# SOC code shared with i915
+xe-$(CONFIG_DRM_XE_DISPLAY) += \
+ i915-soc/intel_dram.o \
+ i915-soc/intel_pch.o
+
+# Display code shared with i915
+xe-$(CONFIG_DRM_XE_DISPLAY) += \
+ i915-display/icl_dsi.o \
+ i915-display/intel_atomic.o \
+ i915-display/intel_atomic_plane.o \
+ i915-display/intel_audio.o \
+ i915-display/intel_backlight.o \
+ i915-display/intel_bios.o \
+ i915-display/intel_bw.o \
+ i915-display/intel_cdclk.o \
+ i915-display/intel_color.o \
+ i915-display/intel_combo_phy.o \
+ i915-display/intel_connector.o \
+ i915-display/intel_crtc.o \
+ i915-display/intel_crtc_state_dump.o \
+ i915-display/intel_cursor.o \
+ i915-display/intel_cx0_phy.o \
+ i915-display/intel_ddi.o \
+ i915-display/intel_ddi_buf_trans.o \
+ i915-display/intel_display.o \
+ i915-display/intel_display_debugfs.o \
+ i915-display/intel_display_debugfs_params.o \
+ i915-display/intel_display_device.o \
+ i915-display/intel_display_driver.o \
+ i915-display/intel_display_irq.o \
+ i915-display/intel_display_params.o \
+ i915-display/intel_display_power.o \
+ i915-display/intel_display_power_map.o \
+ i915-display/intel_display_power_well.o \
+ i915-display/intel_display_trace.o \
+ i915-display/intel_display_wa.o \
+ i915-display/intel_dkl_phy.o \
+ i915-display/intel_dmc.o \
+ i915-display/intel_dp.o \
+ i915-display/intel_dp_aux.o \
+ i915-display/intel_dp_aux_backlight.o \
+ i915-display/intel_dp_hdcp.o \
+ i915-display/intel_dp_link_training.o \
+ i915-display/intel_dp_mst.o \
+ i915-display/intel_dpll.o \
+ i915-display/intel_dpll_mgr.o \
+ i915-display/intel_dpt_common.o \
+ i915-display/intel_drrs.o \
+ i915-display/intel_dsb.o \
+ i915-display/intel_dsi.o \
+ i915-display/intel_dsi_dcs_backlight.o \
+ i915-display/intel_dsi_vbt.o \
+ i915-display/intel_fb.o \
+ i915-display/intel_fbc.o \
+ i915-display/intel_fdi.o \
+ i915-display/intel_fifo_underrun.o \
+ i915-display/intel_frontbuffer.o \
+ i915-display/intel_global_state.o \
+ i915-display/intel_gmbus.o \
+ i915-display/intel_hdcp.o \
+ i915-display/intel_hdmi.o \
+ i915-display/intel_hotplug.o \
+ i915-display/intel_hotplug_irq.o \
+ i915-display/intel_hti.o \
+ i915-display/intel_link_bw.o \
+ i915-display/intel_lspcon.o \
+ i915-display/intel_modeset_lock.o \
+ i915-display/intel_modeset_setup.o \
+ i915-display/intel_modeset_verify.o \
+ i915-display/intel_panel.o \
+ i915-display/intel_pipe_crc.o \
+ i915-display/intel_pmdemand.o \
+ i915-display/intel_pps.o \
+ i915-display/intel_psr.o \
+ i915-display/intel_qp_tables.o \
+ i915-display/intel_quirks.o \
+ i915-display/intel_snps_phy.o \
+ i915-display/intel_tc.o \
+ i915-display/intel_vblank.o \
+ i915-display/intel_vdsc.o \
+ i915-display/intel_vga.o \
+ i915-display/intel_vrr.o \
+ i915-display/intel_wm.o \
+ i915-display/skl_scaler.o \
+ i915-display/skl_universal_plane.o \
+ i915-display/skl_watermark.o
+
+ifeq ($(CONFIG_ACPI),y)
+ xe-$(CONFIG_DRM_XE_DISPLAY) += \
+ i915-display/intel_acpi.o \
+ i915-display/intel_opregion.o
+endif
+
+ifeq ($(CONFIG_DRM_FBDEV_EMULATION),y)
+ xe-$(CONFIG_DRM_XE_DISPLAY) += i915-display/intel_fbdev.o
+endif
+
+obj-$(CONFIG_DRM_XE) += xe.o
+obj-$(CONFIG_DRM_XE_KUNIT_TEST) += tests/
+
+# header test
+hdrtest_find_args := -not -path xe_rtp_helpers.h
+ifneq ($(CONFIG_DRM_XE_DISPLAY),y)
+ hdrtest_find_args += -not -path display/\* -not -path compat-i915-headers/\* -not -path xe_display.h
+endif
+
+always-$(CONFIG_DRM_XE_WERROR) += \
+ $(patsubst %.h,%.hdrtest, $(shell cd $(srctree)/$(src) && find * -name '*.h' $(hdrtest_find_args)))
+
+quiet_cmd_hdrtest = HDRTEST $(patsubst %.hdrtest,%.h,$@)
+ cmd_hdrtest = $(CC) -DHDRTEST $(filter-out $(CFLAGS_GCOV), $(c_flags)) -S -o /dev/null -x c /dev/null -include $<; touch $@
+
+$(obj)/%.hdrtest: $(src)/%.h FORCE
+ $(call if_changed_dep,hdrtest)
diff --git a/drivers/gpu/drm/xe/abi/gsc_command_header_abi.h b/drivers/gpu/drm/xe/abi/gsc_command_header_abi.h
new file mode 100644
index 000000000000..a4c2646803b5
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/gsc_command_header_abi.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _ABI_GSC_COMMAND_HEADER_ABI_H
+#define _ABI_GSC_COMMAND_HEADER_ABI_H
+
+#include <linux/types.h>
+
+struct intel_gsc_mtl_header {
+ u32 validity_marker;
+#define GSC_HECI_VALIDITY_MARKER 0xA578875A
+
+ u8 heci_client_id;
+
+ u8 reserved1;
+
+ u16 header_version;
+#define MTL_GSC_HEADER_VERSION 1
+
+ /* FW allows host to decide host_session handle as it sees fit. */
+ u64 host_session_handle;
+
+ /* handle generated by FW for messages that need to be re-submitted */
+ u64 gsc_message_handle;
+
+ u32 message_size; /* lower 20 bits only, upper 12 are reserved */
+
+ /*
+ * Flags mask:
+ * Bit 0: Pending
+ * Bit 1: Session Cleanup;
+ * Bits 2-15: Flags
+ * Bits 16-31: Extension Size
+ * According to internal spec flags are either input or output
+ * we distinguish the flags using OUTFLAG or INFLAG
+ */
+ u32 flags;
+#define GSC_OUTFLAG_MSG_PENDING BIT(0)
+#define GSC_INFLAG_MSG_CLEANUP BIT(1)
+
+ u32 status;
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/gsc_mkhi_commands_abi.h b/drivers/gpu/drm/xe/abi/gsc_mkhi_commands_abi.h
new file mode 100644
index 000000000000..ad4d041873ab
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/gsc_mkhi_commands_abi.h
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _ABI_GSC_MKHI_COMMANDS_ABI_H
+#define _ABI_GSC_MKHI_COMMANDS_ABI_H
+
+#include <linux/types.h>
+
+/* Heci client ID for MKHI commands */
+#define HECI_MEADDRESS_MKHI 7
+
+/* Generic MKHI header */
+struct gsc_mkhi_header {
+ u8 group_id;
+ u8 command;
+ u8 reserved;
+ u8 result;
+} __packed;
+
+/* GFX_SRV commands */
+#define MKHI_GROUP_ID_GFX_SRV 0x30
+
+#define MKHI_GFX_SRV_GET_HOST_COMPATIBILITY_VERSION (0x42)
+
+struct gsc_get_compatibility_version_in {
+ struct gsc_mkhi_header header;
+} __packed;
+
+struct gsc_get_compatibility_version_out {
+ struct gsc_mkhi_header header;
+ u16 proj_major;
+ u16 compat_major;
+ u16 compat_minor;
+ u16 reserved[5];
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/gsc_pxp_commands_abi.h b/drivers/gpu/drm/xe/abi/gsc_pxp_commands_abi.h
new file mode 100644
index 000000000000..57520809e48d
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/gsc_pxp_commands_abi.h
@@ -0,0 +1,59 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _ABI_GSC_PXP_COMMANDS_ABI_H
+#define _ABI_GSC_PXP_COMMANDS_ABI_H
+
+#include <linux/types.h>
+
+/* Heci client ID for PXP commands */
+#define HECI_MEADDRESS_PXP 17
+
+#define PXP_APIVER(x, y) (((x) & 0xFFFF) << 16 | ((y) & 0xFFFF))
+
+/*
+ * there are a lot of status codes for PXP, but we only define the cross-API
+ * common ones that we actually can handle in the kernel driver. Other failure
+ * codes should be printed to error msg for debug.
+ */
+enum pxp_status {
+ PXP_STATUS_SUCCESS = 0x0,
+ PXP_STATUS_ERROR_API_VERSION = 0x1002,
+ PXP_STATUS_NOT_READY = 0x100e,
+ PXP_STATUS_PLATFCONFIG_KF1_NOVERIF = 0x101a,
+ PXP_STATUS_PLATFCONFIG_KF1_BAD = 0x101f,
+ PXP_STATUS_OP_NOT_PERMITTED = 0x4013
+};
+
+/* Common PXP FW message header */
+struct pxp_cmd_header {
+ u32 api_version;
+ u32 command_id;
+ union {
+ u32 status; /* out */
+ u32 stream_id; /* in */
+#define PXP_CMDHDR_EXTDATA_SESSION_VALID GENMASK(0, 0)
+#define PXP_CMDHDR_EXTDATA_APP_TYPE GENMASK(1, 1)
+#define PXP_CMDHDR_EXTDATA_SESSION_ID GENMASK(17, 2)
+ };
+ /* Length of the message (excluding the header) */
+ u32 buffer_len;
+} __packed;
+
+#define PXP43_CMDID_NEW_HUC_AUTH 0x0000003F /* MTL+ */
+
+/* PXP-Input-Packet: HUC Auth-only */
+struct pxp43_new_huc_auth_in {
+ struct pxp_cmd_header header;
+ u64 huc_base_address;
+ u32 huc_size;
+} __packed;
+
+/* PXP-Output-Packet: HUC Load and Authentication or Auth-only */
+struct pxp43_huc_auth_out {
+ struct pxp_cmd_header header;
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_actions_abi.h b/drivers/gpu/drm/xe/abi/guc_actions_abi.h
new file mode 100644
index 000000000000..3062e0e0d467
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_actions_abi.h
@@ -0,0 +1,219 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2014-2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_ACTIONS_ABI_H
+#define _ABI_GUC_ACTIONS_ABI_H
+
+/**
+ * DOC: HOST2GUC_SELF_CFG
+ *
+ * This message is used by Host KMD to setup of the `GuC Self Config KLVs`_.
+ *
+ * This message must be sent as `MMIO HXG Message`_.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN = GUC_HXG_ORIGIN_HOST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_REQUEST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | DATA0 = MBZ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | ACTION = _`GUC_ACTION_HOST2GUC_SELF_CFG` = 0x0508 |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:16 | **KLV_KEY** - KLV key, see `GuC Self Config KLVs`_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | **KLV_LEN** - KLV length |
+ * | | | |
+ * | | | - 32 bit KLV = 1 |
+ * | | | - 64 bit KLV = 2 |
+ * +---+-------+--------------------------------------------------------------+
+ * | 2 | 31:0 | **VALUE32** - Bits 31-0 of the KLV value |
+ * +---+-------+--------------------------------------------------------------+
+ * | 3 | 31:0 | **VALUE64** - Bits 63-32 of the KLV value (**KLV_LEN** = 2) |
+ * +---+-------+--------------------------------------------------------------+
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN = GUC_HXG_ORIGIN_GUC_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_RESPONSE_SUCCESS_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | DATA0 = **NUM** - 1 if KLV was parsed, 0 if not recognized |
+ * +---+-------+--------------------------------------------------------------+
+ */
+#define GUC_ACTION_HOST2GUC_SELF_CFG 0x0508
+
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_LEN (GUC_HXG_REQUEST_MSG_MIN_LEN + 3u)
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_0_MBZ GUC_HXG_REQUEST_MSG_0_DATA0
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY (0xffff << 16)
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN (0xffff << 0)
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_2_VALUE32 GUC_HXG_REQUEST_MSG_n_DATAn
+#define HOST2GUC_SELF_CFG_REQUEST_MSG_3_VALUE64 GUC_HXG_REQUEST_MSG_n_DATAn
+
+#define HOST2GUC_SELF_CFG_RESPONSE_MSG_LEN GUC_HXG_RESPONSE_MSG_MIN_LEN
+#define HOST2GUC_SELF_CFG_RESPONSE_MSG_0_NUM GUC_HXG_RESPONSE_MSG_0_DATA0
+
+/**
+ * DOC: HOST2GUC_CONTROL_CTB
+ *
+ * This H2G action allows Vf Host to enable or disable H2G and G2H `CT Buffer`_.
+ *
+ * This message must be sent as `MMIO HXG Message`_.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN = GUC_HXG_ORIGIN_HOST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_REQUEST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | DATA0 = MBZ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | ACTION = _`GUC_ACTION_HOST2GUC_CONTROL_CTB` = 0x4509 |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | **CONTROL** - control `CTB based communication`_ |
+ * | | | |
+ * | | | - _`GUC_CTB_CONTROL_DISABLE` = 0 |
+ * | | | - _`GUC_CTB_CONTROL_ENABLE` = 1 |
+ * +---+-------+--------------------------------------------------------------+
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN = GUC_HXG_ORIGIN_GUC_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_RESPONSE_SUCCESS_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | DATA0 = MBZ |
+ * +---+-------+--------------------------------------------------------------+
+ */
+#define GUC_ACTION_HOST2GUC_CONTROL_CTB 0x4509
+
+#define HOST2GUC_CONTROL_CTB_REQUEST_MSG_LEN (GUC_HXG_REQUEST_MSG_MIN_LEN + 1u)
+#define HOST2GUC_CONTROL_CTB_REQUEST_MSG_0_MBZ GUC_HXG_REQUEST_MSG_0_DATA0
+#define HOST2GUC_CONTROL_CTB_REQUEST_MSG_1_CONTROL GUC_HXG_REQUEST_MSG_n_DATAn
+#define GUC_CTB_CONTROL_DISABLE 0u
+#define GUC_CTB_CONTROL_ENABLE 1u
+
+#define HOST2GUC_CONTROL_CTB_RESPONSE_MSG_LEN GUC_HXG_RESPONSE_MSG_MIN_LEN
+#define HOST2GUC_CONTROL_CTB_RESPONSE_MSG_0_MBZ GUC_HXG_RESPONSE_MSG_0_DATA0
+
+/* legacy definitions */
+
+enum xe_guc_action {
+ XE_GUC_ACTION_DEFAULT = 0x0,
+ XE_GUC_ACTION_REQUEST_PREEMPTION = 0x2,
+ XE_GUC_ACTION_REQUEST_ENGINE_RESET = 0x3,
+ XE_GUC_ACTION_ALLOCATE_DOORBELL = 0x10,
+ XE_GUC_ACTION_DEALLOCATE_DOORBELL = 0x20,
+ XE_GUC_ACTION_LOG_BUFFER_FILE_FLUSH_COMPLETE = 0x30,
+ XE_GUC_ACTION_UK_LOG_ENABLE_LOGGING = 0x40,
+ XE_GUC_ACTION_FORCE_LOG_BUFFER_FLUSH = 0x302,
+ XE_GUC_ACTION_ENTER_S_STATE = 0x501,
+ XE_GUC_ACTION_EXIT_S_STATE = 0x502,
+ XE_GUC_ACTION_GLOBAL_SCHED_POLICY_CHANGE = 0x506,
+ XE_GUC_ACTION_SCHED_CONTEXT = 0x1000,
+ XE_GUC_ACTION_SCHED_CONTEXT_MODE_SET = 0x1001,
+ XE_GUC_ACTION_SCHED_CONTEXT_MODE_DONE = 0x1002,
+ XE_GUC_ACTION_SCHED_ENGINE_MODE_SET = 0x1003,
+ XE_GUC_ACTION_SCHED_ENGINE_MODE_DONE = 0x1004,
+ XE_GUC_ACTION_SET_CONTEXT_PRIORITY = 0x1005,
+ XE_GUC_ACTION_SET_CONTEXT_EXECUTION_QUANTUM = 0x1006,
+ XE_GUC_ACTION_SET_CONTEXT_PREEMPTION_TIMEOUT = 0x1007,
+ XE_GUC_ACTION_CONTEXT_RESET_NOTIFICATION = 0x1008,
+ XE_GUC_ACTION_ENGINE_FAILURE_NOTIFICATION = 0x1009,
+ XE_GUC_ACTION_HOST2GUC_UPDATE_CONTEXT_POLICIES = 0x100B,
+ XE_GUC_ACTION_SETUP_PC_GUCRC = 0x3004,
+ XE_GUC_ACTION_AUTHENTICATE_HUC = 0x4000,
+ XE_GUC_ACTION_GET_HWCONFIG = 0x4100,
+ XE_GUC_ACTION_REGISTER_CONTEXT = 0x4502,
+ XE_GUC_ACTION_DEREGISTER_CONTEXT = 0x4503,
+ XE_GUC_ACTION_REGISTER_COMMAND_TRANSPORT_BUFFER = 0x4505,
+ XE_GUC_ACTION_DEREGISTER_COMMAND_TRANSPORT_BUFFER = 0x4506,
+ XE_GUC_ACTION_DEREGISTER_CONTEXT_DONE = 0x4600,
+ XE_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC = 0x4601,
+ XE_GUC_ACTION_CLIENT_SOFT_RESET = 0x5507,
+ XE_GUC_ACTION_SET_ENG_UTIL_BUFF = 0x550A,
+ XE_GUC_ACTION_NOTIFY_MEMORY_CAT_ERROR = 0x6000,
+ XE_GUC_ACTION_REPORT_PAGE_FAULT_REQ_DESC = 0x6002,
+ XE_GUC_ACTION_PAGE_FAULT_RES_DESC = 0x6003,
+ XE_GUC_ACTION_ACCESS_COUNTER_NOTIFY = 0x6004,
+ XE_GUC_ACTION_TLB_INVALIDATION = 0x7000,
+ XE_GUC_ACTION_TLB_INVALIDATION_DONE = 0x7001,
+ XE_GUC_ACTION_TLB_INVALIDATION_ALL = 0x7002,
+ XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION = 0x8002,
+ XE_GUC_ACTION_NOTIFY_FLUSH_LOG_BUFFER_TO_FILE = 0x8003,
+ XE_GUC_ACTION_NOTIFY_CRASH_DUMP_POSTED = 0x8004,
+ XE_GUC_ACTION_NOTIFY_EXCEPTION = 0x8005,
+ XE_GUC_ACTION_LIMIT
+};
+
+enum xe_guc_rc_options {
+ XE_GUCRC_HOST_CONTROL,
+ XE_GUCRC_FIRMWARE_CONTROL,
+};
+
+enum xe_guc_preempt_options {
+ XE_GUC_PREEMPT_OPTION_DROP_WORK_Q = 0x4,
+ XE_GUC_PREEMPT_OPTION_DROP_SUBMIT_Q = 0x8,
+};
+
+enum xe_guc_report_status {
+ XE_GUC_REPORT_STATUS_UNKNOWN = 0x0,
+ XE_GUC_REPORT_STATUS_ACKED = 0x1,
+ XE_GUC_REPORT_STATUS_ERROR = 0x2,
+ XE_GUC_REPORT_STATUS_COMPLETE = 0x4,
+};
+
+enum xe_guc_sleep_state_status {
+ XE_GUC_SLEEP_STATE_SUCCESS = 0x1,
+ XE_GUC_SLEEP_STATE_PREEMPT_TO_IDLE_FAILED = 0x2,
+ XE_GUC_SLEEP_STATE_ENGINE_RESET_FAILED = 0x3
+#define XE_GUC_SLEEP_STATE_INVALID_MASK 0x80000000
+};
+
+#define GUC_LOG_CONTROL_LOGGING_ENABLED (1 << 0)
+#define GUC_LOG_CONTROL_VERBOSITY_SHIFT 4
+#define GUC_LOG_CONTROL_VERBOSITY_MASK (0xF << GUC_LOG_CONTROL_VERBOSITY_SHIFT)
+#define GUC_LOG_CONTROL_DEFAULT_LOGGING (1 << 8)
+
+#define XE_GUC_TLB_INVAL_TYPE_SHIFT 0
+#define XE_GUC_TLB_INVAL_MODE_SHIFT 8
+/* Flush PPC or SMRO caches along with TLB invalidation request */
+#define XE_GUC_TLB_INVAL_FLUSH_CACHE (1 << 31)
+
+enum xe_guc_tlb_invalidation_type {
+ XE_GUC_TLB_INVAL_FULL = 0x0,
+ XE_GUC_TLB_INVAL_PAGE_SELECTIVE = 0x1,
+ XE_GUC_TLB_INVAL_PAGE_SELECTIVE_CTX = 0x2,
+ XE_GUC_TLB_INVAL_GUC = 0x3,
+};
+
+/*
+ * 0: Heavy mode of Invalidation:
+ * The pipeline of the engine(s) for which the invalidation is targeted to is
+ * blocked, and all the in-flight transactions are guaranteed to be Globally
+ * Observed before completing the TLB invalidation
+ * 1: Lite mode of Invalidation:
+ * TLBs of the targeted engine(s) are immediately invalidated.
+ * In-flight transactions are NOT guaranteed to be Globally Observed before
+ * completing TLB invalidation.
+ * Light Invalidation Mode is to be used only when
+ * it can be guaranteed (by SW) that the address translations remain invariant
+ * for the in-flight transactions across the TLB invalidation. In other words,
+ * this mode can be used when the TLB invalidation is intended to clear out the
+ * stale cached translations that are no longer in use. Light Invalidation Mode
+ * is much faster than the Heavy Invalidation Mode, as it does not wait for the
+ * in-flight transactions to be GOd.
+ */
+enum xe_guc_tlb_inval_mode {
+ XE_GUC_TLB_INVAL_MODE_HEAVY = 0x0,
+ XE_GUC_TLB_INVAL_MODE_LITE = 0x1,
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h b/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h
new file mode 100644
index 000000000000..811add10c30d
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_actions_slpc_abi.h
@@ -0,0 +1,249 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _GUC_ACTIONS_SLPC_ABI_H_
+#define _GUC_ACTIONS_SLPC_ABI_H_
+
+#include <linux/types.h>
+
+/**
+ * DOC: SLPC SHARED DATA STRUCTURE
+ *
+ * +----+------+--------------------------------------------------------------+
+ * | CL | Bytes| Description |
+ * +====+======+==============================================================+
+ * | 1 | 0-3 | SHARED DATA SIZE |
+ * | +------+--------------------------------------------------------------+
+ * | | 4-7 | GLOBAL STATE |
+ * | +------+--------------------------------------------------------------+
+ * | | 8-11 | DISPLAY DATA ADDRESS |
+ * | +------+--------------------------------------------------------------+
+ * | | 12:63| PADDING |
+ * +----+------+--------------------------------------------------------------+
+ * | | 0:63 | PADDING(PLATFORM INFO) |
+ * +----+------+--------------------------------------------------------------+
+ * | 3 | 0-3 | TASK STATE DATA |
+ * + +------+--------------------------------------------------------------+
+ * | | 4:63 | PADDING |
+ * +----+------+--------------------------------------------------------------+
+ * |4-21|0:1087| OVERRIDE PARAMS AND BIT FIELDS |
+ * +----+------+--------------------------------------------------------------+
+ * | | | PADDING + EXTRA RESERVED PAGE |
+ * +----+------+--------------------------------------------------------------+
+ */
+
+/*
+ * SLPC exposes certain parameters for global configuration by the host.
+ * These are referred to as override parameters, because in most cases
+ * the host will not need to modify the default values used by SLPC.
+ * SLPC remembers the default values which allows the host to easily restore
+ * them by simply unsetting the override. The host can set or unset override
+ * parameters during SLPC (re-)initialization using the SLPC Reset event.
+ * The host can also set or unset override parameters on the fly using the
+ * Parameter Set and Parameter Unset events
+ */
+
+#define SLPC_MAX_OVERRIDE_PARAMETERS 256
+#define SLPC_OVERRIDE_BITFIELD_SIZE \
+ (SLPC_MAX_OVERRIDE_PARAMETERS / 32)
+
+#define SLPC_PAGE_SIZE_BYTES 4096
+#define SLPC_CACHELINE_SIZE_BYTES 64
+#define SLPC_SHARED_DATA_SIZE_BYTE_HEADER SLPC_CACHELINE_SIZE_BYTES
+#define SLPC_SHARED_DATA_SIZE_BYTE_PLATFORM_INFO SLPC_CACHELINE_SIZE_BYTES
+#define SLPC_SHARED_DATA_SIZE_BYTE_TASK_STATE SLPC_CACHELINE_SIZE_BYTES
+#define SLPC_SHARED_DATA_MODE_DEFN_TABLE_SIZE SLPC_PAGE_SIZE_BYTES
+#define SLPC_SHARED_DATA_SIZE_BYTE_MAX (2 * SLPC_PAGE_SIZE_BYTES)
+
+/*
+ * Cacheline size aligned (Total size needed for
+ * SLPM_KMD_MAX_OVERRIDE_PARAMETERS=256 is 1088 bytes)
+ */
+#define SLPC_OVERRIDE_PARAMS_TOTAL_BYTES (((((SLPC_MAX_OVERRIDE_PARAMETERS * 4) \
+ + ((SLPC_MAX_OVERRIDE_PARAMETERS / 32) * 4)) \
+ + (SLPC_CACHELINE_SIZE_BYTES - 1)) / SLPC_CACHELINE_SIZE_BYTES) * \
+ SLPC_CACHELINE_SIZE_BYTES)
+
+#define SLPC_SHARED_DATA_SIZE_BYTE_OTHER (SLPC_SHARED_DATA_SIZE_BYTE_MAX - \
+ (SLPC_SHARED_DATA_SIZE_BYTE_HEADER \
+ + SLPC_SHARED_DATA_SIZE_BYTE_PLATFORM_INFO \
+ + SLPC_SHARED_DATA_SIZE_BYTE_TASK_STATE \
+ + SLPC_OVERRIDE_PARAMS_TOTAL_BYTES \
+ + SLPC_SHARED_DATA_MODE_DEFN_TABLE_SIZE))
+
+enum slpc_task_enable {
+ SLPC_PARAM_TASK_DEFAULT = 0,
+ SLPC_PARAM_TASK_ENABLED,
+ SLPC_PARAM_TASK_DISABLED,
+ SLPC_PARAM_TASK_UNKNOWN
+};
+
+enum slpc_global_state {
+ SLPC_GLOBAL_STATE_NOT_RUNNING = 0,
+ SLPC_GLOBAL_STATE_INITIALIZING = 1,
+ SLPC_GLOBAL_STATE_RESETTING = 2,
+ SLPC_GLOBAL_STATE_RUNNING = 3,
+ SLPC_GLOBAL_STATE_SHUTTING_DOWN = 4,
+ SLPC_GLOBAL_STATE_ERROR = 5
+};
+
+enum slpc_param_id {
+ SLPC_PARAM_TASK_ENABLE_GTPERF = 0,
+ SLPC_PARAM_TASK_DISABLE_GTPERF = 1,
+ SLPC_PARAM_TASK_ENABLE_BALANCER = 2,
+ SLPC_PARAM_TASK_DISABLE_BALANCER = 3,
+ SLPC_PARAM_TASK_ENABLE_DCC = 4,
+ SLPC_PARAM_TASK_DISABLE_DCC = 5,
+ SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ = 6,
+ SLPC_PARAM_GLOBAL_MAX_GT_UNSLICE_FREQ_MHZ = 7,
+ SLPC_PARAM_GLOBAL_MIN_GT_SLICE_FREQ_MHZ = 8,
+ SLPC_PARAM_GLOBAL_MAX_GT_SLICE_FREQ_MHZ = 9,
+ SLPC_PARAM_GTPERF_THRESHOLD_MAX_FPS = 10,
+ SLPC_PARAM_GLOBAL_DISABLE_GT_FREQ_MANAGEMENT = 11,
+ SLPC_PARAM_GTPERF_ENABLE_FRAMERATE_STALLING = 12,
+ SLPC_PARAM_GLOBAL_DISABLE_RC6_MODE_CHANGE = 13,
+ SLPC_PARAM_GLOBAL_OC_UNSLICE_FREQ_MHZ = 14,
+ SLPC_PARAM_GLOBAL_OC_SLICE_FREQ_MHZ = 15,
+ SLPC_PARAM_GLOBAL_ENABLE_IA_GT_BALANCING = 16,
+ SLPC_PARAM_GLOBAL_ENABLE_ADAPTIVE_BURST_TURBO = 17,
+ SLPC_PARAM_GLOBAL_ENABLE_EVAL_MODE = 18,
+ SLPC_PARAM_GLOBAL_ENABLE_BALANCER_IN_NON_GAMING_MODE = 19,
+ SLPC_PARAM_GLOBAL_RT_MODE_TURBO_FREQ_DELTA_MHZ = 20,
+ SLPC_PARAM_PWRGATE_RC_MODE = 21,
+ SLPC_PARAM_EDR_MODE_COMPUTE_TIMEOUT_MS = 22,
+ SLPC_PARAM_EDR_QOS_FREQ_MHZ = 23,
+ SLPC_PARAM_MEDIA_FF_RATIO_MODE = 24,
+ SLPC_PARAM_ENABLE_IA_FREQ_LIMITING = 25,
+ SLPC_PARAM_STRATEGIES = 26,
+ SLPC_PARAM_POWER_PROFILE = 27,
+ SLPC_PARAM_IGNORE_EFFICIENT_FREQUENCY = 28,
+ SLPC_MAX_PARAM = 32,
+};
+
+enum slpc_media_ratio_mode {
+ SLPC_MEDIA_RATIO_MODE_DYNAMIC_CONTROL = 0,
+ SLPC_MEDIA_RATIO_MODE_FIXED_ONE_TO_ONE = 1,
+ SLPC_MEDIA_RATIO_MODE_FIXED_ONE_TO_TWO = 2,
+};
+
+enum slpc_gucrc_mode {
+ SLPC_GUCRC_MODE_HW = 0,
+ SLPC_GUCRC_MODE_GUCRC_NO_RC6 = 1,
+ SLPC_GUCRC_MODE_GUCRC_STATIC_TIMEOUT = 2,
+ SLPC_GUCRC_MODE_GUCRC_DYNAMIC_HYSTERESIS = 3,
+
+ SLPC_GUCRC_MODE_MAX,
+};
+
+enum slpc_event_id {
+ SLPC_EVENT_RESET = 0,
+ SLPC_EVENT_SHUTDOWN = 1,
+ SLPC_EVENT_PLATFORM_INFO_CHANGE = 2,
+ SLPC_EVENT_DISPLAY_MODE_CHANGE = 3,
+ SLPC_EVENT_FLIP_COMPLETE = 4,
+ SLPC_EVENT_QUERY_TASK_STATE = 5,
+ SLPC_EVENT_PARAMETER_SET = 6,
+ SLPC_EVENT_PARAMETER_UNSET = 7,
+};
+
+struct slpc_task_state_data {
+ union {
+ u32 task_status_padding;
+ struct {
+ u32 status;
+#define SLPC_GTPERF_TASK_ENABLED REG_BIT(0)
+#define SLPC_DCC_TASK_ENABLED REG_BIT(11)
+#define SLPC_IN_DCC REG_BIT(12)
+#define SLPC_BALANCER_ENABLED REG_BIT(15)
+#define SLPC_IBC_TASK_ENABLED REG_BIT(16)
+#define SLPC_BALANCER_IA_LMT_ENABLED REG_BIT(17)
+#define SLPC_BALANCER_IA_LMT_ACTIVE REG_BIT(18)
+ };
+ };
+ union {
+ u32 freq_padding;
+ struct {
+#define SLPC_MAX_UNSLICE_FREQ_MASK REG_GENMASK(7, 0)
+#define SLPC_MIN_UNSLICE_FREQ_MASK REG_GENMASK(15, 8)
+#define SLPC_MAX_SLICE_FREQ_MASK REG_GENMASK(23, 16)
+#define SLPC_MIN_SLICE_FREQ_MASK REG_GENMASK(31, 24)
+ u32 freq;
+ };
+ };
+} __packed;
+
+struct slpc_shared_data_header {
+ /* Total size in bytes of this shared buffer. */
+ u32 size;
+ u32 global_state;
+ u32 display_data_addr;
+} __packed;
+
+struct slpc_override_params {
+ u32 bits[SLPC_OVERRIDE_BITFIELD_SIZE];
+ u32 values[SLPC_MAX_OVERRIDE_PARAMETERS];
+} __packed;
+
+struct slpc_shared_data {
+ struct slpc_shared_data_header header;
+ u8 shared_data_header_pad[SLPC_SHARED_DATA_SIZE_BYTE_HEADER -
+ sizeof(struct slpc_shared_data_header)];
+
+ u8 platform_info_pad[SLPC_SHARED_DATA_SIZE_BYTE_PLATFORM_INFO];
+
+ struct slpc_task_state_data task_state_data;
+ u8 task_state_data_pad[SLPC_SHARED_DATA_SIZE_BYTE_TASK_STATE -
+ sizeof(struct slpc_task_state_data)];
+
+ struct slpc_override_params override_params;
+ u8 override_params_pad[SLPC_OVERRIDE_PARAMS_TOTAL_BYTES -
+ sizeof(struct slpc_override_params)];
+
+ u8 shared_data_pad[SLPC_SHARED_DATA_SIZE_BYTE_OTHER];
+
+ /* PAGE 2 (4096 bytes), mode based parameter will be removed soon */
+ u8 reserved_mode_definition[4096];
+} __packed;
+
+/**
+ * DOC: SLPC H2G MESSAGE FORMAT
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN = GUC_HXG_ORIGIN_HOST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_REQUEST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | DATA0 = MBZ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | ACTION = _`GUC_ACTION_HOST2GUC_PC_SLPM_REQUEST` = 0x3003 |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:8 | **EVENT_ID** |
+ * + +-------+--------------------------------------------------------------+
+ * | | 7:0 | **EVENT_ARGC** - number of data arguments |
+ * +---+-------+--------------------------------------------------------------+
+ * | 2 | 31:0 | **EVENT_DATA1** |
+ * +---+-------+--------------------------------------------------------------+
+ * |...| 31:0 | ... |
+ * +---+-------+--------------------------------------------------------------+
+ * |2+n| 31:0 | **EVENT_DATAn** |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_ACTION_HOST2GUC_PC_SLPC_REQUEST 0x3003
+
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_MIN_LEN \
+ (GUC_HXG_REQUEST_MSG_MIN_LEN + 1u)
+#define HOST2GUC_PC_SLPC_EVENT_MAX_INPUT_ARGS 9
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_MAX_LEN \
+ (HOST2GUC_PC_SLPC_REQUEST_REQUEST_MSG_MIN_LEN + \
+ HOST2GUC_PC_SLPC_EVENT_MAX_INPUT_ARGS)
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_0_MBZ GUC_HXG_REQUEST_MSG_0_DATA0
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ID (0xff << 8)
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ARGC (0xff << 0)
+#define HOST2GUC_PC_SLPC_REQUEST_MSG_N_EVENT_DATA_N GUC_HXG_REQUEST_MSG_n_DATAn
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h
new file mode 100644
index 000000000000..3b83f907ece4
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h
@@ -0,0 +1,127 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2014-2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_COMMUNICATION_CTB_ABI_H
+#define _ABI_GUC_COMMUNICATION_CTB_ABI_H
+
+#include <linux/types.h>
+#include <linux/build_bug.h>
+
+#include "guc_messages_abi.h"
+
+/**
+ * DOC: CT Buffer
+ *
+ * Circular buffer used to send `CTB Message`_
+ */
+
+/**
+ * DOC: CTB Descriptor
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31:0 | **HEAD** - offset (in dwords) to the last dword that was |
+ * | | | read from the `CT Buffer`_. |
+ * | | | It can only be updated by the receiver. |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | **TAIL** - offset (in dwords) to the last dword that was |
+ * | | | written to the `CT Buffer`_. |
+ * | | | It can only be updated by the sender. |
+ * +---+-------+--------------------------------------------------------------+
+ * | 2 | 31:0 | **STATUS** - status of the CTB |
+ * | | | |
+ * | | | - _`GUC_CTB_STATUS_NO_ERROR` = 0 (normal operation) |
+ * | | | - _`GUC_CTB_STATUS_OVERFLOW` = 1 (head/tail too large) |
+ * | | | - _`GUC_CTB_STATUS_UNDERFLOW` = 2 (truncated message) |
+ * | | | - _`GUC_CTB_STATUS_MISMATCH` = 4 (head/tail modified) |
+ * +---+-------+--------------------------------------------------------------+
+ * |...| | RESERVED = MBZ |
+ * +---+-------+--------------------------------------------------------------+
+ * | 15| 31:0 | RESERVED = MBZ |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+struct guc_ct_buffer_desc {
+ u32 head;
+ u32 tail;
+ u32 status;
+#define GUC_CTB_STATUS_NO_ERROR 0
+#define GUC_CTB_STATUS_OVERFLOW (1 << 0)
+#define GUC_CTB_STATUS_UNDERFLOW (1 << 1)
+#define GUC_CTB_STATUS_MISMATCH (1 << 2)
+ u32 reserved[13];
+} __packed;
+static_assert(sizeof(struct guc_ct_buffer_desc) == 64);
+
+/**
+ * DOC: CTB Message
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31:16 | **FENCE** - message identifier |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:12 | **FORMAT** - format of the CTB message |
+ * | | | - _`GUC_CTB_FORMAT_HXG` = 0 - see `CTB HXG Message`_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 11:8 | **RESERVED** |
+ * | +-------+--------------------------------------------------------------+
+ * | | 7:0 | **NUM_DWORDS** - length of the CTB message (w/o header) |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | optional (depends on FORMAT) |
+ * +---+-------+ |
+ * |...| | |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_CTB_HDR_LEN 1u
+#define GUC_CTB_MSG_MIN_LEN GUC_CTB_HDR_LEN
+#define GUC_CTB_MSG_MAX_LEN 256u
+#define GUC_CTB_MSG_0_FENCE (0xffff << 16)
+#define GUC_CTB_MSG_0_FORMAT (0xf << 12)
+#define GUC_CTB_FORMAT_HXG 0u
+#define GUC_CTB_MSG_0_RESERVED (0xf << 8)
+#define GUC_CTB_MSG_0_NUM_DWORDS (0xff << 0)
+
+/**
+ * DOC: CTB HXG Message
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31:16 | FENCE |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:12 | FORMAT = GUC_CTB_FORMAT_HXG_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 11:8 | RESERVED = MBZ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 7:0 | NUM_DWORDS = length (in dwords) of the embedded HXG message |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | |
+ * +---+-------+ |
+ * |...| | [Embedded `HXG Message`_] |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_CTB_HXG_MSG_MIN_LEN (GUC_CTB_MSG_MIN_LEN + GUC_HXG_MSG_MIN_LEN)
+#define GUC_CTB_HXG_MSG_MAX_LEN GUC_CTB_MSG_MAX_LEN
+
+/**
+ * DOC: CTB based communication
+ *
+ * The CTB (command transport buffer) communication between Host and GuC
+ * is based on u32 data stream written to the shared buffer. One buffer can
+ * be used to transmit data only in one direction (one-directional channel).
+ *
+ * Current status of the each buffer is maintained in the `CTB Descriptor`_.
+ * Each message in data stream is encoded as `CTB HXG Message`_.
+ */
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_communication_mmio_abi.h b/drivers/gpu/drm/xe/abi/guc_communication_mmio_abi.h
new file mode 100644
index 000000000000..ef538e34f894
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_communication_mmio_abi.h
@@ -0,0 +1,49 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2014-2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_COMMUNICATION_MMIO_ABI_H
+#define _ABI_GUC_COMMUNICATION_MMIO_ABI_H
+
+/**
+ * DOC: GuC MMIO based communication
+ *
+ * The MMIO based communication between Host and GuC relies on special
+ * hardware registers which format could be defined by the software
+ * (so called scratch registers).
+ *
+ * Each MMIO based message, both Host to GuC (H2G) and GuC to Host (G2H)
+ * messages, which maximum length depends on number of available scratch
+ * registers, is directly written into those scratch registers.
+ *
+ * For Gen9+, there are 16 software scratch registers 0xC180-0xC1B8,
+ * but no H2G command takes more than 4 parameters and the GuC firmware
+ * itself uses an 4-element array to store the H2G message.
+ *
+ * For Gen11+, there are additional 4 registers 0x190240-0x19024C, which
+ * are, regardless on lower count, preferred over legacy ones.
+ *
+ * The MMIO based communication is mainly used during driver initialization
+ * phase to setup the `CTB based communication`_ that will be used afterwards.
+ */
+
+#define GUC_MAX_MMIO_MSG_LEN 4
+
+/**
+ * DOC: MMIO HXG Message
+ *
+ * Format of the MMIO messages follows definitions of `HXG Message`_.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31:0 | |
+ * +---+-------+ |
+ * |...| | [Embedded `HXG Message`_] |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_errors_abi.h b/drivers/gpu/drm/xe/abi/guc_errors_abi.h
new file mode 100644
index 000000000000..ec83551bf9c0
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_errors_abi.h
@@ -0,0 +1,37 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2014-2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_ERRORS_ABI_H
+#define _ABI_GUC_ERRORS_ABI_H
+
+enum xe_guc_response_status {
+ XE_GUC_RESPONSE_STATUS_SUCCESS = 0x0,
+ XE_GUC_RESPONSE_STATUS_GENERIC_FAIL = 0xF000,
+};
+
+enum xe_guc_load_status {
+ XE_GUC_LOAD_STATUS_DEFAULT = 0x00,
+ XE_GUC_LOAD_STATUS_START = 0x01,
+ XE_GUC_LOAD_STATUS_ERROR_DEVID_BUILD_MISMATCH = 0x02,
+ XE_GUC_LOAD_STATUS_GUC_PREPROD_BUILD_MISMATCH = 0x03,
+ XE_GUC_LOAD_STATUS_ERROR_DEVID_INVALID_GUCTYPE = 0x04,
+ XE_GUC_LOAD_STATUS_GDT_DONE = 0x10,
+ XE_GUC_LOAD_STATUS_IDT_DONE = 0x20,
+ XE_GUC_LOAD_STATUS_LAPIC_DONE = 0x30,
+ XE_GUC_LOAD_STATUS_GUCINT_DONE = 0x40,
+ XE_GUC_LOAD_STATUS_DPC_READY = 0x50,
+ XE_GUC_LOAD_STATUS_DPC_ERROR = 0x60,
+ XE_GUC_LOAD_STATUS_EXCEPTION = 0x70,
+ XE_GUC_LOAD_STATUS_INIT_DATA_INVALID = 0x71,
+ XE_GUC_LOAD_STATUS_PXP_TEARDOWN_CTRL_ENABLED = 0x72,
+ XE_GUC_LOAD_STATUS_INVALID_INIT_DATA_RANGE_START,
+ XE_GUC_LOAD_STATUS_MPU_DATA_INVALID = 0x73,
+ XE_GUC_LOAD_STATUS_INIT_MMIO_SAVE_RESTORE_INVALID = 0x74,
+ XE_GUC_LOAD_STATUS_INVALID_INIT_DATA_RANGE_END,
+
+ XE_GUC_LOAD_STATUS_READY = 0xF0,
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_klvs_abi.h b/drivers/gpu/drm/xe/abi/guc_klvs_abi.h
new file mode 100644
index 000000000000..47094b9b044c
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_klvs_abi.h
@@ -0,0 +1,322 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_KLVS_ABI_H
+#define _ABI_GUC_KLVS_ABI_H
+
+#include <linux/types.h>
+
+/**
+ * DOC: GuC KLV
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31:16 | **KEY** - KLV key identifier |
+ * | | | - `GuC Self Config KLVs`_ |
+ * | | | - `GuC VGT Policy KLVs`_ |
+ * | | | - `GuC VF Configuration KLVs`_ |
+ * | | | |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | **LEN** - length of VALUE (in 32bit dwords) |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | **VALUE** - actual value of the KLV (format depends on KEY) |
+ * +---+-------+ |
+ * |...| | |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_KLV_LEN_MIN 1u
+#define GUC_KLV_0_KEY (0xffff << 16)
+#define GUC_KLV_0_LEN (0xffff << 0)
+#define GUC_KLV_n_VALUE (0xffffffff << 0)
+
+/**
+ * DOC: GuC Self Config KLVs
+ *
+ * `GuC KLV`_ keys available for use with HOST2GUC_SELF_CFG_.
+ *
+ * _`GUC_KLV_SELF_CFG_MEMIRQ_STATUS_ADDR` : 0x0900
+ * Refers to 64 bit Global Gfx address (in bytes) of memory based interrupts
+ * status vector for use by the GuC.
+ *
+ * _`GUC_KLV_SELF_CFG_MEMIRQ_SOURCE_ADDR` : 0x0901
+ * Refers to 64 bit Global Gfx address (in bytes) of memory based interrupts
+ * source vector for use by the GuC.
+ *
+ * _`GUC_KLV_SELF_CFG_H2G_CTB_ADDR` : 0x0902
+ * Refers to 64 bit Global Gfx address of H2G `CT Buffer`_.
+ * Should be above WOPCM address but below APIC base address for native mode.
+ *
+ * _`GUC_KLV_SELF_CFG_H2G_CTB_DESCRIPTOR_ADDR : 0x0903
+ * Refers to 64 bit Global Gfx address of H2G `CTB Descriptor`_.
+ * Should be above WOPCM address but below APIC base address for native mode.
+ *
+ * _`GUC_KLV_SELF_CFG_H2G_CTB_SIZE : 0x0904
+ * Refers to size of H2G `CT Buffer`_ in bytes.
+ * Should be a multiple of 4K.
+ *
+ * _`GUC_KLV_SELF_CFG_G2H_CTB_ADDR : 0x0905
+ * Refers to 64 bit Global Gfx address of G2H `CT Buffer`_.
+ * Should be above WOPCM address but below APIC base address for native mode.
+ *
+ * _GUC_KLV_SELF_CFG_G2H_CTB_DESCRIPTOR_ADDR : 0x0906
+ * Refers to 64 bit Global Gfx address of G2H `CTB Descriptor`_.
+ * Should be above WOPCM address but below APIC base address for native mode.
+ *
+ * _GUC_KLV_SELF_CFG_G2H_CTB_SIZE : 0x0907
+ * Refers to size of G2H `CT Buffer`_ in bytes.
+ * Should be a multiple of 4K.
+ */
+
+#define GUC_KLV_SELF_CFG_MEMIRQ_STATUS_ADDR_KEY 0x0900
+#define GUC_KLV_SELF_CFG_MEMIRQ_STATUS_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_MEMIRQ_SOURCE_ADDR_KEY 0x0901
+#define GUC_KLV_SELF_CFG_MEMIRQ_SOURCE_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_H2G_CTB_ADDR_KEY 0x0902
+#define GUC_KLV_SELF_CFG_H2G_CTB_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_H2G_CTB_DESCRIPTOR_ADDR_KEY 0x0903
+#define GUC_KLV_SELF_CFG_H2G_CTB_DESCRIPTOR_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_H2G_CTB_SIZE_KEY 0x0904
+#define GUC_KLV_SELF_CFG_H2G_CTB_SIZE_LEN 1u
+
+#define GUC_KLV_SELF_CFG_G2H_CTB_ADDR_KEY 0x0905
+#define GUC_KLV_SELF_CFG_G2H_CTB_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_G2H_CTB_DESCRIPTOR_ADDR_KEY 0x0906
+#define GUC_KLV_SELF_CFG_G2H_CTB_DESCRIPTOR_ADDR_LEN 2u
+
+#define GUC_KLV_SELF_CFG_G2H_CTB_SIZE_KEY 0x0907
+#define GUC_KLV_SELF_CFG_G2H_CTB_SIZE_LEN 1u
+
+/*
+ * Per context scheduling policy update keys.
+ */
+enum {
+ GUC_CONTEXT_POLICIES_KLV_ID_EXECUTION_QUANTUM = 0x2001,
+ GUC_CONTEXT_POLICIES_KLV_ID_PREEMPTION_TIMEOUT = 0x2002,
+ GUC_CONTEXT_POLICIES_KLV_ID_SCHEDULING_PRIORITY = 0x2003,
+ GUC_CONTEXT_POLICIES_KLV_ID_PREEMPT_TO_IDLE_ON_QUANTUM_EXPIRY = 0x2004,
+ GUC_CONTEXT_POLICIES_KLV_ID_SLPM_GT_FREQUENCY = 0x2005,
+
+ GUC_CONTEXT_POLICIES_KLV_NUM_IDS = 5,
+};
+
+/**
+ * DOC: GuC VGT Policy KLVs
+ *
+ * `GuC KLV`_ keys available for use with PF2GUC_UPDATE_VGT_POLICY.
+ *
+ * _`GUC_KLV_VGT_POLICY_SCHED_IF_IDLE` : 0x8001
+ * This config sets whether strict scheduling is enabled whereby any VF
+ * that doesn’t have work to submit is still allocated a fixed execution
+ * time-slice to ensure active VFs execution is always consitent even
+ * during other VF reprovisiong / rebooting events. Changing this KLV
+ * impacts all VFs and takes effect on the next VF-Switch event.
+ *
+ * :0: don't schedule idle (default)
+ * :1: schedule if idle
+ *
+ * _`GUC_KLV_VGT_POLICY_ADVERSE_SAMPLE_PERIOD` : 0x8002
+ * This config sets the sample period for tracking adverse event counters.
+ * A sample period is the period in millisecs during which events are counted.
+ * This is applicable for all the VFs.
+ *
+ * :0: adverse events are not counted (default)
+ * :n: sample period in milliseconds
+ *
+ * _`GUC_KLV_VGT_POLICY_RESET_AFTER_VF_SWITCH` : 0x8D00
+ * This enum is to reset utilized HW engine after VF Switch (i.e to clean
+ * up Stale HW register left behind by previous VF)
+ *
+ * :0: don't reset (default)
+ * :1: reset
+ */
+
+#define GUC_KLV_VGT_POLICY_SCHED_IF_IDLE_KEY 0x8001
+#define GUC_KLV_VGT_POLICY_SCHED_IF_IDLE_LEN 1u
+
+#define GUC_KLV_VGT_POLICY_ADVERSE_SAMPLE_PERIOD_KEY 0x8002
+#define GUC_KLV_VGT_POLICY_ADVERSE_SAMPLE_PERIOD_LEN 1u
+
+#define GUC_KLV_VGT_POLICY_RESET_AFTER_VF_SWITCH_KEY 0x8D00
+#define GUC_KLV_VGT_POLICY_RESET_AFTER_VF_SWITCH_LEN 1u
+
+/**
+ * DOC: GuC VF Configuration KLVs
+ *
+ * `GuC KLV`_ keys available for use with PF2GUC_UPDATE_VF_CFG.
+ *
+ * _`GUC_KLV_VF_CFG_GGTT_START` : 0x0001
+ * A 4K aligned start GTT address/offset assigned to VF.
+ * Value is 64 bits.
+ *
+ * _`GUC_KLV_VF_CFG_GGTT_SIZE` : 0x0002
+ * A 4K aligned size of GGTT assigned to VF.
+ * Value is 64 bits.
+ *
+ * _`GUC_KLV_VF_CFG_LMEM_SIZE` : 0x0003
+ * A 2M aligned size of local memory assigned to VF.
+ * Value is 64 bits.
+ *
+ * _`GUC_KLV_VF_CFG_NUM_CONTEXTS` : 0x0004
+ * Refers to the number of contexts allocated to this VF.
+ *
+ * :0: no contexts (default)
+ * :1-65535: number of contexts (Gen12)
+ *
+ * _`GUC_KLV_VF_CFG_TILE_MASK` : 0x0005
+ * For multi-tiled products, this field contains the bitwise-OR of tiles
+ * assigned to the VF. Bit-0-set means VF has access to Tile-0,
+ * Bit-31-set means VF has access to Tile-31, and etc.
+ * At least one tile will always be allocated.
+ * If all bits are zero, VF KMD should treat this as a fatal error.
+ * For, single-tile products this KLV config is ignored.
+ *
+ * _`GUC_KLV_VF_CFG_NUM_DOORBELLS` : 0x0006
+ * Refers to the number of doorbells allocated to this VF.
+ *
+ * :0: no doorbells (default)
+ * :1-255: number of doorbells (Gen12)
+ *
+ * _`GUC_KLV_VF_CFG_EXEC_QUANTUM` : 0x8A01
+ * This config sets the VFs-execution-quantum in milliseconds.
+ * GUC will attempt to obey the maximum values as much as HW is capable
+ * of and this will never be perfectly-exact (accumulated nano-second
+ * granularity) since the GPUs clock time runs off a different crystal
+ * from the CPUs clock. Changing this KLV on a VF that is currently
+ * running a context wont take effect until a new context is scheduled in.
+ * That said, when the PF is changing this value from 0xFFFFFFFF to
+ * something else, it might never take effect if the VF is running an
+ * inifinitely long compute or shader kernel. In such a scenario, the
+ * PF would need to trigger a VM PAUSE and then change the KLV to force
+ * it to take effect. Such cases might typically happen on a 1PF+1VF
+ * Virtualization config enabled for heavier workloads like AI/ML.
+ *
+ * :0: infinite exec quantum (default)
+ *
+ * _`GUC_KLV_VF_CFG_PREEMPT_TIMEOUT` : 0x8A02
+ * This config sets the VF-preemption-timeout in microseconds.
+ * GUC will attempt to obey the minimum and maximum values as much as
+ * HW is capable and this will never be perfectly-exact (accumulated
+ * nano-second granularity) since the GPUs clock time runs off a
+ * different crystal from the CPUs clock. Changing this KLV on a VF
+ * that is currently running a context wont take effect until a new
+ * context is scheduled in.
+ * That said, when the PF is changing this value from 0xFFFFFFFF to
+ * something else, it might never take effect if the VF is running an
+ * inifinitely long compute or shader kernel.
+ * In this case, the PF would need to trigger a VM PAUSE and then change
+ * the KLV to force it to take effect. Such cases might typically happen
+ * on a 1PF+1VF Virtualization config enabled for heavier workloads like
+ * AI/ML.
+ *
+ * :0: no preemption timeout (default)
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_CAT_ERR` : 0x8A03
+ * This config sets threshold for CAT errors caused by the VF.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: event occurrence count per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_ENGINE_RESET` : 0x8A04
+ * This config sets threshold for engine reset caused by the VF.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: event occurrence count per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_PAGE_FAULT` : 0x8A05
+ * This config sets threshold for page fault errors caused by the VF.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: event occurrence count per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_H2G_STORM` : 0x8A06
+ * This config sets threshold for H2G interrupts triggered by the VF.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: time (us) per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_IRQ_STORM` : 0x8A07
+ * This config sets threshold for GT interrupts triggered by the VF's
+ * workloads.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: time (us) per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_THRESHOLD_DOORBELL_STORM` : 0x8A08
+ * This config sets threshold for doorbell's ring triggered by the VF.
+ *
+ * :0: adverse events or error will not be reported (default)
+ * :n: time (us) per sampling interval
+ *
+ * _`GUC_KLV_VF_CFG_BEGIN_DOORBELL_ID` : 0x8A0A
+ * Refers to the start index of doorbell assigned to this VF.
+ *
+ * :0: (default)
+ * :1-255: number of doorbells (Gen12)
+ *
+ * _`GUC_KLV_VF_CFG_BEGIN_CONTEXT_ID` : 0x8A0B
+ * Refers to the start index in context array allocated to this VF’s use.
+ *
+ * :0: (default)
+ * :1-65535: number of contexts (Gen12)
+ */
+
+#define GUC_KLV_VF_CFG_GGTT_START_KEY 0x0001
+#define GUC_KLV_VF_CFG_GGTT_START_LEN 2u
+
+#define GUC_KLV_VF_CFG_GGTT_SIZE_KEY 0x0002
+#define GUC_KLV_VF_CFG_GGTT_SIZE_LEN 2u
+
+#define GUC_KLV_VF_CFG_LMEM_SIZE_KEY 0x0003
+#define GUC_KLV_VF_CFG_LMEM_SIZE_LEN 2u
+
+#define GUC_KLV_VF_CFG_NUM_CONTEXTS_KEY 0x0004
+#define GUC_KLV_VF_CFG_NUM_CONTEXTS_LEN 1u
+
+#define GUC_KLV_VF_CFG_TILE_MASK_KEY 0x0005
+#define GUC_KLV_VF_CFG_TILE_MASK_LEN 1u
+
+#define GUC_KLV_VF_CFG_NUM_DOORBELLS_KEY 0x0006
+#define GUC_KLV_VF_CFG_NUM_DOORBELLS_LEN 1u
+
+#define GUC_KLV_VF_CFG_EXEC_QUANTUM_KEY 0x8a01
+#define GUC_KLV_VF_CFG_EXEC_QUANTUM_LEN 1u
+
+#define GUC_KLV_VF_CFG_PREEMPT_TIMEOUT_KEY 0x8a02
+#define GUC_KLV_VF_CFG_PREEMPT_TIMEOUT_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_CAT_ERR_KEY 0x8a03
+#define GUC_KLV_VF_CFG_THRESHOLD_CAT_ERR_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_ENGINE_RESET_KEY 0x8a04
+#define GUC_KLV_VF_CFG_THRESHOLD_ENGINE_RESET_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_PAGE_FAULT_KEY 0x8a05
+#define GUC_KLV_VF_CFG_THRESHOLD_PAGE_FAULT_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_H2G_STORM_KEY 0x8a06
+#define GUC_KLV_VF_CFG_THRESHOLD_H2G_STORM_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_IRQ_STORM_KEY 0x8a07
+#define GUC_KLV_VF_CFG_THRESHOLD_IRQ_STORM_LEN 1u
+
+#define GUC_KLV_VF_CFG_THRESHOLD_DOORBELL_STORM_KEY 0x8a08
+#define GUC_KLV_VF_CFG_THRESHOLD_DOORBELL_STORM_LEN 1u
+
+#define GUC_KLV_VF_CFG_BEGIN_DOORBELL_ID_KEY 0x8a0a
+#define GUC_KLV_VF_CFG_BEGIN_DOORBELL_ID_LEN 1u
+
+#define GUC_KLV_VF_CFG_BEGIN_CONTEXT_ID_KEY 0x8a0b
+#define GUC_KLV_VF_CFG_BEGIN_CONTEXT_ID_LEN 1u
+
+#endif
diff --git a/drivers/gpu/drm/xe/abi/guc_messages_abi.h b/drivers/gpu/drm/xe/abi/guc_messages_abi.h
new file mode 100644
index 000000000000..3d199016cf88
--- /dev/null
+++ b/drivers/gpu/drm/xe/abi/guc_messages_abi.h
@@ -0,0 +1,234 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2014-2021 Intel Corporation
+ */
+
+#ifndef _ABI_GUC_MESSAGES_ABI_H
+#define _ABI_GUC_MESSAGES_ABI_H
+
+/**
+ * DOC: HXG Message
+ *
+ * All messages exchanged with GuC are defined using 32 bit dwords.
+ * First dword is treated as a message header. Remaining dwords are optional.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | | | |
+ * | 0 | 31 | **ORIGIN** - originator of the message |
+ * | | | - _`GUC_HXG_ORIGIN_HOST` = 0 |
+ * | | | - _`GUC_HXG_ORIGIN_GUC` = 1 |
+ * | | | |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | **TYPE** - message type |
+ * | | | - _`GUC_HXG_TYPE_REQUEST` = 0 |
+ * | | | - _`GUC_HXG_TYPE_EVENT` = 1 |
+ * | | | - _`GUC_HXG_TYPE_NO_RESPONSE_BUSY` = 3 |
+ * | | | - _`GUC_HXG_TYPE_NO_RESPONSE_RETRY` = 5 |
+ * | | | - _`GUC_HXG_TYPE_RESPONSE_FAILURE` = 6 |
+ * | | | - _`GUC_HXG_TYPE_RESPONSE_SUCCESS` = 7 |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | **AUX** - auxiliary data (depends on TYPE) |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | |
+ * +---+-------+ |
+ * |...| | **PAYLOAD** - optional payload (depends on TYPE) |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_MSG_MIN_LEN 1u
+#define GUC_HXG_MSG_0_ORIGIN (0x1 << 31)
+#define GUC_HXG_ORIGIN_HOST 0u
+#define GUC_HXG_ORIGIN_GUC 1u
+#define GUC_HXG_MSG_0_TYPE (0x7 << 28)
+#define GUC_HXG_TYPE_REQUEST 0u
+#define GUC_HXG_TYPE_EVENT 1u
+#define GUC_HXG_TYPE_NO_RESPONSE_BUSY 3u
+#define GUC_HXG_TYPE_NO_RESPONSE_RETRY 5u
+#define GUC_HXG_TYPE_RESPONSE_FAILURE 6u
+#define GUC_HXG_TYPE_RESPONSE_SUCCESS 7u
+#define GUC_HXG_MSG_0_AUX (0xfffffff << 0)
+#define GUC_HXG_MSG_n_PAYLOAD (0xffffffff << 0)
+
+/**
+ * DOC: HXG Request
+ *
+ * The `HXG Request`_ message should be used to initiate synchronous activity
+ * for which confirmation or return data is expected.
+ *
+ * The recipient of this message shall use `HXG Response`_, `HXG Failure`_
+ * or `HXG Retry`_ message as a definite reply, and may use `HXG Busy`_
+ * message as a intermediate reply.
+ *
+ * Format of @DATA0 and all @DATAn fields depends on the @ACTION code.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_REQUEST_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | **DATA0** - request data (depends on ACTION) |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | **ACTION** - requested action code |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | |
+ * +---+-------+ |
+ * |...| | **DATAn** - optional data (depends on ACTION) |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_REQUEST_MSG_MIN_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_REQUEST_MSG_0_DATA0 (0xfff << 16)
+#define GUC_HXG_REQUEST_MSG_0_ACTION (0xffff << 0)
+#define GUC_HXG_REQUEST_MSG_n_DATAn GUC_HXG_MSG_n_PAYLOAD
+
+/**
+ * DOC: HXG Event
+ *
+ * The `HXG Event`_ message should be used to initiate asynchronous activity
+ * that does not involves immediate confirmation nor data.
+ *
+ * Format of @DATA0 and all @DATAn fields depends on the @ACTION code.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_EVENT_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | **DATA0** - event data (depends on ACTION) |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | **ACTION** - event action code |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | |
+ * +---+-------+ |
+ * |...| | **DATAn** - optional event data (depends on ACTION) |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_EVENT_MSG_MIN_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_EVENT_MSG_0_DATA0 (0xfff << 16)
+#define GUC_HXG_EVENT_MSG_0_ACTION (0xffff << 0)
+#define GUC_HXG_EVENT_MSG_n_DATAn GUC_HXG_MSG_n_PAYLOAD
+
+/**
+ * DOC: HXG Busy
+ *
+ * The `HXG Busy`_ message may be used to acknowledge reception of the `HXG Request`_
+ * message if the recipient expects that it processing will be longer than default
+ * timeout.
+ *
+ * The @COUNTER field may be used as a progress indicator.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_NO_RESPONSE_BUSY_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | **COUNTER** - progress indicator |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_BUSY_MSG_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_BUSY_MSG_0_COUNTER GUC_HXG_MSG_0_AUX
+
+/**
+ * DOC: HXG Retry
+ *
+ * The `HXG Retry`_ message should be used by recipient to indicate that the
+ * `HXG Request`_ message was dropped and it should be resent again.
+ *
+ * The @REASON field may be used to provide additional information.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_NO_RESPONSE_RETRY_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | **REASON** - reason for retry |
+ * | | | - _`GUC_HXG_RETRY_REASON_UNSPECIFIED` = 0 |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_RETRY_MSG_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_RETRY_MSG_0_REASON GUC_HXG_MSG_0_AUX
+#define GUC_HXG_RETRY_REASON_UNSPECIFIED 0u
+
+/**
+ * DOC: HXG Failure
+ *
+ * The `HXG Failure`_ message shall be used as a reply to the `HXG Request`_
+ * message that could not be processed due to an error.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_RESPONSE_FAILURE_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:16 | **HINT** - additional error hint |
+ * | +-------+--------------------------------------------------------------+
+ * | | 15:0 | **ERROR** - error/result code |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_FAILURE_MSG_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_FAILURE_MSG_0_HINT (0xfff << 16)
+#define GUC_HXG_FAILURE_MSG_0_ERROR (0xffff << 0)
+
+/**
+ * DOC: HXG Response
+ *
+ * The `HXG Response`_ message shall be used as a reply to the `HXG Request`_
+ * message that was successfully processed without an error.
+ *
+ * +---+-------+--------------------------------------------------------------+
+ * | | Bits | Description |
+ * +===+=======+==============================================================+
+ * | 0 | 31 | ORIGIN |
+ * | +-------+--------------------------------------------------------------+
+ * | | 30:28 | TYPE = GUC_HXG_TYPE_RESPONSE_SUCCESS_ |
+ * | +-------+--------------------------------------------------------------+
+ * | | 27:0 | **DATA0** - data (depends on ACTION from `HXG Request`_) |
+ * +---+-------+--------------------------------------------------------------+
+ * | 1 | 31:0 | |
+ * +---+-------+ |
+ * |...| | **DATAn** - data (depends on ACTION from `HXG Request`_) |
+ * +---+-------+ |
+ * | n | 31:0 | |
+ * +---+-------+--------------------------------------------------------------+
+ */
+
+#define GUC_HXG_RESPONSE_MSG_MIN_LEN GUC_HXG_MSG_MIN_LEN
+#define GUC_HXG_RESPONSE_MSG_0_DATA0 GUC_HXG_MSG_0_AUX
+#define GUC_HXG_RESPONSE_MSG_n_DATAn GUC_HXG_MSG_n_PAYLOAD
+
+/* deprecated */
+#define INTEL_GUC_MSG_TYPE_SHIFT 28
+#define INTEL_GUC_MSG_TYPE_MASK (0xF << INTEL_GUC_MSG_TYPE_SHIFT)
+#define INTEL_GUC_MSG_DATA_SHIFT 16
+#define INTEL_GUC_MSG_DATA_MASK (0xFFF << INTEL_GUC_MSG_DATA_SHIFT)
+#define INTEL_GUC_MSG_CODE_SHIFT 0
+#define INTEL_GUC_MSG_CODE_MASK (0xFFFF << INTEL_GUC_MSG_CODE_SHIFT)
+
+enum intel_guc_msg_type {
+ INTEL_GUC_MSG_TYPE_REQUEST = 0x0,
+ INTEL_GUC_MSG_TYPE_RESPONSE = 0xF,
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_lmem.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_lmem.h
new file mode 100644
index 000000000000..710cecca972d
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_lmem.h
@@ -0,0 +1 @@
+/* Empty */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_mman.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_mman.h
new file mode 100644
index 000000000000..650ea2803a97
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_mman.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _I915_GEM_MMAN_H_
+#define _I915_GEM_MMAN_H_
+
+#include "xe_bo_types.h"
+#include <drm/drm_prime.h>
+
+static inline int i915_gem_fb_mmap(struct xe_bo *bo, struct vm_area_struct *vma)
+{
+ return drm_gem_prime_mmap(&bo->ttm.base, vma);
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h
new file mode 100644
index 000000000000..5f19550cc845
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object.h
@@ -0,0 +1,65 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _I915_GEM_OBJECT_H_
+#define _I915_GEM_OBJECT_H_
+
+#include <linux/types.h>
+
+#include "xe_bo.h"
+
+#define i915_gem_object_is_shmem(obj) ((obj)->flags & XE_BO_CREATE_SYSTEM_BIT)
+
+static inline dma_addr_t i915_gem_object_get_dma_address(const struct xe_bo *bo, pgoff_t n)
+{
+ /* Should never be called */
+ WARN_ON(1);
+ return n;
+}
+
+static inline bool i915_gem_object_is_tiled(const struct xe_bo *bo)
+{
+ /* legacy tiling is unused */
+ return false;
+}
+
+static inline bool i915_gem_object_is_userptr(const struct xe_bo *bo)
+{
+ /* legacy tiling is unused */
+ return false;
+}
+
+static inline int i915_gem_object_read_from_page(struct xe_bo *bo,
+ u32 ofs, u64 *ptr, u32 size)
+{
+ struct ttm_bo_kmap_obj map;
+ void *virtual;
+ bool is_iomem;
+ int ret;
+
+ XE_WARN_ON(size != 8);
+
+ ret = xe_bo_lock(bo, true);
+ if (ret)
+ return ret;
+
+ ret = ttm_bo_kmap(&bo->ttm, ofs >> PAGE_SHIFT, 1, &map);
+ if (ret)
+ goto out_unlock;
+
+ ofs &= ~PAGE_MASK;
+ virtual = ttm_kmap_obj_virtual(&map, &is_iomem);
+ if (is_iomem)
+ *ptr = readq((void __iomem *)(virtual + ofs));
+ else
+ *ptr = *(u64 *)(virtual + ofs);
+
+ ttm_bo_kunmap(&map);
+out_unlock:
+ xe_bo_unlock(bo);
+ return ret;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object_frontbuffer.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object_frontbuffer.h
new file mode 100644
index 000000000000..2a3f12d2978c
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_object_frontbuffer.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _I915_GEM_OBJECT_FRONTBUFFER_H_
+#define _I915_GEM_OBJECT_FRONTBUFFER_H_
+
+#define i915_gem_object_get_frontbuffer(obj) NULL
+#define i915_gem_object_set_frontbuffer(obj, front) (front)
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gt/intel_rps.h b/drivers/gpu/drm/xe/compat-i915-headers/gt/intel_rps.h
new file mode 100644
index 000000000000..21fec9cc837c
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gt/intel_rps.h
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_RPS_H__
+#define __INTEL_RPS_H__
+
+#define gen5_rps_irq_handler(x) ({})
+
+#endif /* __INTEL_RPS_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_active.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_active.h
new file mode 100644
index 000000000000..6f0ab3753563
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_active.h
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _I915_ACTIVE_H_
+#define _I915_ACTIVE_H_
+
+#include "i915_active_types.h"
+
+static inline void i915_active_init(struct i915_active *ref,
+ int (*active)(struct i915_active *ref),
+ void (*retire)(struct i915_active *ref),
+ unsigned long flags)
+{
+ (void) active;
+ (void) retire;
+}
+
+#define i915_active_fini(active) do { } while (0)
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_active_types.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_active_types.h
new file mode 100644
index 000000000000..8c31f9a8b168
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_active_types.h
@@ -0,0 +1,13 @@
+/*
+ * SPDX-License-Identifier: MIT
+ *
+ * Copyright © 2019 Intel Corporation
+ */
+
+#ifndef _I915_ACTIVE_TYPES_H_
+#define _I915_ACTIVE_TYPES_H_
+
+struct i915_active {};
+#define I915_ACTIVE_RETIRE_SLEEPS 0
+
+#endif /* _I915_ACTIVE_TYPES_H_ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_config.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_config.h
new file mode 100644
index 000000000000..e835bea08d1b
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_config.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __I915_CONFIG_H__
+#define __I915_CONFIG_H__
+
+#include <linux/sched.h>
+
+struct drm_i915_private;
+
+static inline unsigned long
+i915_fence_timeout(const struct drm_i915_private *i915)
+{
+ return MAX_SCHEDULE_TIMEOUT;
+}
+
+#endif /* __I915_CONFIG_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_debugfs.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_debugfs.h
new file mode 100644
index 000000000000..b4c47617b64b
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_debugfs.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __I915_DEBUGFS_H__
+#define __I915_DEBUGFS_H__
+
+struct drm_i915_gem_object;
+struct seq_file;
+
+static inline void i915_debugfs_describe_obj(struct seq_file *m, struct drm_i915_gem_object *obj) {}
+
+#endif /* __I915_DEBUGFS_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_drv.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_drv.h
new file mode 100644
index 000000000000..5d2a77b52db4
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_drv.h
@@ -0,0 +1,233 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+#ifndef _XE_I915_DRV_H_
+#define _XE_I915_DRV_H_
+
+/*
+ * "Adaptation header" to allow i915 display to also build for xe driver.
+ * TODO: refactor i915 and xe so this can cease to exist
+ */
+
+#include <drm/drm_drv.h>
+
+#include "gem/i915_gem_object.h"
+
+#include "soc/intel_pch.h"
+#include "xe_device.h"
+#include "xe_bo.h"
+#include "xe_pm.h"
+#include "xe_step.h"
+#include "i915_gem.h"
+#include "i915_gem_stolen.h"
+#include "i915_gpu_error.h"
+#include "i915_reg_defs.h"
+#include "i915_utils.h"
+#include "intel_gt_types.h"
+#include "intel_step.h"
+#include "intel_uc_fw.h"
+#include "intel_uncore.h"
+#include "intel_runtime_pm.h"
+#include <linux/pm_runtime.h>
+
+static inline struct drm_i915_private *to_i915(const struct drm_device *dev)
+{
+ return container_of(dev, struct drm_i915_private, drm);
+}
+
+static inline struct drm_i915_private *kdev_to_i915(struct device *kdev)
+{
+ return dev_get_drvdata(kdev);
+}
+
+
+#define INTEL_JASPERLAKE 0
+#define INTEL_ELKHARTLAKE 0
+#define IS_PLATFORM(xe, x) ((xe)->info.platform == x)
+#define INTEL_INFO(dev_priv) (&((dev_priv)->info))
+#define INTEL_DEVID(dev_priv) ((dev_priv)->info.devid)
+#define IS_I830(dev_priv) (dev_priv && 0)
+#define IS_I845G(dev_priv) (dev_priv && 0)
+#define IS_I85X(dev_priv) (dev_priv && 0)
+#define IS_I865G(dev_priv) (dev_priv && 0)
+#define IS_I915G(dev_priv) (dev_priv && 0)
+#define IS_I915GM(dev_priv) (dev_priv && 0)
+#define IS_I945G(dev_priv) (dev_priv && 0)
+#define IS_I945GM(dev_priv) (dev_priv && 0)
+#define IS_I965G(dev_priv) (dev_priv && 0)
+#define IS_I965GM(dev_priv) (dev_priv && 0)
+#define IS_G45(dev_priv) (dev_priv && 0)
+#define IS_GM45(dev_priv) (dev_priv && 0)
+#define IS_G4X(dev_priv) (dev_priv && 0)
+#define IS_PINEVIEW(dev_priv) (dev_priv && 0)
+#define IS_G33(dev_priv) (dev_priv && 0)
+#define IS_IRONLAKE(dev_priv) (dev_priv && 0)
+#define IS_IRONLAKE_M(dev_priv) (dev_priv && 0)
+#define IS_SANDYBRIDGE(dev_priv) (dev_priv && 0)
+#define IS_IVYBRIDGE(dev_priv) (dev_priv && 0)
+#define IS_IVB_GT1(dev_priv) (dev_priv && 0)
+#define IS_VALLEYVIEW(dev_priv) (dev_priv && 0)
+#define IS_CHERRYVIEW(dev_priv) (dev_priv && 0)
+#define IS_HASWELL(dev_priv) (dev_priv && 0)
+#define IS_BROADWELL(dev_priv) (dev_priv && 0)
+#define IS_SKYLAKE(dev_priv) (dev_priv && 0)
+#define IS_BROXTON(dev_priv) (dev_priv && 0)
+#define IS_KABYLAKE(dev_priv) (dev_priv && 0)
+#define IS_GEMINILAKE(dev_priv) (dev_priv && 0)
+#define IS_COFFEELAKE(dev_priv) (dev_priv && 0)
+#define IS_COMETLAKE(dev_priv) (dev_priv && 0)
+#define IS_ICELAKE(dev_priv) (dev_priv && 0)
+#define IS_JASPERLAKE(dev_priv) (dev_priv && 0)
+#define IS_ELKHARTLAKE(dev_priv) (dev_priv && 0)
+#define IS_TIGERLAKE(dev_priv) IS_PLATFORM(dev_priv, XE_TIGERLAKE)
+#define IS_ROCKETLAKE(dev_priv) IS_PLATFORM(dev_priv, XE_ROCKETLAKE)
+#define IS_DG1(dev_priv) IS_PLATFORM(dev_priv, XE_DG1)
+#define IS_ALDERLAKE_S(dev_priv) IS_PLATFORM(dev_priv, XE_ALDERLAKE_S)
+#define IS_ALDERLAKE_P(dev_priv) IS_PLATFORM(dev_priv, XE_ALDERLAKE_P)
+#define IS_XEHPSDV(dev_priv) (dev_priv && 0)
+#define IS_DG2(dev_priv) IS_PLATFORM(dev_priv, XE_DG2)
+#define IS_PONTEVECCHIO(dev_priv) IS_PLATFORM(dev_priv, XE_PVC)
+#define IS_METEORLAKE(dev_priv) IS_PLATFORM(dev_priv, XE_METEORLAKE)
+#define IS_LUNARLAKE(dev_priv) IS_PLATFORM(dev_priv, XE_LUNARLAKE)
+
+#define IS_HASWELL_ULT(dev_priv) (dev_priv && 0)
+#define IS_BROADWELL_ULT(dev_priv) (dev_priv && 0)
+#define IS_BROADWELL_ULX(dev_priv) (dev_priv && 0)
+
+#define IP_VER(ver, rel) ((ver) << 8 | (rel))
+
+#define INTEL_DISPLAY_ENABLED(xe) (HAS_DISPLAY((xe)) && !intel_opregion_headless_sku((xe)))
+
+#define IS_GRAPHICS_VER(xe, first, last) \
+ ((xe)->info.graphics_verx100 >= first * 100 && \
+ (xe)->info.graphics_verx100 <= (last*100 + 99))
+#define IS_MOBILE(xe) (xe && 0)
+#define HAS_LLC(xe) (!IS_DGFX((xe)))
+
+#define HAS_GMD_ID(xe) GRAPHICS_VERx100(xe) >= 1270
+
+/* Workarounds not handled yet */
+#define IS_DISPLAY_STEP(xe, first, last) ({u8 __step = (xe)->info.step.display; first <= __step && __step <= last; })
+#define IS_GRAPHICS_STEP(xe, first, last) ({u8 __step = (xe)->info.step.graphics; first <= __step && __step <= last; })
+
+#define IS_LP(xe) (0)
+#define IS_GEN9_LP(xe) (0)
+#define IS_GEN9_BC(xe) (0)
+
+#define IS_TIGERLAKE_UY(xe) (xe && 0)
+#define IS_COMETLAKE_ULX(xe) (xe && 0)
+#define IS_COFFEELAKE_ULX(xe) (xe && 0)
+#define IS_KABYLAKE_ULX(xe) (xe && 0)
+#define IS_SKYLAKE_ULX(xe) (xe && 0)
+#define IS_HASWELL_ULX(xe) (xe && 0)
+#define IS_COMETLAKE_ULT(xe) (xe && 0)
+#define IS_COFFEELAKE_ULT(xe) (xe && 0)
+#define IS_KABYLAKE_ULT(xe) (xe && 0)
+#define IS_SKYLAKE_ULT(xe) (xe && 0)
+
+#define IS_DG1_GRAPHICS_STEP(xe, first, last) (IS_DG1(xe) && IS_GRAPHICS_STEP(xe, first, last))
+#define IS_DG2_GRAPHICS_STEP(xe, variant, first, last) \
+ ((xe)->info.subplatform == XE_SUBPLATFORM_DG2_ ## variant && \
+ IS_GRAPHICS_STEP(xe, first, last))
+#define IS_XEHPSDV_GRAPHICS_STEP(xe, first, last) (IS_XEHPSDV(xe) && IS_GRAPHICS_STEP(xe, first, last))
+
+/* XXX: No basedie stepping support yet */
+#define IS_PVC_BD_STEP(xe, first, last) (!WARN_ON(1) && IS_PONTEVECCHIO(xe))
+
+#define IS_TIGERLAKE_DISPLAY_STEP(xe, first, last) (IS_TIGERLAKE(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_ROCKETLAKE_DISPLAY_STEP(xe, first, last) (IS_ROCKETLAKE(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_DG1_DISPLAY_STEP(xe, first, last) (IS_DG1(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_DG2_DISPLAY_STEP(xe, first, last) (IS_DG2(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_ADLP_DISPLAY_STEP(xe, first, last) (IS_ALDERLAKE_P(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_ADLS_DISPLAY_STEP(xe, first, last) (IS_ALDERLAKE_S(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_JSL_EHL_DISPLAY_STEP(xe, first, last) (IS_JSL_EHL(xe) && IS_DISPLAY_STEP(xe, first, last))
+#define IS_MTL_DISPLAY_STEP(xe, first, last) (IS_METEORLAKE(xe) && IS_DISPLAY_STEP(xe, first, last))
+
+/* FIXME: Add subplatform here */
+#define IS_MTL_GRAPHICS_STEP(xe, sub, first, last) (IS_METEORLAKE(xe) && IS_DISPLAY_STEP(xe, first, last))
+
+#define IS_DG2_G10(xe) ((xe)->info.subplatform == XE_SUBPLATFORM_DG2_G10)
+#define IS_DG2_G11(xe) ((xe)->info.subplatform == XE_SUBPLATFORM_DG2_G11)
+#define IS_DG2_G12(xe) ((xe)->info.subplatform == XE_SUBPLATFORM_DG2_G12)
+#define IS_RAPTORLAKE_U(xe) ((xe)->info.subplatform == XE_SUBPLATFORM_ALDERLAKE_P_RPLU)
+#define IS_ICL_WITH_PORT_F(xe) (xe && 0)
+#define HAS_FLAT_CCS(xe) (xe_device_has_flat_ccs(xe))
+#define to_intel_bo(x) gem_to_xe_bo((x))
+#define mkwrite_device_info(xe) (INTEL_INFO(xe))
+
+#define HAS_128_BYTE_Y_TILING(xe) (xe || 1)
+
+#define intel_has_gpu_reset(a) (a && 0)
+
+#include "intel_wakeref.h"
+
+static inline bool intel_runtime_pm_get(struct xe_runtime_pm *pm)
+{
+ struct xe_device *xe = container_of(pm, struct xe_device, runtime_pm);
+
+ if (xe_pm_runtime_get(xe) < 0) {
+ xe_pm_runtime_put(xe);
+ return false;
+ }
+ return true;
+}
+
+static inline bool intel_runtime_pm_get_if_in_use(struct xe_runtime_pm *pm)
+{
+ struct xe_device *xe = container_of(pm, struct xe_device, runtime_pm);
+
+ return xe_pm_runtime_get_if_active(xe);
+}
+
+static inline void intel_runtime_pm_put_unchecked(struct xe_runtime_pm *pm)
+{
+ struct xe_device *xe = container_of(pm, struct xe_device, runtime_pm);
+
+ xe_pm_runtime_put(xe);
+}
+
+static inline void intel_runtime_pm_put(struct xe_runtime_pm *pm, bool wakeref)
+{
+ if (wakeref)
+ intel_runtime_pm_put_unchecked(pm);
+}
+
+#define intel_runtime_pm_get_raw intel_runtime_pm_get
+#define intel_runtime_pm_put_raw intel_runtime_pm_put
+#define assert_rpm_wakelock_held(x) do { } while (0)
+#define assert_rpm_raw_wakeref_held(x) do { } while (0)
+
+#define intel_uncore_forcewake_get(x, y) do { } while (0)
+#define intel_uncore_forcewake_put(x, y) do { } while (0)
+
+#define intel_uncore_arm_unclaimed_mmio_detection(x) do { } while (0)
+
+#define I915_PRIORITY_DISPLAY 0
+struct i915_sched_attr {
+ int priority;
+};
+#define i915_gem_fence_wait_priority(fence, attr) do { (void) attr; } while (0)
+
+#define with_intel_runtime_pm(rpm, wf) \
+ for ((wf) = intel_runtime_pm_get(rpm); (wf); \
+ intel_runtime_pm_put((rpm), (wf)), (wf) = 0)
+
+#define pdev_to_i915 pdev_to_xe_device
+#define RUNTIME_INFO(xe) (&(xe)->info.i915_runtime)
+
+#define FORCEWAKE_ALL XE_FORCEWAKE_ALL
+#define HPD_STORM_DEFAULT_THRESHOLD 50
+
+#ifdef CONFIG_ARM64
+/*
+ * arm64 indirectly includes linux/rtc.h,
+ * which defines a irq_lock, so include it
+ * here before #define-ing it
+ */
+#include <linux/rtc.h>
+#endif
+
+#define irq_lock irq.lock
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_fixed.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_fixed.h
new file mode 100644
index 000000000000..12c671fd5235
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_fixed.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/i915_fixed.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_gem.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_gem.h
new file mode 100644
index 000000000000..06b723a479c5
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_gem.h
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __I915_GEM_H__
+#define __I915_GEM_H__
+#define GEM_BUG_ON
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_gem_stolen.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_gem_stolen.h
new file mode 100644
index 000000000000..888e7a87a925
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_gem_stolen.h
@@ -0,0 +1,79 @@
+#ifndef _I915_GEM_STOLEN_H_
+#define _I915_GEM_STOLEN_H_
+
+#include "xe_ttm_stolen_mgr.h"
+#include "xe_res_cursor.h"
+
+struct xe_bo;
+
+struct i915_stolen_fb {
+ struct xe_bo *bo;
+};
+
+static inline int i915_gem_stolen_insert_node_in_range(struct xe_device *xe,
+ struct i915_stolen_fb *fb,
+ u32 size, u32 align,
+ u32 start, u32 end)
+{
+ struct xe_bo *bo;
+ int err;
+ u32 flags = XE_BO_CREATE_PINNED_BIT | XE_BO_CREATE_STOLEN_BIT;
+
+ bo = xe_bo_create_locked_range(xe, xe_device_get_root_tile(xe),
+ NULL, size, start, end,
+ ttm_bo_type_kernel, flags);
+ if (IS_ERR(bo)) {
+ err = PTR_ERR(bo);
+ bo = NULL;
+ return err;
+ }
+ err = xe_bo_pin(bo);
+ xe_bo_unlock_vm_held(bo);
+
+ if (err) {
+ xe_bo_put(fb->bo);
+ bo = NULL;
+ }
+
+ fb->bo = bo;
+
+ return err;
+}
+
+static inline int i915_gem_stolen_insert_node(struct xe_device *xe,
+ struct i915_stolen_fb *fb,
+ u32 size, u32 align)
+{
+ /* Not used on xe */
+ BUG_ON(1);
+ return -ENODEV;
+}
+
+static inline void i915_gem_stolen_remove_node(struct xe_device *xe,
+ struct i915_stolen_fb *fb)
+{
+ xe_bo_unpin_map_no_vm(fb->bo);
+ fb->bo = NULL;
+}
+
+#define i915_gem_stolen_initialized(xe) (!!ttm_manager_type(&(xe)->ttm, XE_PL_STOLEN))
+#define i915_gem_stolen_node_allocated(fb) (!!((fb)->bo))
+
+static inline u32 i915_gem_stolen_node_offset(struct i915_stolen_fb *fb)
+{
+ struct xe_res_cursor res;
+
+ xe_res_first(fb->bo->ttm.resource, 0, 4096, &res);
+ return res.start;
+}
+
+/* Used for < gen4. These are not supported by Xe */
+#define i915_gem_stolen_area_address(xe) (!WARN_ON(1))
+/* Used for gen9 specific WA. Gen9 is not supported by Xe */
+#define i915_gem_stolen_area_size(xe) (!WARN_ON(1))
+
+#define i915_gem_stolen_node_address(xe, fb) (xe_ttm_stolen_gpu_offset(xe) + \
+ i915_gem_stolen_node_offset(fb))
+#define i915_gem_stolen_node_size(fb) ((u64)((fb)->bo->ttm.base.size))
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_gpu_error.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_gpu_error.h
new file mode 100644
index 000000000000..98e9dd78f670
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_gpu_error.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _I915_GPU_ERROR_H_
+#define _I915_GPU_ERROR_H_
+
+struct drm_i915_error_state_buf;
+
+__printf(2, 3)
+static inline void
+i915_error_printf(struct drm_i915_error_state_buf *e, const char *f, ...)
+{
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_irq.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_irq.h
new file mode 100644
index 000000000000..61707a07f91f
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_irq.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/i915_irq.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_reg.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_reg.h
new file mode 100644
index 000000000000..8619ec015ad4
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_reg.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/i915_reg.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_reg_defs.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_reg_defs.h
new file mode 100644
index 000000000000..723279c975b1
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_reg_defs.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/i915_reg_defs.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_trace.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_trace.h
new file mode 100644
index 000000000000..d429d421ac70
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_trace.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#define trace_i915_reg_rw(a...) do { } while (0)
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_utils.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_utils.h
new file mode 100644
index 000000000000..1d7c4360e5c0
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_utils.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/i915_utils.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_vgpu.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_vgpu.h
new file mode 100644
index 000000000000..80b024d435dc
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_vgpu.h
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _I915_VGPU_H_
+#define _I915_VGPU_H_
+
+#include <linux/types.h>
+
+struct drm_i915_private;
+struct i915_ggtt;
+
+static inline void intel_vgpu_detect(struct drm_i915_private *i915)
+{
+}
+static inline bool intel_vgpu_active(struct drm_i915_private *i915)
+{
+ return false;
+}
+static inline void intel_vgpu_register(struct drm_i915_private *i915)
+{
+}
+static inline bool intel_vgpu_has_full_ppgtt(struct drm_i915_private *i915)
+{
+ return false;
+}
+static inline bool intel_vgpu_has_hwsp_emulation(struct drm_i915_private *i915)
+{
+ return false;
+}
+static inline bool intel_vgpu_has_huge_gtt(struct drm_i915_private *i915)
+{
+ return false;
+}
+static inline int intel_vgt_balloon(struct i915_ggtt *ggtt)
+{
+ return 0;
+}
+static inline void intel_vgt_deballoon(struct i915_ggtt *ggtt)
+{
+}
+
+#endif /* _I915_VGPU_H_ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h
new file mode 100644
index 000000000000..a20d2638ea7a
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma.h
@@ -0,0 +1,34 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef I915_VMA_H
+#define I915_VMA_H
+
+#include <uapi/drm/i915_drm.h>
+#include <drm/drm_mm.h>
+
+/* We don't want these from i915_drm.h in case of Xe */
+#undef I915_TILING_X
+#undef I915_TILING_Y
+#define I915_TILING_X 0
+#define I915_TILING_Y 0
+
+struct xe_bo;
+
+struct i915_vma {
+ struct xe_bo *bo, *dpt;
+ struct drm_mm_node node;
+};
+
+#define i915_ggtt_clear_scanout(bo) do { } while (0)
+
+#define i915_vma_fence_id(vma) -1
+
+static inline u32 i915_ggtt_offset(const struct i915_vma *vma)
+{
+ return vma->node.start;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/i915_vma_types.h b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma_types.h
new file mode 100644
index 000000000000..e7aaf50f5485
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/i915_vma_types.h
@@ -0,0 +1,74 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/types.h>
+#include <linux/build_bug.h>
+
+/* XX: Figure out how to handle this vma mapping in xe */
+struct intel_remapped_plane_info {
+ /* in gtt pages */
+ u32 offset:31;
+ u32 linear:1;
+ union {
+ /* in gtt pages for !linear */
+ struct {
+ u16 width;
+ u16 height;
+ u16 src_stride;
+ u16 dst_stride;
+ };
+
+ /* in gtt pages for linear */
+ u32 size;
+ };
+} __packed;
+
+struct intel_remapped_info {
+ struct intel_remapped_plane_info plane[4];
+ /* in gtt pages */
+ u32 plane_alignment;
+} __packed;
+
+struct intel_rotation_info {
+ struct intel_remapped_plane_info plane[2];
+} __packed;
+
+enum i915_gtt_view_type {
+ I915_GTT_VIEW_NORMAL = 0,
+ I915_GTT_VIEW_ROTATED = sizeof(struct intel_rotation_info),
+ I915_GTT_VIEW_REMAPPED = sizeof(struct intel_remapped_info),
+};
+
+static inline void assert_i915_gem_gtt_types(void)
+{
+ BUILD_BUG_ON(sizeof(struct intel_rotation_info) != 2 * sizeof(u32) + 8 * sizeof(u16));
+ BUILD_BUG_ON(sizeof(struct intel_remapped_info) != 5 * sizeof(u32) + 16 * sizeof(u16));
+
+ /* Check that rotation/remapped shares offsets for simplicity */
+ BUILD_BUG_ON(offsetof(struct intel_remapped_info, plane[0]) !=
+ offsetof(struct intel_rotation_info, plane[0]));
+ BUILD_BUG_ON(offsetofend(struct intel_remapped_info, plane[1]) !=
+ offsetofend(struct intel_rotation_info, plane[1]));
+
+ /* As we encode the size of each branch inside the union into its type,
+ * we have to be careful that each branch has a unique size.
+ */
+ switch ((enum i915_gtt_view_type)0) {
+ case I915_GTT_VIEW_NORMAL:
+ case I915_GTT_VIEW_ROTATED:
+ case I915_GTT_VIEW_REMAPPED:
+ /* gcc complains if these are identical cases */
+ break;
+ }
+}
+
+struct i915_gtt_view {
+ enum i915_gtt_view_type type;
+ union {
+ /* Members need to contain no holes/padding */
+ struct intel_rotation_info rotated;
+ struct intel_remapped_info remapped;
+ };
+};
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_clock_gating.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_clock_gating.h
new file mode 100644
index 000000000000..ce986f0e8f38
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_clock_gating.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/intel_clock_gating.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_gt_types.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_gt_types.h
new file mode 100644
index 000000000000..c15806d6c4f7
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_gt_types.h
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_GT_TYPES__
+#define __INTEL_GT_TYPES__
+
+#define intel_gt_support_legacy_fencing(gt) 0
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_mchbar_regs.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_mchbar_regs.h
new file mode 100644
index 000000000000..55b316985340
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_mchbar_regs.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/intel_mchbar_regs.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_pci_config.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_pci_config.h
new file mode 100644
index 000000000000..8c15867fd613
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_pci_config.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/intel_pci_config.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_pcode.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_pcode.h
new file mode 100644
index 000000000000..0c47661bdc6a
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_pcode.h
@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_PCODE_H__
+#define __INTEL_PCODE_H__
+
+#include "intel_uncore.h"
+#include "xe_pcode.h"
+
+static inline int
+snb_pcode_write_timeout(struct intel_uncore *uncore, u32 mbox, u32 val,
+ int fast_timeout_us, int slow_timeout_ms)
+{
+ return xe_pcode_write_timeout(__compat_uncore_to_gt(uncore), mbox, val,
+ slow_timeout_ms ?: 1);
+}
+
+static inline int
+snb_pcode_write(struct intel_uncore *uncore, u32 mbox, u32 val)
+{
+
+ return xe_pcode_write(__compat_uncore_to_gt(uncore), mbox, val);
+}
+
+static inline int
+snb_pcode_read(struct intel_uncore *uncore, u32 mbox, u32 *val, u32 *val1)
+{
+ return xe_pcode_read(__compat_uncore_to_gt(uncore), mbox, val, val1);
+}
+
+static inline int
+skl_pcode_request(struct intel_uncore *uncore, u32 mbox,
+ u32 request, u32 reply_mask, u32 reply,
+ int timeout_base_ms)
+{
+ return xe_pcode_request(__compat_uncore_to_gt(uncore), mbox, request, reply_mask, reply,
+ timeout_base_ms);
+}
+
+#endif /* __INTEL_PCODE_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_runtime_pm.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_runtime_pm.h
new file mode 100644
index 000000000000..89da3cc62f39
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_runtime_pm.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "intel_wakeref.h"
+
+#define intel_runtime_pm xe_runtime_pm
+
+static inline void disable_rpm_wakeref_asserts(void *rpm)
+{
+}
+
+static inline void enable_rpm_wakeref_asserts(void *rpm)
+{
+}
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_step.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_step.h
new file mode 100644
index 000000000000..0006ef812346
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_step.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_STEP_H__
+#define __INTEL_STEP_H__
+
+#include "xe_device_types.h"
+#include "xe_step.h"
+
+#define intel_display_step_name xe_display_step_name
+
+static inline
+const char *xe_display_step_name(struct xe_device *xe)
+{
+ return xe_step_name(xe->info.step.display);
+}
+
+#endif /* __INTEL_STEP_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_uc_fw.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_uc_fw.h
new file mode 100644
index 000000000000..009745328992
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_uc_fw.h
@@ -0,0 +1,11 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _INTEL_UC_FW_H_
+#define _INTEL_UC_FW_H_
+
+#define INTEL_UC_FIRMWARE_URL "https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git"
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h
new file mode 100644
index 000000000000..cd26ddc0f69e
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_uncore.h
@@ -0,0 +1,175 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_UNCORE_H__
+#define __INTEL_UNCORE_H__
+
+#include "xe_device.h"
+#include "xe_device_types.h"
+#include "xe_mmio.h"
+
+static inline struct xe_gt *__compat_uncore_to_gt(struct intel_uncore *uncore)
+{
+ struct xe_device *xe = container_of(uncore, struct xe_device, uncore);
+
+ return xe_root_mmio_gt(xe);
+}
+
+static inline u32 intel_uncore_read(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_read32(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline u32 intel_uncore_read8(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_read8(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline u32 intel_uncore_read16(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_read16(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline u64
+intel_uncore_read64_2x32(struct intel_uncore *uncore,
+ i915_reg_t i915_lower_reg, i915_reg_t i915_upper_reg)
+{
+ struct xe_reg lower_reg = XE_REG(i915_mmio_reg_offset(i915_lower_reg));
+ struct xe_reg upper_reg = XE_REG(i915_mmio_reg_offset(i915_upper_reg));
+ u32 upper, lower, old_upper;
+ int loop = 0;
+
+ upper = xe_mmio_read32(__compat_uncore_to_gt(uncore), upper_reg);
+ do {
+ old_upper = upper;
+ lower = xe_mmio_read32(__compat_uncore_to_gt(uncore), lower_reg);
+ upper = xe_mmio_read32(__compat_uncore_to_gt(uncore), upper_reg);
+ } while (upper != old_upper && loop++ < 2);
+
+ return (u64)upper << 32 | lower;
+}
+
+static inline void intel_uncore_posting_read(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ xe_mmio_read32(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline void intel_uncore_write(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 val)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ xe_mmio_write32(__compat_uncore_to_gt(uncore), reg, val);
+}
+
+static inline u32 intel_uncore_rmw(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 clear, u32 set)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_rmw32(__compat_uncore_to_gt(uncore), reg, clear, set);
+}
+
+static inline int intel_wait_for_register(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 mask,
+ u32 value, unsigned int timeout)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_wait32(__compat_uncore_to_gt(uncore), reg, mask, value,
+ timeout * USEC_PER_MSEC, NULL, false);
+}
+
+static inline int intel_wait_for_register_fw(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 mask,
+ u32 value, unsigned int timeout)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_wait32(__compat_uncore_to_gt(uncore), reg, mask, value,
+ timeout * USEC_PER_MSEC, NULL, false);
+}
+
+static inline int
+__intel_wait_for_register(struct intel_uncore *uncore, i915_reg_t i915_reg,
+ u32 mask, u32 value, unsigned int fast_timeout_us,
+ unsigned int slow_timeout_ms, u32 *out_value)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_wait32(__compat_uncore_to_gt(uncore), reg, mask, value,
+ fast_timeout_us + 1000 * slow_timeout_ms,
+ out_value, false);
+}
+
+static inline u32 intel_uncore_read_fw(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_read32(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline void intel_uncore_write_fw(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 val)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ xe_mmio_write32(__compat_uncore_to_gt(uncore), reg, val);
+}
+
+static inline u32 intel_uncore_read_notrace(struct intel_uncore *uncore,
+ i915_reg_t i915_reg)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ return xe_mmio_read32(__compat_uncore_to_gt(uncore), reg);
+}
+
+static inline void intel_uncore_write_notrace(struct intel_uncore *uncore,
+ i915_reg_t i915_reg, u32 val)
+{
+ struct xe_reg reg = XE_REG(i915_mmio_reg_offset(i915_reg));
+
+ xe_mmio_write32(__compat_uncore_to_gt(uncore), reg, val);
+}
+
+static inline void __iomem *intel_uncore_regs(struct intel_uncore *uncore)
+{
+ struct xe_device *xe = container_of(uncore, struct xe_device, uncore);
+
+ return xe_device_get_root_tile(xe)->mmio.regs;
+}
+
+/*
+ * The raw_reg_{read,write} macros are intended as a micro-optimization for
+ * interrupt handlers so that the pointer indirection on uncore->regs can
+ * be computed once (and presumably cached in a register) instead of generating
+ * extra load instructions for each MMIO access.
+ *
+ * Given that these macros are only intended for non-GSI interrupt registers
+ * (and the goal is to avoid extra instructions generated by the compiler),
+ * these macros do not account for uncore->gsi_offset. Any caller that needs
+ * to use these macros on a GSI register is responsible for adding the
+ * appropriate GSI offset to the 'base' parameter.
+ */
+#define raw_reg_read(base, reg) \
+ readl(base + i915_mmio_reg_offset(reg))
+#define raw_reg_write(base, reg, value) \
+ writel(value, base + i915_mmio_reg_offset(reg))
+
+#endif /* __INTEL_UNCORE_H__ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/intel_wakeref.h b/drivers/gpu/drm/xe/compat-i915-headers/intel_wakeref.h
new file mode 100644
index 000000000000..ecb1c0707706
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/intel_wakeref.h
@@ -0,0 +1,8 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/types.h>
+
+typedef unsigned long intel_wakeref_t;
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/pxp/intel_pxp.h b/drivers/gpu/drm/xe/compat-i915-headers/pxp/intel_pxp.h
new file mode 100644
index 000000000000..c2c30ece8f77
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/pxp/intel_pxp.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_PXP_H__
+#define __INTEL_PXP_H__
+
+#include <linux/errno.h>
+#include <linux/types.h>
+
+struct drm_i915_gem_object;
+struct intel_pxp;
+
+static inline int intel_pxp_key_check(struct intel_pxp *pxp,
+ struct drm_i915_gem_object *obj,
+ bool assign)
+{
+ return -ENODEV;
+}
+
+static inline bool
+i915_gem_object_is_protected(const struct drm_i915_gem_object *obj)
+{
+ return false;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_dram.h b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_dram.h
new file mode 100644
index 000000000000..65707e20c557
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_dram.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../../i915/soc/intel_dram.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h
new file mode 100644
index 000000000000..33c5257b3a71
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_gmch.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../../i915/soc/intel_gmch.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_pch.h b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_pch.h
new file mode 100644
index 000000000000..9c46556d33a4
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/soc/intel_pch.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../../i915/soc/intel_pch.h"
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband.h b/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband.h
new file mode 100644
index 000000000000..ec6f12de5727
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband.h
@@ -0,0 +1,132 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2013-2021 Intel Corporation
+ */
+
+#ifndef _VLV_SIDEBAND_H_
+#define _VLV_SIDEBAND_H_
+
+#include <linux/types.h>
+
+#include "vlv_sideband_reg.h"
+
+enum pipe;
+struct drm_i915_private;
+
+enum {
+ VLV_IOSF_SB_BUNIT,
+ VLV_IOSF_SB_CCK,
+ VLV_IOSF_SB_CCU,
+ VLV_IOSF_SB_DPIO,
+ VLV_IOSF_SB_FLISDSI,
+ VLV_IOSF_SB_GPIO,
+ VLV_IOSF_SB_NC,
+ VLV_IOSF_SB_PUNIT,
+};
+
+static inline void vlv_iosf_sb_get(struct drm_i915_private *i915, unsigned long ports)
+{
+}
+static inline u32 vlv_iosf_sb_read(struct drm_i915_private *i915, u8 port, u32 reg)
+{
+ return 0;
+}
+static inline void vlv_iosf_sb_write(struct drm_i915_private *i915,
+ u8 port, u32 reg, u32 val)
+{
+}
+static inline void vlv_iosf_sb_put(struct drm_i915_private *i915, unsigned long ports)
+{
+}
+static inline void vlv_bunit_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_bunit_read(struct drm_i915_private *i915, u32 reg)
+{
+ return 0;
+}
+static inline void vlv_bunit_write(struct drm_i915_private *i915, u32 reg, u32 val)
+{
+}
+static inline void vlv_bunit_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_cck_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_cck_read(struct drm_i915_private *i915, u32 reg)
+{
+ return 0;
+}
+static inline void vlv_cck_write(struct drm_i915_private *i915, u32 reg, u32 val)
+{
+}
+static inline void vlv_cck_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_ccu_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_ccu_read(struct drm_i915_private *i915, u32 reg)
+{
+ return 0;
+}
+static inline void vlv_ccu_write(struct drm_i915_private *i915, u32 reg, u32 val)
+{
+}
+static inline void vlv_ccu_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_dpio_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_dpio_read(struct drm_i915_private *i915, int pipe, int reg)
+{
+ return 0;
+}
+static inline void vlv_dpio_write(struct drm_i915_private *i915,
+ int pipe, int reg, u32 val)
+{
+}
+static inline void vlv_dpio_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_flisdsi_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_flisdsi_read(struct drm_i915_private *i915, u32 reg)
+{
+ return 0;
+}
+static inline void vlv_flisdsi_write(struct drm_i915_private *i915, u32 reg, u32 val)
+{
+}
+static inline void vlv_flisdsi_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_nc_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_nc_read(struct drm_i915_private *i915, u8 addr)
+{
+ return 0;
+}
+static inline void vlv_nc_put(struct drm_i915_private *i915)
+{
+}
+static inline void vlv_punit_get(struct drm_i915_private *i915)
+{
+}
+static inline u32 vlv_punit_read(struct drm_i915_private *i915, u32 addr)
+{
+ return 0;
+}
+static inline int vlv_punit_write(struct drm_i915_private *i915, u32 addr, u32 val)
+{
+ return 0;
+}
+static inline void vlv_punit_put(struct drm_i915_private *i915)
+{
+}
+
+#endif /* _VLV_SIDEBAND_H_ */
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband_reg.h b/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband_reg.h
new file mode 100644
index 000000000000..949f134ce3cf
--- /dev/null
+++ b/drivers/gpu/drm/xe/compat-i915-headers/vlv_sideband_reg.h
@@ -0,0 +1,6 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "../../i915/vlv_sideband_reg.h"
diff --git a/drivers/gpu/drm/xe/display/ext/i915_irq.c b/drivers/gpu/drm/xe/display/ext/i915_irq.c
new file mode 100644
index 000000000000..bee191a4a97d
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/ext/i915_irq.c
@@ -0,0 +1,77 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "i915_drv.h"
+#include "i915_irq.h"
+#include "i915_reg.h"
+#include "intel_uncore.h"
+
+void gen3_irq_reset(struct intel_uncore *uncore, i915_reg_t imr,
+ i915_reg_t iir, i915_reg_t ier)
+{
+ intel_uncore_write(uncore, imr, 0xffffffff);
+ intel_uncore_posting_read(uncore, imr);
+
+ intel_uncore_write(uncore, ier, 0);
+
+ /* IIR can theoretically queue up two events. Be paranoid. */
+ intel_uncore_write(uncore, iir, 0xffffffff);
+ intel_uncore_posting_read(uncore, iir);
+ intel_uncore_write(uncore, iir, 0xffffffff);
+ intel_uncore_posting_read(uncore, iir);
+}
+
+/*
+ * We should clear IMR at preinstall/uninstall, and just check at postinstall.
+ */
+void gen3_assert_iir_is_zero(struct intel_uncore *uncore, i915_reg_t reg)
+{
+ struct xe_device *xe = container_of(uncore, struct xe_device, uncore);
+ u32 val = intel_uncore_read(uncore, reg);
+
+ if (val == 0)
+ return;
+
+ drm_WARN(&xe->drm, 1,
+ "Interrupt register 0x%x is not zero: 0x%08x\n",
+ i915_mmio_reg_offset(reg), val);
+ intel_uncore_write(uncore, reg, 0xffffffff);
+ intel_uncore_posting_read(uncore, reg);
+ intel_uncore_write(uncore, reg, 0xffffffff);
+ intel_uncore_posting_read(uncore, reg);
+}
+
+void gen3_irq_init(struct intel_uncore *uncore,
+ i915_reg_t imr, u32 imr_val,
+ i915_reg_t ier, u32 ier_val,
+ i915_reg_t iir)
+{
+ gen3_assert_iir_is_zero(uncore, iir);
+
+ intel_uncore_write(uncore, ier, ier_val);
+ intel_uncore_write(uncore, imr, imr_val);
+ intel_uncore_posting_read(uncore, imr);
+}
+
+bool intel_irqs_enabled(struct xe_device *xe)
+{
+ /*
+ * XXX: i915 has a racy handling of the irq.enabled, since it doesn't
+ * lock its transitions. Because of that, the irq.enabled sometimes
+ * is not read with the irq.lock in place.
+ * However, the most critical cases like vblank and page flips are
+ * properly using the locks.
+ * We cannot take the lock in here or run any kind of assert because
+ * of i915 inconsistency.
+ * But at this point the xe irq is better protected against races,
+ * although the full solution would be protecting the i915 side.
+ */
+ return xe->irq.enabled;
+}
+
+void intel_synchronize_irq(struct xe_device *xe)
+{
+ synchronize_irq(to_pci_dev(xe->drm.dev)->irq);
+}
diff --git a/drivers/gpu/drm/xe/display/ext/i915_utils.c b/drivers/gpu/drm/xe/display/ext/i915_utils.c
new file mode 100644
index 000000000000..43b10a2cc508
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/ext/i915_utils.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "i915_drv.h"
+
+bool i915_vtd_active(struct drm_i915_private *i915)
+{
+ if (device_iommu_mapped(i915->drm.dev))
+ return true;
+
+ /* Running as a guest, we assume the host is enforcing VT'd */
+ return i915_run_as_guest();
+}
+
+#if IS_ENABLED(CONFIG_DRM_I915_DEBUG)
+
+/* i915 specific, just put here for shutting it up */
+int __i915_inject_probe_error(struct drm_i915_private *i915, int err,
+ const char *func, int line)
+{
+ return 0;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/display/intel_fb_bo.c b/drivers/gpu/drm/xe/display/intel_fb_bo.c
new file mode 100644
index 000000000000..b21da7b745a5
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/intel_fb_bo.c
@@ -0,0 +1,74 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include <drm/drm_modeset_helper.h>
+
+#include "i915_drv.h"
+#include "intel_display_types.h"
+#include "intel_fb_bo.h"
+
+void intel_fb_bo_framebuffer_fini(struct xe_bo *bo)
+{
+ if (bo->flags & XE_BO_CREATE_PINNED_BIT) {
+ /* Unpin our kernel fb first */
+ xe_bo_lock(bo, false);
+ xe_bo_unpin(bo);
+ xe_bo_unlock(bo);
+ }
+ xe_bo_put(bo);
+}
+
+int intel_fb_bo_framebuffer_init(struct intel_framebuffer *intel_fb,
+ struct xe_bo *bo,
+ struct drm_mode_fb_cmd2 *mode_cmd)
+{
+ struct drm_i915_private *i915 = to_i915(bo->ttm.base.dev);
+ int ret;
+
+ xe_bo_get(bo);
+
+ ret = ttm_bo_reserve(&bo->ttm, true, false, NULL);
+ if (ret)
+ return ret;
+
+ if (!(bo->flags & XE_BO_SCANOUT_BIT)) {
+ /*
+ * XE_BO_SCANOUT_BIT should ideally be set at creation, or is
+ * automatically set when creating FB. We cannot change caching
+ * mode when the boect is VM_BINDed, so we can only set
+ * coherency with display when unbound.
+ */
+ if (XE_IOCTL_DBG(i915, !list_empty(&bo->ttm.base.gpuva.list))) {
+ ttm_bo_unreserve(&bo->ttm);
+ return -EINVAL;
+ }
+ bo->flags |= XE_BO_SCANOUT_BIT;
+ }
+ ttm_bo_unreserve(&bo->ttm);
+
+ return ret;
+}
+
+struct xe_bo *intel_fb_bo_lookup_valid_bo(struct drm_i915_private *i915,
+ struct drm_file *filp,
+ const struct drm_mode_fb_cmd2 *mode_cmd)
+{
+ struct drm_i915_gem_object *bo;
+ struct drm_gem_object *gem = drm_gem_object_lookup(filp, mode_cmd->handles[0]);
+
+ if (!gem)
+ return ERR_PTR(-ENOENT);
+
+ bo = gem_to_xe_bo(gem);
+ /* Require vram placement or dma-buf import */
+ if (IS_DGFX(i915) &&
+ !xe_bo_can_migrate(gem_to_xe_bo(gem), XE_PL_VRAM0) &&
+ bo->ttm.type != ttm_bo_type_sg) {
+ drm_gem_object_put(gem);
+ return ERR_PTR(-EREMOTE);
+ }
+
+ return bo;
+}
diff --git a/drivers/gpu/drm/xe/display/intel_fb_bo.h b/drivers/gpu/drm/xe/display/intel_fb_bo.h
new file mode 100644
index 000000000000..5d365b925b7a
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/intel_fb_bo.h
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef __INTEL_FB_BO_H__
+#define __INTEL_FB_BO_H__
+
+struct drm_file;
+struct drm_mode_fb_cmd2;
+struct drm_i915_private;
+struct intel_framebuffer;
+struct xe_bo;
+
+void intel_fb_bo_framebuffer_fini(struct xe_bo *bo);
+int intel_fb_bo_framebuffer_init(struct intel_framebuffer *intel_fb,
+ struct xe_bo *bo,
+ struct drm_mode_fb_cmd2 *mode_cmd);
+
+struct xe_bo *intel_fb_bo_lookup_valid_bo(struct drm_i915_private *i915,
+ struct drm_file *filp,
+ const struct drm_mode_fb_cmd2 *mode_cmd);
+
+#endif
diff --git a/drivers/gpu/drm/xe/display/intel_fbdev_fb.c b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c
new file mode 100644
index 000000000000..51ae3561fd0d
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/intel_fbdev_fb.c
@@ -0,0 +1,104 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "intel_fbdev_fb.h"
+
+#include <drm/drm_fb_helper.h>
+
+#include "xe_gt.h"
+#include "xe_ttm_stolen_mgr.h"
+
+#include "i915_drv.h"
+#include "intel_display_types.h"
+
+struct drm_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
+ struct drm_fb_helper_surface_size *sizes)
+{
+ struct drm_framebuffer *fb;
+ struct drm_device *dev = helper->dev;
+ struct drm_i915_private *dev_priv = to_i915(dev);
+ struct drm_mode_fb_cmd2 mode_cmd = {};
+ struct drm_i915_gem_object *obj;
+ int size;
+
+ /* we don't do packed 24bpp */
+ if (sizes->surface_bpp == 24)
+ sizes->surface_bpp = 32;
+
+ mode_cmd.width = sizes->surface_width;
+ mode_cmd.height = sizes->surface_height;
+
+ mode_cmd.pitches[0] = ALIGN(mode_cmd.width *
+ DIV_ROUND_UP(sizes->surface_bpp, 8), XE_PAGE_SIZE);
+ mode_cmd.pixel_format = drm_mode_legacy_fb_format(sizes->surface_bpp,
+ sizes->surface_depth);
+
+ size = mode_cmd.pitches[0] * mode_cmd.height;
+ size = PAGE_ALIGN(size);
+ obj = ERR_PTR(-ENODEV);
+
+ if (!IS_DGFX(dev_priv)) {
+ obj = xe_bo_create_pin_map(dev_priv, xe_device_get_root_tile(dev_priv),
+ NULL, size,
+ ttm_bo_type_kernel, XE_BO_SCANOUT_BIT |
+ XE_BO_CREATE_STOLEN_BIT |
+ XE_BO_CREATE_PINNED_BIT);
+ if (!IS_ERR(obj))
+ drm_info(&dev_priv->drm, "Allocated fbdev into stolen\n");
+ else
+ drm_info(&dev_priv->drm, "Allocated fbdev into stolen failed: %li\n", PTR_ERR(obj));
+ }
+ if (IS_ERR(obj)) {
+ obj = xe_bo_create_pin_map(dev_priv, xe_device_get_root_tile(dev_priv), NULL, size,
+ ttm_bo_type_kernel, XE_BO_SCANOUT_BIT |
+ XE_BO_CREATE_VRAM_IF_DGFX(xe_device_get_root_tile(dev_priv)) |
+ XE_BO_CREATE_PINNED_BIT);
+ }
+
+ if (IS_ERR(obj)) {
+ drm_err(&dev_priv->drm, "failed to allocate framebuffer (%pe)\n", obj);
+ fb = ERR_PTR(-ENOMEM);
+ goto err;
+ }
+
+ fb = intel_framebuffer_create(obj, &mode_cmd);
+ if (IS_ERR(fb)) {
+ xe_bo_unpin_map_no_vm(obj);
+ goto err;
+ }
+
+ drm_gem_object_put(intel_bo_to_drm_bo(obj));
+ return fb;
+
+err:
+ return fb;
+}
+
+int intel_fbdev_fb_fill_info(struct drm_i915_private *i915, struct fb_info *info,
+ struct drm_i915_gem_object *obj, struct i915_vma *vma)
+{
+ struct pci_dev *pdev = to_pci_dev(i915->drm.dev);
+
+ if (!(obj->flags & XE_BO_CREATE_SYSTEM_BIT)) {
+ if (obj->flags & XE_BO_CREATE_STOLEN_BIT)
+ info->fix.smem_start = xe_ttm_stolen_io_offset(obj, 0);
+ else
+ info->fix.smem_start =
+ pci_resource_start(pdev, 2) +
+ xe_bo_addr(obj, 0, XE_PAGE_SIZE);
+
+ info->fix.smem_len = obj->ttm.base.size;
+ } else {
+ /* XXX: Pure fiction, as the BO may not be physically accessible.. */
+ info->fix.smem_start = 0;
+ info->fix.smem_len = obj->ttm.base.size;
+ }
+ XE_WARN_ON(iosys_map_is_null(&obj->vmap));
+
+ info->screen_base = obj->vmap.vaddr_iomem;
+ info->screen_size = intel_bo_to_drm_bo(obj)->size;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/display/intel_fbdev_fb.h b/drivers/gpu/drm/xe/display/intel_fbdev_fb.h
new file mode 100644
index 000000000000..ea186772e0bb
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/intel_fbdev_fb.h
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef __INTEL_FBDEV_FB_H__
+#define __INTEL_FBDEV_FB_H__
+
+struct drm_fb_helper;
+struct drm_fb_helper_surface_size;
+struct drm_i915_gem_object;
+struct drm_i915_private;
+struct fb_info;
+struct i915_vma;
+
+struct drm_framebuffer *intel_fbdev_fb_alloc(struct drm_fb_helper *helper,
+ struct drm_fb_helper_surface_size *sizes);
+int intel_fbdev_fb_fill_info(struct drm_i915_private *i915, struct fb_info *info,
+ struct drm_i915_gem_object *obj, struct i915_vma *vma);
+
+#endif
diff --git a/drivers/gpu/drm/xe/display/xe_display_misc.c b/drivers/gpu/drm/xe/display/xe_display_misc.c
new file mode 100644
index 000000000000..242c2ef4ca93
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_display_misc.c
@@ -0,0 +1,16 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "intel_display_types.h"
+
+struct pci_dev;
+
+unsigned int intel_gmch_vga_set_decode(struct pci_dev *pdev, bool enable_decode);
+
+unsigned int intel_gmch_vga_set_decode(struct pci_dev *pdev, bool enable_decode)
+{
+ /* ToDo: Implement the actual handling of vga decode */
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/display/xe_display_rps.c b/drivers/gpu/drm/xe/display/xe_display_rps.c
new file mode 100644
index 000000000000..ab21c581c192
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_display_rps.c
@@ -0,0 +1,17 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "intel_display_rps.h"
+
+void intel_display_rps_boost_after_vblank(struct drm_crtc *crtc,
+ struct dma_fence *fence)
+{
+}
+
+void intel_display_rps_mark_interactive(struct drm_i915_private *i915,
+ struct intel_atomic_state *state,
+ bool interactive)
+{
+}
diff --git a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
new file mode 100644
index 000000000000..27c2fb1c002a
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2023, Intel Corporation.
+ */
+
+#include "i915_drv.h"
+#include "i915_vma.h"
+#include "intel_display_types.h"
+#include "intel_dsb_buffer.h"
+#include "xe_bo.h"
+#include "xe_gt.h"
+
+u32 intel_dsb_buffer_ggtt_offset(struct intel_dsb_buffer *dsb_buf)
+{
+ return xe_bo_ggtt_addr(dsb_buf->vma->bo);
+}
+
+void intel_dsb_buffer_write(struct intel_dsb_buffer *dsb_buf, u32 idx, u32 val)
+{
+ iosys_map_wr(&dsb_buf->vma->bo->vmap, idx * 4, u32, val);
+}
+
+u32 intel_dsb_buffer_read(struct intel_dsb_buffer *dsb_buf, u32 idx)
+{
+ return iosys_map_rd(&dsb_buf->vma->bo->vmap, idx * 4, u32);
+}
+
+void intel_dsb_buffer_memset(struct intel_dsb_buffer *dsb_buf, u32 idx, u32 val, size_t size)
+{
+ WARN_ON(idx > (dsb_buf->buf_size - size) / sizeof(*dsb_buf->cmd_buf));
+
+ iosys_map_memset(&dsb_buf->vma->bo->vmap, idx * 4, val, size);
+}
+
+bool intel_dsb_buffer_create(struct intel_crtc *crtc, struct intel_dsb_buffer *dsb_buf, size_t size)
+{
+ struct drm_i915_private *i915 = to_i915(crtc->base.dev);
+ struct drm_i915_gem_object *obj;
+ struct i915_vma *vma;
+
+ vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+ if (!vma)
+ return false;
+
+ obj = xe_bo_create_pin_map(i915, xe_device_get_root_tile(i915),
+ NULL, PAGE_ALIGN(size),
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(xe_device_get_root_tile(i915)) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(obj)) {
+ kfree(vma);
+ return false;
+ }
+
+ vma->bo = obj;
+ dsb_buf->vma = vma;
+ dsb_buf->buf_size = size;
+
+ return true;
+}
+
+void intel_dsb_buffer_cleanup(struct intel_dsb_buffer *dsb_buf)
+{
+ xe_bo_unpin_map_no_vm(dsb_buf->vma->bo);
+ kfree(dsb_buf->vma);
+}
+
+void intel_dsb_buffer_flush_map(struct intel_dsb_buffer *dsb_buf)
+{
+ /* TODO: add xe specific flush_map() for dsb buffer object. */
+}
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
new file mode 100644
index 000000000000..722c84a56607
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -0,0 +1,384 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "i915_drv.h"
+#include "intel_display_types.h"
+#include "intel_dpt.h"
+#include "intel_fb.h"
+#include "intel_fb_pin.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+
+#include <drm/ttm/ttm_bo.h>
+
+static void
+write_dpt_rotated(struct xe_bo *bo, struct iosys_map *map, u32 *dpt_ofs, u32 bo_ofs,
+ u32 width, u32 height, u32 src_stride, u32 dst_stride)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
+ u32 column, row;
+
+ /* TODO: Maybe rewrite so we can traverse the bo addresses sequentially,
+ * by writing dpt/ggtt in a different order?
+ */
+
+ for (column = 0; column < width; column++) {
+ u32 src_idx = src_stride * (height - 1) + column + bo_ofs;
+
+ for (row = 0; row < height; row++) {
+ u64 pte = ggtt->pt_ops->pte_encode_bo(bo, src_idx * XE_PAGE_SIZE,
+ xe->pat.idx[XE_CACHE_WB]);
+
+ iosys_map_wr(map, *dpt_ofs, u64, pte);
+ *dpt_ofs += 8;
+ src_idx -= src_stride;
+ }
+
+ /* The DE ignores the PTEs for the padding tiles */
+ *dpt_ofs += (dst_stride - height) * 8;
+ }
+
+ /* Align to next page */
+ *dpt_ofs = ALIGN(*dpt_ofs, 4096);
+}
+
+static void
+write_dpt_remapped(struct xe_bo *bo, struct iosys_map *map, u32 *dpt_ofs,
+ u32 bo_ofs, u32 width, u32 height, u32 src_stride,
+ u32 dst_stride)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
+ u64 (*pte_encode_bo)(struct xe_bo *bo, u64 bo_offset, u16 pat_index)
+ = ggtt->pt_ops->pte_encode_bo;
+ u32 column, row;
+
+ for (row = 0; row < height; row++) {
+ u32 src_idx = src_stride * row + bo_ofs;
+
+ for (column = 0; column < width; column++) {
+ iosys_map_wr(map, *dpt_ofs, u64,
+ pte_encode_bo(bo, src_idx * XE_PAGE_SIZE,
+ xe->pat.idx[XE_CACHE_WB]));
+
+ *dpt_ofs += 8;
+ src_idx++;
+ }
+
+ /* The DE ignores the PTEs for the padding tiles */
+ *dpt_ofs += (dst_stride - width) * 8;
+ }
+
+ /* Align to next page */
+ *dpt_ofs = ALIGN(*dpt_ofs, 4096);
+}
+
+static int __xe_pin_fb_vma_dpt(struct intel_framebuffer *fb,
+ const struct i915_gtt_view *view,
+ struct i915_vma *vma)
+{
+ struct xe_device *xe = to_xe_device(fb->base.dev);
+ struct xe_tile *tile0 = xe_device_get_root_tile(xe);
+ struct xe_ggtt *ggtt = tile0->mem.ggtt;
+ struct xe_bo *bo = intel_fb_obj(&fb->base), *dpt;
+ u32 dpt_size, size = bo->ttm.base.size;
+
+ if (view->type == I915_GTT_VIEW_NORMAL)
+ dpt_size = ALIGN(size / XE_PAGE_SIZE * 8, XE_PAGE_SIZE);
+ else if (view->type == I915_GTT_VIEW_REMAPPED)
+ dpt_size = ALIGN(intel_remapped_info_size(&fb->remapped_view.gtt.remapped) * 8,
+ XE_PAGE_SIZE);
+ else
+ /* display uses 4K tiles instead of bytes here, convert to entries.. */
+ dpt_size = ALIGN(intel_rotation_info_size(&view->rotated) * 8,
+ XE_PAGE_SIZE);
+
+ if (IS_DGFX(xe))
+ dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM0_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ else
+ dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_STOLEN_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(dpt))
+ dpt = xe_bo_create_pin_map(xe, tile0, NULL, dpt_size,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_SYSTEM_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(dpt))
+ return PTR_ERR(dpt);
+
+ if (view->type == I915_GTT_VIEW_NORMAL) {
+ u32 x;
+
+ for (x = 0; x < size / XE_PAGE_SIZE; x++) {
+ u64 pte = ggtt->pt_ops->pte_encode_bo(bo, x * XE_PAGE_SIZE,
+ xe->pat.idx[XE_CACHE_WB]);
+
+ iosys_map_wr(&dpt->vmap, x * 8, u64, pte);
+ }
+ } else if (view->type == I915_GTT_VIEW_REMAPPED) {
+ const struct intel_remapped_info *remap_info = &view->remapped;
+ u32 i, dpt_ofs = 0;
+
+ for (i = 0; i < ARRAY_SIZE(remap_info->plane); i++)
+ write_dpt_remapped(bo, &dpt->vmap, &dpt_ofs,
+ remap_info->plane[i].offset,
+ remap_info->plane[i].width,
+ remap_info->plane[i].height,
+ remap_info->plane[i].src_stride,
+ remap_info->plane[i].dst_stride);
+
+ } else {
+ const struct intel_rotation_info *rot_info = &view->rotated;
+ u32 i, dpt_ofs = 0;
+
+ for (i = 0; i < ARRAY_SIZE(rot_info->plane); i++)
+ write_dpt_rotated(bo, &dpt->vmap, &dpt_ofs,
+ rot_info->plane[i].offset,
+ rot_info->plane[i].width,
+ rot_info->plane[i].height,
+ rot_info->plane[i].src_stride,
+ rot_info->plane[i].dst_stride);
+ }
+
+ vma->dpt = dpt;
+ vma->node = dpt->ggtt_node;
+ return 0;
+}
+
+static void
+write_ggtt_rotated(struct xe_bo *bo, struct xe_ggtt *ggtt, u32 *ggtt_ofs, u32 bo_ofs,
+ u32 width, u32 height, u32 src_stride, u32 dst_stride)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ u32 column, row;
+
+ for (column = 0; column < width; column++) {
+ u32 src_idx = src_stride * (height - 1) + column + bo_ofs;
+
+ for (row = 0; row < height; row++) {
+ u64 pte = ggtt->pt_ops->pte_encode_bo(bo, src_idx * XE_PAGE_SIZE,
+ xe->pat.idx[XE_CACHE_WB]);
+
+ xe_ggtt_set_pte(ggtt, *ggtt_ofs, pte);
+ *ggtt_ofs += XE_PAGE_SIZE;
+ src_idx -= src_stride;
+ }
+
+ /* The DE ignores the PTEs for the padding tiles */
+ *ggtt_ofs += (dst_stride - height) * XE_PAGE_SIZE;
+ }
+}
+
+static int __xe_pin_fb_vma_ggtt(struct intel_framebuffer *fb,
+ const struct i915_gtt_view *view,
+ struct i915_vma *vma)
+{
+ struct xe_bo *bo = intel_fb_obj(&fb->base);
+ struct xe_device *xe = to_xe_device(fb->base.dev);
+ struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
+ u32 align;
+ int ret;
+
+ /* TODO: Consider sharing framebuffer mapping?
+ * embed i915_vma inside intel_framebuffer
+ */
+ xe_device_mem_access_get(tile_to_xe(ggtt->tile));
+ ret = mutex_lock_interruptible(&ggtt->lock);
+ if (ret)
+ goto out;
+
+ align = XE_PAGE_SIZE;
+ if (xe_bo_is_vram(bo) && ggtt->flags & XE_GGTT_FLAGS_64K)
+ align = max_t(u32, align, SZ_64K);
+
+ if (bo->ggtt_node.size && view->type == I915_GTT_VIEW_NORMAL) {
+ vma->node = bo->ggtt_node;
+ } else if (view->type == I915_GTT_VIEW_NORMAL) {
+ u32 x, size = bo->ttm.base.size;
+
+ ret = xe_ggtt_insert_special_node_locked(ggtt, &vma->node, size,
+ align, 0);
+ if (ret)
+ goto out_unlock;
+
+ for (x = 0; x < size; x += XE_PAGE_SIZE) {
+ u64 pte = ggtt->pt_ops->pte_encode_bo(bo, x,
+ xe->pat.idx[XE_CACHE_WB]);
+
+ xe_ggtt_set_pte(ggtt, vma->node.start + x, pte);
+ }
+ } else {
+ u32 i, ggtt_ofs;
+ const struct intel_rotation_info *rot_info = &view->rotated;
+
+ /* display seems to use tiles instead of bytes here, so convert it back.. */
+ u32 size = intel_rotation_info_size(rot_info) * XE_PAGE_SIZE;
+
+ ret = xe_ggtt_insert_special_node_locked(ggtt, &vma->node, size,
+ align, 0);
+ if (ret)
+ goto out_unlock;
+
+ ggtt_ofs = vma->node.start;
+
+ for (i = 0; i < ARRAY_SIZE(rot_info->plane); i++)
+ write_ggtt_rotated(bo, ggtt, &ggtt_ofs,
+ rot_info->plane[i].offset,
+ rot_info->plane[i].width,
+ rot_info->plane[i].height,
+ rot_info->plane[i].src_stride,
+ rot_info->plane[i].dst_stride);
+ }
+
+ xe_ggtt_invalidate(ggtt);
+out_unlock:
+ mutex_unlock(&ggtt->lock);
+out:
+ xe_device_mem_access_put(tile_to_xe(ggtt->tile));
+ return ret;
+}
+
+static struct i915_vma *__xe_pin_fb_vma(struct intel_framebuffer *fb,
+ const struct i915_gtt_view *view)
+{
+ struct drm_device *dev = fb->base.dev;
+ struct xe_device *xe = to_xe_device(dev);
+ struct i915_vma *vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+ struct xe_bo *bo = intel_fb_obj(&fb->base);
+ int ret;
+
+ if (!vma)
+ return ERR_PTR(-ENODEV);
+
+ if (IS_DGFX(to_xe_device(bo->ttm.base.dev)) &&
+ intel_fb_rc_ccs_cc_plane(&fb->base) >= 0 &&
+ !(bo->flags & XE_BO_NEEDS_CPU_ACCESS)) {
+ struct xe_tile *tile = xe_device_get_root_tile(xe);
+
+ /*
+ * If we need to able to access the clear-color value stored in
+ * the buffer, then we require that such buffers are also CPU
+ * accessible. This is important on small-bar systems where
+ * only some subset of VRAM is CPU accessible.
+ */
+ if (tile->mem.vram.io_size < tile->mem.vram.usable_size) {
+ ret = -EINVAL;
+ goto err;
+ }
+ }
+
+ /*
+ * Pin the framebuffer, we can't use xe_bo_(un)pin functions as the
+ * assumptions are incorrect for framebuffers
+ */
+ ret = ttm_bo_reserve(&bo->ttm, false, false, NULL);
+ if (ret)
+ goto err;
+
+ if (IS_DGFX(xe))
+ ret = xe_bo_migrate(bo, XE_PL_VRAM0);
+ else
+ ret = xe_bo_validate(bo, NULL, true);
+ if (!ret)
+ ttm_bo_pin(&bo->ttm);
+ ttm_bo_unreserve(&bo->ttm);
+ if (ret)
+ goto err;
+
+ vma->bo = bo;
+ if (intel_fb_uses_dpt(&fb->base))
+ ret = __xe_pin_fb_vma_dpt(fb, view, vma);
+ else
+ ret = __xe_pin_fb_vma_ggtt(fb, view, vma);
+ if (ret)
+ goto err_unpin;
+
+ return vma;
+
+err_unpin:
+ ttm_bo_reserve(&bo->ttm, false, false, NULL);
+ ttm_bo_unpin(&bo->ttm);
+ ttm_bo_unreserve(&bo->ttm);
+err:
+ kfree(vma);
+ return ERR_PTR(ret);
+}
+
+static void __xe_unpin_fb_vma(struct i915_vma *vma)
+{
+ struct xe_device *xe = to_xe_device(vma->bo->ttm.base.dev);
+ struct xe_ggtt *ggtt = xe_device_get_root_tile(xe)->mem.ggtt;
+
+ if (vma->dpt)
+ xe_bo_unpin_map_no_vm(vma->dpt);
+ else if (!drm_mm_node_allocated(&vma->bo->ggtt_node) ||
+ vma->bo->ggtt_node.start != vma->node.start)
+ xe_ggtt_remove_node(ggtt, &vma->node);
+
+ ttm_bo_reserve(&vma->bo->ttm, false, false, NULL);
+ ttm_bo_unpin(&vma->bo->ttm);
+ ttm_bo_unreserve(&vma->bo->ttm);
+ kfree(vma);
+}
+
+struct i915_vma *
+intel_pin_and_fence_fb_obj(struct drm_framebuffer *fb,
+ bool phys_cursor,
+ const struct i915_gtt_view *view,
+ bool uses_fence,
+ unsigned long *out_flags)
+{
+ *out_flags = 0;
+
+ return __xe_pin_fb_vma(to_intel_framebuffer(fb), view);
+}
+
+void intel_unpin_fb_vma(struct i915_vma *vma, unsigned long flags)
+{
+ __xe_unpin_fb_vma(vma);
+}
+
+int intel_plane_pin_fb(struct intel_plane_state *plane_state)
+{
+ struct drm_framebuffer *fb = plane_state->hw.fb;
+ struct xe_bo *bo = intel_fb_obj(fb);
+ struct i915_vma *vma;
+
+ /* We reject creating !SCANOUT fb's, so this is weird.. */
+ drm_WARN_ON(bo->ttm.base.dev, !(bo->flags & XE_BO_SCANOUT_BIT));
+
+ vma = __xe_pin_fb_vma(to_intel_framebuffer(fb), &plane_state->view.gtt);
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
+
+ plane_state->ggtt_vma = vma;
+ return 0;
+}
+
+void intel_plane_unpin_fb(struct intel_plane_state *old_plane_state)
+{
+ __xe_unpin_fb_vma(old_plane_state->ggtt_vma);
+ old_plane_state->ggtt_vma = NULL;
+}
+
+/*
+ * For Xe introduce dummy intel_dpt_create which just return NULL and
+ * intel_dpt_destroy which does nothing.
+ */
+struct i915_address_space *intel_dpt_create(struct intel_framebuffer *fb)
+{
+ return NULL;
+}
+
+void intel_dpt_destroy(struct i915_address_space *vm)
+{
+ return;
+} \ No newline at end of file
diff --git a/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c b/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c
new file mode 100644
index 000000000000..0f11a39333e2
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_hdcp_gsc.c
@@ -0,0 +1,34 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright 2023, Intel Corporation.
+ */
+
+#include "i915_drv.h"
+#include "intel_hdcp_gsc.h"
+
+bool intel_hdcp_gsc_cs_required(struct drm_i915_private *i915)
+{
+ return true;
+}
+
+bool intel_hdcp_gsc_check_status(struct drm_i915_private *i915)
+{
+ return false;
+}
+
+int intel_hdcp_gsc_init(struct drm_i915_private *i915)
+{
+ drm_info(&i915->drm, "HDCP support not yet implemented\n");
+ return -ENODEV;
+}
+
+void intel_hdcp_gsc_fini(struct drm_i915_private *i915)
+{
+}
+
+ssize_t intel_hdcp_gsc_msg_send(struct drm_i915_private *i915, u8 *msg_in,
+ size_t msg_in_len, u8 *msg_out,
+ size_t msg_out_len)
+{
+ return -ENODEV;
+}
diff --git a/drivers/gpu/drm/xe/display/xe_plane_initial.c b/drivers/gpu/drm/xe/display/xe_plane_initial.c
new file mode 100644
index 000000000000..ccf83c12b545
--- /dev/null
+++ b/drivers/gpu/drm/xe/display/xe_plane_initial.c
@@ -0,0 +1,291 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+/* for ioread64 */
+#include <linux/io-64-nonatomic-lo-hi.h>
+
+#include "xe_ggtt.h"
+
+#include "i915_drv.h"
+#include "intel_atomic_plane.h"
+#include "intel_display.h"
+#include "intel_display_types.h"
+#include "intel_fb.h"
+#include "intel_fb_pin.h"
+#include "intel_frontbuffer.h"
+#include "intel_plane_initial.h"
+
+static bool
+intel_reuse_initial_plane_obj(struct drm_i915_private *i915,
+ const struct intel_initial_plane_config *plane_config,
+ struct drm_framebuffer **fb)
+{
+ struct intel_crtc *crtc;
+
+ for_each_intel_crtc(&i915->drm, crtc) {
+ struct intel_crtc_state *crtc_state =
+ to_intel_crtc_state(crtc->base.state);
+ struct intel_plane *plane =
+ to_intel_plane(crtc->base.primary);
+ struct intel_plane_state *plane_state =
+ to_intel_plane_state(plane->base.state);
+
+ if (!crtc_state->uapi.active)
+ continue;
+
+ if (!plane_state->ggtt_vma)
+ continue;
+
+ if (intel_plane_ggtt_offset(plane_state) == plane_config->base) {
+ *fb = plane_state->hw.fb;
+ return true;
+ }
+ }
+
+ return false;
+}
+
+static struct xe_bo *
+initial_plane_bo(struct xe_device *xe,
+ struct intel_initial_plane_config *plane_config)
+{
+ struct xe_tile *tile0 = xe_device_get_root_tile(xe);
+ struct xe_bo *bo;
+ resource_size_t phys_base;
+ u32 base, size, flags;
+ u64 page_size = xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ? SZ_64K : SZ_4K;
+
+ if (plane_config->size == 0)
+ return NULL;
+
+ flags = XE_BO_CREATE_PINNED_BIT | XE_BO_SCANOUT_BIT | XE_BO_CREATE_GGTT_BIT;
+
+ base = round_down(plane_config->base, page_size);
+ if (IS_DGFX(xe)) {
+ u64 __iomem *gte = tile0->mem.ggtt->gsm;
+ u64 pte;
+
+ gte += base / XE_PAGE_SIZE;
+
+ pte = ioread64(gte);
+ if (!(pte & XE_GGTT_PTE_DM)) {
+ drm_err(&xe->drm,
+ "Initial plane programming missing DM bit\n");
+ return NULL;
+ }
+
+ phys_base = pte & ~(page_size - 1);
+ flags |= XE_BO_CREATE_VRAM0_BIT;
+
+ /*
+ * We don't currently expect this to ever be placed in the
+ * stolen portion.
+ */
+ if (phys_base >= tile0->mem.vram.usable_size) {
+ drm_err(&xe->drm,
+ "Initial plane programming using invalid range, phys_base=%pa\n",
+ &phys_base);
+ return NULL;
+ }
+
+ drm_dbg(&xe->drm,
+ "Using phys_base=%pa, based on initial plane programming\n",
+ &phys_base);
+ } else {
+ struct ttm_resource_manager *stolen = ttm_manager_type(&xe->ttm, XE_PL_STOLEN);
+
+ if (!stolen)
+ return NULL;
+ phys_base = base;
+ flags |= XE_BO_CREATE_STOLEN_BIT;
+
+ /*
+ * If the FB is too big, just don't use it since fbdev is not very
+ * important and we should probably use that space with FBC or other
+ * features.
+ */
+ if (IS_ENABLED(CONFIG_FRAMEBUFFER_CONSOLE) &&
+ plane_config->size * 2 >> PAGE_SHIFT >= stolen->size)
+ return NULL;
+ }
+
+ size = round_up(plane_config->base + plane_config->size,
+ page_size);
+ size -= base;
+
+ bo = xe_bo_create_pin_map_at(xe, tile0, NULL, size, phys_base,
+ ttm_bo_type_kernel, flags);
+ if (IS_ERR(bo)) {
+ drm_dbg(&xe->drm,
+ "Failed to create bo phys_base=%pa size %u with flags %x: %li\n",
+ &phys_base, size, flags, PTR_ERR(bo));
+ return NULL;
+ }
+
+ return bo;
+}
+
+static bool
+intel_alloc_initial_plane_obj(struct intel_crtc *crtc,
+ struct intel_initial_plane_config *plane_config)
+{
+ struct drm_device *dev = crtc->base.dev;
+ struct drm_i915_private *dev_priv = to_i915(dev);
+ struct drm_mode_fb_cmd2 mode_cmd = { 0 };
+ struct drm_framebuffer *fb = &plane_config->fb->base;
+ struct xe_bo *bo;
+
+ switch (fb->modifier) {
+ case DRM_FORMAT_MOD_LINEAR:
+ case I915_FORMAT_MOD_X_TILED:
+ case I915_FORMAT_MOD_Y_TILED:
+ case I915_FORMAT_MOD_4_TILED:
+ break;
+ default:
+ drm_dbg(&dev_priv->drm,
+ "Unsupported modifier for initial FB: 0x%llx\n",
+ fb->modifier);
+ return false;
+ }
+
+ mode_cmd.pixel_format = fb->format->format;
+ mode_cmd.width = fb->width;
+ mode_cmd.height = fb->height;
+ mode_cmd.pitches[0] = fb->pitches[0];
+ mode_cmd.modifier[0] = fb->modifier;
+ mode_cmd.flags = DRM_MODE_FB_MODIFIERS;
+
+ bo = initial_plane_bo(dev_priv, plane_config);
+ if (!bo)
+ return false;
+
+ if (intel_framebuffer_init(to_intel_framebuffer(fb),
+ bo, &mode_cmd)) {
+ drm_dbg_kms(&dev_priv->drm, "intel fb init failed\n");
+ goto err_bo;
+ }
+ /* Reference handed over to fb */
+ xe_bo_put(bo);
+
+ return true;
+
+err_bo:
+ xe_bo_unpin_map_no_vm(bo);
+ return false;
+}
+
+static void
+intel_find_initial_plane_obj(struct intel_crtc *crtc,
+ struct intel_initial_plane_config *plane_config)
+{
+ struct drm_device *dev = crtc->base.dev;
+ struct drm_i915_private *dev_priv = to_i915(dev);
+ struct intel_plane *plane =
+ to_intel_plane(crtc->base.primary);
+ struct intel_plane_state *plane_state =
+ to_intel_plane_state(plane->base.state);
+ struct intel_crtc_state *crtc_state =
+ to_intel_crtc_state(crtc->base.state);
+ struct drm_framebuffer *fb;
+ struct i915_vma *vma;
+
+ /*
+ * TODO:
+ * Disable planes if get_initial_plane_config() failed.
+ * Make sure things work if the surface base is not page aligned.
+ */
+ if (!plane_config->fb)
+ return;
+
+ if (intel_alloc_initial_plane_obj(crtc, plane_config))
+ fb = &plane_config->fb->base;
+ else if (!intel_reuse_initial_plane_obj(dev_priv, plane_config, &fb))
+ goto nofb;
+
+ plane_state->uapi.rotation = plane_config->rotation;
+ intel_fb_fill_view(to_intel_framebuffer(fb),
+ plane_state->uapi.rotation, &plane_state->view);
+
+ vma = intel_pin_and_fence_fb_obj(fb, false, &plane_state->view.gtt,
+ false, &plane_state->flags);
+ if (IS_ERR(vma))
+ goto nofb;
+
+ plane_state->ggtt_vma = vma;
+ plane_state->uapi.src_x = 0;
+ plane_state->uapi.src_y = 0;
+ plane_state->uapi.src_w = fb->width << 16;
+ plane_state->uapi.src_h = fb->height << 16;
+
+ plane_state->uapi.crtc_x = 0;
+ plane_state->uapi.crtc_y = 0;
+ plane_state->uapi.crtc_w = fb->width;
+ plane_state->uapi.crtc_h = fb->height;
+
+ plane_state->uapi.fb = fb;
+ drm_framebuffer_get(fb);
+
+ plane_state->uapi.crtc = &crtc->base;
+ intel_plane_copy_uapi_to_hw_state(plane_state, plane_state, crtc);
+
+ atomic_or(plane->frontbuffer_bit, &to_intel_frontbuffer(fb)->bits);
+
+ plane_config->vma = vma;
+
+ /*
+ * Flip to the newly created mapping ASAP, so we can re-use the
+ * first part of GGTT for WOPCM, prevent flickering, and prevent
+ * the lookup of sysmem scratch pages.
+ */
+ plane->check_plane(crtc_state, plane_state);
+ plane->async_flip(plane, crtc_state, plane_state, true);
+ return;
+
+nofb:
+ /*
+ * We've failed to reconstruct the BIOS FB. Current display state
+ * indicates that the primary plane is visible, but has a NULL FB,
+ * which will lead to problems later if we don't fix it up. The
+ * simplest solution is to just disable the primary plane now and
+ * pretend the BIOS never had it enabled.
+ */
+ intel_plane_disable_noatomic(crtc, plane);
+}
+
+static void plane_config_fini(struct intel_initial_plane_config *plane_config)
+{
+ if (plane_config->fb) {
+ struct drm_framebuffer *fb = &plane_config->fb->base;
+
+ /* We may only have the stub and not a full framebuffer */
+ if (drm_framebuffer_read_refcount(fb))
+ drm_framebuffer_put(fb);
+ else
+ kfree(fb);
+ }
+}
+
+void intel_crtc_initial_plane_config(struct intel_crtc *crtc)
+{
+ struct xe_device *xe = to_xe_device(crtc->base.dev);
+ struct intel_initial_plane_config plane_config = {};
+
+ /*
+ * Note that reserving the BIOS fb up front prevents us
+ * from stuffing other stolen allocations like the ring
+ * on top. This prevents some ugliness at boot time, and
+ * can even allow for smooth boot transitions if the BIOS
+ * fb is large enough for the active pipe configuration.
+ */
+ xe->display.funcs.display->get_initial_plane_config(crtc, &plane_config);
+
+ /*
+ * If the fb is shared between multiple heads, we'll
+ * just get the first one.
+ */
+ intel_find_initial_plane_obj(crtc, &plane_config);
+
+ plane_config_fini(&plane_config);
+}
diff --git a/drivers/gpu/drm/xe/instructions/xe_gfxpipe_commands.h b/drivers/gpu/drm/xe/instructions/xe_gfxpipe_commands.h
new file mode 100644
index 000000000000..8e6dd061f2ae
--- /dev/null
+++ b/drivers/gpu/drm/xe/instructions/xe_gfxpipe_commands.h
@@ -0,0 +1,160 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GFXPIPE_COMMANDS_H_
+#define _XE_GFXPIPE_COMMANDS_H_
+
+#include "instructions/xe_instr_defs.h"
+
+#define GFXPIPE_PIPELINE REG_GENMASK(28, 27)
+#define PIPELINE_COMMON REG_FIELD_PREP(GFXPIPE_PIPELINE, 0x0)
+#define PIPELINE_SINGLE_DW REG_FIELD_PREP(GFXPIPE_PIPELINE, 0x1)
+#define PIPELINE_COMPUTE REG_FIELD_PREP(GFXPIPE_PIPELINE, 0x2)
+#define PIPELINE_3D REG_FIELD_PREP(GFXPIPE_PIPELINE, 0x3)
+
+#define GFXPIPE_OPCODE REG_GENMASK(26, 24)
+#define GFXPIPE_SUBOPCODE REG_GENMASK(23, 16)
+
+#define GFXPIPE_MATCH_MASK (XE_INSTR_CMD_TYPE | \
+ GFXPIPE_PIPELINE | \
+ GFXPIPE_OPCODE | \
+ GFXPIPE_SUBOPCODE)
+
+#define GFXPIPE_COMMON_CMD(opcode, subopcode) \
+ (XE_INSTR_GFXPIPE | PIPELINE_COMMON | \
+ REG_FIELD_PREP(GFXPIPE_OPCODE, opcode) | \
+ REG_FIELD_PREP(GFXPIPE_SUBOPCODE, subopcode))
+
+#define GFXPIPE_SINGLE_DW_CMD(opcode, subopcode) \
+ (XE_INSTR_GFXPIPE | PIPELINE_SINGLE_DW | \
+ REG_FIELD_PREP(GFXPIPE_OPCODE, opcode) | \
+ REG_FIELD_PREP(GFXPIPE_SUBOPCODE, subopcode))
+
+#define GFXPIPE_3D_CMD(opcode, subopcode) \
+ (XE_INSTR_GFXPIPE | PIPELINE_3D | \
+ REG_FIELD_PREP(GFXPIPE_OPCODE, opcode) | \
+ REG_FIELD_PREP(GFXPIPE_SUBOPCODE, subopcode))
+
+#define GFXPIPE_COMPUTE_CMD(opcode, subopcode) \
+ (XE_INSTR_GFXPIPE | PIPELINE_COMPUTE | \
+ REG_FIELD_PREP(GFXPIPE_OPCODE, opcode) | \
+ REG_FIELD_PREP(GFXPIPE_SUBOPCODE, subopcode))
+
+#define STATE_BASE_ADDRESS GFXPIPE_COMMON_CMD(0x1, 0x1)
+#define STATE_SIP GFXPIPE_COMMON_CMD(0x1, 0x2)
+#define GPGPU_CSR_BASE_ADDRESS GFXPIPE_COMMON_CMD(0x1, 0x4)
+#define STATE_COMPUTE_MODE GFXPIPE_COMMON_CMD(0x1, 0x5)
+#define CMD_3DSTATE_BTD GFXPIPE_COMMON_CMD(0x1, 0x6)
+
+#define CMD_3DSTATE_VF_STATISTICS GFXPIPE_SINGLE_DW_CMD(0x0, 0xB)
+
+#define PIPELINE_SELECT GFXPIPE_SINGLE_DW_CMD(0x1, 0x4)
+
+#define CMD_3DSTATE_DRAWING_RECTANGLE_FAST GFXPIPE_3D_CMD(0x0, 0x0)
+#define CMD_3DSTATE_CLEAR_PARAMS GFXPIPE_3D_CMD(0x0, 0x4)
+#define CMD_3DSTATE_DEPTH_BUFFER GFXPIPE_3D_CMD(0x0, 0x5)
+#define CMD_3DSTATE_STENCIL_BUFFER GFXPIPE_3D_CMD(0x0, 0x6)
+#define CMD_3DSTATE_HIER_DEPTH_BUFFER GFXPIPE_3D_CMD(0x0, 0x7)
+#define CMD_3DSTATE_VERTEX_BUFFERS GFXPIPE_3D_CMD(0x0, 0x8)
+#define CMD_3DSTATE_VERTEX_ELEMENTS GFXPIPE_3D_CMD(0x0, 0x9)
+#define CMD_3DSTATE_INDEX_BUFFER GFXPIPE_3D_CMD(0x0, 0xA)
+#define CMD_3DSTATE_VF GFXPIPE_3D_CMD(0x0, 0xC)
+#define CMD_3DSTATE_MULTISAMPLE GFXPIPE_3D_CMD(0x0, 0xD)
+#define CMD_3DSTATE_CC_STATE_POINTERS GFXPIPE_3D_CMD(0x0, 0xE)
+#define CMD_3DSTATE_SCISSOR_STATE_POINTERS GFXPIPE_3D_CMD(0x0, 0xF)
+#define CMD_3DSTATE_VS GFXPIPE_3D_CMD(0x0, 0x10)
+#define CMD_3DSTATE_GS GFXPIPE_3D_CMD(0x0, 0x11)
+#define CMD_3DSTATE_CLIP GFXPIPE_3D_CMD(0x0, 0x12)
+#define CMD_3DSTATE_SF GFXPIPE_3D_CMD(0x0, 0x13)
+#define CMD_3DSTATE_WM GFXPIPE_3D_CMD(0x0, 0x14)
+#define CMD_3DSTATE_CONSTANT_VS GFXPIPE_3D_CMD(0x0, 0x15)
+#define CMD_3DSTATE_CONSTANT_GS GFXPIPE_3D_CMD(0x0, 0x16)
+#define CMD_3DSTATE_SAMPLE_MASK GFXPIPE_3D_CMD(0x0, 0x18)
+#define CMD_3DSTATE_CONSTANT_HS GFXPIPE_3D_CMD(0x0, 0x19)
+#define CMD_3DSTATE_CONSTANT_DS GFXPIPE_3D_CMD(0x0, 0x1A)
+#define CMD_3DSTATE_HS GFXPIPE_3D_CMD(0x0, 0x1B)
+#define CMD_3DSTATE_TE GFXPIPE_3D_CMD(0x0, 0x1C)
+#define CMD_3DSTATE_DS GFXPIPE_3D_CMD(0x0, 0x1D)
+#define CMD_3DSTATE_STREAMOUT GFXPIPE_3D_CMD(0x0, 0x1E)
+#define CMD_3DSTATE_SBE GFXPIPE_3D_CMD(0x0, 0x1F)
+#define CMD_3DSTATE_PS GFXPIPE_3D_CMD(0x0, 0x20)
+#define CMD_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP GFXPIPE_3D_CMD(0x0, 0x21)
+#define CMD_3DSTATE_CPS_POINTERS GFXPIPE_3D_CMD(0x0, 0x22)
+#define CMD_3DSTATE_VIEWPORT_STATE_POINTERS_CC GFXPIPE_3D_CMD(0x0, 0x23)
+#define CMD_3DSTATE_BLEND_STATE_POINTERS GFXPIPE_3D_CMD(0x0, 0x24)
+#define CMD_3DSTATE_BINDING_TABLE_POINTERS_VS GFXPIPE_3D_CMD(0x0, 0x26)
+#define CMD_3DSTATE_BINDING_TABLE_POINTERS_HS GFXPIPE_3D_CMD(0x0, 0x27)
+#define CMD_3DSTATE_BINDING_TABLE_POINTERS_DS GFXPIPE_3D_CMD(0x0, 0x28)
+#define CMD_3DSTATE_BINDING_TABLE_POINTERS_GS GFXPIPE_3D_CMD(0x0, 0x29)
+#define CMD_3DSTATE_BINDING_TABLE_POINTERS_PS GFXPIPE_3D_CMD(0x0, 0x2A)
+#define CMD_3DSTATE_SAMPLER_STATE_POINTERS_VS GFXPIPE_3D_CMD(0x0, 0x2B)
+#define CMD_3DSTATE_SAMPLER_STATE_POINTERS_HS GFXPIPE_3D_CMD(0x0, 0x2C)
+#define CMD_3DSTATE_SAMPLER_STATE_POINTERS_DS GFXPIPE_3D_CMD(0x0, 0x2D)
+#define CMD_3DSTATE_SAMPLER_STATE_POINTERS_GS GFXPIPE_3D_CMD(0x0, 0x2E)
+#define CMD_3DSTATE_SAMPLER_STATE_POINTERS_PS GFXPIPE_3D_CMD(0x0, 0x2F)
+#define CMD_3DSTATE_VF_INSTANCING GFXPIPE_3D_CMD(0x0, 0x49)
+#define CMD_3DSTATE_VF_SGVS GFXPIPE_3D_CMD(0x0, 0x4A)
+#define CMD_3DSTATE_VF_TOPOLOGY GFXPIPE_3D_CMD(0x0, 0x4B)
+#define CMD_3DSTATE_WM_CHROMAKEY GFXPIPE_3D_CMD(0x0, 0x4C)
+#define CMD_3DSTATE_PS_BLEND GFXPIPE_3D_CMD(0x0, 0x4D)
+#define CMD_3DSTATE_WM_DEPTH_STENCIL GFXPIPE_3D_CMD(0x0, 0x4E)
+#define CMD_3DSTATE_PS_EXTRA GFXPIPE_3D_CMD(0x0, 0x4F)
+#define CMD_3DSTATE_RASTER GFXPIPE_3D_CMD(0x0, 0x50)
+#define CMD_3DSTATE_SBE_SWIZ GFXPIPE_3D_CMD(0x0, 0x51)
+#define CMD_3DSTATE_WM_HZ_OP GFXPIPE_3D_CMD(0x0, 0x52)
+#define CMD_3DSTATE_VF_COMPONENT_PACKING GFXPIPE_3D_CMD(0x0, 0x55)
+#define CMD_3DSTATE_VF_SGVS_2 GFXPIPE_3D_CMD(0x0, 0x56)
+#define CMD_3DSTATE_VFG GFXPIPE_3D_CMD(0x0, 0x57)
+#define CMD_3DSTATE_URB_ALLOC_VS GFXPIPE_3D_CMD(0x0, 0x58)
+#define CMD_3DSTATE_URB_ALLOC_HS GFXPIPE_3D_CMD(0x0, 0x59)
+#define CMD_3DSTATE_URB_ALLOC_DS GFXPIPE_3D_CMD(0x0, 0x5A)
+#define CMD_3DSTATE_URB_ALLOC_GS GFXPIPE_3D_CMD(0x0, 0x5B)
+#define CMD_3DSTATE_SO_BUFFER_INDEX_0 GFXPIPE_3D_CMD(0x0, 0x60)
+#define CMD_3DSTATE_SO_BUFFER_INDEX_1 GFXPIPE_3D_CMD(0x0, 0x61)
+#define CMD_3DSTATE_SO_BUFFER_INDEX_2 GFXPIPE_3D_CMD(0x0, 0x62)
+#define CMD_3DSTATE_SO_BUFFER_INDEX_3 GFXPIPE_3D_CMD(0x0, 0x63)
+#define CMD_3DSTATE_PRIMITIVE_REPLICATION GFXPIPE_3D_CMD(0x0, 0x6C)
+#define CMD_3DSTATE_TBIMR_TILE_PASS_INFO GFXPIPE_3D_CMD(0x0, 0x6E)
+#define CMD_3DSTATE_AMFS GFXPIPE_3D_CMD(0x0, 0x6F)
+#define CMD_3DSTATE_DEPTH_BOUNDS GFXPIPE_3D_CMD(0x0, 0x71)
+#define CMD_3DSTATE_AMFS_TEXTURE_POINTERS GFXPIPE_3D_CMD(0x0, 0x72)
+#define CMD_3DSTATE_CONSTANT_TS_POINTER GFXPIPE_3D_CMD(0x0, 0x73)
+#define CMD_3DSTATE_MESH_CONTROL GFXPIPE_3D_CMD(0x0, 0x77)
+#define CMD_3DSTATE_MESH_DISTRIB GFXPIPE_3D_CMD(0x0, 0x78)
+#define CMD_3DSTATE_TASK_REDISTRIB GFXPIPE_3D_CMD(0x0, 0x79)
+#define CMD_3DSTATE_MESH_SHADER GFXPIPE_3D_CMD(0x0, 0x7A)
+#define CMD_3DSTATE_MESH_SHADER_DATA GFXPIPE_3D_CMD(0x0, 0x7B)
+#define CMD_3DSTATE_TASK_CONTROL GFXPIPE_3D_CMD(0x0, 0x7C)
+#define CMD_3DSTATE_TASK_SHADER GFXPIPE_3D_CMD(0x0, 0x7D)
+#define CMD_3DSTATE_TASK_SHADER_DATA GFXPIPE_3D_CMD(0x0, 0x7E)
+#define CMD_3DSTATE_URB_ALLOC_MESH GFXPIPE_3D_CMD(0x0, 0x7F)
+#define CMD_3DSTATE_URB_ALLOC_TASK GFXPIPE_3D_CMD(0x0, 0x80)
+#define CMD_3DSTATE_CLIP_MESH GFXPIPE_3D_CMD(0x0, 0x81)
+#define CMD_3DSTATE_SBE_MESH GFXPIPE_3D_CMD(0x0, 0x82)
+#define CMD_3DSTATE_CPSIZE_CONTROL_BUFFER GFXPIPE_3D_CMD(0x0, 0x83)
+
+#define CMD_3DSTATE_DRAWING_RECTANGLE GFXPIPE_3D_CMD(0x1, 0x0)
+#define CMD_3DSTATE_CHROMA_KEY GFXPIPE_3D_CMD(0x1, 0x4)
+#define CMD_3DSTATE_POLY_STIPPLE_OFFSET GFXPIPE_3D_CMD(0x1, 0x6)
+#define CMD_3DSTATE_POLY_STIPPLE_PATTERN GFXPIPE_3D_CMD(0x1, 0x7)
+#define CMD_3DSTATE_LINE_STIPPLE GFXPIPE_3D_CMD(0x1, 0x8)
+#define CMD_3DSTATE_AA_LINE_PARAMETERS GFXPIPE_3D_CMD(0x1, 0xA)
+#define CMD_3DSTATE_MONOFILTER_SIZE GFXPIPE_3D_CMD(0x1, 0x11)
+#define CMD_3DSTATE_PUSH_CONSTANT_ALLOC_VS GFXPIPE_3D_CMD(0x1, 0x12)
+#define CMD_3DSTATE_PUSH_CONSTANT_ALLOC_HS GFXPIPE_3D_CMD(0x1, 0x13)
+#define CMD_3DSTATE_PUSH_CONSTANT_ALLOC_DS GFXPIPE_3D_CMD(0x1, 0x14)
+#define CMD_3DSTATE_PUSH_CONSTANT_ALLOC_GS GFXPIPE_3D_CMD(0x1, 0x15)
+#define CMD_3DSTATE_PUSH_CONSTANT_ALLOC_PS GFXPIPE_3D_CMD(0x1, 0x16)
+#define CMD_3DSTATE_SO_DECL_LIST GFXPIPE_3D_CMD(0x1, 0x17)
+#define CMD_3DSTATE_SO_DECL_LIST_DW_LEN REG_GENMASK(8, 0)
+#define CMD_3DSTATE_SO_BUFFER GFXPIPE_3D_CMD(0x1, 0x18)
+#define CMD_3DSTATE_BINDING_TABLE_POOL_ALLOC GFXPIPE_3D_CMD(0x1, 0x19)
+#define CMD_3DSTATE_SAMPLE_PATTERN GFXPIPE_3D_CMD(0x1, 0x1C)
+#define CMD_3DSTATE_3D_MODE GFXPIPE_3D_CMD(0x1, 0x1E)
+#define CMD_3DSTATE_SUBSLICE_HASH_TABLE GFXPIPE_3D_CMD(0x1, 0x1F)
+#define CMD_3DSTATE_SLICE_TABLE_STATE_POINTERS GFXPIPE_3D_CMD(0x1, 0x20)
+#define CMD_3DSTATE_PTBR_TILE_PASS_INFO GFXPIPE_3D_CMD(0x1, 0x22)
+
+#endif
diff --git a/drivers/gpu/drm/xe/instructions/xe_gsc_commands.h b/drivers/gpu/drm/xe/instructions/xe_gsc_commands.h
new file mode 100644
index 000000000000..f8949cad9d0f
--- /dev/null
+++ b/drivers/gpu/drm/xe/instructions/xe_gsc_commands.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GSC_COMMANDS_H_
+#define _XE_GSC_COMMANDS_H_
+
+#include "instructions/xe_instr_defs.h"
+
+/*
+ * All GSCCS-specific commands have fixed length, so we can include it in the
+ * defines. Note that the generic GSC command header structure includes an
+ * optional data field in bits 9-21, but there are no commands that actually use
+ * it; some of the commands are instead defined as having an extended length
+ * field spanning bits 0-15, even if the extra bits are not required because the
+ * longest GSCCS command is only 8 dwords. To handle this, the defines below use
+ * a single field for both data and len. If we ever get a commands that does
+ * actually have data and this approach doesn't work for it we can re-work it
+ * at that point.
+ */
+
+#define GSC_OPCODE REG_GENMASK(28, 22)
+#define GSC_CMD_DATA_AND_LEN REG_GENMASK(21, 0)
+
+#define __GSC_INSTR(op, dl) \
+ (XE_INSTR_GSC | \
+ REG_FIELD_PREP(GSC_OPCODE, op) | \
+ REG_FIELD_PREP(GSC_CMD_DATA_AND_LEN, dl))
+
+#define GSC_HECI_CMD_PKT __GSC_INSTR(0, 6)
+
+#define GSC_FW_LOAD __GSC_INSTR(1, 2)
+#define GSC_FW_LOAD_LIMIT_VALID REG_BIT(31)
+
+#endif
diff --git a/drivers/gpu/drm/xe/instructions/xe_instr_defs.h b/drivers/gpu/drm/xe/instructions/xe_instr_defs.h
new file mode 100644
index 000000000000..04179b2a48e1
--- /dev/null
+++ b/drivers/gpu/drm/xe/instructions/xe_instr_defs.h
@@ -0,0 +1,33 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_INSTR_DEFS_H_
+#define _XE_INSTR_DEFS_H_
+
+#include "regs/xe_reg_defs.h"
+
+/*
+ * The first dword of any GPU instruction is the "instruction header." Bits
+ * 31:29 identify the general type of the command and determine how exact
+ * opcodes and sub-opcodes will be encoded in the remaining bits.
+ */
+#define XE_INSTR_CMD_TYPE GENMASK(31, 29)
+#define XE_INSTR_MI REG_FIELD_PREP(XE_INSTR_CMD_TYPE, 0x0)
+#define XE_INSTR_GSC REG_FIELD_PREP(XE_INSTR_CMD_TYPE, 0x2)
+#define XE_INSTR_GFXPIPE REG_FIELD_PREP(XE_INSTR_CMD_TYPE, 0x3)
+
+/*
+ * Most (but not all) instructions have a "length" field in the instruction
+ * header. The value expected is the total number of dwords for the
+ * instruction, minus two.
+ *
+ * Some instructions have length fields longer or shorter than 8 bits, but
+ * those are rare. This definition can be used for the common case where
+ * the length field is from 7:0.
+ */
+#define XE_INSTR_LEN_MASK GENMASK(7, 0)
+#define XE_INSTR_NUM_DW(x) REG_FIELD_PREP(XE_INSTR_LEN_MASK, (x) - 2)
+
+#endif
diff --git a/drivers/gpu/drm/xe/instructions/xe_mi_commands.h b/drivers/gpu/drm/xe/instructions/xe_mi_commands.h
new file mode 100644
index 000000000000..1cfa96167fde
--- /dev/null
+++ b/drivers/gpu/drm/xe/instructions/xe_mi_commands.h
@@ -0,0 +1,61 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_MI_COMMANDS_H_
+#define _XE_MI_COMMANDS_H_
+
+#include "instructions/xe_instr_defs.h"
+
+/*
+ * MI (Memory Interface) commands are supported by all GT engines. They
+ * provide general memory operations and command streamer control. MI commands
+ * have a command type of 0x0 (MI_COMMAND) in bits 31:29 of the instruction
+ * header dword and a specific MI opcode in bits 28:23.
+ */
+
+#define MI_OPCODE REG_GENMASK(28, 23)
+#define MI_SUBOPCODE REG_GENMASK(22, 17) /* used with MI_EXPANSION */
+
+#define __MI_INSTR(opcode) \
+ (XE_INSTR_MI | REG_FIELD_PREP(MI_OPCODE, opcode))
+
+#define MI_NOOP __MI_INSTR(0x0)
+#define MI_USER_INTERRUPT __MI_INSTR(0x2)
+#define MI_ARB_CHECK __MI_INSTR(0x5)
+
+#define MI_ARB_ON_OFF __MI_INSTR(0x8)
+#define MI_ARB_ENABLE REG_BIT(0)
+#define MI_ARB_DISABLE 0x0
+
+#define MI_BATCH_BUFFER_END __MI_INSTR(0xA)
+#define MI_TOPOLOGY_FILTER __MI_INSTR(0xD)
+#define MI_FORCE_WAKEUP __MI_INSTR(0x1D)
+
+#define MI_STORE_DATA_IMM __MI_INSTR(0x20)
+#define MI_SDI_GGTT REG_BIT(22)
+#define MI_SDI_LEN_DW GENMASK(9, 0)
+#define MI_SDI_NUM_DW(x) REG_FIELD_PREP(MI_SDI_LEN_DW, (x) + 3 - 2)
+#define MI_SDI_NUM_QW(x) (REG_FIELD_PREP(MI_SDI_LEN_DW, 2 * (x) + 3 - 2) | \
+ REG_BIT(21))
+
+#define MI_LOAD_REGISTER_IMM __MI_INSTR(0x22)
+#define MI_LRI_LRM_CS_MMIO REG_BIT(19)
+#define MI_LRI_MMIO_REMAP_EN REG_BIT(17)
+#define MI_LRI_NUM_REGS(x) XE_INSTR_NUM_DW(2 * (x) + 1)
+#define MI_LRI_FORCE_POSTED REG_BIT(12)
+
+#define MI_FLUSH_DW __MI_INSTR(0x26)
+#define MI_FLUSH_DW_STORE_INDEX REG_BIT(21)
+#define MI_INVALIDATE_TLB REG_BIT(18)
+#define MI_FLUSH_DW_CCS REG_BIT(16)
+#define MI_FLUSH_DW_OP_STOREDW REG_BIT(14)
+#define MI_FLUSH_DW_LEN_DW REG_GENMASK(5, 0)
+#define MI_FLUSH_IMM_DW REG_FIELD_PREP(MI_FLUSH_DW_LEN_DW, 4 - 2)
+#define MI_FLUSH_IMM_QW REG_FIELD_PREP(MI_FLUSH_DW_LEN_DW, 5 - 2)
+#define MI_FLUSH_DW_USE_GTT REG_BIT(2)
+
+#define MI_BATCH_BUFFER_START __MI_INSTR(0x31)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_engine_regs.h b/drivers/gpu/drm/xe/regs/xe_engine_regs.h
new file mode 100644
index 000000000000..5592774fc690
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_engine_regs.h
@@ -0,0 +1,184 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_ENGINE_REGS_H_
+#define _XE_ENGINE_REGS_H_
+
+#include <asm/page.h>
+
+#include "regs/xe_reg_defs.h"
+
+/*
+ * These *_BASE values represent the MMIO offset where each hardware engine's
+ * registers start. The other definitions in this header are parameterized
+ * macros that will take one of these values as a parameter.
+ */
+#define RENDER_RING_BASE 0x02000
+#define BSD_RING_BASE 0x1c0000
+#define BSD2_RING_BASE 0x1c4000
+#define BSD3_RING_BASE 0x1d0000
+#define BSD4_RING_BASE 0x1d4000
+#define XEHP_BSD5_RING_BASE 0x1e0000
+#define XEHP_BSD6_RING_BASE 0x1e4000
+#define XEHP_BSD7_RING_BASE 0x1f0000
+#define XEHP_BSD8_RING_BASE 0x1f4000
+#define VEBOX_RING_BASE 0x1c8000
+#define VEBOX2_RING_BASE 0x1d8000
+#define XEHP_VEBOX3_RING_BASE 0x1e8000
+#define XEHP_VEBOX4_RING_BASE 0x1f8000
+#define COMPUTE0_RING_BASE 0x1a000
+#define COMPUTE1_RING_BASE 0x1c000
+#define COMPUTE2_RING_BASE 0x1e000
+#define COMPUTE3_RING_BASE 0x26000
+#define BLT_RING_BASE 0x22000
+#define XEHPC_BCS1_RING_BASE 0x3e0000
+#define XEHPC_BCS2_RING_BASE 0x3e2000
+#define XEHPC_BCS3_RING_BASE 0x3e4000
+#define XEHPC_BCS4_RING_BASE 0x3e6000
+#define XEHPC_BCS5_RING_BASE 0x3e8000
+#define XEHPC_BCS6_RING_BASE 0x3ea000
+#define XEHPC_BCS7_RING_BASE 0x3ec000
+#define XEHPC_BCS8_RING_BASE 0x3ee000
+#define GSCCS_RING_BASE 0x11a000
+
+#define RING_TAIL(base) XE_REG((base) + 0x30)
+
+#define RING_HEAD(base) XE_REG((base) + 0x34)
+#define HEAD_ADDR 0x001FFFFC
+
+#define RING_START(base) XE_REG((base) + 0x38)
+
+#define RING_CTL(base) XE_REG((base) + 0x3c)
+#define RING_CTL_SIZE(size) ((size) - PAGE_SIZE) /* in bytes -> pages */
+#define RING_CTL_SIZE(size) ((size) - PAGE_SIZE) /* in bytes -> pages */
+
+#define RING_PSMI_CTL(base) XE_REG((base) + 0x50, XE_REG_OPTION_MASKED)
+#define RC_SEMA_IDLE_MSG_DISABLE REG_BIT(12)
+#define WAIT_FOR_EVENT_POWER_DOWN_DISABLE REG_BIT(7)
+#define IDLE_MSG_DISABLE REG_BIT(0)
+
+#define RING_PWRCTX_MAXCNT(base) XE_REG((base) + 0x54)
+#define IDLE_WAIT_TIME REG_GENMASK(19, 0)
+
+#define RING_ACTHD_UDW(base) XE_REG((base) + 0x5c)
+#define RING_DMA_FADD_UDW(base) XE_REG((base) + 0x60)
+#define RING_IPEHR(base) XE_REG((base) + 0x68)
+#define RING_ACTHD(base) XE_REG((base) + 0x74)
+#define RING_DMA_FADD(base) XE_REG((base) + 0x78)
+#define RING_HWS_PGA(base) XE_REG((base) + 0x80)
+#define RING_HWSTAM(base) XE_REG((base) + 0x98)
+#define RING_MI_MODE(base) XE_REG((base) + 0x9c)
+#define RING_NOPID(base) XE_REG((base) + 0x94)
+
+#define FF_THREAD_MODE(base) XE_REG((base) + 0xa0)
+#define FF_TESSELATION_DOP_GATE_DISABLE BIT(19)
+
+#define RING_IMR(base) XE_REG((base) + 0xa8)
+
+#define RING_EIR(base) XE_REG((base) + 0xb0)
+#define RING_EMR(base) XE_REG((base) + 0xb4)
+#define RING_ESR(base) XE_REG((base) + 0xb8)
+
+#define RING_CMD_CCTL(base) XE_REG((base) + 0xc4, XE_REG_OPTION_MASKED)
+/*
+ * CMD_CCTL read/write fields take a MOCS value and _not_ a table index.
+ * The lsb of each can be considered a separate enabling bit for encryption.
+ * 6:0 == default MOCS value for reads => 6:1 == table index for reads.
+ * 13:7 == default MOCS value for writes => 13:8 == table index for writes.
+ * 15:14 == Reserved => 31:30 are set to 0.
+ */
+#define CMD_CCTL_WRITE_OVERRIDE_MASK REG_GENMASK(13, 8)
+#define CMD_CCTL_READ_OVERRIDE_MASK REG_GENMASK(6, 1)
+
+#define CSFE_CHICKEN1(base) XE_REG((base) + 0xd4, XE_REG_OPTION_MASKED)
+#define GHWSP_CSB_REPORT_DIS REG_BIT(15)
+#define PPHWSP_CSB_AND_TIMESTAMP_REPORT_DIS REG_BIT(14)
+
+#define FF_SLICE_CS_CHICKEN1(base) XE_REG((base) + 0xe0, XE_REG_OPTION_MASKED)
+#define FFSC_PERCTX_PREEMPT_CTRL REG_BIT(14)
+
+#define FF_SLICE_CS_CHICKEN2(base) XE_REG((base) + 0xe4, XE_REG_OPTION_MASKED)
+#define PERF_FIX_BALANCING_CFE_DISABLE REG_BIT(15)
+
+#define CS_DEBUG_MODE1(base) XE_REG((base) + 0xec, XE_REG_OPTION_MASKED)
+#define FF_DOP_CLOCK_GATE_DISABLE REG_BIT(1)
+#define REPLAY_MODE_GRANULARITY REG_BIT(0)
+
+#define RING_BBADDR(base) XE_REG((base) + 0x140)
+#define RING_BBADDR_UDW(base) XE_REG((base) + 0x168)
+
+#define BCS_SWCTRL(base) XE_REG((base) + 0x200, XE_REG_OPTION_MASKED)
+#define BCS_SWCTRL_DISABLE_256B REG_BIT(2)
+
+/* Handling MOCS value in BLIT_CCTL like it was done CMD_CCTL */
+#define BLIT_CCTL(base) XE_REG((base) + 0x204)
+#define BLIT_CCTL_DST_MOCS_MASK REG_GENMASK(14, 9)
+#define BLIT_CCTL_SRC_MOCS_MASK REG_GENMASK(6, 1)
+
+#define RING_EXECLIST_STATUS_LO(base) XE_REG((base) + 0x234)
+#define RING_EXECLIST_STATUS_HI(base) XE_REG((base) + 0x234 + 4)
+
+#define RING_CONTEXT_CONTROL(base) XE_REG((base) + 0x244)
+#define CTX_CTRL_INHIBIT_SYN_CTX_SWITCH REG_BIT(3)
+#define CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT REG_BIT(0)
+
+#define RING_MODE(base) XE_REG((base) + 0x29c)
+#define GFX_DISABLE_LEGACY_MODE REG_BIT(3)
+
+#define RING_TIMESTAMP(base) XE_REG((base) + 0x358)
+
+#define RING_TIMESTAMP_UDW(base) XE_REG((base) + 0x358 + 4)
+#define RING_VALID_MASK 0x00000001
+#define RING_VALID 0x00000001
+#define STOP_RING REG_BIT(8)
+#define TAIL_ADDR 0x001FFFF8
+
+#define RING_CTX_TIMESTAMP(base) XE_REG((base) + 0x3a8)
+
+#define RING_FORCE_TO_NONPRIV(base, i) XE_REG(((base) + 0x4d0) + (i) * 4)
+#define RING_FORCE_TO_NONPRIV_DENY REG_BIT(30)
+#define RING_FORCE_TO_NONPRIV_ACCESS_MASK REG_GENMASK(29, 28)
+#define RING_FORCE_TO_NONPRIV_ACCESS_RW REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_ACCESS_MASK, 0)
+#define RING_FORCE_TO_NONPRIV_ACCESS_RD REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_ACCESS_MASK, 1)
+#define RING_FORCE_TO_NONPRIV_ACCESS_WR REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_ACCESS_MASK, 2)
+#define RING_FORCE_TO_NONPRIV_ACCESS_INVALID REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_ACCESS_MASK, 3)
+#define RING_FORCE_TO_NONPRIV_ADDRESS_MASK REG_GENMASK(25, 2)
+#define RING_FORCE_TO_NONPRIV_RANGE_MASK REG_GENMASK(1, 0)
+#define RING_FORCE_TO_NONPRIV_RANGE_1 REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_RANGE_MASK, 0)
+#define RING_FORCE_TO_NONPRIV_RANGE_4 REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_RANGE_MASK, 1)
+#define RING_FORCE_TO_NONPRIV_RANGE_16 REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_RANGE_MASK, 2)
+#define RING_FORCE_TO_NONPRIV_RANGE_64 REG_FIELD_PREP(RING_FORCE_TO_NONPRIV_RANGE_MASK, 3)
+#define RING_FORCE_TO_NONPRIV_MASK_VALID (RING_FORCE_TO_NONPRIV_RANGE_MASK | \
+ RING_FORCE_TO_NONPRIV_ACCESS_MASK | \
+ RING_FORCE_TO_NONPRIV_DENY)
+#define RING_MAX_NONPRIV_SLOTS 12
+
+#define RING_EXECLIST_SQ_CONTENTS_LO(base) XE_REG((base) + 0x510)
+#define RING_EXECLIST_SQ_CONTENTS_HI(base) XE_REG((base) + 0x510 + 4)
+
+#define RING_EXECLIST_CONTROL(base) XE_REG((base) + 0x550)
+#define EL_CTRL_LOAD REG_BIT(0)
+
+#define CS_CHICKEN1(base) XE_REG((base) + 0x580, XE_REG_OPTION_MASKED)
+#define PREEMPT_GPGPU_LEVEL(hi, lo) (((hi) << 2) | ((lo) << 1))
+#define PREEMPT_GPGPU_MID_THREAD_LEVEL PREEMPT_GPGPU_LEVEL(0, 0)
+#define PREEMPT_GPGPU_THREAD_GROUP_LEVEL PREEMPT_GPGPU_LEVEL(0, 1)
+#define PREEMPT_GPGPU_COMMAND_LEVEL PREEMPT_GPGPU_LEVEL(1, 0)
+#define PREEMPT_GPGPU_LEVEL_MASK PREEMPT_GPGPU_LEVEL(1, 1)
+#define PREEMPT_3D_OBJECT_LEVEL REG_BIT(0)
+
+#define VDBOX_CGCTL3F08(base) XE_REG((base) + 0x3f08)
+#define CG3DDISHRS_CLKGATE_DIS REG_BIT(5)
+
+#define VDBOX_CGCTL3F10(base) XE_REG((base) + 0x3f10)
+#define IECPUNIT_CLKGATE_DIS REG_BIT(22)
+
+#define VDBOX_CGCTL3F18(base) XE_REG((base) + 0x3f18)
+#define ALNUNIT_CLKGATE_DIS REG_BIT(13)
+
+#define VDBOX_CGCTL3F1C(base) XE_REG((base) + 0x3f1c)
+#define MFXPIPE_CLKGATE_DIS REG_BIT(3)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_gpu_commands.h b/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
new file mode 100644
index 000000000000..a255946b6f77
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_gpu_commands.h
@@ -0,0 +1,70 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GPU_COMMANDS_H_
+#define _XE_GPU_COMMANDS_H_
+
+#include "regs/xe_reg_defs.h"
+
+#define XY_CTRL_SURF_COPY_BLT ((2 << 29) | (0x48 << 22) | 3)
+#define SRC_ACCESS_TYPE_SHIFT 21
+#define DST_ACCESS_TYPE_SHIFT 20
+#define CCS_SIZE_MASK GENMASK(17, 8)
+#define XE2_CCS_SIZE_MASK GENMASK(18, 9)
+#define XY_CTRL_SURF_MOCS_MASK GENMASK(31, 26)
+#define XE2_XY_CTRL_SURF_MOCS_INDEX_MASK GENMASK(31, 28)
+#define NUM_CCS_BYTES_PER_BLOCK 256
+#define NUM_BYTES_PER_CCS_BYTE(_xe) (GRAPHICS_VER(_xe) >= 20 ? 512 : 256)
+
+#define XY_FAST_COLOR_BLT_CMD (2 << 29 | 0x44 << 22)
+#define XY_FAST_COLOR_BLT_DEPTH_32 (2 << 19)
+#define XY_FAST_COLOR_BLT_DW 16
+#define XY_FAST_COLOR_BLT_MOCS_MASK GENMASK(27, 22)
+#define XE2_XY_FAST_COLOR_BLT_MOCS_INDEX_MASK GENMASK(27, 24)
+#define XY_FAST_COLOR_BLT_MEM_TYPE_SHIFT 31
+
+#define XY_FAST_COPY_BLT_CMD (2 << 29 | 0x42 << 22)
+#define XY_FAST_COPY_BLT_DEPTH_32 (3<<24)
+#define XY_FAST_COPY_BLT_D1_SRC_TILE4 REG_BIT(31)
+#define XY_FAST_COPY_BLT_D1_DST_TILE4 REG_BIT(30)
+#define XE2_XY_FAST_COPY_BLT_MOCS_INDEX_MASK GENMASK(23, 20)
+
+#define PVC_MEM_SET_CMD (2 << 29 | 0x5b << 22)
+#define PVC_MEM_SET_CMD_LEN_DW 7
+#define PVC_MEM_SET_MATRIX REG_BIT(17)
+#define PVC_MEM_SET_DATA_FIELD GENMASK(31, 24)
+/* Bspec lists field as [6:0], but index alone is from [6:1] */
+#define PVC_MEM_SET_MOCS_INDEX_MASK GENMASK(6, 1)
+#define XE2_MEM_SET_MOCS_INDEX_MASK GENMASK(6, 3)
+
+#define GFX_OP_PIPE_CONTROL(len) ((0x3<<29)|(0x3<<27)|(0x2<<24)|((len)-2))
+
+#define PIPE_CONTROL0_HDC_PIPELINE_FLUSH BIT(9) /* gen12 */
+
+#define PIPE_CONTROL_COMMAND_CACHE_INVALIDATE (1<<29)
+#define PIPE_CONTROL_TILE_CACHE_FLUSH (1<<28)
+#define PIPE_CONTROL_AMFS_FLUSH (1<<25)
+#define PIPE_CONTROL_GLOBAL_GTT_IVB (1<<24)
+#define PIPE_CONTROL_LRI_POST_SYNC BIT(23)
+#define PIPE_CONTROL_STORE_DATA_INDEX (1<<21)
+#define PIPE_CONTROL_CS_STALL (1<<20)
+#define PIPE_CONTROL_GLOBAL_SNAPSHOT_RESET (1<<19)
+#define PIPE_CONTROL_TLB_INVALIDATE BIT(18)
+#define PIPE_CONTROL_PSD_SYNC (1<<17)
+#define PIPE_CONTROL_QW_WRITE (1<<14)
+#define PIPE_CONTROL_DEPTH_STALL (1<<13)
+#define PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH (1<<12)
+#define PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE (1<<11)
+#define PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE (1<<10)
+#define PIPE_CONTROL_INDIRECT_STATE_DISABLE (1<<9)
+#define PIPE_CONTROL_FLUSH_ENABLE (1<<7)
+#define PIPE_CONTROL_DC_FLUSH_ENABLE (1<<5)
+#define PIPE_CONTROL_VF_CACHE_INVALIDATE (1<<4)
+#define PIPE_CONTROL_CONST_CACHE_INVALIDATE (1<<3)
+#define PIPE_CONTROL_STATE_CACHE_INVALIDATE (1<<2)
+#define PIPE_CONTROL_STALL_AT_SCOREBOARD (1<<1)
+#define PIPE_CONTROL_DEPTH_CACHE_FLUSH (1<<0)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_gsc_regs.h b/drivers/gpu/drm/xe/regs/xe_gsc_regs.h
new file mode 100644
index 000000000000..9886ec9cb08e
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_gsc_regs.h
@@ -0,0 +1,41 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GSC_REGS_H_
+#define _XE_GSC_REGS_H_
+
+#include <linux/compiler.h>
+#include <linux/types.h>
+
+#include "regs/xe_reg_defs.h"
+
+/* Definitions of GSC H/W registers, bits, etc */
+
+#define MTL_GSC_HECI1_BASE 0x00116000
+#define MTL_GSC_HECI2_BASE 0x00117000
+
+#define HECI_H_CSR(base) XE_REG((base) + 0x4)
+#define HECI_H_CSR_IE REG_BIT(0)
+#define HECI_H_CSR_IS REG_BIT(1)
+#define HECI_H_CSR_IG REG_BIT(2)
+#define HECI_H_CSR_RDY REG_BIT(3)
+#define HECI_H_CSR_RST REG_BIT(4)
+
+/*
+ * The FWSTS register values are FW defined and can be different between
+ * HECI1 and HECI2
+ */
+#define HECI_FWSTS1(base) XE_REG((base) + 0xc40)
+#define HECI1_FWSTS1_CURRENT_STATE REG_GENMASK(3, 0)
+#define HECI1_FWSTS1_CURRENT_STATE_RESET 0
+#define HECI1_FWSTS1_PROXY_STATE_NORMAL 5
+#define HECI1_FWSTS1_INIT_COMPLETE REG_BIT(9)
+#define HECI_FWSTS5(base) XE_REG((base) + 0xc68)
+#define HECI1_FWSTS5_HUC_AUTH_DONE REG_BIT(19)
+
+#define HECI_H_GS1(base) XE_REG((base) + 0xc4c)
+#define HECI_H_GS1_ER_PREP REG_BIT(0)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_gt_regs.h b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
new file mode 100644
index 000000000000..1dd361046b5d
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
@@ -0,0 +1,478 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_REGS_H_
+#define _XE_GT_REGS_H_
+
+#include "regs/xe_reg_defs.h"
+
+/*
+ * The GSI register range [0x0 - 0x40000) is replicated at a higher offset
+ * for the media GT. xe_mmio and xe_gt_mcr functions will automatically
+ * translate offsets by MEDIA_GT_GSI_OFFSET when operating on the media GT.
+ */
+#define MEDIA_GT_GSI_OFFSET 0x380000
+#define MEDIA_GT_GSI_LENGTH 0x40000
+
+/* MTL workpoint reg to get core C state and actual freq of 3D, SAMedia */
+#define MTL_MIRROR_TARGET_WP1 XE_REG(0xc60)
+#define MTL_CAGF_MASK REG_GENMASK(8, 0)
+#define MTL_CC_MASK REG_GENMASK(12, 9)
+
+/* RPM unit config (Gen8+) */
+#define RPM_CONFIG0 XE_REG(0xd00)
+#define RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK REG_GENMASK(5, 3)
+#define RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ 0
+#define RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ 1
+#define RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ 2
+#define RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ 3
+#define RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK REG_GENMASK(2, 1)
+
+#define FORCEWAKE_ACK_MEDIA_VDBOX(n) XE_REG(0xd50 + (n) * 4)
+#define FORCEWAKE_ACK_MEDIA_VEBOX(n) XE_REG(0xd70 + (n) * 4)
+#define FORCEWAKE_ACK_RENDER XE_REG(0xd84)
+
+#define GMD_ID XE_REG(0xd8c)
+#define GMD_ID_ARCH_MASK REG_GENMASK(31, 22)
+#define GMD_ID_RELEASE_MASK REG_GENMASK(21, 14)
+#define GMD_ID_REVID REG_GENMASK(5, 0)
+
+#define FORCEWAKE_ACK_GSC XE_REG(0xdf8)
+#define FORCEWAKE_ACK_GT_MTL XE_REG(0xdfc)
+
+#define MCFG_MCR_SELECTOR XE_REG(0xfd0)
+#define MTL_MCR_SELECTOR XE_REG(0xfd4)
+#define SF_MCR_SELECTOR XE_REG(0xfd8)
+#define MCR_SELECTOR XE_REG(0xfdc)
+#define GAM_MCR_SELECTOR XE_REG(0xfe0)
+#define MCR_MULTICAST REG_BIT(31)
+#define MCR_SLICE_MASK REG_GENMASK(30, 27)
+#define MCR_SLICE(slice) REG_FIELD_PREP(MCR_SLICE_MASK, slice)
+#define MCR_SUBSLICE_MASK REG_GENMASK(26, 24)
+#define MCR_SUBSLICE(subslice) REG_FIELD_PREP(MCR_SUBSLICE_MASK, subslice)
+#define MTL_MCR_GROUPID REG_GENMASK(11, 8)
+#define MTL_MCR_INSTANCEID REG_GENMASK(3, 0)
+
+#define PS_INVOCATION_COUNT XE_REG(0x2348)
+
+#define XELP_GLOBAL_MOCS(i) XE_REG(0x4000 + (i) * 4)
+#define XEHP_GLOBAL_MOCS(i) XE_REG_MCR(0x4000 + (i) * 4)
+#define CCS_AUX_INV XE_REG(0x4208)
+
+#define VD0_AUX_INV XE_REG(0x4218)
+#define VE0_AUX_INV XE_REG(0x4238)
+
+#define VE1_AUX_INV XE_REG(0x42b8)
+#define AUX_INV REG_BIT(0)
+
+#define XEHP_TILE_ADDR_RANGE(_idx) XE_REG_MCR(0x4900 + (_idx) * 4)
+#define XEHP_FLAT_CCS_BASE_ADDR XE_REG_MCR(0x4910)
+
+#define WM_CHICKEN3 XE_REG_MCR(0x5588, XE_REG_OPTION_MASKED)
+#define HIZ_PLANE_COMPRESSION_DIS REG_BIT(10)
+
+#define CHICKEN_RASTER_2 XE_REG_MCR(0x6208, XE_REG_OPTION_MASKED)
+#define TBIMR_FAST_CLIP REG_BIT(5)
+
+#define FF_MODE XE_REG_MCR(0x6210)
+#define DIS_TE_AUTOSTRIP REG_BIT(31)
+#define DIS_MESH_PARTIAL_AUTOSTRIP REG_BIT(16)
+#define DIS_MESH_AUTOSTRIP REG_BIT(15)
+
+#define VFLSKPD XE_REG_MCR(0x62a8, XE_REG_OPTION_MASKED)
+#define DIS_PARTIAL_AUTOSTRIP REG_BIT(9)
+#define DIS_AUTOSTRIP REG_BIT(6)
+#define DIS_OVER_FETCH_CACHE REG_BIT(1)
+#define DIS_MULT_MISS_RD_SQUASH REG_BIT(0)
+
+#define FF_MODE2 XE_REG(0x6604)
+#define XEHP_FF_MODE2 XE_REG_MCR(0x6604)
+#define FF_MODE2_GS_TIMER_MASK REG_GENMASK(31, 24)
+#define FF_MODE2_GS_TIMER_224 REG_FIELD_PREP(FF_MODE2_GS_TIMER_MASK, 224)
+#define FF_MODE2_TDS_TIMER_MASK REG_GENMASK(23, 16)
+#define FF_MODE2_TDS_TIMER_128 REG_FIELD_PREP(FF_MODE2_TDS_TIMER_MASK, 4)
+
+#define CACHE_MODE_1 XE_REG(0x7004, XE_REG_OPTION_MASKED)
+#define MSAA_OPTIMIZATION_REDUC_DISABLE REG_BIT(11)
+
+#define COMMON_SLICE_CHICKEN1 XE_REG(0x7010)
+
+#define HIZ_CHICKEN XE_REG(0x7018, XE_REG_OPTION_MASKED)
+#define DG1_HZ_READ_SUPPRESSION_OPTIMIZATION_DISABLE REG_BIT(14)
+#define HZ_DEPTH_TEST_LE_GE_OPT_DISABLE REG_BIT(13)
+
+#define XEHP_PSS_MODE2 XE_REG_MCR(0x703c, XE_REG_OPTION_MASKED)
+#define SCOREBOARD_STALL_FLUSH_CONTROL REG_BIT(5)
+
+#define XEHP_PSS_CHICKEN XE_REG_MCR(0x7044, XE_REG_OPTION_MASKED)
+#define FLSH_IGNORES_PSD REG_BIT(10)
+#define FD_END_COLLECT REG_BIT(5)
+
+#define COMMON_SLICE_CHICKEN4 XE_REG(0x7300, XE_REG_OPTION_MASKED)
+#define DISABLE_TDC_LOAD_BALANCING_CALC REG_BIT(6)
+
+#define COMMON_SLICE_CHICKEN3 XE_REG(0x7304, XE_REG_OPTION_MASKED)
+#define XEHP_COMMON_SLICE_CHICKEN3 XE_REG_MCR(0x7304, XE_REG_OPTION_MASKED)
+#define DG1_FLOAT_POINT_BLEND_OPT_STRICT_MODE_EN REG_BIT(12)
+#define XEHP_DUAL_SIMD8_SEQ_MERGE_DISABLE REG_BIT(12)
+#define BLEND_EMB_FIX_DISABLE_IN_RCC REG_BIT(11)
+#define DISABLE_CPS_AWARE_COLOR_PIPE REG_BIT(9)
+
+#define XEHP_SLICE_COMMON_ECO_CHICKEN1 XE_REG_MCR(0x731c, XE_REG_OPTION_MASKED)
+#define MSC_MSAA_REODER_BUF_BYPASS_DISABLE REG_BIT(14)
+
+#define VF_PREEMPTION XE_REG(0x83a4, XE_REG_OPTION_MASKED)
+#define PREEMPTION_VERTEX_COUNT REG_GENMASK(15, 0)
+
+#define VF_SCRATCHPAD XE_REG(0x83a8, XE_REG_OPTION_MASKED)
+#define XE2_VFG_TED_CREDIT_INTERFACE_DISABLE REG_BIT(13)
+
+#define VFG_PREEMPTION_CHICKEN XE_REG(0x83b4, XE_REG_OPTION_MASKED)
+#define POLYGON_TRIFAN_LINELOOP_DISABLE REG_BIT(4)
+
+#define SQCNT1 XE_REG_MCR(0x8718)
+#define XELPMP_SQCNT1 XE_REG(0x8718)
+#define ENFORCE_RAR REG_BIT(23)
+
+#define XEHP_SQCM XE_REG_MCR(0x8724)
+#define EN_32B_ACCESS REG_BIT(30)
+
+#define XE2_FLAT_CCS_BASE_RANGE_LOWER XE_REG_MCR(0x8800)
+#define XE2_FLAT_CCS_ENABLE REG_BIT(0)
+
+#define GSCPSMI_BASE XE_REG(0x880c)
+
+/* Fuse readout registers for GT */
+#define XEHP_FUSE4 XE_REG(0x9114)
+#define CCS_EN_MASK REG_GENMASK(19, 16)
+#define GT_L3_EXC_MASK REG_GENMASK(6, 4)
+
+#define MIRROR_FUSE3 XE_REG(0x9118)
+#define XE2_NODE_ENABLE_MASK REG_GENMASK(31, 16)
+#define L3BANK_PAIR_COUNT 4
+#define L3BANK_MASK REG_GENMASK(3, 0)
+/* on Xe_HP the same fuses indicates mslices instead of L3 banks */
+#define MAX_MSLICES 4
+#define MEML3_EN_MASK REG_GENMASK(3, 0)
+
+#define XELP_EU_ENABLE XE_REG(0x9134) /* "_DISABLE" on Xe_LP */
+#define XELP_EU_MASK REG_GENMASK(7, 0)
+#define XELP_GT_GEOMETRY_DSS_ENABLE XE_REG(0x913c)
+
+#define GT_VEBOX_VDBOX_DISABLE XE_REG(0x9140)
+#define GT_VEBOX_DISABLE_MASK REG_GENMASK(19, 16)
+#define GT_VDBOX_DISABLE_MASK REG_GENMASK(7, 0)
+
+#define XEHP_GT_COMPUTE_DSS_ENABLE XE_REG(0x9144)
+#define XEHPC_GT_COMPUTE_DSS_ENABLE_EXT XE_REG(0x9148)
+#define XE2_GT_COMPUTE_DSS_2 XE_REG(0x914c)
+#define XE2_GT_GEOMETRY_DSS_1 XE_REG(0x9150)
+#define XE2_GT_GEOMETRY_DSS_2 XE_REG(0x9154)
+
+#define GDRST XE_REG(0x941c)
+#define GRDOM_GUC REG_BIT(3)
+#define GRDOM_FULL REG_BIT(0)
+
+#define MISCCPCTL XE_REG(0x9424)
+#define DOP_CLOCK_GATE_RENDER_ENABLE REG_BIT(1)
+
+#define UNSLCGCTL9430 XE_REG(0x9430)
+#define MSQDUNIT_CLKGATE_DIS REG_BIT(3)
+
+#define UNSLICE_UNIT_LEVEL_CLKGATE XE_REG(0x9434)
+#define VFUNIT_CLKGATE_DIS REG_BIT(20)
+#define TSGUNIT_CLKGATE_DIS REG_BIT(17) /* XEHPSDV */
+#define CG3DDISCFEG_CLKGATE_DIS REG_BIT(17) /* DG2 */
+#define GAMEDIA_CLKGATE_DIS REG_BIT(11)
+#define HSUNIT_CLKGATE_DIS REG_BIT(8)
+#define VSUNIT_CLKGATE_DIS REG_BIT(3)
+
+#define UNSLCGCTL9440 XE_REG(0x9440)
+#define GAMTLBOACS_CLKGATE_DIS REG_BIT(28)
+#define GAMTLBVDBOX5_CLKGATE_DIS REG_BIT(27)
+#define GAMTLBVDBOX6_CLKGATE_DIS REG_BIT(26)
+#define GAMTLBVDBOX3_CLKGATE_DIS REG_BIT(24)
+#define GAMTLBVDBOX4_CLKGATE_DIS REG_BIT(23)
+#define GAMTLBVDBOX7_CLKGATE_DIS REG_BIT(22)
+#define GAMTLBVDBOX2_CLKGATE_DIS REG_BIT(21)
+#define GAMTLBVDBOX0_CLKGATE_DIS REG_BIT(17)
+#define GAMTLBKCR_CLKGATE_DIS REG_BIT(16)
+#define GAMTLBGUC_CLKGATE_DIS REG_BIT(15)
+#define GAMTLBBLT_CLKGATE_DIS REG_BIT(14)
+#define GAMTLBVDBOX1_CLKGATE_DIS REG_BIT(6)
+
+#define UNSLCGCTL9444 XE_REG(0x9444)
+#define GAMTLBGFXA0_CLKGATE_DIS REG_BIT(30)
+#define GAMTLBGFXA1_CLKGATE_DIS REG_BIT(29)
+#define GAMTLBCOMPA0_CLKGATE_DIS REG_BIT(28)
+#define GAMTLBCOMPA1_CLKGATE_DIS REG_BIT(27)
+#define GAMTLBCOMPB0_CLKGATE_DIS REG_BIT(26)
+#define GAMTLBCOMPB1_CLKGATE_DIS REG_BIT(25)
+#define GAMTLBCOMPC0_CLKGATE_DIS REG_BIT(24)
+#define GAMTLBCOMPC1_CLKGATE_DIS REG_BIT(23)
+#define GAMTLBCOMPD0_CLKGATE_DIS REG_BIT(22)
+#define GAMTLBCOMPD1_CLKGATE_DIS REG_BIT(21)
+#define GAMTLBMERT_CLKGATE_DIS REG_BIT(20)
+#define GAMTLBVEBOX3_CLKGATE_DIS REG_BIT(19)
+#define GAMTLBVEBOX2_CLKGATE_DIS REG_BIT(18)
+#define GAMTLBVEBOX1_CLKGATE_DIS REG_BIT(17)
+#define GAMTLBVEBOX0_CLKGATE_DIS REG_BIT(16)
+#define LTCDD_CLKGATE_DIS REG_BIT(10)
+
+#define XEHP_SLICE_UNIT_LEVEL_CLKGATE XE_REG_MCR(0x94d4)
+#define L3_CR2X_CLKGATE_DIS REG_BIT(17)
+#define L3_CLKGATE_DIS REG_BIT(16)
+#define NODEDSS_CLKGATE_DIS REG_BIT(12)
+#define MSCUNIT_CLKGATE_DIS REG_BIT(10)
+#define RCCUNIT_CLKGATE_DIS REG_BIT(7)
+#define SARBUNIT_CLKGATE_DIS REG_BIT(5)
+#define SBEUNIT_CLKGATE_DIS REG_BIT(4)
+
+#define UNSLICE_UNIT_LEVEL_CLKGATE2 XE_REG(0x94e4)
+#define VSUNIT_CLKGATE2_DIS REG_BIT(19)
+
+#define SUBSLICE_UNIT_LEVEL_CLKGATE XE_REG_MCR(0x9524)
+#define DSS_ROUTER_CLKGATE_DIS REG_BIT(28)
+#define GWUNIT_CLKGATE_DIS REG_BIT(16)
+
+#define SUBSLICE_UNIT_LEVEL_CLKGATE2 XE_REG_MCR(0x9528)
+#define CPSSUNIT_CLKGATE_DIS REG_BIT(9)
+
+#define SSMCGCTL9530 XE_REG_MCR(0x9530)
+#define RTFUNIT_CLKGATE_DIS REG_BIT(18)
+
+#define DFR_RATIO_EN_AND_CHICKEN XE_REG_MCR(0x9550)
+#define DFR_DISABLE REG_BIT(9)
+
+#define RPNSWREQ XE_REG(0xa008)
+#define REQ_RATIO_MASK REG_GENMASK(31, 23)
+
+#define RP_CONTROL XE_REG(0xa024)
+#define RPSWCTL_MASK REG_GENMASK(10, 9)
+#define RPSWCTL_ENABLE REG_FIELD_PREP(RPSWCTL_MASK, 2)
+#define RPSWCTL_DISABLE REG_FIELD_PREP(RPSWCTL_MASK, 0)
+#define RC_CONTROL XE_REG(0xa090)
+#define RC_CTL_HW_ENABLE REG_BIT(31)
+#define RC_CTL_TO_MODE REG_BIT(28)
+#define RC_CTL_RC6_ENABLE REG_BIT(18)
+#define RC_STATE XE_REG(0xa094)
+#define RC_IDLE_HYSTERSIS XE_REG(0xa0ac)
+
+#define PMINTRMSK XE_REG(0xa168)
+#define PMINTR_DISABLE_REDIRECT_TO_GUC REG_BIT(31)
+#define ARAT_EXPIRED_INTRMSK REG_BIT(9)
+
+#define FORCEWAKE_GT XE_REG(0xa188)
+
+#define PG_ENABLE XE_REG(0xa210)
+
+#define CTC_MODE XE_REG(0xa26c)
+#define CTC_SHIFT_PARAMETER_MASK REG_GENMASK(2, 1)
+#define CTC_SOURCE_DIVIDE_LOGIC REG_BIT(0)
+
+#define FORCEWAKE_RENDER XE_REG(0xa278)
+#define FORCEWAKE_MEDIA_VDBOX(n) XE_REG(0xa540 + (n) * 4)
+#define FORCEWAKE_MEDIA_VEBOX(n) XE_REG(0xa560 + (n) * 4)
+#define FORCEWAKE_GSC XE_REG(0xa618)
+
+#define XEHPC_LNCFMISCCFGREG0 XE_REG_MCR(0xb01c, XE_REG_OPTION_MASKED)
+#define XEHPC_OVRLSCCC REG_BIT(0)
+
+/* L3 Cache Control */
+#define XELP_LNCFCMOCS(i) XE_REG(0xb020 + (i) * 4)
+#define XEHP_LNCFCMOCS(i) XE_REG_MCR(0xb020 + (i) * 4)
+#define LNCFCMOCS_REG_COUNT 32
+
+#define XEHP_L3NODEARBCFG XE_REG_MCR(0xb0b4)
+#define XEHP_LNESPARE REG_BIT(19)
+
+#define XEHP_L3SQCREG5 XE_REG_MCR(0xb158)
+#define L3_PWM_TIMER_INIT_VAL_MASK REG_GENMASK(9, 0)
+
+#define XEHP_L3SCQREG7 XE_REG_MCR(0xb188)
+#define BLEND_FILL_CACHING_OPT_DIS REG_BIT(3)
+
+#define XEHPC_L3CLOS_MASK(i) XE_REG_MCR(0xb194 + (i) * 8)
+
+#define XE2LPM_L3SQCREG5 XE_REG_MCR(0xb658)
+
+#define XEHP_MERT_MOD_CTRL XE_REG_MCR(0xcf28)
+#define RENDER_MOD_CTRL XE_REG_MCR(0xcf2c)
+#define COMP_MOD_CTRL XE_REG_MCR(0xcf30)
+#define XEHP_VDBX_MOD_CTRL XE_REG_MCR(0xcf34)
+#define XELPMP_VDBX_MOD_CTRL XE_REG(0xcf34)
+#define XEHP_VEBX_MOD_CTRL XE_REG_MCR(0xcf38)
+#define XELPMP_VEBX_MOD_CTRL XE_REG(0xcf38)
+#define FORCE_MISS_FTLB REG_BIT(3)
+
+#define XEHP_GAMSTLB_CTRL XE_REG_MCR(0xcf4c)
+#define CONTROL_BLOCK_CLKGATE_DIS REG_BIT(12)
+#define EGRESS_BLOCK_CLKGATE_DIS REG_BIT(11)
+#define TAG_BLOCK_CLKGATE_DIS REG_BIT(7)
+
+#define XEHP_GAMCNTRL_CTRL XE_REG_MCR(0xcf54)
+#define INVALIDATION_BROADCAST_MODE_DIS REG_BIT(12)
+#define GLOBAL_INVALIDATION_MODE REG_BIT(2)
+
+#define HALF_SLICE_CHICKEN5 XE_REG_MCR(0xe188, XE_REG_OPTION_MASKED)
+#define DISABLE_SAMPLE_G_PERFORMANCE REG_BIT(0)
+
+#define SAMPLER_MODE XE_REG_MCR(0xe18c, XE_REG_OPTION_MASKED)
+#define ENABLE_SMALLPL REG_BIT(15)
+#define SC_DISABLE_POWER_OPTIMIZATION_EBB REG_BIT(9)
+#define SAMPLER_ENABLE_HEADLESS_MSG REG_BIT(5)
+#define INDIRECT_STATE_BASE_ADDR_OVERRIDE REG_BIT(0)
+
+#define HALF_SLICE_CHICKEN7 XE_REG_MCR(0xe194, XE_REG_OPTION_MASKED)
+#define DG2_DISABLE_ROUND_ENABLE_ALLOW_FOR_SSLA REG_BIT(15)
+
+#define CACHE_MODE_SS XE_REG_MCR(0xe420, XE_REG_OPTION_MASKED)
+#define DISABLE_ECC REG_BIT(5)
+#define ENABLE_PREFETCH_INTO_IC REG_BIT(3)
+
+#define ROW_CHICKEN4 XE_REG_MCR(0xe48c, XE_REG_OPTION_MASKED)
+#define DISABLE_GRF_CLEAR REG_BIT(13)
+#define XEHP_DIS_BBL_SYSPIPE REG_BIT(11)
+#define DISABLE_TDL_PUSH REG_BIT(9)
+#define DIS_PICK_2ND_EU REG_BIT(7)
+#define DISABLE_HDR_PAST_PAYLOAD_HOLD_FIX REG_BIT(4)
+#define THREAD_EX_ARB_MODE REG_GENMASK(3, 2)
+#define THREAD_EX_ARB_MODE_RR_AFTER_DEP REG_FIELD_PREP(THREAD_EX_ARB_MODE, 0x2)
+
+#define ROW_CHICKEN3 XE_REG_MCR(0xe49c, XE_REG_OPTION_MASKED)
+#define DIS_FIX_EOT1_FLUSH REG_BIT(9)
+
+#define ROW_CHICKEN XE_REG_MCR(0xe4f0, XE_REG_OPTION_MASKED)
+#define UGM_BACKUP_MODE REG_BIT(13)
+#define MDQ_ARBITRATION_MODE REG_BIT(12)
+#define EARLY_EOT_DIS REG_BIT(1)
+
+#define ROW_CHICKEN2 XE_REG_MCR(0xe4f4, XE_REG_OPTION_MASKED)
+#define DISABLE_READ_SUPPRESSION REG_BIT(15)
+#define DISABLE_EARLY_READ REG_BIT(14)
+#define ENABLE_LARGE_GRF_MODE REG_BIT(12)
+#define PUSH_CONST_DEREF_HOLD_DIS REG_BIT(8)
+#define DISABLE_DOP_GATING REG_BIT(0)
+
+#define RT_CTRL XE_REG_MCR(0xe530)
+#define DIS_NULL_QUERY REG_BIT(10)
+
+#define XEHP_HDC_CHICKEN0 XE_REG_MCR(0xe5f0, XE_REG_OPTION_MASKED)
+#define LSC_L1_FLUSH_CTL_3D_DATAPORT_FLUSH_EVENTS_MASK REG_GENMASK(13, 11)
+#define DIS_ATOMIC_CHAINING_TYPED_WRITES REG_BIT(3)
+
+#define LSC_CHICKEN_BIT_0 XE_REG_MCR(0xe7c8)
+#define DISABLE_D8_D16_COASLESCE REG_BIT(30)
+#define TGM_WRITE_EOM_FORCE REG_BIT(17)
+#define FORCE_1_SUB_MESSAGE_PER_FRAGMENT REG_BIT(15)
+#define SEQUENTIAL_ACCESS_UPGRADE_DISABLE REG_BIT(13)
+
+#define LSC_CHICKEN_BIT_0_UDW XE_REG_MCR(0xe7c8 + 4)
+#define UGM_FRAGMENT_THRESHOLD_TO_3 REG_BIT(58 - 32)
+#define DIS_CHAIN_2XSIMD8 REG_BIT(55 - 32)
+#define XE2_ALLOC_DPA_STARVE_FIX_DIS REG_BIT(47 - 32)
+#define ENABLE_SMP_LD_RENDER_SURFACE_CONTROL REG_BIT(44 - 32)
+#define FORCE_SLM_FENCE_SCOPE_TO_TILE REG_BIT(42 - 32)
+#define FORCE_UGM_FENCE_SCOPE_TO_TILE REG_BIT(41 - 32)
+#define MAXREQS_PER_BANK REG_GENMASK(39 - 32, 37 - 32)
+#define DISABLE_128B_EVICTION_COMMAND_UDW REG_BIT(36 - 32)
+
+#define SARB_CHICKEN1 XE_REG_MCR(0xe90c)
+#define COMP_CKN_IN REG_GENMASK(30, 29)
+
+#define RCU_MODE XE_REG(0x14800, XE_REG_OPTION_MASKED)
+#define RCU_MODE_FIXED_SLICE_CCS_MODE REG_BIT(1)
+#define RCU_MODE_CCS_ENABLE REG_BIT(0)
+
+/*
+ * Total of 4 cslices, where each cslice is in the form:
+ * [0-3] CCS ID
+ * [4-6] RSVD
+ * [7] Disabled
+ */
+#define CCS_MODE XE_REG(0x14804)
+#define CCS_MODE_CSLICE_0_3_MASK REG_GENMASK(11, 0) /* 3 bits per cslice */
+#define CCS_MODE_CSLICE_MASK 0x7 /* CCS0-3 + rsvd */
+#define CCS_MODE_CSLICE_WIDTH ilog2(CCS_MODE_CSLICE_MASK + 1)
+#define CCS_MODE_CSLICE(cslice, ccs) \
+ ((ccs) << ((cslice) * CCS_MODE_CSLICE_WIDTH))
+
+#define FORCEWAKE_ACK_GT XE_REG(0x130044)
+#define FORCEWAKE_KERNEL BIT(0)
+#define FORCEWAKE_USER BIT(1)
+#define FORCEWAKE_KERNEL_FALLBACK BIT(15)
+
+#define MTL_MEDIA_PERF_LIMIT_REASONS XE_REG(0x138030)
+#define MTL_MEDIA_MC6 XE_REG(0x138048)
+
+#define GT_CORE_STATUS XE_REG(0x138060)
+#define RCN_MASK REG_GENMASK(2, 0)
+#define GT_C0 0
+#define GT_C6 3
+
+#define GT_GFX_RC6_LOCKED XE_REG(0x138104)
+#define GT_GFX_RC6 XE_REG(0x138108)
+
+#define GT0_PERF_LIMIT_REASONS XE_REG(0x1381a8)
+#define GT0_PERF_LIMIT_REASONS_MASK 0xde3
+#define PROCHOT_MASK REG_BIT(0)
+#define THERMAL_LIMIT_MASK REG_BIT(1)
+#define RATL_MASK REG_BIT(5)
+#define VR_THERMALERT_MASK REG_BIT(6)
+#define VR_TDC_MASK REG_BIT(7)
+#define POWER_LIMIT_4_MASK REG_BIT(8)
+#define POWER_LIMIT_1_MASK REG_BIT(10)
+#define POWER_LIMIT_2_MASK REG_BIT(11)
+
+#define GT_PERF_STATUS XE_REG(0x1381b4)
+#define VOLTAGE_MASK REG_GENMASK(10, 0)
+
+#define GT_INTR_DW(x) XE_REG(0x190018 + ((x) * 4))
+
+#define RENDER_COPY_INTR_ENABLE XE_REG(0x190030)
+#define VCS_VECS_INTR_ENABLE XE_REG(0x190034)
+#define GUC_SG_INTR_ENABLE XE_REG(0x190038)
+#define ENGINE1_MASK REG_GENMASK(31, 16)
+#define ENGINE0_MASK REG_GENMASK(15, 0)
+#define GPM_WGBOXPERF_INTR_ENABLE XE_REG(0x19003c)
+#define GUNIT_GSC_INTR_ENABLE XE_REG(0x190044)
+#define CCS_RSVD_INTR_ENABLE XE_REG(0x190048)
+
+#define INTR_IDENTITY_REG(x) XE_REG(0x190060 + ((x) * 4))
+#define INTR_DATA_VALID REG_BIT(31)
+#define INTR_ENGINE_INSTANCE(x) REG_FIELD_GET(GENMASK(25, 20), x)
+#define INTR_ENGINE_CLASS(x) REG_FIELD_GET(GENMASK(18, 16), x)
+#define INTR_ENGINE_INTR(x) REG_FIELD_GET(GENMASK(15, 0), x)
+#define OTHER_GUC_INSTANCE 0
+#define OTHER_GSC_INSTANCE 6
+
+#define IIR_REG_SELECTOR(x) XE_REG(0x190070 + ((x) * 4))
+#define RCS0_RSVD_INTR_MASK XE_REG(0x190090)
+#define BCS_RSVD_INTR_MASK XE_REG(0x1900a0)
+#define VCS0_VCS1_INTR_MASK XE_REG(0x1900a8)
+#define VCS2_VCS3_INTR_MASK XE_REG(0x1900ac)
+#define VECS0_VECS1_INTR_MASK XE_REG(0x1900d0)
+#define GUC_SG_INTR_MASK XE_REG(0x1900e8)
+#define GPM_WGBOXPERF_INTR_MASK XE_REG(0x1900ec)
+#define GUNIT_GSC_INTR_MASK XE_REG(0x1900f4)
+#define CCS0_CCS1_INTR_MASK XE_REG(0x190100)
+#define CCS2_CCS3_INTR_MASK XE_REG(0x190104)
+#define XEHPC_BCS1_BCS2_INTR_MASK XE_REG(0x190110)
+#define XEHPC_BCS3_BCS4_INTR_MASK XE_REG(0x190114)
+#define XEHPC_BCS5_BCS6_INTR_MASK XE_REG(0x190118)
+#define XEHPC_BCS7_BCS8_INTR_MASK XE_REG(0x19011c)
+#define GT_WAIT_SEMAPHORE_INTERRUPT REG_BIT(11)
+#define GT_CONTEXT_SWITCH_INTERRUPT REG_BIT(8)
+#define GT_RENDER_PIPECTL_NOTIFY_INTERRUPT REG_BIT(4)
+#define GT_CS_MASTER_ERROR_INTERRUPT REG_BIT(3)
+#define GT_RENDER_USER_INTERRUPT REG_BIT(0)
+
+#define PVC_GT0_PACKAGE_ENERGY_STATUS XE_REG(0x281004)
+#define PVC_GT0_PACKAGE_RAPL_LIMIT XE_REG(0x281008)
+#define PVC_GT0_PACKAGE_POWER_SKU_UNIT XE_REG(0x281068)
+#define PVC_GT0_PLATFORM_ENERGY_STATUS XE_REG(0x28106c)
+#define PVC_GT0_PACKAGE_POWER_SKU XE_REG(0x281080)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_guc_regs.h b/drivers/gpu/drm/xe/regs/xe_guc_regs.h
new file mode 100644
index 000000000000..92320bbc9d3d
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_guc_regs.h
@@ -0,0 +1,143 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_REGS_H_
+#define _XE_GUC_REGS_H_
+
+#include <linux/compiler.h>
+#include <linux/types.h>
+
+#include "regs/xe_reg_defs.h"
+
+/* Definitions of GuC H/W registers, bits, etc */
+
+#define DIST_DBS_POPULATED XE_REG(0xd08)
+#define DOORBELLS_PER_SQIDI_MASK REG_GENMASK(23, 16)
+#define SQIDIS_DOORBELL_EXIST_MASK REG_GENMASK(15, 0)
+
+#define DRBREGL(x) XE_REG(0x1000 + (x) * 8)
+#define DRB_VALID REG_BIT(0)
+#define DRBREGU(x) XE_REG(0x1000 + (x) * 8 + 4)
+
+#define GTCR XE_REG(0x4274)
+#define GTCR_INVALIDATE REG_BIT(0)
+
+#define GUC_ARAT_C6DIS XE_REG(0xa178)
+
+#define GUC_STATUS XE_REG(0xc000)
+#define GS_AUTH_STATUS_MASK REG_GENMASK(31, 30)
+#define GS_AUTH_STATUS_BAD REG_FIELD_PREP(GS_AUTH_STATUS_MASK, 0x1)
+#define GS_AUTH_STATUS_GOOD REG_FIELD_PREP(GS_AUTH_STATUS_MASK, 0x2)
+#define GS_MIA_MASK REG_GENMASK(18, 16)
+#define GS_MIA_CORE_STATE REG_FIELD_PREP(GS_MIA_MASK, 0x1)
+#define GS_MIA_HALT_REQUESTED REG_FIELD_PREP(GS_MIA_MASK, 0x2)
+#define GS_MIA_ISR_ENTRY REG_FIELD_PREP(GS_MIA_MASK, 0x4)
+#define GS_UKERNEL_MASK REG_GENMASK(15, 8)
+#define GS_BOOTROM_MASK REG_GENMASK(7, 1)
+#define GS_BOOTROM_RSA_FAILED REG_FIELD_PREP(GS_BOOTROM_MASK, 0x50)
+#define GS_BOOTROM_JUMP_PASSED REG_FIELD_PREP(GS_BOOTROM_MASK, 0x76)
+#define GS_MIA_IN_RESET REG_BIT(0)
+
+#define GUC_WOPCM_SIZE XE_REG(0xc050)
+#define GUC_WOPCM_SIZE_MASK REG_GENMASK(31, 12)
+#define GUC_WOPCM_SIZE_LOCKED REG_BIT(0)
+
+#define GUC_SHIM_CONTROL XE_REG(0xc064)
+#define GUC_MOCS_INDEX_MASK REG_GENMASK(27, 24)
+#define GUC_SHIM_WC_ENABLE REG_BIT(21)
+#define GUC_ENABLE_MIA_CLOCK_GATING REG_BIT(15)
+#define GUC_ENABLE_READ_CACHE_FOR_WOPCM_DATA REG_BIT(10)
+#define GUC_ENABLE_READ_CACHE_FOR_SRAM_DATA REG_BIT(9)
+#define GUC_MSGCH_ENABLE REG_BIT(4)
+#define GUC_ENABLE_MIA_CACHING REG_BIT(2)
+#define GUC_ENABLE_READ_CACHE_LOGIC REG_BIT(1)
+#define GUC_DISABLE_SRAM_INIT_TO_ZEROES REG_BIT(0)
+
+#define SOFT_SCRATCH(n) XE_REG(0xc180 + (n) * 4)
+#define SOFT_SCRATCH_COUNT 16
+
+#define HUC_KERNEL_LOAD_INFO XE_REG(0xc1dc)
+#define HUC_LOAD_SUCCESSFUL REG_BIT(0)
+
+#define UOS_RSA_SCRATCH(i) XE_REG(0xc200 + (i) * 4)
+#define UOS_RSA_SCRATCH_COUNT 64
+
+#define DMA_ADDR_0_LOW XE_REG(0xc300)
+#define DMA_ADDR_0_HIGH XE_REG(0xc304)
+#define DMA_ADDR_1_LOW XE_REG(0xc308)
+#define DMA_ADDR_1_HIGH XE_REG(0xc30c)
+#define DMA_ADDR_SPACE_MASK REG_GENMASK(20, 16)
+#define DMA_ADDRESS_SPACE_WOPCM REG_FIELD_PREP(DMA_ADDR_SPACE_MASK, 7)
+#define DMA_ADDRESS_SPACE_GGTT REG_FIELD_PREP(DMA_ADDR_SPACE_MASK, 8)
+#define DMA_COPY_SIZE XE_REG(0xc310)
+#define DMA_CTRL XE_REG(0xc314)
+#define HUC_UKERNEL REG_BIT(9)
+#define UOS_MOVE REG_BIT(4)
+#define START_DMA REG_BIT(0)
+#define DMA_GUC_WOPCM_OFFSET XE_REG(0xc340)
+#define GUC_WOPCM_OFFSET_SHIFT 14
+#define GUC_WOPCM_OFFSET_MASK REG_GENMASK(31, GUC_WOPCM_OFFSET_SHIFT)
+#define HUC_LOADING_AGENT_GUC REG_BIT(1)
+#define GUC_WOPCM_OFFSET_VALID REG_BIT(0)
+#define GUC_MAX_IDLE_COUNT XE_REG(0xc3e4)
+
+#define GUC_SEND_INTERRUPT XE_REG(0xc4c8)
+#define GUC_SEND_TRIGGER REG_BIT(0)
+
+#define GUC_BCS_RCS_IER XE_REG(0xc550)
+#define GUC_VCS2_VCS1_IER XE_REG(0xc554)
+#define GUC_WD_VECS_IER XE_REG(0xc558)
+#define GUC_PM_P24C_IER XE_REG(0xc55c)
+
+#define GUC_TLB_INV_CR XE_REG(0xcee8)
+#define GUC_TLB_INV_CR_INVALIDATE REG_BIT(0)
+
+#define HUC_STATUS2 XE_REG(0xd3b0)
+#define HUC_FW_VERIFIED REG_BIT(7)
+
+#define GT_PM_CONFIG XE_REG(0x13816c)
+#define GT_DOORBELL_ENABLE REG_BIT(0)
+
+#define GUC_HOST_INTERRUPT XE_REG(0x1901f0)
+
+#define VF_SW_FLAG(n) XE_REG(0x190240 + (n) * 4)
+#define VF_SW_FLAG_COUNT 4
+
+#define MED_GUC_HOST_INTERRUPT XE_REG(0x190304)
+
+#define MED_VF_SW_FLAG(n) XE_REG(0x190310 + (n) * 4)
+#define MED_VF_SW_FLAG_COUNT 4
+
+/* GuC Interrupt Vector */
+#define GUC_INTR_GUC2HOST REG_BIT(15)
+#define GUC_INTR_EXEC_ERROR REG_BIT(14)
+#define GUC_INTR_DISPLAY_EVENT REG_BIT(13)
+#define GUC_INTR_SEM_SIG REG_BIT(12)
+#define GUC_INTR_IOMMU2GUC REG_BIT(11)
+#define GUC_INTR_DOORBELL_RANG REG_BIT(10)
+#define GUC_INTR_DMA_DONE REG_BIT(9)
+#define GUC_INTR_FATAL_ERROR REG_BIT(8)
+#define GUC_INTR_NOTIF_ERROR REG_BIT(7)
+#define GUC_INTR_SW_INT_6 REG_BIT(6)
+#define GUC_INTR_SW_INT_5 REG_BIT(5)
+#define GUC_INTR_SW_INT_4 REG_BIT(4)
+#define GUC_INTR_SW_INT_3 REG_BIT(3)
+#define GUC_INTR_SW_INT_2 REG_BIT(2)
+#define GUC_INTR_SW_INT_1 REG_BIT(1)
+#define GUC_INTR_SW_INT_0 REG_BIT(0)
+
+#define GUC_NUM_DOORBELLS 256
+
+/* format of the HW-monitored doorbell cacheline */
+struct guc_doorbell_info {
+ u32 db_status;
+#define GUC_DOORBELL_DISABLED 0
+#define GUC_DOORBELL_ENABLED 1
+
+ u32 cookie;
+ u32 reserved[14];
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_lrc_layout.h b/drivers/gpu/drm/xe/regs/xe_lrc_layout.h
new file mode 100644
index 000000000000..4be81abc86ad
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_lrc_layout.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_LRC_LAYOUT_H_
+#define _XE_LRC_LAYOUT_H_
+
+#define CTX_CONTEXT_CONTROL (0x02 + 1)
+#define CTX_RING_HEAD (0x04 + 1)
+#define CTX_RING_TAIL (0x06 + 1)
+#define CTX_RING_START (0x08 + 1)
+#define CTX_RING_CTL (0x0a + 1)
+#define CTX_PDP0_UDW (0x30 + 1)
+#define CTX_PDP0_LDW (0x32 + 1)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_mchbar_regs.h b/drivers/gpu/drm/xe/regs/xe_mchbar_regs.h
new file mode 100644
index 000000000000..519dd1067a19
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_mchbar_regs.h
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_MCHBAR_REGS_H_
+#define _XE_MCHBAR_REGS_H_
+
+#include "regs/xe_reg_defs.h"
+
+/*
+ * MCHBAR mirror.
+ *
+ * This mirrors the MCHBAR MMIO space whose location is determined by
+ * device 0 function 0's pci config register 0x44 or 0x48 and matches it in
+ * every way.
+ */
+
+#define MCHBAR_MIRROR_BASE_SNB 0x140000
+
+#define PCU_CR_PACKAGE_POWER_SKU XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5930)
+#define PKG_TDP GENMASK_ULL(14, 0)
+#define PKG_MIN_PWR GENMASK_ULL(30, 16)
+#define PKG_MAX_PWR GENMASK_ULL(46, 32)
+#define PKG_MAX_WIN GENMASK_ULL(54, 48)
+#define PKG_MAX_WIN_X GENMASK_ULL(54, 53)
+#define PKG_MAX_WIN_Y GENMASK_ULL(52, 48)
+
+
+#define PCU_CR_PACKAGE_POWER_SKU_UNIT XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5938)
+#define PKG_PWR_UNIT REG_GENMASK(3, 0)
+#define PKG_ENERGY_UNIT REG_GENMASK(12, 8)
+#define PKG_TIME_UNIT REG_GENMASK(19, 16)
+
+#define PCU_CR_PACKAGE_ENERGY_STATUS XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x593c)
+
+#define PCU_CR_PACKAGE_RAPL_LIMIT XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x59a0)
+#define PKG_PWR_LIM_1 REG_GENMASK(14, 0)
+#define PKG_PWR_LIM_1_EN REG_BIT(15)
+#define PKG_PWR_LIM_1_TIME REG_GENMASK(23, 17)
+#define PKG_PWR_LIM_1_TIME_X REG_GENMASK(23, 22)
+#define PKG_PWR_LIM_1_TIME_Y REG_GENMASK(21, 17)
+
+#endif /* _XE_MCHBAR_REGS_H_ */
diff --git a/drivers/gpu/drm/xe/regs/xe_reg_defs.h b/drivers/gpu/drm/xe/regs/xe_reg_defs.h
new file mode 100644
index 000000000000..c50e7650c09a
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_reg_defs.h
@@ -0,0 +1,120 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_REG_DEFS_H_
+#define _XE_REG_DEFS_H_
+
+#include "compat-i915-headers/i915_reg_defs.h"
+
+/**
+ * struct xe_reg - Register definition
+ *
+ * Register defintion to be used by the individual register. Although the same
+ * definition is used for xe_reg and xe_reg_mcr, they use different internal
+ * APIs for accesses.
+ */
+struct xe_reg {
+ union {
+ struct {
+ /** @addr: address */
+ u32 addr:28;
+ /**
+ * @masked: register is "masked", with upper 16bits used
+ * to identify the bits that are updated on the lower
+ * bits
+ */
+ u32 masked:1;
+ /**
+ * @mcr: register is multicast/replicated in the
+ * hardware and needs special handling. Any register
+ * with this set should also use a type of xe_reg_mcr_t.
+ * It's only here so the few places that deal with MCR
+ * registers specially (xe_sr.c) and tests using the raw
+ * value can inspect it.
+ */
+ u32 mcr:1;
+ /**
+ * @ext: access MMIO extension space for current register.
+ */
+ u32 ext:1;
+ };
+ /** @raw: Raw value with both address and options */
+ u32 raw;
+ };
+};
+
+/**
+ * struct xe_reg_mcr - MCR register definition
+ *
+ * MCR register is the same as a regular register, but uses another type since
+ * the internal API used for accessing them is different: it's never correct to
+ * use regular MMIO access.
+ */
+struct xe_reg_mcr {
+ /** @__reg: The register */
+ struct xe_reg __reg;
+};
+
+
+/**
+ * XE_REG_OPTION_MASKED - Register is "masked", with upper 16 bits marking the
+ * written bits on the lower 16 bits.
+ *
+ * It only applies to registers explicitly marked in bspec with
+ * "Access: Masked". Registers with this option can have write operations to
+ * specific lower bits by setting the corresponding upper bits. Other bits will
+ * not be affected. This allows register writes without needing a RMW cycle and
+ * without caching in software the register value.
+ *
+ * Example: a write with value 0x00010001 will set bit 0 and all other bits
+ * retain their previous values.
+ *
+ * To be used with XE_REG(). XE_REG_MCR() and XE_REG_INITIALIZER()
+ */
+#define XE_REG_OPTION_MASKED .masked = 1
+
+/**
+ * XE_REG_INITIALIZER - Initializer for xe_reg_t.
+ * @r_: Register offset
+ * @...: Additional options like access mode. See struct xe_reg for available
+ * options.
+ *
+ * Register field is mandatory, and additional options may be passed as
+ * arguments. Usually ``XE_REG()`` should be preferred since it creates an
+ * object of the right type. However when initializing static const storage,
+ * where a compound statement is not allowed, this can be used instead.
+ */
+#define XE_REG_INITIALIZER(r_, ...) { .addr = r_, __VA_ARGS__ }
+
+
+/**
+ * XE_REG - Create a struct xe_reg from offset and additional flags
+ * @r_: Register offset
+ * @...: Additional options like access mode. See struct xe_reg for available
+ * options.
+ */
+#define XE_REG(r_, ...) ((const struct xe_reg)XE_REG_INITIALIZER(r_, ##__VA_ARGS__))
+
+/**
+ * XE_REG_EXT - Create a struct xe_reg from extension offset and additional
+ * flags
+ * @r_: Register extension offset
+ * @...: Additional options like access mode. See struct xe_reg for available
+ * options.
+ */
+#define XE_REG_EXT(r_, ...) \
+ ((const struct xe_reg)XE_REG_INITIALIZER(r_, ##__VA_ARGS__, .ext = 1))
+
+/**
+ * XE_REG_MCR - Create a struct xe_reg_mcr from offset and additional flags
+ * @r_: Register offset
+ * @...: Additional options like access mode. See struct xe_reg for available
+ * options.
+ */
+#define XE_REG_MCR(r_, ...) ((const struct xe_reg_mcr){ \
+ .__reg = XE_REG_INITIALIZER(r_, ##__VA_ARGS__, .mcr = 1) \
+ })
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_regs.h b/drivers/gpu/drm/xe/regs/xe_regs.h
new file mode 100644
index 000000000000..2c214bb9b671
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_regs.h
@@ -0,0 +1,68 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+#ifndef _XE_REGS_H_
+#define _XE_REGS_H_
+
+#include "regs/xe_reg_defs.h"
+
+#define TIMESTAMP_OVERRIDE XE_REG(0x44074)
+#define TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK REG_GENMASK(15, 12)
+#define TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK REG_GENMASK(9, 0)
+
+#define PCU_IRQ_OFFSET 0x444e0
+#define GU_MISC_IRQ_OFFSET 0x444f0
+#define GU_MISC_GSE REG_BIT(27)
+
+#define SOFTWARE_FLAGS_SPR33 XE_REG(0x4f084)
+
+#define GU_CNTL_PROTECTED XE_REG(0x10100C)
+#define DRIVERINT_FLR_DIS REG_BIT(31)
+
+#define GU_CNTL XE_REG(0x101010)
+#define LMEM_INIT REG_BIT(7)
+#define DRIVERFLR REG_BIT(31)
+
+#define GU_DEBUG XE_REG(0x101018)
+#define DRIVERFLR_STATUS REG_BIT(31)
+
+#define XEHP_CLOCK_GATE_DIS XE_REG(0x101014)
+#define SGSI_SIDECLK_DIS REG_BIT(17)
+
+#define GGC XE_REG(0x108040)
+#define GMS_MASK REG_GENMASK(15, 8)
+#define GGMS_MASK REG_GENMASK(7, 6)
+
+#define DSMBASE XE_REG(0x1080C0)
+#define BDSM_MASK REG_GENMASK64(63, 20)
+
+#define GSMBASE XE_REG(0x108100)
+
+#define STOLEN_RESERVED XE_REG(0x1082c0)
+#define WOPCM_SIZE_MASK REG_GENMASK64(9, 7)
+
+#define MTL_RP_STATE_CAP XE_REG(0x138000)
+
+#define MTL_GT_RPE_FREQUENCY XE_REG(0x13800c)
+
+#define MTL_MEDIAP_STATE_CAP XE_REG(0x138020)
+#define MTL_RPN_CAP_MASK REG_GENMASK(24, 16)
+#define MTL_RP0_CAP_MASK REG_GENMASK(8, 0)
+
+#define MTL_MPE_FREQUENCY XE_REG(0x13802c)
+#define MTL_RPE_MASK REG_GENMASK(8, 0)
+
+#define DG1_MSTR_TILE_INTR XE_REG(0x190008)
+#define DG1_MSTR_IRQ REG_BIT(31)
+#define DG1_MSTR_TILE(t) REG_BIT(t)
+
+#define GFX_MSTR_IRQ XE_REG(0x190010)
+#define MASTER_IRQ REG_BIT(31)
+#define GU_MISC_IRQ REG_BIT(29)
+#define DISPLAY_IRQ REG_BIT(16)
+#define GT_DW_IRQ(x) REG_BIT(x)
+
+#define PVC_RP_STATE_CAP XE_REG(0x281014)
+
+#endif
diff --git a/drivers/gpu/drm/xe/regs/xe_sriov_regs.h b/drivers/gpu/drm/xe/regs/xe_sriov_regs.h
new file mode 100644
index 000000000000..58a4e0fad1e1
--- /dev/null
+++ b/drivers/gpu/drm/xe/regs/xe_sriov_regs.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _REGS_XE_SRIOV_REGS_H_
+#define _REGS_XE_SRIOV_REGS_H_
+
+#include "regs/xe_reg_defs.h"
+
+#define XE2_LMEM_CFG XE_REG(0x48b0)
+
+#define LMEM_CFG XE_REG(0xcf58)
+#define LMEM_EN REG_BIT(31)
+#define LMTT_DIR_PTR REG_GENMASK(30, 0) /* in multiples of 64KB */
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/Makefile b/drivers/gpu/drm/xe/tests/Makefile
new file mode 100644
index 000000000000..39d8a0892274
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/Makefile
@@ -0,0 +1,10 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_DRM_XE_KUNIT_TEST) += \
+ xe_bo_test.o \
+ xe_dma_buf_test.o \
+ xe_migrate_test.o \
+ xe_mocs_test.o \
+ xe_pci_test.o \
+ xe_rtp_test.o \
+ xe_wa_test.o
diff --git a/drivers/gpu/drm/xe/tests/xe_bo.c b/drivers/gpu/drm/xe/tests/xe_bo.c
new file mode 100644
index 000000000000..412b2e7ce40c
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_bo.c
@@ -0,0 +1,353 @@
+// SPDX-License-Identifier: GPL-2.0 AND MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include <kunit/test.h>
+#include <kunit/visibility.h>
+
+#include "tests/xe_bo_test.h"
+#include "tests/xe_pci_test.h"
+#include "tests/xe_test.h"
+
+#include "xe_bo_evict.h"
+#include "xe_pci.h"
+#include "xe_pm.h"
+
+static int ccs_test_migrate(struct xe_tile *tile, struct xe_bo *bo,
+ bool clear, u64 get_val, u64 assign_val,
+ struct kunit *test)
+{
+ struct dma_fence *fence;
+ struct ttm_tt *ttm;
+ struct page *page;
+ pgoff_t ccs_page;
+ long timeout;
+ u64 *cpu_map;
+ int ret;
+ u32 offset;
+
+ /* Move bo to VRAM if not already there. */
+ ret = xe_bo_validate(bo, NULL, false);
+ if (ret) {
+ KUNIT_FAIL(test, "Failed to validate bo.\n");
+ return ret;
+ }
+
+ /* Optionally clear bo *and* CCS data in VRAM. */
+ if (clear) {
+ fence = xe_migrate_clear(tile->migrate, bo, bo->ttm.resource);
+ if (IS_ERR(fence)) {
+ KUNIT_FAIL(test, "Failed to submit bo clear.\n");
+ return PTR_ERR(fence);
+ }
+ dma_fence_put(fence);
+ }
+
+ /* Evict to system. CCS data should be copied. */
+ ret = xe_bo_evict(bo, true);
+ if (ret) {
+ KUNIT_FAIL(test, "Failed to evict bo.\n");
+ return ret;
+ }
+
+ /* Sync all migration blits */
+ timeout = dma_resv_wait_timeout(bo->ttm.base.resv,
+ DMA_RESV_USAGE_KERNEL,
+ true,
+ 5 * HZ);
+ if (timeout <= 0) {
+ KUNIT_FAIL(test, "Failed to sync bo eviction.\n");
+ return -ETIME;
+ }
+
+ /*
+ * Bo with CCS data is now in system memory. Verify backing store
+ * and data integrity. Then assign for the next testing round while
+ * we still have a CPU map.
+ */
+ ttm = bo->ttm.ttm;
+ if (!ttm || !ttm_tt_is_populated(ttm)) {
+ KUNIT_FAIL(test, "Bo was not in expected placement.\n");
+ return -EINVAL;
+ }
+
+ ccs_page = xe_bo_ccs_pages_start(bo) >> PAGE_SHIFT;
+ if (ccs_page >= ttm->num_pages) {
+ KUNIT_FAIL(test, "No TTM CCS pages present.\n");
+ return -EINVAL;
+ }
+
+ page = ttm->pages[ccs_page];
+ cpu_map = kmap_local_page(page);
+
+ /* Check first CCS value */
+ if (cpu_map[0] != get_val) {
+ KUNIT_FAIL(test,
+ "Expected CCS readout 0x%016llx, got 0x%016llx.\n",
+ (unsigned long long)get_val,
+ (unsigned long long)cpu_map[0]);
+ ret = -EINVAL;
+ }
+
+ /* Check last CCS value, or at least last value in page. */
+ offset = xe_device_ccs_bytes(tile_to_xe(tile), bo->size);
+ offset = min_t(u32, offset, PAGE_SIZE) / sizeof(u64) - 1;
+ if (cpu_map[offset] != get_val) {
+ KUNIT_FAIL(test,
+ "Expected CCS readout 0x%016llx, got 0x%016llx.\n",
+ (unsigned long long)get_val,
+ (unsigned long long)cpu_map[offset]);
+ ret = -EINVAL;
+ }
+
+ cpu_map[0] = assign_val;
+ cpu_map[offset] = assign_val;
+ kunmap_local(cpu_map);
+
+ return ret;
+}
+
+static void ccs_test_run_tile(struct xe_device *xe, struct xe_tile *tile,
+ struct kunit *test)
+{
+ struct xe_bo *bo;
+
+ int ret;
+
+ /* TODO: Sanity check */
+ unsigned int bo_flags = XE_BO_CREATE_VRAM_IF_DGFX(tile);
+
+ if (IS_DGFX(xe))
+ kunit_info(test, "Testing vram id %u\n", tile->id);
+ else
+ kunit_info(test, "Testing system memory\n");
+
+ bo = xe_bo_create_user(xe, NULL, NULL, SZ_1M, DRM_XE_GEM_CPU_CACHING_WC,
+ ttm_bo_type_device, bo_flags);
+
+ xe_bo_lock(bo, false);
+
+ if (IS_ERR(bo)) {
+ KUNIT_FAIL(test, "Failed to create bo.\n");
+ return;
+ }
+
+ kunit_info(test, "Verifying that CCS data is cleared on creation.\n");
+ ret = ccs_test_migrate(tile, bo, false, 0ULL, 0xdeadbeefdeadbeefULL,
+ test);
+ if (ret)
+ goto out_unlock;
+
+ kunit_info(test, "Verifying that CCS data survives migration.\n");
+ ret = ccs_test_migrate(tile, bo, false, 0xdeadbeefdeadbeefULL,
+ 0xdeadbeefdeadbeefULL, test);
+ if (ret)
+ goto out_unlock;
+
+ kunit_info(test, "Verifying that CCS data can be properly cleared.\n");
+ ret = ccs_test_migrate(tile, bo, true, 0ULL, 0ULL, test);
+
+out_unlock:
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+}
+
+static int ccs_test_run_device(struct xe_device *xe)
+{
+ struct kunit *test = xe_cur_kunit();
+ struct xe_tile *tile;
+ int id;
+
+ if (!xe_device_has_flat_ccs(xe)) {
+ kunit_info(test, "Skipping non-flat-ccs device.\n");
+ return 0;
+ }
+
+ xe_device_mem_access_get(xe);
+
+ for_each_tile(tile, xe, id) {
+ /* For igfx run only for primary tile */
+ if (!IS_DGFX(xe) && id > 0)
+ continue;
+ ccs_test_run_tile(xe, tile, test);
+ }
+
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+void xe_ccs_migrate_kunit(struct kunit *test)
+{
+ xe_call_for_each_device(ccs_test_run_device);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_ccs_migrate_kunit);
+
+static int evict_test_run_tile(struct xe_device *xe, struct xe_tile *tile, struct kunit *test)
+{
+ struct xe_bo *bo, *external;
+ unsigned int bo_flags = XE_BO_CREATE_VRAM_IF_DGFX(tile);
+ struct xe_vm *vm = xe_migrate_get_vm(xe_device_get_root_tile(xe)->migrate);
+ struct xe_gt *__gt;
+ int err, i, id;
+
+ kunit_info(test, "Testing device %s vram id %u\n",
+ dev_name(xe->drm.dev), tile->id);
+
+ for (i = 0; i < 2; ++i) {
+ xe_vm_lock(vm, false);
+ bo = xe_bo_create_user(xe, NULL, vm, 0x10000,
+ DRM_XE_GEM_CPU_CACHING_WC,
+ ttm_bo_type_device,
+ bo_flags);
+ xe_vm_unlock(vm);
+ if (IS_ERR(bo)) {
+ KUNIT_FAIL(test, "bo create err=%pe\n", bo);
+ break;
+ }
+
+ external = xe_bo_create_user(xe, NULL, NULL, 0x10000,
+ DRM_XE_GEM_CPU_CACHING_WC,
+ ttm_bo_type_device, bo_flags);
+ if (IS_ERR(external)) {
+ KUNIT_FAIL(test, "external bo create err=%pe\n", external);
+ goto cleanup_bo;
+ }
+
+ xe_bo_lock(external, false);
+ err = xe_bo_pin_external(external);
+ xe_bo_unlock(external);
+ if (err) {
+ KUNIT_FAIL(test, "external bo pin err=%pe\n",
+ ERR_PTR(err));
+ goto cleanup_external;
+ }
+
+ err = xe_bo_evict_all(xe);
+ if (err) {
+ KUNIT_FAIL(test, "evict err=%pe\n", ERR_PTR(err));
+ goto cleanup_all;
+ }
+
+ for_each_gt(__gt, xe, id)
+ xe_gt_sanitize(__gt);
+ err = xe_bo_restore_kernel(xe);
+ /*
+ * Snapshotting the CTB and copying back a potentially old
+ * version seems risky, depending on what might have been
+ * inflight. Also it seems snapshotting the ADS object and
+ * copying back results in serious breakage. Normally when
+ * calling xe_bo_restore_kernel() we always fully restart the
+ * GT, which re-intializes such things. We could potentially
+ * skip saving and restoring such objects in xe_bo_evict_all()
+ * however seems quite fragile not to also restart the GT. Try
+ * to do that here by triggering a GT reset.
+ */
+ for_each_gt(__gt, xe, id) {
+ xe_gt_reset_async(__gt);
+ flush_work(&__gt->reset.worker);
+ }
+ if (err) {
+ KUNIT_FAIL(test, "restore kernel err=%pe\n",
+ ERR_PTR(err));
+ goto cleanup_all;
+ }
+
+ err = xe_bo_restore_user(xe);
+ if (err) {
+ KUNIT_FAIL(test, "restore user err=%pe\n", ERR_PTR(err));
+ goto cleanup_all;
+ }
+
+ if (!xe_bo_is_vram(external)) {
+ KUNIT_FAIL(test, "external bo is not vram\n");
+ err = -EPROTO;
+ goto cleanup_all;
+ }
+
+ if (xe_bo_is_vram(bo)) {
+ KUNIT_FAIL(test, "bo is vram\n");
+ err = -EPROTO;
+ goto cleanup_all;
+ }
+
+ if (i) {
+ down_read(&vm->lock);
+ xe_vm_lock(vm, false);
+ err = xe_bo_validate(bo, bo->vm, false);
+ xe_vm_unlock(vm);
+ up_read(&vm->lock);
+ if (err) {
+ KUNIT_FAIL(test, "bo valid err=%pe\n",
+ ERR_PTR(err));
+ goto cleanup_all;
+ }
+ xe_bo_lock(external, false);
+ err = xe_bo_validate(external, NULL, false);
+ xe_bo_unlock(external);
+ if (err) {
+ KUNIT_FAIL(test, "external bo valid err=%pe\n",
+ ERR_PTR(err));
+ goto cleanup_all;
+ }
+ }
+
+ xe_bo_lock(external, false);
+ xe_bo_unpin_external(external);
+ xe_bo_unlock(external);
+
+ xe_bo_put(external);
+
+ xe_bo_lock(bo, false);
+ __xe_bo_unset_bulk_move(bo);
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+ continue;
+
+cleanup_all:
+ xe_bo_lock(external, false);
+ xe_bo_unpin_external(external);
+ xe_bo_unlock(external);
+cleanup_external:
+ xe_bo_put(external);
+cleanup_bo:
+ xe_bo_lock(bo, false);
+ __xe_bo_unset_bulk_move(bo);
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+ break;
+ }
+
+ xe_vm_put(vm);
+
+ return 0;
+}
+
+static int evict_test_run_device(struct xe_device *xe)
+{
+ struct kunit *test = xe_cur_kunit();
+ struct xe_tile *tile;
+ int id;
+
+ if (!IS_DGFX(xe)) {
+ kunit_info(test, "Skipping non-discrete device %s.\n",
+ dev_name(xe->drm.dev));
+ return 0;
+ }
+
+ xe_device_mem_access_get(xe);
+
+ for_each_tile(tile, xe, id)
+ evict_test_run_tile(xe, tile, test);
+
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+void xe_bo_evict_kunit(struct kunit *test)
+{
+ xe_call_for_each_device(evict_test_run_device);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_bo_evict_kunit);
diff --git a/drivers/gpu/drm/xe/tests/xe_bo_test.c b/drivers/gpu/drm/xe/tests/xe_bo_test.c
new file mode 100644
index 000000000000..f408f17f2164
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_bo_test.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_bo_test.h"
+
+#include <kunit/test.h>
+
+static struct kunit_case xe_bo_tests[] = {
+ KUNIT_CASE(xe_ccs_migrate_kunit),
+ KUNIT_CASE(xe_bo_evict_kunit),
+ {}
+};
+
+static struct kunit_suite xe_bo_test_suite = {
+ .name = "xe_bo",
+ .test_cases = xe_bo_tests,
+};
+
+kunit_test_suite(xe_bo_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_bo kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_bo_test.h b/drivers/gpu/drm/xe/tests/xe_bo_test.h
new file mode 100644
index 000000000000..0113ab45066a
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_bo_test.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_BO_TEST_H_
+#define _XE_BO_TEST_H_
+
+struct kunit;
+
+void xe_ccs_migrate_kunit(struct kunit *test);
+void xe_bo_evict_kunit(struct kunit *test);
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_dma_buf.c b/drivers/gpu/drm/xe/tests/xe_dma_buf.c
new file mode 100644
index 000000000000..9f6d571d7fa9
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_dma_buf.c
@@ -0,0 +1,278 @@
+// SPDX-License-Identifier: GPL-2.0 AND MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include <drm/xe_drm.h>
+
+#include <kunit/test.h>
+#include <kunit/visibility.h>
+
+#include "tests/xe_dma_buf_test.h"
+#include "tests/xe_pci_test.h"
+
+#include "xe_pci.h"
+
+static bool p2p_enabled(struct dma_buf_test_params *params)
+{
+ return IS_ENABLED(CONFIG_PCI_P2PDMA) && params->attach_ops &&
+ params->attach_ops->allow_peer2peer;
+}
+
+static bool is_dynamic(struct dma_buf_test_params *params)
+{
+ return IS_ENABLED(CONFIG_DMABUF_MOVE_NOTIFY) && params->attach_ops &&
+ params->attach_ops->move_notify;
+}
+
+static void check_residency(struct kunit *test, struct xe_bo *exported,
+ struct xe_bo *imported, struct dma_buf *dmabuf)
+{
+ struct dma_buf_test_params *params = to_dma_buf_test_params(test->priv);
+ u32 mem_type;
+ int ret;
+
+ xe_bo_assert_held(exported);
+ xe_bo_assert_held(imported);
+
+ mem_type = XE_PL_VRAM0;
+ if (!(params->mem_mask & XE_BO_CREATE_VRAM0_BIT))
+ /* No VRAM allowed */
+ mem_type = XE_PL_TT;
+ else if (params->force_different_devices && !p2p_enabled(params))
+ /* No P2P */
+ mem_type = XE_PL_TT;
+ else if (params->force_different_devices && !is_dynamic(params) &&
+ (params->mem_mask & XE_BO_CREATE_SYSTEM_BIT))
+ /* Pin migrated to TT */
+ mem_type = XE_PL_TT;
+
+ if (!xe_bo_is_mem_type(exported, mem_type)) {
+ KUNIT_FAIL(test, "Exported bo was not in expected memory type.\n");
+ return;
+ }
+
+ if (xe_bo_is_pinned(exported))
+ return;
+
+ /*
+ * Evict exporter. Note that the gem object dma_buf member isn't
+ * set from xe_gem_prime_export(), and it's needed for the move_notify()
+ * functionality, so hack that up here. Evicting the exported bo will
+ * evict also the imported bo through the move_notify() functionality if
+ * importer is on a different device. If they're on the same device,
+ * the exporter and the importer should be the same bo.
+ */
+ swap(exported->ttm.base.dma_buf, dmabuf);
+ ret = xe_bo_evict(exported, true);
+ swap(exported->ttm.base.dma_buf, dmabuf);
+ if (ret) {
+ if (ret != -EINTR && ret != -ERESTARTSYS)
+ KUNIT_FAIL(test, "Evicting exporter failed with err=%d.\n",
+ ret);
+ return;
+ }
+
+ /* Verify that also importer has been evicted to SYSTEM */
+ if (exported != imported && !xe_bo_is_mem_type(imported, XE_PL_SYSTEM)) {
+ KUNIT_FAIL(test, "Importer wasn't properly evicted.\n");
+ return;
+ }
+
+ /* Re-validate the importer. This should move also exporter in. */
+ ret = xe_bo_validate(imported, NULL, false);
+ if (ret) {
+ if (ret != -EINTR && ret != -ERESTARTSYS)
+ KUNIT_FAIL(test, "Validating importer failed with err=%d.\n",
+ ret);
+ return;
+ }
+
+ /*
+ * If on different devices, the exporter is kept in system if
+ * possible, saving a migration step as the transfer is just
+ * likely as fast from system memory.
+ */
+ if (params->mem_mask & XE_BO_CREATE_SYSTEM_BIT)
+ KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(exported, XE_PL_TT));
+ else
+ KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(exported, mem_type));
+
+ if (params->force_different_devices)
+ KUNIT_EXPECT_TRUE(test, xe_bo_is_mem_type(imported, XE_PL_TT));
+ else
+ KUNIT_EXPECT_TRUE(test, exported == imported);
+}
+
+static void xe_test_dmabuf_import_same_driver(struct xe_device *xe)
+{
+ struct kunit *test = xe_cur_kunit();
+ struct dma_buf_test_params *params = to_dma_buf_test_params(test->priv);
+ struct drm_gem_object *import;
+ struct dma_buf *dmabuf;
+ struct xe_bo *bo;
+ size_t size;
+
+ /* No VRAM on this device? */
+ if (!ttm_manager_type(&xe->ttm, XE_PL_VRAM0) &&
+ (params->mem_mask & XE_BO_CREATE_VRAM0_BIT))
+ return;
+
+ size = PAGE_SIZE;
+ if ((params->mem_mask & XE_BO_CREATE_VRAM0_BIT) &&
+ xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)
+ size = SZ_64K;
+
+ kunit_info(test, "running %s\n", __func__);
+ bo = xe_bo_create_user(xe, NULL, NULL, size, DRM_XE_GEM_CPU_CACHING_WC,
+ ttm_bo_type_device, XE_BO_CREATE_USER_BIT | params->mem_mask);
+ if (IS_ERR(bo)) {
+ KUNIT_FAIL(test, "xe_bo_create() failed with err=%ld\n",
+ PTR_ERR(bo));
+ return;
+ }
+
+ dmabuf = xe_gem_prime_export(&bo->ttm.base, 0);
+ if (IS_ERR(dmabuf)) {
+ KUNIT_FAIL(test, "xe_gem_prime_export() failed with err=%ld\n",
+ PTR_ERR(dmabuf));
+ goto out;
+ }
+
+ import = xe_gem_prime_import(&xe->drm, dmabuf);
+ if (!IS_ERR(import)) {
+ struct xe_bo *import_bo = gem_to_xe_bo(import);
+
+ /*
+ * Did import succeed when it shouldn't due to lack of p2p support?
+ */
+ if (params->force_different_devices &&
+ !p2p_enabled(params) &&
+ !(params->mem_mask & XE_BO_CREATE_SYSTEM_BIT)) {
+ KUNIT_FAIL(test,
+ "xe_gem_prime_import() succeeded when it shouldn't have\n");
+ } else {
+ int err;
+
+ /* Is everything where we expect it to be? */
+ xe_bo_lock(import_bo, false);
+ err = xe_bo_validate(import_bo, NULL, false);
+
+ /* Pinning in VRAM is not allowed. */
+ if (!is_dynamic(params) &&
+ params->force_different_devices &&
+ !(params->mem_mask & XE_BO_CREATE_SYSTEM_BIT))
+ KUNIT_EXPECT_EQ(test, err, -EINVAL);
+ /* Otherwise only expect interrupts or success. */
+ else if (err && err != -EINTR && err != -ERESTARTSYS)
+ KUNIT_EXPECT_TRUE(test, !err || err == -EINTR ||
+ err == -ERESTARTSYS);
+
+ if (!err)
+ check_residency(test, bo, import_bo, dmabuf);
+ xe_bo_unlock(import_bo);
+ }
+ drm_gem_object_put(import);
+ } else if (PTR_ERR(import) != -EOPNOTSUPP) {
+ /* Unexpected error code. */
+ KUNIT_FAIL(test,
+ "xe_gem_prime_import failed with the wrong err=%ld\n",
+ PTR_ERR(import));
+ } else if (!params->force_different_devices ||
+ p2p_enabled(params) ||
+ (params->mem_mask & XE_BO_CREATE_SYSTEM_BIT)) {
+ /* Shouldn't fail if we can reuse same bo, use p2p or use system */
+ KUNIT_FAIL(test, "dynamic p2p attachment failed with err=%ld\n",
+ PTR_ERR(import));
+ }
+ dma_buf_put(dmabuf);
+out:
+ drm_gem_object_put(&bo->ttm.base);
+}
+
+static const struct dma_buf_attach_ops nop2p_attach_ops = {
+ .allow_peer2peer = false,
+ .move_notify = xe_dma_buf_move_notify
+};
+
+/*
+ * We test the implementation with bos of different residency and with
+ * importers with different capabilities; some lacking p2p support and some
+ * lacking dynamic capabilities (attach_ops == NULL). We also fake
+ * different devices avoiding the import shortcut that just reuses the same
+ * gem object.
+ */
+static const struct dma_buf_test_params test_params[] = {
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops},
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &nop2p_attach_ops},
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &nop2p_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT},
+ {.mem_mask = XE_BO_CREATE_VRAM0_BIT,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT,
+ .attach_ops = &nop2p_attach_ops},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT,
+ .attach_ops = &nop2p_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &xe_dma_buf_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &nop2p_attach_ops},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT,
+ .attach_ops = &nop2p_attach_ops,
+ .force_different_devices = true},
+
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT},
+ {.mem_mask = XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_VRAM0_BIT,
+ .force_different_devices = true},
+
+ {}
+};
+
+static int dma_buf_run_device(struct xe_device *xe)
+{
+ const struct dma_buf_test_params *params;
+ struct kunit *test = xe_cur_kunit();
+
+ for (params = test_params; params->mem_mask; ++params) {
+ struct dma_buf_test_params p = *params;
+
+ p.base.id = XE_TEST_LIVE_DMA_BUF;
+ test->priv = &p;
+ xe_test_dmabuf_import_same_driver(xe);
+ }
+
+ /* A non-zero return would halt iteration over driver devices */
+ return 0;
+}
+
+void xe_dma_buf_kunit(struct kunit *test)
+{
+ xe_call_for_each_device(dma_buf_run_device);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_dma_buf_kunit);
diff --git a/drivers/gpu/drm/xe/tests/xe_dma_buf_test.c b/drivers/gpu/drm/xe/tests/xe_dma_buf_test.c
new file mode 100644
index 000000000000..9f5a9cda8c0f
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_dma_buf_test.c
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_dma_buf_test.h"
+
+#include <kunit/test.h>
+
+static struct kunit_case xe_dma_buf_tests[] = {
+ KUNIT_CASE(xe_dma_buf_kunit),
+ {}
+};
+
+static struct kunit_suite xe_dma_buf_test_suite = {
+ .name = "xe_dma_buf",
+ .test_cases = xe_dma_buf_tests,
+};
+
+kunit_test_suite(xe_dma_buf_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_dma_buf kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_dma_buf_test.h b/drivers/gpu/drm/xe/tests/xe_dma_buf_test.h
new file mode 100644
index 000000000000..e6b464ddd526
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_dma_buf_test.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DMA_BUF_TEST_H_
+#define _XE_DMA_BUF_TEST_H_
+
+struct kunit;
+
+void xe_dma_buf_kunit(struct kunit *test);
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_lmtt_test.c b/drivers/gpu/drm/xe/tests/xe_lmtt_test.c
new file mode 100644
index 000000000000..1f1557c45ae1
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_lmtt_test.c
@@ -0,0 +1,73 @@
+// SPDX-License-Identifier: GPL-2.0 AND MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <kunit/test.h>
+
+static const struct lmtt_ops_param {
+ const char *desc;
+ const struct xe_lmtt_ops *ops;
+} lmtt_ops_params[] = {
+ { "2-level", &lmtt_2l_ops, },
+ { "multi-level", &lmtt_ml_ops, },
+};
+
+static void lmtt_ops_param_get_desc(const struct lmtt_ops_param *p, char *desc)
+{
+ snprintf(desc, KUNIT_PARAM_DESC_SIZE, "%s", p->desc);
+}
+
+KUNIT_ARRAY_PARAM(lmtt_ops, lmtt_ops_params, lmtt_ops_param_get_desc);
+
+static void test_ops(struct kunit *test)
+{
+ const struct lmtt_ops_param *p = test->param_value;
+ const struct xe_lmtt_ops *ops = p->ops;
+ unsigned int n;
+
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_root_pd_level);
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_pte_num);
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_pte_size);
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_pte_shift);
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_pte_index);
+ KUNIT_ASSERT_NOT_NULL(test, ops->lmtt_pte_encode);
+
+ KUNIT_EXPECT_NE(test, ops->lmtt_root_pd_level(), 0);
+
+ for (n = 0; n <= ops->lmtt_root_pd_level(); n++) {
+ KUNIT_EXPECT_NE_MSG(test, ops->lmtt_pte_num(n), 0,
+ "level=%u", n);
+ KUNIT_EXPECT_NE_MSG(test, ops->lmtt_pte_size(n), 0,
+ "level=%u", n);
+ KUNIT_EXPECT_NE_MSG(test, ops->lmtt_pte_encode(0, n), LMTT_PTE_INVALID,
+ "level=%u", n);
+ }
+
+ for (n = 0; n < ops->lmtt_root_pd_level(); n++) {
+ u64 addr = BIT_ULL(ops->lmtt_pte_shift(n));
+
+ KUNIT_EXPECT_NE_MSG(test, ops->lmtt_pte_shift(n), 0,
+ "level=%u", n);
+ KUNIT_EXPECT_EQ_MSG(test, ops->lmtt_pte_index(addr - 1, n), 0,
+ "addr=%#llx level=%u", addr, n);
+ KUNIT_EXPECT_EQ_MSG(test, ops->lmtt_pte_index(addr + 1, n), 1,
+ "addr=%#llx level=%u", addr, n);
+ KUNIT_EXPECT_EQ_MSG(test, ops->lmtt_pte_index(addr * 2 - 1, n), 1,
+ "addr=%#llx level=%u", addr, n);
+ KUNIT_EXPECT_EQ_MSG(test, ops->lmtt_pte_index(addr * 2, n), 2,
+ "addr=%#llx level=%u", addr, n);
+ }
+}
+
+static struct kunit_case lmtt_test_cases[] = {
+ KUNIT_CASE_PARAM(test_ops, lmtt_ops_gen_params),
+ {}
+};
+
+static struct kunit_suite lmtt_suite = {
+ .name = "lmtt",
+ .test_cases = lmtt_test_cases,
+};
+
+kunit_test_suites(&lmtt_suite);
diff --git a/drivers/gpu/drm/xe/tests/xe_migrate.c b/drivers/gpu/drm/xe/tests/xe_migrate.c
new file mode 100644
index 000000000000..7a32faa2f688
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_migrate.c
@@ -0,0 +1,444 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020-2022 Intel Corporation
+ */
+
+#include <kunit/test.h>
+#include <kunit/visibility.h>
+
+#include "tests/xe_migrate_test.h"
+#include "tests/xe_pci_test.h"
+
+#include "xe_pci.h"
+
+static bool sanity_fence_failed(struct xe_device *xe, struct dma_fence *fence,
+ const char *str, struct kunit *test)
+{
+ long ret;
+
+ if (IS_ERR(fence)) {
+ KUNIT_FAIL(test, "Failed to create fence for %s: %li\n", str,
+ PTR_ERR(fence));
+ return true;
+ }
+ if (!fence)
+ return true;
+
+ ret = dma_fence_wait_timeout(fence, false, 5 * HZ);
+ if (ret <= 0) {
+ KUNIT_FAIL(test, "Fence timed out for %s: %li\n", str, ret);
+ return true;
+ }
+
+ return false;
+}
+
+static int run_sanity_job(struct xe_migrate *m, struct xe_device *xe,
+ struct xe_bb *bb, u32 second_idx, const char *str,
+ struct kunit *test)
+{
+ u64 batch_base = xe_migrate_batch_base(m, xe->info.has_usm);
+ struct xe_sched_job *job = xe_bb_create_migration_job(m->q, bb,
+ batch_base,
+ second_idx);
+ struct dma_fence *fence;
+
+ if (IS_ERR(job)) {
+ KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n",
+ PTR_ERR(job));
+ return PTR_ERR(job);
+ }
+
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ if (sanity_fence_failed(xe, fence, str, test))
+ return -ETIMEDOUT;
+
+ dma_fence_put(fence);
+ kunit_info(test, "%s: Job completed\n", str);
+ return 0;
+}
+
+static void
+sanity_populate_cb(struct xe_migrate_pt_update *pt_update,
+ struct xe_tile *tile, struct iosys_map *map, void *dst,
+ u32 qword_ofs, u32 num_qwords,
+ const struct xe_vm_pgtable_update *update)
+{
+ struct migrate_test_params *p =
+ to_migrate_test_params(xe_cur_kunit_priv(XE_TEST_LIVE_MIGRATE));
+ int i;
+ u64 *ptr = dst;
+ u64 value;
+
+ for (i = 0; i < num_qwords; i++) {
+ value = (qword_ofs + i - update->ofs) * 0x1111111111111111ULL;
+ if (map)
+ xe_map_wr(tile_to_xe(tile), map, (qword_ofs + i) *
+ sizeof(u64), u64, value);
+ else
+ ptr[i] = value;
+ }
+
+ kunit_info(xe_cur_kunit(), "Used %s.\n", map ? "CPU" : "GPU");
+ if (p->force_gpu && map)
+ KUNIT_FAIL(xe_cur_kunit(), "GPU pagetable update used CPU.\n");
+}
+
+static const struct xe_migrate_pt_update_ops sanity_ops = {
+ .populate = sanity_populate_cb,
+};
+
+#define check(_retval, _expected, str, _test) \
+ do { if ((_retval) != (_expected)) { \
+ KUNIT_FAIL(_test, "Sanity check failed: " str \
+ " expected %llx, got %llx\n", \
+ (u64)(_expected), (u64)(_retval)); \
+ } } while (0)
+
+static void test_copy(struct xe_migrate *m, struct xe_bo *bo,
+ struct kunit *test, u32 region)
+{
+ struct xe_device *xe = tile_to_xe(m->tile);
+ u64 retval, expected = 0;
+ bool big = bo->size >= SZ_2M;
+ struct dma_fence *fence;
+ const char *str = big ? "Copying big bo" : "Copying small bo";
+ int err;
+
+ struct xe_bo *remote = xe_bo_create_locked(xe, m->tile, NULL,
+ bo->size,
+ ttm_bo_type_kernel,
+ region |
+ XE_BO_NEEDS_CPU_ACCESS);
+ if (IS_ERR(remote)) {
+ KUNIT_FAIL(test, "Failed to allocate remote bo for %s: %li\n",
+ str, PTR_ERR(remote));
+ return;
+ }
+
+ err = xe_bo_validate(remote, NULL, false);
+ if (err) {
+ KUNIT_FAIL(test, "Failed to validate system bo for %s: %li\n",
+ str, err);
+ goto out_unlock;
+ }
+
+ err = xe_bo_vmap(remote);
+ if (err) {
+ KUNIT_FAIL(test, "Failed to vmap system bo for %s: %li\n",
+ str, err);
+ goto out_unlock;
+ }
+
+ xe_map_memset(xe, &remote->vmap, 0, 0xd0, remote->size);
+ fence = xe_migrate_clear(m, remote, remote->ttm.resource);
+ if (!sanity_fence_failed(xe, fence, big ? "Clearing remote big bo" :
+ "Clearing remote small bo", test)) {
+ retval = xe_map_rd(xe, &remote->vmap, 0, u64);
+ check(retval, expected, "remote first offset should be cleared",
+ test);
+ retval = xe_map_rd(xe, &remote->vmap, remote->size - 8, u64);
+ check(retval, expected, "remote last offset should be cleared",
+ test);
+ }
+ dma_fence_put(fence);
+
+ /* Try to copy 0xc0 from remote to vram with 2MB or 64KiB/4KiB pages */
+ xe_map_memset(xe, &remote->vmap, 0, 0xc0, remote->size);
+ xe_map_memset(xe, &bo->vmap, 0, 0xd0, bo->size);
+
+ expected = 0xc0c0c0c0c0c0c0c0;
+ fence = xe_migrate_copy(m, remote, bo, remote->ttm.resource,
+ bo->ttm.resource, false);
+ if (!sanity_fence_failed(xe, fence, big ? "Copying big bo remote -> vram" :
+ "Copying small bo remote -> vram", test)) {
+ retval = xe_map_rd(xe, &bo->vmap, 0, u64);
+ check(retval, expected,
+ "remote -> vram bo first offset should be copied", test);
+ retval = xe_map_rd(xe, &bo->vmap, bo->size - 8, u64);
+ check(retval, expected,
+ "remote -> vram bo offset should be copied", test);
+ }
+ dma_fence_put(fence);
+
+ /* And other way around.. slightly hacky.. */
+ xe_map_memset(xe, &remote->vmap, 0, 0xd0, remote->size);
+ xe_map_memset(xe, &bo->vmap, 0, 0xc0, bo->size);
+
+ fence = xe_migrate_copy(m, bo, remote, bo->ttm.resource,
+ remote->ttm.resource, false);
+ if (!sanity_fence_failed(xe, fence, big ? "Copying big bo vram -> remote" :
+ "Copying small bo vram -> remote", test)) {
+ retval = xe_map_rd(xe, &remote->vmap, 0, u64);
+ check(retval, expected,
+ "vram -> remote bo first offset should be copied", test);
+ retval = xe_map_rd(xe, &remote->vmap, bo->size - 8, u64);
+ check(retval, expected,
+ "vram -> remote bo last offset should be copied", test);
+ }
+ dma_fence_put(fence);
+
+ xe_bo_vunmap(remote);
+out_unlock:
+ xe_bo_unlock(remote);
+ xe_bo_put(remote);
+}
+
+static void test_copy_sysmem(struct xe_migrate *m, struct xe_bo *bo,
+ struct kunit *test)
+{
+ test_copy(m, bo, test, XE_BO_CREATE_SYSTEM_BIT);
+}
+
+static void test_copy_vram(struct xe_migrate *m, struct xe_bo *bo,
+ struct kunit *test)
+{
+ u32 region;
+
+ if (bo->ttm.resource->mem_type == XE_PL_SYSTEM)
+ return;
+
+ if (bo->ttm.resource->mem_type == XE_PL_VRAM0)
+ region = XE_BO_CREATE_VRAM1_BIT;
+ else
+ region = XE_BO_CREATE_VRAM0_BIT;
+ test_copy(m, bo, test, region);
+}
+
+static void test_pt_update(struct xe_migrate *m, struct xe_bo *pt,
+ struct kunit *test, bool force_gpu)
+{
+ struct xe_device *xe = tile_to_xe(m->tile);
+ struct dma_fence *fence;
+ u64 retval, expected;
+ ktime_t then, now;
+ int i;
+
+ struct xe_vm_pgtable_update update = {
+ .ofs = 1,
+ .qwords = 0x10,
+ .pt_bo = pt,
+ };
+ struct xe_migrate_pt_update pt_update = {
+ .ops = &sanity_ops,
+ };
+ struct migrate_test_params p = {
+ .base.id = XE_TEST_LIVE_MIGRATE,
+ .force_gpu = force_gpu,
+ };
+
+ test->priv = &p;
+ /* Test xe_migrate_update_pgtables() updates the pagetable as expected */
+ expected = 0xf0f0f0f0f0f0f0f0ULL;
+ xe_map_memset(xe, &pt->vmap, 0, (u8)expected, pt->size);
+
+ then = ktime_get();
+ fence = xe_migrate_update_pgtables(m, m->q->vm, NULL, m->q, &update, 1,
+ NULL, 0, &pt_update);
+ now = ktime_get();
+ if (sanity_fence_failed(xe, fence, "Migration pagetable update", test))
+ return;
+
+ kunit_info(test, "Updating without syncing took %llu us,\n",
+ (unsigned long long)ktime_to_us(ktime_sub(now, then)));
+
+ dma_fence_put(fence);
+ retval = xe_map_rd(xe, &pt->vmap, 0, u64);
+ check(retval, expected, "PTE[0] must stay untouched", test);
+
+ for (i = 0; i < update.qwords; i++) {
+ retval = xe_map_rd(xe, &pt->vmap, (update.ofs + i) * 8, u64);
+ check(retval, i * 0x1111111111111111ULL, "PTE update", test);
+ }
+
+ retval = xe_map_rd(xe, &pt->vmap, 8 * (update.ofs + update.qwords),
+ u64);
+ check(retval, expected, "PTE[0x11] must stay untouched", test);
+}
+
+static void xe_migrate_sanity_test(struct xe_migrate *m, struct kunit *test)
+{
+ struct xe_tile *tile = m->tile;
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_bo *pt, *bo = m->pt_bo, *big, *tiny;
+ struct xe_res_cursor src_it;
+ struct dma_fence *fence;
+ u64 retval, expected;
+ struct xe_bb *bb;
+ int err;
+ u8 id = tile->id;
+
+ err = xe_bo_vmap(bo);
+ if (err) {
+ KUNIT_FAIL(test, "Failed to vmap our pagetables: %li\n",
+ PTR_ERR(bo));
+ return;
+ }
+
+ big = xe_bo_create_pin_map(xe, tile, m->q->vm, SZ_4M,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_PINNED_BIT);
+ if (IS_ERR(big)) {
+ KUNIT_FAIL(test, "Failed to allocate bo: %li\n", PTR_ERR(big));
+ goto vunmap;
+ }
+
+ pt = xe_bo_create_pin_map(xe, tile, m->q->vm, XE_PAGE_SIZE,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_PINNED_BIT);
+ if (IS_ERR(pt)) {
+ KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n",
+ PTR_ERR(pt));
+ goto free_big;
+ }
+
+ tiny = xe_bo_create_pin_map(xe, tile, m->q->vm,
+ 2 * SZ_4K,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_PINNED_BIT);
+ if (IS_ERR(tiny)) {
+ KUNIT_FAIL(test, "Failed to allocate fake pt: %li\n",
+ PTR_ERR(pt));
+ goto free_pt;
+ }
+
+ bb = xe_bb_new(tile->primary_gt, 32, xe->info.has_usm);
+ if (IS_ERR(bb)) {
+ KUNIT_FAIL(test, "Failed to create batchbuffer: %li\n",
+ PTR_ERR(bb));
+ goto free_tiny;
+ }
+
+ kunit_info(test, "Starting tests, top level PT addr: %lx, special pagetable base addr: %lx\n",
+ (unsigned long)xe_bo_main_addr(m->q->vm->pt_root[id]->bo, XE_PAGE_SIZE),
+ (unsigned long)xe_bo_main_addr(m->pt_bo, XE_PAGE_SIZE));
+
+ /* First part of the test, are we updating our pagetable bo with a new entry? */
+ xe_map_wr(xe, &bo->vmap, XE_PAGE_SIZE * (NUM_KERNEL_PDE - 1), u64,
+ 0xdeaddeadbeefbeef);
+ expected = m->q->vm->pt_ops->pte_encode_bo(pt, 0, xe->pat.idx[XE_CACHE_WB], 0);
+ if (m->q->vm->flags & XE_VM_FLAG_64K)
+ expected |= XE_PTE_PS64;
+ if (xe_bo_is_vram(pt))
+ xe_res_first(pt->ttm.resource, 0, pt->size, &src_it);
+ else
+ xe_res_first_sg(xe_bo_sg(pt), 0, pt->size, &src_it);
+
+ emit_pte(m, bb, NUM_KERNEL_PDE - 1, xe_bo_is_vram(pt), false,
+ &src_it, XE_PAGE_SIZE, pt);
+
+ run_sanity_job(m, xe, bb, bb->len, "Writing PTE for our fake PT", test);
+
+ retval = xe_map_rd(xe, &bo->vmap, XE_PAGE_SIZE * (NUM_KERNEL_PDE - 1),
+ u64);
+ check(retval, expected, "PTE entry write", test);
+
+ /* Now try to write data to our newly mapped 'pagetable', see if it succeeds */
+ bb->len = 0;
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+ xe_map_wr(xe, &pt->vmap, 0, u32, 0xdeaddead);
+ expected = 0;
+
+ emit_clear(tile->primary_gt, bb, xe_migrate_vm_addr(NUM_KERNEL_PDE - 1, 0), 4, 4,
+ IS_DGFX(xe));
+ run_sanity_job(m, xe, bb, 1, "Writing to our newly mapped pagetable",
+ test);
+
+ retval = xe_map_rd(xe, &pt->vmap, 0, u32);
+ check(retval, expected, "Write to PT after adding PTE", test);
+
+ /* Sanity checks passed, try the full ones! */
+
+ /* Clear a small bo */
+ kunit_info(test, "Clearing small buffer object\n");
+ xe_map_memset(xe, &tiny->vmap, 0, 0x22, tiny->size);
+ expected = 0;
+ fence = xe_migrate_clear(m, tiny, tiny->ttm.resource);
+ if (sanity_fence_failed(xe, fence, "Clearing small bo", test))
+ goto out;
+
+ dma_fence_put(fence);
+ retval = xe_map_rd(xe, &tiny->vmap, 0, u32);
+ check(retval, expected, "Command clear small first value", test);
+ retval = xe_map_rd(xe, &tiny->vmap, tiny->size - 4, u32);
+ check(retval, expected, "Command clear small last value", test);
+
+ kunit_info(test, "Copying small buffer object to system\n");
+ test_copy_sysmem(m, tiny, test);
+ if (xe->info.tile_count > 1) {
+ kunit_info(test, "Copying small buffer object to other vram\n");
+ test_copy_vram(m, tiny, test);
+ }
+
+ /* Clear a big bo */
+ kunit_info(test, "Clearing big buffer object\n");
+ xe_map_memset(xe, &big->vmap, 0, 0x11, big->size);
+ expected = 0;
+ fence = xe_migrate_clear(m, big, big->ttm.resource);
+ if (sanity_fence_failed(xe, fence, "Clearing big bo", test))
+ goto out;
+
+ dma_fence_put(fence);
+ retval = xe_map_rd(xe, &big->vmap, 0, u32);
+ check(retval, expected, "Command clear big first value", test);
+ retval = xe_map_rd(xe, &big->vmap, big->size - 4, u32);
+ check(retval, expected, "Command clear big last value", test);
+
+ kunit_info(test, "Copying big buffer object to system\n");
+ test_copy_sysmem(m, big, test);
+ if (xe->info.tile_count > 1) {
+ kunit_info(test, "Copying big buffer object to other vram\n");
+ test_copy_vram(m, big, test);
+ }
+
+ kunit_info(test, "Testing page table update using CPU if GPU idle.\n");
+ test_pt_update(m, pt, test, false);
+ kunit_info(test, "Testing page table update using GPU\n");
+ test_pt_update(m, pt, test, true);
+
+out:
+ xe_bb_free(bb, NULL);
+free_tiny:
+ xe_bo_unpin(tiny);
+ xe_bo_put(tiny);
+free_pt:
+ xe_bo_unpin(pt);
+ xe_bo_put(pt);
+free_big:
+ xe_bo_unpin(big);
+ xe_bo_put(big);
+vunmap:
+ xe_bo_vunmap(m->pt_bo);
+}
+
+static int migrate_test_run_device(struct xe_device *xe)
+{
+ struct kunit *test = xe_cur_kunit();
+ struct xe_tile *tile;
+ int id;
+
+ for_each_tile(tile, xe, id) {
+ struct xe_migrate *m = tile->migrate;
+
+ kunit_info(test, "Testing tile id %d.\n", id);
+ xe_vm_lock(m->q->vm, true);
+ xe_device_mem_access_get(xe);
+ xe_migrate_sanity_test(m, test);
+ xe_device_mem_access_put(xe);
+ xe_vm_unlock(m->q->vm);
+ }
+
+ return 0;
+}
+
+void xe_migrate_sanity_kunit(struct kunit *test)
+{
+ xe_call_for_each_device(migrate_test_run_device);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_migrate_sanity_kunit);
diff --git a/drivers/gpu/drm/xe/tests/xe_migrate_test.c b/drivers/gpu/drm/xe/tests/xe_migrate_test.c
new file mode 100644
index 000000000000..cf0c173b945f
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_migrate_test.c
@@ -0,0 +1,25 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_migrate_test.h"
+
+#include <kunit/test.h>
+
+static struct kunit_case xe_migrate_tests[] = {
+ KUNIT_CASE(xe_migrate_sanity_kunit),
+ {}
+};
+
+static struct kunit_suite xe_migrate_test_suite = {
+ .name = "xe_migrate",
+ .test_cases = xe_migrate_tests,
+};
+
+kunit_test_suite(xe_migrate_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_migrate kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_migrate_test.h b/drivers/gpu/drm/xe/tests/xe_migrate_test.h
new file mode 100644
index 000000000000..7c645c66824f
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_migrate_test.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_MIGRATE_TEST_H_
+#define _XE_MIGRATE_TEST_H_
+
+struct kunit;
+
+void xe_migrate_sanity_kunit(struct kunit *test);
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_mocs.c b/drivers/gpu/drm/xe/tests/xe_mocs.c
new file mode 100644
index 000000000000..7dd34f94e809
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_mocs.c
@@ -0,0 +1,130 @@
+// SPDX-License-Identifier: GPL-2.0 AND MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include <kunit/test.h>
+#include <kunit/visibility.h>
+
+#include "tests/xe_mocs_test.h"
+#include "tests/xe_pci_test.h"
+#include "tests/xe_test.h"
+
+#include "xe_pci.h"
+#include "xe_gt.h"
+#include "xe_mocs.h"
+#include "xe_device.h"
+
+struct live_mocs {
+ struct xe_mocs_info table;
+};
+
+static int live_mocs_init(struct live_mocs *arg, struct xe_gt *gt)
+{
+ unsigned int flags;
+ struct kunit *test = xe_cur_kunit();
+
+ memset(arg, 0, sizeof(*arg));
+
+ flags = get_mocs_settings(gt_to_xe(gt), &arg->table);
+
+ kunit_info(test, "table size %d", arg->table.size);
+ kunit_info(test, "table uc_index %d", arg->table.uc_index);
+ kunit_info(test, "table n_entries %d", arg->table.n_entries);
+
+ return flags;
+}
+
+static void read_l3cc_table(struct xe_gt *gt,
+ const struct xe_mocs_info *info)
+{
+ unsigned int i;
+ u32 l3cc;
+ u32 reg_val;
+ u32 ret;
+
+ struct kunit *test = xe_cur_kunit();
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ KUNIT_ASSERT_EQ_MSG(test, ret, 0, "Forcewake Failed.\n");
+ mocs_dbg(&gt_to_xe(gt)->drm, "L3CC entries:%d\n", info->n_entries);
+ for (i = 0;
+ i < (info->n_entries + 1) / 2 ?
+ (l3cc = l3cc_combine(get_entry_l3cc(info, 2 * i),
+ get_entry_l3cc(info, 2 * i + 1))), 1 : 0;
+ i++) {
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1250)
+ reg_val = xe_gt_mcr_unicast_read_any(gt, XEHP_LNCFCMOCS(i));
+ else
+ reg_val = xe_mmio_read32(gt, XELP_LNCFCMOCS(i));
+ mocs_dbg(&gt_to_xe(gt)->drm, "%d 0x%x 0x%x 0x%x\n", i,
+ XELP_LNCFCMOCS(i).addr, reg_val, l3cc);
+ if (reg_val != l3cc)
+ KUNIT_FAIL(test, "l3cc reg 0x%x has incorrect val.\n",
+ XELP_LNCFCMOCS(i).addr);
+ }
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ xe_device_mem_access_put(gt_to_xe(gt));
+}
+
+static void read_mocs_table(struct xe_gt *gt,
+ const struct xe_mocs_info *info)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ unsigned int i;
+ u32 mocs;
+ u32 reg_val;
+ u32 ret;
+
+ struct kunit *test = xe_cur_kunit();
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ KUNIT_ASSERT_EQ_MSG(test, ret, 0, "Forcewake Failed.\n");
+ mocs_dbg(&gt_to_xe(gt)->drm, "Global MOCS entries:%d\n", info->n_entries);
+ drm_WARN_ONCE(&xe->drm, !info->unused_entries_index,
+ "Unused entries index should have been defined\n");
+ for (i = 0;
+ i < info->n_entries ? (mocs = get_entry_control(info, i)), 1 : 0;
+ i++) {
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1250)
+ reg_val = xe_gt_mcr_unicast_read_any(gt, XEHP_GLOBAL_MOCS(i));
+ else
+ reg_val = xe_mmio_read32(gt, XELP_GLOBAL_MOCS(i));
+ mocs_dbg(&gt_to_xe(gt)->drm, "%d 0x%x 0x%x 0x%x\n", i,
+ XELP_GLOBAL_MOCS(i).addr, reg_val, mocs);
+ if (reg_val != mocs)
+ KUNIT_FAIL(test, "mocs reg 0x%x has incorrect val.\n",
+ XELP_GLOBAL_MOCS(i).addr);
+ }
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ xe_device_mem_access_put(gt_to_xe(gt));
+}
+
+static int mocs_kernel_test_run_device(struct xe_device *xe)
+{
+ /* Basic check the system is configured with the expected mocs table */
+
+ struct live_mocs mocs;
+ struct xe_gt *gt;
+
+ unsigned int flags;
+ int id;
+
+ for_each_gt(gt, xe, id) {
+ flags = live_mocs_init(&mocs, gt);
+ if (flags & HAS_GLOBAL_MOCS)
+ read_mocs_table(gt, &mocs.table);
+ if (flags & HAS_LNCF_MOCS)
+ read_l3cc_table(gt, &mocs.table);
+ }
+ return 0;
+}
+
+void xe_live_mocs_kernel_kunit(struct kunit *test)
+{
+ xe_call_for_each_device(mocs_kernel_test_run_device);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_live_mocs_kernel_kunit);
diff --git a/drivers/gpu/drm/xe/tests/xe_mocs_test.c b/drivers/gpu/drm/xe/tests/xe_mocs_test.c
new file mode 100644
index 000000000000..ef56bd517b28
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_mocs_test.c
@@ -0,0 +1,24 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_mocs_test.h"
+
+#include <kunit/test.h>
+
+static struct kunit_case xe_mocs_tests[] = {
+ KUNIT_CASE(xe_live_mocs_kernel_kunit),
+ {}
+};
+
+static struct kunit_suite xe_mocs_test_suite = {
+ .name = "xe_mocs",
+ .test_cases = xe_mocs_tests,
+};
+
+kunit_test_suite(xe_mocs_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_mocs_test.h b/drivers/gpu/drm/xe/tests/xe_mocs_test.h
new file mode 100644
index 000000000000..7faa3575e6c3
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_mocs_test.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_MOCS_TEST_H_
+#define _XE_MOCS_TEST_H_
+
+struct kunit;
+
+void xe_live_mocs_kernel_kunit(struct kunit *test);
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_pci.c b/drivers/gpu/drm/xe/tests/xe_pci.c
new file mode 100644
index 000000000000..602793644f61
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_pci.c
@@ -0,0 +1,166 @@
+// SPDX-License-Identifier: GPL-2.0 AND MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "tests/xe_pci_test.h"
+
+#include "tests/xe_test.h"
+
+#include <kunit/test-bug.h>
+#include <kunit/test.h>
+#include <kunit/test-bug.h>
+#include <kunit/visibility.h>
+
+struct kunit_test_data {
+ int ndevs;
+ xe_device_fn xe_fn;
+};
+
+static int dev_to_xe_device_fn(struct device *dev, void *__data)
+
+{
+ struct drm_device *drm = dev_get_drvdata(dev);
+ struct kunit_test_data *data = __data;
+ int ret = 0;
+ int idx;
+
+ data->ndevs++;
+
+ if (drm_dev_enter(drm, &idx))
+ ret = data->xe_fn(to_xe_device(dev_get_drvdata(dev)));
+ drm_dev_exit(idx);
+
+ return ret;
+}
+
+/**
+ * xe_call_for_each_device - Iterate over all devices this driver binds to
+ * @xe_fn: Function to call for each device.
+ *
+ * This function iterated over all devices this driver binds to, and calls
+ * @xe_fn: for each one of them. If the called function returns anything else
+ * than 0, iteration is stopped and the return value is returned by this
+ * function. Across each function call, drm_dev_enter() / drm_dev_exit() is
+ * called for the corresponding drm device.
+ *
+ * Return: Number of devices iterated or
+ * the error code of a call to @xe_fn returning an error code.
+ */
+int xe_call_for_each_device(xe_device_fn xe_fn)
+{
+ int ret;
+ struct kunit_test_data data = {
+ .xe_fn = xe_fn,
+ .ndevs = 0,
+ };
+
+ ret = driver_for_each_device(&xe_pci_driver.driver, NULL,
+ &data, dev_to_xe_device_fn);
+
+ if (!data.ndevs)
+ kunit_skip(current->kunit_test, "test runs only on hardware\n");
+
+ return ret ?: data.ndevs;
+}
+
+/**
+ * xe_call_for_each_graphics_ip - Iterate over all recognized graphics IPs
+ * @xe_fn: Function to call for each device.
+ *
+ * This function iterates over the descriptors for all graphics IPs recognized
+ * by the driver and calls @xe_fn: for each one of them.
+ */
+void xe_call_for_each_graphics_ip(xe_graphics_fn xe_fn)
+{
+ const struct xe_graphics_desc *ip, *last = NULL;
+
+ for (int i = 0; i < ARRAY_SIZE(graphics_ip_map); i++) {
+ ip = graphics_ip_map[i].ip;
+ if (ip == last)
+ continue;
+
+ xe_fn(ip);
+ last = ip;
+ }
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_call_for_each_graphics_ip);
+
+/**
+ * xe_call_for_each_media_ip - Iterate over all recognized media IPs
+ * @xe_fn: Function to call for each device.
+ *
+ * This function iterates over the descriptors for all media IPs recognized
+ * by the driver and calls @xe_fn: for each one of them.
+ */
+void xe_call_for_each_media_ip(xe_media_fn xe_fn)
+{
+ const struct xe_media_desc *ip, *last = NULL;
+
+ for (int i = 0; i < ARRAY_SIZE(media_ip_map); i++) {
+ ip = media_ip_map[i].ip;
+ if (ip == last)
+ continue;
+
+ xe_fn(ip);
+ last = ip;
+ }
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_call_for_each_media_ip);
+
+static void fake_read_gmdid(struct xe_device *xe, enum xe_gmdid_type type,
+ u32 *ver, u32 *revid)
+{
+ struct kunit *test = kunit_get_current_test();
+ struct xe_pci_fake_data *data = test->priv;
+
+ if (type == GMDID_MEDIA) {
+ *ver = data->media_verx100;
+ *revid = xe_step_to_gmdid(data->media_step);
+ } else {
+ *ver = data->graphics_verx100;
+ *revid = xe_step_to_gmdid(data->graphics_step);
+ }
+}
+
+int xe_pci_fake_device_init(struct xe_device *xe)
+{
+ struct kunit *test = kunit_get_current_test();
+ struct xe_pci_fake_data *data = test->priv;
+ const struct pci_device_id *ent = pciidlist;
+ const struct xe_device_desc *desc;
+ const struct xe_subplatform_desc *subplatform_desc;
+
+ if (!data) {
+ desc = (const void *)ent->driver_data;
+ subplatform_desc = NULL;
+ goto done;
+ }
+
+ for (ent = pciidlist; ent->device; ent++) {
+ desc = (const void *)ent->driver_data;
+ if (desc->platform == data->platform)
+ break;
+ }
+
+ if (!ent->device)
+ return -ENODEV;
+
+ for (subplatform_desc = desc->subplatforms;
+ subplatform_desc && subplatform_desc->subplatform;
+ subplatform_desc++)
+ if (subplatform_desc->subplatform == data->subplatform)
+ break;
+
+ if (data->subplatform != XE_SUBPLATFORM_NONE && !subplatform_desc)
+ return -ENODEV;
+
+done:
+ kunit_activate_static_stub(test, read_gmdid, fake_read_gmdid);
+
+ xe_info_init_early(xe, desc, subplatform_desc);
+ xe_info_init(xe, desc->graphics, desc->media);
+
+ return 0;
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_pci_fake_device_init);
diff --git a/drivers/gpu/drm/xe/tests/xe_pci_test.c b/drivers/gpu/drm/xe/tests/xe_pci_test.c
new file mode 100644
index 000000000000..171e4180f1aa
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_pci_test.c
@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_drv.h>
+#include <drm/drm_kunit_helpers.h>
+
+#include <kunit/test.h>
+
+#include "tests/xe_test.h"
+
+#include "xe_device.h"
+#include "xe_pci_test.h"
+#include "xe_pci_types.h"
+
+static void check_graphics_ip(const struct xe_graphics_desc *graphics)
+{
+ struct kunit *test = xe_cur_kunit();
+ u64 mask = graphics->hw_engine_mask;
+
+ /* RCS, CCS, and BCS engines are allowed on the graphics IP */
+ mask &= ~(XE_HW_ENGINE_RCS_MASK |
+ XE_HW_ENGINE_CCS_MASK |
+ XE_HW_ENGINE_BCS_MASK);
+
+ /* Any remaining engines are an error */
+ KUNIT_ASSERT_EQ(test, mask, 0);
+}
+
+static void check_media_ip(const struct xe_media_desc *media)
+{
+ struct kunit *test = xe_cur_kunit();
+ u64 mask = media->hw_engine_mask;
+
+ /* VCS, VECS and GSCCS engines are allowed on the media IP */
+ mask &= ~(XE_HW_ENGINE_VCS_MASK |
+ XE_HW_ENGINE_VECS_MASK |
+ XE_HW_ENGINE_GSCCS_MASK);
+
+ /* Any remaining engines are an error */
+ KUNIT_ASSERT_EQ(test, mask, 0);
+}
+
+static void xe_gmdid_graphics_ip(struct kunit *test)
+{
+ xe_call_for_each_graphics_ip(check_graphics_ip);
+}
+
+static void xe_gmdid_media_ip(struct kunit *test)
+{
+ xe_call_for_each_media_ip(check_media_ip);
+}
+
+static struct kunit_case xe_pci_tests[] = {
+ KUNIT_CASE(xe_gmdid_graphics_ip),
+ KUNIT_CASE(xe_gmdid_media_ip),
+ {}
+};
+
+static struct kunit_suite xe_pci_test_suite = {
+ .name = "xe_pci",
+ .test_cases = xe_pci_tests,
+};
+
+kunit_test_suite(xe_pci_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_pci kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_pci_test.h b/drivers/gpu/drm/xe/tests/xe_pci_test.h
new file mode 100644
index 000000000000..811ffe5bd9fd
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_pci_test.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_PCI_TEST_H_
+#define _XE_PCI_TEST_H_
+
+#include <linux/types.h>
+
+#include "xe_platform_types.h"
+
+struct xe_device;
+struct xe_graphics_desc;
+struct xe_media_desc;
+
+typedef int (*xe_device_fn)(struct xe_device *);
+typedef void (*xe_graphics_fn)(const struct xe_graphics_desc *);
+typedef void (*xe_media_fn)(const struct xe_media_desc *);
+
+int xe_call_for_each_device(xe_device_fn xe_fn);
+void xe_call_for_each_graphics_ip(xe_graphics_fn xe_fn);
+void xe_call_for_each_media_ip(xe_media_fn xe_fn);
+
+struct xe_pci_fake_data {
+ enum xe_platform platform;
+ enum xe_subplatform subplatform;
+ u32 graphics_verx100;
+ u32 media_verx100;
+ u32 graphics_step;
+ u32 media_step;
+};
+
+int xe_pci_fake_device_init(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_rtp_test.c b/drivers/gpu/drm/xe/tests/xe_rtp_test.c
new file mode 100644
index 000000000000..4a6972897675
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_rtp_test.c
@@ -0,0 +1,319 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/string.h>
+#include <linux/xarray.h>
+
+#include <drm/drm_drv.h>
+#include <drm/drm_kunit_helpers.h>
+
+#include <kunit/test.h>
+
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_reg_defs.h"
+#include "xe_device.h"
+#include "xe_device_types.h"
+#include "xe_pci_test.h"
+#include "xe_reg_sr.h"
+#include "xe_rtp.h"
+
+#define REGULAR_REG1 XE_REG(1)
+#define REGULAR_REG2 XE_REG(2)
+#define REGULAR_REG3 XE_REG(3)
+#define MCR_REG1 XE_REG_MCR(1)
+#define MCR_REG2 XE_REG_MCR(2)
+#define MCR_REG3 XE_REG_MCR(3)
+#define MASKED_REG1 XE_REG(1, XE_REG_OPTION_MASKED)
+
+#undef XE_REG_MCR
+#define XE_REG_MCR(...) XE_REG(__VA_ARGS__, .mcr = 1)
+
+struct rtp_test_case {
+ const char *name;
+ struct xe_reg expected_reg;
+ u32 expected_set_bits;
+ u32 expected_clr_bits;
+ unsigned long expected_count;
+ unsigned int expected_sr_errors;
+ const struct xe_rtp_entry_sr *entries;
+};
+
+static bool match_yes(const struct xe_gt *gt, const struct xe_hw_engine *hwe)
+{
+ return true;
+}
+
+static bool match_no(const struct xe_gt *gt, const struct xe_hw_engine *hwe)
+{
+ return false;
+}
+
+static const struct rtp_test_case cases[] = {
+ {
+ .name = "coalesce-same-reg",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0) | REG_BIT(1),
+ .expected_clr_bits = REG_BIT(0) | REG_BIT(1),
+ .expected_count = 1,
+ /* Different bits on the same register: create a single entry */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(1)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "no-match-no-add",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 1,
+ /* Don't coalesce second entry since rules don't match */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_no)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(1)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "no-match-no-add-multiple-rules",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 1,
+ /* Don't coalesce second entry due to one of the rules */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes), FUNC(match_no)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(1)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "two-regs-two-entries",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 2,
+ /* Same bits on different registers are not coalesced */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG2, REG_BIT(0)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "clr-one-set-other",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(1) | REG_BIT(0),
+ .expected_count = 1,
+ /* Check clr vs set actions on different bits */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(CLR(REGULAR_REG1, REG_BIT(1)))
+ },
+ {}
+ },
+ },
+ {
+#define TEMP_MASK REG_GENMASK(10, 8)
+#define TEMP_FIELD REG_FIELD_PREP(TEMP_MASK, 2)
+ .name = "set-field",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = TEMP_FIELD,
+ .expected_clr_bits = TEMP_MASK,
+ .expected_count = 1,
+ /* Check FIELD_SET works */
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(FIELD_SET(REGULAR_REG1,
+ TEMP_MASK, TEMP_FIELD))
+ },
+ {}
+ },
+#undef TEMP_MASK
+#undef TEMP_FIELD
+ },
+ {
+ .name = "conflict-duplicate",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 1,
+ .expected_sr_errors = 1,
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ /* drop: setting same values twice */
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "conflict-not-disjoint",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 1,
+ .expected_sr_errors = 1,
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ /* drop: bits are not disjoint with previous entries */
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(CLR(REGULAR_REG1, REG_GENMASK(1, 0)))
+ },
+ {}
+ },
+ },
+ {
+ .name = "conflict-reg-type",
+ .expected_reg = REGULAR_REG1,
+ .expected_set_bits = REG_BIT(0),
+ .expected_clr_bits = REG_BIT(0),
+ .expected_count = 1,
+ .expected_sr_errors = 2,
+ .entries = (const struct xe_rtp_entry_sr[]) {
+ { XE_RTP_NAME("basic-1"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(REGULAR_REG1, REG_BIT(0)))
+ },
+ /* drop: regular vs MCR */
+ { XE_RTP_NAME("basic-2"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(MCR_REG1, REG_BIT(1)))
+ },
+ /* drop: regular vs masked */
+ { XE_RTP_NAME("basic-3"),
+ XE_RTP_RULES(FUNC(match_yes)),
+ XE_RTP_ACTIONS(SET(MASKED_REG1, REG_BIT(0)))
+ },
+ {}
+ },
+ },
+};
+
+static void xe_rtp_process_tests(struct kunit *test)
+{
+ const struct rtp_test_case *param = test->param_value;
+ struct xe_device *xe = test->priv;
+ struct xe_gt *gt = xe_device_get_root_tile(xe)->primary_gt;
+ struct xe_reg_sr *reg_sr = &gt->reg_sr;
+ const struct xe_reg_sr_entry *sre, *sr_entry = NULL;
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(gt);
+ unsigned long idx, count = 0;
+
+ xe_reg_sr_init(reg_sr, "xe_rtp_tests", xe);
+ xe_rtp_process_to_sr(&ctx, param->entries, reg_sr);
+
+ xa_for_each(&reg_sr->xa, idx, sre) {
+ if (idx == param->expected_reg.addr)
+ sr_entry = sre;
+
+ count++;
+ }
+
+ KUNIT_EXPECT_EQ(test, count, param->expected_count);
+ KUNIT_EXPECT_EQ(test, sr_entry->clr_bits, param->expected_clr_bits);
+ KUNIT_EXPECT_EQ(test, sr_entry->set_bits, param->expected_set_bits);
+ KUNIT_EXPECT_EQ(test, sr_entry->reg.raw, param->expected_reg.raw);
+ KUNIT_EXPECT_EQ(test, reg_sr->errors, param->expected_sr_errors);
+}
+
+static void rtp_desc(const struct rtp_test_case *t, char *desc)
+{
+ strscpy(desc, t->name, KUNIT_PARAM_DESC_SIZE);
+}
+
+KUNIT_ARRAY_PARAM(rtp, cases, rtp_desc);
+
+static int xe_rtp_test_init(struct kunit *test)
+{
+ struct xe_device *xe;
+ struct device *dev;
+ int ret;
+
+ dev = drm_kunit_helper_alloc_device(test);
+ KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev);
+
+ xe = drm_kunit_helper_alloc_drm_device(test, dev,
+ struct xe_device,
+ drm, DRIVER_GEM);
+ KUNIT_ASSERT_NOT_ERR_OR_NULL(test, xe);
+
+ /* Initialize an empty device */
+ test->priv = NULL;
+ ret = xe_pci_fake_device_init(xe);
+ KUNIT_ASSERT_EQ(test, ret, 0);
+
+ xe->drm.dev = dev;
+ test->priv = xe;
+
+ return 0;
+}
+
+static void xe_rtp_test_exit(struct kunit *test)
+{
+ struct xe_device *xe = test->priv;
+
+ drm_kunit_helper_free_device(test, xe->drm.dev);
+}
+
+static struct kunit_case xe_rtp_tests[] = {
+ KUNIT_CASE_PARAM(xe_rtp_process_tests, rtp_gen_params),
+ {}
+};
+
+static struct kunit_suite xe_rtp_test_suite = {
+ .name = "xe_rtp",
+ .init = xe_rtp_test_init,
+ .exit = xe_rtp_test_exit,
+ .test_cases = xe_rtp_tests,
+};
+
+kunit_test_suite(xe_rtp_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_rtp kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/tests/xe_test.h b/drivers/gpu/drm/xe/tests/xe_test.h
new file mode 100644
index 000000000000..7a1ae213e750
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_test.h
@@ -0,0 +1,67 @@
+/* SPDX-License-Identifier: GPL-2.0 AND MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_TEST_H_
+#define _XE_TEST_H_
+
+#include <linux/types.h>
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include <linux/sched.h>
+#include <kunit/test.h>
+
+/*
+ * Each test that provides a kunit private test structure, place a test id
+ * here and point the kunit->priv to an embedded struct xe_test_priv.
+ */
+enum xe_test_priv_id {
+ XE_TEST_LIVE_DMA_BUF,
+ XE_TEST_LIVE_MIGRATE,
+};
+
+/**
+ * struct xe_test_priv - Base class for test private info
+ * @id: enum xe_test_priv_id to identify the subclass.
+ */
+struct xe_test_priv {
+ enum xe_test_priv_id id;
+};
+
+#define XE_TEST_DECLARE(x) x
+#define XE_TEST_ONLY(x) unlikely(x)
+#define XE_TEST_EXPORT
+#define xe_cur_kunit() current->kunit_test
+
+/**
+ * xe_cur_kunit_priv - Obtain the struct xe_test_priv pointed to by
+ * current->kunit->priv if it exists and is embedded in the expected subclass.
+ * @id: Id of the expected subclass.
+ *
+ * Return: NULL if the process is not a kunit test, and NULL if the
+ * current kunit->priv pointer is not pointing to an object of the expected
+ * subclass. A pointer to the embedded struct xe_test_priv otherwise.
+ */
+static inline struct xe_test_priv *
+xe_cur_kunit_priv(enum xe_test_priv_id id)
+{
+ struct xe_test_priv *priv;
+
+ if (!xe_cur_kunit())
+ return NULL;
+
+ priv = xe_cur_kunit()->priv;
+ return priv->id == id ? priv : NULL;
+}
+
+#else /* if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST) */
+
+#define XE_TEST_DECLARE(x)
+#define XE_TEST_ONLY(x) 0
+#define XE_TEST_EXPORT static
+#define xe_cur_kunit() NULL
+#define xe_cur_kunit_priv(_id) NULL
+
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/tests/xe_wa_test.c b/drivers/gpu/drm/xe/tests/xe_wa_test.c
new file mode 100644
index 000000000000..a53c22a19582
--- /dev/null
+++ b/drivers/gpu/drm/xe/tests/xe_wa_test.c
@@ -0,0 +1,170 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_drv.h>
+#include <drm/drm_kunit_helpers.h>
+
+#include <kunit/test.h>
+
+#include "xe_device.h"
+#include "xe_pci_test.h"
+#include "xe_reg_sr.h"
+#include "xe_tuning.h"
+#include "xe_wa.h"
+
+struct platform_test_case {
+ const char *name;
+ enum xe_platform platform;
+ enum xe_subplatform subplatform;
+ u32 graphics_verx100;
+ u32 media_verx100;
+ struct xe_step_info step;
+};
+
+#define PLATFORM_CASE(platform__, graphics_step__) \
+ { \
+ .name = #platform__ " (" #graphics_step__ ")", \
+ .platform = XE_ ## platform__, \
+ .subplatform = XE_SUBPLATFORM_NONE, \
+ .step = { .graphics = STEP_ ## graphics_step__ } \
+ }
+
+
+#define SUBPLATFORM_CASE(platform__, subplatform__, graphics_step__) \
+ { \
+ .name = #platform__ "_" #subplatform__ " (" #graphics_step__ ")", \
+ .platform = XE_ ## platform__, \
+ .subplatform = XE_SUBPLATFORM_ ## platform__ ## _ ## subplatform__, \
+ .step = { .graphics = STEP_ ## graphics_step__ } \
+ }
+
+#define GMDID_CASE(platform__, graphics_verx100__, graphics_step__, \
+ media_verx100__, media_step__) \
+ { \
+ .name = #platform__ " (g:" #graphics_step__ ", m:" #media_step__ ")",\
+ .platform = XE_ ## platform__, \
+ .subplatform = XE_SUBPLATFORM_NONE, \
+ .graphics_verx100 = graphics_verx100__, \
+ .media_verx100 = media_verx100__, \
+ .step = { .graphics = STEP_ ## graphics_step__, \
+ .media = STEP_ ## media_step__ } \
+ }
+
+static const struct platform_test_case cases[] = {
+ PLATFORM_CASE(TIGERLAKE, B0),
+ PLATFORM_CASE(DG1, A0),
+ PLATFORM_CASE(DG1, B0),
+ PLATFORM_CASE(ALDERLAKE_S, A0),
+ PLATFORM_CASE(ALDERLAKE_S, B0),
+ PLATFORM_CASE(ALDERLAKE_S, C0),
+ PLATFORM_CASE(ALDERLAKE_S, D0),
+ PLATFORM_CASE(ALDERLAKE_P, A0),
+ PLATFORM_CASE(ALDERLAKE_P, B0),
+ PLATFORM_CASE(ALDERLAKE_P, C0),
+ SUBPLATFORM_CASE(ALDERLAKE_S, RPLS, D0),
+ SUBPLATFORM_CASE(ALDERLAKE_P, RPLU, E0),
+ SUBPLATFORM_CASE(DG2, G10, A0),
+ SUBPLATFORM_CASE(DG2, G10, A1),
+ SUBPLATFORM_CASE(DG2, G10, B0),
+ SUBPLATFORM_CASE(DG2, G10, C0),
+ SUBPLATFORM_CASE(DG2, G11, A0),
+ SUBPLATFORM_CASE(DG2, G11, B0),
+ SUBPLATFORM_CASE(DG2, G11, B1),
+ SUBPLATFORM_CASE(DG2, G12, A0),
+ SUBPLATFORM_CASE(DG2, G12, A1),
+ PLATFORM_CASE(PVC, B0),
+ PLATFORM_CASE(PVC, B1),
+ PLATFORM_CASE(PVC, C0),
+ GMDID_CASE(METEORLAKE, 1270, A0, 1300, A0),
+ GMDID_CASE(METEORLAKE, 1271, A0, 1300, A0),
+ GMDID_CASE(LUNARLAKE, 2004, A0, 2000, A0),
+ GMDID_CASE(LUNARLAKE, 2004, B0, 2000, A0),
+};
+
+static void platform_desc(const struct platform_test_case *t, char *desc)
+{
+ strscpy(desc, t->name, KUNIT_PARAM_DESC_SIZE);
+}
+
+KUNIT_ARRAY_PARAM(platform, cases, platform_desc);
+
+static int xe_wa_test_init(struct kunit *test)
+{
+ const struct platform_test_case *param = test->param_value;
+ struct xe_pci_fake_data data = {
+ .platform = param->platform,
+ .subplatform = param->subplatform,
+ .graphics_verx100 = param->graphics_verx100,
+ .media_verx100 = param->media_verx100,
+ .graphics_step = param->step.graphics,
+ .media_step = param->step.media,
+ };
+ struct xe_device *xe;
+ struct device *dev;
+ int ret;
+
+ dev = drm_kunit_helper_alloc_device(test);
+ KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev);
+
+ xe = drm_kunit_helper_alloc_drm_device(test, dev,
+ struct xe_device,
+ drm, DRIVER_GEM);
+ KUNIT_ASSERT_NOT_ERR_OR_NULL(test, xe);
+
+ test->priv = &data;
+ ret = xe_pci_fake_device_init(xe);
+ KUNIT_ASSERT_EQ(test, ret, 0);
+
+ if (!param->graphics_verx100)
+ xe->info.step = param->step;
+
+ /* TODO: init hw engines for engine/LRC WAs */
+ xe->drm.dev = dev;
+ test->priv = xe;
+
+ return 0;
+}
+
+static void xe_wa_test_exit(struct kunit *test)
+{
+ struct xe_device *xe = test->priv;
+
+ drm_kunit_helper_free_device(test, xe->drm.dev);
+}
+
+static void xe_wa_gt(struct kunit *test)
+{
+ struct xe_device *xe = test->priv;
+ struct xe_gt *gt;
+ int id;
+
+ for_each_gt(gt, xe, id) {
+ xe_reg_sr_init(&gt->reg_sr, "GT", xe);
+
+ xe_wa_process_gt(gt);
+ xe_tuning_process_gt(gt);
+
+ KUNIT_ASSERT_EQ(test, gt->reg_sr.errors, 0);
+ }
+}
+
+static struct kunit_case xe_wa_tests[] = {
+ KUNIT_CASE_PARAM(xe_wa_gt, platform_gen_params),
+ {}
+};
+
+static struct kunit_suite xe_rtp_test_suite = {
+ .name = "xe_wa",
+ .init = xe_wa_test_init,
+ .exit = xe_wa_test_exit,
+ .test_cases = xe_wa_tests,
+};
+
+kunit_test_suite(xe_rtp_test_suite);
+
+MODULE_AUTHOR("Intel Corporation");
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("xe_wa kunit test");
+MODULE_IMPORT_NS(EXPORTED_FOR_KUNIT_TESTING);
diff --git a/drivers/gpu/drm/xe/xe_assert.h b/drivers/gpu/drm/xe/xe_assert.h
new file mode 100644
index 000000000000..34c142e6cfb0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_assert.h
@@ -0,0 +1,174 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_ASSERT_H_
+#define _XE_ASSERT_H_
+
+#include <linux/string_helpers.h>
+
+#include <drm/drm_print.h>
+
+#include "xe_device_types.h"
+#include "xe_step.h"
+
+/**
+ * DOC: Xe ASSERTs
+ *
+ * While Xe driver aims to be simpler than legacy i915 driver it is still
+ * complex enough that some changes introduced while adding new functionality
+ * could break the existing code.
+ *
+ * Adding &drm_WARN or &drm_err to catch unwanted programming usage could lead
+ * to undesired increased driver footprint and may impact production driver
+ * performance as this additional code will be always present.
+ *
+ * To allow annotate functions with additional detailed debug checks to assert
+ * that all prerequisites are satisfied, without worrying about footprint or
+ * performance penalty on production builds where all potential misuses
+ * introduced during code integration were already fixed, we introduce family
+ * of Xe assert macros that try to follow classic assert() utility:
+ *
+ * * xe_assert()
+ * * xe_tile_assert()
+ * * xe_gt_assert()
+ *
+ * These macros are implemented on top of &drm_WARN, but unlikely to the origin,
+ * warning is triggered when provided condition is false. Additionally all above
+ * assert macros cannot be used in expressions or as a condition, since
+ * underlying code will be compiled out on non-debug builds.
+ *
+ * Note that these macros are not intended for use to cover known gaps in the
+ * implementation; for such cases use regular &drm_WARN or &drm_err and provide
+ * valid safe fallback.
+ *
+ * Also in cases where performance or footprint is not an issue, developers
+ * should continue to use the regular &drm_WARN or &drm_err to ensure that bug
+ * reports from production builds will contain meaningful diagnostics data.
+ *
+ * Below code shows how asserts could help in debug to catch unplanned use::
+ *
+ * static void one_igfx(struct xe_device *xe)
+ * {
+ * xe_assert(xe, xe->info.is_dgfx == false);
+ * xe_assert(xe, xe->info.tile_count == 1);
+ * }
+ *
+ * static void two_dgfx(struct xe_device *xe)
+ * {
+ * xe_assert(xe, xe->info.is_dgfx);
+ * xe_assert(xe, xe->info.tile_count == 2);
+ * }
+ *
+ * void foo(struct xe_device *xe)
+ * {
+ * if (xe->info.dgfx)
+ * return two_dgfx(xe);
+ * return one_igfx(xe);
+ * }
+ *
+ * void bar(struct xe_device *xe)
+ * {
+ * if (drm_WARN_ON(xe->drm, xe->info.tile_count > 2))
+ * return;
+ *
+ * if (xe->info.tile_count == 2)
+ * return two_dgfx(xe);
+ * return one_igfx(xe);
+ * }
+ */
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+#define __xe_assert_msg(xe, condition, msg, arg...) ({ \
+ (void)drm_WARN(&(xe)->drm, !(condition), "[" DRM_NAME "] Assertion `%s` failed!\n" msg, \
+ __stringify(condition), ## arg); \
+})
+#else
+#define __xe_assert_msg(xe, condition, msg, arg...) ({ \
+ typecheck(const struct xe_device *, xe); \
+ BUILD_BUG_ON_INVALID(condition); \
+})
+#endif
+
+/**
+ * xe_assert - warn if condition is false when debugging.
+ * @xe: the &struct xe_device pointer to which &condition applies
+ * @condition: condition to check
+ *
+ * xe_assert() uses &drm_WARN to emit a warning and print additional information
+ * that could be read from the &xe pointer if provided &condition is false.
+ *
+ * Contrary to &drm_WARN, xe_assert() is effective only on debug builds
+ * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
+ * or as a condition.
+ *
+ * See `Xe ASSERTs`_ for general usage guidelines.
+ */
+#define xe_assert(xe, condition) xe_assert_msg((xe), condition, "")
+#define xe_assert_msg(xe, condition, msg, arg...) ({ \
+ const struct xe_device *__xe = (xe); \
+ __xe_assert_msg(__xe, condition, \
+ "platform: %d subplatform: %d\n" \
+ "graphics: %s %u.%02u step %s\n" \
+ "media: %s %u.%02u step %s\n" \
+ msg, \
+ __xe->info.platform, __xe->info.subplatform, \
+ __xe->info.graphics_name, \
+ __xe->info.graphics_verx100 / 100, \
+ __xe->info.graphics_verx100 % 100, \
+ xe_step_name(__xe->info.step.graphics), \
+ __xe->info.media_name, \
+ __xe->info.media_verx100 / 100, \
+ __xe->info.media_verx100 % 100, \
+ xe_step_name(__xe->info.step.media), \
+ ## arg); \
+})
+
+/**
+ * xe_tile_assert - warn if condition is false when debugging.
+ * @tile: the &struct xe_tile pointer to which &condition applies
+ * @condition: condition to check
+ *
+ * xe_tile_assert() uses &drm_WARN to emit a warning and print additional
+ * information that could be read from the &tile pointer if provided &condition
+ * is false.
+ *
+ * Contrary to &drm_WARN, xe_tile_assert() is effective only on debug builds
+ * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
+ * or as a condition.
+ *
+ * See `Xe ASSERTs`_ for general usage guidelines.
+ */
+#define xe_tile_assert(tile, condition) xe_tile_assert_msg((tile), condition, "")
+#define xe_tile_assert_msg(tile, condition, msg, arg...) ({ \
+ const struct xe_tile *__tile = (tile); \
+ char __buf[10] __maybe_unused; \
+ xe_assert_msg(tile_to_xe(__tile), condition, "tile: %u VRAM %s\n" msg, \
+ __tile->id, ({ string_get_size(__tile->mem.vram.actual_physical_size, 1, \
+ STRING_UNITS_2, __buf, sizeof(__buf)); __buf; }), ## arg); \
+})
+
+/**
+ * xe_gt_assert - warn if condition is false when debugging.
+ * @gt: the &struct xe_gt pointer to which &condition applies
+ * @condition: condition to check
+ *
+ * xe_gt_assert() uses &drm_WARN to emit a warning and print additional
+ * information that could be safetely read from the &gt pointer if provided
+ * &condition is false.
+ *
+ * Contrary to &drm_WARN, xe_gt_assert() is effective only on debug builds
+ * (&CONFIG_DRM_XE_DEBUG must be enabled) and cannot be used in expressions
+ * or as a condition.
+ *
+ * See `Xe ASSERTs`_ for general usage guidelines.
+ */
+#define xe_gt_assert(gt, condition) xe_gt_assert_msg((gt), condition, "")
+#define xe_gt_assert_msg(gt, condition, msg, arg...) ({ \
+ const struct xe_gt *__gt = (gt); \
+ xe_tile_assert_msg(gt_to_tile(__gt), condition, "GT: %u type %d\n" msg, \
+ __gt->info.id, __gt->info.type, ## arg); \
+})
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bb.c b/drivers/gpu/drm/xe/xe_bb.c
new file mode 100644
index 000000000000..7c124475c428
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bb.c
@@ -0,0 +1,110 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_bb.h"
+
+#include "instructions/xe_mi_commands.h"
+#include "regs/xe_gpu_commands.h"
+#include "xe_device.h"
+#include "xe_exec_queue_types.h"
+#include "xe_gt.h"
+#include "xe_hw_fence.h"
+#include "xe_sa.h"
+#include "xe_sched_job.h"
+#include "xe_vm_types.h"
+
+static int bb_prefetch(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (GRAPHICS_VERx100(xe) >= 1250 && !xe_gt_is_media_type(gt))
+ /*
+ * RCS and CCS require 1K, although other engines would be
+ * okay with 512.
+ */
+ return SZ_1K;
+ else
+ return SZ_512;
+}
+
+struct xe_bb *xe_bb_new(struct xe_gt *gt, u32 dwords, bool usm)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_bb *bb = kmalloc(sizeof(*bb), GFP_KERNEL);
+ int err;
+
+ if (!bb)
+ return ERR_PTR(-ENOMEM);
+
+ /*
+ * We need to allocate space for the requested number of dwords,
+ * one additional MI_BATCH_BUFFER_END dword, and additional buffer
+ * space to accomodate the platform-specific hardware prefetch
+ * requirements.
+ */
+ bb->bo = xe_sa_bo_new(!usm ? tile->mem.kernel_bb_pool : gt->usm.bb_pool,
+ 4 * (dwords + 1) + bb_prefetch(gt));
+ if (IS_ERR(bb->bo)) {
+ err = PTR_ERR(bb->bo);
+ goto err;
+ }
+
+ bb->cs = xe_sa_bo_cpu_addr(bb->bo);
+ bb->len = 0;
+
+ return bb;
+err:
+ kfree(bb);
+ return ERR_PTR(err);
+}
+
+static struct xe_sched_job *
+__xe_bb_create_job(struct xe_exec_queue *q, struct xe_bb *bb, u64 *addr)
+{
+ u32 size = drm_suballoc_size(bb->bo);
+
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+
+ xe_gt_assert(q->gt, bb->len * 4 + bb_prefetch(q->gt) <= size);
+
+ xe_sa_bo_flush_write(bb->bo);
+
+ return xe_sched_job_create(q, addr);
+}
+
+struct xe_sched_job *xe_bb_create_migration_job(struct xe_exec_queue *q,
+ struct xe_bb *bb,
+ u64 batch_base_ofs,
+ u32 second_idx)
+{
+ u64 addr[2] = {
+ batch_base_ofs + drm_suballoc_soffset(bb->bo),
+ batch_base_ofs + drm_suballoc_soffset(bb->bo) +
+ 4 * second_idx,
+ };
+
+ xe_gt_assert(q->gt, second_idx <= bb->len);
+ xe_gt_assert(q->gt, q->vm->flags & XE_VM_FLAG_MIGRATION);
+
+ return __xe_bb_create_job(q, bb, addr);
+}
+
+struct xe_sched_job *xe_bb_create_job(struct xe_exec_queue *q,
+ struct xe_bb *bb)
+{
+ u64 addr = xe_sa_bo_gpu_addr(bb->bo);
+
+ xe_gt_assert(q->gt, !(q->vm && q->vm->flags & XE_VM_FLAG_MIGRATION));
+ return __xe_bb_create_job(q, bb, &addr);
+}
+
+void xe_bb_free(struct xe_bb *bb, struct dma_fence *fence)
+{
+ if (!bb)
+ return;
+
+ xe_sa_bo_free(bb->bo, fence);
+ kfree(bb);
+}
diff --git a/drivers/gpu/drm/xe/xe_bb.h b/drivers/gpu/drm/xe/xe_bb.h
new file mode 100644
index 000000000000..fafacd73dcc3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bb.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_BB_H_
+#define _XE_BB_H_
+
+#include "xe_bb_types.h"
+
+struct dma_fence;
+
+struct xe_gt;
+struct xe_exec_queue;
+struct xe_sched_job;
+
+struct xe_bb *xe_bb_new(struct xe_gt *gt, u32 size, bool usm);
+struct xe_sched_job *xe_bb_create_job(struct xe_exec_queue *q,
+ struct xe_bb *bb);
+struct xe_sched_job *xe_bb_create_migration_job(struct xe_exec_queue *q,
+ struct xe_bb *bb, u64 batch_ofs,
+ u32 second_idx);
+void xe_bb_free(struct xe_bb *bb, struct dma_fence *fence);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bb_types.h b/drivers/gpu/drm/xe/xe_bb_types.h
new file mode 100644
index 000000000000..b7d30308cf90
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bb_types.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_BB_TYPES_H_
+#define _XE_BB_TYPES_H_
+
+#include <linux/types.h>
+
+struct drm_suballoc;
+
+struct xe_bb {
+ struct drm_suballoc *bo;
+
+ u32 *cs;
+ u32 len; /* in dwords */
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
new file mode 100644
index 000000000000..8e4a3b1f6b93
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -0,0 +1,2269 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_bo.h"
+
+#include <linux/dma-buf.h>
+
+#include <drm/drm_drv.h>
+#include <drm/drm_gem_ttm_helper.h>
+#include <drm/drm_managed.h>
+#include <drm/ttm/ttm_device.h>
+#include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_tt.h>
+#include <drm/xe_drm.h>
+
+#include "xe_device.h"
+#include "xe_dma_buf.h"
+#include "xe_drm_client.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_map.h"
+#include "xe_migrate.h"
+#include "xe_preempt_fence.h"
+#include "xe_res_cursor.h"
+#include "xe_trace.h"
+#include "xe_ttm_stolen_mgr.h"
+#include "xe_vm.h"
+
+static const struct ttm_place sys_placement_flags = {
+ .fpfn = 0,
+ .lpfn = 0,
+ .mem_type = XE_PL_SYSTEM,
+ .flags = 0,
+};
+
+static struct ttm_placement sys_placement = {
+ .num_placement = 1,
+ .placement = &sys_placement_flags,
+ .num_busy_placement = 1,
+ .busy_placement = &sys_placement_flags,
+};
+
+static const struct ttm_place tt_placement_flags = {
+ .fpfn = 0,
+ .lpfn = 0,
+ .mem_type = XE_PL_TT,
+ .flags = 0,
+};
+
+static struct ttm_placement tt_placement = {
+ .num_placement = 1,
+ .placement = &tt_placement_flags,
+ .num_busy_placement = 1,
+ .busy_placement = &sys_placement_flags,
+};
+
+bool mem_type_is_vram(u32 mem_type)
+{
+ return mem_type >= XE_PL_VRAM0 && mem_type != XE_PL_STOLEN;
+}
+
+static bool resource_is_stolen_vram(struct xe_device *xe, struct ttm_resource *res)
+{
+ return res->mem_type == XE_PL_STOLEN && IS_DGFX(xe);
+}
+
+static bool resource_is_vram(struct ttm_resource *res)
+{
+ return mem_type_is_vram(res->mem_type);
+}
+
+bool xe_bo_is_vram(struct xe_bo *bo)
+{
+ return resource_is_vram(bo->ttm.resource) ||
+ resource_is_stolen_vram(xe_bo_device(bo), bo->ttm.resource);
+}
+
+bool xe_bo_is_stolen(struct xe_bo *bo)
+{
+ return bo->ttm.resource->mem_type == XE_PL_STOLEN;
+}
+
+/**
+ * xe_bo_is_stolen_devmem - check if BO is of stolen type accessed via PCI BAR
+ * @bo: The BO
+ *
+ * The stolen memory is accessed through the PCI BAR for both DGFX and some
+ * integrated platforms that have a dedicated bit in the PTE for devmem (DM).
+ *
+ * Returns: true if it's stolen memory accessed via PCI BAR, false otherwise.
+ */
+bool xe_bo_is_stolen_devmem(struct xe_bo *bo)
+{
+ return xe_bo_is_stolen(bo) &&
+ GRAPHICS_VERx100(xe_bo_device(bo)) >= 1270;
+}
+
+static bool xe_bo_is_user(struct xe_bo *bo)
+{
+ return bo->flags & XE_BO_CREATE_USER_BIT;
+}
+
+static struct xe_migrate *
+mem_type_to_migrate(struct xe_device *xe, u32 mem_type)
+{
+ struct xe_tile *tile;
+
+ xe_assert(xe, mem_type == XE_PL_STOLEN || mem_type_is_vram(mem_type));
+ tile = &xe->tiles[mem_type == XE_PL_STOLEN ? 0 : (mem_type - XE_PL_VRAM0)];
+ return tile->migrate;
+}
+
+static struct xe_mem_region *res_to_mem_region(struct ttm_resource *res)
+{
+ struct xe_device *xe = ttm_to_xe_device(res->bo->bdev);
+ struct ttm_resource_manager *mgr;
+
+ xe_assert(xe, resource_is_vram(res));
+ mgr = ttm_manager_type(&xe->ttm, res->mem_type);
+ return to_xe_ttm_vram_mgr(mgr)->vram;
+}
+
+static void try_add_system(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags, u32 *c)
+{
+ xe_assert(xe, *c < ARRAY_SIZE(bo->placements));
+
+ if (bo_flags & XE_BO_CREATE_SYSTEM_BIT) {
+ bo->placements[*c] = (struct ttm_place) {
+ .mem_type = XE_PL_TT,
+ };
+ *c += 1;
+
+ if (bo->props.preferred_mem_type == XE_BO_PROPS_INVALID)
+ bo->props.preferred_mem_type = XE_PL_TT;
+ }
+}
+
+static void add_vram(struct xe_device *xe, struct xe_bo *bo,
+ struct ttm_place *places, u32 bo_flags, u32 mem_type, u32 *c)
+{
+ struct ttm_place place = { .mem_type = mem_type };
+ struct xe_mem_region *vram;
+ u64 io_size;
+
+ vram = to_xe_ttm_vram_mgr(ttm_manager_type(&xe->ttm, mem_type))->vram;
+ xe_assert(xe, vram && vram->usable_size);
+ io_size = vram->io_size;
+
+ /*
+ * For eviction / restore on suspend / resume objects
+ * pinned in VRAM must be contiguous
+ */
+ if (bo_flags & (XE_BO_CREATE_PINNED_BIT |
+ XE_BO_CREATE_GGTT_BIT))
+ place.flags |= TTM_PL_FLAG_CONTIGUOUS;
+
+ if (io_size < vram->usable_size) {
+ if (bo_flags & XE_BO_NEEDS_CPU_ACCESS) {
+ place.fpfn = 0;
+ place.lpfn = io_size >> PAGE_SHIFT;
+ } else {
+ place.flags |= TTM_PL_FLAG_TOPDOWN;
+ }
+ }
+ places[*c] = place;
+ *c += 1;
+
+ if (bo->props.preferred_mem_type == XE_BO_PROPS_INVALID)
+ bo->props.preferred_mem_type = mem_type;
+}
+
+static void try_add_vram(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags, u32 *c)
+{
+ xe_assert(xe, *c < ARRAY_SIZE(bo->placements));
+
+ if (bo->props.preferred_gt == XE_GT1) {
+ if (bo_flags & XE_BO_CREATE_VRAM1_BIT)
+ add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM1, c);
+ if (bo_flags & XE_BO_CREATE_VRAM0_BIT)
+ add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM0, c);
+ } else {
+ if (bo_flags & XE_BO_CREATE_VRAM0_BIT)
+ add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM0, c);
+ if (bo_flags & XE_BO_CREATE_VRAM1_BIT)
+ add_vram(xe, bo, bo->placements, bo_flags, XE_PL_VRAM1, c);
+ }
+}
+
+static void try_add_stolen(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags, u32 *c)
+{
+ xe_assert(xe, *c < ARRAY_SIZE(bo->placements));
+
+ if (bo_flags & XE_BO_CREATE_STOLEN_BIT) {
+ bo->placements[*c] = (struct ttm_place) {
+ .mem_type = XE_PL_STOLEN,
+ .flags = bo_flags & (XE_BO_CREATE_PINNED_BIT |
+ XE_BO_CREATE_GGTT_BIT) ?
+ TTM_PL_FLAG_CONTIGUOUS : 0,
+ };
+ *c += 1;
+ }
+}
+
+static int __xe_bo_placement_for_flags(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags)
+{
+ u32 c = 0;
+
+ bo->props.preferred_mem_type = XE_BO_PROPS_INVALID;
+
+ /* The order of placements should indicate preferred location */
+
+ if (bo->props.preferred_mem_class == DRM_XE_MEM_REGION_CLASS_SYSMEM) {
+ try_add_system(xe, bo, bo_flags, &c);
+ try_add_vram(xe, bo, bo_flags, &c);
+ } else {
+ try_add_vram(xe, bo, bo_flags, &c);
+ try_add_system(xe, bo, bo_flags, &c);
+ }
+ try_add_stolen(xe, bo, bo_flags, &c);
+
+ if (!c)
+ return -EINVAL;
+
+ bo->placement = (struct ttm_placement) {
+ .num_placement = c,
+ .placement = bo->placements,
+ .num_busy_placement = c,
+ .busy_placement = bo->placements,
+ };
+
+ return 0;
+}
+
+int xe_bo_placement_for_flags(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags)
+{
+ xe_bo_assert_held(bo);
+ return __xe_bo_placement_for_flags(xe, bo, bo_flags);
+}
+
+static void xe_evict_flags(struct ttm_buffer_object *tbo,
+ struct ttm_placement *placement)
+{
+ if (!xe_bo_is_xe_bo(tbo)) {
+ /* Don't handle scatter gather BOs */
+ if (tbo->type == ttm_bo_type_sg) {
+ placement->num_placement = 0;
+ placement->num_busy_placement = 0;
+ return;
+ }
+
+ *placement = sys_placement;
+ return;
+ }
+
+ /*
+ * For xe, sg bos that are evicted to system just triggers a
+ * rebind of the sg list upon subsequent validation to XE_PL_TT.
+ */
+ switch (tbo->resource->mem_type) {
+ case XE_PL_VRAM0:
+ case XE_PL_VRAM1:
+ case XE_PL_STOLEN:
+ *placement = tt_placement;
+ break;
+ case XE_PL_TT:
+ default:
+ *placement = sys_placement;
+ break;
+ }
+}
+
+struct xe_ttm_tt {
+ struct ttm_tt ttm;
+ struct device *dev;
+ struct sg_table sgt;
+ struct sg_table *sg;
+};
+
+static int xe_tt_map_sg(struct ttm_tt *tt)
+{
+ struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
+ unsigned long num_pages = tt->num_pages;
+ int ret;
+
+ XE_WARN_ON(tt->page_flags & TTM_TT_FLAG_EXTERNAL);
+
+ if (xe_tt->sg)
+ return 0;
+
+ ret = sg_alloc_table_from_pages_segment(&xe_tt->sgt, tt->pages,
+ num_pages, 0,
+ (u64)num_pages << PAGE_SHIFT,
+ xe_sg_segment_size(xe_tt->dev),
+ GFP_KERNEL);
+ if (ret)
+ return ret;
+
+ xe_tt->sg = &xe_tt->sgt;
+ ret = dma_map_sgtable(xe_tt->dev, xe_tt->sg, DMA_BIDIRECTIONAL,
+ DMA_ATTR_SKIP_CPU_SYNC);
+ if (ret) {
+ sg_free_table(xe_tt->sg);
+ xe_tt->sg = NULL;
+ return ret;
+ }
+
+ return 0;
+}
+
+struct sg_table *xe_bo_sg(struct xe_bo *bo)
+{
+ struct ttm_tt *tt = bo->ttm.ttm;
+ struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
+
+ return xe_tt->sg;
+}
+
+static struct ttm_tt *xe_ttm_tt_create(struct ttm_buffer_object *ttm_bo,
+ u32 page_flags)
+{
+ struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
+ struct xe_device *xe = xe_bo_device(bo);
+ struct xe_ttm_tt *tt;
+ unsigned long extra_pages;
+ enum ttm_caching caching;
+ int err;
+
+ tt = kzalloc(sizeof(*tt), GFP_KERNEL);
+ if (!tt)
+ return NULL;
+
+ tt->dev = xe->drm.dev;
+
+ extra_pages = 0;
+ if (xe_bo_needs_ccs_pages(bo))
+ extra_pages = DIV_ROUND_UP(xe_device_ccs_bytes(xe, bo->size),
+ PAGE_SIZE);
+
+ switch (bo->cpu_caching) {
+ case DRM_XE_GEM_CPU_CACHING_WC:
+ caching = ttm_write_combined;
+ break;
+ default:
+ caching = ttm_cached;
+ break;
+ }
+
+ WARN_ON((bo->flags & XE_BO_CREATE_USER_BIT) && !bo->cpu_caching);
+
+ /*
+ * Display scanout is always non-coherent with the CPU cache.
+ *
+ * For Xe_LPG and beyond, PPGTT PTE lookups are also non-coherent and
+ * require a CPU:WC mapping.
+ */
+ if ((!bo->cpu_caching && bo->flags & XE_BO_SCANOUT_BIT) ||
+ (xe->info.graphics_verx100 >= 1270 && bo->flags & XE_BO_PAGETABLE))
+ caching = ttm_write_combined;
+
+ err = ttm_tt_init(&tt->ttm, &bo->ttm, page_flags, caching, extra_pages);
+ if (err) {
+ kfree(tt);
+ return NULL;
+ }
+
+ return &tt->ttm;
+}
+
+static int xe_ttm_tt_populate(struct ttm_device *ttm_dev, struct ttm_tt *tt,
+ struct ttm_operation_ctx *ctx)
+{
+ int err;
+
+ /*
+ * dma-bufs are not populated with pages, and the dma-
+ * addresses are set up when moved to XE_PL_TT.
+ */
+ if (tt->page_flags & TTM_TT_FLAG_EXTERNAL)
+ return 0;
+
+ err = ttm_pool_alloc(&ttm_dev->pool, tt, ctx);
+ if (err)
+ return err;
+
+ /* A follow up may move this xe_bo_move when BO is moved to XE_PL_TT */
+ err = xe_tt_map_sg(tt);
+ if (err)
+ ttm_pool_free(&ttm_dev->pool, tt);
+
+ return err;
+}
+
+static void xe_ttm_tt_unpopulate(struct ttm_device *ttm_dev, struct ttm_tt *tt)
+{
+ struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
+
+ if (tt->page_flags & TTM_TT_FLAG_EXTERNAL)
+ return;
+
+ if (xe_tt->sg) {
+ dma_unmap_sgtable(xe_tt->dev, xe_tt->sg,
+ DMA_BIDIRECTIONAL, 0);
+ sg_free_table(xe_tt->sg);
+ xe_tt->sg = NULL;
+ }
+
+ return ttm_pool_free(&ttm_dev->pool, tt);
+}
+
+static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct ttm_tt *tt)
+{
+ ttm_tt_fini(tt);
+ kfree(tt);
+}
+
+static int xe_ttm_io_mem_reserve(struct ttm_device *bdev,
+ struct ttm_resource *mem)
+{
+ struct xe_device *xe = ttm_to_xe_device(bdev);
+
+ switch (mem->mem_type) {
+ case XE_PL_SYSTEM:
+ case XE_PL_TT:
+ return 0;
+ case XE_PL_VRAM0:
+ case XE_PL_VRAM1: {
+ struct xe_ttm_vram_mgr_resource *vres =
+ to_xe_ttm_vram_mgr_resource(mem);
+ struct xe_mem_region *vram = res_to_mem_region(mem);
+
+ if (vres->used_visible_size < mem->size)
+ return -EINVAL;
+
+ mem->bus.offset = mem->start << PAGE_SHIFT;
+
+ if (vram->mapping &&
+ mem->placement & TTM_PL_FLAG_CONTIGUOUS)
+ mem->bus.addr = (u8 *)vram->mapping +
+ mem->bus.offset;
+
+ mem->bus.offset += vram->io_start;
+ mem->bus.is_iomem = true;
+
+#if !defined(CONFIG_X86)
+ mem->bus.caching = ttm_write_combined;
+#endif
+ return 0;
+ } case XE_PL_STOLEN:
+ return xe_ttm_stolen_io_mem_reserve(xe, mem);
+ default:
+ return -EINVAL;
+ }
+}
+
+static int xe_bo_trigger_rebind(struct xe_device *xe, struct xe_bo *bo,
+ const struct ttm_operation_ctx *ctx)
+{
+ struct dma_resv_iter cursor;
+ struct dma_fence *fence;
+ struct drm_gem_object *obj = &bo->ttm.base;
+ struct drm_gpuvm_bo *vm_bo;
+ bool idle = false;
+ int ret = 0;
+
+ dma_resv_assert_held(bo->ttm.base.resv);
+
+ if (!list_empty(&bo->ttm.base.gpuva.list)) {
+ dma_resv_iter_begin(&cursor, bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP);
+ dma_resv_for_each_fence_unlocked(&cursor, fence)
+ dma_fence_enable_sw_signaling(fence);
+ dma_resv_iter_end(&cursor);
+ }
+
+ drm_gem_for_each_gpuvm_bo(vm_bo, obj) {
+ struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
+ struct drm_gpuva *gpuva;
+
+ if (!xe_vm_in_fault_mode(vm)) {
+ drm_gpuvm_bo_evict(vm_bo, true);
+ continue;
+ }
+
+ if (!idle) {
+ long timeout;
+
+ if (ctx->no_wait_gpu &&
+ !dma_resv_test_signaled(bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP))
+ return -EBUSY;
+
+ timeout = dma_resv_wait_timeout(bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP,
+ ctx->interruptible,
+ MAX_SCHEDULE_TIMEOUT);
+ if (!timeout)
+ return -ETIME;
+ if (timeout < 0)
+ return timeout;
+
+ idle = true;
+ }
+
+ drm_gpuvm_bo_for_each_va(gpuva, vm_bo) {
+ struct xe_vma *vma = gpuva_to_vma(gpuva);
+
+ trace_xe_vma_evict(vma);
+ ret = xe_vm_invalidate_vma(vma);
+ if (XE_WARN_ON(ret))
+ return ret;
+ }
+ }
+
+ return ret;
+}
+
+/*
+ * The dma-buf map_attachment() / unmap_attachment() is hooked up here.
+ * Note that unmapping the attachment is deferred to the next
+ * map_attachment time, or to bo destroy (after idling) whichever comes first.
+ * This is to avoid syncing before unmap_attachment(), assuming that the
+ * caller relies on idling the reservation object before moving the
+ * backing store out. Should that assumption not hold, then we will be able
+ * to unconditionally call unmap_attachment() when moving out to system.
+ */
+static int xe_bo_move_dmabuf(struct ttm_buffer_object *ttm_bo,
+ struct ttm_resource *new_res)
+{
+ struct dma_buf_attachment *attach = ttm_bo->base.import_attach;
+ struct xe_ttm_tt *xe_tt = container_of(ttm_bo->ttm, struct xe_ttm_tt,
+ ttm);
+ struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+ struct sg_table *sg;
+
+ xe_assert(xe, attach);
+ xe_assert(xe, ttm_bo->ttm);
+
+ if (new_res->mem_type == XE_PL_SYSTEM)
+ goto out;
+
+ if (ttm_bo->sg) {
+ dma_buf_unmap_attachment(attach, ttm_bo->sg, DMA_BIDIRECTIONAL);
+ ttm_bo->sg = NULL;
+ }
+
+ sg = dma_buf_map_attachment(attach, DMA_BIDIRECTIONAL);
+ if (IS_ERR(sg))
+ return PTR_ERR(sg);
+
+ ttm_bo->sg = sg;
+ xe_tt->sg = sg;
+
+out:
+ ttm_bo_move_null(ttm_bo, new_res);
+
+ return 0;
+}
+
+/**
+ * xe_bo_move_notify - Notify subsystems of a pending move
+ * @bo: The buffer object
+ * @ctx: The struct ttm_operation_ctx controlling locking and waits.
+ *
+ * This function notifies subsystems of an upcoming buffer move.
+ * Upon receiving such a notification, subsystems should schedule
+ * halting access to the underlying pages and optionally add a fence
+ * to the buffer object's dma_resv object, that signals when access is
+ * stopped. The caller will wait on all dma_resv fences before
+ * starting the move.
+ *
+ * A subsystem may commence access to the object after obtaining
+ * bindings to the new backing memory under the object lock.
+ *
+ * Return: 0 on success, -EINTR or -ERESTARTSYS if interrupted in fault mode,
+ * negative error code on error.
+ */
+static int xe_bo_move_notify(struct xe_bo *bo,
+ const struct ttm_operation_ctx *ctx)
+{
+ struct ttm_buffer_object *ttm_bo = &bo->ttm;
+ struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+ int ret;
+
+ /*
+ * If this starts to call into many components, consider
+ * using a notification chain here.
+ */
+
+ if (xe_bo_is_pinned(bo))
+ return -EINVAL;
+
+ xe_bo_vunmap(bo);
+ ret = xe_bo_trigger_rebind(xe, bo, ctx);
+ if (ret)
+ return ret;
+
+ /* Don't call move_notify() for imported dma-bufs. */
+ if (ttm_bo->base.dma_buf && !ttm_bo->base.import_attach)
+ dma_buf_move_notify(ttm_bo->base.dma_buf);
+
+ return 0;
+}
+
+static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
+ struct ttm_operation_ctx *ctx,
+ struct ttm_resource *new_mem,
+ struct ttm_place *hop)
+{
+ struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+ struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
+ struct ttm_resource *old_mem = ttm_bo->resource;
+ u32 old_mem_type = old_mem ? old_mem->mem_type : XE_PL_SYSTEM;
+ struct ttm_tt *ttm = ttm_bo->ttm;
+ struct xe_migrate *migrate = NULL;
+ struct dma_fence *fence;
+ bool move_lacks_source;
+ bool tt_has_data;
+ bool needs_clear;
+ bool handle_system_ccs = (!IS_DGFX(xe) && xe_bo_needs_ccs_pages(bo) &&
+ ttm && ttm_tt_is_populated(ttm)) ? true : false;
+ int ret = 0;
+ /* Bo creation path, moving to system or TT. */
+ if ((!old_mem && ttm) && !handle_system_ccs) {
+ ttm_bo_move_null(ttm_bo, new_mem);
+ return 0;
+ }
+
+ if (ttm_bo->type == ttm_bo_type_sg) {
+ ret = xe_bo_move_notify(bo, ctx);
+ if (!ret)
+ ret = xe_bo_move_dmabuf(ttm_bo, new_mem);
+ goto out;
+ }
+
+ tt_has_data = ttm && (ttm_tt_is_populated(ttm) ||
+ (ttm->page_flags & TTM_TT_FLAG_SWAPPED));
+
+ move_lacks_source = handle_system_ccs ? (!bo->ccs_cleared) :
+ (!mem_type_is_vram(old_mem_type) && !tt_has_data);
+
+ needs_clear = (ttm && ttm->page_flags & TTM_TT_FLAG_ZERO_ALLOC) ||
+ (!ttm && ttm_bo->type == ttm_bo_type_device);
+
+ if ((move_lacks_source && !needs_clear)) {
+ ttm_bo_move_null(ttm_bo, new_mem);
+ goto out;
+ }
+
+ if (old_mem_type == XE_PL_SYSTEM && new_mem->mem_type == XE_PL_TT && !handle_system_ccs) {
+ ttm_bo_move_null(ttm_bo, new_mem);
+ goto out;
+ }
+
+ /*
+ * Failed multi-hop where the old_mem is still marked as
+ * TTM_PL_FLAG_TEMPORARY, should just be a dummy move.
+ */
+ if (old_mem_type == XE_PL_TT &&
+ new_mem->mem_type == XE_PL_TT) {
+ ttm_bo_move_null(ttm_bo, new_mem);
+ goto out;
+ }
+
+ if (!move_lacks_source && !xe_bo_is_pinned(bo)) {
+ ret = xe_bo_move_notify(bo, ctx);
+ if (ret)
+ goto out;
+ }
+
+ if (old_mem_type == XE_PL_TT &&
+ new_mem->mem_type == XE_PL_SYSTEM) {
+ long timeout = dma_resv_wait_timeout(ttm_bo->base.resv,
+ DMA_RESV_USAGE_BOOKKEEP,
+ true,
+ MAX_SCHEDULE_TIMEOUT);
+ if (timeout < 0) {
+ ret = timeout;
+ goto out;
+ }
+
+ if (!handle_system_ccs) {
+ ttm_bo_move_null(ttm_bo, new_mem);
+ goto out;
+ }
+ }
+
+ if (!move_lacks_source &&
+ ((old_mem_type == XE_PL_SYSTEM && resource_is_vram(new_mem)) ||
+ (mem_type_is_vram(old_mem_type) &&
+ new_mem->mem_type == XE_PL_SYSTEM))) {
+ hop->fpfn = 0;
+ hop->lpfn = 0;
+ hop->mem_type = XE_PL_TT;
+ hop->flags = TTM_PL_FLAG_TEMPORARY;
+ ret = -EMULTIHOP;
+ goto out;
+ }
+
+ if (bo->tile)
+ migrate = bo->tile->migrate;
+ else if (resource_is_vram(new_mem))
+ migrate = mem_type_to_migrate(xe, new_mem->mem_type);
+ else if (mem_type_is_vram(old_mem_type))
+ migrate = mem_type_to_migrate(xe, old_mem_type);
+ else
+ migrate = xe->tiles[0].migrate;
+
+ xe_assert(xe, migrate);
+
+ trace_xe_bo_move(bo);
+ xe_device_mem_access_get(xe);
+
+ if (xe_bo_is_pinned(bo) && !xe_bo_is_user(bo)) {
+ /*
+ * Kernel memory that is pinned should only be moved on suspend
+ * / resume, some of the pinned memory is required for the
+ * device to resume / use the GPU to move other evicted memory
+ * (user memory) around. This likely could be optimized a bit
+ * futher where we find the minimum set of pinned memory
+ * required for resume but for simplity doing a memcpy for all
+ * pinned memory.
+ */
+ ret = xe_bo_vmap(bo);
+ if (!ret) {
+ ret = ttm_bo_move_memcpy(ttm_bo, ctx, new_mem);
+
+ /* Create a new VMAP once kernel BO back in VRAM */
+ if (!ret && resource_is_vram(new_mem)) {
+ struct xe_mem_region *vram = res_to_mem_region(new_mem);
+ void *new_addr = vram->mapping +
+ (new_mem->start << PAGE_SHIFT);
+
+ if (XE_WARN_ON(new_mem->start == XE_BO_INVALID_OFFSET)) {
+ ret = -EINVAL;
+ xe_device_mem_access_put(xe);
+ goto out;
+ }
+
+ xe_assert(xe, new_mem->start ==
+ bo->placements->fpfn);
+
+ iosys_map_set_vaddr_iomem(&bo->vmap, new_addr);
+ }
+ }
+ } else {
+ if (move_lacks_source)
+ fence = xe_migrate_clear(migrate, bo, new_mem);
+ else
+ fence = xe_migrate_copy(migrate, bo, bo, old_mem,
+ new_mem, handle_system_ccs);
+ if (IS_ERR(fence)) {
+ ret = PTR_ERR(fence);
+ xe_device_mem_access_put(xe);
+ goto out;
+ }
+ if (!move_lacks_source) {
+ ret = ttm_bo_move_accel_cleanup(ttm_bo, fence, evict,
+ true, new_mem);
+ if (ret) {
+ dma_fence_wait(fence, false);
+ ttm_bo_move_null(ttm_bo, new_mem);
+ ret = 0;
+ }
+ } else {
+ /*
+ * ttm_bo_move_accel_cleanup() may blow up if
+ * bo->resource == NULL, so just attach the
+ * fence and set the new resource.
+ */
+ dma_resv_add_fence(ttm_bo->base.resv, fence,
+ DMA_RESV_USAGE_KERNEL);
+ ttm_bo_move_null(ttm_bo, new_mem);
+ }
+
+ dma_fence_put(fence);
+ }
+
+ xe_device_mem_access_put(xe);
+
+out:
+ return ret;
+
+}
+
+/**
+ * xe_bo_evict_pinned() - Evict a pinned VRAM object to system memory
+ * @bo: The buffer object to move.
+ *
+ * On successful completion, the object memory will be moved to sytem memory.
+ * This function blocks until the object has been fully moved.
+ *
+ * This is needed to for special handling of pinned VRAM object during
+ * suspend-resume.
+ *
+ * Return: 0 on success. Negative error code on failure.
+ */
+int xe_bo_evict_pinned(struct xe_bo *bo)
+{
+ struct ttm_place place = {
+ .mem_type = XE_PL_TT,
+ };
+ struct ttm_placement placement = {
+ .placement = &place,
+ .num_placement = 1,
+ };
+ struct ttm_operation_ctx ctx = {
+ .interruptible = false,
+ };
+ struct ttm_resource *new_mem;
+ int ret;
+
+ xe_bo_assert_held(bo);
+
+ if (WARN_ON(!bo->ttm.resource))
+ return -EINVAL;
+
+ if (WARN_ON(!xe_bo_is_pinned(bo)))
+ return -EINVAL;
+
+ if (WARN_ON(!xe_bo_is_vram(bo)))
+ return -EINVAL;
+
+ ret = ttm_bo_mem_space(&bo->ttm, &placement, &new_mem, &ctx);
+ if (ret)
+ return ret;
+
+ if (!bo->ttm.ttm) {
+ bo->ttm.ttm = xe_ttm_tt_create(&bo->ttm, 0);
+ if (!bo->ttm.ttm) {
+ ret = -ENOMEM;
+ goto err_res_free;
+ }
+ }
+
+ ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx);
+ if (ret)
+ goto err_res_free;
+
+ ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1);
+ if (ret)
+ goto err_res_free;
+
+ ret = xe_bo_move(&bo->ttm, false, &ctx, new_mem, NULL);
+ if (ret)
+ goto err_res_free;
+
+ dma_resv_wait_timeout(bo->ttm.base.resv, DMA_RESV_USAGE_KERNEL,
+ false, MAX_SCHEDULE_TIMEOUT);
+
+ return 0;
+
+err_res_free:
+ ttm_resource_free(&bo->ttm, &new_mem);
+ return ret;
+}
+
+/**
+ * xe_bo_restore_pinned() - Restore a pinned VRAM object
+ * @bo: The buffer object to move.
+ *
+ * On successful completion, the object memory will be moved back to VRAM.
+ * This function blocks until the object has been fully moved.
+ *
+ * This is needed to for special handling of pinned VRAM object during
+ * suspend-resume.
+ *
+ * Return: 0 on success. Negative error code on failure.
+ */
+int xe_bo_restore_pinned(struct xe_bo *bo)
+{
+ struct ttm_operation_ctx ctx = {
+ .interruptible = false,
+ };
+ struct ttm_resource *new_mem;
+ int ret;
+
+ xe_bo_assert_held(bo);
+
+ if (WARN_ON(!bo->ttm.resource))
+ return -EINVAL;
+
+ if (WARN_ON(!xe_bo_is_pinned(bo)))
+ return -EINVAL;
+
+ if (WARN_ON(xe_bo_is_vram(bo) || !bo->ttm.ttm))
+ return -EINVAL;
+
+ ret = ttm_bo_mem_space(&bo->ttm, &bo->placement, &new_mem, &ctx);
+ if (ret)
+ return ret;
+
+ ret = ttm_tt_populate(bo->ttm.bdev, bo->ttm.ttm, &ctx);
+ if (ret)
+ goto err_res_free;
+
+ ret = dma_resv_reserve_fences(bo->ttm.base.resv, 1);
+ if (ret)
+ goto err_res_free;
+
+ ret = xe_bo_move(&bo->ttm, false, &ctx, new_mem, NULL);
+ if (ret)
+ goto err_res_free;
+
+ dma_resv_wait_timeout(bo->ttm.base.resv, DMA_RESV_USAGE_KERNEL,
+ false, MAX_SCHEDULE_TIMEOUT);
+
+ return 0;
+
+err_res_free:
+ ttm_resource_free(&bo->ttm, &new_mem);
+ return ret;
+}
+
+static unsigned long xe_ttm_io_mem_pfn(struct ttm_buffer_object *ttm_bo,
+ unsigned long page_offset)
+{
+ struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
+ struct xe_res_cursor cursor;
+ struct xe_mem_region *vram;
+
+ if (ttm_bo->resource->mem_type == XE_PL_STOLEN)
+ return xe_ttm_stolen_io_offset(bo, page_offset << PAGE_SHIFT) >> PAGE_SHIFT;
+
+ vram = res_to_mem_region(ttm_bo->resource);
+ xe_res_first(ttm_bo->resource, (u64)page_offset << PAGE_SHIFT, 0, &cursor);
+ return (vram->io_start + cursor.start) >> PAGE_SHIFT;
+}
+
+static void __xe_bo_vunmap(struct xe_bo *bo);
+
+/*
+ * TODO: Move this function to TTM so we don't rely on how TTM does its
+ * locking, thereby abusing TTM internals.
+ */
+static bool xe_ttm_bo_lock_in_destructor(struct ttm_buffer_object *ttm_bo)
+{
+ struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+ bool locked;
+
+ xe_assert(xe, !kref_read(&ttm_bo->kref));
+
+ /*
+ * We can typically only race with TTM trylocking under the
+ * lru_lock, which will immediately be unlocked again since
+ * the ttm_bo refcount is zero at this point. So trylocking *should*
+ * always succeed here, as long as we hold the lru lock.
+ */
+ spin_lock(&ttm_bo->bdev->lru_lock);
+ locked = dma_resv_trylock(ttm_bo->base.resv);
+ spin_unlock(&ttm_bo->bdev->lru_lock);
+ xe_assert(xe, locked);
+
+ return locked;
+}
+
+static void xe_ttm_bo_release_notify(struct ttm_buffer_object *ttm_bo)
+{
+ struct dma_resv_iter cursor;
+ struct dma_fence *fence;
+ struct dma_fence *replacement = NULL;
+ struct xe_bo *bo;
+
+ if (!xe_bo_is_xe_bo(ttm_bo))
+ return;
+
+ bo = ttm_to_xe_bo(ttm_bo);
+ xe_assert(xe_bo_device(bo), !(bo->created && kref_read(&ttm_bo->base.refcount)));
+
+ /*
+ * Corner case where TTM fails to allocate memory and this BOs resv
+ * still points the VMs resv
+ */
+ if (ttm_bo->base.resv != &ttm_bo->base._resv)
+ return;
+
+ if (!xe_ttm_bo_lock_in_destructor(ttm_bo))
+ return;
+
+ /*
+ * Scrub the preempt fences if any. The unbind fence is already
+ * attached to the resv.
+ * TODO: Don't do this for external bos once we scrub them after
+ * unbind.
+ */
+ dma_resv_for_each_fence(&cursor, ttm_bo->base.resv,
+ DMA_RESV_USAGE_BOOKKEEP, fence) {
+ if (xe_fence_is_xe_preempt(fence) &&
+ !dma_fence_is_signaled(fence)) {
+ if (!replacement)
+ replacement = dma_fence_get_stub();
+
+ dma_resv_replace_fences(ttm_bo->base.resv,
+ fence->context,
+ replacement,
+ DMA_RESV_USAGE_BOOKKEEP);
+ }
+ }
+ dma_fence_put(replacement);
+
+ dma_resv_unlock(ttm_bo->base.resv);
+}
+
+static void xe_ttm_bo_delete_mem_notify(struct ttm_buffer_object *ttm_bo)
+{
+ if (!xe_bo_is_xe_bo(ttm_bo))
+ return;
+
+ /*
+ * Object is idle and about to be destroyed. Release the
+ * dma-buf attachment.
+ */
+ if (ttm_bo->type == ttm_bo_type_sg && ttm_bo->sg) {
+ struct xe_ttm_tt *xe_tt = container_of(ttm_bo->ttm,
+ struct xe_ttm_tt, ttm);
+
+ dma_buf_unmap_attachment(ttm_bo->base.import_attach, ttm_bo->sg,
+ DMA_BIDIRECTIONAL);
+ ttm_bo->sg = NULL;
+ xe_tt->sg = NULL;
+ }
+}
+
+struct ttm_device_funcs xe_ttm_funcs = {
+ .ttm_tt_create = xe_ttm_tt_create,
+ .ttm_tt_populate = xe_ttm_tt_populate,
+ .ttm_tt_unpopulate = xe_ttm_tt_unpopulate,
+ .ttm_tt_destroy = xe_ttm_tt_destroy,
+ .evict_flags = xe_evict_flags,
+ .move = xe_bo_move,
+ .io_mem_reserve = xe_ttm_io_mem_reserve,
+ .io_mem_pfn = xe_ttm_io_mem_pfn,
+ .release_notify = xe_ttm_bo_release_notify,
+ .eviction_valuable = ttm_bo_eviction_valuable,
+ .delete_mem_notify = xe_ttm_bo_delete_mem_notify,
+};
+
+static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo)
+{
+ struct xe_bo *bo = ttm_to_xe_bo(ttm_bo);
+ struct xe_device *xe = ttm_to_xe_device(ttm_bo->bdev);
+
+ if (bo->ttm.base.import_attach)
+ drm_prime_gem_destroy(&bo->ttm.base, NULL);
+ drm_gem_object_release(&bo->ttm.base);
+
+ xe_assert(xe, list_empty(&ttm_bo->base.gpuva.list));
+
+ if (bo->ggtt_node.size)
+ xe_ggtt_remove_bo(bo->tile->mem.ggtt, bo);
+
+#ifdef CONFIG_PROC_FS
+ if (bo->client)
+ xe_drm_client_remove_bo(bo);
+#endif
+
+ if (bo->vm && xe_bo_is_user(bo))
+ xe_vm_put(bo->vm);
+
+ kfree(bo);
+}
+
+static void xe_gem_object_free(struct drm_gem_object *obj)
+{
+ /* Our BO reference counting scheme works as follows:
+ *
+ * The gem object kref is typically used throughout the driver,
+ * and the gem object holds a ttm_buffer_object refcount, so
+ * that when the last gem object reference is put, which is when
+ * we end up in this function, we put also that ttm_buffer_object
+ * refcount. Anything using gem interfaces is then no longer
+ * allowed to access the object in a way that requires a gem
+ * refcount, including locking the object.
+ *
+ * driver ttm callbacks is allowed to use the ttm_buffer_object
+ * refcount directly if needed.
+ */
+ __xe_bo_vunmap(gem_to_xe_bo(obj));
+ ttm_bo_put(container_of(obj, struct ttm_buffer_object, base));
+}
+
+static void xe_gem_object_close(struct drm_gem_object *obj,
+ struct drm_file *file_priv)
+{
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+
+ if (bo->vm && !xe_vm_in_fault_mode(bo->vm)) {
+ xe_assert(xe_bo_device(bo), xe_bo_is_user(bo));
+
+ xe_bo_lock(bo, false);
+ ttm_bo_set_bulk_move(&bo->ttm, NULL);
+ xe_bo_unlock(bo);
+ }
+}
+
+static bool should_migrate_to_system(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+
+ return xe_device_in_fault_mode(xe) && bo->props.cpu_atomic;
+}
+
+static vm_fault_t xe_gem_fault(struct vm_fault *vmf)
+{
+ struct ttm_buffer_object *tbo = vmf->vma->vm_private_data;
+ struct drm_device *ddev = tbo->base.dev;
+ vm_fault_t ret;
+ int idx, r = 0;
+
+ ret = ttm_bo_vm_reserve(tbo, vmf);
+ if (ret)
+ return ret;
+
+ if (drm_dev_enter(ddev, &idx)) {
+ struct xe_bo *bo = ttm_to_xe_bo(tbo);
+
+ trace_xe_bo_cpu_fault(bo);
+
+ if (should_migrate_to_system(bo)) {
+ r = xe_bo_migrate(bo, XE_PL_TT);
+ if (r == -EBUSY || r == -ERESTARTSYS || r == -EINTR)
+ ret = VM_FAULT_NOPAGE;
+ else if (r)
+ ret = VM_FAULT_SIGBUS;
+ }
+ if (!ret)
+ ret = ttm_bo_vm_fault_reserved(vmf,
+ vmf->vma->vm_page_prot,
+ TTM_BO_VM_NUM_PREFAULT);
+ drm_dev_exit(idx);
+ } else {
+ ret = ttm_bo_vm_dummy_page(vmf, vmf->vma->vm_page_prot);
+ }
+ if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
+ return ret;
+
+ dma_resv_unlock(tbo->base.resv);
+ return ret;
+}
+
+static const struct vm_operations_struct xe_gem_vm_ops = {
+ .fault = xe_gem_fault,
+ .open = ttm_bo_vm_open,
+ .close = ttm_bo_vm_close,
+ .access = ttm_bo_vm_access
+};
+
+static const struct drm_gem_object_funcs xe_gem_object_funcs = {
+ .free = xe_gem_object_free,
+ .close = xe_gem_object_close,
+ .mmap = drm_gem_ttm_mmap,
+ .export = xe_gem_prime_export,
+ .vm_ops = &xe_gem_vm_ops,
+};
+
+/**
+ * xe_bo_alloc - Allocate storage for a struct xe_bo
+ *
+ * This funcition is intended to allocate storage to be used for input
+ * to __xe_bo_create_locked(), in the case a pointer to the bo to be
+ * created is needed before the call to __xe_bo_create_locked().
+ * If __xe_bo_create_locked ends up never to be called, then the
+ * storage allocated with this function needs to be freed using
+ * xe_bo_free().
+ *
+ * Return: A pointer to an uninitialized struct xe_bo on success,
+ * ERR_PTR(-ENOMEM) on error.
+ */
+struct xe_bo *xe_bo_alloc(void)
+{
+ struct xe_bo *bo = kzalloc(sizeof(*bo), GFP_KERNEL);
+
+ if (!bo)
+ return ERR_PTR(-ENOMEM);
+
+ return bo;
+}
+
+/**
+ * xe_bo_free - Free storage allocated using xe_bo_alloc()
+ * @bo: The buffer object storage.
+ *
+ * Refer to xe_bo_alloc() documentation for valid use-cases.
+ */
+void xe_bo_free(struct xe_bo *bo)
+{
+ kfree(bo);
+}
+
+struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo,
+ struct xe_tile *tile, struct dma_resv *resv,
+ struct ttm_lru_bulk_move *bulk, size_t size,
+ u16 cpu_caching, enum ttm_bo_type type,
+ u32 flags)
+{
+ struct ttm_operation_ctx ctx = {
+ .interruptible = true,
+ .no_wait_gpu = false,
+ };
+ struct ttm_placement *placement;
+ uint32_t alignment;
+ size_t aligned_size;
+ int err;
+
+ /* Only kernel objects should set GT */
+ xe_assert(xe, !tile || type == ttm_bo_type_kernel);
+
+ if (XE_WARN_ON(!size)) {
+ xe_bo_free(bo);
+ return ERR_PTR(-EINVAL);
+ }
+
+ if (flags & (XE_BO_CREATE_VRAM_MASK | XE_BO_CREATE_STOLEN_BIT) &&
+ !(flags & XE_BO_CREATE_IGNORE_MIN_PAGE_SIZE_BIT) &&
+ xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) {
+ aligned_size = ALIGN(size, SZ_64K);
+ if (type != ttm_bo_type_device)
+ size = ALIGN(size, SZ_64K);
+ flags |= XE_BO_INTERNAL_64K;
+ alignment = SZ_64K >> PAGE_SHIFT;
+
+ } else {
+ aligned_size = ALIGN(size, SZ_4K);
+ flags &= ~XE_BO_INTERNAL_64K;
+ alignment = SZ_4K >> PAGE_SHIFT;
+ }
+
+ if (type == ttm_bo_type_device && aligned_size != size)
+ return ERR_PTR(-EINVAL);
+
+ if (!bo) {
+ bo = xe_bo_alloc();
+ if (IS_ERR(bo))
+ return bo;
+ }
+
+ bo->ccs_cleared = false;
+ bo->tile = tile;
+ bo->size = size;
+ bo->flags = flags;
+ bo->cpu_caching = cpu_caching;
+ bo->ttm.base.funcs = &xe_gem_object_funcs;
+ bo->props.preferred_mem_class = XE_BO_PROPS_INVALID;
+ bo->props.preferred_gt = XE_BO_PROPS_INVALID;
+ bo->props.preferred_mem_type = XE_BO_PROPS_INVALID;
+ bo->ttm.priority = XE_BO_PRIORITY_NORMAL;
+ INIT_LIST_HEAD(&bo->pinned_link);
+#ifdef CONFIG_PROC_FS
+ INIT_LIST_HEAD(&bo->client_link);
+#endif
+
+ drm_gem_private_object_init(&xe->drm, &bo->ttm.base, size);
+
+ if (resv) {
+ ctx.allow_res_evict = !(flags & XE_BO_CREATE_NO_RESV_EVICT);
+ ctx.resv = resv;
+ }
+
+ if (!(flags & XE_BO_FIXED_PLACEMENT_BIT)) {
+ err = __xe_bo_placement_for_flags(xe, bo, bo->flags);
+ if (WARN_ON(err)) {
+ xe_ttm_bo_destroy(&bo->ttm);
+ return ERR_PTR(err);
+ }
+ }
+
+ /* Defer populating type_sg bos */
+ placement = (type == ttm_bo_type_sg ||
+ bo->flags & XE_BO_DEFER_BACKING) ? &sys_placement :
+ &bo->placement;
+ err = ttm_bo_init_reserved(&xe->ttm, &bo->ttm, type,
+ placement, alignment,
+ &ctx, NULL, resv, xe_ttm_bo_destroy);
+ if (err)
+ return ERR_PTR(err);
+
+ /*
+ * The VRAM pages underneath are potentially still being accessed by the
+ * GPU, as per async GPU clearing and async evictions. However TTM makes
+ * sure to add any corresponding move/clear fences into the objects
+ * dma-resv using the DMA_RESV_USAGE_KERNEL slot.
+ *
+ * For KMD internal buffers we don't care about GPU clearing, however we
+ * still need to handle async evictions, where the VRAM is still being
+ * accessed by the GPU. Most internal callers are not expecting this,
+ * since they are missing the required synchronisation before accessing
+ * the memory. To keep things simple just sync wait any kernel fences
+ * here, if the buffer is designated KMD internal.
+ *
+ * For normal userspace objects we should already have the required
+ * pipelining or sync waiting elsewhere, since we already have to deal
+ * with things like async GPU clearing.
+ */
+ if (type == ttm_bo_type_kernel) {
+ long timeout = dma_resv_wait_timeout(bo->ttm.base.resv,
+ DMA_RESV_USAGE_KERNEL,
+ ctx.interruptible,
+ MAX_SCHEDULE_TIMEOUT);
+
+ if (timeout < 0) {
+ if (!resv)
+ dma_resv_unlock(bo->ttm.base.resv);
+ xe_bo_put(bo);
+ return ERR_PTR(timeout);
+ }
+ }
+
+ bo->created = true;
+ if (bulk)
+ ttm_bo_set_bulk_move(&bo->ttm, bulk);
+ else
+ ttm_bo_move_to_lru_tail_unlocked(&bo->ttm);
+
+ return bo;
+}
+
+static int __xe_bo_fixed_placement(struct xe_device *xe,
+ struct xe_bo *bo,
+ u32 flags,
+ u64 start, u64 end, u64 size)
+{
+ struct ttm_place *place = bo->placements;
+
+ if (flags & (XE_BO_CREATE_USER_BIT|XE_BO_CREATE_SYSTEM_BIT))
+ return -EINVAL;
+
+ place->flags = TTM_PL_FLAG_CONTIGUOUS;
+ place->fpfn = start >> PAGE_SHIFT;
+ place->lpfn = end >> PAGE_SHIFT;
+
+ switch (flags & (XE_BO_CREATE_STOLEN_BIT | XE_BO_CREATE_VRAM_MASK)) {
+ case XE_BO_CREATE_VRAM0_BIT:
+ place->mem_type = XE_PL_VRAM0;
+ break;
+ case XE_BO_CREATE_VRAM1_BIT:
+ place->mem_type = XE_PL_VRAM1;
+ break;
+ case XE_BO_CREATE_STOLEN_BIT:
+ place->mem_type = XE_PL_STOLEN;
+ break;
+
+ default:
+ /* 0 or multiple of the above set */
+ return -EINVAL;
+ }
+
+ bo->placement = (struct ttm_placement) {
+ .num_placement = 1,
+ .placement = place,
+ .num_busy_placement = 1,
+ .busy_placement = place,
+ };
+
+ return 0;
+}
+
+static struct xe_bo *
+__xe_bo_create_locked(struct xe_device *xe,
+ struct xe_tile *tile, struct xe_vm *vm,
+ size_t size, u64 start, u64 end,
+ u16 cpu_caching, enum ttm_bo_type type, u32 flags)
+{
+ struct xe_bo *bo = NULL;
+ int err;
+
+ if (vm)
+ xe_vm_assert_held(vm);
+
+ if (start || end != ~0ULL) {
+ bo = xe_bo_alloc();
+ if (IS_ERR(bo))
+ return bo;
+
+ flags |= XE_BO_FIXED_PLACEMENT_BIT;
+ err = __xe_bo_fixed_placement(xe, bo, flags, start, end, size);
+ if (err) {
+ xe_bo_free(bo);
+ return ERR_PTR(err);
+ }
+ }
+
+ bo = ___xe_bo_create_locked(xe, bo, tile, vm ? xe_vm_resv(vm) : NULL,
+ vm && !xe_vm_in_fault_mode(vm) &&
+ flags & XE_BO_CREATE_USER_BIT ?
+ &vm->lru_bulk_move : NULL, size,
+ cpu_caching, type, flags);
+ if (IS_ERR(bo))
+ return bo;
+
+ /*
+ * Note that instead of taking a reference no the drm_gpuvm_resv_bo(),
+ * to ensure the shared resv doesn't disappear under the bo, the bo
+ * will keep a reference to the vm, and avoid circular references
+ * by having all the vm's bo refereferences released at vm close
+ * time.
+ */
+ if (vm && xe_bo_is_user(bo))
+ xe_vm_get(vm);
+ bo->vm = vm;
+
+ if (bo->flags & XE_BO_CREATE_GGTT_BIT) {
+ if (!tile && flags & XE_BO_CREATE_STOLEN_BIT)
+ tile = xe_device_get_root_tile(xe);
+
+ xe_assert(xe, tile);
+
+ if (flags & XE_BO_FIXED_PLACEMENT_BIT) {
+ err = xe_ggtt_insert_bo_at(tile->mem.ggtt, bo,
+ start + bo->size, U64_MAX);
+ } else {
+ err = xe_ggtt_insert_bo(tile->mem.ggtt, bo);
+ }
+ if (err)
+ goto err_unlock_put_bo;
+ }
+
+ return bo;
+
+err_unlock_put_bo:
+ __xe_bo_unset_bulk_move(bo);
+ xe_bo_unlock_vm_held(bo);
+ xe_bo_put(bo);
+ return ERR_PTR(err);
+}
+
+struct xe_bo *
+xe_bo_create_locked_range(struct xe_device *xe,
+ struct xe_tile *tile, struct xe_vm *vm,
+ size_t size, u64 start, u64 end,
+ enum ttm_bo_type type, u32 flags)
+{
+ return __xe_bo_create_locked(xe, tile, vm, size, start, end, 0, type, flags);
+}
+
+struct xe_bo *xe_bo_create_locked(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags)
+{
+ return __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL, 0, type, flags);
+}
+
+struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ u16 cpu_caching,
+ enum ttm_bo_type type,
+ u32 flags)
+{
+ struct xe_bo *bo = __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL,
+ cpu_caching, type,
+ flags | XE_BO_CREATE_USER_BIT);
+ if (!IS_ERR(bo))
+ xe_bo_unlock_vm_held(bo);
+
+ return bo;
+}
+
+struct xe_bo *xe_bo_create(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags)
+{
+ struct xe_bo *bo = xe_bo_create_locked(xe, tile, vm, size, type, flags);
+
+ if (!IS_ERR(bo))
+ xe_bo_unlock_vm_held(bo);
+
+ return bo;
+}
+
+struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm,
+ size_t size, u64 offset,
+ enum ttm_bo_type type, u32 flags)
+{
+ struct xe_bo *bo;
+ int err;
+ u64 start = offset == ~0ull ? 0 : offset;
+ u64 end = offset == ~0ull ? offset : start + size;
+
+ if (flags & XE_BO_CREATE_STOLEN_BIT &&
+ xe_ttm_stolen_cpu_access_needs_ggtt(xe))
+ flags |= XE_BO_CREATE_GGTT_BIT;
+
+ bo = xe_bo_create_locked_range(xe, tile, vm, size, start, end, type,
+ flags | XE_BO_NEEDS_CPU_ACCESS);
+ if (IS_ERR(bo))
+ return bo;
+
+ err = xe_bo_pin(bo);
+ if (err)
+ goto err_put;
+
+ err = xe_bo_vmap(bo);
+ if (err)
+ goto err_unpin;
+
+ xe_bo_unlock_vm_held(bo);
+
+ return bo;
+
+err_unpin:
+ xe_bo_unpin(bo);
+err_put:
+ xe_bo_unlock_vm_held(bo);
+ xe_bo_put(bo);
+ return ERR_PTR(err);
+}
+
+struct xe_bo *xe_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags)
+{
+ return xe_bo_create_pin_map_at(xe, tile, vm, size, ~0ull, type, flags);
+}
+
+struct xe_bo *xe_bo_create_from_data(struct xe_device *xe, struct xe_tile *tile,
+ const void *data, size_t size,
+ enum ttm_bo_type type, u32 flags)
+{
+ struct xe_bo *bo = xe_bo_create_pin_map(xe, tile, NULL,
+ ALIGN(size, PAGE_SIZE),
+ type, flags);
+ if (IS_ERR(bo))
+ return bo;
+
+ xe_map_memcpy_to(xe, &bo->vmap, 0, data, size);
+
+ return bo;
+}
+
+static void __xe_bo_unpin_map_no_vm(struct drm_device *drm, void *arg)
+{
+ xe_bo_unpin_map_no_vm(arg);
+}
+
+struct xe_bo *xe_managed_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile,
+ size_t size, u32 flags)
+{
+ struct xe_bo *bo;
+ int ret;
+
+ bo = xe_bo_create_pin_map(xe, tile, NULL, size, ttm_bo_type_kernel, flags);
+ if (IS_ERR(bo))
+ return bo;
+
+ ret = drmm_add_action_or_reset(&xe->drm, __xe_bo_unpin_map_no_vm, bo);
+ if (ret)
+ return ERR_PTR(ret);
+
+ return bo;
+}
+
+struct xe_bo *xe_managed_bo_create_from_data(struct xe_device *xe, struct xe_tile *tile,
+ const void *data, size_t size, u32 flags)
+{
+ struct xe_bo *bo = xe_managed_bo_create_pin_map(xe, tile, ALIGN(size, PAGE_SIZE), flags);
+
+ if (IS_ERR(bo))
+ return bo;
+
+ xe_map_memcpy_to(xe, &bo->vmap, 0, data, size);
+
+ return bo;
+}
+
+/*
+ * XXX: This is in the VM bind data path, likely should calculate this once and
+ * store, with a recalculation if the BO is moved.
+ */
+uint64_t vram_region_gpu_offset(struct ttm_resource *res)
+{
+ struct xe_device *xe = ttm_to_xe_device(res->bo->bdev);
+
+ if (res->mem_type == XE_PL_STOLEN)
+ return xe_ttm_stolen_gpu_offset(xe);
+
+ return res_to_mem_region(res)->dpa_base;
+}
+
+/**
+ * xe_bo_pin_external - pin an external BO
+ * @bo: buffer object to be pinned
+ *
+ * Pin an external (not tied to a VM, can be exported via dma-buf / prime FD)
+ * BO. Unique call compared to xe_bo_pin as this function has it own set of
+ * asserts and code to ensure evict / restore on suspend / resume.
+ *
+ * Returns 0 for success, negative error code otherwise.
+ */
+int xe_bo_pin_external(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ int err;
+
+ xe_assert(xe, !bo->vm);
+ xe_assert(xe, xe_bo_is_user(bo));
+
+ if (!xe_bo_is_pinned(bo)) {
+ err = xe_bo_validate(bo, NULL, false);
+ if (err)
+ return err;
+
+ if (xe_bo_is_vram(bo)) {
+ spin_lock(&xe->pinned.lock);
+ list_add_tail(&bo->pinned_link,
+ &xe->pinned.external_vram);
+ spin_unlock(&xe->pinned.lock);
+ }
+ }
+
+ ttm_bo_pin(&bo->ttm);
+
+ /*
+ * FIXME: If we always use the reserve / unreserve functions for locking
+ * we do not need this.
+ */
+ ttm_bo_move_to_lru_tail_unlocked(&bo->ttm);
+
+ return 0;
+}
+
+int xe_bo_pin(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ int err;
+
+ /* We currently don't expect user BO to be pinned */
+ xe_assert(xe, !xe_bo_is_user(bo));
+
+ /* Pinned object must be in GGTT or have pinned flag */
+ xe_assert(xe, bo->flags & (XE_BO_CREATE_PINNED_BIT |
+ XE_BO_CREATE_GGTT_BIT));
+
+ /*
+ * No reason we can't support pinning imported dma-bufs we just don't
+ * expect to pin an imported dma-buf.
+ */
+ xe_assert(xe, !bo->ttm.base.import_attach);
+
+ /* We only expect at most 1 pin */
+ xe_assert(xe, !xe_bo_is_pinned(bo));
+
+ err = xe_bo_validate(bo, NULL, false);
+ if (err)
+ return err;
+
+ /*
+ * For pinned objects in on DGFX, which are also in vram, we expect
+ * these to be in contiguous VRAM memory. Required eviction / restore
+ * during suspend / resume (force restore to same physical address).
+ */
+ if (IS_DGFX(xe) && !(IS_ENABLED(CONFIG_DRM_XE_DEBUG) &&
+ bo->flags & XE_BO_INTERNAL_TEST)) {
+ struct ttm_place *place = &(bo->placements[0]);
+
+ if (mem_type_is_vram(place->mem_type)) {
+ xe_assert(xe, place->flags & TTM_PL_FLAG_CONTIGUOUS);
+
+ place->fpfn = (xe_bo_addr(bo, 0, PAGE_SIZE) -
+ vram_region_gpu_offset(bo->ttm.resource)) >> PAGE_SHIFT;
+ place->lpfn = place->fpfn + (bo->size >> PAGE_SHIFT);
+
+ spin_lock(&xe->pinned.lock);
+ list_add_tail(&bo->pinned_link, &xe->pinned.kernel_bo_present);
+ spin_unlock(&xe->pinned.lock);
+ }
+ }
+
+ ttm_bo_pin(&bo->ttm);
+
+ /*
+ * FIXME: If we always use the reserve / unreserve functions for locking
+ * we do not need this.
+ */
+ ttm_bo_move_to_lru_tail_unlocked(&bo->ttm);
+
+ return 0;
+}
+
+/**
+ * xe_bo_unpin_external - unpin an external BO
+ * @bo: buffer object to be unpinned
+ *
+ * Unpin an external (not tied to a VM, can be exported via dma-buf / prime FD)
+ * BO. Unique call compared to xe_bo_unpin as this function has it own set of
+ * asserts and code to ensure evict / restore on suspend / resume.
+ *
+ * Returns 0 for success, negative error code otherwise.
+ */
+void xe_bo_unpin_external(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+
+ xe_assert(xe, !bo->vm);
+ xe_assert(xe, xe_bo_is_pinned(bo));
+ xe_assert(xe, xe_bo_is_user(bo));
+
+ if (bo->ttm.pin_count == 1 && !list_empty(&bo->pinned_link)) {
+ spin_lock(&xe->pinned.lock);
+ list_del_init(&bo->pinned_link);
+ spin_unlock(&xe->pinned.lock);
+ }
+
+ ttm_bo_unpin(&bo->ttm);
+
+ /*
+ * FIXME: If we always use the reserve / unreserve functions for locking
+ * we do not need this.
+ */
+ ttm_bo_move_to_lru_tail_unlocked(&bo->ttm);
+}
+
+void xe_bo_unpin(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+
+ xe_assert(xe, !bo->ttm.base.import_attach);
+ xe_assert(xe, xe_bo_is_pinned(bo));
+
+ if (IS_DGFX(xe) && !(IS_ENABLED(CONFIG_DRM_XE_DEBUG) &&
+ bo->flags & XE_BO_INTERNAL_TEST)) {
+ struct ttm_place *place = &(bo->placements[0]);
+
+ if (mem_type_is_vram(place->mem_type)) {
+ xe_assert(xe, !list_empty(&bo->pinned_link));
+
+ spin_lock(&xe->pinned.lock);
+ list_del_init(&bo->pinned_link);
+ spin_unlock(&xe->pinned.lock);
+ }
+ }
+
+ ttm_bo_unpin(&bo->ttm);
+}
+
+/**
+ * xe_bo_validate() - Make sure the bo is in an allowed placement
+ * @bo: The bo,
+ * @vm: Pointer to a the vm the bo shares a locked dma_resv object with, or
+ * NULL. Used together with @allow_res_evict.
+ * @allow_res_evict: Whether it's allowed to evict bos sharing @vm's
+ * reservation object.
+ *
+ * Make sure the bo is in allowed placement, migrating it if necessary. If
+ * needed, other bos will be evicted. If bos selected for eviction shares
+ * the @vm's reservation object, they can be evicted iff @allow_res_evict is
+ * set to true, otherwise they will be bypassed.
+ *
+ * Return: 0 on success, negative error code on failure. May return
+ * -EINTR or -ERESTARTSYS if internal waits are interrupted by a signal.
+ */
+int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict)
+{
+ struct ttm_operation_ctx ctx = {
+ .interruptible = true,
+ .no_wait_gpu = false,
+ };
+
+ if (vm) {
+ lockdep_assert_held(&vm->lock);
+ xe_vm_assert_held(vm);
+
+ ctx.allow_res_evict = allow_res_evict;
+ ctx.resv = xe_vm_resv(vm);
+ }
+
+ return ttm_bo_validate(&bo->ttm, &bo->placement, &ctx);
+}
+
+bool xe_bo_is_xe_bo(struct ttm_buffer_object *bo)
+{
+ if (bo->destroy == &xe_ttm_bo_destroy)
+ return true;
+
+ return false;
+}
+
+/*
+ * Resolve a BO address. There is no assert to check if the proper lock is held
+ * so it should only be used in cases where it is not fatal to get the wrong
+ * address, such as printing debug information, but not in cases where memory is
+ * written based on this result.
+ */
+dma_addr_t __xe_bo_addr(struct xe_bo *bo, u64 offset, size_t page_size)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ struct xe_res_cursor cur;
+ u64 page;
+
+ xe_assert(xe, page_size <= PAGE_SIZE);
+ page = offset >> PAGE_SHIFT;
+ offset &= (PAGE_SIZE - 1);
+
+ if (!xe_bo_is_vram(bo) && !xe_bo_is_stolen(bo)) {
+ xe_assert(xe, bo->ttm.ttm);
+
+ xe_res_first_sg(xe_bo_sg(bo), page << PAGE_SHIFT,
+ page_size, &cur);
+ return xe_res_dma(&cur) + offset;
+ } else {
+ struct xe_res_cursor cur;
+
+ xe_res_first(bo->ttm.resource, page << PAGE_SHIFT,
+ page_size, &cur);
+ return cur.start + offset + vram_region_gpu_offset(bo->ttm.resource);
+ }
+}
+
+dma_addr_t xe_bo_addr(struct xe_bo *bo, u64 offset, size_t page_size)
+{
+ if (!READ_ONCE(bo->ttm.pin_count))
+ xe_bo_assert_held(bo);
+ return __xe_bo_addr(bo, offset, page_size);
+}
+
+int xe_bo_vmap(struct xe_bo *bo)
+{
+ void *virtual;
+ bool is_iomem;
+ int ret;
+
+ xe_bo_assert_held(bo);
+
+ if (!(bo->flags & XE_BO_NEEDS_CPU_ACCESS))
+ return -EINVAL;
+
+ if (!iosys_map_is_null(&bo->vmap))
+ return 0;
+
+ /*
+ * We use this more or less deprecated interface for now since
+ * ttm_bo_vmap() doesn't offer the optimization of kmapping
+ * single page bos, which is done here.
+ * TODO: Fix up ttm_bo_vmap to do that, or fix up ttm_bo_kmap
+ * to use struct iosys_map.
+ */
+ ret = ttm_bo_kmap(&bo->ttm, 0, bo->size >> PAGE_SHIFT, &bo->kmap);
+ if (ret)
+ return ret;
+
+ virtual = ttm_kmap_obj_virtual(&bo->kmap, &is_iomem);
+ if (is_iomem)
+ iosys_map_set_vaddr_iomem(&bo->vmap, (void __iomem *)virtual);
+ else
+ iosys_map_set_vaddr(&bo->vmap, virtual);
+
+ return 0;
+}
+
+static void __xe_bo_vunmap(struct xe_bo *bo)
+{
+ if (!iosys_map_is_null(&bo->vmap)) {
+ iosys_map_clear(&bo->vmap);
+ ttm_bo_kunmap(&bo->kmap);
+ }
+}
+
+void xe_bo_vunmap(struct xe_bo *bo)
+{
+ xe_bo_assert_held(bo);
+ __xe_bo_vunmap(bo);
+}
+
+int xe_gem_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_gem_create *args = data;
+ struct xe_vm *vm = NULL;
+ struct xe_bo *bo;
+ unsigned int bo_flags;
+ u32 handle;
+ int err;
+
+ if (XE_IOCTL_DBG(xe, args->extensions) ||
+ XE_IOCTL_DBG(xe, args->pad[0] || args->pad[1] || args->pad[2]) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ /* at least one valid memory placement must be specified */
+ if (XE_IOCTL_DBG(xe, (args->placement & ~xe->info.mem_region_mask) ||
+ !args->placement))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags &
+ ~(DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING |
+ DRM_XE_GEM_CREATE_FLAG_SCANOUT |
+ DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM)))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->handle))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !args->size))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->size > SIZE_MAX))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->size & ~PAGE_MASK))
+ return -EINVAL;
+
+ bo_flags = 0;
+ if (args->flags & DRM_XE_GEM_CREATE_FLAG_DEFER_BACKING)
+ bo_flags |= XE_BO_DEFER_BACKING;
+
+ if (args->flags & DRM_XE_GEM_CREATE_FLAG_SCANOUT)
+ bo_flags |= XE_BO_SCANOUT_BIT;
+
+ bo_flags |= args->placement << (ffs(XE_BO_CREATE_SYSTEM_BIT) - 1);
+
+ if (args->flags & DRM_XE_GEM_CREATE_FLAG_NEEDS_VISIBLE_VRAM) {
+ if (XE_IOCTL_DBG(xe, !(bo_flags & XE_BO_CREATE_VRAM_MASK)))
+ return -EINVAL;
+
+ bo_flags |= XE_BO_NEEDS_CPU_ACCESS;
+ }
+
+ if (XE_IOCTL_DBG(xe, !args->cpu_caching ||
+ args->cpu_caching > DRM_XE_GEM_CPU_CACHING_WC))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_CREATE_VRAM_MASK &&
+ args->cpu_caching != DRM_XE_GEM_CPU_CACHING_WC))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, bo_flags & XE_BO_SCANOUT_BIT &&
+ args->cpu_caching == DRM_XE_GEM_CPU_CACHING_WB))
+ return -EINVAL;
+
+ if (args->vm_id) {
+ vm = xe_vm_lookup(xef, args->vm_id);
+ if (XE_IOCTL_DBG(xe, !vm))
+ return -ENOENT;
+ err = xe_vm_lock(vm, true);
+ if (err)
+ goto out_vm;
+ }
+
+ bo = xe_bo_create_user(xe, NULL, vm, args->size, args->cpu_caching,
+ ttm_bo_type_device, bo_flags);
+
+ if (vm)
+ xe_vm_unlock(vm);
+
+ if (IS_ERR(bo)) {
+ err = PTR_ERR(bo);
+ goto out_vm;
+ }
+
+ err = drm_gem_handle_create(file, &bo->ttm.base, &handle);
+ if (err)
+ goto out_bulk;
+
+ args->handle = handle;
+ goto out_put;
+
+out_bulk:
+ if (vm && !xe_vm_in_fault_mode(vm)) {
+ xe_vm_lock(vm, false);
+ __xe_bo_unset_bulk_move(bo);
+ xe_vm_unlock(vm);
+ }
+out_put:
+ xe_bo_put(bo);
+out_vm:
+ if (vm)
+ xe_vm_put(vm);
+
+ return err;
+}
+
+int xe_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct drm_xe_gem_mmap_offset *args = data;
+ struct drm_gem_object *gem_obj;
+
+ if (XE_IOCTL_DBG(xe, args->extensions) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags))
+ return -EINVAL;
+
+ gem_obj = drm_gem_object_lookup(file, args->handle);
+ if (XE_IOCTL_DBG(xe, !gem_obj))
+ return -ENOENT;
+
+ /* The mmap offset was set up at BO allocation time. */
+ args->offset = drm_vma_node_offset_addr(&gem_obj->vma_node);
+
+ xe_bo_put(gem_to_xe_bo(gem_obj));
+ return 0;
+}
+
+/**
+ * xe_bo_lock() - Lock the buffer object's dma_resv object
+ * @bo: The struct xe_bo whose lock is to be taken
+ * @intr: Whether to perform any wait interruptible
+ *
+ * Locks the buffer object's dma_resv object. If the buffer object is
+ * pointing to a shared dma_resv object, that shared lock is locked.
+ *
+ * Return: 0 on success, -EINTR if @intr is true and the wait for a
+ * contended lock was interrupted. If @intr is set to false, the
+ * function always returns 0.
+ */
+int xe_bo_lock(struct xe_bo *bo, bool intr)
+{
+ if (intr)
+ return dma_resv_lock_interruptible(bo->ttm.base.resv, NULL);
+
+ dma_resv_lock(bo->ttm.base.resv, NULL);
+
+ return 0;
+}
+
+/**
+ * xe_bo_unlock() - Unlock the buffer object's dma_resv object
+ * @bo: The struct xe_bo whose lock is to be released.
+ *
+ * Unlock a buffer object lock that was locked by xe_bo_lock().
+ */
+void xe_bo_unlock(struct xe_bo *bo)
+{
+ dma_resv_unlock(bo->ttm.base.resv);
+}
+
+/**
+ * xe_bo_can_migrate - Whether a buffer object likely can be migrated
+ * @bo: The buffer object to migrate
+ * @mem_type: The TTM memory type intended to migrate to
+ *
+ * Check whether the buffer object supports migration to the
+ * given memory type. Note that pinning may affect the ability to migrate as
+ * returned by this function.
+ *
+ * This function is primarily intended as a helper for checking the
+ * possibility to migrate buffer objects and can be called without
+ * the object lock held.
+ *
+ * Return: true if migration is possible, false otherwise.
+ */
+bool xe_bo_can_migrate(struct xe_bo *bo, u32 mem_type)
+{
+ unsigned int cur_place;
+
+ if (bo->ttm.type == ttm_bo_type_kernel)
+ return true;
+
+ if (bo->ttm.type == ttm_bo_type_sg)
+ return false;
+
+ for (cur_place = 0; cur_place < bo->placement.num_placement;
+ cur_place++) {
+ if (bo->placements[cur_place].mem_type == mem_type)
+ return true;
+ }
+
+ return false;
+}
+
+static void xe_place_from_ttm_type(u32 mem_type, struct ttm_place *place)
+{
+ memset(place, 0, sizeof(*place));
+ place->mem_type = mem_type;
+}
+
+/**
+ * xe_bo_migrate - Migrate an object to the desired region id
+ * @bo: The buffer object to migrate.
+ * @mem_type: The TTM region type to migrate to.
+ *
+ * Attempt to migrate the buffer object to the desired memory region. The
+ * buffer object may not be pinned, and must be locked.
+ * On successful completion, the object memory type will be updated,
+ * but an async migration task may not have completed yet, and to
+ * accomplish that, the object's kernel fences must be signaled with
+ * the object lock held.
+ *
+ * Return: 0 on success. Negative error code on failure. In particular may
+ * return -EINTR or -ERESTARTSYS if signal pending.
+ */
+int xe_bo_migrate(struct xe_bo *bo, u32 mem_type)
+{
+ struct xe_device *xe = ttm_to_xe_device(bo->ttm.bdev);
+ struct ttm_operation_ctx ctx = {
+ .interruptible = true,
+ .no_wait_gpu = false,
+ };
+ struct ttm_placement placement;
+ struct ttm_place requested;
+
+ xe_bo_assert_held(bo);
+
+ if (bo->ttm.resource->mem_type == mem_type)
+ return 0;
+
+ if (xe_bo_is_pinned(bo))
+ return -EBUSY;
+
+ if (!xe_bo_can_migrate(bo, mem_type))
+ return -EINVAL;
+
+ xe_place_from_ttm_type(mem_type, &requested);
+ placement.num_placement = 1;
+ placement.num_busy_placement = 1;
+ placement.placement = &requested;
+ placement.busy_placement = &requested;
+
+ /*
+ * Stolen needs to be handled like below VRAM handling if we ever need
+ * to support it.
+ */
+ drm_WARN_ON(&xe->drm, mem_type == XE_PL_STOLEN);
+
+ if (mem_type_is_vram(mem_type)) {
+ u32 c = 0;
+
+ add_vram(xe, bo, &requested, bo->flags, mem_type, &c);
+ }
+
+ return ttm_bo_validate(&bo->ttm, &placement, &ctx);
+}
+
+/**
+ * xe_bo_evict - Evict an object to evict placement
+ * @bo: The buffer object to migrate.
+ * @force_alloc: Set force_alloc in ttm_operation_ctx
+ *
+ * On successful completion, the object memory will be moved to evict
+ * placement. Ths function blocks until the object has been fully moved.
+ *
+ * Return: 0 on success. Negative error code on failure.
+ */
+int xe_bo_evict(struct xe_bo *bo, bool force_alloc)
+{
+ struct ttm_operation_ctx ctx = {
+ .interruptible = false,
+ .no_wait_gpu = false,
+ .force_alloc = force_alloc,
+ };
+ struct ttm_placement placement;
+ int ret;
+
+ xe_evict_flags(&bo->ttm, &placement);
+ ret = ttm_bo_validate(&bo->ttm, &placement, &ctx);
+ if (ret)
+ return ret;
+
+ dma_resv_wait_timeout(bo->ttm.base.resv, DMA_RESV_USAGE_KERNEL,
+ false, MAX_SCHEDULE_TIMEOUT);
+
+ return 0;
+}
+
+/**
+ * xe_bo_needs_ccs_pages - Whether a bo needs to back up CCS pages when
+ * placed in system memory.
+ * @bo: The xe_bo
+ *
+ * Return: true if extra pages need to be allocated, false otherwise.
+ */
+bool xe_bo_needs_ccs_pages(struct xe_bo *bo)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+
+ if (!xe_device_has_flat_ccs(xe) || bo->ttm.type != ttm_bo_type_device)
+ return false;
+
+ /* On discrete GPUs, if the GPU can access this buffer from
+ * system memory (i.e., it allows XE_PL_TT placement), FlatCCS
+ * can't be used since there's no CCS storage associated with
+ * non-VRAM addresses.
+ */
+ if (IS_DGFX(xe) && (bo->flags & XE_BO_CREATE_SYSTEM_BIT))
+ return false;
+
+ return true;
+}
+
+/**
+ * __xe_bo_release_dummy() - Dummy kref release function
+ * @kref: The embedded struct kref.
+ *
+ * Dummy release function for xe_bo_put_deferred(). Keep off.
+ */
+void __xe_bo_release_dummy(struct kref *kref)
+{
+}
+
+/**
+ * xe_bo_put_commit() - Put bos whose put was deferred by xe_bo_put_deferred().
+ * @deferred: The lockless list used for the call to xe_bo_put_deferred().
+ *
+ * Puts all bos whose put was deferred by xe_bo_put_deferred().
+ * The @deferred list can be either an onstack local list or a global
+ * shared list used by a workqueue.
+ */
+void xe_bo_put_commit(struct llist_head *deferred)
+{
+ struct llist_node *freed;
+ struct xe_bo *bo, *next;
+
+ if (!deferred)
+ return;
+
+ freed = llist_del_all(deferred);
+ if (!freed)
+ return;
+
+ llist_for_each_entry_safe(bo, next, freed, freed)
+ drm_gem_object_free(&bo->ttm.base.refcount);
+}
+
+/**
+ * xe_bo_dumb_create - Create a dumb bo as backing for a fb
+ * @file_priv: ...
+ * @dev: ...
+ * @args: ...
+ *
+ * See dumb_create() hook in include/drm/drm_drv.h
+ *
+ * Return: ...
+ */
+int xe_bo_dumb_create(struct drm_file *file_priv,
+ struct drm_device *dev,
+ struct drm_mode_create_dumb *args)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_bo *bo;
+ uint32_t handle;
+ int cpp = DIV_ROUND_UP(args->bpp, 8);
+ int err;
+ u32 page_size = max_t(u32, PAGE_SIZE,
+ xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ? SZ_64K : SZ_4K);
+
+ args->pitch = ALIGN(args->width * cpp, 64);
+ args->size = ALIGN(mul_u32_u32(args->pitch, args->height),
+ page_size);
+
+ bo = xe_bo_create_user(xe, NULL, NULL, args->size,
+ DRM_XE_GEM_CPU_CACHING_WC,
+ ttm_bo_type_device,
+ XE_BO_CREATE_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) |
+ XE_BO_CREATE_USER_BIT | XE_BO_SCANOUT_BIT |
+ XE_BO_NEEDS_CPU_ACCESS);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ err = drm_gem_handle_create(file_priv, &bo->ttm.base, &handle);
+ /* drop reference from allocate - handle holds it now */
+ drm_gem_object_put(&bo->ttm.base);
+ if (!err)
+ args->handle = handle;
+ return err;
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_bo.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
new file mode 100644
index 000000000000..9b1279aca127
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -0,0 +1,355 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_BO_H_
+#define _XE_BO_H_
+
+#include <drm/ttm/ttm_tt.h>
+
+#include "xe_bo_types.h"
+#include "xe_macros.h"
+#include "xe_vm_types.h"
+#include "xe_vm.h"
+
+/**
+ * xe_vm_assert_held(vm) - Assert that the vm's reservation object is held.
+ * @vm: The vm
+ */
+#define xe_vm_assert_held(vm) dma_resv_assert_held(xe_vm_resv(vm))
+
+
+
+#define XE_DEFAULT_GTT_SIZE_MB 3072ULL /* 3GB by default */
+
+#define XE_BO_CREATE_USER_BIT BIT(0)
+/* The bits below need to be contiguous, or things break */
+#define XE_BO_CREATE_SYSTEM_BIT BIT(1)
+#define XE_BO_CREATE_VRAM0_BIT BIT(2)
+#define XE_BO_CREATE_VRAM1_BIT BIT(3)
+#define XE_BO_CREATE_VRAM_MASK (XE_BO_CREATE_VRAM0_BIT | \
+ XE_BO_CREATE_VRAM1_BIT)
+/* -- */
+#define XE_BO_CREATE_STOLEN_BIT BIT(4)
+#define XE_BO_CREATE_VRAM_IF_DGFX(tile) \
+ (IS_DGFX(tile_to_xe(tile)) ? XE_BO_CREATE_VRAM0_BIT << (tile)->id : \
+ XE_BO_CREATE_SYSTEM_BIT)
+#define XE_BO_CREATE_GGTT_BIT BIT(5)
+#define XE_BO_CREATE_IGNORE_MIN_PAGE_SIZE_BIT BIT(6)
+#define XE_BO_CREATE_PINNED_BIT BIT(7)
+#define XE_BO_CREATE_NO_RESV_EVICT BIT(8)
+#define XE_BO_DEFER_BACKING BIT(9)
+#define XE_BO_SCANOUT_BIT BIT(10)
+#define XE_BO_FIXED_PLACEMENT_BIT BIT(11)
+#define XE_BO_PAGETABLE BIT(12)
+#define XE_BO_NEEDS_CPU_ACCESS BIT(13)
+/* this one is trigger internally only */
+#define XE_BO_INTERNAL_TEST BIT(30)
+#define XE_BO_INTERNAL_64K BIT(31)
+
+#define XELPG_PPGTT_PTE_PAT3 BIT_ULL(62)
+#define XE2_PPGTT_PTE_PAT4 BIT_ULL(61)
+#define XE_PPGTT_PDE_PDPE_PAT2 BIT_ULL(12)
+#define XE_PPGTT_PTE_PAT2 BIT_ULL(7)
+#define XE_PPGTT_PTE_PAT1 BIT_ULL(4)
+#define XE_PPGTT_PTE_PAT0 BIT_ULL(3)
+
+#define XE_PTE_SHIFT 12
+#define XE_PAGE_SIZE (1 << XE_PTE_SHIFT)
+#define XE_PTE_MASK (XE_PAGE_SIZE - 1)
+#define XE_PDE_SHIFT (XE_PTE_SHIFT - 3)
+#define XE_PDES (1 << XE_PDE_SHIFT)
+#define XE_PDE_MASK (XE_PDES - 1)
+
+#define XE_64K_PTE_SHIFT 16
+#define XE_64K_PAGE_SIZE (1 << XE_64K_PTE_SHIFT)
+#define XE_64K_PTE_MASK (XE_64K_PAGE_SIZE - 1)
+#define XE_64K_PDE_MASK (XE_PDE_MASK >> 4)
+
+#define XE_PDE_PS_2M BIT_ULL(7)
+#define XE_PDPE_PS_1G BIT_ULL(7)
+#define XE_PDE_IPS_64K BIT_ULL(11)
+
+#define XE_GGTT_PTE_DM BIT_ULL(1)
+#define XE_USM_PPGTT_PTE_AE BIT_ULL(10)
+#define XE_PPGTT_PTE_DM BIT_ULL(11)
+#define XE_PDE_64K BIT_ULL(6)
+#define XE_PTE_PS64 BIT_ULL(8)
+#define XE_PTE_NULL BIT_ULL(9)
+
+#define XE_PAGE_PRESENT BIT_ULL(0)
+#define XE_PAGE_RW BIT_ULL(1)
+
+#define XE_PL_SYSTEM TTM_PL_SYSTEM
+#define XE_PL_TT TTM_PL_TT
+#define XE_PL_VRAM0 TTM_PL_VRAM
+#define XE_PL_VRAM1 (XE_PL_VRAM0 + 1)
+#define XE_PL_STOLEN (TTM_NUM_MEM_TYPES - 1)
+
+#define XE_BO_PROPS_INVALID (-1)
+
+struct sg_table;
+
+struct xe_bo *xe_bo_alloc(void);
+void xe_bo_free(struct xe_bo *bo);
+
+struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo,
+ struct xe_tile *tile, struct dma_resv *resv,
+ struct ttm_lru_bulk_move *bulk, size_t size,
+ u16 cpu_caching, enum ttm_bo_type type,
+ u32 flags);
+struct xe_bo *
+xe_bo_create_locked_range(struct xe_device *xe,
+ struct xe_tile *tile, struct xe_vm *vm,
+ size_t size, u64 start, u64 end,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_bo_create_locked(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_bo_create(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ u16 cpu_caching,
+ enum ttm_bo_type type,
+ u32 flags);
+struct xe_bo *xe_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm, size_t size, u64 offset,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_bo_create_from_data(struct xe_device *xe, struct xe_tile *tile,
+ const void *data, size_t size,
+ enum ttm_bo_type type, u32 flags);
+struct xe_bo *xe_managed_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile,
+ size_t size, u32 flags);
+struct xe_bo *xe_managed_bo_create_from_data(struct xe_device *xe, struct xe_tile *tile,
+ const void *data, size_t size, u32 flags);
+
+int xe_bo_placement_for_flags(struct xe_device *xe, struct xe_bo *bo,
+ u32 bo_flags);
+
+static inline struct xe_bo *ttm_to_xe_bo(const struct ttm_buffer_object *bo)
+{
+ return container_of(bo, struct xe_bo, ttm);
+}
+
+static inline struct xe_bo *gem_to_xe_bo(const struct drm_gem_object *obj)
+{
+ return container_of(obj, struct xe_bo, ttm.base);
+}
+
+#define xe_bo_device(bo) ttm_to_xe_device((bo)->ttm.bdev)
+
+static inline struct xe_bo *xe_bo_get(struct xe_bo *bo)
+{
+ if (bo)
+ drm_gem_object_get(&bo->ttm.base);
+
+ return bo;
+}
+
+static inline void xe_bo_put(struct xe_bo *bo)
+{
+ if (bo)
+ drm_gem_object_put(&bo->ttm.base);
+}
+
+static inline void __xe_bo_unset_bulk_move(struct xe_bo *bo)
+{
+ if (bo)
+ ttm_bo_set_bulk_move(&bo->ttm, NULL);
+}
+
+static inline void xe_bo_assert_held(struct xe_bo *bo)
+{
+ if (bo)
+ dma_resv_assert_held((bo)->ttm.base.resv);
+}
+
+int xe_bo_lock(struct xe_bo *bo, bool intr);
+
+void xe_bo_unlock(struct xe_bo *bo);
+
+static inline void xe_bo_unlock_vm_held(struct xe_bo *bo)
+{
+ if (bo) {
+ XE_WARN_ON(bo->vm && bo->ttm.base.resv != xe_vm_resv(bo->vm));
+ if (bo->vm)
+ xe_vm_assert_held(bo->vm);
+ else
+ dma_resv_unlock(bo->ttm.base.resv);
+ }
+}
+
+int xe_bo_pin_external(struct xe_bo *bo);
+int xe_bo_pin(struct xe_bo *bo);
+void xe_bo_unpin_external(struct xe_bo *bo);
+void xe_bo_unpin(struct xe_bo *bo);
+int xe_bo_validate(struct xe_bo *bo, struct xe_vm *vm, bool allow_res_evict);
+
+static inline bool xe_bo_is_pinned(struct xe_bo *bo)
+{
+ return bo->ttm.pin_count;
+}
+
+static inline void xe_bo_unpin_map_no_vm(struct xe_bo *bo)
+{
+ if (likely(bo)) {
+ xe_bo_lock(bo, false);
+ xe_bo_unpin(bo);
+ xe_bo_unlock(bo);
+
+ xe_bo_put(bo);
+ }
+}
+
+bool xe_bo_is_xe_bo(struct ttm_buffer_object *bo);
+dma_addr_t __xe_bo_addr(struct xe_bo *bo, u64 offset, size_t page_size);
+dma_addr_t xe_bo_addr(struct xe_bo *bo, u64 offset, size_t page_size);
+
+static inline dma_addr_t
+xe_bo_main_addr(struct xe_bo *bo, size_t page_size)
+{
+ return xe_bo_addr(bo, 0, page_size);
+}
+
+static inline u32
+xe_bo_ggtt_addr(struct xe_bo *bo)
+{
+ XE_WARN_ON(bo->ggtt_node.size > bo->size);
+ XE_WARN_ON(bo->ggtt_node.start + bo->ggtt_node.size > (1ull << 32));
+ return bo->ggtt_node.start;
+}
+
+int xe_bo_vmap(struct xe_bo *bo);
+void xe_bo_vunmap(struct xe_bo *bo);
+
+bool mem_type_is_vram(u32 mem_type);
+bool xe_bo_is_vram(struct xe_bo *bo);
+bool xe_bo_is_stolen(struct xe_bo *bo);
+bool xe_bo_is_stolen_devmem(struct xe_bo *bo);
+uint64_t vram_region_gpu_offset(struct ttm_resource *res);
+
+bool xe_bo_can_migrate(struct xe_bo *bo, u32 mem_type);
+
+int xe_bo_migrate(struct xe_bo *bo, u32 mem_type);
+int xe_bo_evict(struct xe_bo *bo, bool force_alloc);
+
+int xe_bo_evict_pinned(struct xe_bo *bo);
+int xe_bo_restore_pinned(struct xe_bo *bo);
+
+extern struct ttm_device_funcs xe_ttm_funcs;
+
+int xe_gem_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_gem_mmap_offset_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_bo_dumb_create(struct drm_file *file_priv,
+ struct drm_device *dev,
+ struct drm_mode_create_dumb *args);
+
+bool xe_bo_needs_ccs_pages(struct xe_bo *bo);
+
+static inline size_t xe_bo_ccs_pages_start(struct xe_bo *bo)
+{
+ return PAGE_ALIGN(bo->ttm.base.size);
+}
+
+static inline bool xe_bo_has_pages(struct xe_bo *bo)
+{
+ if ((bo->ttm.ttm && ttm_tt_is_populated(bo->ttm.ttm)) ||
+ xe_bo_is_vram(bo))
+ return true;
+
+ return false;
+}
+
+void __xe_bo_release_dummy(struct kref *kref);
+
+/**
+ * xe_bo_put_deferred() - Put a buffer object with delayed final freeing
+ * @bo: The bo to put.
+ * @deferred: List to which to add the buffer object if we cannot put, or
+ * NULL if the function is to put unconditionally.
+ *
+ * Since the final freeing of an object includes both sleeping and (!)
+ * memory allocation in the dma_resv individualization, it's not ok
+ * to put an object from atomic context nor from within a held lock
+ * tainted by reclaim. In such situations we want to defer the final
+ * freeing until we've exited the restricting context, or in the worst
+ * case to a workqueue.
+ * This function either puts the object if possible without the refcount
+ * reaching zero, or adds it to the @deferred list if that was not possible.
+ * The caller needs to follow up with a call to xe_bo_put_commit() to actually
+ * put the bo iff this function returns true. It's safe to always
+ * follow up with a call to xe_bo_put_commit().
+ * TODO: It's TTM that is the villain here. Perhaps TTM should add an
+ * interface like this.
+ *
+ * Return: true if @bo was the first object put on the @freed list,
+ * false otherwise.
+ */
+static inline bool
+xe_bo_put_deferred(struct xe_bo *bo, struct llist_head *deferred)
+{
+ if (!deferred) {
+ xe_bo_put(bo);
+ return false;
+ }
+
+ if (!kref_put(&bo->ttm.base.refcount, __xe_bo_release_dummy))
+ return false;
+
+ return llist_add(&bo->freed, deferred);
+}
+
+void xe_bo_put_commit(struct llist_head *deferred);
+
+struct sg_table *xe_bo_sg(struct xe_bo *bo);
+
+/*
+ * xe_sg_segment_size() - Provides upper limit for sg segment size.
+ * @dev: device pointer
+ *
+ * Returns the maximum segment size for the 'struct scatterlist'
+ * elements.
+ */
+static inline unsigned int xe_sg_segment_size(struct device *dev)
+{
+ struct scatterlist __maybe_unused sg;
+ size_t max = BIT_ULL(sizeof(sg.length) * 8) - 1;
+
+ max = min_t(size_t, max, dma_max_mapping_size(dev));
+
+ /*
+ * The iommu_dma_map_sg() function ensures iova allocation doesn't
+ * cross dma segment boundary. It does so by padding some sg elements.
+ * This can cause overflow, ending up with sg->length being set to 0.
+ * Avoid this by ensuring maximum segment size is half of 'max'
+ * rounded down to PAGE_SIZE.
+ */
+ return round_down(max / 2, PAGE_SIZE);
+}
+
+#define i915_gem_object_flush_if_display(obj) ((void)(obj))
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+/**
+ * xe_bo_is_mem_type - Whether the bo currently resides in the given
+ * TTM memory type
+ * @bo: The bo to check.
+ * @mem_type: The TTM memory type.
+ *
+ * Return: true iff the bo resides in @mem_type, false otherwise.
+ */
+static inline bool xe_bo_is_mem_type(struct xe_bo *bo, u32 mem_type)
+{
+ xe_bo_assert_held(bo);
+ return bo->ttm.resource->mem_type == mem_type;
+}
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bo_doc.h b/drivers/gpu/drm/xe/xe_bo_doc.h
new file mode 100644
index 000000000000..f57d440cc95a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo_doc.h
@@ -0,0 +1,179 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_BO_DOC_H_
+#define _XE_BO_DOC_H_
+
+/**
+ * DOC: Buffer Objects (BO)
+ *
+ * BO management
+ * =============
+ *
+ * TTM manages (placement, eviction, etc...) all BOs in XE.
+ *
+ * BO creation
+ * ===========
+ *
+ * Create a chunk of memory which can be used by the GPU. Placement rules
+ * (sysmem or vram region) passed in upon creation. TTM handles placement of BO
+ * and can trigger eviction of other BOs to make space for the new BO.
+ *
+ * Kernel BOs
+ * ----------
+ *
+ * A kernel BO is created as part of driver load (e.g. uC firmware images, GuC
+ * ADS, etc...) or a BO created as part of a user operation which requires
+ * a kernel BO (e.g. engine state, memory for page tables, etc...). These BOs
+ * are typically mapped in the GGTT (any kernel BOs aside memory for page tables
+ * are in the GGTT), are pinned (can't move or be evicted at runtime), have a
+ * vmap (XE can access the memory via xe_map layer) and have contiguous physical
+ * memory.
+ *
+ * More details of why kernel BOs are pinned and contiguous below.
+ *
+ * User BOs
+ * --------
+ *
+ * A user BO is created via the DRM_IOCTL_XE_GEM_CREATE IOCTL. Once it is
+ * created the BO can be mmap'd (via DRM_IOCTL_XE_GEM_MMAP_OFFSET) for user
+ * access and it can be bound for GPU access (via DRM_IOCTL_XE_VM_BIND). All
+ * user BOs are evictable and user BOs are never pinned by XE. The allocation of
+ * the backing store can be defered from creation time until first use which is
+ * either mmap, bind, or pagefault.
+ *
+ * Private BOs
+ * ~~~~~~~~~~~
+ *
+ * A private BO is a user BO created with a valid VM argument passed into the
+ * create IOCTL. If a BO is private it cannot be exported via prime FD and
+ * mappings can only be created for the BO within the VM it is tied to. Lastly,
+ * the BO dma-resv slots / lock point to the VM's dma-resv slots / lock (all
+ * private BOs to a VM share common dma-resv slots / lock).
+ *
+ * External BOs
+ * ~~~~~~~~~~~~
+ *
+ * An external BO is a user BO created with a NULL VM argument passed into the
+ * create IOCTL. An external BO can be shared with different UMDs / devices via
+ * prime FD and the BO can be mapped into multiple VMs. An external BO has its
+ * own unique dma-resv slots / lock. An external BO will be in an array of all
+ * VMs which has a mapping of the BO. This allows VMs to lookup and lock all
+ * external BOs mapped in the VM as needed.
+ *
+ * BO placement
+ * ~~~~~~~~~~~~
+ *
+ * When a user BO is created, a mask of valid placements is passed indicating
+ * which memory regions are considered valid.
+ *
+ * The memory region information is available via query uAPI (TODO: add link).
+ *
+ * BO validation
+ * =============
+ *
+ * BO validation (ttm_bo_validate) refers to ensuring a BO has a valid
+ * placement. If a BO was swapped to temporary storage, a validation call will
+ * trigger a move back to a valid (location where GPU can access BO) placement.
+ * Validation of a BO may evict other BOs to make room for the BO being
+ * validated.
+ *
+ * BO eviction / moving
+ * ====================
+ *
+ * All eviction (or in other words, moving a BO from one memory location to
+ * another) is routed through TTM with a callback into XE.
+ *
+ * Runtime eviction
+ * ----------------
+ *
+ * Runtime evictions refers to during normal operations where TTM decides it
+ * needs to move a BO. Typically this is because TTM needs to make room for
+ * another BO and the evicted BO is first BO on LRU list that is not locked.
+ *
+ * An example of this is a new BO which can only be placed in VRAM but there is
+ * not space in VRAM. There could be multiple BOs which have sysmem and VRAM
+ * placement rules which currently reside in VRAM, TTM trigger a will move of
+ * one (or multiple) of these BO(s) until there is room in VRAM to place the new
+ * BO. The evicted BO(s) are valid but still need new bindings before the BO
+ * used again (exec or compute mode rebind worker).
+ *
+ * Another example would be, TTM can't find a BO to evict which has another
+ * valid placement. In this case TTM will evict one (or multiple) unlocked BO(s)
+ * to a temporary unreachable (invalid) placement. The evicted BO(s) are invalid
+ * and before next use need to be moved to a valid placement and rebound.
+ *
+ * In both cases, moves of these BOs are scheduled behind the fences in the BO's
+ * dma-resv slots.
+ *
+ * WW locking tries to ensures if 2 VMs use 51% of the memory forward progress
+ * is made on both VMs.
+ *
+ * Runtime eviction uses per a GT migration engine (TODO: link to migration
+ * engine doc) to do a GPU memcpy from one location to another.
+ *
+ * Rebinds after runtime eviction
+ * ------------------------------
+ *
+ * When BOs are moved, every mapping (VMA) of the BO needs to rebound before
+ * the BO is used again. Every VMA is added to an evicted list of its VM when
+ * the BO is moved. This is safe because of the VM locking structure (TODO: link
+ * to VM locking doc). On the next use of a VM (exec or compute mode rebind
+ * worker) the evicted VMA list is checked and rebinds are triggered. In the
+ * case of faulting VM, the rebind is done in the page fault handler.
+ *
+ * Suspend / resume eviction of VRAM
+ * ---------------------------------
+ *
+ * During device suspend / resume VRAM may lose power which means the contents
+ * of VRAM's memory is blown away. Thus BOs present in VRAM at the time of
+ * suspend must be moved to sysmem in order for their contents to be saved.
+ *
+ * A simple TTM call (ttm_resource_manager_evict_all) can move all non-pinned
+ * (user) BOs to sysmem. External BOs that are pinned need to be manually
+ * evicted with a simple loop + xe_bo_evict call. It gets a little trickier
+ * with kernel BOs.
+ *
+ * Some kernel BOs are used by the GT migration engine to do moves, thus we
+ * can't move all of the BOs via the GT migration engine. For simplity, use a
+ * TTM memcpy (CPU) to move any kernel (pinned) BO on either suspend or resume.
+ *
+ * Some kernel BOs need to be restored to the exact same physical location. TTM
+ * makes this rather easy but the caveat is the memory must be contiguous. Again
+ * for simplity, we enforce that all kernel (pinned) BOs are contiguous and
+ * restored to the same physical location.
+ *
+ * Pinned external BOs in VRAM are restored on resume via the GPU.
+ *
+ * Rebinds after suspend / resume
+ * ------------------------------
+ *
+ * Most kernel BOs have GGTT mappings which must be restored during the resume
+ * process. All user BOs are rebound after validation on their next use.
+ *
+ * Future work
+ * ===========
+ *
+ * Trim the list of BOs which is saved / restored via TTM memcpy on suspend /
+ * resume. All we really need to save / restore via TTM memcpy is the memory
+ * required for the GuC to load and the memory for the GT migrate engine to
+ * operate.
+ *
+ * Do not require kernel BOs to be contiguous in physical memory / restored to
+ * the same physical address on resume. In all likelihood the only memory that
+ * needs to be restored to the same physical address is memory used for page
+ * tables. All of that memory is allocated 1 page at time so the contiguous
+ * requirement isn't needed. Some work on the vmap code would need to be done if
+ * kernel BOs are not contiguous too.
+ *
+ * Make some kernel BO evictable rather than pinned. An example of this would be
+ * engine state, in all likelihood if the dma-slots of these BOs where properly
+ * used rather than pinning we could safely evict + rebind these BOs as needed.
+ *
+ * Some kernel BOs do not need to be restored on resume (e.g. GuC ADS as that is
+ * repopulated on resume), add flag to mark such objects as no save / restore.
+ */
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bo_evict.c b/drivers/gpu/drm/xe/xe_bo_evict.c
new file mode 100644
index 000000000000..7a264a9ca06e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo_evict.c
@@ -0,0 +1,228 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_bo_evict.h"
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_ggtt.h"
+#include "xe_tile.h"
+
+/**
+ * xe_bo_evict_all - evict all BOs from VRAM
+ *
+ * @xe: xe device
+ *
+ * Evict non-pinned user BOs first (via GPU), evict pinned external BOs next
+ * (via GPU), wait for evictions, and finally evict pinned kernel BOs via CPU.
+ * All eviction magic done via TTM calls.
+ *
+ * Evict == move VRAM BOs to temporary (typically system) memory.
+ *
+ * This function should be called before the device goes into a suspend state
+ * where the VRAM loses power.
+ */
+int xe_bo_evict_all(struct xe_device *xe)
+{
+ struct ttm_device *bdev = &xe->ttm;
+ struct xe_bo *bo;
+ struct xe_tile *tile;
+ struct list_head still_in_list;
+ u32 mem_type;
+ u8 id;
+ int ret;
+
+ if (!IS_DGFX(xe))
+ return 0;
+
+ /* User memory */
+ for (mem_type = XE_PL_VRAM0; mem_type <= XE_PL_VRAM1; ++mem_type) {
+ struct ttm_resource_manager *man =
+ ttm_manager_type(bdev, mem_type);
+
+ if (man) {
+ ret = ttm_resource_manager_evict_all(bdev, man);
+ if (ret)
+ return ret;
+ }
+ }
+
+ /* Pinned user memory in VRAM */
+ INIT_LIST_HEAD(&still_in_list);
+ spin_lock(&xe->pinned.lock);
+ for (;;) {
+ bo = list_first_entry_or_null(&xe->pinned.external_vram,
+ typeof(*bo), pinned_link);
+ if (!bo)
+ break;
+ xe_bo_get(bo);
+ list_move_tail(&bo->pinned_link, &still_in_list);
+ spin_unlock(&xe->pinned.lock);
+
+ xe_bo_lock(bo, false);
+ ret = xe_bo_evict_pinned(bo);
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+ if (ret) {
+ spin_lock(&xe->pinned.lock);
+ list_splice_tail(&still_in_list,
+ &xe->pinned.external_vram);
+ spin_unlock(&xe->pinned.lock);
+ return ret;
+ }
+
+ spin_lock(&xe->pinned.lock);
+ }
+ list_splice_tail(&still_in_list, &xe->pinned.external_vram);
+ spin_unlock(&xe->pinned.lock);
+
+ /*
+ * Wait for all user BO to be evicted as those evictions depend on the
+ * memory moved below.
+ */
+ for_each_tile(tile, xe, id)
+ xe_tile_migrate_wait(tile);
+
+ spin_lock(&xe->pinned.lock);
+ for (;;) {
+ bo = list_first_entry_or_null(&xe->pinned.kernel_bo_present,
+ typeof(*bo), pinned_link);
+ if (!bo)
+ break;
+ xe_bo_get(bo);
+ list_move_tail(&bo->pinned_link, &xe->pinned.evicted);
+ spin_unlock(&xe->pinned.lock);
+
+ xe_bo_lock(bo, false);
+ ret = xe_bo_evict_pinned(bo);
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+ if (ret)
+ return ret;
+
+ spin_lock(&xe->pinned.lock);
+ }
+ spin_unlock(&xe->pinned.lock);
+
+ return 0;
+}
+
+/**
+ * xe_bo_restore_kernel - restore kernel BOs to VRAM
+ *
+ * @xe: xe device
+ *
+ * Move kernel BOs from temporary (typically system) memory to VRAM via CPU. All
+ * moves done via TTM calls.
+ *
+ * This function should be called early, before trying to init the GT, on device
+ * resume.
+ */
+int xe_bo_restore_kernel(struct xe_device *xe)
+{
+ struct xe_bo *bo;
+ int ret;
+
+ if (!IS_DGFX(xe))
+ return 0;
+
+ spin_lock(&xe->pinned.lock);
+ for (;;) {
+ bo = list_first_entry_or_null(&xe->pinned.evicted,
+ typeof(*bo), pinned_link);
+ if (!bo)
+ break;
+ xe_bo_get(bo);
+ list_move_tail(&bo->pinned_link, &xe->pinned.kernel_bo_present);
+ spin_unlock(&xe->pinned.lock);
+
+ xe_bo_lock(bo, false);
+ ret = xe_bo_restore_pinned(bo);
+ xe_bo_unlock(bo);
+ if (ret) {
+ xe_bo_put(bo);
+ return ret;
+ }
+
+ if (bo->flags & XE_BO_CREATE_GGTT_BIT) {
+ struct xe_tile *tile = bo->tile;
+
+ mutex_lock(&tile->mem.ggtt->lock);
+ xe_ggtt_map_bo(tile->mem.ggtt, bo);
+ mutex_unlock(&tile->mem.ggtt->lock);
+ }
+
+ /*
+ * We expect validate to trigger a move VRAM and our move code
+ * should setup the iosys map.
+ */
+ xe_assert(xe, !iosys_map_is_null(&bo->vmap));
+ xe_assert(xe, xe_bo_is_vram(bo));
+
+ xe_bo_put(bo);
+
+ spin_lock(&xe->pinned.lock);
+ }
+ spin_unlock(&xe->pinned.lock);
+
+ return 0;
+}
+
+/**
+ * xe_bo_restore_user - restore pinned user BOs to VRAM
+ *
+ * @xe: xe device
+ *
+ * Move pinned user BOs from temporary (typically system) memory to VRAM via
+ * CPU. All moves done via TTM calls.
+ *
+ * This function should be called late, after GT init, on device resume.
+ */
+int xe_bo_restore_user(struct xe_device *xe)
+{
+ struct xe_bo *bo;
+ struct xe_tile *tile;
+ struct list_head still_in_list;
+ u8 id;
+ int ret;
+
+ if (!IS_DGFX(xe))
+ return 0;
+
+ /* Pinned user memory in VRAM should be validated on resume */
+ INIT_LIST_HEAD(&still_in_list);
+ spin_lock(&xe->pinned.lock);
+ for (;;) {
+ bo = list_first_entry_or_null(&xe->pinned.external_vram,
+ typeof(*bo), pinned_link);
+ if (!bo)
+ break;
+ list_move_tail(&bo->pinned_link, &still_in_list);
+ xe_bo_get(bo);
+ spin_unlock(&xe->pinned.lock);
+
+ xe_bo_lock(bo, false);
+ ret = xe_bo_restore_pinned(bo);
+ xe_bo_unlock(bo);
+ xe_bo_put(bo);
+ if (ret) {
+ spin_lock(&xe->pinned.lock);
+ list_splice_tail(&still_in_list,
+ &xe->pinned.external_vram);
+ spin_unlock(&xe->pinned.lock);
+ return ret;
+ }
+
+ spin_lock(&xe->pinned.lock);
+ }
+ list_splice_tail(&still_in_list, &xe->pinned.external_vram);
+ spin_unlock(&xe->pinned.lock);
+
+ /* Wait for validate to complete */
+ for_each_tile(tile, xe, id)
+ xe_tile_migrate_wait(tile);
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_bo_evict.h b/drivers/gpu/drm/xe/xe_bo_evict.h
new file mode 100644
index 000000000000..746894798852
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo_evict.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_BO_EVICT_H_
+#define _XE_BO_EVICT_H_
+
+struct xe_device;
+
+int xe_bo_evict_all(struct xe_device *xe);
+int xe_bo_restore_kernel(struct xe_device *xe);
+int xe_bo_restore_user(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
new file mode 100644
index 000000000000..64c2249a4e40
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_bo_types.h
@@ -0,0 +1,96 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_BO_TYPES_H_
+#define _XE_BO_TYPES_H_
+
+#include <linux/iosys-map.h>
+
+#include <drm/drm_mm.h>
+#include <drm/ttm/ttm_bo.h>
+#include <drm/ttm/ttm_device.h>
+#include <drm/ttm/ttm_execbuf_util.h>
+#include <drm/ttm/ttm_placement.h>
+
+struct xe_device;
+struct xe_vm;
+
+#define XE_BO_MAX_PLACEMENTS 3
+
+/* TODO: To be selected with VM_MADVISE */
+#define XE_BO_PRIORITY_NORMAL 1
+
+/** @xe_bo: XE buffer object */
+struct xe_bo {
+ /** @ttm: TTM base buffer object */
+ struct ttm_buffer_object ttm;
+ /** @size: Size of this buffer object */
+ size_t size;
+ /** @flags: flags for this buffer object */
+ u32 flags;
+ /** @vm: VM this BO is attached to, for extobj this will be NULL */
+ struct xe_vm *vm;
+ /** @tile: Tile this BO is attached to (kernel BO only) */
+ struct xe_tile *tile;
+ /** @placements: valid placements for this BO */
+ struct ttm_place placements[XE_BO_MAX_PLACEMENTS];
+ /** @placement: current placement for this BO */
+ struct ttm_placement placement;
+ /** @ggtt_node: GGTT node if this BO is mapped in the GGTT */
+ struct drm_mm_node ggtt_node;
+ /** @vmap: iosys map of this buffer */
+ struct iosys_map vmap;
+ /** @ttm_kmap: TTM bo kmap object for internal use only. Keep off. */
+ struct ttm_bo_kmap_obj kmap;
+ /** @pinned_link: link to present / evicted list of pinned BO */
+ struct list_head pinned_link;
+#ifdef CONFIG_PROC_FS
+ /**
+ * @client: @xe_drm_client which created the bo
+ */
+ struct xe_drm_client *client;
+ /**
+ * @client_link: Link into @xe_drm_client.objects_list
+ */
+ struct list_head client_link;
+#endif
+ /** @props: BO user controlled properties */
+ struct {
+ /** @preferred_mem: preferred memory class for this BO */
+ s16 preferred_mem_class;
+ /** @prefered_gt: preferred GT for this BO */
+ s16 preferred_gt;
+ /** @preferred_mem_type: preferred memory type */
+ s32 preferred_mem_type;
+ /**
+ * @cpu_atomic: the CPU expects to do atomics operations to
+ * this BO
+ */
+ bool cpu_atomic;
+ /**
+ * @device_atomic: the device expects to do atomics operations
+ * to this BO
+ */
+ bool device_atomic;
+ } props;
+ /** @freed: List node for delayed put. */
+ struct llist_node freed;
+ /** @created: Whether the bo has passed initial creation */
+ bool created;
+
+ /** @ccs_cleared */
+ bool ccs_cleared;
+
+ /**
+ * @cpu_caching: CPU caching mode. Currently only used for userspace
+ * objects.
+ */
+ u16 cpu_caching;
+};
+
+#define intel_bo_to_drm_bo(bo) (&(bo)->ttm.base)
+#define intel_bo_to_i915(bo) to_i915(intel_bo_to_drm_bo(bo)->dev)
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_debugfs.c b/drivers/gpu/drm/xe/xe_debugfs.c
new file mode 100644
index 000000000000..c56fd7d59f05
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_debugfs.c
@@ -0,0 +1,148 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_debugfs.h"
+
+#include <linux/string_helpers.h>
+
+#include <drm/drm_debugfs.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt_debugfs.h"
+#include "xe_step.h"
+
+#ifdef CONFIG_DRM_XE_DEBUG
+#include "xe_bo_evict.h"
+#include "xe_migrate.h"
+#include "xe_vm.h"
+#endif
+
+#ifdef CONFIG_FAULT_INJECTION
+#include <linux/fault-inject.h> /* XXX: fault-inject.h is broken */
+DECLARE_FAULT_ATTR(gt_reset_failure);
+#endif
+
+static struct xe_device *node_to_xe(struct drm_info_node *node)
+{
+ return to_xe_device(node->minor->dev);
+}
+
+static int info(struct seq_file *m, void *data)
+{
+ struct xe_device *xe = node_to_xe(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+ struct xe_gt *gt;
+ u8 id;
+
+ drm_printf(&p, "graphics_verx100 %d\n", xe->info.graphics_verx100);
+ drm_printf(&p, "media_verx100 %d\n", xe->info.media_verx100);
+ drm_printf(&p, "stepping G:%s M:%s D:%s B:%s\n",
+ xe_step_name(xe->info.step.graphics),
+ xe_step_name(xe->info.step.media),
+ xe_step_name(xe->info.step.display),
+ xe_step_name(xe->info.step.basedie));
+ drm_printf(&p, "is_dgfx %s\n", str_yes_no(xe->info.is_dgfx));
+ drm_printf(&p, "platform %d\n", xe->info.platform);
+ drm_printf(&p, "subplatform %d\n",
+ xe->info.subplatform > XE_SUBPLATFORM_NONE ? xe->info.subplatform : 0);
+ drm_printf(&p, "devid 0x%x\n", xe->info.devid);
+ drm_printf(&p, "revid %d\n", xe->info.revid);
+ drm_printf(&p, "tile_count %d\n", xe->info.tile_count);
+ drm_printf(&p, "vm_max_level %d\n", xe->info.vm_max_level);
+ drm_printf(&p, "force_execlist %s\n", str_yes_no(xe->info.force_execlist));
+ drm_printf(&p, "has_flat_ccs %s\n", str_yes_no(xe->info.has_flat_ccs));
+ drm_printf(&p, "has_usm %s\n", str_yes_no(xe->info.has_usm));
+ for_each_gt(gt, xe, id) {
+ drm_printf(&p, "gt%d force wake %d\n", id,
+ xe_force_wake_ref(gt_to_fw(gt), XE_FW_GT));
+ drm_printf(&p, "gt%d engine_mask 0x%llx\n", id,
+ gt->info.engine_mask);
+ }
+
+ return 0;
+}
+
+static const struct drm_info_list debugfs_list[] = {
+ {"info", info, 0},
+};
+
+static int forcewake_open(struct inode *inode, struct file *file)
+{
+ struct xe_device *xe = inode->i_private;
+ struct xe_gt *gt;
+ u8 id;
+
+ xe_device_mem_access_get(xe);
+
+ for_each_gt(gt, xe, id)
+ XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+
+ return 0;
+}
+
+static int forcewake_release(struct inode *inode, struct file *file)
+{
+ struct xe_device *xe = inode->i_private;
+ struct xe_gt *gt;
+ u8 id;
+
+ for_each_gt(gt, xe, id)
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+static const struct file_operations forcewake_all_fops = {
+ .owner = THIS_MODULE,
+ .open = forcewake_open,
+ .release = forcewake_release,
+};
+
+void xe_debugfs_register(struct xe_device *xe)
+{
+ struct ttm_device *bdev = &xe->ttm;
+ struct drm_minor *minor = xe->drm.primary;
+ struct dentry *root = minor->debugfs_root;
+ struct ttm_resource_manager *man;
+ struct xe_gt *gt;
+ u32 mem_type;
+ u8 id;
+
+ drm_debugfs_create_files(debugfs_list,
+ ARRAY_SIZE(debugfs_list),
+ root, minor);
+
+ debugfs_create_file("forcewake_all", 0400, root, xe,
+ &forcewake_all_fops);
+
+ for (mem_type = XE_PL_VRAM0; mem_type <= XE_PL_VRAM1; ++mem_type) {
+ man = ttm_manager_type(bdev, mem_type);
+
+ if (man) {
+ char name[16];
+
+ sprintf(name, "vram%d_mm", mem_type - XE_PL_VRAM0);
+ ttm_resource_manager_create_debugfs(man, root, name);
+ }
+ }
+
+ man = ttm_manager_type(bdev, XE_PL_TT);
+ ttm_resource_manager_create_debugfs(man, root, "gtt_mm");
+
+ man = ttm_manager_type(bdev, XE_PL_STOLEN);
+ if (man)
+ ttm_resource_manager_create_debugfs(man, root, "stolen_mm");
+
+ for_each_gt(gt, xe, id)
+ xe_gt_debugfs_register(gt);
+
+#ifdef CONFIG_FAULT_INJECTION
+ fault_create_debugfs_attr("fail_gt_reset", root, &gt_reset_failure);
+#endif
+
+}
diff --git a/drivers/gpu/drm/xe/xe_debugfs.h b/drivers/gpu/drm/xe/xe_debugfs.h
new file mode 100644
index 000000000000..715b8e2e0bd9
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_debugfs.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_DEBUGFS_H_
+#define _XE_DEBUGFS_H_
+
+struct xe_device;
+
+void xe_debugfs_register(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_devcoredump.c b/drivers/gpu/drm/xe/xe_devcoredump.c
new file mode 100644
index 000000000000..68abc0b195be
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_devcoredump.c
@@ -0,0 +1,196 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_devcoredump.h"
+#include "xe_devcoredump_types.h"
+
+#include <linux/devcoredump.h>
+#include <generated/utsrelease.h>
+
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_guc_ct.h"
+#include "xe_guc_submit.h"
+#include "xe_hw_engine.h"
+
+/**
+ * DOC: Xe device coredump
+ *
+ * Devices overview:
+ * Xe uses dev_coredump infrastructure for exposing the crash errors in a
+ * standardized way.
+ * devcoredump exposes a temporary device under /sys/class/devcoredump/
+ * which is linked with our card device directly.
+ * The core dump can be accessed either from
+ * /sys/class/drm/card<n>/device/devcoredump/ or from
+ * /sys/class/devcoredump/devcd<m> where
+ * /sys/class/devcoredump/devcd<m>/failing_device is a link to
+ * /sys/class/drm/card<n>/device/.
+ *
+ * Snapshot at hang:
+ * The 'data' file is printed with a drm_printer pointer at devcoredump read
+ * time. For this reason, we need to take snapshots from when the hang has
+ * happened, and not only when the user is reading the file. Otherwise the
+ * information is outdated since the resets might have happened in between.
+ *
+ * 'First' failure snapshot:
+ * In general, the first hang is the most critical one since the following hangs
+ * can be a consequence of the initial hang. For this reason we only take the
+ * snapshot of the 'first' failure and ignore subsequent calls of this function,
+ * at least while the coredump device is alive. Dev_coredump has a delayed work
+ * queue that will eventually delete the device and free all the dump
+ * information.
+ */
+
+#ifdef CONFIG_DEV_COREDUMP
+
+static struct xe_device *coredump_to_xe(const struct xe_devcoredump *coredump)
+{
+ return container_of(coredump, struct xe_device, devcoredump);
+}
+
+static struct xe_guc *exec_queue_to_guc(struct xe_exec_queue *q)
+{
+ return &q->gt->uc.guc;
+}
+
+static ssize_t xe_devcoredump_read(char *buffer, loff_t offset,
+ size_t count, void *data, size_t datalen)
+{
+ struct xe_devcoredump *coredump = data;
+ struct xe_devcoredump_snapshot *ss;
+ struct drm_printer p;
+ struct drm_print_iterator iter;
+ struct timespec64 ts;
+ int i;
+
+ /* Our device is gone already... */
+ if (!data || !coredump_to_xe(coredump))
+ return -ENODEV;
+
+ iter.data = buffer;
+ iter.offset = 0;
+ iter.start = offset;
+ iter.remain = count;
+
+ ss = &coredump->snapshot;
+ p = drm_coredump_printer(&iter);
+
+ drm_printf(&p, "**** Xe Device Coredump ****\n");
+ drm_printf(&p, "kernel: " UTS_RELEASE "\n");
+ drm_printf(&p, "module: " KBUILD_MODNAME "\n");
+
+ ts = ktime_to_timespec64(ss->snapshot_time);
+ drm_printf(&p, "Snapshot time: %lld.%09ld\n", ts.tv_sec, ts.tv_nsec);
+ ts = ktime_to_timespec64(ss->boot_time);
+ drm_printf(&p, "Uptime: %lld.%09ld\n", ts.tv_sec, ts.tv_nsec);
+
+ drm_printf(&p, "\n**** GuC CT ****\n");
+ xe_guc_ct_snapshot_print(coredump->snapshot.ct, &p);
+ xe_guc_exec_queue_snapshot_print(coredump->snapshot.ge, &p);
+
+ drm_printf(&p, "\n**** HW Engines ****\n");
+ for (i = 0; i < XE_NUM_HW_ENGINES; i++)
+ if (coredump->snapshot.hwe[i])
+ xe_hw_engine_snapshot_print(coredump->snapshot.hwe[i],
+ &p);
+
+ return count - iter.remain;
+}
+
+static void xe_devcoredump_free(void *data)
+{
+ struct xe_devcoredump *coredump = data;
+ int i;
+
+ /* Our device is gone. Nothing to do... */
+ if (!data || !coredump_to_xe(coredump))
+ return;
+
+ xe_guc_ct_snapshot_free(coredump->snapshot.ct);
+ xe_guc_exec_queue_snapshot_free(coredump->snapshot.ge);
+ for (i = 0; i < XE_NUM_HW_ENGINES; i++)
+ if (coredump->snapshot.hwe[i])
+ xe_hw_engine_snapshot_free(coredump->snapshot.hwe[i]);
+
+ coredump->captured = false;
+ drm_info(&coredump_to_xe(coredump)->drm,
+ "Xe device coredump has been deleted.\n");
+}
+
+static void devcoredump_snapshot(struct xe_devcoredump *coredump,
+ struct xe_exec_queue *q)
+{
+ struct xe_devcoredump_snapshot *ss = &coredump->snapshot;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ u32 adj_logical_mask = q->logical_mask;
+ u32 width_mask = (0x1 << q->width) - 1;
+ int i;
+ bool cookie;
+
+ ss->snapshot_time = ktime_get_real();
+ ss->boot_time = ktime_get_boottime();
+
+ cookie = dma_fence_begin_signalling();
+ for (i = 0; q->width > 1 && i < XE_HW_ENGINE_MAX_INSTANCE;) {
+ if (adj_logical_mask & BIT(i)) {
+ adj_logical_mask |= width_mask << i;
+ i += q->width;
+ } else {
+ ++i;
+ }
+ }
+
+ xe_force_wake_get(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
+
+ coredump->snapshot.ct = xe_guc_ct_snapshot_capture(&guc->ct, true);
+ coredump->snapshot.ge = xe_guc_exec_queue_snapshot_capture(q);
+
+ for_each_hw_engine(hwe, q->gt, id) {
+ if (hwe->class != q->hwe->class ||
+ !(BIT(hwe->logical_instance) & adj_logical_mask)) {
+ coredump->snapshot.hwe[id] = NULL;
+ continue;
+ }
+ coredump->snapshot.hwe[id] = xe_hw_engine_snapshot_capture(hwe);
+ }
+
+ xe_force_wake_put(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
+ dma_fence_end_signalling(cookie);
+}
+
+/**
+ * xe_devcoredump - Take the required snapshots and initialize coredump device.
+ * @q: The faulty xe_exec_queue, where the issue was detected.
+ *
+ * This function should be called at the crash time within the serialized
+ * gt_reset. It is skipped if we still have the core dump device available
+ * with the information of the 'first' snapshot.
+ */
+void xe_devcoredump(struct xe_exec_queue *q)
+{
+ struct xe_device *xe = gt_to_xe(q->gt);
+ struct xe_devcoredump *coredump = &xe->devcoredump;
+
+ if (coredump->captured) {
+ drm_dbg(&xe->drm, "Multiple hangs are occurring, but only the first snapshot was taken\n");
+ return;
+ }
+
+ coredump->captured = true;
+ devcoredump_snapshot(coredump, q);
+
+ drm_info(&xe->drm, "Xe device coredump has been created\n");
+ drm_info(&xe->drm, "Check your /sys/class/drm/card%d/device/devcoredump/data\n",
+ xe->drm.primary->index);
+
+ dev_coredumpm(xe->drm.dev, THIS_MODULE, coredump, 0, GFP_KERNEL,
+ xe_devcoredump_read, xe_devcoredump_free);
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_devcoredump.h b/drivers/gpu/drm/xe/xe_devcoredump.h
new file mode 100644
index 000000000000..6ac218a5c194
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_devcoredump.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DEVCOREDUMP_H_
+#define _XE_DEVCOREDUMP_H_
+
+struct xe_device;
+struct xe_exec_queue;
+
+#ifdef CONFIG_DEV_COREDUMP
+void xe_devcoredump(struct xe_exec_queue *q);
+#else
+static inline void xe_devcoredump(struct xe_exec_queue *q)
+{
+}
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_devcoredump_types.h b/drivers/gpu/drm/xe/xe_devcoredump_types.h
new file mode 100644
index 000000000000..7fdad9c3d3dd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_devcoredump_types.h
@@ -0,0 +1,55 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DEVCOREDUMP_TYPES_H_
+#define _XE_DEVCOREDUMP_TYPES_H_
+
+#include <linux/ktime.h>
+#include <linux/mutex.h>
+
+#include "xe_hw_engine_types.h"
+
+struct xe_device;
+
+/**
+ * struct xe_devcoredump_snapshot - Crash snapshot
+ *
+ * This struct contains all the useful information quickly captured at the time
+ * of the crash. So, any subsequent reads of the coredump points to a data that
+ * shows the state of the GPU of when the issue has happened.
+ */
+struct xe_devcoredump_snapshot {
+ /** @snapshot_time: Time of this capture. */
+ ktime_t snapshot_time;
+ /** @boot_time: Relative boot time so the uptime can be calculated. */
+ ktime_t boot_time;
+
+ /* GuC snapshots */
+ /** @ct: GuC CT snapshot */
+ struct xe_guc_ct_snapshot *ct;
+ /** @ge: Guc Engine snapshot */
+ struct xe_guc_submit_exec_queue_snapshot *ge;
+ /** @hwe: HW Engine snapshot array */
+ struct xe_hw_engine_snapshot *hwe[XE_NUM_HW_ENGINES];
+};
+
+/**
+ * struct xe_devcoredump - Xe devcoredump main structure
+ *
+ * This struct represents the live and active dev_coredump node.
+ * It is created/populated at the time of a crash/error. Then it
+ * is read later when user access the device coredump data file
+ * for reading the information.
+ */
+struct xe_devcoredump {
+ /** @xe: Xe device. */
+ struct xe_device *xe;
+ /** @captured: The snapshot of the first hang has already been taken. */
+ bool captured;
+ /** @snapshot: Snapshot is captured at time of the first crash */
+ struct xe_devcoredump_snapshot snapshot;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
new file mode 100644
index 000000000000..d9ae77fe7382
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -0,0 +1,700 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_device.h"
+
+#include <linux/units.h>
+
+#include <drm/drm_aperture.h>
+#include <drm/drm_atomic_helper.h>
+#include <drm/drm_gem_ttm_helper.h>
+#include <drm/drm_ioctl.h>
+#include <drm/drm_managed.h>
+#include <drm/drm_print.h>
+#include <drm/xe_drm.h>
+
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_bo.h"
+#include "xe_debugfs.h"
+#include "xe_display.h"
+#include "xe_dma_buf.h"
+#include "xe_drm_client.h"
+#include "xe_drv.h"
+#include "xe_exec_queue.h"
+#include "xe_exec.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_irq.h"
+#include "xe_mmio.h"
+#include "xe_module.h"
+#include "xe_pat.h"
+#include "xe_pcode.h"
+#include "xe_pm.h"
+#include "xe_query.h"
+#include "xe_tile.h"
+#include "xe_ttm_stolen_mgr.h"
+#include "xe_ttm_sys_mgr.h"
+#include "xe_vm.h"
+#include "xe_wait_user_fence.h"
+#include "xe_hwmon.h"
+
+#ifdef CONFIG_LOCKDEP
+struct lockdep_map xe_device_mem_access_lockdep_map = {
+ .name = "xe_device_mem_access_lockdep_map"
+};
+#endif
+
+static int xe_file_open(struct drm_device *dev, struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_drm_client *client;
+ struct xe_file *xef;
+ int ret = -ENOMEM;
+
+ xef = kzalloc(sizeof(*xef), GFP_KERNEL);
+ if (!xef)
+ return ret;
+
+ client = xe_drm_client_alloc();
+ if (!client) {
+ kfree(xef);
+ return ret;
+ }
+
+ xef->drm = file;
+ xef->client = client;
+ xef->xe = xe;
+
+ mutex_init(&xef->vm.lock);
+ xa_init_flags(&xef->vm.xa, XA_FLAGS_ALLOC1);
+
+ mutex_init(&xef->exec_queue.lock);
+ xa_init_flags(&xef->exec_queue.xa, XA_FLAGS_ALLOC1);
+
+ spin_lock(&xe->clients.lock);
+ xe->clients.count++;
+ spin_unlock(&xe->clients.lock);
+
+ file->driver_priv = xef;
+ return 0;
+}
+
+static void device_kill_persistent_exec_queues(struct xe_device *xe,
+ struct xe_file *xef);
+
+static void xe_file_close(struct drm_device *dev, struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = file->driver_priv;
+ struct xe_vm *vm;
+ struct xe_exec_queue *q;
+ unsigned long idx;
+
+ mutex_lock(&xef->exec_queue.lock);
+ xa_for_each(&xef->exec_queue.xa, idx, q) {
+ xe_exec_queue_kill(q);
+ xe_exec_queue_put(q);
+ }
+ mutex_unlock(&xef->exec_queue.lock);
+ xa_destroy(&xef->exec_queue.xa);
+ mutex_destroy(&xef->exec_queue.lock);
+ device_kill_persistent_exec_queues(xe, xef);
+
+ mutex_lock(&xef->vm.lock);
+ xa_for_each(&xef->vm.xa, idx, vm)
+ xe_vm_close_and_put(vm);
+ mutex_unlock(&xef->vm.lock);
+ xa_destroy(&xef->vm.xa);
+ mutex_destroy(&xef->vm.lock);
+
+ spin_lock(&xe->clients.lock);
+ xe->clients.count--;
+ spin_unlock(&xe->clients.lock);
+
+ xe_drm_client_put(xef->client);
+ kfree(xef);
+}
+
+static const struct drm_ioctl_desc xe_ioctls[] = {
+ DRM_IOCTL_DEF_DRV(XE_DEVICE_QUERY, xe_query_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_GEM_CREATE, xe_gem_create_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_GEM_MMAP_OFFSET, xe_gem_mmap_offset_ioctl,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_VM_CREATE, xe_vm_create_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_VM_DESTROY, xe_vm_destroy_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_VM_BIND, xe_vm_bind_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_EXEC, xe_exec_ioctl, DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_EXEC_QUEUE_CREATE, xe_exec_queue_create_ioctl,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_EXEC_QUEUE_DESTROY, xe_exec_queue_destroy_ioctl,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_EXEC_QUEUE_GET_PROPERTY, xe_exec_queue_get_property_ioctl,
+ DRM_RENDER_ALLOW),
+ DRM_IOCTL_DEF_DRV(XE_WAIT_USER_FENCE, xe_wait_user_fence_ioctl,
+ DRM_RENDER_ALLOW),
+};
+
+static const struct file_operations xe_driver_fops = {
+ .owner = THIS_MODULE,
+ .open = drm_open,
+ .release = drm_release_noglobal,
+ .unlocked_ioctl = drm_ioctl,
+ .mmap = drm_gem_mmap,
+ .poll = drm_poll,
+ .read = drm_read,
+ .compat_ioctl = drm_compat_ioctl,
+ .llseek = noop_llseek,
+#ifdef CONFIG_PROC_FS
+ .show_fdinfo = drm_show_fdinfo,
+#endif
+};
+
+static void xe_driver_release(struct drm_device *dev)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ pci_set_drvdata(to_pci_dev(xe->drm.dev), NULL);
+}
+
+static struct drm_driver driver = {
+ /* Don't use MTRRs here; the Xserver or userspace app should
+ * deal with them for Intel hardware.
+ */
+ .driver_features =
+ DRIVER_GEM |
+ DRIVER_RENDER | DRIVER_SYNCOBJ |
+ DRIVER_SYNCOBJ_TIMELINE | DRIVER_GEM_GPUVA,
+ .open = xe_file_open,
+ .postclose = xe_file_close,
+
+ .gem_prime_import = xe_gem_prime_import,
+
+ .dumb_create = xe_bo_dumb_create,
+ .dumb_map_offset = drm_gem_ttm_dumb_map_offset,
+#ifdef CONFIG_PROC_FS
+ .show_fdinfo = xe_drm_client_fdinfo,
+#endif
+ .release = &xe_driver_release,
+
+ .ioctls = xe_ioctls,
+ .num_ioctls = ARRAY_SIZE(xe_ioctls),
+ .fops = &xe_driver_fops,
+ .name = DRIVER_NAME,
+ .desc = DRIVER_DESC,
+ .date = DRIVER_DATE,
+ .major = DRIVER_MAJOR,
+ .minor = DRIVER_MINOR,
+ .patchlevel = DRIVER_PATCHLEVEL,
+};
+
+static void xe_device_destroy(struct drm_device *dev, void *dummy)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ if (xe->ordered_wq)
+ destroy_workqueue(xe->ordered_wq);
+
+ if (xe->unordered_wq)
+ destroy_workqueue(xe->unordered_wq);
+
+ ttm_device_fini(&xe->ttm);
+}
+
+struct xe_device *xe_device_create(struct pci_dev *pdev,
+ const struct pci_device_id *ent)
+{
+ struct xe_device *xe;
+ int err;
+
+ xe_display_driver_set_hooks(&driver);
+
+ err = drm_aperture_remove_conflicting_pci_framebuffers(pdev, &driver);
+ if (err)
+ return ERR_PTR(err);
+
+ xe = devm_drm_dev_alloc(&pdev->dev, &driver, struct xe_device, drm);
+ if (IS_ERR(xe))
+ return xe;
+
+ err = ttm_device_init(&xe->ttm, &xe_ttm_funcs, xe->drm.dev,
+ xe->drm.anon_inode->i_mapping,
+ xe->drm.vma_offset_manager, false, false);
+ if (WARN_ON(err))
+ goto err;
+
+ err = drmm_add_action_or_reset(&xe->drm, xe_device_destroy, NULL);
+ if (err)
+ goto err;
+
+ xe->info.devid = pdev->device;
+ xe->info.revid = pdev->revision;
+ xe->info.force_execlist = xe_modparam.force_execlist;
+
+ spin_lock_init(&xe->irq.lock);
+ spin_lock_init(&xe->clients.lock);
+
+ init_waitqueue_head(&xe->ufence_wq);
+
+ drmm_mutex_init(&xe->drm, &xe->usm.lock);
+ xa_init_flags(&xe->usm.asid_to_vm, XA_FLAGS_ALLOC);
+
+ if (IS_ENABLED(CONFIG_DRM_XE_DEBUG)) {
+ /* Trigger a large asid and an early asid wrap. */
+ u32 asid;
+
+ BUILD_BUG_ON(XE_MAX_ASID < 2);
+ err = xa_alloc_cyclic(&xe->usm.asid_to_vm, &asid, NULL,
+ XA_LIMIT(XE_MAX_ASID - 2, XE_MAX_ASID - 1),
+ &xe->usm.next_asid, GFP_KERNEL);
+ drm_WARN_ON(&xe->drm, err);
+ if (err >= 0)
+ xa_erase(&xe->usm.asid_to_vm, asid);
+ }
+
+ drmm_mutex_init(&xe->drm, &xe->persistent_engines.lock);
+ INIT_LIST_HEAD(&xe->persistent_engines.list);
+
+ spin_lock_init(&xe->pinned.lock);
+ INIT_LIST_HEAD(&xe->pinned.kernel_bo_present);
+ INIT_LIST_HEAD(&xe->pinned.external_vram);
+ INIT_LIST_HEAD(&xe->pinned.evicted);
+
+ xe->ordered_wq = alloc_ordered_workqueue("xe-ordered-wq", 0);
+ xe->unordered_wq = alloc_workqueue("xe-unordered-wq", 0, 0);
+ if (!xe->ordered_wq || !xe->unordered_wq) {
+ drm_err(&xe->drm, "Failed to allocate xe workqueues\n");
+ err = -ENOMEM;
+ goto err;
+ }
+
+ err = xe_display_create(xe);
+ if (WARN_ON(err))
+ goto err;
+
+ return xe;
+
+err:
+ return ERR_PTR(err);
+}
+
+/*
+ * The driver-initiated FLR is the highest level of reset that we can trigger
+ * from within the driver. It is different from the PCI FLR in that it doesn't
+ * fully reset the SGUnit and doesn't modify the PCI config space and therefore
+ * it doesn't require a re-enumeration of the PCI BARs. However, the
+ * driver-initiated FLR does still cause a reset of both GT and display and a
+ * memory wipe of local and stolen memory, so recovery would require a full HW
+ * re-init and saving/restoring (or re-populating) the wiped memory. Since we
+ * perform the FLR as the very last action before releasing access to the HW
+ * during the driver release flow, we don't attempt recovery at all, because
+ * if/when a new instance of i915 is bound to the device it will do a full
+ * re-init anyway.
+ */
+static void xe_driver_flr(struct xe_device *xe)
+{
+ const unsigned int flr_timeout = 3 * MICRO; /* specs recommend a 3s wait */
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+ int ret;
+
+ if (xe_mmio_read32(gt, GU_CNTL_PROTECTED) & DRIVERINT_FLR_DIS) {
+ drm_info_once(&xe->drm, "BIOS Disabled Driver-FLR\n");
+ return;
+ }
+
+ drm_dbg(&xe->drm, "Triggering Driver-FLR\n");
+
+ /*
+ * Make sure any pending FLR requests have cleared by waiting for the
+ * FLR trigger bit to go to zero. Also clear GU_DEBUG's DRIVERFLR_STATUS
+ * to make sure it's not still set from a prior attempt (it's a write to
+ * clear bit).
+ * Note that we should never be in a situation where a previous attempt
+ * is still pending (unless the HW is totally dead), but better to be
+ * safe in case something unexpected happens
+ */
+ ret = xe_mmio_wait32(gt, GU_CNTL, DRIVERFLR, 0, flr_timeout, NULL, false);
+ if (ret) {
+ drm_err(&xe->drm, "Driver-FLR-prepare wait for ready failed! %d\n", ret);
+ return;
+ }
+ xe_mmio_write32(gt, GU_DEBUG, DRIVERFLR_STATUS);
+
+ /* Trigger the actual Driver-FLR */
+ xe_mmio_rmw32(gt, GU_CNTL, 0, DRIVERFLR);
+
+ /* Wait for hardware teardown to complete */
+ ret = xe_mmio_wait32(gt, GU_CNTL, DRIVERFLR, 0, flr_timeout, NULL, false);
+ if (ret) {
+ drm_err(&xe->drm, "Driver-FLR-teardown wait completion failed! %d\n", ret);
+ return;
+ }
+
+ /* Wait for hardware/firmware re-init to complete */
+ ret = xe_mmio_wait32(gt, GU_DEBUG, DRIVERFLR_STATUS, DRIVERFLR_STATUS,
+ flr_timeout, NULL, false);
+ if (ret) {
+ drm_err(&xe->drm, "Driver-FLR-reinit wait completion failed! %d\n", ret);
+ return;
+ }
+
+ /* Clear sticky completion status */
+ xe_mmio_write32(gt, GU_DEBUG, DRIVERFLR_STATUS);
+}
+
+static void xe_driver_flr_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = arg;
+
+ if (xe->needs_flr_on_fini)
+ xe_driver_flr(xe);
+}
+
+static void xe_device_sanitize(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = arg;
+ struct xe_gt *gt;
+ u8 id;
+
+ for_each_gt(gt, xe, id)
+ xe_gt_sanitize(gt);
+}
+
+static int xe_set_dma_info(struct xe_device *xe)
+{
+ unsigned int mask_size = xe->info.dma_mask_size;
+ int err;
+
+ dma_set_max_seg_size(xe->drm.dev, xe_sg_segment_size(xe->drm.dev));
+
+ err = dma_set_mask(xe->drm.dev, DMA_BIT_MASK(mask_size));
+ if (err)
+ goto mask_err;
+
+ err = dma_set_coherent_mask(xe->drm.dev, DMA_BIT_MASK(mask_size));
+ if (err)
+ goto mask_err;
+
+ return 0;
+
+mask_err:
+ drm_err(&xe->drm, "Can't set DMA mask/consistent mask (%d)\n", err);
+ return err;
+}
+
+/*
+ * Initialize MMIO resources that don't require any knowledge about tile count.
+ */
+int xe_device_probe_early(struct xe_device *xe)
+{
+ int err;
+
+ err = xe_mmio_init(xe);
+ if (err)
+ return err;
+
+ err = xe_mmio_root_tile_init(xe);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static int xe_device_set_has_flat_ccs(struct xe_device *xe)
+{
+ u32 reg;
+ int err;
+
+ if (GRAPHICS_VER(xe) < 20 || !xe->info.has_flat_ccs)
+ return 0;
+
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return err;
+
+ reg = xe_gt_mcr_unicast_read_any(gt, XE2_FLAT_CCS_BASE_RANGE_LOWER);
+ xe->info.has_flat_ccs = (reg & XE2_FLAT_CCS_ENABLE);
+
+ if (!xe->info.has_flat_ccs)
+ drm_dbg(&xe->drm,
+ "Flat CCS has been disabled in bios, May lead to performance impact");
+
+ return xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+}
+
+int xe_device_probe(struct xe_device *xe)
+{
+ struct xe_tile *tile;
+ struct xe_gt *gt;
+ int err;
+ u8 id;
+
+ xe_pat_init_early(xe);
+
+ xe->info.mem_region_mask = 1;
+ err = xe_display_init_nommio(xe);
+ if (err)
+ return err;
+
+ err = xe_set_dma_info(xe);
+ if (err)
+ return err;
+
+ xe_mmio_probe_tiles(xe);
+
+ xe_ttm_sys_mgr_init(xe);
+
+ for_each_gt(gt, xe, id)
+ xe_force_wake_init_gt(gt, gt_to_fw(gt));
+
+ for_each_tile(tile, xe, id) {
+ err = xe_ggtt_init_early(tile->mem.ggtt);
+ if (err)
+ return err;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, xe_driver_flr_fini, xe);
+ if (err)
+ return err;
+
+ for_each_gt(gt, xe, id) {
+ err = xe_pcode_probe(gt);
+ if (err)
+ return err;
+ }
+
+ err = xe_display_init_noirq(xe);
+ if (err)
+ return err;
+
+ err = xe_irq_install(xe);
+ if (err)
+ goto err;
+
+ for_each_gt(gt, xe, id) {
+ err = xe_gt_init_early(gt);
+ if (err)
+ goto err_irq_shutdown;
+ }
+
+ err = xe_device_set_has_flat_ccs(xe);
+ if (err)
+ return err;
+
+ err = xe_mmio_probe_vram(xe);
+ if (err)
+ goto err_irq_shutdown;
+
+ for_each_tile(tile, xe, id) {
+ err = xe_tile_init_noalloc(tile);
+ if (err)
+ goto err_irq_shutdown;
+ }
+
+ /* Allocate and map stolen after potential VRAM resize */
+ xe_ttm_stolen_mgr_init(xe);
+
+ /*
+ * Now that GT is initialized (TTM in particular),
+ * we can try to init display, and inherit the initial fb.
+ * This is the reason the first allocation needs to be done
+ * inside display.
+ */
+ err = xe_display_init_noaccel(xe);
+ if (err)
+ goto err_irq_shutdown;
+
+ for_each_gt(gt, xe, id) {
+ err = xe_gt_init(gt);
+ if (err)
+ goto err_irq_shutdown;
+ }
+
+ xe_heci_gsc_init(xe);
+
+ err = xe_display_init(xe);
+ if (err)
+ goto err_irq_shutdown;
+
+ err = drm_dev_register(&xe->drm, 0);
+ if (err)
+ goto err_fini_display;
+
+ xe_display_register(xe);
+
+ xe_debugfs_register(xe);
+
+ xe_hwmon_register(xe);
+
+ err = drmm_add_action_or_reset(&xe->drm, xe_device_sanitize, xe);
+ if (err)
+ return err;
+
+ return 0;
+
+err_fini_display:
+ xe_display_driver_remove(xe);
+
+err_irq_shutdown:
+ xe_irq_shutdown(xe);
+err:
+ xe_display_fini(xe);
+ return err;
+}
+
+static void xe_device_remove_display(struct xe_device *xe)
+{
+ xe_display_unregister(xe);
+
+ drm_dev_unplug(&xe->drm);
+ xe_display_driver_remove(xe);
+}
+
+void xe_device_remove(struct xe_device *xe)
+{
+ xe_device_remove_display(xe);
+
+ xe_display_fini(xe);
+
+ xe_heci_gsc_fini(xe);
+
+ xe_irq_shutdown(xe);
+}
+
+void xe_device_shutdown(struct xe_device *xe)
+{
+}
+
+void xe_device_add_persistent_exec_queues(struct xe_device *xe, struct xe_exec_queue *q)
+{
+ mutex_lock(&xe->persistent_engines.lock);
+ list_add_tail(&q->persistent.link, &xe->persistent_engines.list);
+ mutex_unlock(&xe->persistent_engines.lock);
+}
+
+void xe_device_remove_persistent_exec_queues(struct xe_device *xe,
+ struct xe_exec_queue *q)
+{
+ mutex_lock(&xe->persistent_engines.lock);
+ if (!list_empty(&q->persistent.link))
+ list_del(&q->persistent.link);
+ mutex_unlock(&xe->persistent_engines.lock);
+}
+
+static void device_kill_persistent_exec_queues(struct xe_device *xe,
+ struct xe_file *xef)
+{
+ struct xe_exec_queue *q, *next;
+
+ mutex_lock(&xe->persistent_engines.lock);
+ list_for_each_entry_safe(q, next, &xe->persistent_engines.list,
+ persistent.link)
+ if (q->persistent.xef == xef) {
+ xe_exec_queue_kill(q);
+ list_del_init(&q->persistent.link);
+ }
+ mutex_unlock(&xe->persistent_engines.lock);
+}
+
+void xe_device_wmb(struct xe_device *xe)
+{
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+
+ wmb();
+ if (IS_DGFX(xe))
+ xe_mmio_write32(gt, SOFTWARE_FLAGS_SPR33, 0);
+}
+
+u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size)
+{
+ return xe_device_has_flat_ccs(xe) ?
+ DIV_ROUND_UP(size, NUM_BYTES_PER_CCS_BYTE(xe)) : 0;
+}
+
+bool xe_device_mem_access_ongoing(struct xe_device *xe)
+{
+ if (xe_pm_read_callback_task(xe) != NULL)
+ return true;
+
+ return atomic_read(&xe->mem_access.ref);
+}
+
+void xe_device_assert_mem_access(struct xe_device *xe)
+{
+ XE_WARN_ON(!xe_device_mem_access_ongoing(xe));
+}
+
+bool xe_device_mem_access_get_if_ongoing(struct xe_device *xe)
+{
+ bool active;
+
+ if (xe_pm_read_callback_task(xe) == current)
+ return true;
+
+ active = xe_pm_runtime_get_if_active(xe);
+ if (active) {
+ int ref = atomic_inc_return(&xe->mem_access.ref);
+
+ xe_assert(xe, ref != S32_MAX);
+ }
+
+ return active;
+}
+
+void xe_device_mem_access_get(struct xe_device *xe)
+{
+ int ref;
+
+ /*
+ * This looks racy, but should be fine since the pm_callback_task only
+ * transitions from NULL -> current (and back to NULL again), during the
+ * runtime_resume() or runtime_suspend() callbacks, for which there can
+ * only be a single one running for our device. We only need to prevent
+ * recursively calling the runtime_get or runtime_put from those
+ * callbacks, as well as preventing triggering any access_ongoing
+ * asserts.
+ */
+ if (xe_pm_read_callback_task(xe) == current)
+ return;
+
+ /*
+ * Since the resume here is synchronous it can be quite easy to deadlock
+ * if we are not careful. Also in practice it might be quite timing
+ * sensitive to ever see the 0 -> 1 transition with the callers locks
+ * held, so deadlocks might exist but are hard for lockdep to ever see.
+ * With this in mind, help lockdep learn about the potentially scary
+ * stuff that can happen inside the runtime_resume callback by acquiring
+ * a dummy lock (it doesn't protect anything and gets compiled out on
+ * non-debug builds). Lockdep then only needs to see the
+ * mem_access_lockdep_map -> runtime_resume callback once, and then can
+ * hopefully validate all the (callers_locks) -> mem_access_lockdep_map.
+ * For example if the (callers_locks) are ever grabbed in the
+ * runtime_resume callback, lockdep should give us a nice splat.
+ */
+ lock_map_acquire(&xe_device_mem_access_lockdep_map);
+ lock_map_release(&xe_device_mem_access_lockdep_map);
+
+ xe_pm_runtime_get(xe);
+ ref = atomic_inc_return(&xe->mem_access.ref);
+
+ xe_assert(xe, ref != S32_MAX);
+
+}
+
+void xe_device_mem_access_put(struct xe_device *xe)
+{
+ int ref;
+
+ if (xe_pm_read_callback_task(xe) == current)
+ return;
+
+ ref = atomic_dec_return(&xe->mem_access.ref);
+ xe_pm_runtime_put(xe);
+
+ xe_assert(xe, ref >= 0);
+}
diff --git a/drivers/gpu/drm/xe/xe_device.h b/drivers/gpu/drm/xe/xe_device.h
new file mode 100644
index 000000000000..3da83b233206
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_device.h
@@ -0,0 +1,173 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_DEVICE_H_
+#define _XE_DEVICE_H_
+
+struct xe_exec_queue;
+struct xe_file;
+
+#include <drm/drm_util.h>
+
+#include "regs/xe_gpu_commands.h"
+#include "xe_device_types.h"
+#include "xe_force_wake.h"
+#include "xe_macros.h"
+
+#ifdef CONFIG_LOCKDEP
+extern struct lockdep_map xe_device_mem_access_lockdep_map;
+#endif
+
+static inline struct xe_device *to_xe_device(const struct drm_device *dev)
+{
+ return container_of(dev, struct xe_device, drm);
+}
+
+static inline struct xe_device *pdev_to_xe_device(struct pci_dev *pdev)
+{
+ return pci_get_drvdata(pdev);
+}
+
+static inline struct xe_device *ttm_to_xe_device(struct ttm_device *ttm)
+{
+ return container_of(ttm, struct xe_device, ttm);
+}
+
+struct xe_device *xe_device_create(struct pci_dev *pdev,
+ const struct pci_device_id *ent);
+int xe_device_probe_early(struct xe_device *xe);
+int xe_device_probe(struct xe_device *xe);
+void xe_device_remove(struct xe_device *xe);
+void xe_device_shutdown(struct xe_device *xe);
+
+void xe_device_add_persistent_exec_queues(struct xe_device *xe, struct xe_exec_queue *q);
+void xe_device_remove_persistent_exec_queues(struct xe_device *xe,
+ struct xe_exec_queue *q);
+
+void xe_device_wmb(struct xe_device *xe);
+
+static inline struct xe_file *to_xe_file(const struct drm_file *file)
+{
+ return file->driver_priv;
+}
+
+static inline struct xe_tile *xe_device_get_root_tile(struct xe_device *xe)
+{
+ return &xe->tiles[0];
+}
+
+#define XE_MAX_GT_PER_TILE 2
+
+static inline struct xe_gt *xe_tile_get_gt(struct xe_tile *tile, u8 gt_id)
+{
+ if (drm_WARN_ON(&tile_to_xe(tile)->drm, gt_id > XE_MAX_GT_PER_TILE))
+ gt_id = 0;
+
+ return gt_id ? tile->media_gt : tile->primary_gt;
+}
+
+static inline struct xe_gt *xe_device_get_gt(struct xe_device *xe, u8 gt_id)
+{
+ struct xe_tile *root_tile = xe_device_get_root_tile(xe);
+ struct xe_gt *gt;
+
+ /*
+ * FIXME: This only works for now because multi-tile and standalone
+ * media are mutually exclusive on the platforms we have today.
+ *
+ * id => GT mapping may change once we settle on how we want to handle
+ * our UAPI.
+ */
+ if (MEDIA_VER(xe) >= 13) {
+ gt = xe_tile_get_gt(root_tile, gt_id);
+ } else {
+ if (drm_WARN_ON(&xe->drm, gt_id > XE_MAX_TILES_PER_DEVICE))
+ gt_id = 0;
+
+ gt = xe->tiles[gt_id].primary_gt;
+ }
+
+ if (!gt)
+ return NULL;
+
+ drm_WARN_ON(&xe->drm, gt->info.id != gt_id);
+ drm_WARN_ON(&xe->drm, gt->info.type == XE_GT_TYPE_UNINITIALIZED);
+
+ return gt;
+}
+
+/*
+ * Provide a GT structure suitable for performing non-GT MMIO operations against
+ * the primary tile. Primarily intended for early tile initialization, display
+ * handling, top-most interrupt enable/disable, etc. Since anything using the
+ * MMIO handle returned by this function doesn't need GSI offset translation,
+ * we'll return the primary GT from the root tile.
+ *
+ * FIXME: Fix the driver design so that 'gt' isn't the target of all MMIO
+ * operations.
+ *
+ * Returns the primary gt of the root tile.
+ */
+static inline struct xe_gt *xe_root_mmio_gt(struct xe_device *xe)
+{
+ return xe_device_get_root_tile(xe)->primary_gt;
+}
+
+static inline bool xe_device_uc_enabled(struct xe_device *xe)
+{
+ return !xe->info.force_execlist;
+}
+
+#define for_each_tile(tile__, xe__, id__) \
+ for ((id__) = 0; (id__) < (xe__)->info.tile_count; (id__)++) \
+ for_each_if((tile__) = &(xe__)->tiles[(id__)])
+
+#define for_each_remote_tile(tile__, xe__, id__) \
+ for ((id__) = 1; (id__) < (xe__)->info.tile_count; (id__)++) \
+ for_each_if((tile__) = &(xe__)->tiles[(id__)])
+
+/*
+ * FIXME: This only works for now since multi-tile and standalone media
+ * happen to be mutually exclusive. Future platforms may change this...
+ */
+#define for_each_gt(gt__, xe__, id__) \
+ for ((id__) = 0; (id__) < (xe__)->info.gt_count; (id__)++) \
+ for_each_if((gt__) = xe_device_get_gt((xe__), (id__)))
+
+static inline struct xe_force_wake *gt_to_fw(struct xe_gt *gt)
+{
+ return &gt->mmio.fw;
+}
+
+void xe_device_mem_access_get(struct xe_device *xe);
+bool xe_device_mem_access_get_if_ongoing(struct xe_device *xe);
+void xe_device_mem_access_put(struct xe_device *xe);
+
+void xe_device_assert_mem_access(struct xe_device *xe);
+bool xe_device_mem_access_ongoing(struct xe_device *xe);
+
+static inline bool xe_device_in_fault_mode(struct xe_device *xe)
+{
+ return xe->usm.num_vm_in_fault_mode != 0;
+}
+
+static inline bool xe_device_in_non_fault_mode(struct xe_device *xe)
+{
+ return xe->usm.num_vm_in_non_fault_mode != 0;
+}
+
+static inline bool xe_device_has_flat_ccs(struct xe_device *xe)
+{
+ return xe->info.has_flat_ccs;
+}
+
+static inline bool xe_device_has_sriov(struct xe_device *xe)
+{
+ return xe->info.has_sriov;
+}
+
+u32 xe_device_ccs_bytes(struct xe_device *xe, u64 size);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_device_sysfs.c b/drivers/gpu/drm/xe/xe_device_sysfs.c
new file mode 100644
index 000000000000..99113a5a2b84
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_device_sysfs.c
@@ -0,0 +1,89 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/kobject.h>
+#include <linux/pci.h>
+#include <linux/sysfs.h>
+
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_device_sysfs.h"
+#include "xe_pm.h"
+
+/**
+ * DOC: Xe device sysfs
+ * Xe driver requires exposing certain tunable knobs controlled by user space for
+ * each graphics device. Considering this, we need to add sysfs attributes at device
+ * level granularity.
+ * These sysfs attributes will be available under pci device kobj directory.
+ *
+ * vram_d3cold_threshold - Report/change vram used threshold(in MB) below
+ * which vram save/restore is permissible during runtime D3cold entry/exit.
+ */
+
+static ssize_t
+vram_d3cold_threshold_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+ int ret;
+
+ if (!xe)
+ return -EINVAL;
+
+ ret = sysfs_emit(buf, "%d\n", xe->d3cold.vram_threshold);
+
+ return ret;
+}
+
+static ssize_t
+vram_d3cold_threshold_store(struct device *dev, struct device_attribute *attr,
+ const char *buff, size_t count)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+ u32 vram_d3cold_threshold;
+ int ret;
+
+ if (!xe)
+ return -EINVAL;
+
+ ret = kstrtou32(buff, 0, &vram_d3cold_threshold);
+ if (ret)
+ return ret;
+
+ drm_dbg(&xe->drm, "vram_d3cold_threshold: %u\n", vram_d3cold_threshold);
+
+ ret = xe_pm_set_vram_threshold(xe, vram_d3cold_threshold);
+
+ return ret ?: count;
+}
+
+static DEVICE_ATTR_RW(vram_d3cold_threshold);
+
+static void xe_device_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = arg;
+
+ sysfs_remove_file(&xe->drm.dev->kobj, &dev_attr_vram_d3cold_threshold.attr);
+}
+
+void xe_device_sysfs_init(struct xe_device *xe)
+{
+ struct device *dev = xe->drm.dev;
+ int ret;
+
+ ret = sysfs_create_file(&dev->kobj, &dev_attr_vram_d3cold_threshold.attr);
+ if (ret) {
+ drm_warn(&xe->drm, "Failed to create sysfs file\n");
+ return;
+ }
+
+ ret = drmm_add_action_or_reset(&xe->drm, xe_device_sysfs_fini, xe);
+ if (ret)
+ drm_warn(&xe->drm, "Failed to add sysfs fini drm action\n");
+}
diff --git a/drivers/gpu/drm/xe/xe_device_sysfs.h b/drivers/gpu/drm/xe/xe_device_sysfs.h
new file mode 100644
index 000000000000..38b240684bee
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_device_sysfs.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DEVICE_SYSFS_H_
+#define _XE_DEVICE_SYSFS_H_
+
+struct xe_device;
+
+void xe_device_sysfs_init(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_device_types.h b/drivers/gpu/drm/xe/xe_device_types.h
new file mode 100644
index 000000000000..c45ef17b3473
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_device_types.h
@@ -0,0 +1,545 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022-2023 Intel Corporation
+ */
+
+#ifndef _XE_DEVICE_TYPES_H_
+#define _XE_DEVICE_TYPES_H_
+
+#include <linux/pci.h>
+
+#include <drm/drm_device.h>
+#include <drm/drm_file.h>
+#include <drm/ttm/ttm_device.h>
+
+#include "xe_devcoredump_types.h"
+#include "xe_heci_gsc.h"
+#include "xe_gt_types.h"
+#include "xe_lmtt_types.h"
+#include "xe_platform_types.h"
+#include "xe_pt_types.h"
+#include "xe_sriov_types.h"
+#include "xe_step_types.h"
+
+#if IS_ENABLED(CONFIG_DRM_XE_DISPLAY)
+#include "soc/intel_pch.h"
+#include "intel_display_core.h"
+#include "intel_display_device.h"
+#endif
+
+struct xe_ggtt;
+struct xe_pat_ops;
+
+#define XE_BO_INVALID_OFFSET LONG_MAX
+
+#define GRAPHICS_VER(xe) ((xe)->info.graphics_verx100 / 100)
+#define MEDIA_VER(xe) ((xe)->info.media_verx100 / 100)
+#define GRAPHICS_VERx100(xe) ((xe)->info.graphics_verx100)
+#define MEDIA_VERx100(xe) ((xe)->info.media_verx100)
+#define IS_DGFX(xe) ((xe)->info.is_dgfx)
+#define HAS_HECI_GSCFI(xe) ((xe)->info.has_heci_gscfi)
+
+#define XE_VRAM_FLAGS_NEED64K BIT(0)
+
+#define XE_GT0 0
+#define XE_GT1 1
+#define XE_MAX_TILES_PER_DEVICE (XE_GT1 + 1)
+
+#define XE_MAX_ASID (BIT(20))
+
+#define IS_PLATFORM_STEP(_xe, _platform, min_step, max_step) \
+ ((_xe)->info.platform == (_platform) && \
+ (_xe)->info.step.graphics >= (min_step) && \
+ (_xe)->info.step.graphics < (max_step))
+#define IS_SUBPLATFORM_STEP(_xe, _platform, sub, min_step, max_step) \
+ ((_xe)->info.platform == (_platform) && \
+ (_xe)->info.subplatform == (sub) && \
+ (_xe)->info.step.graphics >= (min_step) && \
+ (_xe)->info.step.graphics < (max_step))
+
+#define tile_to_xe(tile__) \
+ _Generic(tile__, \
+ const struct xe_tile * : (const struct xe_device *)((tile__)->xe), \
+ struct xe_tile * : (tile__)->xe)
+
+/**
+ * struct xe_mem_region - memory region structure
+ * This is used to describe a memory region in xe
+ * device, such as HBM memory or CXL extension memory.
+ */
+struct xe_mem_region {
+ /** @io_start: IO start address of this VRAM instance */
+ resource_size_t io_start;
+ /**
+ * @io_size: IO size of this VRAM instance
+ *
+ * This represents how much of this VRAM we can access
+ * via the CPU through the VRAM BAR. This can be smaller
+ * than @usable_size, in which case only part of VRAM is CPU
+ * accessible (typically the first 256M). This
+ * configuration is known as small-bar.
+ */
+ resource_size_t io_size;
+ /** @dpa_base: This memory regions's DPA (device physical address) base */
+ resource_size_t dpa_base;
+ /**
+ * @usable_size: usable size of VRAM
+ *
+ * Usable size of VRAM excluding reserved portions
+ * (e.g stolen mem)
+ */
+ resource_size_t usable_size;
+ /**
+ * @actual_physical_size: Actual VRAM size
+ *
+ * Actual VRAM size including reserved portions
+ * (e.g stolen mem)
+ */
+ resource_size_t actual_physical_size;
+ /** @mapping: pointer to VRAM mappable space */
+ void *__iomem mapping;
+};
+
+/**
+ * struct xe_tile - hardware tile structure
+ *
+ * From a driver perspective, a "tile" is effectively a complete GPU, containing
+ * an SGunit, 1-2 GTs, and (for discrete platforms) VRAM.
+ *
+ * Multi-tile platforms effectively bundle multiple GPUs behind a single PCI
+ * device and designate one "root" tile as being responsible for external PCI
+ * communication. PCI BAR0 exposes the GGTT and MMIO register space for each
+ * tile in a stacked layout, and PCI BAR2 exposes the local memory associated
+ * with each tile similarly. Device-wide interrupts can be enabled/disabled
+ * at the root tile, and the MSTR_TILE_INTR register will report which tiles
+ * have interrupts that need servicing.
+ */
+struct xe_tile {
+ /** @xe: Backpointer to tile's PCI device */
+ struct xe_device *xe;
+
+ /** @id: ID of the tile */
+ u8 id;
+
+ /**
+ * @primary_gt: Primary GT
+ */
+ struct xe_gt *primary_gt;
+
+ /**
+ * @media_gt: Media GT
+ *
+ * Only present on devices with media version >= 13.
+ */
+ struct xe_gt *media_gt;
+
+ /**
+ * @mmio: MMIO info for a tile.
+ *
+ * Each tile has its own 16MB space in BAR0, laid out as:
+ * * 0-4MB: registers
+ * * 4MB-8MB: reserved
+ * * 8MB-16MB: global GTT
+ */
+ struct {
+ /** @size: size of tile's MMIO space */
+ size_t size;
+
+ /** @regs: pointer to tile's MMIO space (starting with registers) */
+ void *regs;
+ } mmio;
+
+ /**
+ * @mmio_ext: MMIO-extension info for a tile.
+ *
+ * Each tile has its own additional 256MB (28-bit) MMIO-extension space.
+ */
+ struct {
+ /** @size: size of tile's additional MMIO-extension space */
+ size_t size;
+
+ /** @regs: pointer to tile's additional MMIO-extension space */
+ void *regs;
+ } mmio_ext;
+
+ /** @mem: memory management info for tile */
+ struct {
+ /**
+ * @vram: VRAM info for tile.
+ *
+ * Although VRAM is associated with a specific tile, it can
+ * still be accessed by all tiles' GTs.
+ */
+ struct xe_mem_region vram;
+
+ /** @vram_mgr: VRAM TTM manager */
+ struct xe_ttm_vram_mgr *vram_mgr;
+
+ /** @ggtt: Global graphics translation table */
+ struct xe_ggtt *ggtt;
+
+ /**
+ * @kernel_bb_pool: Pool from which batchbuffers are allocated.
+ *
+ * Media GT shares a pool with its primary GT.
+ */
+ struct xe_sa_manager *kernel_bb_pool;
+ } mem;
+
+ /** @sriov: tile level virtualization data */
+ union {
+ struct {
+ /** @sriov.pf.lmtt: Local Memory Translation Table. */
+ struct xe_lmtt lmtt;
+ } pf;
+ } sriov;
+
+ /** @migrate: Migration helper for vram blits and clearing */
+ struct xe_migrate *migrate;
+
+ /** @sysfs: sysfs' kobj used by xe_tile_sysfs */
+ struct kobject *sysfs;
+};
+
+/**
+ * struct xe_device - Top level struct of XE device
+ */
+struct xe_device {
+ /** @drm: drm device */
+ struct drm_device drm;
+
+ /** @devcoredump: device coredump */
+ struct xe_devcoredump devcoredump;
+
+ /** @info: device info */
+ struct intel_device_info {
+ /** @graphics_name: graphics IP name */
+ const char *graphics_name;
+ /** @media_name: media IP name */
+ const char *media_name;
+ /** @tile_mmio_ext_size: size of MMIO extension space, per-tile */
+ u32 tile_mmio_ext_size;
+ /** @graphics_verx100: graphics IP version */
+ u32 graphics_verx100;
+ /** @media_verx100: media IP version */
+ u32 media_verx100;
+ /** @mem_region_mask: mask of valid memory regions */
+ u32 mem_region_mask;
+ /** @platform: XE platform enum */
+ enum xe_platform platform;
+ /** @subplatform: XE subplatform enum */
+ enum xe_subplatform subplatform;
+ /** @devid: device ID */
+ u16 devid;
+ /** @revid: device revision */
+ u8 revid;
+ /** @step: stepping information for each IP */
+ struct xe_step_info step;
+ /** @dma_mask_size: DMA address bits */
+ u8 dma_mask_size;
+ /** @vram_flags: Vram flags */
+ u8 vram_flags;
+ /** @tile_count: Number of tiles */
+ u8 tile_count;
+ /** @gt_count: Total number of GTs for entire device */
+ u8 gt_count;
+ /** @vm_max_level: Max VM level */
+ u8 vm_max_level;
+ /** @va_bits: Maximum bits of a virtual address */
+ u8 va_bits;
+
+ /** @is_dgfx: is discrete device */
+ u8 is_dgfx:1;
+ /** @has_asid: Has address space ID */
+ u8 has_asid:1;
+ /** @force_execlist: Forced execlist submission */
+ u8 force_execlist:1;
+ /** @has_flat_ccs: Whether flat CCS metadata is used */
+ u8 has_flat_ccs:1;
+ /** @has_llc: Device has a shared CPU+GPU last level cache */
+ u8 has_llc:1;
+ /** @has_mmio_ext: Device has extra MMIO address range */
+ u8 has_mmio_ext:1;
+ /** @has_range_tlb_invalidation: Has range based TLB invalidations */
+ u8 has_range_tlb_invalidation:1;
+ /** @has_sriov: Supports SR-IOV */
+ u8 has_sriov:1;
+ /** @has_usm: Device has unified shared memory support */
+ u8 has_usm:1;
+ /** @enable_display: display enabled */
+ u8 enable_display:1;
+ /** @skip_mtcfg: skip Multi-Tile configuration from MTCFG register */
+ u8 skip_mtcfg:1;
+ /** @skip_pcode: skip access to PCODE uC */
+ u8 skip_pcode:1;
+ /** @has_heci_gscfi: device has heci gscfi */
+ u8 has_heci_gscfi:1;
+ /** @skip_guc_pc: Skip GuC based PM feature init */
+ u8 skip_guc_pc:1;
+
+#if IS_ENABLED(CONFIG_DRM_XE_DISPLAY)
+ struct {
+ u32 rawclk_freq;
+ } i915_runtime;
+#endif
+ } info;
+
+ /** @irq: device interrupt state */
+ struct {
+ /** @lock: lock for processing irq's on this device */
+ spinlock_t lock;
+
+ /** @enabled: interrupts enabled on this device */
+ bool enabled;
+ } irq;
+
+ /** @ttm: ttm device */
+ struct ttm_device ttm;
+
+ /** @mmio: mmio info for device */
+ struct {
+ /** @size: size of MMIO space for device */
+ size_t size;
+ /** @regs: pointer to MMIO space for device */
+ void *regs;
+ } mmio;
+
+ /** @mem: memory info for device */
+ struct {
+ /** @vram: VRAM info for device */
+ struct xe_mem_region vram;
+ /** @sys_mgr: system TTM manager */
+ struct ttm_resource_manager sys_mgr;
+ } mem;
+
+ /** @sriov: device level virtualization data */
+ struct {
+ /** @sriov.__mode: SR-IOV mode (Don't access directly!) */
+ enum xe_sriov_mode __mode;
+ } sriov;
+
+ /** @clients: drm clients info */
+ struct {
+ /** @lock: Protects drm clients info */
+ spinlock_t lock;
+
+ /** @count: number of drm clients */
+ u64 count;
+ } clients;
+
+ /** @usm: unified memory state */
+ struct {
+ /** @asid: convert a ASID to VM */
+ struct xarray asid_to_vm;
+ /** @next_asid: next ASID, used to cyclical alloc asids */
+ u32 next_asid;
+ /** @num_vm_in_fault_mode: number of VM in fault mode */
+ u32 num_vm_in_fault_mode;
+ /** @num_vm_in_non_fault_mode: number of VM in non-fault mode */
+ u32 num_vm_in_non_fault_mode;
+ /** @lock: protects UM state */
+ struct mutex lock;
+ } usm;
+
+ /** @persistent_engines: engines that are closed but still running */
+ struct {
+ /** @lock: protects persistent engines */
+ struct mutex lock;
+ /** @list: list of persistent engines */
+ struct list_head list;
+ } persistent_engines;
+
+ /** @pinned: pinned BO state */
+ struct {
+ /** @lock: protected pinned BO list state */
+ spinlock_t lock;
+ /** @evicted: pinned kernel BO that are present */
+ struct list_head kernel_bo_present;
+ /** @evicted: pinned BO that have been evicted */
+ struct list_head evicted;
+ /** @external_vram: pinned external BO in vram*/
+ struct list_head external_vram;
+ } pinned;
+
+ /** @ufence_wq: user fence wait queue */
+ wait_queue_head_t ufence_wq;
+
+ /** @ordered_wq: used to serialize compute mode resume */
+ struct workqueue_struct *ordered_wq;
+
+ /** @unordered_wq: used to serialize unordered work, mostly display */
+ struct workqueue_struct *unordered_wq;
+
+ /** @tiles: device tiles */
+ struct xe_tile tiles[XE_MAX_TILES_PER_DEVICE];
+
+ /**
+ * @mem_access: keep track of memory access in the device, possibly
+ * triggering additional actions when they occur.
+ */
+ struct {
+ /** @ref: ref count of memory accesses */
+ atomic_t ref;
+ } mem_access;
+
+ /**
+ * @pat: Encapsulate PAT related stuff
+ */
+ struct {
+ /** Internal operations to abstract platforms */
+ const struct xe_pat_ops *ops;
+ /** PAT table to program in the HW */
+ const struct xe_pat_table_entry *table;
+ /** Number of PAT entries */
+ int n_entries;
+ u32 idx[__XE_CACHE_LEVEL_COUNT];
+ } pat;
+
+ /** @d3cold: Encapsulate d3cold related stuff */
+ struct {
+ /** capable: Indicates if root port is d3cold capable */
+ bool capable;
+
+ /** @allowed: Indicates if d3cold is a valid device state */
+ bool allowed;
+
+ /** @power_lost: Indicates if card has really lost power. */
+ bool power_lost;
+
+ /**
+ * @vram_threshold:
+ *
+ * This represents the permissible threshold(in megabytes)
+ * for vram save/restore. d3cold will be disallowed,
+ * when vram_usages is above or equals the threshold value
+ * to avoid the vram save/restore latency.
+ * Default threshold value is 300mb.
+ */
+ u32 vram_threshold;
+ /** @lock: protect vram_threshold */
+ struct mutex lock;
+ } d3cold;
+
+ /**
+ * @pm_callback_task: Track the active task that is running in either
+ * the runtime_suspend or runtime_resume callbacks.
+ */
+ struct task_struct *pm_callback_task;
+
+ /** @hwmon: hwmon subsystem integration */
+ struct xe_hwmon *hwmon;
+
+ /** @heci_gsc: graphics security controller */
+ struct xe_heci_gsc heci_gsc;
+
+ /** @needs_flr_on_fini: requests function-reset on fini */
+ bool needs_flr_on_fini;
+
+ /* private: */
+
+#if IS_ENABLED(CONFIG_DRM_XE_DISPLAY)
+ /*
+ * Any fields below this point are the ones used by display.
+ * They are temporarily added here so xe_device can be desguised as
+ * drm_i915_private during build. After cleanup these should go away,
+ * migrating to the right sub-structs
+ */
+ struct intel_display display;
+ enum intel_pch pch_type;
+ u16 pch_id;
+
+ struct dram_info {
+ bool wm_lv_0_adjust_needed;
+ u8 num_channels;
+ bool symmetric_memory;
+ enum intel_dram_type {
+ INTEL_DRAM_UNKNOWN,
+ INTEL_DRAM_DDR3,
+ INTEL_DRAM_DDR4,
+ INTEL_DRAM_LPDDR3,
+ INTEL_DRAM_LPDDR4,
+ INTEL_DRAM_DDR5,
+ INTEL_DRAM_LPDDR5,
+ } type;
+ u8 num_qgv_points;
+ u8 num_psf_gv_points;
+ } dram_info;
+
+ /*
+ * edram size in MB.
+ * Cannot be determined by PCIID. You must always read a register.
+ */
+ u32 edram_size_mb;
+
+ /* To shut up runtime pm macros.. */
+ struct xe_runtime_pm {} runtime_pm;
+
+ /* For pcode */
+ struct mutex sb_lock;
+
+ /* Should be in struct intel_display */
+ u32 skl_preferred_vco_freq, max_dotclk_freq, hti_state;
+ u8 snps_phy_failed_calibration;
+ struct drm_atomic_state *modeset_restore_state;
+ struct list_head global_obj_list;
+
+ union {
+ /* only to allow build, not used functionally */
+ u32 irq_mask;
+ u32 de_irq_mask[I915_MAX_PIPES];
+ };
+ u32 pipestat_irq_mask[I915_MAX_PIPES];
+
+ bool display_irqs_enabled;
+ u32 enabled_irq_mask;
+
+ struct intel_uncore {
+ spinlock_t lock;
+ } uncore;
+
+ /* only to allow build, not used functionally */
+ struct {
+ unsigned int hpll_freq;
+ unsigned int czclk_freq;
+ unsigned int fsb_freq, mem_freq, is_ddr3;
+ u8 vblank_enabled;
+ };
+ struct {
+ const char *dmc_firmware_path;
+ } params;
+
+ void *pxp;
+#endif
+};
+
+/**
+ * struct xe_file - file handle for XE driver
+ */
+struct xe_file {
+ /** @xe: xe DEVICE **/
+ struct xe_device *xe;
+
+ /** @drm: base DRM file */
+ struct drm_file *drm;
+
+ /** @vm: VM state for file */
+ struct {
+ /** @xe: xarray to store VMs */
+ struct xarray xa;
+ /** @lock: protects file VM state */
+ struct mutex lock;
+ } vm;
+
+ /** @exec_queue: Submission exec queue state for file */
+ struct {
+ /** @xe: xarray to store engines */
+ struct xarray xa;
+ /** @lock: protects file engine state */
+ struct mutex lock;
+ } exec_queue;
+
+ /** @client: drm client */
+ struct xe_drm_client *client;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_display.c b/drivers/gpu/drm/xe/xe_display.c
new file mode 100644
index 000000000000..74391d9b11ae
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_display.c
@@ -0,0 +1,422 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_display.h"
+#include "regs/xe_regs.h"
+
+#include <linux/fb.h>
+
+#include <drm/drm_drv.h>
+#include <drm/drm_managed.h>
+#include <drm/xe_drm.h>
+
+#include "soc/intel_dram.h"
+#include "i915_drv.h" /* FIXME: HAS_DISPLAY() depends on this */
+#include "intel_acpi.h"
+#include "intel_audio.h"
+#include "intel_bw.h"
+#include "intel_display.h"
+#include "intel_display_driver.h"
+#include "intel_display_irq.h"
+#include "intel_display_types.h"
+#include "intel_dmc.h"
+#include "intel_dp.h"
+#include "intel_fbdev.h"
+#include "intel_hdcp.h"
+#include "intel_hotplug.h"
+#include "intel_opregion.h"
+#include "xe_module.h"
+
+/* Xe device functions */
+
+static bool has_display(struct xe_device *xe)
+{
+ return HAS_DISPLAY(xe);
+}
+
+/**
+ * xe_display_driver_probe_defer - Detect if we need to wait for other drivers
+ * early on
+ * @pdev: PCI device
+ *
+ * Returns: true if probe needs to be deferred, false otherwise
+ */
+bool xe_display_driver_probe_defer(struct pci_dev *pdev)
+{
+ if (!xe_modparam.enable_display)
+ return 0;
+
+ return intel_display_driver_probe_defer(pdev);
+}
+
+static void xe_display_last_close(struct drm_device *dev)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ if (xe->info.enable_display)
+ intel_fbdev_restore_mode(to_xe_device(dev));
+}
+
+/**
+ * xe_display_driver_set_hooks - Add driver flags and hooks for display
+ * @driver: DRM device driver
+ *
+ * Set features and function hooks in @driver that are needed for driving the
+ * display IP. This sets the driver's capability of driving display, regardless
+ * if the device has it enabled
+ */
+void xe_display_driver_set_hooks(struct drm_driver *driver)
+{
+ if (!xe_modparam.enable_display)
+ return;
+
+ driver->driver_features |= DRIVER_MODESET | DRIVER_ATOMIC;
+ driver->lastclose = xe_display_last_close;
+}
+
+static void unset_display_features(struct xe_device *xe)
+{
+ xe->drm.driver_features &= ~(DRIVER_MODESET | DRIVER_ATOMIC);
+}
+
+static void display_destroy(struct drm_device *dev, void *dummy)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ destroy_workqueue(xe->display.hotplug.dp_wq);
+}
+
+/**
+ * xe_display_create - create display struct
+ * @xe: XE device instance
+ *
+ * Initialize all fields used by the display part.
+ *
+ * TODO: once everything can be inside a single struct, make the struct opaque
+ * to the rest of xe and return it to be xe->display.
+ *
+ * Returns: 0 on success
+ */
+int xe_display_create(struct xe_device *xe)
+{
+ int err;
+
+ spin_lock_init(&xe->display.fb_tracking.lock);
+
+ xe->display.hotplug.dp_wq = alloc_ordered_workqueue("xe-dp", 0);
+
+ drmm_mutex_init(&xe->drm, &xe->sb_lock);
+ drmm_mutex_init(&xe->drm, &xe->display.backlight.lock);
+ drmm_mutex_init(&xe->drm, &xe->display.audio.mutex);
+ drmm_mutex_init(&xe->drm, &xe->display.wm.wm_mutex);
+ drmm_mutex_init(&xe->drm, &xe->display.pps.mutex);
+ drmm_mutex_init(&xe->drm, &xe->display.hdcp.hdcp_mutex);
+ xe->enabled_irq_mask = ~0;
+
+ err = drmm_add_action_or_reset(&xe->drm, display_destroy, NULL);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static void xe_display_fini_nommio(struct drm_device *dev, void *dummy)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ if (!xe->info.enable_display)
+ return;
+
+ intel_power_domains_cleanup(xe);
+}
+
+int xe_display_init_nommio(struct xe_device *xe)
+{
+ int err;
+
+ if (!xe->info.enable_display)
+ return 0;
+
+ /* Fake uncore lock */
+ spin_lock_init(&xe->uncore.lock);
+
+ /* This must be called before any calls to HAS_PCH_* */
+ intel_detect_pch(xe);
+
+ err = intel_power_domains_init(xe);
+ if (err)
+ return err;
+
+ return drmm_add_action_or_reset(&xe->drm, xe_display_fini_nommio, xe);
+}
+
+static void xe_display_fini_noirq(struct drm_device *dev, void *dummy)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ if (!xe->info.enable_display)
+ return;
+
+ intel_display_driver_remove_noirq(xe);
+ intel_power_domains_driver_remove(xe);
+}
+
+int xe_display_init_noirq(struct xe_device *xe)
+{
+ int err;
+
+ if (!xe->info.enable_display)
+ return 0;
+
+ intel_display_driver_early_probe(xe);
+
+ /* Early display init.. */
+ intel_opregion_setup(xe);
+
+ /*
+ * Fill the dram structure to get the system dram info. This will be
+ * used for memory latency calculation.
+ */
+ intel_dram_detect(xe);
+
+ intel_bw_init_hw(xe);
+
+ intel_display_device_info_runtime_init(xe);
+
+ err = intel_display_driver_probe_noirq(xe);
+ if (err)
+ return err;
+
+ return drmm_add_action_or_reset(&xe->drm, xe_display_fini_noirq, NULL);
+}
+
+static void xe_display_fini_noaccel(struct drm_device *dev, void *dummy)
+{
+ struct xe_device *xe = to_xe_device(dev);
+
+ if (!xe->info.enable_display)
+ return;
+
+ intel_display_driver_remove_nogem(xe);
+}
+
+int xe_display_init_noaccel(struct xe_device *xe)
+{
+ int err;
+
+ if (!xe->info.enable_display)
+ return 0;
+
+ err = intel_display_driver_probe_nogem(xe);
+ if (err)
+ return err;
+
+ return drmm_add_action_or_reset(&xe->drm, xe_display_fini_noaccel, NULL);
+}
+
+int xe_display_init(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return 0;
+
+ return intel_display_driver_probe(xe);
+}
+
+void xe_display_fini(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ /* poll work can call into fbdev, hence clean that up afterwards */
+ intel_hpd_poll_fini(xe);
+ intel_fbdev_fini(xe);
+
+ intel_hdcp_component_fini(xe);
+ intel_audio_deinit(xe);
+}
+
+void xe_display_register(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ intel_display_driver_register(xe);
+ intel_register_dsm_handler();
+ intel_power_domains_enable(xe);
+}
+
+void xe_display_unregister(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ intel_unregister_dsm_handler();
+ intel_power_domains_disable(xe);
+ intel_display_driver_unregister(xe);
+}
+
+void xe_display_driver_remove(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ intel_display_driver_remove(xe);
+
+ intel_display_device_remove(xe);
+}
+
+/* IRQ-related functions */
+
+void xe_display_irq_handler(struct xe_device *xe, u32 master_ctl)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ if (master_ctl & DISPLAY_IRQ)
+ gen11_display_irq_handler(xe);
+}
+
+void xe_display_irq_enable(struct xe_device *xe, u32 gu_misc_iir)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ if (gu_misc_iir & GU_MISC_GSE)
+ intel_opregion_asle_intr(xe);
+}
+
+void xe_display_irq_reset(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ gen11_display_irq_reset(xe);
+}
+
+void xe_display_irq_postinstall(struct xe_device *xe, struct xe_gt *gt)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ if (gt->info.id == XE_GT0)
+ gen11_de_irq_postinstall(xe);
+}
+
+static void intel_suspend_encoders(struct xe_device *xe)
+{
+ struct drm_device *dev = &xe->drm;
+ struct intel_encoder *encoder;
+
+ if (has_display(xe))
+ return;
+
+ drm_modeset_lock_all(dev);
+ for_each_intel_encoder(dev, encoder)
+ if (encoder->suspend)
+ encoder->suspend(encoder);
+ drm_modeset_unlock_all(dev);
+}
+
+static bool suspend_to_idle(void)
+{
+#if IS_ENABLED(CONFIG_ACPI_SLEEP)
+ if (acpi_target_system_state() < ACPI_STATE_S3)
+ return true;
+#endif
+ return false;
+}
+
+void xe_display_pm_suspend(struct xe_device *xe)
+{
+ bool s2idle = suspend_to_idle();
+ if (!xe->info.enable_display)
+ return;
+
+ /*
+ * We do a lot of poking in a lot of registers, make sure they work
+ * properly.
+ */
+ intel_power_domains_disable(xe);
+ if (has_display(xe))
+ drm_kms_helper_poll_disable(&xe->drm);
+
+ intel_display_driver_suspend(xe);
+
+ intel_dp_mst_suspend(xe);
+
+ intel_hpd_cancel_work(xe);
+
+ intel_suspend_encoders(xe);
+
+ intel_opregion_suspend(xe, s2idle ? PCI_D1 : PCI_D3cold);
+
+ intel_fbdev_set_suspend(&xe->drm, FBINFO_STATE_SUSPENDED, true);
+
+ intel_dmc_suspend(xe);
+}
+
+void xe_display_pm_suspend_late(struct xe_device *xe)
+{
+ bool s2idle = suspend_to_idle();
+ if (!xe->info.enable_display)
+ return;
+
+ intel_power_domains_suspend(xe, s2idle);
+
+ intel_display_power_suspend_late(xe);
+}
+
+void xe_display_pm_resume_early(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ intel_display_power_resume_early(xe);
+
+ intel_power_domains_resume(xe);
+}
+
+void xe_display_pm_resume(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ return;
+
+ intel_dmc_resume(xe);
+
+ if (has_display(xe))
+ drm_mode_config_reset(&xe->drm);
+
+ intel_display_driver_init_hw(xe);
+ intel_hpd_init(xe);
+
+ /* MST sideband requires HPD interrupts enabled */
+ intel_dp_mst_resume(xe);
+ intel_display_driver_resume(xe);
+
+ intel_hpd_poll_disable(xe);
+ if (has_display(xe))
+ drm_kms_helper_poll_enable(&xe->drm);
+
+ intel_opregion_resume(xe);
+
+ intel_fbdev_set_suspend(&xe->drm, FBINFO_STATE_RUNNING, false);
+
+ intel_power_domains_enable(xe);
+}
+
+void xe_display_probe(struct xe_device *xe)
+{
+ if (!xe->info.enable_display)
+ goto no_display;
+
+ intel_display_device_probe(xe);
+
+ if (has_display(xe))
+ return;
+
+no_display:
+ xe->info.enable_display = false;
+ unset_display_features(xe);
+}
diff --git a/drivers/gpu/drm/xe/xe_display.h b/drivers/gpu/drm/xe/xe_display.h
new file mode 100644
index 000000000000..710e56180b52
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_display.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DISPLAY_H_
+#define _XE_DISPLAY_H_
+
+#include "xe_device.h"
+
+struct drm_driver;
+
+#if IS_ENABLED(CONFIG_DRM_XE_DISPLAY)
+
+bool xe_display_driver_probe_defer(struct pci_dev *pdev);
+void xe_display_driver_set_hooks(struct drm_driver *driver);
+void xe_display_driver_remove(struct xe_device *xe);
+
+int xe_display_create(struct xe_device *xe);
+
+void xe_display_probe(struct xe_device *xe);
+
+int xe_display_init_nommio(struct xe_device *xe);
+int xe_display_init_noirq(struct xe_device *xe);
+int xe_display_init_noaccel(struct xe_device *xe);
+int xe_display_init(struct xe_device *xe);
+void xe_display_fini(struct xe_device *xe);
+
+void xe_display_register(struct xe_device *xe);
+void xe_display_unregister(struct xe_device *xe);
+
+void xe_display_irq_handler(struct xe_device *xe, u32 master_ctl);
+void xe_display_irq_enable(struct xe_device *xe, u32 gu_misc_iir);
+void xe_display_irq_reset(struct xe_device *xe);
+void xe_display_irq_postinstall(struct xe_device *xe, struct xe_gt *gt);
+
+void xe_display_pm_suspend(struct xe_device *xe);
+void xe_display_pm_suspend_late(struct xe_device *xe);
+void xe_display_pm_resume_early(struct xe_device *xe);
+void xe_display_pm_resume(struct xe_device *xe);
+
+#else
+
+static inline int xe_display_driver_probe_defer(struct pci_dev *pdev) { return 0; }
+static inline void xe_display_driver_set_hooks(struct drm_driver *driver) { }
+static inline void xe_display_driver_remove(struct xe_device *xe) {}
+
+static inline int xe_display_create(struct xe_device *xe) { return 0; }
+
+static inline void xe_display_probe(struct xe_device *xe) { }
+
+static inline int xe_display_init_nommio(struct xe_device *xe) { return 0; }
+static inline int xe_display_init_noirq(struct xe_device *xe) { return 0; }
+static inline int xe_display_init_noaccel(struct xe_device *xe) { return 0; }
+static inline int xe_display_init(struct xe_device *xe) { return 0; }
+static inline void xe_display_fini(struct xe_device *xe) {}
+
+static inline void xe_display_register(struct xe_device *xe) {}
+static inline void xe_display_unregister(struct xe_device *xe) {}
+
+static inline void xe_display_irq_handler(struct xe_device *xe, u32 master_ctl) {}
+static inline void xe_display_irq_enable(struct xe_device *xe, u32 gu_misc_iir) {}
+static inline void xe_display_irq_reset(struct xe_device *xe) {}
+static inline void xe_display_irq_postinstall(struct xe_device *xe, struct xe_gt *gt) {}
+
+static inline void xe_display_pm_suspend(struct xe_device *xe) {}
+static inline void xe_display_pm_suspend_late(struct xe_device *xe) {}
+static inline void xe_display_pm_resume_early(struct xe_device *xe) {}
+static inline void xe_display_pm_resume(struct xe_device *xe) {}
+
+#endif /* CONFIG_DRM_XE_DISPLAY */
+#endif /* _XE_DISPLAY_H_ */
diff --git a/drivers/gpu/drm/xe/xe_dma_buf.c b/drivers/gpu/drm/xe/xe_dma_buf.c
new file mode 100644
index 000000000000..64ed303728fd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_dma_buf.c
@@ -0,0 +1,322 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_dma_buf.h"
+
+#include <kunit/test.h>
+#include <linux/dma-buf.h>
+#include <linux/pci-p2pdma.h>
+
+#include <drm/drm_device.h>
+#include <drm/drm_prime.h>
+#include <drm/ttm/ttm_tt.h>
+
+#include "tests/xe_test.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_ttm_vram_mgr.h"
+#include "xe_vm.h"
+
+MODULE_IMPORT_NS(DMA_BUF);
+
+static int xe_dma_buf_attach(struct dma_buf *dmabuf,
+ struct dma_buf_attachment *attach)
+{
+ struct drm_gem_object *obj = attach->dmabuf->priv;
+
+ if (attach->peer2peer &&
+ pci_p2pdma_distance(to_pci_dev(obj->dev->dev), attach->dev, false) < 0)
+ attach->peer2peer = false;
+
+ if (!attach->peer2peer && !xe_bo_can_migrate(gem_to_xe_bo(obj), XE_PL_TT))
+ return -EOPNOTSUPP;
+
+ xe_device_mem_access_get(to_xe_device(obj->dev));
+ return 0;
+}
+
+static void xe_dma_buf_detach(struct dma_buf *dmabuf,
+ struct dma_buf_attachment *attach)
+{
+ struct drm_gem_object *obj = attach->dmabuf->priv;
+
+ xe_device_mem_access_put(to_xe_device(obj->dev));
+}
+
+static int xe_dma_buf_pin(struct dma_buf_attachment *attach)
+{
+ struct drm_gem_object *obj = attach->dmabuf->priv;
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+ struct xe_device *xe = xe_bo_device(bo);
+ int ret;
+
+ /*
+ * For now only support pinning in TT memory, for two reasons:
+ * 1) Avoid pinning in a placement not accessible to some importers.
+ * 2) Pinning in VRAM requires PIN accounting which is a to-do.
+ */
+ if (xe_bo_is_pinned(bo) && bo->ttm.resource->placement != XE_PL_TT) {
+ drm_dbg(&xe->drm, "Can't migrate pinned bo for dma-buf pin.\n");
+ return -EINVAL;
+ }
+
+ ret = xe_bo_migrate(bo, XE_PL_TT);
+ if (ret) {
+ if (ret != -EINTR && ret != -ERESTARTSYS)
+ drm_dbg(&xe->drm,
+ "Failed migrating dma-buf to TT memory: %pe\n",
+ ERR_PTR(ret));
+ return ret;
+ }
+
+ ret = xe_bo_pin_external(bo);
+ xe_assert(xe, !ret);
+
+ return 0;
+}
+
+static void xe_dma_buf_unpin(struct dma_buf_attachment *attach)
+{
+ struct drm_gem_object *obj = attach->dmabuf->priv;
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+
+ xe_bo_unpin_external(bo);
+}
+
+static struct sg_table *xe_dma_buf_map(struct dma_buf_attachment *attach,
+ enum dma_data_direction dir)
+{
+ struct dma_buf *dma_buf = attach->dmabuf;
+ struct drm_gem_object *obj = dma_buf->priv;
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+ struct sg_table *sgt;
+ int r = 0;
+
+ if (!attach->peer2peer && !xe_bo_can_migrate(bo, XE_PL_TT))
+ return ERR_PTR(-EOPNOTSUPP);
+
+ if (!xe_bo_is_pinned(bo)) {
+ if (!attach->peer2peer)
+ r = xe_bo_migrate(bo, XE_PL_TT);
+ else
+ r = xe_bo_validate(bo, NULL, false);
+ if (r)
+ return ERR_PTR(r);
+ }
+
+ switch (bo->ttm.resource->mem_type) {
+ case XE_PL_TT:
+ sgt = drm_prime_pages_to_sg(obj->dev,
+ bo->ttm.ttm->pages,
+ bo->ttm.ttm->num_pages);
+ if (IS_ERR(sgt))
+ return sgt;
+
+ if (dma_map_sgtable(attach->dev, sgt, dir,
+ DMA_ATTR_SKIP_CPU_SYNC))
+ goto error_free;
+ break;
+
+ case XE_PL_VRAM0:
+ case XE_PL_VRAM1:
+ r = xe_ttm_vram_mgr_alloc_sgt(xe_bo_device(bo),
+ bo->ttm.resource, 0,
+ bo->ttm.base.size, attach->dev,
+ dir, &sgt);
+ if (r)
+ return ERR_PTR(r);
+ break;
+ default:
+ return ERR_PTR(-EINVAL);
+ }
+
+ return sgt;
+
+error_free:
+ sg_free_table(sgt);
+ kfree(sgt);
+ return ERR_PTR(-EBUSY);
+}
+
+static void xe_dma_buf_unmap(struct dma_buf_attachment *attach,
+ struct sg_table *sgt,
+ enum dma_data_direction dir)
+{
+ struct dma_buf *dma_buf = attach->dmabuf;
+ struct xe_bo *bo = gem_to_xe_bo(dma_buf->priv);
+
+ if (!xe_bo_is_vram(bo)) {
+ dma_unmap_sgtable(attach->dev, sgt, dir, 0);
+ sg_free_table(sgt);
+ kfree(sgt);
+ } else {
+ xe_ttm_vram_mgr_free_sgt(attach->dev, dir, sgt);
+ }
+}
+
+static int xe_dma_buf_begin_cpu_access(struct dma_buf *dma_buf,
+ enum dma_data_direction direction)
+{
+ struct drm_gem_object *obj = dma_buf->priv;
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+ bool reads = (direction == DMA_BIDIRECTIONAL ||
+ direction == DMA_FROM_DEVICE);
+
+ if (!reads)
+ return 0;
+
+ /* Can we do interruptible lock here? */
+ xe_bo_lock(bo, false);
+ (void)xe_bo_migrate(bo, XE_PL_TT);
+ xe_bo_unlock(bo);
+
+ return 0;
+}
+
+const struct dma_buf_ops xe_dmabuf_ops = {
+ .attach = xe_dma_buf_attach,
+ .detach = xe_dma_buf_detach,
+ .pin = xe_dma_buf_pin,
+ .unpin = xe_dma_buf_unpin,
+ .map_dma_buf = xe_dma_buf_map,
+ .unmap_dma_buf = xe_dma_buf_unmap,
+ .release = drm_gem_dmabuf_release,
+ .begin_cpu_access = xe_dma_buf_begin_cpu_access,
+ .mmap = drm_gem_dmabuf_mmap,
+ .vmap = drm_gem_dmabuf_vmap,
+ .vunmap = drm_gem_dmabuf_vunmap,
+};
+
+struct dma_buf *xe_gem_prime_export(struct drm_gem_object *obj, int flags)
+{
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+ struct dma_buf *buf;
+
+ if (bo->vm)
+ return ERR_PTR(-EPERM);
+
+ buf = drm_gem_prime_export(obj, flags);
+ if (!IS_ERR(buf))
+ buf->ops = &xe_dmabuf_ops;
+
+ return buf;
+}
+
+static struct drm_gem_object *
+xe_dma_buf_init_obj(struct drm_device *dev, struct xe_bo *storage,
+ struct dma_buf *dma_buf)
+{
+ struct dma_resv *resv = dma_buf->resv;
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_bo *bo;
+ int ret;
+
+ dma_resv_lock(resv, NULL);
+ bo = ___xe_bo_create_locked(xe, storage, NULL, resv, NULL, dma_buf->size,
+ 0, /* Will require 1way or 2way for vm_bind */
+ ttm_bo_type_sg, XE_BO_CREATE_SYSTEM_BIT);
+ if (IS_ERR(bo)) {
+ ret = PTR_ERR(bo);
+ goto error;
+ }
+ dma_resv_unlock(resv);
+
+ return &bo->ttm.base;
+
+error:
+ dma_resv_unlock(resv);
+ return ERR_PTR(ret);
+}
+
+static void xe_dma_buf_move_notify(struct dma_buf_attachment *attach)
+{
+ struct drm_gem_object *obj = attach->importer_priv;
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+
+ XE_WARN_ON(xe_bo_evict(bo, false));
+}
+
+static const struct dma_buf_attach_ops xe_dma_buf_attach_ops = {
+ .allow_peer2peer = true,
+ .move_notify = xe_dma_buf_move_notify
+};
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+
+struct dma_buf_test_params {
+ struct xe_test_priv base;
+ const struct dma_buf_attach_ops *attach_ops;
+ bool force_different_devices;
+ u32 mem_mask;
+};
+
+#define to_dma_buf_test_params(_priv) \
+ container_of(_priv, struct dma_buf_test_params, base)
+#endif
+
+struct drm_gem_object *xe_gem_prime_import(struct drm_device *dev,
+ struct dma_buf *dma_buf)
+{
+ XE_TEST_DECLARE(struct dma_buf_test_params *test =
+ to_dma_buf_test_params
+ (xe_cur_kunit_priv(XE_TEST_LIVE_DMA_BUF));)
+ const struct dma_buf_attach_ops *attach_ops;
+ struct dma_buf_attachment *attach;
+ struct drm_gem_object *obj;
+ struct xe_bo *bo;
+
+ if (dma_buf->ops == &xe_dmabuf_ops) {
+ obj = dma_buf->priv;
+ if (obj->dev == dev &&
+ !XE_TEST_ONLY(test && test->force_different_devices)) {
+ /*
+ * Importing dmabuf exported from out own gem increases
+ * refcount on gem itself instead of f_count of dmabuf.
+ */
+ drm_gem_object_get(obj);
+ return obj;
+ }
+ }
+
+ /*
+ * Don't publish the bo until we have a valid attachment, and a
+ * valid attachment needs the bo address. So pre-create a bo before
+ * creating the attachment and publish.
+ */
+ bo = xe_bo_alloc();
+ if (IS_ERR(bo))
+ return ERR_CAST(bo);
+
+ attach_ops = &xe_dma_buf_attach_ops;
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+ if (test)
+ attach_ops = test->attach_ops;
+#endif
+
+ attach = dma_buf_dynamic_attach(dma_buf, dev->dev, attach_ops, &bo->ttm.base);
+ if (IS_ERR(attach)) {
+ obj = ERR_CAST(attach);
+ goto out_err;
+ }
+
+ /* Errors here will take care of freeing the bo. */
+ obj = xe_dma_buf_init_obj(dev, bo, dma_buf);
+ if (IS_ERR(obj))
+ return obj;
+
+
+ get_dma_buf(dma_buf);
+ obj->import_attach = attach;
+ return obj;
+
+out_err:
+ xe_bo_free(bo);
+
+ return obj;
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_dma_buf.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_dma_buf.h b/drivers/gpu/drm/xe/xe_dma_buf.h
new file mode 100644
index 000000000000..861dd28a862c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_dma_buf.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_DMA_BUF_H_
+#define _XE_DMA_BUF_H_
+
+#include <drm/drm_gem.h>
+
+struct dma_buf *xe_gem_prime_export(struct drm_gem_object *obj, int flags);
+struct drm_gem_object *xe_gem_prime_import(struct drm_device *dev,
+ struct dma_buf *dma_buf);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_drm_client.c b/drivers/gpu/drm/xe/xe_drm_client.c
new file mode 100644
index 000000000000..82d1305e831f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_drm_client.c
@@ -0,0 +1,204 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_print.h>
+#include <drm/xe_drm.h>
+#include <linux/kernel.h>
+#include <linux/slab.h>
+#include <linux/types.h>
+
+#include "xe_bo.h"
+#include "xe_bo_types.h"
+#include "xe_device_types.h"
+#include "xe_drm_client.h"
+#include "xe_trace.h"
+
+/**
+ * xe_drm_client_alloc() - Allocate drm client
+ * @void: No arg
+ *
+ * Allocate drm client struct to track client memory against
+ * same till client life. Call this API whenever new client
+ * has opened xe device.
+ *
+ * Return: pointer to client struct or NULL if can't allocate
+ */
+struct xe_drm_client *xe_drm_client_alloc(void)
+{
+ struct xe_drm_client *client;
+
+ client = kzalloc(sizeof(*client), GFP_KERNEL);
+ if (!client)
+ return NULL;
+
+ kref_init(&client->kref);
+
+#ifdef CONFIG_PROC_FS
+ spin_lock_init(&client->bos_lock);
+ INIT_LIST_HEAD(&client->bos_list);
+#endif
+ return client;
+}
+
+/**
+ * __xe_drm_client_free() - Free client struct
+ * @kref: The reference
+ *
+ * This frees client struct. Call this API when xe device is closed
+ * by drm client.
+ *
+ * Return: void
+ */
+void __xe_drm_client_free(struct kref *kref)
+{
+ struct xe_drm_client *client =
+ container_of(kref, typeof(*client), kref);
+
+ kfree(client);
+}
+
+#ifdef CONFIG_PROC_FS
+/**
+ * xe_drm_client_add_bo() - Add BO for tracking client mem usage
+ * @client: The drm client ptr
+ * @bo: The xe BO ptr
+ *
+ * Add all BO created by individual drm client by calling this function.
+ * This helps in tracking client memory usage.
+ *
+ * Return: void
+ */
+void xe_drm_client_add_bo(struct xe_drm_client *client,
+ struct xe_bo *bo)
+{
+ XE_WARN_ON(bo->client);
+ XE_WARN_ON(!list_empty(&bo->client_link));
+
+ spin_lock(&client->bos_lock);
+ bo->client = xe_drm_client_get(client);
+ list_add_tail_rcu(&bo->client_link, &client->bos_list);
+ spin_unlock(&client->bos_lock);
+}
+
+/**
+ * xe_drm_client_remove_bo() - Remove BO for tracking client mem usage
+ * @bo: The xe BO ptr
+ *
+ * Remove all BO removed by individual drm client by calling this function.
+ * This helps in tracking client memory usage.
+ *
+ * Return: void
+ */
+void xe_drm_client_remove_bo(struct xe_bo *bo)
+{
+ struct xe_drm_client *client = bo->client;
+
+ spin_lock(&client->bos_lock);
+ list_del_rcu(&bo->client_link);
+ spin_unlock(&client->bos_lock);
+
+ xe_drm_client_put(client);
+}
+
+static void bo_meminfo(struct xe_bo *bo,
+ struct drm_memory_stats stats[TTM_NUM_MEM_TYPES])
+{
+ u64 sz = bo->size;
+ u32 mem_type;
+
+ if (bo->placement.placement)
+ mem_type = bo->placement.placement->mem_type;
+ else
+ mem_type = XE_PL_TT;
+
+ if (bo->ttm.base.handle_count > 1)
+ stats[mem_type].shared += sz;
+ else
+ stats[mem_type].private += sz;
+
+ if (xe_bo_has_pages(bo)) {
+ stats[mem_type].resident += sz;
+
+ if (!dma_resv_test_signaled(bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP))
+ stats[mem_type].active += sz;
+ else if (mem_type == XE_PL_SYSTEM)
+ stats[mem_type].purgeable += sz;
+ }
+}
+
+static void show_meminfo(struct drm_printer *p, struct drm_file *file)
+{
+ static const char *const mem_type_to_name[TTM_NUM_MEM_TYPES] = {
+ [XE_PL_SYSTEM] = "system",
+ [XE_PL_TT] = "gtt",
+ [XE_PL_VRAM0] = "vram0",
+ [XE_PL_VRAM1] = "vram1",
+ [4 ... 6] = NULL,
+ [XE_PL_STOLEN] = "stolen"
+ };
+ struct drm_memory_stats stats[TTM_NUM_MEM_TYPES] = {};
+ struct xe_file *xef = file->driver_priv;
+ struct ttm_device *bdev = &xef->xe->ttm;
+ struct ttm_resource_manager *man;
+ struct xe_drm_client *client;
+ struct drm_gem_object *obj;
+ struct xe_bo *bo;
+ unsigned int id;
+ u32 mem_type;
+
+ client = xef->client;
+
+ /* Public objects. */
+ spin_lock(&file->table_lock);
+ idr_for_each_entry(&file->object_idr, obj, id) {
+ struct xe_bo *bo = gem_to_xe_bo(obj);
+
+ bo_meminfo(bo, stats);
+ }
+ spin_unlock(&file->table_lock);
+
+ /* Internal objects. */
+ spin_lock(&client->bos_lock);
+ list_for_each_entry_rcu(bo, &client->bos_list, client_link) {
+ if (!bo || !kref_get_unless_zero(&bo->ttm.base.refcount))
+ continue;
+ bo_meminfo(bo, stats);
+ xe_bo_put(bo);
+ }
+ spin_unlock(&client->bos_lock);
+
+ for (mem_type = XE_PL_SYSTEM; mem_type < TTM_NUM_MEM_TYPES; ++mem_type) {
+ if (!mem_type_to_name[mem_type])
+ continue;
+
+ man = ttm_manager_type(bdev, mem_type);
+
+ if (man) {
+ drm_print_memory_stats(p,
+ &stats[mem_type],
+ DRM_GEM_OBJECT_RESIDENT |
+ (mem_type != XE_PL_SYSTEM ? 0 :
+ DRM_GEM_OBJECT_PURGEABLE),
+ mem_type_to_name[mem_type]);
+ }
+ }
+}
+
+/**
+ * xe_drm_client_fdinfo() - Callback for fdinfo interface
+ * @p: The drm_printer ptr
+ * @file: The drm_file ptr
+ *
+ * This is callabck for drm fdinfo interface. Register this callback
+ * in drm driver ops for show_fdinfo.
+ *
+ * Return: void
+ */
+void xe_drm_client_fdinfo(struct drm_printer *p, struct drm_file *file)
+{
+ show_meminfo(p, file);
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_drm_client.h b/drivers/gpu/drm/xe/xe_drm_client.h
new file mode 100644
index 000000000000..a9649aa36011
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_drm_client.h
@@ -0,0 +1,70 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_DRM_CLIENT_H_
+#define _XE_DRM_CLIENT_H_
+
+#include <linux/kref.h>
+#include <linux/list.h>
+#include <linux/pid.h>
+#include <linux/rcupdate.h>
+#include <linux/sched.h>
+#include <linux/spinlock.h>
+
+struct drm_file;
+struct drm_printer;
+struct xe_bo;
+
+struct xe_drm_client {
+ struct kref kref;
+ unsigned int id;
+#ifdef CONFIG_PROC_FS
+ /**
+ * @bos_lock: lock protecting @bos_list
+ */
+ spinlock_t bos_lock;
+ /**
+ * @bos_list: list of bos created by this client
+ *
+ * Protected by @bos_lock.
+ */
+ struct list_head bos_list;
+#endif
+};
+
+ static inline struct xe_drm_client *
+xe_drm_client_get(struct xe_drm_client *client)
+{
+ kref_get(&client->kref);
+ return client;
+}
+
+void __xe_drm_client_free(struct kref *kref);
+
+static inline void xe_drm_client_put(struct xe_drm_client *client)
+{
+ kref_put(&client->kref, __xe_drm_client_free);
+}
+
+struct xe_drm_client *xe_drm_client_alloc(void);
+static inline struct xe_drm_client *
+xe_drm_client_get(struct xe_drm_client *client);
+static inline void xe_drm_client_put(struct xe_drm_client *client);
+#ifdef CONFIG_PROC_FS
+void xe_drm_client_fdinfo(struct drm_printer *p, struct drm_file *file);
+void xe_drm_client_add_bo(struct xe_drm_client *client,
+ struct xe_bo *bo);
+void xe_drm_client_remove_bo(struct xe_bo *bo);
+#else
+static inline void xe_drm_client_add_bo(struct xe_drm_client *client,
+ struct xe_bo *bo)
+{
+}
+
+static inline void xe_drm_client_remove_bo(struct xe_bo *bo)
+{
+}
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/xe_drv.h b/drivers/gpu/drm/xe/xe_drv.h
new file mode 100644
index 000000000000..d45b71426cc8
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_drv.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_DRV_H_
+#define _XE_DRV_H_
+
+#include <drm/drm_drv.h>
+
+#define DRIVER_NAME "xe"
+#define DRIVER_DESC "Intel Xe Graphics"
+#define DRIVER_DATE "20201103"
+
+/* Interface history:
+ *
+ * 1.1: Original.
+ */
+#define DRIVER_MAJOR 1
+#define DRIVER_MINOR 1
+#define DRIVER_PATCHLEVEL 0
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
new file mode 100644
index 000000000000..d30c0d0689bc
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -0,0 +1,350 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_exec.h"
+
+#include <drm/drm_device.h>
+#include <drm/drm_exec.h>
+#include <drm/drm_file.h>
+#include <drm/xe_drm.h>
+#include <linux/delay.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_macros.h"
+#include "xe_ring_ops_types.h"
+#include "xe_sched_job.h"
+#include "xe_sync.h"
+#include "xe_vm.h"
+
+/**
+ * DOC: Execbuf (User GPU command submission)
+ *
+ * Execs have historically been rather complicated in DRM drivers (at least in
+ * the i915) because a few things:
+ *
+ * - Passing in a list BO which are read / written to creating implicit syncs
+ * - Binding at exec time
+ * - Flow controlling the ring at exec time
+ *
+ * In XE we avoid all of this complication by not allowing a BO list to be
+ * passed into an exec, using the dma-buf implicit sync uAPI, have binds as
+ * seperate operations, and using the DRM scheduler to flow control the ring.
+ * Let's deep dive on each of these.
+ *
+ * We can get away from a BO list by forcing the user to use in / out fences on
+ * every exec rather than the kernel tracking dependencies of BO (e.g. if the
+ * user knows an exec writes to a BO and reads from the BO in the next exec, it
+ * is the user's responsibility to pass in / out fence between the two execs).
+ *
+ * Implicit dependencies for external BOs are handled by using the dma-buf
+ * implicit dependency uAPI (TODO: add link). To make this works each exec must
+ * install the job's fence into the DMA_RESV_USAGE_WRITE slot of every external
+ * BO mapped in the VM.
+ *
+ * We do not allow a user to trigger a bind at exec time rather we have a VM
+ * bind IOCTL which uses the same in / out fence interface as exec. In that
+ * sense, a VM bind is basically the same operation as an exec from the user
+ * perspective. e.g. If an exec depends on a VM bind use the in / out fence
+ * interface (struct drm_xe_sync) to synchronize like syncing between two
+ * dependent execs.
+ *
+ * Although a user cannot trigger a bind, we still have to rebind userptrs in
+ * the VM that have been invalidated since the last exec, likewise we also have
+ * to rebind BOs that have been evicted by the kernel. We schedule these rebinds
+ * behind any pending kernel operations on any external BOs in VM or any BOs
+ * private to the VM. This is accomplished by the rebinds waiting on BOs
+ * DMA_RESV_USAGE_KERNEL slot (kernel ops) and kernel ops waiting on all BOs
+ * slots (inflight execs are in the DMA_RESV_USAGE_BOOKING for private BOs and
+ * in DMA_RESV_USAGE_WRITE for external BOs).
+ *
+ * Rebinds / dma-resv usage applies to non-compute mode VMs only as for compute
+ * mode VMs we use preempt fences and a rebind worker (TODO: add link).
+ *
+ * There is no need to flow control the ring in the exec as we write the ring at
+ * submission time and set the DRM scheduler max job limit SIZE_OF_RING /
+ * MAX_JOB_SIZE. The DRM scheduler will then hold all jobs until space in the
+ * ring is available.
+ *
+ * All of this results in a rather simple exec implementation.
+ *
+ * Flow
+ * ~~~~
+ *
+ * .. code-block::
+ *
+ * Parse input arguments
+ * Wait for any async VM bind passed as in-fences to start
+ * <----------------------------------------------------------------------|
+ * Lock global VM lock in read mode |
+ * Pin userptrs (also finds userptr invalidated since last exec) |
+ * Lock exec (VM dma-resv lock, external BOs dma-resv locks) |
+ * Validate BOs that have been evicted |
+ * Create job |
+ * Rebind invalidated userptrs + evicted BOs (non-compute-mode) |
+ * Add rebind fence dependency to job |
+ * Add job VM dma-resv bookkeeping slot (non-compute mode) |
+ * Add job to external BOs dma-resv write slots (non-compute mode) |
+ * Check if any userptrs invalidated since pin ------ Drop locks ---------|
+ * Install in / out fences for job
+ * Submit job
+ * Unlock all
+ */
+
+static int xe_exec_fn(struct drm_gpuvm_exec *vm_exec)
+{
+ return drm_gpuvm_validate(vm_exec->vm, &vm_exec->exec);
+}
+
+int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_exec *args = data;
+ struct drm_xe_sync __user *syncs_user = u64_to_user_ptr(args->syncs);
+ u64 __user *addresses_user = u64_to_user_ptr(args->address);
+ struct xe_exec_queue *q;
+ struct xe_sync_entry *syncs = NULL;
+ u64 addresses[XE_HW_ENGINE_MAX_INSTANCE];
+ struct drm_gpuvm_exec vm_exec = {.extra.fn = xe_exec_fn};
+ struct drm_exec *exec = &vm_exec.exec;
+ u32 i, num_syncs = 0;
+ struct xe_sched_job *job;
+ struct dma_fence *rebind_fence;
+ struct xe_vm *vm;
+ bool write_locked;
+ ktime_t end = 0;
+ int err = 0;
+
+ if (XE_IOCTL_DBG(xe, args->extensions) ||
+ XE_IOCTL_DBG(xe, args->pad[0] || args->pad[1] || args->pad[2]) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ q = xe_exec_queue_lookup(xef, args->exec_queue_id);
+ if (XE_IOCTL_DBG(xe, !q))
+ return -ENOENT;
+
+ if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->num_batch_buffer &&
+ q->width != args->num_batch_buffer))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_BANNED)) {
+ err = -ECANCELED;
+ goto err_exec_queue;
+ }
+
+ if (args->num_syncs) {
+ syncs = kcalloc(args->num_syncs, sizeof(*syncs), GFP_KERNEL);
+ if (!syncs) {
+ err = -ENOMEM;
+ goto err_exec_queue;
+ }
+ }
+
+ vm = q->vm;
+
+ for (i = 0; i < args->num_syncs; i++) {
+ err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs++],
+ &syncs_user[i], SYNC_PARSE_FLAG_EXEC |
+ (xe_vm_in_lr_mode(vm) ?
+ SYNC_PARSE_FLAG_LR_MODE : 0));
+ if (err)
+ goto err_syncs;
+ }
+
+ if (xe_exec_queue_is_parallel(q)) {
+ err = __copy_from_user(addresses, addresses_user, sizeof(u64) *
+ q->width);
+ if (err) {
+ err = -EFAULT;
+ goto err_syncs;
+ }
+ }
+
+retry:
+ if (!xe_vm_in_lr_mode(vm) && xe_vm_userptr_check_repin(vm)) {
+ err = down_write_killable(&vm->lock);
+ write_locked = true;
+ } else {
+ /* We don't allow execs while the VM is in error state */
+ err = down_read_interruptible(&vm->lock);
+ write_locked = false;
+ }
+ if (err)
+ goto err_syncs;
+
+ if (write_locked) {
+ err = xe_vm_userptr_pin(vm);
+ downgrade_write(&vm->lock);
+ write_locked = false;
+ if (err)
+ goto err_unlock_list;
+ }
+
+ vm_exec.vm = &vm->gpuvm;
+ vm_exec.num_fences = 1 + vm->xe->info.tile_count;
+ vm_exec.flags = DRM_EXEC_INTERRUPTIBLE_WAIT;
+ if (xe_vm_in_lr_mode(vm)) {
+ drm_exec_init(exec, vm_exec.flags, 0);
+ } else {
+ err = drm_gpuvm_exec_lock(&vm_exec);
+ if (err) {
+ if (xe_vm_validate_should_retry(exec, err, &end))
+ err = -EAGAIN;
+ goto err_unlock_list;
+ }
+ }
+
+ if (xe_vm_is_closed_or_banned(q->vm)) {
+ drm_warn(&xe->drm, "Trying to schedule after vm is closed or banned\n");
+ err = -ECANCELED;
+ goto err_exec;
+ }
+
+ if (!args->num_batch_buffer) {
+ if (!xe_vm_in_lr_mode(vm)) {
+ struct dma_fence *fence;
+
+ fence = xe_sync_in_fence_get(syncs, num_syncs, q, vm);
+ if (IS_ERR(fence)) {
+ err = PTR_ERR(fence);
+ goto err_exec;
+ }
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ xe_exec_queue_last_fence_set(q, vm, fence);
+ dma_fence_put(fence);
+ }
+
+ goto err_exec;
+ }
+
+ if (xe_exec_queue_is_lr(q) && xe_exec_queue_ring_full(q)) {
+ err = -EWOULDBLOCK;
+ goto err_exec;
+ }
+
+ job = xe_sched_job_create(q, xe_exec_queue_is_parallel(q) ?
+ addresses : &args->address);
+ if (IS_ERR(job)) {
+ err = PTR_ERR(job);
+ goto err_exec;
+ }
+
+ /*
+ * Rebind any invalidated userptr or evicted BOs in the VM, non-compute
+ * VM mode only.
+ */
+ rebind_fence = xe_vm_rebind(vm, false);
+ if (IS_ERR(rebind_fence)) {
+ err = PTR_ERR(rebind_fence);
+ goto err_put_job;
+ }
+
+ /*
+ * We store the rebind_fence in the VM so subsequent execs don't get
+ * scheduled before the rebinds of userptrs / evicted BOs is complete.
+ */
+ if (rebind_fence) {
+ dma_fence_put(vm->rebind_fence);
+ vm->rebind_fence = rebind_fence;
+ }
+ if (vm->rebind_fence) {
+ if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+ &vm->rebind_fence->flags)) {
+ dma_fence_put(vm->rebind_fence);
+ vm->rebind_fence = NULL;
+ } else {
+ dma_fence_get(vm->rebind_fence);
+ err = drm_sched_job_add_dependency(&job->drm,
+ vm->rebind_fence);
+ if (err)
+ goto err_put_job;
+ }
+ }
+
+ /* Wait behind munmap style rebinds */
+ if (!xe_vm_in_lr_mode(vm)) {
+ err = drm_sched_job_add_resv_dependencies(&job->drm,
+ xe_vm_resv(vm),
+ DMA_RESV_USAGE_KERNEL);
+ if (err)
+ goto err_put_job;
+ }
+
+ for (i = 0; i < num_syncs && !err; i++)
+ err = xe_sync_entry_add_deps(&syncs[i], job);
+ if (err)
+ goto err_put_job;
+
+ if (!xe_vm_in_lr_mode(vm)) {
+ err = xe_sched_job_last_fence_add_dep(job, vm);
+ if (err)
+ goto err_put_job;
+
+ err = down_read_interruptible(&vm->userptr.notifier_lock);
+ if (err)
+ goto err_put_job;
+
+ err = __xe_vm_userptr_needs_repin(vm);
+ if (err)
+ goto err_repin;
+ }
+
+ /*
+ * Point of no return, if we error after this point just set an error on
+ * the job and let the DRM scheduler / backend clean up the job.
+ */
+ xe_sched_job_arm(job);
+ if (!xe_vm_in_lr_mode(vm))
+ drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, &job->drm.s_fence->finished,
+ DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_WRITE);
+
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], job,
+ &job->drm.s_fence->finished);
+
+ if (xe_exec_queue_is_lr(q))
+ q->ring_ops->emit_job(job);
+ if (!xe_vm_in_lr_mode(vm))
+ xe_exec_queue_last_fence_set(q, vm, &job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+ xe_vm_reactivate_rebind(vm);
+
+ if (!err && !xe_vm_in_lr_mode(vm)) {
+ spin_lock(&xe->ttm.lru_lock);
+ ttm_lru_bulk_move_tail(&vm->lru_bulk_move);
+ spin_unlock(&xe->ttm.lru_lock);
+ }
+
+err_repin:
+ if (!xe_vm_in_lr_mode(vm))
+ up_read(&vm->userptr.notifier_lock);
+err_put_job:
+ if (err)
+ xe_sched_job_put(job);
+err_exec:
+ drm_exec_fini(exec);
+err_unlock_list:
+ if (write_locked)
+ up_write(&vm->lock);
+ else
+ up_read(&vm->lock);
+ if (err == -EAGAIN)
+ goto retry;
+err_syncs:
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_cleanup(&syncs[i]);
+ kfree(syncs);
+err_exec_queue:
+ xe_exec_queue_put(q);
+
+ return err;
+}
diff --git a/drivers/gpu/drm/xe/xe_exec.h b/drivers/gpu/drm/xe/xe_exec.h
new file mode 100644
index 000000000000..e4932494cea3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_exec.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_EXEC_H_
+#define _XE_EXEC_H_
+
+struct drm_device;
+struct drm_file;
+
+int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_exec_queue.c b/drivers/gpu/drm/xe/xe_exec_queue.c
new file mode 100644
index 000000000000..44fe8097b7cd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_exec_queue.c
@@ -0,0 +1,956 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_exec_queue.h"
+
+#include <linux/nospec.h>
+
+#include <drm/drm_device.h>
+#include <drm/drm_file.h>
+#include <drm/xe_drm.h>
+
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_hw_engine_class_sysfs.h"
+#include "xe_hw_fence.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_migrate.h"
+#include "xe_pm.h"
+#include "xe_ring_ops_types.h"
+#include "xe_trace.h"
+#include "xe_vm.h"
+
+enum xe_exec_queue_sched_prop {
+ XE_EXEC_QUEUE_JOB_TIMEOUT = 0,
+ XE_EXEC_QUEUE_TIMESLICE = 1,
+ XE_EXEC_QUEUE_PREEMPT_TIMEOUT = 2,
+ XE_EXEC_QUEUE_SCHED_PROP_MAX = 3,
+};
+
+static struct xe_exec_queue *__xe_exec_queue_create(struct xe_device *xe,
+ struct xe_vm *vm,
+ u32 logical_mask,
+ u16 width, struct xe_hw_engine *hwe,
+ u32 flags)
+{
+ struct xe_exec_queue *q;
+ struct xe_gt *gt = hwe->gt;
+ int err;
+ int i;
+
+ /* only kernel queues can be permanent */
+ XE_WARN_ON((flags & EXEC_QUEUE_FLAG_PERMANENT) && !(flags & EXEC_QUEUE_FLAG_KERNEL));
+
+ q = kzalloc(sizeof(*q) + sizeof(struct xe_lrc) * width, GFP_KERNEL);
+ if (!q)
+ return ERR_PTR(-ENOMEM);
+
+ kref_init(&q->refcount);
+ q->flags = flags;
+ q->hwe = hwe;
+ q->gt = gt;
+ if (vm)
+ q->vm = xe_vm_get(vm);
+ q->class = hwe->class;
+ q->width = width;
+ q->logical_mask = logical_mask;
+ q->fence_irq = &gt->fence_irq[hwe->class];
+ q->ring_ops = gt->ring_ops[hwe->class];
+ q->ops = gt->exec_queue_ops;
+ INIT_LIST_HEAD(&q->persistent.link);
+ INIT_LIST_HEAD(&q->compute.link);
+ INIT_LIST_HEAD(&q->multi_gt_link);
+
+ q->sched_props.timeslice_us = hwe->eclass->sched_props.timeslice_us;
+ q->sched_props.preempt_timeout_us =
+ hwe->eclass->sched_props.preempt_timeout_us;
+
+ if (xe_exec_queue_is_parallel(q)) {
+ q->parallel.composite_fence_ctx = dma_fence_context_alloc(1);
+ q->parallel.composite_fence_seqno = XE_FENCE_INITIAL_SEQNO;
+ }
+ if (q->flags & EXEC_QUEUE_FLAG_VM) {
+ q->bind.fence_ctx = dma_fence_context_alloc(1);
+ q->bind.fence_seqno = XE_FENCE_INITIAL_SEQNO;
+ }
+
+ for (i = 0; i < width; ++i) {
+ err = xe_lrc_init(q->lrc + i, hwe, q, vm, SZ_16K);
+ if (err)
+ goto err_lrc;
+ }
+
+ err = q->ops->init(q);
+ if (err)
+ goto err_lrc;
+
+ /*
+ * Normally the user vm holds an rpm ref to keep the device
+ * awake, and the context holds a ref for the vm, however for
+ * some engines we use the kernels migrate vm underneath which offers no
+ * such rpm ref, or we lack a vm. Make sure we keep a ref here, so we
+ * can perform GuC CT actions when needed. Caller is expected to have
+ * already grabbed the rpm ref outside any sensitive locks.
+ */
+ if (!(q->flags & EXEC_QUEUE_FLAG_PERMANENT) && (q->flags & EXEC_QUEUE_FLAG_VM || !vm))
+ drm_WARN_ON(&xe->drm, !xe_device_mem_access_get_if_ongoing(xe));
+
+ return q;
+
+err_lrc:
+ for (i = i - 1; i >= 0; --i)
+ xe_lrc_finish(q->lrc + i);
+ kfree(q);
+ return ERR_PTR(err);
+}
+
+struct xe_exec_queue *xe_exec_queue_create(struct xe_device *xe, struct xe_vm *vm,
+ u32 logical_mask, u16 width,
+ struct xe_hw_engine *hwe, u32 flags)
+{
+ struct xe_exec_queue *q;
+ int err;
+
+ if (vm) {
+ err = xe_vm_lock(vm, true);
+ if (err)
+ return ERR_PTR(err);
+ }
+ q = __xe_exec_queue_create(xe, vm, logical_mask, width, hwe, flags);
+ if (vm)
+ xe_vm_unlock(vm);
+
+ return q;
+}
+
+struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe_gt *gt,
+ struct xe_vm *vm,
+ enum xe_engine_class class, u32 flags)
+{
+ struct xe_hw_engine *hwe, *hwe0 = NULL;
+ enum xe_hw_engine_id id;
+ u32 logical_mask = 0;
+
+ for_each_hw_engine(hwe, gt, id) {
+ if (xe_hw_engine_is_reserved(hwe))
+ continue;
+
+ if (hwe->class == class) {
+ logical_mask |= BIT(hwe->logical_instance);
+ if (!hwe0)
+ hwe0 = hwe;
+ }
+ }
+
+ if (!logical_mask)
+ return ERR_PTR(-ENODEV);
+
+ return xe_exec_queue_create(xe, vm, logical_mask, 1, hwe0, flags);
+}
+
+void xe_exec_queue_destroy(struct kref *ref)
+{
+ struct xe_exec_queue *q = container_of(ref, struct xe_exec_queue, refcount);
+ struct xe_exec_queue *eq, *next;
+
+ xe_exec_queue_last_fence_put_unlocked(q);
+ if (!(q->flags & EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD)) {
+ list_for_each_entry_safe(eq, next, &q->multi_gt_list,
+ multi_gt_link)
+ xe_exec_queue_put(eq);
+ }
+
+ q->ops->fini(q);
+}
+
+void xe_exec_queue_fini(struct xe_exec_queue *q)
+{
+ int i;
+
+ for (i = 0; i < q->width; ++i)
+ xe_lrc_finish(q->lrc + i);
+ if (!(q->flags & EXEC_QUEUE_FLAG_PERMANENT) && (q->flags & EXEC_QUEUE_FLAG_VM || !q->vm))
+ xe_device_mem_access_put(gt_to_xe(q->gt));
+ if (q->vm)
+ xe_vm_put(q->vm);
+
+ kfree(q);
+}
+
+void xe_exec_queue_assign_name(struct xe_exec_queue *q, u32 instance)
+{
+ switch (q->class) {
+ case XE_ENGINE_CLASS_RENDER:
+ sprintf(q->name, "rcs%d", instance);
+ break;
+ case XE_ENGINE_CLASS_VIDEO_DECODE:
+ sprintf(q->name, "vcs%d", instance);
+ break;
+ case XE_ENGINE_CLASS_VIDEO_ENHANCE:
+ sprintf(q->name, "vecs%d", instance);
+ break;
+ case XE_ENGINE_CLASS_COPY:
+ sprintf(q->name, "bcs%d", instance);
+ break;
+ case XE_ENGINE_CLASS_COMPUTE:
+ sprintf(q->name, "ccs%d", instance);
+ break;
+ case XE_ENGINE_CLASS_OTHER:
+ sprintf(q->name, "gsccs%d", instance);
+ break;
+ default:
+ XE_WARN_ON(q->class);
+ }
+}
+
+struct xe_exec_queue *xe_exec_queue_lookup(struct xe_file *xef, u32 id)
+{
+ struct xe_exec_queue *q;
+
+ mutex_lock(&xef->exec_queue.lock);
+ q = xa_load(&xef->exec_queue.xa, id);
+ if (q)
+ xe_exec_queue_get(q);
+ mutex_unlock(&xef->exec_queue.lock);
+
+ return q;
+}
+
+enum xe_exec_queue_priority
+xe_exec_queue_device_get_max_priority(struct xe_device *xe)
+{
+ return capable(CAP_SYS_NICE) ? XE_EXEC_QUEUE_PRIORITY_HIGH :
+ XE_EXEC_QUEUE_PRIORITY_NORMAL;
+}
+
+static int exec_queue_set_priority(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ if (XE_IOCTL_DBG(xe, value > XE_EXEC_QUEUE_PRIORITY_HIGH))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, value > xe_exec_queue_device_get_max_priority(xe)))
+ return -EPERM;
+
+ return q->ops->set_priority(q, value);
+}
+
+static bool xe_exec_queue_enforce_schedule_limit(void)
+{
+#if IS_ENABLED(CONFIG_DRM_XE_ENABLE_SCHEDTIMEOUT_LIMIT)
+ return true;
+#else
+ return !capable(CAP_SYS_NICE);
+#endif
+}
+
+static void
+xe_exec_queue_get_prop_minmax(struct xe_hw_engine_class_intf *eclass,
+ enum xe_exec_queue_sched_prop prop,
+ u32 *min, u32 *max)
+{
+ switch (prop) {
+ case XE_EXEC_QUEUE_JOB_TIMEOUT:
+ *min = eclass->sched_props.job_timeout_min;
+ *max = eclass->sched_props.job_timeout_max;
+ break;
+ case XE_EXEC_QUEUE_TIMESLICE:
+ *min = eclass->sched_props.timeslice_min;
+ *max = eclass->sched_props.timeslice_max;
+ break;
+ case XE_EXEC_QUEUE_PREEMPT_TIMEOUT:
+ *min = eclass->sched_props.preempt_timeout_min;
+ *max = eclass->sched_props.preempt_timeout_max;
+ break;
+ default:
+ break;
+ }
+#if IS_ENABLED(CONFIG_DRM_XE_ENABLE_SCHEDTIMEOUT_LIMIT)
+ if (capable(CAP_SYS_NICE)) {
+ switch (prop) {
+ case XE_EXEC_QUEUE_JOB_TIMEOUT:
+ *min = XE_HW_ENGINE_JOB_TIMEOUT_MIN;
+ *max = XE_HW_ENGINE_JOB_TIMEOUT_MAX;
+ break;
+ case XE_EXEC_QUEUE_TIMESLICE:
+ *min = XE_HW_ENGINE_TIMESLICE_MIN;
+ *max = XE_HW_ENGINE_TIMESLICE_MAX;
+ break;
+ case XE_EXEC_QUEUE_PREEMPT_TIMEOUT:
+ *min = XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN;
+ *max = XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX;
+ break;
+ default:
+ break;
+ }
+ }
+#endif
+}
+
+static int exec_queue_set_timeslice(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ u32 min = 0, max = 0;
+
+ xe_exec_queue_get_prop_minmax(q->hwe->eclass,
+ XE_EXEC_QUEUE_TIMESLICE, &min, &max);
+
+ if (xe_exec_queue_enforce_schedule_limit() &&
+ !xe_hw_engine_timeout_in_range(value, min, max))
+ return -EINVAL;
+
+ return q->ops->set_timeslice(q, value);
+}
+
+static int exec_queue_set_preemption_timeout(struct xe_device *xe,
+ struct xe_exec_queue *q, u64 value,
+ bool create)
+{
+ u32 min = 0, max = 0;
+
+ xe_exec_queue_get_prop_minmax(q->hwe->eclass,
+ XE_EXEC_QUEUE_PREEMPT_TIMEOUT, &min, &max);
+
+ if (xe_exec_queue_enforce_schedule_limit() &&
+ !xe_hw_engine_timeout_in_range(value, min, max))
+ return -EINVAL;
+
+ return q->ops->set_preempt_timeout(q, value);
+}
+
+static int exec_queue_set_persistence(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ if (XE_IOCTL_DBG(xe, !create))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, xe_vm_in_preempt_fence_mode(q->vm)))
+ return -EINVAL;
+
+ if (value)
+ q->flags |= EXEC_QUEUE_FLAG_PERSISTENT;
+ else
+ q->flags &= ~EXEC_QUEUE_FLAG_PERSISTENT;
+
+ return 0;
+}
+
+static int exec_queue_set_job_timeout(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ u32 min = 0, max = 0;
+
+ if (XE_IOCTL_DBG(xe, !create))
+ return -EINVAL;
+
+ xe_exec_queue_get_prop_minmax(q->hwe->eclass,
+ XE_EXEC_QUEUE_JOB_TIMEOUT, &min, &max);
+
+ if (xe_exec_queue_enforce_schedule_limit() &&
+ !xe_hw_engine_timeout_in_range(value, min, max))
+ return -EINVAL;
+
+ return q->ops->set_job_timeout(q, value);
+}
+
+static int exec_queue_set_acc_trigger(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ if (XE_IOCTL_DBG(xe, !create))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !xe->info.has_usm))
+ return -EINVAL;
+
+ q->usm.acc_trigger = value;
+
+ return 0;
+}
+
+static int exec_queue_set_acc_notify(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ if (XE_IOCTL_DBG(xe, !create))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !xe->info.has_usm))
+ return -EINVAL;
+
+ q->usm.acc_notify = value;
+
+ return 0;
+}
+
+static int exec_queue_set_acc_granularity(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 value, bool create)
+{
+ if (XE_IOCTL_DBG(xe, !create))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !xe->info.has_usm))
+ return -EINVAL;
+
+ if (value > DRM_XE_ACC_GRANULARITY_64M)
+ return -EINVAL;
+
+ q->usm.acc_granularity = value;
+
+ return 0;
+}
+
+typedef int (*xe_exec_queue_set_property_fn)(struct xe_device *xe,
+ struct xe_exec_queue *q,
+ u64 value, bool create);
+
+static const xe_exec_queue_set_property_fn exec_queue_set_property_funcs[] = {
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_PRIORITY] = exec_queue_set_priority,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_TIMESLICE] = exec_queue_set_timeslice,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_PREEMPTION_TIMEOUT] = exec_queue_set_preemption_timeout,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_PERSISTENCE] = exec_queue_set_persistence,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_JOB_TIMEOUT] = exec_queue_set_job_timeout,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_TRIGGER] = exec_queue_set_acc_trigger,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_NOTIFY] = exec_queue_set_acc_notify,
+ [DRM_XE_EXEC_QUEUE_SET_PROPERTY_ACC_GRANULARITY] = exec_queue_set_acc_granularity,
+};
+
+static int exec_queue_user_ext_set_property(struct xe_device *xe,
+ struct xe_exec_queue *q,
+ u64 extension,
+ bool create)
+{
+ u64 __user *address = u64_to_user_ptr(extension);
+ struct drm_xe_ext_set_property ext;
+ int err;
+ u32 idx;
+
+ err = __copy_from_user(&ext, address, sizeof(ext));
+ if (XE_IOCTL_DBG(xe, err))
+ return -EFAULT;
+
+ if (XE_IOCTL_DBG(xe, ext.property >=
+ ARRAY_SIZE(exec_queue_set_property_funcs)) ||
+ XE_IOCTL_DBG(xe, ext.pad))
+ return -EINVAL;
+
+ idx = array_index_nospec(ext.property, ARRAY_SIZE(exec_queue_set_property_funcs));
+ return exec_queue_set_property_funcs[idx](xe, q, ext.value, create);
+}
+
+typedef int (*xe_exec_queue_user_extension_fn)(struct xe_device *xe,
+ struct xe_exec_queue *q,
+ u64 extension,
+ bool create);
+
+static const xe_exec_queue_set_property_fn exec_queue_user_extension_funcs[] = {
+ [DRM_XE_EXEC_QUEUE_EXTENSION_SET_PROPERTY] = exec_queue_user_ext_set_property,
+};
+
+#define MAX_USER_EXTENSIONS 16
+static int exec_queue_user_extensions(struct xe_device *xe, struct xe_exec_queue *q,
+ u64 extensions, int ext_number, bool create)
+{
+ u64 __user *address = u64_to_user_ptr(extensions);
+ struct drm_xe_user_extension ext;
+ int err;
+ u32 idx;
+
+ if (XE_IOCTL_DBG(xe, ext_number >= MAX_USER_EXTENSIONS))
+ return -E2BIG;
+
+ err = __copy_from_user(&ext, address, sizeof(ext));
+ if (XE_IOCTL_DBG(xe, err))
+ return -EFAULT;
+
+ if (XE_IOCTL_DBG(xe, ext.pad) ||
+ XE_IOCTL_DBG(xe, ext.name >=
+ ARRAY_SIZE(exec_queue_user_extension_funcs)))
+ return -EINVAL;
+
+ idx = array_index_nospec(ext.name,
+ ARRAY_SIZE(exec_queue_user_extension_funcs));
+ err = exec_queue_user_extension_funcs[idx](xe, q, extensions, create);
+ if (XE_IOCTL_DBG(xe, err))
+ return err;
+
+ if (ext.next_extension)
+ return exec_queue_user_extensions(xe, q, ext.next_extension,
+ ++ext_number, create);
+
+ return 0;
+}
+
+static const enum xe_engine_class user_to_xe_engine_class[] = {
+ [DRM_XE_ENGINE_CLASS_RENDER] = XE_ENGINE_CLASS_RENDER,
+ [DRM_XE_ENGINE_CLASS_COPY] = XE_ENGINE_CLASS_COPY,
+ [DRM_XE_ENGINE_CLASS_VIDEO_DECODE] = XE_ENGINE_CLASS_VIDEO_DECODE,
+ [DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE] = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ [DRM_XE_ENGINE_CLASS_COMPUTE] = XE_ENGINE_CLASS_COMPUTE,
+};
+
+static struct xe_hw_engine *
+find_hw_engine(struct xe_device *xe,
+ struct drm_xe_engine_class_instance eci)
+{
+ u32 idx;
+
+ if (eci.engine_class > ARRAY_SIZE(user_to_xe_engine_class))
+ return NULL;
+
+ if (eci.gt_id >= xe->info.gt_count)
+ return NULL;
+
+ idx = array_index_nospec(eci.engine_class,
+ ARRAY_SIZE(user_to_xe_engine_class));
+
+ return xe_gt_hw_engine(xe_device_get_gt(xe, eci.gt_id),
+ user_to_xe_engine_class[idx],
+ eci.engine_instance, true);
+}
+
+static u32 bind_exec_queue_logical_mask(struct xe_device *xe, struct xe_gt *gt,
+ struct drm_xe_engine_class_instance *eci,
+ u16 width, u16 num_placements)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ u32 logical_mask = 0;
+
+ if (XE_IOCTL_DBG(xe, width != 1))
+ return 0;
+ if (XE_IOCTL_DBG(xe, num_placements != 1))
+ return 0;
+ if (XE_IOCTL_DBG(xe, eci[0].engine_instance != 0))
+ return 0;
+
+ eci[0].engine_class = DRM_XE_ENGINE_CLASS_COPY;
+
+ for_each_hw_engine(hwe, gt, id) {
+ if (xe_hw_engine_is_reserved(hwe))
+ continue;
+
+ if (hwe->class ==
+ user_to_xe_engine_class[DRM_XE_ENGINE_CLASS_COPY])
+ logical_mask |= BIT(hwe->logical_instance);
+ }
+
+ return logical_mask;
+}
+
+static u32 calc_validate_logical_mask(struct xe_device *xe, struct xe_gt *gt,
+ struct drm_xe_engine_class_instance *eci,
+ u16 width, u16 num_placements)
+{
+ int len = width * num_placements;
+ int i, j, n;
+ u16 class;
+ u16 gt_id;
+ u32 return_mask = 0, prev_mask;
+
+ if (XE_IOCTL_DBG(xe, !xe_device_uc_enabled(xe) &&
+ len > 1))
+ return 0;
+
+ for (i = 0; i < width; ++i) {
+ u32 current_mask = 0;
+
+ for (j = 0; j < num_placements; ++j) {
+ struct xe_hw_engine *hwe;
+
+ n = j * width + i;
+
+ hwe = find_hw_engine(xe, eci[n]);
+ if (XE_IOCTL_DBG(xe, !hwe))
+ return 0;
+
+ if (XE_IOCTL_DBG(xe, xe_hw_engine_is_reserved(hwe)))
+ return 0;
+
+ if (XE_IOCTL_DBG(xe, n && eci[n].gt_id != gt_id) ||
+ XE_IOCTL_DBG(xe, n && eci[n].engine_class != class))
+ return 0;
+
+ class = eci[n].engine_class;
+ gt_id = eci[n].gt_id;
+
+ if (width == 1 || !i)
+ return_mask |= BIT(eci[n].engine_instance);
+ current_mask |= BIT(eci[n].engine_instance);
+ }
+
+ /* Parallel submissions must be logically contiguous */
+ if (i && XE_IOCTL_DBG(xe, current_mask != prev_mask << 1))
+ return 0;
+
+ prev_mask = current_mask;
+ }
+
+ return return_mask;
+}
+
+int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_exec_queue_create *args = data;
+ struct drm_xe_engine_class_instance eci[XE_HW_ENGINE_MAX_INSTANCE];
+ struct drm_xe_engine_class_instance __user *user_eci =
+ u64_to_user_ptr(args->instances);
+ struct xe_hw_engine *hwe;
+ struct xe_vm *vm, *migrate_vm;
+ struct xe_gt *gt;
+ struct xe_exec_queue *q = NULL;
+ u32 logical_mask;
+ u32 id;
+ u32 len;
+ int err;
+
+ if (XE_IOCTL_DBG(xe, args->flags) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ len = args->width * args->num_placements;
+ if (XE_IOCTL_DBG(xe, !len || len > XE_HW_ENGINE_MAX_INSTANCE))
+ return -EINVAL;
+
+ err = __copy_from_user(eci, user_eci,
+ sizeof(struct drm_xe_engine_class_instance) *
+ len);
+ if (XE_IOCTL_DBG(xe, err))
+ return -EFAULT;
+
+ if (XE_IOCTL_DBG(xe, eci[0].gt_id >= xe->info.gt_count))
+ return -EINVAL;
+
+ if (eci[0].engine_class == DRM_XE_ENGINE_CLASS_VM_BIND) {
+ for_each_gt(gt, xe, id) {
+ struct xe_exec_queue *new;
+
+ if (xe_gt_is_media_type(gt))
+ continue;
+
+ eci[0].gt_id = gt->info.id;
+ logical_mask = bind_exec_queue_logical_mask(xe, gt, eci,
+ args->width,
+ args->num_placements);
+ if (XE_IOCTL_DBG(xe, !logical_mask))
+ return -EINVAL;
+
+ hwe = find_hw_engine(xe, eci[0]);
+ if (XE_IOCTL_DBG(xe, !hwe))
+ return -EINVAL;
+
+ /* The migration vm doesn't hold rpm ref */
+ xe_device_mem_access_get(xe);
+
+ migrate_vm = xe_migrate_get_vm(gt_to_tile(gt)->migrate);
+ new = xe_exec_queue_create(xe, migrate_vm, logical_mask,
+ args->width, hwe,
+ EXEC_QUEUE_FLAG_PERSISTENT |
+ EXEC_QUEUE_FLAG_VM |
+ (id ?
+ EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD :
+ 0));
+
+ xe_device_mem_access_put(xe); /* now held by engine */
+
+ xe_vm_put(migrate_vm);
+ if (IS_ERR(new)) {
+ err = PTR_ERR(new);
+ if (q)
+ goto put_exec_queue;
+ return err;
+ }
+ if (id == 0)
+ q = new;
+ else
+ list_add_tail(&new->multi_gt_list,
+ &q->multi_gt_link);
+ }
+ } else {
+ gt = xe_device_get_gt(xe, eci[0].gt_id);
+ logical_mask = calc_validate_logical_mask(xe, gt, eci,
+ args->width,
+ args->num_placements);
+ if (XE_IOCTL_DBG(xe, !logical_mask))
+ return -EINVAL;
+
+ hwe = find_hw_engine(xe, eci[0]);
+ if (XE_IOCTL_DBG(xe, !hwe))
+ return -EINVAL;
+
+ vm = xe_vm_lookup(xef, args->vm_id);
+ if (XE_IOCTL_DBG(xe, !vm))
+ return -ENOENT;
+
+ err = down_read_interruptible(&vm->lock);
+ if (err) {
+ xe_vm_put(vm);
+ return err;
+ }
+
+ if (XE_IOCTL_DBG(xe, xe_vm_is_closed_or_banned(vm))) {
+ up_read(&vm->lock);
+ xe_vm_put(vm);
+ return -ENOENT;
+ }
+
+ q = xe_exec_queue_create(xe, vm, logical_mask,
+ args->width, hwe,
+ xe_vm_in_lr_mode(vm) ? 0 :
+ EXEC_QUEUE_FLAG_PERSISTENT);
+ up_read(&vm->lock);
+ xe_vm_put(vm);
+ if (IS_ERR(q))
+ return PTR_ERR(q);
+
+ if (xe_vm_in_preempt_fence_mode(vm)) {
+ q->compute.context = dma_fence_context_alloc(1);
+ spin_lock_init(&q->compute.lock);
+
+ err = xe_vm_add_compute_exec_queue(vm, q);
+ if (XE_IOCTL_DBG(xe, err))
+ goto put_exec_queue;
+ }
+ }
+
+ if (args->extensions) {
+ err = exec_queue_user_extensions(xe, q, args->extensions, 0, true);
+ if (XE_IOCTL_DBG(xe, err))
+ goto kill_exec_queue;
+ }
+
+ q->persistent.xef = xef;
+
+ mutex_lock(&xef->exec_queue.lock);
+ err = xa_alloc(&xef->exec_queue.xa, &id, q, xa_limit_32b, GFP_KERNEL);
+ mutex_unlock(&xef->exec_queue.lock);
+ if (err)
+ goto kill_exec_queue;
+
+ args->exec_queue_id = id;
+
+ return 0;
+
+kill_exec_queue:
+ xe_exec_queue_kill(q);
+put_exec_queue:
+ xe_exec_queue_put(q);
+ return err;
+}
+
+int xe_exec_queue_get_property_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_exec_queue_get_property *args = data;
+ struct xe_exec_queue *q;
+ int ret;
+
+ if (XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ q = xe_exec_queue_lookup(xef, args->exec_queue_id);
+ if (XE_IOCTL_DBG(xe, !q))
+ return -ENOENT;
+
+ switch (args->property) {
+ case DRM_XE_EXEC_QUEUE_GET_PROPERTY_BAN:
+ args->value = !!(q->flags & EXEC_QUEUE_FLAG_BANNED);
+ ret = 0;
+ break;
+ default:
+ ret = -EINVAL;
+ }
+
+ xe_exec_queue_put(q);
+
+ return ret;
+}
+
+/**
+ * xe_exec_queue_is_lr() - Whether an exec_queue is long-running
+ * @q: The exec_queue
+ *
+ * Return: True if the exec_queue is long-running, false otherwise.
+ */
+bool xe_exec_queue_is_lr(struct xe_exec_queue *q)
+{
+ return q->vm && xe_vm_in_lr_mode(q->vm) &&
+ !(q->flags & EXEC_QUEUE_FLAG_VM);
+}
+
+static s32 xe_exec_queue_num_job_inflight(struct xe_exec_queue *q)
+{
+ return q->lrc->fence_ctx.next_seqno - xe_lrc_seqno(q->lrc) - 1;
+}
+
+/**
+ * xe_exec_queue_ring_full() - Whether an exec_queue's ring is full
+ * @q: The exec_queue
+ *
+ * Return: True if the exec_queue's ring is full, false otherwise.
+ */
+bool xe_exec_queue_ring_full(struct xe_exec_queue *q)
+{
+ struct xe_lrc *lrc = q->lrc;
+ s32 max_job = lrc->ring.size / MAX_JOB_SIZE_BYTES;
+
+ return xe_exec_queue_num_job_inflight(q) >= max_job;
+}
+
+/**
+ * xe_exec_queue_is_idle() - Whether an exec_queue is idle.
+ * @q: The exec_queue
+ *
+ * FIXME: Need to determine what to use as the short-lived
+ * timeline lock for the exec_queues, so that the return value
+ * of this function becomes more than just an advisory
+ * snapshot in time. The timeline lock must protect the
+ * seqno from racing submissions on the same exec_queue.
+ * Typically vm->resv, but user-created timeline locks use the migrate vm
+ * and never grabs the migrate vm->resv so we have a race there.
+ *
+ * Return: True if the exec_queue is idle, false otherwise.
+ */
+bool xe_exec_queue_is_idle(struct xe_exec_queue *q)
+{
+ if (xe_exec_queue_is_parallel(q)) {
+ int i;
+
+ for (i = 0; i < q->width; ++i) {
+ if (xe_lrc_seqno(&q->lrc[i]) !=
+ q->lrc[i].fence_ctx.next_seqno - 1)
+ return false;
+ }
+
+ return true;
+ }
+
+ return xe_lrc_seqno(&q->lrc[0]) ==
+ q->lrc[0].fence_ctx.next_seqno - 1;
+}
+
+void xe_exec_queue_kill(struct xe_exec_queue *q)
+{
+ struct xe_exec_queue *eq = q, *next;
+
+ list_for_each_entry_safe(eq, next, &eq->multi_gt_list,
+ multi_gt_link) {
+ q->ops->kill(eq);
+ xe_vm_remove_compute_exec_queue(q->vm, eq);
+ }
+
+ q->ops->kill(q);
+ xe_vm_remove_compute_exec_queue(q->vm, q);
+}
+
+int xe_exec_queue_destroy_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_exec_queue_destroy *args = data;
+ struct xe_exec_queue *q;
+
+ if (XE_IOCTL_DBG(xe, args->pad) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ mutex_lock(&xef->exec_queue.lock);
+ q = xa_erase(&xef->exec_queue.xa, args->exec_queue_id);
+ mutex_unlock(&xef->exec_queue.lock);
+ if (XE_IOCTL_DBG(xe, !q))
+ return -ENOENT;
+
+ if (!(q->flags & EXEC_QUEUE_FLAG_PERSISTENT))
+ xe_exec_queue_kill(q);
+ else
+ xe_device_add_persistent_exec_queues(xe, q);
+
+ trace_xe_exec_queue_close(q);
+ xe_exec_queue_put(q);
+
+ return 0;
+}
+
+static void xe_exec_queue_last_fence_lockdep_assert(struct xe_exec_queue *q,
+ struct xe_vm *vm)
+{
+ if (q->flags & EXEC_QUEUE_FLAG_VM)
+ lockdep_assert_held(&vm->lock);
+ else
+ xe_vm_assert_held(vm);
+}
+
+/**
+ * xe_exec_queue_last_fence_put() - Drop ref to last fence
+ * @q: The exec queue
+ * @vm: The VM the engine does a bind or exec for
+ */
+void xe_exec_queue_last_fence_put(struct xe_exec_queue *q, struct xe_vm *vm)
+{
+ xe_exec_queue_last_fence_lockdep_assert(q, vm);
+
+ if (q->last_fence) {
+ dma_fence_put(q->last_fence);
+ q->last_fence = NULL;
+ }
+}
+
+/**
+ * xe_exec_queue_last_fence_put_unlocked() - Drop ref to last fence unlocked
+ * @q: The exec queue
+ *
+ * Only safe to be called from xe_exec_queue_destroy().
+ */
+void xe_exec_queue_last_fence_put_unlocked(struct xe_exec_queue *q)
+{
+ if (q->last_fence) {
+ dma_fence_put(q->last_fence);
+ q->last_fence = NULL;
+ }
+}
+
+/**
+ * xe_exec_queue_last_fence_get() - Get last fence
+ * @q: The exec queue
+ * @vm: The VM the engine does a bind or exec for
+ *
+ * Get last fence, does not take a ref
+ *
+ * Returns: last fence if not signaled, dma fence stub if signaled
+ */
+struct dma_fence *xe_exec_queue_last_fence_get(struct xe_exec_queue *q,
+ struct xe_vm *vm)
+{
+ xe_exec_queue_last_fence_lockdep_assert(q, vm);
+
+ if (q->last_fence &&
+ test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &q->last_fence->flags))
+ xe_exec_queue_last_fence_put(q, vm);
+
+ return q->last_fence ? q->last_fence : dma_fence_get_stub();
+}
+
+/**
+ * xe_exec_queue_last_fence_set() - Set last fence
+ * @q: The exec queue
+ * @vm: The VM the engine does a bind or exec for
+ * @fence: The fence
+ *
+ * Set the last fence for the engine. Increases reference count for fence, when
+ * closing engine xe_exec_queue_last_fence_put should be called.
+ */
+void xe_exec_queue_last_fence_set(struct xe_exec_queue *q, struct xe_vm *vm,
+ struct dma_fence *fence)
+{
+ xe_exec_queue_last_fence_lockdep_assert(q, vm);
+
+ xe_exec_queue_last_fence_put(q, vm);
+ q->last_fence = dma_fence_get(fence);
+}
diff --git a/drivers/gpu/drm/xe/xe_exec_queue.h b/drivers/gpu/drm/xe/xe_exec_queue.h
new file mode 100644
index 000000000000..d959cc4a1a82
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_exec_queue.h
@@ -0,0 +1,69 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_EXEC_QUEUE_H_
+#define _XE_EXEC_QUEUE_H_
+
+#include "xe_exec_queue_types.h"
+#include "xe_vm_types.h"
+
+struct drm_device;
+struct drm_file;
+struct xe_device;
+struct xe_file;
+
+struct xe_exec_queue *xe_exec_queue_create(struct xe_device *xe, struct xe_vm *vm,
+ u32 logical_mask, u16 width,
+ struct xe_hw_engine *hw_engine, u32 flags);
+struct xe_exec_queue *xe_exec_queue_create_class(struct xe_device *xe, struct xe_gt *gt,
+ struct xe_vm *vm,
+ enum xe_engine_class class, u32 flags);
+
+void xe_exec_queue_fini(struct xe_exec_queue *q);
+void xe_exec_queue_destroy(struct kref *ref);
+void xe_exec_queue_assign_name(struct xe_exec_queue *q, u32 instance);
+
+struct xe_exec_queue *xe_exec_queue_lookup(struct xe_file *xef, u32 id);
+
+static inline struct xe_exec_queue *xe_exec_queue_get(struct xe_exec_queue *q)
+{
+ kref_get(&q->refcount);
+ return q;
+}
+
+static inline void xe_exec_queue_put(struct xe_exec_queue *q)
+{
+ kref_put(&q->refcount, xe_exec_queue_destroy);
+}
+
+static inline bool xe_exec_queue_is_parallel(struct xe_exec_queue *q)
+{
+ return q->width > 1;
+}
+
+bool xe_exec_queue_is_lr(struct xe_exec_queue *q);
+
+bool xe_exec_queue_ring_full(struct xe_exec_queue *q);
+
+bool xe_exec_queue_is_idle(struct xe_exec_queue *q);
+
+void xe_exec_queue_kill(struct xe_exec_queue *q);
+
+int xe_exec_queue_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_exec_queue_destroy_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_exec_queue_get_property_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+enum xe_exec_queue_priority xe_exec_queue_device_get_max_priority(struct xe_device *xe);
+
+void xe_exec_queue_last_fence_put(struct xe_exec_queue *e, struct xe_vm *vm);
+void xe_exec_queue_last_fence_put_unlocked(struct xe_exec_queue *e);
+struct dma_fence *xe_exec_queue_last_fence_get(struct xe_exec_queue *e,
+ struct xe_vm *vm);
+void xe_exec_queue_last_fence_set(struct xe_exec_queue *e, struct xe_vm *vm,
+ struct dma_fence *fence);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_exec_queue_types.h b/drivers/gpu/drm/xe/xe_exec_queue_types.h
new file mode 100644
index 000000000000..3d7e704ec3d9
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_exec_queue_types.h
@@ -0,0 +1,222 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_EXEC_QUEUE_TYPES_H_
+#define _XE_EXEC_QUEUE_TYPES_H_
+
+#include <linux/kref.h>
+
+#include <drm/gpu_scheduler.h>
+
+#include "xe_gpu_scheduler_types.h"
+#include "xe_hw_engine_types.h"
+#include "xe_hw_fence_types.h"
+#include "xe_lrc_types.h"
+
+struct xe_execlist_exec_queue;
+struct xe_gt;
+struct xe_guc_exec_queue;
+struct xe_hw_engine;
+struct xe_vm;
+
+enum xe_exec_queue_priority {
+ XE_EXEC_QUEUE_PRIORITY_UNSET = -2, /* For execlist usage only */
+ XE_EXEC_QUEUE_PRIORITY_LOW = 0,
+ XE_EXEC_QUEUE_PRIORITY_NORMAL,
+ XE_EXEC_QUEUE_PRIORITY_HIGH,
+ XE_EXEC_QUEUE_PRIORITY_KERNEL,
+
+ XE_EXEC_QUEUE_PRIORITY_COUNT
+};
+
+/**
+ * struct xe_exec_queue - Execution queue
+ *
+ * Contains all state necessary for submissions. Can either be a user object or
+ * a kernel object.
+ */
+struct xe_exec_queue {
+ /** @gt: graphics tile this exec queue can submit to */
+ struct xe_gt *gt;
+ /**
+ * @hwe: A hardware of the same class. May (physical engine) or may not
+ * (virtual engine) be where jobs actual engine up running. Should never
+ * really be used for submissions.
+ */
+ struct xe_hw_engine *hwe;
+ /** @refcount: ref count of this exec queue */
+ struct kref refcount;
+ /** @vm: VM (address space) for this exec queue */
+ struct xe_vm *vm;
+ /** @class: class of this exec queue */
+ enum xe_engine_class class;
+ /** @priority: priority of this exec queue */
+ enum xe_exec_queue_priority priority;
+ /**
+ * @logical_mask: logical mask of where job submitted to exec queue can run
+ */
+ u32 logical_mask;
+ /** @name: name of this exec queue */
+ char name[MAX_FENCE_NAME_LEN];
+ /** @width: width (number BB submitted per exec) of this exec queue */
+ u16 width;
+ /** @fence_irq: fence IRQ used to signal job completion */
+ struct xe_hw_fence_irq *fence_irq;
+
+ /**
+ * @last_fence: last fence on exec queue, protected by vm->lock in write
+ * mode if bind exec queue, protected by dma resv lock if non-bind exec
+ * queue
+ */
+ struct dma_fence *last_fence;
+
+/* queue no longer allowed to submit */
+#define EXEC_QUEUE_FLAG_BANNED BIT(0)
+/* queue used for kernel submission only */
+#define EXEC_QUEUE_FLAG_KERNEL BIT(1)
+/* kernel engine only destroyed at driver unload */
+#define EXEC_QUEUE_FLAG_PERMANENT BIT(2)
+/* queue keeps running pending jobs after destroy ioctl */
+#define EXEC_QUEUE_FLAG_PERSISTENT BIT(3)
+/* for VM jobs. Caller needs to hold rpm ref when creating queue with this flag */
+#define EXEC_QUEUE_FLAG_VM BIT(4)
+/* child of VM queue for multi-tile VM jobs */
+#define EXEC_QUEUE_FLAG_BIND_ENGINE_CHILD BIT(5)
+
+ /**
+ * @flags: flags for this exec queue, should statically setup aside from ban
+ * bit
+ */
+ unsigned long flags;
+
+ union {
+ /** @multi_gt_list: list head for VM bind engines if multi-GT */
+ struct list_head multi_gt_list;
+ /** @multi_gt_link: link for VM bind engines if multi-GT */
+ struct list_head multi_gt_link;
+ };
+
+ union {
+ /** @execlist: execlist backend specific state for exec queue */
+ struct xe_execlist_exec_queue *execlist;
+ /** @guc: GuC backend specific state for exec queue */
+ struct xe_guc_exec_queue *guc;
+ };
+
+ /**
+ * @persistent: persistent exec queue state
+ */
+ struct {
+ /** @xef: file which this exec queue belongs to */
+ struct xe_file *xef;
+ /** @link: link in list of persistent exec queues */
+ struct list_head link;
+ } persistent;
+
+ union {
+ /**
+ * @parallel: parallel submission state
+ */
+ struct {
+ /** @composite_fence_ctx: context composite fence */
+ u64 composite_fence_ctx;
+ /** @composite_fence_seqno: seqno for composite fence */
+ u32 composite_fence_seqno;
+ } parallel;
+ /**
+ * @bind: bind submission state
+ */
+ struct {
+ /** @fence_ctx: context bind fence */
+ u64 fence_ctx;
+ /** @fence_seqno: seqno for bind fence */
+ u32 fence_seqno;
+ } bind;
+ };
+
+ /** @sched_props: scheduling properties */
+ struct {
+ /** @timeslice_us: timeslice period in micro-seconds */
+ u32 timeslice_us;
+ /** @preempt_timeout_us: preemption timeout in micro-seconds */
+ u32 preempt_timeout_us;
+ } sched_props;
+
+ /** @compute: compute exec queue state */
+ struct {
+ /** @pfence: preemption fence */
+ struct dma_fence *pfence;
+ /** @context: preemption fence context */
+ u64 context;
+ /** @seqno: preemption fence seqno */
+ u32 seqno;
+ /** @link: link into VM's list of exec queues */
+ struct list_head link;
+ /** @lock: preemption fences lock */
+ spinlock_t lock;
+ } compute;
+
+ /** @usm: unified shared memory state */
+ struct {
+ /** @acc_trigger: access counter trigger */
+ u32 acc_trigger;
+ /** @acc_notify: access counter notify */
+ u32 acc_notify;
+ /** @acc_granularity: access counter granularity */
+ u32 acc_granularity;
+ } usm;
+
+ /** @ops: submission backend exec queue operations */
+ const struct xe_exec_queue_ops *ops;
+
+ /** @ring_ops: ring operations for this exec queue */
+ const struct xe_ring_ops *ring_ops;
+ /** @entity: DRM sched entity for this exec queue (1 to 1 relationship) */
+ struct drm_sched_entity *entity;
+ /** @lrc: logical ring context for this exec queue */
+ struct xe_lrc lrc[];
+};
+
+/**
+ * struct xe_exec_queue_ops - Submission backend exec queue operations
+ */
+struct xe_exec_queue_ops {
+ /** @init: Initialize exec queue for submission backend */
+ int (*init)(struct xe_exec_queue *q);
+ /** @kill: Kill inflight submissions for backend */
+ void (*kill)(struct xe_exec_queue *q);
+ /** @fini: Fini exec queue for submission backend */
+ void (*fini)(struct xe_exec_queue *q);
+ /** @set_priority: Set priority for exec queue */
+ int (*set_priority)(struct xe_exec_queue *q,
+ enum xe_exec_queue_priority priority);
+ /** @set_timeslice: Set timeslice for exec queue */
+ int (*set_timeslice)(struct xe_exec_queue *q, u32 timeslice_us);
+ /** @set_preempt_timeout: Set preemption timeout for exec queue */
+ int (*set_preempt_timeout)(struct xe_exec_queue *q, u32 preempt_timeout_us);
+ /** @set_job_timeout: Set job timeout for exec queue */
+ int (*set_job_timeout)(struct xe_exec_queue *q, u32 job_timeout_ms);
+ /**
+ * @suspend: Suspend exec queue from executing, allowed to be called
+ * multiple times in a row before resume with the caveat that
+ * suspend_wait returns before calling suspend again.
+ */
+ int (*suspend)(struct xe_exec_queue *q);
+ /**
+ * @suspend_wait: Wait for an exec queue to suspend executing, should be
+ * call after suspend.
+ */
+ void (*suspend_wait)(struct xe_exec_queue *q);
+ /**
+ * @resume: Resume exec queue execution, exec queue must be in a suspended
+ * state and dma fence returned from most recent suspend call must be
+ * signalled when this function is called.
+ */
+ void (*resume)(struct xe_exec_queue *q);
+ /** @reset_status: check exec queue reset status */
+ bool (*reset_status)(struct xe_exec_queue *q);
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_execlist.c b/drivers/gpu/drm/xe/xe_execlist.c
new file mode 100644
index 000000000000..96b5224eb478
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_execlist.c
@@ -0,0 +1,474 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_execlist.h"
+
+#include <drm/drm_managed.h>
+
+#include "instructions/xe_mi_commands.h"
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gpu_commands.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_lrc_layout.h"
+#include "xe_assert.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_gt.h"
+#include "xe_hw_fence.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+#include "xe_mocs.h"
+#include "xe_ring_ops_types.h"
+#include "xe_sched_job.h"
+
+#define XE_EXECLIST_HANG_LIMIT 1
+
+#define SW_CTX_ID_SHIFT 37
+#define SW_CTX_ID_WIDTH 11
+#define XEHP_SW_CTX_ID_SHIFT 39
+#define XEHP_SW_CTX_ID_WIDTH 16
+
+#define SW_CTX_ID \
+ GENMASK_ULL(SW_CTX_ID_WIDTH + SW_CTX_ID_SHIFT - 1, \
+ SW_CTX_ID_SHIFT)
+
+#define XEHP_SW_CTX_ID \
+ GENMASK_ULL(XEHP_SW_CTX_ID_WIDTH + XEHP_SW_CTX_ID_SHIFT - 1, \
+ XEHP_SW_CTX_ID_SHIFT)
+
+
+static void __start_lrc(struct xe_hw_engine *hwe, struct xe_lrc *lrc,
+ u32 ctx_id)
+{
+ struct xe_gt *gt = hwe->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ u64 lrc_desc;
+
+ lrc_desc = xe_lrc_descriptor(lrc);
+
+ if (GRAPHICS_VERx100(xe) >= 1250) {
+ xe_gt_assert(hwe->gt, FIELD_FIT(XEHP_SW_CTX_ID, ctx_id));
+ lrc_desc |= FIELD_PREP(XEHP_SW_CTX_ID, ctx_id);
+ } else {
+ xe_gt_assert(hwe->gt, FIELD_FIT(SW_CTX_ID, ctx_id));
+ lrc_desc |= FIELD_PREP(SW_CTX_ID, ctx_id);
+ }
+
+ if (hwe->class == XE_ENGINE_CLASS_COMPUTE)
+ xe_mmio_write32(hwe->gt, RCU_MODE,
+ _MASKED_BIT_ENABLE(RCU_MODE_CCS_ENABLE));
+
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_TAIL, lrc->ring.tail);
+ lrc->ring.old_tail = lrc->ring.tail;
+
+ /*
+ * Make sure the context image is complete before we submit it to HW.
+ *
+ * Ostensibly, writes (including the WCB) should be flushed prior to
+ * an uncached write such as our mmio register access, the empirical
+ * evidence (esp. on Braswell) suggests that the WC write into memory
+ * may not be visible to the HW prior to the completion of the UC
+ * register write and that we may begin execution from the context
+ * before its image is complete leading to invalid PD chasing.
+ */
+ wmb();
+
+ xe_mmio_write32(gt, RING_HWS_PGA(hwe->mmio_base),
+ xe_bo_ggtt_addr(hwe->hwsp));
+ xe_mmio_read32(gt, RING_HWS_PGA(hwe->mmio_base));
+ xe_mmio_write32(gt, RING_MODE(hwe->mmio_base),
+ _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE));
+
+ xe_mmio_write32(gt, RING_EXECLIST_SQ_CONTENTS_LO(hwe->mmio_base),
+ lower_32_bits(lrc_desc));
+ xe_mmio_write32(gt, RING_EXECLIST_SQ_CONTENTS_HI(hwe->mmio_base),
+ upper_32_bits(lrc_desc));
+ xe_mmio_write32(gt, RING_EXECLIST_CONTROL(hwe->mmio_base),
+ EL_CTRL_LOAD);
+}
+
+static void __xe_execlist_port_start(struct xe_execlist_port *port,
+ struct xe_execlist_exec_queue *exl)
+{
+ struct xe_device *xe = gt_to_xe(port->hwe->gt);
+ int max_ctx = FIELD_MAX(SW_CTX_ID);
+
+ if (GRAPHICS_VERx100(xe) >= 1250)
+ max_ctx = FIELD_MAX(XEHP_SW_CTX_ID);
+
+ xe_execlist_port_assert_held(port);
+
+ if (port->running_exl != exl || !exl->has_run) {
+ port->last_ctx_id++;
+
+ /* 0 is reserved for the kernel context */
+ if (port->last_ctx_id > max_ctx)
+ port->last_ctx_id = 1;
+ }
+
+ __start_lrc(port->hwe, exl->q->lrc, port->last_ctx_id);
+ port->running_exl = exl;
+ exl->has_run = true;
+}
+
+static void __xe_execlist_port_idle(struct xe_execlist_port *port)
+{
+ u32 noop[2] = { MI_NOOP, MI_NOOP };
+
+ xe_execlist_port_assert_held(port);
+
+ if (!port->running_exl)
+ return;
+
+ xe_lrc_write_ring(&port->hwe->kernel_lrc, noop, sizeof(noop));
+ __start_lrc(port->hwe, &port->hwe->kernel_lrc, 0);
+ port->running_exl = NULL;
+}
+
+static bool xe_execlist_is_idle(struct xe_execlist_exec_queue *exl)
+{
+ struct xe_lrc *lrc = exl->q->lrc;
+
+ return lrc->ring.tail == lrc->ring.old_tail;
+}
+
+static void __xe_execlist_port_start_next_active(struct xe_execlist_port *port)
+{
+ struct xe_execlist_exec_queue *exl = NULL;
+ int i;
+
+ xe_execlist_port_assert_held(port);
+
+ for (i = ARRAY_SIZE(port->active) - 1; i >= 0; i--) {
+ while (!list_empty(&port->active[i])) {
+ exl = list_first_entry(&port->active[i],
+ struct xe_execlist_exec_queue,
+ active_link);
+ list_del(&exl->active_link);
+
+ if (xe_execlist_is_idle(exl)) {
+ exl->active_priority = XE_EXEC_QUEUE_PRIORITY_UNSET;
+ continue;
+ }
+
+ list_add_tail(&exl->active_link, &port->active[i]);
+ __xe_execlist_port_start(port, exl);
+ return;
+ }
+ }
+
+ __xe_execlist_port_idle(port);
+}
+
+static u64 read_execlist_status(struct xe_hw_engine *hwe)
+{
+ struct xe_gt *gt = hwe->gt;
+ u32 hi, lo;
+
+ lo = xe_mmio_read32(gt, RING_EXECLIST_STATUS_LO(hwe->mmio_base));
+ hi = xe_mmio_read32(gt, RING_EXECLIST_STATUS_HI(hwe->mmio_base));
+
+ return lo | (u64)hi << 32;
+}
+
+static void xe_execlist_port_irq_handler_locked(struct xe_execlist_port *port)
+{
+ u64 status;
+
+ xe_execlist_port_assert_held(port);
+
+ status = read_execlist_status(port->hwe);
+ if (status & BIT(7))
+ return;
+
+ __xe_execlist_port_start_next_active(port);
+}
+
+static void xe_execlist_port_irq_handler(struct xe_hw_engine *hwe,
+ u16 intr_vec)
+{
+ struct xe_execlist_port *port = hwe->exl_port;
+
+ spin_lock(&port->lock);
+ xe_execlist_port_irq_handler_locked(port);
+ spin_unlock(&port->lock);
+}
+
+static void xe_execlist_port_wake_locked(struct xe_execlist_port *port,
+ enum xe_exec_queue_priority priority)
+{
+ xe_execlist_port_assert_held(port);
+
+ if (port->running_exl && port->running_exl->active_priority >= priority)
+ return;
+
+ __xe_execlist_port_start_next_active(port);
+}
+
+static void xe_execlist_make_active(struct xe_execlist_exec_queue *exl)
+{
+ struct xe_execlist_port *port = exl->port;
+ enum xe_exec_queue_priority priority = exl->active_priority;
+
+ XE_WARN_ON(priority == XE_EXEC_QUEUE_PRIORITY_UNSET);
+ XE_WARN_ON(priority < 0);
+ XE_WARN_ON(priority >= ARRAY_SIZE(exl->port->active));
+
+ spin_lock_irq(&port->lock);
+
+ if (exl->active_priority != priority &&
+ exl->active_priority != XE_EXEC_QUEUE_PRIORITY_UNSET) {
+ /* Priority changed, move it to the right list */
+ list_del(&exl->active_link);
+ exl->active_priority = XE_EXEC_QUEUE_PRIORITY_UNSET;
+ }
+
+ if (exl->active_priority == XE_EXEC_QUEUE_PRIORITY_UNSET) {
+ exl->active_priority = priority;
+ list_add_tail(&exl->active_link, &port->active[priority]);
+ }
+
+ xe_execlist_port_wake_locked(exl->port, priority);
+
+ spin_unlock_irq(&port->lock);
+}
+
+static void xe_execlist_port_irq_fail_timer(struct timer_list *timer)
+{
+ struct xe_execlist_port *port =
+ container_of(timer, struct xe_execlist_port, irq_fail);
+
+ spin_lock_irq(&port->lock);
+ xe_execlist_port_irq_handler_locked(port);
+ spin_unlock_irq(&port->lock);
+
+ port->irq_fail.expires = jiffies + msecs_to_jiffies(1000);
+ add_timer(&port->irq_fail);
+}
+
+struct xe_execlist_port *xe_execlist_port_create(struct xe_device *xe,
+ struct xe_hw_engine *hwe)
+{
+ struct drm_device *drm = &xe->drm;
+ struct xe_execlist_port *port;
+ int i;
+
+ port = drmm_kzalloc(drm, sizeof(*port), GFP_KERNEL);
+ if (!port)
+ return ERR_PTR(-ENOMEM);
+
+ port->hwe = hwe;
+
+ spin_lock_init(&port->lock);
+ for (i = 0; i < ARRAY_SIZE(port->active); i++)
+ INIT_LIST_HEAD(&port->active[i]);
+
+ port->last_ctx_id = 1;
+ port->running_exl = NULL;
+
+ hwe->irq_handler = xe_execlist_port_irq_handler;
+
+ /* TODO: Fix the interrupt code so it doesn't race like mad */
+ timer_setup(&port->irq_fail, xe_execlist_port_irq_fail_timer, 0);
+ port->irq_fail.expires = jiffies + msecs_to_jiffies(1000);
+ add_timer(&port->irq_fail);
+
+ return port;
+}
+
+void xe_execlist_port_destroy(struct xe_execlist_port *port)
+{
+ del_timer(&port->irq_fail);
+
+ /* Prevent an interrupt while we're destroying */
+ spin_lock_irq(&gt_to_xe(port->hwe->gt)->irq.lock);
+ port->hwe->irq_handler = NULL;
+ spin_unlock_irq(&gt_to_xe(port->hwe->gt)->irq.lock);
+}
+
+static struct dma_fence *
+execlist_run_job(struct drm_sched_job *drm_job)
+{
+ struct xe_sched_job *job = to_xe_sched_job(drm_job);
+ struct xe_exec_queue *q = job->q;
+ struct xe_execlist_exec_queue *exl = job->q->execlist;
+
+ q->ring_ops->emit_job(job);
+ xe_execlist_make_active(exl);
+
+ return dma_fence_get(job->fence);
+}
+
+static void execlist_job_free(struct drm_sched_job *drm_job)
+{
+ struct xe_sched_job *job = to_xe_sched_job(drm_job);
+
+ xe_sched_job_put(job);
+}
+
+static const struct drm_sched_backend_ops drm_sched_ops = {
+ .run_job = execlist_run_job,
+ .free_job = execlist_job_free,
+};
+
+static int execlist_exec_queue_init(struct xe_exec_queue *q)
+{
+ struct drm_gpu_scheduler *sched;
+ struct xe_execlist_exec_queue *exl;
+ struct xe_device *xe = gt_to_xe(q->gt);
+ int err;
+
+ xe_assert(xe, !xe_device_uc_enabled(xe));
+
+ drm_info(&xe->drm, "Enabling execlist submission (GuC submission disabled)\n");
+
+ exl = kzalloc(sizeof(*exl), GFP_KERNEL);
+ if (!exl)
+ return -ENOMEM;
+
+ exl->q = q;
+
+ err = drm_sched_init(&exl->sched, &drm_sched_ops, NULL, 1,
+ q->lrc[0].ring.size / MAX_JOB_SIZE_BYTES,
+ XE_SCHED_HANG_LIMIT, XE_SCHED_JOB_TIMEOUT,
+ NULL, NULL, q->hwe->name,
+ gt_to_xe(q->gt)->drm.dev);
+ if (err)
+ goto err_free;
+
+ sched = &exl->sched;
+ err = drm_sched_entity_init(&exl->entity, 0, &sched, 1, NULL);
+ if (err)
+ goto err_sched;
+
+ exl->port = q->hwe->exl_port;
+ exl->has_run = false;
+ exl->active_priority = XE_EXEC_QUEUE_PRIORITY_UNSET;
+ q->execlist = exl;
+ q->entity = &exl->entity;
+
+ xe_exec_queue_assign_name(q, ffs(q->logical_mask) - 1);
+
+ return 0;
+
+err_sched:
+ drm_sched_fini(&exl->sched);
+err_free:
+ kfree(exl);
+ return err;
+}
+
+static void execlist_exec_queue_fini_async(struct work_struct *w)
+{
+ struct xe_execlist_exec_queue *ee =
+ container_of(w, struct xe_execlist_exec_queue, fini_async);
+ struct xe_exec_queue *q = ee->q;
+ struct xe_execlist_exec_queue *exl = q->execlist;
+ struct xe_device *xe = gt_to_xe(q->gt);
+ unsigned long flags;
+
+ xe_assert(xe, !xe_device_uc_enabled(xe));
+
+ spin_lock_irqsave(&exl->port->lock, flags);
+ if (WARN_ON(exl->active_priority != XE_EXEC_QUEUE_PRIORITY_UNSET))
+ list_del(&exl->active_link);
+ spin_unlock_irqrestore(&exl->port->lock, flags);
+
+ if (q->flags & EXEC_QUEUE_FLAG_PERSISTENT)
+ xe_device_remove_persistent_exec_queues(xe, q);
+ drm_sched_entity_fini(&exl->entity);
+ drm_sched_fini(&exl->sched);
+ kfree(exl);
+
+ xe_exec_queue_fini(q);
+}
+
+static void execlist_exec_queue_kill(struct xe_exec_queue *q)
+{
+ /* NIY */
+}
+
+static void execlist_exec_queue_fini(struct xe_exec_queue *q)
+{
+ INIT_WORK(&q->execlist->fini_async, execlist_exec_queue_fini_async);
+ queue_work(system_unbound_wq, &q->execlist->fini_async);
+}
+
+static int execlist_exec_queue_set_priority(struct xe_exec_queue *q,
+ enum xe_exec_queue_priority priority)
+{
+ /* NIY */
+ return 0;
+}
+
+static int execlist_exec_queue_set_timeslice(struct xe_exec_queue *q, u32 timeslice_us)
+{
+ /* NIY */
+ return 0;
+}
+
+static int execlist_exec_queue_set_preempt_timeout(struct xe_exec_queue *q,
+ u32 preempt_timeout_us)
+{
+ /* NIY */
+ return 0;
+}
+
+static int execlist_exec_queue_set_job_timeout(struct xe_exec_queue *q,
+ u32 job_timeout_ms)
+{
+ /* NIY */
+ return 0;
+}
+
+static int execlist_exec_queue_suspend(struct xe_exec_queue *q)
+{
+ /* NIY */
+ return 0;
+}
+
+static void execlist_exec_queue_suspend_wait(struct xe_exec_queue *q)
+
+{
+ /* NIY */
+}
+
+static void execlist_exec_queue_resume(struct xe_exec_queue *q)
+{
+ /* NIY */
+}
+
+static bool execlist_exec_queue_reset_status(struct xe_exec_queue *q)
+{
+ /* NIY */
+ return false;
+}
+
+static const struct xe_exec_queue_ops execlist_exec_queue_ops = {
+ .init = execlist_exec_queue_init,
+ .kill = execlist_exec_queue_kill,
+ .fini = execlist_exec_queue_fini,
+ .set_priority = execlist_exec_queue_set_priority,
+ .set_timeslice = execlist_exec_queue_set_timeslice,
+ .set_preempt_timeout = execlist_exec_queue_set_preempt_timeout,
+ .set_job_timeout = execlist_exec_queue_set_job_timeout,
+ .suspend = execlist_exec_queue_suspend,
+ .suspend_wait = execlist_exec_queue_suspend_wait,
+ .resume = execlist_exec_queue_resume,
+ .reset_status = execlist_exec_queue_reset_status,
+};
+
+int xe_execlist_init(struct xe_gt *gt)
+{
+ /* GuC submission enabled, nothing to do */
+ if (xe_device_uc_enabled(gt_to_xe(gt)))
+ return 0;
+
+ gt->exec_queue_ops = &execlist_exec_queue_ops;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_execlist.h b/drivers/gpu/drm/xe/xe_execlist.h
new file mode 100644
index 000000000000..26f600ac8552
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_execlist.h
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_EXECLIST_H_
+#define _XE_EXECLIST_H_
+
+#include "xe_execlist_types.h"
+
+struct xe_device;
+struct xe_gt;
+
+#define xe_execlist_port_assert_held(port) lockdep_assert_held(&(port)->lock)
+
+int xe_execlist_init(struct xe_gt *gt);
+struct xe_execlist_port *xe_execlist_port_create(struct xe_device *xe,
+ struct xe_hw_engine *hwe);
+void xe_execlist_port_destroy(struct xe_execlist_port *port);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_execlist_types.h b/drivers/gpu/drm/xe/xe_execlist_types.h
new file mode 100644
index 000000000000..f94bbf4c53e4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_execlist_types.h
@@ -0,0 +1,49 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_EXECLIST_TYPES_H_
+#define _XE_EXECLIST_TYPES_H_
+
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+
+#include "xe_exec_queue_types.h"
+
+struct xe_hw_engine;
+struct xe_execlist_exec_queue;
+
+struct xe_execlist_port {
+ struct xe_hw_engine *hwe;
+
+ spinlock_t lock;
+
+ struct list_head active[XE_EXEC_QUEUE_PRIORITY_COUNT];
+
+ u32 last_ctx_id;
+
+ struct xe_execlist_exec_queue *running_exl;
+
+ struct timer_list irq_fail;
+};
+
+struct xe_execlist_exec_queue {
+ struct xe_exec_queue *q;
+
+ struct drm_gpu_scheduler sched;
+
+ struct drm_sched_entity entity;
+
+ struct xe_execlist_port *port;
+
+ bool has_run;
+
+ struct work_struct fini_async;
+
+ enum xe_exec_queue_priority active_priority;
+ struct list_head active_link;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_force_wake.c b/drivers/gpu/drm/xe/xe_force_wake.c
new file mode 100644
index 000000000000..9bbe8a5040da
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_force_wake.c
@@ -0,0 +1,199 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_force_wake.h"
+
+#include <drm/drm_util.h>
+
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_reg_defs.h"
+#include "xe_gt.h"
+#include "xe_mmio.h"
+
+#define XE_FORCE_WAKE_ACK_TIMEOUT_MS 50
+
+static struct xe_gt *
+fw_to_gt(struct xe_force_wake *fw)
+{
+ return fw->gt;
+}
+
+static struct xe_device *
+fw_to_xe(struct xe_force_wake *fw)
+{
+ return gt_to_xe(fw_to_gt(fw));
+}
+
+static void domain_init(struct xe_force_wake_domain *domain,
+ enum xe_force_wake_domain_id id,
+ struct xe_reg reg, struct xe_reg ack, u32 val, u32 mask)
+{
+ domain->id = id;
+ domain->reg_ctl = reg;
+ domain->reg_ack = ack;
+ domain->val = val;
+ domain->mask = mask;
+}
+
+void xe_force_wake_init_gt(struct xe_gt *gt, struct xe_force_wake *fw)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ fw->gt = gt;
+ spin_lock_init(&fw->lock);
+
+ /* Assuming gen11+ so assert this assumption is correct */
+ xe_gt_assert(gt, GRAPHICS_VER(gt_to_xe(gt)) >= 11);
+
+ if (xe->info.graphics_verx100 >= 1270) {
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_GT],
+ XE_FW_DOMAIN_ID_GT,
+ FORCEWAKE_GT,
+ FORCEWAKE_ACK_GT_MTL,
+ BIT(0), BIT(16));
+ } else {
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_GT],
+ XE_FW_DOMAIN_ID_GT,
+ FORCEWAKE_GT,
+ FORCEWAKE_ACK_GT,
+ BIT(0), BIT(16));
+ }
+}
+
+void xe_force_wake_init_engines(struct xe_gt *gt, struct xe_force_wake *fw)
+{
+ int i, j;
+
+ /* Assuming gen11+ so assert this assumption is correct */
+ xe_gt_assert(gt, GRAPHICS_VER(gt_to_xe(gt)) >= 11);
+
+ if (!xe_gt_is_media_type(gt))
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_RENDER],
+ XE_FW_DOMAIN_ID_RENDER,
+ FORCEWAKE_RENDER,
+ FORCEWAKE_ACK_RENDER,
+ BIT(0), BIT(16));
+
+ for (i = XE_HW_ENGINE_VCS0, j = 0; i <= XE_HW_ENGINE_VCS7; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_MEDIA_VDBOX0 + j],
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX0 + j,
+ FORCEWAKE_MEDIA_VDBOX(j),
+ FORCEWAKE_ACK_MEDIA_VDBOX(j),
+ BIT(0), BIT(16));
+ }
+
+ for (i = XE_HW_ENGINE_VECS0, j = 0; i <= XE_HW_ENGINE_VECS3; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_MEDIA_VEBOX0 + j],
+ XE_FW_DOMAIN_ID_MEDIA_VEBOX0 + j,
+ FORCEWAKE_MEDIA_VEBOX(j),
+ FORCEWAKE_ACK_MEDIA_VEBOX(j),
+ BIT(0), BIT(16));
+ }
+
+ if (gt->info.engine_mask & BIT(XE_HW_ENGINE_GSCCS0))
+ domain_init(&fw->domains[XE_FW_DOMAIN_ID_GSC],
+ XE_FW_DOMAIN_ID_GSC,
+ FORCEWAKE_GSC,
+ FORCEWAKE_ACK_GSC,
+ BIT(0), BIT(16));
+}
+
+static void domain_wake(struct xe_gt *gt, struct xe_force_wake_domain *domain)
+{
+ xe_mmio_write32(gt, domain->reg_ctl, domain->mask | domain->val);
+}
+
+static int domain_wake_wait(struct xe_gt *gt,
+ struct xe_force_wake_domain *domain)
+{
+ return xe_mmio_wait32(gt, domain->reg_ack, domain->val, domain->val,
+ XE_FORCE_WAKE_ACK_TIMEOUT_MS * USEC_PER_MSEC,
+ NULL, true);
+}
+
+static void domain_sleep(struct xe_gt *gt, struct xe_force_wake_domain *domain)
+{
+ xe_mmio_write32(gt, domain->reg_ctl, domain->mask);
+}
+
+static int domain_sleep_wait(struct xe_gt *gt,
+ struct xe_force_wake_domain *domain)
+{
+ return xe_mmio_wait32(gt, domain->reg_ack, domain->val, 0,
+ XE_FORCE_WAKE_ACK_TIMEOUT_MS * USEC_PER_MSEC,
+ NULL, true);
+}
+
+#define for_each_fw_domain_masked(domain__, mask__, fw__, tmp__) \
+ for (tmp__ = (mask__); tmp__; tmp__ &= ~BIT(ffs(tmp__) - 1)) \
+ for_each_if((domain__ = ((fw__)->domains + \
+ (ffs(tmp__) - 1))) && \
+ domain__->reg_ctl.addr)
+
+int xe_force_wake_get(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domains)
+{
+ struct xe_device *xe = fw_to_xe(fw);
+ struct xe_gt *gt = fw_to_gt(fw);
+ struct xe_force_wake_domain *domain;
+ enum xe_force_wake_domains tmp, woken = 0;
+ unsigned long flags;
+ int ret, ret2 = 0;
+
+ spin_lock_irqsave(&fw->lock, flags);
+ for_each_fw_domain_masked(domain, domains, fw, tmp) {
+ if (!domain->ref++) {
+ woken |= BIT(domain->id);
+ domain_wake(gt, domain);
+ }
+ }
+ for_each_fw_domain_masked(domain, woken, fw, tmp) {
+ ret = domain_wake_wait(gt, domain);
+ ret2 |= ret;
+ if (ret)
+ drm_notice(&xe->drm, "Force wake domain (%d) failed to ack wake, ret=%d\n",
+ domain->id, ret);
+ }
+ fw->awake_domains |= woken;
+ spin_unlock_irqrestore(&fw->lock, flags);
+
+ return ret2;
+}
+
+int xe_force_wake_put(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domains)
+{
+ struct xe_device *xe = fw_to_xe(fw);
+ struct xe_gt *gt = fw_to_gt(fw);
+ struct xe_force_wake_domain *domain;
+ enum xe_force_wake_domains tmp, sleep = 0;
+ unsigned long flags;
+ int ret, ret2 = 0;
+
+ spin_lock_irqsave(&fw->lock, flags);
+ for_each_fw_domain_masked(domain, domains, fw, tmp) {
+ if (!--domain->ref) {
+ sleep |= BIT(domain->id);
+ domain_sleep(gt, domain);
+ }
+ }
+ for_each_fw_domain_masked(domain, sleep, fw, tmp) {
+ ret = domain_sleep_wait(gt, domain);
+ ret2 |= ret;
+ if (ret)
+ drm_notice(&xe->drm, "Force wake domain (%d) failed to ack sleep, ret=%d\n",
+ domain->id, ret);
+ }
+ fw->awake_domains &= ~sleep;
+ spin_unlock_irqrestore(&fw->lock, flags);
+
+ return ret2;
+}
diff --git a/drivers/gpu/drm/xe/xe_force_wake.h b/drivers/gpu/drm/xe/xe_force_wake.h
new file mode 100644
index 000000000000..83cb157da7cc
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_force_wake.h
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_FORCE_WAKE_H_
+#define _XE_FORCE_WAKE_H_
+
+#include "xe_assert.h"
+#include "xe_force_wake_types.h"
+
+struct xe_gt;
+
+void xe_force_wake_init_gt(struct xe_gt *gt,
+ struct xe_force_wake *fw);
+void xe_force_wake_init_engines(struct xe_gt *gt,
+ struct xe_force_wake *fw);
+int xe_force_wake_get(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domains);
+int xe_force_wake_put(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domains);
+
+static inline int
+xe_force_wake_ref(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domain)
+{
+ xe_gt_assert(fw->gt, domain);
+ return fw->domains[ffs(domain) - 1].ref;
+}
+
+static inline void
+xe_force_wake_assert_held(struct xe_force_wake *fw,
+ enum xe_force_wake_domains domain)
+{
+ xe_gt_assert(fw->gt, fw->awake_domains & domain);
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_force_wake_types.h b/drivers/gpu/drm/xe/xe_force_wake_types.h
new file mode 100644
index 000000000000..ed0edc2cdf9f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_force_wake_types.h
@@ -0,0 +1,86 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_FORCE_WAKE_TYPES_H_
+#define _XE_FORCE_WAKE_TYPES_H_
+
+#include <linux/mutex.h>
+#include <linux/types.h>
+
+#include "regs/xe_reg_defs.h"
+
+enum xe_force_wake_domain_id {
+ XE_FW_DOMAIN_ID_GT = 0,
+ XE_FW_DOMAIN_ID_RENDER,
+ XE_FW_DOMAIN_ID_MEDIA,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX0,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX1,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX2,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX3,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX4,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX5,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX6,
+ XE_FW_DOMAIN_ID_MEDIA_VDBOX7,
+ XE_FW_DOMAIN_ID_MEDIA_VEBOX0,
+ XE_FW_DOMAIN_ID_MEDIA_VEBOX1,
+ XE_FW_DOMAIN_ID_MEDIA_VEBOX2,
+ XE_FW_DOMAIN_ID_MEDIA_VEBOX3,
+ XE_FW_DOMAIN_ID_GSC,
+ XE_FW_DOMAIN_ID_COUNT
+};
+
+enum xe_force_wake_domains {
+ XE_FW_GT = BIT(XE_FW_DOMAIN_ID_GT),
+ XE_FW_RENDER = BIT(XE_FW_DOMAIN_ID_RENDER),
+ XE_FW_MEDIA = BIT(XE_FW_DOMAIN_ID_MEDIA),
+ XE_FW_MEDIA_VDBOX0 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX0),
+ XE_FW_MEDIA_VDBOX1 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX1),
+ XE_FW_MEDIA_VDBOX2 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX2),
+ XE_FW_MEDIA_VDBOX3 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX3),
+ XE_FW_MEDIA_VDBOX4 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX4),
+ XE_FW_MEDIA_VDBOX5 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX5),
+ XE_FW_MEDIA_VDBOX6 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX6),
+ XE_FW_MEDIA_VDBOX7 = BIT(XE_FW_DOMAIN_ID_MEDIA_VDBOX7),
+ XE_FW_MEDIA_VEBOX0 = BIT(XE_FW_DOMAIN_ID_MEDIA_VEBOX0),
+ XE_FW_MEDIA_VEBOX1 = BIT(XE_FW_DOMAIN_ID_MEDIA_VEBOX1),
+ XE_FW_MEDIA_VEBOX2 = BIT(XE_FW_DOMAIN_ID_MEDIA_VEBOX2),
+ XE_FW_MEDIA_VEBOX3 = BIT(XE_FW_DOMAIN_ID_MEDIA_VEBOX3),
+ XE_FW_GSC = BIT(XE_FW_DOMAIN_ID_GSC),
+ XE_FORCEWAKE_ALL = BIT(XE_FW_DOMAIN_ID_COUNT) - 1
+};
+
+/**
+ * struct xe_force_wake_domain - XE force wake domains
+ */
+struct xe_force_wake_domain {
+ /** @id: domain force wake id */
+ enum xe_force_wake_domain_id id;
+ /** @reg_ctl: domain wake control register address */
+ struct xe_reg reg_ctl;
+ /** @reg_ack: domain ack register address */
+ struct xe_reg reg_ack;
+ /** @val: domain wake write value */
+ u32 val;
+ /** @mask: domain mask */
+ u32 mask;
+ /** @ref: domain reference */
+ u32 ref;
+};
+
+/**
+ * struct xe_force_wake - XE force wake
+ */
+struct xe_force_wake {
+ /** @gt: back pointers to GT */
+ struct xe_gt *gt;
+ /** @lock: protects everything force wake struct */
+ spinlock_t lock;
+ /** @awake_domains: mask of all domains awake */
+ enum xe_force_wake_domains awake_domains;
+ /** @domains: force wake domains */
+ struct xe_force_wake_domain domains[XE_FW_DOMAIN_ID_COUNT];
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gen_wa_oob.c b/drivers/gpu/drm/xe/xe_gen_wa_oob.c
new file mode 100644
index 000000000000..106ee2b027f0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gen_wa_oob.c
@@ -0,0 +1,165 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#define _GNU_SOURCE
+#include <ctype.h>
+#include <errno.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#define HEADER \
+ "// SPDX-License-Identifier: MIT\n" \
+ "\n" \
+ "/*\n" \
+ " * DO NOT MODIFY.\n" \
+ " *\n" \
+ " * This file was generated from rules: %s\n" \
+ " */\n" \
+ "#ifndef _GENERATED_XE_WA_OOB_\n" \
+ "#define _GENERATED_XE_WA_OOB_\n" \
+ "\n" \
+ "enum {\n"
+
+#define FOOTER \
+ "};\n" \
+ "\n" \
+ "#endif\n"
+
+static void print_usage(FILE *f)
+{
+ fprintf(f, "usage: %s <input-rule-file> <generated-c-source-file> <generated-c-header-file>\n",
+ program_invocation_short_name);
+}
+
+static void print_parse_error(const char *err_msg, const char *line,
+ unsigned int lineno)
+{
+ fprintf(stderr, "ERROR: %s\nERROR: %u: %.60s\n",
+ err_msg, lineno, line);
+}
+
+static char *strip(char *line, size_t linelen)
+{
+ while (isspace(*(line + linelen)))
+ linelen--;
+
+ line[linelen - 1] = '\0';
+
+ return line + strspn(line, " \f\n\r\t\v");
+}
+
+#define MAX_LINE_LEN 4096
+static int parse(FILE *input, FILE *csource, FILE *cheader)
+{
+ char line[MAX_LINE_LEN + 1];
+ char *name, *prev_name = NULL, *rules;
+ unsigned int lineno = 0, idx = 0;
+
+ while (fgets(line, sizeof(line), input)) {
+ size_t linelen;
+ bool is_continuation;
+
+ if (line[0] == '\0' || line[0] == '#' || line[0] == '\n') {
+ lineno++;
+ continue;
+ }
+
+ linelen = strlen(line);
+ if (linelen == MAX_LINE_LEN) {
+ print_parse_error("line too long", line, lineno);
+ return -EINVAL;
+ }
+
+ is_continuation = isspace(line[0]);
+ name = strip(line, linelen);
+
+ if (!is_continuation) {
+ name = strtok(name, " \t");
+ rules = strtok(NULL, "");
+ } else {
+ if (!prev_name) {
+ print_parse_error("invalid rule continuation",
+ line, lineno);
+ return -EINVAL;
+ }
+
+ rules = name;
+ name = NULL;
+ }
+
+ if (rules[0] == '\0') {
+ print_parse_error("invalid empty rule\n", line, lineno);
+ return -EINVAL;
+ }
+
+ if (name) {
+ fprintf(cheader, "\tXE_WA_OOB_%s = %u,\n", name, idx);
+ fprintf(csource, "{ XE_RTP_NAME(\"%s\"), XE_RTP_RULES(%s) },\n",
+ name, rules);
+ } else {
+ fprintf(csource, "{ XE_RTP_NAME(NULL), XE_RTP_RULES(%s) },\n",
+ rules);
+ }
+
+ idx++;
+ lineno++;
+ if (!is_continuation)
+ prev_name = name;
+ }
+
+ fprintf(cheader, "\t_XE_WA_OOB_COUNT = %u\n", idx);
+
+ return 0;
+}
+
+int main(int argc, const char *argv[])
+{
+ enum {
+ ARGS_INPUT,
+ ARGS_CSOURCE,
+ ARGS_CHEADER,
+ _ARGS_COUNT
+ };
+ struct {
+ const char *fn;
+ const char *mode;
+ FILE *f;
+ } args[] = {
+ [ARGS_INPUT] = { .fn = argv[1], .mode = "r" },
+ [ARGS_CSOURCE] = { .fn = argv[2], .mode = "w" },
+ [ARGS_CHEADER] = { .fn = argv[3], .mode = "w" },
+ };
+ int ret = 1;
+
+ if (argc < 3) {
+ fprintf(stderr, "ERROR: wrong arguments\n");
+ print_usage(stderr);
+ return 1;
+ }
+
+ for (int i = 0; i < _ARGS_COUNT; i++) {
+ args[i].f = fopen(args[i].fn, args[i].mode);
+ if (!args[i].f) {
+ fprintf(stderr, "ERROR: Can't open %s: %m\n",
+ args[i].fn);
+ goto err;
+ }
+ }
+
+ fprintf(args[ARGS_CHEADER].f, HEADER, args[ARGS_INPUT].fn);
+ ret = parse(args[ARGS_INPUT].f, args[ARGS_CSOURCE].f,
+ args[ARGS_CHEADER].f);
+ if (!ret)
+ fprintf(args[ARGS_CHEADER].f, FOOTER);
+
+err:
+ for (int i = 0; i < _ARGS_COUNT; i++) {
+ if (args[i].f)
+ fclose(args[i].f);
+ }
+
+ return ret;
+}
diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c
new file mode 100644
index 000000000000..3efd2d066bf7
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ggtt.c
@@ -0,0 +1,428 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_ggtt.h"
+
+#include <linux/sizes.h>
+
+#include <drm/drm_managed.h>
+#include <drm/i915_drm.h>
+
+#include "regs/xe_gt_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_wopcm.h"
+
+#define XELPG_GGTT_PTE_PAT0 BIT_ULL(52)
+#define XELPG_GGTT_PTE_PAT1 BIT_ULL(53)
+
+/* GuC addresses above GUC_GGTT_TOP also don't map through the GTT */
+#define GUC_GGTT_TOP 0xFEE00000
+
+static u64 xelp_ggtt_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
+ u16 pat_index)
+{
+ u64 pte;
+
+ pte = xe_bo_addr(bo, bo_offset, XE_PAGE_SIZE);
+ pte |= XE_PAGE_PRESENT;
+
+ if (xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo))
+ pte |= XE_GGTT_PTE_DM;
+
+ return pte;
+}
+
+static u64 xelpg_ggtt_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
+ u16 pat_index)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ u64 pte;
+
+ pte = xelp_ggtt_pte_encode_bo(bo, bo_offset, pat_index);
+
+ xe_assert(xe, pat_index <= 3);
+
+ if (pat_index & BIT(0))
+ pte |= XELPG_GGTT_PTE_PAT0;
+
+ if (pat_index & BIT(1))
+ pte |= XELPG_GGTT_PTE_PAT1;
+
+ return pte;
+}
+
+static unsigned int probe_gsm_size(struct pci_dev *pdev)
+{
+ u16 gmch_ctl, ggms;
+
+ pci_read_config_word(pdev, SNB_GMCH_CTRL, &gmch_ctl);
+ ggms = (gmch_ctl >> BDW_GMCH_GGMS_SHIFT) & BDW_GMCH_GGMS_MASK;
+ return ggms ? SZ_1M << ggms : 0;
+}
+
+void xe_ggtt_set_pte(struct xe_ggtt *ggtt, u64 addr, u64 pte)
+{
+ xe_tile_assert(ggtt->tile, !(addr & XE_PTE_MASK));
+ xe_tile_assert(ggtt->tile, addr < ggtt->size);
+
+ writeq(pte, &ggtt->gsm[addr >> XE_PTE_SHIFT]);
+}
+
+static void xe_ggtt_clear(struct xe_ggtt *ggtt, u64 start, u64 size)
+{
+ u16 pat_index = tile_to_xe(ggtt->tile)->pat.idx[XE_CACHE_WB];
+ u64 end = start + size - 1;
+ u64 scratch_pte;
+
+ xe_tile_assert(ggtt->tile, start < end);
+
+ if (ggtt->scratch)
+ scratch_pte = ggtt->pt_ops->pte_encode_bo(ggtt->scratch, 0,
+ pat_index);
+ else
+ scratch_pte = 0;
+
+ while (start < end) {
+ xe_ggtt_set_pte(ggtt, start, scratch_pte);
+ start += XE_PAGE_SIZE;
+ }
+}
+
+static void ggtt_fini_early(struct drm_device *drm, void *arg)
+{
+ struct xe_ggtt *ggtt = arg;
+
+ mutex_destroy(&ggtt->lock);
+ drm_mm_takedown(&ggtt->mm);
+}
+
+static void ggtt_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_ggtt *ggtt = arg;
+
+ ggtt->scratch = NULL;
+}
+
+static void primelockdep(struct xe_ggtt *ggtt)
+{
+ if (!IS_ENABLED(CONFIG_LOCKDEP))
+ return;
+
+ fs_reclaim_acquire(GFP_KERNEL);
+ might_lock(&ggtt->lock);
+ fs_reclaim_release(GFP_KERNEL);
+}
+
+static const struct xe_ggtt_pt_ops xelp_pt_ops = {
+ .pte_encode_bo = xelp_ggtt_pte_encode_bo,
+};
+
+static const struct xe_ggtt_pt_ops xelpg_pt_ops = {
+ .pte_encode_bo = xelpg_ggtt_pte_encode_bo,
+};
+
+/*
+ * Early GGTT initialization, which allows to create new mappings usable by the
+ * GuC.
+ * Mappings are not usable by the HW engines, as it doesn't have scratch /
+ * initial clear done to it yet. That will happen in the regular, non-early
+ * GGTT init.
+ */
+int xe_ggtt_init_early(struct xe_ggtt *ggtt)
+{
+ struct xe_device *xe = tile_to_xe(ggtt->tile);
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ unsigned int gsm_size;
+
+ gsm_size = probe_gsm_size(pdev);
+ if (gsm_size == 0) {
+ drm_err(&xe->drm, "Hardware reported no preallocated GSM\n");
+ return -ENOMEM;
+ }
+
+ ggtt->gsm = ggtt->tile->mmio.regs + SZ_8M;
+ ggtt->size = (gsm_size / 8) * (u64) XE_PAGE_SIZE;
+
+ if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)
+ ggtt->flags |= XE_GGTT_FLAGS_64K;
+
+ /*
+ * 8B per entry, each points to a 4KB page.
+ *
+ * The GuC address space is limited on both ends of the GGTT, because
+ * the GuC shim HW redirects accesses to those addresses to other HW
+ * areas instead of going through the GGTT. On the bottom end, the GuC
+ * can't access offsets below the WOPCM size, while on the top side the
+ * limit is fixed at GUC_GGTT_TOP. To keep things simple, instead of
+ * checking each object to see if they are accessed by GuC or not, we
+ * just exclude those areas from the allocator. Additionally, to
+ * simplify the driver load, we use the maximum WOPCM size in this logic
+ * instead of the programmed one, so we don't need to wait until the
+ * actual size to be programmed is determined (which requires FW fetch)
+ * before initializing the GGTT. These simplifications might waste space
+ * in the GGTT (about 20-25 MBs depending on the platform) but we can
+ * live with this.
+ *
+ * Another benifit of this is the GuC bootrom can't access anything
+ * below the WOPCM max size so anything the bootom needs to access (e.g.
+ * a RSA key) needs to be placed in the GGTT above the WOPCM max size.
+ * Starting the GGTT allocations above the WOPCM max give us the correct
+ * placement for free.
+ */
+ if (ggtt->size > GUC_GGTT_TOP)
+ ggtt->size = GUC_GGTT_TOP;
+
+ if (GRAPHICS_VERx100(xe) >= 1270)
+ ggtt->pt_ops = &xelpg_pt_ops;
+ else
+ ggtt->pt_ops = &xelp_pt_ops;
+
+ drm_mm_init(&ggtt->mm, xe_wopcm_size(xe),
+ ggtt->size - xe_wopcm_size(xe));
+ mutex_init(&ggtt->lock);
+ primelockdep(ggtt);
+
+ return drmm_add_action_or_reset(&xe->drm, ggtt_fini_early, ggtt);
+}
+
+static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt)
+{
+ struct drm_mm_node *hole;
+ u64 start, end;
+
+ /* Display may have allocated inside ggtt, so be careful with clearing here */
+ xe_device_mem_access_get(tile_to_xe(ggtt->tile));
+ mutex_lock(&ggtt->lock);
+ drm_mm_for_each_hole(hole, &ggtt->mm, start, end)
+ xe_ggtt_clear(ggtt, start, end - start);
+
+ xe_ggtt_invalidate(ggtt);
+ mutex_unlock(&ggtt->lock);
+ xe_device_mem_access_put(tile_to_xe(ggtt->tile));
+}
+
+int xe_ggtt_init(struct xe_ggtt *ggtt)
+{
+ struct xe_device *xe = tile_to_xe(ggtt->tile);
+ unsigned int flags;
+ int err;
+
+ /*
+ * So we don't need to worry about 64K GGTT layout when dealing with
+ * scratch entires, rather keep the scratch page in system memory on
+ * platforms where 64K pages are needed for VRAM.
+ */
+ flags = XE_BO_CREATE_PINNED_BIT;
+ if (ggtt->flags & XE_GGTT_FLAGS_64K)
+ flags |= XE_BO_CREATE_SYSTEM_BIT;
+ else
+ flags |= XE_BO_CREATE_VRAM_IF_DGFX(ggtt->tile);
+
+ ggtt->scratch = xe_managed_bo_create_pin_map(xe, ggtt->tile, XE_PAGE_SIZE, flags);
+ if (IS_ERR(ggtt->scratch)) {
+ err = PTR_ERR(ggtt->scratch);
+ goto err;
+ }
+
+ xe_map_memset(xe, &ggtt->scratch->vmap, 0, 0, ggtt->scratch->size);
+
+ xe_ggtt_initial_clear(ggtt);
+
+ return drmm_add_action_or_reset(&xe->drm, ggtt_fini, ggtt);
+err:
+ ggtt->scratch = NULL;
+ return err;
+}
+
+#define GUC_TLB_INV_CR XE_REG(0xcee8)
+#define GUC_TLB_INV_CR_INVALIDATE REG_BIT(0)
+#define PVC_GUC_TLB_INV_DESC0 XE_REG(0xcf7c)
+#define PVC_GUC_TLB_INV_DESC0_VALID REG_BIT(0)
+#define PVC_GUC_TLB_INV_DESC1 XE_REG(0xcf80)
+#define PVC_GUC_TLB_INV_DESC1_INVALIDATE REG_BIT(6)
+
+static void ggtt_invalidate_gt_tlb(struct xe_gt *gt)
+{
+ if (!gt)
+ return;
+
+ /*
+ * Invalidation can happen when there's no in-flight work keeping the
+ * GT awake. We need to explicitly grab forcewake to ensure the GT
+ * and GuC are accessible.
+ */
+ xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+
+ /* TODO: vfunc for GuC vs. non-GuC */
+
+ if (gt->uc.guc.submission_state.enabled) {
+ int seqno;
+
+ seqno = xe_gt_tlb_invalidation_guc(gt);
+ xe_gt_assert(gt, seqno > 0);
+ if (seqno > 0)
+ xe_gt_tlb_invalidation_wait(gt, seqno);
+ } else if (xe_device_uc_enabled(gt_to_xe(gt))) {
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (xe->info.platform == XE_PVC || GRAPHICS_VER(xe) >= 20) {
+ xe_mmio_write32(gt, PVC_GUC_TLB_INV_DESC1,
+ PVC_GUC_TLB_INV_DESC1_INVALIDATE);
+ xe_mmio_write32(gt, PVC_GUC_TLB_INV_DESC0,
+ PVC_GUC_TLB_INV_DESC0_VALID);
+ } else
+ xe_mmio_write32(gt, GUC_TLB_INV_CR,
+ GUC_TLB_INV_CR_INVALIDATE);
+ }
+
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+}
+
+void xe_ggtt_invalidate(struct xe_ggtt *ggtt)
+{
+ /* Each GT in a tile has its own TLB to cache GGTT lookups */
+ ggtt_invalidate_gt_tlb(ggtt->tile->primary_gt);
+ ggtt_invalidate_gt_tlb(ggtt->tile->media_gt);
+}
+
+void xe_ggtt_printk(struct xe_ggtt *ggtt, const char *prefix)
+{
+ u16 pat_index = tile_to_xe(ggtt->tile)->pat.idx[XE_CACHE_WB];
+ u64 addr, scratch_pte;
+
+ scratch_pte = ggtt->pt_ops->pte_encode_bo(ggtt->scratch, 0, pat_index);
+
+ printk("%sGlobal GTT:", prefix);
+ for (addr = 0; addr < ggtt->size; addr += XE_PAGE_SIZE) {
+ unsigned int i = addr / XE_PAGE_SIZE;
+
+ xe_tile_assert(ggtt->tile, addr <= U32_MAX);
+ if (ggtt->gsm[i] == scratch_pte)
+ continue;
+
+ printk("%s ggtt[0x%08x] = 0x%016llx",
+ prefix, (u32)addr, ggtt->gsm[i]);
+ }
+}
+
+int xe_ggtt_insert_special_node_locked(struct xe_ggtt *ggtt, struct drm_mm_node *node,
+ u32 size, u32 align, u32 mm_flags)
+{
+ return drm_mm_insert_node_generic(&ggtt->mm, node, size, align, 0,
+ mm_flags);
+}
+
+int xe_ggtt_insert_special_node(struct xe_ggtt *ggtt, struct drm_mm_node *node,
+ u32 size, u32 align)
+{
+ int ret;
+
+ mutex_lock(&ggtt->lock);
+ ret = xe_ggtt_insert_special_node_locked(ggtt, node, size,
+ align, DRM_MM_INSERT_HIGH);
+ mutex_unlock(&ggtt->lock);
+
+ return ret;
+}
+
+void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
+{
+ u16 pat_index = tile_to_xe(ggtt->tile)->pat.idx[XE_CACHE_WB];
+ u64 start = bo->ggtt_node.start;
+ u64 offset, pte;
+
+ for (offset = 0; offset < bo->size; offset += XE_PAGE_SIZE) {
+ pte = ggtt->pt_ops->pte_encode_bo(bo, offset, pat_index);
+ xe_ggtt_set_pte(ggtt, start + offset, pte);
+ }
+
+ xe_ggtt_invalidate(ggtt);
+}
+
+static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
+ u64 start, u64 end)
+{
+ int err;
+ u64 alignment = XE_PAGE_SIZE;
+
+ if (xe_bo_is_vram(bo) && ggtt->flags & XE_GGTT_FLAGS_64K)
+ alignment = SZ_64K;
+
+ if (XE_WARN_ON(bo->ggtt_node.size)) {
+ /* Someone's already inserted this BO in the GGTT */
+ xe_tile_assert(ggtt->tile, bo->ggtt_node.size == bo->size);
+ return 0;
+ }
+
+ err = xe_bo_validate(bo, NULL, false);
+ if (err)
+ return err;
+
+ xe_device_mem_access_get(tile_to_xe(ggtt->tile));
+ mutex_lock(&ggtt->lock);
+ err = drm_mm_insert_node_in_range(&ggtt->mm, &bo->ggtt_node, bo->size,
+ alignment, 0, start, end, 0);
+ if (!err)
+ xe_ggtt_map_bo(ggtt, bo);
+ mutex_unlock(&ggtt->lock);
+ xe_device_mem_access_put(tile_to_xe(ggtt->tile));
+
+ return err;
+}
+
+int xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
+ u64 start, u64 end)
+{
+ return __xe_ggtt_insert_bo_at(ggtt, bo, start, end);
+}
+
+int xe_ggtt_insert_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
+{
+ return __xe_ggtt_insert_bo_at(ggtt, bo, 0, U64_MAX);
+}
+
+void xe_ggtt_remove_node(struct xe_ggtt *ggtt, struct drm_mm_node *node)
+{
+ xe_device_mem_access_get(tile_to_xe(ggtt->tile));
+ mutex_lock(&ggtt->lock);
+
+ xe_ggtt_clear(ggtt, node->start, node->size);
+ drm_mm_remove_node(node);
+ node->size = 0;
+
+ xe_ggtt_invalidate(ggtt);
+
+ mutex_unlock(&ggtt->lock);
+ xe_device_mem_access_put(tile_to_xe(ggtt->tile));
+}
+
+void xe_ggtt_remove_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
+{
+ if (XE_WARN_ON(!bo->ggtt_node.size))
+ return;
+
+ /* This BO is not currently in the GGTT */
+ xe_tile_assert(ggtt->tile, bo->ggtt_node.size == bo->size);
+
+ xe_ggtt_remove_node(ggtt, &bo->ggtt_node);
+}
+
+int xe_ggtt_dump(struct xe_ggtt *ggtt, struct drm_printer *p)
+{
+ int err;
+
+ err = mutex_lock_interruptible(&ggtt->lock);
+ if (err)
+ return err;
+
+ drm_mm_print(&ggtt->mm, p);
+ mutex_unlock(&ggtt->lock);
+ return err;
+}
diff --git a/drivers/gpu/drm/xe/xe_ggtt.h b/drivers/gpu/drm/xe/xe_ggtt.h
new file mode 100644
index 000000000000..a09c166dff70
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ggtt.h
@@ -0,0 +1,33 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_GGTT_H_
+#define _XE_GGTT_H_
+
+#include "xe_ggtt_types.h"
+
+struct drm_printer;
+
+void xe_ggtt_set_pte(struct xe_ggtt *ggtt, u64 addr, u64 pte);
+void xe_ggtt_invalidate(struct xe_ggtt *ggtt);
+int xe_ggtt_init_early(struct xe_ggtt *ggtt);
+int xe_ggtt_init(struct xe_ggtt *ggtt);
+void xe_ggtt_printk(struct xe_ggtt *ggtt, const char *prefix);
+
+int xe_ggtt_insert_special_node(struct xe_ggtt *ggtt, struct drm_mm_node *node,
+ u32 size, u32 align);
+int xe_ggtt_insert_special_node_locked(struct xe_ggtt *ggtt,
+ struct drm_mm_node *node,
+ u32 size, u32 align, u32 mm_flags);
+void xe_ggtt_remove_node(struct xe_ggtt *ggtt, struct drm_mm_node *node);
+void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo);
+int xe_ggtt_insert_bo(struct xe_ggtt *ggtt, struct xe_bo *bo);
+int xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
+ u64 start, u64 end);
+void xe_ggtt_remove_bo(struct xe_ggtt *ggtt, struct xe_bo *bo);
+
+int xe_ggtt_dump(struct xe_ggtt *ggtt, struct drm_printer *p);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ggtt_types.h b/drivers/gpu/drm/xe/xe_ggtt_types.h
new file mode 100644
index 000000000000..d8c584d9a8c3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ggtt_types.h
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GGTT_TYPES_H_
+#define _XE_GGTT_TYPES_H_
+
+#include <drm/drm_mm.h>
+
+#include "xe_pt_types.h"
+
+struct xe_bo;
+struct xe_gt;
+
+struct xe_ggtt_pt_ops {
+ u64 (*pte_encode_bo)(struct xe_bo *bo, u64 bo_offset, u16 pat_index);
+};
+
+struct xe_ggtt {
+ struct xe_tile *tile;
+
+ u64 size;
+
+#define XE_GGTT_FLAGS_64K BIT(0)
+ unsigned int flags;
+
+ struct xe_bo *scratch;
+
+ struct mutex lock;
+
+ u64 __iomem *gsm;
+
+ const struct xe_ggtt_pt_ops *pt_ops;
+
+ struct drm_mm mm;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gpu_scheduler.c b/drivers/gpu/drm/xe/xe_gpu_scheduler.c
new file mode 100644
index 000000000000..e4ad1d6ce1d5
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gpu_scheduler.c
@@ -0,0 +1,101 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_gpu_scheduler.h"
+
+static void xe_sched_process_msg_queue(struct xe_gpu_scheduler *sched)
+{
+ if (!READ_ONCE(sched->base.pause_submit))
+ queue_work(sched->base.submit_wq, &sched->work_process_msg);
+}
+
+static void xe_sched_process_msg_queue_if_ready(struct xe_gpu_scheduler *sched)
+{
+ struct xe_sched_msg *msg;
+
+ spin_lock(&sched->base.job_list_lock);
+ msg = list_first_entry_or_null(&sched->msgs, struct xe_sched_msg, link);
+ if (msg)
+ xe_sched_process_msg_queue(sched);
+ spin_unlock(&sched->base.job_list_lock);
+}
+
+static struct xe_sched_msg *
+xe_sched_get_msg(struct xe_gpu_scheduler *sched)
+{
+ struct xe_sched_msg *msg;
+
+ spin_lock(&sched->base.job_list_lock);
+ msg = list_first_entry_or_null(&sched->msgs,
+ struct xe_sched_msg, link);
+ if (msg)
+ list_del(&msg->link);
+ spin_unlock(&sched->base.job_list_lock);
+
+ return msg;
+}
+
+static void xe_sched_process_msg_work(struct work_struct *w)
+{
+ struct xe_gpu_scheduler *sched =
+ container_of(w, struct xe_gpu_scheduler, work_process_msg);
+ struct xe_sched_msg *msg;
+
+ if (READ_ONCE(sched->base.pause_submit))
+ return;
+
+ msg = xe_sched_get_msg(sched);
+ if (msg) {
+ sched->ops->process_msg(msg);
+
+ xe_sched_process_msg_queue_if_ready(sched);
+ }
+}
+
+int xe_sched_init(struct xe_gpu_scheduler *sched,
+ const struct drm_sched_backend_ops *ops,
+ const struct xe_sched_backend_ops *xe_ops,
+ struct workqueue_struct *submit_wq,
+ uint32_t hw_submission, unsigned hang_limit,
+ long timeout, struct workqueue_struct *timeout_wq,
+ atomic_t *score, const char *name,
+ struct device *dev)
+{
+ sched->ops = xe_ops;
+ INIT_LIST_HEAD(&sched->msgs);
+ INIT_WORK(&sched->work_process_msg, xe_sched_process_msg_work);
+
+ return drm_sched_init(&sched->base, ops, submit_wq, 1, hw_submission,
+ hang_limit, timeout, timeout_wq, score, name,
+ dev);
+}
+
+void xe_sched_fini(struct xe_gpu_scheduler *sched)
+{
+ xe_sched_submission_stop(sched);
+ drm_sched_fini(&sched->base);
+}
+
+void xe_sched_submission_start(struct xe_gpu_scheduler *sched)
+{
+ drm_sched_wqueue_start(&sched->base);
+ queue_work(sched->base.submit_wq, &sched->work_process_msg);
+}
+
+void xe_sched_submission_stop(struct xe_gpu_scheduler *sched)
+{
+ drm_sched_wqueue_stop(&sched->base);
+ cancel_work_sync(&sched->work_process_msg);
+}
+
+void xe_sched_add_msg(struct xe_gpu_scheduler *sched,
+ struct xe_sched_msg *msg)
+{
+ spin_lock(&sched->base.job_list_lock);
+ list_add_tail(&msg->link, &sched->msgs);
+ spin_unlock(&sched->base.job_list_lock);
+
+ xe_sched_process_msg_queue(sched);
+}
diff --git a/drivers/gpu/drm/xe/xe_gpu_scheduler.h b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
new file mode 100644
index 000000000000..10c6bb9c9386
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
@@ -0,0 +1,73 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GPU_SCHEDULER_H_
+#define _XE_GPU_SCHEDULER_H_
+
+#include "xe_gpu_scheduler_types.h"
+#include "xe_sched_job_types.h"
+
+int xe_sched_init(struct xe_gpu_scheduler *sched,
+ const struct drm_sched_backend_ops *ops,
+ const struct xe_sched_backend_ops *xe_ops,
+ struct workqueue_struct *submit_wq,
+ uint32_t hw_submission, unsigned hang_limit,
+ long timeout, struct workqueue_struct *timeout_wq,
+ atomic_t *score, const char *name,
+ struct device *dev);
+void xe_sched_fini(struct xe_gpu_scheduler *sched);
+
+void xe_sched_submission_start(struct xe_gpu_scheduler *sched);
+void xe_sched_submission_stop(struct xe_gpu_scheduler *sched);
+
+void xe_sched_add_msg(struct xe_gpu_scheduler *sched,
+ struct xe_sched_msg *msg);
+
+static inline void xe_sched_stop(struct xe_gpu_scheduler *sched)
+{
+ drm_sched_stop(&sched->base, NULL);
+}
+
+static inline void xe_sched_tdr_queue_imm(struct xe_gpu_scheduler *sched)
+{
+ drm_sched_tdr_queue_imm(&sched->base);
+}
+
+static inline void xe_sched_resubmit_jobs(struct xe_gpu_scheduler *sched)
+{
+ drm_sched_resubmit_jobs(&sched->base);
+}
+
+static inline bool
+xe_sched_invalidate_job(struct xe_sched_job *job, int threshold)
+{
+ return drm_sched_invalidate_job(&job->drm, threshold);
+}
+
+static inline void xe_sched_add_pending_job(struct xe_gpu_scheduler *sched,
+ struct xe_sched_job *job)
+{
+ list_add(&job->drm.list, &sched->base.pending_list);
+}
+
+static inline
+struct xe_sched_job *xe_sched_first_pending_job(struct xe_gpu_scheduler *sched)
+{
+ return list_first_entry_or_null(&sched->base.pending_list,
+ struct xe_sched_job, drm.list);
+}
+
+static inline int
+xe_sched_entity_init(struct xe_sched_entity *entity,
+ struct xe_gpu_scheduler *sched)
+{
+ return drm_sched_entity_init(entity, 0,
+ (struct drm_gpu_scheduler **)&sched,
+ 1, NULL);
+}
+
+#define xe_sched_entity_fini drm_sched_entity_fini
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gpu_scheduler_types.h b/drivers/gpu/drm/xe/xe_gpu_scheduler_types.h
new file mode 100644
index 000000000000..6731b13da8bb
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gpu_scheduler_types.h
@@ -0,0 +1,57 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GPU_SCHEDULER_TYPES_H_
+#define _XE_GPU_SCHEDULER_TYPES_H_
+
+#include <drm/gpu_scheduler.h>
+
+/**
+ * struct xe_sched_msg - an in-band (relative to GPU scheduler run queue)
+ * message
+ *
+ * Generic enough for backend defined messages, backend can expand if needed.
+ */
+struct xe_sched_msg {
+ /** @link: list link into the gpu scheduler list of messages */
+ struct list_head link;
+ /**
+ * @private_data: opaque pointer to message private data (backend defined)
+ */
+ void *private_data;
+ /** @opcode: opcode of message (backend defined) */
+ unsigned int opcode;
+};
+
+/**
+ * struct xe_sched_backend_ops - Define the backend operations called by the
+ * scheduler
+ */
+struct xe_sched_backend_ops {
+ /**
+ * @process_msg: Process a message. Allowed to block, it is this
+ * function's responsibility to free message if dynamically allocated.
+ */
+ void (*process_msg)(struct xe_sched_msg *msg);
+};
+
+/**
+ * struct xe_gpu_scheduler - Xe GPU scheduler
+ */
+struct xe_gpu_scheduler {
+ /** @base: DRM GPU scheduler */
+ struct drm_gpu_scheduler base;
+ /** @ops: Xe scheduler ops */
+ const struct xe_sched_backend_ops *ops;
+ /** @msgs: list of messages to be processed in @work_process_msg */
+ struct list_head msgs;
+ /** @work_process_msg: processes messages */
+ struct work_struct work_process_msg;
+};
+
+#define xe_sched_entity drm_sched_entity
+#define xe_sched_policy drm_sched_policy
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gsc.c b/drivers/gpu/drm/xe/xe_gsc.c
new file mode 100644
index 000000000000..a8a895cf4b44
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gsc.c
@@ -0,0 +1,438 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_gsc.h"
+
+#include <drm/drm_managed.h>
+
+#include "abi/gsc_mkhi_commands_abi.h"
+#include "generated/xe_wa_oob.h"
+#include "xe_bb.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_gsc_submit.h"
+#include "xe_gt.h"
+#include "xe_gt_printk.h"
+#include "xe_huc.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_sched_job.h"
+#include "xe_uc_fw.h"
+#include "xe_wa.h"
+#include "instructions/xe_gsc_commands.h"
+#include "regs/xe_gsc_regs.h"
+
+static struct xe_gt *
+gsc_to_gt(struct xe_gsc *gsc)
+{
+ return container_of(gsc, struct xe_gt, uc.gsc);
+}
+
+static int memcpy_fw(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 fw_size = gsc->fw.size;
+ void *storage;
+
+ /*
+ * FIXME: xe_migrate_copy does not work with stolen mem yet, so we use
+ * a memcpy for now.
+ */
+ storage = kmalloc(fw_size, GFP_KERNEL);
+ if (!storage)
+ return -ENOMEM;
+
+ xe_map_memcpy_from(xe, storage, &gsc->fw.bo->vmap, 0, fw_size);
+ xe_map_memcpy_to(xe, &gsc->private->vmap, 0, storage, fw_size);
+ xe_map_memset(xe, &gsc->private->vmap, fw_size, 0, gsc->private->size - fw_size);
+
+ kfree(storage);
+
+ return 0;
+}
+
+static int emit_gsc_upload(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ u64 offset = xe_bo_ggtt_addr(gsc->private);
+ struct xe_bb *bb;
+ struct xe_sched_job *job;
+ struct dma_fence *fence;
+ long timeout;
+
+ bb = xe_bb_new(gt, 4, false);
+ if (IS_ERR(bb))
+ return PTR_ERR(bb);
+
+ bb->cs[bb->len++] = GSC_FW_LOAD;
+ bb->cs[bb->len++] = lower_32_bits(offset);
+ bb->cs[bb->len++] = upper_32_bits(offset);
+ bb->cs[bb->len++] = (gsc->private->size / SZ_4K) | GSC_FW_LOAD_LIMIT_VALID;
+
+ job = xe_bb_create_job(gsc->q, bb);
+ if (IS_ERR(job)) {
+ xe_bb_free(bb, NULL);
+ return PTR_ERR(job);
+ }
+
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ timeout = dma_fence_wait_timeout(fence, false, HZ);
+ dma_fence_put(fence);
+ xe_bb_free(bb, NULL);
+ if (timeout < 0)
+ return timeout;
+ else if (!timeout)
+ return -ETIME;
+
+ return 0;
+}
+
+#define version_query_wr(xe_, map_, offset_, field_, val_) \
+ xe_map_wr_field(xe_, map_, offset_, struct gsc_get_compatibility_version_in, field_, val_)
+#define version_query_rd(xe_, map_, offset_, field_) \
+ xe_map_rd_field(xe_, map_, offset_, struct gsc_get_compatibility_version_out, field_)
+
+static u32 emit_version_query_msg(struct xe_device *xe, struct iosys_map *map, u32 wr_offset)
+{
+ xe_map_memset(xe, map, wr_offset, 0, sizeof(struct gsc_get_compatibility_version_in));
+
+ version_query_wr(xe, map, wr_offset, header.group_id, MKHI_GROUP_ID_GFX_SRV);
+ version_query_wr(xe, map, wr_offset, header.command,
+ MKHI_GFX_SRV_GET_HOST_COMPATIBILITY_VERSION);
+
+ return wr_offset + sizeof(struct gsc_get_compatibility_version_in);
+}
+
+#define GSC_VER_PKT_SZ SZ_4K /* 4K each for input and output */
+static int query_compatibility_version(struct xe_gsc *gsc)
+{
+ struct xe_uc_fw_version *compat = &gsc->fw.versions.found[XE_UC_FW_VER_COMPATIBILITY];
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_bo *bo;
+ u32 wr_offset;
+ u32 rd_offset;
+ u64 ggtt_offset;
+ int err;
+
+ bo = xe_bo_create_pin_map(xe, tile, NULL, GSC_VER_PKT_SZ * 2,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_SYSTEM_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo)) {
+ xe_gt_err(gt, "failed to allocate bo for GSC version query\n");
+ return PTR_ERR(bo);
+ }
+
+ ggtt_offset = xe_bo_ggtt_addr(bo);
+
+ wr_offset = xe_gsc_emit_header(xe, &bo->vmap, 0, HECI_MEADDRESS_MKHI, 0,
+ sizeof(struct gsc_get_compatibility_version_in));
+ wr_offset = emit_version_query_msg(xe, &bo->vmap, wr_offset);
+
+ err = xe_gsc_pkt_submit_kernel(gsc, ggtt_offset, wr_offset,
+ ggtt_offset + GSC_VER_PKT_SZ,
+ GSC_VER_PKT_SZ);
+ if (err) {
+ xe_gt_err(gt,
+ "failed to submit GSC request for compatibility version: %d\n",
+ err);
+ goto out_bo;
+ }
+
+ err = xe_gsc_read_out_header(xe, &bo->vmap, GSC_VER_PKT_SZ,
+ sizeof(struct gsc_get_compatibility_version_out),
+ &rd_offset);
+ if (err) {
+ xe_gt_err(gt, "HuC: invalid GSC reply for version query (err=%d)\n", err);
+ return err;
+ }
+
+ compat->major = version_query_rd(xe, &bo->vmap, rd_offset, compat_major);
+ compat->minor = version_query_rd(xe, &bo->vmap, rd_offset, compat_minor);
+
+ xe_gt_info(gt, "found GSC cv%u.%u\n", compat->major, compat->minor);
+
+out_bo:
+ xe_bo_unpin_map_no_vm(bo);
+ return err;
+}
+
+static int gsc_fw_is_loaded(struct xe_gt *gt)
+{
+ return xe_mmio_read32(gt, HECI_FWSTS1(MTL_GSC_HECI1_BASE)) &
+ HECI1_FWSTS1_INIT_COMPLETE;
+}
+
+static int gsc_fw_wait(struct xe_gt *gt)
+{
+ /*
+ * GSC load can take up to 250ms from the moment the instruction is
+ * executed by the GSCCS. To account for possible submission delays or
+ * other issues, we use a 500ms timeout in the wait here.
+ */
+ return xe_mmio_wait32(gt, HECI_FWSTS1(MTL_GSC_HECI1_BASE),
+ HECI1_FWSTS1_INIT_COMPLETE,
+ HECI1_FWSTS1_INIT_COMPLETE,
+ 500 * USEC_PER_MSEC, NULL, false);
+}
+
+static int gsc_upload(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_device *xe = gt_to_xe(gt);
+ int err;
+
+ /* we should only be here if the init step were successful */
+ xe_assert(xe, xe_uc_fw_is_loadable(&gsc->fw) && gsc->q);
+
+ if (gsc_fw_is_loaded(gt)) {
+ xe_gt_err(gt, "GSC already loaded at upload time\n");
+ return -EEXIST;
+ }
+
+ err = memcpy_fw(gsc);
+ if (err) {
+ xe_gt_err(gt, "Failed to memcpy GSC FW\n");
+ return err;
+ }
+
+ /*
+ * GSC is only killed by an FLR, so we need to trigger one on unload to
+ * make sure we stop it. This is because we assign a chunk of memory to
+ * the GSC as part of the FW load, so we need to make sure it stops
+ * using it when we release it to the system on driver unload. Note that
+ * this is not a problem of the unload per-se, because the GSC will not
+ * touch that memory unless there are requests for it coming from the
+ * driver; therefore, no accesses will happen while Xe is not loaded,
+ * but if we re-load the driver then the GSC might wake up and try to
+ * access that old memory location again.
+ * Given that an FLR is a very disruptive action (see the FLR function
+ * for details), we want to do it as the last action before releasing
+ * the access to the MMIO bar, which means we need to do it as part of
+ * mmio cleanup.
+ */
+ xe->needs_flr_on_fini = true;
+
+ err = emit_gsc_upload(gsc);
+ if (err) {
+ xe_gt_err(gt, "Failed to emit GSC FW upload (%pe)\n", ERR_PTR(err));
+ return err;
+ }
+
+ err = gsc_fw_wait(gt);
+ if (err) {
+ xe_gt_err(gt, "Failed to wait for GSC load (%pe)\n", ERR_PTR(err));
+ return err;
+ }
+
+ err = query_compatibility_version(gsc);
+ if (err)
+ return err;
+
+ err = xe_uc_fw_check_version_requirements(&gsc->fw);
+ if (err)
+ return err;
+
+ xe_gt_dbg(gt, "GSC FW async load completed\n");
+
+ return 0;
+}
+
+static void gsc_work(struct work_struct *work)
+{
+ struct xe_gsc *gsc = container_of(work, typeof(*gsc), work);
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_device *xe = gt_to_xe(gt);
+ int ret;
+
+ xe_device_mem_access_get(xe);
+ xe_force_wake_get(gt_to_fw(gt), XE_FW_GSC);
+
+ ret = gsc_upload(gsc);
+ if (ret && ret != -EEXIST) {
+ xe_uc_fw_change_status(&gsc->fw, XE_UC_FIRMWARE_LOAD_FAIL);
+ goto out;
+ }
+
+ xe_uc_fw_change_status(&gsc->fw, XE_UC_FIRMWARE_TRANSFERRED);
+
+ /* HuC auth failure is not fatal */
+ if (xe_huc_is_authenticated(&gt->uc.huc, XE_HUC_AUTH_VIA_GUC))
+ xe_huc_auth(&gt->uc.huc, XE_HUC_AUTH_VIA_GSC);
+
+out:
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GSC);
+ xe_device_mem_access_put(xe);
+}
+
+int xe_gsc_init(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ int ret;
+
+ gsc->fw.type = XE_UC_FW_TYPE_GSC;
+ INIT_WORK(&gsc->work, gsc_work);
+
+ /* The GSC uC is only available on the media GT */
+ if (tile->media_gt && (gt != tile->media_gt)) {
+ xe_uc_fw_change_status(&gsc->fw, XE_UC_FIRMWARE_NOT_SUPPORTED);
+ return 0;
+ }
+
+ /*
+ * Some platforms can have GuC but not GSC. That would cause
+ * xe_uc_fw_init(gsc) to return a "not supported" failure code and abort
+ * all firmware loading. So check for GSC being enabled before
+ * propagating the failure back up. That way the higher level will keep
+ * going and load GuC as appropriate.
+ */
+ ret = xe_uc_fw_init(&gsc->fw);
+ if (!xe_uc_fw_is_enabled(&gsc->fw))
+ return 0;
+ else if (ret)
+ goto out;
+
+ return 0;
+
+out:
+ xe_gt_err(gt, "GSC init failed with %d", ret);
+ return ret;
+}
+
+static void free_resources(struct drm_device *drm, void *arg)
+{
+ struct xe_gsc *gsc = arg;
+
+ if (gsc->wq) {
+ destroy_workqueue(gsc->wq);
+ gsc->wq = NULL;
+ }
+
+ if (gsc->q) {
+ xe_exec_queue_put(gsc->q);
+ gsc->q = NULL;
+ }
+
+ if (gsc->private) {
+ xe_bo_unpin_map_no_vm(gsc->private);
+ gsc->private = NULL;
+ }
+}
+
+int xe_gsc_init_post_hwconfig(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_hw_engine *hwe = xe_gt_hw_engine(gt, XE_ENGINE_CLASS_OTHER, 0, true);
+ struct xe_exec_queue *q;
+ struct workqueue_struct *wq;
+ struct xe_bo *bo;
+ int err;
+
+ if (!xe_uc_fw_is_available(&gsc->fw))
+ return 0;
+
+ if (!hwe)
+ return -ENODEV;
+
+ bo = xe_bo_create_pin_map(xe, tile, NULL, SZ_4M,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_STOLEN_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ q = xe_exec_queue_create(xe, NULL,
+ BIT(hwe->logical_instance), 1, hwe,
+ EXEC_QUEUE_FLAG_KERNEL |
+ EXEC_QUEUE_FLAG_PERMANENT);
+ if (IS_ERR(q)) {
+ xe_gt_err(gt, "Failed to create queue for GSC submission\n");
+ err = PTR_ERR(q);
+ goto out_bo;
+ }
+
+ wq = alloc_ordered_workqueue("gsc-ordered-wq", 0);
+ if (!wq) {
+ err = -ENOMEM;
+ goto out_q;
+ }
+
+ gsc->private = bo;
+ gsc->q = q;
+ gsc->wq = wq;
+
+ err = drmm_add_action_or_reset(&xe->drm, free_resources, gsc);
+ if (err)
+ return err;
+
+ xe_uc_fw_change_status(&gsc->fw, XE_UC_FIRMWARE_LOADABLE);
+
+ return 0;
+
+out_q:
+ xe_exec_queue_put(q);
+out_bo:
+ xe_bo_unpin_map_no_vm(bo);
+ return err;
+}
+
+void xe_gsc_load_start(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+
+ if (!xe_uc_fw_is_loadable(&gsc->fw) || !gsc->q)
+ return;
+
+ /* GSC FW survives GT reset and D3Hot */
+ if (gsc_fw_is_loaded(gt)) {
+ xe_uc_fw_change_status(&gsc->fw, XE_UC_FIRMWARE_TRANSFERRED);
+ return;
+ }
+
+ queue_work(gsc->wq, &gsc->work);
+}
+
+void xe_gsc_wait_for_worker_completion(struct xe_gsc *gsc)
+{
+ if (xe_uc_fw_is_loadable(&gsc->fw) && gsc->wq)
+ flush_work(&gsc->work);
+}
+
+/*
+ * wa_14015076503: if the GSC FW is loaded, we need to alert it before doing a
+ * GSC engine reset by writing a notification bit in the GS1 register and then
+ * triggering an interrupt to GSC; from the interrupt it will take up to 200ms
+ * for the FW to get prepare for the reset, so we need to wait for that amount
+ * of time.
+ * After the reset is complete we need to then clear the GS1 register.
+ */
+void xe_gsc_wa_14015076503(struct xe_gt *gt, bool prep)
+{
+ u32 gs1_set = prep ? HECI_H_GS1_ER_PREP : 0;
+ u32 gs1_clr = prep ? 0 : HECI_H_GS1_ER_PREP;
+
+ /* WA only applies if the GSC is loaded */
+ if (!XE_WA(gt, 14015076503) || !gsc_fw_is_loaded(gt))
+ return;
+
+ xe_mmio_rmw32(gt, HECI_H_GS1(MTL_GSC_HECI2_BASE), gs1_clr, gs1_set);
+
+ if (prep) {
+ /* make sure the reset bit is clear when writing the CSR reg */
+ xe_mmio_rmw32(gt, HECI_H_CSR(MTL_GSC_HECI2_BASE),
+ HECI_H_CSR_RST, HECI_H_CSR_IG);
+ msleep(200);
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_gsc.h b/drivers/gpu/drm/xe/xe_gsc.h
new file mode 100644
index 000000000000..bc1ef7f31ea2
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gsc.h
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GSC_H_
+#define _XE_GSC_H_
+
+#include "xe_gsc_types.h"
+
+struct xe_gt;
+
+int xe_gsc_init(struct xe_gsc *gsc);
+int xe_gsc_init_post_hwconfig(struct xe_gsc *gsc);
+void xe_gsc_wait_for_worker_completion(struct xe_gsc *gsc);
+void xe_gsc_load_start(struct xe_gsc *gsc);
+
+void xe_gsc_wa_14015076503(struct xe_gt *gt, bool prep);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gsc_submit.c b/drivers/gpu/drm/xe/xe_gsc_submit.c
new file mode 100644
index 000000000000..8c5381e5913f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gsc_submit.c
@@ -0,0 +1,184 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_gsc_submit.h"
+
+#include "abi/gsc_command_header_abi.h"
+#include "xe_bb.h"
+#include "xe_exec_queue.h"
+#include "xe_gt_printk.h"
+#include "xe_gt_types.h"
+#include "xe_map.h"
+#include "xe_sched_job.h"
+#include "instructions/xe_gsc_commands.h"
+#include "regs/xe_gsc_regs.h"
+
+#define GSC_HDR_SIZE (sizeof(struct intel_gsc_mtl_header)) /* shorthand define */
+
+#define mtl_gsc_header_wr(xe_, map_, offset_, field_, val_) \
+ xe_map_wr_field(xe_, map_, offset_, struct intel_gsc_mtl_header, field_, val_)
+
+#define mtl_gsc_header_rd(xe_, map_, offset_, field_) \
+ xe_map_rd_field(xe_, map_, offset_, struct intel_gsc_mtl_header, field_)
+
+/*
+ * GSC FW allows us to define the host_session_handle as we see fit, as long
+ * as we use unique identifier for each user, with handle 0 being reserved for
+ * kernel usage.
+ * To be able to differentiate which client subsystem owns the given session, we
+ * include the client id in the top 8 bits of the handle.
+ */
+#define HOST_SESSION_CLIENT_MASK GENMASK_ULL(63, 56)
+
+static struct xe_gt *
+gsc_to_gt(struct xe_gsc *gsc)
+{
+ return container_of(gsc, struct xe_gt, uc.gsc);
+}
+
+/**
+ * xe_gsc_emit_header - write the MTL GSC header in memory
+ * @xe: the Xe device
+ * @map: the iosys map to write to
+ * @offset: offset from the start of the map at which to write the header
+ * @heci_client_id: client id identifying the type of command (see abi for values)
+ * @host_session_id: host session ID of the caller
+ * @payload_size: size of the payload that follows the header
+ *
+ * Returns: offset memory location following the header
+ */
+u32 xe_gsc_emit_header(struct xe_device *xe, struct iosys_map *map, u32 offset,
+ u8 heci_client_id, u64 host_session_id, u32 payload_size)
+{
+ xe_assert(xe, !(host_session_id & HOST_SESSION_CLIENT_MASK));
+
+ if (host_session_id)
+ host_session_id |= FIELD_PREP(HOST_SESSION_CLIENT_MASK, heci_client_id);
+
+ xe_map_memset(xe, map, offset, 0, GSC_HDR_SIZE);
+
+ mtl_gsc_header_wr(xe, map, offset, validity_marker, GSC_HECI_VALIDITY_MARKER);
+ mtl_gsc_header_wr(xe, map, offset, heci_client_id, heci_client_id);
+ mtl_gsc_header_wr(xe, map, offset, host_session_handle, host_session_id);
+ mtl_gsc_header_wr(xe, map, offset, header_version, MTL_GSC_HEADER_VERSION);
+ mtl_gsc_header_wr(xe, map, offset, message_size, payload_size + GSC_HDR_SIZE);
+
+ return offset + GSC_HDR_SIZE;
+};
+
+/**
+ * xe_gsc_check_and_update_pending - check the pending bit and update the input
+ * header with the retry handle from the output header
+ * @xe: the Xe device
+ * @in: the iosys map containing the input buffer
+ * @offset_in: offset within the iosys at which the input buffer is located
+ * @out: the iosys map containing the output buffer
+ * @offset_out: offset within the iosys at which the output buffer is located
+ *
+ * Returns: true if the pending bit was set, false otherwise
+ */
+bool xe_gsc_check_and_update_pending(struct xe_device *xe,
+ struct iosys_map *in, u32 offset_in,
+ struct iosys_map *out, u32 offset_out)
+{
+ if (mtl_gsc_header_rd(xe, out, offset_out, flags) & GSC_OUTFLAG_MSG_PENDING) {
+ u64 handle = mtl_gsc_header_rd(xe, out, offset_out, gsc_message_handle);
+
+ mtl_gsc_header_wr(xe, in, offset_in, gsc_message_handle, handle);
+
+ return true;
+ }
+
+ return false;
+}
+
+/**
+ * xe_gsc_read_out_header - reads and validates the output header and returns
+ * the offset of the reply following the header
+ * @xe: the Xe device
+ * @map: the iosys map containing the output buffer
+ * @offset: offset within the iosys at which the output buffer is located
+ * @min_payload_size: minimum size of the message excluding the gsc header
+ * @payload_offset: optional pointer to be set to the payload offset
+ *
+ * Returns: -errno value on failure, 0 otherwise
+ */
+int xe_gsc_read_out_header(struct xe_device *xe,
+ struct iosys_map *map, u32 offset,
+ u32 min_payload_size,
+ u32 *payload_offset)
+{
+ u32 marker = mtl_gsc_header_rd(xe, map, offset, validity_marker);
+ u32 size = mtl_gsc_header_rd(xe, map, offset, message_size);
+ u32 payload_size = size - GSC_HDR_SIZE;
+
+ if (marker != GSC_HECI_VALIDITY_MARKER)
+ return -EPROTO;
+
+ if (size < GSC_HDR_SIZE || payload_size < min_payload_size)
+ return -ENODATA;
+
+ if (payload_offset)
+ *payload_offset = offset + GSC_HDR_SIZE;
+
+ return 0;
+}
+
+/**
+ * xe_gsc_pkt_submit_kernel - submit a kernel heci pkt to the GSC
+ * @gsc: the GSC uC
+ * @addr_in: GGTT address of the message to send to the GSC
+ * @size_in: size of the message to send to the GSC
+ * @addr_out: GGTT address for the GSC to write the reply to
+ * @size_out: size of the memory reserved for the reply
+ */
+int xe_gsc_pkt_submit_kernel(struct xe_gsc *gsc, u64 addr_in, u32 size_in,
+ u64 addr_out, u32 size_out)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ struct xe_bb *bb;
+ struct xe_sched_job *job;
+ struct dma_fence *fence;
+ long timeout;
+
+ if (size_in < GSC_HDR_SIZE)
+ return -ENODATA;
+
+ if (size_out < GSC_HDR_SIZE)
+ return -ENOMEM;
+
+ bb = xe_bb_new(gt, 8, false);
+ if (IS_ERR(bb))
+ return PTR_ERR(bb);
+
+ bb->cs[bb->len++] = GSC_HECI_CMD_PKT;
+ bb->cs[bb->len++] = lower_32_bits(addr_in);
+ bb->cs[bb->len++] = upper_32_bits(addr_in);
+ bb->cs[bb->len++] = size_in;
+ bb->cs[bb->len++] = lower_32_bits(addr_out);
+ bb->cs[bb->len++] = upper_32_bits(addr_out);
+ bb->cs[bb->len++] = size_out;
+ bb->cs[bb->len++] = 0;
+
+ job = xe_bb_create_job(gsc->q, bb);
+ if (IS_ERR(job)) {
+ xe_bb_free(bb, NULL);
+ return PTR_ERR(job);
+ }
+
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ timeout = dma_fence_wait_timeout(fence, false, HZ);
+ dma_fence_put(fence);
+ xe_bb_free(bb, NULL);
+ if (timeout < 0)
+ return timeout;
+ else if (!timeout)
+ return -ETIME;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_gsc_submit.h b/drivers/gpu/drm/xe/xe_gsc_submit.h
new file mode 100644
index 000000000000..0801da5d446a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gsc_submit.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GSC_SUBMIT_H_
+#define _XE_GSC_SUBMIT_H_
+
+#include <linux/types.h>
+
+struct iosys_map;
+struct xe_device;
+struct xe_gsc;
+
+u32 xe_gsc_emit_header(struct xe_device *xe, struct iosys_map *map, u32 offset,
+ u8 heci_client_id, u64 host_session_id, u32 payload_size);
+
+bool xe_gsc_check_and_update_pending(struct xe_device *xe,
+ struct iosys_map *in, u32 offset_in,
+ struct iosys_map *out, u32 offset_out);
+
+int xe_gsc_read_out_header(struct xe_device *xe,
+ struct iosys_map *map, u32 offset,
+ u32 min_payload_size,
+ u32 *payload_offset);
+
+int xe_gsc_pkt_submit_kernel(struct xe_gsc *gsc, u64 addr_in, u32 size_in,
+ u64 addr_out, u32 size_out);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gsc_types.h b/drivers/gpu/drm/xe/xe_gsc_types.h
new file mode 100644
index 000000000000..57fefd66a7ea
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gsc_types.h
@@ -0,0 +1,39 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GSC_TYPES_H_
+#define _XE_GSC_TYPES_H_
+
+#include <linux/workqueue.h>
+
+#include "xe_uc_fw_types.h"
+
+struct xe_bo;
+struct xe_exec_queue;
+
+/**
+ * struct xe_gsc - GSC
+ */
+struct xe_gsc {
+ /** @fw: Generic uC firmware management */
+ struct xe_uc_fw fw;
+
+ /** @security_version: SVN found in the fetched blob */
+ u32 security_version;
+
+ /** @private: Private data for use by the GSC FW */
+ struct xe_bo *private;
+
+ /** @q: Default queue used for submissions to GSC FW */
+ struct xe_exec_queue *q;
+
+ /** @wq: workqueue to handle jobs for delayed load and proxy handling */
+ struct workqueue_struct *wq;
+
+ /** @work: delayed load and proxy handling work */
+ struct work_struct work;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
new file mode 100644
index 000000000000..3af2adec1295
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt.c
@@ -0,0 +1,778 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt.h"
+
+#include <linux/minmax.h>
+
+#include <drm/drm_managed.h>
+#include <drm/xe_drm.h>
+
+#include "instructions/xe_gfxpipe_commands.h"
+#include "instructions/xe_mi_commands.h"
+#include "regs/xe_gt_regs.h"
+#include "xe_assert.h"
+#include "xe_bb.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_execlist.h"
+#include "xe_force_wake.h"
+#include "xe_ggtt.h"
+#include "xe_gsc.h"
+#include "xe_gt_ccs_mode.h"
+#include "xe_gt_clock.h"
+#include "xe_gt_freq.h"
+#include "xe_gt_idle.h"
+#include "xe_gt_mcr.h"
+#include "xe_gt_pagefault.h"
+#include "xe_gt_printk.h"
+#include "xe_gt_sysfs.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_gt_topology.h"
+#include "xe_guc_exec_queue_types.h"
+#include "xe_guc_pc.h"
+#include "xe_hw_fence.h"
+#include "xe_hw_engine_class_sysfs.h"
+#include "xe_irq.h"
+#include "xe_lmtt.h"
+#include "xe_lrc.h"
+#include "xe_map.h"
+#include "xe_migrate.h"
+#include "xe_mmio.h"
+#include "xe_pat.h"
+#include "xe_mocs.h"
+#include "xe_reg_sr.h"
+#include "xe_ring_ops.h"
+#include "xe_sa.h"
+#include "xe_sched_job.h"
+#include "xe_sriov.h"
+#include "xe_tuning.h"
+#include "xe_uc.h"
+#include "xe_vm.h"
+#include "xe_wa.h"
+#include "xe_wopcm.h"
+
+struct xe_gt *xe_gt_alloc(struct xe_tile *tile)
+{
+ struct xe_gt *gt;
+
+ gt = drmm_kzalloc(&tile_to_xe(tile)->drm, sizeof(*gt), GFP_KERNEL);
+ if (!gt)
+ return ERR_PTR(-ENOMEM);
+
+ gt->tile = tile;
+ gt->ordered_wq = alloc_ordered_workqueue("gt-ordered-wq", 0);
+
+ return gt;
+}
+
+void xe_gt_sanitize(struct xe_gt *gt)
+{
+ /*
+ * FIXME: if xe_uc_sanitize is called here, on TGL driver will not
+ * reload
+ */
+ gt->uc.guc.submission_state.enabled = false;
+}
+
+static void gt_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_gt *gt = arg;
+ int i;
+
+ destroy_workqueue(gt->ordered_wq);
+
+ for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i)
+ xe_hw_fence_irq_finish(&gt->fence_irq[i]);
+}
+
+static void gt_reset_worker(struct work_struct *w);
+
+static int emit_nop_job(struct xe_gt *gt, struct xe_exec_queue *q)
+{
+ struct xe_sched_job *job;
+ struct xe_bb *bb;
+ struct dma_fence *fence;
+ long timeout;
+
+ bb = xe_bb_new(gt, 4, false);
+ if (IS_ERR(bb))
+ return PTR_ERR(bb);
+
+ job = xe_bb_create_job(q, bb);
+ if (IS_ERR(job)) {
+ xe_bb_free(bb, NULL);
+ return PTR_ERR(job);
+ }
+
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ timeout = dma_fence_wait_timeout(fence, false, HZ);
+ dma_fence_put(fence);
+ xe_bb_free(bb, NULL);
+ if (timeout < 0)
+ return timeout;
+ else if (!timeout)
+ return -ETIME;
+
+ return 0;
+}
+
+/*
+ * Convert back from encoded value to type-safe, only to be used when reg.mcr
+ * is true
+ */
+static struct xe_reg_mcr to_xe_reg_mcr(const struct xe_reg reg)
+{
+ return (const struct xe_reg_mcr){.__reg.raw = reg.raw };
+}
+
+static int emit_wa_job(struct xe_gt *gt, struct xe_exec_queue *q)
+{
+ struct xe_reg_sr *sr = &q->hwe->reg_lrc;
+ struct xe_reg_sr_entry *entry;
+ unsigned long idx;
+ struct xe_sched_job *job;
+ struct xe_bb *bb;
+ struct dma_fence *fence;
+ long timeout;
+ int count = 0;
+
+ if (q->hwe->class == XE_ENGINE_CLASS_RENDER)
+ /* Big enough to emit all of the context's 3DSTATE */
+ bb = xe_bb_new(gt, xe_lrc_size(gt_to_xe(gt), q->hwe->class), false);
+ else
+ /* Just pick a large BB size */
+ bb = xe_bb_new(gt, SZ_4K, false);
+
+ if (IS_ERR(bb))
+ return PTR_ERR(bb);
+
+ xa_for_each(&sr->xa, idx, entry)
+ ++count;
+
+ if (count) {
+ xe_gt_dbg(gt, "LRC WA %s save-restore batch\n", sr->name);
+
+ bb->cs[bb->len++] = MI_LOAD_REGISTER_IMM | MI_LRI_NUM_REGS(count);
+
+ xa_for_each(&sr->xa, idx, entry) {
+ struct xe_reg reg = entry->reg;
+ struct xe_reg_mcr reg_mcr = to_xe_reg_mcr(reg);
+ u32 val;
+
+ /*
+ * Skip reading the register if it's not really needed
+ */
+ if (reg.masked)
+ val = entry->clr_bits << 16;
+ else if (entry->clr_bits + 1)
+ val = (reg.mcr ?
+ xe_gt_mcr_unicast_read_any(gt, reg_mcr) :
+ xe_mmio_read32(gt, reg)) & (~entry->clr_bits);
+ else
+ val = 0;
+
+ val |= entry->set_bits;
+
+ bb->cs[bb->len++] = reg.addr;
+ bb->cs[bb->len++] = val;
+ xe_gt_dbg(gt, "REG[0x%x] = 0x%08x", reg.addr, val);
+ }
+ }
+
+ xe_lrc_emit_hwe_state_instructions(q, bb);
+
+ job = xe_bb_create_job(q, bb);
+ if (IS_ERR(job)) {
+ xe_bb_free(bb, NULL);
+ return PTR_ERR(job);
+ }
+
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ timeout = dma_fence_wait_timeout(fence, false, HZ);
+ dma_fence_put(fence);
+ xe_bb_free(bb, NULL);
+ if (timeout < 0)
+ return timeout;
+ else if (!timeout)
+ return -ETIME;
+
+ return 0;
+}
+
+int xe_gt_record_default_lrcs(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ int err = 0;
+
+ for_each_hw_engine(hwe, gt, id) {
+ struct xe_exec_queue *q, *nop_q;
+ void *default_lrc;
+
+ if (gt->default_lrc[hwe->class])
+ continue;
+
+ xe_reg_sr_init(&hwe->reg_lrc, hwe->name, xe);
+ xe_wa_process_lrc(hwe);
+ xe_hw_engine_setup_default_lrc_state(hwe);
+ xe_tuning_process_lrc(hwe);
+
+ default_lrc = drmm_kzalloc(&xe->drm,
+ xe_lrc_size(xe, hwe->class),
+ GFP_KERNEL);
+ if (!default_lrc)
+ return -ENOMEM;
+
+ q = xe_exec_queue_create(xe, NULL, BIT(hwe->logical_instance), 1,
+ hwe, EXEC_QUEUE_FLAG_KERNEL);
+ if (IS_ERR(q)) {
+ err = PTR_ERR(q);
+ xe_gt_err(gt, "hwe %s: xe_exec_queue_create failed (%pe)\n",
+ hwe->name, q);
+ return err;
+ }
+
+ /* Prime golden LRC with known good state */
+ err = emit_wa_job(gt, q);
+ if (err) {
+ xe_gt_err(gt, "hwe %s: emit_wa_job failed (%pe) guc_id=%u\n",
+ hwe->name, ERR_PTR(err), q->guc->id);
+ goto put_exec_queue;
+ }
+
+ nop_q = xe_exec_queue_create(xe, NULL, BIT(hwe->logical_instance),
+ 1, hwe, EXEC_QUEUE_FLAG_KERNEL);
+ if (IS_ERR(nop_q)) {
+ err = PTR_ERR(nop_q);
+ xe_gt_err(gt, "hwe %s: nop xe_exec_queue_create failed (%pe)\n",
+ hwe->name, nop_q);
+ goto put_exec_queue;
+ }
+
+ /* Switch to different LRC */
+ err = emit_nop_job(gt, nop_q);
+ if (err) {
+ xe_gt_err(gt, "hwe %s: nop emit_nop_job failed (%pe) guc_id=%u\n",
+ hwe->name, ERR_PTR(err), nop_q->guc->id);
+ goto put_nop_q;
+ }
+
+ /* Reload golden LRC to record the effect of any indirect W/A */
+ err = emit_nop_job(gt, q);
+ if (err) {
+ xe_gt_err(gt, "hwe %s: emit_nop_job failed (%pe) guc_id=%u\n",
+ hwe->name, ERR_PTR(err), q->guc->id);
+ goto put_nop_q;
+ }
+
+ xe_map_memcpy_from(xe, default_lrc,
+ &q->lrc[0].bo->vmap,
+ xe_lrc_pphwsp_offset(&q->lrc[0]),
+ xe_lrc_size(xe, hwe->class));
+
+ gt->default_lrc[hwe->class] = default_lrc;
+put_nop_q:
+ xe_exec_queue_put(nop_q);
+put_exec_queue:
+ xe_exec_queue_put(q);
+ if (err)
+ break;
+ }
+
+ return err;
+}
+
+int xe_gt_init_early(struct xe_gt *gt)
+{
+ int err;
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return err;
+
+ xe_gt_topology_init(gt);
+ xe_gt_mcr_init(gt);
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return err;
+
+ xe_reg_sr_init(&gt->reg_sr, "GT", gt_to_xe(gt));
+
+ err = xe_wa_init(gt);
+ if (err)
+ return err;
+
+ xe_wa_process_gt(gt);
+ xe_wa_process_oob(gt);
+ xe_tuning_process_gt(gt);
+
+ return 0;
+}
+
+static void dump_pat_on_error(struct xe_gt *gt)
+{
+ struct drm_printer p;
+ char prefix[32];
+
+ snprintf(prefix, sizeof(prefix), "[GT%u Error]", gt->info.id);
+ p = drm_debug_printer(prefix);
+
+ xe_pat_dump(gt, &p);
+}
+
+static int gt_fw_domain_init(struct xe_gt *gt)
+{
+ int err, i;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_hw_fence_irq;
+
+ xe_pat_init(gt);
+
+ if (!xe_gt_is_media_type(gt)) {
+ err = xe_ggtt_init(gt_to_tile(gt)->mem.ggtt);
+ if (err)
+ goto err_force_wake;
+ if (IS_SRIOV_PF(gt_to_xe(gt)))
+ xe_lmtt_init(&gt_to_tile(gt)->sriov.pf.lmtt);
+ }
+
+ err = xe_uc_init(&gt->uc);
+ if (err)
+ goto err_force_wake;
+
+ /* Raise GT freq to speed up HuC/GuC load */
+ xe_guc_pc_init_early(&gt->uc.guc.pc);
+
+ err = xe_uc_init_hwconfig(&gt->uc);
+ if (err)
+ goto err_force_wake;
+
+ xe_gt_idle_sysfs_init(&gt->gtidle);
+
+ /* XXX: Fake that we pull the engine mask from hwconfig blob */
+ gt->info.engine_mask = gt->info.__engine_mask;
+
+ /* Enable per hw engine IRQs */
+ xe_irq_enable_hwe(gt);
+
+ /* Rerun MCR init as we now have hw engine list */
+ xe_gt_mcr_init(gt);
+
+ err = xe_hw_engines_init_early(gt);
+ if (err)
+ goto err_force_wake;
+
+ err = xe_hw_engine_class_sysfs_init(gt);
+ if (err)
+ drm_warn(&gt_to_xe(gt)->drm,
+ "failed to register engines sysfs directory, err: %d\n",
+ err);
+
+ /* Initialize CCS mode sysfs after early initialization of HW engines */
+ xe_gt_ccs_mode_sysfs_init(gt);
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ XE_WARN_ON(err);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return 0;
+
+err_force_wake:
+ dump_pat_on_error(gt);
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_hw_fence_irq:
+ for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i)
+ xe_hw_fence_irq_finish(&gt->fence_irq[i]);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return err;
+}
+
+static int all_fw_domain_init(struct xe_gt *gt)
+{
+ int err, i;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_hw_fence_irq;
+
+ xe_gt_mcr_set_implicit_defaults(gt);
+ xe_reg_sr_apply_mmio(&gt->reg_sr, gt);
+
+ err = xe_gt_clock_init(gt);
+ if (err)
+ goto err_force_wake;
+
+ xe_mocs_init(gt);
+ err = xe_execlist_init(gt);
+ if (err)
+ goto err_force_wake;
+
+ err = xe_hw_engines_init(gt);
+ if (err)
+ goto err_force_wake;
+
+ err = xe_uc_init_post_hwconfig(&gt->uc);
+ if (err)
+ goto err_force_wake;
+
+ if (!xe_gt_is_media_type(gt)) {
+ /*
+ * USM has its only SA pool to non-block behind user operations
+ */
+ if (gt_to_xe(gt)->info.has_usm) {
+ gt->usm.bb_pool = xe_sa_bo_manager_init(gt_to_tile(gt), SZ_1M, 16);
+ if (IS_ERR(gt->usm.bb_pool)) {
+ err = PTR_ERR(gt->usm.bb_pool);
+ goto err_force_wake;
+ }
+ }
+ }
+
+ if (!xe_gt_is_media_type(gt)) {
+ struct xe_tile *tile = gt_to_tile(gt);
+
+ tile->migrate = xe_migrate_init(tile);
+ if (IS_ERR(tile->migrate)) {
+ err = PTR_ERR(tile->migrate);
+ goto err_force_wake;
+ }
+ }
+
+ err = xe_uc_init_hw(&gt->uc);
+ if (err)
+ goto err_force_wake;
+
+ /* Configure default CCS mode of 1 engine with all resources */
+ if (xe_gt_ccs_mode_enabled(gt)) {
+ gt->ccs_mode = 1;
+ xe_gt_apply_ccs_mode(gt);
+ }
+
+ if (IS_SRIOV_PF(gt_to_xe(gt)) && !xe_gt_is_media_type(gt))
+ xe_lmtt_init_hw(&gt_to_tile(gt)->sriov.pf.lmtt);
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ XE_WARN_ON(err);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return 0;
+
+err_force_wake:
+ xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+err_hw_fence_irq:
+ for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i)
+ xe_hw_fence_irq_finish(&gt->fence_irq[i]);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return err;
+}
+
+int xe_gt_init(struct xe_gt *gt)
+{
+ int err;
+ int i;
+
+ INIT_WORK(&gt->reset.worker, gt_reset_worker);
+
+ for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i) {
+ gt->ring_ops[i] = xe_ring_ops_get(gt, i);
+ xe_hw_fence_irq_init(&gt->fence_irq[i]);
+ }
+
+ err = xe_gt_tlb_invalidation_init(gt);
+ if (err)
+ return err;
+
+ err = xe_gt_pagefault_init(gt);
+ if (err)
+ return err;
+
+ xe_mocs_init_early(gt);
+
+ xe_gt_sysfs_init(gt);
+
+ err = gt_fw_domain_init(gt);
+ if (err)
+ return err;
+
+ xe_gt_freq_init(gt);
+
+ xe_force_wake_init_engines(gt, gt_to_fw(gt));
+
+ err = all_fw_domain_init(gt);
+ if (err)
+ return err;
+
+ err = drmm_add_action_or_reset(&gt_to_xe(gt)->drm, gt_fini, gt);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static int do_gt_reset(struct xe_gt *gt)
+{
+ int err;
+
+ xe_gsc_wa_14015076503(gt, true);
+
+ xe_mmio_write32(gt, GDRST, GRDOM_FULL);
+ err = xe_mmio_wait32(gt, GDRST, GRDOM_FULL, 0, 5000, NULL, false);
+ if (err)
+ xe_gt_err(gt, "failed to clear GRDOM_FULL (%pe)\n",
+ ERR_PTR(err));
+
+ xe_gsc_wa_14015076503(gt, false);
+
+ return err;
+}
+
+static int do_gt_restart(struct xe_gt *gt)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ int err;
+
+ xe_pat_init(gt);
+
+ xe_gt_mcr_set_implicit_defaults(gt);
+ xe_reg_sr_apply_mmio(&gt->reg_sr, gt);
+
+ err = xe_wopcm_init(&gt->uc.wopcm);
+ if (err)
+ return err;
+
+ for_each_hw_engine(hwe, gt, id)
+ xe_hw_engine_enable_ring(hwe);
+
+ err = xe_uc_sanitize_reset(&gt->uc);
+ if (err)
+ return err;
+
+ err = xe_uc_init_hw(&gt->uc);
+ if (err)
+ return err;
+
+ if (IS_SRIOV_PF(gt_to_xe(gt)) && !xe_gt_is_media_type(gt))
+ xe_lmtt_init_hw(&gt_to_tile(gt)->sriov.pf.lmtt);
+
+ xe_mocs_init(gt);
+ err = xe_uc_start(&gt->uc);
+ if (err)
+ return err;
+
+ for_each_hw_engine(hwe, gt, id) {
+ xe_reg_sr_apply_mmio(&hwe->reg_sr, gt);
+ xe_reg_sr_apply_whitelist(hwe);
+ }
+
+ /* Get CCS mode in sync between sw/hw */
+ xe_gt_apply_ccs_mode(gt);
+
+ return 0;
+}
+
+static int gt_reset(struct xe_gt *gt)
+{
+ int err;
+
+ /* We only support GT resets with GuC submission */
+ if (!xe_device_uc_enabled(gt_to_xe(gt)))
+ return -ENODEV;
+
+ xe_gt_info(gt, "reset started\n");
+
+ if (xe_fault_inject_gt_reset()) {
+ err = -ECANCELED;
+ goto err_fail;
+ }
+
+ xe_gt_sanitize(gt);
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_msg;
+
+ xe_uc_gucrc_disable(&gt->uc);
+ xe_uc_stop_prepare(&gt->uc);
+ xe_gt_pagefault_reset(gt);
+
+ err = xe_uc_stop(&gt->uc);
+ if (err)
+ goto err_out;
+
+ err = do_gt_reset(gt);
+ if (err)
+ goto err_out;
+
+ xe_gt_tlb_invalidation_reset(gt);
+
+ err = do_gt_restart(gt);
+ if (err)
+ goto err_out;
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ xe_device_mem_access_put(gt_to_xe(gt));
+ XE_WARN_ON(err);
+
+ xe_gt_info(gt, "reset done\n");
+
+ return 0;
+
+err_out:
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+err_msg:
+ XE_WARN_ON(xe_uc_start(&gt->uc));
+ xe_device_mem_access_put(gt_to_xe(gt));
+err_fail:
+ xe_gt_err(gt, "reset failed (%pe)\n", ERR_PTR(err));
+
+ gt_to_xe(gt)->needs_flr_on_fini = true;
+
+ return err;
+}
+
+static void gt_reset_worker(struct work_struct *w)
+{
+ struct xe_gt *gt = container_of(w, typeof(*gt), reset.worker);
+
+ gt_reset(gt);
+}
+
+void xe_gt_reset_async(struct xe_gt *gt)
+{
+ xe_gt_info(gt, "trying reset\n");
+
+ /* Don't do a reset while one is already in flight */
+ if (!xe_fault_inject_gt_reset() && xe_uc_reset_prepare(&gt->uc))
+ return;
+
+ xe_gt_info(gt, "reset queued\n");
+ queue_work(gt->ordered_wq, &gt->reset.worker);
+}
+
+void xe_gt_suspend_prepare(struct xe_gt *gt)
+{
+ xe_device_mem_access_get(gt_to_xe(gt));
+ XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+
+ xe_uc_stop_prepare(&gt->uc);
+
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+ xe_device_mem_access_put(gt_to_xe(gt));
+}
+
+int xe_gt_suspend(struct xe_gt *gt)
+{
+ int err;
+
+ xe_gt_sanitize(gt);
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_msg;
+
+ err = xe_uc_suspend(&gt->uc);
+ if (err)
+ goto err_force_wake;
+
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+ xe_device_mem_access_put(gt_to_xe(gt));
+ xe_gt_info(gt, "suspended\n");
+
+ return 0;
+
+err_force_wake:
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+err_msg:
+ xe_device_mem_access_put(gt_to_xe(gt));
+ xe_gt_err(gt, "suspend failed (%pe)\n", ERR_PTR(err));
+
+ return err;
+}
+
+int xe_gt_resume(struct xe_gt *gt)
+{
+ int err;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_msg;
+
+ err = do_gt_restart(gt);
+ if (err)
+ goto err_force_wake;
+
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+ xe_device_mem_access_put(gt_to_xe(gt));
+ xe_gt_info(gt, "resumed\n");
+
+ return 0;
+
+err_force_wake:
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+err_msg:
+ xe_device_mem_access_put(gt_to_xe(gt));
+ xe_gt_err(gt, "resume failed (%pe)\n", ERR_PTR(err));
+
+ return err;
+}
+
+struct xe_hw_engine *xe_gt_hw_engine(struct xe_gt *gt,
+ enum xe_engine_class class,
+ u16 instance, bool logical)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(hwe, gt, id)
+ if (hwe->class == class &&
+ ((!logical && hwe->instance == instance) ||
+ (logical && hwe->logical_instance == instance)))
+ return hwe;
+
+ return NULL;
+}
+
+struct xe_hw_engine *xe_gt_any_hw_engine_by_reset_domain(struct xe_gt *gt,
+ enum xe_engine_class class)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(hwe, gt, id) {
+ switch (class) {
+ case XE_ENGINE_CLASS_RENDER:
+ case XE_ENGINE_CLASS_COMPUTE:
+ if (hwe->class == XE_ENGINE_CLASS_RENDER ||
+ hwe->class == XE_ENGINE_CLASS_COMPUTE)
+ return hwe;
+ break;
+ default:
+ if (hwe->class == class)
+ return hwe;
+ }
+ }
+
+ return NULL;
+}
diff --git a/drivers/gpu/drm/xe/xe_gt.h b/drivers/gpu/drm/xe/xe_gt.h
new file mode 100644
index 000000000000..4486e083f5ef
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_H_
+#define _XE_GT_H_
+
+#include <drm/drm_util.h>
+
+#include "xe_device_types.h"
+#include "xe_hw_engine.h"
+
+#define for_each_hw_engine(hwe__, gt__, id__) \
+ for ((id__) = 0; (id__) < ARRAY_SIZE((gt__)->hw_engines); (id__)++) \
+ for_each_if(((hwe__) = (gt__)->hw_engines + (id__)) && \
+ xe_hw_engine_is_valid((hwe__)))
+
+#define CCS_MASK(gt) (((gt)->info.engine_mask & XE_HW_ENGINE_CCS_MASK) >> XE_HW_ENGINE_CCS0)
+
+#ifdef CONFIG_FAULT_INJECTION
+#include <linux/fault-inject.h> /* XXX: fault-inject.h is broken */
+extern struct fault_attr gt_reset_failure;
+static inline bool xe_fault_inject_gt_reset(void)
+{
+ return should_fail(&gt_reset_failure, 1);
+}
+#else
+static inline bool xe_fault_inject_gt_reset(void)
+{
+ return false;
+}
+#endif
+
+struct xe_gt *xe_gt_alloc(struct xe_tile *tile);
+int xe_gt_init_early(struct xe_gt *gt);
+int xe_gt_init(struct xe_gt *gt);
+int xe_gt_record_default_lrcs(struct xe_gt *gt);
+void xe_gt_suspend_prepare(struct xe_gt *gt);
+int xe_gt_suspend(struct xe_gt *gt);
+int xe_gt_resume(struct xe_gt *gt);
+void xe_gt_reset_async(struct xe_gt *gt);
+void xe_gt_sanitize(struct xe_gt *gt);
+
+/**
+ * xe_gt_any_hw_engine_by_reset_domain - scan the list of engines and return the
+ * first that matches the same reset domain as @class
+ * @gt: GT structure
+ * @class: hw engine class to lookup
+ */
+struct xe_hw_engine *
+xe_gt_any_hw_engine_by_reset_domain(struct xe_gt *gt, enum xe_engine_class class);
+
+struct xe_hw_engine *xe_gt_hw_engine(struct xe_gt *gt,
+ enum xe_engine_class class,
+ u16 instance,
+ bool logical);
+
+static inline bool xe_gt_is_media_type(struct xe_gt *gt)
+{
+ return gt->info.type == XE_GT_TYPE_MEDIA;
+}
+
+static inline bool xe_gt_is_usm_hwe(struct xe_gt *gt, struct xe_hw_engine *hwe)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ return xe->info.has_usm && hwe->class == XE_ENGINE_CLASS_COPY &&
+ hwe->instance == gt->usm.reserved_bcs_instance;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_ccs_mode.c b/drivers/gpu/drm/xe/xe_gt_ccs_mode.c
new file mode 100644
index 000000000000..529fc286cd06
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_ccs_mode.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_gt_regs.h"
+#include "xe_assert.h"
+#include "xe_gt.h"
+#include "xe_gt_ccs_mode.h"
+#include "xe_gt_sysfs.h"
+#include "xe_mmio.h"
+
+static void __xe_gt_apply_ccs_mode(struct xe_gt *gt, u32 num_engines)
+{
+ u32 mode = CCS_MODE_CSLICE_0_3_MASK; /* disable all by default */
+ int num_slices = hweight32(CCS_MASK(gt));
+ struct xe_device *xe = gt_to_xe(gt);
+ int width, cslice = 0;
+ u32 config = 0;
+
+ xe_assert(xe, xe_gt_ccs_mode_enabled(gt));
+
+ xe_assert(xe, num_engines && num_engines <= num_slices);
+ xe_assert(xe, !(num_slices % num_engines));
+
+ /*
+ * Loop over all available slices and assign each a user engine.
+ * For example, if there are four compute slices available, the
+ * assignment of compute slices to compute engines would be,
+ *
+ * With 1 engine (ccs0):
+ * slice 0, 1, 2, 3: ccs0
+ *
+ * With 2 engines (ccs0, ccs1):
+ * slice 0, 2: ccs0
+ * slice 1, 3: ccs1
+ *
+ * With 4 engines (ccs0, ccs1, ccs2, ccs3):
+ * slice 0: ccs0
+ * slice 1: ccs1
+ * slice 2: ccs2
+ * slice 3: ccs3
+ */
+ for (width = num_slices / num_engines; width; width--) {
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(hwe, gt, id) {
+ if (hwe->class != XE_ENGINE_CLASS_COMPUTE)
+ continue;
+
+ if (hwe->logical_instance >= num_engines)
+ break;
+
+ config |= BIT(hwe->instance) << XE_HW_ENGINE_CCS0;
+
+ /* If a slice is fused off, leave disabled */
+ while ((CCS_MASK(gt) & BIT(cslice)) == 0)
+ cslice++;
+
+ mode &= ~CCS_MODE_CSLICE(cslice, CCS_MODE_CSLICE_MASK);
+ mode |= CCS_MODE_CSLICE(cslice, hwe->instance);
+ cslice++;
+ }
+ }
+
+ xe_mmio_write32(gt, CCS_MODE, mode);
+
+ xe_gt_info(gt, "CCS_MODE=%x config:%08x, num_engines:%d, num_slices:%d\n",
+ mode, config, num_engines, num_slices);
+}
+
+void xe_gt_apply_ccs_mode(struct xe_gt *gt)
+{
+ if (!gt->ccs_mode)
+ return;
+
+ __xe_gt_apply_ccs_mode(gt, gt->ccs_mode);
+}
+
+static ssize_t
+num_cslices_show(struct device *kdev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
+
+ return sysfs_emit(buf, "%u\n", hweight32(CCS_MASK(gt)));
+}
+
+static DEVICE_ATTR_RO(num_cslices);
+
+static ssize_t
+ccs_mode_show(struct device *kdev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
+
+ return sysfs_emit(buf, "%u\n", gt->ccs_mode);
+}
+
+static ssize_t
+ccs_mode_store(struct device *kdev, struct device_attribute *attr,
+ const char *buff, size_t count)
+{
+ struct xe_gt *gt = kobj_to_gt(&kdev->kobj);
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 num_engines, num_slices;
+ int ret;
+
+ ret = kstrtou32(buff, 0, &num_engines);
+ if (ret)
+ return ret;
+
+ /*
+ * Ensure number of engines specified is valid and there is an
+ * exact multiple of engines for slices.
+ */
+ num_slices = hweight32(CCS_MASK(gt));
+ if (!num_engines || num_engines > num_slices || num_slices % num_engines) {
+ xe_gt_dbg(gt, "Invalid compute config, %d engines %d slices\n",
+ num_engines, num_slices);
+ return -EINVAL;
+ }
+
+ /* CCS mode can only be updated when there are no drm clients */
+ spin_lock(&xe->clients.lock);
+ if (xe->clients.count) {
+ spin_unlock(&xe->clients.lock);
+ return -EBUSY;
+ }
+
+ if (gt->ccs_mode != num_engines) {
+ xe_gt_info(gt, "Setting compute mode to %d\n", num_engines);
+ gt->ccs_mode = num_engines;
+ xe_gt_reset_async(gt);
+ }
+
+ spin_unlock(&xe->clients.lock);
+
+ return count;
+}
+
+static DEVICE_ATTR_RW(ccs_mode);
+
+static const struct attribute *gt_ccs_mode_attrs[] = {
+ &dev_attr_ccs_mode.attr,
+ &dev_attr_num_cslices.attr,
+ NULL,
+};
+
+static void xe_gt_ccs_mode_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_gt *gt = arg;
+
+ sysfs_remove_files(gt->sysfs, gt_ccs_mode_attrs);
+}
+
+/**
+ * xe_gt_ccs_mode_sysfs_init - Initialize CCS mode sysfs interfaces
+ * @gt: GT structure
+ *
+ * Through a per-gt 'ccs_mode' sysfs interface, the user can enable a fixed
+ * number of compute hardware engines to which the available compute slices
+ * are to be allocated. This user configuration change triggers a gt reset
+ * and it is expected that there are no open drm clients while doing so.
+ * The number of available compute slices is exposed to user through a per-gt
+ * 'num_cslices' sysfs interface.
+ */
+void xe_gt_ccs_mode_sysfs_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int err;
+
+ if (!xe_gt_ccs_mode_enabled(gt))
+ return;
+
+ err = sysfs_create_files(gt->sysfs, gt_ccs_mode_attrs);
+ if (err) {
+ drm_warn(&xe->drm, "Sysfs creation for ccs_mode failed err: %d\n", err);
+ return;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, xe_gt_ccs_mode_sysfs_fini, gt);
+ if (err) {
+ sysfs_remove_files(gt->sysfs, gt_ccs_mode_attrs);
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_ccs_mode.h b/drivers/gpu/drm/xe/xe_gt_ccs_mode.h
new file mode 100644
index 000000000000..f39975aaaab0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_ccs_mode.h
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_CCS_MODE_H_
+#define _XE_GT_CCS_MODE_H_
+
+#include "xe_device_types.h"
+#include "xe_gt.h"
+#include "xe_gt_types.h"
+#include "xe_platform_types.h"
+
+void xe_gt_apply_ccs_mode(struct xe_gt *gt);
+void xe_gt_ccs_mode_sysfs_init(struct xe_gt *gt);
+
+static inline bool xe_gt_ccs_mode_enabled(const struct xe_gt *gt)
+{
+ /* Check if there are more than one compute engines available */
+ return hweight32(CCS_MASK(gt)) > 1;
+}
+
+#endif
+
diff --git a/drivers/gpu/drm/xe/xe_gt_clock.c b/drivers/gpu/drm/xe/xe_gt_clock.c
new file mode 100644
index 000000000000..937054e31d72
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_clock.c
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_clock.h"
+
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+
+static u32 read_reference_ts_freq(struct xe_gt *gt)
+{
+ u32 ts_override = xe_mmio_read32(gt, TIMESTAMP_OVERRIDE);
+ u32 base_freq, frac_freq;
+
+ base_freq = REG_FIELD_GET(TIMESTAMP_OVERRIDE_US_COUNTER_DIVIDER_MASK,
+ ts_override) + 1;
+ base_freq *= 1000000;
+
+ frac_freq = REG_FIELD_GET(TIMESTAMP_OVERRIDE_US_COUNTER_DENOMINATOR_MASK,
+ ts_override);
+ frac_freq = 1000000 / (frac_freq + 1);
+
+ return base_freq + frac_freq;
+}
+
+static u32 get_crystal_clock_freq(u32 rpm_config_reg)
+{
+ const u32 f19_2_mhz = 19200000;
+ const u32 f24_mhz = 24000000;
+ const u32 f25_mhz = 25000000;
+ const u32 f38_4_mhz = 38400000;
+ u32 crystal_clock = REG_FIELD_GET(RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_MASK,
+ rpm_config_reg);
+
+ switch (crystal_clock) {
+ case RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_24_MHZ:
+ return f24_mhz;
+ case RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_19_2_MHZ:
+ return f19_2_mhz;
+ case RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_38_4_MHZ:
+ return f38_4_mhz;
+ case RPM_CONFIG0_CRYSTAL_CLOCK_FREQ_25_MHZ:
+ return f25_mhz;
+ default:
+ XE_WARN_ON("NOT_POSSIBLE");
+ return 0;
+ }
+}
+
+int xe_gt_clock_init(struct xe_gt *gt)
+{
+ u32 ctc_reg = xe_mmio_read32(gt, CTC_MODE);
+ u32 freq = 0;
+
+ /* Assuming gen11+ so assert this assumption is correct */
+ xe_gt_assert(gt, GRAPHICS_VER(gt_to_xe(gt)) >= 11);
+
+ if (ctc_reg & CTC_SOURCE_DIVIDE_LOGIC) {
+ freq = read_reference_ts_freq(gt);
+ } else {
+ u32 c0 = xe_mmio_read32(gt, RPM_CONFIG0);
+
+ freq = get_crystal_clock_freq(c0);
+
+ /*
+ * Now figure out how the command stream's timestamp
+ * register increments from this frequency (it might
+ * increment only every few clock cycle).
+ */
+ freq >>= 3 - REG_FIELD_GET(RPM_CONFIG0_CTC_SHIFT_PARAMETER_MASK, c0);
+ }
+
+ gt->info.reference_clock = freq;
+ return 0;
+}
+
+u64 xe_gt_clock_cycles_to_ns(const struct xe_gt *gt, u64 count)
+{
+ return DIV_ROUND_CLOSEST_ULL(count * NSEC_PER_SEC, gt->info.reference_clock);
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_clock.h b/drivers/gpu/drm/xe/xe_gt_clock.h
new file mode 100644
index 000000000000..aa162722f859
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_clock.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_CLOCK_H_
+#define _XE_GT_CLOCK_H_
+
+#include <linux/types.h>
+
+struct xe_gt;
+
+int xe_gt_clock_init(struct xe_gt *gt);
+u64 xe_gt_clock_cycles_to_ns(const struct xe_gt *gt, u64 count);
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_debugfs.c b/drivers/gpu/drm/xe/xe_gt_debugfs.c
new file mode 100644
index 000000000000..c4b67cf09f8f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_debugfs.c
@@ -0,0 +1,249 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_debugfs.h"
+
+#include <drm/drm_debugfs.h>
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_force_wake.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_gt_topology.h"
+#include "xe_hw_engine.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_pat.h"
+#include "xe_reg_sr.h"
+#include "xe_reg_whitelist.h"
+#include "xe_uc_debugfs.h"
+#include "xe_wa.h"
+
+static struct xe_gt *node_to_gt(struct drm_info_node *node)
+{
+ return node->info_ent->data;
+}
+
+static int hw_engines(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct drm_printer p = drm_seq_file_printer(m);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ int err;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (err) {
+ xe_device_mem_access_put(xe);
+ return err;
+ }
+
+ for_each_hw_engine(hwe, gt, id)
+ xe_hw_engine_print(hwe, &p);
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ xe_device_mem_access_put(xe);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static int force_reset(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+
+ xe_gt_reset_async(gt);
+
+ return 0;
+}
+
+static int sa_info(struct seq_file *m, void *data)
+{
+ struct xe_tile *tile = gt_to_tile(node_to_gt(m->private));
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ drm_suballoc_dump_debug_info(&tile->mem.kernel_bb_pool->base, &p,
+ tile->mem.kernel_bb_pool->gpu_addr);
+
+ return 0;
+}
+
+static int topology(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_gt_topology_dump(gt, &p);
+
+ return 0;
+}
+
+static int steering(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_gt_mcr_steering_dump(gt, &p);
+
+ return 0;
+}
+
+static int ggtt(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ return xe_ggtt_dump(gt_to_tile(gt)->mem.ggtt, &p);
+}
+
+static int register_save_restore(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ xe_reg_sr_dump(&gt->reg_sr, &p);
+ drm_printf(&p, "\n");
+
+ drm_printf(&p, "Engine\n");
+ for_each_hw_engine(hwe, gt, id)
+ xe_reg_sr_dump(&hwe->reg_sr, &p);
+ drm_printf(&p, "\n");
+
+ drm_printf(&p, "LRC\n");
+ for_each_hw_engine(hwe, gt, id)
+ xe_reg_sr_dump(&hwe->reg_lrc, &p);
+ drm_printf(&p, "\n");
+
+ drm_printf(&p, "Whitelist\n");
+ for_each_hw_engine(hwe, gt, id)
+ xe_reg_whitelist_dump(&hwe->reg_whitelist, &p);
+
+ return 0;
+}
+
+static int workarounds(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_wa_dump(gt, &p);
+
+ return 0;
+}
+
+static int pat(struct seq_file *m, void *data)
+{
+ struct xe_gt *gt = node_to_gt(m->private);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_pat_dump(gt, &p);
+
+ return 0;
+}
+
+static int rcs_default_lrc(struct seq_file *m, void *data)
+{
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_lrc_dump_default(&p, node_to_gt(m->private), XE_ENGINE_CLASS_RENDER);
+ return 0;
+}
+
+static int ccs_default_lrc(struct seq_file *m, void *data)
+{
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_lrc_dump_default(&p, node_to_gt(m->private), XE_ENGINE_CLASS_COMPUTE);
+ return 0;
+}
+
+static int bcs_default_lrc(struct seq_file *m, void *data)
+{
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_lrc_dump_default(&p, node_to_gt(m->private), XE_ENGINE_CLASS_COPY);
+ return 0;
+}
+
+static int vcs_default_lrc(struct seq_file *m, void *data)
+{
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_lrc_dump_default(&p, node_to_gt(m->private), XE_ENGINE_CLASS_VIDEO_DECODE);
+ return 0;
+}
+
+static int vecs_default_lrc(struct seq_file *m, void *data)
+{
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_lrc_dump_default(&p, node_to_gt(m->private), XE_ENGINE_CLASS_VIDEO_ENHANCE);
+ return 0;
+}
+
+static const struct drm_info_list debugfs_list[] = {
+ {"hw_engines", hw_engines, 0},
+ {"force_reset", force_reset, 0},
+ {"sa_info", sa_info, 0},
+ {"topology", topology, 0},
+ {"steering", steering, 0},
+ {"ggtt", ggtt, 0},
+ {"register-save-restore", register_save_restore, 0},
+ {"workarounds", workarounds, 0},
+ {"pat", pat, 0},
+ {"default_lrc_rcs", rcs_default_lrc},
+ {"default_lrc_ccs", ccs_default_lrc},
+ {"default_lrc_bcs", bcs_default_lrc},
+ {"default_lrc_vcs", vcs_default_lrc},
+ {"default_lrc_vecs", vecs_default_lrc},
+};
+
+void xe_gt_debugfs_register(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct drm_minor *minor = gt_to_xe(gt)->drm.primary;
+ struct dentry *root;
+ struct drm_info_list *local;
+ char name[8];
+ int i;
+
+ xe_gt_assert(gt, minor->debugfs_root);
+
+ sprintf(name, "gt%d", gt->info.id);
+ root = debugfs_create_dir(name, minor->debugfs_root);
+ if (IS_ERR(root)) {
+ drm_warn(&xe->drm, "Create GT directory failed");
+ return;
+ }
+
+ /*
+ * Allocate local copy as we need to pass in the GT to the debugfs
+ * entry and drm_debugfs_create_files just references the drm_info_list
+ * passed in (e.g. can't define this on the stack).
+ */
+#define DEBUGFS_SIZE (ARRAY_SIZE(debugfs_list) * sizeof(struct drm_info_list))
+ local = drmm_kmalloc(&xe->drm, DEBUGFS_SIZE, GFP_KERNEL);
+ if (!local)
+ return;
+
+ memcpy(local, debugfs_list, DEBUGFS_SIZE);
+#undef DEBUGFS_SIZE
+
+ for (i = 0; i < ARRAY_SIZE(debugfs_list); ++i)
+ local[i].data = gt;
+
+ drm_debugfs_create_files(local,
+ ARRAY_SIZE(debugfs_list),
+ root, minor);
+
+ xe_uc_debugfs_register(&gt->uc, root);
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_debugfs.h b/drivers/gpu/drm/xe/xe_gt_debugfs.h
new file mode 100644
index 000000000000..5a329f118a57
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_debugfs.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_DEBUGFS_H_
+#define _XE_GT_DEBUGFS_H_
+
+struct xe_gt;
+
+void xe_gt_debugfs_register(struct xe_gt *gt);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_freq.c b/drivers/gpu/drm/xe/xe_gt_freq.c
new file mode 100644
index 000000000000..3adfa6686e7c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_freq.c
@@ -0,0 +1,219 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_gt_freq.h"
+
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+
+#include <drm/drm_managed.h>
+#include <drm/drm_print.h>
+
+#include "xe_device_types.h"
+#include "xe_gt_sysfs.h"
+#include "xe_gt_throttle_sysfs.h"
+#include "xe_guc_pc.h"
+
+/**
+ * DOC: Xe GT Frequency Management
+ *
+ * This component is responsible for the raw GT frequency management, including
+ * the sysfs API.
+ *
+ * Underneath, Xe enables GuC SLPC automated frequency management. GuC is then
+ * allowed to request PCODE any frequency between the Minimum and the Maximum
+ * selected by this component. Furthermore, it is important to highlight that
+ * PCODE is the ultimate decision maker of the actual running frequency, based
+ * on thermal and other running conditions.
+ *
+ * Xe's Freq provides a sysfs API for frequency management:
+ *
+ * device/tile#/gt#/freq0/<item>_freq *read-only* files:
+ * - act_freq: The actual resolved frequency decided by PCODE.
+ * - cur_freq: The current one requested by GuC PC to the PCODE.
+ * - rpn_freq: The Render Performance (RP) N level, which is the minimal one.
+ * - rpe_freq: The Render Performance (RP) E level, which is the efficient one.
+ * - rp0_freq: The Render Performance (RP) 0 level, which is the maximum one.
+ *
+ * device/tile#/gt#/freq0/<item>_freq *read-write* files:
+ * - min_freq: Min frequency request.
+ * - max_freq: Max frequency request.
+ * If max <= min, then freq_min becomes a fixed frequency request.
+ */
+
+static struct xe_guc_pc *
+dev_to_pc(struct device *dev)
+{
+ return &kobj_to_gt(dev->kobj.parent)->uc.guc.pc;
+}
+
+static ssize_t act_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+
+ return sysfs_emit(buf, "%d\n", xe_guc_pc_get_act_freq(pc));
+}
+static DEVICE_ATTR_RO(act_freq);
+
+static ssize_t cur_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+ u32 freq;
+ ssize_t ret;
+
+ ret = xe_guc_pc_get_cur_freq(pc, &freq);
+ if (ret)
+ return ret;
+
+ return sysfs_emit(buf, "%d\n", freq);
+}
+static DEVICE_ATTR_RO(cur_freq);
+
+static ssize_t rp0_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+
+ return sysfs_emit(buf, "%d\n", xe_guc_pc_get_rp0_freq(pc));
+}
+static DEVICE_ATTR_RO(rp0_freq);
+
+static ssize_t rpe_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+
+ return sysfs_emit(buf, "%d\n", xe_guc_pc_get_rpe_freq(pc));
+}
+static DEVICE_ATTR_RO(rpe_freq);
+
+static ssize_t rpn_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+
+ return sysfs_emit(buf, "%d\n", xe_guc_pc_get_rpn_freq(pc));
+}
+static DEVICE_ATTR_RO(rpn_freq);
+
+static ssize_t min_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+ u32 freq;
+ ssize_t ret;
+
+ ret = xe_guc_pc_get_min_freq(pc, &freq);
+ if (ret)
+ return ret;
+
+ return sysfs_emit(buf, "%d\n", freq);
+}
+
+static ssize_t min_freq_store(struct device *dev, struct device_attribute *attr,
+ const char *buff, size_t count)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+ u32 freq;
+ ssize_t ret;
+
+ ret = kstrtou32(buff, 0, &freq);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_pc_set_min_freq(pc, freq);
+ if (ret)
+ return ret;
+
+ return count;
+}
+static DEVICE_ATTR_RW(min_freq);
+
+static ssize_t max_freq_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+ u32 freq;
+ ssize_t ret;
+
+ ret = xe_guc_pc_get_max_freq(pc, &freq);
+ if (ret)
+ return ret;
+
+ return sysfs_emit(buf, "%d\n", freq);
+}
+
+static ssize_t max_freq_store(struct device *dev, struct device_attribute *attr,
+ const char *buff, size_t count)
+{
+ struct xe_guc_pc *pc = dev_to_pc(dev);
+ u32 freq;
+ ssize_t ret;
+
+ ret = kstrtou32(buff, 0, &freq);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_pc_set_max_freq(pc, freq);
+ if (ret)
+ return ret;
+
+ return count;
+}
+static DEVICE_ATTR_RW(max_freq);
+
+static const struct attribute *freq_attrs[] = {
+ &dev_attr_act_freq.attr,
+ &dev_attr_cur_freq.attr,
+ &dev_attr_rp0_freq.attr,
+ &dev_attr_rpe_freq.attr,
+ &dev_attr_rpn_freq.attr,
+ &dev_attr_min_freq.attr,
+ &dev_attr_max_freq.attr,
+ NULL
+};
+
+static void freq_fini(struct drm_device *drm, void *arg)
+{
+ struct kobject *kobj = arg;
+
+ sysfs_remove_files(kobj, freq_attrs);
+ kobject_put(kobj);
+}
+
+/**
+ * xe_gt_freq_init - Initialize Xe Freq component
+ * @gt: Xe GT object
+ *
+ * It needs to be initialized after GT Sysfs and GuC PC components are ready.
+ */
+void xe_gt_freq_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int err;
+
+ gt->freq = kobject_create_and_add("freq0", gt->sysfs);
+ if (!gt->freq) {
+ drm_warn(&xe->drm, "failed to add freq0 directory to %s\n",
+ kobject_name(gt->sysfs));
+ return;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, freq_fini, gt->freq);
+ if (err) {
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+ return;
+ }
+
+ err = sysfs_create_files(gt->freq, freq_attrs);
+ if (err)
+ drm_warn(&xe->drm, "failed to add freq attrs to %s, err: %d\n",
+ kobject_name(gt->freq), err);
+
+ xe_gt_throttle_sysfs_init(gt);
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_freq.h b/drivers/gpu/drm/xe/xe_gt_freq.h
new file mode 100644
index 000000000000..f3fe3c90491a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_freq.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_FREQ_H_
+#define _XE_GT_FREQ_H_
+
+struct xe_gt;
+
+void xe_gt_freq_init(struct xe_gt *gt);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_idle.c b/drivers/gpu/drm/xe/xe_gt_idle.c
new file mode 100644
index 000000000000..9358f7336889
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_idle.c
@@ -0,0 +1,192 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_idle.h"
+#include "xe_gt_sysfs.h"
+#include "xe_guc_pc.h"
+#include "regs/xe_gt_regs.h"
+#include "xe_mmio.h"
+
+/**
+ * DOC: Xe GT Idle
+ *
+ * Contains functions that init GT idle features like C6
+ *
+ * device/gt#/gtidle/name - name of the state
+ * device/gt#/gtidle/idle_residency_ms - Provides residency of the idle state in ms
+ * device/gt#/gtidle/idle_status - Provides current idle state
+ */
+
+static struct xe_gt_idle *dev_to_gtidle(struct device *dev)
+{
+ struct kobject *kobj = &dev->kobj;
+
+ return &kobj_to_gt(kobj->parent)->gtidle;
+}
+
+static struct xe_gt *gtidle_to_gt(struct xe_gt_idle *gtidle)
+{
+ return container_of(gtidle, struct xe_gt, gtidle);
+}
+
+static struct xe_guc_pc *gtidle_to_pc(struct xe_gt_idle *gtidle)
+{
+ return &gtidle_to_gt(gtidle)->uc.guc.pc;
+}
+
+static const char *gt_idle_state_to_string(enum xe_gt_idle_state state)
+{
+ switch (state) {
+ case GT_IDLE_C0:
+ return "gt-c0";
+ case GT_IDLE_C6:
+ return "gt-c6";
+ default:
+ return "unknown";
+ }
+}
+
+static u64 get_residency_ms(struct xe_gt_idle *gtidle, u64 cur_residency)
+{
+ u64 delta, overflow_residency, prev_residency;
+
+ overflow_residency = BIT_ULL(32);
+
+ /*
+ * Counter wrap handling
+ * Store previous hw counter values for counter wrap-around handling
+ * Relying on sufficient frequency of queries otherwise counters can still wrap.
+ */
+ prev_residency = gtidle->prev_residency;
+ gtidle->prev_residency = cur_residency;
+
+ /* delta */
+ if (cur_residency >= prev_residency)
+ delta = cur_residency - prev_residency;
+ else
+ delta = cur_residency + (overflow_residency - prev_residency);
+
+ /* Add delta to extended raw driver copy of idle residency */
+ cur_residency = gtidle->cur_residency + delta;
+ gtidle->cur_residency = cur_residency;
+
+ /* residency multiplier in ns, convert to ms */
+ cur_residency = mul_u64_u32_div(cur_residency, gtidle->residency_multiplier, 1e6);
+
+ return cur_residency;
+}
+
+static ssize_t name_show(struct device *dev,
+ struct device_attribute *attr, char *buff)
+{
+ struct xe_gt_idle *gtidle = dev_to_gtidle(dev);
+
+ return sysfs_emit(buff, "%s\n", gtidle->name);
+}
+static DEVICE_ATTR_RO(name);
+
+static ssize_t idle_status_show(struct device *dev,
+ struct device_attribute *attr, char *buff)
+{
+ struct xe_gt_idle *gtidle = dev_to_gtidle(dev);
+ struct xe_guc_pc *pc = gtidle_to_pc(gtidle);
+ enum xe_gt_idle_state state;
+
+ state = gtidle->idle_status(pc);
+
+ return sysfs_emit(buff, "%s\n", gt_idle_state_to_string(state));
+}
+static DEVICE_ATTR_RO(idle_status);
+
+static ssize_t idle_residency_ms_show(struct device *dev,
+ struct device_attribute *attr, char *buff)
+{
+ struct xe_gt_idle *gtidle = dev_to_gtidle(dev);
+ struct xe_guc_pc *pc = gtidle_to_pc(gtidle);
+ u64 residency;
+
+ residency = gtidle->idle_residency(pc);
+ return sysfs_emit(buff, "%llu\n", get_residency_ms(gtidle, residency));
+}
+static DEVICE_ATTR_RO(idle_residency_ms);
+
+static const struct attribute *gt_idle_attrs[] = {
+ &dev_attr_name.attr,
+ &dev_attr_idle_status.attr,
+ &dev_attr_idle_residency_ms.attr,
+ NULL,
+};
+
+static void gt_idle_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct kobject *kobj = arg;
+
+ sysfs_remove_files(kobj, gt_idle_attrs);
+ kobject_put(kobj);
+}
+
+void xe_gt_idle_sysfs_init(struct xe_gt_idle *gtidle)
+{
+ struct xe_gt *gt = gtidle_to_gt(gtidle);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct kobject *kobj;
+ int err;
+
+ kobj = kobject_create_and_add("gtidle", gt->sysfs);
+ if (!kobj) {
+ drm_warn(&xe->drm, "%s failed, err: %d\n", __func__, -ENOMEM);
+ return;
+ }
+
+ if (xe_gt_is_media_type(gt)) {
+ sprintf(gtidle->name, "gt%d-mc\n", gt->info.id);
+ gtidle->idle_residency = xe_guc_pc_mc6_residency;
+ } else {
+ sprintf(gtidle->name, "gt%d-rc\n", gt->info.id);
+ gtidle->idle_residency = xe_guc_pc_rc6_residency;
+ }
+
+ /* Multiplier for Residency counter in units of 1.28us */
+ gtidle->residency_multiplier = 1280;
+ gtidle->idle_status = xe_guc_pc_c_status;
+
+ err = sysfs_create_files(kobj, gt_idle_attrs);
+ if (err) {
+ kobject_put(kobj);
+ drm_warn(&xe->drm, "failed to register gtidle sysfs, err: %d\n", err);
+ return;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, gt_idle_sysfs_fini, kobj);
+ if (err)
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+}
+
+void xe_gt_idle_enable_c6(struct xe_gt *gt)
+{
+ xe_device_assert_mem_access(gt_to_xe(gt));
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ /* Units of 1280 ns for a total of 5s */
+ xe_mmio_write32(gt, RC_IDLE_HYSTERSIS, 0x3B9ACA);
+ /* Enable RC6 */
+ xe_mmio_write32(gt, RC_CONTROL,
+ RC_CTL_HW_ENABLE | RC_CTL_TO_MODE | RC_CTL_RC6_ENABLE);
+}
+
+void xe_gt_idle_disable_c6(struct xe_gt *gt)
+{
+ xe_device_assert_mem_access(gt_to_xe(gt));
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+
+ xe_mmio_write32(gt, PG_ENABLE, 0);
+ xe_mmio_write32(gt, RC_CONTROL, 0);
+ xe_mmio_write32(gt, RC_STATE, 0);
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_idle.h b/drivers/gpu/drm/xe/xe_gt_idle.h
new file mode 100644
index 000000000000..69280fd16b03
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_idle.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_IDLE_H_
+#define _XE_GT_IDLE_H_
+
+#include "xe_gt_idle_types.h"
+
+struct xe_gt;
+
+void xe_gt_idle_sysfs_init(struct xe_gt_idle *gtidle);
+void xe_gt_idle_enable_c6(struct xe_gt *gt);
+void xe_gt_idle_disable_c6(struct xe_gt *gt);
+
+#endif /* _XE_GT_IDLE_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_idle_types.h b/drivers/gpu/drm/xe/xe_gt_idle_types.h
new file mode 100644
index 000000000000..f99b447534f3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_idle_types.h
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_IDLE_SYSFS_TYPES_H_
+#define _XE_GT_IDLE_SYSFS_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_guc_pc;
+
+/* States of GT Idle */
+enum xe_gt_idle_state {
+ GT_IDLE_C0,
+ GT_IDLE_C6,
+ GT_IDLE_UNKNOWN,
+};
+
+/**
+ * struct xe_gt_idle - A struct that contains idle properties based of gt
+ */
+struct xe_gt_idle {
+ /** @name: name */
+ char name[16];
+ /** @residency_multiplier: residency multiplier in ns */
+ u32 residency_multiplier;
+ /** @cur_residency: raw driver copy of idle residency */
+ u64 cur_residency;
+ /** @prev_residency: previous residency counter */
+ u64 prev_residency;
+ /** @idle_status: get the current idle state */
+ enum xe_gt_idle_state (*idle_status)(struct xe_guc_pc *pc);
+ /** @idle_residency: get idle residency counter */
+ u64 (*idle_residency)(struct xe_guc_pc *pc);
+};
+
+#endif /* _XE_GT_IDLE_SYSFS_TYPES_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_mcr.c b/drivers/gpu/drm/xe/xe_gt_mcr.c
new file mode 100644
index 000000000000..77925b35cf8d
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_mcr.c
@@ -0,0 +1,685 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_mcr.h"
+
+#include "regs/xe_gt_regs.h"
+#include "xe_gt.h"
+#include "xe_gt_topology.h"
+#include "xe_gt_types.h"
+#include "xe_mmio.h"
+
+/**
+ * DOC: GT Multicast/Replicated (MCR) Register Support
+ *
+ * Some GT registers are designed as "multicast" or "replicated" registers:
+ * multiple instances of the same register share a single MMIO offset. MCR
+ * registers are generally used when the hardware needs to potentially track
+ * independent values of a register per hardware unit (e.g., per-subslice,
+ * per-L3bank, etc.). The specific types of replication that exist vary
+ * per-platform.
+ *
+ * MMIO accesses to MCR registers are controlled according to the settings
+ * programmed in the platform's MCR_SELECTOR register(s). MMIO writes to MCR
+ * registers can be done in either multicast (a single write updates all
+ * instances of the register to the same value) or unicast (a write updates only
+ * one specific instance) form. Reads of MCR registers always operate in a
+ * unicast manner regardless of how the multicast/unicast bit is set in
+ * MCR_SELECTOR. Selection of a specific MCR instance for unicast operations is
+ * referred to as "steering."
+ *
+ * If MCR register operations are steered toward a hardware unit that is
+ * fused off or currently powered down due to power gating, the MMIO operation
+ * is "terminated" by the hardware. Terminated read operations will return a
+ * value of zero and terminated unicast write operations will be silently
+ * ignored. During device initialization, the goal of the various
+ * ``init_steering_*()`` functions is to apply the platform-specific rules for
+ * each MCR register type to identify a steering target that will select a
+ * non-terminated instance.
+ */
+
+#define STEER_SEMAPHORE XE_REG(0xFD0)
+
+static inline struct xe_reg to_xe_reg(struct xe_reg_mcr reg_mcr)
+{
+ return reg_mcr.__reg;
+}
+
+enum {
+ MCR_OP_READ,
+ MCR_OP_WRITE
+};
+
+static const struct xe_mmio_range xelp_l3bank_steering_table[] = {
+ { 0x00B100, 0x00B3FF },
+ {},
+};
+
+static const struct xe_mmio_range xehp_l3bank_steering_table[] = {
+ { 0x008C80, 0x008CFF },
+ { 0x00B100, 0x00B3FF },
+ {},
+};
+
+/*
+ * Although the bspec lists more "MSLICE" ranges than shown here, some of those
+ * are of a "GAM" subclass that has special rules and doesn't need to be
+ * included here.
+ */
+static const struct xe_mmio_range xehp_mslice_steering_table[] = {
+ { 0x00DD00, 0x00DDFF },
+ { 0x00E900, 0x00FFFF }, /* 0xEA00 - OxEFFF is unused */
+ {},
+};
+
+static const struct xe_mmio_range xehp_lncf_steering_table[] = {
+ { 0x00B000, 0x00B0FF },
+ { 0x00D880, 0x00D8FF },
+ {},
+};
+
+/*
+ * We have several types of MCR registers where steering to (0,0) will always
+ * provide us with a non-terminated value. We'll stick them all in the same
+ * table for simplicity.
+ */
+static const struct xe_mmio_range xehpc_instance0_steering_table[] = {
+ { 0x004000, 0x004AFF }, /* HALF-BSLICE */
+ { 0x008800, 0x00887F }, /* CC */
+ { 0x008A80, 0x008AFF }, /* TILEPSMI */
+ { 0x00B000, 0x00B0FF }, /* HALF-BSLICE */
+ { 0x00B100, 0x00B3FF }, /* L3BANK */
+ { 0x00C800, 0x00CFFF }, /* HALF-BSLICE */
+ { 0x00D800, 0x00D8FF }, /* HALF-BSLICE */
+ { 0x00DD00, 0x00DDFF }, /* BSLICE */
+ { 0x00E900, 0x00E9FF }, /* HALF-BSLICE */
+ { 0x00EC00, 0x00EEFF }, /* HALF-BSLICE */
+ { 0x00F000, 0x00FFFF }, /* HALF-BSLICE */
+ { 0x024180, 0x0241FF }, /* HALF-BSLICE */
+ {},
+};
+
+static const struct xe_mmio_range xelpg_instance0_steering_table[] = {
+ { 0x000B00, 0x000BFF }, /* SQIDI */
+ { 0x001000, 0x001FFF }, /* SQIDI */
+ { 0x004000, 0x0048FF }, /* GAM */
+ { 0x008700, 0x0087FF }, /* SQIDI */
+ { 0x00B000, 0x00B0FF }, /* NODE */
+ { 0x00C800, 0x00CFFF }, /* GAM */
+ { 0x00D880, 0x00D8FF }, /* NODE */
+ { 0x00DD00, 0x00DDFF }, /* OAAL2 */
+ {},
+};
+
+static const struct xe_mmio_range xelpg_l3bank_steering_table[] = {
+ { 0x00B100, 0x00B3FF },
+ {},
+};
+
+static const struct xe_mmio_range xelp_dss_steering_table[] = {
+ { 0x008150, 0x00815F },
+ { 0x009520, 0x00955F },
+ { 0x00DE80, 0x00E8FF },
+ { 0x024A00, 0x024A7F },
+ {},
+};
+
+/* DSS steering is used for GSLICE ranges as well */
+static const struct xe_mmio_range xehp_dss_steering_table[] = {
+ { 0x005200, 0x0052FF }, /* GSLICE */
+ { 0x005400, 0x007FFF }, /* GSLICE */
+ { 0x008140, 0x00815F }, /* GSLICE (0x8140-0x814F), DSS (0x8150-0x815F) */
+ { 0x008D00, 0x008DFF }, /* DSS */
+ { 0x0094D0, 0x00955F }, /* GSLICE (0x94D0-0x951F), DSS (0x9520-0x955F) */
+ { 0x009680, 0x0096FF }, /* DSS */
+ { 0x00D800, 0x00D87F }, /* GSLICE */
+ { 0x00DC00, 0x00DCFF }, /* GSLICE */
+ { 0x00DE80, 0x00E8FF }, /* DSS (0xE000-0xE0FF reserved ) */
+ { 0x017000, 0x017FFF }, /* GSLICE */
+ { 0x024A00, 0x024A7F }, /* DSS */
+ {},
+};
+
+/* DSS steering is used for COMPUTE ranges as well */
+static const struct xe_mmio_range xehpc_dss_steering_table[] = {
+ { 0x008140, 0x00817F }, /* COMPUTE (0x8140-0x814F & 0x8160-0x817F), DSS (0x8150-0x815F) */
+ { 0x0094D0, 0x00955F }, /* COMPUTE (0x94D0-0x951F), DSS (0x9520-0x955F) */
+ { 0x009680, 0x0096FF }, /* DSS */
+ { 0x00DC00, 0x00DCFF }, /* COMPUTE */
+ { 0x00DE80, 0x00E7FF }, /* DSS (0xDF00-0xE1FF reserved ) */
+ {},
+};
+
+/* DSS steering is used for SLICE ranges as well */
+static const struct xe_mmio_range xelpg_dss_steering_table[] = {
+ { 0x005200, 0x0052FF }, /* SLICE */
+ { 0x005500, 0x007FFF }, /* SLICE */
+ { 0x008140, 0x00815F }, /* SLICE (0x8140-0x814F), DSS (0x8150-0x815F) */
+ { 0x0094D0, 0x00955F }, /* SLICE (0x94D0-0x951F), DSS (0x9520-0x955F) */
+ { 0x009680, 0x0096FF }, /* DSS */
+ { 0x00D800, 0x00D87F }, /* SLICE */
+ { 0x00DC00, 0x00DCFF }, /* SLICE */
+ { 0x00DE80, 0x00E8FF }, /* DSS (0xE000-0xE0FF reserved) */
+ {},
+};
+
+static const struct xe_mmio_range xelpmp_oaddrm_steering_table[] = {
+ { 0x393200, 0x39323F },
+ { 0x393400, 0x3934FF },
+ {},
+};
+
+static const struct xe_mmio_range dg2_implicit_steering_table[] = {
+ { 0x000B00, 0x000BFF }, /* SF (SQIDI replication) */
+ { 0x001000, 0x001FFF }, /* SF (SQIDI replication) */
+ { 0x004000, 0x004AFF }, /* GAM (MSLICE replication) */
+ { 0x008700, 0x0087FF }, /* MCFG (SQIDI replication) */
+ { 0x00C800, 0x00CFFF }, /* GAM (MSLICE replication) */
+ { 0x00F000, 0x00FFFF }, /* GAM (MSLICE replication) */
+ {},
+};
+
+static const struct xe_mmio_range xe2lpg_dss_steering_table[] = {
+ { 0x005200, 0x0052FF }, /* SLICE */
+ { 0x005500, 0x007FFF }, /* SLICE */
+ { 0x008140, 0x00815F }, /* SLICE (0x8140-0x814F), DSS (0x8150-0x815F) */
+ { 0x0094D0, 0x00955F }, /* SLICE (0x94D0-0x951F), DSS (0x9520-0x955F) */
+ { 0x009680, 0x0096FF }, /* DSS */
+ { 0x00D800, 0x00D87F }, /* SLICE */
+ { 0x00DC00, 0x00DCFF }, /* SLICE */
+ { 0x00DE80, 0x00E8FF }, /* DSS (0xE000-0xE0FF reserved) */
+ { 0x00E980, 0x00E9FF }, /* SLICE */
+ { 0x013000, 0x0133FF }, /* DSS (0x13000-0x131FF), SLICE (0x13200-0x133FF) */
+ {},
+};
+
+static const struct xe_mmio_range xe2lpg_sqidi_psmi_steering_table[] = {
+ { 0x000B00, 0x000BFF },
+ { 0x001000, 0x001FFF },
+ {},
+};
+
+static const struct xe_mmio_range xe2lpg_instance0_steering_table[] = {
+ { 0x004000, 0x004AFF }, /* GAM, rsvd, GAMWKR */
+ { 0x008700, 0x00887F }, /* SQIDI, MEMPIPE */
+ { 0x00B000, 0x00B3FF }, /* NODE, L3BANK */
+ { 0x00C800, 0x00CFFF }, /* GAM */
+ { 0x00D880, 0x00D8FF }, /* NODE */
+ { 0x00DD00, 0x00DDFF }, /* MEMPIPE */
+ { 0x00E900, 0x00E97F }, /* MEMPIPE */
+ { 0x00F000, 0x00FFFF }, /* GAM, GAMWKR */
+ { 0x013400, 0x0135FF }, /* MEMPIPE */
+ {},
+};
+
+static const struct xe_mmio_range xe2lpm_gpmxmt_steering_table[] = {
+ { 0x388160, 0x38817F },
+ { 0x389480, 0x3894CF },
+ {},
+};
+
+static const struct xe_mmio_range xe2lpm_instance0_steering_table[] = {
+ { 0x384000, 0x3847DF }, /* GAM, rsvd, GAM */
+ { 0x384900, 0x384AFF }, /* GAM */
+ { 0x389560, 0x3895FF }, /* MEDIAINF */
+ { 0x38B600, 0x38B8FF }, /* L3BANK */
+ { 0x38C800, 0x38D07F }, /* GAM, MEDIAINF */
+ { 0x38F000, 0x38F0FF }, /* GAM */
+ { 0x393C00, 0x393C7F }, /* MEDIAINF */
+ {},
+};
+
+static void init_steering_l3bank(struct xe_gt *gt)
+{
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1270) {
+ u32 mslice_mask = REG_FIELD_GET(MEML3_EN_MASK,
+ xe_mmio_read32(gt, MIRROR_FUSE3));
+ u32 bank_mask = REG_FIELD_GET(GT_L3_EXC_MASK,
+ xe_mmio_read32(gt, XEHP_FUSE4));
+
+ /*
+ * Group selects mslice, instance selects bank within mslice.
+ * Bank 0 is always valid _except_ when the bank mask is 010b.
+ */
+ gt->steering[L3BANK].group_target = __ffs(mslice_mask);
+ gt->steering[L3BANK].instance_target =
+ bank_mask & BIT(0) ? 0 : 2;
+ } else if (gt_to_xe(gt)->info.platform == XE_DG2) {
+ u32 mslice_mask = REG_FIELD_GET(MEML3_EN_MASK,
+ xe_mmio_read32(gt, MIRROR_FUSE3));
+ u32 bank = __ffs(mslice_mask) * 8;
+
+ /*
+ * Like mslice registers, look for a valid mslice and steer to
+ * the first L3BANK of that quad. Access to the Nth L3 bank is
+ * split between the first bits of group and instance
+ */
+ gt->steering[L3BANK].group_target = (bank >> 2) & 0x7;
+ gt->steering[L3BANK].instance_target = bank & 0x3;
+ } else {
+ u32 fuse = REG_FIELD_GET(L3BANK_MASK,
+ ~xe_mmio_read32(gt, MIRROR_FUSE3));
+
+ gt->steering[L3BANK].group_target = 0; /* unused */
+ gt->steering[L3BANK].instance_target = __ffs(fuse);
+ }
+}
+
+static void init_steering_mslice(struct xe_gt *gt)
+{
+ u32 mask = REG_FIELD_GET(MEML3_EN_MASK,
+ xe_mmio_read32(gt, MIRROR_FUSE3));
+
+ /*
+ * mslice registers are valid (not terminated) if either the meml3
+ * associated with the mslice is present, or at least one DSS associated
+ * with the mslice is present. There will always be at least one meml3
+ * so we can just use that to find a non-terminated mslice and ignore
+ * the DSS fusing.
+ */
+ gt->steering[MSLICE].group_target = __ffs(mask);
+ gt->steering[MSLICE].instance_target = 0; /* unused */
+
+ /*
+ * LNCF termination is also based on mslice presence, so we'll set
+ * it up here. Either LNCF within a non-terminated mslice will work,
+ * so we just always pick LNCF 0 here.
+ */
+ gt->steering[LNCF].group_target = __ffs(mask) << 1;
+ gt->steering[LNCF].instance_target = 0; /* unused */
+}
+
+static void init_steering_dss(struct xe_gt *gt)
+{
+ unsigned int dss = min(xe_dss_mask_group_ffs(gt->fuse_topo.g_dss_mask, 0, 0),
+ xe_dss_mask_group_ffs(gt->fuse_topo.c_dss_mask, 0, 0));
+ unsigned int dss_per_grp = gt_to_xe(gt)->info.platform == XE_PVC ? 8 : 4;
+
+ gt->steering[DSS].group_target = dss / dss_per_grp;
+ gt->steering[DSS].instance_target = dss % dss_per_grp;
+}
+
+static void init_steering_oaddrm(struct xe_gt *gt)
+{
+ /*
+ * First instance is only terminated if the entire first media slice
+ * is absent (i.e., no VCS0 or VECS0).
+ */
+ if (gt->info.engine_mask & (XE_HW_ENGINE_VCS0 | XE_HW_ENGINE_VECS0))
+ gt->steering[OADDRM].group_target = 0;
+ else
+ gt->steering[OADDRM].group_target = 1;
+
+ gt->steering[DSS].instance_target = 0; /* unused */
+}
+
+static void init_steering_sqidi_psmi(struct xe_gt *gt)
+{
+ u32 mask = REG_FIELD_GET(XE2_NODE_ENABLE_MASK,
+ xe_mmio_read32(gt, MIRROR_FUSE3));
+ u32 select = __ffs(mask);
+
+ gt->steering[SQIDI_PSMI].group_target = select >> 1;
+ gt->steering[SQIDI_PSMI].instance_target = select & 0x1;
+}
+
+static void init_steering_inst0(struct xe_gt *gt)
+{
+ gt->steering[DSS].group_target = 0; /* unused */
+ gt->steering[DSS].instance_target = 0; /* unused */
+}
+
+static const struct {
+ const char *name;
+ void (*init)(struct xe_gt *gt);
+} xe_steering_types[] = {
+ [L3BANK] = { "L3BANK", init_steering_l3bank },
+ [MSLICE] = { "MSLICE", init_steering_mslice },
+ [LNCF] = { "LNCF", NULL }, /* initialized by mslice init */
+ [DSS] = { "DSS", init_steering_dss },
+ [OADDRM] = { "OADDRM / GPMXMT", init_steering_oaddrm },
+ [SQIDI_PSMI] = { "SQIDI_PSMI", init_steering_sqidi_psmi },
+ [INSTANCE0] = { "INSTANCE 0", init_steering_inst0 },
+ [IMPLICIT_STEERING] = { "IMPLICIT", NULL },
+};
+
+void xe_gt_mcr_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ BUILD_BUG_ON(IMPLICIT_STEERING + 1 != NUM_STEERING_TYPES);
+ BUILD_BUG_ON(ARRAY_SIZE(xe_steering_types) != NUM_STEERING_TYPES);
+
+ spin_lock_init(&gt->mcr_lock);
+
+ if (gt->info.type == XE_GT_TYPE_MEDIA) {
+ drm_WARN_ON(&xe->drm, MEDIA_VER(xe) < 13);
+
+ if (MEDIA_VER(xe) >= 20) {
+ gt->steering[OADDRM].ranges = xe2lpm_gpmxmt_steering_table;
+ gt->steering[INSTANCE0].ranges = xe2lpm_instance0_steering_table;
+ } else {
+ gt->steering[OADDRM].ranges = xelpmp_oaddrm_steering_table;
+ }
+ } else {
+ if (GRAPHICS_VER(xe) >= 20) {
+ gt->steering[DSS].ranges = xe2lpg_dss_steering_table;
+ gt->steering[SQIDI_PSMI].ranges = xe2lpg_sqidi_psmi_steering_table;
+ gt->steering[INSTANCE0].ranges = xe2lpg_instance0_steering_table;
+ } else if (GRAPHICS_VERx100(xe) >= 1270) {
+ gt->steering[INSTANCE0].ranges = xelpg_instance0_steering_table;
+ gt->steering[L3BANK].ranges = xelpg_l3bank_steering_table;
+ gt->steering[DSS].ranges = xelpg_dss_steering_table;
+ } else if (xe->info.platform == XE_PVC) {
+ gt->steering[INSTANCE0].ranges = xehpc_instance0_steering_table;
+ gt->steering[DSS].ranges = xehpc_dss_steering_table;
+ } else if (xe->info.platform == XE_DG2) {
+ gt->steering[L3BANK].ranges = xehp_l3bank_steering_table;
+ gt->steering[MSLICE].ranges = xehp_mslice_steering_table;
+ gt->steering[LNCF].ranges = xehp_lncf_steering_table;
+ gt->steering[DSS].ranges = xehp_dss_steering_table;
+ gt->steering[IMPLICIT_STEERING].ranges = dg2_implicit_steering_table;
+ } else {
+ gt->steering[L3BANK].ranges = xelp_l3bank_steering_table;
+ gt->steering[DSS].ranges = xelp_dss_steering_table;
+ }
+ }
+
+ /* Select non-terminated steering target for each type */
+ for (int i = 0; i < NUM_STEERING_TYPES; i++)
+ if (gt->steering[i].ranges && xe_steering_types[i].init)
+ xe_steering_types[i].init(gt);
+}
+
+/**
+ * xe_gt_mcr_set_implicit_defaults - Initialize steer control registers
+ * @gt: GT structure
+ *
+ * Some register ranges don't need to have their steering control registers
+ * changed on each access - it's sufficient to set them once on initialization.
+ * This function sets those registers for each platform *
+ */
+void xe_gt_mcr_set_implicit_defaults(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (xe->info.platform == XE_DG2) {
+ u32 steer_val = REG_FIELD_PREP(MCR_SLICE_MASK, 0) |
+ REG_FIELD_PREP(MCR_SUBSLICE_MASK, 2);
+
+ xe_mmio_write32(gt, MCFG_MCR_SELECTOR, steer_val);
+ xe_mmio_write32(gt, SF_MCR_SELECTOR, steer_val);
+ /*
+ * For GAM registers, all reads should be directed to instance 1
+ * (unicast reads against other instances are not allowed),
+ * and instance 1 is already the hardware's default steering
+ * target, which we never change
+ */
+ }
+}
+
+/*
+ * xe_gt_mcr_get_nonterminated_steering - find group/instance values that
+ * will steer a register to a non-terminated instance
+ * @gt: GT structure
+ * @reg: register for which the steering is required
+ * @group: return variable for group steering
+ * @instance: return variable for instance steering
+ *
+ * This function returns a group/instance pair that is guaranteed to work for
+ * read steering of the given register. Note that a value will be returned even
+ * if the register is not replicated and therefore does not actually require
+ * steering.
+ *
+ * Returns true if the caller should steer to the @group/@instance values
+ * returned. Returns false if the caller need not perform any steering
+ */
+static bool xe_gt_mcr_get_nonterminated_steering(struct xe_gt *gt,
+ struct xe_reg_mcr reg_mcr,
+ u8 *group, u8 *instance)
+{
+ const struct xe_reg reg = to_xe_reg(reg_mcr);
+ const struct xe_mmio_range *implicit_ranges;
+
+ for (int type = 0; type < IMPLICIT_STEERING; type++) {
+ if (!gt->steering[type].ranges)
+ continue;
+
+ for (int i = 0; gt->steering[type].ranges[i].end > 0; i++) {
+ if (xe_mmio_in_range(gt, &gt->steering[type].ranges[i], reg)) {
+ *group = gt->steering[type].group_target;
+ *instance = gt->steering[type].instance_target;
+ return true;
+ }
+ }
+ }
+
+ implicit_ranges = gt->steering[IMPLICIT_STEERING].ranges;
+ if (implicit_ranges)
+ for (int i = 0; implicit_ranges[i].end > 0; i++)
+ if (xe_mmio_in_range(gt, &implicit_ranges[i], reg))
+ return false;
+
+ /*
+ * Not found in a steering table and not a register with implicit
+ * steering. Just steer to 0/0 as a guess and raise a warning.
+ */
+ drm_WARN(&gt_to_xe(gt)->drm, true,
+ "Did not find MCR register %#x in any MCR steering table\n",
+ reg.addr);
+ *group = 0;
+ *instance = 0;
+
+ return true;
+}
+
+/*
+ * Obtain exclusive access to MCR steering. On MTL and beyond we also need
+ * to synchronize with external clients (e.g., firmware), so a semaphore
+ * register will also need to be taken.
+ */
+static void mcr_lock(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int ret = 0;
+
+ spin_lock(&gt->mcr_lock);
+
+ /*
+ * Starting with MTL we also need to grab a semaphore register
+ * to synchronize with external agents (e.g., firmware) that now
+ * shares the same steering control register. The semaphore is obtained
+ * when a read to the relevant register returns 1.
+ */
+ if (GRAPHICS_VERx100(xe) >= 1270)
+ ret = xe_mmio_wait32(gt, STEER_SEMAPHORE, 0x1, 0x1, 10, NULL,
+ true);
+
+ drm_WARN_ON_ONCE(&xe->drm, ret == -ETIMEDOUT);
+}
+
+static void mcr_unlock(struct xe_gt *gt)
+{
+ /* Release hardware semaphore - this is done by writing 1 to the register */
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1270)
+ xe_mmio_write32(gt, STEER_SEMAPHORE, 0x1);
+
+ spin_unlock(&gt->mcr_lock);
+}
+
+/*
+ * Access a register with specific MCR steering
+ *
+ * Caller needs to make sure the relevant forcewake wells are up.
+ */
+static u32 rw_with_mcr_steering(struct xe_gt *gt, struct xe_reg_mcr reg_mcr,
+ u8 rw_flag, int group, int instance, u32 value)
+{
+ const struct xe_reg reg = to_xe_reg(reg_mcr);
+ struct xe_reg steer_reg;
+ u32 steer_val, val = 0;
+
+ lockdep_assert_held(&gt->mcr_lock);
+
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1270) {
+ steer_reg = MTL_MCR_SELECTOR;
+ steer_val = REG_FIELD_PREP(MTL_MCR_GROUPID, group) |
+ REG_FIELD_PREP(MTL_MCR_INSTANCEID, instance);
+ } else {
+ steer_reg = MCR_SELECTOR;
+ steer_val = REG_FIELD_PREP(MCR_SLICE_MASK, group) |
+ REG_FIELD_PREP(MCR_SUBSLICE_MASK, instance);
+ }
+
+ /*
+ * Always leave the hardware in multicast mode when doing reads and only
+ * change it to unicast mode when doing writes of a specific instance.
+ *
+ * The setting of the multicast/unicast bit usually wouldn't matter for
+ * read operations (which always return the value from a single register
+ * instance regardless of how that bit is set), but some platforms may
+ * have workarounds requiring us to remain in multicast mode for reads,
+ * e.g. Wa_22013088509 on PVC. There's no real downside to this, so
+ * we'll just go ahead and do so on all platforms; we'll only clear the
+ * multicast bit from the mask when explicitly doing a write operation.
+ *
+ * No need to save old steering reg value.
+ */
+ if (rw_flag == MCR_OP_READ)
+ steer_val |= MCR_MULTICAST;
+
+ xe_mmio_write32(gt, steer_reg, steer_val);
+
+ if (rw_flag == MCR_OP_READ)
+ val = xe_mmio_read32(gt, reg);
+ else
+ xe_mmio_write32(gt, reg, value);
+
+ /*
+ * If we turned off the multicast bit (during a write) we're required
+ * to turn it back on before finishing. The group and instance values
+ * don't matter since they'll be re-programmed on the next MCR
+ * operation.
+ */
+ if (rw_flag == MCR_OP_WRITE)
+ xe_mmio_write32(gt, steer_reg, MCR_MULTICAST);
+
+ return val;
+}
+
+/**
+ * xe_gt_mcr_unicast_read_any - reads a non-terminated instance of an MCR register
+ * @gt: GT structure
+ * @reg_mcr: register to read
+ *
+ * Reads a GT MCR register. The read will be steered to a non-terminated
+ * instance (i.e., one that isn't fused off or powered down by power gating).
+ * This function assumes the caller is already holding any necessary forcewake
+ * domains.
+ *
+ * Returns the value from a non-terminated instance of @reg.
+ */
+u32 xe_gt_mcr_unicast_read_any(struct xe_gt *gt, struct xe_reg_mcr reg_mcr)
+{
+ const struct xe_reg reg = to_xe_reg(reg_mcr);
+ u8 group, instance;
+ u32 val;
+ bool steer;
+
+ steer = xe_gt_mcr_get_nonterminated_steering(gt, reg_mcr,
+ &group, &instance);
+
+ if (steer) {
+ mcr_lock(gt);
+ val = rw_with_mcr_steering(gt, reg_mcr, MCR_OP_READ,
+ group, instance, 0);
+ mcr_unlock(gt);
+ } else {
+ val = xe_mmio_read32(gt, reg);
+ }
+
+ return val;
+}
+
+/**
+ * xe_gt_mcr_unicast_read - read a specific instance of an MCR register
+ * @gt: GT structure
+ * @reg_mcr: the MCR register to read
+ * @group: the MCR group
+ * @instance: the MCR instance
+ *
+ * Returns the value read from an MCR register after steering toward a specific
+ * group/instance.
+ */
+u32 xe_gt_mcr_unicast_read(struct xe_gt *gt,
+ struct xe_reg_mcr reg_mcr,
+ int group, int instance)
+{
+ u32 val;
+
+ mcr_lock(gt);
+ val = rw_with_mcr_steering(gt, reg_mcr, MCR_OP_READ, group, instance, 0);
+ mcr_unlock(gt);
+
+ return val;
+}
+
+/**
+ * xe_gt_mcr_unicast_write - write a specific instance of an MCR register
+ * @gt: GT structure
+ * @reg_mcr: the MCR register to write
+ * @value: value to write
+ * @group: the MCR group
+ * @instance: the MCR instance
+ *
+ * Write an MCR register in unicast mode after steering toward a specific
+ * group/instance.
+ */
+void xe_gt_mcr_unicast_write(struct xe_gt *gt, struct xe_reg_mcr reg_mcr,
+ u32 value, int group, int instance)
+{
+ mcr_lock(gt);
+ rw_with_mcr_steering(gt, reg_mcr, MCR_OP_WRITE, group, instance, value);
+ mcr_unlock(gt);
+}
+
+/**
+ * xe_gt_mcr_multicast_write - write a value to all instances of an MCR register
+ * @gt: GT structure
+ * @reg_mcr: the MCR register to write
+ * @value: value to write
+ *
+ * Write an MCR register in multicast mode to update all instances.
+ */
+void xe_gt_mcr_multicast_write(struct xe_gt *gt, struct xe_reg_mcr reg_mcr,
+ u32 value)
+{
+ struct xe_reg reg = to_xe_reg(reg_mcr);
+
+ /*
+ * Synchronize with any unicast operations. Once we have exclusive
+ * access, the MULTICAST bit should already be set, so there's no need
+ * to touch the steering register.
+ */
+ mcr_lock(gt);
+ xe_mmio_write32(gt, reg, value);
+ mcr_unlock(gt);
+}
+
+void xe_gt_mcr_steering_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ for (int i = 0; i < NUM_STEERING_TYPES; i++) {
+ if (gt->steering[i].ranges) {
+ drm_printf(p, "%s steering: group=%#x, instance=%#x\n",
+ xe_steering_types[i].name,
+ gt->steering[i].group_target,
+ gt->steering[i].instance_target);
+ for (int j = 0; gt->steering[i].ranges[j].end; j++)
+ drm_printf(p, "\t0x%06x - 0x%06x\n",
+ gt->steering[i].ranges[j].start,
+ gt->steering[i].ranges[j].end);
+ }
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_mcr.h b/drivers/gpu/drm/xe/xe_gt_mcr.h
new file mode 100644
index 000000000000..27ca1bc880a0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_mcr.h
@@ -0,0 +1,29 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_MCR_H_
+#define _XE_GT_MCR_H_
+
+#include "regs/xe_reg_defs.h"
+
+struct drm_printer;
+struct xe_gt;
+
+void xe_gt_mcr_init(struct xe_gt *gt);
+
+void xe_gt_mcr_set_implicit_defaults(struct xe_gt *gt);
+
+u32 xe_gt_mcr_unicast_read(struct xe_gt *gt, struct xe_reg_mcr mcr_reg,
+ int group, int instance);
+u32 xe_gt_mcr_unicast_read_any(struct xe_gt *gt, struct xe_reg_mcr mcr_reg);
+
+void xe_gt_mcr_unicast_write(struct xe_gt *gt, struct xe_reg_mcr mcr_reg,
+ u32 value, int group, int instance);
+void xe_gt_mcr_multicast_write(struct xe_gt *gt, struct xe_reg_mcr mcr_reg,
+ u32 value);
+
+void xe_gt_mcr_steering_dump(struct xe_gt *gt, struct drm_printer *p);
+
+#endif /* _XE_GT_MCR_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c b/drivers/gpu/drm/xe/xe_gt_pagefault.c
new file mode 100644
index 000000000000..59a70d2e0a7a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
@@ -0,0 +1,646 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_pagefault.h"
+
+#include <linux/bitfield.h>
+#include <linux/circ_buf.h>
+
+#include <drm/drm_exec.h>
+#include <drm/drm_managed.h>
+#include <drm/ttm/ttm_execbuf_util.h>
+
+#include "abi/guc_actions_abi.h"
+#include "xe_bo.h"
+#include "xe_gt.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_guc.h"
+#include "xe_guc_ct.h"
+#include "xe_migrate.h"
+#include "xe_pt.h"
+#include "xe_trace.h"
+#include "xe_vm.h"
+
+struct pagefault {
+ u64 page_addr;
+ u32 asid;
+ u16 pdata;
+ u8 vfid;
+ u8 access_type;
+ u8 fault_type;
+ u8 fault_level;
+ u8 engine_class;
+ u8 engine_instance;
+ u8 fault_unsuccessful;
+ bool trva_fault;
+};
+
+enum access_type {
+ ACCESS_TYPE_READ = 0,
+ ACCESS_TYPE_WRITE = 1,
+ ACCESS_TYPE_ATOMIC = 2,
+ ACCESS_TYPE_RESERVED = 3,
+};
+
+enum fault_type {
+ NOT_PRESENT = 0,
+ WRITE_ACCESS_VIOLATION = 1,
+ ATOMIC_ACCESS_VIOLATION = 2,
+};
+
+struct acc {
+ u64 va_range_base;
+ u32 asid;
+ u32 sub_granularity;
+ u8 granularity;
+ u8 vfid;
+ u8 access_type;
+ u8 engine_class;
+ u8 engine_instance;
+};
+
+static bool access_is_atomic(enum access_type access_type)
+{
+ return access_type == ACCESS_TYPE_ATOMIC;
+}
+
+static bool vma_is_valid(struct xe_tile *tile, struct xe_vma *vma)
+{
+ return BIT(tile->id) & vma->tile_present &&
+ !(BIT(tile->id) & vma->usm.tile_invalidated);
+}
+
+static bool vma_matches(struct xe_vma *vma, u64 page_addr)
+{
+ if (page_addr > xe_vma_end(vma) - 1 ||
+ page_addr + SZ_4K - 1 < xe_vma_start(vma))
+ return false;
+
+ return true;
+}
+
+static struct xe_vma *lookup_vma(struct xe_vm *vm, u64 page_addr)
+{
+ struct xe_vma *vma = NULL;
+
+ if (vm->usm.last_fault_vma) { /* Fast lookup */
+ if (vma_matches(vm->usm.last_fault_vma, page_addr))
+ vma = vm->usm.last_fault_vma;
+ }
+ if (!vma)
+ vma = xe_vm_find_overlapping_vma(vm, page_addr, SZ_4K);
+
+ return vma;
+}
+
+static int xe_pf_begin(struct drm_exec *exec, struct xe_vma *vma,
+ bool atomic, unsigned int id)
+{
+ struct xe_bo *bo = xe_vma_bo(vma);
+ struct xe_vm *vm = xe_vma_vm(vma);
+ unsigned int num_shared = 2; /* slots for bind + move */
+ int err;
+
+ err = xe_vm_prepare_vma(exec, vma, num_shared);
+ if (err)
+ return err;
+
+ if (atomic && IS_DGFX(vm->xe)) {
+ if (xe_vma_is_userptr(vma)) {
+ err = -EACCES;
+ return err;
+ }
+
+ /* Migrate to VRAM, move should invalidate the VMA first */
+ err = xe_bo_migrate(bo, XE_PL_VRAM0 + id);
+ if (err)
+ return err;
+ } else if (bo) {
+ /* Create backing store if needed */
+ err = xe_bo_validate(bo, vm, true);
+ if (err)
+ return err;
+ }
+
+ return 0;
+}
+
+static int handle_pagefault(struct xe_gt *gt, struct pagefault *pf)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct drm_exec exec;
+ struct xe_vm *vm;
+ struct xe_vma *vma = NULL;
+ struct dma_fence *fence;
+ bool write_locked;
+ int ret = 0;
+ bool atomic;
+
+ /* SW isn't expected to handle TRTT faults */
+ if (pf->trva_fault)
+ return -EFAULT;
+
+ /* ASID to VM */
+ mutex_lock(&xe->usm.lock);
+ vm = xa_load(&xe->usm.asid_to_vm, pf->asid);
+ if (vm)
+ xe_vm_get(vm);
+ mutex_unlock(&xe->usm.lock);
+ if (!vm || !xe_vm_in_fault_mode(vm))
+ return -EINVAL;
+
+retry_userptr:
+ /*
+ * TODO: Avoid exclusive lock if VM doesn't have userptrs, or
+ * start out read-locked?
+ */
+ down_write(&vm->lock);
+ write_locked = true;
+ vma = lookup_vma(vm, pf->page_addr);
+ if (!vma) {
+ ret = -EINVAL;
+ goto unlock_vm;
+ }
+
+ if (!xe_vma_is_userptr(vma) || !xe_vma_userptr_check_repin(vma)) {
+ downgrade_write(&vm->lock);
+ write_locked = false;
+ }
+
+ trace_xe_vma_pagefault(vma);
+
+ atomic = access_is_atomic(pf->access_type);
+
+ /* Check if VMA is valid */
+ if (vma_is_valid(tile, vma) && !atomic)
+ goto unlock_vm;
+
+ /* TODO: Validate fault */
+
+ if (xe_vma_is_userptr(vma) && write_locked) {
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_del_init(&vma->userptr.invalidate_link);
+ spin_unlock(&vm->userptr.invalidated_lock);
+
+ ret = xe_vma_userptr_pin_pages(vma);
+ if (ret)
+ goto unlock_vm;
+
+ downgrade_write(&vm->lock);
+ write_locked = false;
+ }
+
+ /* Lock VM and BOs dma-resv */
+ drm_exec_init(&exec, 0, 0);
+ drm_exec_until_all_locked(&exec) {
+ ret = xe_pf_begin(&exec, vma, atomic, tile->id);
+ drm_exec_retry_on_contention(&exec);
+ if (ret)
+ goto unlock_dma_resv;
+ }
+
+ /* Bind VMA only to the GT that has faulted */
+ trace_xe_vma_pf_bind(vma);
+ fence = __xe_pt_bind_vma(tile, vma, xe_tile_migrate_engine(tile), NULL, 0,
+ vma->tile_present & BIT(tile->id));
+ if (IS_ERR(fence)) {
+ ret = PTR_ERR(fence);
+ goto unlock_dma_resv;
+ }
+
+ /*
+ * XXX: Should we drop the lock before waiting? This only helps if doing
+ * GPU binds which is currently only done if we have to wait for more
+ * than 10ms on a move.
+ */
+ dma_fence_wait(fence, false);
+ dma_fence_put(fence);
+
+ if (xe_vma_is_userptr(vma))
+ ret = xe_vma_userptr_check_repin(vma);
+ vma->usm.tile_invalidated &= ~BIT(tile->id);
+
+unlock_dma_resv:
+ drm_exec_fini(&exec);
+unlock_vm:
+ if (!ret)
+ vm->usm.last_fault_vma = vma;
+ if (write_locked)
+ up_write(&vm->lock);
+ else
+ up_read(&vm->lock);
+ if (ret == -EAGAIN)
+ goto retry_userptr;
+
+ if (!ret) {
+ ret = xe_gt_tlb_invalidation_vma(gt, NULL, vma);
+ if (ret >= 0)
+ ret = 0;
+ }
+ xe_vm_put(vm);
+
+ return ret;
+}
+
+static int send_pagefault_reply(struct xe_guc *guc,
+ struct xe_guc_pagefault_reply *reply)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_PAGE_FAULT_RES_DESC,
+ reply->dw0,
+ reply->dw1,
+ };
+
+ return xe_guc_ct_send(&guc->ct, action, ARRAY_SIZE(action), 0, 0);
+}
+
+static void print_pagefault(struct xe_device *xe, struct pagefault *pf)
+{
+ drm_dbg(&xe->drm, "\n\tASID: %d\n"
+ "\tVFID: %d\n"
+ "\tPDATA: 0x%04x\n"
+ "\tFaulted Address: 0x%08x%08x\n"
+ "\tFaultType: %d\n"
+ "\tAccessType: %d\n"
+ "\tFaultLevel: %d\n"
+ "\tEngineClass: %d\n"
+ "\tEngineInstance: %d\n",
+ pf->asid, pf->vfid, pf->pdata, upper_32_bits(pf->page_addr),
+ lower_32_bits(pf->page_addr),
+ pf->fault_type, pf->access_type, pf->fault_level,
+ pf->engine_class, pf->engine_instance);
+}
+
+#define PF_MSG_LEN_DW 4
+
+static bool get_pagefault(struct pf_queue *pf_queue, struct pagefault *pf)
+{
+ const struct xe_guc_pagefault_desc *desc;
+ bool ret = false;
+
+ spin_lock_irq(&pf_queue->lock);
+ if (pf_queue->head != pf_queue->tail) {
+ desc = (const struct xe_guc_pagefault_desc *)
+ (pf_queue->data + pf_queue->head);
+
+ pf->fault_level = FIELD_GET(PFD_FAULT_LEVEL, desc->dw0);
+ pf->trva_fault = FIELD_GET(XE2_PFD_TRVA_FAULT, desc->dw0);
+ pf->engine_class = FIELD_GET(PFD_ENG_CLASS, desc->dw0);
+ pf->engine_instance = FIELD_GET(PFD_ENG_INSTANCE, desc->dw0);
+ pf->pdata = FIELD_GET(PFD_PDATA_HI, desc->dw1) <<
+ PFD_PDATA_HI_SHIFT;
+ pf->pdata |= FIELD_GET(PFD_PDATA_LO, desc->dw0);
+ pf->asid = FIELD_GET(PFD_ASID, desc->dw1);
+ pf->vfid = FIELD_GET(PFD_VFID, desc->dw2);
+ pf->access_type = FIELD_GET(PFD_ACCESS_TYPE, desc->dw2);
+ pf->fault_type = FIELD_GET(PFD_FAULT_TYPE, desc->dw2);
+ pf->page_addr = (u64)(FIELD_GET(PFD_VIRTUAL_ADDR_HI, desc->dw3)) <<
+ PFD_VIRTUAL_ADDR_HI_SHIFT;
+ pf->page_addr |= FIELD_GET(PFD_VIRTUAL_ADDR_LO, desc->dw2) <<
+ PFD_VIRTUAL_ADDR_LO_SHIFT;
+
+ pf_queue->head = (pf_queue->head + PF_MSG_LEN_DW) %
+ PF_QUEUE_NUM_DW;
+ ret = true;
+ }
+ spin_unlock_irq(&pf_queue->lock);
+
+ return ret;
+}
+
+static bool pf_queue_full(struct pf_queue *pf_queue)
+{
+ lockdep_assert_held(&pf_queue->lock);
+
+ return CIRC_SPACE(pf_queue->tail, pf_queue->head, PF_QUEUE_NUM_DW) <=
+ PF_MSG_LEN_DW;
+}
+
+int xe_guc_pagefault_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct pf_queue *pf_queue;
+ unsigned long flags;
+ u32 asid;
+ bool full;
+
+ if (unlikely(len != PF_MSG_LEN_DW))
+ return -EPROTO;
+
+ asid = FIELD_GET(PFD_ASID, msg[1]);
+ pf_queue = &gt->usm.pf_queue[asid % NUM_PF_QUEUE];
+
+ spin_lock_irqsave(&pf_queue->lock, flags);
+ full = pf_queue_full(pf_queue);
+ if (!full) {
+ memcpy(pf_queue->data + pf_queue->tail, msg, len * sizeof(u32));
+ pf_queue->tail = (pf_queue->tail + len) % PF_QUEUE_NUM_DW;
+ queue_work(gt->usm.pf_wq, &pf_queue->worker);
+ } else {
+ drm_warn(&xe->drm, "PF Queue full, shouldn't be possible");
+ }
+ spin_unlock_irqrestore(&pf_queue->lock, flags);
+
+ return full ? -ENOSPC : 0;
+}
+
+#define USM_QUEUE_MAX_RUNTIME_MS 20
+
+static void pf_queue_work_func(struct work_struct *w)
+{
+ struct pf_queue *pf_queue = container_of(w, struct pf_queue, worker);
+ struct xe_gt *gt = pf_queue->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_guc_pagefault_reply reply = {};
+ struct pagefault pf = {};
+ unsigned long threshold;
+ int ret;
+
+ threshold = jiffies + msecs_to_jiffies(USM_QUEUE_MAX_RUNTIME_MS);
+
+ while (get_pagefault(pf_queue, &pf)) {
+ ret = handle_pagefault(gt, &pf);
+ if (unlikely(ret)) {
+ print_pagefault(xe, &pf);
+ pf.fault_unsuccessful = 1;
+ drm_dbg(&xe->drm, "Fault response: Unsuccessful %d\n", ret);
+ }
+
+ reply.dw0 = FIELD_PREP(PFR_VALID, 1) |
+ FIELD_PREP(PFR_SUCCESS, pf.fault_unsuccessful) |
+ FIELD_PREP(PFR_REPLY, PFR_ACCESS) |
+ FIELD_PREP(PFR_DESC_TYPE, FAULT_RESPONSE_DESC) |
+ FIELD_PREP(PFR_ASID, pf.asid);
+
+ reply.dw1 = FIELD_PREP(PFR_VFID, pf.vfid) |
+ FIELD_PREP(PFR_ENG_INSTANCE, pf.engine_instance) |
+ FIELD_PREP(PFR_ENG_CLASS, pf.engine_class) |
+ FIELD_PREP(PFR_PDATA, pf.pdata);
+
+ send_pagefault_reply(&gt->uc.guc, &reply);
+
+ if (time_after(jiffies, threshold) &&
+ pf_queue->head != pf_queue->tail) {
+ queue_work(gt->usm.pf_wq, w);
+ break;
+ }
+ }
+}
+
+static void acc_queue_work_func(struct work_struct *w);
+
+int xe_gt_pagefault_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i;
+
+ if (!xe->info.has_usm)
+ return 0;
+
+ for (i = 0; i < NUM_PF_QUEUE; ++i) {
+ gt->usm.pf_queue[i].gt = gt;
+ spin_lock_init(&gt->usm.pf_queue[i].lock);
+ INIT_WORK(&gt->usm.pf_queue[i].worker, pf_queue_work_func);
+ }
+ for (i = 0; i < NUM_ACC_QUEUE; ++i) {
+ gt->usm.acc_queue[i].gt = gt;
+ spin_lock_init(&gt->usm.acc_queue[i].lock);
+ INIT_WORK(&gt->usm.acc_queue[i].worker, acc_queue_work_func);
+ }
+
+ gt->usm.pf_wq = alloc_workqueue("xe_gt_page_fault_work_queue",
+ WQ_UNBOUND | WQ_HIGHPRI, NUM_PF_QUEUE);
+ if (!gt->usm.pf_wq)
+ return -ENOMEM;
+
+ gt->usm.acc_wq = alloc_workqueue("xe_gt_access_counter_work_queue",
+ WQ_UNBOUND | WQ_HIGHPRI,
+ NUM_ACC_QUEUE);
+ if (!gt->usm.acc_wq)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void xe_gt_pagefault_reset(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i;
+
+ if (!xe->info.has_usm)
+ return;
+
+ for (i = 0; i < NUM_PF_QUEUE; ++i) {
+ spin_lock_irq(&gt->usm.pf_queue[i].lock);
+ gt->usm.pf_queue[i].head = 0;
+ gt->usm.pf_queue[i].tail = 0;
+ spin_unlock_irq(&gt->usm.pf_queue[i].lock);
+ }
+
+ for (i = 0; i < NUM_ACC_QUEUE; ++i) {
+ spin_lock(&gt->usm.acc_queue[i].lock);
+ gt->usm.acc_queue[i].head = 0;
+ gt->usm.acc_queue[i].tail = 0;
+ spin_unlock(&gt->usm.acc_queue[i].lock);
+ }
+}
+
+static int granularity_in_byte(int val)
+{
+ switch (val) {
+ case 0:
+ return SZ_128K;
+ case 1:
+ return SZ_2M;
+ case 2:
+ return SZ_16M;
+ case 3:
+ return SZ_64M;
+ default:
+ return 0;
+ }
+}
+
+static int sub_granularity_in_byte(int val)
+{
+ return (granularity_in_byte(val) / 32);
+}
+
+static void print_acc(struct xe_device *xe, struct acc *acc)
+{
+ drm_warn(&xe->drm, "Access counter request:\n"
+ "\tType: %s\n"
+ "\tASID: %d\n"
+ "\tVFID: %d\n"
+ "\tEngine: %d:%d\n"
+ "\tGranularity: 0x%x KB Region/ %d KB sub-granularity\n"
+ "\tSub_Granularity Vector: 0x%08x\n"
+ "\tVA Range base: 0x%016llx\n",
+ acc->access_type ? "AC_NTFY_VAL" : "AC_TRIG_VAL",
+ acc->asid, acc->vfid, acc->engine_class, acc->engine_instance,
+ granularity_in_byte(acc->granularity) / SZ_1K,
+ sub_granularity_in_byte(acc->granularity) / SZ_1K,
+ acc->sub_granularity, acc->va_range_base);
+}
+
+static struct xe_vma *get_acc_vma(struct xe_vm *vm, struct acc *acc)
+{
+ u64 page_va = acc->va_range_base + (ffs(acc->sub_granularity) - 1) *
+ sub_granularity_in_byte(acc->granularity);
+
+ return xe_vm_find_overlapping_vma(vm, page_va, SZ_4K);
+}
+
+static int handle_acc(struct xe_gt *gt, struct acc *acc)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct drm_exec exec;
+ struct xe_vm *vm;
+ struct xe_vma *vma;
+ int ret = 0;
+
+ /* We only support ACC_TRIGGER at the moment */
+ if (acc->access_type != ACC_TRIGGER)
+ return -EINVAL;
+
+ /* ASID to VM */
+ mutex_lock(&xe->usm.lock);
+ vm = xa_load(&xe->usm.asid_to_vm, acc->asid);
+ if (vm)
+ xe_vm_get(vm);
+ mutex_unlock(&xe->usm.lock);
+ if (!vm || !xe_vm_in_fault_mode(vm))
+ return -EINVAL;
+
+ down_read(&vm->lock);
+
+ /* Lookup VMA */
+ vma = get_acc_vma(vm, acc);
+ if (!vma) {
+ ret = -EINVAL;
+ goto unlock_vm;
+ }
+
+ trace_xe_vma_acc(vma);
+
+ /* Userptr or null can't be migrated, nothing to do */
+ if (xe_vma_has_no_bo(vma))
+ goto unlock_vm;
+
+ /* Lock VM and BOs dma-resv */
+ drm_exec_init(&exec, 0, 0);
+ drm_exec_until_all_locked(&exec) {
+ ret = xe_pf_begin(&exec, vma, true, tile->id);
+ drm_exec_retry_on_contention(&exec);
+ if (ret)
+ break;
+ }
+
+ drm_exec_fini(&exec);
+unlock_vm:
+ up_read(&vm->lock);
+ xe_vm_put(vm);
+
+ return ret;
+}
+
+#define make_u64(hi__, low__) ((u64)(hi__) << 32 | (u64)(low__))
+
+#define ACC_MSG_LEN_DW 4
+
+static bool get_acc(struct acc_queue *acc_queue, struct acc *acc)
+{
+ const struct xe_guc_acc_desc *desc;
+ bool ret = false;
+
+ spin_lock(&acc_queue->lock);
+ if (acc_queue->head != acc_queue->tail) {
+ desc = (const struct xe_guc_acc_desc *)
+ (acc_queue->data + acc_queue->head);
+
+ acc->granularity = FIELD_GET(ACC_GRANULARITY, desc->dw2);
+ acc->sub_granularity = FIELD_GET(ACC_SUBG_HI, desc->dw1) << 31 |
+ FIELD_GET(ACC_SUBG_LO, desc->dw0);
+ acc->engine_class = FIELD_GET(ACC_ENG_CLASS, desc->dw1);
+ acc->engine_instance = FIELD_GET(ACC_ENG_INSTANCE, desc->dw1);
+ acc->asid = FIELD_GET(ACC_ASID, desc->dw1);
+ acc->vfid = FIELD_GET(ACC_VFID, desc->dw2);
+ acc->access_type = FIELD_GET(ACC_TYPE, desc->dw0);
+ acc->va_range_base = make_u64(desc->dw3 & ACC_VIRTUAL_ADDR_RANGE_HI,
+ desc->dw2 & ACC_VIRTUAL_ADDR_RANGE_LO);
+
+ acc_queue->head = (acc_queue->head + ACC_MSG_LEN_DW) %
+ ACC_QUEUE_NUM_DW;
+ ret = true;
+ }
+ spin_unlock(&acc_queue->lock);
+
+ return ret;
+}
+
+static void acc_queue_work_func(struct work_struct *w)
+{
+ struct acc_queue *acc_queue = container_of(w, struct acc_queue, worker);
+ struct xe_gt *gt = acc_queue->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ struct acc acc = {};
+ unsigned long threshold;
+ int ret;
+
+ threshold = jiffies + msecs_to_jiffies(USM_QUEUE_MAX_RUNTIME_MS);
+
+ while (get_acc(acc_queue, &acc)) {
+ ret = handle_acc(gt, &acc);
+ if (unlikely(ret)) {
+ print_acc(xe, &acc);
+ drm_warn(&xe->drm, "ACC: Unsuccessful %d\n", ret);
+ }
+
+ if (time_after(jiffies, threshold) &&
+ acc_queue->head != acc_queue->tail) {
+ queue_work(gt->usm.acc_wq, w);
+ break;
+ }
+ }
+}
+
+static bool acc_queue_full(struct acc_queue *acc_queue)
+{
+ lockdep_assert_held(&acc_queue->lock);
+
+ return CIRC_SPACE(acc_queue->tail, acc_queue->head, ACC_QUEUE_NUM_DW) <=
+ ACC_MSG_LEN_DW;
+}
+
+int xe_guc_access_counter_notify_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ struct acc_queue *acc_queue;
+ u32 asid;
+ bool full;
+
+ if (unlikely(len != ACC_MSG_LEN_DW))
+ return -EPROTO;
+
+ asid = FIELD_GET(ACC_ASID, msg[1]);
+ acc_queue = &gt->usm.acc_queue[asid % NUM_ACC_QUEUE];
+
+ spin_lock(&acc_queue->lock);
+ full = acc_queue_full(acc_queue);
+ if (!full) {
+ memcpy(acc_queue->data + acc_queue->tail, msg,
+ len * sizeof(u32));
+ acc_queue->tail = (acc_queue->tail + len) % ACC_QUEUE_NUM_DW;
+ queue_work(gt->usm.acc_wq, &acc_queue->worker);
+ } else {
+ drm_warn(&gt_to_xe(gt)->drm, "ACC Queue full, dropping ACC");
+ }
+ spin_unlock(&acc_queue->lock);
+
+ return full ? -ENOSPC : 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.h b/drivers/gpu/drm/xe/xe_gt_pagefault.h
new file mode 100644
index 000000000000..839c065a5e4c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_pagefault.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_PAGEFAULT_H_
+#define _XE_GT_PAGEFAULT_H_
+
+#include <linux/types.h>
+
+struct xe_gt;
+struct xe_guc;
+
+int xe_gt_pagefault_init(struct xe_gt *gt);
+void xe_gt_pagefault_reset(struct xe_gt *gt);
+int xe_guc_pagefault_handler(struct xe_guc *guc, u32 *msg, u32 len);
+int xe_guc_access_counter_notify_handler(struct xe_guc *guc, u32 *msg, u32 len);
+
+#endif /* _XE_GT_PAGEFAULT_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_printk.h b/drivers/gpu/drm/xe/xe_gt_printk.h
new file mode 100644
index 000000000000..5991bcadd47e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_printk.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_PRINTK_H_
+#define _XE_GT_PRINTK_H_
+
+#include <drm/drm_print.h>
+
+#include "xe_device_types.h"
+
+#define xe_gt_printk(_gt, _level, _fmt, ...) \
+ drm_##_level(&gt_to_xe(_gt)->drm, "GT%u: " _fmt, (_gt)->info.id, ##__VA_ARGS__)
+
+#define xe_gt_err(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), err, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_warn(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), warn, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_notice(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), notice, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_info(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), info, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_dbg(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), dbg, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_err_ratelimited(_gt, _fmt, ...) \
+ xe_gt_printk((_gt), err_ratelimited, _fmt, ##__VA_ARGS__)
+
+#define xe_gt_WARN(_gt, _condition, _fmt, ...) \
+ drm_WARN(&gt_to_xe(_gt)->drm, _condition, "GT%u: " _fmt, (_gt)->info.id, ##__VA_ARGS__)
+
+#define xe_gt_WARN_ONCE(_gt, _condition, _fmt, ...) \
+ drm_WARN_ONCE(&gt_to_xe(_gt)->drm, _condition, "GT%u: " _fmt, (_gt)->info.id, ##__VA_ARGS__)
+
+#define xe_gt_WARN_ON(_gt, _condition) \
+ xe_gt_WARN((_gt), _condition, "%s(%s)", "gt_WARN_ON", __stringify(_condition))
+
+#define xe_gt_WARN_ON_ONCE(_gt, _condition) \
+ xe_gt_WARN_ONCE((_gt), _condition, "%s(%s)", "gt_WARN_ON_ONCE", __stringify(_condition))
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_sysfs.c b/drivers/gpu/drm/xe/xe_gt_sysfs.c
new file mode 100644
index 000000000000..c69d2e8a0fe1
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_sysfs.c
@@ -0,0 +1,61 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_sysfs.h"
+
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+
+#include <drm/drm_managed.h>
+
+#include "xe_gt.h"
+
+static void xe_gt_sysfs_kobj_release(struct kobject *kobj)
+{
+ kfree(kobj);
+}
+
+static const struct kobj_type xe_gt_sysfs_kobj_type = {
+ .release = xe_gt_sysfs_kobj_release,
+ .sysfs_ops = &kobj_sysfs_ops,
+};
+
+static void gt_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_gt *gt = arg;
+
+ kobject_put(gt->sysfs);
+}
+
+void xe_gt_sysfs_init(struct xe_gt *gt)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct kobj_gt *kg;
+ int err;
+
+ kg = kzalloc(sizeof(*kg), GFP_KERNEL);
+ if (!kg)
+ return;
+
+ kobject_init(&kg->base, &xe_gt_sysfs_kobj_type);
+ kg->gt = gt;
+
+ err = kobject_add(&kg->base, tile->sysfs, "gt%d", gt->info.id);
+ if (err) {
+ drm_warn(&xe->drm, "failed to add GT sysfs directory, err: %d\n", err);
+ kobject_put(&kg->base);
+ return;
+ }
+
+ gt->sysfs = &kg->base;
+
+ err = drmm_add_action_or_reset(&xe->drm, gt_sysfs_fini, gt);
+ if (err) {
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+ return;
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_sysfs.h b/drivers/gpu/drm/xe/xe_gt_sysfs.h
new file mode 100644
index 000000000000..e3ec278ca0be
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_sysfs.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_SYSFS_H_
+#define _XE_GT_SYSFS_H_
+
+#include "xe_gt_sysfs_types.h"
+
+void xe_gt_sysfs_init(struct xe_gt *gt);
+
+static inline struct xe_gt *
+kobj_to_gt(struct kobject *kobj)
+{
+ return container_of(kobj, struct kobj_gt, base)->gt;
+}
+
+#endif /* _XE_GT_SYSFS_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_sysfs_types.h b/drivers/gpu/drm/xe/xe_gt_sysfs_types.h
new file mode 100644
index 000000000000..d3bc6b83360f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_sysfs_types.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_SYSFS_TYPES_H_
+#define _XE_GT_SYSFS_TYPES_H_
+
+#include <linux/kobject.h>
+
+struct xe_gt;
+
+/**
+ * struct kobj_gt - A GT's kobject struct that connects the kobject and the GT
+ *
+ * When dealing with multiple GTs, this struct helps to understand which GT
+ * needs to be addressed on a given sysfs call.
+ */
+struct kobj_gt {
+ /** @base: The actual kobject */
+ struct kobject base;
+ /** @gt: A pointer to the GT itself */
+ struct xe_gt *gt;
+};
+
+#endif /* _XE_GT_SYSFS_TYPES_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.c b/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.c
new file mode 100644
index 000000000000..63d640591a52
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.c
@@ -0,0 +1,251 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_managed.h>
+
+#include <regs/xe_gt_regs.h>
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_sysfs.h"
+#include "xe_gt_throttle_sysfs.h"
+#include "xe_mmio.h"
+
+/**
+ * DOC: Xe GT Throttle
+ *
+ * Provides sysfs entries for frequency throttle reasons in GT
+ *
+ * device/gt#/freq0/throttle/status - Overall status
+ * device/gt#/freq0/throttle/reason_pl1 - Frequency throttle due to PL1
+ * device/gt#/freq0/throttle/reason_pl2 - Frequency throttle due to PL2
+ * device/gt#/freq0/throttle/reason_pl4 - Frequency throttle due to PL4, Iccmax etc.
+ * device/gt#/freq0/throttle/reason_thermal - Frequency throttle due to thermal
+ * device/gt#/freq0/throttle/reason_prochot - Frequency throttle due to prochot
+ * device/gt#/freq0/throttle/reason_ratl - Frequency throttle due to RATL
+ * device/gt#/freq0/throttle/reason_vr_thermalert - Frequency throttle due to VR THERMALERT
+ * device/gt#/freq0/throttle/reason_vr_tdc - Frequency throttle due to VR TDC
+ */
+
+static struct xe_gt *
+dev_to_gt(struct device *dev)
+{
+ return kobj_to_gt(dev->kobj.parent);
+}
+
+static u32 read_perf_limit_reasons(struct xe_gt *gt)
+{
+ u32 reg;
+
+ if (xe_gt_is_media_type(gt))
+ reg = xe_mmio_read32(gt, MTL_MEDIA_PERF_LIMIT_REASONS);
+ else
+ reg = xe_mmio_read32(gt, GT0_PERF_LIMIT_REASONS);
+
+ return reg;
+}
+
+static u32 read_status(struct xe_gt *gt)
+{
+ u32 status = read_perf_limit_reasons(gt) & GT0_PERF_LIMIT_REASONS_MASK;
+
+ return status;
+}
+
+static u32 read_reason_pl1(struct xe_gt *gt)
+{
+ u32 pl1 = read_perf_limit_reasons(gt) & POWER_LIMIT_1_MASK;
+
+ return pl1;
+}
+
+static u32 read_reason_pl2(struct xe_gt *gt)
+{
+ u32 pl2 = read_perf_limit_reasons(gt) & POWER_LIMIT_2_MASK;
+
+ return pl2;
+}
+
+static u32 read_reason_pl4(struct xe_gt *gt)
+{
+ u32 pl4 = read_perf_limit_reasons(gt) & POWER_LIMIT_4_MASK;
+
+ return pl4;
+}
+
+static u32 read_reason_thermal(struct xe_gt *gt)
+{
+ u32 thermal = read_perf_limit_reasons(gt) & THERMAL_LIMIT_MASK;
+
+ return thermal;
+}
+
+static u32 read_reason_prochot(struct xe_gt *gt)
+{
+ u32 prochot = read_perf_limit_reasons(gt) & PROCHOT_MASK;
+
+ return prochot;
+}
+
+static u32 read_reason_ratl(struct xe_gt *gt)
+{
+ u32 ratl = read_perf_limit_reasons(gt) & RATL_MASK;
+
+ return ratl;
+}
+
+static u32 read_reason_vr_thermalert(struct xe_gt *gt)
+{
+ u32 thermalert = read_perf_limit_reasons(gt) & VR_THERMALERT_MASK;
+
+ return thermalert;
+}
+
+static u32 read_reason_vr_tdc(struct xe_gt *gt)
+{
+ u32 tdc = read_perf_limit_reasons(gt) & VR_TDC_MASK;
+
+ return tdc;
+}
+
+static ssize_t status_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool status = !!read_status(gt);
+
+ return sysfs_emit(buff, "%u\n", status);
+}
+static DEVICE_ATTR_RO(status);
+
+static ssize_t reason_pl1_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool pl1 = !!read_reason_pl1(gt);
+
+ return sysfs_emit(buff, "%u\n", pl1);
+}
+static DEVICE_ATTR_RO(reason_pl1);
+
+static ssize_t reason_pl2_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool pl2 = !!read_reason_pl2(gt);
+
+ return sysfs_emit(buff, "%u\n", pl2);
+}
+static DEVICE_ATTR_RO(reason_pl2);
+
+static ssize_t reason_pl4_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool pl4 = !!read_reason_pl4(gt);
+
+ return sysfs_emit(buff, "%u\n", pl4);
+}
+static DEVICE_ATTR_RO(reason_pl4);
+
+static ssize_t reason_thermal_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool thermal = !!read_reason_thermal(gt);
+
+ return sysfs_emit(buff, "%u\n", thermal);
+}
+static DEVICE_ATTR_RO(reason_thermal);
+
+static ssize_t reason_prochot_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool prochot = !!read_reason_prochot(gt);
+
+ return sysfs_emit(buff, "%u\n", prochot);
+}
+static DEVICE_ATTR_RO(reason_prochot);
+
+static ssize_t reason_ratl_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool ratl = !!read_reason_ratl(gt);
+
+ return sysfs_emit(buff, "%u\n", ratl);
+}
+static DEVICE_ATTR_RO(reason_ratl);
+
+static ssize_t reason_vr_thermalert_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool thermalert = !!read_reason_vr_thermalert(gt);
+
+ return sysfs_emit(buff, "%u\n", thermalert);
+}
+static DEVICE_ATTR_RO(reason_vr_thermalert);
+
+static ssize_t reason_vr_tdc_show(struct device *dev,
+ struct device_attribute *attr,
+ char *buff)
+{
+ struct xe_gt *gt = dev_to_gt(dev);
+ bool tdc = !!read_reason_vr_tdc(gt);
+
+ return sysfs_emit(buff, "%u\n", tdc);
+}
+static DEVICE_ATTR_RO(reason_vr_tdc);
+
+static struct attribute *throttle_attrs[] = {
+ &dev_attr_status.attr,
+ &dev_attr_reason_pl1.attr,
+ &dev_attr_reason_pl2.attr,
+ &dev_attr_reason_pl4.attr,
+ &dev_attr_reason_thermal.attr,
+ &dev_attr_reason_prochot.attr,
+ &dev_attr_reason_ratl.attr,
+ &dev_attr_reason_vr_thermalert.attr,
+ &dev_attr_reason_vr_tdc.attr,
+ NULL
+};
+
+static const struct attribute_group throttle_group_attrs = {
+ .name = "throttle",
+ .attrs = throttle_attrs,
+};
+
+static void gt_throttle_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_gt *gt = arg;
+
+ sysfs_remove_group(gt->freq, &throttle_group_attrs);
+}
+
+void xe_gt_throttle_sysfs_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int err;
+
+ err = sysfs_create_group(gt->freq, &throttle_group_attrs);
+ if (err) {
+ drm_warn(&xe->drm, "failed to register throttle sysfs, err: %d\n", err);
+ return;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, gt_throttle_sysfs_fini, gt);
+ if (err)
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.h b/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.h
new file mode 100644
index 000000000000..3ecfd4beffe1
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_throttle_sysfs.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_THROTTLE_SYSFS_H_
+#define _XE_GT_THROTTLE_SYSFS_H_
+
+#include <drm/drm_managed.h>
+
+struct xe_gt;
+
+void xe_gt_throttle_sysfs_init(struct xe_gt *gt);
+
+#endif /* _XE_GT_THROTTLE_SYSFS_H_ */
+
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
new file mode 100644
index 000000000000..7eef23a00d77
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
@@ -0,0 +1,406 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_gt_tlb_invalidation.h"
+
+#include "abi/guc_actions_abi.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_guc_ct.h"
+#include "xe_trace.h"
+
+#define TLB_TIMEOUT (HZ / 4)
+
+static void xe_gt_tlb_fence_timeout(struct work_struct *work)
+{
+ struct xe_gt *gt = container_of(work, struct xe_gt,
+ tlb_invalidation.fence_tdr.work);
+ struct xe_gt_tlb_invalidation_fence *fence, *next;
+
+ spin_lock_irq(&gt->tlb_invalidation.pending_lock);
+ list_for_each_entry_safe(fence, next,
+ &gt->tlb_invalidation.pending_fences, link) {
+ s64 since_inval_ms = ktime_ms_delta(ktime_get(),
+ fence->invalidation_time);
+
+ if (msecs_to_jiffies(since_inval_ms) < TLB_TIMEOUT)
+ break;
+
+ trace_xe_gt_tlb_invalidation_fence_timeout(fence);
+ drm_err(&gt_to_xe(gt)->drm, "gt%d: TLB invalidation fence timeout, seqno=%d recv=%d",
+ gt->info.id, fence->seqno, gt->tlb_invalidation.seqno_recv);
+
+ list_del(&fence->link);
+ fence->base.error = -ETIME;
+ dma_fence_signal(&fence->base);
+ dma_fence_put(&fence->base);
+ }
+ if (!list_empty(&gt->tlb_invalidation.pending_fences))
+ queue_delayed_work(system_wq,
+ &gt->tlb_invalidation.fence_tdr,
+ TLB_TIMEOUT);
+ spin_unlock_irq(&gt->tlb_invalidation.pending_lock);
+}
+
+/**
+ * xe_gt_tlb_invalidation_init - Initialize GT TLB invalidation state
+ * @gt: graphics tile
+ *
+ * Initialize GT TLB invalidation state, purely software initialization, should
+ * be called once during driver load.
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_gt_tlb_invalidation_init(struct xe_gt *gt)
+{
+ gt->tlb_invalidation.seqno = 1;
+ INIT_LIST_HEAD(&gt->tlb_invalidation.pending_fences);
+ spin_lock_init(&gt->tlb_invalidation.pending_lock);
+ spin_lock_init(&gt->tlb_invalidation.lock);
+ gt->tlb_invalidation.fence_context = dma_fence_context_alloc(1);
+ INIT_DELAYED_WORK(&gt->tlb_invalidation.fence_tdr,
+ xe_gt_tlb_fence_timeout);
+
+ return 0;
+}
+
+static void
+__invalidation_fence_signal(struct xe_gt_tlb_invalidation_fence *fence)
+{
+ trace_xe_gt_tlb_invalidation_fence_signal(fence);
+ dma_fence_signal(&fence->base);
+ dma_fence_put(&fence->base);
+}
+
+static void
+invalidation_fence_signal(struct xe_gt_tlb_invalidation_fence *fence)
+{
+ list_del(&fence->link);
+ __invalidation_fence_signal(fence);
+}
+
+/**
+ * xe_gt_tlb_invalidation_reset - Initialize GT TLB invalidation reset
+ * @gt: graphics tile
+ *
+ * Signal any pending invalidation fences, should be called during a GT reset
+ */
+void xe_gt_tlb_invalidation_reset(struct xe_gt *gt)
+{
+ struct xe_gt_tlb_invalidation_fence *fence, *next;
+ struct xe_guc *guc = &gt->uc.guc;
+ int pending_seqno;
+
+ /*
+ * CT channel is already disabled at this point. No new TLB requests can
+ * appear.
+ */
+
+ mutex_lock(&gt->uc.guc.ct.lock);
+ spin_lock_irq(&gt->tlb_invalidation.pending_lock);
+ cancel_delayed_work(&gt->tlb_invalidation.fence_tdr);
+ /*
+ * We might have various kworkers waiting for TLB flushes to complete
+ * which are not tracked with an explicit TLB fence, however at this
+ * stage that will never happen since the CT is already disabled, so
+ * make sure we signal them here under the assumption that we have
+ * completed a full GT reset.
+ */
+ if (gt->tlb_invalidation.seqno == 1)
+ pending_seqno = TLB_INVALIDATION_SEQNO_MAX - 1;
+ else
+ pending_seqno = gt->tlb_invalidation.seqno - 1;
+ WRITE_ONCE(gt->tlb_invalidation.seqno_recv, pending_seqno);
+ wake_up_all(&guc->ct.wq);
+
+ list_for_each_entry_safe(fence, next,
+ &gt->tlb_invalidation.pending_fences, link)
+ invalidation_fence_signal(fence);
+ spin_unlock_irq(&gt->tlb_invalidation.pending_lock);
+ mutex_unlock(&gt->uc.guc.ct.lock);
+}
+
+static bool tlb_invalidation_seqno_past(struct xe_gt *gt, int seqno)
+{
+ int seqno_recv = READ_ONCE(gt->tlb_invalidation.seqno_recv);
+
+ if (seqno - seqno_recv < -(TLB_INVALIDATION_SEQNO_MAX / 2))
+ return false;
+
+ if (seqno - seqno_recv > (TLB_INVALIDATION_SEQNO_MAX / 2))
+ return true;
+
+ return seqno_recv >= seqno;
+}
+
+static int send_tlb_invalidation(struct xe_guc *guc,
+ struct xe_gt_tlb_invalidation_fence *fence,
+ u32 *action, int len)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ int seqno;
+ int ret;
+
+ /*
+ * XXX: The seqno algorithm relies on TLB invalidation being processed
+ * in order which they currently are, if that changes the algorithm will
+ * need to be updated.
+ */
+
+ mutex_lock(&guc->ct.lock);
+ seqno = gt->tlb_invalidation.seqno;
+ if (fence) {
+ fence->seqno = seqno;
+ trace_xe_gt_tlb_invalidation_fence_send(fence);
+ }
+ action[1] = seqno;
+ ret = xe_guc_ct_send_locked(&guc->ct, action, len,
+ G2H_LEN_DW_TLB_INVALIDATE, 1);
+ if (!ret && fence) {
+ spin_lock_irq(&gt->tlb_invalidation.pending_lock);
+ /*
+ * We haven't actually published the TLB fence as per
+ * pending_fences, but in theory our seqno could have already
+ * been written as we acquired the pending_lock. In such a case
+ * we can just go ahead and signal the fence here.
+ */
+ if (tlb_invalidation_seqno_past(gt, seqno)) {
+ __invalidation_fence_signal(fence);
+ } else {
+ fence->invalidation_time = ktime_get();
+ list_add_tail(&fence->link,
+ &gt->tlb_invalidation.pending_fences);
+
+ if (list_is_singular(&gt->tlb_invalidation.pending_fences))
+ queue_delayed_work(system_wq,
+ &gt->tlb_invalidation.fence_tdr,
+ TLB_TIMEOUT);
+ }
+ spin_unlock_irq(&gt->tlb_invalidation.pending_lock);
+ } else if (ret < 0 && fence) {
+ __invalidation_fence_signal(fence);
+ }
+ if (!ret) {
+ gt->tlb_invalidation.seqno = (gt->tlb_invalidation.seqno + 1) %
+ TLB_INVALIDATION_SEQNO_MAX;
+ if (!gt->tlb_invalidation.seqno)
+ gt->tlb_invalidation.seqno = 1;
+ ret = seqno;
+ }
+ mutex_unlock(&guc->ct.lock);
+
+ return ret;
+}
+
+#define MAKE_INVAL_OP(type) ((type << XE_GUC_TLB_INVAL_TYPE_SHIFT) | \
+ XE_GUC_TLB_INVAL_MODE_HEAVY << XE_GUC_TLB_INVAL_MODE_SHIFT | \
+ XE_GUC_TLB_INVAL_FLUSH_CACHE)
+
+/**
+ * xe_gt_tlb_invalidation_guc - Issue a TLB invalidation on this GT for the GuC
+ * @gt: graphics tile
+ *
+ * Issue a TLB invalidation for the GuC. Completion of TLB is asynchronous and
+ * caller can use seqno + xe_gt_tlb_invalidation_wait to wait for completion.
+ *
+ * Return: Seqno which can be passed to xe_gt_tlb_invalidation_wait on success,
+ * negative error code on error.
+ */
+int xe_gt_tlb_invalidation_guc(struct xe_gt *gt)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_TLB_INVALIDATION,
+ 0, /* seqno, replaced in send_tlb_invalidation */
+ MAKE_INVAL_OP(XE_GUC_TLB_INVAL_GUC),
+ };
+
+ return send_tlb_invalidation(&gt->uc.guc, NULL, action,
+ ARRAY_SIZE(action));
+}
+
+/**
+ * xe_gt_tlb_invalidation_vma - Issue a TLB invalidation on this GT for a VMA
+ * @gt: graphics tile
+ * @fence: invalidation fence which will be signal on TLB invalidation
+ * completion, can be NULL
+ * @vma: VMA to invalidate
+ *
+ * Issue a range based TLB invalidation if supported, if not fallback to a full
+ * TLB invalidation. Completion of TLB is asynchronous and caller can either use
+ * the invalidation fence or seqno + xe_gt_tlb_invalidation_wait to wait for
+ * completion.
+ *
+ * Return: Seqno which can be passed to xe_gt_tlb_invalidation_wait on success,
+ * negative error code on error.
+ */
+int xe_gt_tlb_invalidation_vma(struct xe_gt *gt,
+ struct xe_gt_tlb_invalidation_fence *fence,
+ struct xe_vma *vma)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+#define MAX_TLB_INVALIDATION_LEN 7
+ u32 action[MAX_TLB_INVALIDATION_LEN];
+ int len = 0;
+
+ xe_gt_assert(gt, vma);
+
+ action[len++] = XE_GUC_ACTION_TLB_INVALIDATION;
+ action[len++] = 0; /* seqno, replaced in send_tlb_invalidation */
+ if (!xe->info.has_range_tlb_invalidation) {
+ action[len++] = MAKE_INVAL_OP(XE_GUC_TLB_INVAL_FULL);
+ } else {
+ u64 start = xe_vma_start(vma);
+ u64 length = xe_vma_size(vma);
+ u64 align, end;
+
+ if (length < SZ_4K)
+ length = SZ_4K;
+
+ /*
+ * We need to invalidate a higher granularity if start address
+ * is not aligned to length. When start is not aligned with
+ * length we need to find the length large enough to create an
+ * address mask covering the required range.
+ */
+ align = roundup_pow_of_two(length);
+ start = ALIGN_DOWN(xe_vma_start(vma), align);
+ end = ALIGN(xe_vma_end(vma), align);
+ length = align;
+ while (start + length < end) {
+ length <<= 1;
+ start = ALIGN_DOWN(xe_vma_start(vma), length);
+ }
+
+ /*
+ * Minimum invalidation size for a 2MB page that the hardware
+ * expects is 16MB
+ */
+ if (length >= SZ_2M) {
+ length = max_t(u64, SZ_16M, length);
+ start = ALIGN_DOWN(xe_vma_start(vma), length);
+ }
+
+ xe_gt_assert(gt, length >= SZ_4K);
+ xe_gt_assert(gt, is_power_of_2(length));
+ xe_gt_assert(gt, !(length & GENMASK(ilog2(SZ_16M) - 1, ilog2(SZ_2M) + 1)));
+ xe_gt_assert(gt, IS_ALIGNED(start, length));
+
+ action[len++] = MAKE_INVAL_OP(XE_GUC_TLB_INVAL_PAGE_SELECTIVE);
+ action[len++] = xe_vma_vm(vma)->usm.asid;
+ action[len++] = lower_32_bits(start);
+ action[len++] = upper_32_bits(start);
+ action[len++] = ilog2(length) - ilog2(SZ_4K);
+ }
+
+ xe_gt_assert(gt, len <= MAX_TLB_INVALIDATION_LEN);
+
+ return send_tlb_invalidation(&gt->uc.guc, fence, action, len);
+}
+
+/**
+ * xe_gt_tlb_invalidation_wait - Wait for TLB to complete
+ * @gt: graphics tile
+ * @seqno: seqno to wait which was returned from xe_gt_tlb_invalidation
+ *
+ * Wait for 200ms for a TLB invalidation to complete, in practice we always
+ * should receive the TLB invalidation within 200ms.
+ *
+ * Return: 0 on success, -ETIME on TLB invalidation timeout
+ */
+int xe_gt_tlb_invalidation_wait(struct xe_gt *gt, int seqno)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_guc *guc = &gt->uc.guc;
+ struct drm_printer p = drm_err_printer(__func__);
+ int ret;
+
+ /*
+ * XXX: See above, this algorithm only works if seqno are always in
+ * order
+ */
+ ret = wait_event_timeout(guc->ct.wq,
+ tlb_invalidation_seqno_past(gt, seqno),
+ TLB_TIMEOUT);
+ if (!ret) {
+ drm_err(&xe->drm, "gt%d: TLB invalidation time'd out, seqno=%d, recv=%d\n",
+ gt->info.id, seqno, gt->tlb_invalidation.seqno_recv);
+ xe_guc_ct_print(&guc->ct, &p, true);
+ return -ETIME;
+ }
+
+ return 0;
+}
+
+/**
+ * xe_guc_tlb_invalidation_done_handler - TLB invalidation done handler
+ * @guc: guc
+ * @msg: message indicating TLB invalidation done
+ * @len: length of message
+ *
+ * Parse seqno of TLB invalidation, wake any waiters for seqno, and signal any
+ * invalidation fences for seqno. Algorithm for this depends on seqno being
+ * received in-order and asserts this assumption.
+ *
+ * Return: 0 on success, -EPROTO for malformed messages.
+ */
+int xe_guc_tlb_invalidation_done_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ struct xe_gt_tlb_invalidation_fence *fence, *next;
+ unsigned long flags;
+
+ if (unlikely(len != 1))
+ return -EPROTO;
+
+ /*
+ * This can also be run both directly from the IRQ handler and also in
+ * process_g2h_msg(). Only one may process any individual CT message,
+ * however the order they are processed here could result in skipping a
+ * seqno. To handle that we just process all the seqnos from the last
+ * seqno_recv up to and including the one in msg[0]. The delta should be
+ * very small so there shouldn't be much of pending_fences we actually
+ * need to iterate over here.
+ *
+ * From GuC POV we expect the seqnos to always appear in-order, so if we
+ * see something later in the timeline we can be sure that anything
+ * appearing earlier has already signalled, just that we have yet to
+ * officially process the CT message like if racing against
+ * process_g2h_msg().
+ */
+ spin_lock_irqsave(&gt->tlb_invalidation.pending_lock, flags);
+ if (tlb_invalidation_seqno_past(gt, msg[0])) {
+ spin_unlock_irqrestore(&gt->tlb_invalidation.pending_lock, flags);
+ return 0;
+ }
+
+ /*
+ * wake_up_all() and wait_event_timeout() already have the correct
+ * barriers.
+ */
+ WRITE_ONCE(gt->tlb_invalidation.seqno_recv, msg[0]);
+ wake_up_all(&guc->ct.wq);
+
+ list_for_each_entry_safe(fence, next,
+ &gt->tlb_invalidation.pending_fences, link) {
+ trace_xe_gt_tlb_invalidation_fence_recv(fence);
+
+ if (!tlb_invalidation_seqno_past(gt, fence->seqno))
+ break;
+
+ invalidation_fence_signal(fence);
+ }
+
+ if (!list_empty(&gt->tlb_invalidation.pending_fences))
+ mod_delayed_work(system_wq,
+ &gt->tlb_invalidation.fence_tdr,
+ TLB_TIMEOUT);
+ else
+ cancel_delayed_work(&gt->tlb_invalidation.fence_tdr);
+
+ spin_unlock_irqrestore(&gt->tlb_invalidation.pending_lock, flags);
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h
new file mode 100644
index 000000000000..b333c1709397
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_TLB_INVALIDATION_H_
+#define _XE_GT_TLB_INVALIDATION_H_
+
+#include <linux/types.h>
+
+#include "xe_gt_tlb_invalidation_types.h"
+
+struct xe_gt;
+struct xe_guc;
+struct xe_vma;
+
+int xe_gt_tlb_invalidation_init(struct xe_gt *gt);
+void xe_gt_tlb_invalidation_reset(struct xe_gt *gt);
+int xe_gt_tlb_invalidation_guc(struct xe_gt *gt);
+int xe_gt_tlb_invalidation_vma(struct xe_gt *gt,
+ struct xe_gt_tlb_invalidation_fence *fence,
+ struct xe_vma *vma);
+int xe_gt_tlb_invalidation_wait(struct xe_gt *gt, int seqno);
+int xe_guc_tlb_invalidation_done_handler(struct xe_guc *guc, u32 *msg, u32 len);
+
+#endif /* _XE_GT_TLB_INVALIDATION_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation_types.h b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation_types.h
new file mode 100644
index 000000000000..934c828efe31
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation_types.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_TLB_INVALIDATION_TYPES_H_
+#define _XE_GT_TLB_INVALIDATION_TYPES_H_
+
+#include <linux/dma-fence.h>
+
+/**
+ * struct xe_gt_tlb_invalidation_fence - XE GT TLB invalidation fence
+ *
+ * Optionally passed to xe_gt_tlb_invalidation and will be signaled upon TLB
+ * invalidation completion.
+ */
+struct xe_gt_tlb_invalidation_fence {
+ /** @base: dma fence base */
+ struct dma_fence base;
+ /** @link: link into list of pending tlb fences */
+ struct list_head link;
+ /** @seqno: seqno of TLB invalidation to signal fence one */
+ int seqno;
+ /** @invalidation_time: time of TLB invalidation */
+ ktime_t invalidation_time;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_gt_topology.c b/drivers/gpu/drm/xe/xe_gt_topology.c
new file mode 100644
index 000000000000..a8d7f272c30a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_topology.c
@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_gt_topology.h"
+
+#include <linux/bitmap.h>
+
+#include "regs/xe_gt_regs.h"
+#include "xe_gt.h"
+#include "xe_mmio.h"
+
+#define XE_MAX_DSS_FUSE_BITS (32 * XE_MAX_DSS_FUSE_REGS)
+#define XE_MAX_EU_FUSE_BITS (32 * XE_MAX_EU_FUSE_REGS)
+
+static void
+load_dss_mask(struct xe_gt *gt, xe_dss_mask_t mask, int numregs, ...)
+{
+ va_list argp;
+ u32 fuse_val[XE_MAX_DSS_FUSE_REGS] = {};
+ int i;
+
+ if (drm_WARN_ON(&gt_to_xe(gt)->drm, numregs > XE_MAX_DSS_FUSE_REGS))
+ numregs = XE_MAX_DSS_FUSE_REGS;
+
+ va_start(argp, numregs);
+ for (i = 0; i < numregs; i++)
+ fuse_val[i] = xe_mmio_read32(gt, va_arg(argp, struct xe_reg));
+ va_end(argp);
+
+ bitmap_from_arr32(mask, fuse_val, numregs * 32);
+}
+
+static void
+load_eu_mask(struct xe_gt *gt, xe_eu_mask_t mask)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 reg_val = xe_mmio_read32(gt, XELP_EU_ENABLE);
+ u32 val = 0;
+ int i;
+
+ BUILD_BUG_ON(XE_MAX_EU_FUSE_REGS > 1);
+
+ /*
+ * Pre-Xe_HP platforms inverted the bit meaning (disable instead
+ * of enable).
+ */
+ if (GRAPHICS_VERx100(xe) < 1250)
+ reg_val = ~reg_val & XELP_EU_MASK;
+
+ /* On PVC, one bit = one EU */
+ if (GRAPHICS_VERx100(xe) == 1260) {
+ val = reg_val;
+ } else {
+ /* All other platforms, one bit = 2 EU */
+ for (i = 0; i < fls(reg_val); i++)
+ if (reg_val & BIT(i))
+ val |= 0x3 << 2 * i;
+ }
+
+ bitmap_from_arr32(mask, &val, XE_MAX_EU_FUSE_BITS);
+}
+
+static void
+get_num_dss_regs(struct xe_device *xe, int *geometry_regs, int *compute_regs)
+{
+ if (GRAPHICS_VER(xe) > 20) {
+ *geometry_regs = 3;
+ *compute_regs = 3;
+ } else if (GRAPHICS_VERx100(xe) == 1260) {
+ *geometry_regs = 0;
+ *compute_regs = 2;
+ } else if (GRAPHICS_VERx100(xe) >= 1250) {
+ *geometry_regs = 1;
+ *compute_regs = 1;
+ } else {
+ *geometry_regs = 1;
+ *compute_regs = 0;
+ }
+}
+
+void
+xe_gt_topology_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct drm_printer p = drm_debug_printer("GT topology");
+ int num_geometry_regs, num_compute_regs;
+
+ get_num_dss_regs(xe, &num_geometry_regs, &num_compute_regs);
+
+ /*
+ * Register counts returned shouldn't exceed the number of registers
+ * passed as parameters below.
+ */
+ drm_WARN_ON(&xe->drm, num_geometry_regs > 3);
+ drm_WARN_ON(&xe->drm, num_compute_regs > 3);
+
+ load_dss_mask(gt, gt->fuse_topo.g_dss_mask,
+ num_geometry_regs,
+ XELP_GT_GEOMETRY_DSS_ENABLE,
+ XE2_GT_GEOMETRY_DSS_1,
+ XE2_GT_GEOMETRY_DSS_2);
+ load_dss_mask(gt, gt->fuse_topo.c_dss_mask, num_compute_regs,
+ XEHP_GT_COMPUTE_DSS_ENABLE,
+ XEHPC_GT_COMPUTE_DSS_ENABLE_EXT,
+ XE2_GT_COMPUTE_DSS_2);
+ load_eu_mask(gt, gt->fuse_topo.eu_mask_per_dss);
+
+ xe_gt_topology_dump(gt, &p);
+}
+
+void
+xe_gt_topology_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ drm_printf(p, "dss mask (geometry): %*pb\n", XE_MAX_DSS_FUSE_BITS,
+ gt->fuse_topo.g_dss_mask);
+ drm_printf(p, "dss mask (compute): %*pb\n", XE_MAX_DSS_FUSE_BITS,
+ gt->fuse_topo.c_dss_mask);
+
+ drm_printf(p, "EU mask per DSS: %*pb\n", XE_MAX_EU_FUSE_BITS,
+ gt->fuse_topo.eu_mask_per_dss);
+
+}
+
+/*
+ * Used to obtain the index of the first DSS. Can start searching from the
+ * beginning of a specific dss group (e.g., gslice, cslice, etc.) if
+ * groupsize and groupnum are non-zero.
+ */
+unsigned int
+xe_dss_mask_group_ffs(const xe_dss_mask_t mask, int groupsize, int groupnum)
+{
+ return find_next_bit(mask, XE_MAX_DSS_FUSE_BITS, groupnum * groupsize);
+}
+
+bool xe_dss_mask_empty(const xe_dss_mask_t mask)
+{
+ return bitmap_empty(mask, XE_MAX_DSS_FUSE_BITS);
+}
+
+/**
+ * xe_gt_topology_has_dss_in_quadrant - check fusing of DSS in GT quadrant
+ * @gt: GT to check
+ * @quad: Which quadrant of the DSS space to check
+ *
+ * Since Xe_HP platforms can have up to four CCS engines, those engines
+ * are each logically associated with a quarter of the possible DSS. If there
+ * are no DSS present in one of the four quadrants of the DSS space, the
+ * corresponding CCS engine is also not available for use.
+ *
+ * Returns false if all DSS in a quadrant of the GT are fused off, else true.
+ */
+bool xe_gt_topology_has_dss_in_quadrant(struct xe_gt *gt, int quad)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ xe_dss_mask_t all_dss;
+ int g_dss_regs, c_dss_regs, dss_per_quad, quad_first;
+
+ bitmap_or(all_dss, gt->fuse_topo.g_dss_mask, gt->fuse_topo.c_dss_mask,
+ XE_MAX_DSS_FUSE_BITS);
+
+ get_num_dss_regs(xe, &g_dss_regs, &c_dss_regs);
+ dss_per_quad = 32 * max(g_dss_regs, c_dss_regs) / 4;
+
+ quad_first = xe_dss_mask_group_ffs(all_dss, dss_per_quad, quad);
+
+ return quad_first < (quad + 1) * dss_per_quad;
+}
diff --git a/drivers/gpu/drm/xe/xe_gt_topology.h b/drivers/gpu/drm/xe/xe_gt_topology.h
new file mode 100644
index 000000000000..d1b54fb52ea6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_topology.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GT_TOPOLOGY_H_
+#define _XE_GT_TOPOLOGY_H_
+
+#include "xe_gt_types.h"
+
+struct drm_printer;
+
+void xe_gt_topology_init(struct xe_gt *gt);
+
+void xe_gt_topology_dump(struct xe_gt *gt, struct drm_printer *p);
+
+unsigned int
+xe_dss_mask_group_ffs(const xe_dss_mask_t mask, int groupsize, int groupnum);
+
+bool xe_dss_mask_empty(const xe_dss_mask_t mask);
+
+bool
+xe_gt_topology_has_dss_in_quadrant(struct xe_gt *gt, int quad);
+
+#endif /* _XE_GT_TOPOLOGY_H_ */
diff --git a/drivers/gpu/drm/xe/xe_gt_types.h b/drivers/gpu/drm/xe/xe_gt_types.h
new file mode 100644
index 000000000000..f74684660475
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_gt_types.h
@@ -0,0 +1,363 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022-2023 Intel Corporation
+ */
+
+#ifndef _XE_GT_TYPES_H_
+#define _XE_GT_TYPES_H_
+
+#include "xe_force_wake_types.h"
+#include "xe_gt_idle_types.h"
+#include "xe_hw_engine_types.h"
+#include "xe_hw_fence_types.h"
+#include "xe_reg_sr_types.h"
+#include "xe_sa_types.h"
+#include "xe_uc_types.h"
+
+struct xe_exec_queue_ops;
+struct xe_migrate;
+struct xe_ring_ops;
+
+enum xe_gt_type {
+ XE_GT_TYPE_UNINITIALIZED,
+ XE_GT_TYPE_MAIN,
+ XE_GT_TYPE_MEDIA,
+};
+
+#define XE_MAX_DSS_FUSE_REGS 3
+#define XE_MAX_EU_FUSE_REGS 1
+
+typedef unsigned long xe_dss_mask_t[BITS_TO_LONGS(32 * XE_MAX_DSS_FUSE_REGS)];
+typedef unsigned long xe_eu_mask_t[BITS_TO_LONGS(32 * XE_MAX_EU_FUSE_REGS)];
+
+struct xe_mmio_range {
+ u32 start;
+ u32 end;
+};
+
+/*
+ * The hardware has multiple kinds of multicast register ranges that need
+ * special register steering (and future platforms are expected to add
+ * additional types).
+ *
+ * During driver startup, we initialize the steering control register to
+ * direct reads to a slice/subslice that are valid for the 'subslice' class
+ * of multicast registers. If another type of steering does not have any
+ * overlap in valid steering targets with 'subslice' style registers, we will
+ * need to explicitly re-steer reads of registers of the other type.
+ *
+ * Only the replication types that may need additional non-default steering
+ * are listed here.
+ */
+enum xe_steering_type {
+ L3BANK,
+ MSLICE,
+ LNCF,
+ DSS,
+ OADDRM,
+ SQIDI_PSMI,
+
+ /*
+ * On some platforms there are multiple types of MCR registers that
+ * will always return a non-terminated value at instance (0, 0). We'll
+ * lump those all into a single category to keep things simple.
+ */
+ INSTANCE0,
+
+ /*
+ * Register ranges that don't need special steering for each register:
+ * it's sufficient to keep the HW-default for the selector, or only
+ * change it once, on GT initialization. This needs to be the last
+ * steering type.
+ */
+ IMPLICIT_STEERING,
+ NUM_STEERING_TYPES
+};
+
+#define gt_to_tile(gt__) \
+ _Generic(gt__, \
+ const struct xe_gt * : (const struct xe_tile *)((gt__)->tile), \
+ struct xe_gt * : (gt__)->tile)
+
+#define gt_to_xe(gt__) \
+ _Generic(gt__, \
+ const struct xe_gt * : (const struct xe_device *)(gt_to_tile(gt__)->xe), \
+ struct xe_gt * : gt_to_tile(gt__)->xe)
+
+/**
+ * struct xe_gt - A "Graphics Technology" unit of the GPU
+ *
+ * A GT ("Graphics Technology") is the subset of a GPU primarily responsible
+ * for implementing the graphics, compute, and/or media IP. It encapsulates
+ * the hardware engines, programmable execution units, and GuC. Each GT has
+ * its own handling of power management (RC6+forcewake) and multicast register
+ * steering.
+ *
+ * A GPU/tile may have a single GT that supplies all graphics, compute, and
+ * media functionality, or the graphics/compute and media may be split into
+ * separate GTs within a tile.
+ */
+struct xe_gt {
+ /** @tile: Backpointer to GT's tile */
+ struct xe_tile *tile;
+
+ /** @info: GT info */
+ struct {
+ /** @type: type of GT */
+ enum xe_gt_type type;
+ /** @id: Unique ID of this GT within the PCI Device */
+ u8 id;
+ /** @reference_clock: clock frequency */
+ u32 reference_clock;
+ /** @engine_mask: mask of engines present on GT */
+ u64 engine_mask;
+ /**
+ * @__engine_mask: mask of engines present on GT read from
+ * xe_pci.c, used to fake reading the engine_mask from the
+ * hwconfig blob.
+ */
+ u64 __engine_mask;
+ } info;
+
+ /**
+ * @mmio: mmio info for GT. All GTs within a tile share the same
+ * register space, but have their own copy of GSI registers at a
+ * specific offset, as well as their own forcewake handling.
+ */
+ struct {
+ /** @fw: force wake for GT */
+ struct xe_force_wake fw;
+ /**
+ * @adj_limit: adjust MMIO address if address is below this
+ * value
+ */
+ u32 adj_limit;
+ /** @adj_offset: offect to add to MMIO address when adjusting */
+ u32 adj_offset;
+ } mmio;
+
+ /**
+ * @reg_sr: table with registers to be restored on GT init/resume/reset
+ */
+ struct xe_reg_sr reg_sr;
+
+ /** @reset: state for GT resets */
+ struct {
+ /**
+ * @worker: work so GT resets can done async allowing to reset
+ * code to safely flush all code paths
+ */
+ struct work_struct worker;
+ } reset;
+
+ /** @tlb_invalidation: TLB invalidation state */
+ struct {
+ /** @seqno: TLB invalidation seqno, protected by CT lock */
+#define TLB_INVALIDATION_SEQNO_MAX 0x100000
+ int seqno;
+ /**
+ * @seqno_recv: last received TLB invalidation seqno, protected by CT lock
+ */
+ int seqno_recv;
+ /**
+ * @pending_fences: list of pending fences waiting TLB
+ * invaliations, protected by CT lock
+ */
+ struct list_head pending_fences;
+ /**
+ * @pending_lock: protects @pending_fences and updating
+ * @seqno_recv.
+ */
+ spinlock_t pending_lock;
+ /**
+ * @fence_tdr: schedules a delayed call to
+ * xe_gt_tlb_fence_timeout after the timeut interval is over.
+ */
+ struct delayed_work fence_tdr;
+ /** @fence_context: context for TLB invalidation fences */
+ u64 fence_context;
+ /**
+ * @fence_seqno: seqno to TLB invalidation fences, protected by
+ * tlb_invalidation.lock
+ */
+ u32 fence_seqno;
+ /** @lock: protects TLB invalidation fences */
+ spinlock_t lock;
+ } tlb_invalidation;
+
+ /**
+ * @ccs_mode: Number of compute engines enabled.
+ * Allows fixed mapping of available compute slices to compute engines.
+ * By default only the first available compute engine is enabled and all
+ * available compute slices are allocated to it.
+ */
+ u32 ccs_mode;
+
+ /** @usm: unified shared memory state */
+ struct {
+ /**
+ * @bb_pool: Pool from which batchbuffers, for USM operations
+ * (e.g. migrations, fixing page tables), are allocated.
+ * Dedicated pool needed so USM operations to not get blocked
+ * behind any user operations which may have resulted in a
+ * fault.
+ */
+ struct xe_sa_manager *bb_pool;
+ /**
+ * @reserved_bcs_instance: reserved BCS instance used for USM
+ * operations (e.g. mmigrations, fixing page tables)
+ */
+ u16 reserved_bcs_instance;
+ /** @pf_wq: page fault work queue, unbound, high priority */
+ struct workqueue_struct *pf_wq;
+ /** @acc_wq: access counter work queue, unbound, high priority */
+ struct workqueue_struct *acc_wq;
+ /**
+ * @pf_queue: Page fault queue used to sync faults so faults can
+ * be processed not under the GuC CT lock. The queue is sized so
+ * it can sync all possible faults (1 per physical engine).
+ * Multiple queues exists for page faults from different VMs are
+ * be processed in parallel.
+ */
+ struct pf_queue {
+ /** @gt: back pointer to GT */
+ struct xe_gt *gt;
+#define PF_QUEUE_NUM_DW 128
+ /** @data: data in the page fault queue */
+ u32 data[PF_QUEUE_NUM_DW];
+ /**
+ * @head: head pointer in DWs for page fault queue,
+ * moved by worker which processes faults.
+ */
+ u16 head;
+ /**
+ * @tail: tail pointer in DWs for page fault queue,
+ * moved by G2H handler.
+ */
+ u16 tail;
+ /** @lock: protects page fault queue */
+ spinlock_t lock;
+ /** @worker: to process page faults */
+ struct work_struct worker;
+#define NUM_PF_QUEUE 4
+ } pf_queue[NUM_PF_QUEUE];
+ /**
+ * @acc_queue: Same as page fault queue, cannot process access
+ * counters under CT lock.
+ */
+ struct acc_queue {
+ /** @gt: back pointer to GT */
+ struct xe_gt *gt;
+#define ACC_QUEUE_NUM_DW 128
+ /** @data: data in the page fault queue */
+ u32 data[ACC_QUEUE_NUM_DW];
+ /**
+ * @head: head pointer in DWs for page fault queue,
+ * moved by worker which processes faults.
+ */
+ u16 head;
+ /**
+ * @tail: tail pointer in DWs for page fault queue,
+ * moved by G2H handler.
+ */
+ u16 tail;
+ /** @lock: protects page fault queue */
+ spinlock_t lock;
+ /** @worker: to process access counters */
+ struct work_struct worker;
+#define NUM_ACC_QUEUE 4
+ } acc_queue[NUM_ACC_QUEUE];
+ } usm;
+
+ /** @ordered_wq: used to serialize GT resets and TDRs */
+ struct workqueue_struct *ordered_wq;
+
+ /** @uc: micro controllers on the GT */
+ struct xe_uc uc;
+
+ /** @gtidle: idle properties of GT */
+ struct xe_gt_idle gtidle;
+
+ /** @exec_queue_ops: submission backend exec queue operations */
+ const struct xe_exec_queue_ops *exec_queue_ops;
+
+ /**
+ * @ring_ops: ring operations for this hw engine (1 per engine class)
+ */
+ const struct xe_ring_ops *ring_ops[XE_ENGINE_CLASS_MAX];
+
+ /** @fence_irq: fence IRQs (1 per engine class) */
+ struct xe_hw_fence_irq fence_irq[XE_ENGINE_CLASS_MAX];
+
+ /** @default_lrc: default LRC state */
+ void *default_lrc[XE_ENGINE_CLASS_MAX];
+
+ /** @hw_engines: hardware engines on the GT */
+ struct xe_hw_engine hw_engines[XE_NUM_HW_ENGINES];
+
+ /** @eclass: per hardware engine class interface on the GT */
+ struct xe_hw_engine_class_intf eclass[XE_ENGINE_CLASS_MAX];
+
+ /** @pcode: GT's PCODE */
+ struct {
+ /** @lock: protecting GT's PCODE mailbox data */
+ struct mutex lock;
+ } pcode;
+
+ /** @sysfs: sysfs' kobj used by xe_gt_sysfs */
+ struct kobject *sysfs;
+
+ /** @freq: Main GT freq sysfs control */
+ struct kobject *freq;
+
+ /** @mocs: info */
+ struct {
+ /** @uc_index: UC index */
+ u8 uc_index;
+ /** @wb_index: WB index, only used on L3_CCS platforms */
+ u8 wb_index;
+ } mocs;
+
+ /** @fuse_topo: GT topology reported by fuse registers */
+ struct {
+ /** @g_dss_mask: dual-subslices usable by geometry */
+ xe_dss_mask_t g_dss_mask;
+
+ /** @c_dss_mask: dual-subslices usable by compute */
+ xe_dss_mask_t c_dss_mask;
+
+ /** @eu_mask_per_dss: EU mask per DSS*/
+ xe_eu_mask_t eu_mask_per_dss;
+ } fuse_topo;
+
+ /** @steering: register steering for individual HW units */
+ struct {
+ /* @ranges: register ranges used for this steering type */
+ const struct xe_mmio_range *ranges;
+
+ /** @group_target: target to steer accesses to */
+ u16 group_target;
+ /** @instance_target: instance to steer accesses to */
+ u16 instance_target;
+ } steering[NUM_STEERING_TYPES];
+
+ /**
+ * @mcr_lock: protects the MCR_SELECTOR register for the duration
+ * of a steered operation
+ */
+ spinlock_t mcr_lock;
+
+ /** @wa_active: keep track of active workarounds */
+ struct {
+ /** @gt: bitmap with active GT workarounds */
+ unsigned long *gt;
+ /** @engine: bitmap with active engine workarounds */
+ unsigned long *engine;
+ /** @lrc: bitmap with active LRC workarounds */
+ unsigned long *lrc;
+ /** @oob: bitmap with active OOB workaroudns */
+ unsigned long *oob;
+ } wa_active;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
new file mode 100644
index 000000000000..482cb0df9f15
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc.c
@@ -0,0 +1,911 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc.h"
+
+#include <drm/drm_managed.h>
+
+#include "abi/guc_actions_abi.h"
+#include "abi/guc_errors_abi.h"
+#include "generated/xe_wa_oob.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_guc_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_guc_ads.h"
+#include "xe_guc_ct.h"
+#include "xe_guc_hwconfig.h"
+#include "xe_guc_log.h"
+#include "xe_guc_pc.h"
+#include "xe_guc_submit.h"
+#include "xe_mmio.h"
+#include "xe_platform_types.h"
+#include "xe_uc.h"
+#include "xe_uc_fw.h"
+#include "xe_wa.h"
+#include "xe_wopcm.h"
+
+/* GuC addresses above GUC_GGTT_TOP also don't map through the GTT */
+#define GUC_GGTT_TOP 0xFEE00000
+static u32 guc_bo_ggtt_addr(struct xe_guc *guc,
+ struct xe_bo *bo)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 addr = xe_bo_ggtt_addr(bo);
+
+ xe_assert(xe, addr >= xe_wopcm_size(guc_to_xe(guc)));
+ xe_assert(xe, addr < GUC_GGTT_TOP);
+ xe_assert(xe, bo->size <= GUC_GGTT_TOP - addr);
+
+ return addr;
+}
+
+static u32 guc_ctl_debug_flags(struct xe_guc *guc)
+{
+ u32 level = xe_guc_log_get_level(&guc->log);
+ u32 flags = 0;
+
+ if (!GUC_LOG_LEVEL_IS_VERBOSE(level))
+ flags |= GUC_LOG_DISABLED;
+ else
+ flags |= GUC_LOG_LEVEL_TO_VERBOSITY(level) <<
+ GUC_LOG_VERBOSITY_SHIFT;
+
+ return flags;
+}
+
+static u32 guc_ctl_feature_flags(struct xe_guc *guc)
+{
+ return GUC_CTL_ENABLE_SLPC;
+}
+
+static u32 guc_ctl_log_params_flags(struct xe_guc *guc)
+{
+ u32 offset = guc_bo_ggtt_addr(guc, guc->log.bo) >> PAGE_SHIFT;
+ u32 flags;
+
+ #if (((CRASH_BUFFER_SIZE) % SZ_1M) == 0)
+ #define LOG_UNIT SZ_1M
+ #define LOG_FLAG GUC_LOG_LOG_ALLOC_UNITS
+ #else
+ #define LOG_UNIT SZ_4K
+ #define LOG_FLAG 0
+ #endif
+
+ #if (((CAPTURE_BUFFER_SIZE) % SZ_1M) == 0)
+ #define CAPTURE_UNIT SZ_1M
+ #define CAPTURE_FLAG GUC_LOG_CAPTURE_ALLOC_UNITS
+ #else
+ #define CAPTURE_UNIT SZ_4K
+ #define CAPTURE_FLAG 0
+ #endif
+
+ BUILD_BUG_ON(!CRASH_BUFFER_SIZE);
+ BUILD_BUG_ON(!IS_ALIGNED(CRASH_BUFFER_SIZE, LOG_UNIT));
+ BUILD_BUG_ON(!DEBUG_BUFFER_SIZE);
+ BUILD_BUG_ON(!IS_ALIGNED(DEBUG_BUFFER_SIZE, LOG_UNIT));
+ BUILD_BUG_ON(!CAPTURE_BUFFER_SIZE);
+ BUILD_BUG_ON(!IS_ALIGNED(CAPTURE_BUFFER_SIZE, CAPTURE_UNIT));
+
+ BUILD_BUG_ON((CRASH_BUFFER_SIZE / LOG_UNIT - 1) >
+ (GUC_LOG_CRASH_MASK >> GUC_LOG_CRASH_SHIFT));
+ BUILD_BUG_ON((DEBUG_BUFFER_SIZE / LOG_UNIT - 1) >
+ (GUC_LOG_DEBUG_MASK >> GUC_LOG_DEBUG_SHIFT));
+ BUILD_BUG_ON((CAPTURE_BUFFER_SIZE / CAPTURE_UNIT - 1) >
+ (GUC_LOG_CAPTURE_MASK >> GUC_LOG_CAPTURE_SHIFT));
+
+ flags = GUC_LOG_VALID |
+ GUC_LOG_NOTIFY_ON_HALF_FULL |
+ CAPTURE_FLAG |
+ LOG_FLAG |
+ ((CRASH_BUFFER_SIZE / LOG_UNIT - 1) << GUC_LOG_CRASH_SHIFT) |
+ ((DEBUG_BUFFER_SIZE / LOG_UNIT - 1) << GUC_LOG_DEBUG_SHIFT) |
+ ((CAPTURE_BUFFER_SIZE / CAPTURE_UNIT - 1) <<
+ GUC_LOG_CAPTURE_SHIFT) |
+ (offset << GUC_LOG_BUF_ADDR_SHIFT);
+
+ #undef LOG_UNIT
+ #undef LOG_FLAG
+ #undef CAPTURE_UNIT
+ #undef CAPTURE_FLAG
+
+ return flags;
+}
+
+static u32 guc_ctl_ads_flags(struct xe_guc *guc)
+{
+ u32 ads = guc_bo_ggtt_addr(guc, guc->ads.bo) >> PAGE_SHIFT;
+ u32 flags = ads << GUC_ADS_ADDR_SHIFT;
+
+ return flags;
+}
+
+static u32 guc_ctl_wa_flags(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 flags = 0;
+
+ if (XE_WA(gt, 22012773006))
+ flags |= GUC_WA_POLLCS;
+
+ if (XE_WA(gt, 16011759253))
+ flags |= GUC_WA_GAM_CREDITS;
+
+ if (XE_WA(gt, 14014475959))
+ flags |= GUC_WA_HOLD_CCS_SWITCHOUT;
+
+ if (XE_WA(gt, 22011391025) || XE_WA(gt, 14012197797))
+ flags |= GUC_WA_DUAL_QUEUE;
+
+ /*
+ * Wa_22011802037: FIXME - there's more to be done than simply setting
+ * this flag: make sure each CS is stopped when preparing for GT reset
+ * and wait for pending MI_FW.
+ */
+ if (GRAPHICS_VERx100(xe) < 1270)
+ flags |= GUC_WA_PRE_PARSER;
+
+ if (XE_WA(gt, 16011777198))
+ flags |= GUC_WA_RCS_RESET_BEFORE_RC6;
+
+ if (XE_WA(gt, 22012727170) || XE_WA(gt, 22012727685))
+ flags |= GUC_WA_CONTEXT_ISOLATION;
+
+ if ((XE_WA(gt, 16015675438) || XE_WA(gt, 18020744125)) &&
+ !xe_hw_engine_mask_per_class(gt, XE_ENGINE_CLASS_RENDER))
+ flags |= GUC_WA_RCS_REGS_IN_CCS_REGS_LIST;
+
+ if (XE_WA(gt, 1509372804))
+ flags |= GUC_WA_RENDER_RST_RC6_EXIT;
+
+ return flags;
+}
+
+static u32 guc_ctl_devid(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+
+ return (((u32)xe->info.devid) << 16) | xe->info.revid;
+}
+
+static void guc_init_params(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 *params = guc->params;
+ int i;
+
+ BUILD_BUG_ON(sizeof(guc->params) != GUC_CTL_MAX_DWORDS * sizeof(u32));
+ BUILD_BUG_ON(GUC_CTL_MAX_DWORDS + 2 != SOFT_SCRATCH_COUNT);
+
+ params[GUC_CTL_LOG_PARAMS] = guc_ctl_log_params_flags(guc);
+ params[GUC_CTL_FEATURE] = 0;
+ params[GUC_CTL_DEBUG] = guc_ctl_debug_flags(guc);
+ params[GUC_CTL_ADS] = guc_ctl_ads_flags(guc);
+ params[GUC_CTL_WA] = 0;
+ params[GUC_CTL_DEVID] = guc_ctl_devid(guc);
+
+ for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
+ drm_dbg(&xe->drm, "GuC param[%2d] = 0x%08x\n", i, params[i]);
+}
+
+static void guc_init_params_post_hwconfig(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 *params = guc->params;
+ int i;
+
+ BUILD_BUG_ON(sizeof(guc->params) != GUC_CTL_MAX_DWORDS * sizeof(u32));
+ BUILD_BUG_ON(GUC_CTL_MAX_DWORDS + 2 != SOFT_SCRATCH_COUNT);
+
+ params[GUC_CTL_LOG_PARAMS] = guc_ctl_log_params_flags(guc);
+ params[GUC_CTL_FEATURE] = guc_ctl_feature_flags(guc);
+ params[GUC_CTL_DEBUG] = guc_ctl_debug_flags(guc);
+ params[GUC_CTL_ADS] = guc_ctl_ads_flags(guc);
+ params[GUC_CTL_WA] = guc_ctl_wa_flags(guc);
+ params[GUC_CTL_DEVID] = guc_ctl_devid(guc);
+
+ for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
+ drm_dbg(&xe->drm, "GuC param[%2d] = 0x%08x\n", i, params[i]);
+}
+
+/*
+ * Initialize the GuC parameter block before starting the firmware
+ * transfer. These parameters are read by the firmware on startup
+ * and cannot be changed thereafter.
+ */
+static void guc_write_params(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ int i;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ xe_mmio_write32(gt, SOFT_SCRATCH(0), 0);
+
+ for (i = 0; i < GUC_CTL_MAX_DWORDS; i++)
+ xe_mmio_write32(gt, SOFT_SCRATCH(1 + i), guc->params[i]);
+}
+
+static void guc_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_guc *guc = arg;
+
+ xe_force_wake_get(gt_to_fw(guc_to_gt(guc)), XE_FORCEWAKE_ALL);
+ xe_guc_pc_fini(&guc->pc);
+ xe_uc_fini_hw(&guc_to_gt(guc)->uc);
+ xe_force_wake_put(gt_to_fw(guc_to_gt(guc)), XE_FORCEWAKE_ALL);
+}
+
+int xe_guc_init(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ int ret;
+
+ guc->fw.type = XE_UC_FW_TYPE_GUC;
+ ret = xe_uc_fw_init(&guc->fw);
+ if (ret)
+ goto out;
+
+ if (!xe_uc_fw_is_enabled(&guc->fw))
+ return 0;
+
+ ret = xe_guc_log_init(&guc->log);
+ if (ret)
+ goto out;
+
+ ret = xe_guc_ads_init(&guc->ads);
+ if (ret)
+ goto out;
+
+ ret = xe_guc_ct_init(&guc->ct);
+ if (ret)
+ goto out;
+
+ ret = xe_guc_pc_init(&guc->pc);
+ if (ret)
+ goto out;
+
+ ret = drmm_add_action_or_reset(&gt_to_xe(gt)->drm, guc_fini, guc);
+ if (ret)
+ goto out;
+
+ guc_init_params(guc);
+
+ if (xe_gt_is_media_type(gt))
+ guc->notify_reg = MED_GUC_HOST_INTERRUPT;
+ else
+ guc->notify_reg = GUC_HOST_INTERRUPT;
+
+ xe_uc_fw_change_status(&guc->fw, XE_UC_FIRMWARE_LOADABLE);
+
+ return 0;
+
+out:
+ drm_err(&xe->drm, "GuC init failed with %d", ret);
+ return ret;
+}
+
+/**
+ * xe_guc_init_post_hwconfig - initialize GuC post hwconfig load
+ * @guc: The GuC object
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_guc_init_post_hwconfig(struct xe_guc *guc)
+{
+ guc_init_params_post_hwconfig(guc);
+
+ return xe_guc_ads_init_post_hwconfig(&guc->ads);
+}
+
+int xe_guc_post_load_init(struct xe_guc *guc)
+{
+ xe_guc_ads_populate_post_load(&guc->ads);
+ guc->submission_state.enabled = true;
+
+ return 0;
+}
+
+int xe_guc_reset(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 guc_status, gdrst;
+ int ret;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ xe_mmio_write32(gt, GDRST, GRDOM_GUC);
+
+ ret = xe_mmio_wait32(gt, GDRST, GRDOM_GUC, 0, 5000, &gdrst, false);
+ if (ret) {
+ drm_err(&xe->drm, "GuC reset timed out, GDRST=0x%8x\n",
+ gdrst);
+ goto err_out;
+ }
+
+ guc_status = xe_mmio_read32(gt, GUC_STATUS);
+ if (!(guc_status & GS_MIA_IN_RESET)) {
+ drm_err(&xe->drm,
+ "GuC status: 0x%x, MIA core expected to be in reset\n",
+ guc_status);
+ ret = -EIO;
+ goto err_out;
+ }
+
+ return 0;
+
+err_out:
+
+ return ret;
+}
+
+static void guc_prepare_xfer(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 shim_flags = GUC_ENABLE_READ_CACHE_LOGIC |
+ GUC_ENABLE_READ_CACHE_FOR_SRAM_DATA |
+ GUC_ENABLE_READ_CACHE_FOR_WOPCM_DATA |
+ GUC_ENABLE_MIA_CLOCK_GATING;
+
+ if (GRAPHICS_VERx100(xe) < 1250)
+ shim_flags |= GUC_DISABLE_SRAM_INIT_TO_ZEROES |
+ GUC_ENABLE_MIA_CACHING;
+
+ if (GRAPHICS_VER(xe) >= 20 || xe->info.platform == XE_PVC)
+ shim_flags |= REG_FIELD_PREP(GUC_MOCS_INDEX_MASK, gt->mocs.uc_index);
+
+ /* Must program this register before loading the ucode with DMA */
+ xe_mmio_write32(gt, GUC_SHIM_CONTROL, shim_flags);
+
+ xe_mmio_write32(gt, GT_PM_CONFIG, GT_DOORBELL_ENABLE);
+}
+
+/*
+ * Supporting MMIO & in memory RSA
+ */
+static int guc_xfer_rsa(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 rsa[UOS_RSA_SCRATCH_COUNT];
+ size_t copied;
+ int i;
+
+ if (guc->fw.rsa_size > 256) {
+ u32 rsa_ggtt_addr = xe_bo_ggtt_addr(guc->fw.bo) +
+ xe_uc_fw_rsa_offset(&guc->fw);
+ xe_mmio_write32(gt, UOS_RSA_SCRATCH(0), rsa_ggtt_addr);
+ return 0;
+ }
+
+ copied = xe_uc_fw_copy_rsa(&guc->fw, rsa, sizeof(rsa));
+ if (copied < sizeof(rsa))
+ return -ENOMEM;
+
+ for (i = 0; i < UOS_RSA_SCRATCH_COUNT; i++)
+ xe_mmio_write32(gt, UOS_RSA_SCRATCH(i), rsa[i]);
+
+ return 0;
+}
+
+static int guc_wait_ucode(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 status;
+ int ret;
+
+ /*
+ * Wait for the GuC to start up.
+ * NB: Docs recommend not using the interrupt for completion.
+ * Measurements indicate this should take no more than 20ms
+ * (assuming the GT clock is at maximum frequency). So, a
+ * timeout here indicates that the GuC has failed and is unusable.
+ * (Higher levels of the driver may decide to reset the GuC and
+ * attempt the ucode load again if this happens.)
+ *
+ * FIXME: There is a known (but exceedingly unlikely) race condition
+ * where the asynchronous frequency management code could reduce
+ * the GT clock while a GuC reload is in progress (during a full
+ * GT reset). A fix is in progress but there are complex locking
+ * issues to be resolved. In the meantime bump the timeout to
+ * 200ms. Even at slowest clock, this should be sufficient. And
+ * in the working case, a larger timeout makes no difference.
+ */
+ ret = xe_mmio_wait32(guc_to_gt(guc), GUC_STATUS, GS_UKERNEL_MASK,
+ FIELD_PREP(GS_UKERNEL_MASK, XE_GUC_LOAD_STATUS_READY),
+ 200000, &status, false);
+
+ if (ret) {
+ struct drm_device *drm = &xe->drm;
+ struct drm_printer p = drm_info_printer(drm->dev);
+
+ drm_info(drm, "GuC load failed: status = 0x%08X\n", status);
+ drm_info(drm, "GuC load failed: status: Reset = %d, BootROM = 0x%02X, UKernel = 0x%02X, MIA = 0x%02X, Auth = 0x%02X\n",
+ REG_FIELD_GET(GS_MIA_IN_RESET, status),
+ REG_FIELD_GET(GS_BOOTROM_MASK, status),
+ REG_FIELD_GET(GS_UKERNEL_MASK, status),
+ REG_FIELD_GET(GS_MIA_MASK, status),
+ REG_FIELD_GET(GS_AUTH_STATUS_MASK, status));
+
+ if ((status & GS_BOOTROM_MASK) == GS_BOOTROM_RSA_FAILED) {
+ drm_info(drm, "GuC firmware signature verification failed\n");
+ ret = -ENOEXEC;
+ }
+
+ if (REG_FIELD_GET(GS_UKERNEL_MASK, status) ==
+ XE_GUC_LOAD_STATUS_EXCEPTION) {
+ drm_info(drm, "GuC firmware exception. EIP: %#x\n",
+ xe_mmio_read32(guc_to_gt(guc),
+ SOFT_SCRATCH(13)));
+ ret = -ENXIO;
+ }
+
+ xe_guc_log_print(&guc->log, &p);
+ } else {
+ drm_dbg(&xe->drm, "GuC successfully loaded");
+ }
+
+ return ret;
+}
+
+static int __xe_guc_upload(struct xe_guc *guc)
+{
+ int ret;
+
+ guc_write_params(guc);
+ guc_prepare_xfer(guc);
+
+ /*
+ * Note that GuC needs the CSS header plus uKernel code to be copied
+ * by the DMA engine in one operation, whereas the RSA signature is
+ * loaded separately, either by copying it to the UOS_RSA_SCRATCH
+ * register (if key size <= 256) or through a ggtt-pinned vma (if key
+ * size > 256). The RSA size and therefore the way we provide it to the
+ * HW is fixed for each platform and hard-coded in the bootrom.
+ */
+ ret = guc_xfer_rsa(guc);
+ if (ret)
+ goto out;
+ /*
+ * Current uCode expects the code to be loaded at 8k; locations below
+ * this are used for the stack.
+ */
+ ret = xe_uc_fw_upload(&guc->fw, 0x2000, UOS_MOVE);
+ if (ret)
+ goto out;
+
+ /* Wait for authentication */
+ ret = guc_wait_ucode(guc);
+ if (ret)
+ goto out;
+
+ xe_uc_fw_change_status(&guc->fw, XE_UC_FIRMWARE_RUNNING);
+ return 0;
+
+out:
+ xe_uc_fw_change_status(&guc->fw, XE_UC_FIRMWARE_LOAD_FAIL);
+ return 0 /* FIXME: ret, don't want to stop load currently */;
+}
+
+/**
+ * xe_guc_min_load_for_hwconfig - load minimal GuC and read hwconfig table
+ * @guc: The GuC object
+ *
+ * This function uploads a minimal GuC that does not support submissions but
+ * in a state where the hwconfig table can be read. Next, it reads and parses
+ * the hwconfig table so it can be used for subsequent steps in the driver load.
+ * Lastly, it enables CT communication (XXX: this is needed for PFs/VFs only).
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_guc_min_load_for_hwconfig(struct xe_guc *guc)
+{
+ int ret;
+
+ xe_guc_ads_populate_minimal(&guc->ads);
+
+ ret = __xe_guc_upload(guc);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_hwconfig_init(guc);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_enable_communication(guc);
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+int xe_guc_upload(struct xe_guc *guc)
+{
+ xe_guc_ads_populate(&guc->ads);
+
+ return __xe_guc_upload(guc);
+}
+
+static void guc_handle_mmio_msg(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 msg;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ msg = xe_mmio_read32(gt, SOFT_SCRATCH(15));
+ msg &= XE_GUC_RECV_MSG_EXCEPTION |
+ XE_GUC_RECV_MSG_CRASH_DUMP_POSTED;
+ xe_mmio_write32(gt, SOFT_SCRATCH(15), 0);
+
+ if (msg & XE_GUC_RECV_MSG_CRASH_DUMP_POSTED)
+ drm_err(&guc_to_xe(guc)->drm,
+ "Received early GuC crash dump notification!\n");
+
+ if (msg & XE_GUC_RECV_MSG_EXCEPTION)
+ drm_err(&guc_to_xe(guc)->drm,
+ "Received early GuC exception notification!\n");
+}
+
+static void guc_enable_irq(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 events = xe_gt_is_media_type(gt) ?
+ REG_FIELD_PREP(ENGINE0_MASK, GUC_INTR_GUC2HOST) :
+ REG_FIELD_PREP(ENGINE1_MASK, GUC_INTR_GUC2HOST);
+
+ /* Primary GuC and media GuC share a single enable bit */
+ xe_mmio_write32(gt, GUC_SG_INTR_ENABLE,
+ REG_FIELD_PREP(ENGINE1_MASK, GUC_INTR_GUC2HOST));
+
+ /*
+ * There are separate mask bits for primary and media GuCs, so use
+ * a RMW operation to avoid clobbering the other GuC's setting.
+ */
+ xe_mmio_rmw32(gt, GUC_SG_INTR_MASK, events, 0);
+}
+
+int xe_guc_enable_communication(struct xe_guc *guc)
+{
+ int err;
+
+ guc_enable_irq(guc);
+
+ xe_mmio_rmw32(guc_to_gt(guc), PMINTRMSK,
+ ARAT_EXPIRED_INTRMSK, 0);
+
+ err = xe_guc_ct_enable(&guc->ct);
+ if (err)
+ return err;
+
+ guc_handle_mmio_msg(guc);
+
+ return 0;
+}
+
+int xe_guc_suspend(struct xe_guc *guc)
+{
+ int ret;
+ u32 action[] = {
+ XE_GUC_ACTION_CLIENT_SOFT_RESET,
+ };
+
+ ret = xe_guc_mmio_send(guc, action, ARRAY_SIZE(action));
+ if (ret) {
+ drm_err(&guc_to_xe(guc)->drm,
+ "GuC suspend: CLIENT_SOFT_RESET fail: %d!\n", ret);
+ return ret;
+ }
+
+ xe_guc_sanitize(guc);
+ return 0;
+}
+
+void xe_guc_notify(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ const u32 default_notify_data = 0;
+
+ /*
+ * Both GUC_HOST_INTERRUPT and MED_GUC_HOST_INTERRUPT can pass
+ * additional payload data to the GuC but this capability is not
+ * used by the firmware yet. Use default value in the meantime.
+ */
+ xe_mmio_write32(gt, guc->notify_reg, default_notify_data);
+}
+
+int xe_guc_auth_huc(struct xe_guc *guc, u32 rsa_addr)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_AUTHENTICATE_HUC,
+ rsa_addr
+ };
+
+ return xe_guc_ct_send_block(&guc->ct, action, ARRAY_SIZE(action));
+}
+
+int xe_guc_mmio_send_recv(struct xe_guc *guc, const u32 *request,
+ u32 len, u32 *response_buf)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 header, reply;
+ struct xe_reg reply_reg = xe_gt_is_media_type(gt) ?
+ MED_VF_SW_FLAG(0) : VF_SW_FLAG(0);
+ const u32 LAST_INDEX = VF_SW_FLAG_COUNT - 1;
+ int ret;
+ int i;
+
+ BUILD_BUG_ON(VF_SW_FLAG_COUNT != MED_VF_SW_FLAG_COUNT);
+
+ xe_assert(xe, !guc->ct.enabled);
+ xe_assert(xe, len);
+ xe_assert(xe, len <= VF_SW_FLAG_COUNT);
+ xe_assert(xe, len <= MED_VF_SW_FLAG_COUNT);
+ xe_assert(xe, FIELD_GET(GUC_HXG_MSG_0_ORIGIN, request[0]) ==
+ GUC_HXG_ORIGIN_HOST);
+ xe_assert(xe, FIELD_GET(GUC_HXG_MSG_0_TYPE, request[0]) ==
+ GUC_HXG_TYPE_REQUEST);
+
+retry:
+ /* Not in critical data-path, just do if else for GT type */
+ if (xe_gt_is_media_type(gt)) {
+ for (i = 0; i < len; ++i)
+ xe_mmio_write32(gt, MED_VF_SW_FLAG(i),
+ request[i]);
+ xe_mmio_read32(gt, MED_VF_SW_FLAG(LAST_INDEX));
+ } else {
+ for (i = 0; i < len; ++i)
+ xe_mmio_write32(gt, VF_SW_FLAG(i),
+ request[i]);
+ xe_mmio_read32(gt, VF_SW_FLAG(LAST_INDEX));
+ }
+
+ xe_guc_notify(guc);
+
+ ret = xe_mmio_wait32(gt, reply_reg, GUC_HXG_MSG_0_ORIGIN,
+ FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_GUC),
+ 50000, &reply, false);
+ if (ret) {
+timeout:
+ drm_err(&xe->drm, "mmio request %#x: no reply %#x\n",
+ request[0], reply);
+ return ret;
+ }
+
+ header = xe_mmio_read32(gt, reply_reg);
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) ==
+ GUC_HXG_TYPE_NO_RESPONSE_BUSY) {
+ /*
+ * Once we got a BUSY reply we must wait again for the final
+ * response but this time we can't use ORIGIN mask anymore.
+ * To spot a right change in the reply, we take advantage that
+ * response SUCCESS and FAILURE differ only by the single bit
+ * and all other bits are set and can be used as a new mask.
+ */
+ u32 resp_bits = GUC_HXG_TYPE_RESPONSE_SUCCESS & GUC_HXG_TYPE_RESPONSE_FAILURE;
+ u32 resp_mask = FIELD_PREP(GUC_HXG_MSG_0_TYPE, resp_bits);
+
+ BUILD_BUG_ON(FIELD_MAX(GUC_HXG_MSG_0_TYPE) != GUC_HXG_TYPE_RESPONSE_SUCCESS);
+ BUILD_BUG_ON((GUC_HXG_TYPE_RESPONSE_SUCCESS ^ GUC_HXG_TYPE_RESPONSE_FAILURE) != 1);
+
+ ret = xe_mmio_wait32(gt, reply_reg, resp_mask, resp_mask,
+ 1000000, &header, false);
+
+ if (unlikely(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) !=
+ GUC_HXG_ORIGIN_GUC))
+ goto proto;
+ if (unlikely(ret))
+ goto timeout;
+ }
+
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) ==
+ GUC_HXG_TYPE_NO_RESPONSE_RETRY) {
+ u32 reason = FIELD_GET(GUC_HXG_RETRY_MSG_0_REASON, header);
+
+ drm_dbg(&xe->drm, "mmio request %#x: retrying, reason %#x\n",
+ request[0], reason);
+ goto retry;
+ }
+
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) ==
+ GUC_HXG_TYPE_RESPONSE_FAILURE) {
+ u32 hint = FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, header);
+ u32 error = FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, header);
+
+ drm_err(&xe->drm, "mmio request %#x: failure %#x/%#x\n",
+ request[0], error, hint);
+ return -ENXIO;
+ }
+
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, header) !=
+ GUC_HXG_TYPE_RESPONSE_SUCCESS) {
+proto:
+ drm_err(&xe->drm, "mmio request %#x: unexpected reply %#x\n",
+ request[0], header);
+ return -EPROTO;
+ }
+
+ /* Just copy entire possible message response */
+ if (response_buf) {
+ response_buf[0] = header;
+
+ for (i = 1; i < VF_SW_FLAG_COUNT; i++) {
+ reply_reg.addr += sizeof(u32);
+ response_buf[i] = xe_mmio_read32(gt, reply_reg);
+ }
+ }
+
+ /* Use data from the GuC response as our return value */
+ return FIELD_GET(GUC_HXG_RESPONSE_MSG_0_DATA0, header);
+}
+
+int xe_guc_mmio_send(struct xe_guc *guc, const u32 *request, u32 len)
+{
+ return xe_guc_mmio_send_recv(guc, request, len, NULL);
+}
+
+static int guc_self_cfg(struct xe_guc *guc, u16 key, u16 len, u64 val)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 request[HOST2GUC_SELF_CFG_REQUEST_MSG_LEN] = {
+ FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
+ FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
+ FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION,
+ GUC_ACTION_HOST2GUC_SELF_CFG),
+ FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_KEY, key) |
+ FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_1_KLV_LEN, len),
+ FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_2_VALUE32,
+ lower_32_bits(val)),
+ FIELD_PREP(HOST2GUC_SELF_CFG_REQUEST_MSG_3_VALUE64,
+ upper_32_bits(val)),
+ };
+ int ret;
+
+ xe_assert(xe, len <= 2);
+ xe_assert(xe, len != 1 || !upper_32_bits(val));
+
+ /* Self config must go over MMIO */
+ ret = xe_guc_mmio_send(guc, request, ARRAY_SIZE(request));
+
+ if (unlikely(ret < 0))
+ return ret;
+ if (unlikely(ret > 1))
+ return -EPROTO;
+ if (unlikely(!ret))
+ return -ENOKEY;
+
+ return 0;
+}
+
+int xe_guc_self_cfg32(struct xe_guc *guc, u16 key, u32 val)
+{
+ return guc_self_cfg(guc, key, 1, val);
+}
+
+int xe_guc_self_cfg64(struct xe_guc *guc, u16 key, u64 val)
+{
+ return guc_self_cfg(guc, key, 2, val);
+}
+
+void xe_guc_irq_handler(struct xe_guc *guc, const u16 iir)
+{
+ if (iir & GUC_INTR_GUC2HOST)
+ xe_guc_ct_irq_handler(&guc->ct);
+}
+
+void xe_guc_sanitize(struct xe_guc *guc)
+{
+ xe_uc_fw_change_status(&guc->fw, XE_UC_FIRMWARE_LOADABLE);
+ xe_guc_ct_disable(&guc->ct);
+ guc->submission_state.enabled = false;
+}
+
+int xe_guc_reset_prepare(struct xe_guc *guc)
+{
+ return xe_guc_submit_reset_prepare(guc);
+}
+
+void xe_guc_reset_wait(struct xe_guc *guc)
+{
+ xe_guc_submit_reset_wait(guc);
+}
+
+void xe_guc_stop_prepare(struct xe_guc *guc)
+{
+ XE_WARN_ON(xe_guc_pc_stop(&guc->pc));
+}
+
+int xe_guc_stop(struct xe_guc *guc)
+{
+ int ret;
+
+ xe_guc_ct_disable(&guc->ct);
+
+ ret = xe_guc_submit_stop(guc);
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+int xe_guc_start(struct xe_guc *guc)
+{
+ int ret;
+
+ ret = xe_guc_pc_start(&guc->pc);
+ XE_WARN_ON(ret);
+
+ return xe_guc_submit_start(guc);
+}
+
+void xe_guc_print_info(struct xe_guc *guc, struct drm_printer *p)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 status;
+ int err;
+ int i;
+
+ xe_uc_fw_print(&guc->fw, p);
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return;
+
+ status = xe_mmio_read32(gt, GUC_STATUS);
+
+ drm_printf(p, "\nGuC status 0x%08x:\n", status);
+ drm_printf(p, "\tBootrom status = 0x%x\n",
+ REG_FIELD_GET(GS_BOOTROM_MASK, status));
+ drm_printf(p, "\tuKernel status = 0x%x\n",
+ REG_FIELD_GET(GS_UKERNEL_MASK, status));
+ drm_printf(p, "\tMIA Core status = 0x%x\n",
+ REG_FIELD_GET(GS_MIA_MASK, status));
+ drm_printf(p, "\tLog level = %d\n",
+ xe_guc_log_get_level(&guc->log));
+
+ drm_puts(p, "\nScratch registers:\n");
+ for (i = 0; i < SOFT_SCRATCH_COUNT; i++) {
+ drm_printf(p, "\t%2d: \t0x%x\n",
+ i, xe_mmio_read32(gt, SOFT_SCRATCH(i)));
+ }
+
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+
+ xe_guc_ct_print(&guc->ct, p, false);
+ xe_guc_submit_print(guc, p);
+}
+
+/**
+ * xe_guc_in_reset() - Detect if GuC MIA is in reset.
+ * @guc: The GuC object
+ *
+ * This function detects runtime resume from d3cold by leveraging
+ * GUC_STATUS, GUC doesn't get reset during d3hot,
+ * it strictly to be called from RPM resume handler.
+ *
+ * Return: true if failed to get forcewake or GuC MIA is in Reset,
+ * otherwise false.
+ */
+bool xe_guc_in_reset(struct xe_guc *guc)
+{
+ struct xe_gt *gt = guc_to_gt(guc);
+ u32 status;
+ int err;
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return true;
+
+ status = xe_mmio_read32(gt, GUC_STATUS);
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+
+ return status & GS_MIA_IN_RESET;
+}
diff --git a/drivers/gpu/drm/xe/xe_guc.h b/drivers/gpu/drm/xe/xe_guc.h
new file mode 100644
index 000000000000..d3e49e7fd7c3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_H_
+#define _XE_GUC_H_
+
+#include "xe_gt.h"
+#include "xe_guc_types.h"
+#include "xe_hw_engine_types.h"
+#include "xe_macros.h"
+
+struct drm_printer;
+
+int xe_guc_init(struct xe_guc *guc);
+int xe_guc_init_post_hwconfig(struct xe_guc *guc);
+int xe_guc_post_load_init(struct xe_guc *guc);
+int xe_guc_reset(struct xe_guc *guc);
+int xe_guc_upload(struct xe_guc *guc);
+int xe_guc_min_load_for_hwconfig(struct xe_guc *guc);
+int xe_guc_enable_communication(struct xe_guc *guc);
+int xe_guc_suspend(struct xe_guc *guc);
+void xe_guc_notify(struct xe_guc *guc);
+int xe_guc_auth_huc(struct xe_guc *guc, u32 rsa_addr);
+int xe_guc_mmio_send(struct xe_guc *guc, const u32 *request, u32 len);
+int xe_guc_mmio_send_recv(struct xe_guc *guc, const u32 *request, u32 len,
+ u32 *response_buf);
+int xe_guc_self_cfg32(struct xe_guc *guc, u16 key, u32 val);
+int xe_guc_self_cfg64(struct xe_guc *guc, u16 key, u64 val);
+void xe_guc_irq_handler(struct xe_guc *guc, const u16 iir);
+void xe_guc_sanitize(struct xe_guc *guc);
+void xe_guc_print_info(struct xe_guc *guc, struct drm_printer *p);
+int xe_guc_reset_prepare(struct xe_guc *guc);
+void xe_guc_reset_wait(struct xe_guc *guc);
+void xe_guc_stop_prepare(struct xe_guc *guc);
+int xe_guc_stop(struct xe_guc *guc);
+int xe_guc_start(struct xe_guc *guc);
+bool xe_guc_in_reset(struct xe_guc *guc);
+
+static inline u16 xe_engine_class_to_guc_class(enum xe_engine_class class)
+{
+ switch (class) {
+ case XE_ENGINE_CLASS_RENDER:
+ return GUC_RENDER_CLASS;
+ case XE_ENGINE_CLASS_VIDEO_DECODE:
+ return GUC_VIDEO_CLASS;
+ case XE_ENGINE_CLASS_VIDEO_ENHANCE:
+ return GUC_VIDEOENHANCE_CLASS;
+ case XE_ENGINE_CLASS_COPY:
+ return GUC_BLITTER_CLASS;
+ case XE_ENGINE_CLASS_COMPUTE:
+ return GUC_COMPUTE_CLASS;
+ case XE_ENGINE_CLASS_OTHER:
+ return GUC_GSC_OTHER_CLASS;
+ default:
+ XE_WARN_ON(class);
+ return -1;
+ }
+}
+
+static inline struct xe_gt *guc_to_gt(struct xe_guc *guc)
+{
+ return container_of(guc, struct xe_gt, uc.guc);
+}
+
+static inline struct xe_device *guc_to_xe(struct xe_guc *guc)
+{
+ return gt_to_xe(guc_to_gt(guc));
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ads.c b/drivers/gpu/drm/xe/xe_guc_ads.c
new file mode 100644
index 000000000000..390e6f1bf4e1
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ads.c
@@ -0,0 +1,672 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_ads.h"
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_guc_regs.h"
+#include "xe_bo.h"
+#include "xe_gt.h"
+#include "xe_gt_ccs_mode.h"
+#include "xe_guc.h"
+#include "xe_hw_engine.h"
+#include "xe_lrc.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_platform_types.h"
+
+/* Slack of a few additional entries per engine */
+#define ADS_REGSET_EXTRA_MAX 8
+
+static struct xe_guc *
+ads_to_guc(struct xe_guc_ads *ads)
+{
+ return container_of(ads, struct xe_guc, ads);
+}
+
+static struct xe_gt *
+ads_to_gt(struct xe_guc_ads *ads)
+{
+ return container_of(ads, struct xe_gt, uc.guc.ads);
+}
+
+static struct xe_device *
+ads_to_xe(struct xe_guc_ads *ads)
+{
+ return gt_to_xe(ads_to_gt(ads));
+}
+
+static struct iosys_map *
+ads_to_map(struct xe_guc_ads *ads)
+{
+ return &ads->bo->vmap;
+}
+
+/* UM Queue parameters: */
+#define GUC_UM_QUEUE_SIZE (SZ_64K)
+#define GUC_PAGE_RES_TIMEOUT_US (-1)
+
+/*
+ * The Additional Data Struct (ADS) has pointers for different buffers used by
+ * the GuC. One single gem object contains the ADS struct itself (guc_ads) and
+ * all the extra buffers indirectly linked via the ADS struct's entries.
+ *
+ * Layout of the ADS blob allocated for the GuC:
+ *
+ * +---------------------------------------+ <== base
+ * | guc_ads |
+ * +---------------------------------------+
+ * | guc_policies |
+ * +---------------------------------------+
+ * | guc_gt_system_info |
+ * +---------------------------------------+
+ * | guc_engine_usage |
+ * +---------------------------------------+
+ * | guc_um_init_params |
+ * +---------------------------------------+ <== static
+ * | guc_mmio_reg[countA] (engine 0.0) |
+ * | guc_mmio_reg[countB] (engine 0.1) |
+ * | guc_mmio_reg[countC] (engine 1.0) |
+ * | ... |
+ * +---------------------------------------+ <== dynamic
+ * | padding |
+ * +---------------------------------------+ <== 4K aligned
+ * | golden contexts |
+ * +---------------------------------------+
+ * | padding |
+ * +---------------------------------------+ <== 4K aligned
+ * | capture lists |
+ * +---------------------------------------+
+ * | padding |
+ * +---------------------------------------+ <== 4K aligned
+ * | UM queues |
+ * +---------------------------------------+
+ * | padding |
+ * +---------------------------------------+ <== 4K aligned
+ * | private data |
+ * +---------------------------------------+
+ * | padding |
+ * +---------------------------------------+ <== 4K aligned
+ */
+struct __guc_ads_blob {
+ struct guc_ads ads;
+ struct guc_policies policies;
+ struct guc_gt_system_info system_info;
+ struct guc_engine_usage engine_usage;
+ struct guc_um_init_params um_init_params;
+ /* From here on, location is dynamic! Refer to above diagram. */
+ struct guc_mmio_reg regset[0];
+} __packed;
+
+#define ads_blob_read(ads_, field_) \
+ xe_map_rd_field(ads_to_xe(ads_), ads_to_map(ads_), 0, \
+ struct __guc_ads_blob, field_)
+
+#define ads_blob_write(ads_, field_, val_) \
+ xe_map_wr_field(ads_to_xe(ads_), ads_to_map(ads_), 0, \
+ struct __guc_ads_blob, field_, val_)
+
+#define info_map_write(xe_, map_, field_, val_) \
+ xe_map_wr_field(xe_, map_, 0, struct guc_gt_system_info, field_, val_)
+
+#define info_map_read(xe_, map_, field_) \
+ xe_map_rd_field(xe_, map_, 0, struct guc_gt_system_info, field_)
+
+static size_t guc_ads_regset_size(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+
+ xe_assert(xe, ads->regset_size);
+
+ return ads->regset_size;
+}
+
+static size_t guc_ads_golden_lrc_size(struct xe_guc_ads *ads)
+{
+ return PAGE_ALIGN(ads->golden_lrc_size);
+}
+
+static size_t guc_ads_capture_size(struct xe_guc_ads *ads)
+{
+ /* FIXME: Allocate a proper capture list */
+ return PAGE_ALIGN(PAGE_SIZE);
+}
+
+static size_t guc_ads_um_queues_size(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+
+ if (!xe->info.has_usm)
+ return 0;
+
+ return GUC_UM_QUEUE_SIZE * GUC_UM_HW_QUEUE_MAX;
+}
+
+static size_t guc_ads_private_data_size(struct xe_guc_ads *ads)
+{
+ return PAGE_ALIGN(ads_to_guc(ads)->fw.private_data_size);
+}
+
+static size_t guc_ads_regset_offset(struct xe_guc_ads *ads)
+{
+ return offsetof(struct __guc_ads_blob, regset);
+}
+
+static size_t guc_ads_golden_lrc_offset(struct xe_guc_ads *ads)
+{
+ size_t offset;
+
+ offset = guc_ads_regset_offset(ads) +
+ guc_ads_regset_size(ads);
+
+ return PAGE_ALIGN(offset);
+}
+
+static size_t guc_ads_capture_offset(struct xe_guc_ads *ads)
+{
+ size_t offset;
+
+ offset = guc_ads_golden_lrc_offset(ads) +
+ guc_ads_golden_lrc_size(ads);
+
+ return PAGE_ALIGN(offset);
+}
+
+static size_t guc_ads_um_queues_offset(struct xe_guc_ads *ads)
+{
+ u32 offset;
+
+ offset = guc_ads_capture_offset(ads) +
+ guc_ads_capture_size(ads);
+
+ return PAGE_ALIGN(offset);
+}
+
+static size_t guc_ads_private_data_offset(struct xe_guc_ads *ads)
+{
+ size_t offset;
+
+ offset = guc_ads_um_queues_offset(ads) +
+ guc_ads_um_queues_size(ads);
+
+ return PAGE_ALIGN(offset);
+}
+
+static size_t guc_ads_size(struct xe_guc_ads *ads)
+{
+ return guc_ads_private_data_offset(ads) +
+ guc_ads_private_data_size(ads);
+}
+
+static bool needs_wa_1607983814(struct xe_device *xe)
+{
+ return GRAPHICS_VERx100(xe) < 1250;
+}
+
+static size_t calculate_regset_size(struct xe_gt *gt)
+{
+ struct xe_reg_sr_entry *sr_entry;
+ unsigned long sr_idx;
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ unsigned int count = 0;
+
+ for_each_hw_engine(hwe, gt, id)
+ xa_for_each(&hwe->reg_sr.xa, sr_idx, sr_entry)
+ count++;
+
+ count += ADS_REGSET_EXTRA_MAX * XE_NUM_HW_ENGINES;
+
+ if (needs_wa_1607983814(gt_to_xe(gt)))
+ count += LNCFCMOCS_REG_COUNT;
+
+ return count * sizeof(struct guc_mmio_reg);
+}
+
+static u32 engine_enable_mask(struct xe_gt *gt, enum xe_engine_class class)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ u32 mask = 0;
+
+ for_each_hw_engine(hwe, gt, id)
+ if (hwe->class == class)
+ mask |= BIT(hwe->instance);
+
+ return mask;
+}
+
+static size_t calculate_golden_lrc_size(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+ size_t total_size = 0, alloc_size, real_size;
+ int class;
+
+ for (class = 0; class < XE_ENGINE_CLASS_MAX; ++class) {
+ if (!engine_enable_mask(gt, class))
+ continue;
+
+ real_size = xe_lrc_size(xe, class);
+ alloc_size = PAGE_ALIGN(real_size);
+ total_size += alloc_size;
+ }
+
+ return total_size;
+}
+
+#define MAX_GOLDEN_LRC_SIZE (SZ_4K * 64)
+
+int xe_guc_ads_init(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_bo *bo;
+
+ ads->golden_lrc_size = calculate_golden_lrc_size(ads);
+ ads->regset_size = calculate_regset_size(gt);
+
+ bo = xe_managed_bo_create_pin_map(xe, tile, guc_ads_size(ads) + MAX_GOLDEN_LRC_SIZE,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ ads->bo = bo;
+
+ return 0;
+}
+
+/**
+ * xe_guc_ads_init_post_hwconfig - initialize ADS post hwconfig load
+ * @ads: Additional data structures object
+ *
+ * Recalcuate golden_lrc_size & regset_size as the number hardware engines may
+ * have changed after the hwconfig was loaded. Also verify the new sizes fit in
+ * the already allocated ADS buffer object.
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_guc_ads_init_post_hwconfig(struct xe_guc_ads *ads)
+{
+ struct xe_gt *gt = ads_to_gt(ads);
+ u32 prev_regset_size = ads->regset_size;
+
+ xe_gt_assert(gt, ads->bo);
+
+ ads->golden_lrc_size = calculate_golden_lrc_size(ads);
+ ads->regset_size = calculate_regset_size(gt);
+
+ xe_gt_assert(gt, ads->golden_lrc_size +
+ (ads->regset_size - prev_regset_size) <=
+ MAX_GOLDEN_LRC_SIZE);
+
+ return 0;
+}
+
+static void guc_policies_init(struct xe_guc_ads *ads)
+{
+ ads_blob_write(ads, policies.dpc_promote_time,
+ GLOBAL_POLICY_DEFAULT_DPC_PROMOTE_TIME_US);
+ ads_blob_write(ads, policies.max_num_work_items,
+ GLOBAL_POLICY_MAX_NUM_WI);
+ ads_blob_write(ads, policies.global_flags, 0);
+ ads_blob_write(ads, policies.is_valid, 1);
+}
+
+static void fill_engine_enable_masks(struct xe_gt *gt,
+ struct iosys_map *info_map)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ info_map_write(xe, info_map, engine_enabled_masks[GUC_RENDER_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_RENDER));
+ info_map_write(xe, info_map, engine_enabled_masks[GUC_BLITTER_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_COPY));
+ info_map_write(xe, info_map, engine_enabled_masks[GUC_VIDEO_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_VIDEO_DECODE));
+ info_map_write(xe, info_map,
+ engine_enabled_masks[GUC_VIDEOENHANCE_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_VIDEO_ENHANCE));
+ info_map_write(xe, info_map, engine_enabled_masks[GUC_COMPUTE_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_COMPUTE));
+ info_map_write(xe, info_map, engine_enabled_masks[GUC_GSC_OTHER_CLASS],
+ engine_enable_mask(gt, XE_ENGINE_CLASS_OTHER));
+}
+
+static void guc_prep_golden_lrc_null(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads),
+ offsetof(struct __guc_ads_blob, system_info));
+ u8 guc_class;
+
+ for (guc_class = 0; guc_class <= GUC_MAX_ENGINE_CLASSES; ++guc_class) {
+ if (!info_map_read(xe, &info_map,
+ engine_enabled_masks[guc_class]))
+ continue;
+
+ ads_blob_write(ads, ads.eng_state_size[guc_class],
+ guc_ads_golden_lrc_size(ads) -
+ xe_lrc_skip_size(xe));
+ ads_blob_write(ads, ads.golden_context_lrca[guc_class],
+ xe_bo_ggtt_addr(ads->bo) +
+ guc_ads_golden_lrc_offset(ads));
+ }
+}
+
+static void guc_mapping_table_init_invalid(struct xe_gt *gt,
+ struct iosys_map *info_map)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ unsigned int i, j;
+
+ /* Table must be set to invalid values for entries not used */
+ for (i = 0; i < GUC_MAX_ENGINE_CLASSES; ++i)
+ for (j = 0; j < GUC_MAX_INSTANCES_PER_CLASS; ++j)
+ info_map_write(xe, info_map, mapping_table[i][j],
+ GUC_MAX_INSTANCES_PER_CLASS);
+}
+
+static void guc_mapping_table_init(struct xe_gt *gt,
+ struct iosys_map *info_map)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ guc_mapping_table_init_invalid(gt, info_map);
+
+ for_each_hw_engine(hwe, gt, id) {
+ u8 guc_class;
+
+ guc_class = xe_engine_class_to_guc_class(hwe->class);
+ info_map_write(xe, info_map,
+ mapping_table[guc_class][hwe->logical_instance],
+ hwe->instance);
+ }
+}
+
+static void guc_capture_list_init(struct xe_guc_ads *ads)
+{
+ int i, j;
+ u32 addr = xe_bo_ggtt_addr(ads->bo) + guc_ads_capture_offset(ads);
+
+ /* FIXME: Populate a proper capture list */
+ for (i = 0; i < GUC_CAPTURE_LIST_INDEX_MAX; i++) {
+ for (j = 0; j < GUC_MAX_ENGINE_CLASSES; j++) {
+ ads_blob_write(ads, ads.capture_instance[i][j], addr);
+ ads_blob_write(ads, ads.capture_class[i][j], addr);
+ }
+
+ ads_blob_write(ads, ads.capture_global[i], addr);
+ }
+}
+
+static void guc_mmio_regset_write_one(struct xe_guc_ads *ads,
+ struct iosys_map *regset_map,
+ struct xe_reg reg,
+ unsigned int n_entry)
+{
+ struct guc_mmio_reg entry = {
+ .offset = reg.addr,
+ .flags = reg.masked ? GUC_REGSET_MASKED : 0,
+ };
+
+ xe_map_memcpy_to(ads_to_xe(ads), regset_map, n_entry * sizeof(entry),
+ &entry, sizeof(entry));
+}
+
+static unsigned int guc_mmio_regset_write(struct xe_guc_ads *ads,
+ struct iosys_map *regset_map,
+ struct xe_hw_engine *hwe)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_hw_engine *hwe_rcs_reset_domain =
+ xe_gt_any_hw_engine_by_reset_domain(hwe->gt, XE_ENGINE_CLASS_RENDER);
+ struct xe_reg_sr_entry *entry;
+ unsigned long idx;
+ unsigned int count = 0;
+ const struct {
+ struct xe_reg reg;
+ bool skip;
+ } *e, extra_regs[] = {
+ { .reg = RING_MODE(hwe->mmio_base), },
+ { .reg = RING_HWS_PGA(hwe->mmio_base), },
+ { .reg = RING_IMR(hwe->mmio_base), },
+ { .reg = RCU_MODE, .skip = hwe != hwe_rcs_reset_domain },
+ { .reg = CCS_MODE,
+ .skip = hwe != hwe_rcs_reset_domain || !xe_gt_ccs_mode_enabled(hwe->gt) },
+ };
+ u32 i;
+
+ BUILD_BUG_ON(ARRAY_SIZE(extra_regs) > ADS_REGSET_EXTRA_MAX);
+
+ xa_for_each(&hwe->reg_sr.xa, idx, entry)
+ guc_mmio_regset_write_one(ads, regset_map, entry->reg, count++);
+
+ for (e = extra_regs; e < extra_regs + ARRAY_SIZE(extra_regs); e++) {
+ if (e->skip)
+ continue;
+
+ guc_mmio_regset_write_one(ads, regset_map, e->reg, count++);
+ }
+
+ /* Wa_1607983814 */
+ if (needs_wa_1607983814(xe) && hwe->class == XE_ENGINE_CLASS_RENDER) {
+ for (i = 0; i < LNCFCMOCS_REG_COUNT; i++) {
+ guc_mmio_regset_write_one(ads, regset_map,
+ XELP_LNCFCMOCS(i), count++);
+ }
+ }
+
+ return count;
+}
+
+static void guc_mmio_reg_state_init(struct xe_guc_ads *ads)
+{
+ size_t regset_offset = guc_ads_regset_offset(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ u32 addr = xe_bo_ggtt_addr(ads->bo) + regset_offset;
+ struct iosys_map regset_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads),
+ regset_offset);
+ unsigned int regset_used = 0;
+
+ for_each_hw_engine(hwe, gt, id) {
+ unsigned int count;
+ u8 gc;
+
+ /*
+ * 1. Write all MMIO entries for this exec queue to the table. No
+ * need to worry about fused-off engines and when there are
+ * entries in the regset: the reg_state_list has been zero'ed
+ * by xe_guc_ads_populate()
+ */
+ count = guc_mmio_regset_write(ads, &regset_map, hwe);
+ if (!count)
+ continue;
+
+ /*
+ * 2. Record in the header (ads.reg_state_list) the address
+ * location and number of entries
+ */
+ gc = xe_engine_class_to_guc_class(hwe->class);
+ ads_blob_write(ads, ads.reg_state_list[gc][hwe->instance].address, addr);
+ ads_blob_write(ads, ads.reg_state_list[gc][hwe->instance].count, count);
+
+ addr += count * sizeof(struct guc_mmio_reg);
+ iosys_map_incr(&regset_map, count * sizeof(struct guc_mmio_reg));
+
+ regset_used += count * sizeof(struct guc_mmio_reg);
+ }
+
+ xe_gt_assert(gt, regset_used <= ads->regset_size);
+}
+
+static void guc_um_init_params(struct xe_guc_ads *ads)
+{
+ u32 um_queue_offset = guc_ads_um_queues_offset(ads);
+ u64 base_dpa;
+ u32 base_ggtt;
+ int i;
+
+ base_ggtt = xe_bo_ggtt_addr(ads->bo) + um_queue_offset;
+ base_dpa = xe_bo_main_addr(ads->bo, PAGE_SIZE) + um_queue_offset;
+
+ for (i = 0; i < GUC_UM_HW_QUEUE_MAX; ++i) {
+ ads_blob_write(ads, um_init_params.queue_params[i].base_dpa,
+ base_dpa + (i * GUC_UM_QUEUE_SIZE));
+ ads_blob_write(ads, um_init_params.queue_params[i].base_ggtt_address,
+ base_ggtt + (i * GUC_UM_QUEUE_SIZE));
+ ads_blob_write(ads, um_init_params.queue_params[i].size_in_bytes,
+ GUC_UM_QUEUE_SIZE);
+ }
+
+ ads_blob_write(ads, um_init_params.page_response_timeout_in_us,
+ GUC_PAGE_RES_TIMEOUT_US);
+}
+
+static void guc_doorbell_init(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+
+ if (GRAPHICS_VER(xe) >= 12 && !IS_DGFX(xe)) {
+ u32 distdbreg =
+ xe_mmio_read32(gt, DIST_DBS_POPULATED);
+
+ ads_blob_write(ads,
+ system_info.generic_gt_sysinfo[GUC_GENERIC_GT_SYSINFO_DOORBELL_COUNT_PER_SQIDI],
+ REG_FIELD_GET(DOORBELLS_PER_SQIDI_MASK, distdbreg) + 1);
+ }
+}
+
+/**
+ * xe_guc_ads_populate_minimal - populate minimal ADS
+ * @ads: Additional data structures object
+ *
+ * This function populates a minimal ADS that does not support submissions but
+ * enough so the GuC can load and the hwconfig table can be read.
+ */
+void xe_guc_ads_populate_minimal(struct xe_guc_ads *ads)
+{
+ struct xe_gt *gt = ads_to_gt(ads);
+ struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads),
+ offsetof(struct __guc_ads_blob, system_info));
+ u32 base = xe_bo_ggtt_addr(ads->bo);
+
+ xe_gt_assert(gt, ads->bo);
+
+ xe_map_memset(ads_to_xe(ads), ads_to_map(ads), 0, 0, ads->bo->size);
+ guc_policies_init(ads);
+ guc_prep_golden_lrc_null(ads);
+ guc_mapping_table_init_invalid(gt, &info_map);
+ guc_doorbell_init(ads);
+
+ ads_blob_write(ads, ads.scheduler_policies, base +
+ offsetof(struct __guc_ads_blob, policies));
+ ads_blob_write(ads, ads.gt_system_info, base +
+ offsetof(struct __guc_ads_blob, system_info));
+ ads_blob_write(ads, ads.private_data, base +
+ guc_ads_private_data_offset(ads));
+}
+
+void xe_guc_ads_populate(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+ struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads),
+ offsetof(struct __guc_ads_blob, system_info));
+ u32 base = xe_bo_ggtt_addr(ads->bo);
+
+ xe_gt_assert(gt, ads->bo);
+
+ xe_map_memset(ads_to_xe(ads), ads_to_map(ads), 0, 0, ads->bo->size);
+ guc_policies_init(ads);
+ fill_engine_enable_masks(gt, &info_map);
+ guc_mmio_reg_state_init(ads);
+ guc_prep_golden_lrc_null(ads);
+ guc_mapping_table_init(gt, &info_map);
+ guc_capture_list_init(ads);
+ guc_doorbell_init(ads);
+
+ if (xe->info.has_usm) {
+ guc_um_init_params(ads);
+ ads_blob_write(ads, ads.um_init_data, base +
+ offsetof(struct __guc_ads_blob, um_init_params));
+ }
+
+ ads_blob_write(ads, ads.scheduler_policies, base +
+ offsetof(struct __guc_ads_blob, policies));
+ ads_blob_write(ads, ads.gt_system_info, base +
+ offsetof(struct __guc_ads_blob, system_info));
+ ads_blob_write(ads, ads.private_data, base +
+ guc_ads_private_data_offset(ads));
+}
+
+static void guc_populate_golden_lrc(struct xe_guc_ads *ads)
+{
+ struct xe_device *xe = ads_to_xe(ads);
+ struct xe_gt *gt = ads_to_gt(ads);
+ struct iosys_map info_map = IOSYS_MAP_INIT_OFFSET(ads_to_map(ads),
+ offsetof(struct __guc_ads_blob, system_info));
+ size_t total_size = 0, alloc_size, real_size;
+ u32 addr_ggtt, offset;
+ int class;
+
+ offset = guc_ads_golden_lrc_offset(ads);
+ addr_ggtt = xe_bo_ggtt_addr(ads->bo) + offset;
+
+ for (class = 0; class < XE_ENGINE_CLASS_MAX; ++class) {
+ u8 guc_class;
+
+ guc_class = xe_engine_class_to_guc_class(class);
+
+ if (!info_map_read(xe, &info_map,
+ engine_enabled_masks[guc_class]))
+ continue;
+
+ xe_gt_assert(gt, gt->default_lrc[class]);
+
+ real_size = xe_lrc_size(xe, class);
+ alloc_size = PAGE_ALIGN(real_size);
+ total_size += alloc_size;
+
+ /*
+ * This interface is slightly confusing. We need to pass the
+ * base address of the full golden context and the size of just
+ * the engine state, which is the section of the context image
+ * that starts after the execlists LRC registers. This is
+ * required to allow the GuC to restore just the engine state
+ * when a watchdog reset occurs.
+ * We calculate the engine state size by removing the size of
+ * what comes before it in the context image (which is identical
+ * on all engines).
+ */
+ ads_blob_write(ads, ads.eng_state_size[guc_class],
+ real_size - xe_lrc_skip_size(xe));
+ ads_blob_write(ads, ads.golden_context_lrca[guc_class],
+ addr_ggtt);
+
+ xe_map_memcpy_to(xe, ads_to_map(ads), offset,
+ gt->default_lrc[class], real_size);
+
+ addr_ggtt += alloc_size;
+ offset += alloc_size;
+ }
+
+ xe_gt_assert(gt, total_size == ads->golden_lrc_size);
+}
+
+void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads)
+{
+ guc_populate_golden_lrc(ads);
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_ads.h b/drivers/gpu/drm/xe/xe_guc_ads.h
new file mode 100644
index 000000000000..138ef6267671
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ads.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_ADS_H_
+#define _XE_GUC_ADS_H_
+
+#include "xe_guc_ads_types.h"
+
+int xe_guc_ads_init(struct xe_guc_ads *ads);
+int xe_guc_ads_init_post_hwconfig(struct xe_guc_ads *ads);
+void xe_guc_ads_populate(struct xe_guc_ads *ads);
+void xe_guc_ads_populate_minimal(struct xe_guc_ads *ads);
+void xe_guc_ads_populate_post_load(struct xe_guc_ads *ads);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ads_types.h b/drivers/gpu/drm/xe/xe_guc_ads_types.h
new file mode 100644
index 000000000000..4afe44bece4b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ads_types.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_ADS_TYPES_H_
+#define _XE_GUC_ADS_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_bo;
+
+/**
+ * struct xe_guc_ads - GuC additional data structures (ADS)
+ */
+struct xe_guc_ads {
+ /** @bo: XE BO for GuC ads blob */
+ struct xe_bo *bo;
+ /** @golden_lrc_size: golden LRC size */
+ size_t golden_lrc_size;
+ /** @regset_size: size of register set passed to GuC for save/restore */
+ u32 regset_size;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
new file mode 100644
index 000000000000..24a33fa36496
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ct.c
@@ -0,0 +1,1320 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_ct.h"
+
+#include <linux/bitfield.h>
+#include <linux/circ_buf.h>
+#include <linux/delay.h>
+
+#include <drm/drm_managed.h>
+
+#include "abi/guc_actions_abi.h"
+#include "abi/guc_klvs_abi.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_pagefault.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_guc.h"
+#include "xe_guc_submit.h"
+#include "xe_map.h"
+#include "xe_pm.h"
+#include "xe_trace.h"
+
+/* Used when a CT send wants to block and / or receive data */
+struct g2h_fence {
+ u32 *response_buffer;
+ u32 seqno;
+ u16 response_len;
+ u16 error;
+ u16 hint;
+ u16 reason;
+ bool retry;
+ bool fail;
+ bool done;
+};
+
+static void g2h_fence_init(struct g2h_fence *g2h_fence, u32 *response_buffer)
+{
+ g2h_fence->response_buffer = response_buffer;
+ g2h_fence->response_len = 0;
+ g2h_fence->fail = false;
+ g2h_fence->retry = false;
+ g2h_fence->done = false;
+ g2h_fence->seqno = ~0x0;
+}
+
+static bool g2h_fence_needs_alloc(struct g2h_fence *g2h_fence)
+{
+ return g2h_fence->seqno == ~0x0;
+}
+
+static struct xe_guc *
+ct_to_guc(struct xe_guc_ct *ct)
+{
+ return container_of(ct, struct xe_guc, ct);
+}
+
+static struct xe_gt *
+ct_to_gt(struct xe_guc_ct *ct)
+{
+ return container_of(ct, struct xe_gt, uc.guc.ct);
+}
+
+static struct xe_device *
+ct_to_xe(struct xe_guc_ct *ct)
+{
+ return gt_to_xe(ct_to_gt(ct));
+}
+
+/**
+ * DOC: GuC CTB Blob
+ *
+ * We allocate single blob to hold both CTB descriptors and buffers:
+ *
+ * +--------+-----------------------------------------------+------+
+ * | offset | contents | size |
+ * +========+===============================================+======+
+ * | 0x0000 | H2G CTB Descriptor (send) | |
+ * +--------+-----------------------------------------------+ 4K |
+ * | 0x0800 | G2H CTB Descriptor (g2h) | |
+ * +--------+-----------------------------------------------+------+
+ * | 0x1000 | H2G CT Buffer (send) | n*4K |
+ * | | | |
+ * +--------+-----------------------------------------------+------+
+ * | 0x1000 | G2H CT Buffer (g2h) | m*4K |
+ * | + n*4K | | |
+ * +--------+-----------------------------------------------+------+
+ *
+ * Size of each ``CT Buffer`` must be multiple of 4K.
+ * We don't expect too many messages in flight at any time, unless we are
+ * using the GuC submission. In that case each request requires a minimum
+ * 2 dwords which gives us a maximum 256 queue'd requests. Hopefully this
+ * enough space to avoid backpressure on the driver. We increase the size
+ * of the receive buffer (relative to the send) to ensure a G2H response
+ * CTB has a landing spot.
+ */
+
+#define CTB_DESC_SIZE ALIGN(sizeof(struct guc_ct_buffer_desc), SZ_2K)
+#define CTB_H2G_BUFFER_SIZE (SZ_4K)
+#define CTB_G2H_BUFFER_SIZE (4 * CTB_H2G_BUFFER_SIZE)
+#define G2H_ROOM_BUFFER_SIZE (CTB_G2H_BUFFER_SIZE / 4)
+
+static size_t guc_ct_size(void)
+{
+ return 2 * CTB_DESC_SIZE + CTB_H2G_BUFFER_SIZE +
+ CTB_G2H_BUFFER_SIZE;
+}
+
+static void guc_ct_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_guc_ct *ct = arg;
+
+ xa_destroy(&ct->fence_lookup);
+}
+
+static void g2h_worker_func(struct work_struct *w);
+
+static void primelockdep(struct xe_guc_ct *ct)
+{
+ if (!IS_ENABLED(CONFIG_LOCKDEP))
+ return;
+
+ fs_reclaim_acquire(GFP_KERNEL);
+ might_lock(&ct->lock);
+ fs_reclaim_release(GFP_KERNEL);
+}
+
+int xe_guc_ct_init(struct xe_guc_ct *ct)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct xe_gt *gt = ct_to_gt(ct);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_bo *bo;
+ int err;
+
+ xe_assert(xe, !(guc_ct_size() % PAGE_SIZE));
+
+ drmm_mutex_init(&xe->drm, &ct->lock);
+ spin_lock_init(&ct->fast_lock);
+ xa_init(&ct->fence_lookup);
+ INIT_WORK(&ct->g2h_worker, g2h_worker_func);
+ init_waitqueue_head(&ct->wq);
+ init_waitqueue_head(&ct->g2h_fence_wq);
+
+ primelockdep(ct);
+
+ bo = xe_managed_bo_create_pin_map(xe, tile, guc_ct_size(),
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ ct->bo = bo;
+
+ err = drmm_add_action_or_reset(&xe->drm, guc_ct_fini, ct);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+#define desc_read(xe_, guc_ctb__, field_) \
+ xe_map_rd_field(xe_, &guc_ctb__->desc, 0, \
+ struct guc_ct_buffer_desc, field_)
+
+#define desc_write(xe_, guc_ctb__, field_, val_) \
+ xe_map_wr_field(xe_, &guc_ctb__->desc, 0, \
+ struct guc_ct_buffer_desc, field_, val_)
+
+static void guc_ct_ctb_h2g_init(struct xe_device *xe, struct guc_ctb *h2g,
+ struct iosys_map *map)
+{
+ h2g->info.size = CTB_H2G_BUFFER_SIZE / sizeof(u32);
+ h2g->info.resv_space = 0;
+ h2g->info.tail = 0;
+ h2g->info.head = 0;
+ h2g->info.space = CIRC_SPACE(h2g->info.tail, h2g->info.head,
+ h2g->info.size) -
+ h2g->info.resv_space;
+ h2g->info.broken = false;
+
+ h2g->desc = *map;
+ xe_map_memset(xe, &h2g->desc, 0, 0, sizeof(struct guc_ct_buffer_desc));
+
+ h2g->cmds = IOSYS_MAP_INIT_OFFSET(map, CTB_DESC_SIZE * 2);
+}
+
+static void guc_ct_ctb_g2h_init(struct xe_device *xe, struct guc_ctb *g2h,
+ struct iosys_map *map)
+{
+ g2h->info.size = CTB_G2H_BUFFER_SIZE / sizeof(u32);
+ g2h->info.resv_space = G2H_ROOM_BUFFER_SIZE / sizeof(u32);
+ g2h->info.head = 0;
+ g2h->info.tail = 0;
+ g2h->info.space = CIRC_SPACE(g2h->info.tail, g2h->info.head,
+ g2h->info.size) -
+ g2h->info.resv_space;
+ g2h->info.broken = false;
+
+ g2h->desc = IOSYS_MAP_INIT_OFFSET(map, CTB_DESC_SIZE);
+ xe_map_memset(xe, &g2h->desc, 0, 0, sizeof(struct guc_ct_buffer_desc));
+
+ g2h->cmds = IOSYS_MAP_INIT_OFFSET(map, CTB_DESC_SIZE * 2 +
+ CTB_H2G_BUFFER_SIZE);
+}
+
+static int guc_ct_ctb_h2g_register(struct xe_guc_ct *ct)
+{
+ struct xe_guc *guc = ct_to_guc(ct);
+ u32 desc_addr, ctb_addr, size;
+ int err;
+
+ desc_addr = xe_bo_ggtt_addr(ct->bo);
+ ctb_addr = xe_bo_ggtt_addr(ct->bo) + CTB_DESC_SIZE * 2;
+ size = ct->ctbs.h2g.info.size * sizeof(u32);
+
+ err = xe_guc_self_cfg64(guc,
+ GUC_KLV_SELF_CFG_H2G_CTB_DESCRIPTOR_ADDR_KEY,
+ desc_addr);
+ if (err)
+ return err;
+
+ err = xe_guc_self_cfg64(guc,
+ GUC_KLV_SELF_CFG_H2G_CTB_ADDR_KEY,
+ ctb_addr);
+ if (err)
+ return err;
+
+ return xe_guc_self_cfg32(guc,
+ GUC_KLV_SELF_CFG_H2G_CTB_SIZE_KEY,
+ size);
+}
+
+static int guc_ct_ctb_g2h_register(struct xe_guc_ct *ct)
+{
+ struct xe_guc *guc = ct_to_guc(ct);
+ u32 desc_addr, ctb_addr, size;
+ int err;
+
+ desc_addr = xe_bo_ggtt_addr(ct->bo) + CTB_DESC_SIZE;
+ ctb_addr = xe_bo_ggtt_addr(ct->bo) + CTB_DESC_SIZE * 2 +
+ CTB_H2G_BUFFER_SIZE;
+ size = ct->ctbs.g2h.info.size * sizeof(u32);
+
+ err = xe_guc_self_cfg64(guc,
+ GUC_KLV_SELF_CFG_G2H_CTB_DESCRIPTOR_ADDR_KEY,
+ desc_addr);
+ if (err)
+ return err;
+
+ err = xe_guc_self_cfg64(guc,
+ GUC_KLV_SELF_CFG_G2H_CTB_ADDR_KEY,
+ ctb_addr);
+ if (err)
+ return err;
+
+ return xe_guc_self_cfg32(guc,
+ GUC_KLV_SELF_CFG_G2H_CTB_SIZE_KEY,
+ size);
+}
+
+static int guc_ct_control_toggle(struct xe_guc_ct *ct, bool enable)
+{
+ u32 request[HOST2GUC_CONTROL_CTB_REQUEST_MSG_LEN] = {
+ FIELD_PREP(GUC_HXG_MSG_0_ORIGIN, GUC_HXG_ORIGIN_HOST) |
+ FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
+ FIELD_PREP(GUC_HXG_REQUEST_MSG_0_ACTION,
+ GUC_ACTION_HOST2GUC_CONTROL_CTB),
+ FIELD_PREP(HOST2GUC_CONTROL_CTB_REQUEST_MSG_1_CONTROL,
+ enable ? GUC_CTB_CONTROL_ENABLE :
+ GUC_CTB_CONTROL_DISABLE),
+ };
+ int ret = xe_guc_mmio_send(ct_to_guc(ct), request, ARRAY_SIZE(request));
+
+ return ret > 0 ? -EPROTO : ret;
+}
+
+int xe_guc_ct_enable(struct xe_guc_ct *ct)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ int err;
+
+ xe_assert(xe, !ct->enabled);
+
+ guc_ct_ctb_h2g_init(xe, &ct->ctbs.h2g, &ct->bo->vmap);
+ guc_ct_ctb_g2h_init(xe, &ct->ctbs.g2h, &ct->bo->vmap);
+
+ err = guc_ct_ctb_h2g_register(ct);
+ if (err)
+ goto err_out;
+
+ err = guc_ct_ctb_g2h_register(ct);
+ if (err)
+ goto err_out;
+
+ err = guc_ct_control_toggle(ct, true);
+ if (err)
+ goto err_out;
+
+ mutex_lock(&ct->lock);
+ spin_lock_irq(&ct->fast_lock);
+ ct->g2h_outstanding = 0;
+ ct->enabled = true;
+ spin_unlock_irq(&ct->fast_lock);
+ mutex_unlock(&ct->lock);
+
+ smp_mb();
+ wake_up_all(&ct->wq);
+ drm_dbg(&xe->drm, "GuC CT communication channel enabled\n");
+
+ return 0;
+
+err_out:
+ drm_err(&xe->drm, "Failed to enable CT (%d)\n", err);
+
+ return err;
+}
+
+void xe_guc_ct_disable(struct xe_guc_ct *ct)
+{
+ mutex_lock(&ct->lock); /* Serialise dequeue_one_g2h() */
+ spin_lock_irq(&ct->fast_lock); /* Serialise CT fast-path */
+ ct->enabled = false; /* Finally disable CT communication */
+ spin_unlock_irq(&ct->fast_lock);
+ mutex_unlock(&ct->lock);
+
+ xa_destroy(&ct->fence_lookup);
+}
+
+static bool h2g_has_room(struct xe_guc_ct *ct, u32 cmd_len)
+{
+ struct guc_ctb *h2g = &ct->ctbs.h2g;
+
+ lockdep_assert_held(&ct->lock);
+
+ if (cmd_len > h2g->info.space) {
+ h2g->info.head = desc_read(ct_to_xe(ct), h2g, head);
+ h2g->info.space = CIRC_SPACE(h2g->info.tail, h2g->info.head,
+ h2g->info.size) -
+ h2g->info.resv_space;
+ if (cmd_len > h2g->info.space)
+ return false;
+ }
+
+ return true;
+}
+
+static bool g2h_has_room(struct xe_guc_ct *ct, u32 g2h_len)
+{
+ if (!g2h_len)
+ return true;
+
+ lockdep_assert_held(&ct->fast_lock);
+
+ return ct->ctbs.g2h.info.space > g2h_len;
+}
+
+static int has_room(struct xe_guc_ct *ct, u32 cmd_len, u32 g2h_len)
+{
+ lockdep_assert_held(&ct->lock);
+
+ if (!g2h_has_room(ct, g2h_len) || !h2g_has_room(ct, cmd_len))
+ return -EBUSY;
+
+ return 0;
+}
+
+static void h2g_reserve_space(struct xe_guc_ct *ct, u32 cmd_len)
+{
+ lockdep_assert_held(&ct->lock);
+ ct->ctbs.h2g.info.space -= cmd_len;
+}
+
+static void __g2h_reserve_space(struct xe_guc_ct *ct, u32 g2h_len, u32 num_g2h)
+{
+ xe_assert(ct_to_xe(ct), g2h_len <= ct->ctbs.g2h.info.space);
+
+ if (g2h_len) {
+ lockdep_assert_held(&ct->fast_lock);
+
+ ct->ctbs.g2h.info.space -= g2h_len;
+ ct->g2h_outstanding += num_g2h;
+ }
+}
+
+static void __g2h_release_space(struct xe_guc_ct *ct, u32 g2h_len)
+{
+ lockdep_assert_held(&ct->fast_lock);
+ xe_assert(ct_to_xe(ct), ct->ctbs.g2h.info.space + g2h_len <=
+ ct->ctbs.g2h.info.size - ct->ctbs.g2h.info.resv_space);
+
+ ct->ctbs.g2h.info.space += g2h_len;
+ --ct->g2h_outstanding;
+}
+
+static void g2h_release_space(struct xe_guc_ct *ct, u32 g2h_len)
+{
+ spin_lock_irq(&ct->fast_lock);
+ __g2h_release_space(ct, g2h_len);
+ spin_unlock_irq(&ct->fast_lock);
+}
+
+#define H2G_CT_HEADERS (GUC_CTB_HDR_LEN + 1) /* one DW CTB header and one DW HxG header */
+
+static int h2g_write(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 ct_fence_value, bool want_response)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct guc_ctb *h2g = &ct->ctbs.h2g;
+ u32 cmd[H2G_CT_HEADERS];
+ u32 tail = h2g->info.tail;
+ u32 full_len;
+ struct iosys_map map = IOSYS_MAP_INIT_OFFSET(&h2g->cmds,
+ tail * sizeof(u32));
+
+ full_len = len + GUC_CTB_HDR_LEN;
+
+ lockdep_assert_held(&ct->lock);
+ xe_assert(xe, full_len <= GUC_CTB_MSG_MAX_LEN);
+ xe_assert(xe, tail <= h2g->info.size);
+
+ /* Command will wrap, zero fill (NOPs), return and check credits again */
+ if (tail + full_len > h2g->info.size) {
+ xe_map_memset(xe, &map, 0, 0,
+ (h2g->info.size - tail) * sizeof(u32));
+ h2g_reserve_space(ct, (h2g->info.size - tail));
+ h2g->info.tail = 0;
+ desc_write(xe, h2g, tail, h2g->info.tail);
+
+ return -EAGAIN;
+ }
+
+ /*
+ * dw0: CT header (including fence)
+ * dw1: HXG header (including action code)
+ * dw2+: action data
+ */
+ cmd[0] = FIELD_PREP(GUC_CTB_MSG_0_FORMAT, GUC_CTB_FORMAT_HXG) |
+ FIELD_PREP(GUC_CTB_MSG_0_NUM_DWORDS, len) |
+ FIELD_PREP(GUC_CTB_MSG_0_FENCE, ct_fence_value);
+ if (want_response) {
+ cmd[1] =
+ FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_REQUEST) |
+ FIELD_PREP(GUC_HXG_EVENT_MSG_0_ACTION |
+ GUC_HXG_EVENT_MSG_0_DATA0, action[0]);
+ } else {
+ cmd[1] =
+ FIELD_PREP(GUC_HXG_MSG_0_TYPE, GUC_HXG_TYPE_EVENT) |
+ FIELD_PREP(GUC_HXG_EVENT_MSG_0_ACTION |
+ GUC_HXG_EVENT_MSG_0_DATA0, action[0]);
+ }
+
+ /* H2G header in cmd[1] replaces action[0] so: */
+ --len;
+ ++action;
+
+ /* Write H2G ensuring visable before descriptor update */
+ xe_map_memcpy_to(xe, &map, 0, cmd, H2G_CT_HEADERS * sizeof(u32));
+ xe_map_memcpy_to(xe, &map, H2G_CT_HEADERS * sizeof(u32), action, len * sizeof(u32));
+ xe_device_wmb(xe);
+
+ /* Update local copies */
+ h2g->info.tail = (tail + full_len) % h2g->info.size;
+ h2g_reserve_space(ct, full_len);
+
+ /* Update descriptor */
+ desc_write(xe, h2g, tail, h2g->info.tail);
+
+ trace_xe_guc_ctb_h2g(ct_to_gt(ct)->info.id, *(action - 1), full_len,
+ desc_read(xe, h2g, head), h2g->info.tail);
+
+ return 0;
+}
+
+static int __guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action,
+ u32 len, u32 g2h_len, u32 num_g2h,
+ struct g2h_fence *g2h_fence)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ int ret;
+
+ xe_assert(xe, !g2h_len || !g2h_fence);
+ xe_assert(xe, !num_g2h || !g2h_fence);
+ xe_assert(xe, !g2h_len || num_g2h);
+ xe_assert(xe, g2h_len || !num_g2h);
+ lockdep_assert_held(&ct->lock);
+
+ if (unlikely(ct->ctbs.h2g.info.broken)) {
+ ret = -EPIPE;
+ goto out;
+ }
+
+ if (unlikely(!ct->enabled)) {
+ ret = -ENODEV;
+ goto out;
+ }
+
+ if (g2h_fence) {
+ g2h_len = GUC_CTB_HXG_MSG_MAX_LEN;
+ num_g2h = 1;
+
+ if (g2h_fence_needs_alloc(g2h_fence)) {
+ void *ptr;
+
+ g2h_fence->seqno = (ct->fence_seqno++ & 0xffff);
+ ptr = xa_store(&ct->fence_lookup,
+ g2h_fence->seqno,
+ g2h_fence, GFP_ATOMIC);
+ if (IS_ERR(ptr)) {
+ ret = PTR_ERR(ptr);
+ goto out;
+ }
+ }
+ }
+
+ if (g2h_len)
+ spin_lock_irq(&ct->fast_lock);
+retry:
+ ret = has_room(ct, len + GUC_CTB_HDR_LEN, g2h_len);
+ if (unlikely(ret))
+ goto out_unlock;
+
+ ret = h2g_write(ct, action, len, g2h_fence ? g2h_fence->seqno : 0,
+ !!g2h_fence);
+ if (unlikely(ret)) {
+ if (ret == -EAGAIN)
+ goto retry;
+ goto out_unlock;
+ }
+
+ __g2h_reserve_space(ct, g2h_len, num_g2h);
+ xe_guc_notify(ct_to_guc(ct));
+out_unlock:
+ if (g2h_len)
+ spin_unlock_irq(&ct->fast_lock);
+out:
+ return ret;
+}
+
+static void kick_reset(struct xe_guc_ct *ct)
+{
+ xe_gt_reset_async(ct_to_gt(ct));
+}
+
+static int dequeue_one_g2h(struct xe_guc_ct *ct);
+
+static int guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h,
+ struct g2h_fence *g2h_fence)
+{
+ struct drm_device *drm = &ct_to_xe(ct)->drm;
+ struct drm_printer p = drm_info_printer(drm->dev);
+ unsigned int sleep_period_ms = 1;
+ int ret;
+
+ xe_assert(ct_to_xe(ct), !g2h_len || !g2h_fence);
+ lockdep_assert_held(&ct->lock);
+ xe_device_assert_mem_access(ct_to_xe(ct));
+
+try_again:
+ ret = __guc_ct_send_locked(ct, action, len, g2h_len, num_g2h,
+ g2h_fence);
+
+ /*
+ * We wait to try to restore credits for about 1 second before bailing.
+ * In the case of H2G credits we have no choice but just to wait for the
+ * GuC to consume H2Gs in the channel so we use a wait / sleep loop. In
+ * the case of G2H we process any G2H in the channel, hopefully freeing
+ * credits as we consume the G2H messages.
+ */
+ if (unlikely(ret == -EBUSY &&
+ !h2g_has_room(ct, len + GUC_CTB_HDR_LEN))) {
+ struct guc_ctb *h2g = &ct->ctbs.h2g;
+
+ if (sleep_period_ms == 1024)
+ goto broken;
+
+ trace_xe_guc_ct_h2g_flow_control(h2g->info.head, h2g->info.tail,
+ h2g->info.size,
+ h2g->info.space,
+ len + GUC_CTB_HDR_LEN);
+ msleep(sleep_period_ms);
+ sleep_period_ms <<= 1;
+
+ goto try_again;
+ } else if (unlikely(ret == -EBUSY)) {
+ struct xe_device *xe = ct_to_xe(ct);
+ struct guc_ctb *g2h = &ct->ctbs.g2h;
+
+ trace_xe_guc_ct_g2h_flow_control(g2h->info.head,
+ desc_read(xe, g2h, tail),
+ g2h->info.size,
+ g2h->info.space,
+ g2h_fence ?
+ GUC_CTB_HXG_MSG_MAX_LEN :
+ g2h_len);
+
+#define g2h_avail(ct) \
+ (desc_read(ct_to_xe(ct), (&ct->ctbs.g2h), tail) != ct->ctbs.g2h.info.head)
+ if (!wait_event_timeout(ct->wq, !ct->g2h_outstanding ||
+ g2h_avail(ct), HZ))
+ goto broken;
+#undef g2h_avail
+
+ if (dequeue_one_g2h(ct) < 0)
+ goto broken;
+
+ goto try_again;
+ }
+
+ return ret;
+
+broken:
+ drm_err(drm, "No forward process on H2G, reset required");
+ xe_guc_ct_print(ct, &p, true);
+ ct->ctbs.h2g.info.broken = true;
+
+ return -EDEADLK;
+}
+
+static int guc_ct_send(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h, struct g2h_fence *g2h_fence)
+{
+ int ret;
+
+ xe_assert(ct_to_xe(ct), !g2h_len || !g2h_fence);
+
+ mutex_lock(&ct->lock);
+ ret = guc_ct_send_locked(ct, action, len, g2h_len, num_g2h, g2h_fence);
+ mutex_unlock(&ct->lock);
+
+ return ret;
+}
+
+int xe_guc_ct_send(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h)
+{
+ int ret;
+
+ ret = guc_ct_send(ct, action, len, g2h_len, num_g2h, NULL);
+ if (ret == -EDEADLK)
+ kick_reset(ct);
+
+ return ret;
+}
+
+int xe_guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h)
+{
+ int ret;
+
+ ret = guc_ct_send_locked(ct, action, len, g2h_len, num_g2h, NULL);
+ if (ret == -EDEADLK)
+ kick_reset(ct);
+
+ return ret;
+}
+
+int xe_guc_ct_send_g2h_handler(struct xe_guc_ct *ct, const u32 *action, u32 len)
+{
+ int ret;
+
+ lockdep_assert_held(&ct->lock);
+
+ ret = guc_ct_send_locked(ct, action, len, 0, 0, NULL);
+ if (ret == -EDEADLK)
+ kick_reset(ct);
+
+ return ret;
+}
+
+/*
+ * Check if a GT reset is in progress or will occur and if GT reset brought the
+ * CT back up. Randomly picking 5 seconds for an upper limit to do a GT a reset.
+ */
+static bool retry_failure(struct xe_guc_ct *ct, int ret)
+{
+ if (!(ret == -EDEADLK || ret == -EPIPE || ret == -ENODEV))
+ return false;
+
+#define ct_alive(ct) \
+ (ct->enabled && !ct->ctbs.h2g.info.broken && !ct->ctbs.g2h.info.broken)
+ if (!wait_event_interruptible_timeout(ct->wq, ct_alive(ct), HZ * 5))
+ return false;
+#undef ct_alive
+
+ return true;
+}
+
+static int guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 *response_buffer, bool no_fail)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct g2h_fence g2h_fence;
+ int ret = 0;
+
+ /*
+ * We use a fence to implement blocking sends / receiving response data.
+ * The seqno of the fence is sent in the H2G, returned in the G2H, and
+ * an xarray is used as storage media with the seqno being to key.
+ * Fields in the fence hold success, failure, retry status and the
+ * response data. Safe to allocate on the stack as the xarray is the
+ * only reference and it cannot be present after this function exits.
+ */
+retry:
+ g2h_fence_init(&g2h_fence, response_buffer);
+retry_same_fence:
+ ret = guc_ct_send(ct, action, len, 0, 0, &g2h_fence);
+ if (unlikely(ret == -ENOMEM)) {
+ void *ptr;
+
+ /* Retry allocation /w GFP_KERNEL */
+ ptr = xa_store(&ct->fence_lookup,
+ g2h_fence.seqno,
+ &g2h_fence, GFP_KERNEL);
+ if (IS_ERR(ptr))
+ return PTR_ERR(ptr);
+
+ goto retry_same_fence;
+ } else if (unlikely(ret)) {
+ if (ret == -EDEADLK)
+ kick_reset(ct);
+
+ if (no_fail && retry_failure(ct, ret))
+ goto retry_same_fence;
+
+ if (!g2h_fence_needs_alloc(&g2h_fence))
+ xa_erase_irq(&ct->fence_lookup, g2h_fence.seqno);
+
+ return ret;
+ }
+
+ ret = wait_event_timeout(ct->g2h_fence_wq, g2h_fence.done, HZ);
+ if (!ret) {
+ drm_err(&xe->drm, "Timed out wait for G2H, fence %u, action %04x",
+ g2h_fence.seqno, action[0]);
+ xa_erase_irq(&ct->fence_lookup, g2h_fence.seqno);
+ return -ETIME;
+ }
+
+ if (g2h_fence.retry) {
+ drm_warn(&xe->drm, "Send retry, action 0x%04x, reason %d",
+ action[0], g2h_fence.reason);
+ goto retry;
+ }
+ if (g2h_fence.fail) {
+ drm_err(&xe->drm, "Send failed, action 0x%04x, error %d, hint %d",
+ action[0], g2h_fence.error, g2h_fence.hint);
+ ret = -EIO;
+ }
+
+ return ret > 0 ? 0 : ret;
+}
+
+int xe_guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 *response_buffer)
+{
+ return guc_ct_send_recv(ct, action, len, response_buffer, false);
+}
+
+int xe_guc_ct_send_recv_no_fail(struct xe_guc_ct *ct, const u32 *action,
+ u32 len, u32 *response_buffer)
+{
+ return guc_ct_send_recv(ct, action, len, response_buffer, true);
+}
+
+static int parse_g2h_event(struct xe_guc_ct *ct, u32 *msg, u32 len)
+{
+ u32 action = FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[1]);
+
+ lockdep_assert_held(&ct->lock);
+
+ switch (action) {
+ case XE_GUC_ACTION_SCHED_CONTEXT_MODE_DONE:
+ case XE_GUC_ACTION_DEREGISTER_CONTEXT_DONE:
+ case XE_GUC_ACTION_SCHED_ENGINE_MODE_DONE:
+ case XE_GUC_ACTION_TLB_INVALIDATION_DONE:
+ g2h_release_space(ct, len);
+ }
+
+ return 0;
+}
+
+static int parse_g2h_response(struct xe_guc_ct *ct, u32 *msg, u32 len)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ u32 response_len = len - GUC_CTB_MSG_MIN_LEN;
+ u32 fence = FIELD_GET(GUC_CTB_MSG_0_FENCE, msg[0]);
+ u32 type = FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[1]);
+ struct g2h_fence *g2h_fence;
+
+ lockdep_assert_held(&ct->lock);
+
+ g2h_fence = xa_erase(&ct->fence_lookup, fence);
+ if (unlikely(!g2h_fence)) {
+ /* Don't tear down channel, as send could've timed out */
+ drm_warn(&xe->drm, "G2H fence (%u) not found!\n", fence);
+ g2h_release_space(ct, GUC_CTB_HXG_MSG_MAX_LEN);
+ return 0;
+ }
+
+ xe_assert(xe, fence == g2h_fence->seqno);
+
+ if (type == GUC_HXG_TYPE_RESPONSE_FAILURE) {
+ g2h_fence->fail = true;
+ g2h_fence->error =
+ FIELD_GET(GUC_HXG_FAILURE_MSG_0_ERROR, msg[1]);
+ g2h_fence->hint =
+ FIELD_GET(GUC_HXG_FAILURE_MSG_0_HINT, msg[1]);
+ } else if (type == GUC_HXG_TYPE_NO_RESPONSE_RETRY) {
+ g2h_fence->retry = true;
+ g2h_fence->reason =
+ FIELD_GET(GUC_HXG_RETRY_MSG_0_REASON, msg[1]);
+ } else if (g2h_fence->response_buffer) {
+ g2h_fence->response_len = response_len;
+ memcpy(g2h_fence->response_buffer, msg + GUC_CTB_MSG_MIN_LEN,
+ response_len * sizeof(u32));
+ }
+
+ g2h_release_space(ct, GUC_CTB_HXG_MSG_MAX_LEN);
+
+ g2h_fence->done = true;
+ smp_mb();
+
+ wake_up_all(&ct->g2h_fence_wq);
+
+ return 0;
+}
+
+static int parse_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ u32 hxg, origin, type;
+ int ret;
+
+ lockdep_assert_held(&ct->lock);
+
+ hxg = msg[1];
+
+ origin = FIELD_GET(GUC_HXG_MSG_0_ORIGIN, hxg);
+ if (unlikely(origin != GUC_HXG_ORIGIN_GUC)) {
+ drm_err(&xe->drm,
+ "G2H channel broken on read, origin=%d, reset required\n",
+ origin);
+ ct->ctbs.g2h.info.broken = true;
+
+ return -EPROTO;
+ }
+
+ type = FIELD_GET(GUC_HXG_MSG_0_TYPE, hxg);
+ switch (type) {
+ case GUC_HXG_TYPE_EVENT:
+ ret = parse_g2h_event(ct, msg, len);
+ break;
+ case GUC_HXG_TYPE_RESPONSE_SUCCESS:
+ case GUC_HXG_TYPE_RESPONSE_FAILURE:
+ case GUC_HXG_TYPE_NO_RESPONSE_RETRY:
+ ret = parse_g2h_response(ct, msg, len);
+ break;
+ default:
+ drm_err(&xe->drm,
+ "G2H channel broken on read, type=%d, reset required\n",
+ type);
+ ct->ctbs.g2h.info.broken = true;
+
+ ret = -EOPNOTSUPP;
+ }
+
+ return ret;
+}
+
+static int process_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct xe_guc *guc = ct_to_guc(ct);
+ u32 action = FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[1]);
+ u32 *payload = msg + GUC_CTB_HXG_MSG_MIN_LEN;
+ u32 adj_len = len - GUC_CTB_HXG_MSG_MIN_LEN;
+ int ret = 0;
+
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[1]) != GUC_HXG_TYPE_EVENT)
+ return 0;
+
+ switch (action) {
+ case XE_GUC_ACTION_SCHED_CONTEXT_MODE_DONE:
+ ret = xe_guc_sched_done_handler(guc, payload, adj_len);
+ break;
+ case XE_GUC_ACTION_DEREGISTER_CONTEXT_DONE:
+ ret = xe_guc_deregister_done_handler(guc, payload, adj_len);
+ break;
+ case XE_GUC_ACTION_CONTEXT_RESET_NOTIFICATION:
+ ret = xe_guc_exec_queue_reset_handler(guc, payload, adj_len);
+ break;
+ case XE_GUC_ACTION_ENGINE_FAILURE_NOTIFICATION:
+ ret = xe_guc_exec_queue_reset_failure_handler(guc, payload,
+ adj_len);
+ break;
+ case XE_GUC_ACTION_SCHED_ENGINE_MODE_DONE:
+ /* Selftest only at the moment */
+ break;
+ case XE_GUC_ACTION_STATE_CAPTURE_NOTIFICATION:
+ case XE_GUC_ACTION_NOTIFY_FLUSH_LOG_BUFFER_TO_FILE:
+ /* FIXME: Handle this */
+ break;
+ case XE_GUC_ACTION_NOTIFY_MEMORY_CAT_ERROR:
+ ret = xe_guc_exec_queue_memory_cat_error_handler(guc, payload,
+ adj_len);
+ break;
+ case XE_GUC_ACTION_REPORT_PAGE_FAULT_REQ_DESC:
+ ret = xe_guc_pagefault_handler(guc, payload, adj_len);
+ break;
+ case XE_GUC_ACTION_TLB_INVALIDATION_DONE:
+ ret = xe_guc_tlb_invalidation_done_handler(guc, payload,
+ adj_len);
+ break;
+ case XE_GUC_ACTION_ACCESS_COUNTER_NOTIFY:
+ ret = xe_guc_access_counter_notify_handler(guc, payload,
+ adj_len);
+ break;
+ default:
+ drm_err(&xe->drm, "unexpected action 0x%04x\n", action);
+ }
+
+ if (ret)
+ drm_err(&xe->drm, "action 0x%04x failed processing, ret=%d\n",
+ action, ret);
+
+ return 0;
+}
+
+static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct guc_ctb *g2h = &ct->ctbs.g2h;
+ u32 tail, head, len;
+ s32 avail;
+ u32 action;
+
+ lockdep_assert_held(&ct->fast_lock);
+
+ if (!ct->enabled)
+ return -ENODEV;
+
+ if (g2h->info.broken)
+ return -EPIPE;
+
+ /* Calculate DW available to read */
+ tail = desc_read(xe, g2h, tail);
+ avail = tail - g2h->info.head;
+ if (unlikely(avail == 0))
+ return 0;
+
+ if (avail < 0)
+ avail += g2h->info.size;
+
+ /* Read header */
+ xe_map_memcpy_from(xe, msg, &g2h->cmds, sizeof(u32) * g2h->info.head,
+ sizeof(u32));
+ len = FIELD_GET(GUC_CTB_MSG_0_NUM_DWORDS, msg[0]) + GUC_CTB_MSG_MIN_LEN;
+ if (len > avail) {
+ drm_err(&xe->drm,
+ "G2H channel broken on read, avail=%d, len=%d, reset required\n",
+ avail, len);
+ g2h->info.broken = true;
+
+ return -EPROTO;
+ }
+
+ head = (g2h->info.head + 1) % g2h->info.size;
+ avail = len - 1;
+
+ /* Read G2H message */
+ if (avail + head > g2h->info.size) {
+ u32 avail_til_wrap = g2h->info.size - head;
+
+ xe_map_memcpy_from(xe, msg + 1,
+ &g2h->cmds, sizeof(u32) * head,
+ avail_til_wrap * sizeof(u32));
+ xe_map_memcpy_from(xe, msg + 1 + avail_til_wrap,
+ &g2h->cmds, 0,
+ (avail - avail_til_wrap) * sizeof(u32));
+ } else {
+ xe_map_memcpy_from(xe, msg + 1,
+ &g2h->cmds, sizeof(u32) * head,
+ avail * sizeof(u32));
+ }
+
+ action = FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[1]);
+
+ if (fast_path) {
+ if (FIELD_GET(GUC_HXG_MSG_0_TYPE, msg[1]) != GUC_HXG_TYPE_EVENT)
+ return 0;
+
+ switch (action) {
+ case XE_GUC_ACTION_REPORT_PAGE_FAULT_REQ_DESC:
+ case XE_GUC_ACTION_TLB_INVALIDATION_DONE:
+ break; /* Process these in fast-path */
+ default:
+ return 0;
+ }
+ }
+
+ /* Update local / descriptor header */
+ g2h->info.head = (head + avail) % g2h->info.size;
+ desc_write(xe, g2h, head, g2h->info.head);
+
+ trace_xe_guc_ctb_g2h(ct_to_gt(ct)->info.id, action, len,
+ g2h->info.head, tail);
+
+ return len;
+}
+
+static void g2h_fast_path(struct xe_guc_ct *ct, u32 *msg, u32 len)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct xe_guc *guc = ct_to_guc(ct);
+ u32 action = FIELD_GET(GUC_HXG_EVENT_MSG_0_ACTION, msg[1]);
+ u32 *payload = msg + GUC_CTB_HXG_MSG_MIN_LEN;
+ u32 adj_len = len - GUC_CTB_HXG_MSG_MIN_LEN;
+ int ret = 0;
+
+ switch (action) {
+ case XE_GUC_ACTION_REPORT_PAGE_FAULT_REQ_DESC:
+ ret = xe_guc_pagefault_handler(guc, payload, adj_len);
+ break;
+ case XE_GUC_ACTION_TLB_INVALIDATION_DONE:
+ __g2h_release_space(ct, len);
+ ret = xe_guc_tlb_invalidation_done_handler(guc, payload,
+ adj_len);
+ break;
+ default:
+ drm_warn(&xe->drm, "NOT_POSSIBLE");
+ }
+
+ if (ret)
+ drm_err(&xe->drm, "action 0x%04x failed processing, ret=%d\n",
+ action, ret);
+}
+
+/**
+ * xe_guc_ct_fast_path - process critical G2H in the IRQ handler
+ * @ct: GuC CT object
+ *
+ * Anything related to page faults is critical for performance, process these
+ * critical G2H in the IRQ. This is safe as these handlers either just wake up
+ * waiters or queue another worker.
+ */
+void xe_guc_ct_fast_path(struct xe_guc_ct *ct)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ bool ongoing;
+ int len;
+
+ ongoing = xe_device_mem_access_get_if_ongoing(ct_to_xe(ct));
+ if (!ongoing && xe_pm_read_callback_task(ct_to_xe(ct)) == NULL)
+ return;
+
+ spin_lock(&ct->fast_lock);
+ do {
+ len = g2h_read(ct, ct->fast_msg, true);
+ if (len > 0)
+ g2h_fast_path(ct, ct->fast_msg, len);
+ } while (len > 0);
+ spin_unlock(&ct->fast_lock);
+
+ if (ongoing)
+ xe_device_mem_access_put(xe);
+}
+
+/* Returns less than zero on error, 0 on done, 1 on more available */
+static int dequeue_one_g2h(struct xe_guc_ct *ct)
+{
+ int len;
+ int ret;
+
+ lockdep_assert_held(&ct->lock);
+
+ spin_lock_irq(&ct->fast_lock);
+ len = g2h_read(ct, ct->msg, false);
+ spin_unlock_irq(&ct->fast_lock);
+ if (len <= 0)
+ return len;
+
+ ret = parse_g2h_msg(ct, ct->msg, len);
+ if (unlikely(ret < 0))
+ return ret;
+
+ ret = process_g2h_msg(ct, ct->msg, len);
+ if (unlikely(ret < 0))
+ return ret;
+
+ return 1;
+}
+
+static void g2h_worker_func(struct work_struct *w)
+{
+ struct xe_guc_ct *ct = container_of(w, struct xe_guc_ct, g2h_worker);
+ bool ongoing;
+ int ret;
+
+ /*
+ * Normal users must always hold mem_access.ref around CT calls. However
+ * during the runtime pm callbacks we rely on CT to talk to the GuC, but
+ * at this stage we can't rely on mem_access.ref and even the
+ * callback_task will be different than current. For such cases we just
+ * need to ensure we always process the responses from any blocking
+ * ct_send requests or where we otherwise expect some response when
+ * initiated from those callbacks (which will need to wait for the below
+ * dequeue_one_g2h()). The dequeue_one_g2h() will gracefully fail if
+ * the device has suspended to the point that the CT communication has
+ * been disabled.
+ *
+ * If we are inside the runtime pm callback, we can be the only task
+ * still issuing CT requests (since that requires having the
+ * mem_access.ref). It seems like it might in theory be possible to
+ * receive unsolicited events from the GuC just as we are
+ * suspending-resuming, but those will currently anyway be lost when
+ * eventually exiting from suspend, hence no need to wake up the device
+ * here. If we ever need something stronger than get_if_ongoing() then
+ * we need to be careful with blocking the pm callbacks from getting CT
+ * responses, if the worker here is blocked on those callbacks
+ * completing, creating a deadlock.
+ */
+ ongoing = xe_device_mem_access_get_if_ongoing(ct_to_xe(ct));
+ if (!ongoing && xe_pm_read_callback_task(ct_to_xe(ct)) == NULL)
+ return;
+
+ do {
+ mutex_lock(&ct->lock);
+ ret = dequeue_one_g2h(ct);
+ mutex_unlock(&ct->lock);
+
+ if (unlikely(ret == -EPROTO || ret == -EOPNOTSUPP)) {
+ struct drm_device *drm = &ct_to_xe(ct)->drm;
+ struct drm_printer p = drm_info_printer(drm->dev);
+
+ xe_guc_ct_print(ct, &p, false);
+ kick_reset(ct);
+ }
+ } while (ret == 1);
+
+ if (ongoing)
+ xe_device_mem_access_put(ct_to_xe(ct));
+}
+
+static void guc_ctb_snapshot_capture(struct xe_device *xe, struct guc_ctb *ctb,
+ struct guc_ctb_snapshot *snapshot,
+ bool atomic)
+{
+ u32 head, tail;
+
+ xe_map_memcpy_from(xe, &snapshot->desc, &ctb->desc, 0,
+ sizeof(struct guc_ct_buffer_desc));
+ memcpy(&snapshot->info, &ctb->info, sizeof(struct guc_ctb_info));
+
+ snapshot->cmds = kmalloc_array(ctb->info.size, sizeof(u32),
+ atomic ? GFP_ATOMIC : GFP_KERNEL);
+
+ if (!snapshot->cmds) {
+ drm_err(&xe->drm, "Skipping CTB commands snapshot. Only CTB info will be available.\n");
+ return;
+ }
+
+ head = snapshot->desc.head;
+ tail = snapshot->desc.tail;
+
+ if (head != tail) {
+ struct iosys_map map =
+ IOSYS_MAP_INIT_OFFSET(&ctb->cmds, head * sizeof(u32));
+
+ while (head != tail) {
+ snapshot->cmds[head] = xe_map_rd(xe, &map, 0, u32);
+ ++head;
+ if (head == ctb->info.size) {
+ head = 0;
+ map = ctb->cmds;
+ } else {
+ iosys_map_incr(&map, sizeof(u32));
+ }
+ }
+ }
+}
+
+static void guc_ctb_snapshot_print(struct guc_ctb_snapshot *snapshot,
+ struct drm_printer *p)
+{
+ u32 head, tail;
+
+ drm_printf(p, "\tsize: %d\n", snapshot->info.size);
+ drm_printf(p, "\tresv_space: %d\n", snapshot->info.resv_space);
+ drm_printf(p, "\thead: %d\n", snapshot->info.head);
+ drm_printf(p, "\ttail: %d\n", snapshot->info.tail);
+ drm_printf(p, "\tspace: %d\n", snapshot->info.space);
+ drm_printf(p, "\tbroken: %d\n", snapshot->info.broken);
+ drm_printf(p, "\thead (memory): %d\n", snapshot->desc.head);
+ drm_printf(p, "\ttail (memory): %d\n", snapshot->desc.tail);
+ drm_printf(p, "\tstatus (memory): 0x%x\n", snapshot->desc.status);
+
+ if (!snapshot->cmds)
+ return;
+
+ head = snapshot->desc.head;
+ tail = snapshot->desc.tail;
+
+ while (head != tail) {
+ drm_printf(p, "\tcmd[%d]: 0x%08x\n", head,
+ snapshot->cmds[head]);
+ ++head;
+ if (head == snapshot->info.size)
+ head = 0;
+ }
+}
+
+static void guc_ctb_snapshot_free(struct guc_ctb_snapshot *snapshot)
+{
+ kfree(snapshot->cmds);
+}
+
+/**
+ * xe_guc_ct_snapshot_capture - Take a quick snapshot of the CT state.
+ * @ct: GuC CT object.
+ * @atomic: Boolean to indicate if this is called from atomic context like
+ * reset or CTB handler or from some regular path like debugfs.
+ *
+ * This can be printed out in a later stage like during dev_coredump
+ * analysis.
+ *
+ * Returns: a GuC CT snapshot object that must be freed by the caller
+ * by using `xe_guc_ct_snapshot_free`.
+ */
+struct xe_guc_ct_snapshot *xe_guc_ct_snapshot_capture(struct xe_guc_ct *ct,
+ bool atomic)
+{
+ struct xe_device *xe = ct_to_xe(ct);
+ struct xe_guc_ct_snapshot *snapshot;
+
+ snapshot = kzalloc(sizeof(*snapshot),
+ atomic ? GFP_ATOMIC : GFP_KERNEL);
+
+ if (!snapshot) {
+ drm_err(&xe->drm, "Skipping CTB snapshot entirely.\n");
+ return NULL;
+ }
+
+ if (ct->enabled) {
+ snapshot->ct_enabled = true;
+ snapshot->g2h_outstanding = READ_ONCE(ct->g2h_outstanding);
+ guc_ctb_snapshot_capture(xe, &ct->ctbs.h2g,
+ &snapshot->h2g, atomic);
+ guc_ctb_snapshot_capture(xe, &ct->ctbs.g2h,
+ &snapshot->g2h, atomic);
+ }
+
+ return snapshot;
+}
+
+/**
+ * xe_guc_ct_snapshot_print - Print out a given GuC CT snapshot.
+ * @snapshot: GuC CT snapshot object.
+ * @p: drm_printer where it will be printed out.
+ *
+ * This function prints out a given GuC CT snapshot object.
+ */
+void xe_guc_ct_snapshot_print(struct xe_guc_ct_snapshot *snapshot,
+ struct drm_printer *p)
+{
+ if (!snapshot)
+ return;
+
+ if (snapshot->ct_enabled) {
+ drm_puts(p, "\nH2G CTB (all sizes in DW):\n");
+ guc_ctb_snapshot_print(&snapshot->h2g, p);
+
+ drm_puts(p, "\nG2H CTB (all sizes in DW):\n");
+ guc_ctb_snapshot_print(&snapshot->g2h, p);
+
+ drm_printf(p, "\tg2h outstanding: %d\n",
+ snapshot->g2h_outstanding);
+ } else {
+ drm_puts(p, "\nCT disabled\n");
+ }
+}
+
+/**
+ * xe_guc_ct_snapshot_free - Free all allocated objects for a given snapshot.
+ * @snapshot: GuC CT snapshot object.
+ *
+ * This function free all the memory that needed to be allocated at capture
+ * time.
+ */
+void xe_guc_ct_snapshot_free(struct xe_guc_ct_snapshot *snapshot)
+{
+ if (!snapshot)
+ return;
+
+ guc_ctb_snapshot_free(&snapshot->h2g);
+ guc_ctb_snapshot_free(&snapshot->g2h);
+ kfree(snapshot);
+}
+
+/**
+ * xe_guc_ct_print - GuC CT Print.
+ * @ct: GuC CT.
+ * @p: drm_printer where it will be printed out.
+ * @atomic: Boolean to indicate if this is called from atomic context like
+ * reset or CTB handler or from some regular path like debugfs.
+ *
+ * This function quickly capture a snapshot and immediately print it out.
+ */
+void xe_guc_ct_print(struct xe_guc_ct *ct, struct drm_printer *p, bool atomic)
+{
+ struct xe_guc_ct_snapshot *snapshot;
+
+ snapshot = xe_guc_ct_snapshot_capture(ct, atomic);
+ xe_guc_ct_snapshot_print(snapshot, p);
+ xe_guc_ct_snapshot_free(snapshot);
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.h b/drivers/gpu/drm/xe/xe_guc_ct.h
new file mode 100644
index 000000000000..f15f8a4857e0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ct.h
@@ -0,0 +1,59 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_CT_H_
+#define _XE_GUC_CT_H_
+
+#include "xe_guc_ct_types.h"
+
+struct drm_printer;
+
+int xe_guc_ct_init(struct xe_guc_ct *ct);
+int xe_guc_ct_enable(struct xe_guc_ct *ct);
+void xe_guc_ct_disable(struct xe_guc_ct *ct);
+void xe_guc_ct_fast_path(struct xe_guc_ct *ct);
+
+struct xe_guc_ct_snapshot *
+xe_guc_ct_snapshot_capture(struct xe_guc_ct *ct, bool atomic);
+void xe_guc_ct_snapshot_print(struct xe_guc_ct_snapshot *snapshot,
+ struct drm_printer *p);
+void xe_guc_ct_snapshot_free(struct xe_guc_ct_snapshot *snapshot);
+void xe_guc_ct_print(struct xe_guc_ct *ct, struct drm_printer *p, bool atomic);
+
+static inline void xe_guc_ct_irq_handler(struct xe_guc_ct *ct)
+{
+ wake_up_all(&ct->wq);
+ if (ct->enabled)
+ queue_work(system_unbound_wq, &ct->g2h_worker);
+ xe_guc_ct_fast_path(ct);
+}
+
+/* Basic CT send / receives */
+int xe_guc_ct_send(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h);
+int xe_guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 g2h_len, u32 num_g2h);
+int xe_guc_ct_send_recv(struct xe_guc_ct *ct, const u32 *action, u32 len,
+ u32 *response_buffer);
+static inline int
+xe_guc_ct_send_block(struct xe_guc_ct *ct, const u32 *action, u32 len)
+{
+ return xe_guc_ct_send_recv(ct, action, len, NULL);
+}
+
+/* This is only version of the send CT you can call from a G2H handler */
+int xe_guc_ct_send_g2h_handler(struct xe_guc_ct *ct, const u32 *action,
+ u32 len);
+
+/* Can't fail because a GT reset is in progress */
+int xe_guc_ct_send_recv_no_fail(struct xe_guc_ct *ct, const u32 *action,
+ u32 len, u32 *response_buffer);
+static inline int
+xe_guc_ct_send_block_no_fail(struct xe_guc_ct *ct, const u32 *action, u32 len)
+{
+ return xe_guc_ct_send_recv_no_fail(ct, action, len, NULL);
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ct_types.h b/drivers/gpu/drm/xe/xe_guc_ct_types.h
new file mode 100644
index 000000000000..d814d4ee3fc6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_ct_types.h
@@ -0,0 +1,115 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_CT_TYPES_H_
+#define _XE_GUC_CT_TYPES_H_
+
+#include <linux/interrupt.h>
+#include <linux/iosys-map.h>
+#include <linux/spinlock_types.h>
+#include <linux/wait.h>
+#include <linux/xarray.h>
+
+#include "abi/guc_communication_ctb_abi.h"
+
+struct xe_bo;
+
+/**
+ * struct guc_ctb_info - GuC command transport buffer (CTB) info
+ */
+struct guc_ctb_info {
+ /** @size: size of CTB commands (DW) */
+ u32 size;
+ /** @resv_space: reserved space of CTB commands (DW) */
+ u32 resv_space;
+ /** @head: head of CTB commands (DW) */
+ u32 head;
+ /** @tail: tail of CTB commands (DW) */
+ u32 tail;
+ /** @space: space in CTB commands (DW) */
+ u32 space;
+ /** @broken: channel broken */
+ bool broken;
+};
+
+/**
+ * struct guc_ctb - GuC command transport buffer (CTB)
+ */
+struct guc_ctb {
+ /** @desc: dma buffer map for CTB descriptor */
+ struct iosys_map desc;
+ /** @cmds: dma buffer map for CTB commands */
+ struct iosys_map cmds;
+ /** @info: CTB info */
+ struct guc_ctb_info info;
+};
+
+/**
+ * struct guc_ctb_snapshot - GuC command transport buffer (CTB) snapshot
+ */
+struct guc_ctb_snapshot {
+ /** @desc: snapshot of the CTB descriptor */
+ struct guc_ct_buffer_desc desc;
+ /** @cmds: snapshot of the CTB commands */
+ u32 *cmds;
+ /** @info: snapshot of the CTB info */
+ struct guc_ctb_info info;
+};
+
+/**
+ * struct xe_guc_ct_snapshot - GuC command transport (CT) snapshot
+ */
+struct xe_guc_ct_snapshot {
+ /** @ct_enabled: CT enabled info at capture time. */
+ bool ct_enabled;
+ /** @g2h_outstanding: G2H outstanding info at the capture time */
+ u32 g2h_outstanding;
+ /** @g2h: G2H CTB snapshot */
+ struct guc_ctb_snapshot g2h;
+ /** @h2g: H2G CTB snapshot */
+ struct guc_ctb_snapshot h2g;
+};
+
+/**
+ * struct xe_guc_ct - GuC command transport (CT) layer
+ *
+ * Includes a pair of CT buffers for bi-directional communication and tracking
+ * for the H2G and G2H requests sent and received through the buffers.
+ */
+struct xe_guc_ct {
+ /** @bo: XE BO for CT */
+ struct xe_bo *bo;
+ /** @lock: protects everything in CT layer */
+ struct mutex lock;
+ /** @fast_lock: protects G2H channel and credits */
+ spinlock_t fast_lock;
+ /** @ctbs: buffers for sending and receiving commands */
+ struct {
+ /** @send: Host to GuC (H2G, send) channel */
+ struct guc_ctb h2g;
+ /** @recv: GuC to Host (G2H, receive) channel */
+ struct guc_ctb g2h;
+ } ctbs;
+ /** @g2h_outstanding: number of outstanding G2H */
+ u32 g2h_outstanding;
+ /** @g2h_worker: worker to process G2H messages */
+ struct work_struct g2h_worker;
+ /** @enabled: CT enabled */
+ bool enabled;
+ /** @fence_seqno: G2H fence seqno - 16 bits used by CT */
+ u32 fence_seqno;
+ /** @fence_lookup: G2H fence lookup */
+ struct xarray fence_lookup;
+ /** @wq: wait queue used for reliable CT sends and freeing G2H credits */
+ wait_queue_head_t wq;
+ /** @g2h_fence_wq: wait queue used for G2H fencing */
+ wait_queue_head_t g2h_fence_wq;
+ /** @msg: Message buffer */
+ u32 msg[GUC_CTB_MSG_MAX_LEN];
+ /** @fast_msg: Message buffer */
+ u32 fast_msg[GUC_CTB_MSG_MAX_LEN];
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_debugfs.c b/drivers/gpu/drm/xe/xe_guc_debugfs.c
new file mode 100644
index 000000000000..ffd7d53bcc42
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_debugfs.c
@@ -0,0 +1,74 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_debugfs.h"
+
+#include <drm/drm_debugfs.h>
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_guc_ct.h"
+#include "xe_guc_log.h"
+#include "xe_macros.h"
+
+static struct xe_guc *node_to_guc(struct drm_info_node *node)
+{
+ return node->info_ent->data;
+}
+
+static int guc_info(struct seq_file *m, void *data)
+{
+ struct xe_guc *guc = node_to_guc(m->private);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_device_mem_access_get(xe);
+ xe_guc_print_info(guc, &p);
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+static int guc_log(struct seq_file *m, void *data)
+{
+ struct xe_guc *guc = node_to_guc(m->private);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_device_mem_access_get(xe);
+ xe_guc_log_print(&guc->log, &p);
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+static const struct drm_info_list debugfs_list[] = {
+ {"guc_info", guc_info, 0},
+ {"guc_log", guc_log, 0},
+};
+
+void xe_guc_debugfs_register(struct xe_guc *guc, struct dentry *parent)
+{
+ struct drm_minor *minor = guc_to_xe(guc)->drm.primary;
+ struct drm_info_list *local;
+ int i;
+
+#define DEBUGFS_SIZE (ARRAY_SIZE(debugfs_list) * sizeof(struct drm_info_list))
+ local = drmm_kmalloc(&guc_to_xe(guc)->drm, DEBUGFS_SIZE, GFP_KERNEL);
+ if (!local)
+ return;
+
+ memcpy(local, debugfs_list, DEBUGFS_SIZE);
+#undef DEBUGFS_SIZE
+
+ for (i = 0; i < ARRAY_SIZE(debugfs_list); ++i)
+ local[i].data = guc;
+
+ drm_debugfs_create_files(local,
+ ARRAY_SIZE(debugfs_list),
+ parent, minor);
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_debugfs.h b/drivers/gpu/drm/xe/xe_guc_debugfs.h
new file mode 100644
index 000000000000..4756dff26fca
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_debugfs.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_DEBUGFS_H_
+#define _XE_GUC_DEBUGFS_H_
+
+struct dentry;
+struct xe_guc;
+
+void xe_guc_debugfs_register(struct xe_guc *guc, struct dentry *parent);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
new file mode 100644
index 000000000000..4c39f01e4f52
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_exec_queue_types.h
@@ -0,0 +1,54 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_ENGINE_TYPES_H_
+#define _XE_GUC_ENGINE_TYPES_H_
+
+#include <linux/spinlock.h>
+#include <linux/workqueue.h>
+
+#include "xe_gpu_scheduler_types.h"
+
+struct dma_fence;
+struct xe_exec_queue;
+
+/**
+ * struct xe_guc_exec_queue - GuC specific state for an xe_exec_queue
+ */
+struct xe_guc_exec_queue {
+ /** @q: Backpointer to parent xe_exec_queue */
+ struct xe_exec_queue *q;
+ /** @sched: GPU scheduler for this xe_exec_queue */
+ struct xe_gpu_scheduler sched;
+ /** @entity: Scheduler entity for this xe_exec_queue */
+ struct xe_sched_entity entity;
+ /**
+ * @static_msgs: Static messages for this xe_exec_queue, used when
+ * a message needs to sent through the GPU scheduler but memory
+ * allocations are not allowed.
+ */
+#define MAX_STATIC_MSG_TYPE 3
+ struct xe_sched_msg static_msgs[MAX_STATIC_MSG_TYPE];
+ /** @lr_tdr: long running TDR worker */
+ struct work_struct lr_tdr;
+ /** @fini_async: do final fini async from this worker */
+ struct work_struct fini_async;
+ /** @resume_time: time of last resume */
+ u64 resume_time;
+ /** @state: GuC specific state for this xe_exec_queue */
+ atomic_t state;
+ /** @wqi_head: work queue item tail */
+ u32 wqi_head;
+ /** @wqi_tail: work queue item tail */
+ u32 wqi_tail;
+ /** @id: GuC id for this exec_queue */
+ u16 id;
+ /** @suspend_wait: wait queue used to wait on pending suspends */
+ wait_queue_head_t suspend_wait;
+ /** @suspend_pending: a suspend of the exec_queue is pending */
+ bool suspend_pending;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_fwif.h b/drivers/gpu/drm/xe/xe_guc_fwif.h
new file mode 100644
index 000000000000..4dd5a88a7826
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_fwif.h
@@ -0,0 +1,361 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_FWIF_H
+#define _XE_GUC_FWIF_H
+
+#include <linux/bits.h>
+
+#include "abi/guc_klvs_abi.h"
+
+#define G2H_LEN_DW_SCHED_CONTEXT_MODE_SET 4
+#define G2H_LEN_DW_DEREGISTER_CONTEXT 3
+#define G2H_LEN_DW_TLB_INVALIDATE 3
+
+#define GUC_CONTEXT_DISABLE 0
+#define GUC_CONTEXT_ENABLE 1
+
+#define GUC_CLIENT_PRIORITY_KMD_HIGH 0
+#define GUC_CLIENT_PRIORITY_HIGH 1
+#define GUC_CLIENT_PRIORITY_KMD_NORMAL 2
+#define GUC_CLIENT_PRIORITY_NORMAL 3
+#define GUC_CLIENT_PRIORITY_NUM 4
+
+#define GUC_RENDER_ENGINE 0
+#define GUC_VIDEO_ENGINE 1
+#define GUC_BLITTER_ENGINE 2
+#define GUC_VIDEOENHANCE_ENGINE 3
+#define GUC_VIDEO_ENGINE2 4
+#define GUC_MAX_ENGINES_NUM (GUC_VIDEO_ENGINE2 + 1)
+
+#define GUC_RENDER_CLASS 0
+#define GUC_VIDEO_CLASS 1
+#define GUC_VIDEOENHANCE_CLASS 2
+#define GUC_BLITTER_CLASS 3
+#define GUC_COMPUTE_CLASS 4
+#define GUC_GSC_OTHER_CLASS 5
+#define GUC_LAST_ENGINE_CLASS GUC_GSC_OTHER_CLASS
+#define GUC_MAX_ENGINE_CLASSES 16
+#define GUC_MAX_INSTANCES_PER_CLASS 32
+
+/* Helper for context registration H2G */
+struct guc_ctxt_registration_info {
+ u32 flags;
+ u32 context_idx;
+ u32 engine_class;
+ u32 engine_submit_mask;
+ u32 wq_desc_lo;
+ u32 wq_desc_hi;
+ u32 wq_base_lo;
+ u32 wq_base_hi;
+ u32 wq_size;
+ u32 hwlrca_lo;
+ u32 hwlrca_hi;
+};
+#define CONTEXT_REGISTRATION_FLAG_KMD BIT(0)
+
+/* 32-bit KLV structure as used by policy updates and others */
+struct guc_klv_generic_dw_t {
+ u32 kl;
+ u32 value;
+} __packed;
+
+/* Format of the UPDATE_CONTEXT_POLICIES H2G data packet */
+struct guc_update_exec_queue_policy_header {
+ u32 action;
+ u32 guc_id;
+} __packed;
+
+struct guc_update_exec_queue_policy {
+ struct guc_update_exec_queue_policy_header header;
+ struct guc_klv_generic_dw_t klv[GUC_CONTEXT_POLICIES_KLV_NUM_IDS];
+} __packed;
+
+/* GUC_CTL_* - Parameters for loading the GuC */
+#define GUC_CTL_LOG_PARAMS 0
+#define GUC_LOG_VALID BIT(0)
+#define GUC_LOG_NOTIFY_ON_HALF_FULL BIT(1)
+#define GUC_LOG_CAPTURE_ALLOC_UNITS BIT(2)
+#define GUC_LOG_LOG_ALLOC_UNITS BIT(3)
+#define GUC_LOG_CRASH_SHIFT 4
+#define GUC_LOG_CRASH_MASK (0x3 << GUC_LOG_CRASH_SHIFT)
+#define GUC_LOG_DEBUG_SHIFT 6
+#define GUC_LOG_DEBUG_MASK (0xF << GUC_LOG_DEBUG_SHIFT)
+#define GUC_LOG_CAPTURE_SHIFT 10
+#define GUC_LOG_CAPTURE_MASK (0x3 << GUC_LOG_CAPTURE_SHIFT)
+#define GUC_LOG_BUF_ADDR_SHIFT 12
+
+#define GUC_CTL_WA 1
+#define GUC_WA_GAM_CREDITS BIT(10)
+#define GUC_WA_DUAL_QUEUE BIT(11)
+#define GUC_WA_RCS_RESET_BEFORE_RC6 BIT(13)
+#define GUC_WA_CONTEXT_ISOLATION BIT(15)
+#define GUC_WA_PRE_PARSER BIT(14)
+#define GUC_WA_HOLD_CCS_SWITCHOUT BIT(17)
+#define GUC_WA_POLLCS BIT(18)
+#define GUC_WA_RENDER_RST_RC6_EXIT BIT(19)
+#define GUC_WA_RCS_REGS_IN_CCS_REGS_LIST BIT(21)
+
+#define GUC_CTL_FEATURE 2
+#define GUC_CTL_ENABLE_SLPC BIT(2)
+#define GUC_CTL_DISABLE_SCHEDULER BIT(14)
+
+#define GUC_CTL_DEBUG 3
+#define GUC_LOG_VERBOSITY_SHIFT 0
+#define GUC_LOG_VERBOSITY_LOW (0 << GUC_LOG_VERBOSITY_SHIFT)
+#define GUC_LOG_VERBOSITY_MED (1 << GUC_LOG_VERBOSITY_SHIFT)
+#define GUC_LOG_VERBOSITY_HIGH (2 << GUC_LOG_VERBOSITY_SHIFT)
+#define GUC_LOG_VERBOSITY_ULTRA (3 << GUC_LOG_VERBOSITY_SHIFT)
+#define GUC_LOG_VERBOSITY_MIN 0
+#define GUC_LOG_VERBOSITY_MAX 3
+#define GUC_LOG_VERBOSITY_MASK 0x0000000f
+#define GUC_LOG_DESTINATION_MASK (3 << 4)
+#define GUC_LOG_DISABLED (1 << 6)
+#define GUC_PROFILE_ENABLED (1 << 7)
+
+#define GUC_CTL_ADS 4
+#define GUC_ADS_ADDR_SHIFT 1
+#define GUC_ADS_ADDR_MASK (0xFFFFF << GUC_ADS_ADDR_SHIFT)
+
+#define GUC_CTL_DEVID 5
+
+#define GUC_CTL_MAX_DWORDS 14
+
+/* Scheduling policy settings */
+
+#define GLOBAL_POLICY_MAX_NUM_WI 15
+
+/* Don't reset an engine upon preemption failure */
+#define GLOBAL_POLICY_DISABLE_ENGINE_RESET BIT(0)
+
+#define GLOBAL_POLICY_DEFAULT_DPC_PROMOTE_TIME_US 500000
+
+struct guc_policies {
+ u32 submission_queue_depth[GUC_MAX_ENGINE_CLASSES];
+ /*
+ * In micro seconds. How much time to allow before DPC processing is
+ * called back via interrupt (to prevent DPC queue drain starving).
+ * Typically 1000s of micro seconds (example only, not granularity).
+ */
+ u32 dpc_promote_time;
+
+ /* Must be set to take these new values. */
+ u32 is_valid;
+
+ /*
+ * Max number of WIs to process per call. A large value may keep CS
+ * idle.
+ */
+ u32 max_num_work_items;
+
+ u32 global_flags;
+ u32 reserved[4];
+} __packed;
+
+/* GuC MMIO reg state struct */
+struct guc_mmio_reg {
+ u32 offset;
+ u32 value;
+ u32 flags;
+ u32 mask;
+#define GUC_REGSET_MASKED BIT(0)
+#define GUC_REGSET_MASKED_WITH_VALUE BIT(2)
+#define GUC_REGSET_RESTORE_ONLY BIT(3)
+} __packed;
+
+/* GuC register sets */
+struct guc_mmio_reg_set {
+ u32 address;
+ u16 count;
+ u16 reserved;
+} __packed;
+
+/* Generic GT SysInfo data types */
+#define GUC_GENERIC_GT_SYSINFO_SLICE_ENABLED 0
+#define GUC_GENERIC_GT_SYSINFO_VDBOX_SFC_SUPPORT_MASK 1
+#define GUC_GENERIC_GT_SYSINFO_DOORBELL_COUNT_PER_SQIDI 2
+#define GUC_GENERIC_GT_SYSINFO_MAX 16
+
+/* HW info */
+struct guc_gt_system_info {
+ u8 mapping_table[GUC_MAX_ENGINE_CLASSES][GUC_MAX_INSTANCES_PER_CLASS];
+ u32 engine_enabled_masks[GUC_MAX_ENGINE_CLASSES];
+ u32 generic_gt_sysinfo[GUC_GENERIC_GT_SYSINFO_MAX];
+} __packed;
+
+enum {
+ GUC_CAPTURE_LIST_INDEX_PF = 0,
+ GUC_CAPTURE_LIST_INDEX_VF = 1,
+ GUC_CAPTURE_LIST_INDEX_MAX = 2,
+};
+
+/* GuC Additional Data Struct */
+struct guc_ads {
+ struct guc_mmio_reg_set reg_state_list[GUC_MAX_ENGINE_CLASSES][GUC_MAX_INSTANCES_PER_CLASS];
+ u32 reserved0;
+ u32 scheduler_policies;
+ u32 gt_system_info;
+ u32 reserved1;
+ u32 control_data;
+ u32 golden_context_lrca[GUC_MAX_ENGINE_CLASSES];
+ u32 eng_state_size[GUC_MAX_ENGINE_CLASSES];
+ u32 private_data;
+ u32 um_init_data;
+ u32 capture_instance[GUC_CAPTURE_LIST_INDEX_MAX][GUC_MAX_ENGINE_CLASSES];
+ u32 capture_class[GUC_CAPTURE_LIST_INDEX_MAX][GUC_MAX_ENGINE_CLASSES];
+ u32 capture_global[GUC_CAPTURE_LIST_INDEX_MAX];
+ u32 reserved[14];
+} __packed;
+
+/* Engine usage stats */
+struct guc_engine_usage_record {
+ u32 current_context_index;
+ u32 last_switch_in_stamp;
+ u32 reserved0;
+ u32 total_runtime;
+ u32 reserved1[4];
+} __packed;
+
+struct guc_engine_usage {
+ struct guc_engine_usage_record engines[GUC_MAX_ENGINE_CLASSES][GUC_MAX_INSTANCES_PER_CLASS];
+} __packed;
+
+/* This action will be programmed in C1BC - SOFT_SCRATCH_15_REG */
+enum xe_guc_recv_message {
+ XE_GUC_RECV_MSG_CRASH_DUMP_POSTED = BIT(1),
+ XE_GUC_RECV_MSG_EXCEPTION = BIT(30),
+};
+
+/* Page fault structures */
+struct access_counter_desc {
+ u32 dw0;
+#define ACCESS_COUNTER_TYPE BIT(0)
+#define ACCESS_COUNTER_SUBG_LO GENMASK(31, 1)
+
+ u32 dw1;
+#define ACCESS_COUNTER_SUBG_HI BIT(0)
+#define ACCESS_COUNTER_RSVD0 GENMASK(2, 1)
+#define ACCESS_COUNTER_ENG_INSTANCE GENMASK(8, 3)
+#define ACCESS_COUNTER_ENG_CLASS GENMASK(11, 9)
+#define ACCESS_COUNTER_ASID GENMASK(31, 12)
+
+ u32 dw2;
+#define ACCESS_COUNTER_VFID GENMASK(5, 0)
+#define ACCESS_COUNTER_RSVD1 GENMASK(7, 6)
+#define ACCESS_COUNTER_GRANULARITY GENMASK(10, 8)
+#define ACCESS_COUNTER_RSVD2 GENMASK(16, 11)
+#define ACCESS_COUNTER_VIRTUAL_ADDR_RANGE_LO GENMASK(31, 17)
+
+ u32 dw3;
+#define ACCESS_COUNTER_VIRTUAL_ADDR_RANGE_HI GENMASK(31, 0)
+} __packed;
+
+enum guc_um_queue_type {
+ GUC_UM_HW_QUEUE_PAGE_FAULT = 0,
+ GUC_UM_HW_QUEUE_PAGE_FAULT_RESPONSE,
+ GUC_UM_HW_QUEUE_ACCESS_COUNTER,
+ GUC_UM_HW_QUEUE_MAX
+};
+
+struct guc_um_queue_params {
+ u64 base_dpa;
+ u32 base_ggtt_address;
+ u32 size_in_bytes;
+ u32 rsvd[4];
+} __packed;
+
+struct guc_um_init_params {
+ u64 page_response_timeout_in_us;
+ u32 rsvd[6];
+ struct guc_um_queue_params queue_params[GUC_UM_HW_QUEUE_MAX];
+} __packed;
+
+enum xe_guc_fault_reply_type {
+ PFR_ACCESS = 0,
+ PFR_ENGINE,
+ PFR_VFID,
+ PFR_ALL,
+ PFR_INVALID
+};
+
+enum xe_guc_response_desc_type {
+ TLB_INVALIDATION_DESC = 0,
+ FAULT_RESPONSE_DESC
+};
+
+struct xe_guc_pagefault_desc {
+ u32 dw0;
+#define PFD_FAULT_LEVEL GENMASK(2, 0)
+#define PFD_SRC_ID GENMASK(10, 3)
+#define PFD_RSVD_0 GENMASK(17, 11)
+#define XE2_PFD_TRVA_FAULT BIT(18)
+#define PFD_ENG_INSTANCE GENMASK(24, 19)
+#define PFD_ENG_CLASS GENMASK(27, 25)
+#define PFD_PDATA_LO GENMASK(31, 28)
+
+ u32 dw1;
+#define PFD_PDATA_HI GENMASK(11, 0)
+#define PFD_PDATA_HI_SHIFT 4
+#define PFD_ASID GENMASK(31, 12)
+
+ u32 dw2;
+#define PFD_ACCESS_TYPE GENMASK(1, 0)
+#define PFD_FAULT_TYPE GENMASK(3, 2)
+#define PFD_VFID GENMASK(9, 4)
+#define PFD_RSVD_1 GENMASK(11, 10)
+#define PFD_VIRTUAL_ADDR_LO GENMASK(31, 12)
+#define PFD_VIRTUAL_ADDR_LO_SHIFT 12
+
+ u32 dw3;
+#define PFD_VIRTUAL_ADDR_HI GENMASK(31, 0)
+#define PFD_VIRTUAL_ADDR_HI_SHIFT 32
+} __packed;
+
+struct xe_guc_pagefault_reply {
+ u32 dw0;
+#define PFR_VALID BIT(0)
+#define PFR_SUCCESS BIT(1)
+#define PFR_REPLY GENMASK(4, 2)
+#define PFR_RSVD_0 GENMASK(9, 5)
+#define PFR_DESC_TYPE GENMASK(11, 10)
+#define PFR_ASID GENMASK(31, 12)
+
+ u32 dw1;
+#define PFR_VFID GENMASK(5, 0)
+#define PFR_RSVD_1 BIT(6)
+#define PFR_ENG_INSTANCE GENMASK(12, 7)
+#define PFR_ENG_CLASS GENMASK(15, 13)
+#define PFR_PDATA GENMASK(31, 16)
+
+ u32 dw2;
+#define PFR_RSVD_2 GENMASK(31, 0)
+} __packed;
+
+struct xe_guc_acc_desc {
+ u32 dw0;
+#define ACC_TYPE BIT(0)
+#define ACC_TRIGGER 0
+#define ACC_NOTIFY 1
+#define ACC_SUBG_LO GENMASK(31, 1)
+
+ u32 dw1;
+#define ACC_SUBG_HI BIT(0)
+#define ACC_RSVD0 GENMASK(2, 1)
+#define ACC_ENG_INSTANCE GENMASK(8, 3)
+#define ACC_ENG_CLASS GENMASK(11, 9)
+#define ACC_ASID GENMASK(31, 12)
+
+ u32 dw2;
+#define ACC_VFID GENMASK(5, 0)
+#define ACC_RSVD1 GENMASK(7, 6)
+#define ACC_GRANULARITY GENMASK(10, 8)
+#define ACC_RSVD2 GENMASK(16, 11)
+#define ACC_VIRTUAL_ADDR_RANGE_LO GENMASK(31, 17)
+
+ u32 dw3;
+#define ACC_VIRTUAL_ADDR_RANGE_HI GENMASK(31, 0)
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_hwconfig.c b/drivers/gpu/drm/xe/xe_guc_hwconfig.c
new file mode 100644
index 000000000000..2a13a00917f8
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_hwconfig.c
@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_hwconfig.h"
+
+#include <drm/drm_managed.h>
+
+#include "abi/guc_actions_abi.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_map.h"
+
+static int send_get_hwconfig(struct xe_guc *guc, u32 ggtt_addr, u32 size)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_GET_HWCONFIG,
+ lower_32_bits(ggtt_addr),
+ upper_32_bits(ggtt_addr),
+ size,
+ };
+
+ return xe_guc_mmio_send(guc, action, ARRAY_SIZE(action));
+}
+
+static int guc_hwconfig_size(struct xe_guc *guc, u32 *size)
+{
+ int ret = send_get_hwconfig(guc, 0, 0);
+
+ if (ret < 0)
+ return ret;
+
+ *size = ret;
+ return 0;
+}
+
+static int guc_hwconfig_copy(struct xe_guc *guc)
+{
+ int ret = send_get_hwconfig(guc, xe_bo_ggtt_addr(guc->hwconfig.bo),
+ guc->hwconfig.size);
+
+ if (ret < 0)
+ return ret;
+
+ return 0;
+}
+
+int xe_guc_hwconfig_init(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_bo *bo;
+ u32 size;
+ int err;
+
+ /* Initialization already done */
+ if (guc->hwconfig.bo)
+ return 0;
+
+ /*
+ * All hwconfig the same across GTs so only GT0 needs to be configured
+ */
+ if (gt->info.id != XE_GT0)
+ return 0;
+
+ /* ADL_P, DG2+ supports hwconfig table */
+ if (GRAPHICS_VERx100(xe) < 1255 && xe->info.platform != XE_ALDERLAKE_P)
+ return 0;
+
+ err = guc_hwconfig_size(guc, &size);
+ if (err)
+ return err;
+ if (!size)
+ return -EINVAL;
+
+ bo = xe_managed_bo_create_pin_map(xe, tile, PAGE_ALIGN(size),
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+ guc->hwconfig.bo = bo;
+ guc->hwconfig.size = size;
+
+ return guc_hwconfig_copy(guc);
+}
+
+u32 xe_guc_hwconfig_size(struct xe_guc *guc)
+{
+ return !guc->hwconfig.bo ? 0 : guc->hwconfig.size;
+}
+
+void xe_guc_hwconfig_copy(struct xe_guc *guc, void *dst)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+
+ XE_WARN_ON(!guc->hwconfig.bo);
+
+ xe_map_memcpy_from(xe, dst, &guc->hwconfig.bo->vmap, 0,
+ guc->hwconfig.size);
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_hwconfig.h b/drivers/gpu/drm/xe/xe_guc_hwconfig.h
new file mode 100644
index 000000000000..b5794d641900
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_hwconfig.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_HWCONFIG_H_
+#define _XE_GUC_HWCONFIG_H_
+
+#include <linux/types.h>
+
+struct xe_guc;
+
+int xe_guc_hwconfig_init(struct xe_guc *guc);
+u32 xe_guc_hwconfig_size(struct xe_guc *guc);
+void xe_guc_hwconfig_copy(struct xe_guc *guc, void *dst);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_log.c b/drivers/gpu/drm/xe/xe_guc_log.c
new file mode 100644
index 000000000000..bcd2f4d34081
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_log.c
@@ -0,0 +1,97 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_log.h"
+
+#include <drm/drm_managed.h>
+
+#include "xe_bo.h"
+#include "xe_gt.h"
+#include "xe_map.h"
+#include "xe_module.h"
+
+static struct xe_gt *
+log_to_gt(struct xe_guc_log *log)
+{
+ return container_of(log, struct xe_gt, uc.guc.log);
+}
+
+static struct xe_device *
+log_to_xe(struct xe_guc_log *log)
+{
+ return gt_to_xe(log_to_gt(log));
+}
+
+static size_t guc_log_size(void)
+{
+ /*
+ * GuC Log buffer Layout
+ *
+ * +===============================+ 00B
+ * | Crash dump state header |
+ * +-------------------------------+ 32B
+ * | Debug state header |
+ * +-------------------------------+ 64B
+ * | Capture state header |
+ * +-------------------------------+ 96B
+ * | |
+ * +===============================+ PAGE_SIZE (4KB)
+ * | Crash Dump logs |
+ * +===============================+ + CRASH_SIZE
+ * | Debug logs |
+ * +===============================+ + DEBUG_SIZE
+ * | Capture logs |
+ * +===============================+ + CAPTURE_SIZE
+ */
+ return PAGE_SIZE + CRASH_BUFFER_SIZE + DEBUG_BUFFER_SIZE +
+ CAPTURE_BUFFER_SIZE;
+}
+
+void xe_guc_log_print(struct xe_guc_log *log, struct drm_printer *p)
+{
+ struct xe_device *xe = log_to_xe(log);
+ size_t size;
+ int i, j;
+
+ xe_assert(xe, log->bo);
+
+ size = log->bo->size;
+
+#define DW_PER_READ 128
+ xe_assert(xe, !(size % (DW_PER_READ * sizeof(u32))));
+ for (i = 0; i < size / sizeof(u32); i += DW_PER_READ) {
+ u32 read[DW_PER_READ];
+
+ xe_map_memcpy_from(xe, read, &log->bo->vmap, i * sizeof(u32),
+ DW_PER_READ * sizeof(u32));
+#define DW_PER_PRINT 4
+ for (j = 0; j < DW_PER_READ / DW_PER_PRINT; ++j) {
+ u32 *print = read + j * DW_PER_PRINT;
+
+ drm_printf(p, "0x%08x 0x%08x 0x%08x 0x%08x\n",
+ *(print + 0), *(print + 1),
+ *(print + 2), *(print + 3));
+ }
+ }
+}
+
+int xe_guc_log_init(struct xe_guc_log *log)
+{
+ struct xe_device *xe = log_to_xe(log);
+ struct xe_tile *tile = gt_to_tile(log_to_gt(log));
+ struct xe_bo *bo;
+
+ bo = xe_managed_bo_create_pin_map(xe, tile, guc_log_size(),
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ xe_map_memset(xe, &bo->vmap, 0, 0, guc_log_size());
+ log->bo = bo;
+ log->level = xe_modparam.guc_log_level;
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_log.h b/drivers/gpu/drm/xe/xe_guc_log.h
new file mode 100644
index 000000000000..2d25ab28b4b3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_log.h
@@ -0,0 +1,48 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_LOG_H_
+#define _XE_GUC_LOG_H_
+
+#include "xe_guc_log_types.h"
+
+struct drm_printer;
+
+#if IS_ENABLED(CONFIG_DRM_XE_LARGE_GUC_BUFFER)
+#define CRASH_BUFFER_SIZE SZ_1M
+#define DEBUG_BUFFER_SIZE SZ_8M
+#define CAPTURE_BUFFER_SIZE SZ_2M
+#else
+#define CRASH_BUFFER_SIZE SZ_8K
+#define DEBUG_BUFFER_SIZE SZ_64K
+#define CAPTURE_BUFFER_SIZE SZ_16K
+#endif
+/*
+ * While we're using plain log level in i915, GuC controls are much more...
+ * "elaborate"? We have a couple of bits for verbosity, separate bit for actual
+ * log enabling, and separate bit for default logging - which "conveniently"
+ * ignores the enable bit.
+ */
+#define GUC_LOG_LEVEL_DISABLED 0
+#define GUC_LOG_LEVEL_NON_VERBOSE 1
+#define GUC_LOG_LEVEL_IS_ENABLED(x) ((x) > GUC_LOG_LEVEL_DISABLED)
+#define GUC_LOG_LEVEL_IS_VERBOSE(x) ((x) > GUC_LOG_LEVEL_NON_VERBOSE)
+#define GUC_LOG_LEVEL_TO_VERBOSITY(x) ({ \
+ typeof(x) _x = (x); \
+ GUC_LOG_LEVEL_IS_VERBOSE(_x) ? _x - 2 : 0; \
+})
+#define GUC_VERBOSITY_TO_LOG_LEVEL(x) ((x) + 2)
+#define GUC_LOG_LEVEL_MAX GUC_VERBOSITY_TO_LOG_LEVEL(GUC_LOG_VERBOSITY_MAX)
+
+int xe_guc_log_init(struct xe_guc_log *log);
+void xe_guc_log_print(struct xe_guc_log *log, struct drm_printer *p);
+
+static inline u32
+xe_guc_log_get_level(struct xe_guc_log *log)
+{
+ return log->level;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_log_types.h b/drivers/gpu/drm/xe/xe_guc_log_types.h
new file mode 100644
index 000000000000..125080d138a7
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_log_types.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_LOG_TYPES_H_
+#define _XE_GUC_LOG_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_bo;
+
+/**
+ * struct xe_guc_log - GuC log
+ */
+struct xe_guc_log {
+ /** @level: GuC log level */
+ u32 level;
+ /** @bo: XE BO for GuC log */
+ struct xe_bo *bo;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_pc.c b/drivers/gpu/drm/xe/xe_guc_pc.c
new file mode 100644
index 000000000000..f71085228cb3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_pc.c
@@ -0,0 +1,1000 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_pc.h"
+
+#include <linux/delay.h>
+
+#include <drm/drm_managed.h>
+
+#include "abi/guc_actions_abi.h"
+#include "abi/guc_actions_slpc_abi.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_idle.h"
+#include "xe_gt_sysfs.h"
+#include "xe_gt_types.h"
+#include "xe_guc_ct.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_pcode.h"
+
+#define MCHBAR_MIRROR_BASE_SNB 0x140000
+
+#define RP_STATE_CAP XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5998)
+#define RP0_MASK REG_GENMASK(7, 0)
+#define RP1_MASK REG_GENMASK(15, 8)
+#define RPN_MASK REG_GENMASK(23, 16)
+
+#define FREQ_INFO_REC XE_REG(MCHBAR_MIRROR_BASE_SNB + 0x5ef0)
+#define RPE_MASK REG_GENMASK(15, 8)
+
+#define GT_PERF_STATUS XE_REG(0x1381b4)
+#define CAGF_MASK REG_GENMASK(19, 11)
+
+#define GT_FREQUENCY_MULTIPLIER 50
+#define GT_FREQUENCY_SCALER 3
+
+/**
+ * DOC: GuC Power Conservation (PC)
+ *
+ * GuC Power Conservation (PC) supports multiple features for the most
+ * efficient and performing use of the GT when GuC submission is enabled,
+ * including frequency management, Render-C states management, and various
+ * algorithms for power balancing.
+ *
+ * Single Loop Power Conservation (SLPC) is the name given to the suite of
+ * connected power conservation features in the GuC firmware. The firmware
+ * exposes a programming interface to the host for the control of SLPC.
+ *
+ * Frequency management:
+ * =====================
+ *
+ * Xe driver enables SLPC with all of its defaults features and frequency
+ * selection, which varies per platform.
+ *
+ * Render-C States:
+ * ================
+ *
+ * Render-C states is also a GuC PC feature that is now enabled in Xe for
+ * all platforms.
+ *
+ */
+
+static struct xe_guc *
+pc_to_guc(struct xe_guc_pc *pc)
+{
+ return container_of(pc, struct xe_guc, pc);
+}
+
+static struct xe_device *
+pc_to_xe(struct xe_guc_pc *pc)
+{
+ struct xe_guc *guc = pc_to_guc(pc);
+ struct xe_gt *gt = container_of(guc, struct xe_gt, uc.guc);
+
+ return gt_to_xe(gt);
+}
+
+static struct xe_gt *
+pc_to_gt(struct xe_guc_pc *pc)
+{
+ return container_of(pc, struct xe_gt, uc.guc.pc);
+}
+
+static struct iosys_map *
+pc_to_maps(struct xe_guc_pc *pc)
+{
+ return &pc->bo->vmap;
+}
+
+#define slpc_shared_data_read(pc_, field_) \
+ xe_map_rd_field(pc_to_xe(pc_), pc_to_maps(pc_), 0, \
+ struct slpc_shared_data, field_)
+
+#define slpc_shared_data_write(pc_, field_, val_) \
+ xe_map_wr_field(pc_to_xe(pc_), pc_to_maps(pc_), 0, \
+ struct slpc_shared_data, field_, val_)
+
+#define SLPC_EVENT(id, count) \
+ (FIELD_PREP(HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ID, id) | \
+ FIELD_PREP(HOST2GUC_PC_SLPC_REQUEST_MSG_1_EVENT_ARGC, count))
+
+static int wait_for_pc_state(struct xe_guc_pc *pc,
+ enum slpc_global_state state)
+{
+ int timeout_us = 5000; /* rought 5ms, but no need for precision */
+ int slept, wait = 10;
+
+ xe_device_assert_mem_access(pc_to_xe(pc));
+
+ for (slept = 0; slept < timeout_us;) {
+ if (slpc_shared_data_read(pc, header.global_state) == state)
+ return 0;
+
+ usleep_range(wait, wait << 1);
+ slept += wait;
+ wait <<= 1;
+ if (slept + wait > timeout_us)
+ wait = timeout_us - slept;
+ }
+
+ return -ETIMEDOUT;
+}
+
+static int pc_action_reset(struct xe_guc_pc *pc)
+{
+ struct xe_guc_ct *ct = &pc_to_guc(pc)->ct;
+ int ret;
+ u32 action[] = {
+ GUC_ACTION_HOST2GUC_PC_SLPC_REQUEST,
+ SLPC_EVENT(SLPC_EVENT_RESET, 2),
+ xe_bo_ggtt_addr(pc->bo),
+ 0,
+ };
+
+ ret = xe_guc_ct_send(ct, action, ARRAY_SIZE(action), 0, 0);
+ if (ret)
+ drm_err(&pc_to_xe(pc)->drm, "GuC PC reset: %pe", ERR_PTR(ret));
+
+ return ret;
+}
+
+static int pc_action_shutdown(struct xe_guc_pc *pc)
+{
+ struct xe_guc_ct *ct = &pc_to_guc(pc)->ct;
+ int ret;
+ u32 action[] = {
+ GUC_ACTION_HOST2GUC_PC_SLPC_REQUEST,
+ SLPC_EVENT(SLPC_EVENT_SHUTDOWN, 2),
+ xe_bo_ggtt_addr(pc->bo),
+ 0,
+ };
+
+ ret = xe_guc_ct_send(ct, action, ARRAY_SIZE(action), 0, 0);
+ if (ret)
+ drm_err(&pc_to_xe(pc)->drm, "GuC PC shutdown %pe",
+ ERR_PTR(ret));
+
+ return ret;
+}
+
+static int pc_action_query_task_state(struct xe_guc_pc *pc)
+{
+ struct xe_guc_ct *ct = &pc_to_guc(pc)->ct;
+ int ret;
+ u32 action[] = {
+ GUC_ACTION_HOST2GUC_PC_SLPC_REQUEST,
+ SLPC_EVENT(SLPC_EVENT_QUERY_TASK_STATE, 2),
+ xe_bo_ggtt_addr(pc->bo),
+ 0,
+ };
+
+ if (wait_for_pc_state(pc, SLPC_GLOBAL_STATE_RUNNING))
+ return -EAGAIN;
+
+ /* Blocking here to ensure the results are ready before reading them */
+ ret = xe_guc_ct_send_block(ct, action, ARRAY_SIZE(action));
+ if (ret)
+ drm_err(&pc_to_xe(pc)->drm,
+ "GuC PC query task state failed: %pe", ERR_PTR(ret));
+
+ return ret;
+}
+
+static int pc_action_set_param(struct xe_guc_pc *pc, u8 id, u32 value)
+{
+ struct xe_guc_ct *ct = &pc_to_guc(pc)->ct;
+ int ret;
+ u32 action[] = {
+ GUC_ACTION_HOST2GUC_PC_SLPC_REQUEST,
+ SLPC_EVENT(SLPC_EVENT_PARAMETER_SET, 2),
+ id,
+ value,
+ };
+
+ if (wait_for_pc_state(pc, SLPC_GLOBAL_STATE_RUNNING))
+ return -EAGAIN;
+
+ ret = xe_guc_ct_send(ct, action, ARRAY_SIZE(action), 0, 0);
+ if (ret)
+ drm_err(&pc_to_xe(pc)->drm, "GuC PC set param failed: %pe",
+ ERR_PTR(ret));
+
+ return ret;
+}
+
+static int pc_action_setup_gucrc(struct xe_guc_pc *pc, u32 mode)
+{
+ struct xe_guc_ct *ct = &pc_to_guc(pc)->ct;
+ u32 action[] = {
+ XE_GUC_ACTION_SETUP_PC_GUCRC,
+ mode,
+ };
+ int ret;
+
+ ret = xe_guc_ct_send(ct, action, ARRAY_SIZE(action), 0, 0);
+ if (ret)
+ drm_err(&pc_to_xe(pc)->drm, "GuC RC enable failed: %pe",
+ ERR_PTR(ret));
+ return ret;
+}
+
+static u32 decode_freq(u32 raw)
+{
+ return DIV_ROUND_CLOSEST(raw * GT_FREQUENCY_MULTIPLIER,
+ GT_FREQUENCY_SCALER);
+}
+
+static u32 encode_freq(u32 freq)
+{
+ return DIV_ROUND_CLOSEST(freq * GT_FREQUENCY_SCALER,
+ GT_FREQUENCY_MULTIPLIER);
+}
+
+static u32 pc_get_min_freq(struct xe_guc_pc *pc)
+{
+ u32 freq;
+
+ freq = FIELD_GET(SLPC_MIN_UNSLICE_FREQ_MASK,
+ slpc_shared_data_read(pc, task_state_data.freq));
+
+ return decode_freq(freq);
+}
+
+static void pc_set_manual_rp_ctrl(struct xe_guc_pc *pc, bool enable)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 state = enable ? RPSWCTL_ENABLE : RPSWCTL_DISABLE;
+
+ /* Allow/Disallow punit to process software freq requests */
+ xe_mmio_write32(gt, RP_CONTROL, state);
+}
+
+static void pc_set_cur_freq(struct xe_guc_pc *pc, u32 freq)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 rpnswreq;
+
+ pc_set_manual_rp_ctrl(pc, true);
+
+ /* Req freq is in units of 16.66 Mhz */
+ rpnswreq = REG_FIELD_PREP(REQ_RATIO_MASK, encode_freq(freq));
+ xe_mmio_write32(gt, RPNSWREQ, rpnswreq);
+
+ /* Sleep for a small time to allow pcode to respond */
+ usleep_range(100, 300);
+
+ pc_set_manual_rp_ctrl(pc, false);
+}
+
+static int pc_set_min_freq(struct xe_guc_pc *pc, u32 freq)
+{
+ /*
+ * Let's only check for the rpn-rp0 range. If max < min,
+ * min becomes a fixed request.
+ */
+ if (freq < pc->rpn_freq || freq > pc->rp0_freq)
+ return -EINVAL;
+
+ /*
+ * GuC policy is to elevate minimum frequency to the efficient levels
+ * Our goal is to have the admin choices respected.
+ */
+ pc_action_set_param(pc, SLPC_PARAM_IGNORE_EFFICIENT_FREQUENCY,
+ freq < pc->rpe_freq);
+
+ return pc_action_set_param(pc,
+ SLPC_PARAM_GLOBAL_MIN_GT_UNSLICE_FREQ_MHZ,
+ freq);
+}
+
+static int pc_get_max_freq(struct xe_guc_pc *pc)
+{
+ u32 freq;
+
+ freq = FIELD_GET(SLPC_MAX_UNSLICE_FREQ_MASK,
+ slpc_shared_data_read(pc, task_state_data.freq));
+
+ return decode_freq(freq);
+}
+
+static int pc_set_max_freq(struct xe_guc_pc *pc, u32 freq)
+{
+ /*
+ * Let's only check for the rpn-rp0 range. If max < min,
+ * min becomes a fixed request.
+ * Also, overclocking is not supported.
+ */
+ if (freq < pc->rpn_freq || freq > pc->rp0_freq)
+ return -EINVAL;
+
+ return pc_action_set_param(pc,
+ SLPC_PARAM_GLOBAL_MAX_GT_UNSLICE_FREQ_MHZ,
+ freq);
+}
+
+static void mtl_update_rpe_value(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 reg;
+
+ if (xe_gt_is_media_type(gt))
+ reg = xe_mmio_read32(gt, MTL_MPE_FREQUENCY);
+ else
+ reg = xe_mmio_read32(gt, MTL_GT_RPE_FREQUENCY);
+
+ pc->rpe_freq = decode_freq(REG_FIELD_GET(MTL_RPE_MASK, reg));
+}
+
+static void tgl_update_rpe_value(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 reg;
+
+ /*
+ * For PVC we still need to use fused RP1 as the approximation for RPe
+ * For other platforms than PVC we get the resolved RPe directly from
+ * PCODE at a different register
+ */
+ if (xe->info.platform == XE_PVC)
+ reg = xe_mmio_read32(gt, PVC_RP_STATE_CAP);
+ else
+ reg = xe_mmio_read32(gt, FREQ_INFO_REC);
+
+ pc->rpe_freq = REG_FIELD_GET(RPE_MASK, reg) * GT_FREQUENCY_MULTIPLIER;
+}
+
+static void pc_update_rp_values(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (GRAPHICS_VERx100(xe) >= 1270)
+ mtl_update_rpe_value(pc);
+ else
+ tgl_update_rpe_value(pc);
+
+ /*
+ * RPe is decided at runtime by PCODE. In the rare case where that's
+ * smaller than the fused min, we will trust the PCODE and use that
+ * as our minimum one.
+ */
+ pc->rpn_freq = min(pc->rpn_freq, pc->rpe_freq);
+}
+
+/**
+ * xe_guc_pc_get_act_freq - Get Actual running frequency
+ * @pc: The GuC PC
+ *
+ * Returns: The Actual running frequency. Which might be 0 if GT is in Render-C sleep state (RC6).
+ */
+u32 xe_guc_pc_get_act_freq(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 freq;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+
+ /* When in RC6, actual frequency reported will be 0. */
+ if (GRAPHICS_VERx100(xe) >= 1270) {
+ freq = xe_mmio_read32(gt, MTL_MIRROR_TARGET_WP1);
+ freq = REG_FIELD_GET(MTL_CAGF_MASK, freq);
+ } else {
+ freq = xe_mmio_read32(gt, GT_PERF_STATUS);
+ freq = REG_FIELD_GET(CAGF_MASK, freq);
+ }
+
+ freq = decode_freq(freq);
+
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return freq;
+}
+
+/**
+ * xe_guc_pc_get_cur_freq - Get Current requested frequency
+ * @pc: The GuC PC
+ * @freq: A pointer to a u32 where the freq value will be returned
+ *
+ * Returns: 0 on success,
+ * -EAGAIN if GuC PC not ready (likely in middle of a reset).
+ */
+int xe_guc_pc_get_cur_freq(struct xe_guc_pc *pc, u32 *freq)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ int ret;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ /*
+ * GuC SLPC plays with cur freq request when GuCRC is enabled
+ * Block RC6 for a more reliable read.
+ */
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (ret)
+ goto out;
+
+ *freq = xe_mmio_read32(gt, RPNSWREQ);
+
+ *freq = REG_FIELD_GET(REQ_RATIO_MASK, *freq);
+ *freq = decode_freq(*freq);
+
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+out:
+ xe_device_mem_access_put(gt_to_xe(gt));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_get_rp0_freq - Get the RP0 freq
+ * @pc: The GuC PC
+ *
+ * Returns: RP0 freq.
+ */
+u32 xe_guc_pc_get_rp0_freq(struct xe_guc_pc *pc)
+{
+ return pc->rp0_freq;
+}
+
+/**
+ * xe_guc_pc_get_rpe_freq - Get the RPe freq
+ * @pc: The GuC PC
+ *
+ * Returns: RPe freq.
+ */
+u32 xe_guc_pc_get_rpe_freq(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+
+ xe_device_mem_access_get(xe);
+ pc_update_rp_values(pc);
+ xe_device_mem_access_put(xe);
+
+ return pc->rpe_freq;
+}
+
+/**
+ * xe_guc_pc_get_rpn_freq - Get the RPn freq
+ * @pc: The GuC PC
+ *
+ * Returns: RPn freq.
+ */
+u32 xe_guc_pc_get_rpn_freq(struct xe_guc_pc *pc)
+{
+ return pc->rpn_freq;
+}
+
+/**
+ * xe_guc_pc_get_min_freq - Get the min operational frequency
+ * @pc: The GuC PC
+ * @freq: A pointer to a u32 where the freq value will be returned
+ *
+ * Returns: 0 on success,
+ * -EAGAIN if GuC PC not ready (likely in middle of a reset).
+ */
+int xe_guc_pc_get_min_freq(struct xe_guc_pc *pc, u32 *freq)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ int ret;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+ mutex_lock(&pc->freq_lock);
+ if (!pc->freq_ready) {
+ /* Might be in the middle of a gt reset */
+ ret = -EAGAIN;
+ goto out;
+ }
+
+ /*
+ * GuC SLPC plays with min freq request when GuCRC is enabled
+ * Block RC6 for a more reliable read.
+ */
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (ret)
+ goto out;
+
+ ret = pc_action_query_task_state(pc);
+ if (ret)
+ goto fw;
+
+ *freq = pc_get_min_freq(pc);
+
+fw:
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+out:
+ mutex_unlock(&pc->freq_lock);
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_set_min_freq - Set the minimal operational frequency
+ * @pc: The GuC PC
+ * @freq: The selected minimal frequency
+ *
+ * Returns: 0 on success,
+ * -EAGAIN if GuC PC not ready (likely in middle of a reset),
+ * -EINVAL if value out of bounds.
+ */
+int xe_guc_pc_set_min_freq(struct xe_guc_pc *pc, u32 freq)
+{
+ int ret;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+ mutex_lock(&pc->freq_lock);
+ if (!pc->freq_ready) {
+ /* Might be in the middle of a gt reset */
+ ret = -EAGAIN;
+ goto out;
+ }
+
+ ret = pc_set_min_freq(pc, freq);
+ if (ret)
+ goto out;
+
+ pc->user_requested_min = freq;
+
+out:
+ mutex_unlock(&pc->freq_lock);
+ xe_device_mem_access_put(pc_to_xe(pc));
+
+ return ret;
+}
+
+/**
+ * xe_guc_pc_get_max_freq - Get Maximum operational frequency
+ * @pc: The GuC PC
+ * @freq: A pointer to a u32 where the freq value will be returned
+ *
+ * Returns: 0 on success,
+ * -EAGAIN if GuC PC not ready (likely in middle of a reset).
+ */
+int xe_guc_pc_get_max_freq(struct xe_guc_pc *pc, u32 *freq)
+{
+ int ret;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+ mutex_lock(&pc->freq_lock);
+ if (!pc->freq_ready) {
+ /* Might be in the middle of a gt reset */
+ ret = -EAGAIN;
+ goto out;
+ }
+
+ ret = pc_action_query_task_state(pc);
+ if (ret)
+ goto out;
+
+ *freq = pc_get_max_freq(pc);
+
+out:
+ mutex_unlock(&pc->freq_lock);
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_set_max_freq - Set the maximum operational frequency
+ * @pc: The GuC PC
+ * @freq: The selected maximum frequency value
+ *
+ * Returns: 0 on success,
+ * -EAGAIN if GuC PC not ready (likely in middle of a reset),
+ * -EINVAL if value out of bounds.
+ */
+int xe_guc_pc_set_max_freq(struct xe_guc_pc *pc, u32 freq)
+{
+ int ret;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+ mutex_lock(&pc->freq_lock);
+ if (!pc->freq_ready) {
+ /* Might be in the middle of a gt reset */
+ ret = -EAGAIN;
+ goto out;
+ }
+
+ ret = pc_set_max_freq(pc, freq);
+ if (ret)
+ goto out;
+
+ pc->user_requested_max = freq;
+
+out:
+ mutex_unlock(&pc->freq_lock);
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_c_status - get the current GT C state
+ * @pc: XE_GuC_PC instance
+ */
+enum xe_gt_idle_state xe_guc_pc_c_status(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 reg, gt_c_state;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1270) {
+ reg = xe_mmio_read32(gt, MTL_MIRROR_TARGET_WP1);
+ gt_c_state = REG_FIELD_GET(MTL_CC_MASK, reg);
+ } else {
+ reg = xe_mmio_read32(gt, GT_CORE_STATUS);
+ gt_c_state = REG_FIELD_GET(RCN_MASK, reg);
+ }
+
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ switch (gt_c_state) {
+ case GT_C6:
+ return GT_IDLE_C6;
+ case GT_C0:
+ return GT_IDLE_C0;
+ default:
+ return GT_IDLE_UNKNOWN;
+ }
+}
+
+/**
+ * xe_guc_pc_rc6_residency - rc6 residency counter
+ * @pc: Xe_GuC_PC instance
+ */
+u64 xe_guc_pc_rc6_residency(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 reg;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ reg = xe_mmio_read32(gt, GT_GFX_RC6);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return reg;
+}
+
+/**
+ * xe_guc_pc_mc6_residency - mc6 residency counter
+ * @pc: Xe_GuC_PC instance
+ */
+u64 xe_guc_pc_mc6_residency(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u64 reg;
+
+ xe_device_mem_access_get(gt_to_xe(gt));
+ reg = xe_mmio_read32(gt, MTL_MEDIA_MC6);
+ xe_device_mem_access_put(gt_to_xe(gt));
+
+ return reg;
+}
+
+static void mtl_init_fused_rp_values(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 reg;
+
+ xe_device_assert_mem_access(pc_to_xe(pc));
+
+ if (xe_gt_is_media_type(gt))
+ reg = xe_mmio_read32(gt, MTL_MEDIAP_STATE_CAP);
+ else
+ reg = xe_mmio_read32(gt, MTL_RP_STATE_CAP);
+
+ pc->rp0_freq = decode_freq(REG_FIELD_GET(MTL_RP0_CAP_MASK, reg));
+
+ pc->rpn_freq = decode_freq(REG_FIELD_GET(MTL_RPN_CAP_MASK, reg));
+}
+
+static void tgl_init_fused_rp_values(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 reg;
+
+ xe_device_assert_mem_access(pc_to_xe(pc));
+
+ if (xe->info.platform == XE_PVC)
+ reg = xe_mmio_read32(gt, PVC_RP_STATE_CAP);
+ else
+ reg = xe_mmio_read32(gt, RP_STATE_CAP);
+ pc->rp0_freq = REG_FIELD_GET(RP0_MASK, reg) * GT_FREQUENCY_MULTIPLIER;
+ pc->rpn_freq = REG_FIELD_GET(RPN_MASK, reg) * GT_FREQUENCY_MULTIPLIER;
+}
+
+static void pc_init_fused_rp_values(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (GRAPHICS_VERx100(xe) >= 1270)
+ mtl_init_fused_rp_values(pc);
+ else
+ tgl_init_fused_rp_values(pc);
+}
+
+/**
+ * xe_guc_pc_init_early - Initialize RPx values and request a higher GT
+ * frequency to allow faster GuC load times
+ * @pc: Xe_GuC_PC instance
+ */
+void xe_guc_pc_init_early(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+ pc_init_fused_rp_values(pc);
+ pc_set_cur_freq(pc, pc->rp0_freq);
+}
+
+static int pc_adjust_freq_bounds(struct xe_guc_pc *pc)
+{
+ int ret;
+
+ lockdep_assert_held(&pc->freq_lock);
+
+ ret = pc_action_query_task_state(pc);
+ if (ret)
+ return ret;
+
+ /*
+ * GuC defaults to some RPmax that is not actually achievable without
+ * overclocking. Let's adjust it to the Hardware RP0, which is the
+ * regular maximum
+ */
+ if (pc_get_max_freq(pc) > pc->rp0_freq)
+ pc_set_max_freq(pc, pc->rp0_freq);
+
+ /*
+ * Same thing happens for Server platforms where min is listed as
+ * RPMax
+ */
+ if (pc_get_min_freq(pc) > pc->rp0_freq)
+ pc_set_min_freq(pc, pc->rp0_freq);
+
+ return 0;
+}
+
+static int pc_adjust_requested_freq(struct xe_guc_pc *pc)
+{
+ int ret = 0;
+
+ lockdep_assert_held(&pc->freq_lock);
+
+ if (pc->user_requested_min != 0) {
+ ret = pc_set_min_freq(pc, pc->user_requested_min);
+ if (ret)
+ return ret;
+ }
+
+ if (pc->user_requested_max != 0) {
+ ret = pc_set_max_freq(pc, pc->user_requested_max);
+ if (ret)
+ return ret;
+ }
+
+ return ret;
+}
+
+/**
+ * xe_guc_pc_gucrc_disable - Disable GuC RC
+ * @pc: Xe_GuC_PC instance
+ *
+ * Disables GuC RC by taking control of RC6 back from GuC.
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_guc_pc_gucrc_disable(struct xe_guc_pc *pc)
+{
+ struct xe_device *xe = pc_to_xe(pc);
+ struct xe_gt *gt = pc_to_gt(pc);
+ int ret = 0;
+
+ if (xe->info.skip_guc_pc)
+ return 0;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+
+ ret = pc_action_setup_gucrc(pc, XE_GUCRC_HOST_CONTROL);
+ if (ret)
+ goto out;
+
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (ret)
+ goto out;
+
+ xe_gt_idle_disable_c6(gt);
+
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+
+out:
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+static void pc_init_pcode_freq(struct xe_guc_pc *pc)
+{
+ u32 min = DIV_ROUND_CLOSEST(pc->rpn_freq, GT_FREQUENCY_MULTIPLIER);
+ u32 max = DIV_ROUND_CLOSEST(pc->rp0_freq, GT_FREQUENCY_MULTIPLIER);
+
+ XE_WARN_ON(xe_pcode_init_min_freq_table(pc_to_gt(pc), min, max));
+}
+
+static int pc_init_freqs(struct xe_guc_pc *pc)
+{
+ int ret;
+
+ mutex_lock(&pc->freq_lock);
+
+ ret = pc_adjust_freq_bounds(pc);
+ if (ret)
+ goto out;
+
+ ret = pc_adjust_requested_freq(pc);
+ if (ret)
+ goto out;
+
+ pc_update_rp_values(pc);
+
+ pc_init_pcode_freq(pc);
+
+ /*
+ * The frequencies are really ready for use only after the user
+ * requested ones got restored.
+ */
+ pc->freq_ready = true;
+
+out:
+ mutex_unlock(&pc->freq_lock);
+ return ret;
+}
+
+/**
+ * xe_guc_pc_start - Start GuC's Power Conservation component
+ * @pc: Xe_GuC_PC instance
+ */
+int xe_guc_pc_start(struct xe_guc_pc *pc)
+{
+ struct xe_device *xe = pc_to_xe(pc);
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 size = PAGE_ALIGN(sizeof(struct slpc_shared_data));
+ int ret;
+
+ xe_gt_assert(gt, xe_device_uc_enabled(xe));
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ if (ret)
+ goto out_fail_force_wake;
+
+ if (xe->info.skip_guc_pc) {
+ if (xe->info.platform != XE_PVC)
+ xe_gt_idle_enable_c6(gt);
+
+ /* Request max possible since dynamic freq mgmt is not enabled */
+ pc_set_cur_freq(pc, UINT_MAX);
+
+ ret = 0;
+ goto out;
+ }
+
+ memset(pc->bo->vmap.vaddr, 0, size);
+ slpc_shared_data_write(pc, header.size, size);
+
+ ret = pc_action_reset(pc);
+ if (ret)
+ goto out;
+
+ if (wait_for_pc_state(pc, SLPC_GLOBAL_STATE_RUNNING)) {
+ drm_err(&pc_to_xe(pc)->drm, "GuC PC Start failed\n");
+ ret = -EIO;
+ goto out;
+ }
+
+ ret = pc_init_freqs(pc);
+ if (ret)
+ goto out;
+
+ if (xe->info.platform == XE_PVC) {
+ xe_guc_pc_gucrc_disable(pc);
+ ret = 0;
+ goto out;
+ }
+
+ ret = pc_action_setup_gucrc(pc, XE_GUCRC_FIRMWARE_CONTROL);
+
+out:
+ XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
+out_fail_force_wake:
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_stop - Stop GuC's Power Conservation component
+ * @pc: Xe_GuC_PC instance
+ */
+int xe_guc_pc_stop(struct xe_guc_pc *pc)
+{
+ struct xe_device *xe = pc_to_xe(pc);
+ int ret;
+
+ xe_device_mem_access_get(pc_to_xe(pc));
+
+ if (xe->info.skip_guc_pc) {
+ xe_gt_idle_disable_c6(pc_to_gt(pc));
+ ret = 0;
+ goto out;
+ }
+
+ mutex_lock(&pc->freq_lock);
+ pc->freq_ready = false;
+ mutex_unlock(&pc->freq_lock);
+
+ ret = pc_action_shutdown(pc);
+ if (ret)
+ goto out;
+
+ if (wait_for_pc_state(pc, SLPC_GLOBAL_STATE_NOT_RUNNING)) {
+ drm_err(&pc_to_xe(pc)->drm, "GuC PC Shutdown failed\n");
+ ret = -EIO;
+ }
+
+out:
+ xe_device_mem_access_put(pc_to_xe(pc));
+ return ret;
+}
+
+/**
+ * xe_guc_pc_fini - Finalize GuC's Power Conservation component
+ * @pc: Xe_GuC_PC instance
+ */
+void xe_guc_pc_fini(struct xe_guc_pc *pc)
+{
+ struct xe_device *xe = pc_to_xe(pc);
+
+ if (xe->info.skip_guc_pc) {
+ xe_gt_idle_disable_c6(pc_to_gt(pc));
+ return;
+ }
+
+ XE_WARN_ON(xe_guc_pc_gucrc_disable(pc));
+ XE_WARN_ON(xe_guc_pc_stop(pc));
+ mutex_destroy(&pc->freq_lock);
+}
+
+/**
+ * xe_guc_pc_init - Initialize GuC's Power Conservation component
+ * @pc: Xe_GuC_PC instance
+ */
+int xe_guc_pc_init(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_bo *bo;
+ u32 size = PAGE_ALIGN(sizeof(struct slpc_shared_data));
+
+ if (xe->info.skip_guc_pc)
+ return 0;
+
+ mutex_init(&pc->freq_lock);
+
+ bo = xe_managed_bo_create_pin_map(xe, tile, size,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ pc->bo = bo;
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_pc.h b/drivers/gpu/drm/xe/xe_guc_pc.h
new file mode 100644
index 000000000000..cecad8e9300b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_pc.h
@@ -0,0 +1,31 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_PC_H_
+#define _XE_GUC_PC_H_
+
+#include "xe_guc_pc_types.h"
+
+int xe_guc_pc_init(struct xe_guc_pc *pc);
+void xe_guc_pc_fini(struct xe_guc_pc *pc);
+int xe_guc_pc_start(struct xe_guc_pc *pc);
+int xe_guc_pc_stop(struct xe_guc_pc *pc);
+int xe_guc_pc_gucrc_disable(struct xe_guc_pc *pc);
+
+u32 xe_guc_pc_get_act_freq(struct xe_guc_pc *pc);
+int xe_guc_pc_get_cur_freq(struct xe_guc_pc *pc, u32 *freq);
+u32 xe_guc_pc_get_rp0_freq(struct xe_guc_pc *pc);
+u32 xe_guc_pc_get_rpe_freq(struct xe_guc_pc *pc);
+u32 xe_guc_pc_get_rpn_freq(struct xe_guc_pc *pc);
+int xe_guc_pc_get_min_freq(struct xe_guc_pc *pc, u32 *freq);
+int xe_guc_pc_set_min_freq(struct xe_guc_pc *pc, u32 freq);
+int xe_guc_pc_get_max_freq(struct xe_guc_pc *pc, u32 *freq);
+int xe_guc_pc_set_max_freq(struct xe_guc_pc *pc, u32 freq);
+
+enum xe_gt_idle_state xe_guc_pc_c_status(struct xe_guc_pc *pc);
+u64 xe_guc_pc_rc6_residency(struct xe_guc_pc *pc);
+u64 xe_guc_pc_mc6_residency(struct xe_guc_pc *pc);
+void xe_guc_pc_init_early(struct xe_guc_pc *pc);
+#endif /* _XE_GUC_PC_H_ */
diff --git a/drivers/gpu/drm/xe/xe_guc_pc_types.h b/drivers/gpu/drm/xe/xe_guc_pc_types.h
new file mode 100644
index 000000000000..2afd0dbc3542
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_pc_types.h
@@ -0,0 +1,34 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_PC_TYPES_H_
+#define _XE_GUC_PC_TYPES_H_
+
+#include <linux/mutex.h>
+#include <linux/types.h>
+
+/**
+ * struct xe_guc_pc - GuC Power Conservation (PC)
+ */
+struct xe_guc_pc {
+ /** @bo: GGTT buffer object that is shared with GuC PC */
+ struct xe_bo *bo;
+ /** @rp0_freq: HW RP0 frequency - The Maximum one */
+ u32 rp0_freq;
+ /** @rpe_freq: HW RPe frequency - The Efficient one */
+ u32 rpe_freq;
+ /** @rpn_freq: HW RPN frequency - The Minimum one */
+ u32 rpn_freq;
+ /** @user_requested_min: Stash the minimum requested freq by user */
+ u32 user_requested_min;
+ /** @user_requested_max: Stash the maximum requested freq by user */
+ u32 user_requested_max;
+ /** @freq_lock: Let's protect the frequencies */
+ struct mutex freq_lock;
+ /** @freq_ready: Only handle freq changes, if they are really ready */
+ bool freq_ready;
+};
+
+#endif /* _XE_GUC_PC_TYPES_H_ */
diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
new file mode 100644
index 000000000000..21ac68e3246f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -0,0 +1,1990 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_guc_submit.h"
+
+#include <linux/bitfield.h>
+#include <linux/bitmap.h>
+#include <linux/circ_buf.h>
+#include <linux/delay.h>
+#include <linux/dma-fence-array.h>
+
+#include <drm/drm_managed.h>
+
+#include "abi/guc_actions_abi.h"
+#include "abi/guc_klvs_abi.h"
+#include "regs/xe_lrc_layout.h"
+#include "xe_assert.h"
+#include "xe_devcoredump.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_force_wake.h"
+#include "xe_gpu_scheduler.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_guc_ct.h"
+#include "xe_guc_exec_queue_types.h"
+#include "xe_guc_submit_types.h"
+#include "xe_hw_engine.h"
+#include "xe_hw_fence.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_map.h"
+#include "xe_mocs.h"
+#include "xe_ring_ops_types.h"
+#include "xe_sched_job.h"
+#include "xe_trace.h"
+#include "xe_vm.h"
+
+static struct xe_guc *
+exec_queue_to_guc(struct xe_exec_queue *q)
+{
+ return &q->gt->uc.guc;
+}
+
+/*
+ * Helpers for engine state, using an atomic as some of the bits can transition
+ * as the same time (e.g. a suspend can be happning at the same time as schedule
+ * engine done being processed).
+ */
+#define EXEC_QUEUE_STATE_REGISTERED (1 << 0)
+#define ENGINE_STATE_ENABLED (1 << 1)
+#define EXEC_QUEUE_STATE_PENDING_ENABLE (1 << 2)
+#define EXEC_QUEUE_STATE_PENDING_DISABLE (1 << 3)
+#define EXEC_QUEUE_STATE_DESTROYED (1 << 4)
+#define ENGINE_STATE_SUSPENDED (1 << 5)
+#define EXEC_QUEUE_STATE_RESET (1 << 6)
+#define ENGINE_STATE_KILLED (1 << 7)
+
+static bool exec_queue_registered(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & EXEC_QUEUE_STATE_REGISTERED;
+}
+
+static void set_exec_queue_registered(struct xe_exec_queue *q)
+{
+ atomic_or(EXEC_QUEUE_STATE_REGISTERED, &q->guc->state);
+}
+
+static void clear_exec_queue_registered(struct xe_exec_queue *q)
+{
+ atomic_and(~EXEC_QUEUE_STATE_REGISTERED, &q->guc->state);
+}
+
+static bool exec_queue_enabled(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & ENGINE_STATE_ENABLED;
+}
+
+static void set_exec_queue_enabled(struct xe_exec_queue *q)
+{
+ atomic_or(ENGINE_STATE_ENABLED, &q->guc->state);
+}
+
+static void clear_exec_queue_enabled(struct xe_exec_queue *q)
+{
+ atomic_and(~ENGINE_STATE_ENABLED, &q->guc->state);
+}
+
+static bool exec_queue_pending_enable(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & EXEC_QUEUE_STATE_PENDING_ENABLE;
+}
+
+static void set_exec_queue_pending_enable(struct xe_exec_queue *q)
+{
+ atomic_or(EXEC_QUEUE_STATE_PENDING_ENABLE, &q->guc->state);
+}
+
+static void clear_exec_queue_pending_enable(struct xe_exec_queue *q)
+{
+ atomic_and(~EXEC_QUEUE_STATE_PENDING_ENABLE, &q->guc->state);
+}
+
+static bool exec_queue_pending_disable(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & EXEC_QUEUE_STATE_PENDING_DISABLE;
+}
+
+static void set_exec_queue_pending_disable(struct xe_exec_queue *q)
+{
+ atomic_or(EXEC_QUEUE_STATE_PENDING_DISABLE, &q->guc->state);
+}
+
+static void clear_exec_queue_pending_disable(struct xe_exec_queue *q)
+{
+ atomic_and(~EXEC_QUEUE_STATE_PENDING_DISABLE, &q->guc->state);
+}
+
+static bool exec_queue_destroyed(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & EXEC_QUEUE_STATE_DESTROYED;
+}
+
+static void set_exec_queue_destroyed(struct xe_exec_queue *q)
+{
+ atomic_or(EXEC_QUEUE_STATE_DESTROYED, &q->guc->state);
+}
+
+static bool exec_queue_banned(struct xe_exec_queue *q)
+{
+ return (q->flags & EXEC_QUEUE_FLAG_BANNED);
+}
+
+static void set_exec_queue_banned(struct xe_exec_queue *q)
+{
+ q->flags |= EXEC_QUEUE_FLAG_BANNED;
+}
+
+static bool exec_queue_suspended(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & ENGINE_STATE_SUSPENDED;
+}
+
+static void set_exec_queue_suspended(struct xe_exec_queue *q)
+{
+ atomic_or(ENGINE_STATE_SUSPENDED, &q->guc->state);
+}
+
+static void clear_exec_queue_suspended(struct xe_exec_queue *q)
+{
+ atomic_and(~ENGINE_STATE_SUSPENDED, &q->guc->state);
+}
+
+static bool exec_queue_reset(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & EXEC_QUEUE_STATE_RESET;
+}
+
+static void set_exec_queue_reset(struct xe_exec_queue *q)
+{
+ atomic_or(EXEC_QUEUE_STATE_RESET, &q->guc->state);
+}
+
+static bool exec_queue_killed(struct xe_exec_queue *q)
+{
+ return atomic_read(&q->guc->state) & ENGINE_STATE_KILLED;
+}
+
+static void set_exec_queue_killed(struct xe_exec_queue *q)
+{
+ atomic_or(ENGINE_STATE_KILLED, &q->guc->state);
+}
+
+static bool exec_queue_killed_or_banned(struct xe_exec_queue *q)
+{
+ return exec_queue_killed(q) || exec_queue_banned(q);
+}
+
+#ifdef CONFIG_PROVE_LOCKING
+static int alloc_submit_wq(struct xe_guc *guc)
+{
+ int i;
+
+ for (i = 0; i < NUM_SUBMIT_WQ; ++i) {
+ guc->submission_state.submit_wq_pool[i] =
+ alloc_ordered_workqueue("submit_wq", 0);
+ if (!guc->submission_state.submit_wq_pool[i])
+ goto err_free;
+ }
+
+ return 0;
+
+err_free:
+ while (i)
+ destroy_workqueue(guc->submission_state.submit_wq_pool[--i]);
+
+ return -ENOMEM;
+}
+
+static void free_submit_wq(struct xe_guc *guc)
+{
+ int i;
+
+ for (i = 0; i < NUM_SUBMIT_WQ; ++i)
+ destroy_workqueue(guc->submission_state.submit_wq_pool[i]);
+}
+
+static struct workqueue_struct *get_submit_wq(struct xe_guc *guc)
+{
+ int idx = guc->submission_state.submit_wq_idx++ % NUM_SUBMIT_WQ;
+
+ return guc->submission_state.submit_wq_pool[idx];
+}
+#else
+static int alloc_submit_wq(struct xe_guc *guc)
+{
+ return 0;
+}
+
+static void free_submit_wq(struct xe_guc *guc)
+{
+
+}
+
+static struct workqueue_struct *get_submit_wq(struct xe_guc *guc)
+{
+ return NULL;
+}
+#endif
+
+static void guc_submit_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_guc *guc = arg;
+
+ xa_destroy(&guc->submission_state.exec_queue_lookup);
+ ida_destroy(&guc->submission_state.guc_ids);
+ bitmap_free(guc->submission_state.guc_ids_bitmap);
+ free_submit_wq(guc);
+ mutex_destroy(&guc->submission_state.lock);
+}
+
+#define GUC_ID_MAX 65535
+#define GUC_ID_NUMBER_MLRC 4096
+#define GUC_ID_NUMBER_SLRC (GUC_ID_MAX - GUC_ID_NUMBER_MLRC)
+#define GUC_ID_START_MLRC GUC_ID_NUMBER_SLRC
+
+static const struct xe_exec_queue_ops guc_exec_queue_ops;
+
+static void primelockdep(struct xe_guc *guc)
+{
+ if (!IS_ENABLED(CONFIG_LOCKDEP))
+ return;
+
+ fs_reclaim_acquire(GFP_KERNEL);
+
+ mutex_lock(&guc->submission_state.lock);
+ might_lock(&guc->submission_state.suspend.lock);
+ mutex_unlock(&guc->submission_state.lock);
+
+ fs_reclaim_release(GFP_KERNEL);
+}
+
+int xe_guc_submit_init(struct xe_guc *guc)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ int err;
+
+ guc->submission_state.guc_ids_bitmap =
+ bitmap_zalloc(GUC_ID_NUMBER_MLRC, GFP_KERNEL);
+ if (!guc->submission_state.guc_ids_bitmap)
+ return -ENOMEM;
+
+ err = alloc_submit_wq(guc);
+ if (err) {
+ bitmap_free(guc->submission_state.guc_ids_bitmap);
+ return err;
+ }
+
+ gt->exec_queue_ops = &guc_exec_queue_ops;
+
+ mutex_init(&guc->submission_state.lock);
+ xa_init(&guc->submission_state.exec_queue_lookup);
+ ida_init(&guc->submission_state.guc_ids);
+
+ spin_lock_init(&guc->submission_state.suspend.lock);
+ guc->submission_state.suspend.context = dma_fence_context_alloc(1);
+
+ primelockdep(guc);
+
+ err = drmm_add_action_or_reset(&xe->drm, guc_submit_fini, guc);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static void __release_guc_id(struct xe_guc *guc, struct xe_exec_queue *q, u32 xa_count)
+{
+ int i;
+
+ lockdep_assert_held(&guc->submission_state.lock);
+
+ for (i = 0; i < xa_count; ++i)
+ xa_erase(&guc->submission_state.exec_queue_lookup, q->guc->id + i);
+
+ if (xe_exec_queue_is_parallel(q))
+ bitmap_release_region(guc->submission_state.guc_ids_bitmap,
+ q->guc->id - GUC_ID_START_MLRC,
+ order_base_2(q->width));
+ else
+ ida_simple_remove(&guc->submission_state.guc_ids, q->guc->id);
+}
+
+static int alloc_guc_id(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ int ret;
+ void *ptr;
+ int i;
+
+ /*
+ * Must use GFP_NOWAIT as this lock is in the dma fence signalling path,
+ * worse case user gets -ENOMEM on engine create and has to try again.
+ *
+ * FIXME: Have caller pre-alloc or post-alloc /w GFP_KERNEL to prevent
+ * failure.
+ */
+ lockdep_assert_held(&guc->submission_state.lock);
+
+ if (xe_exec_queue_is_parallel(q)) {
+ void *bitmap = guc->submission_state.guc_ids_bitmap;
+
+ ret = bitmap_find_free_region(bitmap, GUC_ID_NUMBER_MLRC,
+ order_base_2(q->width));
+ } else {
+ ret = ida_simple_get(&guc->submission_state.guc_ids, 0,
+ GUC_ID_NUMBER_SLRC, GFP_NOWAIT);
+ }
+ if (ret < 0)
+ return ret;
+
+ q->guc->id = ret;
+ if (xe_exec_queue_is_parallel(q))
+ q->guc->id += GUC_ID_START_MLRC;
+
+ for (i = 0; i < q->width; ++i) {
+ ptr = xa_store(&guc->submission_state.exec_queue_lookup,
+ q->guc->id + i, q, GFP_NOWAIT);
+ if (IS_ERR(ptr)) {
+ ret = PTR_ERR(ptr);
+ goto err_release;
+ }
+ }
+
+ return 0;
+
+err_release:
+ __release_guc_id(guc, q, i);
+
+ return ret;
+}
+
+static void release_guc_id(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ mutex_lock(&guc->submission_state.lock);
+ __release_guc_id(guc, q, q->width);
+ mutex_unlock(&guc->submission_state.lock);
+}
+
+struct exec_queue_policy {
+ u32 count;
+ struct guc_update_exec_queue_policy h2g;
+};
+
+static u32 __guc_exec_queue_policy_action_size(struct exec_queue_policy *policy)
+{
+ size_t bytes = sizeof(policy->h2g.header) +
+ (sizeof(policy->h2g.klv[0]) * policy->count);
+
+ return bytes / sizeof(u32);
+}
+
+static void __guc_exec_queue_policy_start_klv(struct exec_queue_policy *policy,
+ u16 guc_id)
+{
+ policy->h2g.header.action =
+ XE_GUC_ACTION_HOST2GUC_UPDATE_CONTEXT_POLICIES;
+ policy->h2g.header.guc_id = guc_id;
+ policy->count = 0;
+}
+
+#define MAKE_EXEC_QUEUE_POLICY_ADD(func, id) \
+static void __guc_exec_queue_policy_add_##func(struct exec_queue_policy *policy, \
+ u32 data) \
+{ \
+ XE_WARN_ON(policy->count >= GUC_CONTEXT_POLICIES_KLV_NUM_IDS); \
+\
+ policy->h2g.klv[policy->count].kl = \
+ FIELD_PREP(GUC_KLV_0_KEY, \
+ GUC_CONTEXT_POLICIES_KLV_ID_##id) | \
+ FIELD_PREP(GUC_KLV_0_LEN, 1); \
+ policy->h2g.klv[policy->count].value = data; \
+ policy->count++; \
+}
+
+MAKE_EXEC_QUEUE_POLICY_ADD(execution_quantum, EXECUTION_QUANTUM)
+MAKE_EXEC_QUEUE_POLICY_ADD(preemption_timeout, PREEMPTION_TIMEOUT)
+MAKE_EXEC_QUEUE_POLICY_ADD(priority, SCHEDULING_PRIORITY)
+#undef MAKE_EXEC_QUEUE_POLICY_ADD
+
+static const int xe_exec_queue_prio_to_guc[] = {
+ [XE_EXEC_QUEUE_PRIORITY_LOW] = GUC_CLIENT_PRIORITY_NORMAL,
+ [XE_EXEC_QUEUE_PRIORITY_NORMAL] = GUC_CLIENT_PRIORITY_KMD_NORMAL,
+ [XE_EXEC_QUEUE_PRIORITY_HIGH] = GUC_CLIENT_PRIORITY_HIGH,
+ [XE_EXEC_QUEUE_PRIORITY_KERNEL] = GUC_CLIENT_PRIORITY_KMD_HIGH,
+};
+
+static void init_policies(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ struct exec_queue_policy policy;
+ struct xe_device *xe = guc_to_xe(guc);
+ enum xe_exec_queue_priority prio = q->priority;
+ u32 timeslice_us = q->sched_props.timeslice_us;
+ u32 preempt_timeout_us = q->sched_props.preempt_timeout_us;
+
+ xe_assert(xe, exec_queue_registered(q));
+
+ __guc_exec_queue_policy_start_klv(&policy, q->guc->id);
+ __guc_exec_queue_policy_add_priority(&policy, xe_exec_queue_prio_to_guc[prio]);
+ __guc_exec_queue_policy_add_execution_quantum(&policy, timeslice_us);
+ __guc_exec_queue_policy_add_preemption_timeout(&policy, preempt_timeout_us);
+
+ xe_guc_ct_send(&guc->ct, (u32 *)&policy.h2g,
+ __guc_exec_queue_policy_action_size(&policy), 0, 0);
+}
+
+static void set_min_preemption_timeout(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ struct exec_queue_policy policy;
+
+ __guc_exec_queue_policy_start_klv(&policy, q->guc->id);
+ __guc_exec_queue_policy_add_preemption_timeout(&policy, 1);
+
+ xe_guc_ct_send(&guc->ct, (u32 *)&policy.h2g,
+ __guc_exec_queue_policy_action_size(&policy), 0, 0);
+}
+
+#define parallel_read(xe_, map_, field_) \
+ xe_map_rd_field(xe_, &map_, 0, struct guc_submit_parallel_scratch, \
+ field_)
+#define parallel_write(xe_, map_, field_, val_) \
+ xe_map_wr_field(xe_, &map_, 0, struct guc_submit_parallel_scratch, \
+ field_, val_)
+
+static void __register_mlrc_engine(struct xe_guc *guc,
+ struct xe_exec_queue *q,
+ struct guc_ctxt_registration_info *info)
+{
+#define MAX_MLRC_REG_SIZE (13 + XE_HW_ENGINE_MAX_INSTANCE * 2)
+ struct xe_device *xe = guc_to_xe(guc);
+ u32 action[MAX_MLRC_REG_SIZE];
+ int len = 0;
+ int i;
+
+ xe_assert(xe, xe_exec_queue_is_parallel(q));
+
+ action[len++] = XE_GUC_ACTION_REGISTER_CONTEXT_MULTI_LRC;
+ action[len++] = info->flags;
+ action[len++] = info->context_idx;
+ action[len++] = info->engine_class;
+ action[len++] = info->engine_submit_mask;
+ action[len++] = info->wq_desc_lo;
+ action[len++] = info->wq_desc_hi;
+ action[len++] = info->wq_base_lo;
+ action[len++] = info->wq_base_hi;
+ action[len++] = info->wq_size;
+ action[len++] = q->width;
+ action[len++] = info->hwlrca_lo;
+ action[len++] = info->hwlrca_hi;
+
+ for (i = 1; i < q->width; ++i) {
+ struct xe_lrc *lrc = q->lrc + i;
+
+ action[len++] = lower_32_bits(xe_lrc_descriptor(lrc));
+ action[len++] = upper_32_bits(xe_lrc_descriptor(lrc));
+ }
+
+ xe_assert(xe, len <= MAX_MLRC_REG_SIZE);
+#undef MAX_MLRC_REG_SIZE
+
+ xe_guc_ct_send(&guc->ct, action, len, 0, 0);
+}
+
+static void __register_engine(struct xe_guc *guc,
+ struct guc_ctxt_registration_info *info)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_REGISTER_CONTEXT,
+ info->flags,
+ info->context_idx,
+ info->engine_class,
+ info->engine_submit_mask,
+ info->wq_desc_lo,
+ info->wq_desc_hi,
+ info->wq_base_lo,
+ info->wq_base_hi,
+ info->wq_size,
+ info->hwlrca_lo,
+ info->hwlrca_hi,
+ };
+
+ xe_guc_ct_send(&guc->ct, action, ARRAY_SIZE(action), 0, 0);
+}
+
+static void register_engine(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_lrc *lrc = q->lrc;
+ struct guc_ctxt_registration_info info;
+
+ xe_assert(xe, !exec_queue_registered(q));
+
+ memset(&info, 0, sizeof(info));
+ info.context_idx = q->guc->id;
+ info.engine_class = xe_engine_class_to_guc_class(q->class);
+ info.engine_submit_mask = q->logical_mask;
+ info.hwlrca_lo = lower_32_bits(xe_lrc_descriptor(lrc));
+ info.hwlrca_hi = upper_32_bits(xe_lrc_descriptor(lrc));
+ info.flags = CONTEXT_REGISTRATION_FLAG_KMD;
+
+ if (xe_exec_queue_is_parallel(q)) {
+ u32 ggtt_addr = xe_lrc_parallel_ggtt_addr(lrc);
+ struct iosys_map map = xe_lrc_parallel_map(lrc);
+
+ info.wq_desc_lo = lower_32_bits(ggtt_addr +
+ offsetof(struct guc_submit_parallel_scratch, wq_desc));
+ info.wq_desc_hi = upper_32_bits(ggtt_addr +
+ offsetof(struct guc_submit_parallel_scratch, wq_desc));
+ info.wq_base_lo = lower_32_bits(ggtt_addr +
+ offsetof(struct guc_submit_parallel_scratch, wq[0]));
+ info.wq_base_hi = upper_32_bits(ggtt_addr +
+ offsetof(struct guc_submit_parallel_scratch, wq[0]));
+ info.wq_size = WQ_SIZE;
+
+ q->guc->wqi_head = 0;
+ q->guc->wqi_tail = 0;
+ xe_map_memset(xe, &map, 0, 0, PARALLEL_SCRATCH_SIZE - WQ_SIZE);
+ parallel_write(xe, map, wq_desc.wq_status, WQ_STATUS_ACTIVE);
+ }
+
+ /*
+ * We must keep a reference for LR engines if engine is registered with
+ * the GuC as jobs signal immediately and can't destroy an engine if the
+ * GuC has a reference to it.
+ */
+ if (xe_exec_queue_is_lr(q))
+ xe_exec_queue_get(q);
+
+ set_exec_queue_registered(q);
+ trace_xe_exec_queue_register(q);
+ if (xe_exec_queue_is_parallel(q))
+ __register_mlrc_engine(guc, q, &info);
+ else
+ __register_engine(guc, &info);
+ init_policies(guc, q);
+}
+
+static u32 wq_space_until_wrap(struct xe_exec_queue *q)
+{
+ return (WQ_SIZE - q->guc->wqi_tail);
+}
+
+static int wq_wait_for_space(struct xe_exec_queue *q, u32 wqi_size)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct iosys_map map = xe_lrc_parallel_map(q->lrc);
+ unsigned int sleep_period_ms = 1;
+
+#define AVAILABLE_SPACE \
+ CIRC_SPACE(q->guc->wqi_tail, q->guc->wqi_head, WQ_SIZE)
+ if (wqi_size > AVAILABLE_SPACE) {
+try_again:
+ q->guc->wqi_head = parallel_read(xe, map, wq_desc.head);
+ if (wqi_size > AVAILABLE_SPACE) {
+ if (sleep_period_ms == 1024) {
+ xe_gt_reset_async(q->gt);
+ return -ENODEV;
+ }
+
+ msleep(sleep_period_ms);
+ sleep_period_ms <<= 1;
+ goto try_again;
+ }
+ }
+#undef AVAILABLE_SPACE
+
+ return 0;
+}
+
+static int wq_noop_append(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct iosys_map map = xe_lrc_parallel_map(q->lrc);
+ u32 len_dw = wq_space_until_wrap(q) / sizeof(u32) - 1;
+
+ if (wq_wait_for_space(q, wq_space_until_wrap(q)))
+ return -ENODEV;
+
+ xe_assert(xe, FIELD_FIT(WQ_LEN_MASK, len_dw));
+
+ parallel_write(xe, map, wq[q->guc->wqi_tail / sizeof(u32)],
+ FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_NOOP) |
+ FIELD_PREP(WQ_LEN_MASK, len_dw));
+ q->guc->wqi_tail = 0;
+
+ return 0;
+}
+
+static void wq_item_append(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct iosys_map map = xe_lrc_parallel_map(q->lrc);
+#define WQ_HEADER_SIZE 4 /* Includes 1 LRC address too */
+ u32 wqi[XE_HW_ENGINE_MAX_INSTANCE + (WQ_HEADER_SIZE - 1)];
+ u32 wqi_size = (q->width + (WQ_HEADER_SIZE - 1)) * sizeof(u32);
+ u32 len_dw = (wqi_size / sizeof(u32)) - 1;
+ int i = 0, j;
+
+ if (wqi_size > wq_space_until_wrap(q)) {
+ if (wq_noop_append(q))
+ return;
+ }
+ if (wq_wait_for_space(q, wqi_size))
+ return;
+
+ wqi[i++] = FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_MULTI_LRC) |
+ FIELD_PREP(WQ_LEN_MASK, len_dw);
+ wqi[i++] = xe_lrc_descriptor(q->lrc);
+ wqi[i++] = FIELD_PREP(WQ_GUC_ID_MASK, q->guc->id) |
+ FIELD_PREP(WQ_RING_TAIL_MASK, q->lrc->ring.tail / sizeof(u64));
+ wqi[i++] = 0;
+ for (j = 1; j < q->width; ++j) {
+ struct xe_lrc *lrc = q->lrc + j;
+
+ wqi[i++] = lrc->ring.tail / sizeof(u64);
+ }
+
+ xe_assert(xe, i == wqi_size / sizeof(u32));
+
+ iosys_map_incr(&map, offsetof(struct guc_submit_parallel_scratch,
+ wq[q->guc->wqi_tail / sizeof(u32)]));
+ xe_map_memcpy_to(xe, &map, 0, wqi, wqi_size);
+ q->guc->wqi_tail += wqi_size;
+ xe_assert(xe, q->guc->wqi_tail <= WQ_SIZE);
+
+ xe_device_wmb(xe);
+
+ map = xe_lrc_parallel_map(q->lrc);
+ parallel_write(xe, map, wq_desc.tail, q->guc->wqi_tail);
+}
+
+#define RESUME_PENDING ~0x0ull
+static void submit_exec_queue(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_lrc *lrc = q->lrc;
+ u32 action[3];
+ u32 g2h_len = 0;
+ u32 num_g2h = 0;
+ int len = 0;
+ bool extra_submit = false;
+
+ xe_assert(xe, exec_queue_registered(q));
+
+ if (xe_exec_queue_is_parallel(q))
+ wq_item_append(q);
+ else
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_TAIL, lrc->ring.tail);
+
+ if (exec_queue_suspended(q) && !xe_exec_queue_is_parallel(q))
+ return;
+
+ if (!exec_queue_enabled(q) && !exec_queue_suspended(q)) {
+ action[len++] = XE_GUC_ACTION_SCHED_CONTEXT_MODE_SET;
+ action[len++] = q->guc->id;
+ action[len++] = GUC_CONTEXT_ENABLE;
+ g2h_len = G2H_LEN_DW_SCHED_CONTEXT_MODE_SET;
+ num_g2h = 1;
+ if (xe_exec_queue_is_parallel(q))
+ extra_submit = true;
+
+ q->guc->resume_time = RESUME_PENDING;
+ set_exec_queue_pending_enable(q);
+ set_exec_queue_enabled(q);
+ trace_xe_exec_queue_scheduling_enable(q);
+ } else {
+ action[len++] = XE_GUC_ACTION_SCHED_CONTEXT;
+ action[len++] = q->guc->id;
+ trace_xe_exec_queue_submit(q);
+ }
+
+ xe_guc_ct_send(&guc->ct, action, len, g2h_len, num_g2h);
+
+ if (extra_submit) {
+ len = 0;
+ action[len++] = XE_GUC_ACTION_SCHED_CONTEXT;
+ action[len++] = q->guc->id;
+ trace_xe_exec_queue_submit(q);
+
+ xe_guc_ct_send(&guc->ct, action, len, 0, 0);
+ }
+}
+
+static struct dma_fence *
+guc_exec_queue_run_job(struct drm_sched_job *drm_job)
+{
+ struct xe_sched_job *job = to_xe_sched_job(drm_job);
+ struct xe_exec_queue *q = job->q;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ bool lr = xe_exec_queue_is_lr(q);
+
+ xe_assert(xe, !(exec_queue_destroyed(q) || exec_queue_pending_disable(q)) ||
+ exec_queue_banned(q) || exec_queue_suspended(q));
+
+ trace_xe_sched_job_run(job);
+
+ if (!exec_queue_killed_or_banned(q) && !xe_sched_job_is_error(job)) {
+ if (!exec_queue_registered(q))
+ register_engine(q);
+ if (!lr) /* LR jobs are emitted in the exec IOCTL */
+ q->ring_ops->emit_job(job);
+ submit_exec_queue(q);
+ }
+
+ if (lr) {
+ xe_sched_job_set_error(job, -EOPNOTSUPP);
+ return NULL;
+ } else if (test_and_set_bit(JOB_FLAG_SUBMIT, &job->fence->flags)) {
+ return job->fence;
+ } else {
+ return dma_fence_get(job->fence);
+ }
+}
+
+static void guc_exec_queue_free_job(struct drm_sched_job *drm_job)
+{
+ struct xe_sched_job *job = to_xe_sched_job(drm_job);
+
+ trace_xe_sched_job_free(job);
+ xe_sched_job_put(job);
+}
+
+static int guc_read_stopped(struct xe_guc *guc)
+{
+ return atomic_read(&guc->submission_state.stopped);
+}
+
+#define MAKE_SCHED_CONTEXT_ACTION(q, enable_disable) \
+ u32 action[] = { \
+ XE_GUC_ACTION_SCHED_CONTEXT_MODE_SET, \
+ q->guc->id, \
+ GUC_CONTEXT_##enable_disable, \
+ }
+
+static void disable_scheduling_deregister(struct xe_guc *guc,
+ struct xe_exec_queue *q)
+{
+ MAKE_SCHED_CONTEXT_ACTION(q, DISABLE);
+ struct xe_device *xe = guc_to_xe(guc);
+ int ret;
+
+ set_min_preemption_timeout(guc, q);
+ smp_rmb();
+ ret = wait_event_timeout(guc->ct.wq, !exec_queue_pending_enable(q) ||
+ guc_read_stopped(guc), HZ * 5);
+ if (!ret) {
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+
+ drm_warn(&xe->drm, "Pending enable failed to respond");
+ xe_sched_submission_start(sched);
+ xe_gt_reset_async(q->gt);
+ xe_sched_tdr_queue_imm(sched);
+ return;
+ }
+
+ clear_exec_queue_enabled(q);
+ set_exec_queue_pending_disable(q);
+ set_exec_queue_destroyed(q);
+ trace_xe_exec_queue_scheduling_disable(q);
+
+ /*
+ * Reserve space for both G2H here as the 2nd G2H is sent from a G2H
+ * handler and we are not allowed to reserved G2H space in handlers.
+ */
+ xe_guc_ct_send(&guc->ct, action, ARRAY_SIZE(action),
+ G2H_LEN_DW_SCHED_CONTEXT_MODE_SET +
+ G2H_LEN_DW_DEREGISTER_CONTEXT, 2);
+}
+
+static void guc_exec_queue_print(struct xe_exec_queue *q, struct drm_printer *p);
+
+#if IS_ENABLED(CONFIG_DRM_XE_SIMPLE_ERROR_CAPTURE)
+static void simple_error_capture(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct drm_printer p = drm_err_printer("");
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ u32 adj_logical_mask = q->logical_mask;
+ u32 width_mask = (0x1 << q->width) - 1;
+ int i;
+ bool cookie;
+
+ if (q->vm && !q->vm->error_capture.capture_once) {
+ q->vm->error_capture.capture_once = true;
+ cookie = dma_fence_begin_signalling();
+ for (i = 0; q->width > 1 && i < XE_HW_ENGINE_MAX_INSTANCE;) {
+ if (adj_logical_mask & BIT(i)) {
+ adj_logical_mask |= width_mask << i;
+ i += q->width;
+ } else {
+ ++i;
+ }
+ }
+
+ xe_force_wake_get(gt_to_fw(guc_to_gt(guc)), XE_FORCEWAKE_ALL);
+ xe_guc_ct_print(&guc->ct, &p, true);
+ guc_exec_queue_print(q, &p);
+ for_each_hw_engine(hwe, guc_to_gt(guc), id) {
+ if (hwe->class != q->hwe->class ||
+ !(BIT(hwe->logical_instance) & adj_logical_mask))
+ continue;
+ xe_hw_engine_print(hwe, &p);
+ }
+ xe_analyze_vm(&p, q->vm, q->gt->info.id);
+ xe_force_wake_put(gt_to_fw(guc_to_gt(guc)), XE_FORCEWAKE_ALL);
+ dma_fence_end_signalling(cookie);
+ }
+}
+#else
+static void simple_error_capture(struct xe_exec_queue *q)
+{
+}
+#endif
+
+static void xe_guc_exec_queue_trigger_cleanup(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+
+ /** to wakeup xe_wait_user_fence ioctl if exec queue is reset */
+ wake_up_all(&xe->ufence_wq);
+
+ if (xe_exec_queue_is_lr(q))
+ queue_work(guc_to_gt(guc)->ordered_wq, &q->guc->lr_tdr);
+ else
+ xe_sched_tdr_queue_imm(&q->guc->sched);
+}
+
+static void xe_guc_exec_queue_lr_cleanup(struct work_struct *w)
+{
+ struct xe_guc_exec_queue *ge =
+ container_of(w, struct xe_guc_exec_queue, lr_tdr);
+ struct xe_exec_queue *q = ge->q;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gpu_scheduler *sched = &ge->sched;
+
+ xe_assert(xe, xe_exec_queue_is_lr(q));
+ trace_xe_exec_queue_lr_cleanup(q);
+
+ /* Kill the run_job / process_msg entry points */
+ xe_sched_submission_stop(sched);
+
+ /*
+ * Engine state now mostly stable, disable scheduling / deregister if
+ * needed. This cleanup routine might be called multiple times, where
+ * the actual async engine deregister drops the final engine ref.
+ * Calling disable_scheduling_deregister will mark the engine as
+ * destroyed and fire off the CT requests to disable scheduling /
+ * deregister, which we only want to do once. We also don't want to mark
+ * the engine as pending_disable again as this may race with the
+ * xe_guc_deregister_done_handler() which treats it as an unexpected
+ * state.
+ */
+ if (exec_queue_registered(q) && !exec_queue_destroyed(q)) {
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ int ret;
+
+ set_exec_queue_banned(q);
+ disable_scheduling_deregister(guc, q);
+
+ /*
+ * Must wait for scheduling to be disabled before signalling
+ * any fences, if GT broken the GT reset code should signal us.
+ */
+ ret = wait_event_timeout(guc->ct.wq,
+ !exec_queue_pending_disable(q) ||
+ guc_read_stopped(guc), HZ * 5);
+ if (!ret) {
+ drm_warn(&xe->drm, "Schedule disable failed to respond");
+ xe_sched_submission_start(sched);
+ xe_gt_reset_async(q->gt);
+ return;
+ }
+ }
+
+ xe_sched_submission_start(sched);
+}
+
+static enum drm_gpu_sched_stat
+guc_exec_queue_timedout_job(struct drm_sched_job *drm_job)
+{
+ struct xe_sched_job *job = to_xe_sched_job(drm_job);
+ struct xe_sched_job *tmp_job;
+ struct xe_exec_queue *q = job->q;
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+ struct xe_device *xe = guc_to_xe(exec_queue_to_guc(q));
+ int err = -ETIME;
+ int i = 0;
+
+ if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &job->fence->flags)) {
+ xe_assert(xe, !(q->flags & EXEC_QUEUE_FLAG_KERNEL));
+ xe_assert(xe, !(q->flags & EXEC_QUEUE_FLAG_VM && !exec_queue_killed(q)));
+
+ drm_notice(&xe->drm, "Timedout job: seqno=%u, guc_id=%d, flags=0x%lx",
+ xe_sched_job_seqno(job), q->guc->id, q->flags);
+ simple_error_capture(q);
+ xe_devcoredump(q);
+ } else {
+ drm_dbg(&xe->drm, "Timedout signaled job: seqno=%u, guc_id=%d, flags=0x%lx",
+ xe_sched_job_seqno(job), q->guc->id, q->flags);
+ }
+ trace_xe_sched_job_timedout(job);
+
+ /* Kill the run_job entry point */
+ xe_sched_submission_stop(sched);
+
+ /*
+ * Kernel jobs should never fail, nor should VM jobs if they do
+ * somethings has gone wrong and the GT needs a reset
+ */
+ if (q->flags & EXEC_QUEUE_FLAG_KERNEL ||
+ (q->flags & EXEC_QUEUE_FLAG_VM && !exec_queue_killed(q))) {
+ if (!xe_sched_invalidate_job(job, 2)) {
+ xe_sched_add_pending_job(sched, job);
+ xe_sched_submission_start(sched);
+ xe_gt_reset_async(q->gt);
+ goto out;
+ }
+ }
+
+ /* Engine state now stable, disable scheduling if needed */
+ if (exec_queue_registered(q)) {
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ int ret;
+
+ if (exec_queue_reset(q))
+ err = -EIO;
+ set_exec_queue_banned(q);
+ if (!exec_queue_destroyed(q)) {
+ xe_exec_queue_get(q);
+ disable_scheduling_deregister(guc, q);
+ }
+
+ /*
+ * Must wait for scheduling to be disabled before signalling
+ * any fences, if GT broken the GT reset code should signal us.
+ *
+ * FIXME: Tests can generate a ton of 0x6000 (IOMMU CAT fault
+ * error) messages which can cause the schedule disable to get
+ * lost. If this occurs, trigger a GT reset to recover.
+ */
+ smp_rmb();
+ ret = wait_event_timeout(guc->ct.wq,
+ !exec_queue_pending_disable(q) ||
+ guc_read_stopped(guc), HZ * 5);
+ if (!ret || guc_read_stopped(guc)) {
+ drm_warn(&xe->drm, "Schedule disable failed to respond");
+ xe_sched_add_pending_job(sched, job);
+ xe_sched_submission_start(sched);
+ xe_gt_reset_async(q->gt);
+ xe_sched_tdr_queue_imm(sched);
+ goto out;
+ }
+ }
+
+ /* Stop fence signaling */
+ xe_hw_fence_irq_stop(q->fence_irq);
+
+ /*
+ * Fence state now stable, stop / start scheduler which cleans up any
+ * fences that are complete
+ */
+ xe_sched_add_pending_job(sched, job);
+ xe_sched_submission_start(sched);
+ xe_guc_exec_queue_trigger_cleanup(q);
+
+ /* Mark all outstanding jobs as bad, thus completing them */
+ spin_lock(&sched->base.job_list_lock);
+ list_for_each_entry(tmp_job, &sched->base.pending_list, drm.list)
+ xe_sched_job_set_error(tmp_job, !i++ ? err : -ECANCELED);
+ spin_unlock(&sched->base.job_list_lock);
+
+ /* Start fence signaling */
+ xe_hw_fence_irq_start(q->fence_irq);
+
+out:
+ return DRM_GPU_SCHED_STAT_NOMINAL;
+}
+
+static void __guc_exec_queue_fini_async(struct work_struct *w)
+{
+ struct xe_guc_exec_queue *ge =
+ container_of(w, struct xe_guc_exec_queue, fini_async);
+ struct xe_exec_queue *q = ge->q;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+
+ trace_xe_exec_queue_destroy(q);
+
+ if (xe_exec_queue_is_lr(q))
+ cancel_work_sync(&ge->lr_tdr);
+ if (q->flags & EXEC_QUEUE_FLAG_PERSISTENT)
+ xe_device_remove_persistent_exec_queues(gt_to_xe(q->gt), q);
+ release_guc_id(guc, q);
+ xe_sched_entity_fini(&ge->entity);
+ xe_sched_fini(&ge->sched);
+
+ kfree(ge);
+ xe_exec_queue_fini(q);
+}
+
+static void guc_exec_queue_fini_async(struct xe_exec_queue *q)
+{
+ INIT_WORK(&q->guc->fini_async, __guc_exec_queue_fini_async);
+
+ /* We must block on kernel engines so slabs are empty on driver unload */
+ if (q->flags & EXEC_QUEUE_FLAG_PERMANENT)
+ __guc_exec_queue_fini_async(&q->guc->fini_async);
+ else
+ queue_work(system_wq, &q->guc->fini_async);
+}
+
+static void __guc_exec_queue_fini(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ /*
+ * Might be done from within the GPU scheduler, need to do async as we
+ * fini the scheduler when the engine is fini'd, the scheduler can't
+ * complete fini within itself (circular dependency). Async resolves
+ * this we and don't really care when everything is fini'd, just that it
+ * is.
+ */
+ guc_exec_queue_fini_async(q);
+}
+
+static void __guc_exec_queue_process_msg_cleanup(struct xe_sched_msg *msg)
+{
+ struct xe_exec_queue *q = msg->private_data;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, !(q->flags & EXEC_QUEUE_FLAG_PERMANENT));
+ trace_xe_exec_queue_cleanup_entity(q);
+
+ if (exec_queue_registered(q))
+ disable_scheduling_deregister(guc, q);
+ else
+ __guc_exec_queue_fini(guc, q);
+}
+
+static bool guc_exec_queue_allowed_to_change_state(struct xe_exec_queue *q)
+{
+ return !exec_queue_killed_or_banned(q) && exec_queue_registered(q);
+}
+
+static void __guc_exec_queue_process_msg_set_sched_props(struct xe_sched_msg *msg)
+{
+ struct xe_exec_queue *q = msg->private_data;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+
+ if (guc_exec_queue_allowed_to_change_state(q))
+ init_policies(guc, q);
+ kfree(msg);
+}
+
+static void suspend_fence_signal(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, exec_queue_suspended(q) || exec_queue_killed(q) ||
+ guc_read_stopped(guc));
+ xe_assert(xe, q->guc->suspend_pending);
+
+ q->guc->suspend_pending = false;
+ smp_wmb();
+ wake_up(&q->guc->suspend_wait);
+}
+
+static void __guc_exec_queue_process_msg_suspend(struct xe_sched_msg *msg)
+{
+ struct xe_exec_queue *q = msg->private_data;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+
+ if (guc_exec_queue_allowed_to_change_state(q) && !exec_queue_suspended(q) &&
+ exec_queue_enabled(q)) {
+ wait_event(guc->ct.wq, q->guc->resume_time != RESUME_PENDING ||
+ guc_read_stopped(guc));
+
+ if (!guc_read_stopped(guc)) {
+ MAKE_SCHED_CONTEXT_ACTION(q, DISABLE);
+ s64 since_resume_ms =
+ ktime_ms_delta(ktime_get(),
+ q->guc->resume_time);
+ s64 wait_ms = q->vm->preempt.min_run_period_ms -
+ since_resume_ms;
+
+ if (wait_ms > 0 && q->guc->resume_time)
+ msleep(wait_ms);
+
+ set_exec_queue_suspended(q);
+ clear_exec_queue_enabled(q);
+ set_exec_queue_pending_disable(q);
+ trace_xe_exec_queue_scheduling_disable(q);
+
+ xe_guc_ct_send(&guc->ct, action, ARRAY_SIZE(action),
+ G2H_LEN_DW_SCHED_CONTEXT_MODE_SET, 1);
+ }
+ } else if (q->guc->suspend_pending) {
+ set_exec_queue_suspended(q);
+ suspend_fence_signal(q);
+ }
+}
+
+static void __guc_exec_queue_process_msg_resume(struct xe_sched_msg *msg)
+{
+ struct xe_exec_queue *q = msg->private_data;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+
+ if (guc_exec_queue_allowed_to_change_state(q)) {
+ MAKE_SCHED_CONTEXT_ACTION(q, ENABLE);
+
+ q->guc->resume_time = RESUME_PENDING;
+ clear_exec_queue_suspended(q);
+ set_exec_queue_pending_enable(q);
+ set_exec_queue_enabled(q);
+ trace_xe_exec_queue_scheduling_enable(q);
+
+ xe_guc_ct_send(&guc->ct, action, ARRAY_SIZE(action),
+ G2H_LEN_DW_SCHED_CONTEXT_MODE_SET, 1);
+ } else {
+ clear_exec_queue_suspended(q);
+ }
+}
+
+#define CLEANUP 1 /* Non-zero values to catch uninitialized msg */
+#define SET_SCHED_PROPS 2
+#define SUSPEND 3
+#define RESUME 4
+
+static void guc_exec_queue_process_msg(struct xe_sched_msg *msg)
+{
+ trace_xe_sched_msg_recv(msg);
+
+ switch (msg->opcode) {
+ case CLEANUP:
+ __guc_exec_queue_process_msg_cleanup(msg);
+ break;
+ case SET_SCHED_PROPS:
+ __guc_exec_queue_process_msg_set_sched_props(msg);
+ break;
+ case SUSPEND:
+ __guc_exec_queue_process_msg_suspend(msg);
+ break;
+ case RESUME:
+ __guc_exec_queue_process_msg_resume(msg);
+ break;
+ default:
+ XE_WARN_ON("Unknown message type");
+ }
+}
+
+static const struct drm_sched_backend_ops drm_sched_ops = {
+ .run_job = guc_exec_queue_run_job,
+ .free_job = guc_exec_queue_free_job,
+ .timedout_job = guc_exec_queue_timedout_job,
+};
+
+static const struct xe_sched_backend_ops xe_sched_ops = {
+ .process_msg = guc_exec_queue_process_msg,
+};
+
+static int guc_exec_queue_init(struct xe_exec_queue *q)
+{
+ struct xe_gpu_scheduler *sched;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_guc_exec_queue *ge;
+ long timeout;
+ int err;
+
+ xe_assert(xe, xe_device_uc_enabled(guc_to_xe(guc)));
+
+ ge = kzalloc(sizeof(*ge), GFP_KERNEL);
+ if (!ge)
+ return -ENOMEM;
+
+ q->guc = ge;
+ ge->q = q;
+ init_waitqueue_head(&ge->suspend_wait);
+
+ timeout = (q->vm && xe_vm_in_lr_mode(q->vm)) ? MAX_SCHEDULE_TIMEOUT :
+ q->hwe->eclass->sched_props.job_timeout_ms;
+ err = xe_sched_init(&ge->sched, &drm_sched_ops, &xe_sched_ops,
+ get_submit_wq(guc),
+ q->lrc[0].ring.size / MAX_JOB_SIZE_BYTES, 64,
+ timeout, guc_to_gt(guc)->ordered_wq, NULL,
+ q->name, gt_to_xe(q->gt)->drm.dev);
+ if (err)
+ goto err_free;
+
+ sched = &ge->sched;
+ err = xe_sched_entity_init(&ge->entity, sched);
+ if (err)
+ goto err_sched;
+ q->priority = XE_EXEC_QUEUE_PRIORITY_NORMAL;
+
+ if (xe_exec_queue_is_lr(q))
+ INIT_WORK(&q->guc->lr_tdr, xe_guc_exec_queue_lr_cleanup);
+
+ mutex_lock(&guc->submission_state.lock);
+
+ err = alloc_guc_id(guc, q);
+ if (err)
+ goto err_entity;
+
+ q->entity = &ge->entity;
+
+ if (guc_read_stopped(guc))
+ xe_sched_stop(sched);
+
+ mutex_unlock(&guc->submission_state.lock);
+
+ xe_exec_queue_assign_name(q, q->guc->id);
+
+ trace_xe_exec_queue_create(q);
+
+ return 0;
+
+err_entity:
+ xe_sched_entity_fini(&ge->entity);
+err_sched:
+ xe_sched_fini(&ge->sched);
+err_free:
+ kfree(ge);
+
+ return err;
+}
+
+static void guc_exec_queue_kill(struct xe_exec_queue *q)
+{
+ trace_xe_exec_queue_kill(q);
+ set_exec_queue_killed(q);
+ xe_guc_exec_queue_trigger_cleanup(q);
+}
+
+static void guc_exec_queue_add_msg(struct xe_exec_queue *q, struct xe_sched_msg *msg,
+ u32 opcode)
+{
+ INIT_LIST_HEAD(&msg->link);
+ msg->opcode = opcode;
+ msg->private_data = q;
+
+ trace_xe_sched_msg_add(msg);
+ xe_sched_add_msg(&q->guc->sched, msg);
+}
+
+#define STATIC_MSG_CLEANUP 0
+#define STATIC_MSG_SUSPEND 1
+#define STATIC_MSG_RESUME 2
+static void guc_exec_queue_fini(struct xe_exec_queue *q)
+{
+ struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_CLEANUP;
+
+ if (!(q->flags & EXEC_QUEUE_FLAG_PERMANENT))
+ guc_exec_queue_add_msg(q, msg, CLEANUP);
+ else
+ __guc_exec_queue_fini(exec_queue_to_guc(q), q);
+}
+
+static int guc_exec_queue_set_priority(struct xe_exec_queue *q,
+ enum xe_exec_queue_priority priority)
+{
+ struct xe_sched_msg *msg;
+
+ if (q->priority == priority || exec_queue_killed_or_banned(q))
+ return 0;
+
+ msg = kmalloc(sizeof(*msg), GFP_KERNEL);
+ if (!msg)
+ return -ENOMEM;
+
+ guc_exec_queue_add_msg(q, msg, SET_SCHED_PROPS);
+ q->priority = priority;
+
+ return 0;
+}
+
+static int guc_exec_queue_set_timeslice(struct xe_exec_queue *q, u32 timeslice_us)
+{
+ struct xe_sched_msg *msg;
+
+ if (q->sched_props.timeslice_us == timeslice_us ||
+ exec_queue_killed_or_banned(q))
+ return 0;
+
+ msg = kmalloc(sizeof(*msg), GFP_KERNEL);
+ if (!msg)
+ return -ENOMEM;
+
+ q->sched_props.timeslice_us = timeslice_us;
+ guc_exec_queue_add_msg(q, msg, SET_SCHED_PROPS);
+
+ return 0;
+}
+
+static int guc_exec_queue_set_preempt_timeout(struct xe_exec_queue *q,
+ u32 preempt_timeout_us)
+{
+ struct xe_sched_msg *msg;
+
+ if (q->sched_props.preempt_timeout_us == preempt_timeout_us ||
+ exec_queue_killed_or_banned(q))
+ return 0;
+
+ msg = kmalloc(sizeof(*msg), GFP_KERNEL);
+ if (!msg)
+ return -ENOMEM;
+
+ q->sched_props.preempt_timeout_us = preempt_timeout_us;
+ guc_exec_queue_add_msg(q, msg, SET_SCHED_PROPS);
+
+ return 0;
+}
+
+static int guc_exec_queue_set_job_timeout(struct xe_exec_queue *q, u32 job_timeout_ms)
+{
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, !exec_queue_registered(q));
+ xe_assert(xe, !exec_queue_banned(q));
+ xe_assert(xe, !exec_queue_killed(q));
+
+ sched->base.timeout = job_timeout_ms;
+
+ return 0;
+}
+
+static int guc_exec_queue_suspend(struct xe_exec_queue *q)
+{
+ struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_SUSPEND;
+
+ if (exec_queue_killed_or_banned(q) || q->guc->suspend_pending)
+ return -EINVAL;
+
+ q->guc->suspend_pending = true;
+ guc_exec_queue_add_msg(q, msg, SUSPEND);
+
+ return 0;
+}
+
+static void guc_exec_queue_suspend_wait(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+
+ wait_event(q->guc->suspend_wait, !q->guc->suspend_pending ||
+ guc_read_stopped(guc));
+}
+
+static void guc_exec_queue_resume(struct xe_exec_queue *q)
+{
+ struct xe_sched_msg *msg = q->guc->static_msgs + STATIC_MSG_RESUME;
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, !q->guc->suspend_pending);
+
+ guc_exec_queue_add_msg(q, msg, RESUME);
+}
+
+static bool guc_exec_queue_reset_status(struct xe_exec_queue *q)
+{
+ return exec_queue_reset(q);
+}
+
+/*
+ * All of these functions are an abstraction layer which other parts of XE can
+ * use to trap into the GuC backend. All of these functions, aside from init,
+ * really shouldn't do much other than trap into the DRM scheduler which
+ * synchronizes these operations.
+ */
+static const struct xe_exec_queue_ops guc_exec_queue_ops = {
+ .init = guc_exec_queue_init,
+ .kill = guc_exec_queue_kill,
+ .fini = guc_exec_queue_fini,
+ .set_priority = guc_exec_queue_set_priority,
+ .set_timeslice = guc_exec_queue_set_timeslice,
+ .set_preempt_timeout = guc_exec_queue_set_preempt_timeout,
+ .set_job_timeout = guc_exec_queue_set_job_timeout,
+ .suspend = guc_exec_queue_suspend,
+ .suspend_wait = guc_exec_queue_suspend_wait,
+ .resume = guc_exec_queue_resume,
+ .reset_status = guc_exec_queue_reset_status,
+};
+
+static void guc_exec_queue_stop(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+
+ /* Stop scheduling + flush any DRM scheduler operations */
+ xe_sched_submission_stop(sched);
+
+ /* Clean up lost G2H + reset engine state */
+ if (exec_queue_registered(q)) {
+ if ((exec_queue_banned(q) && exec_queue_destroyed(q)) ||
+ xe_exec_queue_is_lr(q))
+ xe_exec_queue_put(q);
+ else if (exec_queue_destroyed(q))
+ __guc_exec_queue_fini(guc, q);
+ }
+ if (q->guc->suspend_pending) {
+ set_exec_queue_suspended(q);
+ suspend_fence_signal(q);
+ }
+ atomic_and(EXEC_QUEUE_STATE_DESTROYED | ENGINE_STATE_SUSPENDED,
+ &q->guc->state);
+ q->guc->resume_time = 0;
+ trace_xe_exec_queue_stop(q);
+
+ /*
+ * Ban any engine (aside from kernel and engines used for VM ops) with a
+ * started but not complete job or if a job has gone through a GT reset
+ * more than twice.
+ */
+ if (!(q->flags & (EXEC_QUEUE_FLAG_KERNEL | EXEC_QUEUE_FLAG_VM))) {
+ struct xe_sched_job *job = xe_sched_first_pending_job(sched);
+
+ if (job) {
+ if ((xe_sched_job_started(job) &&
+ !xe_sched_job_completed(job)) ||
+ xe_sched_invalidate_job(job, 2)) {
+ trace_xe_sched_job_ban(job);
+ xe_sched_tdr_queue_imm(&q->guc->sched);
+ set_exec_queue_banned(q);
+ }
+ }
+ }
+}
+
+int xe_guc_submit_reset_prepare(struct xe_guc *guc)
+{
+ int ret;
+
+ /*
+ * Using an atomic here rather than submission_state.lock as this
+ * function can be called while holding the CT lock (engine reset
+ * failure). submission_state.lock needs the CT lock to resubmit jobs.
+ * Atomic is not ideal, but it works to prevent against concurrent reset
+ * and releasing any TDRs waiting on guc->submission_state.stopped.
+ */
+ ret = atomic_fetch_or(1, &guc->submission_state.stopped);
+ smp_wmb();
+ wake_up_all(&guc->ct.wq);
+
+ return ret;
+}
+
+void xe_guc_submit_reset_wait(struct xe_guc *guc)
+{
+ wait_event(guc->ct.wq, !guc_read_stopped(guc));
+}
+
+int xe_guc_submit_stop(struct xe_guc *guc)
+{
+ struct xe_exec_queue *q;
+ unsigned long index;
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, guc_read_stopped(guc) == 1);
+
+ mutex_lock(&guc->submission_state.lock);
+
+ xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
+ guc_exec_queue_stop(guc, q);
+
+ mutex_unlock(&guc->submission_state.lock);
+
+ /*
+ * No one can enter the backend at this point, aside from new engine
+ * creation which is protected by guc->submission_state.lock.
+ */
+
+ return 0;
+}
+
+static void guc_exec_queue_start(struct xe_exec_queue *q)
+{
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+
+ if (!exec_queue_killed_or_banned(q)) {
+ int i;
+
+ trace_xe_exec_queue_resubmit(q);
+ for (i = 0; i < q->width; ++i)
+ xe_lrc_set_ring_head(q->lrc + i, q->lrc[i].ring.tail);
+ xe_sched_resubmit_jobs(sched);
+ }
+
+ xe_sched_submission_start(sched);
+}
+
+int xe_guc_submit_start(struct xe_guc *guc)
+{
+ struct xe_exec_queue *q;
+ unsigned long index;
+ struct xe_device *xe = guc_to_xe(guc);
+
+ xe_assert(xe, guc_read_stopped(guc) == 1);
+
+ mutex_lock(&guc->submission_state.lock);
+ atomic_dec(&guc->submission_state.stopped);
+ xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
+ guc_exec_queue_start(q);
+ mutex_unlock(&guc->submission_state.lock);
+
+ wake_up_all(&guc->ct.wq);
+
+ return 0;
+}
+
+static struct xe_exec_queue *
+g2h_exec_queue_lookup(struct xe_guc *guc, u32 guc_id)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_exec_queue *q;
+
+ if (unlikely(guc_id >= GUC_ID_MAX)) {
+ drm_err(&xe->drm, "Invalid guc_id %u", guc_id);
+ return NULL;
+ }
+
+ q = xa_load(&guc->submission_state.exec_queue_lookup, guc_id);
+ if (unlikely(!q)) {
+ drm_err(&xe->drm, "Not engine present for guc_id %u", guc_id);
+ return NULL;
+ }
+
+ xe_assert(xe, guc_id >= q->guc->id);
+ xe_assert(xe, guc_id < (q->guc->id + q->width));
+
+ return q;
+}
+
+static void deregister_exec_queue(struct xe_guc *guc, struct xe_exec_queue *q)
+{
+ u32 action[] = {
+ XE_GUC_ACTION_DEREGISTER_CONTEXT,
+ q->guc->id,
+ };
+
+ trace_xe_exec_queue_deregister(q);
+
+ xe_guc_ct_send_g2h_handler(&guc->ct, action, ARRAY_SIZE(action));
+}
+
+int xe_guc_sched_done_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_exec_queue *q;
+ u32 guc_id = msg[0];
+
+ if (unlikely(len < 2)) {
+ drm_err(&xe->drm, "Invalid length %u", len);
+ return -EPROTO;
+ }
+
+ q = g2h_exec_queue_lookup(guc, guc_id);
+ if (unlikely(!q))
+ return -EPROTO;
+
+ if (unlikely(!exec_queue_pending_enable(q) &&
+ !exec_queue_pending_disable(q))) {
+ drm_err(&xe->drm, "Unexpected engine state 0x%04x",
+ atomic_read(&q->guc->state));
+ return -EPROTO;
+ }
+
+ trace_xe_exec_queue_scheduling_done(q);
+
+ if (exec_queue_pending_enable(q)) {
+ q->guc->resume_time = ktime_get();
+ clear_exec_queue_pending_enable(q);
+ smp_wmb();
+ wake_up_all(&guc->ct.wq);
+ } else {
+ clear_exec_queue_pending_disable(q);
+ if (q->guc->suspend_pending) {
+ suspend_fence_signal(q);
+ } else {
+ if (exec_queue_banned(q)) {
+ smp_wmb();
+ wake_up_all(&guc->ct.wq);
+ }
+ deregister_exec_queue(guc, q);
+ }
+ }
+
+ return 0;
+}
+
+int xe_guc_deregister_done_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_exec_queue *q;
+ u32 guc_id = msg[0];
+
+ if (unlikely(len < 1)) {
+ drm_err(&xe->drm, "Invalid length %u", len);
+ return -EPROTO;
+ }
+
+ q = g2h_exec_queue_lookup(guc, guc_id);
+ if (unlikely(!q))
+ return -EPROTO;
+
+ if (!exec_queue_destroyed(q) || exec_queue_pending_disable(q) ||
+ exec_queue_pending_enable(q) || exec_queue_enabled(q)) {
+ drm_err(&xe->drm, "Unexpected engine state 0x%04x",
+ atomic_read(&q->guc->state));
+ return -EPROTO;
+ }
+
+ trace_xe_exec_queue_deregister_done(q);
+
+ clear_exec_queue_registered(q);
+
+ if (exec_queue_banned(q) || xe_exec_queue_is_lr(q))
+ xe_exec_queue_put(q);
+ else
+ __guc_exec_queue_fini(guc, q);
+
+ return 0;
+}
+
+int xe_guc_exec_queue_reset_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_exec_queue *q;
+ u32 guc_id = msg[0];
+
+ if (unlikely(len < 1)) {
+ drm_err(&xe->drm, "Invalid length %u", len);
+ return -EPROTO;
+ }
+
+ q = g2h_exec_queue_lookup(guc, guc_id);
+ if (unlikely(!q))
+ return -EPROTO;
+
+ drm_info(&xe->drm, "Engine reset: guc_id=%d", guc_id);
+
+ /* FIXME: Do error capture, most likely async */
+
+ trace_xe_exec_queue_reset(q);
+
+ /*
+ * A banned engine is a NOP at this point (came from
+ * guc_exec_queue_timedout_job). Otherwise, kick drm scheduler to cancel
+ * jobs by setting timeout of the job to the minimum value kicking
+ * guc_exec_queue_timedout_job.
+ */
+ set_exec_queue_reset(q);
+ if (!exec_queue_banned(q))
+ xe_guc_exec_queue_trigger_cleanup(q);
+
+ return 0;
+}
+
+int xe_guc_exec_queue_memory_cat_error_handler(struct xe_guc *guc, u32 *msg,
+ u32 len)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_exec_queue *q;
+ u32 guc_id = msg[0];
+
+ if (unlikely(len < 1)) {
+ drm_err(&xe->drm, "Invalid length %u", len);
+ return -EPROTO;
+ }
+
+ q = g2h_exec_queue_lookup(guc, guc_id);
+ if (unlikely(!q))
+ return -EPROTO;
+
+ drm_dbg(&xe->drm, "Engine memory cat error: guc_id=%d", guc_id);
+ trace_xe_exec_queue_memory_cat_error(q);
+
+ /* Treat the same as engine reset */
+ set_exec_queue_reset(q);
+ if (!exec_queue_banned(q))
+ xe_guc_exec_queue_trigger_cleanup(q);
+
+ return 0;
+}
+
+int xe_guc_exec_queue_reset_failure_handler(struct xe_guc *guc, u32 *msg, u32 len)
+{
+ struct xe_device *xe = guc_to_xe(guc);
+ u8 guc_class, instance;
+ u32 reason;
+
+ if (unlikely(len != 3)) {
+ drm_err(&xe->drm, "Invalid length %u", len);
+ return -EPROTO;
+ }
+
+ guc_class = msg[0];
+ instance = msg[1];
+ reason = msg[2];
+
+ /* Unexpected failure of a hardware feature, log an actual error */
+ drm_err(&xe->drm, "GuC engine reset request failed on %d:%d because 0x%08X",
+ guc_class, instance, reason);
+
+ xe_gt_reset_async(guc_to_gt(guc));
+
+ return 0;
+}
+
+static void
+guc_exec_queue_wq_snapshot_capture(struct xe_exec_queue *q,
+ struct xe_guc_submit_exec_queue_snapshot *snapshot)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct iosys_map map = xe_lrc_parallel_map(q->lrc);
+ int i;
+
+ snapshot->guc.wqi_head = q->guc->wqi_head;
+ snapshot->guc.wqi_tail = q->guc->wqi_tail;
+ snapshot->parallel.wq_desc.head = parallel_read(xe, map, wq_desc.head);
+ snapshot->parallel.wq_desc.tail = parallel_read(xe, map, wq_desc.tail);
+ snapshot->parallel.wq_desc.status = parallel_read(xe, map,
+ wq_desc.wq_status);
+
+ if (snapshot->parallel.wq_desc.head !=
+ snapshot->parallel.wq_desc.tail) {
+ for (i = snapshot->parallel.wq_desc.head;
+ i != snapshot->parallel.wq_desc.tail;
+ i = (i + sizeof(u32)) % WQ_SIZE)
+ snapshot->parallel.wq[i / sizeof(u32)] =
+ parallel_read(xe, map, wq[i / sizeof(u32)]);
+ }
+}
+
+static void
+guc_exec_queue_wq_snapshot_print(struct xe_guc_submit_exec_queue_snapshot *snapshot,
+ struct drm_printer *p)
+{
+ int i;
+
+ drm_printf(p, "\tWQ head: %u (internal), %d (memory)\n",
+ snapshot->guc.wqi_head, snapshot->parallel.wq_desc.head);
+ drm_printf(p, "\tWQ tail: %u (internal), %d (memory)\n",
+ snapshot->guc.wqi_tail, snapshot->parallel.wq_desc.tail);
+ drm_printf(p, "\tWQ status: %u\n", snapshot->parallel.wq_desc.status);
+
+ if (snapshot->parallel.wq_desc.head !=
+ snapshot->parallel.wq_desc.tail) {
+ for (i = snapshot->parallel.wq_desc.head;
+ i != snapshot->parallel.wq_desc.tail;
+ i = (i + sizeof(u32)) % WQ_SIZE)
+ drm_printf(p, "\tWQ[%zu]: 0x%08x\n", i / sizeof(u32),
+ snapshot->parallel.wq[i / sizeof(u32)]);
+ }
+}
+
+/**
+ * xe_guc_exec_queue_snapshot_capture - Take a quick snapshot of the GuC Engine.
+ * @q: Xe exec queue.
+ *
+ * This can be printed out in a later stage like during dev_coredump
+ * analysis.
+ *
+ * Returns: a GuC Submit Engine snapshot object that must be freed by the
+ * caller, using `xe_guc_exec_queue_snapshot_free`.
+ */
+struct xe_guc_submit_exec_queue_snapshot *
+xe_guc_exec_queue_snapshot_capture(struct xe_exec_queue *q)
+{
+ struct xe_guc *guc = exec_queue_to_guc(q);
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gpu_scheduler *sched = &q->guc->sched;
+ struct xe_sched_job *job;
+ struct xe_guc_submit_exec_queue_snapshot *snapshot;
+ int i;
+
+ snapshot = kzalloc(sizeof(*snapshot), GFP_ATOMIC);
+
+ if (!snapshot) {
+ drm_err(&xe->drm, "Skipping GuC Engine snapshot entirely.\n");
+ return NULL;
+ }
+
+ snapshot->guc.id = q->guc->id;
+ memcpy(&snapshot->name, &q->name, sizeof(snapshot->name));
+ snapshot->class = q->class;
+ snapshot->logical_mask = q->logical_mask;
+ snapshot->width = q->width;
+ snapshot->refcount = kref_read(&q->refcount);
+ snapshot->sched_timeout = sched->base.timeout;
+ snapshot->sched_props.timeslice_us = q->sched_props.timeslice_us;
+ snapshot->sched_props.preempt_timeout_us =
+ q->sched_props.preempt_timeout_us;
+
+ snapshot->lrc = kmalloc_array(q->width, sizeof(struct lrc_snapshot),
+ GFP_ATOMIC);
+
+ if (!snapshot->lrc) {
+ drm_err(&xe->drm, "Skipping GuC Engine LRC snapshot.\n");
+ } else {
+ for (i = 0; i < q->width; ++i) {
+ struct xe_lrc *lrc = q->lrc + i;
+
+ snapshot->lrc[i].context_desc =
+ lower_32_bits(xe_lrc_ggtt_addr(lrc));
+ snapshot->lrc[i].head = xe_lrc_ring_head(lrc);
+ snapshot->lrc[i].tail.internal = lrc->ring.tail;
+ snapshot->lrc[i].tail.memory =
+ xe_lrc_read_ctx_reg(lrc, CTX_RING_TAIL);
+ snapshot->lrc[i].start_seqno = xe_lrc_start_seqno(lrc);
+ snapshot->lrc[i].seqno = xe_lrc_seqno(lrc);
+ }
+ }
+
+ snapshot->schedule_state = atomic_read(&q->guc->state);
+ snapshot->exec_queue_flags = q->flags;
+
+ snapshot->parallel_execution = xe_exec_queue_is_parallel(q);
+ if (snapshot->parallel_execution)
+ guc_exec_queue_wq_snapshot_capture(q, snapshot);
+
+ spin_lock(&sched->base.job_list_lock);
+ snapshot->pending_list_size = list_count_nodes(&sched->base.pending_list);
+ snapshot->pending_list = kmalloc_array(snapshot->pending_list_size,
+ sizeof(struct pending_list_snapshot),
+ GFP_ATOMIC);
+
+ if (!snapshot->pending_list) {
+ drm_err(&xe->drm, "Skipping GuC Engine pending_list snapshot.\n");
+ } else {
+ i = 0;
+ list_for_each_entry(job, &sched->base.pending_list, drm.list) {
+ snapshot->pending_list[i].seqno =
+ xe_sched_job_seqno(job);
+ snapshot->pending_list[i].fence =
+ dma_fence_is_signaled(job->fence) ? 1 : 0;
+ snapshot->pending_list[i].finished =
+ dma_fence_is_signaled(&job->drm.s_fence->finished)
+ ? 1 : 0;
+ i++;
+ }
+ }
+
+ spin_unlock(&sched->base.job_list_lock);
+
+ return snapshot;
+}
+
+/**
+ * xe_guc_exec_queue_snapshot_print - Print out a given GuC Engine snapshot.
+ * @snapshot: GuC Submit Engine snapshot object.
+ * @p: drm_printer where it will be printed out.
+ *
+ * This function prints out a given GuC Submit Engine snapshot object.
+ */
+void
+xe_guc_exec_queue_snapshot_print(struct xe_guc_submit_exec_queue_snapshot *snapshot,
+ struct drm_printer *p)
+{
+ int i;
+
+ if (!snapshot)
+ return;
+
+ drm_printf(p, "\nGuC ID: %d\n", snapshot->guc.id);
+ drm_printf(p, "\tName: %s\n", snapshot->name);
+ drm_printf(p, "\tClass: %d\n", snapshot->class);
+ drm_printf(p, "\tLogical mask: 0x%x\n", snapshot->logical_mask);
+ drm_printf(p, "\tWidth: %d\n", snapshot->width);
+ drm_printf(p, "\tRef: %d\n", snapshot->refcount);
+ drm_printf(p, "\tTimeout: %ld (ms)\n", snapshot->sched_timeout);
+ drm_printf(p, "\tTimeslice: %u (us)\n",
+ snapshot->sched_props.timeslice_us);
+ drm_printf(p, "\tPreempt timeout: %u (us)\n",
+ snapshot->sched_props.preempt_timeout_us);
+
+ for (i = 0; snapshot->lrc && i < snapshot->width; ++i) {
+ drm_printf(p, "\tHW Context Desc: 0x%08x\n",
+ snapshot->lrc[i].context_desc);
+ drm_printf(p, "\tLRC Head: (memory) %u\n",
+ snapshot->lrc[i].head);
+ drm_printf(p, "\tLRC Tail: (internal) %u, (memory) %u\n",
+ snapshot->lrc[i].tail.internal,
+ snapshot->lrc[i].tail.memory);
+ drm_printf(p, "\tStart seqno: (memory) %d\n",
+ snapshot->lrc[i].start_seqno);
+ drm_printf(p, "\tSeqno: (memory) %d\n", snapshot->lrc[i].seqno);
+ }
+ drm_printf(p, "\tSchedule State: 0x%x\n", snapshot->schedule_state);
+ drm_printf(p, "\tFlags: 0x%lx\n", snapshot->exec_queue_flags);
+
+ if (snapshot->parallel_execution)
+ guc_exec_queue_wq_snapshot_print(snapshot, p);
+
+ for (i = 0; snapshot->pending_list && i < snapshot->pending_list_size;
+ i++)
+ drm_printf(p, "\tJob: seqno=%d, fence=%d, finished=%d\n",
+ snapshot->pending_list[i].seqno,
+ snapshot->pending_list[i].fence,
+ snapshot->pending_list[i].finished);
+}
+
+/**
+ * xe_guc_exec_queue_snapshot_free - Free all allocated objects for a given
+ * snapshot.
+ * @snapshot: GuC Submit Engine snapshot object.
+ *
+ * This function free all the memory that needed to be allocated at capture
+ * time.
+ */
+void xe_guc_exec_queue_snapshot_free(struct xe_guc_submit_exec_queue_snapshot *snapshot)
+{
+ if (!snapshot)
+ return;
+
+ kfree(snapshot->lrc);
+ kfree(snapshot->pending_list);
+ kfree(snapshot);
+}
+
+static void guc_exec_queue_print(struct xe_exec_queue *q, struct drm_printer *p)
+{
+ struct xe_guc_submit_exec_queue_snapshot *snapshot;
+
+ snapshot = xe_guc_exec_queue_snapshot_capture(q);
+ xe_guc_exec_queue_snapshot_print(snapshot, p);
+ xe_guc_exec_queue_snapshot_free(snapshot);
+}
+
+/**
+ * xe_guc_submit_print - GuC Submit Print.
+ * @guc: GuC.
+ * @p: drm_printer where it will be printed out.
+ *
+ * This function capture and prints snapshots of **all** GuC Engines.
+ */
+void xe_guc_submit_print(struct xe_guc *guc, struct drm_printer *p)
+{
+ struct xe_exec_queue *q;
+ unsigned long index;
+
+ if (!xe_device_uc_enabled(guc_to_xe(guc)))
+ return;
+
+ mutex_lock(&guc->submission_state.lock);
+ xa_for_each(&guc->submission_state.exec_queue_lookup, index, q)
+ guc_exec_queue_print(q, p);
+ mutex_unlock(&guc->submission_state.lock);
+}
diff --git a/drivers/gpu/drm/xe/xe_guc_submit.h b/drivers/gpu/drm/xe/xe_guc_submit.h
new file mode 100644
index 000000000000..fc97869c5b86
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_submit.h
@@ -0,0 +1,38 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_SUBMIT_H_
+#define _XE_GUC_SUBMIT_H_
+
+#include <linux/types.h>
+
+struct drm_printer;
+struct xe_exec_queue;
+struct xe_guc;
+
+int xe_guc_submit_init(struct xe_guc *guc);
+
+int xe_guc_submit_reset_prepare(struct xe_guc *guc);
+void xe_guc_submit_reset_wait(struct xe_guc *guc);
+int xe_guc_submit_stop(struct xe_guc *guc);
+int xe_guc_submit_start(struct xe_guc *guc);
+
+int xe_guc_sched_done_handler(struct xe_guc *guc, u32 *msg, u32 len);
+int xe_guc_deregister_done_handler(struct xe_guc *guc, u32 *msg, u32 len);
+int xe_guc_exec_queue_reset_handler(struct xe_guc *guc, u32 *msg, u32 len);
+int xe_guc_exec_queue_memory_cat_error_handler(struct xe_guc *guc, u32 *msg,
+ u32 len);
+int xe_guc_exec_queue_reset_failure_handler(struct xe_guc *guc, u32 *msg, u32 len);
+
+struct xe_guc_submit_exec_queue_snapshot *
+xe_guc_exec_queue_snapshot_capture(struct xe_exec_queue *q);
+void
+xe_guc_exec_queue_snapshot_print(struct xe_guc_submit_exec_queue_snapshot *snapshot,
+ struct drm_printer *p);
+void
+xe_guc_exec_queue_snapshot_free(struct xe_guc_submit_exec_queue_snapshot *snapshot);
+void xe_guc_submit_print(struct xe_guc *guc, struct drm_printer *p);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_submit_types.h b/drivers/gpu/drm/xe/xe_guc_submit_types.h
new file mode 100644
index 000000000000..649b0a852692
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_submit_types.h
@@ -0,0 +1,155 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_GUC_SUBMIT_TYPES_H_
+#define _XE_GUC_SUBMIT_TYPES_H_
+
+#include "xe_hw_engine_types.h"
+
+/* Work item for submitting workloads into work queue of GuC. */
+#define WQ_STATUS_ACTIVE 1
+#define WQ_STATUS_SUSPENDED 2
+#define WQ_STATUS_CMD_ERROR 3
+#define WQ_STATUS_ENGINE_ID_NOT_USED 4
+#define WQ_STATUS_SUSPENDED_FROM_RESET 5
+#define WQ_TYPE_NOOP 0x4
+#define WQ_TYPE_MULTI_LRC 0x5
+#define WQ_TYPE_MASK GENMASK(7, 0)
+#define WQ_LEN_MASK GENMASK(26, 16)
+
+#define WQ_GUC_ID_MASK GENMASK(15, 0)
+#define WQ_RING_TAIL_MASK GENMASK(28, 18)
+
+#define PARALLEL_SCRATCH_SIZE 2048
+#define WQ_SIZE (PARALLEL_SCRATCH_SIZE / 2)
+#define WQ_OFFSET (PARALLEL_SCRATCH_SIZE - WQ_SIZE)
+#define CACHELINE_BYTES 64
+
+struct guc_sched_wq_desc {
+ u32 head;
+ u32 tail;
+ u32 error_offset;
+ u32 wq_status;
+ u32 reserved[28];
+} __packed;
+
+struct sync_semaphore {
+ u32 semaphore;
+ u8 unused[CACHELINE_BYTES - sizeof(u32)];
+};
+
+/**
+ * struct guc_submit_parallel_scratch - A scratch shared mapped buffer.
+ */
+struct guc_submit_parallel_scratch {
+ /** @wq_desc: Guc scheduler workqueue descriptor */
+ struct guc_sched_wq_desc wq_desc;
+
+ /** @go: Go Semaphore */
+ struct sync_semaphore go;
+ /** @join: Joined semaphore for the relevant hw engine instances */
+ struct sync_semaphore join[XE_HW_ENGINE_MAX_INSTANCE];
+
+ /** @unused: Unused/Reserved memory space */
+ u8 unused[WQ_OFFSET - sizeof(struct guc_sched_wq_desc) -
+ sizeof(struct sync_semaphore) *
+ (XE_HW_ENGINE_MAX_INSTANCE + 1)];
+
+ /** @wq: Workqueue info */
+ u32 wq[WQ_SIZE / sizeof(u32)];
+};
+
+struct lrc_snapshot {
+ u32 context_desc;
+ u32 head;
+ struct {
+ u32 internal;
+ u32 memory;
+ } tail;
+ u32 start_seqno;
+ u32 seqno;
+};
+
+struct pending_list_snapshot {
+ u32 seqno;
+ bool fence;
+ bool finished;
+};
+
+/**
+ * struct xe_guc_submit_exec_queue_snapshot - Snapshot for devcoredump
+ */
+struct xe_guc_submit_exec_queue_snapshot {
+ /** @name: name of this exec queue */
+ char name[MAX_FENCE_NAME_LEN];
+ /** @class: class of this exec queue */
+ enum xe_engine_class class;
+ /**
+ * @logical_mask: logical mask of where job submitted to exec queue can run
+ */
+ u32 logical_mask;
+ /** @width: width (number BB submitted per exec) of this exec queue */
+ u16 width;
+ /** @refcount: ref count of this exec queue */
+ u32 refcount;
+ /**
+ * @sched_timeout: the time after which a job is removed from the
+ * scheduler.
+ */
+ long sched_timeout;
+
+ /** @sched_props: scheduling properties */
+ struct {
+ /** @timeslice_us: timeslice period in micro-seconds */
+ u32 timeslice_us;
+ /** @preempt_timeout_us: preemption timeout in micro-seconds */
+ u32 preempt_timeout_us;
+ } sched_props;
+
+ /** @lrc: LRC Snapshot */
+ struct lrc_snapshot *lrc;
+
+ /** @schedule_state: Schedule State at the moment of Crash */
+ u32 schedule_state;
+ /** @exec_queue_flags: Flags of the faulty exec_queue */
+ unsigned long exec_queue_flags;
+
+ /** @guc: GuC Engine Snapshot */
+ struct {
+ /** @wqi_head: work queue item head */
+ u32 wqi_head;
+ /** @wqi_tail: work queue item tail */
+ u32 wqi_tail;
+ /** @id: GuC id for this exec_queue */
+ u16 id;
+ } guc;
+
+ /**
+ * @parallel_execution: Indication if the failure was during parallel
+ * execution
+ */
+ bool parallel_execution;
+ /** @parallel: snapshot of the useful parallel scratch */
+ struct {
+ /** @wq_desc: Workqueue description */
+ struct {
+ /** @head: Workqueue Head */
+ u32 head;
+ /** @tail: Workqueue Tail */
+ u32 tail;
+ /** @status: Workqueue Status */
+ u32 status;
+ } wq_desc;
+ /** @wq: Workqueue Items */
+ u32 wq[WQ_SIZE / sizeof(u32)];
+ } parallel;
+
+ /** @pending_list_size: Size of the pending list snapshot array */
+ int pending_list_size;
+ /** @pending_list: snapshot of the pending list info */
+ struct pending_list_snapshot *pending_list;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_types.h b/drivers/gpu/drm/xe/xe_guc_types.h
new file mode 100644
index 000000000000..cd80802e8918
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_guc_types.h
@@ -0,0 +1,81 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_GUC_TYPES_H_
+#define _XE_GUC_TYPES_H_
+
+#include <linux/idr.h>
+#include <linux/xarray.h>
+
+#include "regs/xe_reg_defs.h"
+#include "xe_guc_ads_types.h"
+#include "xe_guc_ct_types.h"
+#include "xe_guc_fwif.h"
+#include "xe_guc_log_types.h"
+#include "xe_guc_pc_types.h"
+#include "xe_uc_fw_types.h"
+
+/**
+ * struct xe_guc - Graphic micro controller
+ */
+struct xe_guc {
+ /** @fw: Generic uC firmware management */
+ struct xe_uc_fw fw;
+ /** @log: GuC log */
+ struct xe_guc_log log;
+ /** @ads: GuC ads */
+ struct xe_guc_ads ads;
+ /** @ct: GuC ct */
+ struct xe_guc_ct ct;
+ /** @pc: GuC Power Conservation */
+ struct xe_guc_pc pc;
+ /** @submission_state: GuC submission state */
+ struct {
+ /** @exec_queue_lookup: Lookup an xe_engine from guc_id */
+ struct xarray exec_queue_lookup;
+ /** @guc_ids: used to allocate new guc_ids, single-lrc */
+ struct ida guc_ids;
+ /** @guc_ids_bitmap: used to allocate new guc_ids, multi-lrc */
+ unsigned long *guc_ids_bitmap;
+ /** @stopped: submissions are stopped */
+ atomic_t stopped;
+ /** @lock: protects submission state */
+ struct mutex lock;
+ /** @suspend: suspend fence state */
+ struct {
+ /** @lock: suspend fences lock */
+ spinlock_t lock;
+ /** @context: suspend fences context */
+ u64 context;
+ /** @seqno: suspend fences seqno */
+ u32 seqno;
+ } suspend;
+#ifdef CONFIG_PROVE_LOCKING
+#define NUM_SUBMIT_WQ 256
+ /** @submit_wq_pool: submission ordered workqueues pool */
+ struct workqueue_struct *submit_wq_pool[NUM_SUBMIT_WQ];
+ /** @submit_wq_idx: submission ordered workqueue index */
+ int submit_wq_idx;
+#endif
+ /** @enabled: submission is enabled */
+ bool enabled;
+ } submission_state;
+ /** @hwconfig: Hardware config state */
+ struct {
+ /** @bo: buffer object of the hardware config */
+ struct xe_bo *bo;
+ /** @size: size of the hardware config */
+ u32 size;
+ } hwconfig;
+
+ /**
+ * @notify_reg: Register which is written to notify GuC of H2G messages
+ */
+ struct xe_reg notify_reg;
+ /** @params: Control params for fw initialization */
+ u32 params[GUC_CTL_MAX_DWORDS];
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_heci_gsc.c b/drivers/gpu/drm/xe/xe_heci_gsc.c
new file mode 100644
index 000000000000..bfdd33b9b23b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_heci_gsc.c
@@ -0,0 +1,234 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright(c) 2023, Intel Corporation. All rights reserved.
+ */
+
+#include <linux/irq.h>
+#include <linux/mei_aux.h>
+#include <linux/pci.h>
+#include <linux/sizes.h>
+
+#include "xe_device_types.h"
+#include "xe_drv.h"
+#include "xe_heci_gsc.h"
+#include "xe_platform_types.h"
+
+#define GSC_BAR_LENGTH 0x00000FFC
+
+#define DG1_GSC_HECI2_BASE 0x259000
+#define PVC_GSC_HECI2_BASE 0x285000
+#define DG2_GSC_HECI2_BASE 0x374000
+
+static void heci_gsc_irq_mask(struct irq_data *d)
+{
+ /* generic irq handling */
+}
+
+static void heci_gsc_irq_unmask(struct irq_data *d)
+{
+ /* generic irq handling */
+}
+
+static struct irq_chip heci_gsc_irq_chip = {
+ .name = "gsc_irq_chip",
+ .irq_mask = heci_gsc_irq_mask,
+ .irq_unmask = heci_gsc_irq_unmask,
+};
+
+static int heci_gsc_irq_init(int irq)
+{
+ irq_set_chip_and_handler_name(irq, &heci_gsc_irq_chip,
+ handle_simple_irq, "heci_gsc_irq_handler");
+
+ return irq_set_chip_data(irq, NULL);
+}
+
+/**
+ * struct heci_gsc_def - graphics security controller heci interface definitions
+ *
+ * @name: name of the heci device
+ * @bar: address of the mmio bar
+ * @bar_size: size of the mmio bar
+ * @use_polling: indication of using polling mode for the device
+ * @slow_firmware: indication of whether the device is slow (needs longer timeouts)
+ */
+struct heci_gsc_def {
+ const char *name;
+ unsigned long bar;
+ size_t bar_size;
+ bool use_polling;
+ bool slow_firmware;
+};
+
+/* gsc resources and definitions */
+static const struct heci_gsc_def heci_gsc_def_dg1 = {
+ .name = "mei-gscfi",
+ .bar = DG1_GSC_HECI2_BASE,
+ .bar_size = GSC_BAR_LENGTH,
+};
+
+static const struct heci_gsc_def heci_gsc_def_dg2 = {
+ .name = "mei-gscfi",
+ .bar = DG2_GSC_HECI2_BASE,
+ .bar_size = GSC_BAR_LENGTH,
+};
+
+static const struct heci_gsc_def heci_gsc_def_pvc = {
+ .name = "mei-gscfi",
+ .bar = PVC_GSC_HECI2_BASE,
+ .bar_size = GSC_BAR_LENGTH,
+ .slow_firmware = true,
+};
+
+static void heci_gsc_release_dev(struct device *dev)
+{
+ struct auxiliary_device *aux_dev = to_auxiliary_dev(dev);
+ struct mei_aux_device *adev = auxiliary_dev_to_mei_aux_dev(aux_dev);
+
+ kfree(adev);
+}
+
+void xe_heci_gsc_fini(struct xe_device *xe)
+{
+ struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
+
+ if (!HAS_HECI_GSCFI(xe))
+ return;
+
+ if (heci_gsc->adev) {
+ struct auxiliary_device *aux_dev = &heci_gsc->adev->aux_dev;
+
+ auxiliary_device_delete(aux_dev);
+ auxiliary_device_uninit(aux_dev);
+ heci_gsc->adev = NULL;
+ }
+
+ if (heci_gsc->irq >= 0)
+ irq_free_desc(heci_gsc->irq);
+ heci_gsc->irq = -1;
+}
+
+static int heci_gsc_irq_setup(struct xe_device *xe)
+{
+ struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
+ int ret;
+
+ heci_gsc->irq = irq_alloc_desc(0);
+ if (heci_gsc->irq < 0) {
+ drm_err(&xe->drm, "gsc irq error %d\n", heci_gsc->irq);
+ return heci_gsc->irq;
+ }
+
+ ret = heci_gsc_irq_init(heci_gsc->irq);
+ if (ret < 0)
+ drm_err(&xe->drm, "gsc irq init failed %d\n", ret);
+
+ return ret;
+}
+
+static int heci_gsc_add_device(struct xe_device *xe, const struct heci_gsc_def *def)
+{
+ struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ struct auxiliary_device *aux_dev;
+ struct mei_aux_device *adev;
+ int ret;
+
+ adev = kzalloc(sizeof(*adev), GFP_KERNEL);
+ if (!adev)
+ return -ENOMEM;
+ adev->irq = heci_gsc->irq;
+ adev->bar.parent = &pdev->resource[0];
+ adev->bar.start = def->bar + pdev->resource[0].start;
+ adev->bar.end = adev->bar.start + def->bar_size - 1;
+ adev->bar.flags = IORESOURCE_MEM;
+ adev->bar.desc = IORES_DESC_NONE;
+ adev->slow_firmware = def->slow_firmware;
+
+ aux_dev = &adev->aux_dev;
+ aux_dev->name = def->name;
+ aux_dev->id = (pci_domain_nr(pdev->bus) << 16) |
+ PCI_DEVID(pdev->bus->number, pdev->devfn);
+ aux_dev->dev.parent = &pdev->dev;
+ aux_dev->dev.release = heci_gsc_release_dev;
+
+ ret = auxiliary_device_init(aux_dev);
+ if (ret < 0) {
+ drm_err(&xe->drm, "gsc aux init failed %d\n", ret);
+ kfree(adev);
+ return ret;
+ }
+
+ heci_gsc->adev = adev; /* needed by the notifier */
+ ret = auxiliary_device_add(aux_dev);
+ if (ret < 0) {
+ drm_err(&xe->drm, "gsc aux add failed %d\n", ret);
+ heci_gsc->adev = NULL;
+
+ /* adev will be freed with the put_device() and .release sequence */
+ auxiliary_device_uninit(aux_dev);
+ }
+ return ret;
+}
+
+void xe_heci_gsc_init(struct xe_device *xe)
+{
+ struct xe_heci_gsc *heci_gsc = &xe->heci_gsc;
+ const struct heci_gsc_def *def;
+ int ret;
+
+ if (!HAS_HECI_GSCFI(xe))
+ return;
+
+ heci_gsc->irq = -1;
+
+ if (xe->info.platform == XE_PVC) {
+ def = &heci_gsc_def_pvc;
+ } else if (xe->info.platform == XE_DG2) {
+ def = &heci_gsc_def_dg2;
+ } else if (xe->info.platform == XE_DG1) {
+ def = &heci_gsc_def_dg1;
+ } else {
+ drm_warn_once(&xe->drm, "Unknown platform\n");
+ return;
+ }
+
+ if (!def->name) {
+ drm_warn_once(&xe->drm, "HECI is not implemented!\n");
+ return;
+ }
+
+ if (!def->use_polling) {
+ ret = heci_gsc_irq_setup(xe);
+ if (ret)
+ goto fail;
+ }
+
+ ret = heci_gsc_add_device(xe, def);
+ if (ret)
+ goto fail;
+
+ return;
+fail:
+ xe_heci_gsc_fini(xe);
+}
+
+void xe_heci_gsc_irq_handler(struct xe_device *xe, u32 iir)
+{
+ int ret;
+
+ if ((iir & GSC_IRQ_INTF(1)) == 0)
+ return;
+
+ if (!HAS_HECI_GSCFI(xe)) {
+ drm_warn_once(&xe->drm, "GSC irq: not supported");
+ return;
+ }
+
+ if (xe->heci_gsc.irq < 0)
+ return;
+
+ ret = generic_handle_irq(xe->heci_gsc.irq);
+ if (ret)
+ drm_err_ratelimited(&xe->drm, "error handling GSC irq: %d\n", ret);
+}
diff --git a/drivers/gpu/drm/xe/xe_heci_gsc.h b/drivers/gpu/drm/xe/xe_heci_gsc.h
new file mode 100644
index 000000000000..9db454478fae
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_heci_gsc.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright(c) 2023, Intel Corporation. All rights reserved.
+ */
+#ifndef __XE_HECI_GSC_DEV_H__
+#define __XE_HECI_GSC_DEV_H__
+
+#include <linux/types.h>
+
+struct xe_device;
+struct mei_aux_device;
+
+/*
+ * The HECI1 bit corresponds to bit15 and HECI2 to bit14.
+ * The reason for this is to allow growth for more interfaces in the future.
+ */
+#define GSC_IRQ_INTF(_x) BIT(15 - (_x))
+
+/**
+ * struct xe_heci_gsc - graphics security controller for xe, HECI interface
+ *
+ * @adev : pointer to mei auxiliary device structure
+ * @irq : irq number
+ *
+ */
+struct xe_heci_gsc {
+ struct mei_aux_device *adev;
+ int irq;
+};
+
+void xe_heci_gsc_init(struct xe_device *xe);
+void xe_heci_gsc_fini(struct xe_device *xe);
+void xe_heci_gsc_irq_handler(struct xe_device *xe, u32 iir);
+
+#endif /* __XE_HECI_GSC_DEV_H__ */
diff --git a/drivers/gpu/drm/xe/xe_huc.c b/drivers/gpu/drm/xe/xe_huc.c
new file mode 100644
index 000000000000..eca109791c6a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_huc.c
@@ -0,0 +1,307 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_huc.h"
+
+#include <drm/drm_managed.h>
+
+#include "abi/gsc_pxp_commands_abi.h"
+#include "regs/xe_gsc_regs.h"
+#include "regs/xe_guc_regs.h"
+#include "xe_assert.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_force_wake.h"
+#include "xe_gsc_submit.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_uc_fw.h"
+
+static struct xe_gt *
+huc_to_gt(struct xe_huc *huc)
+{
+ return container_of(huc, struct xe_gt, uc.huc);
+}
+
+static struct xe_device *
+huc_to_xe(struct xe_huc *huc)
+{
+ return gt_to_xe(huc_to_gt(huc));
+}
+
+static struct xe_guc *
+huc_to_guc(struct xe_huc *huc)
+{
+ return &container_of(huc, struct xe_uc, huc)->guc;
+}
+
+static void free_gsc_pkt(struct drm_device *drm, void *arg)
+{
+ struct xe_huc *huc = arg;
+
+ xe_bo_unpin_map_no_vm(huc->gsc_pkt);
+ huc->gsc_pkt = NULL;
+}
+
+#define PXP43_HUC_AUTH_INOUT_SIZE SZ_4K
+static int huc_alloc_gsc_pkt(struct xe_huc *huc)
+{
+ struct xe_gt *gt = huc_to_gt(huc);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_bo *bo;
+ int err;
+
+ /* we use a single object for both input and output */
+ bo = xe_bo_create_pin_map(xe, gt_to_tile(gt), NULL,
+ PXP43_HUC_AUTH_INOUT_SIZE * 2,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_SYSTEM_BIT |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ huc->gsc_pkt = bo;
+
+ err = drmm_add_action_or_reset(&xe->drm, free_gsc_pkt, huc);
+ if (err) {
+ free_gsc_pkt(&xe->drm, huc);
+ return err;
+ }
+
+ return 0;
+}
+
+int xe_huc_init(struct xe_huc *huc)
+{
+ struct xe_gt *gt = huc_to_gt(huc);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ int ret;
+
+ huc->fw.type = XE_UC_FW_TYPE_HUC;
+
+ /* On platforms with a media GT the HuC is only available there */
+ if (tile->media_gt && (gt != tile->media_gt)) {
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_NOT_SUPPORTED);
+ return 0;
+ }
+
+ ret = xe_uc_fw_init(&huc->fw);
+ if (ret)
+ goto out;
+
+ if (!xe_uc_fw_is_enabled(&huc->fw))
+ return 0;
+
+ if (huc->fw.has_gsc_headers) {
+ ret = huc_alloc_gsc_pkt(huc);
+ if (ret)
+ goto out;
+ }
+
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_LOADABLE);
+
+ return 0;
+
+out:
+ drm_err(&xe->drm, "HuC init failed with %d", ret);
+ return ret;
+}
+
+int xe_huc_upload(struct xe_huc *huc)
+{
+ if (!xe_uc_fw_is_loadable(&huc->fw))
+ return 0;
+ return xe_uc_fw_upload(&huc->fw, 0, HUC_UKERNEL);
+}
+
+#define huc_auth_msg_wr(xe_, map_, offset_, field_, val_) \
+ xe_map_wr_field(xe_, map_, offset_, struct pxp43_new_huc_auth_in, field_, val_)
+#define huc_auth_msg_rd(xe_, map_, offset_, field_) \
+ xe_map_rd_field(xe_, map_, offset_, struct pxp43_huc_auth_out, field_)
+
+static u32 huc_emit_pxp_auth_msg(struct xe_device *xe, struct iosys_map *map,
+ u32 wr_offset, u32 huc_offset, u32 huc_size)
+{
+ xe_map_memset(xe, map, wr_offset, 0, sizeof(struct pxp43_new_huc_auth_in));
+
+ huc_auth_msg_wr(xe, map, wr_offset, header.api_version, PXP_APIVER(4, 3));
+ huc_auth_msg_wr(xe, map, wr_offset, header.command_id, PXP43_CMDID_NEW_HUC_AUTH);
+ huc_auth_msg_wr(xe, map, wr_offset, header.status, 0);
+ huc_auth_msg_wr(xe, map, wr_offset, header.buffer_len,
+ sizeof(struct pxp43_new_huc_auth_in) - sizeof(struct pxp_cmd_header));
+ huc_auth_msg_wr(xe, map, wr_offset, huc_base_address, huc_offset);
+ huc_auth_msg_wr(xe, map, wr_offset, huc_size, huc_size);
+
+ return wr_offset + sizeof(struct pxp43_new_huc_auth_in);
+}
+
+static int huc_auth_via_gsccs(struct xe_huc *huc)
+{
+ struct xe_gt *gt = huc_to_gt(huc);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_bo *pkt = huc->gsc_pkt;
+ u32 wr_offset;
+ u32 rd_offset;
+ u64 ggtt_offset;
+ u32 out_status;
+ int retry = 5;
+ int err = 0;
+
+ if (!pkt)
+ return -ENODEV;
+
+ ggtt_offset = xe_bo_ggtt_addr(pkt);
+
+ wr_offset = xe_gsc_emit_header(xe, &pkt->vmap, 0, HECI_MEADDRESS_PXP, 0,
+ sizeof(struct pxp43_new_huc_auth_in));
+ wr_offset = huc_emit_pxp_auth_msg(xe, &pkt->vmap, wr_offset,
+ xe_bo_ggtt_addr(huc->fw.bo),
+ huc->fw.bo->size);
+ do {
+ err = xe_gsc_pkt_submit_kernel(&gt->uc.gsc, ggtt_offset, wr_offset,
+ ggtt_offset + PXP43_HUC_AUTH_INOUT_SIZE,
+ PXP43_HUC_AUTH_INOUT_SIZE);
+ if (err)
+ break;
+
+ if (xe_gsc_check_and_update_pending(xe, &pkt->vmap, 0, &pkt->vmap,
+ PXP43_HUC_AUTH_INOUT_SIZE)) {
+ err = -EBUSY;
+ msleep(50);
+ }
+ } while (--retry && err == -EBUSY);
+
+ if (err) {
+ drm_err(&xe->drm, "failed to submit GSC request to auth: %d\n", err);
+ return err;
+ }
+
+ err = xe_gsc_read_out_header(xe, &pkt->vmap, PXP43_HUC_AUTH_INOUT_SIZE,
+ sizeof(struct pxp43_huc_auth_out), &rd_offset);
+ if (err) {
+ drm_err(&xe->drm, "HuC: invalid GSC reply for auth (err=%d)\n", err);
+ return err;
+ }
+
+ /*
+ * The GSC will return PXP_STATUS_OP_NOT_PERMITTED if the HuC is already
+ * authenticated. If the same error is ever returned with HuC not loaded
+ * we'll still catch it when we check the authentication bit later.
+ */
+ out_status = huc_auth_msg_rd(xe, &pkt->vmap, rd_offset, header.status);
+ if (out_status != PXP_STATUS_SUCCESS && out_status != PXP_STATUS_OP_NOT_PERMITTED) {
+ drm_err(&xe->drm, "auth failed with GSC error = 0x%x\n", out_status);
+ return -EIO;
+ }
+
+ return 0;
+}
+
+static const struct {
+ const char *name;
+ struct xe_reg reg;
+ u32 val;
+} huc_auth_modes[XE_HUC_AUTH_TYPES_COUNT] = {
+ [XE_HUC_AUTH_VIA_GUC] = { "GuC",
+ HUC_KERNEL_LOAD_INFO,
+ HUC_LOAD_SUCCESSFUL },
+ [XE_HUC_AUTH_VIA_GSC] = { "GSC",
+ HECI_FWSTS5(MTL_GSC_HECI1_BASE),
+ HECI1_FWSTS5_HUC_AUTH_DONE },
+};
+
+bool xe_huc_is_authenticated(struct xe_huc *huc, enum xe_huc_auth_types type)
+{
+ struct xe_gt *gt = huc_to_gt(huc);
+
+ return xe_mmio_read32(gt, huc_auth_modes[type].reg) & huc_auth_modes[type].val;
+}
+
+int xe_huc_auth(struct xe_huc *huc, enum xe_huc_auth_types type)
+{
+ struct xe_device *xe = huc_to_xe(huc);
+ struct xe_gt *gt = huc_to_gt(huc);
+ struct xe_guc *guc = huc_to_guc(huc);
+ int ret;
+
+ if (!xe_uc_fw_is_loadable(&huc->fw))
+ return 0;
+
+ /* On newer platforms the HuC survives reset, so no need to re-auth */
+ if (xe_huc_is_authenticated(huc, type)) {
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_RUNNING);
+ return 0;
+ }
+
+ if (!xe_uc_fw_is_loaded(&huc->fw))
+ return -ENOEXEC;
+
+ switch (type) {
+ case XE_HUC_AUTH_VIA_GUC:
+ ret = xe_guc_auth_huc(guc, xe_bo_ggtt_addr(huc->fw.bo) +
+ xe_uc_fw_rsa_offset(&huc->fw));
+ break;
+ case XE_HUC_AUTH_VIA_GSC:
+ ret = huc_auth_via_gsccs(huc);
+ break;
+ default:
+ XE_WARN_ON(type);
+ return -EINVAL;
+ }
+ if (ret) {
+ drm_err(&xe->drm, "Failed to trigger HuC auth via %s: %d\n",
+ huc_auth_modes[type].name, ret);
+ goto fail;
+ }
+
+ ret = xe_mmio_wait32(gt, huc_auth_modes[type].reg, huc_auth_modes[type].val,
+ huc_auth_modes[type].val, 100000, NULL, false);
+ if (ret) {
+ drm_err(&xe->drm, "HuC: Firmware not verified %d\n", ret);
+ goto fail;
+ }
+
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_RUNNING);
+ drm_dbg(&xe->drm, "HuC authenticated via %s\n", huc_auth_modes[type].name);
+
+ return 0;
+
+fail:
+ drm_err(&xe->drm, "HuC: Auth via %s failed: %d\n",
+ huc_auth_modes[type].name, ret);
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_LOAD_FAIL);
+
+ return ret;
+}
+
+void xe_huc_sanitize(struct xe_huc *huc)
+{
+ if (!xe_uc_fw_is_loadable(&huc->fw))
+ return;
+ xe_uc_fw_change_status(&huc->fw, XE_UC_FIRMWARE_LOADABLE);
+}
+
+void xe_huc_print_info(struct xe_huc *huc, struct drm_printer *p)
+{
+ struct xe_gt *gt = huc_to_gt(huc);
+ int err;
+
+ xe_uc_fw_print(&huc->fw, p);
+
+ if (!xe_uc_fw_is_enabled(&huc->fw))
+ return;
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return;
+
+ drm_printf(p, "\nHuC status: 0x%08x\n",
+ xe_mmio_read32(gt, HUC_KERNEL_LOAD_INFO));
+
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+}
diff --git a/drivers/gpu/drm/xe/xe_huc.h b/drivers/gpu/drm/xe/xe_huc.h
new file mode 100644
index 000000000000..532017230287
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_huc.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_HUC_H_
+#define _XE_HUC_H_
+
+#include "xe_huc_types.h"
+
+struct drm_printer;
+
+enum xe_huc_auth_types {
+ XE_HUC_AUTH_VIA_GUC = 0,
+ XE_HUC_AUTH_VIA_GSC,
+ XE_HUC_AUTH_TYPES_COUNT
+};
+
+int xe_huc_init(struct xe_huc *huc);
+int xe_huc_upload(struct xe_huc *huc);
+int xe_huc_auth(struct xe_huc *huc, enum xe_huc_auth_types type);
+bool xe_huc_is_authenticated(struct xe_huc *huc, enum xe_huc_auth_types type);
+void xe_huc_sanitize(struct xe_huc *huc);
+void xe_huc_print_info(struct xe_huc *huc, struct drm_printer *p);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_huc_debugfs.c b/drivers/gpu/drm/xe/xe_huc_debugfs.c
new file mode 100644
index 000000000000..18585a7eeb9d
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_huc_debugfs.c
@@ -0,0 +1,70 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_huc_debugfs.h"
+
+#include <drm/drm_debugfs.h>
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_huc.h"
+#include "xe_macros.h"
+
+static struct xe_gt *
+huc_to_gt(struct xe_huc *huc)
+{
+ return container_of(huc, struct xe_gt, uc.huc);
+}
+
+static struct xe_device *
+huc_to_xe(struct xe_huc *huc)
+{
+ return gt_to_xe(huc_to_gt(huc));
+}
+
+static struct xe_huc *node_to_huc(struct drm_info_node *node)
+{
+ return node->info_ent->data;
+}
+
+static int huc_info(struct seq_file *m, void *data)
+{
+ struct xe_huc *huc = node_to_huc(m->private);
+ struct xe_device *xe = huc_to_xe(huc);
+ struct drm_printer p = drm_seq_file_printer(m);
+
+ xe_device_mem_access_get(xe);
+ xe_huc_print_info(huc, &p);
+ xe_device_mem_access_put(xe);
+
+ return 0;
+}
+
+static const struct drm_info_list debugfs_list[] = {
+ {"huc_info", huc_info, 0},
+};
+
+void xe_huc_debugfs_register(struct xe_huc *huc, struct dentry *parent)
+{
+ struct drm_minor *minor = huc_to_xe(huc)->drm.primary;
+ struct drm_info_list *local;
+ int i;
+
+#define DEBUGFS_SIZE (ARRAY_SIZE(debugfs_list) * sizeof(struct drm_info_list))
+ local = drmm_kmalloc(&huc_to_xe(huc)->drm, DEBUGFS_SIZE, GFP_KERNEL);
+ if (!local)
+ return;
+
+ memcpy(local, debugfs_list, DEBUGFS_SIZE);
+#undef DEBUGFS_SIZE
+
+ for (i = 0; i < ARRAY_SIZE(debugfs_list); ++i)
+ local[i].data = huc;
+
+ drm_debugfs_create_files(local,
+ ARRAY_SIZE(debugfs_list),
+ parent, minor);
+}
diff --git a/drivers/gpu/drm/xe/xe_huc_debugfs.h b/drivers/gpu/drm/xe/xe_huc_debugfs.h
new file mode 100644
index 000000000000..ec58f1818804
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_huc_debugfs.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_HUC_DEBUGFS_H_
+#define _XE_HUC_DEBUGFS_H_
+
+struct dentry;
+struct xe_huc;
+
+void xe_huc_debugfs_register(struct xe_huc *huc, struct dentry *parent);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_huc_types.h b/drivers/gpu/drm/xe/xe_huc_types.h
new file mode 100644
index 000000000000..cfbaa5e0dfca
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_huc_types.h
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_HUC_TYPES_H_
+#define _XE_HUC_TYPES_H_
+
+#include "xe_uc_fw_types.h"
+
+struct xe_bo;
+
+/**
+ * struct xe_huc - HuC
+ */
+struct xe_huc {
+ /** @fw: Generic uC firmware management */
+ struct xe_uc_fw fw;
+
+ /** @gsc_pkt: bo to store the packet for auth via GSC */
+ struct xe_bo *gsc_pkt;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hw_engine.c b/drivers/gpu/drm/xe/xe_hw_engine.c
new file mode 100644
index 000000000000..1fa5cf5eea97
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_engine.c
@@ -0,0 +1,883 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_hw_engine.h"
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "xe_assert.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_execlist.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_gt_ccs_mode.h"
+#include "xe_gt_topology.h"
+#include "xe_hw_fence.h"
+#include "xe_irq.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+#include "xe_reg_sr.h"
+#include "xe_rtp.h"
+#include "xe_sched_job.h"
+#include "xe_tuning.h"
+#include "xe_uc_fw.h"
+#include "xe_wa.h"
+
+#define MAX_MMIO_BASES 3
+struct engine_info {
+ const char *name;
+ unsigned int class : 8;
+ unsigned int instance : 8;
+ enum xe_force_wake_domains domain;
+ u32 mmio_base;
+};
+
+static const struct engine_info engine_infos[] = {
+ [XE_HW_ENGINE_RCS0] = {
+ .name = "rcs0",
+ .class = XE_ENGINE_CLASS_RENDER,
+ .instance = 0,
+ .domain = XE_FW_RENDER,
+ .mmio_base = RENDER_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS0] = {
+ .name = "bcs0",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 0,
+ .domain = XE_FW_RENDER,
+ .mmio_base = BLT_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS1] = {
+ .name = "bcs1",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 1,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS1_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS2] = {
+ .name = "bcs2",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 2,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS2_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS3] = {
+ .name = "bcs3",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 3,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS3_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS4] = {
+ .name = "bcs4",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 4,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS4_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS5] = {
+ .name = "bcs5",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 5,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS5_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS6] = {
+ .name = "bcs6",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 6,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS6_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS7] = {
+ .name = "bcs7",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 7,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS7_RING_BASE,
+ },
+ [XE_HW_ENGINE_BCS8] = {
+ .name = "bcs8",
+ .class = XE_ENGINE_CLASS_COPY,
+ .instance = 8,
+ .domain = XE_FW_RENDER,
+ .mmio_base = XEHPC_BCS8_RING_BASE,
+ },
+
+ [XE_HW_ENGINE_VCS0] = {
+ .name = "vcs0",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 0,
+ .domain = XE_FW_MEDIA_VDBOX0,
+ .mmio_base = BSD_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS1] = {
+ .name = "vcs1",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 1,
+ .domain = XE_FW_MEDIA_VDBOX1,
+ .mmio_base = BSD2_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS2] = {
+ .name = "vcs2",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 2,
+ .domain = XE_FW_MEDIA_VDBOX2,
+ .mmio_base = BSD3_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS3] = {
+ .name = "vcs3",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 3,
+ .domain = XE_FW_MEDIA_VDBOX3,
+ .mmio_base = BSD4_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS4] = {
+ .name = "vcs4",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 4,
+ .domain = XE_FW_MEDIA_VDBOX4,
+ .mmio_base = XEHP_BSD5_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS5] = {
+ .name = "vcs5",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 5,
+ .domain = XE_FW_MEDIA_VDBOX5,
+ .mmio_base = XEHP_BSD6_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS6] = {
+ .name = "vcs6",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 6,
+ .domain = XE_FW_MEDIA_VDBOX6,
+ .mmio_base = XEHP_BSD7_RING_BASE,
+ },
+ [XE_HW_ENGINE_VCS7] = {
+ .name = "vcs7",
+ .class = XE_ENGINE_CLASS_VIDEO_DECODE,
+ .instance = 7,
+ .domain = XE_FW_MEDIA_VDBOX7,
+ .mmio_base = XEHP_BSD8_RING_BASE,
+ },
+ [XE_HW_ENGINE_VECS0] = {
+ .name = "vecs0",
+ .class = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ .instance = 0,
+ .domain = XE_FW_MEDIA_VEBOX0,
+ .mmio_base = VEBOX_RING_BASE,
+ },
+ [XE_HW_ENGINE_VECS1] = {
+ .name = "vecs1",
+ .class = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ .instance = 1,
+ .domain = XE_FW_MEDIA_VEBOX1,
+ .mmio_base = VEBOX2_RING_BASE,
+ },
+ [XE_HW_ENGINE_VECS2] = {
+ .name = "vecs2",
+ .class = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ .instance = 2,
+ .domain = XE_FW_MEDIA_VEBOX2,
+ .mmio_base = XEHP_VEBOX3_RING_BASE,
+ },
+ [XE_HW_ENGINE_VECS3] = {
+ .name = "vecs3",
+ .class = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ .instance = 3,
+ .domain = XE_FW_MEDIA_VEBOX3,
+ .mmio_base = XEHP_VEBOX4_RING_BASE,
+ },
+ [XE_HW_ENGINE_CCS0] = {
+ .name = "ccs0",
+ .class = XE_ENGINE_CLASS_COMPUTE,
+ .instance = 0,
+ .domain = XE_FW_RENDER,
+ .mmio_base = COMPUTE0_RING_BASE,
+ },
+ [XE_HW_ENGINE_CCS1] = {
+ .name = "ccs1",
+ .class = XE_ENGINE_CLASS_COMPUTE,
+ .instance = 1,
+ .domain = XE_FW_RENDER,
+ .mmio_base = COMPUTE1_RING_BASE,
+ },
+ [XE_HW_ENGINE_CCS2] = {
+ .name = "ccs2",
+ .class = XE_ENGINE_CLASS_COMPUTE,
+ .instance = 2,
+ .domain = XE_FW_RENDER,
+ .mmio_base = COMPUTE2_RING_BASE,
+ },
+ [XE_HW_ENGINE_CCS3] = {
+ .name = "ccs3",
+ .class = XE_ENGINE_CLASS_COMPUTE,
+ .instance = 3,
+ .domain = XE_FW_RENDER,
+ .mmio_base = COMPUTE3_RING_BASE,
+ },
+ [XE_HW_ENGINE_GSCCS0] = {
+ .name = "gsccs0",
+ .class = XE_ENGINE_CLASS_OTHER,
+ .instance = OTHER_GSC_INSTANCE,
+ .domain = XE_FW_GSC,
+ .mmio_base = GSCCS_RING_BASE,
+ },
+};
+
+static void hw_engine_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_hw_engine *hwe = arg;
+
+ if (hwe->exl_port)
+ xe_execlist_port_destroy(hwe->exl_port);
+ xe_lrc_finish(&hwe->kernel_lrc);
+
+ hwe->gt = NULL;
+}
+
+static void hw_engine_mmio_write32(struct xe_hw_engine *hwe, struct xe_reg reg,
+ u32 val)
+{
+ xe_gt_assert(hwe->gt, !(reg.addr & hwe->mmio_base));
+ xe_force_wake_assert_held(gt_to_fw(hwe->gt), hwe->domain);
+
+ reg.addr += hwe->mmio_base;
+
+ xe_mmio_write32(hwe->gt, reg, val);
+}
+
+static u32 hw_engine_mmio_read32(struct xe_hw_engine *hwe, struct xe_reg reg)
+{
+ xe_gt_assert(hwe->gt, !(reg.addr & hwe->mmio_base));
+ xe_force_wake_assert_held(gt_to_fw(hwe->gt), hwe->domain);
+
+ reg.addr += hwe->mmio_base;
+
+ return xe_mmio_read32(hwe->gt, reg);
+}
+
+void xe_hw_engine_enable_ring(struct xe_hw_engine *hwe)
+{
+ u32 ccs_mask =
+ xe_hw_engine_mask_per_class(hwe->gt, XE_ENGINE_CLASS_COMPUTE);
+
+ if (hwe->class == XE_ENGINE_CLASS_COMPUTE && ccs_mask)
+ xe_mmio_write32(hwe->gt, RCU_MODE,
+ _MASKED_BIT_ENABLE(RCU_MODE_CCS_ENABLE));
+
+ hw_engine_mmio_write32(hwe, RING_HWSTAM(0), ~0x0);
+ hw_engine_mmio_write32(hwe, RING_HWS_PGA(0),
+ xe_bo_ggtt_addr(hwe->hwsp));
+ hw_engine_mmio_write32(hwe, RING_MODE(0),
+ _MASKED_BIT_ENABLE(GFX_DISABLE_LEGACY_MODE));
+ hw_engine_mmio_write32(hwe, RING_MI_MODE(0),
+ _MASKED_BIT_DISABLE(STOP_RING));
+ hw_engine_mmio_read32(hwe, RING_MI_MODE(0));
+}
+
+static bool xe_hw_engine_match_fixed_cslice_mode(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe)
+{
+ return xe_gt_ccs_mode_enabled(gt) &&
+ xe_rtp_match_first_render_or_compute(gt, hwe);
+}
+
+void
+xe_hw_engine_setup_default_lrc_state(struct xe_hw_engine *hwe)
+{
+ struct xe_gt *gt = hwe->gt;
+ const u8 mocs_write_idx = gt->mocs.uc_index;
+ const u8 mocs_read_idx = gt->mocs.uc_index;
+ u32 blit_cctl_val = REG_FIELD_PREP(BLIT_CCTL_DST_MOCS_MASK, mocs_write_idx) |
+ REG_FIELD_PREP(BLIT_CCTL_SRC_MOCS_MASK, mocs_read_idx);
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+ const struct xe_rtp_entry_sr lrc_was[] = {
+ /*
+ * Some blitter commands do not have a field for MOCS, those
+ * commands will use MOCS index pointed by BLIT_CCTL.
+ * BLIT_CCTL registers are needed to be programmed to un-cached.
+ */
+ { XE_RTP_NAME("BLIT_CCTL_default_MOCS"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, XE_RTP_END_VERSION_UNDEFINED),
+ ENGINE_CLASS(COPY)),
+ XE_RTP_ACTIONS(FIELD_SET(BLIT_CCTL(0),
+ BLIT_CCTL_DST_MOCS_MASK |
+ BLIT_CCTL_SRC_MOCS_MASK,
+ blit_cctl_val,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+ /* Use Fixed slice CCS mode */
+ { XE_RTP_NAME("RCU_MODE_FIXED_SLICE_CCS_MODE"),
+ XE_RTP_RULES(FUNC(xe_hw_engine_match_fixed_cslice_mode)),
+ XE_RTP_ACTIONS(FIELD_SET(RCU_MODE, RCU_MODE_FIXED_SLICE_CCS_MODE,
+ RCU_MODE_FIXED_SLICE_CCS_MODE))
+ },
+ {}
+ };
+
+ xe_rtp_process_to_sr(&ctx, lrc_was, &hwe->reg_lrc);
+}
+
+static void
+hw_engine_setup_default_state(struct xe_hw_engine *hwe)
+{
+ struct xe_gt *gt = hwe->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ /*
+ * RING_CMD_CCTL specifies the default MOCS entry that will be
+ * used by the command streamer when executing commands that
+ * don't have a way to explicitly specify a MOCS setting.
+ * The default should usually reference whichever MOCS entry
+ * corresponds to uncached behavior, although use of a WB cached
+ * entry is recommended by the spec in certain circumstances on
+ * specific platforms.
+ * Bspec: 72161
+ */
+ const u8 mocs_write_idx = gt->mocs.uc_index;
+ const u8 mocs_read_idx = hwe->class == XE_ENGINE_CLASS_COMPUTE &&
+ (GRAPHICS_VER(xe) >= 20 || xe->info.platform == XE_PVC) ?
+ gt->mocs.wb_index : gt->mocs.uc_index;
+ u32 ring_cmd_cctl_val = REG_FIELD_PREP(CMD_CCTL_WRITE_OVERRIDE_MASK, mocs_write_idx) |
+ REG_FIELD_PREP(CMD_CCTL_READ_OVERRIDE_MASK, mocs_read_idx);
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+ const struct xe_rtp_entry_sr engine_entries[] = {
+ { XE_RTP_NAME("RING_CMD_CCTL_default_MOCS"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, XE_RTP_END_VERSION_UNDEFINED)),
+ XE_RTP_ACTIONS(FIELD_SET(RING_CMD_CCTL(0),
+ CMD_CCTL_WRITE_OVERRIDE_MASK |
+ CMD_CCTL_READ_OVERRIDE_MASK,
+ ring_cmd_cctl_val,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+ /*
+ * To allow the GSC engine to go idle on MTL we need to enable
+ * idle messaging and set the hysteresis value (we use 0xA=5us
+ * as recommended in spec). On platforms after MTL this is
+ * enabled by default.
+ */
+ { XE_RTP_NAME("MTL GSCCS IDLE MSG enable"),
+ XE_RTP_RULES(MEDIA_VERSION(1300), ENGINE_CLASS(OTHER)),
+ XE_RTP_ACTIONS(CLR(RING_PSMI_CTL(0),
+ IDLE_MSG_DISABLE,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)),
+ FIELD_SET(RING_PWRCTX_MAXCNT(0),
+ IDLE_WAIT_TIME,
+ 0xA,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+ {}
+ };
+
+ xe_rtp_process_to_sr(&ctx, engine_entries, &hwe->reg_sr);
+}
+
+static void hw_engine_init_early(struct xe_gt *gt, struct xe_hw_engine *hwe,
+ enum xe_hw_engine_id id)
+{
+ const struct engine_info *info;
+
+ if (WARN_ON(id >= ARRAY_SIZE(engine_infos) || !engine_infos[id].name))
+ return;
+
+ if (!(gt->info.engine_mask & BIT(id)))
+ return;
+
+ info = &engine_infos[id];
+
+ xe_gt_assert(gt, !hwe->gt);
+
+ hwe->gt = gt;
+ hwe->class = info->class;
+ hwe->instance = info->instance;
+ hwe->mmio_base = info->mmio_base;
+ hwe->domain = info->domain;
+ hwe->name = info->name;
+ hwe->fence_irq = &gt->fence_irq[info->class];
+ hwe->engine_id = id;
+
+ hwe->eclass = &gt->eclass[hwe->class];
+ if (!hwe->eclass->sched_props.job_timeout_ms) {
+ hwe->eclass->sched_props.job_timeout_ms = 5 * 1000;
+ hwe->eclass->sched_props.job_timeout_min = XE_HW_ENGINE_JOB_TIMEOUT_MIN;
+ hwe->eclass->sched_props.job_timeout_max = XE_HW_ENGINE_JOB_TIMEOUT_MAX;
+ hwe->eclass->sched_props.timeslice_us = 1 * 1000;
+ hwe->eclass->sched_props.timeslice_min = XE_HW_ENGINE_TIMESLICE_MIN;
+ hwe->eclass->sched_props.timeslice_max = XE_HW_ENGINE_TIMESLICE_MAX;
+ hwe->eclass->sched_props.preempt_timeout_us = XE_HW_ENGINE_PREEMPT_TIMEOUT;
+ hwe->eclass->sched_props.preempt_timeout_min = XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN;
+ hwe->eclass->sched_props.preempt_timeout_max = XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX;
+ /* Record default props */
+ hwe->eclass->defaults = hwe->eclass->sched_props;
+ }
+
+ xe_reg_sr_init(&hwe->reg_sr, hwe->name, gt_to_xe(gt));
+ xe_tuning_process_engine(hwe);
+ xe_wa_process_engine(hwe);
+ hw_engine_setup_default_state(hwe);
+
+ xe_reg_sr_init(&hwe->reg_whitelist, hwe->name, gt_to_xe(gt));
+ xe_reg_whitelist_process_engine(hwe);
+}
+
+static int hw_engine_init(struct xe_gt *gt, struct xe_hw_engine *hwe,
+ enum xe_hw_engine_id id)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_tile *tile = gt_to_tile(gt);
+ int err;
+
+ xe_gt_assert(gt, id < ARRAY_SIZE(engine_infos) && engine_infos[id].name);
+ xe_gt_assert(gt, gt->info.engine_mask & BIT(id));
+
+ xe_reg_sr_apply_mmio(&hwe->reg_sr, gt);
+ xe_reg_sr_apply_whitelist(hwe);
+
+ hwe->hwsp = xe_managed_bo_create_pin_map(xe, tile, SZ_4K,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(hwe->hwsp)) {
+ err = PTR_ERR(hwe->hwsp);
+ goto err_name;
+ }
+
+ err = xe_lrc_init(&hwe->kernel_lrc, hwe, NULL, NULL, SZ_16K);
+ if (err)
+ goto err_hwsp;
+
+ if (!xe_device_uc_enabled(xe)) {
+ hwe->exl_port = xe_execlist_port_create(xe, hwe);
+ if (IS_ERR(hwe->exl_port)) {
+ err = PTR_ERR(hwe->exl_port);
+ goto err_kernel_lrc;
+ }
+ }
+
+ if (xe_device_uc_enabled(xe))
+ xe_hw_engine_enable_ring(hwe);
+
+ /* We reserve the highest BCS instance for USM */
+ if (xe->info.has_usm && hwe->class == XE_ENGINE_CLASS_COPY)
+ gt->usm.reserved_bcs_instance = hwe->instance;
+
+ err = drmm_add_action_or_reset(&xe->drm, hw_engine_fini, hwe);
+ if (err)
+ return err;
+
+ return 0;
+
+err_kernel_lrc:
+ xe_lrc_finish(&hwe->kernel_lrc);
+err_hwsp:
+ xe_bo_unpin_map_no_vm(hwe->hwsp);
+err_name:
+ hwe->name = NULL;
+
+ return err;
+}
+
+static void hw_engine_setup_logical_mapping(struct xe_gt *gt)
+{
+ int class;
+
+ /* FIXME: Doing a simple logical mapping that works for most hardware */
+ for (class = 0; class < XE_ENGINE_CLASS_MAX; ++class) {
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ int logical_instance = 0;
+
+ for_each_hw_engine(hwe, gt, id)
+ if (hwe->class == class)
+ hwe->logical_instance = logical_instance++;
+ }
+}
+
+static void read_media_fuses(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 media_fuse;
+ u16 vdbox_mask;
+ u16 vebox_mask;
+ int i, j;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ media_fuse = xe_mmio_read32(gt, GT_VEBOX_VDBOX_DISABLE);
+
+ /*
+ * Pre-Xe_HP platforms had register bits representing absent engines,
+ * whereas Xe_HP and beyond have bits representing present engines.
+ * Invert the polarity on old platforms so that we can use common
+ * handling below.
+ */
+ if (GRAPHICS_VERx100(xe) < 1250)
+ media_fuse = ~media_fuse;
+
+ vdbox_mask = REG_FIELD_GET(GT_VDBOX_DISABLE_MASK, media_fuse);
+ vebox_mask = REG_FIELD_GET(GT_VEBOX_DISABLE_MASK, media_fuse);
+
+ for (i = XE_HW_ENGINE_VCS0, j = 0; i <= XE_HW_ENGINE_VCS7; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ if (!(BIT(j) & vdbox_mask)) {
+ gt->info.engine_mask &= ~BIT(i);
+ drm_info(&xe->drm, "vcs%u fused off\n", j);
+ }
+ }
+
+ for (i = XE_HW_ENGINE_VECS0, j = 0; i <= XE_HW_ENGINE_VECS3; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ if (!(BIT(j) & vebox_mask)) {
+ gt->info.engine_mask &= ~BIT(i);
+ drm_info(&xe->drm, "vecs%u fused off\n", j);
+ }
+ }
+}
+
+static void read_copy_fuses(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 bcs_mask;
+
+ if (GRAPHICS_VERx100(xe) < 1260 || GRAPHICS_VERx100(xe) >= 1270)
+ return;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ bcs_mask = xe_mmio_read32(gt, MIRROR_FUSE3);
+ bcs_mask = REG_FIELD_GET(MEML3_EN_MASK, bcs_mask);
+
+ /* BCS0 is always present; only BCS1-BCS8 may be fused off */
+ for (int i = XE_HW_ENGINE_BCS1, j = 0; i <= XE_HW_ENGINE_BCS8; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ if (!(BIT(j / 2) & bcs_mask)) {
+ gt->info.engine_mask &= ~BIT(i);
+ drm_info(&xe->drm, "bcs%u fused off\n", j);
+ }
+ }
+}
+
+static void read_compute_fuses_from_dss(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ /*
+ * CCS fusing based on DSS masks only applies to platforms that can
+ * have more than one CCS.
+ */
+ if (hweight64(gt->info.engine_mask &
+ GENMASK_ULL(XE_HW_ENGINE_CCS3, XE_HW_ENGINE_CCS0)) <= 1)
+ return;
+
+ /*
+ * CCS availability on Xe_HP is inferred from the presence of DSS in
+ * each quadrant.
+ */
+ for (int i = XE_HW_ENGINE_CCS0, j = 0; i <= XE_HW_ENGINE_CCS3; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ if (!xe_gt_topology_has_dss_in_quadrant(gt, j)) {
+ gt->info.engine_mask &= ~BIT(i);
+ drm_info(&xe->drm, "ccs%u fused off\n", j);
+ }
+ }
+}
+
+static void read_compute_fuses_from_reg(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 ccs_mask;
+
+ ccs_mask = xe_mmio_read32(gt, XEHP_FUSE4);
+ ccs_mask = REG_FIELD_GET(CCS_EN_MASK, ccs_mask);
+
+ for (int i = XE_HW_ENGINE_CCS0, j = 0; i <= XE_HW_ENGINE_CCS3; ++i, ++j) {
+ if (!(gt->info.engine_mask & BIT(i)))
+ continue;
+
+ if ((ccs_mask & BIT(j)) == 0) {
+ gt->info.engine_mask &= ~BIT(i);
+ drm_info(&xe->drm, "ccs%u fused off\n", j);
+ }
+ }
+}
+
+static void read_compute_fuses(struct xe_gt *gt)
+{
+ if (GRAPHICS_VER(gt_to_xe(gt)) >= 20)
+ read_compute_fuses_from_reg(gt);
+ else
+ read_compute_fuses_from_dss(gt);
+}
+
+static void check_gsc_availability(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (!(gt->info.engine_mask & BIT(XE_HW_ENGINE_GSCCS0)))
+ return;
+
+ /*
+ * The GSCCS is only used to communicate with the GSC FW, so if we don't
+ * have the FW there is nothing we need the engine for and can therefore
+ * skip its initialization.
+ */
+ if (!xe_uc_fw_is_available(&gt->uc.gsc.fw)) {
+ gt->info.engine_mask &= ~BIT(XE_HW_ENGINE_GSCCS0);
+ drm_info(&xe->drm, "gsccs disabled due to lack of FW\n");
+ }
+}
+
+int xe_hw_engines_init_early(struct xe_gt *gt)
+{
+ int i;
+
+ read_media_fuses(gt);
+ read_copy_fuses(gt);
+ read_compute_fuses(gt);
+ check_gsc_availability(gt);
+
+ BUILD_BUG_ON(XE_HW_ENGINE_PREEMPT_TIMEOUT < XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN);
+ BUILD_BUG_ON(XE_HW_ENGINE_PREEMPT_TIMEOUT > XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX);
+
+ for (i = 0; i < ARRAY_SIZE(gt->hw_engines); i++)
+ hw_engine_init_early(gt, &gt->hw_engines[i], i);
+
+ return 0;
+}
+
+int xe_hw_engines_init(struct xe_gt *gt)
+{
+ int err;
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(hwe, gt, id) {
+ err = hw_engine_init(gt, hwe, id);
+ if (err)
+ return err;
+ }
+
+ hw_engine_setup_logical_mapping(gt);
+
+ return 0;
+}
+
+void xe_hw_engine_handle_irq(struct xe_hw_engine *hwe, u16 intr_vec)
+{
+ wake_up_all(&gt_to_xe(hwe->gt)->ufence_wq);
+
+ if (hwe->irq_handler)
+ hwe->irq_handler(hwe, intr_vec);
+
+ if (intr_vec & GT_RENDER_USER_INTERRUPT)
+ xe_hw_fence_irq_run(hwe->fence_irq);
+}
+
+/**
+ * xe_hw_engine_snapshot_capture - Take a quick snapshot of the HW Engine.
+ * @hwe: Xe HW Engine.
+ *
+ * This can be printed out in a later stage like during dev_coredump
+ * analysis.
+ *
+ * Returns: a Xe HW Engine snapshot object that must be freed by the
+ * caller, using `xe_hw_engine_snapshot_free`.
+ */
+struct xe_hw_engine_snapshot *
+xe_hw_engine_snapshot_capture(struct xe_hw_engine *hwe)
+{
+ struct xe_hw_engine_snapshot *snapshot;
+ int len;
+
+ if (!xe_hw_engine_is_valid(hwe))
+ return NULL;
+
+ snapshot = kzalloc(sizeof(*snapshot), GFP_ATOMIC);
+
+ if (!snapshot)
+ return NULL;
+
+ len = strlen(hwe->name) + 1;
+ snapshot->name = kzalloc(len, GFP_ATOMIC);
+ if (snapshot->name)
+ strscpy(snapshot->name, hwe->name, len);
+
+ snapshot->class = hwe->class;
+ snapshot->logical_instance = hwe->logical_instance;
+ snapshot->forcewake.domain = hwe->domain;
+ snapshot->forcewake.ref = xe_force_wake_ref(gt_to_fw(hwe->gt),
+ hwe->domain);
+ snapshot->mmio_base = hwe->mmio_base;
+
+ snapshot->reg.ring_hwstam = hw_engine_mmio_read32(hwe, RING_HWSTAM(0));
+ snapshot->reg.ring_hws_pga = hw_engine_mmio_read32(hwe,
+ RING_HWS_PGA(0));
+ snapshot->reg.ring_execlist_status_lo =
+ hw_engine_mmio_read32(hwe, RING_EXECLIST_STATUS_LO(0));
+ snapshot->reg.ring_execlist_status_hi =
+ hw_engine_mmio_read32(hwe, RING_EXECLIST_STATUS_HI(0));
+ snapshot->reg.ring_execlist_sq_contents_lo =
+ hw_engine_mmio_read32(hwe,
+ RING_EXECLIST_SQ_CONTENTS_LO(0));
+ snapshot->reg.ring_execlist_sq_contents_hi =
+ hw_engine_mmio_read32(hwe,
+ RING_EXECLIST_SQ_CONTENTS_HI(0));
+ snapshot->reg.ring_start = hw_engine_mmio_read32(hwe, RING_START(0));
+ snapshot->reg.ring_head =
+ hw_engine_mmio_read32(hwe, RING_HEAD(0)) & HEAD_ADDR;
+ snapshot->reg.ring_tail =
+ hw_engine_mmio_read32(hwe, RING_TAIL(0)) & TAIL_ADDR;
+ snapshot->reg.ring_ctl = hw_engine_mmio_read32(hwe, RING_CTL(0));
+ snapshot->reg.ring_mi_mode =
+ hw_engine_mmio_read32(hwe, RING_MI_MODE(0));
+ snapshot->reg.ring_mode = hw_engine_mmio_read32(hwe, RING_MODE(0));
+ snapshot->reg.ring_imr = hw_engine_mmio_read32(hwe, RING_IMR(0));
+ snapshot->reg.ring_esr = hw_engine_mmio_read32(hwe, RING_ESR(0));
+ snapshot->reg.ring_emr = hw_engine_mmio_read32(hwe, RING_EMR(0));
+ snapshot->reg.ring_eir = hw_engine_mmio_read32(hwe, RING_EIR(0));
+ snapshot->reg.ring_acthd_udw =
+ hw_engine_mmio_read32(hwe, RING_ACTHD_UDW(0));
+ snapshot->reg.ring_acthd = hw_engine_mmio_read32(hwe, RING_ACTHD(0));
+ snapshot->reg.ring_bbaddr_udw =
+ hw_engine_mmio_read32(hwe, RING_BBADDR_UDW(0));
+ snapshot->reg.ring_bbaddr = hw_engine_mmio_read32(hwe, RING_BBADDR(0));
+ snapshot->reg.ring_dma_fadd_udw =
+ hw_engine_mmio_read32(hwe, RING_DMA_FADD_UDW(0));
+ snapshot->reg.ring_dma_fadd =
+ hw_engine_mmio_read32(hwe, RING_DMA_FADD(0));
+ snapshot->reg.ipehr = hw_engine_mmio_read32(hwe, RING_IPEHR(0));
+
+ if (snapshot->class == XE_ENGINE_CLASS_COMPUTE)
+ snapshot->reg.rcu_mode = xe_mmio_read32(hwe->gt, RCU_MODE);
+
+ return snapshot;
+}
+
+/**
+ * xe_hw_engine_snapshot_print - Print out a given Xe HW Engine snapshot.
+ * @snapshot: Xe HW Engine snapshot object.
+ * @p: drm_printer where it will be printed out.
+ *
+ * This function prints out a given Xe HW Engine snapshot object.
+ */
+void xe_hw_engine_snapshot_print(struct xe_hw_engine_snapshot *snapshot,
+ struct drm_printer *p)
+{
+ if (!snapshot)
+ return;
+
+ drm_printf(p, "%s (physical), logical instance=%d\n",
+ snapshot->name ? snapshot->name : "",
+ snapshot->logical_instance);
+ drm_printf(p, "\tForcewake: domain 0x%x, ref %d\n",
+ snapshot->forcewake.domain, snapshot->forcewake.ref);
+ drm_printf(p, "\tHWSTAM: 0x%08x\n", snapshot->reg.ring_hwstam);
+ drm_printf(p, "\tRING_HWS_PGA: 0x%08x\n", snapshot->reg.ring_hws_pga);
+ drm_printf(p, "\tRING_EXECLIST_STATUS_LO: 0x%08x\n",
+ snapshot->reg.ring_execlist_status_lo);
+ drm_printf(p, "\tRING_EXECLIST_STATUS_HI: 0x%08x\n",
+ snapshot->reg.ring_execlist_status_hi);
+ drm_printf(p, "\tRING_EXECLIST_SQ_CONTENTS_LO: 0x%08x\n",
+ snapshot->reg.ring_execlist_sq_contents_lo);
+ drm_printf(p, "\tRING_EXECLIST_SQ_CONTENTS_HI: 0x%08x\n",
+ snapshot->reg.ring_execlist_sq_contents_hi);
+ drm_printf(p, "\tRING_START: 0x%08x\n", snapshot->reg.ring_start);
+ drm_printf(p, "\tRING_HEAD: 0x%08x\n", snapshot->reg.ring_head);
+ drm_printf(p, "\tRING_TAIL: 0x%08x\n", snapshot->reg.ring_tail);
+ drm_printf(p, "\tRING_CTL: 0x%08x\n", snapshot->reg.ring_ctl);
+ drm_printf(p, "\tRING_MI_MODE: 0x%08x\n", snapshot->reg.ring_mi_mode);
+ drm_printf(p, "\tRING_MODE: 0x%08x\n",
+ snapshot->reg.ring_mode);
+ drm_printf(p, "\tRING_IMR: 0x%08x\n", snapshot->reg.ring_imr);
+ drm_printf(p, "\tRING_ESR: 0x%08x\n", snapshot->reg.ring_esr);
+ drm_printf(p, "\tRING_EMR: 0x%08x\n", snapshot->reg.ring_emr);
+ drm_printf(p, "\tRING_EIR: 0x%08x\n", snapshot->reg.ring_eir);
+ drm_printf(p, "\tACTHD: 0x%08x_%08x\n", snapshot->reg.ring_acthd_udw,
+ snapshot->reg.ring_acthd);
+ drm_printf(p, "\tBBADDR: 0x%08x_%08x\n", snapshot->reg.ring_bbaddr_udw,
+ snapshot->reg.ring_bbaddr);
+ drm_printf(p, "\tDMA_FADDR: 0x%08x_%08x\n",
+ snapshot->reg.ring_dma_fadd_udw,
+ snapshot->reg.ring_dma_fadd);
+ drm_printf(p, "\tIPEHR: 0x%08x\n\n", snapshot->reg.ipehr);
+ if (snapshot->class == XE_ENGINE_CLASS_COMPUTE)
+ drm_printf(p, "\tRCU_MODE: 0x%08x\n",
+ snapshot->reg.rcu_mode);
+}
+
+/**
+ * xe_hw_engine_snapshot_free - Free all allocated objects for a given snapshot.
+ * @snapshot: Xe HW Engine snapshot object.
+ *
+ * This function free all the memory that needed to be allocated at capture
+ * time.
+ */
+void xe_hw_engine_snapshot_free(struct xe_hw_engine_snapshot *snapshot)
+{
+ if (!snapshot)
+ return;
+
+ kfree(snapshot->name);
+ kfree(snapshot);
+}
+
+/**
+ * xe_hw_engine_print - Xe HW Engine Print.
+ * @hwe: Hardware Engine.
+ * @p: drm_printer.
+ *
+ * This function quickly capture a snapshot and immediately print it out.
+ */
+void xe_hw_engine_print(struct xe_hw_engine *hwe, struct drm_printer *p)
+{
+ struct xe_hw_engine_snapshot *snapshot;
+
+ snapshot = xe_hw_engine_snapshot_capture(hwe);
+ xe_hw_engine_snapshot_print(snapshot, p);
+ xe_hw_engine_snapshot_free(snapshot);
+}
+
+u32 xe_hw_engine_mask_per_class(struct xe_gt *gt,
+ enum xe_engine_class engine_class)
+{
+ u32 mask = 0;
+ enum xe_hw_engine_id id;
+
+ for (id = 0; id < XE_NUM_HW_ENGINES; ++id) {
+ if (engine_infos[id].class == engine_class &&
+ gt->info.engine_mask & BIT(id))
+ mask |= BIT(engine_infos[id].instance);
+ }
+ return mask;
+}
+
+bool xe_hw_engine_is_reserved(struct xe_hw_engine *hwe)
+{
+ struct xe_gt *gt = hwe->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (hwe->class == XE_ENGINE_CLASS_OTHER)
+ return true;
+
+ /* Check for engines disabled by ccs_mode setting */
+ if (xe_gt_ccs_mode_enabled(gt) &&
+ hwe->class == XE_ENGINE_CLASS_COMPUTE &&
+ hwe->logical_instance >= gt->ccs_mode)
+ return true;
+
+ return xe->info.has_usm && hwe->class == XE_ENGINE_CLASS_COPY &&
+ hwe->instance == gt->usm.reserved_bcs_instance;
+}
diff --git a/drivers/gpu/drm/xe/xe_hw_engine.h b/drivers/gpu/drm/xe/xe_hw_engine.h
new file mode 100644
index 000000000000..71968ee2f600
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_engine.h
@@ -0,0 +1,70 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_HW_ENGINE_H_
+#define _XE_HW_ENGINE_H_
+
+#include "xe_hw_engine_types.h"
+
+struct drm_printer;
+
+#ifdef CONFIG_DRM_XE_JOB_TIMEOUT_MIN
+#define XE_HW_ENGINE_JOB_TIMEOUT_MIN CONFIG_DRM_XE_JOB_TIMEOUT_MIN
+#else
+#define XE_HW_ENGINE_JOB_TIMEOUT_MIN 1
+#endif
+#ifdef CONFIG_DRM_XE_JOB_TIMEOUT_MAX
+#define XE_HW_ENGINE_JOB_TIMEOUT_MAX CONFIG_DRM_XE_JOB_TIMEOUT_MAX
+#else
+#define XE_HW_ENGINE_JOB_TIMEOUT_MAX (10 * 1000)
+#endif
+#ifdef CONFIG_DRM_XE_TIMESLICE_MIN
+#define XE_HW_ENGINE_TIMESLICE_MIN CONFIG_DRM_XE_TIMESLICE_MIN
+#else
+#define XE_HW_ENGINE_TIMESLICE_MIN 1
+#endif
+#ifdef CONFIG_DRM_XE_TIMESLICE_MAX
+#define XE_HW_ENGINE_TIMESLICE_MAX CONFIG_DRM_XE_TIMESLICE_MAX
+#else
+#define XE_HW_ENGINE_TIMESLICE_MAX (10 * 1000 * 1000)
+#endif
+#ifdef CONFIG_DRM_XE_PREEMPT_TIMEOUT
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT CONFIG_DRM_XE_PREEMPT_TIMEOUT
+#else
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT (640 * 1000)
+#endif
+#ifdef CONFIG_DRM_XE_PREEMPT_TIMEOUT_MIN
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN CONFIG_DRM_XE_PREEMPT_TIMEOUT_MIN
+#else
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN 1
+#endif
+#ifdef CONFIG_DRM_XE_PREEMPT_TIMEOUT_MAX
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX CONFIG_DRM_XE_PREEMPT_TIMEOUT_MAX
+#else
+#define XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX (10 * 1000 * 1000)
+#endif
+
+int xe_hw_engines_init_early(struct xe_gt *gt);
+int xe_hw_engines_init(struct xe_gt *gt);
+void xe_hw_engine_handle_irq(struct xe_hw_engine *hwe, u16 intr_vec);
+void xe_hw_engine_enable_ring(struct xe_hw_engine *hwe);
+u32 xe_hw_engine_mask_per_class(struct xe_gt *gt,
+ enum xe_engine_class engine_class);
+
+struct xe_hw_engine_snapshot *
+xe_hw_engine_snapshot_capture(struct xe_hw_engine *hwe);
+void xe_hw_engine_snapshot_free(struct xe_hw_engine_snapshot *snapshot);
+void xe_hw_engine_snapshot_print(struct xe_hw_engine_snapshot *snapshot,
+ struct drm_printer *p);
+void xe_hw_engine_print(struct xe_hw_engine *hwe, struct drm_printer *p);
+void xe_hw_engine_setup_default_lrc_state(struct xe_hw_engine *hwe);
+
+bool xe_hw_engine_is_reserved(struct xe_hw_engine *hwe);
+static inline bool xe_hw_engine_is_valid(struct xe_hw_engine *hwe)
+{
+ return hwe->name;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.c b/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.c
new file mode 100644
index 000000000000..e49bc14f0ecf
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.c
@@ -0,0 +1,675 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_managed.h>
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+
+#include "xe_gt.h"
+#include "xe_hw_engine_class_sysfs.h"
+
+#define MAX_ENGINE_CLASS_NAME_LEN 16
+static int xe_add_hw_engine_class_defaults(struct xe_device *xe,
+ struct kobject *parent);
+
+/**
+ * xe_hw_engine_timeout_in_range - Helper to check if timeout is in range
+ * @timeout: timeout to validate
+ * @min: min value of valid range
+ * @max: max value of valid range
+ *
+ * This helper helps to validate if timeout is in min-max range of HW engine
+ * scheduler.
+ *
+ * Returns: Returns false value for failure and true for success.
+ */
+bool xe_hw_engine_timeout_in_range(u64 timeout, u64 min, u64 max)
+{
+ return timeout >= min && timeout <= max;
+}
+
+static void kobj_xe_hw_engine_release(struct kobject *kobj)
+{
+ kfree(kobj);
+}
+
+static const struct kobj_type kobj_xe_hw_engine_type = {
+ .release = kobj_xe_hw_engine_release,
+ .sysfs_ops = &kobj_sysfs_ops
+};
+
+static ssize_t job_timeout_max_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (timeout < eclass->sched_props.job_timeout_min)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(timeout,
+ XE_HW_ENGINE_JOB_TIMEOUT_MIN,
+ XE_HW_ENGINE_JOB_TIMEOUT_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.job_timeout_max, timeout);
+
+ return count;
+}
+
+static ssize_t job_timeout_max_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.job_timeout_max);
+}
+
+static struct kobj_attribute job_timeout_max_attr =
+__ATTR(job_timeout_max, 0644, job_timeout_max_show, job_timeout_max_store);
+
+static ssize_t job_timeout_min_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (timeout > eclass->sched_props.job_timeout_max)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(timeout,
+ XE_HW_ENGINE_JOB_TIMEOUT_MIN,
+ XE_HW_ENGINE_JOB_TIMEOUT_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.job_timeout_min, timeout);
+
+ return count;
+}
+
+static ssize_t job_timeout_min_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.job_timeout_min);
+}
+
+static struct kobj_attribute job_timeout_min_attr =
+__ATTR(job_timeout_min, 0644, job_timeout_min_show, job_timeout_min_store);
+
+static ssize_t job_timeout_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 min = eclass->sched_props.job_timeout_min;
+ u32 max = eclass->sched_props.job_timeout_max;
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (!xe_hw_engine_timeout_in_range(timeout, min, max))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.job_timeout_ms, timeout);
+
+ return count;
+}
+
+static ssize_t job_timeout_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.job_timeout_ms);
+}
+
+static struct kobj_attribute job_timeout_attr =
+__ATTR(job_timeout_ms, 0644, job_timeout_show, job_timeout_store);
+
+static ssize_t job_timeout_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.job_timeout_ms);
+}
+
+static struct kobj_attribute job_timeout_def =
+__ATTR(job_timeout_ms, 0444, job_timeout_default, NULL);
+
+static ssize_t job_timeout_min_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.job_timeout_min);
+}
+
+static struct kobj_attribute job_timeout_min_def =
+__ATTR(job_timeout_min, 0444, job_timeout_min_default, NULL);
+
+static ssize_t job_timeout_max_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.job_timeout_max);
+}
+
+static struct kobj_attribute job_timeout_max_def =
+__ATTR(job_timeout_max, 0444, job_timeout_max_default, NULL);
+
+static ssize_t timeslice_duration_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 min = eclass->sched_props.timeslice_min;
+ u32 max = eclass->sched_props.timeslice_max;
+ u32 duration;
+ int err;
+
+ err = kstrtou32(buf, 0, &duration);
+ if (err)
+ return err;
+
+ if (!xe_hw_engine_timeout_in_range(duration, min, max))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.timeslice_us, duration);
+
+ return count;
+}
+
+static ssize_t timeslice_duration_max_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 duration;
+ int err;
+
+ err = kstrtou32(buf, 0, &duration);
+ if (err)
+ return err;
+
+ if (duration < eclass->sched_props.timeslice_min)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(duration,
+ XE_HW_ENGINE_TIMESLICE_MIN,
+ XE_HW_ENGINE_TIMESLICE_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.timeslice_max, duration);
+
+ return count;
+}
+
+static ssize_t timeslice_duration_max_show(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.timeslice_max);
+}
+
+static struct kobj_attribute timeslice_duration_max_attr =
+ __ATTR(timeslice_duration_max, 0644, timeslice_duration_max_show,
+ timeslice_duration_max_store);
+
+static ssize_t timeslice_duration_min_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 duration;
+ int err;
+
+ err = kstrtou32(buf, 0, &duration);
+ if (err)
+ return err;
+
+ if (duration > eclass->sched_props.timeslice_max)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(duration,
+ XE_HW_ENGINE_TIMESLICE_MIN,
+ XE_HW_ENGINE_TIMESLICE_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.timeslice_min, duration);
+
+ return count;
+}
+
+static ssize_t timeslice_duration_min_show(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.timeslice_min);
+}
+
+static struct kobj_attribute timeslice_duration_min_attr =
+ __ATTR(timeslice_duration_min, 0644, timeslice_duration_min_show,
+ timeslice_duration_min_store);
+
+static ssize_t timeslice_duration_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.timeslice_us);
+}
+
+static struct kobj_attribute timeslice_duration_attr =
+ __ATTR(timeslice_duration_us, 0644, timeslice_duration_show,
+ timeslice_duration_store);
+
+static ssize_t timeslice_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.timeslice_us);
+}
+
+static struct kobj_attribute timeslice_duration_def =
+__ATTR(timeslice_duration_us, 0444, timeslice_default, NULL);
+
+static ssize_t timeslice_min_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.timeslice_min);
+}
+
+static struct kobj_attribute timeslice_duration_min_def =
+__ATTR(timeslice_duration_min, 0444, timeslice_min_default, NULL);
+
+static ssize_t timeslice_max_default(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.timeslice_max);
+}
+
+static struct kobj_attribute timeslice_duration_max_def =
+__ATTR(timeslice_duration_max, 0444, timeslice_max_default, NULL);
+
+static ssize_t preempt_timeout_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 min = eclass->sched_props.preempt_timeout_min;
+ u32 max = eclass->sched_props.preempt_timeout_max;
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (!xe_hw_engine_timeout_in_range(timeout, min, max))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.preempt_timeout_us, timeout);
+
+ return count;
+}
+
+static ssize_t preempt_timeout_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.preempt_timeout_us);
+}
+
+static struct kobj_attribute preempt_timeout_attr =
+__ATTR(preempt_timeout_us, 0644, preempt_timeout_show, preempt_timeout_store);
+
+static ssize_t preempt_timeout_default(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.preempt_timeout_us);
+}
+
+static struct kobj_attribute preempt_timeout_def =
+__ATTR(preempt_timeout_us, 0444, preempt_timeout_default, NULL);
+
+static ssize_t preempt_timeout_min_default(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.preempt_timeout_min);
+}
+
+static struct kobj_attribute preempt_timeout_min_def =
+__ATTR(preempt_timeout_min, 0444, preempt_timeout_min_default, NULL);
+
+static ssize_t preempt_timeout_max_default(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj->parent);
+
+ return sprintf(buf, "%u\n", eclass->defaults.preempt_timeout_max);
+}
+
+static struct kobj_attribute preempt_timeout_max_def =
+__ATTR(preempt_timeout_max, 0444, preempt_timeout_max_default, NULL);
+
+static ssize_t preempt_timeout_max_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (timeout < eclass->sched_props.preempt_timeout_min)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(timeout,
+ XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN,
+ XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.preempt_timeout_max, timeout);
+
+ return count;
+}
+
+static ssize_t preempt_timeout_max_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.preempt_timeout_max);
+}
+
+static struct kobj_attribute preempt_timeout_max_attr =
+ __ATTR(preempt_timeout_max, 0644, preempt_timeout_max_show,
+ preempt_timeout_max_store);
+
+static ssize_t preempt_timeout_min_store(struct kobject *kobj,
+ struct kobj_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+ u32 timeout;
+ int err;
+
+ err = kstrtou32(buf, 0, &timeout);
+ if (err)
+ return err;
+
+ if (timeout > eclass->sched_props.preempt_timeout_max)
+ return -EINVAL;
+
+ if (!xe_hw_engine_timeout_in_range(timeout,
+ XE_HW_ENGINE_PREEMPT_TIMEOUT_MIN,
+ XE_HW_ENGINE_PREEMPT_TIMEOUT_MAX))
+ return -EINVAL;
+
+ WRITE_ONCE(eclass->sched_props.preempt_timeout_min, timeout);
+
+ return count;
+}
+
+static ssize_t preempt_timeout_min_show(struct kobject *kobj,
+ struct kobj_attribute *attr, char *buf)
+{
+ struct xe_hw_engine_class_intf *eclass = kobj_to_eclass(kobj);
+
+ return sprintf(buf, "%u\n", eclass->sched_props.preempt_timeout_min);
+}
+
+static struct kobj_attribute preempt_timeout_min_attr =
+ __ATTR(preempt_timeout_min, 0644, preempt_timeout_min_show,
+ preempt_timeout_min_store);
+
+static const struct attribute *defaults[] = {
+ &job_timeout_def.attr,
+ &job_timeout_min_def.attr,
+ &job_timeout_max_def.attr,
+ &timeslice_duration_def.attr,
+ &timeslice_duration_min_def.attr,
+ &timeslice_duration_max_def.attr,
+ &preempt_timeout_def.attr,
+ &preempt_timeout_min_def.attr,
+ &preempt_timeout_max_def.attr,
+ NULL
+};
+
+static const struct attribute *files[] = {
+ &job_timeout_attr.attr,
+ &job_timeout_min_attr.attr,
+ &job_timeout_max_attr.attr,
+ &timeslice_duration_attr.attr,
+ &timeslice_duration_min_attr.attr,
+ &timeslice_duration_max_attr.attr,
+ &preempt_timeout_attr.attr,
+ &preempt_timeout_min_attr.attr,
+ &preempt_timeout_max_attr.attr,
+ NULL
+};
+
+static void kobj_xe_hw_engine_class_fini(struct drm_device *drm, void *arg)
+{
+ struct kobject *kobj = arg;
+
+ sysfs_remove_files(kobj, files);
+ kobject_put(kobj);
+}
+
+ static struct kobj_eclass *
+kobj_xe_hw_engine_class(struct xe_device *xe, struct kobject *parent, char *name)
+{
+ struct kobj_eclass *keclass;
+ int err = 0;
+
+ keclass = kzalloc(sizeof(*keclass), GFP_KERNEL);
+ if (!keclass)
+ return NULL;
+
+ kobject_init(&keclass->base, &kobj_xe_hw_engine_type);
+ if (kobject_add(&keclass->base, parent, "%s", name)) {
+ kobject_put(&keclass->base);
+ return NULL;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, kobj_xe_hw_engine_class_fini,
+ &keclass->base);
+ if (err)
+ drm_warn(&xe->drm,
+ "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+ return keclass;
+}
+
+static void hw_engine_class_defaults_fini(struct drm_device *drm, void *arg)
+{
+ struct kobject *kobj = arg;
+
+ sysfs_remove_files(kobj, defaults);
+ kobject_put(kobj);
+}
+
+static int xe_add_hw_engine_class_defaults(struct xe_device *xe,
+ struct kobject *parent)
+{
+ struct kobject *kobj;
+ int err = 0;
+
+ kobj = kzalloc(sizeof(*kobj), GFP_KERNEL);
+ if (!kobj)
+ return -ENOMEM;
+
+ kobject_init(kobj, &kobj_xe_hw_engine_type);
+ err = kobject_add(kobj, parent, "%s", ".defaults");
+ if (err)
+ goto err_object;
+
+ err = sysfs_create_files(kobj, defaults);
+ if (err)
+ goto err_object;
+
+ err = drmm_add_action_or_reset(&xe->drm, hw_engine_class_defaults_fini,
+ kobj);
+ if (err)
+ drm_warn(&xe->drm,
+ "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+ return err;
+err_object:
+ kobject_put(kobj);
+ return err;
+}
+
+static void xe_hw_engine_sysfs_kobj_release(struct kobject *kobj)
+{
+ kfree(kobj);
+}
+
+static const struct kobj_type xe_hw_engine_sysfs_kobj_type = {
+ .release = xe_hw_engine_sysfs_kobj_release,
+ .sysfs_ops = &kobj_sysfs_ops,
+};
+
+static void hw_engine_class_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct kobject *kobj = arg;
+
+ kobject_put(kobj);
+}
+
+/**
+ * xe_hw_engine_class_sysfs_init - Init HW engine classes on GT.
+ * @gt: Xe GT.
+ *
+ * This routine creates sysfs for HW engine classes and adds methods
+ * to get/set different scheduling properties for HW engines class.
+ *
+ * Returns: Returns error value for failure and 0 for success.
+ */
+int xe_hw_engine_class_sysfs_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ struct kobject *kobj;
+ u16 class_mask = 0;
+ int err = 0;
+
+ kobj = kzalloc(sizeof(*kobj), GFP_KERNEL);
+ if (!kobj)
+ return -ENOMEM;
+
+ kobject_init(kobj, &xe_hw_engine_sysfs_kobj_type);
+
+ err = kobject_add(kobj, gt->sysfs, "engines");
+ if (err)
+ goto err_object;
+
+ for_each_hw_engine(hwe, gt, id) {
+ char name[MAX_ENGINE_CLASS_NAME_LEN];
+ struct kobj_eclass *keclass;
+
+ if (hwe->class == XE_ENGINE_CLASS_OTHER ||
+ hwe->class == XE_ENGINE_CLASS_MAX)
+ continue;
+
+ if ((class_mask >> hwe->class) & 1)
+ continue;
+
+ class_mask |= 1 << hwe->class;
+
+ switch (hwe->class) {
+ case XE_ENGINE_CLASS_RENDER:
+ strcpy(name, "rcs");
+ break;
+ case XE_ENGINE_CLASS_VIDEO_DECODE:
+ strcpy(name, "vcs");
+ break;
+ case XE_ENGINE_CLASS_VIDEO_ENHANCE:
+ strcpy(name, "vecs");
+ break;
+ case XE_ENGINE_CLASS_COPY:
+ strcpy(name, "bcs");
+ break;
+ case XE_ENGINE_CLASS_COMPUTE:
+ strcpy(name, "ccs");
+ break;
+ default:
+ err = -EINVAL;
+ goto err_object;
+ }
+
+ keclass = kobj_xe_hw_engine_class(xe, kobj, name);
+ if (!keclass) {
+ err = -EINVAL;
+ goto err_object;
+ }
+
+ keclass->eclass = hwe->eclass;
+ err = xe_add_hw_engine_class_defaults(xe, &keclass->base);
+ if (err) {
+ drm_warn(&xe->drm,
+ "Add .defaults to engines failed!, err: %d\n",
+ err);
+ goto err_object;
+ }
+
+ err = sysfs_create_files(&keclass->base, files);
+ if (err)
+ goto err_object;
+ }
+
+ err = drmm_add_action_or_reset(&xe->drm, hw_engine_class_sysfs_fini,
+ kobj);
+ if (err)
+ drm_warn(&xe->drm,
+ "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+
+ return err;
+err_object:
+ kobject_put(kobj);
+ return err;
+}
diff --git a/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.h b/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.h
new file mode 100644
index 000000000000..ec5ba673b314
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_engine_class_sysfs.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_ENGINE_CLASS_SYSFS_H_
+#define _XE_ENGINE_CLASS_SYSFS_H_
+
+#include <linux/kobject.h>
+
+struct xe_gt;
+struct xe_hw_engine_class_intf;
+
+int xe_hw_engine_class_sysfs_init(struct xe_gt *gt);
+bool xe_hw_engine_timeout_in_range(u64 timeout, u64 min, u64 max);
+
+/**
+ * struct kobj_eclass - A eclass's kobject struct that connects the kobject and the
+ * eclass.
+ *
+ * When dealing with multiple eclass, this struct helps to understand which eclass
+ * needs to be addressed on a given sysfs call.
+ */
+struct kobj_eclass {
+ /** @base: The actual kobject */
+ struct kobject base;
+ /** @eclass: A pointer to the hw engine class interface */
+ struct xe_hw_engine_class_intf *eclass;
+};
+
+static inline struct xe_hw_engine_class_intf *kobj_to_eclass(struct kobject *kobj)
+{
+ return container_of(kobj, struct kobj_eclass, base)->eclass;
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hw_engine_types.h b/drivers/gpu/drm/xe/xe_hw_engine_types.h
new file mode 100644
index 000000000000..39908dec042a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_engine_types.h
@@ -0,0 +1,225 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_HW_ENGINE_TYPES_H_
+#define _XE_HW_ENGINE_TYPES_H_
+
+#include "xe_force_wake_types.h"
+#include "xe_lrc_types.h"
+#include "xe_reg_sr_types.h"
+
+/* See "Engine ID Definition" struct in the Icelake PRM */
+enum xe_engine_class {
+ XE_ENGINE_CLASS_RENDER = 0,
+ XE_ENGINE_CLASS_VIDEO_DECODE = 1,
+ XE_ENGINE_CLASS_VIDEO_ENHANCE = 2,
+ XE_ENGINE_CLASS_COPY = 3,
+ XE_ENGINE_CLASS_OTHER = 4,
+ XE_ENGINE_CLASS_COMPUTE = 5,
+ XE_ENGINE_CLASS_MAX = 6,
+};
+
+enum xe_hw_engine_id {
+ XE_HW_ENGINE_RCS0,
+#define XE_HW_ENGINE_RCS_MASK GENMASK_ULL(XE_HW_ENGINE_RCS0, XE_HW_ENGINE_RCS0)
+ XE_HW_ENGINE_BCS0,
+ XE_HW_ENGINE_BCS1,
+ XE_HW_ENGINE_BCS2,
+ XE_HW_ENGINE_BCS3,
+ XE_HW_ENGINE_BCS4,
+ XE_HW_ENGINE_BCS5,
+ XE_HW_ENGINE_BCS6,
+ XE_HW_ENGINE_BCS7,
+ XE_HW_ENGINE_BCS8,
+#define XE_HW_ENGINE_BCS_MASK GENMASK_ULL(XE_HW_ENGINE_BCS8, XE_HW_ENGINE_BCS0)
+ XE_HW_ENGINE_VCS0,
+ XE_HW_ENGINE_VCS1,
+ XE_HW_ENGINE_VCS2,
+ XE_HW_ENGINE_VCS3,
+ XE_HW_ENGINE_VCS4,
+ XE_HW_ENGINE_VCS5,
+ XE_HW_ENGINE_VCS6,
+ XE_HW_ENGINE_VCS7,
+#define XE_HW_ENGINE_VCS_MASK GENMASK_ULL(XE_HW_ENGINE_VCS7, XE_HW_ENGINE_VCS0)
+ XE_HW_ENGINE_VECS0,
+ XE_HW_ENGINE_VECS1,
+ XE_HW_ENGINE_VECS2,
+ XE_HW_ENGINE_VECS3,
+#define XE_HW_ENGINE_VECS_MASK GENMASK_ULL(XE_HW_ENGINE_VECS3, XE_HW_ENGINE_VECS0)
+ XE_HW_ENGINE_CCS0,
+ XE_HW_ENGINE_CCS1,
+ XE_HW_ENGINE_CCS2,
+ XE_HW_ENGINE_CCS3,
+#define XE_HW_ENGINE_CCS_MASK GENMASK_ULL(XE_HW_ENGINE_CCS3, XE_HW_ENGINE_CCS0)
+ XE_HW_ENGINE_GSCCS0,
+#define XE_HW_ENGINE_GSCCS_MASK GENMASK_ULL(XE_HW_ENGINE_GSCCS0, XE_HW_ENGINE_GSCCS0)
+ XE_NUM_HW_ENGINES,
+};
+
+/* FIXME: s/XE_HW_ENGINE_MAX_INSTANCE/XE_HW_ENGINE_MAX_COUNT */
+#define XE_HW_ENGINE_MAX_INSTANCE 9
+
+struct xe_bo;
+struct xe_execlist_port;
+struct xe_gt;
+
+/**
+ * struct xe_hw_engine_class_intf - per hw engine class struct interface
+ *
+ * Contains all the hw engine properties per engine class.
+ *
+ * @sched_props: scheduling properties
+ * @defaults: default scheduling properties
+ */
+struct xe_hw_engine_class_intf {
+ /**
+ * @sched_props: scheduling properties
+ * @defaults: default scheduling properties
+ */
+ struct {
+ /** @set_job_timeout: Set job timeout in ms for engine */
+ u32 job_timeout_ms;
+ /** @job_timeout_min: Min job timeout in ms for engine */
+ u32 job_timeout_min;
+ /** @job_timeout_max: Max job timeout in ms for engine */
+ u32 job_timeout_max;
+ /** @timeslice_us: timeslice period in micro-seconds */
+ u32 timeslice_us;
+ /** @timeslice_min: min timeslice period in micro-seconds */
+ u32 timeslice_min;
+ /** @timeslice_max: max timeslice period in micro-seconds */
+ u32 timeslice_max;
+ /** @preempt_timeout_us: preemption timeout in micro-seconds */
+ u32 preempt_timeout_us;
+ /** @preempt_timeout_min: min preemption timeout in micro-seconds */
+ u32 preempt_timeout_min;
+ /** @preempt_timeout_max: max preemption timeout in micro-seconds */
+ u32 preempt_timeout_max;
+ } sched_props, defaults;
+};
+
+/**
+ * struct xe_hw_engine - Hardware engine
+ *
+ * Contains all the hardware engine state for physical instances.
+ */
+struct xe_hw_engine {
+ /** @gt: graphics tile this hw engine belongs to */
+ struct xe_gt *gt;
+ /** @name: name of this hw engine */
+ const char *name;
+ /** @class: class of this hw engine */
+ enum xe_engine_class class;
+ /** @instance: physical instance of this hw engine */
+ u16 instance;
+ /** @logical_instance: logical instance of this hw engine */
+ u16 logical_instance;
+ /** @mmio_base: MMIO base address of this hw engine*/
+ u32 mmio_base;
+ /**
+ * @reg_sr: table with registers to be restored on GT init/resume/reset
+ */
+ struct xe_reg_sr reg_sr;
+ /**
+ * @reg_whitelist: table with registers to be whitelisted
+ */
+ struct xe_reg_sr reg_whitelist;
+ /**
+ * @reg_lrc: LRC workaround registers
+ */
+ struct xe_reg_sr reg_lrc;
+ /** @domain: force wake domain of this hw engine */
+ enum xe_force_wake_domains domain;
+ /** @hwsp: hardware status page buffer object */
+ struct xe_bo *hwsp;
+ /** @kernel_lrc: Kernel LRC (should be replaced /w an xe_engine) */
+ struct xe_lrc kernel_lrc;
+ /** @exl_port: execlists port */
+ struct xe_execlist_port *exl_port;
+ /** @fence_irq: fence IRQ to run when a hw engine IRQ is received */
+ struct xe_hw_fence_irq *fence_irq;
+ /** @irq_handler: IRQ handler to run when hw engine IRQ is received */
+ void (*irq_handler)(struct xe_hw_engine *hwe, u16 intr_vec);
+ /** @engine_id: id for this hw engine */
+ enum xe_hw_engine_id engine_id;
+ /** @eclass: pointer to per hw engine class interface */
+ struct xe_hw_engine_class_intf *eclass;
+};
+
+/**
+ * struct xe_hw_engine_snapshot - Hardware engine snapshot
+ *
+ * Contains the snapshot of useful hardware engine info and registers.
+ */
+struct xe_hw_engine_snapshot {
+ /** @name: name of the hw engine */
+ char *name;
+ /** @class: class of this hw engine */
+ enum xe_engine_class class;
+ /** @logical_instance: logical instance of this hw engine */
+ u16 logical_instance;
+ /** @forcewake: Force Wake information snapshot */
+ struct {
+ /** @domain: force wake domain of this hw engine */
+ enum xe_force_wake_domains domain;
+ /** @ref: Forcewake ref for the above domain */
+ int ref;
+ } forcewake;
+ /** @mmio_base: MMIO base address of this hw engine*/
+ u32 mmio_base;
+ /** @reg: Useful MMIO register snapshot */
+ struct {
+ /** @ring_hwstam: RING_HWSTAM */
+ u32 ring_hwstam;
+ /** @ring_hws_pga: RING_HWS_PGA */
+ u32 ring_hws_pga;
+ /** @ring_execlist_status_lo: RING_EXECLIST_STATUS_LO */
+ u32 ring_execlist_status_lo;
+ /** @ring_execlist_status_hi: RING_EXECLIST_STATUS_HI */
+ u32 ring_execlist_status_hi;
+ /** @ring_execlist_sq_contents_lo: RING_EXECLIST_SQ_CONTENTS */
+ u32 ring_execlist_sq_contents_lo;
+ /** @ring_execlist_sq_contents_hi: RING_EXECLIST_SQ_CONTENTS + 4 */
+ u32 ring_execlist_sq_contents_hi;
+ /** @ring_start: RING_START */
+ u32 ring_start;
+ /** @ring_head: RING_HEAD */
+ u32 ring_head;
+ /** @ring_tail: RING_TAIL */
+ u32 ring_tail;
+ /** @ring_ctl: RING_CTL */
+ u32 ring_ctl;
+ /** @ring_mi_mode: RING_MI_MODE */
+ u32 ring_mi_mode;
+ /** @ring_mode: RING_MODE */
+ u32 ring_mode;
+ /** @ring_imr: RING_IMR */
+ u32 ring_imr;
+ /** @ring_esr: RING_ESR */
+ u32 ring_esr;
+ /** @ring_emr: RING_EMR */
+ u32 ring_emr;
+ /** @ring_eir: RING_EIR */
+ u32 ring_eir;
+ /** @ring_acthd_udw: RING_ACTHD_UDW */
+ u32 ring_acthd_udw;
+ /** @ring_acthd: RING_ACTHD */
+ u32 ring_acthd;
+ /** @ring_bbaddr_udw: RING_BBADDR_UDW */
+ u32 ring_bbaddr_udw;
+ /** @ring_bbaddr: RING_BBADDR */
+ u32 ring_bbaddr;
+ /** @ring_dma_fadd_udw: RING_DMA_FADD_UDW */
+ u32 ring_dma_fadd_udw;
+ /** @ring_dma_fadd: RING_DMA_FADD */
+ u32 ring_dma_fadd;
+ /** @ipehr: IPEHR */
+ u32 ipehr;
+ /** @rcu_mode: RCU_MODE */
+ u32 rcu_mode;
+ } reg;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hw_fence.c b/drivers/gpu/drm/xe/xe_hw_fence.c
new file mode 100644
index 000000000000..a6094c81f2ad
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_fence.c
@@ -0,0 +1,230 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_hw_fence.h"
+
+#include <linux/device.h>
+#include <linux/slab.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_hw_engine.h"
+#include "xe_macros.h"
+#include "xe_map.h"
+#include "xe_trace.h"
+
+static struct kmem_cache *xe_hw_fence_slab;
+
+int __init xe_hw_fence_module_init(void)
+{
+ xe_hw_fence_slab = kmem_cache_create("xe_hw_fence",
+ sizeof(struct xe_hw_fence), 0,
+ SLAB_HWCACHE_ALIGN, NULL);
+ if (!xe_hw_fence_slab)
+ return -ENOMEM;
+
+ return 0;
+}
+
+void xe_hw_fence_module_exit(void)
+{
+ rcu_barrier();
+ kmem_cache_destroy(xe_hw_fence_slab);
+}
+
+static struct xe_hw_fence *fence_alloc(void)
+{
+ return kmem_cache_zalloc(xe_hw_fence_slab, GFP_KERNEL);
+}
+
+static void fence_free(struct rcu_head *rcu)
+{
+ struct xe_hw_fence *fence =
+ container_of(rcu, struct xe_hw_fence, dma.rcu);
+
+ if (!WARN_ON_ONCE(!fence))
+ kmem_cache_free(xe_hw_fence_slab, fence);
+}
+
+static void hw_fence_irq_run_cb(struct irq_work *work)
+{
+ struct xe_hw_fence_irq *irq = container_of(work, typeof(*irq), work);
+ struct xe_hw_fence *fence, *next;
+ bool tmp;
+
+ tmp = dma_fence_begin_signalling();
+ spin_lock(&irq->lock);
+ if (irq->enabled) {
+ list_for_each_entry_safe(fence, next, &irq->pending, irq_link) {
+ struct dma_fence *dma_fence = &fence->dma;
+
+ trace_xe_hw_fence_try_signal(fence);
+ if (dma_fence_is_signaled_locked(dma_fence)) {
+ trace_xe_hw_fence_signal(fence);
+ list_del_init(&fence->irq_link);
+ dma_fence_put(dma_fence);
+ }
+ }
+ }
+ spin_unlock(&irq->lock);
+ dma_fence_end_signalling(tmp);
+}
+
+void xe_hw_fence_irq_init(struct xe_hw_fence_irq *irq)
+{
+ spin_lock_init(&irq->lock);
+ init_irq_work(&irq->work, hw_fence_irq_run_cb);
+ INIT_LIST_HEAD(&irq->pending);
+ irq->enabled = true;
+}
+
+void xe_hw_fence_irq_finish(struct xe_hw_fence_irq *irq)
+{
+ struct xe_hw_fence *fence, *next;
+ unsigned long flags;
+ int err;
+ bool tmp;
+
+ if (XE_WARN_ON(!list_empty(&irq->pending))) {
+ tmp = dma_fence_begin_signalling();
+ spin_lock_irqsave(&irq->lock, flags);
+ list_for_each_entry_safe(fence, next, &irq->pending, irq_link) {
+ list_del_init(&fence->irq_link);
+ err = dma_fence_signal_locked(&fence->dma);
+ dma_fence_put(&fence->dma);
+ XE_WARN_ON(err);
+ }
+ spin_unlock_irqrestore(&irq->lock, flags);
+ dma_fence_end_signalling(tmp);
+ }
+}
+
+void xe_hw_fence_irq_run(struct xe_hw_fence_irq *irq)
+{
+ irq_work_queue(&irq->work);
+}
+
+void xe_hw_fence_irq_stop(struct xe_hw_fence_irq *irq)
+{
+ spin_lock_irq(&irq->lock);
+ irq->enabled = false;
+ spin_unlock_irq(&irq->lock);
+}
+
+void xe_hw_fence_irq_start(struct xe_hw_fence_irq *irq)
+{
+ spin_lock_irq(&irq->lock);
+ irq->enabled = true;
+ spin_unlock_irq(&irq->lock);
+
+ irq_work_queue(&irq->work);
+}
+
+void xe_hw_fence_ctx_init(struct xe_hw_fence_ctx *ctx, struct xe_gt *gt,
+ struct xe_hw_fence_irq *irq, const char *name)
+{
+ ctx->gt = gt;
+ ctx->irq = irq;
+ ctx->dma_fence_ctx = dma_fence_context_alloc(1);
+ ctx->next_seqno = XE_FENCE_INITIAL_SEQNO;
+ sprintf(ctx->name, "%s", name);
+}
+
+void xe_hw_fence_ctx_finish(struct xe_hw_fence_ctx *ctx)
+{
+}
+
+static struct xe_hw_fence *to_xe_hw_fence(struct dma_fence *fence);
+
+static struct xe_hw_fence_irq *xe_hw_fence_irq(struct xe_hw_fence *fence)
+{
+ return container_of(fence->dma.lock, struct xe_hw_fence_irq, lock);
+}
+
+static const char *xe_hw_fence_get_driver_name(struct dma_fence *dma_fence)
+{
+ struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
+
+ return dev_name(gt_to_xe(fence->ctx->gt)->drm.dev);
+}
+
+static const char *xe_hw_fence_get_timeline_name(struct dma_fence *dma_fence)
+{
+ struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
+
+ return fence->ctx->name;
+}
+
+static bool xe_hw_fence_signaled(struct dma_fence *dma_fence)
+{
+ struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
+ struct xe_device *xe = gt_to_xe(fence->ctx->gt);
+ u32 seqno = xe_map_rd(xe, &fence->seqno_map, 0, u32);
+
+ return dma_fence->error ||
+ !__dma_fence_is_later(dma_fence->seqno, seqno, dma_fence->ops);
+}
+
+static bool xe_hw_fence_enable_signaling(struct dma_fence *dma_fence)
+{
+ struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
+ struct xe_hw_fence_irq *irq = xe_hw_fence_irq(fence);
+
+ dma_fence_get(dma_fence);
+ list_add_tail(&fence->irq_link, &irq->pending);
+
+ /* SW completed (no HW IRQ) so kick handler to signal fence */
+ if (xe_hw_fence_signaled(dma_fence))
+ xe_hw_fence_irq_run(irq);
+
+ return true;
+}
+
+static void xe_hw_fence_release(struct dma_fence *dma_fence)
+{
+ struct xe_hw_fence *fence = to_xe_hw_fence(dma_fence);
+
+ trace_xe_hw_fence_free(fence);
+ XE_WARN_ON(!list_empty(&fence->irq_link));
+ call_rcu(&dma_fence->rcu, fence_free);
+}
+
+static const struct dma_fence_ops xe_hw_fence_ops = {
+ .get_driver_name = xe_hw_fence_get_driver_name,
+ .get_timeline_name = xe_hw_fence_get_timeline_name,
+ .enable_signaling = xe_hw_fence_enable_signaling,
+ .signaled = xe_hw_fence_signaled,
+ .release = xe_hw_fence_release,
+};
+
+static struct xe_hw_fence *to_xe_hw_fence(struct dma_fence *fence)
+{
+ if (XE_WARN_ON(fence->ops != &xe_hw_fence_ops))
+ return NULL;
+
+ return container_of(fence, struct xe_hw_fence, dma);
+}
+
+struct xe_hw_fence *xe_hw_fence_create(struct xe_hw_fence_ctx *ctx,
+ struct iosys_map seqno_map)
+{
+ struct xe_hw_fence *fence;
+
+ fence = fence_alloc();
+ if (!fence)
+ return ERR_PTR(-ENOMEM);
+
+ dma_fence_init(&fence->dma, &xe_hw_fence_ops, &ctx->irq->lock,
+ ctx->dma_fence_ctx, ctx->next_seqno++);
+
+ fence->ctx = ctx;
+ fence->seqno_map = seqno_map;
+ INIT_LIST_HEAD(&fence->irq_link);
+
+ trace_xe_hw_fence_create(fence);
+
+ return fence;
+}
diff --git a/drivers/gpu/drm/xe/xe_hw_fence.h b/drivers/gpu/drm/xe/xe_hw_fence.h
new file mode 100644
index 000000000000..cfe5fd603787
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_fence.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_HW_FENCE_H_
+#define _XE_HW_FENCE_H_
+
+#include "xe_hw_fence_types.h"
+
+/* Cause an early wrap to catch wrapping errors */
+#define XE_FENCE_INITIAL_SEQNO (-127)
+
+int xe_hw_fence_module_init(void);
+void xe_hw_fence_module_exit(void);
+
+void xe_hw_fence_irq_init(struct xe_hw_fence_irq *irq);
+void xe_hw_fence_irq_finish(struct xe_hw_fence_irq *irq);
+void xe_hw_fence_irq_run(struct xe_hw_fence_irq *irq);
+void xe_hw_fence_irq_stop(struct xe_hw_fence_irq *irq);
+void xe_hw_fence_irq_start(struct xe_hw_fence_irq *irq);
+
+void xe_hw_fence_ctx_init(struct xe_hw_fence_ctx *ctx, struct xe_gt *gt,
+ struct xe_hw_fence_irq *irq, const char *name);
+void xe_hw_fence_ctx_finish(struct xe_hw_fence_ctx *ctx);
+
+struct xe_hw_fence *xe_hw_fence_create(struct xe_hw_fence_ctx *ctx,
+ struct iosys_map seqno_map);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hw_fence_types.h b/drivers/gpu/drm/xe/xe_hw_fence_types.h
new file mode 100644
index 000000000000..b33c4956e8ea
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hw_fence_types.h
@@ -0,0 +1,72 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_HW_FENCE_TYPES_H_
+#define _XE_HW_FENCE_TYPES_H_
+
+#include <linux/dma-fence.h>
+#include <linux/iosys-map.h>
+#include <linux/irq_work.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+
+struct xe_gt;
+
+/**
+ * struct xe_hw_fence_irq - hardware fence IRQ handler
+ *
+ * One per engine class, signals completed xe_hw_fences, triggered via hw engine
+ * interrupt. On each trigger, search list of pending fences and signal.
+ */
+struct xe_hw_fence_irq {
+ /** @lock: protects all xe_hw_fences + pending list */
+ spinlock_t lock;
+ /** @work: IRQ worker run to signal the fences */
+ struct irq_work work;
+ /** @pending: list of pending xe_hw_fences */
+ struct list_head pending;
+ /** @enabled: fence signaling enabled */
+ bool enabled;
+};
+
+#define MAX_FENCE_NAME_LEN 16
+
+/**
+ * struct xe_hw_fence_ctx - hardware fence context
+ *
+ * The context for a hardware fence. 1 to 1 relationship with xe_engine. Points
+ * to a xe_hw_fence_irq, maintains serial seqno.
+ */
+struct xe_hw_fence_ctx {
+ /** @gt: graphics tile of hardware fence context */
+ struct xe_gt *gt;
+ /** @irq: fence irq handler */
+ struct xe_hw_fence_irq *irq;
+ /** @dma_fence_ctx: dma fence context for hardware fence */
+ u64 dma_fence_ctx;
+ /** @next_seqno: next seqno for hardware fence */
+ u32 next_seqno;
+ /** @name: name of hardware fence context */
+ char name[MAX_FENCE_NAME_LEN];
+};
+
+/**
+ * struct xe_hw_fence - hardware fence
+ *
+ * Used to indicate a xe_sched_job is complete via a seqno written to memory.
+ * Signals on error or seqno past.
+ */
+struct xe_hw_fence {
+ /** @dma: base dma fence for hardware fence context */
+ struct dma_fence dma;
+ /** @ctx: hardware fence context */
+ struct xe_hw_fence_ctx *ctx;
+ /** @seqno_map: I/O map for seqno */
+ struct iosys_map seqno_map;
+ /** @irq_link: Link in struct xe_hw_fence_irq.pending */
+ struct list_head irq_link;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_hwmon.c b/drivers/gpu/drm/xe/xe_hwmon.c
new file mode 100644
index 000000000000..6ef2aa1eae8b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hwmon.c
@@ -0,0 +1,776 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/hwmon-sysfs.h>
+#include <linux/hwmon.h>
+#include <linux/types.h>
+
+#include <drm/drm_managed.h>
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_mchbar_regs.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_hwmon.h"
+#include "xe_mmio.h"
+#include "xe_pcode.h"
+#include "xe_pcode_api.h"
+
+enum xe_hwmon_reg {
+ REG_PKG_RAPL_LIMIT,
+ REG_PKG_POWER_SKU,
+ REG_PKG_POWER_SKU_UNIT,
+ REG_GT_PERF_STATUS,
+ REG_PKG_ENERGY_STATUS,
+};
+
+enum xe_hwmon_reg_operation {
+ REG_READ32,
+ REG_RMW32,
+ REG_READ64,
+};
+
+/*
+ * SF_* - scale factors for particular quantities according to hwmon spec.
+ */
+#define SF_POWER 1000000 /* microwatts */
+#define SF_CURR 1000 /* milliamperes */
+#define SF_VOLTAGE 1000 /* millivolts */
+#define SF_ENERGY 1000000 /* microjoules */
+#define SF_TIME 1000 /* milliseconds */
+
+/**
+ * struct xe_hwmon_energy_info - to accumulate energy
+ */
+struct xe_hwmon_energy_info {
+ /** @reg_val_prev: previous energy reg val */
+ u32 reg_val_prev;
+ /** @accum_energy: accumulated energy */
+ long accum_energy;
+};
+
+/**
+ * struct xe_hwmon - xe hwmon data structure
+ */
+struct xe_hwmon {
+ /** @hwmon_dev: hwmon device for xe */
+ struct device *hwmon_dev;
+ /** @gt: primary gt */
+ struct xe_gt *gt;
+ /** @hwmon_lock: lock for rw attributes*/
+ struct mutex hwmon_lock;
+ /** @scl_shift_power: pkg power unit */
+ int scl_shift_power;
+ /** @scl_shift_energy: pkg energy unit */
+ int scl_shift_energy;
+ /** @scl_shift_time: pkg time unit */
+ int scl_shift_time;
+ /** @ei: Energy info for energy1_input */
+ struct xe_hwmon_energy_info ei;
+};
+
+static u32 xe_hwmon_get_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg hwmon_reg)
+{
+ struct xe_device *xe = gt_to_xe(hwmon->gt);
+ struct xe_reg reg = XE_REG(0);
+
+ switch (hwmon_reg) {
+ case REG_PKG_RAPL_LIMIT:
+ if (xe->info.platform == XE_DG2)
+ reg = PCU_CR_PACKAGE_RAPL_LIMIT;
+ else if (xe->info.platform == XE_PVC)
+ reg = PVC_GT0_PACKAGE_RAPL_LIMIT;
+ break;
+ case REG_PKG_POWER_SKU:
+ if (xe->info.platform == XE_DG2)
+ reg = PCU_CR_PACKAGE_POWER_SKU;
+ else if (xe->info.platform == XE_PVC)
+ reg = PVC_GT0_PACKAGE_POWER_SKU;
+ break;
+ case REG_PKG_POWER_SKU_UNIT:
+ if (xe->info.platform == XE_DG2)
+ reg = PCU_CR_PACKAGE_POWER_SKU_UNIT;
+ else if (xe->info.platform == XE_PVC)
+ reg = PVC_GT0_PACKAGE_POWER_SKU_UNIT;
+ break;
+ case REG_GT_PERF_STATUS:
+ if (xe->info.platform == XE_DG2)
+ reg = GT_PERF_STATUS;
+ break;
+ case REG_PKG_ENERGY_STATUS:
+ if (xe->info.platform == XE_DG2)
+ reg = PCU_CR_PACKAGE_ENERGY_STATUS;
+ else if (xe->info.platform == XE_PVC)
+ reg = PVC_GT0_PLATFORM_ENERGY_STATUS;
+ break;
+ default:
+ drm_warn(&xe->drm, "Unknown xe hwmon reg id: %d\n", hwmon_reg);
+ break;
+ }
+
+ return reg.raw;
+}
+
+static void xe_hwmon_process_reg(struct xe_hwmon *hwmon, enum xe_hwmon_reg hwmon_reg,
+ enum xe_hwmon_reg_operation operation, u64 *value,
+ u32 clr, u32 set)
+{
+ struct xe_reg reg;
+
+ reg.raw = xe_hwmon_get_reg(hwmon, hwmon_reg);
+
+ if (!reg.raw)
+ return;
+
+ switch (operation) {
+ case REG_READ32:
+ *value = xe_mmio_read32(hwmon->gt, reg);
+ break;
+ case REG_RMW32:
+ *value = xe_mmio_rmw32(hwmon->gt, reg, clr, set);
+ break;
+ case REG_READ64:
+ *value = xe_mmio_read64_2x32(hwmon->gt, reg);
+ break;
+ default:
+ drm_warn(&gt_to_xe(hwmon->gt)->drm, "Invalid xe hwmon reg operation: %d\n",
+ operation);
+ break;
+ }
+}
+
+#define PL1_DISABLE 0
+
+/*
+ * HW allows arbitrary PL1 limits to be set but silently clamps these values to
+ * "typical but not guaranteed" min/max values in REG_PKG_POWER_SKU. Follow the
+ * same pattern for sysfs, allow arbitrary PL1 limits to be set but display
+ * clamped values when read.
+ */
+static void xe_hwmon_power_max_read(struct xe_hwmon *hwmon, long *value)
+{
+ u64 reg_val, min, max;
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT, REG_READ32, &reg_val, 0, 0);
+ /* Check if PL1 limit is disabled */
+ if (!(reg_val & PKG_PWR_LIM_1_EN)) {
+ *value = PL1_DISABLE;
+ goto unlock;
+ }
+
+ reg_val = REG_FIELD_GET(PKG_PWR_LIM_1, reg_val);
+ *value = mul_u64_u32_shr(reg_val, SF_POWER, hwmon->scl_shift_power);
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_POWER_SKU, REG_READ64, &reg_val, 0, 0);
+ min = REG_FIELD_GET(PKG_MIN_PWR, reg_val);
+ min = mul_u64_u32_shr(min, SF_POWER, hwmon->scl_shift_power);
+ max = REG_FIELD_GET(PKG_MAX_PWR, reg_val);
+ max = mul_u64_u32_shr(max, SF_POWER, hwmon->scl_shift_power);
+
+ if (min && max)
+ *value = clamp_t(u64, *value, min, max);
+unlock:
+ mutex_unlock(&hwmon->hwmon_lock);
+}
+
+static int xe_hwmon_power_max_write(struct xe_hwmon *hwmon, long value)
+{
+ int ret = 0;
+ u64 reg_val;
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ /* Disable PL1 limit and verify, as limit cannot be disabled on all platforms */
+ if (value == PL1_DISABLE) {
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT, REG_RMW32, &reg_val,
+ PKG_PWR_LIM_1_EN, 0);
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT, REG_READ32, &reg_val,
+ PKG_PWR_LIM_1_EN, 0);
+
+ if (reg_val & PKG_PWR_LIM_1_EN) {
+ ret = -EOPNOTSUPP;
+ goto unlock;
+ }
+ }
+
+ /* Computation in 64-bits to avoid overflow. Round to nearest. */
+ reg_val = DIV_ROUND_CLOSEST_ULL((u64)value << hwmon->scl_shift_power, SF_POWER);
+ reg_val = PKG_PWR_LIM_1_EN | REG_FIELD_PREP(PKG_PWR_LIM_1, reg_val);
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT, REG_RMW32, &reg_val,
+ PKG_PWR_LIM_1_EN | PKG_PWR_LIM_1, reg_val);
+unlock:
+ mutex_unlock(&hwmon->hwmon_lock);
+ return ret;
+}
+
+static void xe_hwmon_power_rated_max_read(struct xe_hwmon *hwmon, long *value)
+{
+ u64 reg_val;
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_POWER_SKU, REG_READ32, &reg_val, 0, 0);
+ reg_val = REG_FIELD_GET(PKG_TDP, reg_val);
+ *value = mul_u64_u32_shr(reg_val, SF_POWER, hwmon->scl_shift_power);
+}
+
+/*
+ * xe_hwmon_energy_get - Obtain energy value
+ *
+ * The underlying energy hardware register is 32-bits and is subject to
+ * overflow. How long before overflow? For example, with an example
+ * scaling bit shift of 14 bits (see register *PACKAGE_POWER_SKU_UNIT) and
+ * a power draw of 1000 watts, the 32-bit counter will overflow in
+ * approximately 4.36 minutes.
+ *
+ * Examples:
+ * 1 watt: (2^32 >> 14) / 1 W / (60 * 60 * 24) secs/day -> 3 days
+ * 1000 watts: (2^32 >> 14) / 1000 W / 60 secs/min -> 4.36 minutes
+ *
+ * The function significantly increases overflow duration (from 4.36
+ * minutes) by accumulating the energy register into a 'long' as allowed by
+ * the hwmon API. Using x86_64 128 bit arithmetic (see mul_u64_u32_shr()),
+ * a 'long' of 63 bits, SF_ENERGY of 1e6 (~20 bits) and
+ * hwmon->scl_shift_energy of 14 bits we have 57 (63 - 20 + 14) bits before
+ * energy1_input overflows. This at 1000 W is an overflow duration of 278 years.
+ */
+static void
+xe_hwmon_energy_get(struct xe_hwmon *hwmon, long *energy)
+{
+ struct xe_hwmon_energy_info *ei = &hwmon->ei;
+ u64 reg_val;
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_ENERGY_STATUS, REG_READ32,
+ &reg_val, 0, 0);
+
+ if (reg_val >= ei->reg_val_prev)
+ ei->accum_energy += reg_val - ei->reg_val_prev;
+ else
+ ei->accum_energy += UINT_MAX - ei->reg_val_prev + reg_val;
+
+ ei->reg_val_prev = reg_val;
+
+ *energy = mul_u64_u32_shr(ei->accum_energy, SF_ENERGY,
+ hwmon->scl_shift_energy);
+}
+
+static ssize_t
+xe_hwmon_power1_max_interval_show(struct device *dev, struct device_attribute *attr,
+ char *buf)
+{
+ struct xe_hwmon *hwmon = dev_get_drvdata(dev);
+ u32 x, y, x_w = 2; /* 2 bits */
+ u64 r, tau4, out;
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT,
+ REG_READ32, &r, 0, 0);
+
+ mutex_unlock(&hwmon->hwmon_lock);
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ x = REG_FIELD_GET(PKG_PWR_LIM_1_TIME_X, r);
+ y = REG_FIELD_GET(PKG_PWR_LIM_1_TIME_Y, r);
+
+ /*
+ * tau = 1.x * power(2,y), x = bits(23:22), y = bits(21:17)
+ * = (4 | x) << (y - 2)
+ *
+ * Here (y - 2) ensures a 1.x fixed point representation of 1.x
+ * As x is 2 bits so 1.x can be 1.0, 1.25, 1.50, 1.75
+ *
+ * As y can be < 2, we compute tau4 = (4 | x) << y
+ * and then add 2 when doing the final right shift to account for units
+ */
+ tau4 = ((1 << x_w) | x) << y;
+
+ /* val in hwmon interface units (millisec) */
+ out = mul_u64_u32_shr(tau4, SF_TIME, hwmon->scl_shift_time + x_w);
+
+ return sysfs_emit(buf, "%llu\n", out);
+}
+
+static ssize_t
+xe_hwmon_power1_max_interval_store(struct device *dev, struct device_attribute *attr,
+ const char *buf, size_t count)
+{
+ struct xe_hwmon *hwmon = dev_get_drvdata(dev);
+ u32 x, y, rxy, x_w = 2; /* 2 bits */
+ u64 tau4, r, max_win;
+ unsigned long val;
+ int ret;
+
+ ret = kstrtoul(buf, 0, &val);
+ if (ret)
+ return ret;
+
+ /*
+ * Max HW supported tau in '1.x * power(2,y)' format, x = 0, y = 0x12.
+ * The hwmon->scl_shift_time default of 0xa results in a max tau of 256 seconds.
+ *
+ * The ideal scenario is for PKG_MAX_WIN to be read from the PKG_PWR_SKU register.
+ * However, it is observed that existing discrete GPUs does not provide correct
+ * PKG_MAX_WIN value, therefore a using default constant value. For future discrete GPUs
+ * this may get resolved, in which case PKG_MAX_WIN should be obtained from PKG_PWR_SKU.
+ */
+#define PKG_MAX_WIN_DEFAULT 0x12ull
+
+ /*
+ * val must be < max in hwmon interface units. The steps below are
+ * explained in xe_hwmon_power1_max_interval_show()
+ */
+ r = FIELD_PREP(PKG_MAX_WIN, PKG_MAX_WIN_DEFAULT);
+ x = REG_FIELD_GET(PKG_MAX_WIN_X, r);
+ y = REG_FIELD_GET(PKG_MAX_WIN_Y, r);
+ tau4 = ((1 << x_w) | x) << y;
+ max_win = mul_u64_u32_shr(tau4, SF_TIME, hwmon->scl_shift_time + x_w);
+
+ if (val > max_win)
+ return -EINVAL;
+
+ /* val in hw units */
+ val = DIV_ROUND_CLOSEST_ULL((u64)val << hwmon->scl_shift_time, SF_TIME);
+
+ /*
+ * Convert val to 1.x * power(2,y)
+ * y = ilog2(val)
+ * x = (val - (1 << y)) >> (y - 2)
+ */
+ if (!val) {
+ y = 0;
+ x = 0;
+ } else {
+ y = ilog2(val);
+ x = (val - (1ul << y)) << x_w >> y;
+ }
+
+ rxy = REG_FIELD_PREP(PKG_PWR_LIM_1_TIME_X, x) | REG_FIELD_PREP(PKG_PWR_LIM_1_TIME_Y, y);
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ xe_hwmon_process_reg(hwmon, REG_PKG_RAPL_LIMIT, REG_RMW32, (u64 *)&r,
+ PKG_PWR_LIM_1_TIME, rxy);
+
+ mutex_unlock(&hwmon->hwmon_lock);
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ return count;
+}
+
+static SENSOR_DEVICE_ATTR(power1_max_interval, 0664,
+ xe_hwmon_power1_max_interval_show,
+ xe_hwmon_power1_max_interval_store, 0);
+
+static struct attribute *hwmon_attributes[] = {
+ &sensor_dev_attr_power1_max_interval.dev_attr.attr,
+ NULL
+};
+
+static umode_t xe_hwmon_attributes_visible(struct kobject *kobj,
+ struct attribute *attr, int index)
+{
+ struct device *dev = kobj_to_dev(kobj);
+ struct xe_hwmon *hwmon = dev_get_drvdata(dev);
+ int ret = 0;
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ if (attr == &sensor_dev_attr_power1_max_interval.dev_attr.attr)
+ ret = xe_hwmon_get_reg(hwmon, REG_PKG_RAPL_LIMIT) ? attr->mode : 0;
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ return ret;
+}
+
+static const struct attribute_group hwmon_attrgroup = {
+ .attrs = hwmon_attributes,
+ .is_visible = xe_hwmon_attributes_visible,
+};
+
+static const struct attribute_group *hwmon_groups[] = {
+ &hwmon_attrgroup,
+ NULL
+};
+
+static const struct hwmon_channel_info *hwmon_info[] = {
+ HWMON_CHANNEL_INFO(power, HWMON_P_MAX | HWMON_P_RATED_MAX | HWMON_P_CRIT),
+ HWMON_CHANNEL_INFO(curr, HWMON_C_CRIT),
+ HWMON_CHANNEL_INFO(in, HWMON_I_INPUT),
+ HWMON_CHANNEL_INFO(energy, HWMON_E_INPUT),
+ NULL
+};
+
+/* I1 is exposed as power_crit or as curr_crit depending on bit 31 */
+static int xe_hwmon_pcode_read_i1(struct xe_gt *gt, u32 *uval)
+{
+ /* Avoid Illegal Subcommand error */
+ if (gt_to_xe(gt)->info.platform == XE_DG2)
+ return -ENXIO;
+
+ return xe_pcode_read(gt, PCODE_MBOX(PCODE_POWER_SETUP,
+ POWER_SETUP_SUBCOMMAND_READ_I1, 0),
+ uval, 0);
+}
+
+static int xe_hwmon_pcode_write_i1(struct xe_gt *gt, u32 uval)
+{
+ return xe_pcode_write(gt, PCODE_MBOX(PCODE_POWER_SETUP,
+ POWER_SETUP_SUBCOMMAND_WRITE_I1, 0),
+ uval);
+}
+
+static int xe_hwmon_power_curr_crit_read(struct xe_hwmon *hwmon, long *value, u32 scale_factor)
+{
+ int ret;
+ u32 uval;
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ ret = xe_hwmon_pcode_read_i1(hwmon->gt, &uval);
+ if (ret)
+ goto unlock;
+
+ *value = mul_u64_u32_shr(REG_FIELD_GET(POWER_SETUP_I1_DATA_MASK, uval),
+ scale_factor, POWER_SETUP_I1_SHIFT);
+unlock:
+ mutex_unlock(&hwmon->hwmon_lock);
+ return ret;
+}
+
+static int xe_hwmon_power_curr_crit_write(struct xe_hwmon *hwmon, long value, u32 scale_factor)
+{
+ int ret;
+ u32 uval;
+
+ mutex_lock(&hwmon->hwmon_lock);
+
+ uval = DIV_ROUND_CLOSEST_ULL(value << POWER_SETUP_I1_SHIFT, scale_factor);
+ ret = xe_hwmon_pcode_write_i1(hwmon->gt, uval);
+
+ mutex_unlock(&hwmon->hwmon_lock);
+ return ret;
+}
+
+static void xe_hwmon_get_voltage(struct xe_hwmon *hwmon, long *value)
+{
+ u64 reg_val;
+
+ xe_hwmon_process_reg(hwmon, REG_GT_PERF_STATUS,
+ REG_READ32, &reg_val, 0, 0);
+ /* HW register value in units of 2.5 millivolt */
+ *value = DIV_ROUND_CLOSEST(REG_FIELD_GET(VOLTAGE_MASK, reg_val) * 2500, SF_VOLTAGE);
+}
+
+static umode_t
+xe_hwmon_power_is_visible(struct xe_hwmon *hwmon, u32 attr, int chan)
+{
+ u32 uval;
+
+ switch (attr) {
+ case hwmon_power_max:
+ return xe_hwmon_get_reg(hwmon, REG_PKG_RAPL_LIMIT) ? 0664 : 0;
+ case hwmon_power_rated_max:
+ return xe_hwmon_get_reg(hwmon, REG_PKG_POWER_SKU) ? 0444 : 0;
+ case hwmon_power_crit:
+ return (xe_hwmon_pcode_read_i1(hwmon->gt, &uval) ||
+ !(uval & POWER_SETUP_I1_WATTS)) ? 0 : 0644;
+ default:
+ return 0;
+ }
+}
+
+static int
+xe_hwmon_power_read(struct xe_hwmon *hwmon, u32 attr, int chan, long *val)
+{
+ switch (attr) {
+ case hwmon_power_max:
+ xe_hwmon_power_max_read(hwmon, val);
+ return 0;
+ case hwmon_power_rated_max:
+ xe_hwmon_power_rated_max_read(hwmon, val);
+ return 0;
+ case hwmon_power_crit:
+ return xe_hwmon_power_curr_crit_read(hwmon, val, SF_POWER);
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static int
+xe_hwmon_power_write(struct xe_hwmon *hwmon, u32 attr, int chan, long val)
+{
+ switch (attr) {
+ case hwmon_power_max:
+ return xe_hwmon_power_max_write(hwmon, val);
+ case hwmon_power_crit:
+ return xe_hwmon_power_curr_crit_write(hwmon, val, SF_POWER);
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static umode_t
+xe_hwmon_curr_is_visible(const struct xe_hwmon *hwmon, u32 attr)
+{
+ u32 uval;
+
+ switch (attr) {
+ case hwmon_curr_crit:
+ return (xe_hwmon_pcode_read_i1(hwmon->gt, &uval) ||
+ (uval & POWER_SETUP_I1_WATTS)) ? 0 : 0644;
+ default:
+ return 0;
+ }
+}
+
+static int
+xe_hwmon_curr_read(struct xe_hwmon *hwmon, u32 attr, long *val)
+{
+ switch (attr) {
+ case hwmon_curr_crit:
+ return xe_hwmon_power_curr_crit_read(hwmon, val, SF_CURR);
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static int
+xe_hwmon_curr_write(struct xe_hwmon *hwmon, u32 attr, long val)
+{
+ switch (attr) {
+ case hwmon_curr_crit:
+ return xe_hwmon_power_curr_crit_write(hwmon, val, SF_CURR);
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static umode_t
+xe_hwmon_in_is_visible(struct xe_hwmon *hwmon, u32 attr)
+{
+ switch (attr) {
+ case hwmon_in_input:
+ return xe_hwmon_get_reg(hwmon, REG_GT_PERF_STATUS) ? 0444 : 0;
+ default:
+ return 0;
+ }
+}
+
+static int
+xe_hwmon_in_read(struct xe_hwmon *hwmon, u32 attr, long *val)
+{
+ switch (attr) {
+ case hwmon_in_input:
+ xe_hwmon_get_voltage(hwmon, val);
+ return 0;
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static umode_t
+xe_hwmon_energy_is_visible(struct xe_hwmon *hwmon, u32 attr)
+{
+ switch (attr) {
+ case hwmon_energy_input:
+ return xe_hwmon_get_reg(hwmon, REG_PKG_ENERGY_STATUS) ? 0444 : 0;
+ default:
+ return 0;
+ }
+}
+
+static int
+xe_hwmon_energy_read(struct xe_hwmon *hwmon, u32 attr, long *val)
+{
+ switch (attr) {
+ case hwmon_energy_input:
+ xe_hwmon_energy_get(hwmon, val);
+ return 0;
+ default:
+ return -EOPNOTSUPP;
+ }
+}
+
+static umode_t
+xe_hwmon_is_visible(const void *drvdata, enum hwmon_sensor_types type,
+ u32 attr, int channel)
+{
+ struct xe_hwmon *hwmon = (struct xe_hwmon *)drvdata;
+ int ret;
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ switch (type) {
+ case hwmon_power:
+ ret = xe_hwmon_power_is_visible(hwmon, attr, channel);
+ break;
+ case hwmon_curr:
+ ret = xe_hwmon_curr_is_visible(hwmon, attr);
+ break;
+ case hwmon_in:
+ ret = xe_hwmon_in_is_visible(hwmon, attr);
+ break;
+ case hwmon_energy:
+ ret = xe_hwmon_energy_is_visible(hwmon, attr);
+ break;
+ default:
+ ret = 0;
+ break;
+ }
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ return ret;
+}
+
+static int
+xe_hwmon_read(struct device *dev, enum hwmon_sensor_types type, u32 attr,
+ int channel, long *val)
+{
+ struct xe_hwmon *hwmon = dev_get_drvdata(dev);
+ int ret;
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ switch (type) {
+ case hwmon_power:
+ ret = xe_hwmon_power_read(hwmon, attr, channel, val);
+ break;
+ case hwmon_curr:
+ ret = xe_hwmon_curr_read(hwmon, attr, val);
+ break;
+ case hwmon_in:
+ ret = xe_hwmon_in_read(hwmon, attr, val);
+ break;
+ case hwmon_energy:
+ ret = xe_hwmon_energy_read(hwmon, attr, val);
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ return ret;
+}
+
+static int
+xe_hwmon_write(struct device *dev, enum hwmon_sensor_types type, u32 attr,
+ int channel, long val)
+{
+ struct xe_hwmon *hwmon = dev_get_drvdata(dev);
+ int ret;
+
+ xe_device_mem_access_get(gt_to_xe(hwmon->gt));
+
+ switch (type) {
+ case hwmon_power:
+ ret = xe_hwmon_power_write(hwmon, attr, channel, val);
+ break;
+ case hwmon_curr:
+ ret = xe_hwmon_curr_write(hwmon, attr, val);
+ break;
+ default:
+ ret = -EOPNOTSUPP;
+ break;
+ }
+
+ xe_device_mem_access_put(gt_to_xe(hwmon->gt));
+
+ return ret;
+}
+
+static const struct hwmon_ops hwmon_ops = {
+ .is_visible = xe_hwmon_is_visible,
+ .read = xe_hwmon_read,
+ .write = xe_hwmon_write,
+};
+
+static const struct hwmon_chip_info hwmon_chip_info = {
+ .ops = &hwmon_ops,
+ .info = hwmon_info,
+};
+
+static void
+xe_hwmon_get_preregistration_info(struct xe_device *xe)
+{
+ struct xe_hwmon *hwmon = xe->hwmon;
+ long energy;
+ u64 val_sku_unit = 0;
+
+ /*
+ * The contents of register PKG_POWER_SKU_UNIT do not change,
+ * so read it once and store the shift values.
+ */
+ if (xe_hwmon_get_reg(hwmon, REG_PKG_POWER_SKU_UNIT)) {
+ xe_hwmon_process_reg(hwmon, REG_PKG_POWER_SKU_UNIT,
+ REG_READ32, &val_sku_unit, 0, 0);
+ hwmon->scl_shift_power = REG_FIELD_GET(PKG_PWR_UNIT, val_sku_unit);
+ hwmon->scl_shift_energy = REG_FIELD_GET(PKG_ENERGY_UNIT, val_sku_unit);
+ hwmon->scl_shift_time = REG_FIELD_GET(PKG_TIME_UNIT, val_sku_unit);
+ }
+
+ /*
+ * Initialize 'struct xe_hwmon_energy_info', i.e. set fields to the
+ * first value of the energy register read
+ */
+ if (xe_hwmon_is_visible(hwmon, hwmon_energy, hwmon_energy_input, 0))
+ xe_hwmon_energy_get(hwmon, &energy);
+}
+
+static void xe_hwmon_mutex_destroy(void *arg)
+{
+ struct xe_hwmon *hwmon = arg;
+
+ mutex_destroy(&hwmon->hwmon_lock);
+}
+
+void xe_hwmon_register(struct xe_device *xe)
+{
+ struct device *dev = xe->drm.dev;
+ struct xe_hwmon *hwmon;
+
+ /* hwmon is available only for dGfx */
+ if (!IS_DGFX(xe))
+ return;
+
+ hwmon = devm_kzalloc(dev, sizeof(*hwmon), GFP_KERNEL);
+ if (!hwmon)
+ return;
+
+ xe->hwmon = hwmon;
+
+ mutex_init(&hwmon->hwmon_lock);
+ if (devm_add_action_or_reset(dev, xe_hwmon_mutex_destroy, hwmon))
+ return;
+
+ /* primary GT to access device level properties */
+ hwmon->gt = xe->tiles[0].primary_gt;
+
+ xe_hwmon_get_preregistration_info(xe);
+
+ drm_dbg(&xe->drm, "Register xe hwmon interface\n");
+
+ /* hwmon_dev points to device hwmon<i> */
+ hwmon->hwmon_dev = devm_hwmon_device_register_with_info(dev, "xe", hwmon,
+ &hwmon_chip_info,
+ hwmon_groups);
+
+ if (IS_ERR(hwmon->hwmon_dev)) {
+ drm_warn(&xe->drm, "Failed to register xe hwmon (%pe)\n", hwmon->hwmon_dev);
+ xe->hwmon = NULL;
+ return;
+ }
+}
+
diff --git a/drivers/gpu/drm/xe/xe_hwmon.h b/drivers/gpu/drm/xe/xe_hwmon.h
new file mode 100644
index 000000000000..c42a1de2cd7a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_hwmon.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_HWMON_H_
+#define _XE_HWMON_H_
+
+#include <linux/types.h>
+
+struct xe_device;
+
+#if IS_REACHABLE(CONFIG_HWMON)
+void xe_hwmon_register(struct xe_device *xe);
+#else
+static inline void xe_hwmon_register(struct xe_device *xe) { };
+#endif
+
+#endif /* _XE_HWMON_H_ */
diff --git a/drivers/gpu/drm/xe/xe_irq.c b/drivers/gpu/drm/xe/xe_irq.c
new file mode 100644
index 000000000000..d1f5ba4bb745
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_irq.c
@@ -0,0 +1,666 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_irq.h"
+
+#include <linux/sched/clock.h>
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_device.h"
+#include "xe_display.h"
+#include "xe_drv.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_hw_engine.h"
+#include "xe_mmio.h"
+
+/*
+ * Interrupt registers for a unit are always consecutive and ordered
+ * ISR, IMR, IIR, IER.
+ */
+#define IMR(offset) XE_REG(offset + 0x4)
+#define IIR(offset) XE_REG(offset + 0x8)
+#define IER(offset) XE_REG(offset + 0xc)
+
+static void assert_iir_is_zero(struct xe_gt *mmio, struct xe_reg reg)
+{
+ u32 val = xe_mmio_read32(mmio, reg);
+
+ if (val == 0)
+ return;
+
+ drm_WARN(&gt_to_xe(mmio)->drm, 1,
+ "Interrupt register 0x%x is not zero: 0x%08x\n",
+ reg.addr, val);
+ xe_mmio_write32(mmio, reg, 0xffffffff);
+ xe_mmio_read32(mmio, reg);
+ xe_mmio_write32(mmio, reg, 0xffffffff);
+ xe_mmio_read32(mmio, reg);
+}
+
+/*
+ * Unmask and enable the specified interrupts. Does not check current state,
+ * so any bits not specified here will become masked and disabled.
+ */
+static void unmask_and_enable(struct xe_tile *tile, u32 irqregs, u32 bits)
+{
+ struct xe_gt *mmio = tile->primary_gt;
+
+ /*
+ * If we're just enabling an interrupt now, it shouldn't already
+ * be raised in the IIR.
+ */
+ assert_iir_is_zero(mmio, IIR(irqregs));
+
+ xe_mmio_write32(mmio, IER(irqregs), bits);
+ xe_mmio_write32(mmio, IMR(irqregs), ~bits);
+
+ /* Posting read */
+ xe_mmio_read32(mmio, IMR(irqregs));
+}
+
+/* Mask and disable all interrupts. */
+static void mask_and_disable(struct xe_tile *tile, u32 irqregs)
+{
+ struct xe_gt *mmio = tile->primary_gt;
+
+ xe_mmio_write32(mmio, IMR(irqregs), ~0);
+ /* Posting read */
+ xe_mmio_read32(mmio, IMR(irqregs));
+
+ xe_mmio_write32(mmio, IER(irqregs), 0);
+
+ /* IIR can theoretically queue up two events. Be paranoid. */
+ xe_mmio_write32(mmio, IIR(irqregs), ~0);
+ xe_mmio_read32(mmio, IIR(irqregs));
+ xe_mmio_write32(mmio, IIR(irqregs), ~0);
+ xe_mmio_read32(mmio, IIR(irqregs));
+}
+
+static u32 xelp_intr_disable(struct xe_device *xe)
+{
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+
+ xe_mmio_write32(mmio, GFX_MSTR_IRQ, 0);
+
+ /*
+ * Now with master disabled, get a sample of level indications
+ * for this interrupt. Indications will be cleared on related acks.
+ * New indications can and will light up during processing,
+ * and will generate new interrupt after enabling master.
+ */
+ return xe_mmio_read32(mmio, GFX_MSTR_IRQ);
+}
+
+static u32
+gu_misc_irq_ack(struct xe_device *xe, const u32 master_ctl)
+{
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+ u32 iir;
+
+ if (!(master_ctl & GU_MISC_IRQ))
+ return 0;
+
+ iir = xe_mmio_read32(mmio, IIR(GU_MISC_IRQ_OFFSET));
+ if (likely(iir))
+ xe_mmio_write32(mmio, IIR(GU_MISC_IRQ_OFFSET), iir);
+
+ return iir;
+}
+
+static inline void xelp_intr_enable(struct xe_device *xe, bool stall)
+{
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+
+ xe_mmio_write32(mmio, GFX_MSTR_IRQ, MASTER_IRQ);
+ if (stall)
+ xe_mmio_read32(mmio, GFX_MSTR_IRQ);
+}
+
+/* Enable/unmask the HWE interrupts for a specific GT's engines. */
+void xe_irq_enable_hwe(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 ccs_mask, bcs_mask;
+ u32 irqs, dmask, smask;
+ u32 gsc_mask = 0;
+
+ if (xe_device_uc_enabled(xe)) {
+ irqs = GT_RENDER_USER_INTERRUPT |
+ GT_RENDER_PIPECTL_NOTIFY_INTERRUPT;
+ } else {
+ irqs = GT_RENDER_USER_INTERRUPT |
+ GT_CS_MASTER_ERROR_INTERRUPT |
+ GT_CONTEXT_SWITCH_INTERRUPT |
+ GT_WAIT_SEMAPHORE_INTERRUPT;
+ }
+
+ ccs_mask = xe_hw_engine_mask_per_class(gt, XE_ENGINE_CLASS_COMPUTE);
+ bcs_mask = xe_hw_engine_mask_per_class(gt, XE_ENGINE_CLASS_COPY);
+
+ dmask = irqs << 16 | irqs;
+ smask = irqs << 16;
+
+ if (!xe_gt_is_media_type(gt)) {
+ /* Enable interrupts for each engine class */
+ xe_mmio_write32(gt, RENDER_COPY_INTR_ENABLE, dmask);
+ if (ccs_mask)
+ xe_mmio_write32(gt, CCS_RSVD_INTR_ENABLE, smask);
+
+ /* Unmask interrupts for each engine instance */
+ xe_mmio_write32(gt, RCS0_RSVD_INTR_MASK, ~smask);
+ xe_mmio_write32(gt, BCS_RSVD_INTR_MASK, ~smask);
+ if (bcs_mask & (BIT(1)|BIT(2)))
+ xe_mmio_write32(gt, XEHPC_BCS1_BCS2_INTR_MASK, ~dmask);
+ if (bcs_mask & (BIT(3)|BIT(4)))
+ xe_mmio_write32(gt, XEHPC_BCS3_BCS4_INTR_MASK, ~dmask);
+ if (bcs_mask & (BIT(5)|BIT(6)))
+ xe_mmio_write32(gt, XEHPC_BCS5_BCS6_INTR_MASK, ~dmask);
+ if (bcs_mask & (BIT(7)|BIT(8)))
+ xe_mmio_write32(gt, XEHPC_BCS7_BCS8_INTR_MASK, ~dmask);
+ if (ccs_mask & (BIT(0)|BIT(1)))
+ xe_mmio_write32(gt, CCS0_CCS1_INTR_MASK, ~dmask);
+ if (ccs_mask & (BIT(2)|BIT(3)))
+ xe_mmio_write32(gt, CCS2_CCS3_INTR_MASK, ~dmask);
+ }
+
+ if (xe_gt_is_media_type(gt) || MEDIA_VER(xe) < 13) {
+ /* Enable interrupts for each engine class */
+ xe_mmio_write32(gt, VCS_VECS_INTR_ENABLE, dmask);
+
+ /* Unmask interrupts for each engine instance */
+ xe_mmio_write32(gt, VCS0_VCS1_INTR_MASK, ~dmask);
+ xe_mmio_write32(gt, VCS2_VCS3_INTR_MASK, ~dmask);
+ xe_mmio_write32(gt, VECS0_VECS1_INTR_MASK, ~dmask);
+
+ if (xe_hw_engine_mask_per_class(gt, XE_ENGINE_CLASS_OTHER))
+ gsc_mask = irqs;
+ else if (HAS_HECI_GSCFI(xe))
+ gsc_mask = GSC_IRQ_INTF(1);
+ if (gsc_mask) {
+ xe_mmio_write32(gt, GUNIT_GSC_INTR_ENABLE, gsc_mask);
+ xe_mmio_write32(gt, GUNIT_GSC_INTR_MASK, ~gsc_mask);
+ }
+ }
+}
+
+static u32
+gt_engine_identity(struct xe_device *xe,
+ struct xe_gt *mmio,
+ const unsigned int bank,
+ const unsigned int bit)
+{
+ u32 timeout_ts;
+ u32 ident;
+
+ lockdep_assert_held(&xe->irq.lock);
+
+ xe_mmio_write32(mmio, IIR_REG_SELECTOR(bank), BIT(bit));
+
+ /*
+ * NB: Specs do not specify how long to spin wait,
+ * so we do ~100us as an educated guess.
+ */
+ timeout_ts = (local_clock() >> 10) + 100;
+ do {
+ ident = xe_mmio_read32(mmio, INTR_IDENTITY_REG(bank));
+ } while (!(ident & INTR_DATA_VALID) &&
+ !time_after32(local_clock() >> 10, timeout_ts));
+
+ if (unlikely(!(ident & INTR_DATA_VALID))) {
+ drm_err(&xe->drm, "INTR_IDENTITY_REG%u:%u 0x%08x not valid!\n",
+ bank, bit, ident);
+ return 0;
+ }
+
+ xe_mmio_write32(mmio, INTR_IDENTITY_REG(bank), ident);
+
+ return ident;
+}
+
+#define OTHER_MEDIA_GUC_INSTANCE 16
+
+static void
+gt_other_irq_handler(struct xe_gt *gt, const u8 instance, const u16 iir)
+{
+ if (instance == OTHER_GUC_INSTANCE && !xe_gt_is_media_type(gt))
+ return xe_guc_irq_handler(&gt->uc.guc, iir);
+ if (instance == OTHER_MEDIA_GUC_INSTANCE && xe_gt_is_media_type(gt))
+ return xe_guc_irq_handler(&gt->uc.guc, iir);
+
+ if (instance != OTHER_GUC_INSTANCE &&
+ instance != OTHER_MEDIA_GUC_INSTANCE) {
+ WARN_ONCE(1, "unhandled other interrupt instance=0x%x, iir=0x%x\n",
+ instance, iir);
+ }
+}
+
+static struct xe_gt *pick_engine_gt(struct xe_tile *tile,
+ enum xe_engine_class class,
+ unsigned int instance)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+
+ if (MEDIA_VER(xe) < 13)
+ return tile->primary_gt;
+
+ if (class == XE_ENGINE_CLASS_VIDEO_DECODE ||
+ class == XE_ENGINE_CLASS_VIDEO_ENHANCE)
+ return tile->media_gt;
+
+ if (class == XE_ENGINE_CLASS_OTHER &&
+ (instance == OTHER_MEDIA_GUC_INSTANCE || instance == OTHER_GSC_INSTANCE))
+ return tile->media_gt;
+
+ return tile->primary_gt;
+}
+
+static void gt_irq_handler(struct xe_tile *tile,
+ u32 master_ctl, unsigned long *intr_dw,
+ u32 *identity)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_gt *mmio = tile->primary_gt;
+ unsigned int bank, bit;
+ u16 instance, intr_vec;
+ enum xe_engine_class class;
+ struct xe_hw_engine *hwe;
+
+ spin_lock(&xe->irq.lock);
+
+ for (bank = 0; bank < 2; bank++) {
+ if (!(master_ctl & GT_DW_IRQ(bank)))
+ continue;
+
+ intr_dw[bank] = xe_mmio_read32(mmio, GT_INTR_DW(bank));
+ for_each_set_bit(bit, intr_dw + bank, 32)
+ identity[bit] = gt_engine_identity(xe, mmio, bank, bit);
+ xe_mmio_write32(mmio, GT_INTR_DW(bank), intr_dw[bank]);
+
+ for_each_set_bit(bit, intr_dw + bank, 32) {
+ struct xe_gt *engine_gt;
+
+ class = INTR_ENGINE_CLASS(identity[bit]);
+ instance = INTR_ENGINE_INSTANCE(identity[bit]);
+ intr_vec = INTR_ENGINE_INTR(identity[bit]);
+
+ engine_gt = pick_engine_gt(tile, class, instance);
+
+ hwe = xe_gt_hw_engine(engine_gt, class, instance, false);
+ if (hwe) {
+ xe_hw_engine_handle_irq(hwe, intr_vec);
+ continue;
+ }
+
+ if (class == XE_ENGINE_CLASS_OTHER) {
+ /* HECI GSCFI interrupts come from outside of GT */
+ if (HAS_HECI_GSCFI(xe) && instance == OTHER_GSC_INSTANCE)
+ xe_heci_gsc_irq_handler(xe, intr_vec);
+ else
+ gt_other_irq_handler(engine_gt, instance, intr_vec);
+ continue;
+ }
+ }
+ }
+
+ spin_unlock(&xe->irq.lock);
+}
+
+/*
+ * Top-level interrupt handler for Xe_LP platforms (which did not have
+ * a "master tile" interrupt register.
+ */
+static irqreturn_t xelp_irq_handler(int irq, void *arg)
+{
+ struct xe_device *xe = arg;
+ struct xe_tile *tile = xe_device_get_root_tile(xe);
+ u32 master_ctl, gu_misc_iir;
+ unsigned long intr_dw[2];
+ u32 identity[32];
+
+ spin_lock(&xe->irq.lock);
+ if (!xe->irq.enabled) {
+ spin_unlock(&xe->irq.lock);
+ return IRQ_NONE;
+ }
+ spin_unlock(&xe->irq.lock);
+
+ master_ctl = xelp_intr_disable(xe);
+ if (!master_ctl) {
+ xelp_intr_enable(xe, false);
+ return IRQ_NONE;
+ }
+
+ gt_irq_handler(tile, master_ctl, intr_dw, identity);
+
+ xe_display_irq_handler(xe, master_ctl);
+
+ gu_misc_iir = gu_misc_irq_ack(xe, master_ctl);
+
+ xelp_intr_enable(xe, false);
+
+ xe_display_irq_enable(xe, gu_misc_iir);
+
+ return IRQ_HANDLED;
+}
+
+static u32 dg1_intr_disable(struct xe_device *xe)
+{
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+ u32 val;
+
+ /* First disable interrupts */
+ xe_mmio_write32(mmio, DG1_MSTR_TILE_INTR, 0);
+
+ /* Get the indication levels and ack the master unit */
+ val = xe_mmio_read32(mmio, DG1_MSTR_TILE_INTR);
+ if (unlikely(!val))
+ return 0;
+
+ xe_mmio_write32(mmio, DG1_MSTR_TILE_INTR, val);
+
+ return val;
+}
+
+static void dg1_intr_enable(struct xe_device *xe, bool stall)
+{
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+
+ xe_mmio_write32(mmio, DG1_MSTR_TILE_INTR, DG1_MSTR_IRQ);
+ if (stall)
+ xe_mmio_read32(mmio, DG1_MSTR_TILE_INTR);
+}
+
+/*
+ * Top-level interrupt handler for Xe_LP+ and beyond. These platforms have
+ * a "master tile" interrupt register which must be consulted before the
+ * "graphics master" interrupt register.
+ */
+static irqreturn_t dg1_irq_handler(int irq, void *arg)
+{
+ struct xe_device *xe = arg;
+ struct xe_tile *tile;
+ u32 master_tile_ctl, master_ctl = 0, gu_misc_iir = 0;
+ unsigned long intr_dw[2];
+ u32 identity[32];
+ u8 id;
+
+ /* TODO: This really shouldn't be copied+pasted */
+
+ spin_lock(&xe->irq.lock);
+ if (!xe->irq.enabled) {
+ spin_unlock(&xe->irq.lock);
+ return IRQ_NONE;
+ }
+ spin_unlock(&xe->irq.lock);
+
+ master_tile_ctl = dg1_intr_disable(xe);
+ if (!master_tile_ctl) {
+ dg1_intr_enable(xe, false);
+ return IRQ_NONE;
+ }
+
+ for_each_tile(tile, xe, id) {
+ struct xe_gt *mmio = tile->primary_gt;
+
+ if ((master_tile_ctl & DG1_MSTR_TILE(tile->id)) == 0)
+ continue;
+
+ master_ctl = xe_mmio_read32(mmio, GFX_MSTR_IRQ);
+
+ /*
+ * We might be in irq handler just when PCIe DPC is initiated
+ * and all MMIO reads will be returned with all 1's. Ignore this
+ * irq as device is inaccessible.
+ */
+ if (master_ctl == REG_GENMASK(31, 0)) {
+ dev_dbg(tile_to_xe(tile)->drm.dev,
+ "Ignore this IRQ as device might be in DPC containment.\n");
+ return IRQ_HANDLED;
+ }
+
+ xe_mmio_write32(mmio, GFX_MSTR_IRQ, master_ctl);
+
+ gt_irq_handler(tile, master_ctl, intr_dw, identity);
+
+ /*
+ * Display interrupts (including display backlight operations
+ * that get reported as Gunit GSE) would only be hooked up to
+ * the primary tile.
+ */
+ if (id == 0) {
+ xe_display_irq_handler(xe, master_ctl);
+ gu_misc_iir = gu_misc_irq_ack(xe, master_ctl);
+ }
+ }
+
+ dg1_intr_enable(xe, false);
+ xe_display_irq_enable(xe, gu_misc_iir);
+
+ return IRQ_HANDLED;
+}
+
+static void gt_irq_reset(struct xe_tile *tile)
+{
+ struct xe_gt *mmio = tile->primary_gt;
+
+ u32 ccs_mask = xe_hw_engine_mask_per_class(tile->primary_gt,
+ XE_ENGINE_CLASS_COMPUTE);
+ u32 bcs_mask = xe_hw_engine_mask_per_class(tile->primary_gt,
+ XE_ENGINE_CLASS_COPY);
+
+ /* Disable RCS, BCS, VCS and VECS class engines. */
+ xe_mmio_write32(mmio, RENDER_COPY_INTR_ENABLE, 0);
+ xe_mmio_write32(mmio, VCS_VECS_INTR_ENABLE, 0);
+ if (ccs_mask)
+ xe_mmio_write32(mmio, CCS_RSVD_INTR_ENABLE, 0);
+
+ /* Restore masks irqs on RCS, BCS, VCS and VECS engines. */
+ xe_mmio_write32(mmio, RCS0_RSVD_INTR_MASK, ~0);
+ xe_mmio_write32(mmio, BCS_RSVD_INTR_MASK, ~0);
+ if (bcs_mask & (BIT(1)|BIT(2)))
+ xe_mmio_write32(mmio, XEHPC_BCS1_BCS2_INTR_MASK, ~0);
+ if (bcs_mask & (BIT(3)|BIT(4)))
+ xe_mmio_write32(mmio, XEHPC_BCS3_BCS4_INTR_MASK, ~0);
+ if (bcs_mask & (BIT(5)|BIT(6)))
+ xe_mmio_write32(mmio, XEHPC_BCS5_BCS6_INTR_MASK, ~0);
+ if (bcs_mask & (BIT(7)|BIT(8)))
+ xe_mmio_write32(mmio, XEHPC_BCS7_BCS8_INTR_MASK, ~0);
+ xe_mmio_write32(mmio, VCS0_VCS1_INTR_MASK, ~0);
+ xe_mmio_write32(mmio, VCS2_VCS3_INTR_MASK, ~0);
+ xe_mmio_write32(mmio, VECS0_VECS1_INTR_MASK, ~0);
+ if (ccs_mask & (BIT(0)|BIT(1)))
+ xe_mmio_write32(mmio, CCS0_CCS1_INTR_MASK, ~0);
+ if (ccs_mask & (BIT(2)|BIT(3)))
+ xe_mmio_write32(mmio, CCS2_CCS3_INTR_MASK, ~0);
+
+ if ((tile->media_gt &&
+ xe_hw_engine_mask_per_class(tile->media_gt, XE_ENGINE_CLASS_OTHER)) ||
+ HAS_HECI_GSCFI(tile_to_xe(tile))) {
+ xe_mmio_write32(mmio, GUNIT_GSC_INTR_ENABLE, 0);
+ xe_mmio_write32(mmio, GUNIT_GSC_INTR_MASK, ~0);
+ }
+
+ xe_mmio_write32(mmio, GPM_WGBOXPERF_INTR_ENABLE, 0);
+ xe_mmio_write32(mmio, GPM_WGBOXPERF_INTR_MASK, ~0);
+ xe_mmio_write32(mmio, GUC_SG_INTR_ENABLE, 0);
+ xe_mmio_write32(mmio, GUC_SG_INTR_MASK, ~0);
+}
+
+static void xelp_irq_reset(struct xe_tile *tile)
+{
+ xelp_intr_disable(tile_to_xe(tile));
+
+ gt_irq_reset(tile);
+
+ mask_and_disable(tile, PCU_IRQ_OFFSET);
+}
+
+static void dg1_irq_reset(struct xe_tile *tile)
+{
+ if (tile->id == 0)
+ dg1_intr_disable(tile_to_xe(tile));
+
+ gt_irq_reset(tile);
+
+ mask_and_disable(tile, PCU_IRQ_OFFSET);
+}
+
+static void dg1_irq_reset_mstr(struct xe_tile *tile)
+{
+ struct xe_gt *mmio = tile->primary_gt;
+
+ xe_mmio_write32(mmio, GFX_MSTR_IRQ, ~0);
+}
+
+static void xe_irq_reset(struct xe_device *xe)
+{
+ struct xe_tile *tile;
+ u8 id;
+
+ for_each_tile(tile, xe, id) {
+ if (GRAPHICS_VERx100(xe) >= 1210)
+ dg1_irq_reset(tile);
+ else
+ xelp_irq_reset(tile);
+ }
+
+ tile = xe_device_get_root_tile(xe);
+ mask_and_disable(tile, GU_MISC_IRQ_OFFSET);
+ xe_display_irq_reset(xe);
+
+ /*
+ * The tile's top-level status register should be the last one
+ * to be reset to avoid possible bit re-latching from lower
+ * level interrupts.
+ */
+ if (GRAPHICS_VERx100(xe) >= 1210) {
+ for_each_tile(tile, xe, id)
+ dg1_irq_reset_mstr(tile);
+ }
+}
+
+static void xe_irq_postinstall(struct xe_device *xe)
+{
+ xe_display_irq_postinstall(xe, xe_root_mmio_gt(xe));
+
+ /*
+ * ASLE backlight operations are reported via GUnit GSE interrupts
+ * on the root tile.
+ */
+ unmask_and_enable(xe_device_get_root_tile(xe),
+ GU_MISC_IRQ_OFFSET, GU_MISC_GSE);
+
+ /* Enable top-level interrupts */
+ if (GRAPHICS_VERx100(xe) >= 1210)
+ dg1_intr_enable(xe, true);
+ else
+ xelp_intr_enable(xe, true);
+}
+
+static irq_handler_t xe_irq_handler(struct xe_device *xe)
+{
+ if (GRAPHICS_VERx100(xe) >= 1210)
+ return dg1_irq_handler;
+ else
+ return xelp_irq_handler;
+}
+
+static void irq_uninstall(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = arg;
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ int irq;
+
+ if (!xe->irq.enabled)
+ return;
+
+ xe->irq.enabled = false;
+ xe_irq_reset(xe);
+
+ irq = pci_irq_vector(pdev, 0);
+ free_irq(irq, xe);
+}
+
+int xe_irq_install(struct xe_device *xe)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ irq_handler_t irq_handler;
+ int err, irq;
+
+ irq_handler = xe_irq_handler(xe);
+ if (!irq_handler) {
+ drm_err(&xe->drm, "No supported interrupt handler");
+ return -EINVAL;
+ }
+
+ xe_irq_reset(xe);
+
+ err = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI | PCI_IRQ_MSIX);
+ if (err < 0) {
+ drm_err(&xe->drm, "MSI/MSIX: Failed to enable support %d\n", err);
+ return err;
+ }
+
+ irq = pci_irq_vector(pdev, 0);
+ err = request_irq(irq, irq_handler, IRQF_SHARED, DRIVER_NAME, xe);
+ if (err < 0) {
+ drm_err(&xe->drm, "Failed to request MSI/MSIX IRQ %d\n", err);
+ return err;
+ }
+
+ xe->irq.enabled = true;
+
+ xe_irq_postinstall(xe);
+
+ err = drmm_add_action_or_reset(&xe->drm, irq_uninstall, xe);
+ if (err)
+ goto free_irq_handler;
+
+ return 0;
+
+free_irq_handler:
+ free_irq(irq, xe);
+
+ return err;
+}
+
+void xe_irq_shutdown(struct xe_device *xe)
+{
+ irq_uninstall(&xe->drm, xe);
+}
+
+void xe_irq_suspend(struct xe_device *xe)
+{
+ int irq = to_pci_dev(xe->drm.dev)->irq;
+
+ spin_lock_irq(&xe->irq.lock);
+ xe->irq.enabled = false; /* no new irqs */
+ spin_unlock_irq(&xe->irq.lock);
+
+ synchronize_irq(irq); /* flush irqs */
+ xe_irq_reset(xe); /* turn irqs off */
+}
+
+void xe_irq_resume(struct xe_device *xe)
+{
+ struct xe_gt *gt;
+ int id;
+
+ /*
+ * lock not needed:
+ * 1. no irq will arrive before the postinstall
+ * 2. display is not yet resumed
+ */
+ xe->irq.enabled = true;
+ xe_irq_reset(xe);
+ xe_irq_postinstall(xe); /* turn irqs on */
+
+ for_each_gt(gt, xe, id)
+ xe_irq_enable_hwe(gt);
+}
diff --git a/drivers/gpu/drm/xe/xe_irq.h b/drivers/gpu/drm/xe/xe_irq.h
new file mode 100644
index 000000000000..bc42bc90d967
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_irq.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_IRQ_H_
+#define _XE_IRQ_H_
+
+struct xe_device;
+struct xe_tile;
+struct xe_gt;
+
+int xe_irq_install(struct xe_device *xe);
+void xe_irq_shutdown(struct xe_device *xe);
+void xe_irq_suspend(struct xe_device *xe);
+void xe_irq_resume(struct xe_device *xe);
+void xe_irq_enable_hwe(struct xe_gt *gt);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_lmtt.c b/drivers/gpu/drm/xe/xe_lmtt.c
new file mode 100644
index 000000000000..0d7c5514e092
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lmtt.c
@@ -0,0 +1,506 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/align.h>
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_sriov_regs.h"
+
+#include "xe_assert.h"
+#include "xe_bo.h"
+#include "xe_lmtt.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_res_cursor.h"
+#include "xe_sriov.h"
+#include "xe_sriov_printk.h"
+
+/**
+ * DOC: Local Memory Translation Table
+ *
+ * The Local Memory Translation Table (LMTT) provides additional abstraction
+ * when Virtual Function (VF) is accessing device Local Memory (VRAM).
+ *
+ * The Root LMTT Page Directory contains one entry for each VF. Entries are
+ * indexed by the function number (1-based, index 0 is unused).
+ *
+ * See `Two-Level LMTT Structure`_ and `Multi-Level LMTT Structure`_.
+ */
+
+#define lmtt_assert(lmtt, condition) xe_tile_assert(lmtt_to_tile(lmtt), condition)
+#define lmtt_debug(lmtt, msg...) xe_sriov_dbg_verbose(lmtt_to_xe(lmtt), "LMTT: " msg)
+
+static bool xe_has_multi_level_lmtt(struct xe_device *xe)
+{
+ return xe->info.platform == XE_PVC;
+}
+
+static struct xe_tile *lmtt_to_tile(struct xe_lmtt *lmtt)
+{
+ return container_of(lmtt, struct xe_tile, sriov.pf.lmtt);
+}
+
+static struct xe_device *lmtt_to_xe(struct xe_lmtt *lmtt)
+{
+ return tile_to_xe(lmtt_to_tile(lmtt));
+}
+
+static u64 lmtt_page_size(struct xe_lmtt *lmtt)
+{
+ return BIT_ULL(lmtt->ops->lmtt_pte_shift(0));
+}
+
+static struct xe_lmtt_pt *lmtt_pt_alloc(struct xe_lmtt *lmtt, unsigned int level)
+{
+ unsigned int num_entries = level ? lmtt->ops->lmtt_pte_num(level) : 0;
+ struct xe_lmtt_pt *pt;
+ struct xe_bo *bo;
+ int err;
+
+ pt = kzalloc(struct_size(pt, entries, num_entries), GFP_KERNEL);
+ if (!pt) {
+ err = -ENOMEM;
+ goto out;
+ }
+
+ bo = xe_bo_create_pin_map(lmtt_to_xe(lmtt), lmtt_to_tile(lmtt), NULL,
+ PAGE_ALIGN(lmtt->ops->lmtt_pte_size(level) *
+ lmtt->ops->lmtt_pte_num(level)),
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(lmtt_to_tile(lmtt)) |
+ XE_BO_CREATE_PINNED_BIT);
+ if (IS_ERR(bo)) {
+ err = PTR_ERR(bo);
+ goto out_free_pt;
+ }
+
+ lmtt_assert(lmtt, xe_bo_is_vram(bo));
+
+ pt->level = level;
+ pt->bo = bo;
+ return pt;
+
+out_free_pt:
+ kfree(pt);
+out:
+ return ERR_PTR(err);
+}
+
+static void lmtt_pt_free(struct xe_lmtt_pt *pt)
+{
+ xe_bo_unpin_map_no_vm(pt->bo);
+ kfree(pt);
+}
+
+static int lmtt_init_pd(struct xe_lmtt *lmtt)
+{
+ struct xe_lmtt_pt *pd;
+
+ lmtt_assert(lmtt, !lmtt->pd);
+ lmtt_assert(lmtt, lmtt->ops->lmtt_root_pd_level());
+
+ pd = lmtt_pt_alloc(lmtt, lmtt->ops->lmtt_root_pd_level());
+ if (IS_ERR(pd))
+ return PTR_ERR(pd);
+
+ lmtt->pd = pd;
+ return 0;
+}
+
+static void lmtt_fini_pd(struct xe_lmtt *lmtt)
+{
+ struct xe_lmtt_pt *pd = lmtt->pd;
+ unsigned int num_entries = lmtt->ops->lmtt_pte_num(pd->level);
+ unsigned int n = 0;
+
+ /* make sure we don't leak */
+ for (n = 0; n < num_entries; n++)
+ lmtt_assert(lmtt, !pd->entries[n]);
+
+ lmtt->pd = NULL;
+ lmtt_pt_free(pd);
+}
+
+static void fini_lmtt(struct drm_device *drm, void *arg)
+{
+ struct xe_lmtt *lmtt = arg;
+
+ lmtt_assert(lmtt, !(!!lmtt->ops ^ !!lmtt->pd));
+
+ if (!lmtt->pd)
+ return;
+
+ lmtt_fini_pd(lmtt);
+ lmtt->ops = NULL;
+}
+
+/**
+ * xe_lmtt_init - LMTT software initialization.
+ * @lmtt: the &xe_lmtt to initialize
+ *
+ * The LMTT initialization requires two steps.
+ *
+ * The xe_lmtt_init() checks if LMTT is required on current device and selects
+ * and initialize proper variant of the LMTT Root Directory. Currently supported
+ * variants are `Two-Level LMTT Structure`_ and `Multi-Level LMTT Structure`_.
+ *
+ * In next step xe_lmtt_init_hw() will register this directory on the hardware.
+ *
+ * Notes:
+ * The LMTT allocations are managed and will be implicitly released on driver unload.
+ * This function shall be called only once and only when running as a PF driver.
+ * Any LMTT initialization failure should block VFs enabling.
+ *
+ * Return: 0 on success or a negative error code on failure.
+ */
+int xe_lmtt_init(struct xe_lmtt *lmtt)
+{
+ struct xe_device *xe = lmtt_to_xe(lmtt);
+ int err;
+
+ lmtt_assert(lmtt, IS_SRIOV_PF(xe));
+ lmtt_assert(lmtt, !lmtt->ops);
+
+ if (!IS_DGFX(xe))
+ return 0;
+
+ if (xe_has_multi_level_lmtt(xe))
+ lmtt->ops = &lmtt_ml_ops;
+ else
+ lmtt->ops = &lmtt_2l_ops;
+
+ err = lmtt_init_pd(lmtt);
+ if (unlikely(err))
+ goto fail;
+
+ return drmm_add_action_or_reset(&xe->drm, fini_lmtt, lmtt);
+
+fail:
+ lmtt->ops = NULL;
+ return err;
+}
+
+static void lmtt_setup_dir_ptr(struct xe_lmtt *lmtt)
+{
+ struct xe_tile *tile = lmtt_to_tile(lmtt);
+ struct xe_device *xe = tile_to_xe(tile);
+ dma_addr_t offset = xe_bo_main_addr(lmtt->pd->bo, XE_PAGE_SIZE);
+
+ lmtt_debug(lmtt, "DIR offset %pad\n", &offset);
+ lmtt_assert(lmtt, xe_bo_is_vram(lmtt->pd->bo));
+ lmtt_assert(lmtt, IS_ALIGNED(offset, SZ_64K));
+
+ xe_mmio_write32(tile->primary_gt,
+ GRAPHICS_VER(xe) >= 20 ? XE2_LMEM_CFG : LMEM_CFG,
+ LMEM_EN | REG_FIELD_PREP(LMTT_DIR_PTR, offset / SZ_64K));
+}
+
+/**
+ * xe_lmtt_init_hw - Perform LMTT hardware initialization.
+ * @lmtt: the &xe_lmtt to initialize
+ *
+ * This function is a second step of the LMTT initialization.
+ * This function registers LMTT Root Directory prepared in xe_lmtt_init().
+ *
+ * This function shall be called after every hardware reset.
+ * This function shall be called only when running as a PF driver.
+ */
+void xe_lmtt_init_hw(struct xe_lmtt *lmtt)
+{
+ if (!lmtt->pd)
+ return;
+
+ lmtt_setup_dir_ptr(lmtt);
+}
+
+static void lmtt_write_pte(struct xe_lmtt *lmtt, struct xe_lmtt_pt *pt,
+ u64 pte, unsigned int idx)
+{
+ unsigned int level = pt->level;
+
+ lmtt_assert(lmtt, idx <= lmtt->ops->lmtt_pte_num(level));
+ lmtt_debug(lmtt, "WRITE level=%u index=%u pte=%#llx\n", level, idx, pte);
+
+ switch (lmtt->ops->lmtt_pte_size(level)) {
+ case sizeof(u32):
+ xe_map_wr(lmtt_to_xe(lmtt), &pt->bo->vmap, idx * sizeof(u32), u32, pte);
+ break;
+ case sizeof(u64):
+ xe_map_wr(lmtt_to_xe(lmtt), &pt->bo->vmap, idx * sizeof(u64), u64, pte);
+ break;
+ default:
+ lmtt_assert(lmtt, !!!"invalid pte size");
+ }
+}
+
+static void lmtt_destroy_pt(struct xe_lmtt *lmtt, struct xe_lmtt_pt *pd)
+{
+ unsigned int num_entries = pd->level ? lmtt->ops->lmtt_pte_num(pd->level) : 0;
+ struct xe_lmtt_pt *pt;
+ unsigned int i;
+
+ for (i = 0; i < num_entries; i++) {
+ pt = pd->entries[i];
+ pd->entries[i] = NULL;
+ if (!pt)
+ continue;
+
+ lmtt_destroy_pt(lmtt, pt);
+ }
+
+ lmtt_pt_free(pd);
+}
+
+static void lmtt_drop_pages(struct xe_lmtt *lmtt, unsigned int vfid)
+{
+ struct xe_lmtt_pt *pd = lmtt->pd;
+ struct xe_lmtt_pt *pt;
+
+ pt = pd->entries[vfid];
+ pd->entries[vfid] = NULL;
+ if (!pt)
+ return;
+
+ lmtt_write_pte(lmtt, pd, LMTT_PTE_INVALID, vfid);
+
+ lmtt_assert(lmtt, pd->level > 0);
+ lmtt_assert(lmtt, pt->level == pd->level - 1);
+ lmtt_destroy_pt(lmtt, pt);
+}
+
+static int __lmtt_alloc_range(struct xe_lmtt *lmtt, struct xe_lmtt_pt *pd,
+ u64 start, u64 end)
+{
+ u64 pte_addr_shift = BIT_ULL(lmtt->ops->lmtt_pte_shift(pd->level));
+ u64 offset;
+ int err;
+
+ lmtt_assert(lmtt, pd->level > 0);
+
+ offset = start;
+ while (offset < end) {
+ struct xe_lmtt_pt *pt;
+ u64 next, pde, pt_addr;
+ unsigned int idx;
+
+ pt = lmtt_pt_alloc(lmtt, pd->level - 1);
+ if (IS_ERR(pt))
+ return PTR_ERR(pt);
+
+ pt_addr = xe_bo_main_addr(pt->bo, XE_PAGE_SIZE);
+
+ idx = lmtt->ops->lmtt_pte_index(offset, pd->level);
+ pde = lmtt->ops->lmtt_pte_encode(pt_addr, pd->level);
+
+ lmtt_write_pte(lmtt, pd, pde, idx);
+
+ pd->entries[idx] = pt;
+
+ next = min(end, round_up(offset + 1, pte_addr_shift));
+
+ if (pt->level != 0) {
+ err = __lmtt_alloc_range(lmtt, pt, offset, next);
+ if (err)
+ return err;
+ }
+
+ offset = next;
+ }
+
+ return 0;
+}
+
+static int lmtt_alloc_range(struct xe_lmtt *lmtt, unsigned int vfid, u64 start, u64 end)
+{
+ struct xe_lmtt_pt *pd = lmtt->pd;
+ struct xe_lmtt_pt *pt;
+ u64 pt_addr;
+ u64 pde;
+ int err;
+
+ lmtt_assert(lmtt, pd->level > 0);
+ lmtt_assert(lmtt, vfid <= lmtt->ops->lmtt_pte_num(pd->level));
+ lmtt_assert(lmtt, IS_ALIGNED(start, lmtt_page_size(lmtt)));
+ lmtt_assert(lmtt, IS_ALIGNED(end, lmtt_page_size(lmtt)));
+
+ if (pd->entries[vfid])
+ return -ENOTEMPTY;
+
+ pt = lmtt_pt_alloc(lmtt, pd->level - 1);
+ if (IS_ERR(pt))
+ return PTR_ERR(pt);
+
+ pt_addr = xe_bo_main_addr(pt->bo, XE_PAGE_SIZE);
+
+ pde = lmtt->ops->lmtt_pte_encode(pt_addr, pd->level);
+
+ lmtt_write_pte(lmtt, pd, pde, vfid);
+
+ pd->entries[vfid] = pt;
+
+ if (pt->level != 0) {
+ err = __lmtt_alloc_range(lmtt, pt, start, end);
+ if (err)
+ goto out_free_pt;
+ }
+
+ return 0;
+
+out_free_pt:
+ lmtt_pt_free(pt);
+ return err;
+}
+
+static struct xe_lmtt_pt *lmtt_leaf_pt(struct xe_lmtt *lmtt, unsigned int vfid, u64 addr)
+{
+ struct xe_lmtt_pt *pd = lmtt->pd;
+ struct xe_lmtt_pt *pt;
+
+ lmtt_assert(lmtt, vfid <= lmtt->ops->lmtt_pte_num(pd->level));
+ pt = pd->entries[vfid];
+
+ while (pt->level) {
+ lmtt_assert(lmtt, lmtt->ops->lmtt_pte_index(addr, pt->level) <=
+ lmtt->ops->lmtt_pte_num(pt->level));
+
+ pt = pt->entries[lmtt->ops->lmtt_pte_index(addr, pt->level)];
+
+ addr >>= lmtt->ops->lmtt_pte_shift(pt->level);
+ }
+
+ lmtt_assert(lmtt, lmtt->ops->lmtt_pte_index(addr, pt->level) <=
+ lmtt->ops->lmtt_pte_num(pt->level));
+ lmtt_assert(lmtt, pt->level != pd->level);
+ lmtt_assert(lmtt, pt->level == 0);
+ return pt;
+}
+
+static void lmtt_insert_bo(struct xe_lmtt *lmtt, unsigned int vfid, struct xe_bo *bo, u64 start)
+{
+ u64 page_size = lmtt_page_size(lmtt);
+ struct xe_res_cursor cur;
+ struct xe_lmtt_pt *pt;
+ u64 addr, vram_offset;
+
+ lmtt_assert(lmtt, IS_ALIGNED(start, page_size));
+ lmtt_assert(lmtt, IS_ALIGNED(bo->size, page_size));
+ lmtt_assert(lmtt, xe_bo_is_vram(bo));
+
+ vram_offset = vram_region_gpu_offset(bo->ttm.resource);
+ xe_res_first(bo->ttm.resource, 0, bo->size, &cur);
+ while (cur.remaining) {
+ addr = xe_res_dma(&cur);
+ addr += vram_offset; /* XXX */
+
+ pt = lmtt_leaf_pt(lmtt, vfid, start);
+
+ lmtt_write_pte(lmtt, pt, lmtt->ops->lmtt_pte_encode(addr, 0),
+ lmtt->ops->lmtt_pte_index(start, 0));
+
+ xe_res_next(&cur, page_size);
+ start += page_size;
+ }
+}
+
+/**
+ * xe_lmtt_prepare_pages - Create VF's LMTT Page Tables.
+ * @lmtt: the &xe_lmtt to update
+ * @vfid: the VF identifier (1-based)
+ * @range: top range of LMEM offset to be supported
+ *
+ * This function creates empty LMTT page tables for given VF to support
+ * up to maximum #range LMEM offset. The LMTT page tables created by this
+ * function must be released using xe_lmtt_drop_pages() function.
+ *
+ * Notes:
+ * This function shall be called only after successful LMTT initialization.
+ * See xe_lmtt_init().
+ *
+ * Return: 0 on success or a negative error code on failure.
+ */
+int xe_lmtt_prepare_pages(struct xe_lmtt *lmtt, unsigned int vfid, u64 range)
+{
+ lmtt_assert(lmtt, lmtt->pd);
+ lmtt_assert(lmtt, vfid);
+
+ return lmtt_alloc_range(lmtt, vfid, 0, range);
+}
+
+/**
+ * xe_lmtt_populate_pages - Update VF's LMTT Page Table Entries.
+ * @lmtt: the &xe_lmtt to update
+ * @vfid: the VF identifier (1-based)
+ * @bo: the buffer object with LMEM allocation to be mapped
+ * @offset: the offset at which #bo should be mapped
+ *
+ * This function updates VF's LMTT entries to use given buffer object as a backstore.
+ *
+ * Notes:
+ * This function shall be called only after successful preparation of the
+ * VF's LMTT Page Tables. See xe_lmtt_prepare().
+ *
+ * Return: 0 on success or a negative error code on failure.
+ */
+int xe_lmtt_populate_pages(struct xe_lmtt *lmtt, unsigned int vfid, struct xe_bo *bo, u64 offset)
+{
+ lmtt_assert(lmtt, lmtt->pd);
+ lmtt_assert(lmtt, vfid);
+
+ lmtt_insert_bo(lmtt, vfid, bo, offset);
+ return 0;
+}
+
+/**
+ * xe_lmtt_drop_pages - Remove VF's LMTT Pages.
+ * @lmtt: the &xe_lmtt to update
+ * @vfid: the VF identifier (1-based)
+ *
+ * This function removes all LMTT Page Tables prepared by xe_lmtt_prepare_pages().
+ *
+ * This function shall be called only after successful LMTT initialization.
+ * See xe_lmtt_init().
+ */
+void xe_lmtt_drop_pages(struct xe_lmtt *lmtt, unsigned int vfid)
+{
+ lmtt_assert(lmtt, lmtt->pd);
+ lmtt_assert(lmtt, vfid);
+
+ lmtt_drop_pages(lmtt, vfid);
+}
+
+/**
+ * xe_lmtt_estimate_pt_size - Estimate size of LMTT PT allocations.
+ * @lmtt: the &xe_lmtt
+ * @size: the size of the LMEM to be mapped over LMTT (including any offset)
+ *
+ * This function shall be called only by PF.
+ *
+ * Return: size of the PT allocation(s) needed to support given LMEM size.
+ */
+u64 xe_lmtt_estimate_pt_size(struct xe_lmtt *lmtt, u64 size)
+{
+ unsigned int level = 0;
+ u64 pt_size;
+
+ lmtt_assert(lmtt, IS_SRIOV_PF(lmtt_to_xe(lmtt)));
+ lmtt_assert(lmtt, IS_DGFX(lmtt_to_xe(lmtt)));
+ lmtt_assert(lmtt, lmtt->ops);
+
+ pt_size = PAGE_ALIGN(lmtt->ops->lmtt_pte_size(level) *
+ lmtt->ops->lmtt_pte_num(level));
+
+ while (++level < lmtt->ops->lmtt_root_pd_level()) {
+ pt_size *= lmtt->ops->lmtt_pte_index(size, level) + 1;
+ pt_size += PAGE_ALIGN(lmtt->ops->lmtt_pte_size(level) *
+ lmtt->ops->lmtt_pte_num(level));
+ }
+
+ return pt_size;
+}
+
+#if IS_BUILTIN(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_lmtt_test.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_lmtt.h b/drivers/gpu/drm/xe/xe_lmtt.h
new file mode 100644
index 000000000000..cb10ef994db6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lmtt.h
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_LMTT_H_
+#define _XE_LMTT_H_
+
+#include <linux/types.h>
+
+struct xe_bo;
+struct xe_lmtt;
+struct xe_lmtt_ops;
+
+#ifdef CONFIG_PCI_IOV
+int xe_lmtt_init(struct xe_lmtt *lmtt);
+void xe_lmtt_init_hw(struct xe_lmtt *lmtt);
+int xe_lmtt_prepare_pages(struct xe_lmtt *lmtt, unsigned int vfid, u64 range);
+int xe_lmtt_populate_pages(struct xe_lmtt *lmtt, unsigned int vfid, struct xe_bo *bo, u64 offset);
+void xe_lmtt_drop_pages(struct xe_lmtt *lmtt, unsigned int vfid);
+u64 xe_lmtt_estimate_pt_size(struct xe_lmtt *lmtt, u64 size);
+#else
+static inline int xe_lmtt_init(struct xe_lmtt *lmtt) { return 0; }
+static inline void xe_lmtt_init_hw(struct xe_lmtt *lmtt) { }
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_lmtt_2l.c b/drivers/gpu/drm/xe/xe_lmtt_2l.c
new file mode 100644
index 000000000000..84bc5c4212b5
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lmtt_2l.c
@@ -0,0 +1,150 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/align.h>
+#include <linux/bitfield.h>
+#include <linux/log2.h>
+#include <linux/sizes.h>
+
+#include "xe_lmtt_types.h"
+#include "xe_macros.h"
+
+/**
+ * DOC: Two-Level LMTT Structure
+ *
+ * LMHAW (Local Memory Host Address Width) is 37 bit (128GB)
+ *
+ * LMGAW (Local Memory Guest Address Width) is 37 bit (128GB)
+ *
+ * The following figure illustrates the structure and function of the 2L LMTT::
+ *
+ * LMTT Directory
+ * (1 Entry per VF)
+ * +-----------+ LMTT (per VF)
+ * | | +-----------+
+ * | | | |
+ * | | index: | |
+ * | | LMEM VF +===========+
+ * | | offset --> | PTE | ==> LMEM PF offset
+ * | | +===========+
+ * index: +===========+ | |
+ * VFID --> | PDE | -----------------> +-----------+
+ * +===========+ / \.
+ * | | / \.
+ * | | / \.
+ * | | / \.
+ * +-----------+ <== [LMTT Directory Ptr] \.
+ * / \ / \.
+ * / \ +-----------+-----------------+------+---+
+ * / \ | 31:HAW-16 | HAW-17:5 | 4:1 | 0 |
+ * / \ +===========+=================+======+===+
+ * / \ | Reserved | LMEM Page (2MB) | Rsvd | V |
+ * / \ +-----------+-----------------+------+---+
+ * / \.
+ * +-----------+-----------------+------+---+
+ * | 31:HAW-12 | HAW-13:4 | 3:1 | 0 |
+ * +===========+=================+======+===+
+ * | Reserved | LMTT Ptr (64KB) | Rsvd | V |
+ * +-----------+-----------------+------+---+
+ *
+ */
+
+typedef u32 lmtt_2l_pde_t;
+typedef u32 lmtt_2l_pte_t;
+
+#if IS_ENABLED(CONFIG_DRM_XE_LMTT_2L_128GB)
+#define LMTT_2L_HAW 37 /* 128 GiB */
+#else
+#define LMTT_2L_HAW 35 /* 32 GiB */
+#endif
+
+#define LMTT_2L_PDE_MAX_NUM 64 /* SRIOV with PF and 63 VFs, index 0 (PF) is unused */
+#define LMTT_2L_PDE_LMTT_PTR GENMASK(LMTT_2L_HAW - 13, 4)
+#define LMTT_2L_PDE_VALID BIT(0)
+
+#define LMTT_2L_PTE_MAX_NUM BIT(LMTT_2L_HAW - ilog2(SZ_2M))
+#define LMTT_2L_PTE_LMEM_PAGE GENMASK(LMTT_2L_HAW - 17, 5)
+#define LMTT_2L_PTE_VALID BIT(0)
+
+static unsigned int lmtt_2l_root_pd_level(void)
+{
+ return 1; /* implementation is 0-based */
+}
+
+static unsigned int lmtt_2l_pte_num(unsigned int level)
+{
+ switch (level) {
+ case 1:
+ return LMTT_2L_PDE_MAX_NUM;
+ case 0:
+ BUILD_BUG_ON(LMTT_2L_HAW == 37 && LMTT_2L_PTE_MAX_NUM != SZ_64K);
+ BUILD_BUG_ON(LMTT_2L_HAW == 35 && LMTT_2L_PTE_MAX_NUM != SZ_16K);
+ return LMTT_2L_PTE_MAX_NUM;
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_2l_pte_size(unsigned int level)
+{
+ switch (level) {
+ case 1:
+ return sizeof(lmtt_2l_pde_t);
+ case 0:
+ return sizeof(lmtt_2l_pte_t);
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_2l_pte_shift(unsigned int level)
+{
+ switch (level) {
+ case 0:
+ return ilog2(SZ_2M);
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_2l_pte_index(u64 addr, unsigned int level)
+{
+ addr >>= lmtt_2l_pte_shift(level);
+
+ switch (level) {
+ case 0:
+ /* SZ_2M increments */
+ BUILD_BUG_ON_NOT_POWER_OF_2(LMTT_2L_PTE_MAX_NUM);
+ return addr & (LMTT_2L_PTE_MAX_NUM - 1);
+ default:
+ return 0;
+ }
+}
+
+static u64 lmtt_2l_pte_encode(unsigned long offset, unsigned int level)
+{
+ switch (level) {
+ case 0:
+ XE_WARN_ON(!IS_ALIGNED(offset, SZ_2M));
+ XE_WARN_ON(!FIELD_FIT(LMTT_2L_PTE_LMEM_PAGE, offset / SZ_2M));
+ return FIELD_PREP(LMTT_2L_PTE_LMEM_PAGE, offset / SZ_2M) | LMTT_2L_PTE_VALID;
+ case 1:
+ XE_WARN_ON(!IS_ALIGNED(offset, SZ_64K));
+ XE_WARN_ON(!FIELD_FIT(LMTT_2L_PDE_LMTT_PTR, offset / SZ_64K));
+ return FIELD_PREP(LMTT_2L_PDE_LMTT_PTR, offset / SZ_64K) | LMTT_2L_PDE_VALID;
+ default:
+ XE_WARN_ON(true);
+ return 0;
+ }
+}
+
+const struct xe_lmtt_ops lmtt_2l_ops = {
+ .lmtt_root_pd_level = lmtt_2l_root_pd_level,
+ .lmtt_pte_num = lmtt_2l_pte_num,
+ .lmtt_pte_size = lmtt_2l_pte_size,
+ .lmtt_pte_shift = lmtt_2l_pte_shift,
+ .lmtt_pte_index = lmtt_2l_pte_index,
+ .lmtt_pte_encode = lmtt_2l_pte_encode,
+};
diff --git a/drivers/gpu/drm/xe/xe_lmtt_ml.c b/drivers/gpu/drm/xe/xe_lmtt_ml.c
new file mode 100644
index 000000000000..b21215a2edd6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lmtt_ml.c
@@ -0,0 +1,161 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/align.h>
+#include <linux/bitfield.h>
+#include <linux/log2.h>
+#include <linux/sizes.h>
+
+#include "xe_lmtt_types.h"
+#include "xe_macros.h"
+
+/**
+ * DOC: Multi-Level LMTT Structure
+ *
+ * LMHAW (Local Memory Host Address Width) is 48 bit (256TB)
+ *
+ * LMGAW (Local Memory Guest Address Width) is 48 bit (256TB)
+ *
+ * The following figure illustrates the structure and function of the ML LMTT::
+ *
+ * LMTT L3 Directory
+ * (1 Entry per VF) LMTT L1 Leaf
+ * +-----------+ +-----------+
+ * | | LMTT L2 (per VF) | |
+ * | | +-----------+ | |
+ * | | | | index: +===========+
+ * | | | | GDPA --> | PTE | => LMEM PF offset
+ * | | | | 34:21 +===========+
+ * | | index: | | | |
+ * | | LMEM VF +===========+ | |
+ * | | offset -> | PTE | ----------> +-----------+
+ * | | GAW-1:35 +===========+ / \.
+ * index: +===========+ | | / \.
+ * VFID --> | PDE | ---------> +-----------+ / \.
+ * +===========+ / / / \.
+ * | | / / / \.
+ * +-----------+ <== [LMTT Directory Ptr] / \.
+ * / \ / / / \.
+ * / \ / / +-----------+-----------------+------+---+
+ * / /\ / | 31:HAW-16 | HAW-17:5 | 4:1 | 0 |
+ * / / \ / +===========+=================+======+===+
+ * / / \ / | Reserved | LMEM Page (2MB) | Rsvd | V |
+ * / / +-----------+-----------------+------+---+
+ * / /
+ * +-----------+-----------------+------+---+
+ * | 63:HAW-12 | HAW-13:4 | 3:1 | 0 |
+ * +===========+=================+======+===+
+ * | Reserved | LMTT Ptr (64KB) | Rsvd | V |
+ * +-----------+-----------------+------+---+
+ *
+ */
+
+typedef u64 lmtt_ml_pde_t;
+typedef u32 lmtt_ml_pte_t;
+
+#define LMTT_ML_HAW 48 /* 256 TiB */
+
+#define LMTT_ML_PDE_MAX_NUM 64 /* SRIOV with PF and 63 VFs, index 0 (PF) is unused */
+#define LMTT_ML_PDE_LMTT_PTR GENMASK_ULL(LMTT_ML_HAW - 13, 4)
+#define LMTT_ML_PDE_VALID BIT(0)
+
+#define LMTT_ML_PDE_L2_SHIFT 35
+#define LMTT_ML_PDE_L2_MAX_NUM BIT_ULL(LMTT_ML_HAW - 35)
+
+#define LMTT_ML_PTE_MAX_NUM BIT(35 - ilog2(SZ_2M))
+#define LMTT_ML_PTE_LMEM_PAGE GENMASK(LMTT_ML_HAW - 17, 5)
+#define LMTT_ML_PTE_VALID BIT(0)
+
+static unsigned int lmtt_ml_root_pd_level(void)
+{
+ return 2; /* implementation is 0-based */
+}
+
+static unsigned int lmtt_ml_pte_num(unsigned int level)
+{
+ switch (level) {
+ case 2:
+ return LMTT_ML_PDE_MAX_NUM;
+ case 1:
+ BUILD_BUG_ON(LMTT_ML_HAW == 48 && LMTT_ML_PDE_L2_MAX_NUM != SZ_8K);
+ return LMTT_ML_PDE_L2_MAX_NUM;
+ case 0:
+ BUILD_BUG_ON(LMTT_ML_PTE_MAX_NUM != SZ_16K);
+ return LMTT_ML_PTE_MAX_NUM;
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_ml_pte_size(unsigned int level)
+{
+ switch (level) {
+ case 2:
+ case 1:
+ return sizeof(lmtt_ml_pde_t);
+ case 0:
+ return sizeof(lmtt_ml_pte_t);
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_ml_pte_shift(unsigned int level)
+{
+ switch (level) {
+ case 1:
+ BUILD_BUG_ON(BIT_ULL(LMTT_ML_PDE_L2_SHIFT) != SZ_32G);
+ return ilog2(SZ_32G);
+ case 0:
+ return ilog2(SZ_2M);
+ default:
+ return 0;
+ }
+}
+
+static unsigned int lmtt_ml_pte_index(u64 addr, unsigned int level)
+{
+ addr >>= lmtt_ml_pte_shift(level);
+
+ switch (level) {
+ case 1:
+ /* SZ_32G increments */
+ BUILD_BUG_ON_NOT_POWER_OF_2(LMTT_ML_PDE_L2_MAX_NUM);
+ return addr & (LMTT_ML_PDE_L2_MAX_NUM - 1);
+ case 0:
+ /* SZ_2M increments */
+ BUILD_BUG_ON_NOT_POWER_OF_2(LMTT_ML_PTE_MAX_NUM);
+ return addr & (LMTT_ML_PTE_MAX_NUM - 1);
+ default:
+ return 0;
+ }
+}
+
+static u64 lmtt_ml_pte_encode(unsigned long offset, unsigned int level)
+{
+ switch (level) {
+ case 0:
+ XE_WARN_ON(!IS_ALIGNED(offset, SZ_2M));
+ XE_WARN_ON(!FIELD_FIT(LMTT_ML_PTE_LMEM_PAGE, offset / SZ_2M));
+ return FIELD_PREP(LMTT_ML_PTE_LMEM_PAGE, offset / SZ_2M) | LMTT_ML_PTE_VALID;
+ case 1:
+ case 2:
+ XE_WARN_ON(!IS_ALIGNED(offset, SZ_64K));
+ XE_WARN_ON(!FIELD_FIT(LMTT_ML_PDE_LMTT_PTR, offset / SZ_64K));
+ return FIELD_PREP(LMTT_ML_PDE_LMTT_PTR, offset / SZ_64K) | LMTT_ML_PDE_VALID;
+ default:
+ XE_WARN_ON(true);
+ return 0;
+ }
+}
+
+const struct xe_lmtt_ops lmtt_ml_ops = {
+ .lmtt_root_pd_level = lmtt_ml_root_pd_level,
+ .lmtt_pte_num = lmtt_ml_pte_num,
+ .lmtt_pte_size = lmtt_ml_pte_size,
+ .lmtt_pte_shift = lmtt_ml_pte_shift,
+ .lmtt_pte_index = lmtt_ml_pte_index,
+ .lmtt_pte_encode = lmtt_ml_pte_encode,
+};
diff --git a/drivers/gpu/drm/xe/xe_lmtt_types.h b/drivers/gpu/drm/xe/xe_lmtt_types.h
new file mode 100644
index 000000000000..b37abad23416
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lmtt_types.h
@@ -0,0 +1,63 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_LMTT_TYPES_H_
+#define _XE_LMTT_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_bo;
+struct xe_lmtt;
+struct xe_lmtt_pt;
+struct xe_lmtt_ops;
+
+#define LMTT_PTE_INVALID ULL(0)
+
+/**
+ * struct xe_lmtt - Local Memory Translation Table Manager
+ */
+struct xe_lmtt {
+ /** @pd: root LMTT Directory */
+ struct xe_lmtt_pt *pd;
+
+ /** @ops: LMTT functions */
+ const struct xe_lmtt_ops *ops;
+};
+
+/**
+ * struct xe_lmtt_pt - Local Memory Translation Table Page Table
+ *
+ * Represents single level of the LMTT.
+ */
+struct xe_lmtt_pt {
+ /** @level: page table level, 0 is leaf */
+ unsigned int level;
+
+ /** @bo: buffer object with actual LMTT PTE values */
+ struct xe_bo *bo;
+
+ /** @entries: leaf page tables, exist only for root/non-leaf */
+ struct xe_lmtt_pt *entries[];
+};
+
+/**
+ * struct xe_lmtt_ops - Local Memory Translation Table Operations
+ *
+ * Provides abstraction of the LMTT variants.
+ */
+struct xe_lmtt_ops {
+ /* private: */
+ unsigned int (*lmtt_root_pd_level)(void);
+ unsigned int (*lmtt_pte_num)(unsigned int level);
+ unsigned int (*lmtt_pte_size)(unsigned int level);
+ unsigned int (*lmtt_pte_shift)(unsigned int level);
+ unsigned int (*lmtt_pte_index)(u64 addr, unsigned int level);
+ u64 (*lmtt_pte_encode)(unsigned long offset, unsigned int level);
+};
+
+extern const struct xe_lmtt_ops lmtt_2l_ops;
+extern const struct xe_lmtt_ops lmtt_ml_ops;
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
new file mode 100644
index 000000000000..b7fa3831b684
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lrc.c
@@ -0,0 +1,1272 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_lrc.h"
+
+#include "instructions/xe_mi_commands.h"
+#include "instructions/xe_gfxpipe_commands.h"
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gpu_commands.h"
+#include "regs/xe_lrc_layout.h"
+#include "xe_bb.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_drm_client.h"
+#include "xe_exec_queue_types.h"
+#include "xe_gt.h"
+#include "xe_gt_printk.h"
+#include "xe_hw_fence.h"
+#include "xe_map.h"
+#include "xe_vm.h"
+
+#define CTX_VALID (1 << 0)
+#define CTX_PRIVILEGE (1 << 8)
+#define CTX_ADDRESSING_MODE_SHIFT 3
+#define LEGACY_64B_CONTEXT 3
+
+#define ENGINE_CLASS_SHIFT 61
+#define ENGINE_INSTANCE_SHIFT 48
+
+static struct xe_device *
+lrc_to_xe(struct xe_lrc *lrc)
+{
+ return gt_to_xe(lrc->fence_ctx.gt);
+}
+
+size_t xe_lrc_size(struct xe_device *xe, enum xe_engine_class class)
+{
+ switch (class) {
+ case XE_ENGINE_CLASS_RENDER:
+ if (GRAPHICS_VER(xe) >= 20)
+ return 4 * SZ_4K;
+ else
+ return 14 * SZ_4K;
+ case XE_ENGINE_CLASS_COMPUTE:
+ /* 14 pages since graphics_ver == 11 */
+ if (GRAPHICS_VER(xe) >= 20)
+ return 3 * SZ_4K;
+ else
+ return 14 * SZ_4K;
+ default:
+ WARN(1, "Unknown engine class: %d", class);
+ fallthrough;
+ case XE_ENGINE_CLASS_COPY:
+ case XE_ENGINE_CLASS_VIDEO_DECODE:
+ case XE_ENGINE_CLASS_VIDEO_ENHANCE:
+ case XE_ENGINE_CLASS_OTHER:
+ return 2 * SZ_4K;
+ }
+}
+
+/*
+ * The per-platform tables are u8-encoded in @data. Decode @data and set the
+ * addresses' offset and commands in @regs. The following encoding is used
+ * for each byte. There are 2 steps: decoding commands and decoding addresses.
+ *
+ * Commands:
+ * [7]: create NOPs - number of NOPs are set in lower bits
+ * [6]: When creating MI_LOAD_REGISTER_IMM command, allow to set
+ * MI_LRI_FORCE_POSTED
+ * [5:0]: Number of NOPs or registers to set values to in case of
+ * MI_LOAD_REGISTER_IMM
+ *
+ * Addresses: these are decoded after a MI_LOAD_REGISTER_IMM command by "count"
+ * number of registers. They are set by using the REG/REG16 macros: the former
+ * is used for offsets smaller than 0x200 while the latter is for values bigger
+ * than that. Those macros already set all the bits documented below correctly:
+ *
+ * [7]: When a register offset needs more than 6 bits, use additional bytes, to
+ * follow, for the lower bits
+ * [6:0]: Register offset, without considering the engine base.
+ *
+ * This function only tweaks the commands and register offsets. Values are not
+ * filled out.
+ */
+static void set_offsets(u32 *regs,
+ const u8 *data,
+ const struct xe_hw_engine *hwe)
+#define NOP(x) (BIT(7) | (x))
+#define LRI(count, flags) ((flags) << 6 | (count) | \
+ BUILD_BUG_ON_ZERO(count >= BIT(6)))
+#define POSTED BIT(0)
+#define REG(x) (((x) >> 2) | BUILD_BUG_ON_ZERO(x >= 0x200))
+#define REG16(x) \
+ (((x) >> 9) | BIT(7) | BUILD_BUG_ON_ZERO(x >= 0x10000)), \
+ (((x) >> 2) & 0x7f)
+#define END 0
+{
+ const u32 base = hwe->mmio_base;
+
+ while (*data) {
+ u8 count, flags;
+
+ if (*data & BIT(7)) { /* skip */
+ count = *data++ & ~BIT(7);
+ regs += count;
+ continue;
+ }
+
+ count = *data & 0x3f;
+ flags = *data >> 6;
+ data++;
+
+ *regs = MI_LOAD_REGISTER_IMM | MI_LRI_NUM_REGS(count);
+ if (flags & POSTED)
+ *regs |= MI_LRI_FORCE_POSTED;
+ *regs |= MI_LRI_LRM_CS_MMIO;
+ regs++;
+
+ xe_gt_assert(hwe->gt, count);
+ do {
+ u32 offset = 0;
+ u8 v;
+
+ do {
+ v = *data++;
+ offset <<= 7;
+ offset |= v & ~BIT(7);
+ } while (v & BIT(7));
+
+ regs[0] = base + (offset << 2);
+ regs += 2;
+ } while (--count);
+ }
+
+ *regs = MI_BATCH_BUFFER_END | BIT(0);
+}
+
+static const u8 gen12_xcs_offsets[] = {
+ NOP(1),
+ LRI(13, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+
+ NOP(5),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ END
+};
+
+static const u8 dg2_xcs_offsets[] = {
+ NOP(1),
+ LRI(15, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+ REG(0x120),
+ REG(0x124),
+
+ NOP(1),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ END
+};
+
+static const u8 gen12_rcs_offsets[] = {
+ NOP(1),
+ LRI(13, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+
+ NOP(5),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ LRI(3, POSTED),
+ REG(0x1b0),
+ REG16(0x5a8),
+ REG16(0x5ac),
+
+ NOP(6),
+ LRI(1, 0),
+ REG(0x0c8),
+ NOP(3 + 9 + 1),
+
+ LRI(51, POSTED),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG16(0x588),
+ REG(0x028),
+ REG(0x09c),
+ REG(0x0c0),
+ REG(0x178),
+ REG(0x17c),
+ REG16(0x358),
+ REG(0x170),
+ REG(0x150),
+ REG(0x154),
+ REG(0x158),
+ REG16(0x41c),
+ REG16(0x600),
+ REG16(0x604),
+ REG16(0x608),
+ REG16(0x60c),
+ REG16(0x610),
+ REG16(0x614),
+ REG16(0x618),
+ REG16(0x61c),
+ REG16(0x620),
+ REG16(0x624),
+ REG16(0x628),
+ REG16(0x62c),
+ REG16(0x630),
+ REG16(0x634),
+ REG16(0x638),
+ REG16(0x63c),
+ REG16(0x640),
+ REG16(0x644),
+ REG16(0x648),
+ REG16(0x64c),
+ REG16(0x650),
+ REG16(0x654),
+ REG16(0x658),
+ REG16(0x65c),
+ REG16(0x660),
+ REG16(0x664),
+ REG16(0x668),
+ REG16(0x66c),
+ REG16(0x670),
+ REG16(0x674),
+ REG16(0x678),
+ REG16(0x67c),
+ REG(0x068),
+ REG(0x084),
+ NOP(1),
+
+ END
+};
+
+static const u8 xehp_rcs_offsets[] = {
+ NOP(1),
+ LRI(13, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+
+ NOP(5),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ LRI(3, POSTED),
+ REG(0x1b0),
+ REG16(0x5a8),
+ REG16(0x5ac),
+
+ NOP(6),
+ LRI(1, 0),
+ REG(0x0c8),
+
+ END
+};
+
+static const u8 dg2_rcs_offsets[] = {
+ NOP(1),
+ LRI(15, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+ REG(0x120),
+ REG(0x124),
+
+ NOP(1),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ LRI(3, POSTED),
+ REG(0x1b0),
+ REG16(0x5a8),
+ REG16(0x5ac),
+
+ NOP(6),
+ LRI(1, 0),
+ REG(0x0c8),
+
+ END
+};
+
+static const u8 mtl_rcs_offsets[] = {
+ NOP(1),
+ LRI(15, POSTED),
+ REG16(0x244),
+ REG(0x034),
+ REG(0x030),
+ REG(0x038),
+ REG(0x03c),
+ REG(0x168),
+ REG(0x140),
+ REG(0x110),
+ REG(0x1c0),
+ REG(0x1c4),
+ REG(0x1c8),
+ REG(0x180),
+ REG16(0x2b4),
+ REG(0x120),
+ REG(0x124),
+
+ NOP(1),
+ LRI(9, POSTED),
+ REG16(0x3a8),
+ REG16(0x28c),
+ REG16(0x288),
+ REG16(0x284),
+ REG16(0x280),
+ REG16(0x27c),
+ REG16(0x278),
+ REG16(0x274),
+ REG16(0x270),
+
+ NOP(2),
+ LRI(2, POSTED),
+ REG16(0x5a8),
+ REG16(0x5ac),
+
+ NOP(6),
+ LRI(1, 0),
+ REG(0x0c8),
+
+ END
+};
+
+#define XE2_CTX_COMMON \
+ NOP(1), /* [0x00] */ \
+ LRI(15, POSTED), /* [0x01] */ \
+ REG16(0x244), /* [0x02] CTXT_SR_CTL */ \
+ REG(0x034), /* [0x04] RING_BUFFER_HEAD */ \
+ REG(0x030), /* [0x06] RING_BUFFER_TAIL */ \
+ REG(0x038), /* [0x08] RING_BUFFER_START */ \
+ REG(0x03c), /* [0x0a] RING_BUFFER_CONTROL */ \
+ REG(0x168), /* [0x0c] BB_ADDR_UDW */ \
+ REG(0x140), /* [0x0e] BB_ADDR */ \
+ REG(0x110), /* [0x10] BB_STATE */ \
+ REG(0x1c0), /* [0x12] BB_PER_CTX_PTR */ \
+ REG(0x1c4), /* [0x14] RCS_INDIRECT_CTX */ \
+ REG(0x1c8), /* [0x16] RCS_INDIRECT_CTX_OFFSET */ \
+ REG(0x180), /* [0x18] CCID */ \
+ REG16(0x2b4), /* [0x1a] SEMAPHORE_TOKEN */ \
+ REG(0x120), /* [0x1c] PRT_BB_STATE */ \
+ REG(0x124), /* [0x1e] PRT_BB_STATE_UDW */ \
+ \
+ NOP(1), /* [0x20] */ \
+ LRI(9, POSTED), /* [0x21] */ \
+ REG16(0x3a8), /* [0x22] CTX_TIMESTAMP */ \
+ REG16(0x3ac), /* [0x24] CTX_TIMESTAMP_UDW */ \
+ REG(0x108), /* [0x26] INDIRECT_RING_STATE */ \
+ REG16(0x284), /* [0x28] dummy reg */ \
+ REG16(0x280), /* [0x2a] CS_ACC_CTR_THOLD */ \
+ REG16(0x27c), /* [0x2c] CS_CTX_SYS_PASID */ \
+ REG16(0x278), /* [0x2e] CS_CTX_ASID */ \
+ REG16(0x274), /* [0x30] PTBP_UDW */ \
+ REG16(0x270) /* [0x32] PTBP_LDW */
+
+static const u8 xe2_rcs_offsets[] = {
+ XE2_CTX_COMMON,
+
+ NOP(2), /* [0x34] */
+ LRI(2, POSTED), /* [0x36] */
+ REG16(0x5a8), /* [0x37] CONTEXT_SCHEDULING_ATTRIBUTES */
+ REG16(0x5ac), /* [0x39] PREEMPTION_STATUS */
+
+ NOP(6), /* [0x41] */
+ LRI(1, 0), /* [0x47] */
+ REG(0x0c8), /* [0x48] R_PWR_CLK_STATE */
+
+ END
+};
+
+static const u8 xe2_bcs_offsets[] = {
+ XE2_CTX_COMMON,
+
+ NOP(4 + 8 + 1), /* [0x34] */
+ LRI(2, POSTED), /* [0x41] */
+ REG16(0x200), /* [0x42] BCS_SWCTRL */
+ REG16(0x204), /* [0x44] BLIT_CCTL */
+
+ END
+};
+
+static const u8 xe2_xcs_offsets[] = {
+ XE2_CTX_COMMON,
+
+ END
+};
+
+#undef END
+#undef REG16
+#undef REG
+#undef LRI
+#undef NOP
+
+static const u8 *reg_offsets(struct xe_device *xe, enum xe_engine_class class)
+{
+ if (class == XE_ENGINE_CLASS_RENDER) {
+ if (GRAPHICS_VER(xe) >= 20)
+ return xe2_rcs_offsets;
+ else if (GRAPHICS_VERx100(xe) >= 1270)
+ return mtl_rcs_offsets;
+ else if (GRAPHICS_VERx100(xe) >= 1255)
+ return dg2_rcs_offsets;
+ else if (GRAPHICS_VERx100(xe) >= 1250)
+ return xehp_rcs_offsets;
+ else
+ return gen12_rcs_offsets;
+ } else if (class == XE_ENGINE_CLASS_COPY) {
+ if (GRAPHICS_VER(xe) >= 20)
+ return xe2_bcs_offsets;
+ else
+ return gen12_xcs_offsets;
+ } else {
+ if (GRAPHICS_VER(xe) >= 20)
+ return xe2_xcs_offsets;
+ else if (GRAPHICS_VERx100(xe) >= 1255)
+ return dg2_xcs_offsets;
+ else
+ return gen12_xcs_offsets;
+ }
+}
+
+static void set_context_control(u32 *regs, struct xe_hw_engine *hwe)
+{
+ regs[CTX_CONTEXT_CONTROL] = _MASKED_BIT_ENABLE(CTX_CTRL_INHIBIT_SYN_CTX_SWITCH) |
+ _MASKED_BIT_DISABLE(CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT) |
+ CTX_CTRL_ENGINE_CTX_RESTORE_INHIBIT;
+
+ /* TODO: Timestamp */
+}
+
+static int lrc_ring_mi_mode(struct xe_hw_engine *hwe)
+{
+ struct xe_device *xe = gt_to_xe(hwe->gt);
+
+ if (GRAPHICS_VERx100(xe) >= 1250)
+ return 0x70;
+ else
+ return 0x60;
+}
+
+static void reset_stop_ring(u32 *regs, struct xe_hw_engine *hwe)
+{
+ int x;
+
+ x = lrc_ring_mi_mode(hwe);
+ regs[x + 1] &= ~STOP_RING;
+ regs[x + 1] |= STOP_RING << 16;
+}
+
+static inline u32 __xe_lrc_ring_offset(struct xe_lrc *lrc)
+{
+ return 0;
+}
+
+u32 xe_lrc_pphwsp_offset(struct xe_lrc *lrc)
+{
+ return lrc->ring.size;
+}
+
+/* Make the magic macros work */
+#define __xe_lrc_pphwsp_offset xe_lrc_pphwsp_offset
+
+#define LRC_SEQNO_PPHWSP_OFFSET 512
+#define LRC_START_SEQNO_PPHWSP_OFFSET (LRC_SEQNO_PPHWSP_OFFSET + 8)
+#define LRC_PARALLEL_PPHWSP_OFFSET 2048
+#define LRC_PPHWSP_SIZE SZ_4K
+
+static size_t lrc_reg_size(struct xe_device *xe)
+{
+ if (GRAPHICS_VERx100(xe) >= 1250)
+ return 96 * sizeof(u32);
+ else
+ return 80 * sizeof(u32);
+}
+
+size_t xe_lrc_skip_size(struct xe_device *xe)
+{
+ return LRC_PPHWSP_SIZE + lrc_reg_size(xe);
+}
+
+static inline u32 __xe_lrc_seqno_offset(struct xe_lrc *lrc)
+{
+ /* The seqno is stored in the driver-defined portion of PPHWSP */
+ return xe_lrc_pphwsp_offset(lrc) + LRC_SEQNO_PPHWSP_OFFSET;
+}
+
+static inline u32 __xe_lrc_start_seqno_offset(struct xe_lrc *lrc)
+{
+ /* The start seqno is stored in the driver-defined portion of PPHWSP */
+ return xe_lrc_pphwsp_offset(lrc) + LRC_START_SEQNO_PPHWSP_OFFSET;
+}
+
+static inline u32 __xe_lrc_parallel_offset(struct xe_lrc *lrc)
+{
+ /* The parallel is stored in the driver-defined portion of PPHWSP */
+ return xe_lrc_pphwsp_offset(lrc) + LRC_PARALLEL_PPHWSP_OFFSET;
+}
+
+static inline u32 __xe_lrc_regs_offset(struct xe_lrc *lrc)
+{
+ return xe_lrc_pphwsp_offset(lrc) + LRC_PPHWSP_SIZE;
+}
+
+#define DECL_MAP_ADDR_HELPERS(elem) \
+static inline struct iosys_map __xe_lrc_##elem##_map(struct xe_lrc *lrc) \
+{ \
+ struct iosys_map map = lrc->bo->vmap; \
+\
+ xe_assert(lrc_to_xe(lrc), !iosys_map_is_null(&map)); \
+ iosys_map_incr(&map, __xe_lrc_##elem##_offset(lrc)); \
+ return map; \
+} \
+static inline u32 __xe_lrc_##elem##_ggtt_addr(struct xe_lrc *lrc) \
+{ \
+ return xe_bo_ggtt_addr(lrc->bo) + __xe_lrc_##elem##_offset(lrc); \
+} \
+
+DECL_MAP_ADDR_HELPERS(ring)
+DECL_MAP_ADDR_HELPERS(pphwsp)
+DECL_MAP_ADDR_HELPERS(seqno)
+DECL_MAP_ADDR_HELPERS(regs)
+DECL_MAP_ADDR_HELPERS(start_seqno)
+DECL_MAP_ADDR_HELPERS(parallel)
+
+#undef DECL_MAP_ADDR_HELPERS
+
+u32 xe_lrc_ggtt_addr(struct xe_lrc *lrc)
+{
+ return __xe_lrc_pphwsp_ggtt_addr(lrc);
+}
+
+u32 xe_lrc_read_ctx_reg(struct xe_lrc *lrc, int reg_nr)
+{
+ struct xe_device *xe = lrc_to_xe(lrc);
+ struct iosys_map map;
+
+ map = __xe_lrc_regs_map(lrc);
+ iosys_map_incr(&map, reg_nr * sizeof(u32));
+ return xe_map_read32(xe, &map);
+}
+
+void xe_lrc_write_ctx_reg(struct xe_lrc *lrc, int reg_nr, u32 val)
+{
+ struct xe_device *xe = lrc_to_xe(lrc);
+ struct iosys_map map;
+
+ map = __xe_lrc_regs_map(lrc);
+ iosys_map_incr(&map, reg_nr * sizeof(u32));
+ xe_map_write32(xe, &map, val);
+}
+
+static void *empty_lrc_data(struct xe_hw_engine *hwe)
+{
+ struct xe_device *xe = gt_to_xe(hwe->gt);
+ void *data;
+ u32 *regs;
+
+ data = kzalloc(xe_lrc_size(xe, hwe->class), GFP_KERNEL);
+ if (!data)
+ return NULL;
+
+ /* 1st page: Per-Process of HW status Page */
+ regs = data + LRC_PPHWSP_SIZE;
+ set_offsets(regs, reg_offsets(xe, hwe->class), hwe);
+ set_context_control(regs, hwe);
+ reset_stop_ring(regs, hwe);
+
+ return data;
+}
+
+static void xe_lrc_set_ppgtt(struct xe_lrc *lrc, struct xe_vm *vm)
+{
+ u64 desc = xe_vm_pdp4_descriptor(vm, lrc->tile);
+
+ xe_lrc_write_ctx_reg(lrc, CTX_PDP0_UDW, upper_32_bits(desc));
+ xe_lrc_write_ctx_reg(lrc, CTX_PDP0_LDW, lower_32_bits(desc));
+}
+
+#define PVC_CTX_ASID (0x2e + 1)
+#define PVC_CTX_ACC_CTR_THOLD (0x2a + 1)
+#define ACC_GRANULARITY_S 20
+#define ACC_NOTIFY_S 16
+
+int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
+ struct xe_exec_queue *q, struct xe_vm *vm, u32 ring_size)
+{
+ struct xe_gt *gt = hwe->gt;
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_device *xe = gt_to_xe(gt);
+ struct iosys_map map;
+ void *init_data = NULL;
+ u32 arb_enable;
+ int err;
+
+ lrc->flags = 0;
+
+ /*
+ * FIXME: Perma-pinning LRC as we don't yet support moving GGTT address
+ * via VM bind calls.
+ */
+ lrc->bo = xe_bo_create_pin_map(xe, tile, vm,
+ ring_size + xe_lrc_size(xe, hwe->class),
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(lrc->bo))
+ return PTR_ERR(lrc->bo);
+
+ lrc->tile = gt_to_tile(hwe->gt);
+ lrc->ring.size = ring_size;
+ lrc->ring.tail = 0;
+
+ xe_hw_fence_ctx_init(&lrc->fence_ctx, hwe->gt,
+ hwe->fence_irq, hwe->name);
+
+ if (!gt->default_lrc[hwe->class]) {
+ init_data = empty_lrc_data(hwe);
+ if (!init_data) {
+ err = -ENOMEM;
+ goto err_lrc_finish;
+ }
+ }
+
+ /*
+ * Init Per-Process of HW status Page, LRC / context state to known
+ * values
+ */
+ map = __xe_lrc_pphwsp_map(lrc);
+ if (!init_data) {
+ xe_map_memset(xe, &map, 0, 0, LRC_PPHWSP_SIZE); /* PPHWSP */
+ xe_map_memcpy_to(xe, &map, LRC_PPHWSP_SIZE,
+ gt->default_lrc[hwe->class] + LRC_PPHWSP_SIZE,
+ xe_lrc_size(xe, hwe->class) - LRC_PPHWSP_SIZE);
+ } else {
+ xe_map_memcpy_to(xe, &map, 0, init_data,
+ xe_lrc_size(xe, hwe->class));
+ kfree(init_data);
+ }
+
+ if (vm) {
+ xe_lrc_set_ppgtt(lrc, vm);
+
+ if (vm->xef)
+ xe_drm_client_add_bo(vm->xef->client, lrc->bo);
+ }
+
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_START, __xe_lrc_ring_ggtt_addr(lrc));
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_HEAD, 0);
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_TAIL, lrc->ring.tail);
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_CTL,
+ RING_CTL_SIZE(lrc->ring.size) | RING_VALID);
+ if (xe->info.has_asid && vm)
+ xe_lrc_write_ctx_reg(lrc, PVC_CTX_ASID,
+ (q->usm.acc_granularity <<
+ ACC_GRANULARITY_S) | vm->usm.asid);
+ if (xe->info.has_usm && vm)
+ xe_lrc_write_ctx_reg(lrc, PVC_CTX_ACC_CTR_THOLD,
+ (q->usm.acc_notify << ACC_NOTIFY_S) |
+ q->usm.acc_trigger);
+
+ lrc->desc = CTX_VALID;
+ lrc->desc |= LEGACY_64B_CONTEXT << CTX_ADDRESSING_MODE_SHIFT;
+ /* TODO: Priority */
+
+ /* While this appears to have something about privileged batches or
+ * some such, it really just means PPGTT mode.
+ */
+ if (vm)
+ lrc->desc |= CTX_PRIVILEGE;
+
+ if (GRAPHICS_VERx100(xe) < 1250) {
+ lrc->desc |= (u64)hwe->instance << ENGINE_INSTANCE_SHIFT;
+ lrc->desc |= (u64)hwe->class << ENGINE_CLASS_SHIFT;
+ }
+
+ arb_enable = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+ xe_lrc_write_ring(lrc, &arb_enable, sizeof(arb_enable));
+
+ map = __xe_lrc_seqno_map(lrc);
+ xe_map_write32(lrc_to_xe(lrc), &map, lrc->fence_ctx.next_seqno - 1);
+
+ map = __xe_lrc_start_seqno_map(lrc);
+ xe_map_write32(lrc_to_xe(lrc), &map, lrc->fence_ctx.next_seqno - 1);
+
+ return 0;
+
+err_lrc_finish:
+ xe_lrc_finish(lrc);
+ return err;
+}
+
+void xe_lrc_finish(struct xe_lrc *lrc)
+{
+ xe_hw_fence_ctx_finish(&lrc->fence_ctx);
+ xe_bo_lock(lrc->bo, false);
+ xe_bo_unpin(lrc->bo);
+ xe_bo_unlock(lrc->bo);
+ xe_bo_put(lrc->bo);
+}
+
+void xe_lrc_set_ring_head(struct xe_lrc *lrc, u32 head)
+{
+ xe_lrc_write_ctx_reg(lrc, CTX_RING_HEAD, head);
+}
+
+u32 xe_lrc_ring_head(struct xe_lrc *lrc)
+{
+ return xe_lrc_read_ctx_reg(lrc, CTX_RING_HEAD) & HEAD_ADDR;
+}
+
+u32 xe_lrc_ring_space(struct xe_lrc *lrc)
+{
+ const u32 head = xe_lrc_ring_head(lrc);
+ const u32 tail = lrc->ring.tail;
+ const u32 size = lrc->ring.size;
+
+ return ((head - tail - 1) & (size - 1)) + 1;
+}
+
+static void __xe_lrc_write_ring(struct xe_lrc *lrc, struct iosys_map ring,
+ const void *data, size_t size)
+{
+ struct xe_device *xe = lrc_to_xe(lrc);
+
+ iosys_map_incr(&ring, lrc->ring.tail);
+ xe_map_memcpy_to(xe, &ring, 0, data, size);
+ lrc->ring.tail = (lrc->ring.tail + size) & (lrc->ring.size - 1);
+}
+
+void xe_lrc_write_ring(struct xe_lrc *lrc, const void *data, size_t size)
+{
+ struct xe_device *xe = lrc_to_xe(lrc);
+ struct iosys_map ring;
+ u32 rhs;
+ size_t aligned_size;
+
+ xe_assert(xe, IS_ALIGNED(size, 4));
+ aligned_size = ALIGN(size, 8);
+
+ ring = __xe_lrc_ring_map(lrc);
+
+ xe_assert(xe, lrc->ring.tail < lrc->ring.size);
+ rhs = lrc->ring.size - lrc->ring.tail;
+ if (size > rhs) {
+ __xe_lrc_write_ring(lrc, ring, data, rhs);
+ __xe_lrc_write_ring(lrc, ring, data + rhs, size - rhs);
+ } else {
+ __xe_lrc_write_ring(lrc, ring, data, size);
+ }
+
+ if (aligned_size > size) {
+ u32 noop = MI_NOOP;
+
+ __xe_lrc_write_ring(lrc, ring, &noop, sizeof(noop));
+ }
+}
+
+u64 xe_lrc_descriptor(struct xe_lrc *lrc)
+{
+ return lrc->desc | xe_lrc_ggtt_addr(lrc);
+}
+
+u32 xe_lrc_seqno_ggtt_addr(struct xe_lrc *lrc)
+{
+ return __xe_lrc_seqno_ggtt_addr(lrc);
+}
+
+struct dma_fence *xe_lrc_create_seqno_fence(struct xe_lrc *lrc)
+{
+ return &xe_hw_fence_create(&lrc->fence_ctx,
+ __xe_lrc_seqno_map(lrc))->dma;
+}
+
+s32 xe_lrc_seqno(struct xe_lrc *lrc)
+{
+ struct iosys_map map = __xe_lrc_seqno_map(lrc);
+
+ return xe_map_read32(lrc_to_xe(lrc), &map);
+}
+
+s32 xe_lrc_start_seqno(struct xe_lrc *lrc)
+{
+ struct iosys_map map = __xe_lrc_start_seqno_map(lrc);
+
+ return xe_map_read32(lrc_to_xe(lrc), &map);
+}
+
+u32 xe_lrc_start_seqno_ggtt_addr(struct xe_lrc *lrc)
+{
+ return __xe_lrc_start_seqno_ggtt_addr(lrc);
+}
+
+u32 xe_lrc_parallel_ggtt_addr(struct xe_lrc *lrc)
+{
+ return __xe_lrc_parallel_ggtt_addr(lrc);
+}
+
+struct iosys_map xe_lrc_parallel_map(struct xe_lrc *lrc)
+{
+ return __xe_lrc_parallel_map(lrc);
+}
+
+static int instr_dw(u32 cmd_header)
+{
+ /* GFXPIPE "SINGLE_DW" opcodes are a single dword */
+ if ((cmd_header & (XE_INSTR_CMD_TYPE | GFXPIPE_PIPELINE)) ==
+ GFXPIPE_SINGLE_DW_CMD(0, 0))
+ return 1;
+
+ /* 3DSTATE_SO_DECL_LIST has a 9-bit dword length rather than 8 */
+ if ((cmd_header & GFXPIPE_MATCH_MASK) == CMD_3DSTATE_SO_DECL_LIST)
+ return REG_FIELD_GET(CMD_3DSTATE_SO_DECL_LIST_DW_LEN, cmd_header) + 2;
+
+ /* Most instructions have the # of dwords (minus 2) in 7:0 */
+ return REG_FIELD_GET(XE_INSTR_LEN_MASK, cmd_header) + 2;
+}
+
+static int dump_mi_command(struct drm_printer *p,
+ struct xe_gt *gt,
+ u32 *dw,
+ int remaining_dw)
+{
+ u32 inst_header = *dw;
+ u32 numdw = instr_dw(inst_header);
+ u32 opcode = REG_FIELD_GET(MI_OPCODE, inst_header);
+ int num_noop;
+
+ /* First check for commands that don't have/use a '# DW' field */
+ switch (inst_header & MI_OPCODE) {
+ case MI_NOOP:
+ num_noop = 1;
+ while (num_noop < remaining_dw &&
+ (*(++dw) & REG_GENMASK(31, 23)) == MI_NOOP)
+ num_noop++;
+ drm_printf(p, "[%#010x] MI_NOOP (%d dwords)\n", inst_header, num_noop);
+ return num_noop;
+
+ case MI_TOPOLOGY_FILTER:
+ drm_printf(p, "[%#010x] MI_TOPOLOGY_FILTER\n", inst_header);
+ return 1;
+
+ case MI_BATCH_BUFFER_END:
+ drm_printf(p, "[%#010x] MI_BATCH_BUFFER_END\n", inst_header);
+ /* Return 'remaining_dw' to consume the rest of the LRC */
+ return remaining_dw;
+ }
+
+ /*
+ * Any remaining commands include a # of dwords. We should make sure
+ * it doesn't exceed the remaining size of the LRC.
+ */
+ if (xe_gt_WARN_ON(gt, numdw > remaining_dw))
+ numdw = remaining_dw;
+
+ switch (inst_header & MI_OPCODE) {
+ case MI_LOAD_REGISTER_IMM:
+ drm_printf(p, "[%#010x] MI_LOAD_REGISTER_IMM: %d regs\n",
+ inst_header, (numdw - 1) / 2);
+ for (int i = 1; i < numdw; i += 2)
+ drm_printf(p, " - %#6x = %#010x\n", dw[i], dw[i + 1]);
+ return numdw;
+
+ case MI_FORCE_WAKEUP:
+ drm_printf(p, "[%#010x] MI_FORCE_WAKEUP\n", inst_header);
+ return numdw;
+
+ default:
+ drm_printf(p, "[%#010x] unknown MI opcode %#x, likely %d dwords\n",
+ inst_header, opcode, numdw);
+ return numdw;
+ }
+}
+
+static int dump_gfxpipe_command(struct drm_printer *p,
+ struct xe_gt *gt,
+ u32 *dw,
+ int remaining_dw)
+{
+ u32 numdw = instr_dw(*dw);
+ u32 pipeline = REG_FIELD_GET(GFXPIPE_PIPELINE, *dw);
+ u32 opcode = REG_FIELD_GET(GFXPIPE_OPCODE, *dw);
+ u32 subopcode = REG_FIELD_GET(GFXPIPE_SUBOPCODE, *dw);
+
+ /*
+ * Make sure we haven't mis-parsed a number of dwords that exceeds the
+ * remaining size of the LRC.
+ */
+ if (xe_gt_WARN_ON(gt, numdw > remaining_dw))
+ numdw = remaining_dw;
+
+ switch (*dw & GFXPIPE_MATCH_MASK) {
+#define MATCH(cmd) \
+ case cmd: \
+ drm_printf(p, "[%#010x] " #cmd " (%d dwords)\n", *dw, numdw); \
+ return numdw
+#define MATCH3D(cmd) \
+ case CMD_##cmd: \
+ drm_printf(p, "[%#010x] " #cmd " (%d dwords)\n", *dw, numdw); \
+ return numdw
+
+ MATCH(STATE_BASE_ADDRESS);
+ MATCH(STATE_SIP);
+ MATCH(GPGPU_CSR_BASE_ADDRESS);
+ MATCH(STATE_COMPUTE_MODE);
+ MATCH3D(3DSTATE_BTD);
+
+ MATCH3D(3DSTATE_VF_STATISTICS);
+
+ MATCH(PIPELINE_SELECT);
+
+ MATCH3D(3DSTATE_DRAWING_RECTANGLE_FAST);
+ MATCH3D(3DSTATE_CLEAR_PARAMS);
+ MATCH3D(3DSTATE_DEPTH_BUFFER);
+ MATCH3D(3DSTATE_STENCIL_BUFFER);
+ MATCH3D(3DSTATE_HIER_DEPTH_BUFFER);
+ MATCH3D(3DSTATE_VERTEX_BUFFERS);
+ MATCH3D(3DSTATE_VERTEX_ELEMENTS);
+ MATCH3D(3DSTATE_INDEX_BUFFER);
+ MATCH3D(3DSTATE_VF);
+ MATCH3D(3DSTATE_MULTISAMPLE);
+ MATCH3D(3DSTATE_CC_STATE_POINTERS);
+ MATCH3D(3DSTATE_SCISSOR_STATE_POINTERS);
+ MATCH3D(3DSTATE_VS);
+ MATCH3D(3DSTATE_GS);
+ MATCH3D(3DSTATE_CLIP);
+ MATCH3D(3DSTATE_SF);
+ MATCH3D(3DSTATE_WM);
+ MATCH3D(3DSTATE_CONSTANT_VS);
+ MATCH3D(3DSTATE_CONSTANT_GS);
+ MATCH3D(3DSTATE_SAMPLE_MASK);
+ MATCH3D(3DSTATE_CONSTANT_HS);
+ MATCH3D(3DSTATE_CONSTANT_DS);
+ MATCH3D(3DSTATE_HS);
+ MATCH3D(3DSTATE_TE);
+ MATCH3D(3DSTATE_DS);
+ MATCH3D(3DSTATE_STREAMOUT);
+ MATCH3D(3DSTATE_SBE);
+ MATCH3D(3DSTATE_PS);
+ MATCH3D(3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP);
+ MATCH3D(3DSTATE_CPS_POINTERS);
+ MATCH3D(3DSTATE_VIEWPORT_STATE_POINTERS_CC);
+ MATCH3D(3DSTATE_BLEND_STATE_POINTERS);
+ MATCH3D(3DSTATE_BINDING_TABLE_POINTERS_VS);
+ MATCH3D(3DSTATE_BINDING_TABLE_POINTERS_HS);
+ MATCH3D(3DSTATE_BINDING_TABLE_POINTERS_DS);
+ MATCH3D(3DSTATE_BINDING_TABLE_POINTERS_GS);
+ MATCH3D(3DSTATE_BINDING_TABLE_POINTERS_PS);
+ MATCH3D(3DSTATE_SAMPLER_STATE_POINTERS_VS);
+ MATCH3D(3DSTATE_SAMPLER_STATE_POINTERS_HS);
+ MATCH3D(3DSTATE_SAMPLER_STATE_POINTERS_DS);
+ MATCH3D(3DSTATE_SAMPLER_STATE_POINTERS_GS);
+ MATCH3D(3DSTATE_SAMPLER_STATE_POINTERS_PS);
+ MATCH3D(3DSTATE_VF_INSTANCING);
+ MATCH3D(3DSTATE_VF_SGVS);
+ MATCH3D(3DSTATE_VF_TOPOLOGY);
+ MATCH3D(3DSTATE_WM_CHROMAKEY);
+ MATCH3D(3DSTATE_PS_BLEND);
+ MATCH3D(3DSTATE_WM_DEPTH_STENCIL);
+ MATCH3D(3DSTATE_PS_EXTRA);
+ MATCH3D(3DSTATE_RASTER);
+ MATCH3D(3DSTATE_SBE_SWIZ);
+ MATCH3D(3DSTATE_WM_HZ_OP);
+ MATCH3D(3DSTATE_VF_COMPONENT_PACKING);
+ MATCH3D(3DSTATE_VF_SGVS_2);
+ MATCH3D(3DSTATE_VFG);
+ MATCH3D(3DSTATE_URB_ALLOC_VS);
+ MATCH3D(3DSTATE_URB_ALLOC_HS);
+ MATCH3D(3DSTATE_URB_ALLOC_DS);
+ MATCH3D(3DSTATE_URB_ALLOC_GS);
+ MATCH3D(3DSTATE_SO_BUFFER_INDEX_0);
+ MATCH3D(3DSTATE_SO_BUFFER_INDEX_1);
+ MATCH3D(3DSTATE_SO_BUFFER_INDEX_2);
+ MATCH3D(3DSTATE_SO_BUFFER_INDEX_3);
+ MATCH3D(3DSTATE_PRIMITIVE_REPLICATION);
+ MATCH3D(3DSTATE_TBIMR_TILE_PASS_INFO);
+ MATCH3D(3DSTATE_AMFS);
+ MATCH3D(3DSTATE_DEPTH_BOUNDS);
+ MATCH3D(3DSTATE_AMFS_TEXTURE_POINTERS);
+ MATCH3D(3DSTATE_CONSTANT_TS_POINTER);
+ MATCH3D(3DSTATE_MESH_CONTROL);
+ MATCH3D(3DSTATE_MESH_DISTRIB);
+ MATCH3D(3DSTATE_TASK_REDISTRIB);
+ MATCH3D(3DSTATE_MESH_SHADER);
+ MATCH3D(3DSTATE_MESH_SHADER_DATA);
+ MATCH3D(3DSTATE_TASK_CONTROL);
+ MATCH3D(3DSTATE_TASK_SHADER);
+ MATCH3D(3DSTATE_TASK_SHADER_DATA);
+ MATCH3D(3DSTATE_URB_ALLOC_MESH);
+ MATCH3D(3DSTATE_URB_ALLOC_TASK);
+ MATCH3D(3DSTATE_CLIP_MESH);
+ MATCH3D(3DSTATE_SBE_MESH);
+ MATCH3D(3DSTATE_CPSIZE_CONTROL_BUFFER);
+
+ MATCH3D(3DSTATE_DRAWING_RECTANGLE);
+ MATCH3D(3DSTATE_CHROMA_KEY);
+ MATCH3D(3DSTATE_POLY_STIPPLE_OFFSET);
+ MATCH3D(3DSTATE_POLY_STIPPLE_PATTERN);
+ MATCH3D(3DSTATE_LINE_STIPPLE);
+ MATCH3D(3DSTATE_AA_LINE_PARAMETERS);
+ MATCH3D(3DSTATE_MONOFILTER_SIZE);
+ MATCH3D(3DSTATE_PUSH_CONSTANT_ALLOC_VS);
+ MATCH3D(3DSTATE_PUSH_CONSTANT_ALLOC_HS);
+ MATCH3D(3DSTATE_PUSH_CONSTANT_ALLOC_DS);
+ MATCH3D(3DSTATE_PUSH_CONSTANT_ALLOC_GS);
+ MATCH3D(3DSTATE_PUSH_CONSTANT_ALLOC_PS);
+ MATCH3D(3DSTATE_SO_DECL_LIST);
+ MATCH3D(3DSTATE_SO_BUFFER);
+ MATCH3D(3DSTATE_BINDING_TABLE_POOL_ALLOC);
+ MATCH3D(3DSTATE_SAMPLE_PATTERN);
+ MATCH3D(3DSTATE_3D_MODE);
+ MATCH3D(3DSTATE_SUBSLICE_HASH_TABLE);
+ MATCH3D(3DSTATE_SLICE_TABLE_STATE_POINTERS);
+ MATCH3D(3DSTATE_PTBR_TILE_PASS_INFO);
+
+ default:
+ drm_printf(p, "[%#010x] unknown GFXPIPE command (pipeline=%#x, opcode=%#x, subopcode=%#x), likely %d dwords\n",
+ *dw, pipeline, opcode, subopcode, numdw);
+ return numdw;
+ }
+}
+
+void xe_lrc_dump_default(struct drm_printer *p,
+ struct xe_gt *gt,
+ enum xe_engine_class hwe_class)
+{
+ u32 *dw;
+ int remaining_dw, num_dw;
+
+ if (!gt->default_lrc[hwe_class]) {
+ drm_printf(p, "No default LRC for class %d\n", hwe_class);
+ return;
+ }
+
+ /*
+ * Skip the beginning of the LRC since it contains the per-process
+ * hardware status page.
+ */
+ dw = gt->default_lrc[hwe_class] + LRC_PPHWSP_SIZE;
+ remaining_dw = (xe_lrc_size(gt_to_xe(gt), hwe_class) - LRC_PPHWSP_SIZE) / 4;
+
+ while (remaining_dw > 0) {
+ if ((*dw & XE_INSTR_CMD_TYPE) == XE_INSTR_MI) {
+ num_dw = dump_mi_command(p, gt, dw, remaining_dw);
+ } else if ((*dw & XE_INSTR_CMD_TYPE) == XE_INSTR_GFXPIPE) {
+ num_dw = dump_gfxpipe_command(p, gt, dw, remaining_dw);
+ } else {
+ num_dw = min(instr_dw(*dw), remaining_dw);
+ drm_printf(p, "[%#10x] Unknown instruction of type %#x, likely %d dwords\n",
+ *dw, REG_FIELD_GET(XE_INSTR_CMD_TYPE, *dw),
+ num_dw);
+ }
+
+ dw += num_dw;
+ remaining_dw -= num_dw;
+ }
+}
+
+struct instr_state {
+ u32 instr;
+ u16 num_dw;
+};
+
+static const struct instr_state xe_hpg_svg_state[] = {
+ { .instr = CMD_3DSTATE_CONSTANT_VS, .num_dw = 11 },
+ { .instr = CMD_3DSTATE_CONSTANT_HS, .num_dw = 11 },
+ { .instr = CMD_3DSTATE_CONSTANT_DS, .num_dw = 11 },
+ { .instr = CMD_3DSTATE_CONSTANT_GS, .num_dw = 11 },
+ { .instr = CMD_3DSTATE_VERTEX_ELEMENTS, .num_dw = 69 },
+ { .instr = CMD_3DSTATE_VF_COMPONENT_PACKING, .num_dw = 5 },
+ { .instr = CMD_3DSTATE_VF_SGVS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_VF_SGVS_2, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_VS, .num_dw = 9 },
+ { .instr = CMD_3DSTATE_BINDING_TABLE_POINTERS_VS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_SAMPLER_STATE_POINTERS_VS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_VS, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_STREAMOUT, .num_dw = 5 },
+ { .instr = CMD_3DSTATE_SO_BUFFER_INDEX_0, .num_dw = 8 },
+ { .instr = CMD_3DSTATE_SO_BUFFER_INDEX_1, .num_dw = 8 },
+ { .instr = CMD_3DSTATE_SO_BUFFER_INDEX_2, .num_dw = 8 },
+ { .instr = CMD_3DSTATE_SO_BUFFER_INDEX_3, .num_dw = 8 },
+ { .instr = CMD_3DSTATE_CLIP, .num_dw = 4 },
+ { .instr = CMD_3DSTATE_PRIMITIVE_REPLICATION, .num_dw = 6 },
+ { .instr = CMD_3DSTATE_CLIP_MESH, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_SF, .num_dw = 4 },
+ { .instr = CMD_3DSTATE_SCISSOR_STATE_POINTERS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_VIEWPORT_STATE_POINTERS_SF_CLIP, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_RASTER, .num_dw = 5 },
+ { .instr = CMD_3DSTATE_TBIMR_TILE_PASS_INFO, .num_dw = 4 },
+ { .instr = CMD_3DSTATE_WM_HZ_OP, .num_dw = 6 },
+ { .instr = CMD_3DSTATE_MULTISAMPLE, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_HS, .num_dw = 9 },
+ { .instr = CMD_3DSTATE_BINDING_TABLE_POINTERS_HS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_SAMPLER_STATE_POINTERS_HS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_HS, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_TASK_CONTROL, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_TASK_SHADER, .num_dw = 7 },
+ { .instr = CMD_3DSTATE_TASK_SHADER_DATA, .num_dw = 10 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_TASK, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_TE, .num_dw = 5 },
+ { .instr = CMD_3DSTATE_TASK_REDISTRIB, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_DS, .num_dw = 11 },
+ { .instr = CMD_3DSTATE_BINDING_TABLE_POINTERS_DS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_SAMPLER_STATE_POINTERS_DS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_DS, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_GS, .num_dw = 10 },
+ { .instr = CMD_3DSTATE_BINDING_TABLE_POINTERS_GS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_SAMPLER_STATE_POINTERS_GS, .num_dw = 2 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_GS, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_MESH_CONTROL, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_MESH_SHADER_DATA, .num_dw = 10 },
+ { .instr = CMD_3DSTATE_URB_ALLOC_MESH, .num_dw = 3 },
+ { .instr = CMD_3DSTATE_MESH_SHADER, .num_dw = 8 },
+ { .instr = CMD_3DSTATE_DRAWING_RECTANGLE, .num_dw = 4 },
+};
+
+void xe_lrc_emit_hwe_state_instructions(struct xe_exec_queue *q, struct xe_bb *bb)
+{
+ struct xe_gt *gt = q->hwe->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ const struct instr_state *state_table = NULL;
+ int state_table_size = 0;
+
+ /*
+ * At the moment we only need to emit non-register state for the RCS
+ * engine.
+ */
+ if (q->hwe->class != XE_ENGINE_CLASS_RENDER)
+ return;
+
+ switch (GRAPHICS_VERx100(xe)) {
+ case 1255:
+ case 1270 ... 2004:
+ state_table = xe_hpg_svg_state;
+ state_table_size = ARRAY_SIZE(xe_hpg_svg_state);
+ break;
+ default:
+ xe_gt_dbg(gt, "No non-register state to emit on graphics ver %d.%02d\n",
+ GRAPHICS_VER(xe), GRAPHICS_VERx100(xe) % 100);
+ return;
+ }
+
+ for (int i = 0; i < state_table_size; i++) {
+ u32 instr = state_table[i].instr;
+ u16 num_dw = state_table[i].num_dw;
+ bool is_single_dw = ((instr & GFXPIPE_PIPELINE) == PIPELINE_SINGLE_DW);
+
+ xe_gt_assert(gt, (instr & XE_INSTR_CMD_TYPE) == XE_INSTR_GFXPIPE);
+ xe_gt_assert(gt, num_dw != 0);
+ xe_gt_assert(gt, is_single_dw ^ (num_dw > 1));
+
+ /*
+ * Xe2's SVG context is the same as the one on DG2 / MTL
+ * except that 3DSTATE_DRAWING_RECTANGLE (non-pipelined) has
+ * been replaced by 3DSTATE_DRAWING_RECTANGLE_FAST (pipelined).
+ * Just make the replacement here rather than defining a
+ * whole separate table for the single trivial change.
+ */
+ if (GRAPHICS_VER(xe) >= 20 &&
+ instr == CMD_3DSTATE_DRAWING_RECTANGLE)
+ instr = CMD_3DSTATE_DRAWING_RECTANGLE_FAST;
+
+ bb->cs[bb->len] = instr;
+ if (!is_single_dw)
+ bb->cs[bb->len] |= (num_dw - 2);
+
+ bb->len += num_dw;
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_lrc.h b/drivers/gpu/drm/xe/xe_lrc.h
new file mode 100644
index 000000000000..28b1d3f404d4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lrc.h
@@ -0,0 +1,58 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+#ifndef _XE_LRC_H_
+#define _XE_LRC_H_
+
+#include "xe_lrc_types.h"
+
+struct drm_printer;
+struct xe_bb;
+struct xe_device;
+struct xe_exec_queue;
+enum xe_engine_class;
+struct xe_hw_engine;
+struct xe_vm;
+
+#define LRC_PPHWSP_SCRATCH_ADDR (0x34 * 4)
+
+int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
+ struct xe_exec_queue *q, struct xe_vm *vm, u32 ring_size);
+void xe_lrc_finish(struct xe_lrc *lrc);
+
+size_t xe_lrc_size(struct xe_device *xe, enum xe_engine_class class);
+u32 xe_lrc_pphwsp_offset(struct xe_lrc *lrc);
+
+void xe_lrc_set_ring_head(struct xe_lrc *lrc, u32 head);
+u32 xe_lrc_ring_head(struct xe_lrc *lrc);
+u32 xe_lrc_ring_space(struct xe_lrc *lrc);
+void xe_lrc_write_ring(struct xe_lrc *lrc, const void *data, size_t size);
+
+u32 xe_lrc_ggtt_addr(struct xe_lrc *lrc);
+u32 *xe_lrc_regs(struct xe_lrc *lrc);
+
+u32 xe_lrc_read_ctx_reg(struct xe_lrc *lrc, int reg_nr);
+void xe_lrc_write_ctx_reg(struct xe_lrc *lrc, int reg_nr, u32 val);
+
+u64 xe_lrc_descriptor(struct xe_lrc *lrc);
+
+u32 xe_lrc_seqno_ggtt_addr(struct xe_lrc *lrc);
+struct dma_fence *xe_lrc_create_seqno_fence(struct xe_lrc *lrc);
+s32 xe_lrc_seqno(struct xe_lrc *lrc);
+
+u32 xe_lrc_start_seqno_ggtt_addr(struct xe_lrc *lrc);
+s32 xe_lrc_start_seqno(struct xe_lrc *lrc);
+
+u32 xe_lrc_parallel_ggtt_addr(struct xe_lrc *lrc);
+struct iosys_map xe_lrc_parallel_map(struct xe_lrc *lrc);
+
+size_t xe_lrc_skip_size(struct xe_device *xe);
+
+void xe_lrc_dump_default(struct drm_printer *p,
+ struct xe_gt *gt,
+ enum xe_engine_class);
+
+void xe_lrc_emit_hwe_state_instructions(struct xe_exec_queue *q, struct xe_bb *bb);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_lrc_types.h b/drivers/gpu/drm/xe/xe_lrc_types.h
new file mode 100644
index 000000000000..78220336062c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_lrc_types.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_LRC_TYPES_H_
+#define _XE_LRC_TYPES_H_
+
+#include "xe_hw_fence_types.h"
+
+struct xe_bo;
+
+/**
+ * struct xe_lrc - Logical ring context (LRC) and submission ring object
+ */
+struct xe_lrc {
+ /**
+ * @bo: buffer object (memory) for logical ring context, per process HW
+ * status page, and submission ring.
+ */
+ struct xe_bo *bo;
+
+ /** @tile: tile which this LRC belongs to */
+ struct xe_tile *tile;
+
+ /** @flags: LRC flags */
+ u32 flags;
+
+ /** @ring: submission ring state */
+ struct {
+ /** @size: size of submission ring */
+ u32 size;
+ /** @tail: tail of submission ring */
+ u32 tail;
+ /** @old_tail: shadow of tail */
+ u32 old_tail;
+ } ring;
+
+ /** @desc: LRC descriptor */
+ u64 desc;
+
+ /** @fence_ctx: context for hw fence */
+ struct xe_hw_fence_ctx fence_ctx;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_macros.h b/drivers/gpu/drm/xe/xe_macros.h
new file mode 100644
index 000000000000..daf56c846d03
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_macros.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_MACROS_H_
+#define _XE_MACROS_H_
+
+#include <linux/bug.h>
+
+#define XE_WARN_ON WARN_ON
+
+#define XE_IOCTL_DBG(xe, cond) \
+ ((cond) && (drm_dbg(&(xe)->drm, \
+ "Ioctl argument check failed at %s:%d: %s", \
+ __FILE__, __LINE__, #cond), 1))
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_map.h b/drivers/gpu/drm/xe/xe_map.h
new file mode 100644
index 000000000000..f62e0c8b67ab
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_map.h
@@ -0,0 +1,93 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_MAP_H_
+#define _XE_MAP_H_
+
+#include <linux/iosys-map.h>
+
+#include <xe_device.h>
+
+/**
+ * DOC: Map layer
+ *
+ * All access to any memory shared with a device (both sysmem and vram) in the
+ * XE driver should go through this layer (xe_map). This layer is built on top
+ * of :ref:`driver-api/device-io:Generalizing Access to System and I/O Memory`
+ * and with extra hooks into the XE driver that allows adding asserts to memory
+ * accesses (e.g. for blocking runtime_pm D3Cold on Discrete Graphics).
+ */
+
+static inline void xe_map_memcpy_to(struct xe_device *xe, struct iosys_map *dst,
+ size_t dst_offset, const void *src,
+ size_t len)
+{
+ xe_device_assert_mem_access(xe);
+ iosys_map_memcpy_to(dst, dst_offset, src, len);
+}
+
+static inline void xe_map_memcpy_from(struct xe_device *xe, void *dst,
+ const struct iosys_map *src,
+ size_t src_offset, size_t len)
+{
+ xe_device_assert_mem_access(xe);
+ iosys_map_memcpy_from(dst, src, src_offset, len);
+}
+
+static inline void xe_map_memset(struct xe_device *xe,
+ struct iosys_map *dst, size_t offset,
+ int value, size_t len)
+{
+ xe_device_assert_mem_access(xe);
+ iosys_map_memset(dst, offset, value, len);
+}
+
+/* FIXME: We likely should kill these two functions sooner or later */
+static inline u32 xe_map_read32(struct xe_device *xe, struct iosys_map *map)
+{
+ xe_device_assert_mem_access(xe);
+
+ if (map->is_iomem)
+ return readl(map->vaddr_iomem);
+ else
+ return READ_ONCE(*(u32 *)map->vaddr);
+}
+
+static inline void xe_map_write32(struct xe_device *xe, struct iosys_map *map,
+ u32 val)
+{
+ xe_device_assert_mem_access(xe);
+
+ if (map->is_iomem)
+ writel(val, map->vaddr_iomem);
+ else
+ *(u32 *)map->vaddr = val;
+}
+
+#define xe_map_rd(xe__, map__, offset__, type__) ({ \
+ struct xe_device *__xe = xe__; \
+ xe_device_assert_mem_access(__xe); \
+ iosys_map_rd(map__, offset__, type__); \
+})
+
+#define xe_map_wr(xe__, map__, offset__, type__, val__) ({ \
+ struct xe_device *__xe = xe__; \
+ xe_device_assert_mem_access(__xe); \
+ iosys_map_wr(map__, offset__, type__, val__); \
+})
+
+#define xe_map_rd_field(xe__, map__, struct_offset__, struct_type__, field__) ({ \
+ struct xe_device *__xe = xe__; \
+ xe_device_assert_mem_access(__xe); \
+ iosys_map_rd_field(map__, struct_offset__, struct_type__, field__); \
+})
+
+#define xe_map_wr_field(xe__, map__, struct_offset__, struct_type__, field__, val__) ({ \
+ struct xe_device *__xe = xe__; \
+ xe_device_assert_mem_access(__xe); \
+ iosys_map_wr_field(map__, struct_offset__, struct_type__, field__, val__); \
+})
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
new file mode 100644
index 000000000000..adf1dab5eba2
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_migrate.c
@@ -0,0 +1,1410 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2020 Intel Corporation
+ */
+
+#include "xe_migrate.h"
+
+#include <linux/bitfield.h>
+#include <linux/sizes.h>
+
+#include <drm/drm_managed.h>
+#include <drm/ttm/ttm_tt.h>
+#include <drm/xe_drm.h>
+
+#include "generated/xe_wa_oob.h"
+#include "instructions/xe_mi_commands.h"
+#include "regs/xe_gpu_commands.h"
+#include "tests/xe_test.h"
+#include "xe_assert.h"
+#include "xe_bb.h"
+#include "xe_bo.h"
+#include "xe_exec_queue.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_hw_engine.h"
+#include "xe_lrc.h"
+#include "xe_map.h"
+#include "xe_mocs.h"
+#include "xe_pt.h"
+#include "xe_res_cursor.h"
+#include "xe_sched_job.h"
+#include "xe_sync.h"
+#include "xe_trace.h"
+#include "xe_vm.h"
+#include "xe_wa.h"
+
+/**
+ * struct xe_migrate - migrate context.
+ */
+struct xe_migrate {
+ /** @q: Default exec queue used for migration */
+ struct xe_exec_queue *q;
+ /** @tile: Backpointer to the tile this struct xe_migrate belongs to. */
+ struct xe_tile *tile;
+ /** @job_mutex: Timeline mutex for @eng. */
+ struct mutex job_mutex;
+ /** @pt_bo: Page-table buffer object. */
+ struct xe_bo *pt_bo;
+ /** @batch_base_ofs: VM offset of the migration batch buffer */
+ u64 batch_base_ofs;
+ /** @usm_batch_base_ofs: VM offset of the usm batch buffer */
+ u64 usm_batch_base_ofs;
+ /** @cleared_mem_ofs: VM offset of @cleared_bo. */
+ u64 cleared_mem_ofs;
+ /**
+ * @fence: dma-fence representing the last migration job batch.
+ * Protected by @job_mutex.
+ */
+ struct dma_fence *fence;
+ /**
+ * @vm_update_sa: For integrated, used to suballocate page-tables
+ * out of the pt_bo.
+ */
+ struct drm_suballoc_manager vm_update_sa;
+};
+
+#define MAX_PREEMPTDISABLE_TRANSFER SZ_8M /* Around 1ms. */
+#define MAX_CCS_LIMITED_TRANSFER SZ_4M /* XE_PAGE_SIZE * (FIELD_MAX(XE2_CCS_SIZE_MASK) + 1) */
+#define NUM_KERNEL_PDE 17
+#define NUM_PT_SLOTS 32
+#define LEVEL0_PAGE_TABLE_ENCODE_SIZE SZ_2M
+
+/**
+ * xe_tile_migrate_engine() - Get this tile's migrate engine.
+ * @tile: The tile.
+ *
+ * Returns the default migrate engine of this tile.
+ * TODO: Perhaps this function is slightly misplaced, and even unneeded?
+ *
+ * Return: The default migrate engine
+ */
+struct xe_exec_queue *xe_tile_migrate_engine(struct xe_tile *tile)
+{
+ return tile->migrate->q;
+}
+
+static void xe_migrate_fini(struct drm_device *dev, void *arg)
+{
+ struct xe_migrate *m = arg;
+
+ xe_vm_lock(m->q->vm, false);
+ xe_bo_unpin(m->pt_bo);
+ xe_vm_unlock(m->q->vm);
+
+ dma_fence_put(m->fence);
+ xe_bo_put(m->pt_bo);
+ drm_suballoc_manager_fini(&m->vm_update_sa);
+ mutex_destroy(&m->job_mutex);
+ xe_vm_close_and_put(m->q->vm);
+ xe_exec_queue_put(m->q);
+}
+
+static u64 xe_migrate_vm_addr(u64 slot, u32 level)
+{
+ XE_WARN_ON(slot >= NUM_PT_SLOTS);
+
+ /* First slot is reserved for mapping of PT bo and bb, start from 1 */
+ return (slot + 1ULL) << xe_pt_shift(level + 1);
+}
+
+static u64 xe_migrate_vram_ofs(struct xe_device *xe, u64 addr)
+{
+ /*
+ * Remove the DPA to get a correct offset into identity table for the
+ * migrate offset
+ */
+ addr -= xe->mem.vram.dpa_base;
+ return addr + (256ULL << xe_pt_shift(2));
+}
+
+static int xe_migrate_prepare_vm(struct xe_tile *tile, struct xe_migrate *m,
+ struct xe_vm *vm)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ u16 pat_index = xe->pat.idx[XE_CACHE_WB];
+ u8 id = tile->id;
+ u32 num_entries = NUM_PT_SLOTS, num_level = vm->pt_root[id]->level;
+ u32 map_ofs, level, i;
+ struct xe_bo *bo, *batch = tile->mem.kernel_bb_pool->bo;
+ u64 entry;
+
+ /* Can't bump NUM_PT_SLOTS too high */
+ BUILD_BUG_ON(NUM_PT_SLOTS > SZ_2M/XE_PAGE_SIZE);
+ /* Must be a multiple of 64K to support all platforms */
+ BUILD_BUG_ON(NUM_PT_SLOTS * XE_PAGE_SIZE % SZ_64K);
+ /* And one slot reserved for the 4KiB page table updates */
+ BUILD_BUG_ON(!(NUM_KERNEL_PDE & 1));
+
+ /* Need to be sure everything fits in the first PT, or create more */
+ xe_tile_assert(tile, m->batch_base_ofs + batch->size < SZ_2M);
+
+ bo = xe_bo_create_pin_map(vm->xe, tile, vm,
+ num_entries * XE_PAGE_SIZE,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_PINNED_BIT);
+ if (IS_ERR(bo))
+ return PTR_ERR(bo);
+
+ entry = vm->pt_ops->pde_encode_bo(bo, bo->size - XE_PAGE_SIZE, pat_index);
+ xe_pt_write(xe, &vm->pt_root[id]->bo->vmap, 0, entry);
+
+ map_ofs = (num_entries - num_level) * XE_PAGE_SIZE;
+
+ /* Map the entire BO in our level 0 pt */
+ for (i = 0, level = 0; i < num_entries; level++) {
+ entry = vm->pt_ops->pte_encode_bo(bo, i * XE_PAGE_SIZE,
+ pat_index, 0);
+
+ xe_map_wr(xe, &bo->vmap, map_ofs + level * 8, u64, entry);
+
+ if (vm->flags & XE_VM_FLAG_64K)
+ i += 16;
+ else
+ i += 1;
+ }
+
+ if (!IS_DGFX(xe)) {
+ /* Write out batch too */
+ m->batch_base_ofs = NUM_PT_SLOTS * XE_PAGE_SIZE;
+ if (xe->info.has_usm) {
+ batch = tile->primary_gt->usm.bb_pool->bo;
+ m->usm_batch_base_ofs = m->batch_base_ofs;
+ }
+
+ for (i = 0; i < batch->size;
+ i += vm->flags & XE_VM_FLAG_64K ? XE_64K_PAGE_SIZE :
+ XE_PAGE_SIZE) {
+ entry = vm->pt_ops->pte_encode_bo(batch, i,
+ pat_index, 0);
+
+ xe_map_wr(xe, &bo->vmap, map_ofs + level * 8, u64,
+ entry);
+ level++;
+ }
+ } else {
+ u64 batch_addr = xe_bo_addr(batch, 0, XE_PAGE_SIZE);
+
+ m->batch_base_ofs = xe_migrate_vram_ofs(xe, batch_addr);
+
+ if (xe->info.has_usm) {
+ batch = tile->primary_gt->usm.bb_pool->bo;
+ batch_addr = xe_bo_addr(batch, 0, XE_PAGE_SIZE);
+ m->usm_batch_base_ofs = xe_migrate_vram_ofs(xe, batch_addr);
+ }
+ }
+
+ for (level = 1; level < num_level; level++) {
+ u32 flags = 0;
+
+ if (vm->flags & XE_VM_FLAG_64K && level == 1)
+ flags = XE_PDE_64K;
+
+ entry = vm->pt_ops->pde_encode_bo(bo, map_ofs + (level - 1) *
+ XE_PAGE_SIZE, pat_index);
+ xe_map_wr(xe, &bo->vmap, map_ofs + XE_PAGE_SIZE * level, u64,
+ entry | flags);
+ }
+
+ /* Write PDE's that point to our BO. */
+ for (i = 0; i < num_entries - num_level; i++) {
+ entry = vm->pt_ops->pde_encode_bo(bo, i * XE_PAGE_SIZE,
+ pat_index);
+
+ xe_map_wr(xe, &bo->vmap, map_ofs + XE_PAGE_SIZE +
+ (i + 1) * 8, u64, entry);
+ }
+
+ /* Set up a 1GiB NULL mapping at 255GiB offset. */
+ level = 2;
+ xe_map_wr(xe, &bo->vmap, map_ofs + XE_PAGE_SIZE * level + 255 * 8, u64,
+ vm->pt_ops->pte_encode_addr(xe, 0, pat_index, level, IS_DGFX(xe), 0)
+ | XE_PTE_NULL);
+ m->cleared_mem_ofs = (255ULL << xe_pt_shift(level));
+
+ /* Identity map the entire vram at 256GiB offset */
+ if (IS_DGFX(xe)) {
+ u64 pos, ofs, flags;
+
+ level = 2;
+ ofs = map_ofs + XE_PAGE_SIZE * level + 256 * 8;
+ flags = vm->pt_ops->pte_encode_addr(xe, 0, pat_index, level,
+ true, 0);
+
+ /*
+ * Use 1GB pages, it shouldn't matter the physical amount of
+ * vram is less, when we don't access it.
+ */
+ for (pos = xe->mem.vram.dpa_base;
+ pos < xe->mem.vram.actual_physical_size + xe->mem.vram.dpa_base;
+ pos += SZ_1G, ofs += 8)
+ xe_map_wr(xe, &bo->vmap, ofs, u64, pos | flags);
+ }
+
+ /*
+ * Example layout created above, with root level = 3:
+ * [PT0...PT7]: kernel PT's for copy/clear; 64 or 4KiB PTE's
+ * [PT8]: Kernel PT for VM_BIND, 4 KiB PTE's
+ * [PT9...PT28]: Userspace PT's for VM_BIND, 4 KiB PTE's
+ * [PT29 = PDE 0] [PT30 = PDE 1] [PT31 = PDE 2]
+ *
+ * This makes the lowest part of the VM point to the pagetables.
+ * Hence the lowest 2M in the vm should point to itself, with a few writes
+ * and flushes, other parts of the VM can be used either for copying and
+ * clearing.
+ *
+ * For performance, the kernel reserves PDE's, so about 20 are left
+ * for async VM updates.
+ *
+ * To make it easier to work, each scratch PT is put in slot (1 + PT #)
+ * everywhere, this allows lockless updates to scratch pages by using
+ * the different addresses in VM.
+ */
+#define NUM_VMUSA_UNIT_PER_PAGE 32
+#define VM_SA_UPDATE_UNIT_SIZE (XE_PAGE_SIZE / NUM_VMUSA_UNIT_PER_PAGE)
+#define NUM_VMUSA_WRITES_PER_UNIT (VM_SA_UPDATE_UNIT_SIZE / sizeof(u64))
+ drm_suballoc_manager_init(&m->vm_update_sa,
+ (map_ofs / XE_PAGE_SIZE - NUM_KERNEL_PDE) *
+ NUM_VMUSA_UNIT_PER_PAGE, 0);
+
+ m->pt_bo = bo;
+ return 0;
+}
+
+/*
+ * Due to workaround 16017236439, odd instance hardware copy engines are
+ * faster than even instance ones.
+ * This function returns the mask involving all fast copy engines and the
+ * reserved copy engine to be used as logical mask for migrate engine.
+ * Including the reserved copy engine is required to avoid deadlocks due to
+ * migrate jobs servicing the faults gets stuck behind the job that faulted.
+ */
+static u32 xe_migrate_usm_logical_mask(struct xe_gt *gt)
+{
+ u32 logical_mask = 0;
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(hwe, gt, id) {
+ if (hwe->class != XE_ENGINE_CLASS_COPY)
+ continue;
+
+ if (!XE_WA(gt, 16017236439) ||
+ xe_gt_is_usm_hwe(gt, hwe) || hwe->instance & 1)
+ logical_mask |= BIT(hwe->logical_instance);
+ }
+
+ return logical_mask;
+}
+
+/**
+ * xe_migrate_init() - Initialize a migrate context
+ * @tile: Back-pointer to the tile we're initializing for.
+ *
+ * Return: Pointer to a migrate context on success. Error pointer on error.
+ */
+struct xe_migrate *xe_migrate_init(struct xe_tile *tile)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_gt *primary_gt = tile->primary_gt;
+ struct xe_migrate *m;
+ struct xe_vm *vm;
+ int err;
+
+ m = drmm_kzalloc(&xe->drm, sizeof(*m), GFP_KERNEL);
+ if (!m)
+ return ERR_PTR(-ENOMEM);
+
+ m->tile = tile;
+
+ /* Special layout, prepared below.. */
+ vm = xe_vm_create(xe, XE_VM_FLAG_MIGRATION |
+ XE_VM_FLAG_SET_TILE_ID(tile));
+ if (IS_ERR(vm))
+ return ERR_CAST(vm);
+
+ xe_vm_lock(vm, false);
+ err = xe_migrate_prepare_vm(tile, m, vm);
+ xe_vm_unlock(vm);
+ if (err) {
+ xe_vm_close_and_put(vm);
+ return ERR_PTR(err);
+ }
+
+ if (xe->info.has_usm) {
+ struct xe_hw_engine *hwe = xe_gt_hw_engine(primary_gt,
+ XE_ENGINE_CLASS_COPY,
+ primary_gt->usm.reserved_bcs_instance,
+ false);
+ u32 logical_mask = xe_migrate_usm_logical_mask(primary_gt);
+
+ if (!hwe || !logical_mask)
+ return ERR_PTR(-EINVAL);
+
+ m->q = xe_exec_queue_create(xe, vm, logical_mask, 1, hwe,
+ EXEC_QUEUE_FLAG_KERNEL |
+ EXEC_QUEUE_FLAG_PERMANENT);
+ } else {
+ m->q = xe_exec_queue_create_class(xe, primary_gt, vm,
+ XE_ENGINE_CLASS_COPY,
+ EXEC_QUEUE_FLAG_KERNEL |
+ EXEC_QUEUE_FLAG_PERMANENT);
+ }
+ if (IS_ERR(m->q)) {
+ xe_vm_close_and_put(vm);
+ return ERR_CAST(m->q);
+ }
+ if (xe->info.has_usm)
+ m->q->priority = XE_EXEC_QUEUE_PRIORITY_KERNEL;
+
+ mutex_init(&m->job_mutex);
+
+ err = drmm_add_action_or_reset(&xe->drm, xe_migrate_fini, m);
+ if (err)
+ return ERR_PTR(err);
+
+ return m;
+}
+
+static u64 max_mem_transfer_per_pass(struct xe_device *xe)
+{
+ if (!IS_DGFX(xe) && xe_device_has_flat_ccs(xe))
+ return MAX_CCS_LIMITED_TRANSFER;
+
+ return MAX_PREEMPTDISABLE_TRANSFER;
+}
+
+static u64 xe_migrate_res_sizes(struct xe_device *xe, struct xe_res_cursor *cur)
+{
+ /*
+ * For VRAM we use identity mapped pages so we are limited to current
+ * cursor size. For system we program the pages ourselves so we have no
+ * such limitation.
+ */
+ return min_t(u64, max_mem_transfer_per_pass(xe),
+ mem_type_is_vram(cur->mem_type) ? cur->size :
+ cur->remaining);
+}
+
+static u32 pte_update_size(struct xe_migrate *m,
+ bool is_vram,
+ struct ttm_resource *res,
+ struct xe_res_cursor *cur,
+ u64 *L0, u64 *L0_ofs, u32 *L0_pt,
+ u32 cmd_size, u32 pt_ofs, u32 avail_pts)
+{
+ u32 cmds = 0;
+
+ *L0_pt = pt_ofs;
+ if (!is_vram) {
+ /* Clip L0 to available size */
+ u64 size = min(*L0, (u64)avail_pts * SZ_2M);
+ u64 num_4k_pages = DIV_ROUND_UP(size, XE_PAGE_SIZE);
+
+ *L0 = size;
+ *L0_ofs = xe_migrate_vm_addr(pt_ofs, 0);
+
+ /* MI_STORE_DATA_IMM */
+ cmds += 3 * DIV_ROUND_UP(num_4k_pages, 0x1ff);
+
+ /* PDE qwords */
+ cmds += num_4k_pages * 2;
+
+ /* Each chunk has a single blit command */
+ cmds += cmd_size;
+ } else {
+ /* Offset into identity map. */
+ *L0_ofs = xe_migrate_vram_ofs(tile_to_xe(m->tile),
+ cur->start + vram_region_gpu_offset(res));
+ cmds += cmd_size;
+ }
+
+ return cmds;
+}
+
+static void emit_pte(struct xe_migrate *m,
+ struct xe_bb *bb, u32 at_pt,
+ bool is_vram, bool is_comp_pte,
+ struct xe_res_cursor *cur,
+ u32 size, struct xe_bo *bo)
+{
+ struct xe_device *xe = tile_to_xe(m->tile);
+
+ u16 pat_index;
+ u32 ptes;
+ u64 ofs = at_pt * XE_PAGE_SIZE;
+ u64 cur_ofs;
+
+ /* Indirect access needs compression enabled uncached PAT index */
+ if (GRAPHICS_VERx100(xe) >= 2000)
+ pat_index = is_comp_pte ? xe->pat.idx[XE_CACHE_NONE_COMPRESSION] :
+ xe->pat.idx[XE_CACHE_NONE];
+ else
+ pat_index = xe->pat.idx[XE_CACHE_WB];
+
+ /*
+ * FIXME: Emitting VRAM PTEs to L0 PTs is forbidden. Currently
+ * we're only emitting VRAM PTEs during sanity tests, so when
+ * that's moved to a Kunit test, we should condition VRAM PTEs
+ * on running tests.
+ */
+
+ ptes = DIV_ROUND_UP(size, XE_PAGE_SIZE);
+
+ while (ptes) {
+ u32 chunk = min(0x1ffU, ptes);
+
+ bb->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(chunk);
+ bb->cs[bb->len++] = ofs;
+ bb->cs[bb->len++] = 0;
+
+ cur_ofs = ofs;
+ ofs += chunk * 8;
+ ptes -= chunk;
+
+ while (chunk--) {
+ u64 addr, flags = 0;
+ bool devmem = false;
+
+ addr = xe_res_dma(cur) & PAGE_MASK;
+ if (is_vram) {
+ /* Is this a 64K PTE entry? */
+ if ((m->q->vm->flags & XE_VM_FLAG_64K) &&
+ !(cur_ofs & (16 * 8 - 1))) {
+ xe_tile_assert(m->tile, IS_ALIGNED(addr, SZ_64K));
+ flags |= XE_PTE_PS64;
+ }
+
+ addr += vram_region_gpu_offset(bo->ttm.resource);
+ devmem = true;
+ }
+
+ addr = m->q->vm->pt_ops->pte_encode_addr(m->tile->xe,
+ addr, pat_index,
+ 0, devmem, flags);
+ bb->cs[bb->len++] = lower_32_bits(addr);
+ bb->cs[bb->len++] = upper_32_bits(addr);
+
+ xe_res_next(cur, min_t(u32, size, PAGE_SIZE));
+ cur_ofs += 8;
+ }
+ }
+}
+
+#define EMIT_COPY_CCS_DW 5
+static void emit_copy_ccs(struct xe_gt *gt, struct xe_bb *bb,
+ u64 dst_ofs, bool dst_is_indirect,
+ u64 src_ofs, bool src_is_indirect,
+ u32 size)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 *cs = bb->cs + bb->len;
+ u32 num_ccs_blks;
+ u32 num_pages;
+ u32 ccs_copy_size;
+ u32 mocs;
+
+ if (GRAPHICS_VERx100(xe) >= 2000) {
+ num_pages = DIV_ROUND_UP(size, XE_PAGE_SIZE);
+ xe_gt_assert(gt, FIELD_FIT(XE2_CCS_SIZE_MASK, num_pages - 1));
+
+ ccs_copy_size = REG_FIELD_PREP(XE2_CCS_SIZE_MASK, num_pages - 1);
+ mocs = FIELD_PREP(XE2_XY_CTRL_SURF_MOCS_INDEX_MASK, gt->mocs.uc_index);
+
+ } else {
+ num_ccs_blks = DIV_ROUND_UP(xe_device_ccs_bytes(gt_to_xe(gt), size),
+ NUM_CCS_BYTES_PER_BLOCK);
+ xe_gt_assert(gt, FIELD_FIT(CCS_SIZE_MASK, num_ccs_blks - 1));
+
+ ccs_copy_size = REG_FIELD_PREP(CCS_SIZE_MASK, num_ccs_blks - 1);
+ mocs = FIELD_PREP(XY_CTRL_SURF_MOCS_MASK, gt->mocs.uc_index);
+ }
+
+ *cs++ = XY_CTRL_SURF_COPY_BLT |
+ (src_is_indirect ? 0x0 : 0x1) << SRC_ACCESS_TYPE_SHIFT |
+ (dst_is_indirect ? 0x0 : 0x1) << DST_ACCESS_TYPE_SHIFT |
+ ccs_copy_size;
+ *cs++ = lower_32_bits(src_ofs);
+ *cs++ = upper_32_bits(src_ofs) | mocs;
+ *cs++ = lower_32_bits(dst_ofs);
+ *cs++ = upper_32_bits(dst_ofs) | mocs;
+
+ bb->len = cs - bb->cs;
+}
+
+#define EMIT_COPY_DW 10
+static void emit_copy(struct xe_gt *gt, struct xe_bb *bb,
+ u64 src_ofs, u64 dst_ofs, unsigned int size,
+ unsigned int pitch)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 mocs = 0;
+ u32 tile_y = 0;
+
+ xe_gt_assert(gt, size / pitch <= S16_MAX);
+ xe_gt_assert(gt, pitch / 4 <= S16_MAX);
+ xe_gt_assert(gt, pitch <= U16_MAX);
+
+ if (GRAPHICS_VER(xe) >= 20)
+ mocs = FIELD_PREP(XE2_XY_FAST_COPY_BLT_MOCS_INDEX_MASK, gt->mocs.uc_index);
+
+ if (GRAPHICS_VERx100(xe) >= 1250)
+ tile_y = XY_FAST_COPY_BLT_D1_SRC_TILE4 | XY_FAST_COPY_BLT_D1_DST_TILE4;
+
+ bb->cs[bb->len++] = XY_FAST_COPY_BLT_CMD | (10 - 2);
+ bb->cs[bb->len++] = XY_FAST_COPY_BLT_DEPTH_32 | pitch | tile_y | mocs;
+ bb->cs[bb->len++] = 0;
+ bb->cs[bb->len++] = (size / pitch) << 16 | pitch / 4;
+ bb->cs[bb->len++] = lower_32_bits(dst_ofs);
+ bb->cs[bb->len++] = upper_32_bits(dst_ofs);
+ bb->cs[bb->len++] = 0;
+ bb->cs[bb->len++] = pitch | mocs;
+ bb->cs[bb->len++] = lower_32_bits(src_ofs);
+ bb->cs[bb->len++] = upper_32_bits(src_ofs);
+}
+
+static int job_add_deps(struct xe_sched_job *job, struct dma_resv *resv,
+ enum dma_resv_usage usage)
+{
+ return drm_sched_job_add_resv_dependencies(&job->drm, resv, usage);
+}
+
+static u64 xe_migrate_batch_base(struct xe_migrate *m, bool usm)
+{
+ return usm ? m->usm_batch_base_ofs : m->batch_base_ofs;
+}
+
+static u32 xe_migrate_ccs_copy(struct xe_migrate *m,
+ struct xe_bb *bb,
+ u64 src_ofs, bool src_is_indirect,
+ u64 dst_ofs, bool dst_is_indirect, u32 dst_size,
+ u64 ccs_ofs, bool copy_ccs)
+{
+ struct xe_gt *gt = m->tile->primary_gt;
+ u32 flush_flags = 0;
+
+ if (xe_device_has_flat_ccs(gt_to_xe(gt)) && !copy_ccs && dst_is_indirect) {
+ /*
+ * If the src is already in vram, then it should already
+ * have been cleared by us, or has been populated by the
+ * user. Make sure we copy the CCS aux state as-is.
+ *
+ * Otherwise if the bo doesn't have any CCS metadata attached,
+ * we still need to clear it for security reasons.
+ */
+ u64 ccs_src_ofs = src_is_indirect ? src_ofs : m->cleared_mem_ofs;
+
+ emit_copy_ccs(gt, bb,
+ dst_ofs, true,
+ ccs_src_ofs, src_is_indirect, dst_size);
+
+ flush_flags = MI_FLUSH_DW_CCS;
+ } else if (copy_ccs) {
+ if (!src_is_indirect)
+ src_ofs = ccs_ofs;
+ else if (!dst_is_indirect)
+ dst_ofs = ccs_ofs;
+
+ xe_gt_assert(gt, src_is_indirect || dst_is_indirect);
+
+ emit_copy_ccs(gt, bb, dst_ofs, dst_is_indirect, src_ofs,
+ src_is_indirect, dst_size);
+ if (dst_is_indirect)
+ flush_flags = MI_FLUSH_DW_CCS;
+ }
+
+ return flush_flags;
+}
+
+/**
+ * xe_migrate_copy() - Copy content of TTM resources.
+ * @m: The migration context.
+ * @src_bo: The buffer object @src is currently bound to.
+ * @dst_bo: If copying between resources created for the same bo, set this to
+ * the same value as @src_bo. If copying between buffer objects, set it to
+ * the buffer object @dst is currently bound to.
+ * @src: The source TTM resource.
+ * @dst: The dst TTM resource.
+ * @copy_only_ccs: If true copy only CCS metadata
+ *
+ * Copies the contents of @src to @dst: On flat CCS devices,
+ * the CCS metadata is copied as well if needed, or if not present,
+ * the CCS metadata of @dst is cleared for security reasons.
+ *
+ * Return: Pointer to a dma_fence representing the last copy batch, or
+ * an error pointer on failure. If there is a failure, any copy operation
+ * started by the function call has been synced.
+ */
+struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
+ struct xe_bo *src_bo,
+ struct xe_bo *dst_bo,
+ struct ttm_resource *src,
+ struct ttm_resource *dst,
+ bool copy_only_ccs)
+{
+ struct xe_gt *gt = m->tile->primary_gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ struct dma_fence *fence = NULL;
+ u64 size = src_bo->size;
+ struct xe_res_cursor src_it, dst_it, ccs_it;
+ u64 src_L0_ofs, dst_L0_ofs;
+ u32 src_L0_pt, dst_L0_pt;
+ u64 src_L0, dst_L0;
+ int pass = 0;
+ int err;
+ bool src_is_pltt = src->mem_type == XE_PL_TT;
+ bool dst_is_pltt = dst->mem_type == XE_PL_TT;
+ bool src_is_vram = mem_type_is_vram(src->mem_type);
+ bool dst_is_vram = mem_type_is_vram(dst->mem_type);
+ bool copy_ccs = xe_device_has_flat_ccs(xe) &&
+ xe_bo_needs_ccs_pages(src_bo) && xe_bo_needs_ccs_pages(dst_bo);
+ bool copy_system_ccs = copy_ccs && (!src_is_vram || !dst_is_vram);
+
+ /* Copying CCS between two different BOs is not supported yet. */
+ if (XE_WARN_ON(copy_ccs && src_bo != dst_bo))
+ return ERR_PTR(-EINVAL);
+
+ if (src_bo != dst_bo && XE_WARN_ON(src_bo->size != dst_bo->size))
+ return ERR_PTR(-EINVAL);
+
+ if (!src_is_vram)
+ xe_res_first_sg(xe_bo_sg(src_bo), 0, size, &src_it);
+ else
+ xe_res_first(src, 0, size, &src_it);
+ if (!dst_is_vram)
+ xe_res_first_sg(xe_bo_sg(dst_bo), 0, size, &dst_it);
+ else
+ xe_res_first(dst, 0, size, &dst_it);
+
+ if (copy_system_ccs)
+ xe_res_first_sg(xe_bo_sg(src_bo), xe_bo_ccs_pages_start(src_bo),
+ PAGE_ALIGN(xe_device_ccs_bytes(xe, size)),
+ &ccs_it);
+
+ while (size) {
+ u32 batch_size = 2; /* arb_clear() + MI_BATCH_BUFFER_END */
+ struct xe_sched_job *job;
+ struct xe_bb *bb;
+ u32 flush_flags;
+ u32 update_idx;
+ u64 ccs_ofs, ccs_size;
+ u32 ccs_pt;
+
+ bool usm = xe->info.has_usm;
+ u32 avail_pts = max_mem_transfer_per_pass(xe) / LEVEL0_PAGE_TABLE_ENCODE_SIZE;
+
+ src_L0 = xe_migrate_res_sizes(xe, &src_it);
+ dst_L0 = xe_migrate_res_sizes(xe, &dst_it);
+
+ drm_dbg(&xe->drm, "Pass %u, sizes: %llu & %llu\n",
+ pass++, src_L0, dst_L0);
+
+ src_L0 = min(src_L0, dst_L0);
+
+ batch_size += pte_update_size(m, src_is_vram, src, &src_it, &src_L0,
+ &src_L0_ofs, &src_L0_pt, 0, 0,
+ avail_pts);
+
+ batch_size += pte_update_size(m, dst_is_vram, dst, &dst_it, &src_L0,
+ &dst_L0_ofs, &dst_L0_pt, 0,
+ avail_pts, avail_pts);
+
+ if (copy_system_ccs) {
+ ccs_size = xe_device_ccs_bytes(xe, src_L0);
+ batch_size += pte_update_size(m, false, NULL, &ccs_it, &ccs_size,
+ &ccs_ofs, &ccs_pt, 0,
+ 2 * avail_pts,
+ avail_pts);
+ }
+
+ /* Add copy commands size here */
+ batch_size += ((copy_only_ccs) ? 0 : EMIT_COPY_DW) +
+ ((xe_device_has_flat_ccs(xe) ? EMIT_COPY_CCS_DW : 0));
+
+ bb = xe_bb_new(gt, batch_size, usm);
+ if (IS_ERR(bb)) {
+ err = PTR_ERR(bb);
+ goto err_sync;
+ }
+
+ if (!src_is_vram)
+ emit_pte(m, bb, src_L0_pt, src_is_vram, true, &src_it, src_L0,
+ src_bo);
+ else
+ xe_res_next(&src_it, src_L0);
+
+ if (!dst_is_vram)
+ emit_pte(m, bb, dst_L0_pt, dst_is_vram, true, &dst_it, src_L0,
+ dst_bo);
+ else
+ xe_res_next(&dst_it, src_L0);
+
+ if (copy_system_ccs)
+ emit_pte(m, bb, ccs_pt, false, false, &ccs_it, ccs_size, src_bo);
+
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+ update_idx = bb->len;
+
+ if (!copy_only_ccs)
+ emit_copy(gt, bb, src_L0_ofs, dst_L0_ofs, src_L0, XE_PAGE_SIZE);
+
+ flush_flags = xe_migrate_ccs_copy(m, bb, src_L0_ofs,
+ IS_DGFX(xe) ? src_is_vram : src_is_pltt,
+ dst_L0_ofs,
+ IS_DGFX(xe) ? dst_is_vram : dst_is_pltt,
+ src_L0, ccs_ofs, copy_ccs);
+
+ mutex_lock(&m->job_mutex);
+ job = xe_bb_create_migration_job(m->q, bb,
+ xe_migrate_batch_base(m, usm),
+ update_idx);
+ if (IS_ERR(job)) {
+ err = PTR_ERR(job);
+ goto err;
+ }
+
+ xe_sched_job_add_migrate_flush(job, flush_flags);
+ if (!fence) {
+ err = job_add_deps(job, src_bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP);
+ if (!err && src_bo != dst_bo)
+ err = job_add_deps(job, dst_bo->ttm.base.resv,
+ DMA_RESV_USAGE_BOOKKEEP);
+ if (err)
+ goto err_job;
+ }
+
+ xe_sched_job_arm(job);
+ dma_fence_put(fence);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ dma_fence_put(m->fence);
+ m->fence = dma_fence_get(fence);
+
+ mutex_unlock(&m->job_mutex);
+
+ xe_bb_free(bb, fence);
+ size -= src_L0;
+ continue;
+
+err_job:
+ xe_sched_job_put(job);
+err:
+ mutex_unlock(&m->job_mutex);
+ xe_bb_free(bb, NULL);
+
+err_sync:
+ /* Sync partial copy if any. FIXME: under job_mutex? */
+ if (fence) {
+ dma_fence_wait(fence, false);
+ dma_fence_put(fence);
+ }
+
+ return ERR_PTR(err);
+ }
+
+ return fence;
+}
+
+static void emit_clear_link_copy(struct xe_gt *gt, struct xe_bb *bb, u64 src_ofs,
+ u32 size, u32 pitch)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 *cs = bb->cs + bb->len;
+ u32 len = PVC_MEM_SET_CMD_LEN_DW;
+
+ *cs++ = PVC_MEM_SET_CMD | PVC_MEM_SET_MATRIX | (len - 2);
+ *cs++ = pitch - 1;
+ *cs++ = (size / pitch) - 1;
+ *cs++ = pitch - 1;
+ *cs++ = lower_32_bits(src_ofs);
+ *cs++ = upper_32_bits(src_ofs);
+ if (GRAPHICS_VERx100(xe) >= 2000)
+ *cs++ = FIELD_PREP(XE2_MEM_SET_MOCS_INDEX_MASK, gt->mocs.uc_index);
+ else
+ *cs++ = FIELD_PREP(PVC_MEM_SET_MOCS_INDEX_MASK, gt->mocs.uc_index);
+
+ xe_gt_assert(gt, cs - bb->cs == len + bb->len);
+
+ bb->len += len;
+}
+
+static void emit_clear_main_copy(struct xe_gt *gt, struct xe_bb *bb,
+ u64 src_ofs, u32 size, u32 pitch, bool is_vram)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ u32 *cs = bb->cs + bb->len;
+ u32 len = XY_FAST_COLOR_BLT_DW;
+
+ if (GRAPHICS_VERx100(xe) < 1250)
+ len = 11;
+
+ *cs++ = XY_FAST_COLOR_BLT_CMD | XY_FAST_COLOR_BLT_DEPTH_32 |
+ (len - 2);
+ if (GRAPHICS_VERx100(xe) >= 2000)
+ *cs++ = FIELD_PREP(XE2_XY_FAST_COLOR_BLT_MOCS_INDEX_MASK, gt->mocs.uc_index) |
+ (pitch - 1);
+ else
+ *cs++ = FIELD_PREP(XY_FAST_COLOR_BLT_MOCS_MASK, gt->mocs.uc_index) |
+ (pitch - 1);
+ *cs++ = 0;
+ *cs++ = (size / pitch) << 16 | pitch / 4;
+ *cs++ = lower_32_bits(src_ofs);
+ *cs++ = upper_32_bits(src_ofs);
+ *cs++ = (is_vram ? 0x0 : 0x1) << XY_FAST_COLOR_BLT_MEM_TYPE_SHIFT;
+ *cs++ = 0;
+ *cs++ = 0;
+ *cs++ = 0;
+ *cs++ = 0;
+
+ if (len > 11) {
+ *cs++ = 0;
+ *cs++ = 0;
+ *cs++ = 0;
+ *cs++ = 0;
+ *cs++ = 0;
+ }
+
+ xe_gt_assert(gt, cs - bb->cs == len + bb->len);
+
+ bb->len += len;
+}
+
+static bool has_service_copy_support(struct xe_gt *gt)
+{
+ /*
+ * What we care about is whether the architecture was designed with
+ * service copy functionality (specifically the new MEM_SET / MEM_COPY
+ * instructions) so check the architectural engine list rather than the
+ * actual list since these instructions are usable on BCS0 even if
+ * all of the actual service copy engines (BCS1-BCS8) have been fused
+ * off.
+ */
+ return gt->info.__engine_mask & GENMASK(XE_HW_ENGINE_BCS8,
+ XE_HW_ENGINE_BCS1);
+}
+
+static u32 emit_clear_cmd_len(struct xe_gt *gt)
+{
+ if (has_service_copy_support(gt))
+ return PVC_MEM_SET_CMD_LEN_DW;
+ else
+ return XY_FAST_COLOR_BLT_DW;
+}
+
+static void emit_clear(struct xe_gt *gt, struct xe_bb *bb, u64 src_ofs,
+ u32 size, u32 pitch, bool is_vram)
+{
+ if (has_service_copy_support(gt))
+ emit_clear_link_copy(gt, bb, src_ofs, size, pitch);
+ else
+ emit_clear_main_copy(gt, bb, src_ofs, size, pitch,
+ is_vram);
+}
+
+/**
+ * xe_migrate_clear() - Copy content of TTM resources.
+ * @m: The migration context.
+ * @bo: The buffer object @dst is currently bound to.
+ * @dst: The dst TTM resource to be cleared.
+ *
+ * Clear the contents of @dst to zero. On flat CCS devices,
+ * the CCS metadata is cleared to zero as well on VRAM destinations.
+ * TODO: Eliminate the @bo argument.
+ *
+ * Return: Pointer to a dma_fence representing the last clear batch, or
+ * an error pointer on failure. If there is a failure, any clear operation
+ * started by the function call has been synced.
+ */
+struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
+ struct xe_bo *bo,
+ struct ttm_resource *dst)
+{
+ bool clear_vram = mem_type_is_vram(dst->mem_type);
+ struct xe_gt *gt = m->tile->primary_gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ bool clear_system_ccs = (xe_bo_needs_ccs_pages(bo) && !IS_DGFX(xe)) ? true : false;
+ struct dma_fence *fence = NULL;
+ u64 size = bo->size;
+ struct xe_res_cursor src_it;
+ struct ttm_resource *src = dst;
+ int err;
+ int pass = 0;
+
+ if (!clear_vram)
+ xe_res_first_sg(xe_bo_sg(bo), 0, bo->size, &src_it);
+ else
+ xe_res_first(src, 0, bo->size, &src_it);
+
+ while (size) {
+ u64 clear_L0_ofs;
+ u32 clear_L0_pt;
+ u32 flush_flags = 0;
+ u64 clear_L0;
+ struct xe_sched_job *job;
+ struct xe_bb *bb;
+ u32 batch_size, update_idx;
+
+ bool usm = xe->info.has_usm;
+ u32 avail_pts = max_mem_transfer_per_pass(xe) / LEVEL0_PAGE_TABLE_ENCODE_SIZE;
+
+ clear_L0 = xe_migrate_res_sizes(xe, &src_it);
+
+ drm_dbg(&xe->drm, "Pass %u, size: %llu\n", pass++, clear_L0);
+
+ /* Calculate final sizes and batch size.. */
+ batch_size = 2 +
+ pte_update_size(m, clear_vram, src, &src_it,
+ &clear_L0, &clear_L0_ofs, &clear_L0_pt,
+ clear_system_ccs ? 0 : emit_clear_cmd_len(gt), 0,
+ avail_pts);
+
+ if (xe_device_has_flat_ccs(xe))
+ batch_size += EMIT_COPY_CCS_DW;
+
+ /* Clear commands */
+
+ if (WARN_ON_ONCE(!clear_L0))
+ break;
+
+ bb = xe_bb_new(gt, batch_size, usm);
+ if (IS_ERR(bb)) {
+ err = PTR_ERR(bb);
+ goto err_sync;
+ }
+
+ size -= clear_L0;
+ /* Preemption is enabled again by the ring ops. */
+ if (!clear_vram) {
+ emit_pte(m, bb, clear_L0_pt, clear_vram, true, &src_it, clear_L0,
+ bo);
+ } else {
+ xe_res_next(&src_it, clear_L0);
+ }
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+ update_idx = bb->len;
+
+ if (!clear_system_ccs)
+ emit_clear(gt, bb, clear_L0_ofs, clear_L0, XE_PAGE_SIZE, clear_vram);
+
+ if (xe_device_has_flat_ccs(xe)) {
+ emit_copy_ccs(gt, bb, clear_L0_ofs, true,
+ m->cleared_mem_ofs, false, clear_L0);
+ flush_flags = MI_FLUSH_DW_CCS;
+ }
+
+ mutex_lock(&m->job_mutex);
+ job = xe_bb_create_migration_job(m->q, bb,
+ xe_migrate_batch_base(m, usm),
+ update_idx);
+ if (IS_ERR(job)) {
+ err = PTR_ERR(job);
+ goto err;
+ }
+
+ xe_sched_job_add_migrate_flush(job, flush_flags);
+ if (!fence) {
+ /*
+ * There can't be anything userspace related at this
+ * point, so we just need to respect any potential move
+ * fences, which are always tracked as
+ * DMA_RESV_USAGE_KERNEL.
+ */
+ err = job_add_deps(job, bo->ttm.base.resv,
+ DMA_RESV_USAGE_KERNEL);
+ if (err)
+ goto err_job;
+ }
+
+ xe_sched_job_arm(job);
+ dma_fence_put(fence);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ dma_fence_put(m->fence);
+ m->fence = dma_fence_get(fence);
+
+ mutex_unlock(&m->job_mutex);
+
+ xe_bb_free(bb, fence);
+ continue;
+
+err_job:
+ xe_sched_job_put(job);
+err:
+ mutex_unlock(&m->job_mutex);
+ xe_bb_free(bb, NULL);
+err_sync:
+ /* Sync partial copies if any. FIXME: job_mutex? */
+ if (fence) {
+ dma_fence_wait(m->fence, false);
+ dma_fence_put(fence);
+ }
+
+ return ERR_PTR(err);
+ }
+
+ if (clear_system_ccs)
+ bo->ccs_cleared = true;
+
+ return fence;
+}
+
+static void write_pgtable(struct xe_tile *tile, struct xe_bb *bb, u64 ppgtt_ofs,
+ const struct xe_vm_pgtable_update *update,
+ struct xe_migrate_pt_update *pt_update)
+{
+ const struct xe_migrate_pt_update_ops *ops = pt_update->ops;
+ u32 chunk;
+ u32 ofs = update->ofs, size = update->qwords;
+
+ /*
+ * If we have 512 entries (max), we would populate it ourselves,
+ * and update the PDE above it to the new pointer.
+ * The only time this can only happen if we have to update the top
+ * PDE. This requires a BO that is almost vm->size big.
+ *
+ * This shouldn't be possible in practice.. might change when 16K
+ * pages are used. Hence the assert.
+ */
+ xe_tile_assert(tile, update->qwords <= 0x1ff);
+ if (!ppgtt_ofs)
+ ppgtt_ofs = xe_migrate_vram_ofs(tile_to_xe(tile),
+ xe_bo_addr(update->pt_bo, 0,
+ XE_PAGE_SIZE));
+
+ do {
+ u64 addr = ppgtt_ofs + ofs * 8;
+
+ chunk = min(update->qwords, 0x1ffU);
+
+ /* Ensure populatefn can do memset64 by aligning bb->cs */
+ if (!(bb->len & 1))
+ bb->cs[bb->len++] = MI_NOOP;
+
+ bb->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(chunk);
+ bb->cs[bb->len++] = lower_32_bits(addr);
+ bb->cs[bb->len++] = upper_32_bits(addr);
+ ops->populate(pt_update, tile, NULL, bb->cs + bb->len, ofs, chunk,
+ update);
+
+ bb->len += chunk * 2;
+ ofs += chunk;
+ size -= chunk;
+ } while (size);
+}
+
+struct xe_vm *xe_migrate_get_vm(struct xe_migrate *m)
+{
+ return xe_vm_get(m->q->vm);
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+struct migrate_test_params {
+ struct xe_test_priv base;
+ bool force_gpu;
+};
+
+#define to_migrate_test_params(_priv) \
+ container_of(_priv, struct migrate_test_params, base)
+#endif
+
+static struct dma_fence *
+xe_migrate_update_pgtables_cpu(struct xe_migrate *m,
+ struct xe_vm *vm, struct xe_bo *bo,
+ const struct xe_vm_pgtable_update *updates,
+ u32 num_updates, bool wait_vm,
+ struct xe_migrate_pt_update *pt_update)
+{
+ XE_TEST_DECLARE(struct migrate_test_params *test =
+ to_migrate_test_params
+ (xe_cur_kunit_priv(XE_TEST_LIVE_MIGRATE));)
+ const struct xe_migrate_pt_update_ops *ops = pt_update->ops;
+ struct dma_fence *fence;
+ int err;
+ u32 i;
+
+ if (XE_TEST_ONLY(test && test->force_gpu))
+ return ERR_PTR(-ETIME);
+
+ if (bo && !dma_resv_test_signaled(bo->ttm.base.resv,
+ DMA_RESV_USAGE_KERNEL))
+ return ERR_PTR(-ETIME);
+
+ if (wait_vm && !dma_resv_test_signaled(xe_vm_resv(vm),
+ DMA_RESV_USAGE_BOOKKEEP))
+ return ERR_PTR(-ETIME);
+
+ if (ops->pre_commit) {
+ pt_update->job = NULL;
+ err = ops->pre_commit(pt_update);
+ if (err)
+ return ERR_PTR(err);
+ }
+ for (i = 0; i < num_updates; i++) {
+ const struct xe_vm_pgtable_update *update = &updates[i];
+
+ ops->populate(pt_update, m->tile, &update->pt_bo->vmap, NULL,
+ update->ofs, update->qwords, update);
+ }
+
+ if (vm) {
+ trace_xe_vm_cpu_bind(vm);
+ xe_device_wmb(vm->xe);
+ }
+
+ fence = dma_fence_get_stub();
+
+ return fence;
+}
+
+static bool no_in_syncs(struct xe_vm *vm, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs)
+{
+ struct dma_fence *fence;
+ int i;
+
+ for (i = 0; i < num_syncs; i++) {
+ fence = syncs[i].fence;
+
+ if (fence && !test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+ &fence->flags))
+ return false;
+ }
+ if (q) {
+ fence = xe_exec_queue_last_fence_get(q, vm);
+ if (!test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &fence->flags))
+ return false;
+ }
+
+ return true;
+}
+
+/**
+ * xe_migrate_update_pgtables() - Pipelined page-table update
+ * @m: The migrate context.
+ * @vm: The vm we'll be updating.
+ * @bo: The bo whose dma-resv we will await before updating, or NULL if userptr.
+ * @q: The exec queue to be used for the update or NULL if the default
+ * migration engine is to be used.
+ * @updates: An array of update descriptors.
+ * @num_updates: Number of descriptors in @updates.
+ * @syncs: Array of xe_sync_entry to await before updating. Note that waits
+ * will block the engine timeline.
+ * @num_syncs: Number of entries in @syncs.
+ * @pt_update: Pointer to a struct xe_migrate_pt_update, which contains
+ * pointers to callback functions and, if subclassed, private arguments to
+ * those.
+ *
+ * Perform a pipelined page-table update. The update descriptors are typically
+ * built under the same lock critical section as a call to this function. If
+ * using the default engine for the updates, they will be performed in the
+ * order they grab the job_mutex. If different engines are used, external
+ * synchronization is needed for overlapping updates to maintain page-table
+ * consistency. Note that the meaing of "overlapping" is that the updates
+ * touch the same page-table, which might be a higher-level page-directory.
+ * If no pipelining is needed, then updates may be performed by the cpu.
+ *
+ * Return: A dma_fence that, when signaled, indicates the update completion.
+ */
+struct dma_fence *
+xe_migrate_update_pgtables(struct xe_migrate *m,
+ struct xe_vm *vm,
+ struct xe_bo *bo,
+ struct xe_exec_queue *q,
+ const struct xe_vm_pgtable_update *updates,
+ u32 num_updates,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ struct xe_migrate_pt_update *pt_update)
+{
+ const struct xe_migrate_pt_update_ops *ops = pt_update->ops;
+ struct xe_tile *tile = m->tile;
+ struct xe_gt *gt = tile->primary_gt;
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_sched_job *job;
+ struct dma_fence *fence;
+ struct drm_suballoc *sa_bo = NULL;
+ struct xe_vma *vma = pt_update->vma;
+ struct xe_bb *bb;
+ u32 i, batch_size, ppgtt_ofs, update_idx, page_ofs = 0;
+ u64 addr;
+ int err = 0;
+ bool usm = !q && xe->info.has_usm;
+ bool first_munmap_rebind = vma &&
+ vma->gpuva.flags & XE_VMA_FIRST_REBIND;
+ struct xe_exec_queue *q_override = !q ? m->q : q;
+ u16 pat_index = xe->pat.idx[XE_CACHE_WB];
+
+ /* Use the CPU if no in syncs and engine is idle */
+ if (no_in_syncs(vm, q, syncs, num_syncs) && xe_exec_queue_is_idle(q_override)) {
+ fence = xe_migrate_update_pgtables_cpu(m, vm, bo, updates,
+ num_updates,
+ first_munmap_rebind,
+ pt_update);
+ if (!IS_ERR(fence) || fence == ERR_PTR(-EAGAIN))
+ return fence;
+ }
+
+ /* fixed + PTE entries */
+ if (IS_DGFX(xe))
+ batch_size = 2;
+ else
+ batch_size = 6 + num_updates * 2;
+
+ for (i = 0; i < num_updates; i++) {
+ u32 num_cmds = DIV_ROUND_UP(updates[i].qwords, 0x1ff);
+
+ /* align noop + MI_STORE_DATA_IMM cmd prefix */
+ batch_size += 4 * num_cmds + updates[i].qwords * 2;
+ }
+
+ /*
+ * XXX: Create temp bo to copy from, if batch_size becomes too big?
+ *
+ * Worst case: Sum(2 * (each lower level page size) + (top level page size))
+ * Should be reasonably bound..
+ */
+ xe_tile_assert(tile, batch_size < SZ_128K);
+
+ bb = xe_bb_new(gt, batch_size, !q && xe->info.has_usm);
+ if (IS_ERR(bb))
+ return ERR_CAST(bb);
+
+ /* For sysmem PTE's, need to map them in our hole.. */
+ if (!IS_DGFX(xe)) {
+ ppgtt_ofs = NUM_KERNEL_PDE - 1;
+ if (q) {
+ xe_tile_assert(tile, num_updates <= NUM_VMUSA_WRITES_PER_UNIT);
+
+ sa_bo = drm_suballoc_new(&m->vm_update_sa, 1,
+ GFP_KERNEL, true, 0);
+ if (IS_ERR(sa_bo)) {
+ err = PTR_ERR(sa_bo);
+ goto err;
+ }
+
+ ppgtt_ofs = NUM_KERNEL_PDE +
+ (drm_suballoc_soffset(sa_bo) /
+ NUM_VMUSA_UNIT_PER_PAGE);
+ page_ofs = (drm_suballoc_soffset(sa_bo) %
+ NUM_VMUSA_UNIT_PER_PAGE) *
+ VM_SA_UPDATE_UNIT_SIZE;
+ }
+
+ /* Map our PT's to gtt */
+ bb->cs[bb->len++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(num_updates);
+ bb->cs[bb->len++] = ppgtt_ofs * XE_PAGE_SIZE + page_ofs;
+ bb->cs[bb->len++] = 0; /* upper_32_bits */
+
+ for (i = 0; i < num_updates; i++) {
+ struct xe_bo *pt_bo = updates[i].pt_bo;
+
+ xe_tile_assert(tile, pt_bo->size == SZ_4K);
+
+ addr = vm->pt_ops->pte_encode_bo(pt_bo, 0, pat_index, 0);
+ bb->cs[bb->len++] = lower_32_bits(addr);
+ bb->cs[bb->len++] = upper_32_bits(addr);
+ }
+
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+ update_idx = bb->len;
+
+ addr = xe_migrate_vm_addr(ppgtt_ofs, 0) +
+ (page_ofs / sizeof(u64)) * XE_PAGE_SIZE;
+ for (i = 0; i < num_updates; i++)
+ write_pgtable(tile, bb, addr + i * XE_PAGE_SIZE,
+ &updates[i], pt_update);
+ } else {
+ /* phys pages, no preamble required */
+ bb->cs[bb->len++] = MI_BATCH_BUFFER_END;
+ update_idx = bb->len;
+
+ for (i = 0; i < num_updates; i++)
+ write_pgtable(tile, bb, 0, &updates[i], pt_update);
+ }
+
+ if (!q)
+ mutex_lock(&m->job_mutex);
+
+ job = xe_bb_create_migration_job(q ?: m->q, bb,
+ xe_migrate_batch_base(m, usm),
+ update_idx);
+ if (IS_ERR(job)) {
+ err = PTR_ERR(job);
+ goto err_bb;
+ }
+
+ /* Wait on BO move */
+ if (bo) {
+ err = job_add_deps(job, bo->ttm.base.resv,
+ DMA_RESV_USAGE_KERNEL);
+ if (err)
+ goto err_job;
+ }
+
+ /*
+ * Munmap style VM unbind, need to wait for all jobs to be complete /
+ * trigger preempts before moving forward
+ */
+ if (first_munmap_rebind) {
+ err = job_add_deps(job, xe_vm_resv(vm),
+ DMA_RESV_USAGE_BOOKKEEP);
+ if (err)
+ goto err_job;
+ }
+
+ err = xe_sched_job_last_fence_add_dep(job, vm);
+ for (i = 0; !err && i < num_syncs; i++)
+ err = xe_sync_entry_add_deps(&syncs[i], job);
+
+ if (err)
+ goto err_job;
+
+ if (ops->pre_commit) {
+ pt_update->job = job;
+ err = ops->pre_commit(pt_update);
+ if (err)
+ goto err_job;
+ }
+ xe_sched_job_arm(job);
+ fence = dma_fence_get(&job->drm.s_fence->finished);
+ xe_sched_job_push(job);
+
+ if (!q)
+ mutex_unlock(&m->job_mutex);
+
+ xe_bb_free(bb, fence);
+ drm_suballoc_free(sa_bo, fence);
+
+ return fence;
+
+err_job:
+ xe_sched_job_put(job);
+err_bb:
+ if (!q)
+ mutex_unlock(&m->job_mutex);
+ xe_bb_free(bb, NULL);
+err:
+ drm_suballoc_free(sa_bo, NULL);
+ return ERR_PTR(err);
+}
+
+/**
+ * xe_migrate_wait() - Complete all operations using the xe_migrate context
+ * @m: Migrate context to wait for.
+ *
+ * Waits until the GPU no longer uses the migrate context's default engine
+ * or its page-table objects. FIXME: What about separate page-table update
+ * engines?
+ */
+void xe_migrate_wait(struct xe_migrate *m)
+{
+ if (m->fence)
+ dma_fence_wait(m->fence, false);
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_migrate.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_migrate.h b/drivers/gpu/drm/xe/xe_migrate.h
new file mode 100644
index 000000000000..951f19318ea4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_migrate.h
@@ -0,0 +1,110 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2020 Intel Corporation
+ */
+
+#ifndef _XE_MIGRATE_
+#define _XE_MIGRATE_
+
+#include <drm/drm_mm.h>
+
+struct dma_fence;
+struct iosys_map;
+struct ttm_resource;
+
+struct xe_bo;
+struct xe_gt;
+struct xe_exec_queue;
+struct xe_migrate;
+struct xe_migrate_pt_update;
+struct xe_sync_entry;
+struct xe_pt;
+struct xe_tile;
+struct xe_vm;
+struct xe_vm_pgtable_update;
+struct xe_vma;
+
+/**
+ * struct xe_migrate_pt_update_ops - Callbacks for the
+ * xe_migrate_update_pgtables() function.
+ */
+struct xe_migrate_pt_update_ops {
+ /**
+ * @populate: Populate a command buffer or page-table with ptes.
+ * @pt_update: Embeddable callback argument.
+ * @tile: The tile for the current operation.
+ * @map: struct iosys_map into the memory to be populated.
+ * @pos: If @map is NULL, map into the memory to be populated.
+ * @ofs: qword offset into @map, unused if @map is NULL.
+ * @num_qwords: Number of qwords to write.
+ * @update: Information about the PTEs to be inserted.
+ *
+ * This interface is intended to be used as a callback into the
+ * page-table system to populate command buffers or shared
+ * page-tables with PTEs.
+ */
+ void (*populate)(struct xe_migrate_pt_update *pt_update,
+ struct xe_tile *tile, struct iosys_map *map,
+ void *pos, u32 ofs, u32 num_qwords,
+ const struct xe_vm_pgtable_update *update);
+
+ /**
+ * @pre_commit: Callback to be called just before arming the
+ * sched_job.
+ * @pt_update: Pointer to embeddable callback argument.
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+ int (*pre_commit)(struct xe_migrate_pt_update *pt_update);
+};
+
+/**
+ * struct xe_migrate_pt_update - Argument to the
+ * struct xe_migrate_pt_update_ops callbacks.
+ *
+ * Intended to be subclassed to support additional arguments if necessary.
+ */
+struct xe_migrate_pt_update {
+ /** @ops: Pointer to the struct xe_migrate_pt_update_ops callbacks */
+ const struct xe_migrate_pt_update_ops *ops;
+ /** @vma: The vma we're updating the pagetable for. */
+ struct xe_vma *vma;
+ /** @job: The job if a GPU page-table update. NULL otherwise */
+ struct xe_sched_job *job;
+ /** @start: Start of update for the range fence */
+ u64 start;
+ /** @last: Last of update for the range fence */
+ u64 last;
+ /** @tile_id: Tile ID of the update */
+ u8 tile_id;
+};
+
+struct xe_migrate *xe_migrate_init(struct xe_tile *tile);
+
+struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
+ struct xe_bo *src_bo,
+ struct xe_bo *dst_bo,
+ struct ttm_resource *src,
+ struct ttm_resource *dst,
+ bool copy_only_ccs);
+
+struct dma_fence *xe_migrate_clear(struct xe_migrate *m,
+ struct xe_bo *bo,
+ struct ttm_resource *dst);
+
+struct xe_vm *xe_migrate_get_vm(struct xe_migrate *m);
+
+struct dma_fence *
+xe_migrate_update_pgtables(struct xe_migrate *m,
+ struct xe_vm *vm,
+ struct xe_bo *bo,
+ struct xe_exec_queue *q,
+ const struct xe_vm_pgtable_update *updates,
+ u32 num_updates,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ struct xe_migrate_pt_update *pt_update);
+
+void xe_migrate_wait(struct xe_migrate *m);
+
+struct xe_exec_queue *xe_tile_migrate_engine(struct xe_tile *tile);
+#endif
diff --git a/drivers/gpu/drm/xe/xe_migrate_doc.h b/drivers/gpu/drm/xe/xe_migrate_doc.h
new file mode 100644
index 000000000000..63c7d67b5b62
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_migrate_doc.h
@@ -0,0 +1,88 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_MIGRATE_DOC_H_
+#define _XE_MIGRATE_DOC_H_
+
+/**
+ * DOC: Migrate Layer
+ *
+ * The XE migrate layer is used generate jobs which can copy memory (eviction),
+ * clear memory, or program tables (binds). This layer exists in every GT, has
+ * a migrate engine, and uses a special VM for all generated jobs.
+ *
+ * Special VM details
+ * ==================
+ *
+ * The special VM is configured with a page structure where we can dynamically
+ * map BOs which need to be copied and cleared, dynamically map other VM's page
+ * table BOs for updates, and identity map the entire device's VRAM with 1 GB
+ * pages.
+ *
+ * Currently the page structure consists of 32 physical pages with 16 being
+ * reserved for BO mapping during copies and clear, 1 reserved for kernel binds,
+ * several pages are needed to setup the identity mappings (exact number based
+ * on how many bits of address space the device has), and the rest are reserved
+ * user bind operations.
+ *
+ * TODO: Diagram of layout
+ *
+ * Bind jobs
+ * =========
+ *
+ * A bind job consist of two batches and runs either on the migrate engine
+ * (kernel binds) or the bind engine passed in (user binds). In both cases the
+ * VM of the engine is the migrate VM.
+ *
+ * The first batch is used to update the migration VM page structure to point to
+ * the bind VM page table BOs which need to be updated. A physical page is
+ * required for this. If it is a user bind, the page is allocated from pool of
+ * pages reserved user bind operations with drm_suballoc managing this pool. If
+ * it is a kernel bind, the page reserved for kernel binds is used.
+ *
+ * The first batch is only required for devices without VRAM as when the device
+ * has VRAM the bind VM page table BOs are in VRAM and the identity mapping can
+ * be used.
+ *
+ * The second batch is used to program page table updated in the bind VM. Why
+ * not just one batch? Well the TLBs need to be invalidated between these two
+ * batches and that only can be done from the ring.
+ *
+ * When the bind job complete, the page allocated is returned the pool of pages
+ * reserved for user bind operations if a user bind. No need do this for kernel
+ * binds as the reserved kernel page is serially used by each job.
+ *
+ * Copy / clear jobs
+ * =================
+ *
+ * A copy or clear job consist of two batches and runs on the migrate engine.
+ *
+ * Like binds, the first batch is used update the migration VM page structure.
+ * In copy jobs, we need to map the source and destination of the BO into page
+ * the structure. In clear jobs, we just need to add 1 mapping of BO into the
+ * page structure. We use the 16 reserved pages in migration VM for mappings,
+ * this gives us a maximum copy size of 16 MB and maximum clear size of 32 MB.
+ *
+ * The second batch is used do either do the copy or clear. Again similar to
+ * binds, two batches are required as the TLBs need to be invalidated from the
+ * ring between the batches.
+ *
+ * More than one job will be generated if the BO is larger than maximum copy /
+ * clear size.
+ *
+ * Future work
+ * ===========
+ *
+ * Update copy and clear code to use identity mapped VRAM.
+ *
+ * Can we rework the use of the pages async binds to use all the entries in each
+ * page?
+ *
+ * Using large pages for sysmem mappings.
+ *
+ * Is it possible to identity map the sysmem? We should explore this.
+ */
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_mmio.c b/drivers/gpu/drm/xe/xe_mmio.c
new file mode 100644
index 000000000000..f660cfb79f50
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_mmio.c
@@ -0,0 +1,524 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021-2023 Intel Corporation
+ */
+
+#include <linux/minmax.h>
+
+#include "xe_mmio.h"
+
+#include <drm/drm_managed.h>
+#include <drm/xe_drm.h>
+
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_macros.h"
+#include "xe_module.h"
+#include "xe_tile.h"
+
+#define XEHP_MTCFG_ADDR XE_REG(0x101800)
+#define TILE_COUNT REG_GENMASK(15, 8)
+
+#define BAR_SIZE_SHIFT 20
+
+static void
+_resize_bar(struct xe_device *xe, int resno, resource_size_t size)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ int bar_size = pci_rebar_bytes_to_size(size);
+ int ret;
+
+ if (pci_resource_len(pdev, resno))
+ pci_release_resource(pdev, resno);
+
+ ret = pci_resize_resource(pdev, resno, bar_size);
+ if (ret) {
+ drm_info(&xe->drm, "Failed to resize BAR%d to %dM (%pe). Consider enabling 'Resizable BAR' support in your BIOS\n",
+ resno, 1 << bar_size, ERR_PTR(ret));
+ return;
+ }
+
+ drm_info(&xe->drm, "BAR%d resized to %dM\n", resno, 1 << bar_size);
+}
+
+/*
+ * if force_vram_bar_size is set, attempt to set to the requested size
+ * else set to maximum possible size
+ */
+static void xe_resize_vram_bar(struct xe_device *xe)
+{
+ u64 force_vram_bar_size = xe_modparam.force_vram_bar_size;
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ struct pci_bus *root = pdev->bus;
+ resource_size_t current_size;
+ resource_size_t rebar_size;
+ struct resource *root_res;
+ u32 bar_size_mask;
+ u32 pci_cmd;
+ int i;
+
+ /* gather some relevant info */
+ current_size = pci_resource_len(pdev, LMEM_BAR);
+ bar_size_mask = pci_rebar_get_possible_sizes(pdev, LMEM_BAR);
+
+ if (!bar_size_mask)
+ return;
+
+ /* set to a specific size? */
+ if (force_vram_bar_size) {
+ u32 bar_size_bit;
+
+ rebar_size = force_vram_bar_size * (resource_size_t)SZ_1M;
+
+ bar_size_bit = bar_size_mask & BIT(pci_rebar_bytes_to_size(rebar_size));
+
+ if (!bar_size_bit) {
+ drm_info(&xe->drm,
+ "Requested size: %lluMiB is not supported by rebar sizes: 0x%x. Leaving default: %lluMiB\n",
+ (u64)rebar_size >> 20, bar_size_mask, (u64)current_size >> 20);
+ return;
+ }
+
+ rebar_size = 1ULL << (__fls(bar_size_bit) + BAR_SIZE_SHIFT);
+
+ if (rebar_size == current_size)
+ return;
+ } else {
+ rebar_size = 1ULL << (__fls(bar_size_mask) + BAR_SIZE_SHIFT);
+
+ /* only resize if larger than current */
+ if (rebar_size <= current_size)
+ return;
+ }
+
+ drm_info(&xe->drm, "Attempting to resize bar from %lluMiB -> %lluMiB\n",
+ (u64)current_size >> 20, (u64)rebar_size >> 20);
+
+ while (root->parent)
+ root = root->parent;
+
+ pci_bus_for_each_resource(root, root_res, i) {
+ if (root_res && root_res->flags & (IORESOURCE_MEM | IORESOURCE_MEM_64) &&
+ root_res->start > 0x100000000ull)
+ break;
+ }
+
+ if (!root_res) {
+ drm_info(&xe->drm, "Can't resize VRAM BAR - platform support is missing. Consider enabling 'Resizable BAR' support in your BIOS\n");
+ return;
+ }
+
+ pci_read_config_dword(pdev, PCI_COMMAND, &pci_cmd);
+ pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd & ~PCI_COMMAND_MEMORY);
+
+ _resize_bar(xe, LMEM_BAR, rebar_size);
+
+ pci_assign_unassigned_bus_resources(pdev->bus);
+ pci_write_config_dword(pdev, PCI_COMMAND, pci_cmd);
+}
+
+static bool xe_pci_resource_valid(struct pci_dev *pdev, int bar)
+{
+ if (!pci_resource_flags(pdev, bar))
+ return false;
+
+ if (pci_resource_flags(pdev, bar) & IORESOURCE_UNSET)
+ return false;
+
+ if (!pci_resource_len(pdev, bar))
+ return false;
+
+ return true;
+}
+
+static int xe_determine_lmem_bar_size(struct xe_device *xe)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+
+ if (!xe_pci_resource_valid(pdev, LMEM_BAR)) {
+ drm_err(&xe->drm, "pci resource is not valid\n");
+ return -ENXIO;
+ }
+
+ xe_resize_vram_bar(xe);
+
+ xe->mem.vram.io_start = pci_resource_start(pdev, LMEM_BAR);
+ xe->mem.vram.io_size = pci_resource_len(pdev, LMEM_BAR);
+ if (!xe->mem.vram.io_size)
+ return -EIO;
+
+ /* XXX: Need to change when xe link code is ready */
+ xe->mem.vram.dpa_base = 0;
+
+ /* set up a map to the total memory area. */
+ xe->mem.vram.mapping = ioremap_wc(xe->mem.vram.io_start, xe->mem.vram.io_size);
+
+ return 0;
+}
+
+/**
+ * xe_mmio_tile_vram_size() - Collect vram size and offset information
+ * @tile: tile to get info for
+ * @vram_size: available vram (size - device reserved portions)
+ * @tile_size: actual vram size
+ * @tile_offset: physical start point in the vram address space
+ *
+ * There are 4 places for size information:
+ * - io size (from pci_resource_len of LMEM bar) (only used for small bar and DG1)
+ * - TILEx size (actual vram size)
+ * - GSMBASE offset (TILEx - "stolen")
+ * - CSSBASE offset (TILEx - CSS space necessary)
+ *
+ * CSSBASE is always a lower/smaller offset then GSMBASE.
+ *
+ * The actual available size of memory is to the CCS or GSM base.
+ * NOTE: multi-tile bases will include the tile offset.
+ *
+ */
+static int xe_mmio_tile_vram_size(struct xe_tile *tile, u64 *vram_size,
+ u64 *tile_size, u64 *tile_offset)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_gt *gt = tile->primary_gt;
+ u64 offset;
+ int err;
+ u32 reg;
+
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ return err;
+
+ /* actual size */
+ if (unlikely(xe->info.platform == XE_DG1)) {
+ *tile_size = pci_resource_len(to_pci_dev(xe->drm.dev), LMEM_BAR);
+ *tile_offset = 0;
+ } else {
+ reg = xe_gt_mcr_unicast_read_any(gt, XEHP_TILE_ADDR_RANGE(gt->info.id));
+ *tile_size = (u64)REG_FIELD_GET(GENMASK(14, 8), reg) * SZ_1G;
+ *tile_offset = (u64)REG_FIELD_GET(GENMASK(7, 1), reg) * SZ_1G;
+ }
+
+ /* minus device usage */
+ if (xe->info.has_flat_ccs) {
+ reg = xe_gt_mcr_unicast_read_any(gt, XEHP_FLAT_CCS_BASE_ADDR);
+ offset = (u64)REG_FIELD_GET(GENMASK(31, 8), reg) * SZ_64K;
+ } else {
+ offset = xe_mmio_read64_2x32(gt, GSMBASE);
+ }
+
+ /* remove the tile offset so we have just the available size */
+ *vram_size = offset - *tile_offset;
+
+ return xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+}
+
+int xe_mmio_probe_vram(struct xe_device *xe)
+{
+ struct xe_tile *tile;
+ resource_size_t io_size;
+ u64 available_size = 0;
+ u64 total_size = 0;
+ u64 tile_offset;
+ u64 tile_size;
+ u64 vram_size;
+ int err;
+ u8 id;
+
+ if (!IS_DGFX(xe))
+ return 0;
+
+ /* Get the size of the root tile's vram for later accessibility comparison */
+ tile = xe_device_get_root_tile(xe);
+ err = xe_mmio_tile_vram_size(tile, &vram_size, &tile_size, &tile_offset);
+ if (err)
+ return err;
+
+ err = xe_determine_lmem_bar_size(xe);
+ if (err)
+ return err;
+
+ drm_info(&xe->drm, "VISIBLE VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
+ &xe->mem.vram.io_size);
+
+ io_size = xe->mem.vram.io_size;
+
+ /* tile specific ranges */
+ for_each_tile(tile, xe, id) {
+ err = xe_mmio_tile_vram_size(tile, &vram_size, &tile_size, &tile_offset);
+ if (err)
+ return err;
+
+ tile->mem.vram.actual_physical_size = tile_size;
+ tile->mem.vram.io_start = xe->mem.vram.io_start + tile_offset;
+ tile->mem.vram.io_size = min_t(u64, vram_size, io_size);
+
+ if (!tile->mem.vram.io_size) {
+ drm_err(&xe->drm, "Tile without any CPU visible VRAM. Aborting.\n");
+ return -ENODEV;
+ }
+
+ tile->mem.vram.dpa_base = xe->mem.vram.dpa_base + tile_offset;
+ tile->mem.vram.usable_size = vram_size;
+ tile->mem.vram.mapping = xe->mem.vram.mapping + tile_offset;
+
+ if (tile->mem.vram.io_size < tile->mem.vram.usable_size)
+ drm_info(&xe->drm, "Small BAR device\n");
+ drm_info(&xe->drm, "VRAM[%u, %u]: Actual physical size %pa, usable size exclude stolen %pa, CPU accessible size %pa\n", id,
+ tile->id, &tile->mem.vram.actual_physical_size, &tile->mem.vram.usable_size, &tile->mem.vram.io_size);
+ drm_info(&xe->drm, "VRAM[%u, %u]: DPA range: [%pa-%llx], io range: [%pa-%llx]\n", id, tile->id,
+ &tile->mem.vram.dpa_base, tile->mem.vram.dpa_base + tile->mem.vram.actual_physical_size,
+ &tile->mem.vram.io_start, tile->mem.vram.io_start + tile->mem.vram.io_size);
+
+ /* calculate total size using tile size to get the correct HW sizing */
+ total_size += tile_size;
+ available_size += vram_size;
+
+ if (total_size > xe->mem.vram.io_size) {
+ drm_info(&xe->drm, "VRAM: %pa is larger than resource %pa\n",
+ &total_size, &xe->mem.vram.io_size);
+ }
+
+ io_size -= min_t(u64, tile_size, io_size);
+ }
+
+ xe->mem.vram.actual_physical_size = total_size;
+
+ drm_info(&xe->drm, "Total VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
+ &xe->mem.vram.actual_physical_size);
+ drm_info(&xe->drm, "Available VRAM: %pa, %pa\n", &xe->mem.vram.io_start,
+ &available_size);
+
+ return 0;
+}
+
+void xe_mmio_probe_tiles(struct xe_device *xe)
+{
+ size_t tile_mmio_size = SZ_16M, tile_mmio_ext_size = xe->info.tile_mmio_ext_size;
+ u8 id, tile_count = xe->info.tile_count;
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+ struct xe_tile *tile;
+ void *regs;
+ u32 mtcfg;
+
+ if (tile_count == 1)
+ goto add_mmio_ext;
+
+ if (!xe->info.skip_mtcfg) {
+ mtcfg = xe_mmio_read64_2x32(gt, XEHP_MTCFG_ADDR);
+ tile_count = REG_FIELD_GET(TILE_COUNT, mtcfg) + 1;
+ if (tile_count < xe->info.tile_count) {
+ drm_info(&xe->drm, "tile_count: %d, reduced_tile_count %d\n",
+ xe->info.tile_count, tile_count);
+ xe->info.tile_count = tile_count;
+
+ /*
+ * FIXME: Needs some work for standalone media, but should be impossible
+ * with multi-tile for now.
+ */
+ xe->info.gt_count = xe->info.tile_count;
+ }
+ }
+
+ regs = xe->mmio.regs;
+ for_each_tile(tile, xe, id) {
+ tile->mmio.size = tile_mmio_size;
+ tile->mmio.regs = regs;
+ regs += tile_mmio_size;
+ }
+
+add_mmio_ext:
+ /*
+ * By design, there's a contiguous multi-tile MMIO space (16MB hard coded per tile).
+ * When supported, there could be an additional contiguous multi-tile MMIO extension
+ * space ON TOP of it, and hence the necessity for distinguished MMIO spaces.
+ */
+ if (xe->info.has_mmio_ext) {
+ regs = xe->mmio.regs + tile_mmio_size * tile_count;
+
+ for_each_tile(tile, xe, id) {
+ tile->mmio_ext.size = tile_mmio_ext_size;
+ tile->mmio_ext.regs = regs;
+
+ regs += tile_mmio_ext_size;
+ }
+ }
+}
+
+static void mmio_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = arg;
+
+ pci_iounmap(to_pci_dev(xe->drm.dev), xe->mmio.regs);
+ if (xe->mem.vram.mapping)
+ iounmap(xe->mem.vram.mapping);
+}
+
+static int xe_verify_lmem_ready(struct xe_device *xe)
+{
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+
+ /*
+ * The boot firmware initializes local memory and assesses its health.
+ * If memory training fails, the punit will have been instructed to
+ * keep the GT powered down; we won't be able to communicate with it
+ * and we should not continue with driver initialization.
+ */
+ if (IS_DGFX(xe) && !(xe_mmio_read32(gt, GU_CNTL) & LMEM_INIT)) {
+ drm_err(&xe->drm, "VRAM not initialized by firmware\n");
+ return -ENODEV;
+ }
+
+ return 0;
+}
+
+int xe_mmio_init(struct xe_device *xe)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ const int mmio_bar = 0;
+
+ /*
+ * Map the entire BAR.
+ * The first 16MB of the BAR, belong to the root tile, and include:
+ * registers (0-4MB), reserved space (4MB-8MB) and GGTT (8MB-16MB).
+ */
+ xe->mmio.size = pci_resource_len(pdev, mmio_bar);
+ xe->mmio.regs = pci_iomap(pdev, mmio_bar, 0);
+ if (xe->mmio.regs == NULL) {
+ drm_err(&xe->drm, "failed to map registers\n");
+ return -EIO;
+ }
+
+ return drmm_add_action_or_reset(&xe->drm, mmio_fini, xe);
+}
+
+int xe_mmio_root_tile_init(struct xe_device *xe)
+{
+ struct xe_tile *root_tile = xe_device_get_root_tile(xe);
+ int err;
+
+ /* Setup first tile; other tiles (if present) will be setup later. */
+ root_tile->mmio.size = SZ_16M;
+ root_tile->mmio.regs = xe->mmio.regs;
+
+ err = xe_verify_lmem_ready(xe);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+/**
+ * xe_mmio_read64_2x32() - Read a 64-bit register as two 32-bit reads
+ * @gt: MMIO target GT
+ * @reg: register to read value from
+ *
+ * Although Intel GPUs have some 64-bit registers, the hardware officially
+ * only supports GTTMMADR register reads of 32 bits or smaller. Even if
+ * a readq operation may return a reasonable value, that violation of the
+ * spec shouldn't be relied upon and all 64-bit register reads should be
+ * performed as two 32-bit reads of the upper and lower dwords.
+ *
+ * When reading registers that may be changing (such as
+ * counters), a rollover of the lower dword between the two 32-bit reads
+ * can be problematic. This function attempts to ensure the upper dword has
+ * stabilized before returning the 64-bit value.
+ *
+ * Note that because this function may re-read the register multiple times
+ * while waiting for the value to stabilize it should not be used to read
+ * any registers where read operations have side effects.
+ *
+ * Returns the value of the 64-bit register.
+ */
+u64 xe_mmio_read64_2x32(struct xe_gt *gt, struct xe_reg reg)
+{
+ struct xe_reg reg_udw = { .addr = reg.addr + 0x4 };
+ u32 ldw, udw, oldudw, retries;
+
+ if (reg.addr < gt->mmio.adj_limit) {
+ reg.addr += gt->mmio.adj_offset;
+ reg_udw.addr += gt->mmio.adj_offset;
+ }
+
+ oldudw = xe_mmio_read32(gt, reg_udw);
+ for (retries = 5; retries; --retries) {
+ ldw = xe_mmio_read32(gt, reg);
+ udw = xe_mmio_read32(gt, reg_udw);
+
+ if (udw == oldudw)
+ break;
+
+ oldudw = udw;
+ }
+
+ xe_gt_WARN(gt, retries == 0,
+ "64-bit read of %#x did not stabilize\n", reg.addr);
+
+ return (u64)udw << 32 | ldw;
+}
+
+/**
+ * xe_mmio_wait32() - Wait for a register to match the desired masked value
+ * @gt: MMIO target GT
+ * @reg: register to read value from
+ * @mask: mask to be applied to the value read from the register
+ * @val: desired value after applying the mask
+ * @timeout_us: time out after this period of time. Wait logic tries to be
+ * smart, applying an exponential backoff until @timeout_us is reached.
+ * @out_val: if not NULL, points where to store the last unmasked value
+ * @atomic: needs to be true if calling from an atomic context
+ *
+ * This function polls for the desired masked value and returns zero on success
+ * or -ETIMEDOUT if timed out.
+ *
+ * Note that @timeout_us represents the minimum amount of time to wait before
+ * giving up. The actual time taken by this function can be a little more than
+ * @timeout_us for different reasons, specially in non-atomic contexts. Thus,
+ * it is possible that this function succeeds even after @timeout_us has passed.
+ */
+int xe_mmio_wait32(struct xe_gt *gt, struct xe_reg reg, u32 mask, u32 val, u32 timeout_us,
+ u32 *out_val, bool atomic)
+{
+ ktime_t cur = ktime_get_raw();
+ const ktime_t end = ktime_add_us(cur, timeout_us);
+ int ret = -ETIMEDOUT;
+ s64 wait = 10;
+ u32 read;
+
+ for (;;) {
+ read = xe_mmio_read32(gt, reg);
+ if ((read & mask) == val) {
+ ret = 0;
+ break;
+ }
+
+ cur = ktime_get_raw();
+ if (!ktime_before(cur, end))
+ break;
+
+ if (ktime_after(ktime_add_us(cur, wait), end))
+ wait = ktime_us_delta(end, cur);
+
+ if (atomic)
+ udelay(wait);
+ else
+ usleep_range(wait, wait << 1);
+ wait <<= 1;
+ }
+
+ if (ret != 0) {
+ read = xe_mmio_read32(gt, reg);
+ if ((read & mask) == val)
+ ret = 0;
+ }
+
+ if (out_val)
+ *out_val = read;
+
+ return ret;
+}
diff --git a/drivers/gpu/drm/xe/xe_mmio.h b/drivers/gpu/drm/xe/xe_mmio.h
new file mode 100644
index 000000000000..98de5c13c89b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_mmio.h
@@ -0,0 +1,107 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021-2023 Intel Corporation
+ */
+
+#ifndef _XE_MMIO_H_
+#define _XE_MMIO_H_
+
+#include <linux/delay.h>
+#include <linux/io-64-nonatomic-lo-hi.h>
+
+#include "regs/xe_reg_defs.h"
+#include "xe_device_types.h"
+#include "xe_gt_printk.h"
+#include "xe_gt_types.h"
+
+struct drm_device;
+struct drm_file;
+struct xe_device;
+
+#define LMEM_BAR 2
+
+int xe_mmio_init(struct xe_device *xe);
+int xe_mmio_root_tile_init(struct xe_device *xe);
+void xe_mmio_probe_tiles(struct xe_device *xe);
+
+static inline u8 xe_mmio_read8(struct xe_gt *gt, struct xe_reg reg)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+
+ if (reg.addr < gt->mmio.adj_limit)
+ reg.addr += gt->mmio.adj_offset;
+
+ return readb((reg.ext ? tile->mmio_ext.regs : tile->mmio.regs) + reg.addr);
+}
+
+static inline u16 xe_mmio_read16(struct xe_gt *gt, struct xe_reg reg)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+
+ if (reg.addr < gt->mmio.adj_limit)
+ reg.addr += gt->mmio.adj_offset;
+
+ return readw((reg.ext ? tile->mmio_ext.regs : tile->mmio.regs) + reg.addr);
+}
+
+static inline void xe_mmio_write32(struct xe_gt *gt,
+ struct xe_reg reg, u32 val)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+
+ if (reg.addr < gt->mmio.adj_limit)
+ reg.addr += gt->mmio.adj_offset;
+
+ writel(val, (reg.ext ? tile->mmio_ext.regs : tile->mmio.regs) + reg.addr);
+}
+
+static inline u32 xe_mmio_read32(struct xe_gt *gt, struct xe_reg reg)
+{
+ struct xe_tile *tile = gt_to_tile(gt);
+
+ if (reg.addr < gt->mmio.adj_limit)
+ reg.addr += gt->mmio.adj_offset;
+
+ return readl((reg.ext ? tile->mmio_ext.regs : tile->mmio.regs) + reg.addr);
+}
+
+static inline u32 xe_mmio_rmw32(struct xe_gt *gt, struct xe_reg reg, u32 clr,
+ u32 set)
+{
+ u32 old, reg_val;
+
+ old = xe_mmio_read32(gt, reg);
+ reg_val = (old & ~clr) | set;
+ xe_mmio_write32(gt, reg, reg_val);
+
+ return old;
+}
+
+static inline int xe_mmio_write32_and_verify(struct xe_gt *gt,
+ struct xe_reg reg, u32 val,
+ u32 mask, u32 eval)
+{
+ u32 reg_val;
+
+ xe_mmio_write32(gt, reg, val);
+ reg_val = xe_mmio_read32(gt, reg);
+
+ return (reg_val & mask) != eval ? -EINVAL : 0;
+}
+
+static inline bool xe_mmio_in_range(const struct xe_gt *gt,
+ const struct xe_mmio_range *range,
+ struct xe_reg reg)
+{
+ if (reg.addr < gt->mmio.adj_limit)
+ reg.addr += gt->mmio.adj_offset;
+
+ return range && reg.addr >= range->start && reg.addr <= range->end;
+}
+
+int xe_mmio_probe_vram(struct xe_device *xe);
+u64 xe_mmio_read64_2x32(struct xe_gt *gt, struct xe_reg reg);
+int xe_mmio_wait32(struct xe_gt *gt, struct xe_reg reg, u32 mask, u32 val, u32 timeout_us,
+ u32 *out_val, bool atomic);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_mocs.c b/drivers/gpu/drm/xe/xe_mocs.c
new file mode 100644
index 000000000000..ef79552e4f2f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_mocs.c
@@ -0,0 +1,580 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_mocs.h"
+
+#include "regs/xe_gt_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_mmio.h"
+#include "xe_platform_types.h"
+#include "xe_step_types.h"
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+#define mocs_dbg drm_dbg
+#else
+__printf(2, 3)
+static inline void mocs_dbg(const struct drm_device *dev,
+ const char *format, ...)
+{ /* noop */ }
+#endif
+
+enum {
+ HAS_GLOBAL_MOCS = BIT(0),
+ HAS_LNCF_MOCS = BIT(1),
+};
+
+struct xe_mocs_entry {
+ u32 control_value;
+ u16 l3cc_value;
+ u16 used;
+};
+
+struct xe_mocs_info {
+ unsigned int size;
+ unsigned int n_entries;
+ const struct xe_mocs_entry *table;
+ u8 uc_index;
+ u8 wb_index;
+ u8 unused_entries_index;
+};
+
+/* Defines for the tables (XXX_MOCS_0 - XXX_MOCS_63) */
+#define _LE_CACHEABILITY(value) ((value) << 0)
+#define _LE_TGT_CACHE(value) ((value) << 2)
+#define LE_LRUM(value) ((value) << 4)
+#define LE_AOM(value) ((value) << 6)
+#define LE_RSC(value) ((value) << 7)
+#define LE_SCC(value) ((value) << 8)
+#define LE_PFM(value) ((value) << 11)
+#define LE_SCF(value) ((value) << 14)
+#define LE_COS(value) ((value) << 15)
+#define LE_SSE(value) ((value) << 17)
+
+/* Defines for the tables (LNCFMOCS0 - LNCFMOCS31) - two entries per word */
+#define L3_ESC(value) ((value) << 0)
+#define L3_SCC(value) ((value) << 1)
+#define _L3_CACHEABILITY(value) ((value) << 4)
+#define L3_GLBGO(value) ((value) << 6)
+#define L3_LKUP(value) ((value) << 7)
+
+/* Defines for the tables (GLOB_MOCS_0 - GLOB_MOCS_16) */
+#define IG_PAT REG_BIT(8)
+#define L3_CACHE_POLICY_MASK REG_GENMASK(5, 4)
+#define L4_CACHE_POLICY_MASK REG_GENMASK(3, 2)
+
+/* Helper defines */
+#define XELP_NUM_MOCS_ENTRIES 64 /* 63-64 are reserved, but configured. */
+#define PVC_NUM_MOCS_ENTRIES 3
+#define MTL_NUM_MOCS_ENTRIES 16
+#define XE2_NUM_MOCS_ENTRIES 16
+
+/* (e)LLC caching options */
+/*
+ * Note: LE_0_PAGETABLE works only up to Gen11; for newer gens it means
+ * the same as LE_UC
+ */
+#define LE_0_PAGETABLE _LE_CACHEABILITY(0)
+#define LE_1_UC _LE_CACHEABILITY(1)
+#define LE_2_WT _LE_CACHEABILITY(2)
+#define LE_3_WB _LE_CACHEABILITY(3)
+
+/* Target cache */
+#define LE_TC_0_PAGETABLE _LE_TGT_CACHE(0)
+#define LE_TC_1_LLC _LE_TGT_CACHE(1)
+#define LE_TC_2_LLC_ELLC _LE_TGT_CACHE(2)
+#define LE_TC_3_LLC_ELLC_ALT _LE_TGT_CACHE(3)
+
+/* L3 caching options */
+#define L3_0_DIRECT _L3_CACHEABILITY(0)
+#define L3_1_UC _L3_CACHEABILITY(1)
+#define L3_2_RESERVED _L3_CACHEABILITY(2)
+#define L3_3_WB _L3_CACHEABILITY(3)
+
+/* L4 caching options */
+#define L4_0_WB REG_FIELD_PREP(L4_CACHE_POLICY_MASK, 0)
+#define L4_1_WT REG_FIELD_PREP(L4_CACHE_POLICY_MASK, 1)
+#define L4_3_UC REG_FIELD_PREP(L4_CACHE_POLICY_MASK, 3)
+
+#define XE2_L3_0_WB REG_FIELD_PREP(L3_CACHE_POLICY_MASK, 0)
+/* XD: WB Transient Display */
+#define XE2_L3_1_XD REG_FIELD_PREP(L3_CACHE_POLICY_MASK, 1)
+#define XE2_L3_3_UC REG_FIELD_PREP(L3_CACHE_POLICY_MASK, 3)
+
+#define MOCS_ENTRY(__idx, __control_value, __l3cc_value) \
+ [__idx] = { \
+ .control_value = __control_value, \
+ .l3cc_value = __l3cc_value, \
+ .used = 1, \
+ }
+
+/*
+ * MOCS tables
+ *
+ * These are the MOCS tables that are programmed across all the rings.
+ * The control value is programmed to all the rings that support the
+ * MOCS registers. While the l3cc_values are only programmed to the
+ * LNCFCMOCS0 - LNCFCMOCS32 registers.
+ *
+ * These tables are intended to be kept reasonably consistent across
+ * HW platforms, and for ICL+, be identical across OSes. To achieve
+ * that, the list of entries is published as part of bspec.
+ *
+ * Entries not part of the following tables are undefined as far as userspace is
+ * concerned and shouldn't be relied upon. The last few entries are reserved by
+ * the hardware. They should be initialized according to bspec and never used.
+ *
+ * NOTE1: These tables are part of bspec and defined as part of the hardware
+ * interface. It is expected that, for specific hardware platform, existing
+ * entries will remain constant and the table will only be updated by adding new
+ * entries, filling unused positions.
+ *
+ * NOTE2: Reserved and unspecified MOCS indices have been set to L3 WB. These
+ * reserved entries should never be used. They may be changed to low performant
+ * variants with better coherency in the future if more entries are needed.
+ */
+
+static const struct xe_mocs_entry gen12_mocs_desc[] = {
+ /* Base - L3 + LLC */
+ MOCS_ENTRY(2,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_3_WB),
+ /* Base - Uncached */
+ MOCS_ENTRY(3,
+ LE_1_UC | LE_TC_1_LLC,
+ L3_1_UC),
+ /* Base - L3 */
+ MOCS_ENTRY(4,
+ LE_1_UC | LE_TC_1_LLC,
+ L3_3_WB),
+ /* Base - LLC */
+ MOCS_ENTRY(5,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_1_UC),
+ /* Age 0 - LLC */
+ MOCS_ENTRY(6,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(1),
+ L3_1_UC),
+ /* Age 0 - L3 + LLC */
+ MOCS_ENTRY(7,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(1),
+ L3_3_WB),
+ /* Age: Don't Chg. - LLC */
+ MOCS_ENTRY(8,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(2),
+ L3_1_UC),
+ /* Age: Don't Chg. - L3 + LLC */
+ MOCS_ENTRY(9,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(2),
+ L3_3_WB),
+ /* No AOM - LLC */
+ MOCS_ENTRY(10,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_AOM(1),
+ L3_1_UC),
+ /* No AOM - L3 + LLC */
+ MOCS_ENTRY(11,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_AOM(1),
+ L3_3_WB),
+ /* No AOM; Age 0 - LLC */
+ MOCS_ENTRY(12,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(1) | LE_AOM(1),
+ L3_1_UC),
+ /* No AOM; Age 0 - L3 + LLC */
+ MOCS_ENTRY(13,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(1) | LE_AOM(1),
+ L3_3_WB),
+ /* No AOM; Age:DC - LLC */
+ MOCS_ENTRY(14,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(2) | LE_AOM(1),
+ L3_1_UC),
+ /* No AOM; Age:DC - L3 + LLC */
+ MOCS_ENTRY(15,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(2) | LE_AOM(1),
+ L3_3_WB),
+ /* Self-Snoop - L3 + LLC */
+ MOCS_ENTRY(18,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SSE(3),
+ L3_3_WB),
+ /* Skip Caching - L3 + LLC(12.5%) */
+ MOCS_ENTRY(19,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(7),
+ L3_3_WB),
+ /* Skip Caching - L3 + LLC(25%) */
+ MOCS_ENTRY(20,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(3),
+ L3_3_WB),
+ /* Skip Caching - L3 + LLC(50%) */
+ MOCS_ENTRY(21,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_SCC(1),
+ L3_3_WB),
+ /* Skip Caching - L3 + LLC(75%) */
+ MOCS_ENTRY(22,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_RSC(1) | LE_SCC(3),
+ L3_3_WB),
+ /* Skip Caching - L3 + LLC(87.5%) */
+ MOCS_ENTRY(23,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3) | LE_RSC(1) | LE_SCC(7),
+ L3_3_WB),
+ /* Implicitly enable L1 - HDC:L1 + L3 + LLC */
+ MOCS_ENTRY(48,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_3_WB),
+ /* Implicitly enable L1 - HDC:L1 + L3 */
+ MOCS_ENTRY(49,
+ LE_1_UC | LE_TC_1_LLC,
+ L3_3_WB),
+ /* Implicitly enable L1 - HDC:L1 + LLC */
+ MOCS_ENTRY(50,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_1_UC),
+ /* Implicitly enable L1 - HDC:L1 */
+ MOCS_ENTRY(51,
+ LE_1_UC | LE_TC_1_LLC,
+ L3_1_UC),
+ /* HW Special Case (CCS) */
+ MOCS_ENTRY(60,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_1_UC),
+ /* HW Special Case (Displayable) */
+ MOCS_ENTRY(61,
+ LE_1_UC | LE_TC_1_LLC,
+ L3_3_WB),
+ /* HW Reserved - SW program but never use */
+ MOCS_ENTRY(62,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_1_UC),
+ /* HW Reserved - SW program but never use */
+ MOCS_ENTRY(63,
+ LE_3_WB | LE_TC_1_LLC | LE_LRUM(3),
+ L3_1_UC)
+};
+
+static const struct xe_mocs_entry dg1_mocs_desc[] = {
+ /* UC */
+ MOCS_ENTRY(1, 0, L3_1_UC),
+ /* WB - L3 */
+ MOCS_ENTRY(5, 0, L3_3_WB),
+ /* WB - L3 50% */
+ MOCS_ENTRY(6, 0, L3_ESC(1) | L3_SCC(1) | L3_3_WB),
+ /* WB - L3 25% */
+ MOCS_ENTRY(7, 0, L3_ESC(1) | L3_SCC(3) | L3_3_WB),
+ /* WB - L3 12.5% */
+ MOCS_ENTRY(8, 0, L3_ESC(1) | L3_SCC(7) | L3_3_WB),
+
+ /* HDC:L1 + L3 */
+ MOCS_ENTRY(48, 0, L3_3_WB),
+ /* HDC:L1 */
+ MOCS_ENTRY(49, 0, L3_1_UC),
+
+ /* HW Reserved */
+ MOCS_ENTRY(60, 0, L3_1_UC),
+ MOCS_ENTRY(61, 0, L3_1_UC),
+ MOCS_ENTRY(62, 0, L3_1_UC),
+ MOCS_ENTRY(63, 0, L3_1_UC),
+};
+
+static const struct xe_mocs_entry dg2_mocs_desc[] = {
+ /* UC - Coherent; GO:L3 */
+ MOCS_ENTRY(0, 0, L3_1_UC | L3_LKUP(1)),
+ /* UC - Coherent; GO:Memory */
+ MOCS_ENTRY(1, 0, L3_1_UC | L3_GLBGO(1) | L3_LKUP(1)),
+ /* UC - Non-Coherent; GO:Memory */
+ MOCS_ENTRY(2, 0, L3_1_UC | L3_GLBGO(1)),
+
+ /* WB - LC */
+ MOCS_ENTRY(3, 0, L3_3_WB | L3_LKUP(1)),
+};
+
+static const struct xe_mocs_entry dg2_mocs_desc_g10_ax[] = {
+ /* Wa_14011441408: Set Go to Memory for MOCS#0 */
+ MOCS_ENTRY(0, 0, L3_1_UC | L3_GLBGO(1) | L3_LKUP(1)),
+ /* UC - Coherent; GO:Memory */
+ MOCS_ENTRY(1, 0, L3_1_UC | L3_GLBGO(1) | L3_LKUP(1)),
+ /* UC - Non-Coherent; GO:Memory */
+ MOCS_ENTRY(2, 0, L3_1_UC | L3_GLBGO(1)),
+
+ /* WB - LC */
+ MOCS_ENTRY(3, 0, L3_3_WB | L3_LKUP(1)),
+};
+
+static const struct xe_mocs_entry pvc_mocs_desc[] = {
+ /* Error */
+ MOCS_ENTRY(0, 0, L3_3_WB),
+
+ /* UC */
+ MOCS_ENTRY(1, 0, L3_1_UC),
+
+ /* WB */
+ MOCS_ENTRY(2, 0, L3_3_WB),
+};
+
+static const struct xe_mocs_entry mtl_mocs_desc[] = {
+ /* Error - Reserved for Non-Use */
+ MOCS_ENTRY(0,
+ 0,
+ L3_LKUP(1) | L3_3_WB),
+ /* Cached - L3 + L4 */
+ MOCS_ENTRY(1,
+ IG_PAT,
+ L3_LKUP(1) | L3_3_WB),
+ /* L4 - GO:L3 */
+ MOCS_ENTRY(2,
+ IG_PAT,
+ L3_LKUP(1) | L3_1_UC),
+ /* Uncached - GO:L3 */
+ MOCS_ENTRY(3,
+ IG_PAT | L4_3_UC,
+ L3_LKUP(1) | L3_1_UC),
+ /* L4 - GO:Mem */
+ MOCS_ENTRY(4,
+ IG_PAT,
+ L3_LKUP(1) | L3_GLBGO(1) | L3_1_UC),
+ /* Uncached - GO:Mem */
+ MOCS_ENTRY(5,
+ IG_PAT | L4_3_UC,
+ L3_LKUP(1) | L3_GLBGO(1) | L3_1_UC),
+ /* L4 - L3:NoLKUP; GO:L3 */
+ MOCS_ENTRY(6,
+ IG_PAT,
+ L3_1_UC),
+ /* Uncached - L3:NoLKUP; GO:L3 */
+ MOCS_ENTRY(7,
+ IG_PAT | L4_3_UC,
+ L3_1_UC),
+ /* L4 - L3:NoLKUP; GO:Mem */
+ MOCS_ENTRY(8,
+ IG_PAT,
+ L3_GLBGO(1) | L3_1_UC),
+ /* Uncached - L3:NoLKUP; GO:Mem */
+ MOCS_ENTRY(9,
+ IG_PAT | L4_3_UC,
+ L3_GLBGO(1) | L3_1_UC),
+ /* Display - L3; L4:WT */
+ MOCS_ENTRY(14,
+ IG_PAT | L4_1_WT,
+ L3_LKUP(1) | L3_3_WB),
+ /* CCS - Non-Displayable */
+ MOCS_ENTRY(15,
+ IG_PAT,
+ L3_GLBGO(1) | L3_1_UC),
+};
+
+static const struct xe_mocs_entry xe2_mocs_table[] = {
+ /* Defer to PAT */
+ MOCS_ENTRY(0, XE2_L3_0_WB | L4_3_UC, 0),
+ /* Cached L3, Uncached L4 */
+ MOCS_ENTRY(1, IG_PAT | XE2_L3_0_WB | L4_3_UC, 0),
+ /* Uncached L3, Cached L4 */
+ MOCS_ENTRY(2, IG_PAT | XE2_L3_3_UC | L4_0_WB, 0),
+ /* Uncached L3 + L4 */
+ MOCS_ENTRY(3, IG_PAT | XE2_L3_3_UC | L4_3_UC, 0),
+ /* Cached L3 + L4 */
+ MOCS_ENTRY(4, IG_PAT | XE2_L3_0_WB | L4_0_WB, 0),
+};
+
+static unsigned int get_mocs_settings(struct xe_device *xe,
+ struct xe_mocs_info *info)
+{
+ unsigned int flags = 0;
+
+ memset(info, 0, sizeof(struct xe_mocs_info));
+
+ switch (xe->info.platform) {
+ case XE_LUNARLAKE:
+ info->size = ARRAY_SIZE(xe2_mocs_table);
+ info->table = xe2_mocs_table;
+ info->n_entries = XE2_NUM_MOCS_ENTRIES;
+ info->uc_index = 3;
+ info->wb_index = 4;
+ info->unused_entries_index = 4;
+ break;
+ case XE_PVC:
+ info->size = ARRAY_SIZE(pvc_mocs_desc);
+ info->table = pvc_mocs_desc;
+ info->n_entries = PVC_NUM_MOCS_ENTRIES;
+ info->uc_index = 1;
+ info->wb_index = 2;
+ info->unused_entries_index = 2;
+ break;
+ case XE_METEORLAKE:
+ info->size = ARRAY_SIZE(mtl_mocs_desc);
+ info->table = mtl_mocs_desc;
+ info->n_entries = MTL_NUM_MOCS_ENTRIES;
+ info->uc_index = 9;
+ info->unused_entries_index = 1;
+ break;
+ case XE_DG2:
+ if (xe->info.subplatform == XE_SUBPLATFORM_DG2_G10 &&
+ xe->info.step.graphics >= STEP_A0 &&
+ xe->info.step.graphics <= STEP_B0) {
+ info->size = ARRAY_SIZE(dg2_mocs_desc_g10_ax);
+ info->table = dg2_mocs_desc_g10_ax;
+ } else {
+ info->size = ARRAY_SIZE(dg2_mocs_desc);
+ info->table = dg2_mocs_desc;
+ }
+ info->uc_index = 1;
+ info->n_entries = XELP_NUM_MOCS_ENTRIES;
+ info->unused_entries_index = 3;
+ break;
+ case XE_DG1:
+ info->size = ARRAY_SIZE(dg1_mocs_desc);
+ info->table = dg1_mocs_desc;
+ info->uc_index = 1;
+ info->n_entries = XELP_NUM_MOCS_ENTRIES;
+ info->unused_entries_index = 5;
+ break;
+ case XE_TIGERLAKE:
+ case XE_ROCKETLAKE:
+ case XE_ALDERLAKE_S:
+ case XE_ALDERLAKE_P:
+ case XE_ALDERLAKE_N:
+ info->size = ARRAY_SIZE(gen12_mocs_desc);
+ info->table = gen12_mocs_desc;
+ info->n_entries = XELP_NUM_MOCS_ENTRIES;
+ info->uc_index = 3;
+ info->unused_entries_index = 2;
+ break;
+ default:
+ drm_err(&xe->drm, "Platform that should have a MOCS table does not.\n");
+ return 0;
+ }
+
+ /*
+ * Index 0 is a reserved/unused table entry on most platforms, but
+ * even on those where it does represent a legitimate MOCS entry, it
+ * never represents the "most cached, least coherent" behavior we want
+ * to populate undefined table rows with. So if unused_entries_index
+ * is still 0 at this point, we'll assume that it was omitted by
+ * mistake in the switch statement above.
+ */
+ xe_assert(xe, info->unused_entries_index != 0);
+
+ if (XE_WARN_ON(info->size > info->n_entries)) {
+ info->table = NULL;
+ return 0;
+ }
+
+ if (!IS_DGFX(xe) || GRAPHICS_VER(xe) >= 20)
+ flags |= HAS_GLOBAL_MOCS;
+ if (GRAPHICS_VER(xe) < 20)
+ flags |= HAS_LNCF_MOCS;
+
+ return flags;
+}
+
+/*
+ * Get control_value from MOCS entry. If the table entry is not defined, the
+ * settings from unused_entries_index will be returned.
+ */
+static u32 get_entry_control(const struct xe_mocs_info *info,
+ unsigned int index)
+{
+ if (index < info->size && info->table[index].used)
+ return info->table[index].control_value;
+ return info->table[info->unused_entries_index].control_value;
+}
+
+static void __init_mocs_table(struct xe_gt *gt,
+ const struct xe_mocs_info *info)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ unsigned int i;
+ u32 mocs;
+
+ mocs_dbg(&gt_to_xe(gt)->drm, "entries:%d\n", info->n_entries);
+ drm_WARN_ONCE(&xe->drm, !info->unused_entries_index,
+ "Unused entries index should have been defined\n");
+ for (i = 0;
+ i < info->n_entries ? (mocs = get_entry_control(info, i)), 1 : 0;
+ i++) {
+ mocs_dbg(&gt_to_xe(gt)->drm, "GLOB_MOCS[%d] 0x%x 0x%x\n", i,
+ XELP_GLOBAL_MOCS(i).addr, mocs);
+
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) > 1250)
+ xe_gt_mcr_multicast_write(gt, XEHP_GLOBAL_MOCS(i), mocs);
+ else
+ xe_mmio_write32(gt, XELP_GLOBAL_MOCS(i), mocs);
+ }
+}
+
+/*
+ * Get l3cc_value from MOCS entry taking into account when it's not used
+ * then if unused_entries_index is not zero then its value will be returned
+ * otherwise I915_MOCS_PTE's value is returned in this case.
+ */
+static u16 get_entry_l3cc(const struct xe_mocs_info *info,
+ unsigned int index)
+{
+ if (index < info->size && info->table[index].used)
+ return info->table[index].l3cc_value;
+ return info->table[info->unused_entries_index].l3cc_value;
+}
+
+static u32 l3cc_combine(u16 low, u16 high)
+{
+ return low | (u32)high << 16;
+}
+
+static void init_l3cc_table(struct xe_gt *gt,
+ const struct xe_mocs_info *info)
+{
+ unsigned int i;
+ u32 l3cc;
+
+ mocs_dbg(&gt_to_xe(gt)->drm, "entries:%d\n", info->n_entries);
+ for (i = 0;
+ i < (info->n_entries + 1) / 2 ?
+ (l3cc = l3cc_combine(get_entry_l3cc(info, 2 * i),
+ get_entry_l3cc(info, 2 * i + 1))), 1 : 0;
+ i++) {
+ mocs_dbg(&gt_to_xe(gt)->drm, "LNCFCMOCS[%d] 0x%x 0x%x\n", i, XELP_LNCFCMOCS(i).addr,
+ l3cc);
+
+ if (GRAPHICS_VERx100(gt_to_xe(gt)) >= 1250)
+ xe_gt_mcr_multicast_write(gt, XEHP_LNCFCMOCS(i), l3cc);
+ else
+ xe_mmio_write32(gt, XELP_LNCFCMOCS(i), l3cc);
+ }
+}
+
+void xe_mocs_init_early(struct xe_gt *gt)
+{
+ struct xe_mocs_info table;
+
+ get_mocs_settings(gt_to_xe(gt), &table);
+ gt->mocs.uc_index = table.uc_index;
+ gt->mocs.wb_index = table.wb_index;
+}
+
+void xe_mocs_init(struct xe_gt *gt)
+{
+ struct xe_mocs_info table;
+ unsigned int flags;
+
+ /*
+ * MOCS settings are split between "GLOB_MOCS" and/or "LNCFCMOCS"
+ * registers depending on platform.
+ *
+ * These registers should be programmed before GuC initialization
+ * since their values will affect some of the memory transactions
+ * performed by the GuC.
+ */
+ flags = get_mocs_settings(gt_to_xe(gt), &table);
+ mocs_dbg(&gt_to_xe(gt)->drm, "flag:0x%x\n", flags);
+
+ if (flags & HAS_GLOBAL_MOCS)
+ __init_mocs_table(gt, &table);
+ if (flags & HAS_LNCF_MOCS)
+ init_l3cc_table(gt, &table);
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_mocs.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_mocs.h b/drivers/gpu/drm/xe/xe_mocs.h
new file mode 100644
index 000000000000..053754c5a94e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_mocs.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_MOCS_H_
+#define _XE_MOCS_H_
+
+#include <linux/types.h>
+
+struct xe_exec_queue;
+struct xe_gt;
+
+void xe_mocs_init_early(struct xe_gt *gt);
+void xe_mocs_init(struct xe_gt *gt);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_module.c b/drivers/gpu/drm/xe/xe_module.c
new file mode 100644
index 000000000000..110b69864656
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_module.c
@@ -0,0 +1,101 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_module.h"
+
+#include <linux/init.h>
+#include <linux/module.h>
+
+#include "xe_drv.h"
+#include "xe_hw_fence.h"
+#include "xe_pci.h"
+#include "xe_sched_job.h"
+
+struct xe_modparam xe_modparam = {
+ .enable_display = true,
+ .guc_log_level = 5,
+ .force_probe = CONFIG_DRM_XE_FORCE_PROBE,
+ /* the rest are 0 by default */
+};
+
+module_param_named_unsafe(force_execlist, xe_modparam.force_execlist, bool, 0444);
+MODULE_PARM_DESC(force_execlist, "Force Execlist submission");
+
+module_param_named(enable_display, xe_modparam.enable_display, bool, 0444);
+MODULE_PARM_DESC(enable_display, "Enable display");
+
+module_param_named(vram_bar_size, xe_modparam.force_vram_bar_size, uint, 0600);
+MODULE_PARM_DESC(vram_bar_size, "Set the vram bar size(in MiB)");
+
+module_param_named(guc_log_level, xe_modparam.guc_log_level, int, 0600);
+MODULE_PARM_DESC(guc_log_level, "GuC firmware logging level (0=disable, 1..5=enable with verbosity min..max)");
+
+module_param_named_unsafe(guc_firmware_path, xe_modparam.guc_firmware_path, charp, 0400);
+MODULE_PARM_DESC(guc_firmware_path,
+ "GuC firmware path to use instead of the default one");
+
+module_param_named_unsafe(huc_firmware_path, xe_modparam.huc_firmware_path, charp, 0400);
+MODULE_PARM_DESC(huc_firmware_path,
+ "HuC firmware path to use instead of the default one - empty string disables");
+
+module_param_named_unsafe(gsc_firmware_path, xe_modparam.gsc_firmware_path, charp, 0400);
+MODULE_PARM_DESC(gsc_firmware_path,
+ "GSC firmware path to use instead of the default one - empty string disables");
+
+module_param_named_unsafe(force_probe, xe_modparam.force_probe, charp, 0400);
+MODULE_PARM_DESC(force_probe,
+ "Force probe options for specified devices. See CONFIG_DRM_XE_FORCE_PROBE for details.");
+
+struct init_funcs {
+ int (*init)(void);
+ void (*exit)(void);
+};
+
+static const struct init_funcs init_funcs[] = {
+ {
+ .init = xe_hw_fence_module_init,
+ .exit = xe_hw_fence_module_exit,
+ },
+ {
+ .init = xe_sched_job_module_init,
+ .exit = xe_sched_job_module_exit,
+ },
+ {
+ .init = xe_register_pci_driver,
+ .exit = xe_unregister_pci_driver,
+ },
+};
+
+static int __init xe_init(void)
+{
+ int err, i;
+
+ for (i = 0; i < ARRAY_SIZE(init_funcs); i++) {
+ err = init_funcs[i].init();
+ if (err) {
+ while (i--)
+ init_funcs[i].exit();
+ return err;
+ }
+ }
+
+ return 0;
+}
+
+static void __exit xe_exit(void)
+{
+ int i;
+
+ for (i = ARRAY_SIZE(init_funcs) - 1; i >= 0; i--)
+ init_funcs[i].exit();
+}
+
+module_init(xe_init);
+module_exit(xe_exit);
+
+MODULE_AUTHOR("Intel Corporation");
+
+MODULE_DESCRIPTION(DRIVER_DESC);
+MODULE_LICENSE("GPL and additional rights");
diff --git a/drivers/gpu/drm/xe/xe_module.h b/drivers/gpu/drm/xe/xe_module.h
new file mode 100644
index 000000000000..88ef0e8b2bfd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_module.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_MODULE_H_
+#define _XE_MODULE_H_
+
+#include <linux/types.h>
+
+/* Module modprobe variables */
+struct xe_modparam {
+ bool force_execlist;
+ bool enable_display;
+ u32 force_vram_bar_size;
+ int guc_log_level;
+ char *guc_firmware_path;
+ char *huc_firmware_path;
+ char *gsc_firmware_path;
+ char *force_probe;
+};
+
+extern struct xe_modparam xe_modparam;
+
+#endif
+
diff --git a/drivers/gpu/drm/xe/xe_pat.c b/drivers/gpu/drm/xe/xe_pat.c
new file mode 100644
index 000000000000..1ff6bc79e7d4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pat.c
@@ -0,0 +1,459 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_pat.h"
+
+#include <drm/xe_drm.h>
+
+#include "regs/xe_reg_defs.h"
+#include "xe_assert.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_mmio.h"
+
+#define _PAT_ATS 0x47fc
+#define _PAT_INDEX(index) _PICK_EVEN_2RANGES(index, 8, \
+ 0x4800, 0x4804, \
+ 0x4848, 0x484c)
+#define _PAT_PTA 0x4820
+
+#define XE2_NO_PROMOTE REG_BIT(10)
+#define XE2_COMP_EN REG_BIT(9)
+#define XE2_L3_CLOS REG_GENMASK(7, 6)
+#define XE2_L3_POLICY REG_GENMASK(5, 4)
+#define XE2_L4_POLICY REG_GENMASK(3, 2)
+#define XE2_COH_MODE REG_GENMASK(1, 0)
+
+#define XELPG_L4_POLICY_MASK REG_GENMASK(3, 2)
+#define XELPG_PAT_3_UC REG_FIELD_PREP(XELPG_L4_POLICY_MASK, 3)
+#define XELPG_PAT_1_WT REG_FIELD_PREP(XELPG_L4_POLICY_MASK, 1)
+#define XELPG_PAT_0_WB REG_FIELD_PREP(XELPG_L4_POLICY_MASK, 0)
+#define XELPG_INDEX_COH_MODE_MASK REG_GENMASK(1, 0)
+#define XELPG_3_COH_2W REG_FIELD_PREP(XELPG_INDEX_COH_MODE_MASK, 3)
+#define XELPG_2_COH_1W REG_FIELD_PREP(XELPG_INDEX_COH_MODE_MASK, 2)
+#define XELPG_0_COH_NON REG_FIELD_PREP(XELPG_INDEX_COH_MODE_MASK, 0)
+
+#define XEHPC_CLOS_LEVEL_MASK REG_GENMASK(3, 2)
+#define XEHPC_PAT_CLOS(x) REG_FIELD_PREP(XEHPC_CLOS_LEVEL_MASK, x)
+
+#define XELP_MEM_TYPE_MASK REG_GENMASK(1, 0)
+#define XELP_PAT_WB REG_FIELD_PREP(XELP_MEM_TYPE_MASK, 3)
+#define XELP_PAT_WT REG_FIELD_PREP(XELP_MEM_TYPE_MASK, 2)
+#define XELP_PAT_WC REG_FIELD_PREP(XELP_MEM_TYPE_MASK, 1)
+#define XELP_PAT_UC REG_FIELD_PREP(XELP_MEM_TYPE_MASK, 0)
+
+static const char *XELP_MEM_TYPE_STR_MAP[] = { "UC", "WC", "WT", "WB" };
+
+struct xe_pat_ops {
+ void (*program_graphics)(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries);
+ void (*program_media)(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries);
+ void (*dump)(struct xe_gt *gt, struct drm_printer *p);
+};
+
+static const struct xe_pat_table_entry xelp_pat_table[] = {
+ [0] = { XELP_PAT_WB, XE_COH_AT_LEAST_1WAY },
+ [1] = { XELP_PAT_WC, XE_COH_NONE },
+ [2] = { XELP_PAT_WT, XE_COH_NONE },
+ [3] = { XELP_PAT_UC, XE_COH_NONE },
+};
+
+static const struct xe_pat_table_entry xehpc_pat_table[] = {
+ [0] = { XELP_PAT_UC, XE_COH_NONE },
+ [1] = { XELP_PAT_WC, XE_COH_NONE },
+ [2] = { XELP_PAT_WT, XE_COH_NONE },
+ [3] = { XELP_PAT_WB, XE_COH_AT_LEAST_1WAY },
+ [4] = { XEHPC_PAT_CLOS(1) | XELP_PAT_WT, XE_COH_NONE },
+ [5] = { XEHPC_PAT_CLOS(1) | XELP_PAT_WB, XE_COH_AT_LEAST_1WAY },
+ [6] = { XEHPC_PAT_CLOS(2) | XELP_PAT_WT, XE_COH_NONE },
+ [7] = { XEHPC_PAT_CLOS(2) | XELP_PAT_WB, XE_COH_AT_LEAST_1WAY },
+};
+
+static const struct xe_pat_table_entry xelpg_pat_table[] = {
+ [0] = { XELPG_PAT_0_WB, XE_COH_NONE },
+ [1] = { XELPG_PAT_1_WT, XE_COH_NONE },
+ [2] = { XELPG_PAT_3_UC, XE_COH_NONE },
+ [3] = { XELPG_PAT_0_WB | XELPG_2_COH_1W, XE_COH_AT_LEAST_1WAY },
+ [4] = { XELPG_PAT_0_WB | XELPG_3_COH_2W, XE_COH_AT_LEAST_1WAY },
+};
+
+/*
+ * The Xe2 table is getting large/complicated so it's easier to review if
+ * provided in a form that exactly matches the bspec's formatting. The meaning
+ * of the fields here are:
+ * - no_promote: 0=promotable, 1=no promote
+ * - comp_en: 0=disable, 1=enable
+ * - l3clos: L3 class of service (0-3)
+ * - l3_policy: 0=WB, 1=XD ("WB - Transient Display"), 3=UC
+ * - l4_policy: 0=WB, 1=WT, 3=UC
+ * - coh_mode: 0=no snoop, 2=1-way coherent, 3=2-way coherent
+ *
+ * Reserved entries should be programmed with the maximum caching, minimum
+ * coherency (which matches an all-0's encoding), so we can just omit them
+ * in the table.
+ */
+#define XE2_PAT(no_promote, comp_en, l3clos, l3_policy, l4_policy, __coh_mode) \
+ { \
+ .value = (no_promote ? XE2_NO_PROMOTE : 0) | \
+ (comp_en ? XE2_COMP_EN : 0) | \
+ REG_FIELD_PREP(XE2_L3_CLOS, l3clos) | \
+ REG_FIELD_PREP(XE2_L3_POLICY, l3_policy) | \
+ REG_FIELD_PREP(XE2_L4_POLICY, l4_policy) | \
+ REG_FIELD_PREP(XE2_COH_MODE, __coh_mode), \
+ .coh_mode = __coh_mode ? XE_COH_AT_LEAST_1WAY : XE_COH_NONE \
+ }
+
+static const struct xe_pat_table_entry xe2_pat_table[] = {
+ [ 0] = XE2_PAT( 0, 0, 0, 0, 3, 0 ),
+ [ 1] = XE2_PAT( 0, 0, 0, 0, 3, 2 ),
+ [ 2] = XE2_PAT( 0, 0, 0, 0, 3, 3 ),
+ [ 3] = XE2_PAT( 0, 0, 0, 3, 3, 0 ),
+ [ 4] = XE2_PAT( 0, 0, 0, 3, 0, 2 ),
+ [ 5] = XE2_PAT( 0, 0, 0, 3, 3, 2 ),
+ [ 6] = XE2_PAT( 1, 0, 0, 1, 3, 0 ),
+ [ 7] = XE2_PAT( 0, 0, 0, 3, 0, 3 ),
+ [ 8] = XE2_PAT( 0, 0, 0, 3, 0, 0 ),
+ [ 9] = XE2_PAT( 0, 1, 0, 0, 3, 0 ),
+ [10] = XE2_PAT( 0, 1, 0, 3, 0, 0 ),
+ [11] = XE2_PAT( 1, 1, 0, 1, 3, 0 ),
+ [12] = XE2_PAT( 0, 1, 0, 3, 3, 0 ),
+ [13] = XE2_PAT( 0, 0, 0, 0, 0, 0 ),
+ [14] = XE2_PAT( 0, 1, 0, 0, 0, 0 ),
+ [15] = XE2_PAT( 1, 1, 0, 1, 1, 0 ),
+ /* 16..19 are reserved; leave set to all 0's */
+ [20] = XE2_PAT( 0, 0, 1, 0, 3, 0 ),
+ [21] = XE2_PAT( 0, 1, 1, 0, 3, 0 ),
+ [22] = XE2_PAT( 0, 0, 1, 0, 3, 2 ),
+ [23] = XE2_PAT( 0, 0, 1, 0, 3, 3 ),
+ [24] = XE2_PAT( 0, 0, 2, 0, 3, 0 ),
+ [25] = XE2_PAT( 0, 1, 2, 0, 3, 0 ),
+ [26] = XE2_PAT( 0, 0, 2, 0, 3, 2 ),
+ [27] = XE2_PAT( 0, 0, 2, 0, 3, 3 ),
+ [28] = XE2_PAT( 0, 0, 3, 0, 3, 0 ),
+ [29] = XE2_PAT( 0, 1, 3, 0, 3, 0 ),
+ [30] = XE2_PAT( 0, 0, 3, 0, 3, 2 ),
+ [31] = XE2_PAT( 0, 0, 3, 0, 3, 3 ),
+};
+
+/* Special PAT values programmed outside the main table */
+static const struct xe_pat_table_entry xe2_pat_ats = XE2_PAT( 0, 0, 0, 0, 3, 3 );
+
+u16 xe_pat_index_get_coh_mode(struct xe_device *xe, u16 pat_index)
+{
+ WARN_ON(pat_index >= xe->pat.n_entries);
+ return xe->pat.table[pat_index].coh_mode;
+}
+
+static void program_pat(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries)
+{
+ for (int i = 0; i < n_entries; i++) {
+ struct xe_reg reg = XE_REG(_PAT_INDEX(i));
+
+ xe_mmio_write32(gt, reg, table[i].value);
+ }
+}
+
+static void program_pat_mcr(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries)
+{
+ for (int i = 0; i < n_entries; i++) {
+ struct xe_reg_mcr reg_mcr = XE_REG_MCR(_PAT_INDEX(i));
+
+ xe_gt_mcr_multicast_write(gt, reg_mcr, table[i].value);
+ }
+}
+
+static void xelp_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i, err;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_fw;
+
+ drm_printf(p, "PAT table:\n");
+
+ for (i = 0; i < xe->pat.n_entries; i++) {
+ u32 pat = xe_mmio_read32(gt, XE_REG(_PAT_INDEX(i)));
+ u8 mem_type = REG_FIELD_GET(XELP_MEM_TYPE_MASK, pat);
+
+ drm_printf(p, "PAT[%2d] = %s (%#8x)\n", i,
+ XELP_MEM_TYPE_STR_MAP[mem_type], pat);
+ }
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_fw:
+ xe_assert(xe, !err);
+ xe_device_mem_access_put(xe);
+}
+
+static const struct xe_pat_ops xelp_pat_ops = {
+ .program_graphics = program_pat,
+ .dump = xelp_dump,
+};
+
+static void xehp_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i, err;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_fw;
+
+ drm_printf(p, "PAT table:\n");
+
+ for (i = 0; i < xe->pat.n_entries; i++) {
+ u32 pat = xe_gt_mcr_unicast_read_any(gt, XE_REG_MCR(_PAT_INDEX(i)));
+ u8 mem_type;
+
+ mem_type = REG_FIELD_GET(XELP_MEM_TYPE_MASK, pat);
+
+ drm_printf(p, "PAT[%2d] = %s (%#8x)\n", i,
+ XELP_MEM_TYPE_STR_MAP[mem_type], pat);
+ }
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_fw:
+ xe_assert(xe, !err);
+ xe_device_mem_access_put(xe);
+}
+
+static const struct xe_pat_ops xehp_pat_ops = {
+ .program_graphics = program_pat_mcr,
+ .dump = xehp_dump,
+};
+
+static void xehpc_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i, err;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_fw;
+
+ drm_printf(p, "PAT table:\n");
+
+ for (i = 0; i < xe->pat.n_entries; i++) {
+ u32 pat = xe_gt_mcr_unicast_read_any(gt, XE_REG_MCR(_PAT_INDEX(i)));
+
+ drm_printf(p, "PAT[%2d] = [ %u, %u ] (%#8x)\n", i,
+ REG_FIELD_GET(XELP_MEM_TYPE_MASK, pat),
+ REG_FIELD_GET(XEHPC_CLOS_LEVEL_MASK, pat), pat);
+ }
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_fw:
+ xe_assert(xe, !err);
+ xe_device_mem_access_put(xe);
+}
+
+static const struct xe_pat_ops xehpc_pat_ops = {
+ .program_graphics = program_pat_mcr,
+ .dump = xehpc_dump,
+};
+
+static void xelpg_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i, err;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_fw;
+
+ drm_printf(p, "PAT table:\n");
+
+ for (i = 0; i < xe->pat.n_entries; i++) {
+ u32 pat;
+
+ if (xe_gt_is_media_type(gt))
+ pat = xe_mmio_read32(gt, XE_REG(_PAT_INDEX(i)));
+ else
+ pat = xe_gt_mcr_unicast_read_any(gt, XE_REG_MCR(_PAT_INDEX(i)));
+
+ drm_printf(p, "PAT[%2d] = [ %u, %u ] (%#8x)\n", i,
+ REG_FIELD_GET(XELPG_L4_POLICY_MASK, pat),
+ REG_FIELD_GET(XELPG_INDEX_COH_MODE_MASK, pat), pat);
+ }
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_fw:
+ xe_assert(xe, !err);
+ xe_device_mem_access_put(xe);
+}
+
+/*
+ * SAMedia register offsets are adjusted by the write methods and they target
+ * registers that are not MCR, while for normal GT they are MCR
+ */
+static const struct xe_pat_ops xelpg_pat_ops = {
+ .program_graphics = program_pat,
+ .program_media = program_pat_mcr,
+ .dump = xelpg_dump,
+};
+
+static void xe2lpg_program_pat(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries)
+{
+ program_pat_mcr(gt, table, n_entries);
+ xe_gt_mcr_multicast_write(gt, XE_REG_MCR(_PAT_ATS), xe2_pat_ats.value);
+}
+
+static void xe2lpm_program_pat(struct xe_gt *gt, const struct xe_pat_table_entry table[],
+ int n_entries)
+{
+ program_pat(gt, table, n_entries);
+ xe_mmio_write32(gt, XE_REG(_PAT_ATS), xe2_pat_ats.value);
+}
+
+static void xe2_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ int i, err;
+ u32 pat;
+
+ xe_device_mem_access_get(xe);
+ err = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ if (err)
+ goto err_fw;
+
+ drm_printf(p, "PAT table:\n");
+
+ for (i = 0; i < xe->pat.n_entries; i++) {
+ if (xe_gt_is_media_type(gt))
+ pat = xe_mmio_read32(gt, XE_REG(_PAT_INDEX(i)));
+ else
+ pat = xe_gt_mcr_unicast_read_any(gt, XE_REG_MCR(_PAT_INDEX(i)));
+
+ drm_printf(p, "PAT[%2d] = [ %u, %u, %u, %u, %u, %u ] (%#8x)\n", i,
+ !!(pat & XE2_NO_PROMOTE),
+ !!(pat & XE2_COMP_EN),
+ REG_FIELD_GET(XE2_L3_CLOS, pat),
+ REG_FIELD_GET(XE2_L3_POLICY, pat),
+ REG_FIELD_GET(XE2_L4_POLICY, pat),
+ REG_FIELD_GET(XE2_COH_MODE, pat),
+ pat);
+ }
+
+ /*
+ * Also print PTA_MODE, which describes how the hardware accesses
+ * PPGTT entries.
+ */
+ if (xe_gt_is_media_type(gt))
+ pat = xe_mmio_read32(gt, XE_REG(_PAT_PTA));
+ else
+ pat = xe_gt_mcr_unicast_read_any(gt, XE_REG_MCR(_PAT_PTA));
+
+ drm_printf(p, "Page Table Access:\n");
+ drm_printf(p, "PTA_MODE= [ %u, %u, %u, %u, %u, %u ] (%#8x)\n",
+ !!(pat & XE2_NO_PROMOTE),
+ !!(pat & XE2_COMP_EN),
+ REG_FIELD_GET(XE2_L3_CLOS, pat),
+ REG_FIELD_GET(XE2_L3_POLICY, pat),
+ REG_FIELD_GET(XE2_L4_POLICY, pat),
+ REG_FIELD_GET(XE2_COH_MODE, pat),
+ pat);
+
+ err = xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+err_fw:
+ xe_assert(xe, !err);
+ xe_device_mem_access_put(xe);
+}
+
+static const struct xe_pat_ops xe2_pat_ops = {
+ .program_graphics = xe2lpg_program_pat,
+ .program_media = xe2lpm_program_pat,
+ .dump = xe2_dump,
+};
+
+void xe_pat_init_early(struct xe_device *xe)
+{
+ if (GRAPHICS_VER(xe) == 20) {
+ xe->pat.ops = &xe2_pat_ops;
+ xe->pat.table = xe2_pat_table;
+ xe->pat.n_entries = ARRAY_SIZE(xe2_pat_table);
+ xe->pat.idx[XE_CACHE_NONE] = 3;
+ xe->pat.idx[XE_CACHE_WT] = 15;
+ xe->pat.idx[XE_CACHE_WB] = 2;
+ xe->pat.idx[XE_CACHE_NONE_COMPRESSION] = 12; /*Applicable on xe2 and beyond */
+ } else if (xe->info.platform == XE_METEORLAKE) {
+ xe->pat.ops = &xelpg_pat_ops;
+ xe->pat.table = xelpg_pat_table;
+ xe->pat.n_entries = ARRAY_SIZE(xelpg_pat_table);
+ xe->pat.idx[XE_CACHE_NONE] = 2;
+ xe->pat.idx[XE_CACHE_WT] = 1;
+ xe->pat.idx[XE_CACHE_WB] = 3;
+ } else if (xe->info.platform == XE_PVC) {
+ xe->pat.ops = &xehpc_pat_ops;
+ xe->pat.table = xehpc_pat_table;
+ xe->pat.n_entries = ARRAY_SIZE(xehpc_pat_table);
+ xe->pat.idx[XE_CACHE_NONE] = 0;
+ xe->pat.idx[XE_CACHE_WT] = 2;
+ xe->pat.idx[XE_CACHE_WB] = 3;
+ } else if (xe->info.platform == XE_DG2) {
+ /*
+ * Table is the same as previous platforms, but programming
+ * method has changed.
+ */
+ xe->pat.ops = &xehp_pat_ops;
+ xe->pat.table = xelp_pat_table;
+ xe->pat.n_entries = ARRAY_SIZE(xelp_pat_table);
+ xe->pat.idx[XE_CACHE_NONE] = 3;
+ xe->pat.idx[XE_CACHE_WT] = 2;
+ xe->pat.idx[XE_CACHE_WB] = 0;
+ } else if (GRAPHICS_VERx100(xe) <= 1210) {
+ WARN_ON_ONCE(!IS_DGFX(xe) && !xe->info.has_llc);
+ xe->pat.ops = &xelp_pat_ops;
+ xe->pat.table = xelp_pat_table;
+ xe->pat.n_entries = ARRAY_SIZE(xelp_pat_table);
+ xe->pat.idx[XE_CACHE_NONE] = 3;
+ xe->pat.idx[XE_CACHE_WT] = 2;
+ xe->pat.idx[XE_CACHE_WB] = 0;
+ } else {
+ /*
+ * Going forward we expect to need new PAT settings for most
+ * new platforms; failure to provide a new table can easily
+ * lead to subtle, hard-to-debug problems. If none of the
+ * conditions above match the platform we're running on we'll
+ * raise an error rather than trying to silently inherit the
+ * most recent platform's behavior.
+ */
+ drm_err(&xe->drm, "Missing PAT table for platform with graphics version %d.%02d!\n",
+ GRAPHICS_VER(xe), GRAPHICS_VERx100(xe) % 100);
+ }
+}
+
+void xe_pat_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (!xe->pat.ops)
+ return;
+
+ if (xe_gt_is_media_type(gt))
+ xe->pat.ops->program_media(gt, xe->pat.table, xe->pat.n_entries);
+ else
+ xe->pat.ops->program_graphics(gt, xe->pat.table, xe->pat.n_entries);
+}
+
+void xe_pat_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ if (!xe->pat.ops->dump)
+ return;
+
+ xe->pat.ops->dump(gt, p);
+}
diff --git a/drivers/gpu/drm/xe/xe_pat.h b/drivers/gpu/drm/xe/xe_pat.h
new file mode 100644
index 000000000000..fa0dfbe525cd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pat.h
@@ -0,0 +1,61 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_PAT_H_
+#define _XE_PAT_H_
+
+#include <linux/types.h>
+
+struct drm_printer;
+struct xe_device;
+struct xe_gt;
+
+/**
+ * struct xe_pat_table_entry - The pat_index encoding and other meta information.
+ */
+struct xe_pat_table_entry {
+ /**
+ * @value: The platform specific value encoding the various memory
+ * attributes (this maps to some fixed pat_index). So things like
+ * caching, coherency, compression etc can be encoded here.
+ */
+ u32 value;
+
+ /**
+ * @coh_mode: The GPU coherency mode that @value maps to.
+ */
+#define XE_COH_NONE 1
+#define XE_COH_AT_LEAST_1WAY 2
+ u16 coh_mode;
+};
+
+/**
+ * xe_pat_init_early - SW initialization, setting up data based on device
+ * @xe: xe device
+ */
+void xe_pat_init_early(struct xe_device *xe);
+
+/**
+ * xe_pat_init - Program HW PAT table
+ * @gt: GT structure
+ */
+void xe_pat_init(struct xe_gt *gt);
+
+/**
+ * xe_pat_dump - Dump PAT table
+ * @gt: GT structure
+ * @p: Printer to dump info to
+ */
+void xe_pat_dump(struct xe_gt *gt, struct drm_printer *p);
+
+/**
+ * xe_pat_index_get_coh_mode - Extract the coherency mode for the given
+ * pat_index.
+ * @xe: xe device
+ * @pat_index: The pat_index to query
+ */
+u16 xe_pat_index_get_coh_mode(struct xe_device *xe, u16 pat_index);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
new file mode 100644
index 000000000000..dcc5ded1558e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pci.c
@@ -0,0 +1,951 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_pci.h"
+
+#include <kunit/static_stub.h>
+#include <linux/device/driver.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/pm_runtime.h>
+
+#include <drm/drm_color_mgmt.h>
+#include <drm/drm_drv.h>
+#include <drm/xe_pciids.h>
+
+#include "regs/xe_gt_regs.h"
+#include "xe_device.h"
+#include "xe_display.h"
+#include "xe_drv.h"
+#include "xe_gt.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+#include "xe_module.h"
+#include "xe_pci_types.h"
+#include "xe_pm.h"
+#include "xe_sriov.h"
+#include "xe_step.h"
+#include "xe_tile.h"
+
+enum toggle_d3cold {
+ D3COLD_DISABLE,
+ D3COLD_ENABLE,
+};
+
+struct xe_subplatform_desc {
+ enum xe_subplatform subplatform;
+ const char *name;
+ const u16 *pciidlist;
+};
+
+struct xe_gt_desc {
+ enum xe_gt_type type;
+ u32 mmio_adj_limit;
+ u32 mmio_adj_offset;
+};
+
+struct xe_device_desc {
+ /* Should only ever be set for platforms without GMD_ID */
+ const struct xe_graphics_desc *graphics;
+ /* Should only ever be set for platforms without GMD_ID */
+ const struct xe_media_desc *media;
+
+ const char *platform_name;
+ const struct xe_subplatform_desc *subplatforms;
+
+ enum xe_platform platform;
+
+ u8 require_force_probe:1;
+ u8 is_dgfx:1;
+
+ u8 has_display:1;
+ u8 has_heci_gscfi:1;
+ u8 has_llc:1;
+ u8 has_mmio_ext:1;
+ u8 has_sriov:1;
+ u8 skip_guc_pc:1;
+ u8 skip_mtcfg:1;
+ u8 skip_pcode:1;
+};
+
+__diag_push();
+__diag_ignore_all("-Woverride-init", "Allow field overrides in table");
+
+#define PLATFORM(x) \
+ .platform = (x), \
+ .platform_name = #x
+
+#define NOP(x) x
+
+static const struct xe_graphics_desc graphics_xelp = {
+ .name = "Xe_LP",
+ .ver = 12,
+ .rel = 0,
+
+ .hw_engine_mask = BIT(XE_HW_ENGINE_RCS0) | BIT(XE_HW_ENGINE_BCS0),
+
+ .dma_mask_size = 39,
+ .va_bits = 48,
+ .vm_max_level = 3,
+};
+
+static const struct xe_graphics_desc graphics_xelpp = {
+ .name = "Xe_LP+",
+ .ver = 12,
+ .rel = 10,
+
+ .hw_engine_mask = BIT(XE_HW_ENGINE_RCS0) | BIT(XE_HW_ENGINE_BCS0),
+
+ .dma_mask_size = 39,
+ .va_bits = 48,
+ .vm_max_level = 3,
+};
+
+#define XE_HP_FEATURES \
+ .has_range_tlb_invalidation = true, \
+ .has_flat_ccs = true, \
+ .dma_mask_size = 46, \
+ .va_bits = 48, \
+ .vm_max_level = 3
+
+static const struct xe_graphics_desc graphics_xehpg = {
+ .name = "Xe_HPG",
+ .ver = 12,
+ .rel = 55,
+
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_RCS0) | BIT(XE_HW_ENGINE_BCS0) |
+ BIT(XE_HW_ENGINE_CCS0) | BIT(XE_HW_ENGINE_CCS1) |
+ BIT(XE_HW_ENGINE_CCS2) | BIT(XE_HW_ENGINE_CCS3),
+
+ XE_HP_FEATURES,
+ .vram_flags = XE_VRAM_FLAGS_NEED64K,
+};
+
+static const struct xe_graphics_desc graphics_xehpc = {
+ .name = "Xe_HPC",
+ .ver = 12,
+ .rel = 60,
+
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_BCS0) | BIT(XE_HW_ENGINE_BCS1) |
+ BIT(XE_HW_ENGINE_BCS2) | BIT(XE_HW_ENGINE_BCS3) |
+ BIT(XE_HW_ENGINE_BCS4) | BIT(XE_HW_ENGINE_BCS5) |
+ BIT(XE_HW_ENGINE_BCS6) | BIT(XE_HW_ENGINE_BCS7) |
+ BIT(XE_HW_ENGINE_BCS8) |
+ BIT(XE_HW_ENGINE_CCS0) | BIT(XE_HW_ENGINE_CCS1) |
+ BIT(XE_HW_ENGINE_CCS2) | BIT(XE_HW_ENGINE_CCS3),
+
+ XE_HP_FEATURES,
+ .dma_mask_size = 52,
+ .max_remote_tiles = 1,
+ .va_bits = 57,
+ .vm_max_level = 4,
+ .vram_flags = XE_VRAM_FLAGS_NEED64K,
+
+ .has_asid = 1,
+ .has_flat_ccs = 0,
+ .has_usm = 1,
+};
+
+static const struct xe_graphics_desc graphics_xelpg = {
+ .name = "Xe_LPG",
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_RCS0) | BIT(XE_HW_ENGINE_BCS0) |
+ BIT(XE_HW_ENGINE_CCS0),
+
+ XE_HP_FEATURES,
+ .has_flat_ccs = 0,
+};
+
+#define XE2_GFX_FEATURES \
+ .dma_mask_size = 46, \
+ .has_asid = 1, \
+ .has_flat_ccs = 1, \
+ .has_range_tlb_invalidation = 1, \
+ .has_usm = 0 /* FIXME: implementation missing */, \
+ .va_bits = 48, \
+ .vm_max_level = 4, \
+ .hw_engine_mask = \
+ BIT(XE_HW_ENGINE_RCS0) | \
+ BIT(XE_HW_ENGINE_BCS8) | BIT(XE_HW_ENGINE_BCS0) | \
+ GENMASK(XE_HW_ENGINE_CCS3, XE_HW_ENGINE_CCS0)
+
+static const struct xe_graphics_desc graphics_xe2 = {
+ .name = "Xe2_LPG",
+
+ XE2_GFX_FEATURES,
+};
+
+static const struct xe_media_desc media_xem = {
+ .name = "Xe_M",
+ .ver = 12,
+ .rel = 0,
+
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_VCS0) | BIT(XE_HW_ENGINE_VCS2) |
+ BIT(XE_HW_ENGINE_VECS0),
+};
+
+static const struct xe_media_desc media_xehpm = {
+ .name = "Xe_HPM",
+ .ver = 12,
+ .rel = 55,
+
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_VCS0) | BIT(XE_HW_ENGINE_VCS2) |
+ BIT(XE_HW_ENGINE_VECS0) | BIT(XE_HW_ENGINE_VECS1),
+};
+
+static const struct xe_media_desc media_xelpmp = {
+ .name = "Xe_LPM+",
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_VCS0) | BIT(XE_HW_ENGINE_VCS2) |
+ BIT(XE_HW_ENGINE_VECS0) | BIT(XE_HW_ENGINE_GSCCS0)
+};
+
+static const struct xe_media_desc media_xe2 = {
+ .name = "Xe2_LPM",
+ .hw_engine_mask =
+ BIT(XE_HW_ENGINE_VCS0) | BIT(XE_HW_ENGINE_VECS0), /* TODO: GSC0 */
+};
+
+static const struct xe_device_desc tgl_desc = {
+ .graphics = &graphics_xelp,
+ .media = &media_xem,
+ PLATFORM(XE_TIGERLAKE),
+ .has_display = true,
+ .has_llc = true,
+ .require_force_probe = true,
+};
+
+static const struct xe_device_desc rkl_desc = {
+ .graphics = &graphics_xelp,
+ .media = &media_xem,
+ PLATFORM(XE_ROCKETLAKE),
+ .has_display = true,
+ .has_llc = true,
+ .require_force_probe = true,
+};
+
+static const u16 adls_rpls_ids[] = { XE_RPLS_IDS(NOP), 0 };
+
+static const struct xe_device_desc adl_s_desc = {
+ .graphics = &graphics_xelp,
+ .media = &media_xem,
+ PLATFORM(XE_ALDERLAKE_S),
+ .has_display = true,
+ .has_llc = true,
+ .require_force_probe = true,
+ .subplatforms = (const struct xe_subplatform_desc[]) {
+ { XE_SUBPLATFORM_ALDERLAKE_S_RPLS, "RPLS", adls_rpls_ids },
+ {},
+ },
+};
+
+static const u16 adlp_rplu_ids[] = { XE_RPLU_IDS(NOP), 0 };
+
+static const struct xe_device_desc adl_p_desc = {
+ .graphics = &graphics_xelp,
+ .media = &media_xem,
+ PLATFORM(XE_ALDERLAKE_P),
+ .has_display = true,
+ .has_llc = true,
+ .require_force_probe = true,
+ .subplatforms = (const struct xe_subplatform_desc[]) {
+ { XE_SUBPLATFORM_ALDERLAKE_P_RPLU, "RPLU", adlp_rplu_ids },
+ {},
+ },
+};
+
+static const struct xe_device_desc adl_n_desc = {
+ .graphics = &graphics_xelp,
+ .media = &media_xem,
+ PLATFORM(XE_ALDERLAKE_N),
+ .has_display = true,
+ .has_llc = true,
+ .require_force_probe = true,
+};
+
+#define DGFX_FEATURES \
+ .is_dgfx = 1
+
+static const struct xe_device_desc dg1_desc = {
+ .graphics = &graphics_xelpp,
+ .media = &media_xem,
+ DGFX_FEATURES,
+ PLATFORM(XE_DG1),
+ .has_display = true,
+ .has_heci_gscfi = 1,
+ .require_force_probe = true,
+};
+
+static const u16 dg2_g10_ids[] = { XE_DG2_G10_IDS(NOP), XE_ATS_M150_IDS(NOP), 0 };
+static const u16 dg2_g11_ids[] = { XE_DG2_G11_IDS(NOP), XE_ATS_M75_IDS(NOP), 0 };
+static const u16 dg2_g12_ids[] = { XE_DG2_G12_IDS(NOP), 0 };
+
+#define DG2_FEATURES \
+ DGFX_FEATURES, \
+ PLATFORM(XE_DG2), \
+ .has_heci_gscfi = 1, \
+ .subplatforms = (const struct xe_subplatform_desc[]) { \
+ { XE_SUBPLATFORM_DG2_G10, "G10", dg2_g10_ids }, \
+ { XE_SUBPLATFORM_DG2_G11, "G11", dg2_g11_ids }, \
+ { XE_SUBPLATFORM_DG2_G12, "G12", dg2_g12_ids }, \
+ { } \
+ }
+
+static const struct xe_device_desc ats_m_desc = {
+ .graphics = &graphics_xehpg,
+ .media = &media_xehpm,
+ .require_force_probe = true,
+
+ DG2_FEATURES,
+ .has_display = false,
+};
+
+static const struct xe_device_desc dg2_desc = {
+ .graphics = &graphics_xehpg,
+ .media = &media_xehpm,
+ .require_force_probe = true,
+
+ DG2_FEATURES,
+ .has_display = true,
+};
+
+static const __maybe_unused struct xe_device_desc pvc_desc = {
+ .graphics = &graphics_xehpc,
+ DGFX_FEATURES,
+ PLATFORM(XE_PVC),
+ .has_display = false,
+ .has_heci_gscfi = 1,
+ .require_force_probe = true,
+};
+
+static const struct xe_device_desc mtl_desc = {
+ /* .graphics and .media determined via GMD_ID */
+ .require_force_probe = true,
+ PLATFORM(XE_METEORLAKE),
+ .has_display = true,
+};
+
+static const struct xe_device_desc lnl_desc = {
+ PLATFORM(XE_LUNARLAKE),
+ .require_force_probe = true,
+};
+
+#undef PLATFORM
+__diag_pop();
+
+/* Map of GMD_ID values to graphics IP */
+static struct gmdid_map graphics_ip_map[] = {
+ { 1270, &graphics_xelpg },
+ { 1271, &graphics_xelpg },
+ { 2004, &graphics_xe2 },
+};
+
+/* Map of GMD_ID values to media IP */
+static struct gmdid_map media_ip_map[] = {
+ { 1300, &media_xelpmp },
+ { 2000, &media_xe2 },
+};
+
+#define INTEL_VGA_DEVICE(id, info) { \
+ PCI_DEVICE(PCI_VENDOR_ID_INTEL, id), \
+ PCI_BASE_CLASS_DISPLAY << 16, 0xff << 16, \
+ (unsigned long) info }
+
+/*
+ * Make sure any device matches here are from most specific to most
+ * general. For example, since the Quanta match is based on the subsystem
+ * and subvendor IDs, we need it to come before the more general IVB
+ * PCI ID matches, otherwise we'll use the wrong info struct above.
+ */
+static const struct pci_device_id pciidlist[] = {
+ XE_TGL_IDS(INTEL_VGA_DEVICE, &tgl_desc),
+ XE_RKL_IDS(INTEL_VGA_DEVICE, &rkl_desc),
+ XE_ADLS_IDS(INTEL_VGA_DEVICE, &adl_s_desc),
+ XE_ADLP_IDS(INTEL_VGA_DEVICE, &adl_p_desc),
+ XE_ADLN_IDS(INTEL_VGA_DEVICE, &adl_n_desc),
+ XE_RPLP_IDS(INTEL_VGA_DEVICE, &adl_p_desc),
+ XE_RPLS_IDS(INTEL_VGA_DEVICE, &adl_s_desc),
+ XE_DG1_IDS(INTEL_VGA_DEVICE, &dg1_desc),
+ XE_ATS_M_IDS(INTEL_VGA_DEVICE, &ats_m_desc),
+ XE_DG2_IDS(INTEL_VGA_DEVICE, &dg2_desc),
+ XE_MTL_IDS(INTEL_VGA_DEVICE, &mtl_desc),
+ XE_LNL_IDS(INTEL_VGA_DEVICE, &lnl_desc),
+ { }
+};
+MODULE_DEVICE_TABLE(pci, pciidlist);
+
+#undef INTEL_VGA_DEVICE
+
+/* is device_id present in comma separated list of ids */
+static bool device_id_in_list(u16 device_id, const char *devices, bool negative)
+{
+ char *s, *p, *tok;
+ bool ret;
+
+ if (!devices || !*devices)
+ return false;
+
+ /* match everything */
+ if (negative && strcmp(devices, "!*") == 0)
+ return true;
+ if (!negative && strcmp(devices, "*") == 0)
+ return true;
+
+ s = kstrdup(devices, GFP_KERNEL);
+ if (!s)
+ return false;
+
+ for (p = s, ret = false; (tok = strsep(&p, ",")) != NULL; ) {
+ u16 val;
+
+ if (negative && tok[0] == '!')
+ tok++;
+ else if ((negative && tok[0] != '!') ||
+ (!negative && tok[0] == '!'))
+ continue;
+
+ if (kstrtou16(tok, 16, &val) == 0 && val == device_id) {
+ ret = true;
+ break;
+ }
+ }
+
+ kfree(s);
+
+ return ret;
+}
+
+static bool id_forced(u16 device_id)
+{
+ return device_id_in_list(device_id, xe_modparam.force_probe, false);
+}
+
+static bool id_blocked(u16 device_id)
+{
+ return device_id_in_list(device_id, xe_modparam.force_probe, true);
+}
+
+static const struct xe_subplatform_desc *
+find_subplatform(const struct xe_device *xe, const struct xe_device_desc *desc)
+{
+ const struct xe_subplatform_desc *sp;
+ const u16 *id;
+
+ for (sp = desc->subplatforms; sp && sp->subplatform; sp++)
+ for (id = sp->pciidlist; *id; id++)
+ if (*id == xe->info.devid)
+ return sp;
+
+ return NULL;
+}
+
+enum xe_gmdid_type {
+ GMDID_GRAPHICS,
+ GMDID_MEDIA
+};
+
+static void read_gmdid(struct xe_device *xe, enum xe_gmdid_type type, u32 *ver, u32 *revid)
+{
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+ struct xe_reg gmdid_reg = GMD_ID;
+ u32 val;
+
+ KUNIT_STATIC_STUB_REDIRECT(read_gmdid, xe, type, ver, revid);
+
+ if (type == GMDID_MEDIA)
+ gmdid_reg.addr += MEDIA_GT_GSI_OFFSET;
+
+ val = xe_mmio_read32(gt, gmdid_reg);
+ *ver = REG_FIELD_GET(GMD_ID_ARCH_MASK, val) * 100 + REG_FIELD_GET(GMD_ID_RELEASE_MASK, val);
+ *revid = REG_FIELD_GET(GMD_ID_REVID, val);
+}
+
+/*
+ * Pre-GMD_ID platform: device descriptor already points to the appropriate
+ * graphics descriptor. Simply forward the description and calculate the version
+ * appropriately. "graphics" should be present in all such platforms, while
+ * media is optional.
+ */
+static void handle_pre_gmdid(struct xe_device *xe,
+ const struct xe_graphics_desc *graphics,
+ const struct xe_media_desc *media)
+{
+ xe->info.graphics_verx100 = graphics->ver * 100 + graphics->rel;
+
+ if (media)
+ xe->info.media_verx100 = media->ver * 100 + media->rel;
+
+}
+
+/*
+ * GMD_ID platform: read IP version from hardware and select graphics descriptor
+ * based on the result.
+ */
+static void handle_gmdid(struct xe_device *xe,
+ const struct xe_graphics_desc **graphics,
+ const struct xe_media_desc **media,
+ u32 *graphics_revid,
+ u32 *media_revid)
+{
+ u32 ver;
+
+ read_gmdid(xe, GMDID_GRAPHICS, &ver, graphics_revid);
+
+ for (int i = 0; i < ARRAY_SIZE(graphics_ip_map); i++) {
+ if (ver == graphics_ip_map[i].ver) {
+ xe->info.graphics_verx100 = ver;
+ *graphics = graphics_ip_map[i].ip;
+
+ break;
+ }
+ }
+
+ if (!xe->info.graphics_verx100) {
+ drm_err(&xe->drm, "Hardware reports unknown graphics version %u.%02u\n",
+ ver / 100, ver % 100);
+ }
+
+ read_gmdid(xe, GMDID_MEDIA, &ver, media_revid);
+
+ /* Media may legitimately be fused off / not present */
+ if (ver == 0)
+ return;
+
+ for (int i = 0; i < ARRAY_SIZE(media_ip_map); i++) {
+ if (ver == media_ip_map[i].ver) {
+ xe->info.media_verx100 = ver;
+ *media = media_ip_map[i].ip;
+
+ break;
+ }
+ }
+
+ if (!xe->info.media_verx100) {
+ drm_err(&xe->drm, "Hardware reports unknown media version %u.%02u\n",
+ ver / 100, ver % 100);
+ }
+}
+
+/*
+ * Initialize device info content that only depends on static driver_data
+ * passed to the driver at probe time from PCI ID table.
+ */
+static int xe_info_init_early(struct xe_device *xe,
+ const struct xe_device_desc *desc,
+ const struct xe_subplatform_desc *subplatform_desc)
+{
+ int err;
+
+ xe->info.platform = desc->platform;
+ xe->info.subplatform = subplatform_desc ?
+ subplatform_desc->subplatform : XE_SUBPLATFORM_NONE;
+
+ xe->info.is_dgfx = desc->is_dgfx;
+ xe->info.has_heci_gscfi = desc->has_heci_gscfi;
+ xe->info.has_llc = desc->has_llc;
+ xe->info.has_mmio_ext = desc->has_mmio_ext;
+ xe->info.has_sriov = desc->has_sriov;
+ xe->info.skip_guc_pc = desc->skip_guc_pc;
+ xe->info.skip_mtcfg = desc->skip_mtcfg;
+ xe->info.skip_pcode = desc->skip_pcode;
+
+ xe->info.enable_display = IS_ENABLED(CONFIG_DRM_XE_DISPLAY) &&
+ xe_modparam.enable_display &&
+ desc->has_display;
+
+ err = xe_tile_init_early(xe_device_get_root_tile(xe), xe, 0);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+/*
+ * Initialize device info content that does require knowledge about
+ * graphics / media IP version.
+ * Make sure that GT / tile structures allocated by the driver match the data
+ * present in device info.
+ */
+static int xe_info_init(struct xe_device *xe,
+ const struct xe_graphics_desc *graphics_desc,
+ const struct xe_media_desc *media_desc)
+{
+ u32 graphics_gmdid_revid = 0, media_gmdid_revid = 0;
+ struct xe_tile *tile;
+ struct xe_gt *gt;
+ u8 id;
+
+ /*
+ * If this platform supports GMD_ID, we'll detect the proper IP
+ * descriptor to use from hardware registers. desc->graphics will only
+ * ever be set at this point for platforms before GMD_ID. In that case
+ * the IP descriptions and versions are simply derived from that.
+ */
+ if (graphics_desc) {
+ handle_pre_gmdid(xe, graphics_desc, media_desc);
+ xe->info.step = xe_step_pre_gmdid_get(xe);
+ } else {
+ xe_assert(xe, !media_desc);
+ handle_gmdid(xe, &graphics_desc, &media_desc,
+ &graphics_gmdid_revid, &media_gmdid_revid);
+ xe->info.step = xe_step_gmdid_get(xe,
+ graphics_gmdid_revid,
+ media_gmdid_revid);
+ }
+
+ /*
+ * If we couldn't detect the graphics IP, that's considered a fatal
+ * error and we should abort driver load. Failing to detect media
+ * IP is non-fatal; we'll just proceed without enabling media support.
+ */
+ if (!graphics_desc)
+ return -ENODEV;
+
+ xe->info.graphics_name = graphics_desc->name;
+ xe->info.media_name = media_desc ? media_desc->name : "none";
+ xe->info.tile_mmio_ext_size = graphics_desc->tile_mmio_ext_size;
+
+ xe->info.dma_mask_size = graphics_desc->dma_mask_size;
+ xe->info.vram_flags = graphics_desc->vram_flags;
+ xe->info.va_bits = graphics_desc->va_bits;
+ xe->info.vm_max_level = graphics_desc->vm_max_level;
+ xe->info.has_asid = graphics_desc->has_asid;
+ xe->info.has_flat_ccs = graphics_desc->has_flat_ccs;
+ xe->info.has_range_tlb_invalidation = graphics_desc->has_range_tlb_invalidation;
+ xe->info.has_usm = graphics_desc->has_usm;
+
+ /*
+ * All platforms have at least one primary GT. Any platform with media
+ * version 13 or higher has an additional dedicated media GT. And
+ * depending on the graphics IP there may be additional "remote tiles."
+ * All of these together determine the overall GT count.
+ *
+ * FIXME: 'tile_count' here is misnamed since the rest of the driver
+ * treats it as the number of GTs rather than just the number of tiles.
+ */
+ xe->info.tile_count = 1 + graphics_desc->max_remote_tiles;
+
+ for_each_remote_tile(tile, xe, id) {
+ int err;
+
+ err = xe_tile_init_early(tile, xe, id);
+ if (err)
+ return err;
+ }
+
+ for_each_tile(tile, xe, id) {
+ gt = tile->primary_gt;
+ gt->info.id = xe->info.gt_count++;
+ gt->info.type = XE_GT_TYPE_MAIN;
+ gt->info.__engine_mask = graphics_desc->hw_engine_mask;
+ if (MEDIA_VER(xe) < 13 && media_desc)
+ gt->info.__engine_mask |= media_desc->hw_engine_mask;
+
+ if (MEDIA_VER(xe) < 13 || !media_desc)
+ continue;
+
+ /*
+ * Allocate and setup media GT for platforms with standalone
+ * media.
+ */
+ tile->media_gt = xe_gt_alloc(tile);
+ if (IS_ERR(tile->media_gt))
+ return PTR_ERR(tile->media_gt);
+
+ gt = tile->media_gt;
+ gt->info.type = XE_GT_TYPE_MEDIA;
+ gt->info.__engine_mask = media_desc->hw_engine_mask;
+ gt->mmio.adj_offset = MEDIA_GT_GSI_OFFSET;
+ gt->mmio.adj_limit = MEDIA_GT_GSI_LENGTH;
+
+ /*
+ * FIXME: At the moment multi-tile and standalone media are
+ * mutually exclusive on current platforms. We'll need to
+ * come up with a better way to number GTs if we ever wind
+ * up with platforms that support both together.
+ */
+ drm_WARN_ON(&xe->drm, id != 0);
+ gt->info.id = xe->info.gt_count++;
+ }
+
+ return 0;
+}
+
+static void xe_pci_remove(struct pci_dev *pdev)
+{
+ struct xe_device *xe;
+
+ xe = pci_get_drvdata(pdev);
+ if (!xe) /* driver load aborted, nothing to cleanup */
+ return;
+
+ xe_device_remove(xe);
+ xe_pm_runtime_fini(xe);
+ pci_set_drvdata(pdev, NULL);
+}
+
+static int xe_pci_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+ const struct xe_device_desc *desc = (const void *)ent->driver_data;
+ const struct xe_subplatform_desc *subplatform_desc;
+ struct xe_device *xe;
+ int err;
+
+ if (desc->require_force_probe && !id_forced(pdev->device)) {
+ dev_info(&pdev->dev,
+ "Your graphics device %04x is not officially supported\n"
+ "by xe driver in this kernel version. To force Xe probe,\n"
+ "use xe.force_probe='%04x' and i915.force_probe='!%04x'\n"
+ "module parameters or CONFIG_DRM_XE_FORCE_PROBE='%04x' and\n"
+ "CONFIG_DRM_I915_FORCE_PROBE='!%04x' configuration options.\n",
+ pdev->device, pdev->device, pdev->device,
+ pdev->device, pdev->device);
+ return -ENODEV;
+ }
+
+ if (id_blocked(pdev->device)) {
+ dev_info(&pdev->dev, "Probe blocked for device [%04x:%04x].\n",
+ pdev->vendor, pdev->device);
+ return -ENODEV;
+ }
+
+ if (xe_display_driver_probe_defer(pdev))
+ return -EPROBE_DEFER;
+
+ err = pcim_enable_device(pdev);
+ if (err)
+ return err;
+
+ xe = xe_device_create(pdev, ent);
+ if (IS_ERR(xe))
+ return PTR_ERR(xe);
+
+ pci_set_drvdata(pdev, xe);
+
+ xe_pm_assert_unbounded_bridge(xe);
+ subplatform_desc = find_subplatform(xe, desc);
+
+ pci_set_master(pdev);
+
+ err = xe_info_init_early(xe, desc, subplatform_desc);
+ if (err)
+ return err;
+
+ xe_sriov_probe_early(xe, desc->has_sriov);
+
+ err = xe_device_probe_early(xe);
+ if (err)
+ return err;
+
+ err = xe_info_init(xe, desc->graphics, desc->media);
+ if (err)
+ return err;
+
+ xe_display_probe(xe);
+
+ drm_dbg(&xe->drm, "%s %s %04x:%04x dgfx:%d gfx:%s (%d.%02d) media:%s (%d.%02d) display:%s dma_m_s:%d tc:%d gscfi:%d",
+ desc->platform_name,
+ subplatform_desc ? subplatform_desc->name : "",
+ xe->info.devid, xe->info.revid,
+ xe->info.is_dgfx,
+ xe->info.graphics_name,
+ xe->info.graphics_verx100 / 100,
+ xe->info.graphics_verx100 % 100,
+ xe->info.media_name,
+ xe->info.media_verx100 / 100,
+ xe->info.media_verx100 % 100,
+ str_yes_no(xe->info.enable_display),
+ xe->info.dma_mask_size, xe->info.tile_count,
+ xe->info.has_heci_gscfi);
+
+ drm_dbg(&xe->drm, "Stepping = (G:%s, M:%s, D:%s, B:%s)\n",
+ xe_step_name(xe->info.step.graphics),
+ xe_step_name(xe->info.step.media),
+ xe_step_name(xe->info.step.display),
+ xe_step_name(xe->info.step.basedie));
+
+ drm_dbg(&xe->drm, "SR-IOV support: %s (mode: %s)\n",
+ str_yes_no(xe_device_has_sriov(xe)),
+ xe_sriov_mode_to_string(xe_device_sriov_mode(xe)));
+
+ err = xe_device_probe(xe);
+ if (err)
+ return err;
+
+ xe_pm_init(xe);
+
+ drm_dbg(&xe->drm, "d3cold: capable=%s\n",
+ str_yes_no(xe->d3cold.capable));
+
+ return 0;
+}
+
+static void xe_pci_shutdown(struct pci_dev *pdev)
+{
+ xe_device_shutdown(pdev_to_xe_device(pdev));
+}
+
+#ifdef CONFIG_PM_SLEEP
+static void d3cold_toggle(struct pci_dev *pdev, enum toggle_d3cold toggle)
+{
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+ struct pci_dev *root_pdev;
+
+ if (!xe->d3cold.capable)
+ return;
+
+ root_pdev = pcie_find_root_port(pdev);
+ if (!root_pdev)
+ return;
+
+ switch (toggle) {
+ case D3COLD_DISABLE:
+ pci_d3cold_disable(root_pdev);
+ break;
+ case D3COLD_ENABLE:
+ pci_d3cold_enable(root_pdev);
+ break;
+ }
+}
+
+static int xe_pci_suspend(struct device *dev)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ int err;
+
+ err = xe_pm_suspend(pdev_to_xe_device(pdev));
+ if (err)
+ return err;
+
+ /*
+ * Enabling D3Cold is needed for S2Idle/S0ix.
+ * It is save to allow here since xe_pm_suspend has evicted
+ * the local memory and the direct complete optimization is disabled.
+ */
+ d3cold_toggle(pdev, D3COLD_ENABLE);
+
+ pci_save_state(pdev);
+ pci_disable_device(pdev);
+
+ return 0;
+}
+
+static int xe_pci_resume(struct device *dev)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ int err;
+
+ /* Give back the D3Cold decision to the runtime P M*/
+ d3cold_toggle(pdev, D3COLD_DISABLE);
+
+ err = pci_set_power_state(pdev, PCI_D0);
+ if (err)
+ return err;
+
+ err = pci_enable_device(pdev);
+ if (err)
+ return err;
+
+ pci_set_master(pdev);
+
+ err = xe_pm_resume(pdev_to_xe_device(pdev));
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static int xe_pci_runtime_suspend(struct device *dev)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+ int err;
+
+ err = xe_pm_runtime_suspend(xe);
+ if (err)
+ return err;
+
+ pci_save_state(pdev);
+
+ if (xe->d3cold.allowed) {
+ d3cold_toggle(pdev, D3COLD_ENABLE);
+ pci_disable_device(pdev);
+ pci_ignore_hotplug(pdev);
+ pci_set_power_state(pdev, PCI_D3cold);
+ } else {
+ d3cold_toggle(pdev, D3COLD_DISABLE);
+ pci_set_power_state(pdev, PCI_D3hot);
+ }
+
+ return 0;
+}
+
+static int xe_pci_runtime_resume(struct device *dev)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+ int err;
+
+ err = pci_set_power_state(pdev, PCI_D0);
+ if (err)
+ return err;
+
+ pci_restore_state(pdev);
+
+ if (xe->d3cold.allowed) {
+ err = pci_enable_device(pdev);
+ if (err)
+ return err;
+
+ pci_set_master(pdev);
+ }
+
+ return xe_pm_runtime_resume(xe);
+}
+
+static int xe_pci_runtime_idle(struct device *dev)
+{
+ struct pci_dev *pdev = to_pci_dev(dev);
+ struct xe_device *xe = pdev_to_xe_device(pdev);
+
+ xe_pm_d3cold_allowed_toggle(xe);
+
+ return 0;
+}
+
+static const struct dev_pm_ops xe_pm_ops = {
+ SET_SYSTEM_SLEEP_PM_OPS(xe_pci_suspend, xe_pci_resume)
+ SET_RUNTIME_PM_OPS(xe_pci_runtime_suspend, xe_pci_runtime_resume, xe_pci_runtime_idle)
+};
+#endif
+
+static struct pci_driver xe_pci_driver = {
+ .name = DRIVER_NAME,
+ .id_table = pciidlist,
+ .probe = xe_pci_probe,
+ .remove = xe_pci_remove,
+ .shutdown = xe_pci_shutdown,
+#ifdef CONFIG_PM_SLEEP
+ .driver.pm = &xe_pm_ops,
+#endif
+};
+
+int xe_register_pci_driver(void)
+{
+ return pci_register_driver(&xe_pci_driver);
+}
+
+void xe_unregister_pci_driver(void)
+{
+ pci_unregister_driver(&xe_pci_driver);
+}
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+#include "tests/xe_pci.c"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pci.h b/drivers/gpu/drm/xe/xe_pci.h
new file mode 100644
index 000000000000..611c1209b14c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pci.h
@@ -0,0 +1,12 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_PCI_H_
+#define _XE_PCI_H_
+
+int xe_register_pci_driver(void);
+void xe_unregister_pci_driver(void);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pci_types.h b/drivers/gpu/drm/xe/xe_pci_types.h
new file mode 100644
index 000000000000..b1ad12fa22d6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pci_types.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_PCI_TYPES_H_
+#define _XE_PCI_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_graphics_desc {
+ const char *name;
+ u8 ver;
+ u8 rel;
+
+ u8 dma_mask_size; /* available DMA address bits */
+ u8 va_bits;
+ u8 vm_max_level;
+ u8 vram_flags;
+
+ u64 hw_engine_mask; /* hardware engines provided by graphics IP */
+
+ u32 tile_mmio_ext_size; /* size of MMIO extension space, per-tile */
+
+ u8 max_remote_tiles:2;
+
+ u8 has_asid:1;
+ u8 has_flat_ccs:1;
+ u8 has_range_tlb_invalidation:1;
+ u8 has_usm:1;
+};
+
+struct xe_media_desc {
+ const char *name;
+ u8 ver;
+ u8 rel;
+
+ u64 hw_engine_mask; /* hardware engines provided by media IP */
+};
+
+struct gmdid_map {
+ unsigned int ver;
+ const void *ip;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pcode.c b/drivers/gpu/drm/xe/xe_pcode.c
new file mode 100644
index 000000000000..b324dc2a5deb
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pcode.c
@@ -0,0 +1,296 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_pcode.h"
+
+#include <linux/delay.h>
+#include <linux/errno.h>
+
+#include <drm/drm_managed.h>
+
+#include "xe_gt.h"
+#include "xe_mmio.h"
+#include "xe_pcode_api.h"
+
+/**
+ * DOC: PCODE
+ *
+ * Xe PCODE is the component responsible for interfacing with the PCODE
+ * firmware.
+ * It shall provide a very simple ABI to other Xe components, but be the
+ * single and consolidated place that will communicate with PCODE. All read
+ * and write operations to PCODE will be internal and private to this component.
+ *
+ * What's next:
+ * - PCODE hw metrics
+ * - PCODE for display operations
+ */
+
+static int pcode_mailbox_status(struct xe_gt *gt)
+{
+ u32 err;
+ static const struct pcode_err_decode err_decode[] = {
+ [PCODE_ILLEGAL_CMD] = {-ENXIO, "Illegal Command"},
+ [PCODE_TIMEOUT] = {-ETIMEDOUT, "Timed out"},
+ [PCODE_ILLEGAL_DATA] = {-EINVAL, "Illegal Data"},
+ [PCODE_ILLEGAL_SUBCOMMAND] = {-ENXIO, "Illegal Subcommand"},
+ [PCODE_LOCKED] = {-EBUSY, "PCODE Locked"},
+ [PCODE_GT_RATIO_OUT_OF_RANGE] = {-EOVERFLOW,
+ "GT ratio out of range"},
+ [PCODE_REJECTED] = {-EACCES, "PCODE Rejected"},
+ [PCODE_ERROR_MASK] = {-EPROTO, "Unknown"},
+ };
+
+ lockdep_assert_held(&gt->pcode.lock);
+
+ err = xe_mmio_read32(gt, PCODE_MAILBOX) & PCODE_ERROR_MASK;
+ if (err) {
+ drm_err(&gt_to_xe(gt)->drm, "PCODE Mailbox failed: %d %s", err,
+ err_decode[err].str ?: "Unknown");
+ return err_decode[err].errno ?: -EPROTO;
+ }
+
+ return 0;
+}
+
+static int pcode_mailbox_rw(struct xe_gt *gt, u32 mbox, u32 *data0, u32 *data1,
+ unsigned int timeout_ms, bool return_data,
+ bool atomic)
+{
+ int err;
+
+ if (gt_to_xe(gt)->info.skip_pcode)
+ return 0;
+
+ lockdep_assert_held(&gt->pcode.lock);
+
+ if ((xe_mmio_read32(gt, PCODE_MAILBOX) & PCODE_READY) != 0)
+ return -EAGAIN;
+
+ xe_mmio_write32(gt, PCODE_DATA0, *data0);
+ xe_mmio_write32(gt, PCODE_DATA1, data1 ? *data1 : 0);
+ xe_mmio_write32(gt, PCODE_MAILBOX, PCODE_READY | mbox);
+
+ err = xe_mmio_wait32(gt, PCODE_MAILBOX, PCODE_READY, 0,
+ timeout_ms * 1000, NULL, atomic);
+ if (err)
+ return err;
+
+ if (return_data) {
+ *data0 = xe_mmio_read32(gt, PCODE_DATA0);
+ if (data1)
+ *data1 = xe_mmio_read32(gt, PCODE_DATA1);
+ }
+
+ return pcode_mailbox_status(gt);
+}
+
+int xe_pcode_write_timeout(struct xe_gt *gt, u32 mbox, u32 data, int timeout)
+{
+ int err;
+
+ mutex_lock(&gt->pcode.lock);
+ err = pcode_mailbox_rw(gt, mbox, &data, NULL, timeout, false, false);
+ mutex_unlock(&gt->pcode.lock);
+
+ return err;
+}
+
+int xe_pcode_read(struct xe_gt *gt, u32 mbox, u32 *val, u32 *val1)
+{
+ int err;
+
+ mutex_lock(&gt->pcode.lock);
+ err = pcode_mailbox_rw(gt, mbox, val, val1, 1, true, false);
+ mutex_unlock(&gt->pcode.lock);
+
+ return err;
+}
+
+static int xe_pcode_try_request(struct xe_gt *gt, u32 mbox,
+ u32 request, u32 reply_mask, u32 reply,
+ u32 *status, bool atomic, int timeout_us)
+{
+ int slept, wait = 10;
+
+ for (slept = 0; slept < timeout_us; slept += wait) {
+ *status = pcode_mailbox_rw(gt, mbox, &request, NULL, 1, true,
+ atomic);
+ if ((*status == 0) && ((request & reply_mask) == reply))
+ return 0;
+
+ if (atomic)
+ udelay(wait);
+ else
+ usleep_range(wait, wait << 1);
+ wait <<= 1;
+ }
+
+ return -ETIMEDOUT;
+}
+
+/**
+ * xe_pcode_request - send PCODE request until acknowledgment
+ * @gt: gt
+ * @mbox: PCODE mailbox ID the request is targeted for
+ * @request: request ID
+ * @reply_mask: mask used to check for request acknowledgment
+ * @reply: value used to check for request acknowledgment
+ * @timeout_base_ms: timeout for polling with preemption enabled
+ *
+ * Keep resending the @request to @mbox until PCODE acknowledges it, PCODE
+ * reports an error or an overall timeout of @timeout_base_ms+50 ms expires.
+ * The request is acknowledged once the PCODE reply dword equals @reply after
+ * applying @reply_mask. Polling is first attempted with preemption enabled
+ * for @timeout_base_ms and if this times out for another 50 ms with
+ * preemption disabled.
+ *
+ * Returns 0 on success, %-ETIMEDOUT in case of a timeout, <0 in case of some
+ * other error as reported by PCODE.
+ */
+int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
+ u32 reply_mask, u32 reply, int timeout_base_ms)
+{
+ u32 status;
+ int ret;
+
+ mutex_lock(&gt->pcode.lock);
+
+ ret = xe_pcode_try_request(gt, mbox, request, reply_mask, reply, &status,
+ false, timeout_base_ms * 1000);
+ if (!ret)
+ goto out;
+
+ /*
+ * The above can time out if the number of requests was low (2 in the
+ * worst case) _and_ PCODE was busy for some reason even after a
+ * (queued) request and @timeout_base_ms delay. As a workaround retry
+ * the poll with preemption disabled to maximize the number of
+ * requests. Increase the timeout from @timeout_base_ms to 50ms to
+ * account for interrupts that could reduce the number of these
+ * requests, and for any quirks of the PCODE firmware that delays
+ * the request completion.
+ */
+ drm_err(&gt_to_xe(gt)->drm,
+ "PCODE timeout, retrying with preemption disabled\n");
+ drm_WARN_ON_ONCE(&gt_to_xe(gt)->drm, timeout_base_ms > 1);
+ preempt_disable();
+ ret = xe_pcode_try_request(gt, mbox, request, reply_mask, reply, &status,
+ true, timeout_base_ms * 1000);
+ preempt_enable();
+
+out:
+ mutex_unlock(&gt->pcode.lock);
+ return status ? status : ret;
+}
+/**
+ * xe_pcode_init_min_freq_table - Initialize PCODE's QOS frequency table
+ * @gt: gt instance
+ * @min_gt_freq: Minimal (RPn) GT frequency in units of 50MHz.
+ * @max_gt_freq: Maximal (RP0) GT frequency in units of 50MHz.
+ *
+ * This function initialize PCODE's QOS frequency table for a proper minimal
+ * frequency/power steering decision, depending on the current requested GT
+ * frequency. For older platforms this was a more complete table including
+ * the IA freq. However for the latest platforms this table become a simple
+ * 1-1 Ring vs GT frequency. Even though, without setting it, PCODE might
+ * not take the right decisions for some memory frequencies and affect latency.
+ *
+ * It returns 0 on success, and -ERROR number on failure, -EINVAL if max
+ * frequency is higher then the minimal, and other errors directly translated
+ * from the PCODE Error returs:
+ * - -ENXIO: "Illegal Command"
+ * - -ETIMEDOUT: "Timed out"
+ * - -EINVAL: "Illegal Data"
+ * - -ENXIO, "Illegal Subcommand"
+ * - -EBUSY: "PCODE Locked"
+ * - -EOVERFLOW, "GT ratio out of range"
+ * - -EACCES, "PCODE Rejected"
+ * - -EPROTO, "Unknown"
+ */
+int xe_pcode_init_min_freq_table(struct xe_gt *gt, u32 min_gt_freq,
+ u32 max_gt_freq)
+{
+ int ret;
+ u32 freq;
+
+ if (!gt_to_xe(gt)->info.has_llc)
+ return 0;
+
+ if (max_gt_freq <= min_gt_freq)
+ return -EINVAL;
+
+ mutex_lock(&gt->pcode.lock);
+ for (freq = min_gt_freq; freq <= max_gt_freq; freq++) {
+ u32 data = freq << PCODE_FREQ_RING_RATIO_SHIFT | freq;
+
+ ret = pcode_mailbox_rw(gt, PCODE_WRITE_MIN_FREQ_TABLE,
+ &data, NULL, 1, false, false);
+ if (ret)
+ goto unlock;
+ }
+
+unlock:
+ mutex_unlock(&gt->pcode.lock);
+ return ret;
+}
+
+/**
+ * xe_pcode_init - Ensure PCODE is initialized
+ * @gt: gt instance
+ *
+ * This function ensures that PCODE is properly initialized. To be called during
+ * probe and resume paths.
+ *
+ * It returns 0 on success, and -error number on failure.
+ */
+int xe_pcode_init(struct xe_gt *gt)
+{
+ u32 status, request = DGFX_GET_INIT_STATUS;
+ int timeout_us = 180000000; /* 3 min */
+ int ret;
+
+ if (gt_to_xe(gt)->info.skip_pcode)
+ return 0;
+
+ if (!IS_DGFX(gt_to_xe(gt)))
+ return 0;
+
+ mutex_lock(&gt->pcode.lock);
+ ret = xe_pcode_try_request(gt, DGFX_PCODE_STATUS, request,
+ DGFX_INIT_STATUS_COMPLETE,
+ DGFX_INIT_STATUS_COMPLETE,
+ &status, false, timeout_us);
+ mutex_unlock(&gt->pcode.lock);
+
+ if (ret)
+ drm_err(&gt_to_xe(gt)->drm,
+ "PCODE initialization timedout after: 3 min\n");
+
+ return ret;
+}
+
+/**
+ * xe_pcode_probe - Prepare xe_pcode and also ensure PCODE is initialized.
+ * @gt: gt instance
+ *
+ * This function initializes the xe_pcode component, and when needed, it ensures
+ * that PCODE has properly performed its initialization and it is really ready
+ * to go. To be called once only during probe.
+ *
+ * It returns 0 on success, and -error number on failure.
+ */
+int xe_pcode_probe(struct xe_gt *gt)
+{
+ drmm_mutex_init(&gt_to_xe(gt)->drm, &gt->pcode.lock);
+
+ if (gt_to_xe(gt)->info.skip_pcode)
+ return 0;
+
+ if (!IS_DGFX(gt_to_xe(gt)))
+ return 0;
+
+ return xe_pcode_init(gt);
+}
diff --git a/drivers/gpu/drm/xe/xe_pcode.h b/drivers/gpu/drm/xe/xe_pcode.h
new file mode 100644
index 000000000000..08cb1d047cba
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pcode.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PCODE_H_
+#define _XE_PCODE_H_
+
+#include <linux/types.h>
+struct xe_gt;
+
+int xe_pcode_probe(struct xe_gt *gt);
+int xe_pcode_init(struct xe_gt *gt);
+int xe_pcode_init_min_freq_table(struct xe_gt *gt, u32 min_gt_freq,
+ u32 max_gt_freq);
+int xe_pcode_read(struct xe_gt *gt, u32 mbox, u32 *val, u32 *val1);
+int xe_pcode_write_timeout(struct xe_gt *gt, u32 mbox, u32 val,
+ int timeout_ms);
+#define xe_pcode_write(gt, mbox, val) \
+ xe_pcode_write_timeout(gt, mbox, val, 1)
+
+int xe_pcode_request(struct xe_gt *gt, u32 mbox, u32 request,
+ u32 reply_mask, u32 reply, int timeout_ms);
+
+#define PCODE_MBOX(mbcmd, param1, param2)\
+ (FIELD_PREP(PCODE_MB_COMMAND, mbcmd)\
+ | FIELD_PREP(PCODE_MB_PARAM1, param1)\
+ | FIELD_PREP(PCODE_MB_PARAM2, param2))
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pcode_api.h b/drivers/gpu/drm/xe/xe_pcode_api.h
new file mode 100644
index 000000000000..5935cfe30204
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pcode_api.h
@@ -0,0 +1,49 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+/* Internal to xe_pcode */
+
+#include "regs/xe_reg_defs.h"
+
+#define PCODE_MAILBOX XE_REG(0x138124)
+#define PCODE_READY REG_BIT(31)
+#define PCODE_MB_PARAM2 REG_GENMASK(23, 16)
+#define PCODE_MB_PARAM1 REG_GENMASK(15, 8)
+#define PCODE_MB_COMMAND REG_GENMASK(7, 0)
+#define PCODE_ERROR_MASK 0xFF
+#define PCODE_SUCCESS 0x0
+#define PCODE_ILLEGAL_CMD 0x1
+#define PCODE_TIMEOUT 0x2
+#define PCODE_ILLEGAL_DATA 0x3
+#define PCODE_ILLEGAL_SUBCOMMAND 0x4
+#define PCODE_LOCKED 0x6
+#define PCODE_GT_RATIO_OUT_OF_RANGE 0x10
+#define PCODE_REJECTED 0x11
+
+#define PCODE_DATA0 XE_REG(0x138128)
+#define PCODE_DATA1 XE_REG(0x13812C)
+
+/* Min Freq QOS Table */
+#define PCODE_WRITE_MIN_FREQ_TABLE 0x8
+#define PCODE_READ_MIN_FREQ_TABLE 0x9
+#define PCODE_FREQ_RING_RATIO_SHIFT 16
+
+/* PCODE Init */
+#define DGFX_PCODE_STATUS 0x7E
+#define DGFX_GET_INIT_STATUS 0x0
+#define DGFX_INIT_STATUS_COMPLETE 0x1
+
+#define PCODE_POWER_SETUP 0x7C
+#define POWER_SETUP_SUBCOMMAND_READ_I1 0x4
+#define POWER_SETUP_SUBCOMMAND_WRITE_I1 0x5
+#define POWER_SETUP_I1_WATTS REG_BIT(31)
+#define POWER_SETUP_I1_SHIFT 6 /* 10.6 fixed point format */
+#define POWER_SETUP_I1_DATA_MASK REG_GENMASK(15, 0)
+
+struct pcode_err_decode {
+ int errno;
+ const char *str;
+};
+
diff --git a/drivers/gpu/drm/xe/xe_platform_types.h b/drivers/gpu/drm/xe/xe_platform_types.h
new file mode 100644
index 000000000000..553f53dbd093
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_platform_types.h
@@ -0,0 +1,37 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PLATFORM_INFO_TYPES_H_
+#define _XE_PLATFORM_INFO_TYPES_H_
+
+/*
+ * Keep this in graphics version based order and chronological order within a
+ * version
+ */
+enum xe_platform {
+ XE_PLATFORM_UNINITIALIZED = 0,
+ XE_TIGERLAKE,
+ XE_ROCKETLAKE,
+ XE_ALDERLAKE_S,
+ XE_ALDERLAKE_P,
+ XE_ALDERLAKE_N,
+ XE_DG1,
+ XE_DG2,
+ XE_PVC,
+ XE_METEORLAKE,
+ XE_LUNARLAKE,
+};
+
+enum xe_subplatform {
+ XE_SUBPLATFORM_UNINITIALIZED = 0,
+ XE_SUBPLATFORM_NONE,
+ XE_SUBPLATFORM_ALDERLAKE_P_RPLU,
+ XE_SUBPLATFORM_ALDERLAKE_S_RPLS,
+ XE_SUBPLATFORM_DG2_G10,
+ XE_SUBPLATFORM_DG2_G11,
+ XE_SUBPLATFORM_DG2_G12,
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pm.c b/drivers/gpu/drm/xe/xe_pm.c
new file mode 100644
index 000000000000..b429c2876a76
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pm.c
@@ -0,0 +1,405 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_pm.h"
+
+#include <linux/pm_runtime.h>
+
+#include <drm/drm_managed.h>
+#include <drm/ttm/ttm_placement.h>
+
+#include "xe_bo.h"
+#include "xe_bo_evict.h"
+#include "xe_device.h"
+#include "xe_device_sysfs.h"
+#include "xe_display.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_irq.h"
+#include "xe_pcode.h"
+#include "xe_wa.h"
+
+/**
+ * DOC: Xe Power Management
+ *
+ * Xe PM shall be guided by the simplicity.
+ * Use the simplest hook options whenever possible.
+ * Let's not reinvent the runtime_pm references and hooks.
+ * Shall have a clear separation of display and gt underneath this component.
+ *
+ * What's next:
+ *
+ * For now s2idle and s3 are only working in integrated devices. The next step
+ * is to iterate through all VRAM's BO backing them up into the system memory
+ * before allowing the system suspend.
+ *
+ * Also runtime_pm needs to be here from the beginning.
+ *
+ * RC6/RPS are also critical PM features. Let's start with GuCRC and GuC SLPC
+ * and no wait boost. Frequency optimizations should come on a next stage.
+ */
+
+/**
+ * xe_pm_suspend - Helper for System suspend, i.e. S0->S3 / S0->S2idle
+ * @xe: xe device instance
+ *
+ * Return: 0 on success
+ */
+int xe_pm_suspend(struct xe_device *xe)
+{
+ struct xe_gt *gt;
+ u8 id;
+ int err;
+
+ for_each_gt(gt, xe, id)
+ xe_gt_suspend_prepare(gt);
+
+ /* FIXME: Super racey... */
+ err = xe_bo_evict_all(xe);
+ if (err)
+ return err;
+
+ xe_display_pm_suspend(xe);
+
+ for_each_gt(gt, xe, id) {
+ err = xe_gt_suspend(gt);
+ if (err) {
+ xe_display_pm_resume(xe);
+ return err;
+ }
+ }
+
+ xe_irq_suspend(xe);
+
+ xe_display_pm_suspend_late(xe);
+
+ return 0;
+}
+
+/**
+ * xe_pm_resume - Helper for System resume S3->S0 / S2idle->S0
+ * @xe: xe device instance
+ *
+ * Return: 0 on success
+ */
+int xe_pm_resume(struct xe_device *xe)
+{
+ struct xe_tile *tile;
+ struct xe_gt *gt;
+ u8 id;
+ int err;
+
+ for_each_tile(tile, xe, id)
+ xe_wa_apply_tile_workarounds(tile);
+
+ for_each_gt(gt, xe, id) {
+ err = xe_pcode_init(gt);
+ if (err)
+ return err;
+ }
+
+ xe_display_pm_resume_early(xe);
+
+ /*
+ * This only restores pinned memory which is the memory required for the
+ * GT(s) to resume.
+ */
+ err = xe_bo_restore_kernel(xe);
+ if (err)
+ return err;
+
+ xe_irq_resume(xe);
+
+ xe_display_pm_resume(xe);
+
+ for_each_gt(gt, xe, id)
+ xe_gt_resume(gt);
+
+ err = xe_bo_restore_user(xe);
+ if (err)
+ return err;
+
+ return 0;
+}
+
+static bool xe_pm_pci_d3cold_capable(struct pci_dev *pdev)
+{
+ struct pci_dev *root_pdev;
+
+ root_pdev = pcie_find_root_port(pdev);
+ if (!root_pdev)
+ return false;
+
+ /* D3Cold requires PME capability and _PR3 power resource */
+ if (!pci_pme_capable(root_pdev, PCI_D3cold) || !pci_pr3_present(root_pdev))
+ return false;
+
+ return true;
+}
+
+static void xe_pm_runtime_init(struct xe_device *xe)
+{
+ struct device *dev = xe->drm.dev;
+
+ /*
+ * Disable the system suspend direct complete optimization.
+ * We need to ensure that the regular device suspend/resume functions
+ * are called since our runtime_pm cannot guarantee local memory
+ * eviction for d3cold.
+ * TODO: Check HDA audio dependencies claimed by i915, and then enforce
+ * this option to integrated graphics as well.
+ */
+ if (IS_DGFX(xe))
+ dev_pm_set_driver_flags(dev, DPM_FLAG_NO_DIRECT_COMPLETE);
+
+ pm_runtime_use_autosuspend(dev);
+ pm_runtime_set_autosuspend_delay(dev, 1000);
+ pm_runtime_set_active(dev);
+ pm_runtime_allow(dev);
+ pm_runtime_mark_last_busy(dev);
+ pm_runtime_put(dev);
+}
+
+void xe_pm_init(struct xe_device *xe)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+
+ /* For now suspend/resume is only allowed with GuC */
+ if (!xe_device_uc_enabled(xe))
+ return;
+
+ drmm_mutex_init(&xe->drm, &xe->d3cold.lock);
+
+ xe->d3cold.capable = xe_pm_pci_d3cold_capable(pdev);
+
+ if (xe->d3cold.capable) {
+ xe_device_sysfs_init(xe);
+ xe_pm_set_vram_threshold(xe, DEFAULT_VRAM_THRESHOLD);
+ }
+
+ xe_pm_runtime_init(xe);
+}
+
+void xe_pm_runtime_fini(struct xe_device *xe)
+{
+ struct device *dev = xe->drm.dev;
+
+ pm_runtime_get_sync(dev);
+ pm_runtime_forbid(dev);
+}
+
+static void xe_pm_write_callback_task(struct xe_device *xe,
+ struct task_struct *task)
+{
+ WRITE_ONCE(xe->pm_callback_task, task);
+
+ /*
+ * Just in case it's somehow possible for our writes to be reordered to
+ * the extent that something else re-uses the task written in
+ * pm_callback_task. For example after returning from the callback, but
+ * before the reordered write that resets pm_callback_task back to NULL.
+ */
+ smp_mb(); /* pairs with xe_pm_read_callback_task */
+}
+
+struct task_struct *xe_pm_read_callback_task(struct xe_device *xe)
+{
+ smp_mb(); /* pairs with xe_pm_write_callback_task */
+
+ return READ_ONCE(xe->pm_callback_task);
+}
+
+int xe_pm_runtime_suspend(struct xe_device *xe)
+{
+ struct xe_gt *gt;
+ u8 id;
+ int err = 0;
+
+ if (xe->d3cold.allowed && xe_device_mem_access_ongoing(xe))
+ return -EBUSY;
+
+ /* Disable access_ongoing asserts and prevent recursive pm calls */
+ xe_pm_write_callback_task(xe, current);
+
+ /*
+ * The actual xe_device_mem_access_put() is always async underneath, so
+ * exactly where that is called should makes no difference to us. However
+ * we still need to be very careful with the locks that this callback
+ * acquires and the locks that are acquired and held by any callers of
+ * xe_device_mem_access_get(). We already have the matching annotation
+ * on that side, but we also need it here. For example lockdep should be
+ * able to tell us if the following scenario is in theory possible:
+ *
+ * CPU0 | CPU1 (kworker)
+ * lock(A) |
+ * | xe_pm_runtime_suspend()
+ * | lock(A)
+ * xe_device_mem_access_get() |
+ *
+ * This will clearly deadlock since rpm core needs to wait for
+ * xe_pm_runtime_suspend() to complete, but here we are holding lock(A)
+ * on CPU0 which prevents CPU1 making forward progress. With the
+ * annotation here and in xe_device_mem_access_get() lockdep will see
+ * the potential lock inversion and give us a nice splat.
+ */
+ lock_map_acquire(&xe_device_mem_access_lockdep_map);
+
+ if (xe->d3cold.allowed) {
+ err = xe_bo_evict_all(xe);
+ if (err)
+ goto out;
+ }
+
+ for_each_gt(gt, xe, id) {
+ err = xe_gt_suspend(gt);
+ if (err)
+ goto out;
+ }
+
+ xe_irq_suspend(xe);
+out:
+ lock_map_release(&xe_device_mem_access_lockdep_map);
+ xe_pm_write_callback_task(xe, NULL);
+ return err;
+}
+
+int xe_pm_runtime_resume(struct xe_device *xe)
+{
+ struct xe_gt *gt;
+ u8 id;
+ int err = 0;
+
+ /* Disable access_ongoing asserts and prevent recursive pm calls */
+ xe_pm_write_callback_task(xe, current);
+
+ lock_map_acquire(&xe_device_mem_access_lockdep_map);
+
+ /*
+ * It can be possible that xe has allowed d3cold but other pcie devices
+ * in gfx card soc would have blocked d3cold, therefore card has not
+ * really lost power. Detecting primary Gt power is sufficient.
+ */
+ gt = xe_device_get_gt(xe, 0);
+ xe->d3cold.power_lost = xe_guc_in_reset(&gt->uc.guc);
+
+ if (xe->d3cold.allowed && xe->d3cold.power_lost) {
+ for_each_gt(gt, xe, id) {
+ err = xe_pcode_init(gt);
+ if (err)
+ goto out;
+ }
+
+ /*
+ * This only restores pinned memory which is the memory
+ * required for the GT(s) to resume.
+ */
+ err = xe_bo_restore_kernel(xe);
+ if (err)
+ goto out;
+ }
+
+ xe_irq_resume(xe);
+
+ for_each_gt(gt, xe, id)
+ xe_gt_resume(gt);
+
+ if (xe->d3cold.allowed && xe->d3cold.power_lost) {
+ err = xe_bo_restore_user(xe);
+ if (err)
+ goto out;
+ }
+out:
+ lock_map_release(&xe_device_mem_access_lockdep_map);
+ xe_pm_write_callback_task(xe, NULL);
+ return err;
+}
+
+int xe_pm_runtime_get(struct xe_device *xe)
+{
+ return pm_runtime_get_sync(xe->drm.dev);
+}
+
+int xe_pm_runtime_put(struct xe_device *xe)
+{
+ pm_runtime_mark_last_busy(xe->drm.dev);
+ return pm_runtime_put(xe->drm.dev);
+}
+
+int xe_pm_runtime_get_if_active(struct xe_device *xe)
+{
+ return pm_runtime_get_if_active(xe->drm.dev, true);
+}
+
+void xe_pm_assert_unbounded_bridge(struct xe_device *xe)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ struct pci_dev *bridge = pci_upstream_bridge(pdev);
+
+ if (!bridge)
+ return;
+
+ if (!bridge->driver) {
+ drm_warn(&xe->drm, "unbounded parent pci bridge, device won't support any PM support.\n");
+ device_set_pm_not_required(&pdev->dev);
+ }
+}
+
+int xe_pm_set_vram_threshold(struct xe_device *xe, u32 threshold)
+{
+ struct ttm_resource_manager *man;
+ u32 vram_total_mb = 0;
+ int i;
+
+ for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i) {
+ man = ttm_manager_type(&xe->ttm, i);
+ if (man)
+ vram_total_mb += DIV_ROUND_UP_ULL(man->size, 1024 * 1024);
+ }
+
+ drm_dbg(&xe->drm, "Total vram %u mb\n", vram_total_mb);
+
+ if (threshold > vram_total_mb)
+ return -EINVAL;
+
+ mutex_lock(&xe->d3cold.lock);
+ xe->d3cold.vram_threshold = threshold;
+ mutex_unlock(&xe->d3cold.lock);
+
+ return 0;
+}
+
+void xe_pm_d3cold_allowed_toggle(struct xe_device *xe)
+{
+ struct ttm_resource_manager *man;
+ u32 total_vram_used_mb = 0;
+ u64 vram_used;
+ int i;
+
+ if (!xe->d3cold.capable) {
+ xe->d3cold.allowed = false;
+ return;
+ }
+
+ for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i) {
+ man = ttm_manager_type(&xe->ttm, i);
+ if (man) {
+ vram_used = ttm_resource_manager_usage(man);
+ total_vram_used_mb += DIV_ROUND_UP_ULL(vram_used, 1024 * 1024);
+ }
+ }
+
+ mutex_lock(&xe->d3cold.lock);
+
+ if (total_vram_used_mb < xe->d3cold.vram_threshold)
+ xe->d3cold.allowed = true;
+ else
+ xe->d3cold.allowed = false;
+
+ mutex_unlock(&xe->d3cold.lock);
+
+ drm_dbg(&xe->drm,
+ "d3cold: allowed=%s\n", str_yes_no(xe->d3cold.allowed));
+}
diff --git a/drivers/gpu/drm/xe/xe_pm.h b/drivers/gpu/drm/xe/xe_pm.h
new file mode 100644
index 000000000000..6b9031f7af24
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pm.h
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PM_H_
+#define _XE_PM_H_
+
+#include <linux/pm_runtime.h>
+
+/*
+ * TODO: Threshold = 0 will block D3Cold.
+ * Before we can move this to a higher value (like 300), we need to:
+ * 1. rewrite the VRAM save / restore to avoid buffer object locks
+ */
+#define DEFAULT_VRAM_THRESHOLD 0 /* in MB */
+
+struct xe_device;
+
+int xe_pm_suspend(struct xe_device *xe);
+int xe_pm_resume(struct xe_device *xe);
+
+void xe_pm_init(struct xe_device *xe);
+void xe_pm_runtime_fini(struct xe_device *xe);
+int xe_pm_runtime_suspend(struct xe_device *xe);
+int xe_pm_runtime_resume(struct xe_device *xe);
+int xe_pm_runtime_get(struct xe_device *xe);
+int xe_pm_runtime_put(struct xe_device *xe);
+int xe_pm_runtime_get_if_active(struct xe_device *xe);
+void xe_pm_assert_unbounded_bridge(struct xe_device *xe);
+int xe_pm_set_vram_threshold(struct xe_device *xe, u32 threshold);
+void xe_pm_d3cold_allowed_toggle(struct xe_device *xe);
+struct task_struct *xe_pm_read_callback_task(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_preempt_fence.c b/drivers/gpu/drm/xe/xe_preempt_fence.c
new file mode 100644
index 000000000000..7bce2a332603
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_preempt_fence.c
@@ -0,0 +1,158 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_preempt_fence.h"
+
+#include <linux/slab.h>
+
+#include "xe_exec_queue.h"
+#include "xe_vm.h"
+
+static void preempt_fence_work_func(struct work_struct *w)
+{
+ bool cookie = dma_fence_begin_signalling();
+ struct xe_preempt_fence *pfence =
+ container_of(w, typeof(*pfence), preempt_work);
+ struct xe_exec_queue *q = pfence->q;
+
+ if (pfence->error)
+ dma_fence_set_error(&pfence->base, pfence->error);
+ else
+ q->ops->suspend_wait(q);
+
+ dma_fence_signal(&pfence->base);
+ dma_fence_end_signalling(cookie);
+
+ xe_vm_queue_rebind_worker(q->vm);
+
+ xe_exec_queue_put(q);
+}
+
+static const char *
+preempt_fence_get_driver_name(struct dma_fence *fence)
+{
+ return "xe";
+}
+
+static const char *
+preempt_fence_get_timeline_name(struct dma_fence *fence)
+{
+ return "preempt";
+}
+
+static bool preempt_fence_enable_signaling(struct dma_fence *fence)
+{
+ struct xe_preempt_fence *pfence =
+ container_of(fence, typeof(*pfence), base);
+ struct xe_exec_queue *q = pfence->q;
+
+ pfence->error = q->ops->suspend(q);
+ queue_work(system_unbound_wq, &pfence->preempt_work);
+ return true;
+}
+
+static const struct dma_fence_ops preempt_fence_ops = {
+ .get_driver_name = preempt_fence_get_driver_name,
+ .get_timeline_name = preempt_fence_get_timeline_name,
+ .enable_signaling = preempt_fence_enable_signaling,
+};
+
+/**
+ * xe_preempt_fence_alloc() - Allocate a preempt fence with minimal
+ * initialization
+ *
+ * Allocate a preempt fence, and initialize its list head.
+ * If the preempt_fence allocated has been armed with
+ * xe_preempt_fence_arm(), it must be freed using dma_fence_put(). If not,
+ * it must be freed using xe_preempt_fence_free().
+ *
+ * Return: A struct xe_preempt_fence pointer used for calling into
+ * xe_preempt_fence_arm() or xe_preempt_fence_free().
+ * An error pointer on error.
+ */
+struct xe_preempt_fence *xe_preempt_fence_alloc(void)
+{
+ struct xe_preempt_fence *pfence;
+
+ pfence = kmalloc(sizeof(*pfence), GFP_KERNEL);
+ if (!pfence)
+ return ERR_PTR(-ENOMEM);
+
+ INIT_LIST_HEAD(&pfence->link);
+ INIT_WORK(&pfence->preempt_work, preempt_fence_work_func);
+
+ return pfence;
+}
+
+/**
+ * xe_preempt_fence_free() - Free a preempt fence allocated using
+ * xe_preempt_fence_alloc().
+ * @pfence: pointer obtained from xe_preempt_fence_alloc();
+ *
+ * Free a preempt fence that has not yet been armed.
+ */
+void xe_preempt_fence_free(struct xe_preempt_fence *pfence)
+{
+ list_del(&pfence->link);
+ kfree(pfence);
+}
+
+/**
+ * xe_preempt_fence_arm() - Arm a preempt fence allocated using
+ * xe_preempt_fence_alloc().
+ * @pfence: The struct xe_preempt_fence pointer returned from
+ * xe_preempt_fence_alloc().
+ * @q: The struct xe_exec_queue used for arming.
+ * @context: The dma-fence context used for arming.
+ * @seqno: The dma-fence seqno used for arming.
+ *
+ * Inserts the preempt fence into @context's timeline, takes @link off any
+ * list, and registers the struct xe_exec_queue as the xe_engine to be preempted.
+ *
+ * Return: A pointer to a struct dma_fence embedded into the preempt fence.
+ * This function doesn't error.
+ */
+struct dma_fence *
+xe_preempt_fence_arm(struct xe_preempt_fence *pfence, struct xe_exec_queue *q,
+ u64 context, u32 seqno)
+{
+ list_del_init(&pfence->link);
+ pfence->q = xe_exec_queue_get(q);
+ dma_fence_init(&pfence->base, &preempt_fence_ops,
+ &q->compute.lock, context, seqno);
+
+ return &pfence->base;
+}
+
+/**
+ * xe_preempt_fence_create() - Helper to create and arm a preempt fence.
+ * @q: The struct xe_exec_queue used for arming.
+ * @context: The dma-fence context used for arming.
+ * @seqno: The dma-fence seqno used for arming.
+ *
+ * Allocates and inserts the preempt fence into @context's timeline,
+ * and registers @e as the struct xe_exec_queue to be preempted.
+ *
+ * Return: A pointer to the resulting struct dma_fence on success. An error
+ * pointer on error. In particular if allocation fails it returns
+ * ERR_PTR(-ENOMEM);
+ */
+struct dma_fence *
+xe_preempt_fence_create(struct xe_exec_queue *q,
+ u64 context, u32 seqno)
+{
+ struct xe_preempt_fence *pfence;
+
+ pfence = xe_preempt_fence_alloc();
+ if (IS_ERR(pfence))
+ return ERR_CAST(pfence);
+
+ return xe_preempt_fence_arm(pfence, q, context, seqno);
+}
+
+bool xe_fence_is_xe_preempt(const struct dma_fence *fence)
+{
+ return fence->ops == &preempt_fence_ops;
+}
diff --git a/drivers/gpu/drm/xe/xe_preempt_fence.h b/drivers/gpu/drm/xe/xe_preempt_fence.h
new file mode 100644
index 000000000000..9406c6fea525
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_preempt_fence.h
@@ -0,0 +1,61 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PREEMPT_FENCE_H_
+#define _XE_PREEMPT_FENCE_H_
+
+#include "xe_preempt_fence_types.h"
+
+struct list_head;
+
+struct dma_fence *
+xe_preempt_fence_create(struct xe_exec_queue *q,
+ u64 context, u32 seqno);
+
+struct xe_preempt_fence *xe_preempt_fence_alloc(void);
+
+void xe_preempt_fence_free(struct xe_preempt_fence *pfence);
+
+struct dma_fence *
+xe_preempt_fence_arm(struct xe_preempt_fence *pfence, struct xe_exec_queue *q,
+ u64 context, u32 seqno);
+
+static inline struct xe_preempt_fence *
+to_preempt_fence(struct dma_fence *fence)
+{
+ return container_of(fence, struct xe_preempt_fence, base);
+}
+
+/**
+ * xe_preempt_fence_link() - Return a link used to keep unarmed preempt
+ * fences on a list.
+ * @pfence: Pointer to the preempt fence.
+ *
+ * The link is embedded in the struct xe_preempt_fence. Use
+ * link_to_preempt_fence() to convert back to the preempt fence.
+ *
+ * Return: A pointer to an embedded struct list_head.
+ */
+static inline struct list_head *
+xe_preempt_fence_link(struct xe_preempt_fence *pfence)
+{
+ return &pfence->link;
+}
+
+/**
+ * to_preempt_fence_from_link() - Convert back to a preempt fence pointer
+ * from a link obtained with xe_preempt_fence_link().
+ * @link: The struct list_head obtained from xe_preempt_fence_link().
+ *
+ * Return: A pointer to the embedding struct xe_preempt_fence.
+ */
+static inline struct xe_preempt_fence *
+to_preempt_fence_from_link(struct list_head *link)
+{
+ return container_of(link, struct xe_preempt_fence, link);
+}
+
+bool xe_fence_is_xe_preempt(const struct dma_fence *fence);
+#endif
diff --git a/drivers/gpu/drm/xe/xe_preempt_fence_types.h b/drivers/gpu/drm/xe/xe_preempt_fence_types.h
new file mode 100644
index 000000000000..b54b5c29b533
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_preempt_fence_types.h
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PREEMPT_FENCE_TYPES_H_
+#define _XE_PREEMPT_FENCE_TYPES_H_
+
+#include <linux/dma-fence.h>
+#include <linux/workqueue.h>
+
+struct xe_exec_queue;
+
+/**
+ * struct xe_preempt_fence - XE preempt fence
+ *
+ * hardware and triggers a callback once the xe_engine is complete.
+ */
+struct xe_preempt_fence {
+ /** @base: dma fence base */
+ struct dma_fence base;
+ /** @link: link into list of pending preempt fences */
+ struct list_head link;
+ /** @q: exec queue for this preempt fence */
+ struct xe_exec_queue *q;
+ /** @preempt_work: work struct which issues preemption */
+ struct work_struct preempt_work;
+ /** @error: preempt fence is in error state */
+ int error;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
new file mode 100644
index 000000000000..de1030a47588
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -0,0 +1,1653 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_pt.h"
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_drm_client.h"
+#include "xe_gt.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_migrate.h"
+#include "xe_pt_types.h"
+#include "xe_pt_walk.h"
+#include "xe_res_cursor.h"
+#include "xe_trace.h"
+#include "xe_ttm_stolen_mgr.h"
+#include "xe_vm.h"
+
+struct xe_pt_dir {
+ struct xe_pt pt;
+ /** @dir: Directory structure for the xe_pt_walk functionality */
+ struct xe_ptw_dir dir;
+};
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
+#define xe_pt_set_addr(__xe_pt, __addr) ((__xe_pt)->addr = (__addr))
+#define xe_pt_addr(__xe_pt) ((__xe_pt)->addr)
+#else
+#define xe_pt_set_addr(__xe_pt, __addr)
+#define xe_pt_addr(__xe_pt) 0ull
+#endif
+
+static const u64 xe_normal_pt_shifts[] = {12, 21, 30, 39, 48};
+static const u64 xe_compact_pt_shifts[] = {16, 21, 30, 39, 48};
+
+#define XE_PT_HIGHEST_LEVEL (ARRAY_SIZE(xe_normal_pt_shifts) - 1)
+
+static struct xe_pt_dir *as_xe_pt_dir(struct xe_pt *pt)
+{
+ return container_of(pt, struct xe_pt_dir, pt);
+}
+
+static struct xe_pt *xe_pt_entry(struct xe_pt_dir *pt_dir, unsigned int index)
+{
+ return container_of(pt_dir->dir.entries[index], struct xe_pt, base);
+}
+
+static u64 __xe_pt_empty_pte(struct xe_tile *tile, struct xe_vm *vm,
+ unsigned int level)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ u16 pat_index = xe->pat.idx[XE_CACHE_WB];
+ u8 id = tile->id;
+
+ if (!xe_vm_has_scratch(vm))
+ return 0;
+
+ if (level > MAX_HUGEPTE_LEVEL)
+ return vm->pt_ops->pde_encode_bo(vm->scratch_pt[id][level - 1]->bo,
+ 0, pat_index);
+
+ return vm->pt_ops->pte_encode_addr(xe, 0, pat_index, level, IS_DGFX(xe), 0) |
+ XE_PTE_NULL;
+}
+
+/**
+ * xe_pt_create() - Create a page-table.
+ * @vm: The vm to create for.
+ * @tile: The tile to create for.
+ * @level: The page-table level.
+ *
+ * Allocate and initialize a single struct xe_pt metadata structure. Also
+ * create the corresponding page-table bo, but don't initialize it. If the
+ * level is grater than zero, then it's assumed to be a directory page-
+ * table and the directory structure is also allocated and initialized to
+ * NULL pointers.
+ *
+ * Return: A valid struct xe_pt pointer on success, Pointer error code on
+ * error.
+ */
+struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile,
+ unsigned int level)
+{
+ struct xe_pt *pt;
+ struct xe_bo *bo;
+ size_t size;
+ int err;
+
+ size = !level ? sizeof(struct xe_pt) : sizeof(struct xe_pt_dir) +
+ XE_PDES * sizeof(struct xe_ptw *);
+ pt = kzalloc(size, GFP_KERNEL);
+ if (!pt)
+ return ERR_PTR(-ENOMEM);
+
+ bo = xe_bo_create_pin_map(vm->xe, tile, vm, SZ_4K,
+ ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_IGNORE_MIN_PAGE_SIZE_BIT |
+ XE_BO_CREATE_PINNED_BIT |
+ XE_BO_CREATE_NO_RESV_EVICT |
+ XE_BO_PAGETABLE);
+ if (IS_ERR(bo)) {
+ err = PTR_ERR(bo);
+ goto err_kfree;
+ }
+ pt->bo = bo;
+ pt->level = level;
+ pt->base.dir = level ? &as_xe_pt_dir(pt)->dir : NULL;
+
+ if (vm->xef)
+ xe_drm_client_add_bo(vm->xef->client, pt->bo);
+ xe_tile_assert(tile, level <= XE_VM_MAX_LEVEL);
+
+ return pt;
+
+err_kfree:
+ kfree(pt);
+ return ERR_PTR(err);
+}
+
+/**
+ * xe_pt_populate_empty() - Populate a page-table bo with scratch- or zero
+ * entries.
+ * @tile: The tile the scratch pagetable of which to use.
+ * @vm: The vm we populate for.
+ * @pt: The pagetable the bo of which to initialize.
+ *
+ * Populate the page-table bo of @pt with entries pointing into the tile's
+ * scratch page-table tree if any. Otherwise populate with zeros.
+ */
+void xe_pt_populate_empty(struct xe_tile *tile, struct xe_vm *vm,
+ struct xe_pt *pt)
+{
+ struct iosys_map *map = &pt->bo->vmap;
+ u64 empty;
+ int i;
+
+ if (!xe_vm_has_scratch(vm)) {
+ /*
+ * FIXME: Some memory is allocated already allocated to zero?
+ * Find out which memory that is and avoid this memset...
+ */
+ xe_map_memset(vm->xe, map, 0, 0, SZ_4K);
+ } else {
+ empty = __xe_pt_empty_pte(tile, vm, pt->level);
+ for (i = 0; i < XE_PDES; i++)
+ xe_pt_write(vm->xe, map, i, empty);
+ }
+}
+
+/**
+ * xe_pt_shift() - Return the ilog2 value of the size of the address range of
+ * a page-table at a certain level.
+ * @level: The level.
+ *
+ * Return: The ilog2 value of the size of the address range of a page-table
+ * at level @level.
+ */
+unsigned int xe_pt_shift(unsigned int level)
+{
+ return XE_PTE_SHIFT + XE_PDE_SHIFT * level;
+}
+
+/**
+ * xe_pt_destroy() - Destroy a page-table tree.
+ * @pt: The root of the page-table tree to destroy.
+ * @flags: vm flags. Currently unused.
+ * @deferred: List head of lockless list for deferred putting. NULL for
+ * immediate putting.
+ *
+ * Puts the page-table bo, recursively calls xe_pt_destroy on all children
+ * and finally frees @pt. TODO: Can we remove the @flags argument?
+ */
+void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred)
+{
+ int i;
+
+ if (!pt)
+ return;
+
+ XE_WARN_ON(!list_empty(&pt->bo->ttm.base.gpuva.list));
+ xe_bo_unpin(pt->bo);
+ xe_bo_put_deferred(pt->bo, deferred);
+
+ if (pt->level > 0 && pt->num_live) {
+ struct xe_pt_dir *pt_dir = as_xe_pt_dir(pt);
+
+ for (i = 0; i < XE_PDES; i++) {
+ if (xe_pt_entry(pt_dir, i))
+ xe_pt_destroy(xe_pt_entry(pt_dir, i), flags,
+ deferred);
+ }
+ }
+ kfree(pt);
+}
+
+/**
+ * DOC: Pagetable building
+ *
+ * Below we use the term "page-table" for both page-directories, containing
+ * pointers to lower level page-directories or page-tables, and level 0
+ * page-tables that contain only page-table-entries pointing to memory pages.
+ *
+ * When inserting an address range in an already existing page-table tree
+ * there will typically be a set of page-tables that are shared with other
+ * address ranges, and a set that are private to this address range.
+ * The set of shared page-tables can be at most two per level,
+ * and those can't be updated immediately because the entries of those
+ * page-tables may still be in use by the gpu for other mappings. Therefore
+ * when inserting entries into those, we instead stage those insertions by
+ * adding insertion data into struct xe_vm_pgtable_update structures. This
+ * data, (subtrees for the cpu and page-table-entries for the gpu) is then
+ * added in a separate commit step. CPU-data is committed while still under the
+ * vm lock, the object lock and for userptr, the notifier lock in read mode.
+ * The GPU async data is committed either by the GPU or CPU after fulfilling
+ * relevant dependencies.
+ * For non-shared page-tables (and, in fact, for shared ones that aren't
+ * existing at the time of staging), we add the data in-place without the
+ * special update structures. This private part of the page-table tree will
+ * remain disconnected from the vm page-table tree until data is committed to
+ * the shared page tables of the vm tree in the commit phase.
+ */
+
+struct xe_pt_update {
+ /** @update: The update structure we're building for this parent. */
+ struct xe_vm_pgtable_update *update;
+ /** @parent: The parent. Used to detect a parent change. */
+ struct xe_pt *parent;
+ /** @preexisting: Whether the parent was pre-existing or allocated */
+ bool preexisting;
+};
+
+struct xe_pt_stage_bind_walk {
+ /** base: The base class. */
+ struct xe_pt_walk base;
+
+ /* Input parameters for the walk */
+ /** @vm: The vm we're building for. */
+ struct xe_vm *vm;
+ /** @tile: The tile we're building for. */
+ struct xe_tile *tile;
+ /** @default_pte: PTE flag only template. No address is associated */
+ u64 default_pte;
+ /** @dma_offset: DMA offset to add to the PTE. */
+ u64 dma_offset;
+ /**
+ * @needs_64k: This address range enforces 64K alignment and
+ * granularity.
+ */
+ bool needs_64K;
+ /**
+ * @vma: VMA being mapped
+ */
+ struct xe_vma *vma;
+
+ /* Also input, but is updated during the walk*/
+ /** @curs: The DMA address cursor. */
+ struct xe_res_cursor *curs;
+ /** @va_curs_start: The Virtual address coresponding to @curs->start */
+ u64 va_curs_start;
+
+ /* Output */
+ struct xe_walk_update {
+ /** @wupd.entries: Caller provided storage. */
+ struct xe_vm_pgtable_update *entries;
+ /** @wupd.num_used_entries: Number of update @entries used. */
+ unsigned int num_used_entries;
+ /** @wupd.updates: Tracks the update entry at a given level */
+ struct xe_pt_update updates[XE_VM_MAX_LEVEL + 1];
+ } wupd;
+
+ /* Walk state */
+ /**
+ * @l0_end_addr: The end address of the current l0 leaf. Used for
+ * 64K granularity detection.
+ */
+ u64 l0_end_addr;
+ /** @addr_64K: The start address of the current 64K chunk. */
+ u64 addr_64K;
+ /** @found_64: Whether @add_64K actually points to a 64K chunk. */
+ bool found_64K;
+};
+
+static int
+xe_pt_new_shared(struct xe_walk_update *wupd, struct xe_pt *parent,
+ pgoff_t offset, bool alloc_entries)
+{
+ struct xe_pt_update *upd = &wupd->updates[parent->level];
+ struct xe_vm_pgtable_update *entry;
+
+ /*
+ * For *each level*, we could only have one active
+ * struct xt_pt_update at any one time. Once we move on to a
+ * new parent and page-directory, the old one is complete, and
+ * updates are either already stored in the build tree or in
+ * @wupd->entries
+ */
+ if (likely(upd->parent == parent))
+ return 0;
+
+ upd->parent = parent;
+ upd->preexisting = true;
+
+ if (wupd->num_used_entries == XE_VM_MAX_LEVEL * 2 + 1)
+ return -EINVAL;
+
+ entry = wupd->entries + wupd->num_used_entries++;
+ upd->update = entry;
+ entry->ofs = offset;
+ entry->pt_bo = parent->bo;
+ entry->pt = parent;
+ entry->flags = 0;
+ entry->qwords = 0;
+
+ if (alloc_entries) {
+ entry->pt_entries = kmalloc_array(XE_PDES,
+ sizeof(*entry->pt_entries),
+ GFP_KERNEL);
+ if (!entry->pt_entries)
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+/*
+ * NOTE: This is a very frequently called function so we allow ourselves
+ * to annotate (using branch prediction hints) the fastpath of updating a
+ * non-pre-existing pagetable with leaf ptes.
+ */
+static int
+xe_pt_insert_entry(struct xe_pt_stage_bind_walk *xe_walk, struct xe_pt *parent,
+ pgoff_t offset, struct xe_pt *xe_child, u64 pte)
+{
+ struct xe_pt_update *upd = &xe_walk->wupd.updates[parent->level];
+ struct xe_pt_update *child_upd = xe_child ?
+ &xe_walk->wupd.updates[xe_child->level] : NULL;
+ int ret;
+
+ ret = xe_pt_new_shared(&xe_walk->wupd, parent, offset, true);
+ if (unlikely(ret))
+ return ret;
+
+ /*
+ * Register this new pagetable so that it won't be recognized as
+ * a shared pagetable by a subsequent insertion.
+ */
+ if (unlikely(child_upd)) {
+ child_upd->update = NULL;
+ child_upd->parent = xe_child;
+ child_upd->preexisting = false;
+ }
+
+ if (likely(!upd->preexisting)) {
+ /* Continue building a non-connected subtree. */
+ struct iosys_map *map = &parent->bo->vmap;
+
+ if (unlikely(xe_child))
+ parent->base.dir->entries[offset] = &xe_child->base;
+
+ xe_pt_write(xe_walk->vm->xe, map, offset, pte);
+ parent->num_live++;
+ } else {
+ /* Shared pt. Stage update. */
+ unsigned int idx;
+ struct xe_vm_pgtable_update *entry = upd->update;
+
+ idx = offset - entry->ofs;
+ entry->pt_entries[idx].pt = xe_child;
+ entry->pt_entries[idx].pte = pte;
+ entry->qwords++;
+ }
+
+ return 0;
+}
+
+static bool xe_pt_hugepte_possible(u64 addr, u64 next, unsigned int level,
+ struct xe_pt_stage_bind_walk *xe_walk)
+{
+ u64 size, dma;
+
+ if (level > MAX_HUGEPTE_LEVEL)
+ return false;
+
+ /* Does the virtual range requested cover a huge pte? */
+ if (!xe_pt_covers(addr, next, level, &xe_walk->base))
+ return false;
+
+ /* Does the DMA segment cover the whole pte? */
+ if (next - xe_walk->va_curs_start > xe_walk->curs->size)
+ return false;
+
+ /* null VMA's do not have dma addresses */
+ if (xe_vma_is_null(xe_walk->vma))
+ return true;
+
+ /* Is the DMA address huge PTE size aligned? */
+ size = next - addr;
+ dma = addr - xe_walk->va_curs_start + xe_res_dma(xe_walk->curs);
+
+ return IS_ALIGNED(dma, size);
+}
+
+/*
+ * Scan the requested mapping to check whether it can be done entirely
+ * with 64K PTEs.
+ */
+static bool
+xe_pt_scan_64K(u64 addr, u64 next, struct xe_pt_stage_bind_walk *xe_walk)
+{
+ struct xe_res_cursor curs = *xe_walk->curs;
+
+ if (!IS_ALIGNED(addr, SZ_64K))
+ return false;
+
+ if (next > xe_walk->l0_end_addr)
+ return false;
+
+ /* null VMA's do not have dma addresses */
+ if (xe_vma_is_null(xe_walk->vma))
+ return true;
+
+ xe_res_next(&curs, addr - xe_walk->va_curs_start);
+ for (; addr < next; addr += SZ_64K) {
+ if (!IS_ALIGNED(xe_res_dma(&curs), SZ_64K) || curs.size < SZ_64K)
+ return false;
+
+ xe_res_next(&curs, SZ_64K);
+ }
+
+ return addr == next;
+}
+
+/*
+ * For non-compact "normal" 4K level-0 pagetables, we want to try to group
+ * addresses together in 64K-contigous regions to add a 64K TLB hint for the
+ * device to the PTE.
+ * This function determines whether the address is part of such a
+ * segment. For VRAM in normal pagetables, this is strictly necessary on
+ * some devices.
+ */
+static bool
+xe_pt_is_pte_ps64K(u64 addr, u64 next, struct xe_pt_stage_bind_walk *xe_walk)
+{
+ /* Address is within an already found 64k region */
+ if (xe_walk->found_64K && addr - xe_walk->addr_64K < SZ_64K)
+ return true;
+
+ xe_walk->found_64K = xe_pt_scan_64K(addr, addr + SZ_64K, xe_walk);
+ xe_walk->addr_64K = addr;
+
+ return xe_walk->found_64K;
+}
+
+static int
+xe_pt_stage_bind_entry(struct xe_ptw *parent, pgoff_t offset,
+ unsigned int level, u64 addr, u64 next,
+ struct xe_ptw **child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk)
+{
+ struct xe_pt_stage_bind_walk *xe_walk =
+ container_of(walk, typeof(*xe_walk), base);
+ u16 pat_index = xe_walk->vma->pat_index;
+ struct xe_pt *xe_parent = container_of(parent, typeof(*xe_parent), base);
+ struct xe_vm *vm = xe_walk->vm;
+ struct xe_pt *xe_child;
+ bool covers;
+ int ret = 0;
+ u64 pte;
+
+ /* Is this a leaf entry ?*/
+ if (level == 0 || xe_pt_hugepte_possible(addr, next, level, xe_walk)) {
+ struct xe_res_cursor *curs = xe_walk->curs;
+ bool is_null = xe_vma_is_null(xe_walk->vma);
+
+ XE_WARN_ON(xe_walk->va_curs_start != addr);
+
+ pte = vm->pt_ops->pte_encode_vma(is_null ? 0 :
+ xe_res_dma(curs) + xe_walk->dma_offset,
+ xe_walk->vma, pat_index, level);
+ pte |= xe_walk->default_pte;
+
+ /*
+ * Set the XE_PTE_PS64 hint if possible, otherwise if
+ * this device *requires* 64K PTE size for VRAM, fail.
+ */
+ if (level == 0 && !xe_parent->is_compact) {
+ if (xe_pt_is_pte_ps64K(addr, next, xe_walk))
+ pte |= XE_PTE_PS64;
+ else if (XE_WARN_ON(xe_walk->needs_64K))
+ return -EINVAL;
+ }
+
+ ret = xe_pt_insert_entry(xe_walk, xe_parent, offset, NULL, pte);
+ if (unlikely(ret))
+ return ret;
+
+ if (!is_null)
+ xe_res_next(curs, next - addr);
+ xe_walk->va_curs_start = next;
+ xe_walk->vma->gpuva.flags |= (XE_VMA_PTE_4K << level);
+ *action = ACTION_CONTINUE;
+
+ return ret;
+ }
+
+ /*
+ * Descending to lower level. Determine if we need to allocate a
+ * new page table or -directory, which we do if there is no
+ * previous one or there is one we can completely replace.
+ */
+ if (level == 1) {
+ walk->shifts = xe_normal_pt_shifts;
+ xe_walk->l0_end_addr = next;
+ }
+
+ covers = xe_pt_covers(addr, next, level, &xe_walk->base);
+ if (covers || !*child) {
+ u64 flags = 0;
+
+ xe_child = xe_pt_create(xe_walk->vm, xe_walk->tile, level - 1);
+ if (IS_ERR(xe_child))
+ return PTR_ERR(xe_child);
+
+ xe_pt_set_addr(xe_child,
+ round_down(addr, 1ull << walk->shifts[level]));
+
+ if (!covers)
+ xe_pt_populate_empty(xe_walk->tile, xe_walk->vm, xe_child);
+
+ *child = &xe_child->base;
+
+ /*
+ * Prefer the compact pagetable layout for L0 if possible.
+ * TODO: Suballocate the pt bo to avoid wasting a lot of
+ * memory.
+ */
+ if (GRAPHICS_VERx100(tile_to_xe(xe_walk->tile)) >= 1250 && level == 1 &&
+ covers && xe_pt_scan_64K(addr, next, xe_walk)) {
+ walk->shifts = xe_compact_pt_shifts;
+ flags |= XE_PDE_64K;
+ xe_child->is_compact = true;
+ }
+
+ pte = vm->pt_ops->pde_encode_bo(xe_child->bo, 0, pat_index) | flags;
+ ret = xe_pt_insert_entry(xe_walk, xe_parent, offset, xe_child,
+ pte);
+ }
+
+ *action = ACTION_SUBTREE;
+ return ret;
+}
+
+static const struct xe_pt_walk_ops xe_pt_stage_bind_ops = {
+ .pt_entry = xe_pt_stage_bind_entry,
+};
+
+/**
+ * xe_pt_stage_bind() - Build a disconnected page-table tree for a given address
+ * range.
+ * @tile: The tile we're building for.
+ * @vma: The vma indicating the address range.
+ * @entries: Storage for the update entries used for connecting the tree to
+ * the main tree at commit time.
+ * @num_entries: On output contains the number of @entries used.
+ *
+ * This function builds a disconnected page-table tree for a given address
+ * range. The tree is connected to the main vm tree for the gpu using
+ * xe_migrate_update_pgtables() and for the cpu using xe_pt_commit_bind().
+ * The function builds xe_vm_pgtable_update structures for already existing
+ * shared page-tables, and non-existing shared and non-shared page-tables
+ * are built and populated directly.
+ *
+ * Return 0 on success, negative error code on error.
+ */
+static int
+xe_pt_stage_bind(struct xe_tile *tile, struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries, u32 *num_entries)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_bo *bo = xe_vma_bo(vma);
+ bool is_devmem = !xe_vma_is_userptr(vma) && bo &&
+ (xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo));
+ struct xe_res_cursor curs;
+ struct xe_pt_stage_bind_walk xe_walk = {
+ .base = {
+ .ops = &xe_pt_stage_bind_ops,
+ .shifts = xe_normal_pt_shifts,
+ .max_level = XE_PT_HIGHEST_LEVEL,
+ },
+ .vm = xe_vma_vm(vma),
+ .tile = tile,
+ .curs = &curs,
+ .va_curs_start = xe_vma_start(vma),
+ .vma = vma,
+ .wupd.entries = entries,
+ .needs_64K = (xe_vma_vm(vma)->flags & XE_VM_FLAG_64K) && is_devmem,
+ };
+ struct xe_pt *pt = xe_vma_vm(vma)->pt_root[tile->id];
+ int ret;
+
+ if (vma && (vma->gpuva.flags & XE_VMA_ATOMIC_PTE_BIT) &&
+ (is_devmem || !IS_DGFX(xe)))
+ xe_walk.default_pte |= XE_USM_PPGTT_PTE_AE;
+
+ if (is_devmem) {
+ xe_walk.default_pte |= XE_PPGTT_PTE_DM;
+ xe_walk.dma_offset = vram_region_gpu_offset(bo->ttm.resource);
+ }
+
+ if (!xe_vma_has_no_bo(vma) && xe_bo_is_stolen(bo))
+ xe_walk.dma_offset = xe_ttm_stolen_gpu_offset(xe_bo_device(bo));
+
+ xe_bo_assert_held(bo);
+
+ if (!xe_vma_is_null(vma)) {
+ if (xe_vma_is_userptr(vma))
+ xe_res_first_sg(vma->userptr.sg, 0, xe_vma_size(vma),
+ &curs);
+ else if (xe_bo_is_vram(bo) || xe_bo_is_stolen(bo))
+ xe_res_first(bo->ttm.resource, xe_vma_bo_offset(vma),
+ xe_vma_size(vma), &curs);
+ else
+ xe_res_first_sg(xe_bo_sg(bo), xe_vma_bo_offset(vma),
+ xe_vma_size(vma), &curs);
+ } else {
+ curs.size = xe_vma_size(vma);
+ }
+
+ ret = xe_pt_walk_range(&pt->base, pt->level, xe_vma_start(vma),
+ xe_vma_end(vma), &xe_walk.base);
+
+ *num_entries = xe_walk.wupd.num_used_entries;
+ return ret;
+}
+
+/**
+ * xe_pt_nonshared_offsets() - Determine the non-shared entry offsets of a
+ * shared pagetable.
+ * @addr: The start address within the non-shared pagetable.
+ * @end: The end address within the non-shared pagetable.
+ * @level: The level of the non-shared pagetable.
+ * @walk: Walk info. The function adjusts the walk action.
+ * @action: next action to perform (see enum page_walk_action)
+ * @offset: Ignored on input, First non-shared entry on output.
+ * @end_offset: Ignored on input, Last non-shared entry + 1 on output.
+ *
+ * A non-shared page-table has some entries that belong to the address range
+ * and others that don't. This function determines the entries that belong
+ * fully to the address range. Depending on level, some entries may
+ * partially belong to the address range (that can't happen at level 0).
+ * The function detects that and adjust those offsets to not include those
+ * partial entries. Iff it does detect partial entries, we know that there must
+ * be shared page tables also at lower levels, so it adjusts the walk action
+ * accordingly.
+ *
+ * Return: true if there were non-shared entries, false otherwise.
+ */
+static bool xe_pt_nonshared_offsets(u64 addr, u64 end, unsigned int level,
+ struct xe_pt_walk *walk,
+ enum page_walk_action *action,
+ pgoff_t *offset, pgoff_t *end_offset)
+{
+ u64 size = 1ull << walk->shifts[level];
+
+ *offset = xe_pt_offset(addr, level, walk);
+ *end_offset = xe_pt_num_entries(addr, end, level, walk) + *offset;
+
+ if (!level)
+ return true;
+
+ /*
+ * If addr or next are not size aligned, there are shared pts at lower
+ * level, so in that case traverse down the subtree
+ */
+ *action = ACTION_CONTINUE;
+ if (!IS_ALIGNED(addr, size)) {
+ *action = ACTION_SUBTREE;
+ (*offset)++;
+ }
+
+ if (!IS_ALIGNED(end, size)) {
+ *action = ACTION_SUBTREE;
+ (*end_offset)--;
+ }
+
+ return *end_offset > *offset;
+}
+
+struct xe_pt_zap_ptes_walk {
+ /** @base: The walk base-class */
+ struct xe_pt_walk base;
+
+ /* Input parameters for the walk */
+ /** @tile: The tile we're building for */
+ struct xe_tile *tile;
+
+ /* Output */
+ /** @needs_invalidate: Whether we need to invalidate TLB*/
+ bool needs_invalidate;
+};
+
+static int xe_pt_zap_ptes_entry(struct xe_ptw *parent, pgoff_t offset,
+ unsigned int level, u64 addr, u64 next,
+ struct xe_ptw **child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk)
+{
+ struct xe_pt_zap_ptes_walk *xe_walk =
+ container_of(walk, typeof(*xe_walk), base);
+ struct xe_pt *xe_child = container_of(*child, typeof(*xe_child), base);
+ pgoff_t end_offset;
+
+ XE_WARN_ON(!*child);
+ XE_WARN_ON(!level && xe_child->is_compact);
+
+ /*
+ * Note that we're called from an entry callback, and we're dealing
+ * with the child of that entry rather than the parent, so need to
+ * adjust level down.
+ */
+ if (xe_pt_nonshared_offsets(addr, next, --level, walk, action, &offset,
+ &end_offset)) {
+ xe_map_memset(tile_to_xe(xe_walk->tile), &xe_child->bo->vmap,
+ offset * sizeof(u64), 0,
+ (end_offset - offset) * sizeof(u64));
+ xe_walk->needs_invalidate = true;
+ }
+
+ return 0;
+}
+
+static const struct xe_pt_walk_ops xe_pt_zap_ptes_ops = {
+ .pt_entry = xe_pt_zap_ptes_entry,
+};
+
+/**
+ * xe_pt_zap_ptes() - Zap (zero) gpu ptes of an address range
+ * @tile: The tile we're zapping for.
+ * @vma: GPU VMA detailing address range.
+ *
+ * Eviction and Userptr invalidation needs to be able to zap the
+ * gpu ptes of a given address range in pagefaulting mode.
+ * In order to be able to do that, that function needs access to the shared
+ * page-table entrieaso it can either clear the leaf PTEs or
+ * clear the pointers to lower-level page-tables. The caller is required
+ * to hold the necessary locks to ensure neither the page-table connectivity
+ * nor the page-table entries of the range is updated from under us.
+ *
+ * Return: Whether ptes were actually updated and a TLB invalidation is
+ * required.
+ */
+bool xe_pt_zap_ptes(struct xe_tile *tile, struct xe_vma *vma)
+{
+ struct xe_pt_zap_ptes_walk xe_walk = {
+ .base = {
+ .ops = &xe_pt_zap_ptes_ops,
+ .shifts = xe_normal_pt_shifts,
+ .max_level = XE_PT_HIGHEST_LEVEL,
+ },
+ .tile = tile,
+ };
+ struct xe_pt *pt = xe_vma_vm(vma)->pt_root[tile->id];
+
+ if (!(vma->tile_present & BIT(tile->id)))
+ return false;
+
+ (void)xe_pt_walk_shared(&pt->base, pt->level, xe_vma_start(vma),
+ xe_vma_end(vma), &xe_walk.base);
+
+ return xe_walk.needs_invalidate;
+}
+
+static void
+xe_vm_populate_pgtable(struct xe_migrate_pt_update *pt_update, struct xe_tile *tile,
+ struct iosys_map *map, void *data,
+ u32 qword_ofs, u32 num_qwords,
+ const struct xe_vm_pgtable_update *update)
+{
+ struct xe_pt_entry *ptes = update->pt_entries;
+ u64 *ptr = data;
+ u32 i;
+
+ for (i = 0; i < num_qwords; i++) {
+ if (map)
+ xe_map_wr(tile_to_xe(tile), map, (qword_ofs + i) *
+ sizeof(u64), u64, ptes[i].pte);
+ else
+ ptr[i] = ptes[i].pte;
+ }
+}
+
+static void xe_pt_abort_bind(struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries,
+ u32 num_entries)
+{
+ u32 i, j;
+
+ for (i = 0; i < num_entries; i++) {
+ if (!entries[i].pt_entries)
+ continue;
+
+ for (j = 0; j < entries[i].qwords; j++)
+ xe_pt_destroy(entries[i].pt_entries[j].pt, xe_vma_vm(vma)->flags, NULL);
+ kfree(entries[i].pt_entries);
+ }
+}
+
+static void xe_pt_commit_locks_assert(struct xe_vma *vma)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+
+ lockdep_assert_held(&vm->lock);
+
+ if (xe_vma_is_userptr(vma))
+ lockdep_assert_held_read(&vm->userptr.notifier_lock);
+ else if (!xe_vma_is_null(vma))
+ dma_resv_assert_held(xe_vma_bo(vma)->ttm.base.resv);
+
+ xe_vm_assert_held(vm);
+}
+
+static void xe_pt_commit_bind(struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries,
+ u32 num_entries, bool rebind,
+ struct llist_head *deferred)
+{
+ u32 i, j;
+
+ xe_pt_commit_locks_assert(vma);
+
+ for (i = 0; i < num_entries; i++) {
+ struct xe_pt *pt = entries[i].pt;
+ struct xe_pt_dir *pt_dir;
+
+ if (!rebind)
+ pt->num_live += entries[i].qwords;
+
+ if (!pt->level) {
+ kfree(entries[i].pt_entries);
+ continue;
+ }
+
+ pt_dir = as_xe_pt_dir(pt);
+ for (j = 0; j < entries[i].qwords; j++) {
+ u32 j_ = j + entries[i].ofs;
+ struct xe_pt *newpte = entries[i].pt_entries[j].pt;
+
+ if (xe_pt_entry(pt_dir, j_))
+ xe_pt_destroy(xe_pt_entry(pt_dir, j_),
+ xe_vma_vm(vma)->flags, deferred);
+
+ pt_dir->dir.entries[j_] = &newpte->base;
+ }
+ kfree(entries[i].pt_entries);
+ }
+}
+
+static int
+xe_pt_prepare_bind(struct xe_tile *tile, struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries, u32 *num_entries,
+ bool rebind)
+{
+ int err;
+
+ *num_entries = 0;
+ err = xe_pt_stage_bind(tile, vma, entries, num_entries);
+ if (!err)
+ xe_tile_assert(tile, *num_entries);
+ else /* abort! */
+ xe_pt_abort_bind(vma, entries, *num_entries);
+
+ return err;
+}
+
+static void xe_vm_dbg_print_entries(struct xe_device *xe,
+ const struct xe_vm_pgtable_update *entries,
+ unsigned int num_entries)
+#if (IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM))
+{
+ unsigned int i;
+
+ vm_dbg(&xe->drm, "%u entries to update\n", num_entries);
+ for (i = 0; i < num_entries; i++) {
+ const struct xe_vm_pgtable_update *entry = &entries[i];
+ struct xe_pt *xe_pt = entry->pt;
+ u64 page_size = 1ull << xe_pt_shift(xe_pt->level);
+ u64 end;
+ u64 start;
+
+ xe_assert(xe, !entry->pt->is_compact);
+ start = entry->ofs * page_size;
+ end = start + page_size * entry->qwords;
+ vm_dbg(&xe->drm,
+ "\t%u: Update level %u at (%u + %u) [%llx...%llx) f:%x\n",
+ i, xe_pt->level, entry->ofs, entry->qwords,
+ xe_pt_addr(xe_pt) + start, xe_pt_addr(xe_pt) + end, 0);
+ }
+}
+#else
+{}
+#endif
+
+#ifdef CONFIG_DRM_XE_USERPTR_INVAL_INJECT
+
+static int xe_pt_userptr_inject_eagain(struct xe_vma *vma)
+{
+ u32 divisor = vma->userptr.divisor ? vma->userptr.divisor : 2;
+ static u32 count;
+
+ if (count++ % divisor == divisor - 1) {
+ struct xe_vm *vm = xe_vma_vm(vma);
+
+ vma->userptr.divisor = divisor << 1;
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_move_tail(&vma->userptr.invalidate_link,
+ &vm->userptr.invalidated);
+ spin_unlock(&vm->userptr.invalidated_lock);
+ return true;
+ }
+
+ return false;
+}
+
+#else
+
+static bool xe_pt_userptr_inject_eagain(struct xe_vma *vma)
+{
+ return false;
+}
+
+#endif
+
+/**
+ * struct xe_pt_migrate_pt_update - Callback argument for pre-commit callbacks
+ * @base: Base we derive from.
+ * @bind: Whether this is a bind or an unbind operation. A bind operation
+ * makes the pre-commit callback error with -EAGAIN if it detects a
+ * pending invalidation.
+ * @locked: Whether the pre-commit callback locked the userptr notifier lock
+ * and it needs unlocking.
+ */
+struct xe_pt_migrate_pt_update {
+ struct xe_migrate_pt_update base;
+ bool bind;
+ bool locked;
+};
+
+/*
+ * This function adds the needed dependencies to a page-table update job
+ * to make sure racing jobs for separate bind engines don't race writing
+ * to the same page-table range, wreaking havoc. Initially use a single
+ * fence for the entire VM. An optimization would use smaller granularity.
+ */
+static int xe_pt_vm_dependencies(struct xe_sched_job *job,
+ struct xe_range_fence_tree *rftree,
+ u64 start, u64 last)
+{
+ struct xe_range_fence *rtfence;
+ struct dma_fence *fence;
+ int err;
+
+ rtfence = xe_range_fence_tree_first(rftree, start, last);
+ while (rtfence) {
+ fence = rtfence->fence;
+
+ if (!dma_fence_is_signaled(fence)) {
+ /*
+ * Is this a CPU update? GPU is busy updating, so return
+ * an error
+ */
+ if (!job)
+ return -ETIME;
+
+ dma_fence_get(fence);
+ err = drm_sched_job_add_dependency(&job->drm, fence);
+ if (err)
+ return err;
+ }
+
+ rtfence = xe_range_fence_tree_next(rtfence, start, last);
+ }
+
+ return 0;
+}
+
+static int xe_pt_pre_commit(struct xe_migrate_pt_update *pt_update)
+{
+ struct xe_range_fence_tree *rftree =
+ &xe_vma_vm(pt_update->vma)->rftree[pt_update->tile_id];
+
+ return xe_pt_vm_dependencies(pt_update->job, rftree,
+ pt_update->start, pt_update->last);
+}
+
+static int xe_pt_userptr_pre_commit(struct xe_migrate_pt_update *pt_update)
+{
+ struct xe_pt_migrate_pt_update *userptr_update =
+ container_of(pt_update, typeof(*userptr_update), base);
+ struct xe_vma *vma = pt_update->vma;
+ unsigned long notifier_seq = vma->userptr.notifier_seq;
+ struct xe_vm *vm = xe_vma_vm(vma);
+ int err = xe_pt_vm_dependencies(pt_update->job,
+ &vm->rftree[pt_update->tile_id],
+ pt_update->start,
+ pt_update->last);
+
+ if (err)
+ return err;
+
+ userptr_update->locked = false;
+
+ /*
+ * Wait until nobody is running the invalidation notifier, and
+ * since we're exiting the loop holding the notifier lock,
+ * nobody can proceed invalidating either.
+ *
+ * Note that we don't update the vma->userptr.notifier_seq since
+ * we don't update the userptr pages.
+ */
+ do {
+ down_read(&vm->userptr.notifier_lock);
+ if (!mmu_interval_read_retry(&vma->userptr.notifier,
+ notifier_seq))
+ break;
+
+ up_read(&vm->userptr.notifier_lock);
+
+ if (userptr_update->bind)
+ return -EAGAIN;
+
+ notifier_seq = mmu_interval_read_begin(&vma->userptr.notifier);
+ } while (true);
+
+ /* Inject errors to test_whether they are handled correctly */
+ if (userptr_update->bind && xe_pt_userptr_inject_eagain(vma)) {
+ up_read(&vm->userptr.notifier_lock);
+ return -EAGAIN;
+ }
+
+ userptr_update->locked = true;
+
+ return 0;
+}
+
+static const struct xe_migrate_pt_update_ops bind_ops = {
+ .populate = xe_vm_populate_pgtable,
+ .pre_commit = xe_pt_pre_commit,
+};
+
+static const struct xe_migrate_pt_update_ops userptr_bind_ops = {
+ .populate = xe_vm_populate_pgtable,
+ .pre_commit = xe_pt_userptr_pre_commit,
+};
+
+struct invalidation_fence {
+ struct xe_gt_tlb_invalidation_fence base;
+ struct xe_gt *gt;
+ struct xe_vma *vma;
+ struct dma_fence *fence;
+ struct dma_fence_cb cb;
+ struct work_struct work;
+};
+
+static const char *
+invalidation_fence_get_driver_name(struct dma_fence *dma_fence)
+{
+ return "xe";
+}
+
+static const char *
+invalidation_fence_get_timeline_name(struct dma_fence *dma_fence)
+{
+ return "invalidation_fence";
+}
+
+static const struct dma_fence_ops invalidation_fence_ops = {
+ .get_driver_name = invalidation_fence_get_driver_name,
+ .get_timeline_name = invalidation_fence_get_timeline_name,
+};
+
+static void invalidation_fence_cb(struct dma_fence *fence,
+ struct dma_fence_cb *cb)
+{
+ struct invalidation_fence *ifence =
+ container_of(cb, struct invalidation_fence, cb);
+
+ trace_xe_gt_tlb_invalidation_fence_cb(&ifence->base);
+ if (!ifence->fence->error) {
+ queue_work(system_wq, &ifence->work);
+ } else {
+ ifence->base.base.error = ifence->fence->error;
+ dma_fence_signal(&ifence->base.base);
+ dma_fence_put(&ifence->base.base);
+ }
+ dma_fence_put(ifence->fence);
+}
+
+static void invalidation_fence_work_func(struct work_struct *w)
+{
+ struct invalidation_fence *ifence =
+ container_of(w, struct invalidation_fence, work);
+
+ trace_xe_gt_tlb_invalidation_fence_work_func(&ifence->base);
+ xe_gt_tlb_invalidation_vma(ifence->gt, &ifence->base, ifence->vma);
+}
+
+static int invalidation_fence_init(struct xe_gt *gt,
+ struct invalidation_fence *ifence,
+ struct dma_fence *fence,
+ struct xe_vma *vma)
+{
+ int ret;
+
+ trace_xe_gt_tlb_invalidation_fence_create(&ifence->base);
+
+ spin_lock_irq(&gt->tlb_invalidation.lock);
+ dma_fence_init(&ifence->base.base, &invalidation_fence_ops,
+ &gt->tlb_invalidation.lock,
+ gt->tlb_invalidation.fence_context,
+ ++gt->tlb_invalidation.fence_seqno);
+ spin_unlock_irq(&gt->tlb_invalidation.lock);
+
+ INIT_LIST_HEAD(&ifence->base.link);
+
+ dma_fence_get(&ifence->base.base); /* Ref for caller */
+ ifence->fence = fence;
+ ifence->gt = gt;
+ ifence->vma = vma;
+
+ INIT_WORK(&ifence->work, invalidation_fence_work_func);
+ ret = dma_fence_add_callback(fence, &ifence->cb, invalidation_fence_cb);
+ if (ret == -ENOENT) {
+ dma_fence_put(ifence->fence); /* Usually dropped in CB */
+ invalidation_fence_work_func(&ifence->work);
+ } else if (ret) {
+ dma_fence_put(&ifence->base.base); /* Caller ref */
+ dma_fence_put(&ifence->base.base); /* Creation ref */
+ }
+
+ xe_gt_assert(gt, !ret || ret == -ENOENT);
+
+ return ret && ret != -ENOENT ? ret : 0;
+}
+
+static void xe_pt_calc_rfence_interval(struct xe_vma *vma,
+ struct xe_pt_migrate_pt_update *update,
+ struct xe_vm_pgtable_update *entries,
+ u32 num_entries)
+{
+ int i, level = 0;
+
+ for (i = 0; i < num_entries; i++) {
+ const struct xe_vm_pgtable_update *entry = &entries[i];
+
+ if (entry->pt->level > level)
+ level = entry->pt->level;
+ }
+
+ /* Greedy (non-optimal) calculation but simple */
+ update->base.start = ALIGN_DOWN(xe_vma_start(vma),
+ 0x1ull << xe_pt_shift(level));
+ update->base.last = ALIGN(xe_vma_end(vma),
+ 0x1ull << xe_pt_shift(level)) - 1;
+}
+
+/**
+ * __xe_pt_bind_vma() - Build and connect a page-table tree for the vma
+ * address range.
+ * @tile: The tile to bind for.
+ * @vma: The vma to bind.
+ * @q: The exec_queue with which to do pipelined page-table updates.
+ * @syncs: Entries to sync on before binding the built tree to the live vm tree.
+ * @num_syncs: Number of @sync entries.
+ * @rebind: Whether we're rebinding this vma to the same address range without
+ * an unbind in-between.
+ *
+ * This function builds a page-table tree (see xe_pt_stage_bind() for more
+ * information on page-table building), and the xe_vm_pgtable_update entries
+ * abstracting the operations needed to attach it to the main vm tree. It
+ * then takes the relevant locks and updates the metadata side of the main
+ * vm tree and submits the operations for pipelined attachment of the
+ * gpu page-table to the vm main tree, (which can be done either by the
+ * cpu and the GPU).
+ *
+ * Return: A valid dma-fence representing the pipelined attachment operation
+ * on success, an error pointer on error.
+ */
+struct dma_fence *
+__xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool rebind)
+{
+ struct xe_vm_pgtable_update entries[XE_VM_MAX_LEVEL * 2 + 1];
+ struct xe_pt_migrate_pt_update bind_pt_update = {
+ .base = {
+ .ops = xe_vma_is_userptr(vma) ? &userptr_bind_ops : &bind_ops,
+ .vma = vma,
+ .tile_id = tile->id,
+ },
+ .bind = true,
+ };
+ struct xe_vm *vm = xe_vma_vm(vma);
+ u32 num_entries;
+ struct dma_fence *fence;
+ struct invalidation_fence *ifence = NULL;
+ struct xe_range_fence *rfence;
+ int err;
+
+ bind_pt_update.locked = false;
+ xe_bo_assert_held(xe_vma_bo(vma));
+ xe_vm_assert_held(vm);
+
+ vm_dbg(&xe_vma_vm(vma)->xe->drm,
+ "Preparing bind, with range [%llx...%llx) engine %p.\n",
+ xe_vma_start(vma), xe_vma_end(vma), q);
+
+ err = xe_pt_prepare_bind(tile, vma, entries, &num_entries, rebind);
+ if (err)
+ goto err;
+ xe_tile_assert(tile, num_entries <= ARRAY_SIZE(entries));
+
+ xe_vm_dbg_print_entries(tile_to_xe(tile), entries, num_entries);
+ xe_pt_calc_rfence_interval(vma, &bind_pt_update, entries,
+ num_entries);
+
+ /*
+ * If rebind, we have to invalidate TLB on !LR vms to invalidate
+ * cached PTEs point to freed memory. on LR vms this is done
+ * automatically when the context is re-enabled by the rebind worker,
+ * or in fault mode it was invalidated on PTE zapping.
+ *
+ * If !rebind, and scratch enabled VMs, there is a chance the scratch
+ * PTE is already cached in the TLB so it needs to be invalidated.
+ * on !LR VMs this is done in the ring ops preceding a batch, but on
+ * non-faulting LR, in particular on user-space batch buffer chaining,
+ * it needs to be done here.
+ */
+ if ((rebind && !xe_vm_in_lr_mode(vm) && !vm->batch_invalidate_tlb) ||
+ (!rebind && xe_vm_has_scratch(vm) && xe_vm_in_preempt_fence_mode(vm))) {
+ ifence = kzalloc(sizeof(*ifence), GFP_KERNEL);
+ if (!ifence)
+ return ERR_PTR(-ENOMEM);
+ }
+
+ rfence = kzalloc(sizeof(*rfence), GFP_KERNEL);
+ if (!rfence) {
+ kfree(ifence);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ fence = xe_migrate_update_pgtables(tile->migrate,
+ vm, xe_vma_bo(vma), q,
+ entries, num_entries,
+ syncs, num_syncs,
+ &bind_pt_update.base);
+ if (!IS_ERR(fence)) {
+ bool last_munmap_rebind = vma->gpuva.flags & XE_VMA_LAST_REBIND;
+ LLIST_HEAD(deferred);
+ int err;
+
+ err = xe_range_fence_insert(&vm->rftree[tile->id], rfence,
+ &xe_range_fence_kfree_ops,
+ bind_pt_update.base.start,
+ bind_pt_update.base.last, fence);
+ if (err)
+ dma_fence_wait(fence, false);
+
+ /* TLB invalidation must be done before signaling rebind */
+ if (ifence) {
+ int err = invalidation_fence_init(tile->primary_gt, ifence, fence,
+ vma);
+ if (err) {
+ dma_fence_put(fence);
+ kfree(ifence);
+ return ERR_PTR(err);
+ }
+ fence = &ifence->base.base;
+ }
+
+ /* add shared fence now for pagetable delayed destroy */
+ dma_resv_add_fence(xe_vm_resv(vm), fence, !rebind &&
+ last_munmap_rebind ?
+ DMA_RESV_USAGE_KERNEL :
+ DMA_RESV_USAGE_BOOKKEEP);
+
+ if (!xe_vma_has_no_bo(vma) && !xe_vma_bo(vma)->vm)
+ dma_resv_add_fence(xe_vma_bo(vma)->ttm.base.resv, fence,
+ DMA_RESV_USAGE_BOOKKEEP);
+ xe_pt_commit_bind(vma, entries, num_entries, rebind,
+ bind_pt_update.locked ? &deferred : NULL);
+
+ /* This vma is live (again?) now */
+ vma->tile_present |= BIT(tile->id);
+
+ if (bind_pt_update.locked) {
+ vma->userptr.initial_bind = true;
+ up_read(&vm->userptr.notifier_lock);
+ xe_bo_put_commit(&deferred);
+ }
+ if (!rebind && last_munmap_rebind &&
+ xe_vm_in_preempt_fence_mode(vm))
+ xe_vm_queue_rebind_worker(vm);
+ } else {
+ kfree(rfence);
+ kfree(ifence);
+ if (bind_pt_update.locked)
+ up_read(&vm->userptr.notifier_lock);
+ xe_pt_abort_bind(vma, entries, num_entries);
+ }
+
+ return fence;
+
+err:
+ return ERR_PTR(err);
+}
+
+struct xe_pt_stage_unbind_walk {
+ /** @base: The pagewalk base-class. */
+ struct xe_pt_walk base;
+
+ /* Input parameters for the walk */
+ /** @tile: The tile we're unbinding from. */
+ struct xe_tile *tile;
+
+ /**
+ * @modified_start: Walk range start, modified to include any
+ * shared pagetables that we're the only user of and can thus
+ * treat as private.
+ */
+ u64 modified_start;
+ /** @modified_end: Walk range start, modified like @modified_start. */
+ u64 modified_end;
+
+ /* Output */
+ /* @wupd: Structure to track the page-table updates we're building */
+ struct xe_walk_update wupd;
+};
+
+/*
+ * Check whether this range is the only one populating this pagetable,
+ * and in that case, update the walk range checks so that higher levels don't
+ * view us as a shared pagetable.
+ */
+static bool xe_pt_check_kill(u64 addr, u64 next, unsigned int level,
+ const struct xe_pt *child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk)
+{
+ struct xe_pt_stage_unbind_walk *xe_walk =
+ container_of(walk, typeof(*xe_walk), base);
+ unsigned int shift = walk->shifts[level];
+ u64 size = 1ull << shift;
+
+ if (IS_ALIGNED(addr, size) && IS_ALIGNED(next, size) &&
+ ((next - addr) >> shift) == child->num_live) {
+ u64 size = 1ull << walk->shifts[level + 1];
+
+ *action = ACTION_CONTINUE;
+
+ if (xe_walk->modified_start >= addr)
+ xe_walk->modified_start = round_down(addr, size);
+ if (xe_walk->modified_end <= next)
+ xe_walk->modified_end = round_up(next, size);
+
+ return true;
+ }
+
+ return false;
+}
+
+static int xe_pt_stage_unbind_entry(struct xe_ptw *parent, pgoff_t offset,
+ unsigned int level, u64 addr, u64 next,
+ struct xe_ptw **child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk)
+{
+ struct xe_pt *xe_child = container_of(*child, typeof(*xe_child), base);
+
+ XE_WARN_ON(!*child);
+ XE_WARN_ON(!level && xe_child->is_compact);
+
+ xe_pt_check_kill(addr, next, level - 1, xe_child, action, walk);
+
+ return 0;
+}
+
+static int
+xe_pt_stage_unbind_post_descend(struct xe_ptw *parent, pgoff_t offset,
+ unsigned int level, u64 addr, u64 next,
+ struct xe_ptw **child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk)
+{
+ struct xe_pt_stage_unbind_walk *xe_walk =
+ container_of(walk, typeof(*xe_walk), base);
+ struct xe_pt *xe_child = container_of(*child, typeof(*xe_child), base);
+ pgoff_t end_offset;
+ u64 size = 1ull << walk->shifts[--level];
+
+ if (!IS_ALIGNED(addr, size))
+ addr = xe_walk->modified_start;
+ if (!IS_ALIGNED(next, size))
+ next = xe_walk->modified_end;
+
+ /* Parent == *child is the root pt. Don't kill it. */
+ if (parent != *child &&
+ xe_pt_check_kill(addr, next, level, xe_child, action, walk))
+ return 0;
+
+ if (!xe_pt_nonshared_offsets(addr, next, level, walk, action, &offset,
+ &end_offset))
+ return 0;
+
+ (void)xe_pt_new_shared(&xe_walk->wupd, xe_child, offset, false);
+ xe_walk->wupd.updates[level].update->qwords = end_offset - offset;
+
+ return 0;
+}
+
+static const struct xe_pt_walk_ops xe_pt_stage_unbind_ops = {
+ .pt_entry = xe_pt_stage_unbind_entry,
+ .pt_post_descend = xe_pt_stage_unbind_post_descend,
+};
+
+/**
+ * xe_pt_stage_unbind() - Build page-table update structures for an unbind
+ * operation
+ * @tile: The tile we're unbinding for.
+ * @vma: The vma we're unbinding.
+ * @entries: Caller-provided storage for the update structures.
+ *
+ * Builds page-table update structures for an unbind operation. The function
+ * will attempt to remove all page-tables that we're the only user
+ * of, and for that to work, the unbind operation must be committed in the
+ * same critical section that blocks racing binds to the same page-table tree.
+ *
+ * Return: The number of entries used.
+ */
+static unsigned int xe_pt_stage_unbind(struct xe_tile *tile, struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries)
+{
+ struct xe_pt_stage_unbind_walk xe_walk = {
+ .base = {
+ .ops = &xe_pt_stage_unbind_ops,
+ .shifts = xe_normal_pt_shifts,
+ .max_level = XE_PT_HIGHEST_LEVEL,
+ },
+ .tile = tile,
+ .modified_start = xe_vma_start(vma),
+ .modified_end = xe_vma_end(vma),
+ .wupd.entries = entries,
+ };
+ struct xe_pt *pt = xe_vma_vm(vma)->pt_root[tile->id];
+
+ (void)xe_pt_walk_shared(&pt->base, pt->level, xe_vma_start(vma),
+ xe_vma_end(vma), &xe_walk.base);
+
+ return xe_walk.wupd.num_used_entries;
+}
+
+static void
+xe_migrate_clear_pgtable_callback(struct xe_migrate_pt_update *pt_update,
+ struct xe_tile *tile, struct iosys_map *map,
+ void *ptr, u32 qword_ofs, u32 num_qwords,
+ const struct xe_vm_pgtable_update *update)
+{
+ struct xe_vma *vma = pt_update->vma;
+ u64 empty = __xe_pt_empty_pte(tile, xe_vma_vm(vma), update->pt->level);
+ int i;
+
+ if (map && map->is_iomem)
+ for (i = 0; i < num_qwords; ++i)
+ xe_map_wr(tile_to_xe(tile), map, (qword_ofs + i) *
+ sizeof(u64), u64, empty);
+ else if (map)
+ memset64(map->vaddr + qword_ofs * sizeof(u64), empty,
+ num_qwords);
+ else
+ memset64(ptr, empty, num_qwords);
+}
+
+static void
+xe_pt_commit_unbind(struct xe_vma *vma,
+ struct xe_vm_pgtable_update *entries, u32 num_entries,
+ struct llist_head *deferred)
+{
+ u32 j;
+
+ xe_pt_commit_locks_assert(vma);
+
+ for (j = 0; j < num_entries; ++j) {
+ struct xe_vm_pgtable_update *entry = &entries[j];
+ struct xe_pt *pt = entry->pt;
+
+ pt->num_live -= entry->qwords;
+ if (pt->level) {
+ struct xe_pt_dir *pt_dir = as_xe_pt_dir(pt);
+ u32 i;
+
+ for (i = entry->ofs; i < entry->ofs + entry->qwords;
+ i++) {
+ if (xe_pt_entry(pt_dir, i))
+ xe_pt_destroy(xe_pt_entry(pt_dir, i),
+ xe_vma_vm(vma)->flags, deferred);
+
+ pt_dir->dir.entries[i] = NULL;
+ }
+ }
+ }
+}
+
+static const struct xe_migrate_pt_update_ops unbind_ops = {
+ .populate = xe_migrate_clear_pgtable_callback,
+ .pre_commit = xe_pt_pre_commit,
+};
+
+static const struct xe_migrate_pt_update_ops userptr_unbind_ops = {
+ .populate = xe_migrate_clear_pgtable_callback,
+ .pre_commit = xe_pt_userptr_pre_commit,
+};
+
+/**
+ * __xe_pt_unbind_vma() - Disconnect and free a page-table tree for the vma
+ * address range.
+ * @tile: The tile to unbind for.
+ * @vma: The vma to unbind.
+ * @q: The exec_queue with which to do pipelined page-table updates.
+ * @syncs: Entries to sync on before disconnecting the tree to be destroyed.
+ * @num_syncs: Number of @sync entries.
+ *
+ * This function builds a the xe_vm_pgtable_update entries abstracting the
+ * operations needed to detach the page-table tree to be destroyed from the
+ * man vm tree.
+ * It then takes the relevant locks and submits the operations for
+ * pipelined detachment of the gpu page-table from the vm main tree,
+ * (which can be done either by the cpu and the GPU), Finally it frees the
+ * detached page-table tree.
+ *
+ * Return: A valid dma-fence representing the pipelined detachment operation
+ * on success, an error pointer on error.
+ */
+struct dma_fence *
+__xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs)
+{
+ struct xe_vm_pgtable_update entries[XE_VM_MAX_LEVEL * 2 + 1];
+ struct xe_pt_migrate_pt_update unbind_pt_update = {
+ .base = {
+ .ops = xe_vma_is_userptr(vma) ? &userptr_unbind_ops :
+ &unbind_ops,
+ .vma = vma,
+ .tile_id = tile->id,
+ },
+ };
+ struct xe_vm *vm = xe_vma_vm(vma);
+ u32 num_entries;
+ struct dma_fence *fence = NULL;
+ struct invalidation_fence *ifence;
+ struct xe_range_fence *rfence;
+
+ LLIST_HEAD(deferred);
+
+ xe_bo_assert_held(xe_vma_bo(vma));
+ xe_vm_assert_held(vm);
+
+ vm_dbg(&xe_vma_vm(vma)->xe->drm,
+ "Preparing unbind, with range [%llx...%llx) engine %p.\n",
+ xe_vma_start(vma), xe_vma_end(vma), q);
+
+ num_entries = xe_pt_stage_unbind(tile, vma, entries);
+ xe_tile_assert(tile, num_entries <= ARRAY_SIZE(entries));
+
+ xe_vm_dbg_print_entries(tile_to_xe(tile), entries, num_entries);
+ xe_pt_calc_rfence_interval(vma, &unbind_pt_update, entries,
+ num_entries);
+
+ ifence = kzalloc(sizeof(*ifence), GFP_KERNEL);
+ if (!ifence)
+ return ERR_PTR(-ENOMEM);
+
+ rfence = kzalloc(sizeof(*rfence), GFP_KERNEL);
+ if (!rfence) {
+ kfree(ifence);
+ return ERR_PTR(-ENOMEM);
+ }
+
+ /*
+ * Even if we were already evicted and unbind to destroy, we need to
+ * clear again here. The eviction may have updated pagetables at a
+ * lower level, because it needs to be more conservative.
+ */
+ fence = xe_migrate_update_pgtables(tile->migrate,
+ vm, NULL, q ? q :
+ vm->q[tile->id],
+ entries, num_entries,
+ syncs, num_syncs,
+ &unbind_pt_update.base);
+ if (!IS_ERR(fence)) {
+ int err;
+
+ err = xe_range_fence_insert(&vm->rftree[tile->id], rfence,
+ &xe_range_fence_kfree_ops,
+ unbind_pt_update.base.start,
+ unbind_pt_update.base.last, fence);
+ if (err)
+ dma_fence_wait(fence, false);
+
+ /* TLB invalidation must be done before signaling unbind */
+ err = invalidation_fence_init(tile->primary_gt, ifence, fence, vma);
+ if (err) {
+ dma_fence_put(fence);
+ kfree(ifence);
+ return ERR_PTR(err);
+ }
+ fence = &ifence->base.base;
+
+ /* add shared fence now for pagetable delayed destroy */
+ dma_resv_add_fence(xe_vm_resv(vm), fence,
+ DMA_RESV_USAGE_BOOKKEEP);
+
+ /* This fence will be installed by caller when doing eviction */
+ if (!xe_vma_has_no_bo(vma) && !xe_vma_bo(vma)->vm)
+ dma_resv_add_fence(xe_vma_bo(vma)->ttm.base.resv, fence,
+ DMA_RESV_USAGE_BOOKKEEP);
+ xe_pt_commit_unbind(vma, entries, num_entries,
+ unbind_pt_update.locked ? &deferred : NULL);
+ vma->tile_present &= ~BIT(tile->id);
+ } else {
+ kfree(rfence);
+ kfree(ifence);
+ }
+
+ if (!vma->tile_present)
+ list_del_init(&vma->combined_links.rebind);
+
+ if (unbind_pt_update.locked) {
+ xe_tile_assert(tile, xe_vma_is_userptr(vma));
+
+ if (!vma->tile_present) {
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_del_init(&vma->userptr.invalidate_link);
+ spin_unlock(&vm->userptr.invalidated_lock);
+ }
+ up_read(&vm->userptr.notifier_lock);
+ xe_bo_put_commit(&deferred);
+ }
+
+ return fence;
+}
diff --git a/drivers/gpu/drm/xe/xe_pt.h b/drivers/gpu/drm/xe/xe_pt.h
new file mode 100644
index 000000000000..71a4fbfcff43
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pt.h
@@ -0,0 +1,48 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+#ifndef _XE_PT_H_
+#define _XE_PT_H_
+
+#include <linux/types.h>
+
+#include "xe_pt_types.h"
+
+struct dma_fence;
+struct xe_bo;
+struct xe_device;
+struct xe_exec_queue;
+struct xe_sync_entry;
+struct xe_tile;
+struct xe_vm;
+struct xe_vma;
+
+/* Largest huge pte is currently 1GiB. May become device dependent. */
+#define MAX_HUGEPTE_LEVEL 2
+
+#define xe_pt_write(xe, map, idx, data) \
+ xe_map_wr(xe, map, (idx) * sizeof(u64), u64, data)
+
+unsigned int xe_pt_shift(unsigned int level);
+
+struct xe_pt *xe_pt_create(struct xe_vm *vm, struct xe_tile *tile,
+ unsigned int level);
+
+void xe_pt_populate_empty(struct xe_tile *tile, struct xe_vm *vm,
+ struct xe_pt *pt);
+
+void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred);
+
+struct dma_fence *
+__xe_pt_bind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool rebind);
+
+struct dma_fence *
+__xe_pt_unbind_vma(struct xe_tile *tile, struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs);
+
+bool xe_pt_zap_ptes(struct xe_tile *tile, struct xe_vma *vma);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pt_types.h b/drivers/gpu/drm/xe/xe_pt_types.h
new file mode 100644
index 000000000000..cee70cb0f014
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pt_types.h
@@ -0,0 +1,77 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_PT_TYPES_H_
+#define _XE_PT_TYPES_H_
+
+#include <linux/types.h>
+
+#include "xe_pt_walk.h"
+
+struct xe_bo;
+struct xe_device;
+struct xe_vma;
+
+enum xe_cache_level {
+ XE_CACHE_NONE,
+ XE_CACHE_WT,
+ XE_CACHE_WB,
+ XE_CACHE_NONE_COMPRESSION, /*UC + COH_NONE + COMPRESSION */
+ __XE_CACHE_LEVEL_COUNT,
+};
+
+#define XE_VM_MAX_LEVEL 4
+
+struct xe_pt {
+ struct xe_ptw base;
+ struct xe_bo *bo;
+ unsigned int level;
+ unsigned int num_live;
+ bool rebind;
+ bool is_compact;
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
+ /** addr: Virtual address start address of the PT. */
+ u64 addr;
+#endif
+};
+
+struct xe_pt_ops {
+ u64 (*pte_encode_bo)(struct xe_bo *bo, u64 bo_offset,
+ u16 pat_index, u32 pt_level);
+ u64 (*pte_encode_vma)(u64 pte, struct xe_vma *vma,
+ u16 pat_index, u32 pt_level);
+ u64 (*pte_encode_addr)(struct xe_device *xe, u64 addr,
+ u16 pat_index,
+ u32 pt_level, bool devmem, u64 flags);
+ u64 (*pde_encode_bo)(struct xe_bo *bo, u64 bo_offset,
+ u16 pat_index);
+};
+
+struct xe_pt_entry {
+ struct xe_pt *pt;
+ u64 pte;
+};
+
+struct xe_vm_pgtable_update {
+ /** @bo: page table bo to write to */
+ struct xe_bo *pt_bo;
+
+ /** @ofs: offset inside this PTE to begin writing to (in qwords) */
+ u32 ofs;
+
+ /** @qwords: number of PTE's to write */
+ u32 qwords;
+
+ /** @pt: opaque pointer useful for the caller of xe_migrate_update_pgtables */
+ struct xe_pt *pt;
+
+ /** @pt_entries: Newly added pagetable entries */
+ struct xe_pt_entry *pt_entries;
+
+ /** @flags: Target flags */
+ u32 flags;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_pt_walk.c b/drivers/gpu/drm/xe/xe_pt_walk.c
new file mode 100644
index 000000000000..8f6c8d063f39
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pt_walk.c
@@ -0,0 +1,160 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+#include "xe_pt_walk.h"
+
+/**
+ * DOC: GPU page-table tree walking.
+ * The utilities in this file are similar to the CPU page-table walk
+ * utilities in mm/pagewalk.c. The main difference is that we distinguish
+ * the various levels of a page-table tree with an unsigned integer rather
+ * than by name. 0 is the lowest level, and page-tables with level 0 can
+ * not be directories pointing to lower levels, whereas all other levels
+ * can. The user of the utilities determines the highest level.
+ *
+ * Nomenclature:
+ * Each struct xe_ptw, regardless of level is referred to as a page table, and
+ * multiple page tables typically form a page table tree with page tables at
+ * intermediate levels being page directories pointing at page tables at lower
+ * levels. A shared page table for a given address range is a page-table which
+ * is neither fully within nor fully outside the address range and that can
+ * thus be shared by two or more address ranges.
+ *
+ * Please keep this code generic so that it can used as a drm-wide page-
+ * table walker should other drivers find use for it.
+ */
+static u64 xe_pt_addr_end(u64 addr, u64 end, unsigned int level,
+ const struct xe_pt_walk *walk)
+{
+ u64 size = 1ull << walk->shifts[level];
+ u64 tmp = round_up(addr + 1, size);
+
+ return min_t(u64, tmp, end);
+}
+
+static bool xe_pt_next(pgoff_t *offset, u64 *addr, u64 next, u64 end,
+ unsigned int level, const struct xe_pt_walk *walk)
+{
+ pgoff_t step = 1;
+
+ /* Shared pt walk skips to the last pagetable */
+ if (unlikely(walk->shared_pt_mode)) {
+ unsigned int shift = walk->shifts[level];
+ u64 skip_to = round_down(end, 1ull << shift);
+
+ if (skip_to > next) {
+ step += (skip_to - next) >> shift;
+ next = skip_to;
+ }
+ }
+
+ *addr = next;
+ *offset += step;
+
+ return next != end;
+}
+
+/**
+ * xe_pt_walk_range() - Walk a range of a gpu page table tree with callbacks
+ * for each page-table entry in all levels.
+ * @parent: The root page table for walk start.
+ * @level: The root page table level.
+ * @addr: Virtual address start.
+ * @end: Virtual address end + 1.
+ * @walk: Walk info.
+ *
+ * Similar to the CPU page-table walker, this is a helper to walk
+ * a gpu page table and call a provided callback function for each entry.
+ *
+ * Return: 0 on success, negative error code on error. The error is
+ * propagated from the callback and on error the walk is terminated.
+ */
+int xe_pt_walk_range(struct xe_ptw *parent, unsigned int level,
+ u64 addr, u64 end, struct xe_pt_walk *walk)
+{
+ pgoff_t offset = xe_pt_offset(addr, level, walk);
+ struct xe_ptw **entries = parent->dir ? parent->dir->entries : NULL;
+ const struct xe_pt_walk_ops *ops = walk->ops;
+ enum page_walk_action action;
+ struct xe_ptw *child;
+ int err = 0;
+ u64 next;
+
+ do {
+ next = xe_pt_addr_end(addr, end, level, walk);
+ if (walk->shared_pt_mode && xe_pt_covers(addr, next, level,
+ walk))
+ continue;
+again:
+ action = ACTION_SUBTREE;
+ child = entries ? entries[offset] : NULL;
+ err = ops->pt_entry(parent, offset, level, addr, next,
+ &child, &action, walk);
+ if (err)
+ break;
+
+ /* Probably not needed yet for gpu pagetable walk. */
+ if (unlikely(action == ACTION_AGAIN))
+ goto again;
+
+ if (likely(!level || !child || action == ACTION_CONTINUE))
+ continue;
+
+ err = xe_pt_walk_range(child, level - 1, addr, next, walk);
+
+ if (!err && ops->pt_post_descend)
+ err = ops->pt_post_descend(parent, offset, level, addr,
+ next, &child, &action, walk);
+ if (err)
+ break;
+
+ } while (xe_pt_next(&offset, &addr, next, end, level, walk));
+
+ return err;
+}
+
+/**
+ * xe_pt_walk_shared() - Walk shared page tables of a page-table tree.
+ * @parent: Root page table directory.
+ * @level: Level of the root.
+ * @addr: Start address.
+ * @end: Last address + 1.
+ * @walk: Walk info.
+ *
+ * This function is similar to xe_pt_walk_range() but it skips page tables
+ * that are private to the range. Since the root (or @parent) page table is
+ * typically also a shared page table this function is different in that it
+ * calls the pt_entry callback and the post_descend callback also for the
+ * root. The root can be detected in the callbacks by checking whether
+ * parent == *child.
+ * Walking only the shared page tables is common for unbind-type operations
+ * where the page-table entries for an address range are cleared or detached
+ * from the main page-table tree.
+ *
+ * Return: 0 on success, negative error code on error: If a callback
+ * returns an error, the walk will be terminated and the error returned by
+ * this function.
+ */
+int xe_pt_walk_shared(struct xe_ptw *parent, unsigned int level,
+ u64 addr, u64 end, struct xe_pt_walk *walk)
+{
+ const struct xe_pt_walk_ops *ops = walk->ops;
+ enum page_walk_action action = ACTION_SUBTREE;
+ struct xe_ptw *child = parent;
+ int err;
+
+ walk->shared_pt_mode = true;
+ err = walk->ops->pt_entry(parent, 0, level + 1, addr, end,
+ &child, &action, walk);
+
+ if (err || action != ACTION_SUBTREE)
+ return err;
+
+ err = xe_pt_walk_range(parent, level, addr, end, walk);
+ if (!err && ops->pt_post_descend) {
+ err = ops->pt_post_descend(parent, 0, level + 1, addr, end,
+ &child, &action, walk);
+ }
+ return err;
+}
diff --git a/drivers/gpu/drm/xe/xe_pt_walk.h b/drivers/gpu/drm/xe/xe_pt_walk.h
new file mode 100644
index 000000000000..ec3d1e9efa6d
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_pt_walk.h
@@ -0,0 +1,161 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+#ifndef __XE_PT_WALK__
+#define __XE_PT_WALK__
+
+#include <linux/pagewalk.h>
+#include <linux/types.h>
+
+struct xe_ptw_dir;
+
+/**
+ * struct xe_ptw - base class for driver pagetable subclassing.
+ * @dir: Pointer to an array of children if any.
+ *
+ * Drivers could subclass this, and if it's a page-directory, typically
+ * embed the xe_ptw_dir::entries array in the same allocation.
+ */
+struct xe_ptw {
+ struct xe_ptw_dir *dir;
+};
+
+/**
+ * struct xe_ptw_dir - page directory structure
+ * @entries: Array holding page directory children.
+ *
+ * It is the responsibility of the user to ensure @entries is
+ * correctly sized.
+ */
+struct xe_ptw_dir {
+ struct xe_ptw *entries[0];
+};
+
+/**
+ * struct xe_pt_walk - Embeddable struct for walk parameters
+ */
+struct xe_pt_walk {
+ /** @ops: The walk ops used for the pagewalk */
+ const struct xe_pt_walk_ops *ops;
+ /**
+ * @shifts: Array of page-table entry shifts used for the
+ * different levels, starting out with the leaf level 0
+ * page-shift as the first entry. It's legal for this pointer to be
+ * changed during the walk.
+ */
+ const u64 *shifts;
+ /** @max_level: Highest populated level in @sizes */
+ unsigned int max_level;
+ /**
+ * @shared_pt_mode: Whether to skip all entries that are private
+ * to the address range and called only for entries that are
+ * shared with other address ranges. Such entries are referred to
+ * as shared pagetables.
+ */
+ bool shared_pt_mode;
+};
+
+/**
+ * typedef xe_pt_entry_fn - gpu page-table-walk callback-function
+ * @parent: The parent page.table.
+ * @offset: The offset (number of entries) into the page table.
+ * @level: The level of @parent.
+ * @addr: The virtual address.
+ * @next: The virtual address for the next call, or end address.
+ * @child: Pointer to pointer to child page-table at this @offset. The
+ * function may modify the value pointed to if, for example, allocating a
+ * child page table.
+ * @action: The walk action to take upon return. See <linux/pagewalk.h>.
+ * @walk: The walk parameters.
+ */
+typedef int (*xe_pt_entry_fn)(struct xe_ptw *parent, pgoff_t offset,
+ unsigned int level, u64 addr, u64 next,
+ struct xe_ptw **child,
+ enum page_walk_action *action,
+ struct xe_pt_walk *walk);
+
+/**
+ * struct xe_pt_walk_ops - Walk callbacks.
+ */
+struct xe_pt_walk_ops {
+ /**
+ * @pt_entry: Callback to be called for each page table entry prior
+ * to descending to the next level. The returned value of the action
+ * function parameter is honored.
+ */
+ xe_pt_entry_fn pt_entry;
+ /**
+ * @pt_post_descend: Callback to be called for each page table entry
+ * after return from descending to the next level. The returned value
+ * of the action function parameter is ignored.
+ */
+ xe_pt_entry_fn pt_post_descend;
+};
+
+int xe_pt_walk_range(struct xe_ptw *parent, unsigned int level,
+ u64 addr, u64 end, struct xe_pt_walk *walk);
+
+int xe_pt_walk_shared(struct xe_ptw *parent, unsigned int level,
+ u64 addr, u64 end, struct xe_pt_walk *walk);
+
+/**
+ * xe_pt_covers - Whether the address range covers an entire entry in @level
+ * @addr: Start of the range.
+ * @end: End of range + 1.
+ * @level: Page table level.
+ * @walk: Page table walk info.
+ *
+ * This function is a helper to aid in determining whether a leaf page table
+ * entry can be inserted at this @level.
+ *
+ * Return: Whether the range provided covers exactly an entry at this level.
+ */
+static inline bool xe_pt_covers(u64 addr, u64 end, unsigned int level,
+ const struct xe_pt_walk *walk)
+{
+ u64 pt_size = 1ull << walk->shifts[level];
+
+ return end - addr == pt_size && IS_ALIGNED(addr, pt_size);
+}
+
+/**
+ * xe_pt_num_entries: Number of page-table entries of a given range at this
+ * level
+ * @addr: Start address.
+ * @end: End address.
+ * @level: Page table level.
+ * @walk: Walk info.
+ *
+ * Return: The number of page table entries at this level between @start and
+ * @end.
+ */
+static inline pgoff_t
+xe_pt_num_entries(u64 addr, u64 end, unsigned int level,
+ const struct xe_pt_walk *walk)
+{
+ u64 pt_size = 1ull << walk->shifts[level];
+
+ return (round_up(end, pt_size) - round_down(addr, pt_size)) >>
+ walk->shifts[level];
+}
+
+/**
+ * xe_pt_offset: Offset of the page-table entry for a given address.
+ * @addr: The address.
+ * @level: Page table level.
+ * @walk: Walk info.
+ *
+ * Return: The page table entry offset for the given address in a
+ * page table with size indicated by @level.
+ */
+static inline pgoff_t
+xe_pt_offset(u64 addr, unsigned int level, const struct xe_pt_walk *walk)
+{
+ if (level < walk->max_level)
+ addr &= ((1ull << walk->shifts[level + 1]) - 1);
+
+ return addr >> walk->shifts[level];
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_query.c b/drivers/gpu/drm/xe/xe_query.c
new file mode 100644
index 000000000000..9b35673b286c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_query.c
@@ -0,0 +1,552 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_query.h"
+
+#include <linux/nospec.h>
+#include <linux/sched/clock.h>
+
+#include <drm/ttm/ttm_placement.h>
+#include <drm/xe_drm.h>
+
+#include "regs/xe_engine_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_guc_hwconfig.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+#include "xe_ttm_vram_mgr.h"
+
+static const u16 xe_to_user_engine_class[] = {
+ [XE_ENGINE_CLASS_RENDER] = DRM_XE_ENGINE_CLASS_RENDER,
+ [XE_ENGINE_CLASS_COPY] = DRM_XE_ENGINE_CLASS_COPY,
+ [XE_ENGINE_CLASS_VIDEO_DECODE] = DRM_XE_ENGINE_CLASS_VIDEO_DECODE,
+ [XE_ENGINE_CLASS_VIDEO_ENHANCE] = DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ [XE_ENGINE_CLASS_COMPUTE] = DRM_XE_ENGINE_CLASS_COMPUTE,
+};
+
+static const enum xe_engine_class user_to_xe_engine_class[] = {
+ [DRM_XE_ENGINE_CLASS_RENDER] = XE_ENGINE_CLASS_RENDER,
+ [DRM_XE_ENGINE_CLASS_COPY] = XE_ENGINE_CLASS_COPY,
+ [DRM_XE_ENGINE_CLASS_VIDEO_DECODE] = XE_ENGINE_CLASS_VIDEO_DECODE,
+ [DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE] = XE_ENGINE_CLASS_VIDEO_ENHANCE,
+ [DRM_XE_ENGINE_CLASS_COMPUTE] = XE_ENGINE_CLASS_COMPUTE,
+};
+
+static size_t calc_hw_engine_info_size(struct xe_device *xe)
+{
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ struct xe_gt *gt;
+ u8 gt_id;
+ int i = 0;
+
+ for_each_gt(gt, xe, gt_id)
+ for_each_hw_engine(hwe, gt, id) {
+ if (xe_hw_engine_is_reserved(hwe))
+ continue;
+ i++;
+ }
+
+ return sizeof(struct drm_xe_query_engines) +
+ i * sizeof(struct drm_xe_engine);
+}
+
+typedef u64 (*__ktime_func_t)(void);
+static __ktime_func_t __clock_id_to_func(clockid_t clk_id)
+{
+ /*
+ * Use logic same as the perf subsystem to allow user to select the
+ * reference clock id to be used for timestamps.
+ */
+ switch (clk_id) {
+ case CLOCK_MONOTONIC:
+ return &ktime_get_ns;
+ case CLOCK_MONOTONIC_RAW:
+ return &ktime_get_raw_ns;
+ case CLOCK_REALTIME:
+ return &ktime_get_real_ns;
+ case CLOCK_BOOTTIME:
+ return &ktime_get_boottime_ns;
+ case CLOCK_TAI:
+ return &ktime_get_clocktai_ns;
+ default:
+ return NULL;
+ }
+}
+
+static void
+__read_timestamps(struct xe_gt *gt,
+ struct xe_reg lower_reg,
+ struct xe_reg upper_reg,
+ u64 *engine_ts,
+ u64 *cpu_ts,
+ u64 *cpu_delta,
+ __ktime_func_t cpu_clock)
+{
+ u32 upper, lower, old_upper, loop = 0;
+
+ upper = xe_mmio_read32(gt, upper_reg);
+ do {
+ *cpu_delta = local_clock();
+ *cpu_ts = cpu_clock();
+ lower = xe_mmio_read32(gt, lower_reg);
+ *cpu_delta = local_clock() - *cpu_delta;
+ old_upper = upper;
+ upper = xe_mmio_read32(gt, upper_reg);
+ } while (upper != old_upper && loop++ < 2);
+
+ *engine_ts = (u64)upper << 32 | lower;
+}
+
+static int
+query_engine_cycles(struct xe_device *xe,
+ struct drm_xe_device_query *query)
+{
+ struct drm_xe_query_engine_cycles __user *query_ptr;
+ struct drm_xe_engine_class_instance *eci;
+ struct drm_xe_query_engine_cycles resp;
+ size_t size = sizeof(resp);
+ __ktime_func_t cpu_clock;
+ struct xe_hw_engine *hwe;
+ struct xe_gt *gt;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ query_ptr = u64_to_user_ptr(query->data);
+ if (copy_from_user(&resp, query_ptr, size))
+ return -EFAULT;
+
+ cpu_clock = __clock_id_to_func(resp.clockid);
+ if (!cpu_clock)
+ return -EINVAL;
+
+ eci = &resp.eci;
+ if (eci->gt_id > XE_MAX_GT_PER_TILE)
+ return -EINVAL;
+
+ gt = xe_device_get_gt(xe, eci->gt_id);
+ if (!gt)
+ return -EINVAL;
+
+ if (eci->engine_class >= ARRAY_SIZE(user_to_xe_engine_class))
+ return -EINVAL;
+
+ hwe = xe_gt_hw_engine(gt, user_to_xe_engine_class[eci->engine_class],
+ eci->engine_instance, true);
+ if (!hwe)
+ return -EINVAL;
+
+ xe_device_mem_access_get(xe);
+ xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+
+ __read_timestamps(gt,
+ RING_TIMESTAMP(hwe->mmio_base),
+ RING_TIMESTAMP_UDW(hwe->mmio_base),
+ &resp.engine_cycles,
+ &resp.cpu_timestamp,
+ &resp.cpu_delta,
+ cpu_clock);
+
+ xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ xe_device_mem_access_put(xe);
+ resp.width = 36;
+
+ /* Only write to the output fields of user query */
+ if (put_user(resp.cpu_timestamp, &query_ptr->cpu_timestamp))
+ return -EFAULT;
+
+ if (put_user(resp.cpu_delta, &query_ptr->cpu_delta))
+ return -EFAULT;
+
+ if (put_user(resp.engine_cycles, &query_ptr->engine_cycles))
+ return -EFAULT;
+
+ if (put_user(resp.width, &query_ptr->width))
+ return -EFAULT;
+
+ return 0;
+}
+
+static int query_engines(struct xe_device *xe,
+ struct drm_xe_device_query *query)
+{
+ size_t size = calc_hw_engine_info_size(xe);
+ struct drm_xe_query_engines __user *query_ptr =
+ u64_to_user_ptr(query->data);
+ struct drm_xe_query_engines *engines;
+ struct xe_hw_engine *hwe;
+ enum xe_hw_engine_id id;
+ struct xe_gt *gt;
+ u8 gt_id;
+ int i = 0;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ engines = kmalloc(size, GFP_KERNEL);
+ if (!engines)
+ return -ENOMEM;
+
+ for_each_gt(gt, xe, gt_id)
+ for_each_hw_engine(hwe, gt, id) {
+ if (xe_hw_engine_is_reserved(hwe))
+ continue;
+
+ engines->engines[i].instance.engine_class =
+ xe_to_user_engine_class[hwe->class];
+ engines->engines[i].instance.engine_instance =
+ hwe->logical_instance;
+ engines->engines[i].instance.gt_id = gt->info.id;
+ engines->engines[i].instance.pad = 0;
+ memset(engines->engines[i].reserved, 0,
+ sizeof(engines->engines[i].reserved));
+
+ i++;
+ }
+
+ engines->pad = 0;
+ engines->num_engines = i;
+
+ if (copy_to_user(query_ptr, engines, size)) {
+ kfree(engines);
+ return -EFAULT;
+ }
+ kfree(engines);
+
+ return 0;
+}
+
+static size_t calc_mem_regions_size(struct xe_device *xe)
+{
+ u32 num_managers = 1;
+ int i;
+
+ for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i)
+ if (ttm_manager_type(&xe->ttm, i))
+ num_managers++;
+
+ return offsetof(struct drm_xe_query_mem_regions, mem_regions[num_managers]);
+}
+
+static int query_mem_regions(struct xe_device *xe,
+ struct drm_xe_device_query *query)
+{
+ size_t size = calc_mem_regions_size(xe);
+ struct drm_xe_query_mem_regions *mem_regions;
+ struct drm_xe_query_mem_regions __user *query_ptr =
+ u64_to_user_ptr(query->data);
+ struct ttm_resource_manager *man;
+ int ret, i;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ mem_regions = kzalloc(size, GFP_KERNEL);
+ if (XE_IOCTL_DBG(xe, !mem_regions))
+ return -ENOMEM;
+
+ man = ttm_manager_type(&xe->ttm, XE_PL_TT);
+ mem_regions->mem_regions[0].mem_class = DRM_XE_MEM_REGION_CLASS_SYSMEM;
+ /*
+ * The instance needs to be a unique number that represents the index
+ * in the placement mask used at xe_gem_create_ioctl() for the
+ * xe_bo_create() placement.
+ */
+ mem_regions->mem_regions[0].instance = 0;
+ mem_regions->mem_regions[0].min_page_size = PAGE_SIZE;
+ mem_regions->mem_regions[0].total_size = man->size << PAGE_SHIFT;
+ if (perfmon_capable())
+ mem_regions->mem_regions[0].used = ttm_resource_manager_usage(man);
+ mem_regions->num_mem_regions = 1;
+
+ for (i = XE_PL_VRAM0; i <= XE_PL_VRAM1; ++i) {
+ man = ttm_manager_type(&xe->ttm, i);
+ if (man) {
+ mem_regions->mem_regions[mem_regions->num_mem_regions].mem_class =
+ DRM_XE_MEM_REGION_CLASS_VRAM;
+ mem_regions->mem_regions[mem_regions->num_mem_regions].instance =
+ mem_regions->num_mem_regions;
+ mem_regions->mem_regions[mem_regions->num_mem_regions].min_page_size =
+ xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ?
+ SZ_64K : PAGE_SIZE;
+ mem_regions->mem_regions[mem_regions->num_mem_regions].total_size =
+ man->size;
+
+ if (perfmon_capable()) {
+ xe_ttm_vram_get_used(man,
+ &mem_regions->mem_regions
+ [mem_regions->num_mem_regions].used,
+ &mem_regions->mem_regions
+ [mem_regions->num_mem_regions].cpu_visible_used);
+ }
+
+ mem_regions->mem_regions[mem_regions->num_mem_regions].cpu_visible_size =
+ xe_ttm_vram_get_cpu_visible_size(man);
+ mem_regions->num_mem_regions++;
+ }
+ }
+
+ if (!copy_to_user(query_ptr, mem_regions, size))
+ ret = 0;
+ else
+ ret = -ENOSPC;
+
+ kfree(mem_regions);
+ return ret;
+}
+
+static int query_config(struct xe_device *xe, struct drm_xe_device_query *query)
+{
+ const u32 num_params = DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY + 1;
+ size_t size =
+ sizeof(struct drm_xe_query_config) + num_params * sizeof(u64);
+ struct drm_xe_query_config __user *query_ptr =
+ u64_to_user_ptr(query->data);
+ struct drm_xe_query_config *config;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ config = kzalloc(size, GFP_KERNEL);
+ if (!config)
+ return -ENOMEM;
+
+ config->num_params = num_params;
+ config->info[DRM_XE_QUERY_CONFIG_REV_AND_DEVICE_ID] =
+ xe->info.devid | (xe->info.revid << 16);
+ if (xe_device_get_root_tile(xe)->mem.vram.usable_size)
+ config->info[DRM_XE_QUERY_CONFIG_FLAGS] =
+ DRM_XE_QUERY_CONFIG_FLAG_HAS_VRAM;
+ config->info[DRM_XE_QUERY_CONFIG_MIN_ALIGNMENT] =
+ xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ? SZ_64K : SZ_4K;
+ config->info[DRM_XE_QUERY_CONFIG_VA_BITS] = xe->info.va_bits;
+ config->info[DRM_XE_QUERY_CONFIG_MAX_EXEC_QUEUE_PRIORITY] =
+ xe_exec_queue_device_get_max_priority(xe);
+
+ if (copy_to_user(query_ptr, config, size)) {
+ kfree(config);
+ return -EFAULT;
+ }
+ kfree(config);
+
+ return 0;
+}
+
+static int query_gt_list(struct xe_device *xe, struct drm_xe_device_query *query)
+{
+ struct xe_gt *gt;
+ size_t size = sizeof(struct drm_xe_query_gt_list) +
+ xe->info.gt_count * sizeof(struct drm_xe_gt);
+ struct drm_xe_query_gt_list __user *query_ptr =
+ u64_to_user_ptr(query->data);
+ struct drm_xe_query_gt_list *gt_list;
+ u8 id;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ gt_list = kzalloc(size, GFP_KERNEL);
+ if (!gt_list)
+ return -ENOMEM;
+
+ gt_list->num_gt = xe->info.gt_count;
+
+ for_each_gt(gt, xe, id) {
+ if (xe_gt_is_media_type(gt))
+ gt_list->gt_list[id].type = DRM_XE_QUERY_GT_TYPE_MEDIA;
+ else
+ gt_list->gt_list[id].type = DRM_XE_QUERY_GT_TYPE_MAIN;
+ gt_list->gt_list[id].tile_id = gt_to_tile(gt)->id;
+ gt_list->gt_list[id].gt_id = gt->info.id;
+ gt_list->gt_list[id].reference_clock = gt->info.reference_clock;
+ /*
+ * The mem_regions indexes in the mask below need to
+ * directly identify the struct
+ * drm_xe_query_mem_regions' instance constructed at
+ * query_mem_regions()
+ *
+ * For our current platforms:
+ * Bit 0 -> System Memory
+ * Bit 1 -> VRAM0 on Tile0
+ * Bit 2 -> VRAM1 on Tile1
+ * However the uAPI is generic and it's userspace's
+ * responsibility to check the mem_class, without any
+ * assumption.
+ */
+ if (!IS_DGFX(xe))
+ gt_list->gt_list[id].near_mem_regions = 0x1;
+ else
+ gt_list->gt_list[id].near_mem_regions =
+ BIT(gt_to_tile(gt)->id) << 1;
+ gt_list->gt_list[id].far_mem_regions = xe->info.mem_region_mask ^
+ gt_list->gt_list[id].near_mem_regions;
+ }
+
+ if (copy_to_user(query_ptr, gt_list, size)) {
+ kfree(gt_list);
+ return -EFAULT;
+ }
+ kfree(gt_list);
+
+ return 0;
+}
+
+static int query_hwconfig(struct xe_device *xe,
+ struct drm_xe_device_query *query)
+{
+ struct xe_gt *gt = xe_root_mmio_gt(xe);
+ size_t size = xe_guc_hwconfig_size(&gt->uc.guc);
+ void __user *query_ptr = u64_to_user_ptr(query->data);
+ void *hwconfig;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ hwconfig = kzalloc(size, GFP_KERNEL);
+ if (!hwconfig)
+ return -ENOMEM;
+
+ xe_device_mem_access_get(xe);
+ xe_guc_hwconfig_copy(&gt->uc.guc, hwconfig);
+ xe_device_mem_access_put(xe);
+
+ if (copy_to_user(query_ptr, hwconfig, size)) {
+ kfree(hwconfig);
+ return -EFAULT;
+ }
+ kfree(hwconfig);
+
+ return 0;
+}
+
+static size_t calc_topo_query_size(struct xe_device *xe)
+{
+ return xe->info.gt_count *
+ (3 * sizeof(struct drm_xe_query_topology_mask) +
+ sizeof_field(struct xe_gt, fuse_topo.g_dss_mask) +
+ sizeof_field(struct xe_gt, fuse_topo.c_dss_mask) +
+ sizeof_field(struct xe_gt, fuse_topo.eu_mask_per_dss));
+}
+
+static void __user *copy_mask(void __user *ptr,
+ struct drm_xe_query_topology_mask *topo,
+ void *mask, size_t mask_size)
+{
+ topo->num_bytes = mask_size;
+
+ if (copy_to_user(ptr, topo, sizeof(*topo)))
+ return ERR_PTR(-EFAULT);
+ ptr += sizeof(topo);
+
+ if (copy_to_user(ptr, mask, mask_size))
+ return ERR_PTR(-EFAULT);
+ ptr += mask_size;
+
+ return ptr;
+}
+
+static int query_gt_topology(struct xe_device *xe,
+ struct drm_xe_device_query *query)
+{
+ void __user *query_ptr = u64_to_user_ptr(query->data);
+ size_t size = calc_topo_query_size(xe);
+ struct drm_xe_query_topology_mask topo;
+ struct xe_gt *gt;
+ int id;
+
+ if (query->size == 0) {
+ query->size = size;
+ return 0;
+ } else if (XE_IOCTL_DBG(xe, query->size != size)) {
+ return -EINVAL;
+ }
+
+ for_each_gt(gt, xe, id) {
+ topo.gt_id = id;
+
+ topo.type = DRM_XE_TOPO_DSS_GEOMETRY;
+ query_ptr = copy_mask(query_ptr, &topo,
+ gt->fuse_topo.g_dss_mask,
+ sizeof(gt->fuse_topo.g_dss_mask));
+ if (IS_ERR(query_ptr))
+ return PTR_ERR(query_ptr);
+
+ topo.type = DRM_XE_TOPO_DSS_COMPUTE;
+ query_ptr = copy_mask(query_ptr, &topo,
+ gt->fuse_topo.c_dss_mask,
+ sizeof(gt->fuse_topo.c_dss_mask));
+ if (IS_ERR(query_ptr))
+ return PTR_ERR(query_ptr);
+
+ topo.type = DRM_XE_TOPO_EU_PER_DSS;
+ query_ptr = copy_mask(query_ptr, &topo,
+ gt->fuse_topo.eu_mask_per_dss,
+ sizeof(gt->fuse_topo.eu_mask_per_dss));
+ if (IS_ERR(query_ptr))
+ return PTR_ERR(query_ptr);
+ }
+
+ return 0;
+}
+
+static int (* const xe_query_funcs[])(struct xe_device *xe,
+ struct drm_xe_device_query *query) = {
+ query_engines,
+ query_mem_regions,
+ query_config,
+ query_gt_list,
+ query_hwconfig,
+ query_gt_topology,
+ query_engine_cycles,
+};
+
+int xe_query_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct drm_xe_device_query *query = data;
+ u32 idx;
+
+ if (XE_IOCTL_DBG(xe, query->extensions) ||
+ XE_IOCTL_DBG(xe, query->reserved[0] || query->reserved[1]))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, query->query >= ARRAY_SIZE(xe_query_funcs)))
+ return -EINVAL;
+
+ idx = array_index_nospec(query->query, ARRAY_SIZE(xe_query_funcs));
+ if (XE_IOCTL_DBG(xe, !xe_query_funcs[idx]))
+ return -EINVAL;
+
+ return xe_query_funcs[idx](xe, query);
+}
diff --git a/drivers/gpu/drm/xe/xe_query.h b/drivers/gpu/drm/xe/xe_query.h
new file mode 100644
index 000000000000..beeb7a8192b4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_query.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_QUERY_H_
+#define _XE_QUERY_H_
+
+struct drm_device;
+struct drm_file;
+
+int xe_query_ioctl(struct drm_device *dev, void *data, struct drm_file *file);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_range_fence.c b/drivers/gpu/drm/xe/xe_range_fence.c
new file mode 100644
index 000000000000..d35d9ec58e86
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_range_fence.c
@@ -0,0 +1,156 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/dma-fence.h>
+#include <linux/interval_tree_generic.h>
+#include <linux/slab.h>
+
+#include "xe_macros.h"
+#include "xe_range_fence.h"
+
+#define XE_RANGE_TREE_START(_node) ((_node)->start)
+#define XE_RANGE_TREE_LAST(_node) ((_node)->last)
+
+INTERVAL_TREE_DEFINE(struct xe_range_fence, rb, u64, __subtree_last,
+ XE_RANGE_TREE_START, XE_RANGE_TREE_LAST, static,
+ xe_range_fence_tree);
+
+static void
+xe_range_fence_signal_notify(struct dma_fence *fence, struct dma_fence_cb *cb)
+{
+ struct xe_range_fence *rfence = container_of(cb, typeof(*rfence), cb);
+ struct xe_range_fence_tree *tree = rfence->tree;
+
+ llist_add(&rfence->link, &tree->list);
+}
+
+static bool __xe_range_fence_tree_cleanup(struct xe_range_fence_tree *tree)
+{
+ struct llist_node *node = llist_del_all(&tree->list);
+ struct xe_range_fence *rfence, *next;
+
+ llist_for_each_entry_safe(rfence, next, node, link) {
+ xe_range_fence_tree_remove(rfence, &tree->root);
+ dma_fence_put(rfence->fence);
+ kfree(rfence);
+ }
+
+ return !!node;
+}
+
+/**
+ * xe_range_fence_insert() - range fence insert
+ * @tree: range fence tree to insert intoi
+ * @rfence: range fence
+ * @ops: range fence ops
+ * @start: start address of range fence
+ * @last: last address of range fence
+ * @fence: dma fence which signals range fence can be removed + freed
+ *
+ * Return: 0 on success, non-zero on failure
+ */
+int xe_range_fence_insert(struct xe_range_fence_tree *tree,
+ struct xe_range_fence *rfence,
+ const struct xe_range_fence_ops *ops,
+ u64 start, u64 last, struct dma_fence *fence)
+{
+ int err = 0;
+
+ __xe_range_fence_tree_cleanup(tree);
+
+ if (dma_fence_is_signaled(fence))
+ goto free;
+
+ rfence->ops = ops;
+ rfence->start = start;
+ rfence->last = last;
+ rfence->tree = tree;
+ rfence->fence = dma_fence_get(fence);
+ err = dma_fence_add_callback(fence, &rfence->cb,
+ xe_range_fence_signal_notify);
+ if (err == -ENOENT) {
+ dma_fence_put(fence);
+ err = 0;
+ goto free;
+ } else if (err == 0) {
+ xe_range_fence_tree_insert(rfence, &tree->root);
+ return 0;
+ }
+
+free:
+ if (ops->free)
+ ops->free(rfence);
+
+ return err;
+}
+
+static void xe_range_fence_tree_remove_all(struct xe_range_fence_tree *tree)
+{
+ struct xe_range_fence *rfence;
+ bool retry = true;
+
+ rfence = xe_range_fence_tree_iter_first(&tree->root, 0, U64_MAX);
+ while (rfence) {
+ /* Should be ok with the minimalistic callback */
+ if (dma_fence_remove_callback(rfence->fence, &rfence->cb))
+ llist_add(&rfence->link, &tree->list);
+ rfence = xe_range_fence_tree_iter_next(rfence, 0, U64_MAX);
+ }
+
+ while (retry)
+ retry = __xe_range_fence_tree_cleanup(tree);
+}
+
+/**
+ * xe_range_fence_tree_init() - Init range fence tree
+ * @tree: range fence tree
+ */
+void xe_range_fence_tree_init(struct xe_range_fence_tree *tree)
+{
+ memset(tree, 0, sizeof(*tree));
+}
+
+/**
+ * xe_range_fence_tree_fini() - Fini range fence tree
+ * @tree: range fence tree
+ */
+void xe_range_fence_tree_fini(struct xe_range_fence_tree *tree)
+{
+ xe_range_fence_tree_remove_all(tree);
+ XE_WARN_ON(!RB_EMPTY_ROOT(&tree->root.rb_root));
+}
+
+/**
+ * xe_range_fence_tree_first() - range fence tree iterator first
+ * @tree: range fence tree
+ * @start: start address of range fence
+ * @last: last address of range fence
+ *
+ * Return: first range fence found in range or NULL
+ */
+struct xe_range_fence *
+xe_range_fence_tree_first(struct xe_range_fence_tree *tree, u64 start,
+ u64 last)
+{
+ return xe_range_fence_tree_iter_first(&tree->root, start, last);
+}
+
+/**
+ * xe_range_fence_tree_next() - range fence tree iterator next
+ * @rfence: current range fence
+ * @start: start address of range fence
+ * @last: last address of range fence
+ *
+ * Return: next range fence found in range or NULL
+ */
+struct xe_range_fence *
+xe_range_fence_tree_next(struct xe_range_fence *rfence, u64 start, u64 last)
+{
+ return xe_range_fence_tree_iter_next(rfence, start, last);
+}
+
+const struct xe_range_fence_ops xe_range_fence_kfree_ops = {
+ .free = (void (*)(struct xe_range_fence *rfence)) kfree,
+};
diff --git a/drivers/gpu/drm/xe/xe_range_fence.h b/drivers/gpu/drm/xe/xe_range_fence.h
new file mode 100644
index 000000000000..edd58b34f5c0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_range_fence.h
@@ -0,0 +1,75 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_RANGE_FENCE_H_
+#define _XE_RANGE_FENCE_H_
+
+#include <linux/dma-fence.h>
+#include <linux/rbtree.h>
+#include <linux/types.h>
+
+struct xe_range_fence_tree;
+struct xe_range_fence;
+
+/** struct xe_range_fence_ops - XE range fence ops */
+struct xe_range_fence_ops {
+ /** @free: free range fence op */
+ void (*free)(struct xe_range_fence *rfence);
+};
+
+/** struct xe_range_fence - XE range fence (address conflict tracking) */
+struct xe_range_fence {
+ /** @rb: RB tree node inserted into interval tree */
+ struct rb_node rb;
+ /** @start: start address of range fence is interval tree */
+ u64 start;
+ /** @last: last address (inclusive) of range fence is interval tree */
+ u64 last;
+ /** @__subtree_last: interval tree internal usage */
+ u64 __subtree_last;
+ /**
+ * @fence: fence signals address in range fence no longer has conflict
+ */
+ struct dma_fence *fence;
+ /** @tree: interval tree which range fence belongs to */
+ struct xe_range_fence_tree *tree;
+ /**
+ * @cb: callback when fence signals to remove range fence free from interval tree
+ */
+ struct dma_fence_cb cb;
+ /** @link: used to defer free of range fence to non-irq context */
+ struct llist_node link;
+ /** @ops: range fence ops */
+ const struct xe_range_fence_ops *ops;
+};
+
+/** struct xe_range_fence_tree - interval tree to store range fences */
+struct xe_range_fence_tree {
+ /** @root: interval tree root */
+ struct rb_root_cached root;
+ /** @list: list of pending range fences to be freed */
+ struct llist_head list;
+};
+
+extern const struct xe_range_fence_ops xe_range_fence_kfree_ops;
+
+struct xe_range_fence *
+xe_range_fence_tree_first(struct xe_range_fence_tree *tree, u64 start,
+ u64 last);
+
+struct xe_range_fence *
+xe_range_fence_tree_next(struct xe_range_fence *rfence, u64 start, u64 last);
+
+void xe_range_fence_tree_init(struct xe_range_fence_tree *tree);
+
+void xe_range_fence_tree_fini(struct xe_range_fence_tree *tree);
+
+int xe_range_fence_insert(struct xe_range_fence_tree *tree,
+ struct xe_range_fence *rfence,
+ const struct xe_range_fence_ops *ops,
+ u64 start, u64 end,
+ struct dma_fence *fence);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_reg_sr.c b/drivers/gpu/drm/xe/xe_reg_sr.c
new file mode 100644
index 000000000000..87adefb56024
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_reg_sr.c
@@ -0,0 +1,284 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_reg_sr.h"
+
+#include <kunit/visibility.h>
+#include <linux/align.h>
+#include <linux/string_helpers.h>
+#include <linux/xarray.h>
+
+#include <drm/drm_managed.h>
+#include <drm/drm_print.h>
+
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "xe_device_types.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_gt_mcr.h"
+#include "xe_gt_printk.h"
+#include "xe_hw_engine_types.h"
+#include "xe_macros.h"
+#include "xe_mmio.h"
+#include "xe_reg_whitelist.h"
+#include "xe_rtp_types.h"
+
+#define XE_REG_SR_GROW_STEP_DEFAULT 16
+
+static void reg_sr_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_reg_sr *sr = arg;
+
+ xa_destroy(&sr->xa);
+ kfree(sr->pool.arr);
+ memset(&sr->pool, 0, sizeof(sr->pool));
+}
+
+int xe_reg_sr_init(struct xe_reg_sr *sr, const char *name, struct xe_device *xe)
+{
+ xa_init(&sr->xa);
+ memset(&sr->pool, 0, sizeof(sr->pool));
+ sr->pool.grow_step = XE_REG_SR_GROW_STEP_DEFAULT;
+ sr->name = name;
+
+ return drmm_add_action_or_reset(&xe->drm, reg_sr_fini, sr);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_reg_sr_init);
+
+static struct xe_reg_sr_entry *alloc_entry(struct xe_reg_sr *sr)
+{
+ if (sr->pool.used == sr->pool.allocated) {
+ struct xe_reg_sr_entry *arr;
+
+ arr = krealloc_array(sr->pool.arr,
+ ALIGN(sr->pool.allocated + 1, sr->pool.grow_step),
+ sizeof(*arr), GFP_KERNEL);
+ if (!arr)
+ return NULL;
+
+ sr->pool.arr = arr;
+ sr->pool.allocated += sr->pool.grow_step;
+ }
+
+ return &sr->pool.arr[sr->pool.used++];
+}
+
+static bool compatible_entries(const struct xe_reg_sr_entry *e1,
+ const struct xe_reg_sr_entry *e2)
+{
+ /*
+ * Don't allow overwriting values: clr_bits/set_bits should be disjoint
+ * when operating in the same register
+ */
+ if (e1->clr_bits & e2->clr_bits || e1->set_bits & e2->set_bits ||
+ e1->clr_bits & e2->set_bits || e1->set_bits & e2->clr_bits)
+ return false;
+
+ if (e1->reg.raw != e2->reg.raw)
+ return false;
+
+ return true;
+}
+
+static void reg_sr_inc_error(struct xe_reg_sr *sr)
+{
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+ sr->errors++;
+#endif
+}
+
+int xe_reg_sr_add(struct xe_reg_sr *sr,
+ const struct xe_reg_sr_entry *e,
+ struct xe_gt *gt)
+{
+ unsigned long idx = e->reg.addr;
+ struct xe_reg_sr_entry *pentry = xa_load(&sr->xa, idx);
+ int ret;
+
+ if (pentry) {
+ if (!compatible_entries(pentry, e)) {
+ ret = -EINVAL;
+ goto fail;
+ }
+
+ pentry->clr_bits |= e->clr_bits;
+ pentry->set_bits |= e->set_bits;
+ pentry->read_mask |= e->read_mask;
+
+ return 0;
+ }
+
+ pentry = alloc_entry(sr);
+ if (!pentry) {
+ ret = -ENOMEM;
+ goto fail;
+ }
+
+ *pentry = *e;
+ ret = xa_err(xa_store(&sr->xa, idx, pentry, GFP_KERNEL));
+ if (ret)
+ goto fail;
+
+ return 0;
+
+fail:
+ xe_gt_err(gt,
+ "discarding save-restore reg %04lx (clear: %08x, set: %08x, masked: %s, mcr: %s): ret=%d\n",
+ idx, e->clr_bits, e->set_bits,
+ str_yes_no(e->reg.masked),
+ str_yes_no(e->reg.mcr),
+ ret);
+ reg_sr_inc_error(sr);
+
+ return ret;
+}
+
+/*
+ * Convert back from encoded value to type-safe, only to be used when reg.mcr
+ * is true
+ */
+static struct xe_reg_mcr to_xe_reg_mcr(const struct xe_reg reg)
+{
+ return (const struct xe_reg_mcr){.__reg.raw = reg.raw };
+}
+
+static void apply_one_mmio(struct xe_gt *gt, struct xe_reg_sr_entry *entry)
+{
+ struct xe_reg reg = entry->reg;
+ struct xe_reg_mcr reg_mcr = to_xe_reg_mcr(reg);
+ u32 val;
+
+ /*
+ * If this is a masked register, need to set the upper 16 bits.
+ * Set them to clr_bits since that is always a superset of the bits
+ * being modified.
+ *
+ * When it's not masked, we have to read it from hardware, unless we are
+ * supposed to set all bits.
+ */
+ if (reg.masked)
+ val = entry->clr_bits << 16;
+ else if (entry->clr_bits + 1)
+ val = (reg.mcr ?
+ xe_gt_mcr_unicast_read_any(gt, reg_mcr) :
+ xe_mmio_read32(gt, reg)) & (~entry->clr_bits);
+ else
+ val = 0;
+
+ /*
+ * TODO: add selftest to validate all tables, regardless of platform:
+ * - Masked registers can't have set_bits with upper bits set
+ * - set_bits must be contained in clr_bits
+ */
+ val |= entry->set_bits;
+
+ xe_gt_dbg(gt, "REG[0x%x] = 0x%08x", reg.addr, val);
+
+ if (entry->reg.mcr)
+ xe_gt_mcr_multicast_write(gt, reg_mcr, val);
+ else
+ xe_mmio_write32(gt, reg, val);
+}
+
+void xe_reg_sr_apply_mmio(struct xe_reg_sr *sr, struct xe_gt *gt)
+{
+ struct xe_reg_sr_entry *entry;
+ unsigned long reg;
+ int err;
+
+ if (xa_empty(&sr->xa))
+ return;
+
+ xe_gt_dbg(gt, "Applying %s save-restore MMIOs\n", sr->name);
+
+ err = xe_force_wake_get(&gt->mmio.fw, XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_force_wake;
+
+ xa_for_each(&sr->xa, reg, entry)
+ apply_one_mmio(gt, entry);
+
+ err = xe_force_wake_put(&gt->mmio.fw, XE_FORCEWAKE_ALL);
+ XE_WARN_ON(err);
+
+ return;
+
+err_force_wake:
+ xe_gt_err(gt, "Failed to apply, err=%d\n", err);
+}
+
+void xe_reg_sr_apply_whitelist(struct xe_hw_engine *hwe)
+{
+ struct xe_reg_sr *sr = &hwe->reg_whitelist;
+ struct xe_gt *gt = hwe->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ struct xe_reg_sr_entry *entry;
+ struct drm_printer p;
+ u32 mmio_base = hwe->mmio_base;
+ unsigned long reg;
+ unsigned int slot = 0;
+ int err;
+
+ if (xa_empty(&sr->xa))
+ return;
+
+ drm_dbg(&xe->drm, "Whitelisting %s registers\n", sr->name);
+
+ err = xe_force_wake_get(&gt->mmio.fw, XE_FORCEWAKE_ALL);
+ if (err)
+ goto err_force_wake;
+
+ p = drm_debug_printer(KBUILD_MODNAME);
+ xa_for_each(&sr->xa, reg, entry) {
+ if (slot == RING_MAX_NONPRIV_SLOTS) {
+ xe_gt_err(gt,
+ "hwe %s: maximum register whitelist slots (%d) reached, refusing to add more\n",
+ hwe->name, RING_MAX_NONPRIV_SLOTS);
+ break;
+ }
+
+ xe_reg_whitelist_print_entry(&p, 0, reg, entry);
+ xe_mmio_write32(gt, RING_FORCE_TO_NONPRIV(mmio_base, slot),
+ reg | entry->set_bits);
+ slot++;
+ }
+
+ /* And clear the rest just in case of garbage */
+ for (; slot < RING_MAX_NONPRIV_SLOTS; slot++) {
+ u32 addr = RING_NOPID(mmio_base).addr;
+
+ xe_mmio_write32(gt, RING_FORCE_TO_NONPRIV(mmio_base, slot), addr);
+ }
+
+ err = xe_force_wake_put(&gt->mmio.fw, XE_FORCEWAKE_ALL);
+ XE_WARN_ON(err);
+
+ return;
+
+err_force_wake:
+ drm_err(&xe->drm, "Failed to apply, err=%d\n", err);
+}
+
+/**
+ * xe_reg_sr_dump - print all save/restore entries
+ * @sr: Save/restore entries
+ * @p: DRM printer
+ */
+void xe_reg_sr_dump(struct xe_reg_sr *sr, struct drm_printer *p)
+{
+ struct xe_reg_sr_entry *entry;
+ unsigned long reg;
+
+ if (!sr->name || xa_empty(&sr->xa))
+ return;
+
+ drm_printf(p, "%s\n", sr->name);
+ xa_for_each(&sr->xa, reg, entry)
+ drm_printf(p, "\tREG[0x%lx] clr=0x%08x set=0x%08x masked=%s mcr=%s\n",
+ reg, entry->clr_bits, entry->set_bits,
+ str_yes_no(entry->reg.masked),
+ str_yes_no(entry->reg.mcr));
+}
diff --git a/drivers/gpu/drm/xe/xe_reg_sr.h b/drivers/gpu/drm/xe/xe_reg_sr.h
new file mode 100644
index 000000000000..e3197c33afe2
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_reg_sr.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_REG_SR_
+#define _XE_REG_SR_
+
+#include "xe_reg_sr_types.h"
+
+/*
+ * Reg save/restore bookkeeping
+ */
+
+struct xe_device;
+struct xe_gt;
+struct xe_hw_engine;
+struct drm_printer;
+
+int xe_reg_sr_init(struct xe_reg_sr *sr, const char *name, struct xe_device *xe);
+void xe_reg_sr_dump(struct xe_reg_sr *sr, struct drm_printer *p);
+
+int xe_reg_sr_add(struct xe_reg_sr *sr, const struct xe_reg_sr_entry *e,
+ struct xe_gt *gt);
+void xe_reg_sr_apply_mmio(struct xe_reg_sr *sr, struct xe_gt *gt);
+void xe_reg_sr_apply_whitelist(struct xe_hw_engine *hwe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_reg_sr_types.h b/drivers/gpu/drm/xe/xe_reg_sr_types.h
new file mode 100644
index 000000000000..ad48a52b824a
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_reg_sr_types.h
@@ -0,0 +1,37 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_REG_SR_TYPES_
+#define _XE_REG_SR_TYPES_
+
+#include <linux/types.h>
+#include <linux/xarray.h>
+
+#include "regs/xe_reg_defs.h"
+
+struct xe_reg_sr_entry {
+ struct xe_reg reg;
+ u32 clr_bits;
+ u32 set_bits;
+ /* Mask for bits to consider when reading value back */
+ u32 read_mask;
+};
+
+struct xe_reg_sr {
+ struct {
+ struct xe_reg_sr_entry *arr;
+ unsigned int used;
+ unsigned int allocated;
+ unsigned int grow_step;
+ } pool;
+ struct xarray xa;
+ const char *name;
+
+#if IS_ENABLED(CONFIG_DRM_XE_KUNIT_TEST)
+ unsigned int errors;
+#endif
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_reg_whitelist.c b/drivers/gpu/drm/xe/xe_reg_whitelist.c
new file mode 100644
index 000000000000..e66ae1bdaf9c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_reg_whitelist.c
@@ -0,0 +1,146 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_reg_whitelist.h"
+
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "xe_gt_types.h"
+#include "xe_platform_types.h"
+#include "xe_rtp.h"
+
+#undef XE_REG_MCR
+#define XE_REG_MCR(...) XE_REG(__VA_ARGS__, .mcr = 1)
+
+static bool match_not_render(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe)
+{
+ return hwe->class != XE_ENGINE_CLASS_RENDER;
+}
+
+static const struct xe_rtp_entry_sr register_whitelist[] = {
+ { XE_RTP_NAME("WaAllowPMDepthAndInvocationCountAccessFromUMD, 1408556865"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(WHITELIST(PS_INVOCATION_COUNT,
+ RING_FORCE_TO_NONPRIV_ACCESS_RD |
+ RING_FORCE_TO_NONPRIV_RANGE_4))
+ },
+ { XE_RTP_NAME("1508744258, 14012131227, 1808121037"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(WHITELIST(COMMON_SLICE_CHICKEN1, 0))
+ },
+ { XE_RTP_NAME("1806527549"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(WHITELIST(HIZ_CHICKEN, 0))
+ },
+ { XE_RTP_NAME("allow_read_ctx_timestamp"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1260), FUNC(match_not_render)),
+ XE_RTP_ACTIONS(WHITELIST(RING_CTX_TIMESTAMP(0),
+ RING_FORCE_TO_NONPRIV_ACCESS_RD,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+ { XE_RTP_NAME("16014440446"),
+ XE_RTP_RULES(PLATFORM(PVC)),
+ XE_RTP_ACTIONS(WHITELIST(XE_REG(0x4400),
+ RING_FORCE_TO_NONPRIV_DENY |
+ RING_FORCE_TO_NONPRIV_RANGE_64),
+ WHITELIST(XE_REG(0x4500),
+ RING_FORCE_TO_NONPRIV_DENY |
+ RING_FORCE_TO_NONPRIV_RANGE_64))
+ },
+ { XE_RTP_NAME("16017236439"),
+ XE_RTP_RULES(PLATFORM(PVC), ENGINE_CLASS(COPY)),
+ XE_RTP_ACTIONS(WHITELIST(BCS_SWCTRL(0),
+ RING_FORCE_TO_NONPRIV_DENY,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+ {}
+};
+
+/**
+ * xe_reg_whitelist_process_engine - process table of registers to whitelist
+ * @hwe: engine instance to process whitelist for
+ *
+ * Process wwhitelist table for this platform, saving in @hwe all the
+ * registers that need to be whitelisted by the hardware so they can be accessed
+ * by userspace.
+ */
+void xe_reg_whitelist_process_engine(struct xe_hw_engine *hwe)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+
+ xe_rtp_process_to_sr(&ctx, register_whitelist, &hwe->reg_whitelist);
+}
+
+/**
+ * xe_reg_whitelist_print_entry - print one whitelist entry
+ * @p: DRM printer
+ * @indent: indent level
+ * @reg: register allowed/denied
+ * @entry: save-restore entry
+ *
+ * Print details about the entry added to allow/deny access
+ */
+void xe_reg_whitelist_print_entry(struct drm_printer *p, unsigned int indent,
+ u32 reg, struct xe_reg_sr_entry *entry)
+{
+ u32 val = entry->set_bits;
+ const char *access_str = "(invalid)";
+ unsigned int range_bit = 2;
+ u32 range_start, range_end;
+ bool deny;
+
+ deny = val & RING_FORCE_TO_NONPRIV_DENY;
+
+ switch (val & RING_FORCE_TO_NONPRIV_RANGE_MASK) {
+ case RING_FORCE_TO_NONPRIV_RANGE_4:
+ range_bit = 4;
+ break;
+ case RING_FORCE_TO_NONPRIV_RANGE_16:
+ range_bit = 6;
+ break;
+ case RING_FORCE_TO_NONPRIV_RANGE_64:
+ range_bit = 8;
+ break;
+ }
+
+ range_start = reg & REG_GENMASK(25, range_bit);
+ range_end = range_start | REG_GENMASK(range_bit, 0);
+
+ switch (val & RING_FORCE_TO_NONPRIV_ACCESS_MASK) {
+ case RING_FORCE_TO_NONPRIV_ACCESS_RW:
+ access_str = "rw";
+ break;
+ case RING_FORCE_TO_NONPRIV_ACCESS_RD:
+ access_str = "read";
+ break;
+ case RING_FORCE_TO_NONPRIV_ACCESS_WR:
+ access_str = "write";
+ break;
+ }
+
+ drm_printf_indent(p, indent, "REG[0x%x-0x%x]: %s %s access\n",
+ range_start, range_end,
+ deny ? "deny" : "allow",
+ access_str);
+}
+
+/**
+ * xe_reg_whitelist_dump - print all whitelist entries
+ * @sr: Save/restore entries
+ * @p: DRM printer
+ */
+void xe_reg_whitelist_dump(struct xe_reg_sr *sr, struct drm_printer *p)
+{
+ struct xe_reg_sr_entry *entry;
+ unsigned long reg;
+
+ if (!sr->name || xa_empty(&sr->xa))
+ return;
+
+ drm_printf(p, "%s\n", sr->name);
+ xa_for_each(&sr->xa, reg, entry)
+ xe_reg_whitelist_print_entry(p, 1, reg, entry);
+}
diff --git a/drivers/gpu/drm/xe/xe_reg_whitelist.h b/drivers/gpu/drm/xe/xe_reg_whitelist.h
new file mode 100644
index 000000000000..69b121d377da
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_reg_whitelist.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_REG_WHITELIST_
+#define _XE_REG_WHITELIST_
+
+#include <linux/types.h>
+
+struct drm_printer;
+struct xe_hw_engine;
+struct xe_reg_sr;
+struct xe_reg_sr_entry;
+
+void xe_reg_whitelist_process_engine(struct xe_hw_engine *hwe);
+
+void xe_reg_whitelist_print_entry(struct drm_printer *p, unsigned int indent,
+ u32 reg, struct xe_reg_sr_entry *entry);
+
+void xe_reg_whitelist_dump(struct xe_reg_sr *sr, struct drm_printer *p);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_res_cursor.h b/drivers/gpu/drm/xe/xe_res_cursor.h
new file mode 100644
index 000000000000..0a306963aa8e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_res_cursor.h
@@ -0,0 +1,240 @@
+/* SPDX-License-Identifier: GPL-2.0-only OR MIT */
+/*
+ * Copyright 2020 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef _XE_RES_CURSOR_H_
+#define _XE_RES_CURSOR_H_
+
+#include <linux/scatterlist.h>
+
+#include <drm/drm_mm.h>
+#include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_range_manager.h>
+#include <drm/ttm/ttm_resource.h>
+#include <drm/ttm/ttm_tt.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_macros.h"
+#include "xe_ttm_vram_mgr.h"
+
+/* state back for walking over vram_mgr, stolen_mgr, and gtt_mgr allocations */
+struct xe_res_cursor {
+ u64 start;
+ u64 size;
+ u64 remaining;
+ void *node;
+ u32 mem_type;
+ struct scatterlist *sgl;
+ struct drm_buddy *mm;
+};
+
+static struct drm_buddy *xe_res_get_buddy(struct ttm_resource *res)
+{
+ struct ttm_resource_manager *mgr;
+
+ mgr = ttm_manager_type(res->bo->bdev, res->mem_type);
+ return &to_xe_ttm_vram_mgr(mgr)->mm;
+}
+
+/**
+ * xe_res_first - initialize a xe_res_cursor
+ *
+ * @res: TTM resource object to walk
+ * @start: Start of the range
+ * @size: Size of the range
+ * @cur: cursor object to initialize
+ *
+ * Start walking over the range of allocations between @start and @size.
+ */
+static inline void xe_res_first(struct ttm_resource *res,
+ u64 start, u64 size,
+ struct xe_res_cursor *cur)
+{
+ cur->sgl = NULL;
+ if (!res)
+ goto fallback;
+
+ XE_WARN_ON(start + size > res->size);
+
+ cur->mem_type = res->mem_type;
+
+ switch (cur->mem_type) {
+ case XE_PL_STOLEN:
+ case XE_PL_VRAM0:
+ case XE_PL_VRAM1: {
+ struct drm_buddy_block *block;
+ struct list_head *head, *next;
+ struct drm_buddy *mm = xe_res_get_buddy(res);
+
+ head = &to_xe_ttm_vram_mgr_resource(res)->blocks;
+
+ block = list_first_entry_or_null(head,
+ struct drm_buddy_block,
+ link);
+ if (!block)
+ goto fallback;
+
+ while (start >= drm_buddy_block_size(mm, block)) {
+ start -= drm_buddy_block_size(mm, block);
+
+ next = block->link.next;
+ if (next != head)
+ block = list_entry(next, struct drm_buddy_block,
+ link);
+ }
+
+ cur->mm = mm;
+ cur->start = drm_buddy_block_offset(block) + start;
+ cur->size = min(drm_buddy_block_size(mm, block) - start,
+ size);
+ cur->remaining = size;
+ cur->node = block;
+ break;
+ }
+ default:
+ goto fallback;
+ }
+
+ return;
+
+fallback:
+ cur->start = start;
+ cur->size = size;
+ cur->remaining = size;
+ cur->node = NULL;
+ cur->mem_type = XE_PL_TT;
+ XE_WARN_ON(res && start + size > res->size);
+}
+
+static inline void __xe_res_sg_next(struct xe_res_cursor *cur)
+{
+ struct scatterlist *sgl = cur->sgl;
+ u64 start = cur->start;
+
+ while (start >= sg_dma_len(sgl)) {
+ start -= sg_dma_len(sgl);
+ sgl = sg_next(sgl);
+ XE_WARN_ON(!sgl);
+ }
+
+ cur->start = start;
+ cur->size = sg_dma_len(sgl) - start;
+ cur->sgl = sgl;
+}
+
+/**
+ * xe_res_first_sg - initialize a xe_res_cursor with a scatter gather table
+ *
+ * @sg: scatter gather table to walk
+ * @start: Start of the range
+ * @size: Size of the range
+ * @cur: cursor object to initialize
+ *
+ * Start walking over the range of allocations between @start and @size.
+ */
+static inline void xe_res_first_sg(const struct sg_table *sg,
+ u64 start, u64 size,
+ struct xe_res_cursor *cur)
+{
+ XE_WARN_ON(!sg);
+ XE_WARN_ON(!IS_ALIGNED(start, PAGE_SIZE) ||
+ !IS_ALIGNED(size, PAGE_SIZE));
+ cur->node = NULL;
+ cur->start = start;
+ cur->remaining = size;
+ cur->size = 0;
+ cur->sgl = sg->sgl;
+ cur->mem_type = XE_PL_TT;
+ __xe_res_sg_next(cur);
+}
+
+/**
+ * xe_res_next - advance the cursor
+ *
+ * @cur: the cursor to advance
+ * @size: number of bytes to move forward
+ *
+ * Move the cursor @size bytes forwrad, walking to the next node if necessary.
+ */
+static inline void xe_res_next(struct xe_res_cursor *cur, u64 size)
+{
+ struct drm_buddy_block *block;
+ struct list_head *next;
+ u64 start;
+
+ XE_WARN_ON(size > cur->remaining);
+
+ cur->remaining -= size;
+ if (!cur->remaining)
+ return;
+
+ if (cur->size > size) {
+ cur->size -= size;
+ cur->start += size;
+ return;
+ }
+
+ if (cur->sgl) {
+ cur->start += size;
+ __xe_res_sg_next(cur);
+ return;
+ }
+
+ switch (cur->mem_type) {
+ case XE_PL_STOLEN:
+ case XE_PL_VRAM0:
+ case XE_PL_VRAM1:
+ start = size - cur->size;
+ block = cur->node;
+
+ next = block->link.next;
+ block = list_entry(next, struct drm_buddy_block, link);
+
+
+ while (start >= drm_buddy_block_size(cur->mm, block)) {
+ start -= drm_buddy_block_size(cur->mm, block);
+
+ next = block->link.next;
+ block = list_entry(next, struct drm_buddy_block, link);
+ }
+
+ cur->start = drm_buddy_block_offset(block) + start;
+ cur->size = min(drm_buddy_block_size(cur->mm, block) - start,
+ cur->remaining);
+ cur->node = block;
+ break;
+ default:
+ return;
+ }
+}
+
+/**
+ * xe_res_dma - return dma address of cursor at current position
+ *
+ * @cur: the cursor to return the dma address from
+ */
+static inline u64 xe_res_dma(const struct xe_res_cursor *cur)
+{
+ return cur->sgl ? sg_dma_address(cur->sgl) + cur->start : cur->start;
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ring_ops.c b/drivers/gpu/drm/xe/xe_ring_ops.c
new file mode 100644
index 000000000000..1e4c06eacd98
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ring_ops.c
@@ -0,0 +1,482 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_ring_ops.h"
+
+#include "generated/xe_wa_oob.h"
+#include "instructions/xe_mi_commands.h"
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gpu_commands.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_lrc_layout.h"
+#include "xe_exec_queue_types.h"
+#include "xe_gt.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_sched_job.h"
+#include "xe_vm_types.h"
+#include "xe_vm.h"
+#include "xe_wa.h"
+
+/*
+ * 3D-related flags that can't be set on _engines_ that lack access to the 3D
+ * pipeline (i.e., CCS engines).
+ */
+#define PIPE_CONTROL_3D_ENGINE_FLAGS (\
+ PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH | \
+ PIPE_CONTROL_DEPTH_CACHE_FLUSH | \
+ PIPE_CONTROL_TILE_CACHE_FLUSH | \
+ PIPE_CONTROL_DEPTH_STALL | \
+ PIPE_CONTROL_STALL_AT_SCOREBOARD | \
+ PIPE_CONTROL_PSD_SYNC | \
+ PIPE_CONTROL_AMFS_FLUSH | \
+ PIPE_CONTROL_VF_CACHE_INVALIDATE | \
+ PIPE_CONTROL_GLOBAL_SNAPSHOT_RESET)
+
+/* 3D-related flags that can't be set on _platforms_ that lack a 3D pipeline */
+#define PIPE_CONTROL_3D_ARCH_FLAGS ( \
+ PIPE_CONTROL_3D_ENGINE_FLAGS | \
+ PIPE_CONTROL_INDIRECT_STATE_DISABLE | \
+ PIPE_CONTROL_FLUSH_ENABLE | \
+ PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE | \
+ PIPE_CONTROL_DC_FLUSH_ENABLE)
+
+static u32 preparser_disable(bool state)
+{
+ return MI_ARB_CHECK | BIT(8) | state;
+}
+
+static int emit_aux_table_inv(struct xe_gt *gt, struct xe_reg reg,
+ u32 *dw, int i)
+{
+ dw[i++] = MI_LOAD_REGISTER_IMM | MI_LRI_NUM_REGS(1) | MI_LRI_MMIO_REMAP_EN;
+ dw[i++] = reg.addr + gt->mmio.adj_offset;
+ dw[i++] = AUX_INV;
+ dw[i++] = MI_NOOP;
+
+ return i;
+}
+
+static int emit_user_interrupt(u32 *dw, int i)
+{
+ dw[i++] = MI_USER_INTERRUPT;
+ dw[i++] = MI_ARB_ON_OFF | MI_ARB_ENABLE;
+ dw[i++] = MI_ARB_CHECK;
+
+ return i;
+}
+
+static int emit_store_imm_ggtt(u32 addr, u32 value, u32 *dw, int i)
+{
+ dw[i++] = MI_STORE_DATA_IMM | MI_SDI_GGTT | MI_SDI_NUM_DW(1);
+ dw[i++] = addr;
+ dw[i++] = 0;
+ dw[i++] = value;
+
+ return i;
+}
+
+static int emit_flush_imm_ggtt(u32 addr, u32 value, bool invalidate_tlb,
+ u32 *dw, int i)
+{
+ dw[i++] = MI_FLUSH_DW | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_IMM_DW |
+ (invalidate_tlb ? MI_INVALIDATE_TLB : 0);
+ dw[i++] = addr | MI_FLUSH_DW_USE_GTT;
+ dw[i++] = 0;
+ dw[i++] = value;
+
+ return i;
+}
+
+static int emit_bb_start(u64 batch_addr, u32 ppgtt_flag, u32 *dw, int i)
+{
+ dw[i++] = MI_BATCH_BUFFER_START | ppgtt_flag | XE_INSTR_NUM_DW(3);
+ dw[i++] = lower_32_bits(batch_addr);
+ dw[i++] = upper_32_bits(batch_addr);
+
+ return i;
+}
+
+static int emit_flush_invalidate(u32 flag, u32 *dw, int i)
+{
+ dw[i] = MI_FLUSH_DW;
+ dw[i] |= flag;
+ dw[i++] |= MI_INVALIDATE_TLB | MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_IMM_DW |
+ MI_FLUSH_DW_STORE_INDEX;
+
+ dw[i++] = LRC_PPHWSP_SCRATCH_ADDR | MI_FLUSH_DW_USE_GTT;
+ dw[i++] = 0;
+ dw[i++] = ~0U;
+
+ return i;
+}
+
+static int emit_pipe_invalidate(u32 mask_flags, bool invalidate_tlb, u32 *dw,
+ int i)
+{
+ u32 flags = PIPE_CONTROL_CS_STALL |
+ PIPE_CONTROL_COMMAND_CACHE_INVALIDATE |
+ PIPE_CONTROL_INSTRUCTION_CACHE_INVALIDATE |
+ PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
+ PIPE_CONTROL_VF_CACHE_INVALIDATE |
+ PIPE_CONTROL_CONST_CACHE_INVALIDATE |
+ PIPE_CONTROL_STATE_CACHE_INVALIDATE |
+ PIPE_CONTROL_QW_WRITE |
+ PIPE_CONTROL_STORE_DATA_INDEX;
+
+ if (invalidate_tlb)
+ flags |= PIPE_CONTROL_TLB_INVALIDATE;
+
+ flags &= ~mask_flags;
+
+ dw[i++] = GFX_OP_PIPE_CONTROL(6);
+ dw[i++] = flags;
+ dw[i++] = LRC_PPHWSP_SCRATCH_ADDR;
+ dw[i++] = 0;
+ dw[i++] = 0;
+ dw[i++] = 0;
+
+ return i;
+}
+
+static int emit_store_imm_ppgtt_posted(u64 addr, u64 value,
+ u32 *dw, int i)
+{
+ dw[i++] = MI_STORE_DATA_IMM | MI_SDI_NUM_QW(1);
+ dw[i++] = lower_32_bits(addr);
+ dw[i++] = upper_32_bits(addr);
+ dw[i++] = lower_32_bits(value);
+ dw[i++] = upper_32_bits(value);
+
+ return i;
+}
+
+static int emit_render_cache_flush(struct xe_sched_job *job, u32 *dw, int i)
+{
+ struct xe_gt *gt = job->q->gt;
+ bool lacks_render = !(gt->info.engine_mask & XE_HW_ENGINE_RCS_MASK);
+ u32 flags;
+
+ flags = (PIPE_CONTROL_CS_STALL |
+ PIPE_CONTROL_TILE_CACHE_FLUSH |
+ PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH |
+ PIPE_CONTROL_DEPTH_CACHE_FLUSH |
+ PIPE_CONTROL_DC_FLUSH_ENABLE |
+ PIPE_CONTROL_FLUSH_ENABLE);
+
+ if (XE_WA(gt, 1409600907))
+ flags |= PIPE_CONTROL_DEPTH_STALL;
+
+ if (lacks_render)
+ flags &= ~PIPE_CONTROL_3D_ARCH_FLAGS;
+ else if (job->q->class == XE_ENGINE_CLASS_COMPUTE)
+ flags &= ~PIPE_CONTROL_3D_ENGINE_FLAGS;
+
+ dw[i++] = GFX_OP_PIPE_CONTROL(6) | PIPE_CONTROL0_HDC_PIPELINE_FLUSH;
+ dw[i++] = flags;
+ dw[i++] = 0;
+ dw[i++] = 0;
+ dw[i++] = 0;
+ dw[i++] = 0;
+
+ return i;
+}
+
+static int emit_pipe_control_to_ring_end(struct xe_hw_engine *hwe, u32 *dw, int i)
+{
+ if (hwe->class != XE_ENGINE_CLASS_RENDER)
+ return i;
+
+ if (XE_WA(hwe->gt, 16020292621)) {
+ dw[i++] = GFX_OP_PIPE_CONTROL(6);
+ dw[i++] = PIPE_CONTROL_LRI_POST_SYNC;
+ dw[i++] = RING_NOPID(hwe->mmio_base).addr;
+ dw[i++] = 0;
+ dw[i++] = 0;
+ dw[i++] = 0;
+ }
+
+ return i;
+}
+
+static int emit_pipe_imm_ggtt(u32 addr, u32 value, bool stall_only, u32 *dw,
+ int i)
+{
+ dw[i++] = GFX_OP_PIPE_CONTROL(6);
+ dw[i++] = (stall_only ? PIPE_CONTROL_CS_STALL :
+ PIPE_CONTROL_FLUSH_ENABLE | PIPE_CONTROL_CS_STALL) |
+ PIPE_CONTROL_GLOBAL_GTT_IVB | PIPE_CONTROL_QW_WRITE;
+ dw[i++] = addr;
+ dw[i++] = 0;
+ dw[i++] = value;
+ dw[i++] = 0; /* We're thrashing one extra dword. */
+
+ return i;
+}
+
+static u32 get_ppgtt_flag(struct xe_sched_job *job)
+{
+ return job->q->vm ? BIT(8) : 0;
+}
+
+/* for engines that don't require any special HW handling (no EUs, no aux inval, etc) */
+static void __emit_job_gen12_simple(struct xe_sched_job *job, struct xe_lrc *lrc,
+ u64 batch_addr, u32 seqno)
+{
+ u32 dw[MAX_JOB_SIZE_DW], i = 0;
+ u32 ppgtt_flag = get_ppgtt_flag(job);
+ struct xe_vm *vm = job->q->vm;
+ struct xe_gt *gt = job->q->gt;
+
+ if (vm && vm->batch_invalidate_tlb) {
+ dw[i++] = preparser_disable(true);
+ i = emit_flush_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, true, dw, i);
+ dw[i++] = preparser_disable(false);
+ } else {
+ i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, dw, i);
+ }
+
+ i = emit_bb_start(batch_addr, ppgtt_flag, dw, i);
+
+ if (job->user_fence.used)
+ i = emit_store_imm_ppgtt_posted(job->user_fence.addr,
+ job->user_fence.value,
+ dw, i);
+
+ i = emit_flush_imm_ggtt(xe_lrc_seqno_ggtt_addr(lrc), seqno, false, dw, i);
+
+ i = emit_user_interrupt(dw, i);
+
+ xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW);
+
+ xe_lrc_write_ring(lrc, dw, i * sizeof(*dw));
+}
+
+static bool has_aux_ccs(struct xe_device *xe)
+{
+ /*
+ * PVC is a special case that has no compression of either type
+ * (FlatCCS or AuxCCS). Also, AuxCCS is no longer used from Xe2
+ * onward, so any future platforms with no FlatCCS will not have
+ * AuxCCS either.
+ */
+ if (GRAPHICS_VER(xe) >= 20 || xe->info.platform == XE_PVC)
+ return false;
+
+ return !xe->info.has_flat_ccs;
+}
+
+static void __emit_job_gen12_video(struct xe_sched_job *job, struct xe_lrc *lrc,
+ u64 batch_addr, u32 seqno)
+{
+ u32 dw[MAX_JOB_SIZE_DW], i = 0;
+ u32 ppgtt_flag = get_ppgtt_flag(job);
+ struct xe_gt *gt = job->q->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ bool decode = job->q->class == XE_ENGINE_CLASS_VIDEO_DECODE;
+ struct xe_vm *vm = job->q->vm;
+
+ dw[i++] = preparser_disable(true);
+
+ /* hsdes: 1809175790 */
+ if (has_aux_ccs(xe)) {
+ if (decode)
+ i = emit_aux_table_inv(gt, VD0_AUX_INV, dw, i);
+ else
+ i = emit_aux_table_inv(gt, VE0_AUX_INV, dw, i);
+ }
+
+ if (vm && vm->batch_invalidate_tlb)
+ i = emit_flush_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, true, dw, i);
+
+ dw[i++] = preparser_disable(false);
+
+ if (!vm || !vm->batch_invalidate_tlb)
+ i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, dw, i);
+
+ i = emit_bb_start(batch_addr, ppgtt_flag, dw, i);
+
+ if (job->user_fence.used)
+ i = emit_store_imm_ppgtt_posted(job->user_fence.addr,
+ job->user_fence.value,
+ dw, i);
+
+ i = emit_flush_imm_ggtt(xe_lrc_seqno_ggtt_addr(lrc), seqno, false, dw, i);
+
+ i = emit_user_interrupt(dw, i);
+
+ xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW);
+
+ xe_lrc_write_ring(lrc, dw, i * sizeof(*dw));
+}
+
+static void __emit_job_gen12_render_compute(struct xe_sched_job *job,
+ struct xe_lrc *lrc,
+ u64 batch_addr, u32 seqno)
+{
+ u32 dw[MAX_JOB_SIZE_DW], i = 0;
+ u32 ppgtt_flag = get_ppgtt_flag(job);
+ struct xe_gt *gt = job->q->gt;
+ struct xe_device *xe = gt_to_xe(gt);
+ bool lacks_render = !(gt->info.engine_mask & XE_HW_ENGINE_RCS_MASK);
+ struct xe_vm *vm = job->q->vm;
+ u32 mask_flags = 0;
+
+ dw[i++] = preparser_disable(true);
+ if (lacks_render)
+ mask_flags = PIPE_CONTROL_3D_ARCH_FLAGS;
+ else if (job->q->class == XE_ENGINE_CLASS_COMPUTE)
+ mask_flags = PIPE_CONTROL_3D_ENGINE_FLAGS;
+
+ /* See __xe_pt_bind_vma() for a discussion on TLB invalidations. */
+ i = emit_pipe_invalidate(mask_flags, vm && vm->batch_invalidate_tlb, dw, i);
+
+ /* hsdes: 1809175790 */
+ if (has_aux_ccs(xe))
+ i = emit_aux_table_inv(gt, CCS_AUX_INV, dw, i);
+
+ dw[i++] = preparser_disable(false);
+
+ i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, dw, i);
+
+ i = emit_bb_start(batch_addr, ppgtt_flag, dw, i);
+
+ i = emit_render_cache_flush(job, dw, i);
+
+ if (job->user_fence.used)
+ i = emit_store_imm_ppgtt_posted(job->user_fence.addr,
+ job->user_fence.value,
+ dw, i);
+
+ i = emit_pipe_imm_ggtt(xe_lrc_seqno_ggtt_addr(lrc), seqno, lacks_render, dw, i);
+
+ i = emit_user_interrupt(dw, i);
+
+ i = emit_pipe_control_to_ring_end(job->q->hwe, dw, i);
+
+ xe_gt_assert(gt, i <= MAX_JOB_SIZE_DW);
+
+ xe_lrc_write_ring(lrc, dw, i * sizeof(*dw));
+}
+
+static void emit_migration_job_gen12(struct xe_sched_job *job,
+ struct xe_lrc *lrc, u32 seqno)
+{
+ u32 dw[MAX_JOB_SIZE_DW], i = 0;
+
+ i = emit_store_imm_ggtt(xe_lrc_start_seqno_ggtt_addr(lrc),
+ seqno, dw, i);
+
+ dw[i++] = MI_ARB_ON_OFF | MI_ARB_DISABLE; /* Enabled again below */
+
+ i = emit_bb_start(job->batch_addr[0], BIT(8), dw, i);
+
+ /* XXX: Do we need this? Leaving for now. */
+ dw[i++] = preparser_disable(true);
+ i = emit_flush_invalidate(0, dw, i);
+ dw[i++] = preparser_disable(false);
+
+ i = emit_bb_start(job->batch_addr[1], BIT(8), dw, i);
+
+ dw[i++] = MI_FLUSH_DW | MI_INVALIDATE_TLB | job->migrate_flush_flags |
+ MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_IMM_DW;
+ dw[i++] = xe_lrc_seqno_ggtt_addr(lrc) | MI_FLUSH_DW_USE_GTT;
+ dw[i++] = 0;
+ dw[i++] = seqno; /* value */
+
+ i = emit_user_interrupt(dw, i);
+
+ xe_gt_assert(job->q->gt, i <= MAX_JOB_SIZE_DW);
+
+ xe_lrc_write_ring(lrc, dw, i * sizeof(*dw));
+}
+
+static void emit_job_gen12_gsc(struct xe_sched_job *job)
+{
+ struct xe_gt *gt = job->q->gt;
+
+ xe_gt_assert(gt, job->q->width <= 1); /* no parallel submission for GSCCS */
+
+ __emit_job_gen12_simple(job, job->q->lrc,
+ job->batch_addr[0],
+ xe_sched_job_seqno(job));
+}
+
+static void emit_job_gen12_copy(struct xe_sched_job *job)
+{
+ int i;
+
+ if (xe_sched_job_is_migration(job->q)) {
+ emit_migration_job_gen12(job, job->q->lrc,
+ xe_sched_job_seqno(job));
+ return;
+ }
+
+ for (i = 0; i < job->q->width; ++i)
+ __emit_job_gen12_simple(job, job->q->lrc + i,
+ job->batch_addr[i],
+ xe_sched_job_seqno(job));
+}
+
+static void emit_job_gen12_video(struct xe_sched_job *job)
+{
+ int i;
+
+ /* FIXME: Not doing parallel handshake for now */
+ for (i = 0; i < job->q->width; ++i)
+ __emit_job_gen12_video(job, job->q->lrc + i,
+ job->batch_addr[i],
+ xe_sched_job_seqno(job));
+}
+
+static void emit_job_gen12_render_compute(struct xe_sched_job *job)
+{
+ int i;
+
+ for (i = 0; i < job->q->width; ++i)
+ __emit_job_gen12_render_compute(job, job->q->lrc + i,
+ job->batch_addr[i],
+ xe_sched_job_seqno(job));
+}
+
+static const struct xe_ring_ops ring_ops_gen12_gsc = {
+ .emit_job = emit_job_gen12_gsc,
+};
+
+static const struct xe_ring_ops ring_ops_gen12_copy = {
+ .emit_job = emit_job_gen12_copy,
+};
+
+static const struct xe_ring_ops ring_ops_gen12_video = {
+ .emit_job = emit_job_gen12_video,
+};
+
+static const struct xe_ring_ops ring_ops_gen12_render_compute = {
+ .emit_job = emit_job_gen12_render_compute,
+};
+
+const struct xe_ring_ops *
+xe_ring_ops_get(struct xe_gt *gt, enum xe_engine_class class)
+{
+ switch (class) {
+ case XE_ENGINE_CLASS_OTHER:
+ return &ring_ops_gen12_gsc;
+ case XE_ENGINE_CLASS_COPY:
+ return &ring_ops_gen12_copy;
+ case XE_ENGINE_CLASS_VIDEO_DECODE:
+ case XE_ENGINE_CLASS_VIDEO_ENHANCE:
+ return &ring_ops_gen12_video;
+ case XE_ENGINE_CLASS_RENDER:
+ case XE_ENGINE_CLASS_COMPUTE:
+ return &ring_ops_gen12_render_compute;
+ default:
+ return NULL;
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_ring_ops.h b/drivers/gpu/drm/xe/xe_ring_ops.h
new file mode 100644
index 000000000000..e942735d76a6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ring_ops.h
@@ -0,0 +1,17 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_RING_OPS_H_
+#define _XE_RING_OPS_H_
+
+#include "xe_hw_engine_types.h"
+#include "xe_ring_ops_types.h"
+
+struct xe_gt;
+
+const struct xe_ring_ops *
+xe_ring_ops_get(struct xe_gt *gt, enum xe_engine_class class);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ring_ops_types.h b/drivers/gpu/drm/xe/xe_ring_ops_types.h
new file mode 100644
index 000000000000..1ae56e2ee7b4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ring_ops_types.h
@@ -0,0 +1,22 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_RING_OPS_TYPES_H_
+#define _XE_RING_OPS_TYPES_H_
+
+struct xe_sched_job;
+
+#define MAX_JOB_SIZE_DW 48
+#define MAX_JOB_SIZE_BYTES (MAX_JOB_SIZE_DW * 4)
+
+/**
+ * struct xe_ring_ops - Ring operations
+ */
+struct xe_ring_ops {
+ /** @emit_job: Write job to ring */
+ void (*emit_job)(struct xe_sched_job *job);
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_rtp.c b/drivers/gpu/drm/xe/xe_rtp.c
new file mode 100644
index 000000000000..fb44cc7521d8
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_rtp.c
@@ -0,0 +1,325 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_rtp.h"
+
+#include <kunit/visibility.h>
+
+#include <drm/xe_drm.h>
+
+#include "xe_gt.h"
+#include "xe_gt_topology.h"
+#include "xe_macros.h"
+#include "xe_reg_sr.h"
+
+/**
+ * DOC: Register Table Processing
+ *
+ * Internal infrastructure to define how registers should be updated based on
+ * rules and actions. This can be used to define tables with multiple entries
+ * (one per register) that will be walked over at some point in time to apply
+ * the values to the registers that have matching rules.
+ */
+
+static bool has_samedia(const struct xe_device *xe)
+{
+ return xe->info.media_verx100 >= 1300;
+}
+
+static bool rule_matches(const struct xe_device *xe,
+ struct xe_gt *gt,
+ struct xe_hw_engine *hwe,
+ const struct xe_rtp_rule *rules,
+ unsigned int n_rules)
+{
+ const struct xe_rtp_rule *r;
+ unsigned int i;
+ bool match;
+
+ for (r = rules, i = 0; i < n_rules; r = &rules[++i]) {
+ switch (r->match_type) {
+ case XE_RTP_MATCH_PLATFORM:
+ match = xe->info.platform == r->platform;
+ break;
+ case XE_RTP_MATCH_SUBPLATFORM:
+ match = xe->info.platform == r->platform &&
+ xe->info.subplatform == r->subplatform;
+ break;
+ case XE_RTP_MATCH_GRAPHICS_VERSION:
+ match = xe->info.graphics_verx100 == r->ver_start &&
+ (!has_samedia(xe) || !xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_GRAPHICS_VERSION_RANGE:
+ match = xe->info.graphics_verx100 >= r->ver_start &&
+ xe->info.graphics_verx100 <= r->ver_end &&
+ (!has_samedia(xe) || !xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_GRAPHICS_STEP:
+ match = xe->info.step.graphics >= r->step_start &&
+ xe->info.step.graphics < r->step_end &&
+ (!has_samedia(xe) || !xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_MEDIA_VERSION:
+ match = xe->info.media_verx100 == r->ver_start &&
+ (!has_samedia(xe) || xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_MEDIA_VERSION_RANGE:
+ match = xe->info.media_verx100 >= r->ver_start &&
+ xe->info.media_verx100 <= r->ver_end &&
+ (!has_samedia(xe) || xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_MEDIA_STEP:
+ match = xe->info.step.media >= r->step_start &&
+ xe->info.step.media < r->step_end &&
+ (!has_samedia(xe) || xe_gt_is_media_type(gt));
+ break;
+ case XE_RTP_MATCH_INTEGRATED:
+ match = !xe->info.is_dgfx;
+ break;
+ case XE_RTP_MATCH_DISCRETE:
+ match = xe->info.is_dgfx;
+ break;
+ case XE_RTP_MATCH_ENGINE_CLASS:
+ if (drm_WARN_ON(&xe->drm, !hwe))
+ return false;
+
+ match = hwe->class == r->engine_class;
+ break;
+ case XE_RTP_MATCH_NOT_ENGINE_CLASS:
+ if (drm_WARN_ON(&xe->drm, !hwe))
+ return false;
+
+ match = hwe->class != r->engine_class;
+ break;
+ case XE_RTP_MATCH_FUNC:
+ match = r->match_func(gt, hwe);
+ break;
+ default:
+ drm_warn(&xe->drm, "Invalid RTP match %u\n",
+ r->match_type);
+ match = false;
+ }
+
+ if (!match)
+ return false;
+ }
+
+ return true;
+}
+
+static void rtp_add_sr_entry(const struct xe_rtp_action *action,
+ struct xe_gt *gt,
+ u32 mmio_base,
+ struct xe_reg_sr *sr)
+{
+ struct xe_reg_sr_entry sr_entry = {
+ .reg = action->reg,
+ .clr_bits = action->clr_bits,
+ .set_bits = action->set_bits,
+ .read_mask = action->read_mask,
+ };
+
+ sr_entry.reg.addr += mmio_base;
+ xe_reg_sr_add(sr, &sr_entry, gt);
+}
+
+static bool rtp_process_one_sr(const struct xe_rtp_entry_sr *entry,
+ struct xe_device *xe, struct xe_gt *gt,
+ struct xe_hw_engine *hwe, struct xe_reg_sr *sr)
+{
+ const struct xe_rtp_action *action;
+ u32 mmio_base;
+ unsigned int i;
+
+ if (!rule_matches(xe, gt, hwe, entry->rules, entry->n_rules))
+ return false;
+
+ for (i = 0, action = &entry->actions[0]; i < entry->n_actions; action++, i++) {
+ if ((entry->flags & XE_RTP_ENTRY_FLAG_FOREACH_ENGINE) ||
+ (action->flags & XE_RTP_ACTION_FLAG_ENGINE_BASE))
+ mmio_base = hwe->mmio_base;
+ else
+ mmio_base = 0;
+
+ rtp_add_sr_entry(action, gt, mmio_base, sr);
+ }
+
+ return true;
+}
+
+static void rtp_get_context(struct xe_rtp_process_ctx *ctx,
+ struct xe_hw_engine **hwe,
+ struct xe_gt **gt,
+ struct xe_device **xe)
+{
+ switch (ctx->type) {
+ case XE_RTP_PROCESS_TYPE_GT:
+ *hwe = NULL;
+ *gt = ctx->gt;
+ *xe = gt_to_xe(*gt);
+ break;
+ case XE_RTP_PROCESS_TYPE_ENGINE:
+ *hwe = ctx->hwe;
+ *gt = (*hwe)->gt;
+ *xe = gt_to_xe(*gt);
+ break;
+ };
+}
+
+/**
+ * xe_rtp_process_ctx_enable_active_tracking - Enable tracking of active entries
+ *
+ * Set additional metadata to track what entries are considered "active", i.e.
+ * their rules match the condition. Bits are never cleared: entries with
+ * matching rules set the corresponding bit in the bitmap.
+ *
+ * @ctx: The context for processing the table
+ * @active_entries: bitmap to store the active entries
+ * @n_entries: number of entries to be processed
+ */
+void xe_rtp_process_ctx_enable_active_tracking(struct xe_rtp_process_ctx *ctx,
+ unsigned long *active_entries,
+ size_t n_entries)
+{
+ ctx->active_entries = active_entries;
+ ctx->n_entries = n_entries;
+}
+
+static void rtp_mark_active(struct xe_device *xe,
+ struct xe_rtp_process_ctx *ctx,
+ unsigned int first, unsigned int last)
+{
+ if (!ctx->active_entries)
+ return;
+
+ if (drm_WARN_ON(&xe->drm, last > ctx->n_entries))
+ return;
+
+ if (first == last)
+ bitmap_set(ctx->active_entries, first, 1);
+ else
+ bitmap_set(ctx->active_entries, first, last - first + 2);
+}
+
+/**
+ * xe_rtp_process_to_sr - Process all rtp @entries, adding the matching ones to
+ * the save-restore argument.
+ * @ctx: The context for processing the table, with one of device, gt or hwe
+ * @entries: Table with RTP definitions
+ * @sr: Save-restore struct where matching rules execute the action. This can be
+ * viewed as the "coalesced view" of multiple the tables. The bits for each
+ * register set are expected not to collide with previously added entries
+ *
+ * Walk the table pointed by @entries (with an empty sentinel) and add all
+ * entries with matching rules to @sr. If @hwe is not NULL, its mmio_base is
+ * used to calculate the right register offset
+ */
+void xe_rtp_process_to_sr(struct xe_rtp_process_ctx *ctx,
+ const struct xe_rtp_entry_sr *entries,
+ struct xe_reg_sr *sr)
+{
+ const struct xe_rtp_entry_sr *entry;
+ struct xe_hw_engine *hwe = NULL;
+ struct xe_gt *gt = NULL;
+ struct xe_device *xe = NULL;
+
+ rtp_get_context(ctx, &hwe, &gt, &xe);
+
+ for (entry = entries; entry && entry->name; entry++) {
+ bool match = false;
+
+ if (entry->flags & XE_RTP_ENTRY_FLAG_FOREACH_ENGINE) {
+ struct xe_hw_engine *each_hwe;
+ enum xe_hw_engine_id id;
+
+ for_each_hw_engine(each_hwe, gt, id)
+ match |= rtp_process_one_sr(entry, xe, gt,
+ each_hwe, sr);
+ } else {
+ match = rtp_process_one_sr(entry, xe, gt, hwe, sr);
+ }
+
+ if (match)
+ rtp_mark_active(xe, ctx, entry - entries,
+ entry - entries);
+ }
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_rtp_process_to_sr);
+
+/**
+ * xe_rtp_process - Process all rtp @entries, without running any action
+ * @ctx: The context for processing the table, with one of device, gt or hwe
+ * @entries: Table with RTP definitions
+ *
+ * Walk the table pointed by @entries (with an empty sentinel), executing the
+ * rules. A few differences from xe_rtp_process_to_sr():
+ *
+ * 1. There is no action associated with each entry since this uses
+ * struct xe_rtp_entry. Its main use is for marking active workarounds via
+ * xe_rtp_process_ctx_enable_active_tracking().
+ * 2. There is support for OR operations by having entries with no name.
+ */
+void xe_rtp_process(struct xe_rtp_process_ctx *ctx,
+ const struct xe_rtp_entry *entries)
+{
+ const struct xe_rtp_entry *entry, *first_entry;
+ struct xe_hw_engine *hwe;
+ struct xe_gt *gt;
+ struct xe_device *xe;
+
+ rtp_get_context(ctx, &hwe, &gt, &xe);
+
+ first_entry = entries;
+ if (drm_WARN_ON(&xe->drm, !first_entry->name))
+ return;
+
+ for (entry = entries; entry && entry->rules; entry++) {
+ if (entry->name)
+ first_entry = entry;
+
+ if (!rule_matches(xe, gt, hwe, entry->rules, entry->n_rules))
+ continue;
+
+ /* Fast-forward entry, eliminating the OR'ed entries */
+ for (entry++; entry && entry->rules; entry++)
+ if (entry->name)
+ break;
+ entry--;
+
+ rtp_mark_active(xe, ctx, first_entry - entries,
+ entry - entries);
+ }
+}
+
+bool xe_rtp_match_even_instance(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe)
+{
+ return hwe->instance % 2 == 0;
+}
+
+bool xe_rtp_match_first_render_or_compute(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe)
+{
+ u64 render_compute_mask = gt->info.engine_mask &
+ (XE_HW_ENGINE_CCS_MASK | XE_HW_ENGINE_RCS_MASK);
+
+ return render_compute_mask &&
+ hwe->engine_id == __ffs(render_compute_mask);
+}
+
+bool xe_rtp_match_first_gslice_fused_off(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe)
+{
+ unsigned int dss_per_gslice = 4;
+ unsigned int dss;
+
+ if (drm_WARN(&gt_to_xe(gt)->drm, xe_dss_mask_empty(gt->fuse_topo.g_dss_mask),
+ "Checking gslice for platform without geometry pipeline\n"))
+ return false;
+
+ dss = xe_dss_mask_group_ffs(gt->fuse_topo.g_dss_mask, 0, 0);
+
+ return dss >= dss_per_gslice;
+}
diff --git a/drivers/gpu/drm/xe/xe_rtp.h b/drivers/gpu/drm/xe/xe_rtp.h
new file mode 100644
index 000000000000..c56fedd126e6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_rtp.h
@@ -0,0 +1,430 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_RTP_
+#define _XE_RTP_
+
+#include <linux/types.h>
+#include <linux/xarray.h>
+
+#define _XE_RTP_INCLUDE_PRIVATE_HELPERS
+
+#include "xe_rtp_helpers.h"
+#include "xe_rtp_types.h"
+
+#undef _XE_RTP_INCLUDE_PRIVATE_HELPERS
+
+/*
+ * Register table poke infrastructure
+ */
+
+struct xe_hw_engine;
+struct xe_gt;
+struct xe_reg_sr;
+
+/*
+ * Macros to encode rules to match against platform, IP version, stepping, etc.
+ * Shouldn't be used directly - see XE_RTP_RULES()
+ */
+#define _XE_RTP_RULE_PLATFORM(plat__) \
+ { .match_type = XE_RTP_MATCH_PLATFORM, .platform = plat__ }
+
+#define _XE_RTP_RULE_SUBPLATFORM(plat__, sub__) \
+ { .match_type = XE_RTP_MATCH_SUBPLATFORM, \
+ .platform = plat__, .subplatform = sub__ }
+
+#define _XE_RTP_RULE_GRAPHICS_STEP(start__, end__) \
+ { .match_type = XE_RTP_MATCH_GRAPHICS_STEP, \
+ .step_start = start__, .step_end = end__ }
+
+#define _XE_RTP_RULE_MEDIA_STEP(start__, end__) \
+ { .match_type = XE_RTP_MATCH_MEDIA_STEP, \
+ .step_start = start__, .step_end = end__ }
+
+#define _XE_RTP_RULE_ENGINE_CLASS(cls__) \
+ { .match_type = XE_RTP_MATCH_ENGINE_CLASS, \
+ .engine_class = (cls__) }
+
+/**
+ * XE_RTP_RULE_PLATFORM - Create rule matching platform
+ * @plat_: platform to match
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_PLATFORM(plat_) \
+ _XE_RTP_RULE_PLATFORM(XE_##plat_)
+
+/**
+ * XE_RTP_RULE_SUBPLATFORM - Create rule matching platform and sub-platform
+ * @plat_: platform to match
+ * @sub_: sub-platform to match
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_SUBPLATFORM(plat_, sub_) \
+ _XE_RTP_RULE_SUBPLATFORM(XE_##plat_, XE_SUBPLATFORM_##plat_##_##sub_)
+
+/**
+ * XE_RTP_RULE_GRAPHICS_STEP - Create rule matching graphics stepping
+ * @start_: First stepping matching the rule
+ * @end_: First stepping that does not match the rule
+ *
+ * Note that the range matching this rule is [ @start_, @end_ ), i.e. inclusive
+ * on the left, exclusive on the right.
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_GRAPHICS_STEP(start_, end_) \
+ _XE_RTP_RULE_GRAPHICS_STEP(STEP_##start_, STEP_##end_)
+
+/**
+ * XE_RTP_RULE_MEDIA_STEP - Create rule matching media stepping
+ * @start_: First stepping matching the rule
+ * @end_: First stepping that does not match the rule
+ *
+ * Note that the range matching this rule is [ @start_, @end_ ), i.e. inclusive
+ * on the left, exclusive on the right.
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_MEDIA_STEP(start_, end_) \
+ _XE_RTP_RULE_MEDIA_STEP(STEP_##start_, STEP_##end_)
+
+/**
+ * XE_RTP_RULE_ENGINE_CLASS - Create rule matching an engine class
+ * @cls_: Engine class to match
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_ENGINE_CLASS(cls_) \
+ _XE_RTP_RULE_ENGINE_CLASS(XE_ENGINE_CLASS_##cls_)
+
+/**
+ * XE_RTP_RULE_FUNC - Create rule using callback function for match
+ * @func__: Function to call to decide if rule matches
+ *
+ * This allows more complex checks to be performed. The ``XE_RTP``
+ * infrastructure will simply call the function @func_ passed to decide if this
+ * rule matches the device.
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_FUNC(func__) \
+ { .match_type = XE_RTP_MATCH_FUNC, \
+ .match_func = (func__) }
+
+/**
+ * XE_RTP_RULE_GRAPHICS_VERSION - Create rule matching graphics version
+ * @ver__: Graphics IP version to match
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_GRAPHICS_VERSION(ver__) \
+ { .match_type = XE_RTP_MATCH_GRAPHICS_VERSION, \
+ .ver_start = ver__, }
+
+/**
+ * XE_RTP_RULE_GRAPHICS_VERSION_RANGE - Create rule matching a range of graphics version
+ * @ver_start__: First graphics IP version to match
+ * @ver_end__: Last graphics IP version to match
+ *
+ * Note that the range matching this rule is [ @ver_start__, @ver_end__ ], i.e.
+ * inclusive on boths sides
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_GRAPHICS_VERSION_RANGE(ver_start__, ver_end__) \
+ { .match_type = XE_RTP_MATCH_GRAPHICS_VERSION_RANGE, \
+ .ver_start = ver_start__, .ver_end = ver_end__, }
+
+/**
+ * XE_RTP_RULE_MEDIA_VERSION - Create rule matching media version
+ * @ver__: Graphics IP version to match
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_MEDIA_VERSION(ver__) \
+ { .match_type = XE_RTP_MATCH_MEDIA_VERSION, \
+ .ver_start = ver__, }
+
+/**
+ * XE_RTP_RULE_MEDIA_VERSION_RANGE - Create rule matching a range of media version
+ * @ver_start__: First media IP version to match
+ * @ver_end__: Last media IP version to match
+ *
+ * Note that the range matching this rule is [ @ver_start__, @ver_end__ ], i.e.
+ * inclusive on boths sides
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_MEDIA_VERSION_RANGE(ver_start__, ver_end__) \
+ { .match_type = XE_RTP_MATCH_MEDIA_VERSION_RANGE, \
+ .ver_start = ver_start__, .ver_end = ver_end__, }
+
+/**
+ * XE_RTP_RULE_IS_INTEGRATED - Create a rule matching integrated graphics devices
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_IS_INTEGRATED \
+ { .match_type = XE_RTP_MATCH_INTEGRATED }
+
+/**
+ * XE_RTP_RULE_IS_DISCRETE - Create a rule matching discrete graphics devices
+ *
+ * Refer to XE_RTP_RULES() for expected usage.
+ */
+#define XE_RTP_RULE_IS_DISCRETE \
+ { .match_type = XE_RTP_MATCH_DISCRETE }
+
+/**
+ * XE_RTP_ACTION_WR - Helper to write a value to the register, overriding all
+ * the bits
+ * @reg_: Register
+ * @val_: Value to set
+ * @...: Additional fields to override in the struct xe_rtp_action entry
+ *
+ * The correspondent notation in bspec is:
+ *
+ * REGNAME = VALUE
+ */
+#define XE_RTP_ACTION_WR(reg_, val_, ...) \
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .clr_bits = ~0u, .set_bits = (val_), \
+ .read_mask = (~0u), ##__VA_ARGS__ }
+
+/**
+ * XE_RTP_ACTION_SET - Set bits from @val_ in the register.
+ * @reg_: Register
+ * @val_: Bits to set in the register
+ * @...: Additional fields to override in the struct xe_rtp_action entry
+ *
+ * For masked registers this translates to a single write, while for other
+ * registers it's a RMW. The correspondent bspec notation is (example for bits 2
+ * and 5, but could be any):
+ *
+ * REGNAME[2] = 1
+ * REGNAME[5] = 1
+ */
+#define XE_RTP_ACTION_SET(reg_, val_, ...) \
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .clr_bits = val_, .set_bits = val_, \
+ .read_mask = val_, ##__VA_ARGS__ }
+
+/**
+ * XE_RTP_ACTION_CLR: Clear bits from @val_ in the register.
+ * @reg_: Register
+ * @val_: Bits to clear in the register
+ * @...: Additional fields to override in the struct xe_rtp_action entry
+ *
+ * For masked registers this translates to a single write, while for other
+ * registers it's a RMW. The correspondent bspec notation is (example for bits 2
+ * and 5, but could be any):
+ *
+ * REGNAME[2] = 0
+ * REGNAME[5] = 0
+ */
+#define XE_RTP_ACTION_CLR(reg_, val_, ...) \
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .clr_bits = val_, .set_bits = 0, \
+ .read_mask = val_, ##__VA_ARGS__ }
+
+/**
+ * XE_RTP_ACTION_FIELD_SET: Set a bit range
+ * @reg_: Register
+ * @mask_bits_: Mask of bits to be changed in the register, forming a field
+ * @val_: Value to set in the field denoted by @mask_bits_
+ * @...: Additional fields to override in the struct xe_rtp_action entry
+ *
+ * For masked registers this translates to a single write, while for other
+ * registers it's a RMW. The correspondent bspec notation is:
+ *
+ * REGNAME[<end>:<start>] = VALUE
+ */
+#define XE_RTP_ACTION_FIELD_SET(reg_, mask_bits_, val_, ...) \
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .clr_bits = mask_bits_, .set_bits = val_, \
+ .read_mask = mask_bits_, ##__VA_ARGS__ }
+
+#define XE_RTP_ACTION_FIELD_SET_NO_READ_MASK(reg_, mask_bits_, val_, ...) \
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .clr_bits = (mask_bits_), .set_bits = (val_), \
+ .read_mask = 0, ##__VA_ARGS__ }
+
+/**
+ * XE_RTP_ACTION_WHITELIST - Add register to userspace whitelist
+ * @reg_: Register
+ * @val_: Whitelist-specific flags to set
+ * @...: Additional fields to override in the struct xe_rtp_action entry
+ *
+ * Add a register to the whitelist, allowing userspace to modify the ster with
+ * regular user privileges.
+ */
+#define XE_RTP_ACTION_WHITELIST(reg_, val_, ...) \
+ /* TODO fail build if ((flags) & ~(RING_FORCE_TO_NONPRIV_MASK_VALID)) */\
+ { .reg = XE_RTP_DROP_CAST(reg_), \
+ .set_bits = val_, \
+ .clr_bits = RING_FORCE_TO_NONPRIV_MASK_VALID, \
+ ##__VA_ARGS__ }
+
+/**
+ * XE_RTP_NAME - Helper to set the name in xe_rtp_entry
+ * @s_: Name describing this rule, often a HW-specific number
+ *
+ * TODO: maybe move this behind a debug config?
+ */
+#define XE_RTP_NAME(s_) .name = (s_)
+
+/**
+ * XE_RTP_ENTRY_FLAG - Helper to add multiple flags to a struct xe_rtp_entry_sr
+ * @...: Entry flags, without the ``XE_RTP_ENTRY_FLAG_`` prefix
+ *
+ * Helper to automatically add a ``XE_RTP_ENTRY_FLAG_`` prefix to the flags
+ * when defining struct xe_rtp_entry entries. Example:
+ *
+ * .. code-block:: c
+ *
+ * const struct xe_rtp_entry_sr wa_entries[] = {
+ * ...
+ * { XE_RTP_NAME("test-entry"),
+ * ...
+ * XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ * ...
+ * },
+ * ...
+ * };
+ */
+#define XE_RTP_ENTRY_FLAG(...) \
+ .flags = (XE_RTP_PASTE_FOREACH(ENTRY_FLAG_, BITWISE_OR, (__VA_ARGS__)))
+
+/**
+ * XE_RTP_ACTION_FLAG - Helper to add multiple flags to a struct xe_rtp_action
+ * @...: Action flags, without the ``XE_RTP_ACTION_FLAG_`` prefix
+ *
+ * Helper to automatically add a ``XE_RTP_ACTION_FLAG_`` prefix to the flags
+ * when defining struct xe_rtp_action entries. Example:
+ *
+ * .. code-block:: c
+ *
+ * const struct xe_rtp_entry_sr wa_entries[] = {
+ * ...
+ * { XE_RTP_NAME("test-entry"),
+ * ...
+ * XE_RTP_ACTION_SET(..., XE_RTP_ACTION_FLAG(FOREACH_ENGINE)),
+ * ...
+ * },
+ * ...
+ * };
+ */
+#define XE_RTP_ACTION_FLAG(...) \
+ .flags = (XE_RTP_PASTE_FOREACH(ACTION_FLAG_, BITWISE_OR, (__VA_ARGS__)))
+
+/**
+ * XE_RTP_RULES - Helper to set multiple rules to a struct xe_rtp_entry_sr entry
+ * @...: Rules
+ *
+ * At least one rule is needed and up to 4 are supported. Multiple rules are
+ * AND'ed together, i.e. all the rules must evaluate to true for the entry to
+ * be processed. See XE_RTP_MATCH_* for the possible match rules. Example:
+ *
+ * .. code-block:: c
+ *
+ * const struct xe_rtp_entry_sr wa_entries[] = {
+ * ...
+ * { XE_RTP_NAME("test-entry"),
+ * XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ * ...
+ * },
+ * ...
+ * };
+ */
+#define XE_RTP_RULES(...) \
+ .n_rules = _XE_COUNT_ARGS(__VA_ARGS__), \
+ .rules = (const struct xe_rtp_rule[]) { \
+ XE_RTP_PASTE_FOREACH(RULE_, COMMA, (__VA_ARGS__)) \
+ }
+
+/**
+ * XE_RTP_ACTIONS - Helper to set multiple actions to a struct xe_rtp_entry_sr
+ * @...: Actions to be taken
+ *
+ * At least one action is needed and up to 4 are supported. See XE_RTP_ACTION_*
+ * for the possible actions. Example:
+ *
+ * .. code-block:: c
+ *
+ * const struct xe_rtp_entry_sr wa_entries[] = {
+ * ...
+ * { XE_RTP_NAME("test-entry"),
+ * XE_RTP_RULES(...),
+ * XE_RTP_ACTIONS(SET(..), SET(...), CLR(...)),
+ * ...
+ * },
+ * ...
+ * };
+ */
+#define XE_RTP_ACTIONS(...) \
+ .n_actions = _XE_COUNT_ARGS(__VA_ARGS__), \
+ .actions = (const struct xe_rtp_action[]) { \
+ XE_RTP_PASTE_FOREACH(ACTION_, COMMA, (__VA_ARGS__)) \
+ }
+
+#define XE_RTP_PROCESS_CTX_INITIALIZER(arg__) _Generic((arg__), \
+ struct xe_hw_engine * : (struct xe_rtp_process_ctx){ { (void *)(arg__) }, XE_RTP_PROCESS_TYPE_ENGINE }, \
+ struct xe_gt * : (struct xe_rtp_process_ctx){ { (void *)(arg__) }, XE_RTP_PROCESS_TYPE_GT })
+
+void xe_rtp_process_ctx_enable_active_tracking(struct xe_rtp_process_ctx *ctx,
+ unsigned long *active_entries,
+ size_t n_entries);
+
+void xe_rtp_process_to_sr(struct xe_rtp_process_ctx *ctx,
+ const struct xe_rtp_entry_sr *entries,
+ struct xe_reg_sr *sr);
+
+void xe_rtp_process(struct xe_rtp_process_ctx *ctx,
+ const struct xe_rtp_entry *entries);
+
+/* Match functions to be used with XE_RTP_MATCH_FUNC */
+
+/**
+ * xe_rtp_match_even_instance - Match if engine instance is even
+ * @gt: GT structure
+ * @hwe: Engine instance
+ *
+ * Returns: true if engine instance is even, false otherwise
+ */
+bool xe_rtp_match_even_instance(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe);
+
+/*
+ * xe_rtp_match_first_render_or_compute - Match if it's first render or compute
+ * engine in the GT
+ *
+ * @gt: GT structure
+ * @hwe: Engine instance
+ *
+ * Registers on the render reset domain need to have their values re-applied
+ * when any of those engines are reset. Since the engines reset together, a
+ * programming can be set to just one of them. For simplicity the first engine
+ * of either render or compute class can be chosen.
+ *
+ * Returns: true if engine id is the first to match the render reset domain,
+ * false otherwise.
+ */
+bool xe_rtp_match_first_render_or_compute(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe);
+
+/*
+ * xe_rtp_match_first_gslice_fused_off - Match when first gslice is fused off
+ *
+ * @gt: GT structure
+ * @hwe: Engine instance
+ *
+ * Returns: true if first gslice is fused off, false otherwise.
+ */
+bool xe_rtp_match_first_gslice_fused_off(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_rtp_helpers.h b/drivers/gpu/drm/xe/xe_rtp_helpers.h
new file mode 100644
index 000000000000..181b6290fac3
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_rtp_helpers.h
@@ -0,0 +1,81 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_RTP_HELPERS_
+#define _XE_RTP_HELPERS_
+
+#ifndef _XE_RTP_INCLUDE_PRIVATE_HELPERS
+#error "This header is supposed to be included by xe_rtp.h only"
+#endif
+
+/*
+ * Helper macros - not to be used outside this header.
+ */
+#define _XE_ESC(...) __VA_ARGS__
+#define _XE_COUNT_ARGS(...) _XE_ESC(__XE_COUNT_ARGS(__VA_ARGS__, 5, 4, 3, 2, 1,))
+#define __XE_COUNT_ARGS(_, _5, _4, _3, _2, X_, ...) X_
+
+#define _XE_FIRST(...) _XE_ESC(__XE_FIRST(__VA_ARGS__,))
+#define __XE_FIRST(x_, ...) x_
+#define _XE_TUPLE_TAIL(...) _XE_ESC(__XE_TUPLE_TAIL(__VA_ARGS__))
+#define __XE_TUPLE_TAIL(x_, ...) (__VA_ARGS__)
+
+#define _XE_DROP_FIRST(x_, ...) __VA_ARGS__
+
+#define _XE_RTP_CONCAT(a, b) __XE_RTP_CONCAT(a, b)
+#define __XE_RTP_CONCAT(a, b) XE_RTP_ ## a ## b
+
+#define __XE_RTP_PASTE_SEP_COMMA ,
+#define __XE_RTP_PASTE_SEP_BITWISE_OR |
+
+/*
+ * XE_RTP_PASTE_FOREACH - Paste XE_RTP_<@prefix_> on each element of the tuple
+ * @args, with the end result separated by @sep_. @sep must be one of the
+ * previously declared macros __XE_RTP_PASTE_SEP_*, or declared with such
+ * prefix.
+ *
+ * Examples:
+ *
+ * 1) XE_RTP_PASTE_FOREACH(TEST_, COMMA, (FOO, BAR))
+ * expands to:
+ *
+ * XE_RTP_TEST_FOO , XE_RTP_TEST_BAR
+ *
+ * 2) XE_RTP_PASTE_FOREACH(TEST2_, COMMA, (FOO))
+ * expands to:
+ *
+ * XE_RTP_TEST2_FOO
+ *
+ * 3) XE_RTP_PASTE_FOREACH(TEST3, BITWISE_OR, (FOO, BAR))
+ * expands to:
+ *
+ * XE_RTP_TEST3_FOO | XE_RTP_TEST3_BAR
+ *
+ * 4) #define __XE_RTP_PASTE_SEP_MY_SEP BANANA
+ * XE_RTP_PASTE_FOREACH(TEST_, MY_SEP, (FOO, BAR))
+ * expands to:
+ *
+ * XE_RTP_TEST_FOO BANANA XE_RTP_TEST_BAR
+ */
+#define XE_RTP_PASTE_FOREACH(prefix_, sep_, args_) _XE_ESC(_XE_RTP_CONCAT(PASTE_, _XE_COUNT_ARGS args_)(prefix_, sep_, args_))
+#define XE_RTP_PASTE_1(prefix_, sep_, args_) _XE_RTP_CONCAT(prefix_, _XE_FIRST args_)
+#define XE_RTP_PASTE_2(prefix_, sep_, args_) _XE_RTP_CONCAT(prefix_, _XE_FIRST args_) __XE_RTP_PASTE_SEP_ ## sep_ XE_RTP_PASTE_1(prefix_, sep_, _XE_TUPLE_TAIL args_)
+#define XE_RTP_PASTE_3(prefix_, sep_, args_) _XE_RTP_CONCAT(prefix_, _XE_FIRST args_) __XE_RTP_PASTE_SEP_ ## sep_ XE_RTP_PASTE_2(prefix_, sep_, _XE_TUPLE_TAIL args_)
+#define XE_RTP_PASTE_4(prefix_, sep_, args_) _XE_RTP_CONCAT(prefix_, _XE_FIRST args_) __XE_RTP_PASTE_SEP_ ## sep_ XE_RTP_PASTE_3(prefix_, sep_, _XE_TUPLE_TAIL args_)
+
+/*
+ * XE_RTP_DROP_CAST - Drop cast to convert a compound statement to a initializer
+ *
+ * Example:
+ *
+ * #define foo(a_) ((struct foo){ .a = a_ })
+ * XE_RTP_DROP_CAST(foo(10))
+ * expands to:
+ *
+ * { .a = 10 }
+ */
+#define XE_RTP_DROP_CAST(...) _XE_ESC(_XE_DROP_FIRST _XE_ESC __VA_ARGS__)
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_rtp_types.h b/drivers/gpu/drm/xe/xe_rtp_types.h
new file mode 100644
index 000000000000..637acc7626a4
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_rtp_types.h
@@ -0,0 +1,124 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_RTP_TYPES_
+#define _XE_RTP_TYPES_
+
+#include <linux/types.h>
+
+#include "regs/xe_reg_defs.h"
+
+struct xe_hw_engine;
+struct xe_gt;
+
+/**
+ * struct xe_rtp_action - action to take for any matching rule
+ *
+ * This struct records what action should be taken in a register that has a
+ * matching rule. Example of actions: set/clear bits.
+ */
+struct xe_rtp_action {
+ /** @reg: Register */
+ struct xe_reg reg;
+ /**
+ * @clr_bits: bits to clear when updating register. It's always a
+ * superset of bits being modified
+ */
+ u32 clr_bits;
+ /** @set_bits: bits to set when updating register */
+ u32 set_bits;
+#define XE_RTP_NOCHECK .read_mask = 0
+ /** @read_mask: mask for bits to consider when reading value back */
+ u32 read_mask;
+#define XE_RTP_ACTION_FLAG_ENGINE_BASE BIT(0)
+ /** @flags: flags to apply on rule evaluation or action */
+ u8 flags;
+};
+
+enum {
+ XE_RTP_MATCH_PLATFORM,
+ XE_RTP_MATCH_SUBPLATFORM,
+ XE_RTP_MATCH_GRAPHICS_VERSION,
+ XE_RTP_MATCH_GRAPHICS_VERSION_RANGE,
+ XE_RTP_MATCH_GRAPHICS_STEP,
+ XE_RTP_MATCH_MEDIA_VERSION,
+ XE_RTP_MATCH_MEDIA_VERSION_RANGE,
+ XE_RTP_MATCH_MEDIA_STEP,
+ XE_RTP_MATCH_INTEGRATED,
+ XE_RTP_MATCH_DISCRETE,
+ XE_RTP_MATCH_ENGINE_CLASS,
+ XE_RTP_MATCH_NOT_ENGINE_CLASS,
+ XE_RTP_MATCH_FUNC,
+};
+
+/** struct xe_rtp_rule - match rule for processing entry */
+struct xe_rtp_rule {
+ u8 match_type;
+
+ /* match filters */
+ union {
+ /* MATCH_PLATFORM / MATCH_SUBPLATFORM */
+ struct {
+ u8 platform;
+ u8 subplatform;
+ };
+ /*
+ * MATCH_GRAPHICS_VERSION / XE_RTP_MATCH_GRAPHICS_VERSION_RANGE /
+ * MATCH_MEDIA_VERSION / XE_RTP_MATCH_MEDIA_VERSION_RANGE
+ */
+ struct {
+ u32 ver_start;
+#define XE_RTP_END_VERSION_UNDEFINED U32_MAX
+ u32 ver_end;
+ };
+ /* MATCH_STEP */
+ struct {
+ u8 step_start;
+ u8 step_end;
+ };
+ /* MATCH_ENGINE_CLASS / MATCH_NOT_ENGINE_CLASS */
+ struct {
+ u8 engine_class;
+ };
+ /* MATCH_FUNC */
+ bool (*match_func)(const struct xe_gt *gt,
+ const struct xe_hw_engine *hwe);
+ };
+};
+
+/** struct xe_rtp_entry_sr - Entry in an rtp table */
+struct xe_rtp_entry_sr {
+ const char *name;
+ const struct xe_rtp_action *actions;
+ const struct xe_rtp_rule *rules;
+ u8 n_rules;
+ u8 n_actions;
+#define XE_RTP_ENTRY_FLAG_FOREACH_ENGINE BIT(0)
+ u8 flags;
+};
+
+/** struct xe_rtp_entry - Entry in an rtp table, with no action associated */
+struct xe_rtp_entry {
+ const char *name;
+ const struct xe_rtp_rule *rules;
+ u8 n_rules;
+};
+
+enum xe_rtp_process_type {
+ XE_RTP_PROCESS_TYPE_GT,
+ XE_RTP_PROCESS_TYPE_ENGINE,
+};
+
+struct xe_rtp_process_ctx {
+ union {
+ struct xe_gt *gt;
+ struct xe_hw_engine *hwe;
+ };
+ enum xe_rtp_process_type type;
+ unsigned long *active_entries;
+ size_t n_entries;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sa.c b/drivers/gpu/drm/xe/xe_sa.c
new file mode 100644
index 000000000000..2c4632259edd
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sa.c
@@ -0,0 +1,106 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_sa.h"
+
+#include <linux/kernel.h>
+
+#include <drm/drm_managed.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_map.h"
+
+static void xe_sa_bo_manager_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_sa_manager *sa_manager = arg;
+ struct xe_bo *bo = sa_manager->bo;
+
+ if (!bo) {
+ drm_err(drm, "no bo for sa manager\n");
+ return;
+ }
+
+ drm_suballoc_manager_fini(&sa_manager->base);
+
+ if (bo->vmap.is_iomem)
+ kvfree(sa_manager->cpu_ptr);
+
+ xe_bo_unpin_map_no_vm(bo);
+ sa_manager->bo = NULL;
+}
+
+struct xe_sa_manager *xe_sa_bo_manager_init(struct xe_tile *tile, u32 size, u32 align)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ u32 managed_size = size - SZ_4K;
+ struct xe_bo *bo;
+ int ret;
+
+ struct xe_sa_manager *sa_manager = drmm_kzalloc(&tile_to_xe(tile)->drm,
+ sizeof(*sa_manager),
+ GFP_KERNEL);
+ if (!sa_manager)
+ return ERR_PTR(-ENOMEM);
+
+ sa_manager->bo = NULL;
+
+ bo = xe_bo_create_pin_map(xe, tile, NULL, size, ttm_bo_type_kernel,
+ XE_BO_CREATE_VRAM_IF_DGFX(tile) |
+ XE_BO_CREATE_GGTT_BIT);
+ if (IS_ERR(bo)) {
+ drm_err(&xe->drm, "failed to allocate bo for sa manager: %ld\n",
+ PTR_ERR(bo));
+ return (struct xe_sa_manager *)bo;
+ }
+ sa_manager->bo = bo;
+
+ drm_suballoc_manager_init(&sa_manager->base, managed_size, align);
+ sa_manager->gpu_addr = xe_bo_ggtt_addr(bo);
+
+ if (bo->vmap.is_iomem) {
+ sa_manager->cpu_ptr = kvzalloc(managed_size, GFP_KERNEL);
+ if (!sa_manager->cpu_ptr) {
+ xe_bo_unpin_map_no_vm(sa_manager->bo);
+ sa_manager->bo = NULL;
+ return ERR_PTR(-ENOMEM);
+ }
+ } else {
+ sa_manager->cpu_ptr = bo->vmap.vaddr;
+ memset(sa_manager->cpu_ptr, 0, bo->ttm.base.size);
+ }
+
+ ret = drmm_add_action_or_reset(&xe->drm, xe_sa_bo_manager_fini,
+ sa_manager);
+ if (ret)
+ return ERR_PTR(ret);
+
+ return sa_manager;
+}
+
+struct drm_suballoc *xe_sa_bo_new(struct xe_sa_manager *sa_manager,
+ unsigned int size)
+{
+ return drm_suballoc_new(&sa_manager->base, size, GFP_KERNEL, true, 0);
+}
+
+void xe_sa_bo_flush_write(struct drm_suballoc *sa_bo)
+{
+ struct xe_sa_manager *sa_manager = to_xe_sa_manager(sa_bo->manager);
+ struct xe_device *xe = tile_to_xe(sa_manager->bo->tile);
+
+ if (!sa_manager->bo->vmap.is_iomem)
+ return;
+
+ xe_map_memcpy_to(xe, &sa_manager->bo->vmap, drm_suballoc_soffset(sa_bo),
+ xe_sa_bo_cpu_addr(sa_bo),
+ drm_suballoc_size(sa_bo));
+}
+
+void xe_sa_bo_free(struct drm_suballoc *sa_bo,
+ struct dma_fence *fence)
+{
+ drm_suballoc_free(sa_bo, fence);
+}
diff --git a/drivers/gpu/drm/xe/xe_sa.h b/drivers/gpu/drm/xe/xe_sa.h
new file mode 100644
index 000000000000..4e96483057d7
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sa.h
@@ -0,0 +1,40 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+#ifndef _XE_SA_H_
+#define _XE_SA_H_
+
+#include "xe_sa_types.h"
+
+struct dma_fence;
+struct xe_bo;
+struct xe_tile;
+
+struct xe_sa_manager *xe_sa_bo_manager_init(struct xe_tile *tile, u32 size, u32 align);
+
+struct drm_suballoc *xe_sa_bo_new(struct xe_sa_manager *sa_manager,
+ u32 size);
+void xe_sa_bo_flush_write(struct drm_suballoc *sa_bo);
+void xe_sa_bo_free(struct drm_suballoc *sa_bo,
+ struct dma_fence *fence);
+
+static inline struct xe_sa_manager *
+to_xe_sa_manager(struct drm_suballoc_manager *mng)
+{
+ return container_of(mng, struct xe_sa_manager, base);
+}
+
+static inline u64 xe_sa_bo_gpu_addr(struct drm_suballoc *sa)
+{
+ return to_xe_sa_manager(sa->manager)->gpu_addr +
+ drm_suballoc_soffset(sa);
+}
+
+static inline void *xe_sa_bo_cpu_addr(struct drm_suballoc *sa)
+{
+ return to_xe_sa_manager(sa->manager)->cpu_ptr +
+ drm_suballoc_soffset(sa);
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sa_types.h b/drivers/gpu/drm/xe/xe_sa_types.h
new file mode 100644
index 000000000000..2ef896aeca1d
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sa_types.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+#ifndef _XE_SA_TYPES_H_
+#define _XE_SA_TYPES_H_
+
+#include <drm/drm_suballoc.h>
+
+struct xe_bo;
+
+struct xe_sa_manager {
+ struct drm_suballoc_manager base;
+ struct xe_bo *bo;
+ u64 gpu_addr;
+ void *cpu_ptr;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sched_job.c b/drivers/gpu/drm/xe/xe_sched_job.c
new file mode 100644
index 000000000000..01106a1156ad
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sched_job.c
@@ -0,0 +1,280 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_sched_job.h"
+
+#include <linux/dma-fence-array.h>
+#include <linux/slab.h>
+
+#include "xe_device.h"
+#include "xe_exec_queue.h"
+#include "xe_gt.h"
+#include "xe_hw_engine_types.h"
+#include "xe_hw_fence.h"
+#include "xe_lrc.h"
+#include "xe_macros.h"
+#include "xe_trace.h"
+#include "xe_vm.h"
+
+static struct kmem_cache *xe_sched_job_slab;
+static struct kmem_cache *xe_sched_job_parallel_slab;
+
+int __init xe_sched_job_module_init(void)
+{
+ xe_sched_job_slab =
+ kmem_cache_create("xe_sched_job",
+ sizeof(struct xe_sched_job) +
+ sizeof(u64), 0,
+ SLAB_HWCACHE_ALIGN, NULL);
+ if (!xe_sched_job_slab)
+ return -ENOMEM;
+
+ xe_sched_job_parallel_slab =
+ kmem_cache_create("xe_sched_job_parallel",
+ sizeof(struct xe_sched_job) +
+ sizeof(u64) *
+ XE_HW_ENGINE_MAX_INSTANCE, 0,
+ SLAB_HWCACHE_ALIGN, NULL);
+ if (!xe_sched_job_parallel_slab) {
+ kmem_cache_destroy(xe_sched_job_slab);
+ return -ENOMEM;
+ }
+
+ return 0;
+}
+
+void xe_sched_job_module_exit(void)
+{
+ kmem_cache_destroy(xe_sched_job_slab);
+ kmem_cache_destroy(xe_sched_job_parallel_slab);
+}
+
+static struct xe_sched_job *job_alloc(bool parallel)
+{
+ return kmem_cache_zalloc(parallel ? xe_sched_job_parallel_slab :
+ xe_sched_job_slab, GFP_KERNEL);
+}
+
+bool xe_sched_job_is_migration(struct xe_exec_queue *q)
+{
+ return q->vm && (q->vm->flags & XE_VM_FLAG_MIGRATION);
+}
+
+static void job_free(struct xe_sched_job *job)
+{
+ struct xe_exec_queue *q = job->q;
+ bool is_migration = xe_sched_job_is_migration(q);
+
+ kmem_cache_free(xe_exec_queue_is_parallel(job->q) || is_migration ?
+ xe_sched_job_parallel_slab : xe_sched_job_slab, job);
+}
+
+static struct xe_device *job_to_xe(struct xe_sched_job *job)
+{
+ return gt_to_xe(job->q->gt);
+}
+
+struct xe_sched_job *xe_sched_job_create(struct xe_exec_queue *q,
+ u64 *batch_addr)
+{
+ struct xe_sched_job *job;
+ struct dma_fence **fences;
+ bool is_migration = xe_sched_job_is_migration(q);
+ int err;
+ int i, j;
+ u32 width;
+
+ /* only a kernel context can submit a vm-less job */
+ XE_WARN_ON(!q->vm && !(q->flags & EXEC_QUEUE_FLAG_KERNEL));
+
+ /* Migration and kernel engines have their own locking */
+ if (!(q->flags & (EXEC_QUEUE_FLAG_KERNEL | EXEC_QUEUE_FLAG_VM))) {
+ lockdep_assert_held(&q->vm->lock);
+ if (!xe_vm_in_lr_mode(q->vm))
+ xe_vm_assert_held(q->vm);
+ }
+
+ job = job_alloc(xe_exec_queue_is_parallel(q) || is_migration);
+ if (!job)
+ return ERR_PTR(-ENOMEM);
+
+ job->q = q;
+ kref_init(&job->refcount);
+ xe_exec_queue_get(job->q);
+
+ err = drm_sched_job_init(&job->drm, q->entity, 1, NULL);
+ if (err)
+ goto err_free;
+
+ if (!xe_exec_queue_is_parallel(q)) {
+ job->fence = xe_lrc_create_seqno_fence(q->lrc);
+ if (IS_ERR(job->fence)) {
+ err = PTR_ERR(job->fence);
+ goto err_sched_job;
+ }
+ } else {
+ struct dma_fence_array *cf;
+
+ fences = kmalloc_array(q->width, sizeof(*fences), GFP_KERNEL);
+ if (!fences) {
+ err = -ENOMEM;
+ goto err_sched_job;
+ }
+
+ for (j = 0; j < q->width; ++j) {
+ fences[j] = xe_lrc_create_seqno_fence(q->lrc + j);
+ if (IS_ERR(fences[j])) {
+ err = PTR_ERR(fences[j]);
+ goto err_fences;
+ }
+ }
+
+ cf = dma_fence_array_create(q->width, fences,
+ q->parallel.composite_fence_ctx,
+ q->parallel.composite_fence_seqno++,
+ false);
+ if (!cf) {
+ --q->parallel.composite_fence_seqno;
+ err = -ENOMEM;
+ goto err_fences;
+ }
+
+ /* Sanity check */
+ for (j = 0; j < q->width; ++j)
+ xe_assert(job_to_xe(job), cf->base.seqno == fences[j]->seqno);
+
+ job->fence = &cf->base;
+ }
+
+ width = q->width;
+ if (is_migration)
+ width = 2;
+
+ for (i = 0; i < width; ++i)
+ job->batch_addr[i] = batch_addr[i];
+
+ /* All other jobs require a VM to be open which has a ref */
+ if (unlikely(q->flags & EXEC_QUEUE_FLAG_KERNEL))
+ xe_device_mem_access_get(job_to_xe(job));
+ xe_device_assert_mem_access(job_to_xe(job));
+
+ trace_xe_sched_job_create(job);
+ return job;
+
+err_fences:
+ for (j = j - 1; j >= 0; --j) {
+ --q->lrc[j].fence_ctx.next_seqno;
+ dma_fence_put(fences[j]);
+ }
+ kfree(fences);
+err_sched_job:
+ drm_sched_job_cleanup(&job->drm);
+err_free:
+ xe_exec_queue_put(q);
+ job_free(job);
+ return ERR_PTR(err);
+}
+
+/**
+ * xe_sched_job_destroy - Destroy XE schedule job
+ * @ref: reference to XE schedule job
+ *
+ * Called when ref == 0, drop a reference to job's xe_engine + fence, cleanup
+ * base DRM schedule job, and free memory for XE schedule job.
+ */
+void xe_sched_job_destroy(struct kref *ref)
+{
+ struct xe_sched_job *job =
+ container_of(ref, struct xe_sched_job, refcount);
+
+ if (unlikely(job->q->flags & EXEC_QUEUE_FLAG_KERNEL))
+ xe_device_mem_access_put(job_to_xe(job));
+ xe_exec_queue_put(job->q);
+ dma_fence_put(job->fence);
+ drm_sched_job_cleanup(&job->drm);
+ job_free(job);
+}
+
+void xe_sched_job_set_error(struct xe_sched_job *job, int error)
+{
+ if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT, &job->fence->flags))
+ return;
+
+ dma_fence_set_error(job->fence, error);
+
+ if (dma_fence_is_array(job->fence)) {
+ struct dma_fence_array *array =
+ to_dma_fence_array(job->fence);
+ struct dma_fence **child = array->fences;
+ unsigned int nchild = array->num_fences;
+
+ do {
+ struct dma_fence *current_fence = *child++;
+
+ if (test_bit(DMA_FENCE_FLAG_SIGNALED_BIT,
+ &current_fence->flags))
+ continue;
+ dma_fence_set_error(current_fence, error);
+ } while (--nchild);
+ }
+
+ trace_xe_sched_job_set_error(job);
+
+ dma_fence_enable_sw_signaling(job->fence);
+ xe_hw_fence_irq_run(job->q->fence_irq);
+}
+
+bool xe_sched_job_started(struct xe_sched_job *job)
+{
+ struct xe_lrc *lrc = job->q->lrc;
+
+ return !__dma_fence_is_later(xe_sched_job_seqno(job),
+ xe_lrc_start_seqno(lrc),
+ job->fence->ops);
+}
+
+bool xe_sched_job_completed(struct xe_sched_job *job)
+{
+ struct xe_lrc *lrc = job->q->lrc;
+
+ /*
+ * Can safely check just LRC[0] seqno as that is last seqno written when
+ * parallel handshake is done.
+ */
+
+ return !__dma_fence_is_later(xe_sched_job_seqno(job), xe_lrc_seqno(lrc),
+ job->fence->ops);
+}
+
+void xe_sched_job_arm(struct xe_sched_job *job)
+{
+ drm_sched_job_arm(&job->drm);
+}
+
+void xe_sched_job_push(struct xe_sched_job *job)
+{
+ xe_sched_job_get(job);
+ trace_xe_sched_job_exec(job);
+ drm_sched_entity_push_job(&job->drm);
+ xe_sched_job_put(job);
+}
+
+/**
+ * xe_sched_job_last_fence_add_dep - Add last fence dependency to job
+ * @job:job to add the last fence dependency to
+ * @vm: virtual memory job belongs to
+ *
+ * Returns:
+ * 0 on success, or an error on failing to expand the array.
+ */
+int xe_sched_job_last_fence_add_dep(struct xe_sched_job *job, struct xe_vm *vm)
+{
+ struct dma_fence *fence;
+
+ fence = xe_exec_queue_last_fence_get(job->q, vm);
+ dma_fence_get(fence);
+
+ return drm_sched_job_add_dependency(&job->drm, fence);
+}
diff --git a/drivers/gpu/drm/xe/xe_sched_job.h b/drivers/gpu/drm/xe/xe_sched_job.h
new file mode 100644
index 000000000000..34f475ba7f50
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sched_job.h
@@ -0,0 +1,80 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_SCHED_JOB_H_
+#define _XE_SCHED_JOB_H_
+
+#include "xe_sched_job_types.h"
+
+struct xe_vm;
+
+#define XE_SCHED_HANG_LIMIT 1
+#define XE_SCHED_JOB_TIMEOUT LONG_MAX
+
+int xe_sched_job_module_init(void);
+void xe_sched_job_module_exit(void);
+
+struct xe_sched_job *xe_sched_job_create(struct xe_exec_queue *q,
+ u64 *batch_addr);
+void xe_sched_job_destroy(struct kref *ref);
+
+/**
+ * xe_sched_job_get - get reference to XE schedule job
+ * @job: XE schedule job object
+ *
+ * Increment XE schedule job's reference count
+ */
+static inline struct xe_sched_job *xe_sched_job_get(struct xe_sched_job *job)
+{
+ kref_get(&job->refcount);
+ return job;
+}
+
+/**
+ * xe_sched_job_put - put reference to XE schedule job
+ * @job: XE schedule job object
+ *
+ * Decrement XE schedule job's reference count, call xe_sched_job_destroy when
+ * reference count == 0.
+ */
+static inline void xe_sched_job_put(struct xe_sched_job *job)
+{
+ kref_put(&job->refcount, xe_sched_job_destroy);
+}
+
+void xe_sched_job_set_error(struct xe_sched_job *job, int error);
+static inline bool xe_sched_job_is_error(struct xe_sched_job *job)
+{
+ return job->fence->error < 0;
+}
+
+bool xe_sched_job_started(struct xe_sched_job *job);
+bool xe_sched_job_completed(struct xe_sched_job *job);
+
+void xe_sched_job_arm(struct xe_sched_job *job);
+void xe_sched_job_push(struct xe_sched_job *job);
+
+int xe_sched_job_last_fence_add_dep(struct xe_sched_job *job, struct xe_vm *vm);
+
+static inline struct xe_sched_job *
+to_xe_sched_job(struct drm_sched_job *drm)
+{
+ return container_of(drm, struct xe_sched_job, drm);
+}
+
+static inline u32 xe_sched_job_seqno(struct xe_sched_job *job)
+{
+ return job->fence->seqno;
+}
+
+static inline void
+xe_sched_job_add_migrate_flush(struct xe_sched_job *job, u32 flags)
+{
+ job->migrate_flush_flags = flags;
+}
+
+bool xe_sched_job_is_migration(struct xe_exec_queue *q);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sched_job_types.h b/drivers/gpu/drm/xe/xe_sched_job_types.h
new file mode 100644
index 000000000000..71213ba9735b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sched_job_types.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_SCHED_JOB_TYPES_H_
+#define _XE_SCHED_JOB_TYPES_H_
+
+#include <linux/kref.h>
+
+#include <drm/gpu_scheduler.h>
+
+struct xe_exec_queue;
+
+/**
+ * struct xe_sched_job - XE schedule job (batch buffer tracking)
+ */
+struct xe_sched_job {
+ /** @drm: base DRM scheduler job */
+ struct drm_sched_job drm;
+ /** @q: Exec queue */
+ struct xe_exec_queue *q;
+ /** @refcount: ref count of this job */
+ struct kref refcount;
+ /**
+ * @fence: dma fence to indicate completion. 1 way relationship - job
+ * can safely reference fence, fence cannot safely reference job.
+ */
+#define JOB_FLAG_SUBMIT DMA_FENCE_FLAG_USER_BITS
+ struct dma_fence *fence;
+ /** @user_fence: write back value when BB is complete */
+ struct {
+ /** @used: user fence is used */
+ bool used;
+ /** @addr: address to write to */
+ u64 addr;
+ /** @value: write back value */
+ u64 value;
+ } user_fence;
+ /** @migrate_flush_flags: Additional flush flags for migration jobs */
+ u32 migrate_flush_flags;
+ /** @batch_addr: batch buffer address of job */
+ u64 batch_addr[];
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sriov.c b/drivers/gpu/drm/xe/xe_sriov.c
new file mode 100644
index 000000000000..42a0e0c917a0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sriov.c
@@ -0,0 +1,55 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include "xe_assert.h"
+#include "xe_sriov.h"
+
+/**
+ * xe_sriov_mode_to_string - Convert enum value to string.
+ * @mode: the &xe_sriov_mode to convert
+ *
+ * Returns: SR-IOV mode as a user friendly string.
+ */
+const char *xe_sriov_mode_to_string(enum xe_sriov_mode mode)
+{
+ switch (mode) {
+ case XE_SRIOV_MODE_NONE:
+ return "none";
+ case XE_SRIOV_MODE_PF:
+ return "SR-IOV PF";
+ case XE_SRIOV_MODE_VF:
+ return "SR-IOV VF";
+ default:
+ return "<invalid>";
+ }
+}
+
+/**
+ * xe_sriov_probe_early - Probe a SR-IOV mode.
+ * @xe: the &xe_device to probe mode on
+ * @has_sriov: flag indicating hardware support for SR-IOV
+ *
+ * This function should be called only once and as soon as possible during
+ * driver probe to detect whether we are running a SR-IOV Physical Function
+ * (PF) or a Virtual Function (VF) device.
+ *
+ * SR-IOV PF mode detection is based on PCI @dev_is_pf() function.
+ * SR-IOV VF mode detection is based on dedicated MMIO register read.
+ */
+void xe_sriov_probe_early(struct xe_device *xe, bool has_sriov)
+{
+ enum xe_sriov_mode mode = XE_SRIOV_MODE_NONE;
+
+ /* TODO: replace with proper mode detection */
+ xe_assert(xe, !has_sriov);
+
+ xe_assert(xe, !xe->sriov.__mode);
+ xe->sriov.__mode = mode;
+ xe_assert(xe, xe->sriov.__mode);
+
+ if (has_sriov)
+ drm_info(&xe->drm, "Running in %s mode\n",
+ xe_sriov_mode_to_string(xe_device_sriov_mode(xe)));
+}
diff --git a/drivers/gpu/drm/xe/xe_sriov.h b/drivers/gpu/drm/xe/xe_sriov.h
new file mode 100644
index 000000000000..5af73a3172b0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sriov.h
@@ -0,0 +1,42 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_SRIOV_H_
+#define _XE_SRIOV_H_
+
+#include "xe_assert.h"
+#include "xe_device_types.h"
+#include "xe_sriov_types.h"
+
+const char *xe_sriov_mode_to_string(enum xe_sriov_mode mode);
+
+void xe_sriov_probe_early(struct xe_device *xe, bool has_sriov);
+
+static inline enum xe_sriov_mode xe_device_sriov_mode(struct xe_device *xe)
+{
+ xe_assert(xe, xe->sriov.__mode);
+ return xe->sriov.__mode;
+}
+
+static inline bool xe_device_is_sriov_pf(struct xe_device *xe)
+{
+ return xe_device_sriov_mode(xe) == XE_SRIOV_MODE_PF;
+}
+
+static inline bool xe_device_is_sriov_vf(struct xe_device *xe)
+{
+ return xe_device_sriov_mode(xe) == XE_SRIOV_MODE_VF;
+}
+
+#ifdef CONFIG_PCI_IOV
+#define IS_SRIOV_PF(xe) xe_device_is_sriov_pf(xe)
+#else
+#define IS_SRIOV_PF(xe) (typecheck(struct xe_device *, (xe)) && false)
+#endif
+#define IS_SRIOV_VF(xe) xe_device_is_sriov_vf(xe)
+
+#define IS_SRIOV(xe) (IS_SRIOV_PF(xe) || IS_SRIOV_VF(xe))
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sriov_printk.h b/drivers/gpu/drm/xe/xe_sriov_printk.h
new file mode 100644
index 000000000000..117e1d541692
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sriov_printk.h
@@ -0,0 +1,46 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_SRIOV_PRINTK_H_
+#define _XE_SRIOV_PRINTK_H_
+
+#include <drm/drm_print.h>
+
+#include "xe_device_types.h"
+#include "xe_sriov_types.h"
+
+#define xe_sriov_printk_prefix(xe) \
+ ((xe)->sriov.__mode == XE_SRIOV_MODE_PF ? "PF: " : \
+ (xe)->sriov.__mode == XE_SRIOV_MODE_VF ? "VF: " : "")
+
+#define xe_sriov_printk(xe, _level, fmt, ...) \
+ drm_##_level(&(xe)->drm, "%s" fmt, xe_sriov_printk_prefix(xe), ##__VA_ARGS__)
+
+#define xe_sriov_err(xe, fmt, ...) \
+ xe_sriov_printk((xe), err, fmt, ##__VA_ARGS__)
+
+#define xe_sriov_err_ratelimited(xe, fmt, ...) \
+ xe_sriov_printk((xe), err_ratelimited, fmt, ##__VA_ARGS__)
+
+#define xe_sriov_warn(xe, fmt, ...) \
+ xe_sriov_printk((xe), warn, fmt, ##__VA_ARGS__)
+
+#define xe_sriov_notice(xe, fmt, ...) \
+ xe_sriov_printk((xe), notice, fmt, ##__VA_ARGS__)
+
+#define xe_sriov_info(xe, fmt, ...) \
+ xe_sriov_printk((xe), info, fmt, ##__VA_ARGS__)
+
+#define xe_sriov_dbg(xe, fmt, ...) \
+ xe_sriov_printk((xe), dbg, fmt, ##__VA_ARGS__)
+
+/* for low level noisy debug messages */
+#ifdef CONFIG_DRM_XE_DEBUG_SRIOV
+#define xe_sriov_dbg_verbose(xe, fmt, ...) xe_sriov_dbg(xe, fmt, ##__VA_ARGS__)
+#else
+#define xe_sriov_dbg_verbose(xe, fmt, ...) typecheck(struct xe_device *, (xe))
+#endif
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sriov_types.h b/drivers/gpu/drm/xe/xe_sriov_types.h
new file mode 100644
index 000000000000..999a4311b98b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sriov_types.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_SRIOV_TYPES_H_
+#define _XE_SRIOV_TYPES_H_
+
+#include <linux/build_bug.h>
+
+/**
+ * enum xe_sriov_mode - SR-IOV mode
+ * @XE_SRIOV_MODE_NONE: bare-metal mode (non-virtualized)
+ * @XE_SRIOV_MODE_PF: SR-IOV Physical Function (PF) mode
+ * @XE_SRIOV_MODE_VF: SR-IOV Virtual Function (VF) mode
+ */
+enum xe_sriov_mode {
+ /*
+ * Note: We don't use default enum value 0 to allow catch any too early
+ * attempt of checking the SR-IOV mode prior to the actual mode probe.
+ */
+ XE_SRIOV_MODE_NONE = 1,
+ XE_SRIOV_MODE_PF,
+ XE_SRIOV_MODE_VF,
+};
+static_assert(XE_SRIOV_MODE_NONE);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_step.c b/drivers/gpu/drm/xe/xe_step.c
new file mode 100644
index 000000000000..eaf1b718f26c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_step.c
@@ -0,0 +1,264 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_step.h"
+
+#include <linux/bitfield.h>
+
+#include "xe_device.h"
+#include "xe_platform_types.h"
+
+/*
+ * Provide mapping between PCI's revision ID to the individual GMD
+ * (Graphics/Media/Display) stepping values that can be compared numerically.
+ *
+ * Some platforms may have unusual ways of mapping PCI revision ID to GMD
+ * steppings. E.g., in some cases a higher PCI revision may translate to a
+ * lower stepping of the GT and/or display IP.
+ *
+ * Also note that some revisions/steppings may have been set aside as
+ * placeholders but never materialized in real hardware; in those cases there
+ * may be jumps in the revision IDs or stepping values in the tables below.
+ */
+
+/*
+ * Some platforms always have the same stepping value for GT and display;
+ * use a macro to define these to make it easier to identify the platforms
+ * where the two steppings can deviate.
+ */
+#define COMMON_GT_MEDIA_STEP(x_) \
+ .graphics = STEP_##x_, \
+ .media = STEP_##x_
+
+#define COMMON_STEP(x_) \
+ COMMON_GT_MEDIA_STEP(x_), \
+ .graphics = STEP_##x_, \
+ .media = STEP_##x_, \
+ .display = STEP_##x_
+
+__diag_push();
+__diag_ignore_all("-Woverride-init", "Allow field overrides in table");
+
+/* Same GT stepping between tgl_uy_revids and tgl_revids don't mean the same HW */
+static const struct xe_step_info tgl_revids[] = {
+ [0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_B0 },
+ [1] = { COMMON_GT_MEDIA_STEP(B0), .display = STEP_D0 },
+};
+
+static const struct xe_step_info dg1_revids[] = {
+ [0] = { COMMON_STEP(A0) },
+ [1] = { COMMON_STEP(B0) },
+};
+
+static const struct xe_step_info adls_revids[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_A0 },
+ [0x1] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_A2 },
+ [0x4] = { COMMON_GT_MEDIA_STEP(B0), .display = STEP_B0 },
+ [0x8] = { COMMON_GT_MEDIA_STEP(C0), .display = STEP_B0 },
+ [0xC] = { COMMON_GT_MEDIA_STEP(D0), .display = STEP_C0 },
+};
+
+static const struct xe_step_info adls_rpls_revids[] = {
+ [0x4] = { COMMON_GT_MEDIA_STEP(D0), .display = STEP_D0 },
+ [0xC] = { COMMON_GT_MEDIA_STEP(D0), .display = STEP_C0 },
+};
+
+static const struct xe_step_info adlp_revids[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_A0 },
+ [0x4] = { COMMON_GT_MEDIA_STEP(B0), .display = STEP_B0 },
+ [0x8] = { COMMON_GT_MEDIA_STEP(C0), .display = STEP_C0 },
+ [0xC] = { COMMON_GT_MEDIA_STEP(C0), .display = STEP_D0 },
+};
+
+static const struct xe_step_info adlp_rpl_revids[] = {
+ [0x4] = { COMMON_GT_MEDIA_STEP(C0), .display = STEP_E0 },
+};
+
+static const struct xe_step_info adln_revids[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_D0 },
+};
+
+static const struct xe_step_info dg2_g10_revid_step_tbl[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_A0 },
+ [0x1] = { COMMON_GT_MEDIA_STEP(A1), .display = STEP_A0 },
+ [0x4] = { COMMON_GT_MEDIA_STEP(B0), .display = STEP_B0 },
+ [0x8] = { COMMON_GT_MEDIA_STEP(C0), .display = STEP_C0 },
+};
+
+static const struct xe_step_info dg2_g11_revid_step_tbl[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_B0 },
+ [0x4] = { COMMON_GT_MEDIA_STEP(B0), .display = STEP_C0 },
+ [0x5] = { COMMON_GT_MEDIA_STEP(B1), .display = STEP_C0 },
+};
+
+static const struct xe_step_info dg2_g12_revid_step_tbl[] = {
+ [0x0] = { COMMON_GT_MEDIA_STEP(A0), .display = STEP_C0 },
+ [0x1] = { COMMON_GT_MEDIA_STEP(A1), .display = STEP_C0 },
+};
+
+static const struct xe_step_info pvc_revid_step_tbl[] = {
+ [0x5] = { .graphics = STEP_B0 },
+ [0x6] = { .graphics = STEP_B1 },
+ [0x7] = { .graphics = STEP_C0 },
+};
+
+static const int pvc_basedie_subids[] = {
+ [0x3] = STEP_B0,
+ [0x4] = STEP_B1,
+ [0x5] = STEP_B3,
+};
+
+__diag_pop();
+
+/**
+ * xe_step_pre_gmdid_get - Determine IP steppings from PCI revid
+ * @xe: Xe device
+ *
+ * Convert the PCI revid into proper IP steppings. This should only be
+ * used on platforms that do not have GMD_ID support.
+ */
+struct xe_step_info xe_step_pre_gmdid_get(struct xe_device *xe)
+{
+ const struct xe_step_info *revids = NULL;
+ struct xe_step_info step = {};
+ u16 revid = xe->info.revid;
+ int size = 0;
+ const int *basedie_info = NULL;
+ int basedie_size = 0;
+ int baseid = 0;
+
+ if (xe->info.platform == XE_PVC) {
+ baseid = FIELD_GET(GENMASK(5, 3), xe->info.revid);
+ revid = FIELD_GET(GENMASK(2, 0), xe->info.revid);
+ revids = pvc_revid_step_tbl;
+ size = ARRAY_SIZE(pvc_revid_step_tbl);
+ basedie_info = pvc_basedie_subids;
+ basedie_size = ARRAY_SIZE(pvc_basedie_subids);
+ } else if (xe->info.subplatform == XE_SUBPLATFORM_DG2_G10) {
+ revids = dg2_g10_revid_step_tbl;
+ size = ARRAY_SIZE(dg2_g10_revid_step_tbl);
+ } else if (xe->info.subplatform == XE_SUBPLATFORM_DG2_G11) {
+ revids = dg2_g11_revid_step_tbl;
+ size = ARRAY_SIZE(dg2_g11_revid_step_tbl);
+ } else if (xe->info.subplatform == XE_SUBPLATFORM_DG2_G12) {
+ revids = dg2_g12_revid_step_tbl;
+ size = ARRAY_SIZE(dg2_g12_revid_step_tbl);
+ } else if (xe->info.platform == XE_ALDERLAKE_N) {
+ revids = adln_revids;
+ size = ARRAY_SIZE(adln_revids);
+ } else if (xe->info.subplatform == XE_SUBPLATFORM_ALDERLAKE_S_RPLS) {
+ revids = adls_rpls_revids;
+ size = ARRAY_SIZE(adls_rpls_revids);
+ } else if (xe->info.subplatform == XE_SUBPLATFORM_ALDERLAKE_P_RPLU) {
+ revids = adlp_rpl_revids;
+ size = ARRAY_SIZE(adlp_rpl_revids);
+ } else if (xe->info.platform == XE_ALDERLAKE_P) {
+ revids = adlp_revids;
+ size = ARRAY_SIZE(adlp_revids);
+ } else if (xe->info.platform == XE_ALDERLAKE_S) {
+ revids = adls_revids;
+ size = ARRAY_SIZE(adls_revids);
+ } else if (xe->info.platform == XE_DG1) {
+ revids = dg1_revids;
+ size = ARRAY_SIZE(dg1_revids);
+ } else if (xe->info.platform == XE_TIGERLAKE) {
+ revids = tgl_revids;
+ size = ARRAY_SIZE(tgl_revids);
+ }
+
+ /* Not using the stepping scheme for the platform yet. */
+ if (!revids)
+ return step;
+
+ if (revid < size && revids[revid].graphics != STEP_NONE) {
+ step = revids[revid];
+ } else {
+ drm_warn(&xe->drm, "Unknown revid 0x%02x\n", revid);
+
+ /*
+ * If we hit a gap in the revid array, use the information for
+ * the next revid.
+ *
+ * This may be wrong in all sorts of ways, especially if the
+ * steppings in the array are not monotonically increasing, but
+ * it's better than defaulting to 0.
+ */
+ while (revid < size && revids[revid].graphics == STEP_NONE)
+ revid++;
+
+ if (revid < size) {
+ drm_dbg(&xe->drm, "Using steppings for revid 0x%02x\n",
+ revid);
+ step = revids[revid];
+ } else {
+ drm_dbg(&xe->drm, "Using future steppings\n");
+ step.graphics = STEP_FUTURE;
+ step.display = STEP_FUTURE;
+ }
+ }
+
+ drm_WARN_ON(&xe->drm, step.graphics == STEP_NONE);
+
+ if (basedie_info && basedie_size) {
+ if (baseid < basedie_size && basedie_info[baseid] != STEP_NONE) {
+ step.basedie = basedie_info[baseid];
+ } else {
+ drm_warn(&xe->drm, "Unknown baseid 0x%02x\n", baseid);
+ step.basedie = STEP_FUTURE;
+ }
+ }
+
+ return step;
+}
+
+/**
+ * xe_step_gmdid_get - Determine IP steppings from GMD_ID revid fields
+ * @xe: Xe device
+ * @graphics_gmdid_revid: value of graphics GMD_ID register's revid field
+ * @media_gmdid_revid: value of media GMD_ID register's revid field
+ *
+ * Convert the revid fields of the GMD_ID registers into proper IP steppings.
+ *
+ * GMD_ID revid values are currently expected to have consistent meanings on
+ * all platforms: major steppings (A0, B0, etc.) are 4 apart, with minor
+ * steppings (A1, A2, etc.) taking the values in between.
+ */
+struct xe_step_info xe_step_gmdid_get(struct xe_device *xe,
+ u32 graphics_gmdid_revid,
+ u32 media_gmdid_revid)
+{
+ struct xe_step_info step = {
+ .graphics = STEP_A0 + graphics_gmdid_revid,
+ .media = STEP_A0 + media_gmdid_revid,
+ };
+
+ if (step.graphics >= STEP_FUTURE) {
+ step.graphics = STEP_FUTURE;
+ drm_dbg(&xe->drm, "Graphics GMD_ID revid value %d treated as future stepping\n",
+ graphics_gmdid_revid);
+ }
+
+ if (step.media >= STEP_FUTURE) {
+ step.media = STEP_FUTURE;
+ drm_dbg(&xe->drm, "Media GMD_ID revid value %d treated as future stepping\n",
+ media_gmdid_revid);
+ }
+
+ return step;
+}
+
+#define STEP_NAME_CASE(name) \
+ case STEP_##name: \
+ return #name;
+
+const char *xe_step_name(enum xe_step step)
+{
+ switch (step) {
+ STEP_NAME_LIST(STEP_NAME_CASE);
+
+ default:
+ return "**";
+ }
+}
diff --git a/drivers/gpu/drm/xe/xe_step.h b/drivers/gpu/drm/xe/xe_step.h
new file mode 100644
index 000000000000..686cb59200c2
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_step.h
@@ -0,0 +1,23 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_STEP_H_
+#define _XE_STEP_H_
+
+#include <linux/types.h>
+
+#include "xe_step_types.h"
+
+struct xe_device;
+
+struct xe_step_info xe_step_pre_gmdid_get(struct xe_device *xe);
+struct xe_step_info xe_step_gmdid_get(struct xe_device *xe,
+ u32 graphics_gmdid_revid,
+ u32 media_gmdid_revid);
+static inline u32 xe_step_to_gmdid(enum xe_step step) { return step - STEP_A0; }
+
+const char *xe_step_name(enum xe_step step);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_step_types.h b/drivers/gpu/drm/xe/xe_step_types.h
new file mode 100644
index 000000000000..ccc9b4795e95
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_step_types.h
@@ -0,0 +1,50 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_STEP_TYPES_H_
+#define _XE_STEP_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_step_info {
+ u8 graphics;
+ u8 media;
+ u8 display;
+ u8 basedie;
+};
+
+#define STEP_ENUM_VAL(name) STEP_##name,
+
+#define STEP_NAME_LIST(func) \
+ func(A0) \
+ func(A1) \
+ func(A2) \
+ func(A3) \
+ func(B0) \
+ func(B1) \
+ func(B2) \
+ func(B3) \
+ func(C0) \
+ func(C1) \
+ func(C2) \
+ func(C3) \
+ func(D0) \
+ func(D1) \
+ func(D2) \
+ func(D3) \
+ func(E0)
+
+/*
+ * Symbolic steppings that do not match the hardware. These are valid both as gt
+ * and display steppings as symbolic names.
+ */
+enum xe_step {
+ STEP_NONE = 0,
+ STEP_NAME_LIST(STEP_ENUM_VAL)
+ STEP_FUTURE,
+ STEP_FOREVER,
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sync.c b/drivers/gpu/drm/xe/xe_sync.c
new file mode 100644
index 000000000000..e4c220cf9115
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sync.c
@@ -0,0 +1,344 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_sync.h"
+
+#include <linux/dma-fence-array.h>
+#include <linux/kthread.h>
+#include <linux/sched/mm.h>
+#include <linux/uaccess.h>
+
+#include <drm/drm_print.h>
+#include <drm/drm_syncobj.h>
+#include <drm/xe_drm.h>
+
+#include "xe_device_types.h"
+#include "xe_exec_queue.h"
+#include "xe_macros.h"
+#include "xe_sched_job_types.h"
+
+struct user_fence {
+ struct xe_device *xe;
+ struct kref refcount;
+ struct dma_fence_cb cb;
+ struct work_struct worker;
+ struct mm_struct *mm;
+ u64 __user *addr;
+ u64 value;
+};
+
+static void user_fence_destroy(struct kref *kref)
+{
+ struct user_fence *ufence = container_of(kref, struct user_fence,
+ refcount);
+
+ mmdrop(ufence->mm);
+ kfree(ufence);
+}
+
+static void user_fence_get(struct user_fence *ufence)
+{
+ kref_get(&ufence->refcount);
+}
+
+static void user_fence_put(struct user_fence *ufence)
+{
+ kref_put(&ufence->refcount, user_fence_destroy);
+}
+
+static struct user_fence *user_fence_create(struct xe_device *xe, u64 addr,
+ u64 value)
+{
+ struct user_fence *ufence;
+
+ ufence = kmalloc(sizeof(*ufence), GFP_KERNEL);
+ if (!ufence)
+ return NULL;
+
+ ufence->xe = xe;
+ kref_init(&ufence->refcount);
+ ufence->addr = u64_to_user_ptr(addr);
+ ufence->value = value;
+ ufence->mm = current->mm;
+ mmgrab(ufence->mm);
+
+ return ufence;
+}
+
+static void user_fence_worker(struct work_struct *w)
+{
+ struct user_fence *ufence = container_of(w, struct user_fence, worker);
+
+ if (mmget_not_zero(ufence->mm)) {
+ kthread_use_mm(ufence->mm);
+ if (copy_to_user(ufence->addr, &ufence->value, sizeof(ufence->value)))
+ XE_WARN_ON("Copy to user failed");
+ kthread_unuse_mm(ufence->mm);
+ mmput(ufence->mm);
+ }
+
+ wake_up_all(&ufence->xe->ufence_wq);
+ user_fence_put(ufence);
+}
+
+static void kick_ufence(struct user_fence *ufence, struct dma_fence *fence)
+{
+ INIT_WORK(&ufence->worker, user_fence_worker);
+ queue_work(ufence->xe->ordered_wq, &ufence->worker);
+ dma_fence_put(fence);
+}
+
+static void user_fence_cb(struct dma_fence *fence, struct dma_fence_cb *cb)
+{
+ struct user_fence *ufence = container_of(cb, struct user_fence, cb);
+
+ kick_ufence(ufence, fence);
+}
+
+int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,
+ struct xe_sync_entry *sync,
+ struct drm_xe_sync __user *sync_user,
+ unsigned int flags)
+{
+ struct drm_xe_sync sync_in;
+ int err;
+ bool exec = flags & SYNC_PARSE_FLAG_EXEC;
+ bool in_lr_mode = flags & SYNC_PARSE_FLAG_LR_MODE;
+ bool disallow_user_fence = flags & SYNC_PARSE_FLAG_DISALLOW_USER_FENCE;
+ bool signal;
+
+ if (copy_from_user(&sync_in, sync_user, sizeof(*sync_user)))
+ return -EFAULT;
+
+ if (XE_IOCTL_DBG(xe, sync_in.flags & ~DRM_XE_SYNC_FLAG_SIGNAL) ||
+ XE_IOCTL_DBG(xe, sync_in.reserved[0] || sync_in.reserved[1]))
+ return -EINVAL;
+
+ signal = sync_in.flags & DRM_XE_SYNC_FLAG_SIGNAL;
+ switch (sync_in.type) {
+ case DRM_XE_SYNC_TYPE_SYNCOBJ:
+ if (XE_IOCTL_DBG(xe, in_lr_mode && signal))
+ return -EOPNOTSUPP;
+
+ if (XE_IOCTL_DBG(xe, upper_32_bits(sync_in.addr)))
+ return -EINVAL;
+
+ sync->syncobj = drm_syncobj_find(xef->drm, sync_in.handle);
+ if (XE_IOCTL_DBG(xe, !sync->syncobj))
+ return -ENOENT;
+
+ if (!signal) {
+ sync->fence = drm_syncobj_fence_get(sync->syncobj);
+ if (XE_IOCTL_DBG(xe, !sync->fence))
+ return -EINVAL;
+ }
+ break;
+
+ case DRM_XE_SYNC_TYPE_TIMELINE_SYNCOBJ:
+ if (XE_IOCTL_DBG(xe, in_lr_mode && signal))
+ return -EOPNOTSUPP;
+
+ if (XE_IOCTL_DBG(xe, upper_32_bits(sync_in.addr)))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, sync_in.timeline_value == 0))
+ return -EINVAL;
+
+ sync->syncobj = drm_syncobj_find(xef->drm, sync_in.handle);
+ if (XE_IOCTL_DBG(xe, !sync->syncobj))
+ return -ENOENT;
+
+ if (signal) {
+ sync->chain_fence = dma_fence_chain_alloc();
+ if (!sync->chain_fence)
+ return -ENOMEM;
+ } else {
+ sync->fence = drm_syncobj_fence_get(sync->syncobj);
+ if (XE_IOCTL_DBG(xe, !sync->fence))
+ return -EINVAL;
+
+ err = dma_fence_chain_find_seqno(&sync->fence,
+ sync_in.timeline_value);
+ if (err)
+ return err;
+ }
+ break;
+
+ case DRM_XE_SYNC_TYPE_USER_FENCE:
+ if (XE_IOCTL_DBG(xe, disallow_user_fence))
+ return -EOPNOTSUPP;
+
+ if (XE_IOCTL_DBG(xe, !signal))
+ return -EOPNOTSUPP;
+
+ if (XE_IOCTL_DBG(xe, sync_in.addr & 0x7))
+ return -EINVAL;
+
+ if (exec) {
+ sync->addr = sync_in.addr;
+ } else {
+ sync->ufence = user_fence_create(xe, sync_in.addr,
+ sync_in.timeline_value);
+ if (XE_IOCTL_DBG(xe, !sync->ufence))
+ return -ENOMEM;
+ }
+
+ break;
+
+ default:
+ return -EINVAL;
+ }
+
+ sync->type = sync_in.type;
+ sync->flags = sync_in.flags;
+ sync->timeline_value = sync_in.timeline_value;
+
+ return 0;
+}
+
+int xe_sync_entry_wait(struct xe_sync_entry *sync)
+{
+ if (sync->fence)
+ dma_fence_wait(sync->fence, true);
+
+ return 0;
+}
+
+int xe_sync_entry_add_deps(struct xe_sync_entry *sync, struct xe_sched_job *job)
+{
+ int err;
+
+ if (sync->fence) {
+ err = drm_sched_job_add_dependency(&job->drm,
+ dma_fence_get(sync->fence));
+ if (err) {
+ dma_fence_put(sync->fence);
+ return err;
+ }
+ }
+
+ return 0;
+}
+
+void xe_sync_entry_signal(struct xe_sync_entry *sync, struct xe_sched_job *job,
+ struct dma_fence *fence)
+{
+ if (!(sync->flags & DRM_XE_SYNC_FLAG_SIGNAL))
+ return;
+
+ if (sync->chain_fence) {
+ drm_syncobj_add_point(sync->syncobj, sync->chain_fence,
+ fence, sync->timeline_value);
+ /*
+ * The chain's ownership is transferred to the
+ * timeline.
+ */
+ sync->chain_fence = NULL;
+ } else if (sync->syncobj) {
+ drm_syncobj_replace_fence(sync->syncobj, fence);
+ } else if (sync->ufence) {
+ int err;
+
+ dma_fence_get(fence);
+ user_fence_get(sync->ufence);
+ err = dma_fence_add_callback(fence, &sync->ufence->cb,
+ user_fence_cb);
+ if (err == -ENOENT) {
+ kick_ufence(sync->ufence, fence);
+ } else if (err) {
+ XE_WARN_ON("failed to add user fence");
+ user_fence_put(sync->ufence);
+ dma_fence_put(fence);
+ }
+ } else if (sync->type == DRM_XE_SYNC_TYPE_USER_FENCE) {
+ job->user_fence.used = true;
+ job->user_fence.addr = sync->addr;
+ job->user_fence.value = sync->timeline_value;
+ }
+}
+
+void xe_sync_entry_cleanup(struct xe_sync_entry *sync)
+{
+ if (sync->syncobj)
+ drm_syncobj_put(sync->syncobj);
+ if (sync->fence)
+ dma_fence_put(sync->fence);
+ if (sync->chain_fence)
+ dma_fence_put(&sync->chain_fence->base);
+ if (sync->ufence)
+ user_fence_put(sync->ufence);
+}
+
+/**
+ * xe_sync_in_fence_get() - Get a fence from syncs, exec queue, and VM
+ * @sync: input syncs
+ * @num_sync: number of syncs
+ * @q: exec queue
+ * @vm: VM
+ *
+ * Get a fence from syncs, exec queue, and VM. If syncs contain in-fences create
+ * and return a composite fence of all in-fences + last fence. If no in-fences
+ * return last fence on input exec queue. Caller must drop reference to
+ * returned fence.
+ *
+ * Return: fence on success, ERR_PTR(-ENOMEM) on failure
+ */
+struct dma_fence *
+xe_sync_in_fence_get(struct xe_sync_entry *sync, int num_sync,
+ struct xe_exec_queue *q, struct xe_vm *vm)
+{
+ struct dma_fence **fences = NULL;
+ struct dma_fence_array *cf = NULL;
+ struct dma_fence *fence;
+ int i, num_in_fence = 0, current_fence = 0;
+
+ lockdep_assert_held(&vm->lock);
+
+ /* Count in-fences */
+ for (i = 0; i < num_sync; ++i) {
+ if (sync[i].fence) {
+ ++num_in_fence;
+ fence = sync[i].fence;
+ }
+ }
+
+ /* Easy case... */
+ if (!num_in_fence) {
+ fence = xe_exec_queue_last_fence_get(q, vm);
+ dma_fence_get(fence);
+ return fence;
+ }
+
+ /* Create composite fence */
+ fences = kmalloc_array(num_in_fence + 1, sizeof(*fences), GFP_KERNEL);
+ if (!fences)
+ return ERR_PTR(-ENOMEM);
+ for (i = 0; i < num_sync; ++i) {
+ if (sync[i].fence) {
+ dma_fence_get(sync[i].fence);
+ fences[current_fence++] = sync[i].fence;
+ }
+ }
+ fences[current_fence++] = xe_exec_queue_last_fence_get(q, vm);
+ dma_fence_get(fences[current_fence - 1]);
+ cf = dma_fence_array_create(num_in_fence, fences,
+ vm->composite_fence_ctx,
+ vm->composite_fence_seqno++,
+ false);
+ if (!cf) {
+ --vm->composite_fence_seqno;
+ goto err_out;
+ }
+
+ return &cf->base;
+
+err_out:
+ while (current_fence)
+ dma_fence_put(fences[--current_fence]);
+ kfree(fences);
+ kfree(cf);
+
+ return ERR_PTR(-ENOMEM);
+}
diff --git a/drivers/gpu/drm/xe/xe_sync.h b/drivers/gpu/drm/xe/xe_sync.h
new file mode 100644
index 000000000000..d284afbe917c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sync.h
@@ -0,0 +1,36 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_SYNC_H_
+#define _XE_SYNC_H_
+
+#include "xe_sync_types.h"
+
+struct xe_device;
+struct xe_exec_queue;
+struct xe_file;
+struct xe_sched_job;
+struct xe_vm;
+
+#define SYNC_PARSE_FLAG_EXEC BIT(0)
+#define SYNC_PARSE_FLAG_LR_MODE BIT(1)
+#define SYNC_PARSE_FLAG_DISALLOW_USER_FENCE BIT(2)
+
+int xe_sync_entry_parse(struct xe_device *xe, struct xe_file *xef,
+ struct xe_sync_entry *sync,
+ struct drm_xe_sync __user *sync_user,
+ unsigned int flags);
+int xe_sync_entry_wait(struct xe_sync_entry *sync);
+int xe_sync_entry_add_deps(struct xe_sync_entry *sync,
+ struct xe_sched_job *job);
+void xe_sync_entry_signal(struct xe_sync_entry *sync,
+ struct xe_sched_job *job,
+ struct dma_fence *fence);
+void xe_sync_entry_cleanup(struct xe_sync_entry *sync);
+struct dma_fence *
+xe_sync_in_fence_get(struct xe_sync_entry *sync, int num_sync,
+ struct xe_exec_queue *q, struct xe_vm *vm);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_sync_types.h b/drivers/gpu/drm/xe/xe_sync_types.h
new file mode 100644
index 000000000000..852db5e7884f
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_sync_types.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_SYNC_TYPES_H_
+#define _XE_SYNC_TYPES_H_
+
+#include <linux/types.h>
+
+struct drm_syncobj;
+struct dma_fence;
+struct dma_fence_chain;
+struct drm_xe_sync;
+struct user_fence;
+
+struct xe_sync_entry {
+ struct drm_syncobj *syncobj;
+ struct dma_fence *fence;
+ struct dma_fence_chain *chain_fence;
+ struct user_fence *ufence;
+ u64 addr;
+ u64 timeline_value;
+ u32 type;
+ u32 flags;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_tile.c b/drivers/gpu/drm/xe/xe_tile.c
new file mode 100644
index 000000000000..044c20881de7
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tile.c
@@ -0,0 +1,185 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <drm/drm_managed.h>
+
+#include "xe_device.h"
+#include "xe_ggtt.h"
+#include "xe_gt.h"
+#include "xe_migrate.h"
+#include "xe_sa.h"
+#include "xe_tile.h"
+#include "xe_tile_sysfs.h"
+#include "xe_ttm_vram_mgr.h"
+#include "xe_wa.h"
+
+/**
+ * DOC: Multi-tile Design
+ *
+ * Different vendors use the term "tile" a bit differently, but in the Intel
+ * world, a 'tile' is pretty close to what most people would think of as being
+ * a complete GPU. When multiple GPUs are placed behind a single PCI device,
+ * that's what is referred to as a "multi-tile device." In such cases, pretty
+ * much all hardware is replicated per-tile, although certain responsibilities
+ * like PCI communication, reporting of interrupts to the OS, etc. are handled
+ * solely by the "root tile." A multi-tile platform takes care of tying the
+ * tiles together in a way such that interrupt notifications from remote tiles
+ * are forwarded to the root tile, the per-tile vram is combined into a single
+ * address space, etc.
+ *
+ * In contrast, a "GT" (which officially stands for "Graphics Technology") is
+ * the subset of a GPU/tile that is responsible for implementing graphics
+ * and/or media operations. The GT is where a lot of the driver implementation
+ * happens since it's where the hardware engines, the execution units, and the
+ * GuC all reside.
+ *
+ * Historically most Intel devices were single-tile devices that contained a
+ * single GT. PVC is an example of an Intel platform built on a multi-tile
+ * design (i.e., multiple GPUs behind a single PCI device); each PVC tile only
+ * has a single GT. In contrast, platforms like MTL that have separate chips
+ * for render and media IP are still only a single logical GPU, but the
+ * graphics and media IP blocks are each exposed as a separate GT within that
+ * single GPU. This is important from a software perspective because multi-GT
+ * platforms like MTL only replicate a subset of the GPU hardware and behave
+ * differently than multi-tile platforms like PVC where nearly everything is
+ * replicated.
+ *
+ * Per-tile functionality (shared by all GTs within the tile):
+ * - Complete 4MB MMIO space (containing SGunit/SoC registers, GT
+ * registers, display registers, etc.)
+ * - Global GTT
+ * - VRAM (if discrete)
+ * - Interrupt flows
+ * - Migration context
+ * - kernel batchbuffer pool
+ * - Primary GT
+ * - Media GT (if media version >= 13)
+ *
+ * Per-GT functionality:
+ * - GuC
+ * - Hardware engines
+ * - Programmable hardware units (subslices, EUs)
+ * - GSI subset of registers (multiple copies of these registers reside
+ * within the complete MMIO space provided by the tile, but at different
+ * offsets --- 0 for render, 0x380000 for media)
+ * - Multicast register steering
+ * - TLBs to cache page table translations
+ * - Reset capability
+ * - Low-level power management (e.g., C6)
+ * - Clock frequency
+ * - MOCS and PAT programming
+ */
+
+/**
+ * xe_tile_alloc - Perform per-tile memory allocation
+ * @tile: Tile to perform allocations for
+ *
+ * Allocates various per-tile data structures using DRM-managed allocations.
+ * Does not touch the hardware.
+ *
+ * Returns -ENOMEM if allocations fail, otherwise 0.
+ */
+static int xe_tile_alloc(struct xe_tile *tile)
+{
+ struct drm_device *drm = &tile_to_xe(tile)->drm;
+
+ tile->mem.ggtt = drmm_kzalloc(drm, sizeof(*tile->mem.ggtt),
+ GFP_KERNEL);
+ if (!tile->mem.ggtt)
+ return -ENOMEM;
+ tile->mem.ggtt->tile = tile;
+
+ tile->mem.vram_mgr = drmm_kzalloc(drm, sizeof(*tile->mem.vram_mgr), GFP_KERNEL);
+ if (!tile->mem.vram_mgr)
+ return -ENOMEM;
+
+ return 0;
+}
+
+/**
+ * xe_tile_init_early - Initialize the tile and primary GT
+ * @tile: Tile to initialize
+ * @xe: Parent Xe device
+ * @id: Tile ID
+ *
+ * Initializes per-tile resources that don't require any interactions with the
+ * hardware or any knowledge about the Graphics/Media IP version.
+ *
+ * Returns: 0 on success, negative error code on error.
+ */
+int xe_tile_init_early(struct xe_tile *tile, struct xe_device *xe, u8 id)
+{
+ int err;
+
+ tile->xe = xe;
+ tile->id = id;
+
+ err = xe_tile_alloc(tile);
+ if (err)
+ return err;
+
+ tile->primary_gt = xe_gt_alloc(tile);
+ if (IS_ERR(tile->primary_gt))
+ return PTR_ERR(tile->primary_gt);
+
+ return 0;
+}
+
+static int tile_ttm_mgr_init(struct xe_tile *tile)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ int err;
+
+ if (tile->mem.vram.usable_size) {
+ err = xe_ttm_vram_mgr_init(tile, tile->mem.vram_mgr);
+ if (err)
+ return err;
+ xe->info.mem_region_mask |= BIT(tile->id) << 1;
+ }
+
+ return 0;
+}
+
+/**
+ * xe_tile_init_noalloc - Init tile up to the point where allocations can happen.
+ * @tile: The tile to initialize.
+ *
+ * This function prepares the tile to allow memory allocations to VRAM, but is
+ * not allowed to allocate memory itself. This state is useful for display
+ * readout, because the inherited display framebuffer will otherwise be
+ * overwritten as it is usually put at the start of VRAM.
+ *
+ * Note that since this is tile initialization, it should not perform any
+ * GT-specific operations, and thus does not need to hold GT forcewake.
+ *
+ * Returns: 0 on success, negative error code on error.
+ */
+int xe_tile_init_noalloc(struct xe_tile *tile)
+{
+ int err;
+
+ xe_device_mem_access_get(tile_to_xe(tile));
+
+ err = tile_ttm_mgr_init(tile);
+ if (err)
+ goto err_mem_access;
+
+ tile->mem.kernel_bb_pool = xe_sa_bo_manager_init(tile, SZ_1M, 16);
+ if (IS_ERR(tile->mem.kernel_bb_pool))
+ err = PTR_ERR(tile->mem.kernel_bb_pool);
+
+ xe_wa_apply_tile_workarounds(tile);
+
+ xe_tile_sysfs_init(tile);
+
+err_mem_access:
+ xe_device_mem_access_put(tile_to_xe(tile));
+ return err;
+}
+
+void xe_tile_migrate_wait(struct xe_tile *tile)
+{
+ xe_migrate_wait(tile->migrate);
+}
diff --git a/drivers/gpu/drm/xe/xe_tile.h b/drivers/gpu/drm/xe/xe_tile.h
new file mode 100644
index 000000000000..1c9e42ade6b0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tile.h
@@ -0,0 +1,18 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_TILE_H_
+#define _XE_TILE_H_
+
+#include "xe_device_types.h"
+
+struct xe_tile;
+
+int xe_tile_init_early(struct xe_tile *tile, struct xe_device *xe, u8 id);
+int xe_tile_init_noalloc(struct xe_tile *tile);
+
+void xe_tile_migrate_wait(struct xe_tile *tile);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_tile_sysfs.c b/drivers/gpu/drm/xe/xe_tile_sysfs.c
new file mode 100644
index 000000000000..0f8d3e7fce46
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tile_sysfs.c
@@ -0,0 +1,57 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#include <linux/kobject.h>
+#include <linux/sysfs.h>
+#include <drm/drm_managed.h>
+
+#include "xe_tile.h"
+#include "xe_tile_sysfs.h"
+
+static void xe_tile_sysfs_kobj_release(struct kobject *kobj)
+{
+ kfree(kobj);
+}
+
+static const struct kobj_type xe_tile_sysfs_kobj_type = {
+ .release = xe_tile_sysfs_kobj_release,
+ .sysfs_ops = &kobj_sysfs_ops,
+};
+
+static void tile_sysfs_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_tile *tile = arg;
+
+ kobject_put(tile->sysfs);
+}
+
+void xe_tile_sysfs_init(struct xe_tile *tile)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct device *dev = xe->drm.dev;
+ struct kobj_tile *kt;
+ int err;
+
+ kt = kzalloc(sizeof(*kt), GFP_KERNEL);
+ if (!kt)
+ return;
+
+ kobject_init(&kt->base, &xe_tile_sysfs_kobj_type);
+ kt->tile = tile;
+
+ err = kobject_add(&kt->base, &dev->kobj, "tile%d", tile->id);
+ if (err) {
+ kobject_put(&kt->base);
+ drm_warn(&xe->drm, "failed to register TILE sysfs directory, err: %d\n", err);
+ return;
+ }
+
+ tile->sysfs = &kt->base;
+
+ err = drmm_add_action_or_reset(&xe->drm, tile_sysfs_fini, tile);
+ if (err)
+ drm_warn(&xe->drm, "%s: drmm_add_action_or_reset failed, err: %d\n",
+ __func__, err);
+}
diff --git a/drivers/gpu/drm/xe/xe_tile_sysfs.h b/drivers/gpu/drm/xe/xe_tile_sysfs.h
new file mode 100644
index 000000000000..e4f065039eba
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tile_sysfs.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_TILE_SYSFS_H_
+#define _XE_TILE_SYSFS_H_
+
+#include "xe_tile_sysfs_types.h"
+
+void xe_tile_sysfs_init(struct xe_tile *tile);
+
+static inline struct xe_tile *
+kobj_to_tile(struct kobject *kobj)
+{
+ return container_of(kobj, struct kobj_tile, base)->tile;
+}
+
+#endif /* _XE_TILE_SYSFS_H_ */
diff --git a/drivers/gpu/drm/xe/xe_tile_sysfs_types.h b/drivers/gpu/drm/xe/xe_tile_sysfs_types.h
new file mode 100644
index 000000000000..75906ba11a9e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tile_sysfs_types.h
@@ -0,0 +1,27 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_TILE_SYSFS_TYPES_H_
+#define _XE_TILE_SYSFS_TYPES_H_
+
+#include <linux/kobject.h>
+
+struct xe_tile;
+
+/**
+ * struct kobj_tile - A tile's kobject struct that connects the kobject
+ * and the TILE
+ *
+ * When dealing with multiple TILEs, this struct helps to understand which
+ * TILE needs to be addressed on a given sysfs call.
+ */
+struct kobj_tile {
+ /** @base: The actual kobject */
+ struct kobject base;
+ /** @tile: A pointer to the tile itself */
+ struct xe_tile *tile;
+};
+
+#endif /* _XE_TILE_SYSFS_TYPES_H_ */
diff --git a/drivers/gpu/drm/xe/xe_trace.c b/drivers/gpu/drm/xe/xe_trace.c
new file mode 100644
index 000000000000..2527c556bff1
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_trace.c
@@ -0,0 +1,9 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef __CHECKER__
+#define CREATE_TRACE_POINTS
+#include "xe_trace.h"
+#endif
diff --git a/drivers/gpu/drm/xe/xe_trace.h b/drivers/gpu/drm/xe/xe_trace.h
new file mode 100644
index 000000000000..95163c303f3e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_trace.h
@@ -0,0 +1,608 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM xe
+
+#if !defined(_XE_TRACE_H_) || defined(TRACE_HEADER_MULTI_READ)
+#define _XE_TRACE_H_
+
+#include <linux/tracepoint.h>
+#include <linux/types.h>
+
+#include "xe_bo_types.h"
+#include "xe_exec_queue_types.h"
+#include "xe_gpu_scheduler_types.h"
+#include "xe_gt_tlb_invalidation_types.h"
+#include "xe_gt_types.h"
+#include "xe_guc_exec_queue_types.h"
+#include "xe_sched_job.h"
+#include "xe_vm.h"
+
+DECLARE_EVENT_CLASS(xe_gt_tlb_invalidation_fence,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence),
+
+ TP_STRUCT__entry(
+ __field(u64, fence)
+ __field(int, seqno)
+ ),
+
+ TP_fast_assign(
+ __entry->fence = (u64)fence;
+ __entry->seqno = fence->seqno;
+ ),
+
+ TP_printk("fence=0x%016llx, seqno=%d",
+ __entry->fence, __entry->seqno)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_create,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence,
+ xe_gt_tlb_invalidation_fence_work_func,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_cb,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_send,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_recv,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_signal,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_gt_tlb_invalidation_fence, xe_gt_tlb_invalidation_fence_timeout,
+ TP_PROTO(struct xe_gt_tlb_invalidation_fence *fence),
+ TP_ARGS(fence)
+);
+
+DECLARE_EVENT_CLASS(xe_bo,
+ TP_PROTO(struct xe_bo *bo),
+ TP_ARGS(bo),
+
+ TP_STRUCT__entry(
+ __field(size_t, size)
+ __field(u32, flags)
+ __field(u64, vm)
+ ),
+
+ TP_fast_assign(
+ __entry->size = bo->size;
+ __entry->flags = bo->flags;
+ __entry->vm = (unsigned long)bo->vm;
+ ),
+
+ TP_printk("size=%zu, flags=0x%02x, vm=0x%016llx",
+ __entry->size, __entry->flags, __entry->vm)
+);
+
+DEFINE_EVENT(xe_bo, xe_bo_cpu_fault,
+ TP_PROTO(struct xe_bo *bo),
+ TP_ARGS(bo)
+);
+
+DEFINE_EVENT(xe_bo, xe_bo_move,
+ TP_PROTO(struct xe_bo *bo),
+ TP_ARGS(bo)
+);
+
+DECLARE_EVENT_CLASS(xe_exec_queue,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q),
+
+ TP_STRUCT__entry(
+ __field(enum xe_engine_class, class)
+ __field(u32, logical_mask)
+ __field(u8, gt_id)
+ __field(u16, width)
+ __field(u16, guc_id)
+ __field(u32, guc_state)
+ __field(u32, flags)
+ ),
+
+ TP_fast_assign(
+ __entry->class = q->class;
+ __entry->logical_mask = q->logical_mask;
+ __entry->gt_id = q->gt->info.id;
+ __entry->width = q->width;
+ __entry->guc_id = q->guc->id;
+ __entry->guc_state = atomic_read(&q->guc->state);
+ __entry->flags = q->flags;
+ ),
+
+ TP_printk("%d:0x%x, gt=%d, width=%d, guc_id=%d, guc_state=0x%x, flags=0x%x",
+ __entry->class, __entry->logical_mask,
+ __entry->gt_id, __entry->width, __entry->guc_id,
+ __entry->guc_state, __entry->flags)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_create,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_supress_resume,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_submit,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_scheduling_enable,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_scheduling_disable,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_scheduling_done,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_register,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_deregister,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_deregister_done,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_close,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_kill,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_cleanup_entity,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_destroy,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_reset,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_memory_cat_error,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_stop,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_resubmit,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DEFINE_EVENT(xe_exec_queue, xe_exec_queue_lr_cleanup,
+ TP_PROTO(struct xe_exec_queue *q),
+ TP_ARGS(q)
+);
+
+DECLARE_EVENT_CLASS(xe_sched_job,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job),
+
+ TP_STRUCT__entry(
+ __field(u32, seqno)
+ __field(u16, guc_id)
+ __field(u32, guc_state)
+ __field(u32, flags)
+ __field(int, error)
+ __field(u64, fence)
+ __field(u64, batch_addr)
+ ),
+
+ TP_fast_assign(
+ __entry->seqno = xe_sched_job_seqno(job);
+ __entry->guc_id = job->q->guc->id;
+ __entry->guc_state =
+ atomic_read(&job->q->guc->state);
+ __entry->flags = job->q->flags;
+ __entry->error = job->fence->error;
+ __entry->fence = (unsigned long)job->fence;
+ __entry->batch_addr = (u64)job->batch_addr[0];
+ ),
+
+ TP_printk("fence=0x%016llx, seqno=%u, guc_id=%d, batch_addr=0x%012llx, guc_state=0x%x, flags=0x%x, error=%d",
+ __entry->fence, __entry->seqno, __entry->guc_id,
+ __entry->batch_addr, __entry->guc_state,
+ __entry->flags, __entry->error)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_create,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_exec,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_run,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_free,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_timedout,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_set_error,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DEFINE_EVENT(xe_sched_job, xe_sched_job_ban,
+ TP_PROTO(struct xe_sched_job *job),
+ TP_ARGS(job)
+);
+
+DECLARE_EVENT_CLASS(xe_sched_msg,
+ TP_PROTO(struct xe_sched_msg *msg),
+ TP_ARGS(msg),
+
+ TP_STRUCT__entry(
+ __field(u32, opcode)
+ __field(u16, guc_id)
+ ),
+
+ TP_fast_assign(
+ __entry->opcode = msg->opcode;
+ __entry->guc_id =
+ ((struct xe_exec_queue *)msg->private_data)->guc->id;
+ ),
+
+ TP_printk("guc_id=%d, opcode=%u", __entry->guc_id,
+ __entry->opcode)
+);
+
+DEFINE_EVENT(xe_sched_msg, xe_sched_msg_add,
+ TP_PROTO(struct xe_sched_msg *msg),
+ TP_ARGS(msg)
+);
+
+DEFINE_EVENT(xe_sched_msg, xe_sched_msg_recv,
+ TP_PROTO(struct xe_sched_msg *msg),
+ TP_ARGS(msg)
+);
+
+DECLARE_EVENT_CLASS(xe_hw_fence,
+ TP_PROTO(struct xe_hw_fence *fence),
+ TP_ARGS(fence),
+
+ TP_STRUCT__entry(
+ __field(u64, ctx)
+ __field(u32, seqno)
+ __field(u64, fence)
+ ),
+
+ TP_fast_assign(
+ __entry->ctx = fence->dma.context;
+ __entry->seqno = fence->dma.seqno;
+ __entry->fence = (unsigned long)fence;
+ ),
+
+ TP_printk("ctx=0x%016llx, fence=0x%016llx, seqno=%u",
+ __entry->ctx, __entry->fence, __entry->seqno)
+);
+
+DEFINE_EVENT(xe_hw_fence, xe_hw_fence_create,
+ TP_PROTO(struct xe_hw_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_hw_fence, xe_hw_fence_signal,
+ TP_PROTO(struct xe_hw_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_hw_fence, xe_hw_fence_try_signal,
+ TP_PROTO(struct xe_hw_fence *fence),
+ TP_ARGS(fence)
+);
+
+DEFINE_EVENT(xe_hw_fence, xe_hw_fence_free,
+ TP_PROTO(struct xe_hw_fence *fence),
+ TP_ARGS(fence)
+);
+
+DECLARE_EVENT_CLASS(xe_vma,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma),
+
+ TP_STRUCT__entry(
+ __field(u64, vma)
+ __field(u32, asid)
+ __field(u64, start)
+ __field(u64, end)
+ __field(u64, ptr)
+ ),
+
+ TP_fast_assign(
+ __entry->vma = (unsigned long)vma;
+ __entry->asid = xe_vma_vm(vma)->usm.asid;
+ __entry->start = xe_vma_start(vma);
+ __entry->end = xe_vma_end(vma) - 1;
+ __entry->ptr = xe_vma_userptr(vma);
+ ),
+
+ TP_printk("vma=0x%016llx, asid=0x%05x, start=0x%012llx, end=0x%012llx, ptr=0x%012llx,",
+ __entry->vma, __entry->asid, __entry->start,
+ __entry->end, __entry->ptr)
+)
+
+DEFINE_EVENT(xe_vma, xe_vma_flush,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_pagefault,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_acc,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_fail,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_bind,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_pf_bind,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_unbind,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_userptr_rebind_worker,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_userptr_rebind_exec,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_rebind_worker,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_rebind_exec,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_userptr_invalidate,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_usm_invalidate,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_evict,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DEFINE_EVENT(xe_vma, xe_vma_userptr_invalidate_complete,
+ TP_PROTO(struct xe_vma *vma),
+ TP_ARGS(vma)
+);
+
+DECLARE_EVENT_CLASS(xe_vm,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm),
+
+ TP_STRUCT__entry(
+ __field(u64, vm)
+ __field(u32, asid)
+ ),
+
+ TP_fast_assign(
+ __entry->vm = (unsigned long)vm;
+ __entry->asid = vm->usm.asid;
+ ),
+
+ TP_printk("vm=0x%016llx, asid=0x%05x", __entry->vm,
+ __entry->asid)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_kill,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_create,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_free,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_cpu_bind,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_restart,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_rebind_worker_enter,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_rebind_worker_retry,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+DEFINE_EVENT(xe_vm, xe_vm_rebind_worker_exit,
+ TP_PROTO(struct xe_vm *vm),
+ TP_ARGS(vm)
+);
+
+/* GuC */
+DECLARE_EVENT_CLASS(xe_guc_ct_flow_control,
+ TP_PROTO(u32 _head, u32 _tail, u32 size, u32 space, u32 len),
+ TP_ARGS(_head, _tail, size, space, len),
+
+ TP_STRUCT__entry(
+ __field(u32, _head)
+ __field(u32, _tail)
+ __field(u32, size)
+ __field(u32, space)
+ __field(u32, len)
+ ),
+
+ TP_fast_assign(
+ __entry->_head = _head;
+ __entry->_tail = _tail;
+ __entry->size = size;
+ __entry->space = space;
+ __entry->len = len;
+ ),
+
+ TP_printk("h2g flow control: head=%u, tail=%u, size=%u, space=%u, len=%u",
+ __entry->_head, __entry->_tail, __entry->size,
+ __entry->space, __entry->len)
+);
+
+DEFINE_EVENT(xe_guc_ct_flow_control, xe_guc_ct_h2g_flow_control,
+ TP_PROTO(u32 _head, u32 _tail, u32 size, u32 space, u32 len),
+ TP_ARGS(_head, _tail, size, space, len)
+);
+
+DEFINE_EVENT_PRINT(xe_guc_ct_flow_control, xe_guc_ct_g2h_flow_control,
+ TP_PROTO(u32 _head, u32 _tail, u32 size, u32 space, u32 len),
+ TP_ARGS(_head, _tail, size, space, len),
+
+ TP_printk("g2h flow control: head=%u, tail=%u, size=%u, space=%u, len=%u",
+ __entry->_head, __entry->_tail, __entry->size,
+ __entry->space, __entry->len)
+);
+
+DECLARE_EVENT_CLASS(xe_guc_ctb,
+ TP_PROTO(u8 gt_id, u32 action, u32 len, u32 _head, u32 tail),
+ TP_ARGS(gt_id, action, len, _head, tail),
+
+ TP_STRUCT__entry(
+ __field(u8, gt_id)
+ __field(u32, action)
+ __field(u32, len)
+ __field(u32, tail)
+ __field(u32, _head)
+ ),
+
+ TP_fast_assign(
+ __entry->gt_id = gt_id;
+ __entry->action = action;
+ __entry->len = len;
+ __entry->tail = tail;
+ __entry->_head = _head;
+ ),
+
+ TP_printk("gt%d: H2G CTB: action=0x%x, len=%d, tail=%d, head=%d\n",
+ __entry->gt_id, __entry->action, __entry->len,
+ __entry->tail, __entry->_head)
+);
+
+DEFINE_EVENT(xe_guc_ctb, xe_guc_ctb_h2g,
+ TP_PROTO(u8 gt_id, u32 action, u32 len, u32 _head, u32 tail),
+ TP_ARGS(gt_id, action, len, _head, tail)
+);
+
+DEFINE_EVENT_PRINT(xe_guc_ctb, xe_guc_ctb_g2h,
+ TP_PROTO(u8 gt_id, u32 action, u32 len, u32 _head, u32 tail),
+ TP_ARGS(gt_id, action, len, _head, tail),
+
+ TP_printk("gt%d: G2H CTB: action=0x%x, len=%d, tail=%d, head=%d\n",
+ __entry->gt_id, __entry->action, __entry->len,
+ __entry->tail, __entry->_head)
+
+);
+
+#endif
+
+/* This part must be outside protection */
+#undef TRACE_INCLUDE_PATH
+#undef TRACE_INCLUDE_FILE
+#define TRACE_INCLUDE_PATH ../../drivers/gpu/drm/xe
+#define TRACE_INCLUDE_FILE xe_trace
+#include <trace/define_trace.h>
diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
new file mode 100644
index 000000000000..d2b00d0bf1e2
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
@@ -0,0 +1,334 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021-2023 Intel Corporation
+ * Copyright (C) 2021-2002 Red Hat
+ */
+
+#include <drm/drm_managed.h>
+#include <drm/drm_mm.h>
+
+#include <drm/ttm/ttm_device.h>
+#include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_range_manager.h>
+
+#include "generated/xe_wa_oob.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_mmio.h"
+#include "xe_res_cursor.h"
+#include "xe_ttm_stolen_mgr.h"
+#include "xe_ttm_vram_mgr.h"
+#include "xe_wa.h"
+
+struct xe_ttm_stolen_mgr {
+ struct xe_ttm_vram_mgr base;
+
+ /* PCI base offset */
+ resource_size_t io_base;
+ /* GPU base offset */
+ resource_size_t stolen_base;
+
+ void *__iomem mapping;
+};
+
+static inline struct xe_ttm_stolen_mgr *
+to_stolen_mgr(struct ttm_resource_manager *man)
+{
+ return container_of(man, struct xe_ttm_stolen_mgr, base.manager);
+}
+
+/**
+ * xe_ttm_stolen_cpu_access_needs_ggtt() - If we can't directly CPU access
+ * stolen, can we then fallback to mapping through the GGTT.
+ * @xe: xe device
+ *
+ * Some older integrated platforms don't support reliable CPU access for stolen,
+ * however on such hardware we can always use the mappable part of the GGTT for
+ * CPU access. Check if that's the case for this device.
+ */
+bool xe_ttm_stolen_cpu_access_needs_ggtt(struct xe_device *xe)
+{
+ return GRAPHICS_VERx100(xe) < 1270 && !IS_DGFX(xe);
+}
+
+static s64 detect_bar2_dgfx(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
+{
+ struct xe_tile *tile = xe_device_get_root_tile(xe);
+ struct xe_gt *mmio = xe_root_mmio_gt(xe);
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ u64 stolen_size;
+ u64 tile_offset;
+ u64 tile_size;
+
+ tile_offset = tile->mem.vram.io_start - xe->mem.vram.io_start;
+ tile_size = tile->mem.vram.actual_physical_size;
+
+ /* Use DSM base address instead for stolen memory */
+ mgr->stolen_base = (xe_mmio_read64_2x32(mmio, DSMBASE) & BDSM_MASK) - tile_offset;
+ if (drm_WARN_ON(&xe->drm, tile_size < mgr->stolen_base))
+ return 0;
+
+ stolen_size = tile_size - mgr->stolen_base;
+
+ /* Verify usage fits in the actual resource available */
+ if (mgr->stolen_base + stolen_size <= pci_resource_len(pdev, LMEM_BAR))
+ mgr->io_base = tile->mem.vram.io_start + mgr->stolen_base;
+
+ /*
+ * There may be few KB of platform dependent reserved memory at the end
+ * of vram which is not part of the DSM. Such reserved memory portion is
+ * always less then DSM granularity so align down the stolen_size to DSM
+ * granularity to accommodate such reserve vram portion.
+ */
+ return ALIGN_DOWN(stolen_size, SZ_1M);
+}
+
+static u32 get_wopcm_size(struct xe_device *xe)
+{
+ u32 wopcm_size;
+ u64 val;
+
+ val = xe_mmio_read64_2x32(xe_root_mmio_gt(xe), STOLEN_RESERVED);
+ val = REG_FIELD_GET64(WOPCM_SIZE_MASK, val);
+
+ switch (val) {
+ case 0x5 ... 0x6:
+ val--;
+ fallthrough;
+ case 0x0 ... 0x3:
+ wopcm_size = (1U << val) * SZ_1M;
+ break;
+ default:
+ WARN(1, "Missing case wopcm_size=%llx\n", val);
+ wopcm_size = 0;
+ }
+
+ return wopcm_size;
+}
+
+static u32 detect_bar2_integrated(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ struct xe_gt *media_gt = xe_device_get_root_tile(xe)->media_gt;
+ u32 stolen_size, wopcm_size;
+ u32 ggc, gms;
+
+ ggc = xe_mmio_read32(xe_root_mmio_gt(xe), GGC);
+
+ /*
+ * Check GGMS: it should be fixed 0x3 (8MB), which corresponds to the
+ * GTT size
+ */
+ if (drm_WARN_ON(&xe->drm, (ggc & GGMS_MASK) != GGMS_MASK))
+ return 0;
+
+ /*
+ * Graphics >= 1270 uses the offset to the GSMBASE as address in the
+ * PTEs, together with the DM flag being set. Previously there was no
+ * such flag so the address was the io_base.
+ *
+ * DSMBASE = GSMBASE + 8MB
+ */
+ mgr->stolen_base = SZ_8M;
+ mgr->io_base = pci_resource_start(pdev, 2) + mgr->stolen_base;
+
+ /* return valid GMS value, -EIO if invalid */
+ gms = REG_FIELD_GET(GMS_MASK, ggc);
+ switch (gms) {
+ case 0x0 ... 0x04:
+ stolen_size = gms * 32 * SZ_1M;
+ break;
+ case 0xf0 ... 0xfe:
+ stolen_size = (gms - 0xf0 + 1) * 4 * SZ_1M;
+ break;
+ default:
+ return 0;
+ }
+
+ /* Carve out the top of DSM as it contains the reserved WOPCM region */
+ wopcm_size = get_wopcm_size(xe);
+ if (drm_WARN_ON(&xe->drm, !wopcm_size))
+ return 0;
+
+ stolen_size -= wopcm_size;
+
+ if (media_gt && XE_WA(media_gt, 14019821291)) {
+ u64 gscpsmi_base = xe_mmio_read64_2x32(media_gt, GSCPSMI_BASE)
+ & ~GENMASK_ULL(5, 0);
+
+ /*
+ * This workaround is primarily implemented by the BIOS. We
+ * just need to figure out whether the BIOS has applied the
+ * workaround (meaning the programmed address falls within
+ * the DSM) and, if so, reserve that part of the DSM to
+ * prevent accidental reuse. The DSM location should be just
+ * below the WOPCM.
+ */
+ if (gscpsmi_base >= mgr->io_base &&
+ gscpsmi_base < mgr->io_base + stolen_size) {
+ xe_gt_dbg(media_gt,
+ "Reserving %llu bytes of DSM for Wa_14019821291\n",
+ mgr->io_base + stolen_size - gscpsmi_base);
+ stolen_size = gscpsmi_base - mgr->io_base;
+ }
+ }
+
+ if (drm_WARN_ON(&xe->drm, stolen_size + SZ_8M > pci_resource_len(pdev, 2)))
+ return 0;
+
+ return stolen_size;
+}
+
+extern struct resource intel_graphics_stolen_res;
+
+static u64 detect_stolen(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
+{
+#ifdef CONFIG_X86
+ /* Map into GGTT */
+ mgr->io_base = pci_resource_start(to_pci_dev(xe->drm.dev), 2);
+
+ /* Stolen memory is x86 only */
+ mgr->stolen_base = intel_graphics_stolen_res.start;
+ return resource_size(&intel_graphics_stolen_res);
+#else
+ return 0;
+#endif
+}
+
+void xe_ttm_stolen_mgr_init(struct xe_device *xe)
+{
+ struct xe_ttm_stolen_mgr *mgr = drmm_kzalloc(&xe->drm, sizeof(*mgr), GFP_KERNEL);
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+ u64 stolen_size, io_size, pgsize;
+ int err;
+
+ if (IS_DGFX(xe))
+ stolen_size = detect_bar2_dgfx(xe, mgr);
+ else if (GRAPHICS_VERx100(xe) >= 1270)
+ stolen_size = detect_bar2_integrated(xe, mgr);
+ else
+ stolen_size = detect_stolen(xe, mgr);
+
+ if (!stolen_size) {
+ drm_dbg_kms(&xe->drm, "No stolen memory support\n");
+ return;
+ }
+
+ pgsize = xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K ? SZ_64K : SZ_4K;
+ if (pgsize < PAGE_SIZE)
+ pgsize = PAGE_SIZE;
+
+ /*
+ * We don't try to attempt partial visible support for stolen vram,
+ * since stolen is always at the end of vram, and the BAR size is pretty
+ * much always 256M, with small-bar.
+ */
+ io_size = 0;
+ if (mgr->io_base && !xe_ttm_stolen_cpu_access_needs_ggtt(xe))
+ io_size = stolen_size;
+
+ err = __xe_ttm_vram_mgr_init(xe, &mgr->base, XE_PL_STOLEN, stolen_size,
+ io_size, pgsize);
+ if (err) {
+ drm_dbg_kms(&xe->drm, "Stolen mgr init failed: %i\n", err);
+ return;
+ }
+
+ drm_dbg_kms(&xe->drm, "Initialized stolen memory support with %llu bytes\n",
+ stolen_size);
+
+ if (io_size)
+ mgr->mapping = devm_ioremap_wc(&pdev->dev, mgr->io_base, io_size);
+}
+
+u64 xe_ttm_stolen_io_offset(struct xe_bo *bo, u32 offset)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ struct ttm_resource_manager *ttm_mgr = ttm_manager_type(&xe->ttm, XE_PL_STOLEN);
+ struct xe_ttm_stolen_mgr *mgr = to_stolen_mgr(ttm_mgr);
+ struct xe_res_cursor cur;
+
+ XE_WARN_ON(!mgr->io_base);
+
+ if (xe_ttm_stolen_cpu_access_needs_ggtt(xe))
+ return mgr->io_base + xe_bo_ggtt_addr(bo) + offset;
+
+ xe_res_first(bo->ttm.resource, offset, 4096, &cur);
+ return mgr->io_base + cur.start;
+}
+
+static int __xe_ttm_stolen_io_mem_reserve_bar2(struct xe_device *xe,
+ struct xe_ttm_stolen_mgr *mgr,
+ struct ttm_resource *mem)
+{
+ struct xe_res_cursor cur;
+
+ if (!mgr->io_base)
+ return -EIO;
+
+ xe_res_first(mem, 0, 4096, &cur);
+ mem->bus.offset = cur.start;
+
+ drm_WARN_ON(&xe->drm, !(mem->placement & TTM_PL_FLAG_CONTIGUOUS));
+
+ if (mem->placement & TTM_PL_FLAG_CONTIGUOUS && mgr->mapping)
+ mem->bus.addr = (u8 *)mgr->mapping + mem->bus.offset;
+
+ mem->bus.offset += mgr->io_base;
+ mem->bus.is_iomem = true;
+ mem->bus.caching = ttm_write_combined;
+
+ return 0;
+}
+
+static int __xe_ttm_stolen_io_mem_reserve_stolen(struct xe_device *xe,
+ struct xe_ttm_stolen_mgr *mgr,
+ struct ttm_resource *mem)
+{
+#ifdef CONFIG_X86
+ struct xe_bo *bo = ttm_to_xe_bo(mem->bo);
+
+ XE_WARN_ON(IS_DGFX(xe));
+
+ /* XXX: Require BO to be mapped to GGTT? */
+ if (drm_WARN_ON(&xe->drm, !(bo->flags & XE_BO_CREATE_GGTT_BIT)))
+ return -EIO;
+
+ /* GGTT is always contiguously mapped */
+ mem->bus.offset = xe_bo_ggtt_addr(bo) + mgr->io_base;
+
+ mem->bus.is_iomem = true;
+ mem->bus.caching = ttm_write_combined;
+
+ return 0;
+#else
+ /* How is it even possible to get here without gen12 stolen? */
+ drm_WARN_ON(&xe->drm, 1);
+ return -EIO;
+#endif
+}
+
+int xe_ttm_stolen_io_mem_reserve(struct xe_device *xe, struct ttm_resource *mem)
+{
+ struct ttm_resource_manager *ttm_mgr = ttm_manager_type(&xe->ttm, XE_PL_STOLEN);
+ struct xe_ttm_stolen_mgr *mgr = ttm_mgr ? to_stolen_mgr(ttm_mgr) : NULL;
+
+ if (!mgr || !mgr->io_base)
+ return -EIO;
+
+ if (xe_ttm_stolen_cpu_access_needs_ggtt(xe))
+ return __xe_ttm_stolen_io_mem_reserve_stolen(xe, mgr, mem);
+ else
+ return __xe_ttm_stolen_io_mem_reserve_bar2(xe, mgr, mem);
+}
+
+u64 xe_ttm_stolen_gpu_offset(struct xe_device *xe)
+{
+ struct xe_ttm_stolen_mgr *mgr =
+ to_stolen_mgr(ttm_manager_type(&xe->ttm, XE_PL_STOLEN));
+
+ return mgr->stolen_base;
+}
diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h
new file mode 100644
index 000000000000..1777245ff810
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h
@@ -0,0 +1,21 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_TTM_STOLEN_MGR_H_
+#define _XE_TTM_STOLEN_MGR_H_
+
+#include <linux/types.h>
+
+struct ttm_resource;
+struct xe_bo;
+struct xe_device;
+
+void xe_ttm_stolen_mgr_init(struct xe_device *xe);
+int xe_ttm_stolen_io_mem_reserve(struct xe_device *xe, struct ttm_resource *mem);
+bool xe_ttm_stolen_cpu_access_needs_ggtt(struct xe_device *xe);
+u64 xe_ttm_stolen_io_offset(struct xe_bo *bo, u32 offset);
+u64 xe_ttm_stolen_gpu_offset(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
new file mode 100644
index 000000000000..3e1fa0c832ca
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.c
@@ -0,0 +1,118 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021-2022 Intel Corporation
+ * Copyright (C) 2021-2002 Red Hat
+ */
+
+#include "xe_ttm_sys_mgr.h"
+
+#include <drm/drm_managed.h>
+
+#include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_range_manager.h>
+#include <drm/ttm/ttm_tt.h>
+
+#include "xe_bo.h"
+#include "xe_gt.h"
+
+struct xe_ttm_sys_node {
+ struct ttm_buffer_object *tbo;
+ struct ttm_range_mgr_node base;
+};
+
+static inline struct xe_ttm_sys_node *
+to_xe_ttm_sys_node(struct ttm_resource *res)
+{
+ return container_of(res, struct xe_ttm_sys_node, base.base);
+}
+
+static int xe_ttm_sys_mgr_new(struct ttm_resource_manager *man,
+ struct ttm_buffer_object *tbo,
+ const struct ttm_place *place,
+ struct ttm_resource **res)
+{
+ struct xe_ttm_sys_node *node;
+ int r;
+
+ node = kzalloc(struct_size(node, base.mm_nodes, 1), GFP_KERNEL);
+ if (!node)
+ return -ENOMEM;
+
+ node->tbo = tbo;
+ ttm_resource_init(tbo, place, &node->base.base);
+
+ if (!(place->flags & TTM_PL_FLAG_TEMPORARY) &&
+ ttm_resource_manager_usage(man) > (man->size << PAGE_SHIFT)) {
+ r = -ENOSPC;
+ goto err_fini;
+ }
+
+ node->base.mm_nodes[0].start = 0;
+ node->base.mm_nodes[0].size = PFN_UP(node->base.base.size);
+ node->base.base.start = XE_BO_INVALID_OFFSET;
+
+ *res = &node->base.base;
+
+ return 0;
+
+err_fini:
+ ttm_resource_fini(man, &node->base.base);
+ kfree(node);
+ return r;
+}
+
+static void xe_ttm_sys_mgr_del(struct ttm_resource_manager *man,
+ struct ttm_resource *res)
+{
+ struct xe_ttm_sys_node *node = to_xe_ttm_sys_node(res);
+
+ ttm_resource_fini(man, res);
+ kfree(node);
+}
+
+static void xe_ttm_sys_mgr_debug(struct ttm_resource_manager *man,
+ struct drm_printer *printer)
+{
+
+}
+
+static const struct ttm_resource_manager_func xe_ttm_sys_mgr_func = {
+ .alloc = xe_ttm_sys_mgr_new,
+ .free = xe_ttm_sys_mgr_del,
+ .debug = xe_ttm_sys_mgr_debug
+};
+
+static void ttm_sys_mgr_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_device *xe = (struct xe_device *)arg;
+ struct ttm_resource_manager *man = &xe->mem.sys_mgr;
+ int err;
+
+ ttm_resource_manager_set_used(man, false);
+
+ err = ttm_resource_manager_evict_all(&xe->ttm, man);
+ if (err)
+ return;
+
+ ttm_resource_manager_cleanup(man);
+ ttm_set_driver_manager(&xe->ttm, XE_PL_TT, NULL);
+}
+
+int xe_ttm_sys_mgr_init(struct xe_device *xe)
+{
+ struct ttm_resource_manager *man = &xe->mem.sys_mgr;
+ struct sysinfo si;
+ u64 gtt_size;
+
+ si_meminfo(&si);
+ gtt_size = (u64)si.totalram * si.mem_unit;
+ /* TTM limits allocation of all TTM devices by 50% of system memory */
+ gtt_size /= 2;
+
+ man->use_tt = true;
+ man->func = &xe_ttm_sys_mgr_func;
+ ttm_resource_manager_init(man, &xe->ttm, gtt_size >> PAGE_SHIFT);
+ ttm_set_driver_manager(&xe->ttm, XE_PL_TT, man);
+ ttm_resource_manager_set_used(man, true);
+ return drmm_add_action_or_reset(&xe->drm, ttm_sys_mgr_fini, xe);
+}
diff --git a/drivers/gpu/drm/xe/xe_ttm_sys_mgr.h b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.h
new file mode 100644
index 000000000000..e8f5cd395b28
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_sys_mgr.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2023 Intel Corporation
+ */
+
+#ifndef _XE_TTM_SYS_MGR_H_
+#define _XE_TTM_SYS_MGR_H_
+
+struct xe_device;
+
+int xe_ttm_sys_mgr_init(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
new file mode 100644
index 000000000000..115ec745e502
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.c
@@ -0,0 +1,480 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021-2022 Intel Corporation
+ * Copyright (C) 2021-2002 Red Hat
+ */
+
+#include <drm/drm_managed.h>
+
+#include <drm/ttm/ttm_placement.h>
+#include <drm/ttm/ttm_range_manager.h>
+
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_res_cursor.h"
+#include "xe_ttm_vram_mgr.h"
+
+static inline struct drm_buddy_block *
+xe_ttm_vram_mgr_first_block(struct list_head *list)
+{
+ return list_first_entry_or_null(list, struct drm_buddy_block, link);
+}
+
+static inline bool xe_is_vram_mgr_blocks_contiguous(struct drm_buddy *mm,
+ struct list_head *head)
+{
+ struct drm_buddy_block *block;
+ u64 start, size;
+
+ block = xe_ttm_vram_mgr_first_block(head);
+ if (!block)
+ return false;
+
+ while (head != block->link.next) {
+ start = drm_buddy_block_offset(block);
+ size = drm_buddy_block_size(mm, block);
+
+ block = list_entry(block->link.next, struct drm_buddy_block,
+ link);
+ if (start + size != drm_buddy_block_offset(block))
+ return false;
+ }
+
+ return true;
+}
+
+static int xe_ttm_vram_mgr_new(struct ttm_resource_manager *man,
+ struct ttm_buffer_object *tbo,
+ const struct ttm_place *place,
+ struct ttm_resource **res)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+ struct xe_ttm_vram_mgr_resource *vres;
+ struct drm_buddy *mm = &mgr->mm;
+ u64 size, remaining_size, min_page_size;
+ unsigned long lpfn;
+ int err;
+
+ lpfn = place->lpfn;
+ if (!lpfn || lpfn > man->size >> PAGE_SHIFT)
+ lpfn = man->size >> PAGE_SHIFT;
+
+ if (tbo->base.size >> PAGE_SHIFT > (lpfn - place->fpfn))
+ return -E2BIG; /* don't trigger eviction for the impossible */
+
+ vres = kzalloc(sizeof(*vres), GFP_KERNEL);
+ if (!vres)
+ return -ENOMEM;
+
+ ttm_resource_init(tbo, place, &vres->base);
+
+ /* bail out quickly if there's likely not enough VRAM for this BO */
+ if (ttm_resource_manager_usage(man) > man->size) {
+ err = -ENOSPC;
+ goto error_fini;
+ }
+
+ INIT_LIST_HEAD(&vres->blocks);
+
+ if (place->flags & TTM_PL_FLAG_TOPDOWN)
+ vres->flags |= DRM_BUDDY_TOPDOWN_ALLOCATION;
+
+ if (place->fpfn || lpfn != man->size >> PAGE_SHIFT)
+ vres->flags |= DRM_BUDDY_RANGE_ALLOCATION;
+
+ if (WARN_ON(!vres->base.size)) {
+ err = -EINVAL;
+ goto error_fini;
+ }
+ size = vres->base.size;
+
+ min_page_size = mgr->default_page_size;
+ if (tbo->page_alignment)
+ min_page_size = tbo->page_alignment << PAGE_SHIFT;
+
+ if (WARN_ON(min_page_size < mm->chunk_size)) {
+ err = -EINVAL;
+ goto error_fini;
+ }
+
+ if (WARN_ON(min_page_size > SZ_2G)) { /* FIXME: sg limit */
+ err = -EINVAL;
+ goto error_fini;
+ }
+
+ if (WARN_ON((size > SZ_2G &&
+ (vres->base.placement & TTM_PL_FLAG_CONTIGUOUS)))) {
+ err = -EINVAL;
+ goto error_fini;
+ }
+
+ if (WARN_ON(!IS_ALIGNED(size, min_page_size))) {
+ err = -EINVAL;
+ goto error_fini;
+ }
+
+ mutex_lock(&mgr->lock);
+ if (lpfn <= mgr->visible_size >> PAGE_SHIFT && size > mgr->visible_avail) {
+ mutex_unlock(&mgr->lock);
+ err = -ENOSPC;
+ goto error_fini;
+ }
+
+ if (place->fpfn + (size >> PAGE_SHIFT) != place->lpfn &&
+ place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ size = roundup_pow_of_two(size);
+ min_page_size = size;
+
+ lpfn = max_t(unsigned long, place->fpfn + (size >> PAGE_SHIFT), lpfn);
+ }
+
+ remaining_size = size;
+ do {
+ /*
+ * Limit maximum size to 2GiB due to SG table limitations.
+ * FIXME: Should maybe be handled as part of sg construction.
+ */
+ u64 alloc_size = min_t(u64, remaining_size, SZ_2G);
+
+ err = drm_buddy_alloc_blocks(mm, (u64)place->fpfn << PAGE_SHIFT,
+ (u64)lpfn << PAGE_SHIFT,
+ alloc_size,
+ min_page_size,
+ &vres->blocks,
+ vres->flags);
+ if (err)
+ goto error_free_blocks;
+
+ remaining_size -= alloc_size;
+ } while (remaining_size);
+
+ if (place->flags & TTM_PL_FLAG_CONTIGUOUS) {
+ if (!drm_buddy_block_trim(mm, vres->base.size, &vres->blocks))
+ size = vres->base.size;
+ }
+
+ if (lpfn <= mgr->visible_size >> PAGE_SHIFT) {
+ vres->used_visible_size = size;
+ } else {
+ struct drm_buddy_block *block;
+
+ list_for_each_entry(block, &vres->blocks, link) {
+ u64 start = drm_buddy_block_offset(block);
+
+ if (start < mgr->visible_size) {
+ u64 end = start + drm_buddy_block_size(mm, block);
+
+ vres->used_visible_size +=
+ min(end, mgr->visible_size) - start;
+ }
+ }
+ }
+
+ mgr->visible_avail -= vres->used_visible_size;
+ mutex_unlock(&mgr->lock);
+
+ if (!(vres->base.placement & TTM_PL_FLAG_CONTIGUOUS) &&
+ xe_is_vram_mgr_blocks_contiguous(mm, &vres->blocks))
+ vres->base.placement |= TTM_PL_FLAG_CONTIGUOUS;
+
+ /*
+ * For some kernel objects we still rely on the start when io mapping
+ * the object.
+ */
+ if (vres->base.placement & TTM_PL_FLAG_CONTIGUOUS) {
+ struct drm_buddy_block *block = list_first_entry(&vres->blocks,
+ typeof(*block),
+ link);
+
+ vres->base.start = drm_buddy_block_offset(block) >> PAGE_SHIFT;
+ } else {
+ vres->base.start = XE_BO_INVALID_OFFSET;
+ }
+
+ *res = &vres->base;
+ return 0;
+
+error_free_blocks:
+ drm_buddy_free_list(mm, &vres->blocks);
+ mutex_unlock(&mgr->lock);
+error_fini:
+ ttm_resource_fini(man, &vres->base);
+ kfree(vres);
+
+ return err;
+}
+
+static void xe_ttm_vram_mgr_del(struct ttm_resource_manager *man,
+ struct ttm_resource *res)
+{
+ struct xe_ttm_vram_mgr_resource *vres =
+ to_xe_ttm_vram_mgr_resource(res);
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+ struct drm_buddy *mm = &mgr->mm;
+
+ mutex_lock(&mgr->lock);
+ drm_buddy_free_list(mm, &vres->blocks);
+ mgr->visible_avail += vres->used_visible_size;
+ mutex_unlock(&mgr->lock);
+
+ ttm_resource_fini(man, res);
+
+ kfree(vres);
+}
+
+static void xe_ttm_vram_mgr_debug(struct ttm_resource_manager *man,
+ struct drm_printer *printer)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+ struct drm_buddy *mm = &mgr->mm;
+
+ mutex_lock(&mgr->lock);
+ drm_printf(printer, "default_page_size: %lluKiB\n",
+ mgr->default_page_size >> 10);
+ drm_printf(printer, "visible_avail: %lluMiB\n",
+ (u64)mgr->visible_avail >> 20);
+ drm_printf(printer, "visible_size: %lluMiB\n",
+ (u64)mgr->visible_size >> 20);
+
+ drm_buddy_print(mm, printer);
+ mutex_unlock(&mgr->lock);
+ drm_printf(printer, "man size:%llu\n", man->size);
+}
+
+static bool xe_ttm_vram_mgr_intersects(struct ttm_resource_manager *man,
+ struct ttm_resource *res,
+ const struct ttm_place *place,
+ size_t size)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+ struct xe_ttm_vram_mgr_resource *vres =
+ to_xe_ttm_vram_mgr_resource(res);
+ struct drm_buddy *mm = &mgr->mm;
+ struct drm_buddy_block *block;
+
+ if (!place->fpfn && !place->lpfn)
+ return true;
+
+ if (!place->fpfn && place->lpfn == mgr->visible_size >> PAGE_SHIFT)
+ return vres->used_visible_size > 0;
+
+ list_for_each_entry(block, &vres->blocks, link) {
+ unsigned long fpfn =
+ drm_buddy_block_offset(block) >> PAGE_SHIFT;
+ unsigned long lpfn = fpfn +
+ (drm_buddy_block_size(mm, block) >> PAGE_SHIFT);
+
+ if (place->fpfn < lpfn && place->lpfn > fpfn)
+ return true;
+ }
+
+ return false;
+}
+
+static bool xe_ttm_vram_mgr_compatible(struct ttm_resource_manager *man,
+ struct ttm_resource *res,
+ const struct ttm_place *place,
+ size_t size)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+ struct xe_ttm_vram_mgr_resource *vres =
+ to_xe_ttm_vram_mgr_resource(res);
+ struct drm_buddy *mm = &mgr->mm;
+ struct drm_buddy_block *block;
+
+ if (!place->fpfn && !place->lpfn)
+ return true;
+
+ if (!place->fpfn && place->lpfn == mgr->visible_size >> PAGE_SHIFT)
+ return vres->used_visible_size == size;
+
+ list_for_each_entry(block, &vres->blocks, link) {
+ unsigned long fpfn =
+ drm_buddy_block_offset(block) >> PAGE_SHIFT;
+ unsigned long lpfn = fpfn +
+ (drm_buddy_block_size(mm, block) >> PAGE_SHIFT);
+
+ if (fpfn < place->fpfn || lpfn > place->lpfn)
+ return false;
+ }
+
+ return true;
+}
+
+static const struct ttm_resource_manager_func xe_ttm_vram_mgr_func = {
+ .alloc = xe_ttm_vram_mgr_new,
+ .free = xe_ttm_vram_mgr_del,
+ .intersects = xe_ttm_vram_mgr_intersects,
+ .compatible = xe_ttm_vram_mgr_compatible,
+ .debug = xe_ttm_vram_mgr_debug
+};
+
+static void ttm_vram_mgr_fini(struct drm_device *dev, void *arg)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_ttm_vram_mgr *mgr = arg;
+ struct ttm_resource_manager *man = &mgr->manager;
+
+ ttm_resource_manager_set_used(man, false);
+
+ if (ttm_resource_manager_evict_all(&xe->ttm, man))
+ return;
+
+ WARN_ON_ONCE(mgr->visible_avail != mgr->visible_size);
+
+ drm_buddy_fini(&mgr->mm);
+
+ ttm_resource_manager_cleanup(&mgr->manager);
+
+ ttm_set_driver_manager(&xe->ttm, mgr->mem_type, NULL);
+
+ mutex_destroy(&mgr->lock);
+}
+
+int __xe_ttm_vram_mgr_init(struct xe_device *xe, struct xe_ttm_vram_mgr *mgr,
+ u32 mem_type, u64 size, u64 io_size,
+ u64 default_page_size)
+{
+ struct ttm_resource_manager *man = &mgr->manager;
+ int err;
+
+ man->func = &xe_ttm_vram_mgr_func;
+ mgr->mem_type = mem_type;
+ mutex_init(&mgr->lock);
+ mgr->default_page_size = default_page_size;
+ mgr->visible_size = io_size;
+ mgr->visible_avail = io_size;
+
+ ttm_resource_manager_init(man, &xe->ttm, size);
+ err = drm_buddy_init(&mgr->mm, man->size, default_page_size);
+ if (err)
+ return err;
+
+ ttm_set_driver_manager(&xe->ttm, mem_type, &mgr->manager);
+ ttm_resource_manager_set_used(&mgr->manager, true);
+
+ return drmm_add_action_or_reset(&xe->drm, ttm_vram_mgr_fini, mgr);
+}
+
+int xe_ttm_vram_mgr_init(struct xe_tile *tile, struct xe_ttm_vram_mgr *mgr)
+{
+ struct xe_device *xe = tile_to_xe(tile);
+ struct xe_mem_region *vram = &tile->mem.vram;
+
+ mgr->vram = vram;
+ return __xe_ttm_vram_mgr_init(xe, mgr, XE_PL_VRAM0 + tile->id,
+ vram->usable_size, vram->io_size,
+ PAGE_SIZE);
+}
+
+int xe_ttm_vram_mgr_alloc_sgt(struct xe_device *xe,
+ struct ttm_resource *res,
+ u64 offset, u64 length,
+ struct device *dev,
+ enum dma_data_direction dir,
+ struct sg_table **sgt)
+{
+ struct xe_tile *tile = &xe->tiles[res->mem_type - XE_PL_VRAM0];
+ struct xe_ttm_vram_mgr_resource *vres = to_xe_ttm_vram_mgr_resource(res);
+ struct xe_res_cursor cursor;
+ struct scatterlist *sg;
+ int num_entries = 0;
+ int i, r;
+
+ if (vres->used_visible_size < res->size)
+ return -EOPNOTSUPP;
+
+ *sgt = kmalloc(sizeof(**sgt), GFP_KERNEL);
+ if (!*sgt)
+ return -ENOMEM;
+
+ /* Determine the number of DRM_BUDDY blocks to export */
+ xe_res_first(res, offset, length, &cursor);
+ while (cursor.remaining) {
+ num_entries++;
+ xe_res_next(&cursor, cursor.size);
+ }
+
+ r = sg_alloc_table(*sgt, num_entries, GFP_KERNEL);
+ if (r)
+ goto error_free;
+
+ /* Initialize scatterlist nodes of sg_table */
+ for_each_sgtable_sg((*sgt), sg, i)
+ sg->length = 0;
+
+ /*
+ * Walk down DRM_BUDDY blocks to populate scatterlist nodes
+ * @note: Use iterator api to get first the DRM_BUDDY block
+ * and the number of bytes from it. Access the following
+ * DRM_BUDDY block(s) if more buffer needs to exported
+ */
+ xe_res_first(res, offset, length, &cursor);
+ for_each_sgtable_sg((*sgt), sg, i) {
+ phys_addr_t phys = cursor.start + tile->mem.vram.io_start;
+ size_t size = cursor.size;
+ dma_addr_t addr;
+
+ addr = dma_map_resource(dev, phys, size, dir,
+ DMA_ATTR_SKIP_CPU_SYNC);
+ r = dma_mapping_error(dev, addr);
+ if (r)
+ goto error_unmap;
+
+ sg_set_page(sg, NULL, size, 0);
+ sg_dma_address(sg) = addr;
+ sg_dma_len(sg) = size;
+
+ xe_res_next(&cursor, cursor.size);
+ }
+
+ return 0;
+
+error_unmap:
+ for_each_sgtable_sg((*sgt), sg, i) {
+ if (!sg->length)
+ continue;
+
+ dma_unmap_resource(dev, sg->dma_address,
+ sg->length, dir,
+ DMA_ATTR_SKIP_CPU_SYNC);
+ }
+ sg_free_table(*sgt);
+
+error_free:
+ kfree(*sgt);
+ return r;
+}
+
+void xe_ttm_vram_mgr_free_sgt(struct device *dev, enum dma_data_direction dir,
+ struct sg_table *sgt)
+{
+ struct scatterlist *sg;
+ int i;
+
+ for_each_sgtable_sg(sgt, sg, i)
+ dma_unmap_resource(dev, sg->dma_address,
+ sg->length, dir,
+ DMA_ATTR_SKIP_CPU_SYNC);
+ sg_free_table(sgt);
+ kfree(sgt);
+}
+
+u64 xe_ttm_vram_get_cpu_visible_size(struct ttm_resource_manager *man)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+
+ return mgr->visible_size;
+}
+
+void xe_ttm_vram_get_used(struct ttm_resource_manager *man,
+ u64 *used, u64 *used_visible)
+{
+ struct xe_ttm_vram_mgr *mgr = to_xe_ttm_vram_mgr(man);
+
+ mutex_lock(&mgr->lock);
+ *used = mgr->mm.size - mgr->mm.avail;
+ *used_visible = mgr->visible_size - mgr->visible_avail;
+ mutex_unlock(&mgr->lock);
+}
diff --git a/drivers/gpu/drm/xe/xe_ttm_vram_mgr.h b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.h
new file mode 100644
index 000000000000..d184e19a9230
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_vram_mgr.h
@@ -0,0 +1,44 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_TTM_VRAM_MGR_H_
+#define _XE_TTM_VRAM_MGR_H_
+
+#include "xe_ttm_vram_mgr_types.h"
+
+enum dma_data_direction;
+struct xe_device;
+struct xe_tile;
+
+int __xe_ttm_vram_mgr_init(struct xe_device *xe, struct xe_ttm_vram_mgr *mgr,
+ u32 mem_type, u64 size, u64 io_size,
+ u64 default_page_size);
+int xe_ttm_vram_mgr_init(struct xe_tile *tile, struct xe_ttm_vram_mgr *mgr);
+int xe_ttm_vram_mgr_alloc_sgt(struct xe_device *xe,
+ struct ttm_resource *res,
+ u64 offset, u64 length,
+ struct device *dev,
+ enum dma_data_direction dir,
+ struct sg_table **sgt);
+void xe_ttm_vram_mgr_free_sgt(struct device *dev, enum dma_data_direction dir,
+ struct sg_table *sgt);
+
+u64 xe_ttm_vram_get_cpu_visible_size(struct ttm_resource_manager *man);
+void xe_ttm_vram_get_used(struct ttm_resource_manager *man,
+ u64 *used, u64 *used_visible);
+
+static inline struct xe_ttm_vram_mgr_resource *
+to_xe_ttm_vram_mgr_resource(struct ttm_resource *res)
+{
+ return container_of(res, struct xe_ttm_vram_mgr_resource, base);
+}
+
+static inline struct xe_ttm_vram_mgr *
+to_xe_ttm_vram_mgr(struct ttm_resource_manager *man)
+{
+ return container_of(man, struct xe_ttm_vram_mgr, manager);
+}
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h b/drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h
new file mode 100644
index 000000000000..2d75cf126289
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_ttm_vram_mgr_types.h
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_TTM_VRAM_MGR_TYPES_H_
+#define _XE_TTM_VRAM_MGR_TYPES_H_
+
+#include <drm/drm_buddy.h>
+#include <drm/ttm/ttm_device.h>
+
+struct xe_mem_region;
+
+/**
+ * struct xe_ttm_vram_mgr - XE TTM VRAM manager
+ *
+ * Manages placement of TTM resource in VRAM.
+ */
+struct xe_ttm_vram_mgr {
+ /** @manager: Base TTM resource manager */
+ struct ttm_resource_manager manager;
+ /** @mm: DRM buddy allocator which manages the VRAM */
+ struct drm_buddy mm;
+ /** @vram: ptr to details of associated VRAM region */
+ struct xe_mem_region *vram;
+ /** @visible_size: Proped size of the CPU visible portion */
+ u64 visible_size;
+ /** @visible_avail: CPU visible portion still unallocated */
+ u64 visible_avail;
+ /** @default_page_size: default page size */
+ u64 default_page_size;
+ /** @lock: protects allocations of VRAM */
+ struct mutex lock;
+ /** @mem_type: The TTM memory type */
+ u32 mem_type;
+};
+
+/**
+ * struct xe_ttm_vram_mgr_resource - XE TTM VRAM resource
+ */
+struct xe_ttm_vram_mgr_resource {
+ /** @base: Base TTM resource */
+ struct ttm_resource base;
+ /** @blocks: list of DRM buddy blocks */
+ struct list_head blocks;
+ /** @used_visible_size: How many CPU visible bytes this resource is using */
+ u64 used_visible_size;
+ /** @flags: flags associated with the resource */
+ unsigned long flags;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_tuning.c b/drivers/gpu/drm/xe/xe_tuning.c
new file mode 100644
index 000000000000..53ccd338fd8c
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tuning.c
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_tuning.h"
+
+#include <kunit/visibility.h>
+
+#include "regs/xe_gt_regs.h"
+#include "xe_gt_types.h"
+#include "xe_platform_types.h"
+#include "xe_rtp.h"
+
+#undef XE_REG_MCR
+#define XE_REG_MCR(...) XE_REG(__VA_ARGS__, .mcr = 1)
+
+static const struct xe_rtp_entry_sr gt_tunings[] = {
+ { XE_RTP_NAME("Tuning: Blend Fill Caching Optimization Disable"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_L3SCQREG7, BLEND_FILL_CACHING_OPT_DIS))
+ },
+ { XE_RTP_NAME("Tuning: 32B Access Enable"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_SQCM, EN_32B_ACCESS))
+ },
+
+ /* Xe2 */
+
+ { XE_RTP_NAME("Tuning: L3 cache"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004)),
+ XE_RTP_ACTIONS(FIELD_SET(XEHP_L3SQCREG5, L3_PWM_TIMER_INIT_VAL_MASK,
+ REG_FIELD_PREP(L3_PWM_TIMER_INIT_VAL_MASK, 0x7f)))
+ },
+ { XE_RTP_NAME("Tuning: L3 cache - media"),
+ XE_RTP_RULES(MEDIA_VERSION(2000)),
+ XE_RTP_ACTIONS(FIELD_SET(XE2LPM_L3SQCREG5, L3_PWM_TIMER_INIT_VAL_MASK,
+ REG_FIELD_PREP(L3_PWM_TIMER_INIT_VAL_MASK, 0x7f)))
+ },
+
+ {}
+};
+
+static const struct xe_rtp_entry_sr engine_tunings[] = {
+ { XE_RTP_NAME("Tuning: Set Indirect State Override"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1271),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(SAMPLER_MODE, INDIRECT_STATE_BASE_ADDR_OVERRIDE))
+ },
+ {}
+};
+
+static const struct xe_rtp_entry_sr lrc_tunings[] = {
+ { XE_RTP_NAME("Tuning: ganged timer, also known as 16011163337"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ /* read verification is ignored due to 1608008084. */
+ XE_RTP_ACTIONS(FIELD_SET_NO_READ_MASK(FF_MODE2,
+ FF_MODE2_GS_TIMER_MASK,
+ FF_MODE2_GS_TIMER_224))
+ },
+
+ /* DG2 */
+
+ { XE_RTP_NAME("Tuning: L3 cache"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(FIELD_SET(XEHP_L3SQCREG5, L3_PWM_TIMER_INIT_VAL_MASK,
+ REG_FIELD_PREP(L3_PWM_TIMER_INIT_VAL_MASK, 0x7f)))
+ },
+ { XE_RTP_NAME("Tuning: TDS gang timer"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ /* read verification is ignored as in i915 - need to check enabling */
+ XE_RTP_ACTIONS(FIELD_SET_NO_READ_MASK(XEHP_FF_MODE2,
+ FF_MODE2_TDS_TIMER_MASK,
+ FF_MODE2_TDS_TIMER_128))
+ },
+ { XE_RTP_NAME("Tuning: TBIMR fast clip"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(CHICKEN_RASTER_2, TBIMR_FAST_CLIP))
+ },
+
+ /* Xe_LPG */
+
+ { XE_RTP_NAME("Tuning: L3 cache"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(FIELD_SET(XEHP_L3SQCREG5, L3_PWM_TIMER_INIT_VAL_MASK,
+ REG_FIELD_PREP(L3_PWM_TIMER_INIT_VAL_MASK, 0x7f)))
+ },
+
+ {}
+};
+
+void xe_tuning_process_gt(struct xe_gt *gt)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(gt);
+
+ xe_rtp_process_to_sr(&ctx, gt_tunings, &gt->reg_sr);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_tuning_process_gt);
+
+void xe_tuning_process_engine(struct xe_hw_engine *hwe)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+
+ xe_rtp_process_to_sr(&ctx, engine_tunings, &hwe->reg_sr);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_tuning_process_engine);
+
+/**
+ * xe_tuning_process_lrc - process lrc tunings
+ * @hwe: engine instance to process tunings for
+ *
+ * Process LRC table for this platform, saving in @hwe all the tunings that need
+ * to be applied on context restore. These are tunings touching registers that
+ * are part of the HW context image.
+ */
+void xe_tuning_process_lrc(struct xe_hw_engine *hwe)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+
+ xe_rtp_process_to_sr(&ctx, lrc_tunings, &hwe->reg_lrc);
+}
diff --git a/drivers/gpu/drm/xe/xe_tuning.h b/drivers/gpu/drm/xe/xe_tuning.h
new file mode 100644
index 000000000000..4f9c3ac3b516
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_tuning.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_TUNING_
+#define _XE_TUNING_
+
+struct xe_gt;
+struct xe_hw_engine;
+
+void xe_tuning_process_gt(struct xe_gt *gt);
+void xe_tuning_process_engine(struct xe_hw_engine *hwe);
+void xe_tuning_process_lrc(struct xe_hw_engine *hwe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc.c b/drivers/gpu/drm/xe/xe_uc.c
new file mode 100644
index 000000000000..25e1ddfd2f86
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc.c
@@ -0,0 +1,258 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_uc.h"
+
+#include "xe_device.h"
+#include "xe_gsc.h"
+#include "xe_gt.h"
+#include "xe_guc.h"
+#include "xe_guc_pc.h"
+#include "xe_guc_submit.h"
+#include "xe_huc.h"
+#include "xe_uc_fw.h"
+#include "xe_wopcm.h"
+
+static struct xe_gt *
+uc_to_gt(struct xe_uc *uc)
+{
+ return container_of(uc, struct xe_gt, uc);
+}
+
+static struct xe_device *
+uc_to_xe(struct xe_uc *uc)
+{
+ return gt_to_xe(uc_to_gt(uc));
+}
+
+/* Should be called once at driver load only */
+int xe_uc_init(struct xe_uc *uc)
+{
+ int ret;
+
+ /*
+ * We call the GuC/HuC/GSC init functions even if GuC submission is off
+ * to correctly move our tracking of the FW state to "disabled".
+ */
+
+ ret = xe_guc_init(&uc->guc);
+ if (ret)
+ goto err;
+
+ ret = xe_huc_init(&uc->huc);
+ if (ret)
+ goto err;
+
+ ret = xe_gsc_init(&uc->gsc);
+ if (ret)
+ goto err;
+
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ ret = xe_wopcm_init(&uc->wopcm);
+ if (ret)
+ goto err;
+
+ ret = xe_guc_submit_init(&uc->guc);
+ if (ret)
+ goto err;
+
+ return 0;
+
+err:
+ return ret;
+}
+
+/**
+ * xe_uc_init_post_hwconfig - init Uc post hwconfig load
+ * @uc: The UC object
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_uc_init_post_hwconfig(struct xe_uc *uc)
+{
+ int err;
+
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ err = xe_uc_sanitize_reset(uc);
+ if (err)
+ return err;
+
+ err = xe_guc_init_post_hwconfig(&uc->guc);
+ if (err)
+ return err;
+
+ return xe_gsc_init_post_hwconfig(&uc->gsc);
+}
+
+static int uc_reset(struct xe_uc *uc)
+{
+ struct xe_device *xe = uc_to_xe(uc);
+ int ret;
+
+ ret = xe_guc_reset(&uc->guc);
+ if (ret) {
+ drm_err(&xe->drm, "Failed to reset GuC, ret = %d\n", ret);
+ return ret;
+ }
+
+ return 0;
+}
+
+static void xe_uc_sanitize(struct xe_uc *uc)
+{
+ xe_huc_sanitize(&uc->huc);
+ xe_guc_sanitize(&uc->guc);
+}
+
+int xe_uc_sanitize_reset(struct xe_uc *uc)
+{
+ xe_uc_sanitize(uc);
+
+ return uc_reset(uc);
+}
+
+/**
+ * xe_uc_init_hwconfig - minimally init Uc, read and parse hwconfig
+ * @uc: The UC object
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+int xe_uc_init_hwconfig(struct xe_uc *uc)
+{
+ int ret;
+
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ ret = xe_guc_min_load_for_hwconfig(&uc->guc);
+ if (ret)
+ return ret;
+
+ return 0;
+}
+
+/*
+ * Should be called during driver load, after every GT reset, and after every
+ * suspend to reload / auth the firmwares.
+ */
+int xe_uc_init_hw(struct xe_uc *uc)
+{
+ int ret;
+
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ ret = xe_huc_upload(&uc->huc);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_upload(&uc->guc);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_enable_communication(&uc->guc);
+ if (ret)
+ return ret;
+
+ ret = xe_gt_record_default_lrcs(uc_to_gt(uc));
+ if (ret)
+ return ret;
+
+ ret = xe_guc_post_load_init(&uc->guc);
+ if (ret)
+ return ret;
+
+ ret = xe_guc_pc_start(&uc->guc.pc);
+ if (ret)
+ return ret;
+
+ /* We don't fail the driver load if HuC fails to auth, but let's warn */
+ ret = xe_huc_auth(&uc->huc, XE_HUC_AUTH_VIA_GUC);
+ xe_gt_assert(uc_to_gt(uc), !ret);
+
+ /* GSC load is async */
+ xe_gsc_load_start(&uc->gsc);
+
+ return 0;
+}
+
+int xe_uc_fini_hw(struct xe_uc *uc)
+{
+ return xe_uc_sanitize_reset(uc);
+}
+
+int xe_uc_reset_prepare(struct xe_uc *uc)
+{
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ return xe_guc_reset_prepare(&uc->guc);
+}
+
+void xe_uc_gucrc_disable(struct xe_uc *uc)
+{
+ XE_WARN_ON(xe_guc_pc_gucrc_disable(&uc->guc.pc));
+}
+
+void xe_uc_stop_prepare(struct xe_uc *uc)
+{
+ xe_gsc_wait_for_worker_completion(&uc->gsc);
+ xe_guc_stop_prepare(&uc->guc);
+}
+
+int xe_uc_stop(struct xe_uc *uc)
+{
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ return xe_guc_stop(&uc->guc);
+}
+
+int xe_uc_start(struct xe_uc *uc)
+{
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ return xe_guc_start(&uc->guc);
+}
+
+static void uc_reset_wait(struct xe_uc *uc)
+{
+ int ret;
+
+again:
+ xe_guc_reset_wait(&uc->guc);
+
+ ret = xe_uc_reset_prepare(uc);
+ if (ret)
+ goto again;
+}
+
+int xe_uc_suspend(struct xe_uc *uc)
+{
+ int ret;
+
+ /* GuC submission not enabled, nothing to do */
+ if (!xe_device_uc_enabled(uc_to_xe(uc)))
+ return 0;
+
+ uc_reset_wait(uc);
+
+ ret = xe_uc_stop(uc);
+ if (ret)
+ return ret;
+
+ return xe_guc_suspend(&uc->guc);
+}
diff --git a/drivers/gpu/drm/xe/xe_uc.h b/drivers/gpu/drm/xe/xe_uc.h
new file mode 100644
index 000000000000..5d5110c0c834
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc.h
@@ -0,0 +1,24 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_H_
+#define _XE_UC_H_
+
+#include "xe_uc_types.h"
+
+int xe_uc_init(struct xe_uc *uc);
+int xe_uc_init_hwconfig(struct xe_uc *uc);
+int xe_uc_init_post_hwconfig(struct xe_uc *uc);
+int xe_uc_init_hw(struct xe_uc *uc);
+int xe_uc_fini_hw(struct xe_uc *uc);
+void xe_uc_gucrc_disable(struct xe_uc *uc);
+int xe_uc_reset_prepare(struct xe_uc *uc);
+void xe_uc_stop_prepare(struct xe_uc *uc);
+int xe_uc_stop(struct xe_uc *uc);
+int xe_uc_start(struct xe_uc *uc);
+int xe_uc_suspend(struct xe_uc *uc);
+int xe_uc_sanitize_reset(struct xe_uc *uc);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc_debugfs.c b/drivers/gpu/drm/xe/xe_uc_debugfs.c
new file mode 100644
index 000000000000..0a39ec5a6e99
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_debugfs.c
@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include <drm/drm_debugfs.h>
+
+#include "xe_gt.h"
+#include "xe_guc_debugfs.h"
+#include "xe_huc_debugfs.h"
+#include "xe_macros.h"
+#include "xe_uc_debugfs.h"
+
+void xe_uc_debugfs_register(struct xe_uc *uc, struct dentry *parent)
+{
+ struct dentry *root;
+
+ root = debugfs_create_dir("uc", parent);
+ if (IS_ERR(root)) {
+ XE_WARN_ON("Create UC directory failed");
+ return;
+ }
+
+ xe_guc_debugfs_register(&uc->guc, root);
+ xe_huc_debugfs_register(&uc->huc, root);
+}
diff --git a/drivers/gpu/drm/xe/xe_uc_debugfs.h b/drivers/gpu/drm/xe/xe_uc_debugfs.h
new file mode 100644
index 000000000000..a13382df2bd7
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_debugfs.h
@@ -0,0 +1,14 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_DEBUGFS_H_
+#define _XE_UC_DEBUGFS_H_
+
+struct dentry;
+struct xe_uc;
+
+void xe_uc_debugfs_register(struct xe_uc *uc, struct dentry *parent);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc_fw.c b/drivers/gpu/drm/xe/xe_uc_fw.c
new file mode 100644
index 000000000000..9dff96dfe455
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_fw.c
@@ -0,0 +1,882 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include <linux/bitfield.h>
+#include <linux/firmware.h>
+
+#include <drm/drm_managed.h>
+
+#include "regs/xe_guc_regs.h"
+#include "xe_bo.h"
+#include "xe_device_types.h"
+#include "xe_force_wake.h"
+#include "xe_gsc.h"
+#include "xe_gt.h"
+#include "xe_map.h"
+#include "xe_mmio.h"
+#include "xe_module.h"
+#include "xe_uc_fw.h"
+
+/*
+ * List of required GuC and HuC binaries per-platform. They must be ordered
+ * based on platform, from newer to older.
+ *
+ * Versioning follows the guidelines from
+ * Documentation/driver-api/firmware/firmware-usage-guidelines.rst. There is a
+ * distinction for platforms being officially supported by the driver or not.
+ * Platforms not available publicly or not yet officially supported by the
+ * driver (under force-probe), use the mmp_ver(): the firmware autoselect logic
+ * will select the firmware from disk with filename that matches the full
+ * "mpp version", i.e. major.minor.patch. mmp_ver() should only be used for
+ * this case.
+ *
+ * For platforms officially supported by the driver, the filename always only
+ * ever contains the major version (GuC) or no version at all (HuC).
+ *
+ * After loading the file, the driver parses the versions embedded in the blob.
+ * The major version needs to match a major version supported by the driver (if
+ * any). The minor version is also checked and a notice emitted to the log if
+ * the version found is smaller than the version wanted. This is done only for
+ * informational purposes so users may have a chance to upgrade, but the driver
+ * still loads and use the older firmware.
+ *
+ * Examples:
+ *
+ * 1) Platform officially supported by i915 - using Tigerlake as example.
+ * Driver loads the following firmware blobs from disk:
+ *
+ * - i915/tgl_guc_<major>.bin
+ * - i915/tgl_huc.bin
+ *
+ * <major> number for GuC is checked that it matches the version inside
+ * the blob. <minor> version is checked and if smaller than the expected
+ * an info message is emitted about that.
+ *
+ * 1) XE_<FUTUREINTELPLATFORM>, still under require_force_probe. Using
+ * "wipplat" as a short-name. Driver loads the following firmware blobs
+ * from disk:
+ *
+ * - xe/wipplat_guc_<major>.<minor>.<patch>.bin
+ * - xe/wipplat_huc_<major>.<minor>.<patch>.bin
+ *
+ * <major> and <minor> are checked that they match the version inside
+ * the blob. Both of them need to match exactly what the driver is
+ * expecting, otherwise it fails.
+ *
+ * 3) Platform officially supported by xe and out of force-probe. Using
+ * "plat" as a short-name. Except for the different directory, the
+ * behavior is the same as (1). Driver loads the following firmware
+ * blobs from disk:
+ *
+ * - xe/plat_guc_<major>.bin
+ * - xe/plat_huc.bin
+ *
+ * <major> number for GuC is checked that it matches the version inside
+ * the blob. <minor> version is checked and if smaller than the expected
+ * an info message is emitted about that.
+ *
+ * For the platforms already released with a major version, they should never be
+ * removed from the table. Instead new entries with newer versions may be added
+ * before them, so they take precedence.
+ *
+ * TODO: Currently there's no fallback on major version. That's because xe
+ * driver only supports the one major version of each firmware in the table.
+ * This needs to be fixed when the major version of GuC is updated.
+ */
+
+struct uc_fw_entry {
+ enum xe_platform platform;
+ struct {
+ const char *path;
+ u16 major;
+ u16 minor;
+ bool full_ver_required;
+ };
+};
+
+struct fw_blobs_by_type {
+ const struct uc_fw_entry *entries;
+ u32 count;
+};
+
+#define XE_GUC_FIRMWARE_DEFS(fw_def, mmp_ver, major_ver) \
+ fw_def(METEORLAKE, major_ver(i915, guc, mtl, 70, 7)) \
+ fw_def(DG2, major_ver(i915, guc, dg2, 70, 5)) \
+ fw_def(DG1, major_ver(i915, guc, dg1, 70, 5)) \
+ fw_def(ALDERLAKE_N, major_ver(i915, guc, tgl, 70, 5)) \
+ fw_def(ALDERLAKE_P, major_ver(i915, guc, adlp, 70, 5)) \
+ fw_def(ALDERLAKE_S, major_ver(i915, guc, tgl, 70, 5)) \
+ fw_def(ROCKETLAKE, major_ver(i915, guc, tgl, 70, 5)) \
+ fw_def(TIGERLAKE, major_ver(i915, guc, tgl, 70, 5))
+
+#define XE_HUC_FIRMWARE_DEFS(fw_def, mmp_ver, no_ver) \
+ fw_def(METEORLAKE, no_ver(i915, huc_gsc, mtl)) \
+ fw_def(DG1, no_ver(i915, huc, dg1)) \
+ fw_def(ALDERLAKE_P, no_ver(i915, huc, tgl)) \
+ fw_def(ALDERLAKE_S, no_ver(i915, huc, tgl)) \
+ fw_def(ROCKETLAKE, no_ver(i915, huc, tgl)) \
+ fw_def(TIGERLAKE, no_ver(i915, huc, tgl))
+
+/* for the GSC FW we match the compatibility version and not the release one */
+#define XE_GSC_FIRMWARE_DEFS(fw_def, major_ver) \
+ fw_def(METEORLAKE, major_ver(i915, gsc, mtl, 1, 0))
+
+#define MAKE_FW_PATH(dir__, uc__, shortname__, version__) \
+ __stringify(dir__) "/" __stringify(shortname__) "_" __stringify(uc__) version__ ".bin"
+
+#define fw_filename_mmp_ver(dir_, uc_, shortname_, a, b, c) \
+ MAKE_FW_PATH(dir_, uc_, shortname_, "_" __stringify(a ## . ## b ## . ## c))
+#define fw_filename_major_ver(dir_, uc_, shortname_, a, b) \
+ MAKE_FW_PATH(dir_, uc_, shortname_, "_" __stringify(a))
+#define fw_filename_no_ver(dir_, uc_, shortname_) \
+ MAKE_FW_PATH(dir_, uc_, shortname_, "")
+
+#define uc_fw_entry_mmp_ver(dir_, uc_, shortname_, a, b, c) \
+ { fw_filename_mmp_ver(dir_, uc_, shortname_, a, b, c), \
+ a, b, true }
+#define uc_fw_entry_major_ver(dir_, uc_, shortname_, a, b) \
+ { fw_filename_major_ver(dir_, uc_, shortname_, a, b), \
+ a, b }
+#define uc_fw_entry_no_ver(dir_, uc_, shortname_) \
+ { fw_filename_no_ver(dir_, uc_, shortname_), \
+ 0, 0 }
+
+/* All blobs need to be declared via MODULE_FIRMWARE() */
+#define XE_UC_MODULE_FIRMWARE(platform__, fw_filename) \
+ MODULE_FIRMWARE(fw_filename);
+
+#define XE_UC_FW_ENTRY(platform__, entry__) \
+ { \
+ .platform = XE_ ## platform__, \
+ entry__, \
+ },
+
+XE_GUC_FIRMWARE_DEFS(XE_UC_MODULE_FIRMWARE,
+ fw_filename_mmp_ver, fw_filename_major_ver)
+XE_HUC_FIRMWARE_DEFS(XE_UC_MODULE_FIRMWARE,
+ fw_filename_mmp_ver, fw_filename_no_ver)
+XE_GSC_FIRMWARE_DEFS(XE_UC_MODULE_FIRMWARE, fw_filename_major_ver)
+
+static struct xe_gt *
+__uc_fw_to_gt(struct xe_uc_fw *uc_fw, enum xe_uc_fw_type type)
+{
+ XE_WARN_ON(type >= XE_UC_FW_NUM_TYPES);
+
+ switch (type) {
+ case XE_UC_FW_TYPE_GUC:
+ return container_of(uc_fw, struct xe_gt, uc.guc.fw);
+ case XE_UC_FW_TYPE_HUC:
+ return container_of(uc_fw, struct xe_gt, uc.huc.fw);
+ case XE_UC_FW_TYPE_GSC:
+ return container_of(uc_fw, struct xe_gt, uc.gsc.fw);
+ default:
+ return NULL;
+ }
+}
+
+static struct xe_gt *uc_fw_to_gt(struct xe_uc_fw *uc_fw)
+{
+ return __uc_fw_to_gt(uc_fw, uc_fw->type);
+}
+
+static struct xe_device *uc_fw_to_xe(struct xe_uc_fw *uc_fw)
+{
+ return gt_to_xe(uc_fw_to_gt(uc_fw));
+}
+
+static void
+uc_fw_auto_select(struct xe_device *xe, struct xe_uc_fw *uc_fw)
+{
+ static const struct uc_fw_entry entries_guc[] = {
+ XE_GUC_FIRMWARE_DEFS(XE_UC_FW_ENTRY,
+ uc_fw_entry_mmp_ver,
+ uc_fw_entry_major_ver)
+ };
+ static const struct uc_fw_entry entries_huc[] = {
+ XE_HUC_FIRMWARE_DEFS(XE_UC_FW_ENTRY,
+ uc_fw_entry_mmp_ver,
+ uc_fw_entry_no_ver)
+ };
+ static const struct uc_fw_entry entries_gsc[] = {
+ XE_GSC_FIRMWARE_DEFS(XE_UC_FW_ENTRY, uc_fw_entry_major_ver)
+ };
+ static const struct fw_blobs_by_type blobs_all[XE_UC_FW_NUM_TYPES] = {
+ [XE_UC_FW_TYPE_GUC] = { entries_guc, ARRAY_SIZE(entries_guc) },
+ [XE_UC_FW_TYPE_HUC] = { entries_huc, ARRAY_SIZE(entries_huc) },
+ [XE_UC_FW_TYPE_GSC] = { entries_gsc, ARRAY_SIZE(entries_gsc) },
+ };
+ static const struct uc_fw_entry *entries;
+ enum xe_platform p = xe->info.platform;
+ u32 count;
+ int i;
+
+ xe_assert(xe, uc_fw->type < ARRAY_SIZE(blobs_all));
+ entries = blobs_all[uc_fw->type].entries;
+ count = blobs_all[uc_fw->type].count;
+
+ for (i = 0; i < count && p <= entries[i].platform; i++) {
+ if (p == entries[i].platform) {
+ uc_fw->path = entries[i].path;
+ uc_fw->versions.wanted.major = entries[i].major;
+ uc_fw->versions.wanted.minor = entries[i].minor;
+ uc_fw->full_ver_required = entries[i].full_ver_required;
+
+ if (uc_fw->type == XE_UC_FW_TYPE_GSC)
+ uc_fw->versions.wanted_type = XE_UC_FW_VER_COMPATIBILITY;
+ else
+ uc_fw->versions.wanted_type = XE_UC_FW_VER_RELEASE;
+
+ break;
+ }
+ }
+}
+
+static void
+uc_fw_override(struct xe_uc_fw *uc_fw)
+{
+ char *path_override = NULL;
+
+ /* empty string disables, but it's not allowed for GuC */
+ switch (uc_fw->type) {
+ case XE_UC_FW_TYPE_GUC:
+ if (xe_modparam.guc_firmware_path && *xe_modparam.guc_firmware_path)
+ path_override = xe_modparam.guc_firmware_path;
+ break;
+ case XE_UC_FW_TYPE_HUC:
+ path_override = xe_modparam.huc_firmware_path;
+ break;
+ case XE_UC_FW_TYPE_GSC:
+ path_override = xe_modparam.gsc_firmware_path;
+ break;
+ default:
+ break;
+ }
+
+ if (path_override) {
+ uc_fw->path = path_override;
+ uc_fw->user_overridden = true;
+ }
+}
+
+/**
+ * xe_uc_fw_copy_rsa - copy fw RSA to buffer
+ *
+ * @uc_fw: uC firmware
+ * @dst: dst buffer
+ * @max_len: max number of bytes to copy
+ *
+ * Return: number of copied bytes.
+ */
+size_t xe_uc_fw_copy_rsa(struct xe_uc_fw *uc_fw, void *dst, u32 max_len)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ u32 size = min_t(u32, uc_fw->rsa_size, max_len);
+
+ xe_assert(xe, !(size % 4));
+ xe_assert(xe, xe_uc_fw_is_available(uc_fw));
+
+ xe_map_memcpy_from(xe, dst, &uc_fw->bo->vmap,
+ xe_uc_fw_rsa_offset(uc_fw), size);
+
+ return size;
+}
+
+static void uc_fw_fini(struct drm_device *drm, void *arg)
+{
+ struct xe_uc_fw *uc_fw = arg;
+
+ if (!xe_uc_fw_is_available(uc_fw))
+ return;
+
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_SELECTED);
+}
+
+static void guc_read_css_info(struct xe_uc_fw *uc_fw, struct uc_css_header *css)
+{
+ struct xe_gt *gt = uc_fw_to_gt(uc_fw);
+ struct xe_uc_fw_version *release = &uc_fw->versions.found[XE_UC_FW_VER_RELEASE];
+ struct xe_uc_fw_version *compatibility = &uc_fw->versions.found[XE_UC_FW_VER_COMPATIBILITY];
+
+ xe_gt_assert(gt, uc_fw->type == XE_UC_FW_TYPE_GUC);
+ xe_gt_assert(gt, release->major >= 70);
+
+ if (release->major > 70 || release->minor >= 6) {
+ /* v70.6.0 adds CSS header support */
+ compatibility->major = FIELD_GET(CSS_SW_VERSION_UC_MAJOR,
+ css->submission_version);
+ compatibility->minor = FIELD_GET(CSS_SW_VERSION_UC_MINOR,
+ css->submission_version);
+ compatibility->patch = FIELD_GET(CSS_SW_VERSION_UC_PATCH,
+ css->submission_version);
+ } else if (release->minor >= 3) {
+ /* v70.3.0 introduced v1.1.0 */
+ compatibility->major = 1;
+ compatibility->minor = 1;
+ compatibility->patch = 0;
+ } else {
+ /* v70.0.0 introduced v1.0.0 */
+ compatibility->major = 1;
+ compatibility->minor = 0;
+ compatibility->patch = 0;
+ }
+
+ uc_fw->private_data_size = css->private_data_size;
+}
+
+int xe_uc_fw_check_version_requirements(struct xe_uc_fw *uc_fw)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ struct xe_uc_fw_version *wanted = &uc_fw->versions.wanted;
+ struct xe_uc_fw_version *found = &uc_fw->versions.found[uc_fw->versions.wanted_type];
+
+ /* Driver has no requirement on any version, any is good. */
+ if (!wanted->major)
+ return 0;
+
+ /*
+ * If full version is required, both major and minor should match.
+ * Otherwise, at least the major version.
+ */
+ if (wanted->major != found->major ||
+ (uc_fw->full_ver_required && wanted->minor != found->minor)) {
+ drm_notice(&xe->drm, "%s firmware %s: unexpected version: %u.%u != %u.%u\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path,
+ found->major, found->minor,
+ wanted->major, wanted->minor);
+ goto fail;
+ }
+
+ if (wanted->minor > found->minor) {
+ drm_notice(&xe->drm, "%s firmware (%u.%u) is recommended, but only (%u.%u) was found in %s\n",
+ xe_uc_fw_type_repr(uc_fw->type),
+ wanted->major, wanted->minor,
+ found->major, found->minor,
+ uc_fw->path);
+ drm_info(&xe->drm, "Consider updating your linux-firmware pkg or downloading from %s\n",
+ XE_UC_FIRMWARE_URL);
+ }
+
+ return 0;
+
+fail:
+ if (xe_uc_fw_is_overridden(uc_fw))
+ return 0;
+
+ return -ENOEXEC;
+}
+
+/* Refer to the "CSS-based Firmware Layout" documentation entry for details */
+static int parse_css_header(struct xe_uc_fw *uc_fw, const void *fw_data, size_t fw_size)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ struct xe_uc_fw_version *release = &uc_fw->versions.found[XE_UC_FW_VER_RELEASE];
+ struct uc_css_header *css;
+ size_t size;
+
+ /* Check the size of the blob before examining buffer contents */
+ if (unlikely(fw_size < sizeof(struct uc_css_header))) {
+ drm_warn(&xe->drm, "%s firmware %s: invalid size: %zu < %zu\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path,
+ fw_size, sizeof(struct uc_css_header));
+ return -ENODATA;
+ }
+
+ css = (struct uc_css_header *)fw_data;
+
+ /* Check integrity of size values inside CSS header */
+ size = (css->header_size_dw - css->key_size_dw - css->modulus_size_dw -
+ css->exponent_size_dw) * sizeof(u32);
+ if (unlikely(size != sizeof(struct uc_css_header))) {
+ drm_warn(&xe->drm,
+ "%s firmware %s: unexpected header size: %zu != %zu\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path,
+ fw_size, sizeof(struct uc_css_header));
+ return -EPROTO;
+ }
+
+ /* uCode size must calculated from other sizes */
+ uc_fw->ucode_size = (css->size_dw - css->header_size_dw) * sizeof(u32);
+
+ /* now RSA */
+ uc_fw->rsa_size = css->key_size_dw * sizeof(u32);
+
+ /* At least, it should have header, uCode and RSA. Size of all three. */
+ size = sizeof(struct uc_css_header) + uc_fw->ucode_size +
+ uc_fw->rsa_size;
+ if (unlikely(fw_size < size)) {
+ drm_warn(&xe->drm, "%s firmware %s: invalid size: %zu < %zu\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path,
+ fw_size, size);
+ return -ENOEXEC;
+ }
+
+ /* Get version numbers from the CSS header */
+ release->major = FIELD_GET(CSS_SW_VERSION_UC_MAJOR, css->sw_version);
+ release->minor = FIELD_GET(CSS_SW_VERSION_UC_MINOR, css->sw_version);
+ release->patch = FIELD_GET(CSS_SW_VERSION_UC_PATCH, css->sw_version);
+
+ if (uc_fw->type == XE_UC_FW_TYPE_GUC)
+ guc_read_css_info(uc_fw, css);
+
+ return 0;
+}
+
+static bool is_cpd_header(const void *data)
+{
+ const u32 *marker = data;
+
+ return *marker == GSC_CPD_HEADER_MARKER;
+}
+
+static u32 entry_offset(const struct gsc_cpd_header_v2 *header, const char *name)
+{
+ const struct gsc_cpd_entry *entry;
+ int i;
+
+ entry = (void *)header + header->header_length;
+
+ for (i = 0; i < header->num_of_entries; i++, entry++)
+ if (strcmp(entry->name, name) == 0)
+ return entry->offset & GSC_CPD_ENTRY_OFFSET_MASK;
+
+ return 0;
+}
+
+/* Refer to the "GSC-based Firmware Layout" documentation entry for details */
+static int parse_cpd_header(struct xe_uc_fw *uc_fw, const void *data, size_t size,
+ const char *manifest_entry, const char *css_entry)
+{
+ struct xe_gt *gt = uc_fw_to_gt(uc_fw);
+ struct xe_device *xe = gt_to_xe(gt);
+ const struct gsc_cpd_header_v2 *header = data;
+ struct xe_uc_fw_version *release = &uc_fw->versions.found[XE_UC_FW_VER_RELEASE];
+ const struct gsc_manifest_header *manifest;
+ size_t min_size = sizeof(*header);
+ u32 offset;
+
+ /* manifest_entry is mandatory, css_entry is optional */
+ xe_assert(xe, manifest_entry);
+
+ if (size < min_size || !is_cpd_header(header))
+ return -ENOENT;
+
+ if (header->header_length < sizeof(struct gsc_cpd_header_v2)) {
+ xe_gt_err(gt, "invalid CPD header length %u!\n", header->header_length);
+ return -EINVAL;
+ }
+
+ min_size = header->header_length + sizeof(struct gsc_cpd_entry) * header->num_of_entries;
+ if (size < min_size) {
+ xe_gt_err(gt, "FW too small! %zu < %zu\n", size, min_size);
+ return -ENODATA;
+ }
+
+ /* Look for the manifest first */
+ offset = entry_offset(header, manifest_entry);
+ if (!offset) {
+ xe_gt_err(gt, "Failed to find %s manifest!\n",
+ xe_uc_fw_type_repr(uc_fw->type));
+ return -ENODATA;
+ }
+
+ min_size = offset + sizeof(struct gsc_manifest_header);
+ if (size < min_size) {
+ xe_gt_err(gt, "FW too small! %zu < %zu\n", size, min_size);
+ return -ENODATA;
+ }
+
+ manifest = data + offset;
+
+ release->major = manifest->fw_version.major;
+ release->minor = manifest->fw_version.minor;
+ release->patch = manifest->fw_version.hotfix;
+
+ if (uc_fw->type == XE_UC_FW_TYPE_GSC) {
+ struct xe_gsc *gsc = container_of(uc_fw, struct xe_gsc, fw);
+
+ release->build = manifest->fw_version.build;
+ gsc->security_version = manifest->security_version;
+ }
+
+ /* then optionally look for the css header */
+ if (css_entry) {
+ int ret;
+
+ /*
+ * This section does not contain a CSS entry on DG2. We
+ * don't support DG2 HuC right now, so no need to handle
+ * it, just add a reminder in case that changes.
+ */
+ xe_assert(xe, xe->info.platform != XE_DG2);
+
+ offset = entry_offset(header, css_entry);
+
+ /* the CSS header parser will check that the CSS header fits */
+ if (offset > size) {
+ xe_gt_err(gt, "FW too small! %zu < %u\n", size, offset);
+ return -ENODATA;
+ }
+
+ ret = parse_css_header(uc_fw, data + offset, size - offset);
+ if (ret)
+ return ret;
+
+ uc_fw->css_offset = offset;
+ }
+
+ uc_fw->has_gsc_headers = true;
+
+ return 0;
+}
+
+static int parse_gsc_layout(struct xe_uc_fw *uc_fw, const void *data, size_t size)
+{
+ struct xe_gt *gt = uc_fw_to_gt(uc_fw);
+ const struct gsc_layout_pointers *layout = data;
+ const struct gsc_bpdt_header *bpdt_header = NULL;
+ const struct gsc_bpdt_entry *bpdt_entry = NULL;
+ size_t min_size = sizeof(*layout);
+ int i;
+
+ if (size < min_size) {
+ xe_gt_err(gt, "GSC FW too small! %zu < %zu\n", size, min_size);
+ return -ENODATA;
+ }
+
+ min_size = layout->boot1.offset + layout->boot1.size;
+ if (size < min_size) {
+ xe_gt_err(gt, "GSC FW too small for boot section! %zu < %zu\n",
+ size, min_size);
+ return -ENODATA;
+ }
+
+ min_size = sizeof(*bpdt_header);
+ if (layout->boot1.size < min_size) {
+ xe_gt_err(gt, "GSC FW boot section too small for BPDT header: %u < %zu\n",
+ layout->boot1.size, min_size);
+ return -ENODATA;
+ }
+
+ bpdt_header = data + layout->boot1.offset;
+ if (bpdt_header->signature != GSC_BPDT_HEADER_SIGNATURE) {
+ xe_gt_err(gt, "invalid signature for BPDT header: 0x%08x!\n",
+ bpdt_header->signature);
+ return -EINVAL;
+ }
+
+ min_size += sizeof(*bpdt_entry) * bpdt_header->descriptor_count;
+ if (layout->boot1.size < min_size) {
+ xe_gt_err(gt, "GSC FW boot section too small for BPDT entries: %u < %zu\n",
+ layout->boot1.size, min_size);
+ return -ENODATA;
+ }
+
+ bpdt_entry = (void *)bpdt_header + sizeof(*bpdt_header);
+ for (i = 0; i < bpdt_header->descriptor_count; i++, bpdt_entry++) {
+ if ((bpdt_entry->type & GSC_BPDT_ENTRY_TYPE_MASK) !=
+ GSC_BPDT_ENTRY_TYPE_GSC_RBE)
+ continue;
+
+ min_size = bpdt_entry->sub_partition_offset;
+
+ /* the CPD header parser will check that the CPD header fits */
+ if (layout->boot1.size < min_size) {
+ xe_gt_err(gt, "GSC FW boot section too small for CPD offset: %u < %zu\n",
+ layout->boot1.size, min_size);
+ return -ENODATA;
+ }
+
+ return parse_cpd_header(uc_fw,
+ (void *)bpdt_header + min_size,
+ layout->boot1.size - min_size,
+ "RBEP.man", NULL);
+ }
+
+ xe_gt_err(gt, "couldn't find CPD header in GSC binary!\n");
+ return -ENODATA;
+}
+
+static int parse_headers(struct xe_uc_fw *uc_fw, const struct firmware *fw)
+{
+ int ret;
+
+ /*
+ * All GuC releases and older HuC ones use CSS headers, while newer HuC
+ * releases use GSC CPD headers.
+ */
+ switch (uc_fw->type) {
+ case XE_UC_FW_TYPE_GSC:
+ return parse_gsc_layout(uc_fw, fw->data, fw->size);
+ case XE_UC_FW_TYPE_HUC:
+ ret = parse_cpd_header(uc_fw, fw->data, fw->size, "HUCP.man", "huc_fw");
+ if (!ret || ret != -ENOENT)
+ return ret;
+ fallthrough;
+ case XE_UC_FW_TYPE_GUC:
+ return parse_css_header(uc_fw, fw->data, fw->size);
+ default:
+ return -EINVAL;
+ }
+
+ return 0;
+}
+
+#define print_uc_fw_version(p_, version_, prefix_, ...) \
+do { \
+ struct xe_uc_fw_version *ver_ = (version_); \
+ if (ver_->build) \
+ drm_printf(p_, prefix_ " version %u.%u.%u.%u\n", ##__VA_ARGS__, \
+ ver_->major, ver_->minor, \
+ ver_->patch, ver_->build); \
+ else \
+ drm_printf(p_, prefix_ " version %u.%u.%u\n", ##__VA_ARGS__, \
+ ver_->major, ver_->minor, ver_->patch); \
+} while (0)
+
+static int uc_fw_request(struct xe_uc_fw *uc_fw, const struct firmware **firmware_p)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ struct device *dev = xe->drm.dev;
+ struct drm_printer p = drm_info_printer(dev);
+ const struct firmware *fw = NULL;
+ int err;
+
+ /*
+ * we use FIRMWARE_UNINITIALIZED to detect checks against uc_fw->status
+ * before we're looked at the HW caps to see if we have uc support
+ */
+ BUILD_BUG_ON(XE_UC_FIRMWARE_UNINITIALIZED);
+ xe_assert(xe, !uc_fw->status);
+ xe_assert(xe, !uc_fw->path);
+
+ uc_fw_auto_select(xe, uc_fw);
+ xe_uc_fw_change_status(uc_fw, uc_fw->path ?
+ XE_UC_FIRMWARE_SELECTED :
+ XE_UC_FIRMWARE_NOT_SUPPORTED);
+
+ if (!xe_uc_fw_is_supported(uc_fw))
+ return 0;
+
+ uc_fw_override(uc_fw);
+
+ /* an empty path means the firmware is disabled */
+ if (!xe_device_uc_enabled(xe) || !(*uc_fw->path)) {
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_DISABLED);
+ drm_dbg(&xe->drm, "%s disabled", xe_uc_fw_type_repr(uc_fw->type));
+ return 0;
+ }
+
+ err = request_firmware(&fw, uc_fw->path, dev);
+ if (err)
+ goto fail;
+
+ err = parse_headers(uc_fw, fw);
+ if (err)
+ goto fail;
+
+ print_uc_fw_version(&p,
+ &uc_fw->versions.found[XE_UC_FW_VER_RELEASE],
+ "Using %s firmware from %s",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path);
+
+ /* for GSC FW we want the compatibility version, which we query after load */
+ if (uc_fw->type != XE_UC_FW_TYPE_GSC) {
+ err = xe_uc_fw_check_version_requirements(uc_fw);
+ if (err)
+ goto fail;
+ }
+
+ *firmware_p = fw;
+
+ return 0;
+
+fail:
+ xe_uc_fw_change_status(uc_fw, err == -ENOENT ?
+ XE_UC_FIRMWARE_MISSING :
+ XE_UC_FIRMWARE_ERROR);
+
+ drm_notice(&xe->drm, "%s firmware %s: fetch failed with error %d\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path, err);
+ drm_info(&xe->drm, "%s firmware(s) can be downloaded from %s\n",
+ xe_uc_fw_type_repr(uc_fw->type), XE_UC_FIRMWARE_URL);
+
+ release_firmware(fw); /* OK even if fw is NULL */
+
+ return err;
+}
+
+static void uc_fw_release(const struct firmware *fw)
+{
+ release_firmware(fw);
+}
+
+static int uc_fw_copy(struct xe_uc_fw *uc_fw, const void *data, size_t size, u32 flags)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ struct xe_gt *gt = uc_fw_to_gt(uc_fw);
+ struct xe_tile *tile = gt_to_tile(gt);
+ struct xe_bo *obj;
+ int err;
+
+ obj = xe_managed_bo_create_from_data(xe, tile, data, size, flags);
+ if (IS_ERR(obj)) {
+ drm_notice(&xe->drm, "%s firmware %s: failed to create / populate bo",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path);
+ err = PTR_ERR(obj);
+ goto fail;
+ }
+
+ uc_fw->bo = obj;
+ uc_fw->size = size;
+
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_AVAILABLE);
+
+ err = drmm_add_action_or_reset(&xe->drm, uc_fw_fini, uc_fw);
+ if (err)
+ goto fail;
+
+ return 0;
+
+fail:
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_ERROR);
+ drm_notice(&xe->drm, "%s firmware %s: copy failed with error %d\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path, err);
+
+ return err;
+}
+
+int xe_uc_fw_init(struct xe_uc_fw *uc_fw)
+{
+ const struct firmware *fw = NULL;
+ int err;
+
+ err = uc_fw_request(uc_fw, &fw);
+ if (err)
+ return err;
+
+ /* no error and no firmware means nothing to copy */
+ if (!fw)
+ return 0;
+
+ err = uc_fw_copy(uc_fw, fw->data, fw->size,
+ XE_BO_CREATE_SYSTEM_BIT | XE_BO_CREATE_GGTT_BIT);
+
+ uc_fw_release(fw);
+
+ return err;
+}
+
+static u32 uc_fw_ggtt_offset(struct xe_uc_fw *uc_fw)
+{
+ return xe_bo_ggtt_addr(uc_fw->bo);
+}
+
+static int uc_fw_xfer(struct xe_uc_fw *uc_fw, u32 offset, u32 dma_flags)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ struct xe_gt *gt = uc_fw_to_gt(uc_fw);
+ u32 src_offset, dma_ctrl;
+ int ret;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+
+ /* Set the source address for the uCode */
+ src_offset = uc_fw_ggtt_offset(uc_fw) + uc_fw->css_offset;
+ xe_mmio_write32(gt, DMA_ADDR_0_LOW, lower_32_bits(src_offset));
+ xe_mmio_write32(gt, DMA_ADDR_0_HIGH,
+ upper_32_bits(src_offset) | DMA_ADDRESS_SPACE_GGTT);
+
+ /* Set the DMA destination */
+ xe_mmio_write32(gt, DMA_ADDR_1_LOW, offset);
+ xe_mmio_write32(gt, DMA_ADDR_1_HIGH, DMA_ADDRESS_SPACE_WOPCM);
+
+ /*
+ * Set the transfer size. The header plus uCode will be copied to WOPCM
+ * via DMA, excluding any other components
+ */
+ xe_mmio_write32(gt, DMA_COPY_SIZE,
+ sizeof(struct uc_css_header) + uc_fw->ucode_size);
+
+ /* Start the DMA */
+ xe_mmio_write32(gt, DMA_CTRL,
+ _MASKED_BIT_ENABLE(dma_flags | START_DMA));
+
+ /* Wait for DMA to finish */
+ ret = xe_mmio_wait32(gt, DMA_CTRL, START_DMA, 0, 100000, &dma_ctrl,
+ false);
+ if (ret)
+ drm_err(&xe->drm, "DMA for %s fw failed, DMA_CTRL=%u\n",
+ xe_uc_fw_type_repr(uc_fw->type), dma_ctrl);
+
+ /* Disable the bits once DMA is over */
+ xe_mmio_write32(gt, DMA_CTRL, _MASKED_BIT_DISABLE(dma_flags));
+
+ return ret;
+}
+
+int xe_uc_fw_upload(struct xe_uc_fw *uc_fw, u32 offset, u32 dma_flags)
+{
+ struct xe_device *xe = uc_fw_to_xe(uc_fw);
+ int err;
+
+ /* make sure the status was cleared the last time we reset the uc */
+ xe_assert(xe, !xe_uc_fw_is_loaded(uc_fw));
+
+ if (!xe_uc_fw_is_loadable(uc_fw))
+ return -ENOEXEC;
+
+ /* Call custom loader */
+ err = uc_fw_xfer(uc_fw, offset, dma_flags);
+ if (err)
+ goto fail;
+
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_TRANSFERRED);
+ return 0;
+
+fail:
+ drm_err(&xe->drm, "Failed to load %s firmware %s (%d)\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path,
+ err);
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_LOAD_FAIL);
+ return err;
+}
+
+static const char *version_type_repr(enum xe_uc_fw_version_types type)
+{
+ switch (type) {
+ case XE_UC_FW_VER_RELEASE:
+ return "release";
+ case XE_UC_FW_VER_COMPATIBILITY:
+ return "compatibility";
+ default:
+ return "Unknown version type";
+ }
+}
+
+void xe_uc_fw_print(struct xe_uc_fw *uc_fw, struct drm_printer *p)
+{
+ int i;
+
+ drm_printf(p, "%s firmware: %s\n",
+ xe_uc_fw_type_repr(uc_fw->type), uc_fw->path);
+ drm_printf(p, "\tstatus: %s\n",
+ xe_uc_fw_status_repr(uc_fw->status));
+
+ print_uc_fw_version(p, &uc_fw->versions.wanted, "\twanted %s",
+ version_type_repr(uc_fw->versions.wanted_type));
+
+ for (i = 0; i < XE_UC_FW_VER_TYPE_COUNT; i++) {
+ struct xe_uc_fw_version *ver = &uc_fw->versions.found[i];
+
+ if (ver->major)
+ print_uc_fw_version(p, ver, "\tfound %s",
+ version_type_repr(i));
+ }
+
+ if (uc_fw->ucode_size)
+ drm_printf(p, "\tuCode: %u bytes\n", uc_fw->ucode_size);
+ if (uc_fw->rsa_size)
+ drm_printf(p, "\tRSA: %u bytes\n", uc_fw->rsa_size);
+}
diff --git a/drivers/gpu/drm/xe/xe_uc_fw.h b/drivers/gpu/drm/xe/xe_uc_fw.h
new file mode 100644
index 000000000000..85c20795d1f8
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_fw.h
@@ -0,0 +1,184 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_FW_H_
+#define _XE_UC_FW_H_
+
+#include <linux/errno.h>
+
+#include "xe_macros.h"
+#include "xe_uc_fw_abi.h"
+#include "xe_uc_fw_types.h"
+
+struct drm_printer;
+
+int xe_uc_fw_init(struct xe_uc_fw *uc_fw);
+size_t xe_uc_fw_copy_rsa(struct xe_uc_fw *uc_fw, void *dst, u32 max_len);
+int xe_uc_fw_upload(struct xe_uc_fw *uc_fw, u32 offset, u32 dma_flags);
+int xe_uc_fw_check_version_requirements(struct xe_uc_fw *uc_fw);
+void xe_uc_fw_print(struct xe_uc_fw *uc_fw, struct drm_printer *p);
+
+static inline u32 xe_uc_fw_rsa_offset(struct xe_uc_fw *uc_fw)
+{
+ return sizeof(struct uc_css_header) + uc_fw->ucode_size + uc_fw->css_offset;
+}
+
+static inline void xe_uc_fw_change_status(struct xe_uc_fw *uc_fw,
+ enum xe_uc_fw_status status)
+{
+ uc_fw->__status = status;
+}
+
+static inline
+const char *xe_uc_fw_status_repr(enum xe_uc_fw_status status)
+{
+ switch (status) {
+ case XE_UC_FIRMWARE_NOT_SUPPORTED:
+ return "N/A";
+ case XE_UC_FIRMWARE_UNINITIALIZED:
+ return "UNINITIALIZED";
+ case XE_UC_FIRMWARE_DISABLED:
+ return "DISABLED";
+ case XE_UC_FIRMWARE_SELECTED:
+ return "SELECTED";
+ case XE_UC_FIRMWARE_MISSING:
+ return "MISSING";
+ case XE_UC_FIRMWARE_ERROR:
+ return "ERROR";
+ case XE_UC_FIRMWARE_AVAILABLE:
+ return "AVAILABLE";
+ case XE_UC_FIRMWARE_INIT_FAIL:
+ return "INIT FAIL";
+ case XE_UC_FIRMWARE_LOADABLE:
+ return "LOADABLE";
+ case XE_UC_FIRMWARE_LOAD_FAIL:
+ return "LOAD FAIL";
+ case XE_UC_FIRMWARE_TRANSFERRED:
+ return "TRANSFERRED";
+ case XE_UC_FIRMWARE_RUNNING:
+ return "RUNNING";
+ }
+ return "<invalid>";
+}
+
+static inline int xe_uc_fw_status_to_error(enum xe_uc_fw_status status)
+{
+ switch (status) {
+ case XE_UC_FIRMWARE_NOT_SUPPORTED:
+ return -ENODEV;
+ case XE_UC_FIRMWARE_UNINITIALIZED:
+ return -EACCES;
+ case XE_UC_FIRMWARE_DISABLED:
+ return -EPERM;
+ case XE_UC_FIRMWARE_MISSING:
+ return -ENOENT;
+ case XE_UC_FIRMWARE_ERROR:
+ return -ENOEXEC;
+ case XE_UC_FIRMWARE_INIT_FAIL:
+ case XE_UC_FIRMWARE_LOAD_FAIL:
+ return -EIO;
+ case XE_UC_FIRMWARE_SELECTED:
+ return -ESTALE;
+ case XE_UC_FIRMWARE_AVAILABLE:
+ case XE_UC_FIRMWARE_LOADABLE:
+ case XE_UC_FIRMWARE_TRANSFERRED:
+ case XE_UC_FIRMWARE_RUNNING:
+ return 0;
+ }
+ return -EINVAL;
+}
+
+static inline const char *xe_uc_fw_type_repr(enum xe_uc_fw_type type)
+{
+ switch (type) {
+ case XE_UC_FW_TYPE_GUC:
+ return "GuC";
+ case XE_UC_FW_TYPE_HUC:
+ return "HuC";
+ case XE_UC_FW_TYPE_GSC:
+ return "GSC";
+ default:
+ return "uC";
+ }
+}
+
+static inline enum xe_uc_fw_status
+__xe_uc_fw_status(struct xe_uc_fw *uc_fw)
+{
+ /* shouldn't call this before checking hw/blob availability */
+ XE_WARN_ON(uc_fw->status == XE_UC_FIRMWARE_UNINITIALIZED);
+ return uc_fw->status;
+}
+
+static inline bool xe_uc_fw_is_supported(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) != XE_UC_FIRMWARE_NOT_SUPPORTED;
+}
+
+static inline bool xe_uc_fw_is_enabled(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) > XE_UC_FIRMWARE_DISABLED;
+}
+
+static inline bool xe_uc_fw_is_disabled(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) == XE_UC_FIRMWARE_DISABLED;
+}
+
+static inline bool xe_uc_fw_is_available(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) >= XE_UC_FIRMWARE_AVAILABLE;
+}
+
+static inline bool xe_uc_fw_is_loadable(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) >= XE_UC_FIRMWARE_LOADABLE;
+}
+
+static inline bool xe_uc_fw_is_loaded(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) >= XE_UC_FIRMWARE_TRANSFERRED;
+}
+
+static inline bool xe_uc_fw_is_running(struct xe_uc_fw *uc_fw)
+{
+ return __xe_uc_fw_status(uc_fw) == XE_UC_FIRMWARE_RUNNING;
+}
+
+static inline bool xe_uc_fw_is_overridden(const struct xe_uc_fw *uc_fw)
+{
+ return uc_fw->user_overridden;
+}
+
+static inline void xe_uc_fw_sanitize(struct xe_uc_fw *uc_fw)
+{
+ if (xe_uc_fw_is_loaded(uc_fw))
+ xe_uc_fw_change_status(uc_fw, XE_UC_FIRMWARE_LOADABLE);
+}
+
+static inline u32 __xe_uc_fw_get_upload_size(struct xe_uc_fw *uc_fw)
+{
+ return sizeof(struct uc_css_header) + uc_fw->ucode_size;
+}
+
+/**
+ * xe_uc_fw_get_upload_size() - Get size of firmware needed to be uploaded.
+ * @uc_fw: uC firmware.
+ *
+ * Get the size of the firmware and header that will be uploaded to WOPCM.
+ *
+ * Return: Upload firmware size, or zero on firmware fetch failure.
+ */
+static inline u32 xe_uc_fw_get_upload_size(struct xe_uc_fw *uc_fw)
+{
+ if (!xe_uc_fw_is_available(uc_fw))
+ return 0;
+
+ return __xe_uc_fw_get_upload_size(uc_fw);
+}
+
+#define XE_UC_FIRMWARE_URL "https://git.kernel.org/pub/scm/linux/kernel/git/firmware/linux-firmware.git"
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc_fw_abi.h b/drivers/gpu/drm/xe/xe_uc_fw_abi.h
new file mode 100644
index 000000000000..87ade41209d0
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_fw_abi.h
@@ -0,0 +1,321 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_FW_ABI_H
+#define _XE_UC_FW_ABI_H
+
+#include <linux/build_bug.h>
+#include <linux/types.h>
+
+/**
+ * DOC: CSS-based Firmware Layout
+ *
+ * The CSS-based firmware structure is used for GuC releases on all platforms
+ * and for HuC releases up to DG1. Starting from DG2/MTL the HuC uses the GSC
+ * layout instead.
+ * The CSS firmware layout looks like this::
+ *
+ * +======================================================================+
+ * | Firmware blob |
+ * +===============+===============+============+============+============+
+ * | CSS header | uCode | RSA key | modulus | exponent |
+ * +===============+===============+============+============+============+
+ * <-header size-> <---header size continued ----------->
+ * <--- size ----------------------------------------------------------->
+ * <-key size->
+ * <-mod size->
+ * <-exp size->
+ *
+ * The firmware may or may not have modulus key and exponent data. The header,
+ * uCode and RSA signature are must-have components that will be used by driver.
+ * Length of each components, which is all in dwords, can be found in header.
+ * In the case that modulus and exponent are not present in fw, a.k.a truncated
+ * image, the length value still appears in header.
+ *
+ * Driver will do some basic fw size validation based on the following rules:
+ *
+ * 1. Header, uCode and RSA are must-have components.
+ * 2. All firmware components, if they present, are in the sequence illustrated
+ * in the layout table above.
+ * 3. Length info of each component can be found in header, in dwords.
+ * 4. Modulus and exponent key are not required by driver. They may not appear
+ * in fw. So driver will load a truncated firmware in this case.
+ */
+
+struct uc_css_header {
+ u32 module_type;
+ /*
+ * header_size includes all non-uCode bits, including css_header, rsa
+ * key, modulus key and exponent data.
+ */
+ u32 header_size_dw;
+ u32 header_version;
+ u32 module_id;
+ u32 module_vendor;
+ u32 date;
+#define CSS_DATE_DAY (0xFF << 0)
+#define CSS_DATE_MONTH (0xFF << 8)
+#define CSS_DATE_YEAR (0xFFFF << 16)
+ u32 size_dw; /* uCode plus header_size_dw */
+ u32 key_size_dw;
+ u32 modulus_size_dw;
+ u32 exponent_size_dw;
+ u32 time;
+#define CSS_TIME_HOUR (0xFF << 0)
+#define CSS_DATE_MIN (0xFF << 8)
+#define CSS_DATE_SEC (0xFFFF << 16)
+ char username[8];
+ char buildnumber[12];
+ u32 sw_version;
+#define CSS_SW_VERSION_UC_MAJOR (0xFF << 16)
+#define CSS_SW_VERSION_UC_MINOR (0xFF << 8)
+#define CSS_SW_VERSION_UC_PATCH (0xFF << 0)
+ union {
+ u32 submission_version; /* only applies to GuC */
+ u32 reserved2;
+ };
+ u32 reserved0[12];
+ union {
+ u32 private_data_size; /* only applies to GuC */
+ u32 reserved1;
+ };
+ u32 header_info;
+} __packed;
+static_assert(sizeof(struct uc_css_header) == 128);
+
+/**
+ * DOC: GSC-based Firmware Layout
+ *
+ * The GSC-based firmware structure is used for GSC releases on all platforms
+ * and for HuC releases starting from DG2/MTL. Older HuC releases use the
+ * CSS-based layout instead. Differently from the CSS headers, the GSC headers
+ * uses a directory + entries structure (i.e., there is array of addresses
+ * pointing to specific header extensions identified by a name). Although the
+ * header structures are the same, some of the entries are specific to GSC while
+ * others are specific to HuC. The manifest header entry, which includes basic
+ * information about the binary (like the version) is always present, but it is
+ * named differently based on the binary type.
+ *
+ * The HuC binary starts with a Code Partition Directory (CPD) header. The
+ * entries we're interested in for use in the driver are:
+ *
+ * 1. "HUCP.man": points to the manifest header for the HuC.
+ * 2. "huc_fw": points to the FW code. On platforms that support load via DMA
+ * and 2-step HuC authentication (i.e. MTL+) this is a full CSS-based binary,
+ * while if the GSC is the one doing the load (which only happens on DG2)
+ * this section only contains the uCode.
+ *
+ * The GSC-based HuC firmware layout looks like this::
+ *
+ * +================================================+
+ * | CPD Header |
+ * +================================================+
+ * | CPD entries[] |
+ * | entry1 |
+ * | ... |
+ * | entryX |
+ * | "HUCP.man" |
+ * | ... |
+ * | offset >----------------------------|------o
+ * | ... | |
+ * | entryY | |
+ * | "huc_fw" | |
+ * | ... | |
+ * | offset >----------------------------|----------o
+ * +================================================+ | |
+ * | |
+ * +================================================+ | |
+ * | Manifest Header |<-----o |
+ * | ... | |
+ * | FW version | |
+ * | ... | |
+ * +================================================+ |
+ * |
+ * +================================================+ |
+ * | FW binary |<---------o
+ * | CSS (MTL+ only) |
+ * | uCode |
+ * | RSA Key (MTL+ only) |
+ * | ... |
+ * +================================================+
+ *
+ * The GSC binary starts instead with a layout header, which contains the
+ * locations of the various partitions of the binary. The one we're interested
+ * in is the boot1 partition, where we can find a BPDT header followed by
+ * entries, one of which points to the RBE sub-section of the partition, which
+ * contains the CPD. The GSC blob does not contain a CSS-based binary, so we
+ * only need to look for the manifest, which is under the "RBEP.man" CPD entry.
+ * Note that we have no need to find where the actual FW code is inside the
+ * image because the GSC ROM will itself parse the headers to find it and load
+ * it.
+ * The GSC firmware header layout looks like this::
+ *
+ * +================================================+
+ * | Layout Pointers |
+ * | ... |
+ * | Boot1 offset >---------------------------|------o
+ * | ... | |
+ * +================================================+ |
+ * |
+ * +================================================+ |
+ * | BPDT header |<-----o
+ * +================================================+
+ * | BPDT entries[] |
+ * | entry1 |
+ * | ... |
+ * | entryX |
+ * | type == GSC_RBE |
+ * | offset >-----------------------------|------o
+ * | ... | |
+ * +================================================+ |
+ * |
+ * +================================================+ |
+ * | CPD Header |<-----o
+ * +================================================+
+ * | CPD entries[] |
+ * | entry1 |
+ * | ... |
+ * | entryX |
+ * | "RBEP.man" |
+ * | ... |
+ * | offset >----------------------------|------o
+ * | ... | |
+ * +================================================+ |
+ * |
+ * +================================================+ |
+ * | Manifest Header |<-----o
+ * | ... |
+ * | FW version |
+ * | ... |
+ * | Security version |
+ * | ... |
+ * +================================================+
+ */
+
+struct gsc_version {
+ u16 major;
+ u16 minor;
+ u16 hotfix;
+ u16 build;
+} __packed;
+
+struct gsc_partition {
+ u32 offset;
+ u32 size;
+} __packed;
+
+struct gsc_layout_pointers {
+ u8 rom_bypass_vector[16];
+
+ /* size of this header section, not including ROM bypass vector */
+ u16 size;
+
+ /*
+ * bit0: Backup copy of layout pointers exists
+ * bits1-15: reserved
+ */
+ u8 flags;
+
+ u8 reserved;
+
+ u32 crc32;
+
+ struct gsc_partition datap;
+ struct gsc_partition boot1;
+ struct gsc_partition boot2;
+ struct gsc_partition boot3;
+ struct gsc_partition boot4;
+ struct gsc_partition boot5;
+ struct gsc_partition temp_pages;
+} __packed;
+
+/* Boot partition structures */
+struct gsc_bpdt_header {
+ u32 signature;
+#define GSC_BPDT_HEADER_SIGNATURE 0x000055AA
+
+ u16 descriptor_count; /* num of entries after the header */
+
+ u8 version;
+ u8 configuration;
+
+ u32 crc32;
+
+ u32 build_version;
+ struct gsc_version tool_version;
+} __packed;
+
+struct gsc_bpdt_entry {
+ /*
+ * Bits 0-15: BPDT entry type
+ * Bits 16-17: reserved
+ * Bit 18: code sub-partition
+ * Bits 19-31: reserved
+ */
+ u32 type;
+#define GSC_BPDT_ENTRY_TYPE_MASK GENMASK(15, 0)
+#define GSC_BPDT_ENTRY_TYPE_GSC_RBE 0x1
+
+ u32 sub_partition_offset; /* from the base of the BPDT header */
+ u32 sub_partition_size;
+} __packed;
+
+/* Code partition directory (CPD) structures */
+struct gsc_cpd_header_v2 {
+ u32 header_marker;
+#define GSC_CPD_HEADER_MARKER 0x44504324
+
+ u32 num_of_entries;
+ u8 header_version;
+ u8 entry_version;
+ u8 header_length; /* in bytes */
+ u8 flags;
+ u32 partition_name;
+ u32 crc32;
+} __packed;
+
+struct gsc_cpd_entry {
+ u8 name[12];
+
+ /*
+ * Bits 0-24: offset from the beginning of the code partition
+ * Bit 25: huffman compressed
+ * Bits 26-31: reserved
+ */
+ u32 offset;
+#define GSC_CPD_ENTRY_OFFSET_MASK GENMASK(24, 0)
+#define GSC_CPD_ENTRY_HUFFMAN_COMP BIT(25)
+
+ /*
+ * Module/Item length, in bytes. For Huffman-compressed modules, this
+ * refers to the uncompressed size. For software-compressed modules,
+ * this refers to the compressed size.
+ */
+ u32 length;
+
+ u8 reserved[4];
+} __packed;
+
+struct gsc_manifest_header {
+ u32 header_type; /* 0x4 for manifest type */
+ u32 header_length; /* in dwords */
+ u32 header_version;
+ u32 flags;
+ u32 vendor;
+ u32 date;
+ u32 size; /* In dwords, size of entire manifest (header + extensions) */
+ u32 header_id;
+ u32 internal_data;
+ struct gsc_version fw_version;
+ u32 security_version;
+ struct gsc_version meu_kit_version;
+ u32 meu_manifest_version;
+ u8 general_data[4];
+ u8 reserved3[56];
+ u32 modulus_size; /* in dwords */
+ u32 exponent_size; /* in dwords */
+} __packed;
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc_fw_types.h b/drivers/gpu/drm/xe/xe_uc_fw_types.h
new file mode 100644
index 000000000000..ee914a5d8523
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_fw_types.h
@@ -0,0 +1,146 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_FW_TYPES_H_
+#define _XE_UC_FW_TYPES_H_
+
+#include <linux/types.h>
+
+struct xe_bo;
+
+/*
+ * +------------+---------------------------------------------------+
+ * | PHASE | FIRMWARE STATUS TRANSITIONS |
+ * +============+===================================================+
+ * | | UNINITIALIZED |
+ * +------------+- / | \ -+
+ * | | DISABLED <--/ | \--> NOT_SUPPORTED |
+ * | init_early | V |
+ * | | SELECTED |
+ * +------------+- / | \ -+
+ * | | MISSING <--/ | \--> ERROR |
+ * | fetch | V |
+ * | | AVAILABLE |
+ * +------------+- | \ -+
+ * | | | \--> INIT FAIL |
+ * | init | V |
+ * | | /------> LOADABLE <----<-----------\ |
+ * +------------+- \ / \ \ \ -+
+ * | | LOAD FAIL <--< \--> TRANSFERRED \ |
+ * | upload | \ / \ / |
+ * | | \---------/ \--> RUNNING |
+ * +------------+---------------------------------------------------+
+ */
+
+/*
+ * FIXME: Ported from the i915 and this is state machine is way too complicated.
+ * Circle back and simplify this.
+ */
+enum xe_uc_fw_status {
+ XE_UC_FIRMWARE_NOT_SUPPORTED = -1, /* no uc HW */
+ XE_UC_FIRMWARE_UNINITIALIZED = 0, /* used to catch checks done too early */
+ XE_UC_FIRMWARE_DISABLED, /* disabled */
+ XE_UC_FIRMWARE_SELECTED, /* selected the blob we want to load */
+ XE_UC_FIRMWARE_MISSING, /* blob not found on the system */
+ XE_UC_FIRMWARE_ERROR, /* invalid format or version */
+ XE_UC_FIRMWARE_AVAILABLE, /* blob found and copied in mem */
+ XE_UC_FIRMWARE_INIT_FAIL, /* failed to prepare fw objects for load */
+ XE_UC_FIRMWARE_LOADABLE, /* all fw-required objects are ready */
+ XE_UC_FIRMWARE_LOAD_FAIL, /* failed to xfer or init/auth the fw */
+ XE_UC_FIRMWARE_TRANSFERRED, /* dma xfer done */
+ XE_UC_FIRMWARE_RUNNING /* init/auth done */
+};
+
+enum xe_uc_fw_type {
+ XE_UC_FW_TYPE_GUC = 0,
+ XE_UC_FW_TYPE_HUC,
+ XE_UC_FW_TYPE_GSC,
+ XE_UC_FW_NUM_TYPES
+};
+
+/**
+ * struct xe_uc_fw_version - Version for XE micro controller firmware
+ */
+struct xe_uc_fw_version {
+ /** @major: major version of the FW */
+ u16 major;
+ /** @minor: minor version of the FW */
+ u16 minor;
+ /** @patch: patch version of the FW */
+ u16 patch;
+ /** @build: build version of the FW (not always available) */
+ u16 build;
+};
+
+enum xe_uc_fw_version_types {
+ XE_UC_FW_VER_RELEASE,
+ XE_UC_FW_VER_COMPATIBILITY,
+ XE_UC_FW_VER_TYPE_COUNT
+};
+
+/**
+ * struct xe_uc_fw - XE micro controller firmware
+ */
+struct xe_uc_fw {
+ /** @type: type uC firmware */
+ enum xe_uc_fw_type type;
+ union {
+ /** @status: firmware load status */
+ const enum xe_uc_fw_status status;
+ /**
+ * @__status: private firmware load status - only to be used
+ * by firmware laoding code
+ */
+ enum xe_uc_fw_status __status;
+ };
+ /** @path: path to uC firmware */
+ const char *path;
+ /** @user_overridden: user provided path to uC firmware via modparam */
+ bool user_overridden;
+ /**
+ * @full_ver_required: driver still under development and not ready
+ * for backward-compatible firmware. To be used only for **new**
+ * platforms, i.e. still under require_force_probe protection and not
+ * supported by i915.
+ */
+ bool full_ver_required;
+ /** @size: size of uC firmware including css header */
+ size_t size;
+
+ /** @bo: XE BO for uC firmware */
+ struct xe_bo *bo;
+
+ /** @has_gsc_headers: whether the FW image starts with GSC headers */
+ bool has_gsc_headers;
+
+ /*
+ * The firmware build process will generate a version header file with
+ * major and minor version defined. The versions are built into CSS
+ * header of firmware. The xe kernel driver set the minimal firmware
+ * version required per platform.
+ */
+
+ /** @versions: FW versions wanted and found */
+ struct {
+ /** @wanted: firmware version wanted by platform */
+ struct xe_uc_fw_version wanted;
+ /** @wanted_type: type of firmware version wanted (release vs compatibility) */
+ enum xe_uc_fw_version_types wanted_type;
+ /** @found: fw versions found in firmware blob */
+ struct xe_uc_fw_version found[XE_UC_FW_VER_TYPE_COUNT];
+ } versions;
+
+ /** @rsa_size: RSA size */
+ u32 rsa_size;
+ /** @ucode_size: micro kernel size */
+ u32 ucode_size;
+ /** @css_offset: offset within the blob at which the CSS is located */
+ u32 css_offset;
+
+ /** @private_data_size: size of private data found in uC css header */
+ u32 private_data_size;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_uc_types.h b/drivers/gpu/drm/xe/xe_uc_types.h
new file mode 100644
index 000000000000..9924e4484866
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_uc_types.h
@@ -0,0 +1,28 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_UC_TYPES_H_
+#define _XE_UC_TYPES_H_
+
+#include "xe_gsc_types.h"
+#include "xe_guc_types.h"
+#include "xe_huc_types.h"
+#include "xe_wopcm_types.h"
+
+/**
+ * struct xe_uc - XE micro controllers
+ */
+struct xe_uc {
+ /** @guc: Graphics micro controller */
+ struct xe_guc guc;
+ /** @huc: HuC */
+ struct xe_huc huc;
+ /** @gsc: Graphics Security Controller */
+ struct xe_gsc gsc;
+ /** @wopcm: WOPCM */
+ struct xe_wopcm wopcm;
+};
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
new file mode 100644
index 000000000000..0cfe7289b97e
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -0,0 +1,3209 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#include "xe_vm.h"
+
+#include <linux/dma-fence-array.h>
+#include <linux/nospec.h>
+
+#include <drm/drm_exec.h>
+#include <drm/drm_print.h>
+#include <drm/ttm/ttm_execbuf_util.h>
+#include <drm/ttm/ttm_tt.h>
+#include <drm/xe_drm.h>
+#include <linux/delay.h>
+#include <linux/kthread.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+
+#include "xe_assert.h"
+#include "xe_bo.h"
+#include "xe_device.h"
+#include "xe_drm_client.h"
+#include "xe_exec_queue.h"
+#include "xe_gt.h"
+#include "xe_gt_pagefault.h"
+#include "xe_gt_tlb_invalidation.h"
+#include "xe_migrate.h"
+#include "xe_pat.h"
+#include "xe_pm.h"
+#include "xe_preempt_fence.h"
+#include "xe_pt.h"
+#include "xe_res_cursor.h"
+#include "xe_sync.h"
+#include "xe_trace.h"
+#include "generated/xe_wa_oob.h"
+#include "xe_wa.h"
+
+#define TEST_VM_ASYNC_OPS_ERROR
+
+static struct drm_gem_object *xe_vm_obj(struct xe_vm *vm)
+{
+ return vm->gpuvm.r_obj;
+}
+
+/**
+ * xe_vma_userptr_check_repin() - Advisory check for repin needed
+ * @vma: The userptr vma
+ *
+ * Check if the userptr vma has been invalidated since last successful
+ * repin. The check is advisory only and can the function can be called
+ * without the vm->userptr.notifier_lock held. There is no guarantee that the
+ * vma userptr will remain valid after a lockless check, so typically
+ * the call needs to be followed by a proper check under the notifier_lock.
+ *
+ * Return: 0 if userptr vma is valid, -EAGAIN otherwise; repin recommended.
+ */
+int xe_vma_userptr_check_repin(struct xe_vma *vma)
+{
+ return mmu_interval_check_retry(&vma->userptr.notifier,
+ vma->userptr.notifier_seq) ?
+ -EAGAIN : 0;
+}
+
+int xe_vma_userptr_pin_pages(struct xe_vma *vma)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+ struct xe_device *xe = vm->xe;
+ const unsigned long num_pages = xe_vma_size(vma) >> PAGE_SHIFT;
+ struct page **pages;
+ bool in_kthread = !current->mm;
+ unsigned long notifier_seq;
+ int pinned, ret, i;
+ bool read_only = xe_vma_read_only(vma);
+
+ lockdep_assert_held(&vm->lock);
+ xe_assert(xe, xe_vma_is_userptr(vma));
+retry:
+ if (vma->gpuva.flags & XE_VMA_DESTROYED)
+ return 0;
+
+ notifier_seq = mmu_interval_read_begin(&vma->userptr.notifier);
+ if (notifier_seq == vma->userptr.notifier_seq)
+ return 0;
+
+ pages = kvmalloc_array(num_pages, sizeof(*pages), GFP_KERNEL);
+ if (!pages)
+ return -ENOMEM;
+
+ if (vma->userptr.sg) {
+ dma_unmap_sgtable(xe->drm.dev,
+ vma->userptr.sg,
+ read_only ? DMA_TO_DEVICE :
+ DMA_BIDIRECTIONAL, 0);
+ sg_free_table(vma->userptr.sg);
+ vma->userptr.sg = NULL;
+ }
+
+ pinned = ret = 0;
+ if (in_kthread) {
+ if (!mmget_not_zero(vma->userptr.notifier.mm)) {
+ ret = -EFAULT;
+ goto mm_closed;
+ }
+ kthread_use_mm(vma->userptr.notifier.mm);
+ }
+
+ while (pinned < num_pages) {
+ ret = get_user_pages_fast(xe_vma_userptr(vma) +
+ pinned * PAGE_SIZE,
+ num_pages - pinned,
+ read_only ? 0 : FOLL_WRITE,
+ &pages[pinned]);
+ if (ret < 0) {
+ if (in_kthread)
+ ret = 0;
+ break;
+ }
+
+ pinned += ret;
+ ret = 0;
+ }
+
+ if (in_kthread) {
+ kthread_unuse_mm(vma->userptr.notifier.mm);
+ mmput(vma->userptr.notifier.mm);
+ }
+mm_closed:
+ if (ret)
+ goto out;
+
+ ret = sg_alloc_table_from_pages_segment(&vma->userptr.sgt, pages,
+ pinned, 0,
+ (u64)pinned << PAGE_SHIFT,
+ xe_sg_segment_size(xe->drm.dev),
+ GFP_KERNEL);
+ if (ret) {
+ vma->userptr.sg = NULL;
+ goto out;
+ }
+ vma->userptr.sg = &vma->userptr.sgt;
+
+ ret = dma_map_sgtable(xe->drm.dev, vma->userptr.sg,
+ read_only ? DMA_TO_DEVICE :
+ DMA_BIDIRECTIONAL,
+ DMA_ATTR_SKIP_CPU_SYNC |
+ DMA_ATTR_NO_KERNEL_MAPPING);
+ if (ret) {
+ sg_free_table(vma->userptr.sg);
+ vma->userptr.sg = NULL;
+ goto out;
+ }
+
+ for (i = 0; i < pinned; ++i) {
+ if (!read_only) {
+ lock_page(pages[i]);
+ set_page_dirty(pages[i]);
+ unlock_page(pages[i]);
+ }
+
+ mark_page_accessed(pages[i]);
+ }
+
+out:
+ release_pages(pages, pinned);
+ kvfree(pages);
+
+ if (!(ret < 0)) {
+ vma->userptr.notifier_seq = notifier_seq;
+ if (xe_vma_userptr_check_repin(vma) == -EAGAIN)
+ goto retry;
+ }
+
+ return ret < 0 ? ret : 0;
+}
+
+static bool preempt_fences_waiting(struct xe_vm *vm)
+{
+ struct xe_exec_queue *q;
+
+ lockdep_assert_held(&vm->lock);
+ xe_vm_assert_held(vm);
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
+ if (!q->compute.pfence ||
+ (q->compute.pfence && test_bit(DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT,
+ &q->compute.pfence->flags))) {
+ return true;
+ }
+ }
+
+ return false;
+}
+
+static void free_preempt_fences(struct list_head *list)
+{
+ struct list_head *link, *next;
+
+ list_for_each_safe(link, next, list)
+ xe_preempt_fence_free(to_preempt_fence_from_link(link));
+}
+
+static int alloc_preempt_fences(struct xe_vm *vm, struct list_head *list,
+ unsigned int *count)
+{
+ lockdep_assert_held(&vm->lock);
+ xe_vm_assert_held(vm);
+
+ if (*count >= vm->preempt.num_exec_queues)
+ return 0;
+
+ for (; *count < vm->preempt.num_exec_queues; ++(*count)) {
+ struct xe_preempt_fence *pfence = xe_preempt_fence_alloc();
+
+ if (IS_ERR(pfence))
+ return PTR_ERR(pfence);
+
+ list_move_tail(xe_preempt_fence_link(pfence), list);
+ }
+
+ return 0;
+}
+
+static int wait_for_existing_preempt_fences(struct xe_vm *vm)
+{
+ struct xe_exec_queue *q;
+
+ xe_vm_assert_held(vm);
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
+ if (q->compute.pfence) {
+ long timeout = dma_fence_wait(q->compute.pfence, false);
+
+ if (timeout < 0)
+ return -ETIME;
+ dma_fence_put(q->compute.pfence);
+ q->compute.pfence = NULL;
+ }
+ }
+
+ return 0;
+}
+
+static bool xe_vm_is_idle(struct xe_vm *vm)
+{
+ struct xe_exec_queue *q;
+
+ xe_vm_assert_held(vm);
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
+ if (!xe_exec_queue_is_idle(q))
+ return false;
+ }
+
+ return true;
+}
+
+static void arm_preempt_fences(struct xe_vm *vm, struct list_head *list)
+{
+ struct list_head *link;
+ struct xe_exec_queue *q;
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
+ struct dma_fence *fence;
+
+ link = list->next;
+ xe_assert(vm->xe, link != list);
+
+ fence = xe_preempt_fence_arm(to_preempt_fence_from_link(link),
+ q, q->compute.context,
+ ++q->compute.seqno);
+ dma_fence_put(q->compute.pfence);
+ q->compute.pfence = fence;
+ }
+}
+
+static int add_preempt_fences(struct xe_vm *vm, struct xe_bo *bo)
+{
+ struct xe_exec_queue *q;
+ int err;
+
+ if (!vm->preempt.num_exec_queues)
+ return 0;
+
+ err = xe_bo_lock(bo, true);
+ if (err)
+ return err;
+
+ err = dma_resv_reserve_fences(bo->ttm.base.resv, vm->preempt.num_exec_queues);
+ if (err)
+ goto out_unlock;
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link)
+ if (q->compute.pfence) {
+ dma_resv_add_fence(bo->ttm.base.resv,
+ q->compute.pfence,
+ DMA_RESV_USAGE_BOOKKEEP);
+ }
+
+out_unlock:
+ xe_bo_unlock(bo);
+ return err;
+}
+
+static void resume_and_reinstall_preempt_fences(struct xe_vm *vm,
+ struct drm_exec *exec)
+{
+ struct xe_exec_queue *q;
+
+ lockdep_assert_held(&vm->lock);
+ xe_vm_assert_held(vm);
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link) {
+ q->ops->resume(q);
+
+ drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, q->compute.pfence,
+ DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
+ }
+}
+
+int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
+{
+ struct drm_gpuvm_exec vm_exec = {
+ .vm = &vm->gpuvm,
+ .flags = DRM_EXEC_INTERRUPTIBLE_WAIT,
+ .num_fences = 1,
+ };
+ struct drm_exec *exec = &vm_exec.exec;
+ struct dma_fence *pfence;
+ int err;
+ bool wait;
+
+ xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
+
+ down_write(&vm->lock);
+ err = drm_gpuvm_exec_lock(&vm_exec);
+ if (err)
+ return err;
+
+ pfence = xe_preempt_fence_create(q, q->compute.context,
+ ++q->compute.seqno);
+ if (!pfence) {
+ err = -ENOMEM;
+ goto out_unlock;
+ }
+
+ list_add(&q->compute.link, &vm->preempt.exec_queues);
+ ++vm->preempt.num_exec_queues;
+ q->compute.pfence = pfence;
+
+ down_read(&vm->userptr.notifier_lock);
+
+ drm_gpuvm_resv_add_fence(&vm->gpuvm, exec, pfence,
+ DMA_RESV_USAGE_BOOKKEEP, DMA_RESV_USAGE_BOOKKEEP);
+
+ /*
+ * Check to see if a preemption on VM is in flight or userptr
+ * invalidation, if so trigger this preempt fence to sync state with
+ * other preempt fences on the VM.
+ */
+ wait = __xe_vm_userptr_needs_repin(vm) || preempt_fences_waiting(vm);
+ if (wait)
+ dma_fence_enable_sw_signaling(pfence);
+
+ up_read(&vm->userptr.notifier_lock);
+
+out_unlock:
+ drm_exec_fini(exec);
+ up_write(&vm->lock);
+
+ return err;
+}
+
+/**
+ * xe_vm_remove_compute_exec_queue() - Remove compute exec queue from VM
+ * @vm: The VM.
+ * @q: The exec_queue
+ */
+void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
+{
+ if (!xe_vm_in_preempt_fence_mode(vm))
+ return;
+
+ down_write(&vm->lock);
+ list_del(&q->compute.link);
+ --vm->preempt.num_exec_queues;
+ if (q->compute.pfence) {
+ dma_fence_enable_sw_signaling(q->compute.pfence);
+ dma_fence_put(q->compute.pfence);
+ q->compute.pfence = NULL;
+ }
+ up_write(&vm->lock);
+}
+
+/**
+ * __xe_vm_userptr_needs_repin() - Check whether the VM does have userptrs
+ * that need repinning.
+ * @vm: The VM.
+ *
+ * This function checks for whether the VM has userptrs that need repinning,
+ * and provides a release-type barrier on the userptr.notifier_lock after
+ * checking.
+ *
+ * Return: 0 if there are no userptrs needing repinning, -EAGAIN if there are.
+ */
+int __xe_vm_userptr_needs_repin(struct xe_vm *vm)
+{
+ lockdep_assert_held_read(&vm->userptr.notifier_lock);
+
+ return (list_empty(&vm->userptr.repin_list) &&
+ list_empty(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
+}
+
+#define XE_VM_REBIND_RETRY_TIMEOUT_MS 1000
+
+static void xe_vm_kill(struct xe_vm *vm)
+{
+ struct xe_exec_queue *q;
+
+ lockdep_assert_held(&vm->lock);
+
+ xe_vm_lock(vm, false);
+ vm->flags |= XE_VM_FLAG_BANNED;
+ trace_xe_vm_kill(vm);
+
+ list_for_each_entry(q, &vm->preempt.exec_queues, compute.link)
+ q->ops->kill(q);
+ xe_vm_unlock(vm);
+
+ /* TODO: Inform user the VM is banned */
+}
+
+/**
+ * xe_vm_validate_should_retry() - Whether to retry after a validate error.
+ * @exec: The drm_exec object used for locking before validation.
+ * @err: The error returned from ttm_bo_validate().
+ * @end: A ktime_t cookie that should be set to 0 before first use and
+ * that should be reused on subsequent calls.
+ *
+ * With multiple active VMs, under memory pressure, it is possible that
+ * ttm_bo_validate() run into -EDEADLK and in such case returns -ENOMEM.
+ * Until ttm properly handles locking in such scenarios, best thing the
+ * driver can do is retry with a timeout. Check if that is necessary, and
+ * if so unlock the drm_exec's objects while keeping the ticket to prepare
+ * for a rerun.
+ *
+ * Return: true if a retry after drm_exec_init() is recommended;
+ * false otherwise.
+ */
+bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end)
+{
+ ktime_t cur;
+
+ if (err != -ENOMEM)
+ return false;
+
+ cur = ktime_get();
+ *end = *end ? : ktime_add_ms(cur, XE_VM_REBIND_RETRY_TIMEOUT_MS);
+ if (!ktime_before(cur, *end))
+ return false;
+
+ msleep(20);
+ return true;
+}
+
+static int xe_gpuvm_validate(struct drm_gpuvm_bo *vm_bo, struct drm_exec *exec)
+{
+ struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
+ struct drm_gpuva *gpuva;
+ int ret;
+
+ lockdep_assert_held(&vm->lock);
+ drm_gpuvm_bo_for_each_va(gpuva, vm_bo)
+ list_move_tail(&gpuva_to_vma(gpuva)->combined_links.rebind,
+ &vm->rebind_list);
+
+ ret = xe_bo_validate(gem_to_xe_bo(vm_bo->obj), vm, false);
+ if (ret)
+ return ret;
+
+ vm_bo->evicted = false;
+ return 0;
+}
+
+static int xe_preempt_work_begin(struct drm_exec *exec, struct xe_vm *vm,
+ bool *done)
+{
+ int err;
+
+ /*
+ * 1 fence for each preempt fence plus a fence for each tile from a
+ * possible rebind
+ */
+ err = drm_gpuvm_prepare_vm(&vm->gpuvm, exec, vm->preempt.num_exec_queues +
+ vm->xe->info.tile_count);
+ if (err)
+ return err;
+
+ if (xe_vm_is_idle(vm)) {
+ vm->preempt.rebind_deactivated = true;
+ *done = true;
+ return 0;
+ }
+
+ if (!preempt_fences_waiting(vm)) {
+ *done = true;
+ return 0;
+ }
+
+ err = drm_gpuvm_prepare_objects(&vm->gpuvm, exec, vm->preempt.num_exec_queues);
+ if (err)
+ return err;
+
+ err = wait_for_existing_preempt_fences(vm);
+ if (err)
+ return err;
+
+ return drm_gpuvm_validate(&vm->gpuvm, exec);
+}
+
+static void preempt_rebind_work_func(struct work_struct *w)
+{
+ struct xe_vm *vm = container_of(w, struct xe_vm, preempt.rebind_work);
+ struct drm_exec exec;
+ struct dma_fence *rebind_fence;
+ unsigned int fence_count = 0;
+ LIST_HEAD(preempt_fences);
+ ktime_t end = 0;
+ int err = 0;
+ long wait;
+ int __maybe_unused tries = 0;
+
+ xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
+ trace_xe_vm_rebind_worker_enter(vm);
+
+ down_write(&vm->lock);
+
+ if (xe_vm_is_closed_or_banned(vm)) {
+ up_write(&vm->lock);
+ trace_xe_vm_rebind_worker_exit(vm);
+ return;
+ }
+
+retry:
+ if (xe_vm_userptr_check_repin(vm)) {
+ err = xe_vm_userptr_pin(vm);
+ if (err)
+ goto out_unlock_outer;
+ }
+
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+
+ drm_exec_until_all_locked(&exec) {
+ bool done = false;
+
+ err = xe_preempt_work_begin(&exec, vm, &done);
+ drm_exec_retry_on_contention(&exec);
+ if (err || done) {
+ drm_exec_fini(&exec);
+ if (err && xe_vm_validate_should_retry(&exec, err, &end))
+ err = -EAGAIN;
+
+ goto out_unlock_outer;
+ }
+ }
+
+ err = alloc_preempt_fences(vm, &preempt_fences, &fence_count);
+ if (err)
+ goto out_unlock;
+
+ rebind_fence = xe_vm_rebind(vm, true);
+ if (IS_ERR(rebind_fence)) {
+ err = PTR_ERR(rebind_fence);
+ goto out_unlock;
+ }
+
+ if (rebind_fence) {
+ dma_fence_wait(rebind_fence, false);
+ dma_fence_put(rebind_fence);
+ }
+
+ /* Wait on munmap style VM unbinds */
+ wait = dma_resv_wait_timeout(xe_vm_resv(vm),
+ DMA_RESV_USAGE_KERNEL,
+ false, MAX_SCHEDULE_TIMEOUT);
+ if (wait <= 0) {
+ err = -ETIME;
+ goto out_unlock;
+ }
+
+#define retry_required(__tries, __vm) \
+ (IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT) ? \
+ (!(__tries)++ || __xe_vm_userptr_needs_repin(__vm)) : \
+ __xe_vm_userptr_needs_repin(__vm))
+
+ down_read(&vm->userptr.notifier_lock);
+ if (retry_required(tries, vm)) {
+ up_read(&vm->userptr.notifier_lock);
+ err = -EAGAIN;
+ goto out_unlock;
+ }
+
+#undef retry_required
+
+ spin_lock(&vm->xe->ttm.lru_lock);
+ ttm_lru_bulk_move_tail(&vm->lru_bulk_move);
+ spin_unlock(&vm->xe->ttm.lru_lock);
+
+ /* Point of no return. */
+ arm_preempt_fences(vm, &preempt_fences);
+ resume_and_reinstall_preempt_fences(vm, &exec);
+ up_read(&vm->userptr.notifier_lock);
+
+out_unlock:
+ drm_exec_fini(&exec);
+out_unlock_outer:
+ if (err == -EAGAIN) {
+ trace_xe_vm_rebind_worker_retry(vm);
+ goto retry;
+ }
+
+ if (err) {
+ drm_warn(&vm->xe->drm, "VM worker error: %d\n", err);
+ xe_vm_kill(vm);
+ }
+ up_write(&vm->lock);
+
+ free_preempt_fences(&preempt_fences);
+
+ trace_xe_vm_rebind_worker_exit(vm);
+}
+
+static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
+ const struct mmu_notifier_range *range,
+ unsigned long cur_seq)
+{
+ struct xe_vma *vma = container_of(mni, struct xe_vma, userptr.notifier);
+ struct xe_vm *vm = xe_vma_vm(vma);
+ struct dma_resv_iter cursor;
+ struct dma_fence *fence;
+ long err;
+
+ xe_assert(vm->xe, xe_vma_is_userptr(vma));
+ trace_xe_vma_userptr_invalidate(vma);
+
+ if (!mmu_notifier_range_blockable(range))
+ return false;
+
+ down_write(&vm->userptr.notifier_lock);
+ mmu_interval_set_seq(mni, cur_seq);
+
+ /* No need to stop gpu access if the userptr is not yet bound. */
+ if (!vma->userptr.initial_bind) {
+ up_write(&vm->userptr.notifier_lock);
+ return true;
+ }
+
+ /*
+ * Tell exec and rebind worker they need to repin and rebind this
+ * userptr.
+ */
+ if (!xe_vm_in_fault_mode(vm) &&
+ !(vma->gpuva.flags & XE_VMA_DESTROYED) && vma->tile_present) {
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_move_tail(&vma->userptr.invalidate_link,
+ &vm->userptr.invalidated);
+ spin_unlock(&vm->userptr.invalidated_lock);
+ }
+
+ up_write(&vm->userptr.notifier_lock);
+
+ /*
+ * Preempt fences turn into schedule disables, pipeline these.
+ * Note that even in fault mode, we need to wait for binds and
+ * unbinds to complete, and those are attached as BOOKMARK fences
+ * to the vm.
+ */
+ dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
+ DMA_RESV_USAGE_BOOKKEEP);
+ dma_resv_for_each_fence_unlocked(&cursor, fence)
+ dma_fence_enable_sw_signaling(fence);
+ dma_resv_iter_end(&cursor);
+
+ err = dma_resv_wait_timeout(xe_vm_resv(vm),
+ DMA_RESV_USAGE_BOOKKEEP,
+ false, MAX_SCHEDULE_TIMEOUT);
+ XE_WARN_ON(err <= 0);
+
+ if (xe_vm_in_fault_mode(vm)) {
+ err = xe_vm_invalidate_vma(vma);
+ XE_WARN_ON(err);
+ }
+
+ trace_xe_vma_userptr_invalidate_complete(vma);
+
+ return true;
+}
+
+static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = {
+ .invalidate = vma_userptr_invalidate,
+};
+
+int xe_vm_userptr_pin(struct xe_vm *vm)
+{
+ struct xe_vma *vma, *next;
+ int err = 0;
+ LIST_HEAD(tmp_evict);
+
+ lockdep_assert_held_write(&vm->lock);
+
+ /* Collect invalidated userptrs */
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_for_each_entry_safe(vma, next, &vm->userptr.invalidated,
+ userptr.invalidate_link) {
+ list_del_init(&vma->userptr.invalidate_link);
+ list_move_tail(&vma->combined_links.userptr,
+ &vm->userptr.repin_list);
+ }
+ spin_unlock(&vm->userptr.invalidated_lock);
+
+ /* Pin and move to temporary list */
+ list_for_each_entry_safe(vma, next, &vm->userptr.repin_list,
+ combined_links.userptr) {
+ err = xe_vma_userptr_pin_pages(vma);
+ if (err < 0)
+ return err;
+
+ list_move_tail(&vma->combined_links.userptr, &vm->rebind_list);
+ }
+
+ return 0;
+}
+
+/**
+ * xe_vm_userptr_check_repin() - Check whether the VM might have userptrs
+ * that need repinning.
+ * @vm: The VM.
+ *
+ * This function does an advisory check for whether the VM has userptrs that
+ * need repinning.
+ *
+ * Return: 0 if there are no indications of userptrs needing repinning,
+ * -EAGAIN if there are.
+ */
+int xe_vm_userptr_check_repin(struct xe_vm *vm)
+{
+ return (list_empty_careful(&vm->userptr.repin_list) &&
+ list_empty_careful(&vm->userptr.invalidated)) ? 0 : -EAGAIN;
+}
+
+static struct dma_fence *
+xe_vm_bind_vma(struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool first_op, bool last_op);
+
+struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker)
+{
+ struct dma_fence *fence = NULL;
+ struct xe_vma *vma, *next;
+
+ lockdep_assert_held(&vm->lock);
+ if (xe_vm_in_lr_mode(vm) && !rebind_worker)
+ return NULL;
+
+ xe_vm_assert_held(vm);
+ list_for_each_entry_safe(vma, next, &vm->rebind_list,
+ combined_links.rebind) {
+ xe_assert(vm->xe, vma->tile_present);
+
+ list_del_init(&vma->combined_links.rebind);
+ dma_fence_put(fence);
+ if (rebind_worker)
+ trace_xe_vma_rebind_worker(vma);
+ else
+ trace_xe_vma_rebind_exec(vma);
+ fence = xe_vm_bind_vma(vma, NULL, NULL, 0, false, false);
+ if (IS_ERR(fence))
+ return fence;
+ }
+
+ return fence;
+}
+
+#define VMA_CREATE_FLAG_READ_ONLY BIT(0)
+#define VMA_CREATE_FLAG_IS_NULL BIT(1)
+
+static struct xe_vma *xe_vma_create(struct xe_vm *vm,
+ struct xe_bo *bo,
+ u64 bo_offset_or_userptr,
+ u64 start, u64 end,
+ u16 pat_index, unsigned int flags)
+{
+ struct xe_vma *vma;
+ struct xe_tile *tile;
+ u8 id;
+ bool read_only = (flags & VMA_CREATE_FLAG_READ_ONLY);
+ bool is_null = (flags & VMA_CREATE_FLAG_IS_NULL);
+
+ xe_assert(vm->xe, start < end);
+ xe_assert(vm->xe, end < vm->size);
+
+ if (!bo && !is_null) /* userptr */
+ vma = kzalloc(sizeof(*vma), GFP_KERNEL);
+ else
+ vma = kzalloc(sizeof(*vma) - sizeof(struct xe_userptr),
+ GFP_KERNEL);
+ if (!vma) {
+ vma = ERR_PTR(-ENOMEM);
+ return vma;
+ }
+
+ INIT_LIST_HEAD(&vma->combined_links.rebind);
+
+ INIT_LIST_HEAD(&vma->gpuva.gem.entry);
+ vma->gpuva.vm = &vm->gpuvm;
+ vma->gpuva.va.addr = start;
+ vma->gpuva.va.range = end - start + 1;
+ if (read_only)
+ vma->gpuva.flags |= XE_VMA_READ_ONLY;
+ if (is_null)
+ vma->gpuva.flags |= DRM_GPUVA_SPARSE;
+
+ for_each_tile(tile, vm->xe, id)
+ vma->tile_mask |= 0x1 << id;
+
+ if (GRAPHICS_VER(vm->xe) >= 20 || vm->xe->info.platform == XE_PVC)
+ vma->gpuva.flags |= XE_VMA_ATOMIC_PTE_BIT;
+
+ vma->pat_index = pat_index;
+
+ if (bo) {
+ struct drm_gpuvm_bo *vm_bo;
+
+ xe_bo_assert_held(bo);
+
+ vm_bo = drm_gpuvm_bo_obtain(vma->gpuva.vm, &bo->ttm.base);
+ if (IS_ERR(vm_bo)) {
+ kfree(vma);
+ return ERR_CAST(vm_bo);
+ }
+
+ drm_gpuvm_bo_extobj_add(vm_bo);
+ drm_gem_object_get(&bo->ttm.base);
+ vma->gpuva.gem.obj = &bo->ttm.base;
+ vma->gpuva.gem.offset = bo_offset_or_userptr;
+ drm_gpuva_link(&vma->gpuva, vm_bo);
+ drm_gpuvm_bo_put(vm_bo);
+ } else /* userptr or null */ {
+ if (!is_null) {
+ u64 size = end - start + 1;
+ int err;
+
+ INIT_LIST_HEAD(&vma->userptr.invalidate_link);
+ vma->gpuva.gem.offset = bo_offset_or_userptr;
+
+ err = mmu_interval_notifier_insert(&vma->userptr.notifier,
+ current->mm,
+ xe_vma_userptr(vma), size,
+ &vma_userptr_notifier_ops);
+ if (err) {
+ kfree(vma);
+ vma = ERR_PTR(err);
+ return vma;
+ }
+
+ vma->userptr.notifier_seq = LONG_MAX;
+ }
+
+ xe_vm_get(vm);
+ }
+
+ return vma;
+}
+
+static void xe_vma_destroy_late(struct xe_vma *vma)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+ struct xe_device *xe = vm->xe;
+ bool read_only = xe_vma_read_only(vma);
+
+ if (xe_vma_is_userptr(vma)) {
+ if (vma->userptr.sg) {
+ dma_unmap_sgtable(xe->drm.dev,
+ vma->userptr.sg,
+ read_only ? DMA_TO_DEVICE :
+ DMA_BIDIRECTIONAL, 0);
+ sg_free_table(vma->userptr.sg);
+ vma->userptr.sg = NULL;
+ }
+
+ /*
+ * Since userptr pages are not pinned, we can't remove
+ * the notifer until we're sure the GPU is not accessing
+ * them anymore
+ */
+ mmu_interval_notifier_remove(&vma->userptr.notifier);
+ xe_vm_put(vm);
+ } else if (xe_vma_is_null(vma)) {
+ xe_vm_put(vm);
+ } else {
+ xe_bo_put(xe_vma_bo(vma));
+ }
+
+ kfree(vma);
+}
+
+static void vma_destroy_work_func(struct work_struct *w)
+{
+ struct xe_vma *vma =
+ container_of(w, struct xe_vma, destroy_work);
+
+ xe_vma_destroy_late(vma);
+}
+
+static void vma_destroy_cb(struct dma_fence *fence,
+ struct dma_fence_cb *cb)
+{
+ struct xe_vma *vma = container_of(cb, struct xe_vma, destroy_cb);
+
+ INIT_WORK(&vma->destroy_work, vma_destroy_work_func);
+ queue_work(system_unbound_wq, &vma->destroy_work);
+}
+
+static void xe_vma_destroy(struct xe_vma *vma, struct dma_fence *fence)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+
+ lockdep_assert_held_write(&vm->lock);
+ xe_assert(vm->xe, list_empty(&vma->combined_links.destroy));
+
+ if (xe_vma_is_userptr(vma)) {
+ xe_assert(vm->xe, vma->gpuva.flags & XE_VMA_DESTROYED);
+
+ spin_lock(&vm->userptr.invalidated_lock);
+ list_del(&vma->userptr.invalidate_link);
+ spin_unlock(&vm->userptr.invalidated_lock);
+ } else if (!xe_vma_is_null(vma)) {
+ xe_bo_assert_held(xe_vma_bo(vma));
+
+ drm_gpuva_unlink(&vma->gpuva);
+ }
+
+ xe_vm_assert_held(vm);
+ if (fence) {
+ int ret = dma_fence_add_callback(fence, &vma->destroy_cb,
+ vma_destroy_cb);
+
+ if (ret) {
+ XE_WARN_ON(ret != -ENOENT);
+ xe_vma_destroy_late(vma);
+ }
+ } else {
+ xe_vma_destroy_late(vma);
+ }
+}
+
+/**
+ * xe_vm_prepare_vma() - drm_exec utility to lock a vma
+ * @exec: The drm_exec object we're currently locking for.
+ * @vma: The vma for witch we want to lock the vm resv and any attached
+ * object's resv.
+ * @num_shared: The number of dma-fence slots to pre-allocate in the
+ * objects' reservation objects.
+ *
+ * Return: 0 on success, negative error code on error. In particular
+ * may return -EDEADLK on WW transaction contention and -EINTR if
+ * an interruptible wait is terminated by a signal.
+ */
+int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
+ unsigned int num_shared)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+ struct xe_bo *bo = xe_vma_bo(vma);
+ int err;
+
+ XE_WARN_ON(!vm);
+ err = drm_exec_prepare_obj(exec, xe_vm_obj(vm), num_shared);
+ if (!err && bo && !bo->vm)
+ err = drm_exec_prepare_obj(exec, &bo->ttm.base, num_shared);
+
+ return err;
+}
+
+static void xe_vma_destroy_unlocked(struct xe_vma *vma)
+{
+ struct drm_exec exec;
+ int err;
+
+ drm_exec_init(&exec, 0, 0);
+ drm_exec_until_all_locked(&exec) {
+ err = xe_vm_prepare_vma(&exec, vma, 0);
+ drm_exec_retry_on_contention(&exec);
+ if (XE_WARN_ON(err))
+ break;
+ }
+
+ xe_vma_destroy(vma, NULL);
+
+ drm_exec_fini(&exec);
+}
+
+struct xe_vma *
+xe_vm_find_overlapping_vma(struct xe_vm *vm, u64 start, u64 range)
+{
+ struct drm_gpuva *gpuva;
+
+ lockdep_assert_held(&vm->lock);
+
+ if (xe_vm_is_closed_or_banned(vm))
+ return NULL;
+
+ xe_assert(vm->xe, start + range <= vm->size);
+
+ gpuva = drm_gpuva_find_first(&vm->gpuvm, start, range);
+
+ return gpuva ? gpuva_to_vma(gpuva) : NULL;
+}
+
+static int xe_vm_insert_vma(struct xe_vm *vm, struct xe_vma *vma)
+{
+ int err;
+
+ xe_assert(vm->xe, xe_vma_vm(vma) == vm);
+ lockdep_assert_held(&vm->lock);
+
+ err = drm_gpuva_insert(&vm->gpuvm, &vma->gpuva);
+ XE_WARN_ON(err); /* Shouldn't be possible */
+
+ return err;
+}
+
+static void xe_vm_remove_vma(struct xe_vm *vm, struct xe_vma *vma)
+{
+ xe_assert(vm->xe, xe_vma_vm(vma) == vm);
+ lockdep_assert_held(&vm->lock);
+
+ drm_gpuva_remove(&vma->gpuva);
+ if (vm->usm.last_fault_vma == vma)
+ vm->usm.last_fault_vma = NULL;
+}
+
+static struct drm_gpuva_op *xe_vm_op_alloc(void)
+{
+ struct xe_vma_op *op;
+
+ op = kzalloc(sizeof(*op), GFP_KERNEL);
+
+ if (unlikely(!op))
+ return NULL;
+
+ return &op->base;
+}
+
+static void xe_vm_free(struct drm_gpuvm *gpuvm);
+
+static struct drm_gpuvm_ops gpuvm_ops = {
+ .op_alloc = xe_vm_op_alloc,
+ .vm_bo_validate = xe_gpuvm_validate,
+ .vm_free = xe_vm_free,
+};
+
+static u64 pde_encode_pat_index(struct xe_device *xe, u16 pat_index)
+{
+ u64 pte = 0;
+
+ if (pat_index & BIT(0))
+ pte |= XE_PPGTT_PTE_PAT0;
+
+ if (pat_index & BIT(1))
+ pte |= XE_PPGTT_PTE_PAT1;
+
+ return pte;
+}
+
+static u64 pte_encode_pat_index(struct xe_device *xe, u16 pat_index,
+ u32 pt_level)
+{
+ u64 pte = 0;
+
+ if (pat_index & BIT(0))
+ pte |= XE_PPGTT_PTE_PAT0;
+
+ if (pat_index & BIT(1))
+ pte |= XE_PPGTT_PTE_PAT1;
+
+ if (pat_index & BIT(2)) {
+ if (pt_level)
+ pte |= XE_PPGTT_PDE_PDPE_PAT2;
+ else
+ pte |= XE_PPGTT_PTE_PAT2;
+ }
+
+ if (pat_index & BIT(3))
+ pte |= XELPG_PPGTT_PTE_PAT3;
+
+ if (pat_index & (BIT(4)))
+ pte |= XE2_PPGTT_PTE_PAT4;
+
+ return pte;
+}
+
+static u64 pte_encode_ps(u32 pt_level)
+{
+ XE_WARN_ON(pt_level > MAX_HUGEPTE_LEVEL);
+
+ if (pt_level == 1)
+ return XE_PDE_PS_2M;
+ else if (pt_level == 2)
+ return XE_PDPE_PS_1G;
+
+ return 0;
+}
+
+static u64 xelp_pde_encode_bo(struct xe_bo *bo, u64 bo_offset,
+ const u16 pat_index)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ u64 pde;
+
+ pde = xe_bo_addr(bo, bo_offset, XE_PAGE_SIZE);
+ pde |= XE_PAGE_PRESENT | XE_PAGE_RW;
+ pde |= pde_encode_pat_index(xe, pat_index);
+
+ return pde;
+}
+
+static u64 xelp_pte_encode_bo(struct xe_bo *bo, u64 bo_offset,
+ u16 pat_index, u32 pt_level)
+{
+ struct xe_device *xe = xe_bo_device(bo);
+ u64 pte;
+
+ pte = xe_bo_addr(bo, bo_offset, XE_PAGE_SIZE);
+ pte |= XE_PAGE_PRESENT | XE_PAGE_RW;
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
+ pte |= pte_encode_ps(pt_level);
+
+ if (xe_bo_is_vram(bo) || xe_bo_is_stolen_devmem(bo))
+ pte |= XE_PPGTT_PTE_DM;
+
+ return pte;
+}
+
+static u64 xelp_pte_encode_vma(u64 pte, struct xe_vma *vma,
+ u16 pat_index, u32 pt_level)
+{
+ struct xe_device *xe = xe_vma_vm(vma)->xe;
+
+ pte |= XE_PAGE_PRESENT;
+
+ if (likely(!xe_vma_read_only(vma)))
+ pte |= XE_PAGE_RW;
+
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
+ pte |= pte_encode_ps(pt_level);
+
+ if (unlikely(xe_vma_is_null(vma)))
+ pte |= XE_PTE_NULL;
+
+ return pte;
+}
+
+static u64 xelp_pte_encode_addr(struct xe_device *xe, u64 addr,
+ u16 pat_index,
+ u32 pt_level, bool devmem, u64 flags)
+{
+ u64 pte;
+
+ /* Avoid passing random bits directly as flags */
+ xe_assert(xe, !(flags & ~XE_PTE_PS64));
+
+ pte = addr;
+ pte |= XE_PAGE_PRESENT | XE_PAGE_RW;
+ pte |= pte_encode_pat_index(xe, pat_index, pt_level);
+ pte |= pte_encode_ps(pt_level);
+
+ if (devmem)
+ pte |= XE_PPGTT_PTE_DM;
+
+ pte |= flags;
+
+ return pte;
+}
+
+static const struct xe_pt_ops xelp_pt_ops = {
+ .pte_encode_bo = xelp_pte_encode_bo,
+ .pte_encode_vma = xelp_pte_encode_vma,
+ .pte_encode_addr = xelp_pte_encode_addr,
+ .pde_encode_bo = xelp_pde_encode_bo,
+};
+
+static void vm_destroy_work_func(struct work_struct *w);
+
+/**
+ * xe_vm_create_scratch() - Setup a scratch memory pagetable tree for the
+ * given tile and vm.
+ * @xe: xe device.
+ * @tile: tile to set up for.
+ * @vm: vm to set up for.
+ *
+ * Sets up a pagetable tree with one page-table per level and a single
+ * leaf PTE. All pagetable entries point to the single page-table or,
+ * for MAX_HUGEPTE_LEVEL, a NULL huge PTE returning 0 on read and
+ * writes become NOPs.
+ *
+ * Return: 0 on success, negative error code on error.
+ */
+static int xe_vm_create_scratch(struct xe_device *xe, struct xe_tile *tile,
+ struct xe_vm *vm)
+{
+ u8 id = tile->id;
+ int i;
+
+ for (i = MAX_HUGEPTE_LEVEL; i < vm->pt_root[id]->level; i++) {
+ vm->scratch_pt[id][i] = xe_pt_create(vm, tile, i);
+ if (IS_ERR(vm->scratch_pt[id][i]))
+ return PTR_ERR(vm->scratch_pt[id][i]);
+
+ xe_pt_populate_empty(tile, vm, vm->scratch_pt[id][i]);
+ }
+
+ return 0;
+}
+
+static void xe_vm_free_scratch(struct xe_vm *vm)
+{
+ struct xe_tile *tile;
+ u8 id;
+
+ if (!xe_vm_has_scratch(vm))
+ return;
+
+ for_each_tile(tile, vm->xe, id) {
+ u32 i;
+
+ if (!vm->pt_root[id])
+ continue;
+
+ for (i = MAX_HUGEPTE_LEVEL; i < vm->pt_root[id]->level; ++i)
+ if (vm->scratch_pt[id][i])
+ xe_pt_destroy(vm->scratch_pt[id][i], vm->flags, NULL);
+ }
+}
+
+struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
+{
+ struct drm_gem_object *vm_resv_obj;
+ struct xe_vm *vm;
+ int err, number_tiles = 0;
+ struct xe_tile *tile;
+ u8 id;
+
+ vm = kzalloc(sizeof(*vm), GFP_KERNEL);
+ if (!vm)
+ return ERR_PTR(-ENOMEM);
+
+ vm->xe = xe;
+
+ vm->size = 1ull << xe->info.va_bits;
+
+ vm->flags = flags;
+
+ init_rwsem(&vm->lock);
+
+ INIT_LIST_HEAD(&vm->rebind_list);
+
+ INIT_LIST_HEAD(&vm->userptr.repin_list);
+ INIT_LIST_HEAD(&vm->userptr.invalidated);
+ init_rwsem(&vm->userptr.notifier_lock);
+ spin_lock_init(&vm->userptr.invalidated_lock);
+
+ INIT_WORK(&vm->destroy_work, vm_destroy_work_func);
+
+ INIT_LIST_HEAD(&vm->preempt.exec_queues);
+ vm->preempt.min_run_period_ms = 10; /* FIXME: Wire up to uAPI */
+
+ for_each_tile(tile, xe, id)
+ xe_range_fence_tree_init(&vm->rftree[id]);
+
+ vm->pt_ops = &xelp_pt_ops;
+
+ if (!(flags & XE_VM_FLAG_MIGRATION))
+ xe_device_mem_access_get(xe);
+
+ vm_resv_obj = drm_gpuvm_resv_object_alloc(&xe->drm);
+ if (!vm_resv_obj) {
+ err = -ENOMEM;
+ goto err_no_resv;
+ }
+
+ drm_gpuvm_init(&vm->gpuvm, "Xe VM", DRM_GPUVM_RESV_PROTECTED, &xe->drm,
+ vm_resv_obj, 0, vm->size, 0, 0, &gpuvm_ops);
+
+ drm_gem_object_put(vm_resv_obj);
+
+ err = dma_resv_lock_interruptible(xe_vm_resv(vm), NULL);
+ if (err)
+ goto err_close;
+
+ if (IS_DGFX(xe) && xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K)
+ vm->flags |= XE_VM_FLAG_64K;
+
+ for_each_tile(tile, xe, id) {
+ if (flags & XE_VM_FLAG_MIGRATION &&
+ tile->id != XE_VM_FLAG_TILE_ID(flags))
+ continue;
+
+ vm->pt_root[id] = xe_pt_create(vm, tile, xe->info.vm_max_level);
+ if (IS_ERR(vm->pt_root[id])) {
+ err = PTR_ERR(vm->pt_root[id]);
+ vm->pt_root[id] = NULL;
+ goto err_unlock_close;
+ }
+ }
+
+ if (xe_vm_has_scratch(vm)) {
+ for_each_tile(tile, xe, id) {
+ if (!vm->pt_root[id])
+ continue;
+
+ err = xe_vm_create_scratch(xe, tile, vm);
+ if (err)
+ goto err_unlock_close;
+ }
+ vm->batch_invalidate_tlb = true;
+ }
+
+ if (flags & XE_VM_FLAG_LR_MODE) {
+ INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func);
+ vm->flags |= XE_VM_FLAG_LR_MODE;
+ vm->batch_invalidate_tlb = false;
+ }
+
+ /* Fill pt_root after allocating scratch tables */
+ for_each_tile(tile, xe, id) {
+ if (!vm->pt_root[id])
+ continue;
+
+ xe_pt_populate_empty(tile, vm, vm->pt_root[id]);
+ }
+ dma_resv_unlock(xe_vm_resv(vm));
+
+ /* Kernel migration VM shouldn't have a circular loop.. */
+ if (!(flags & XE_VM_FLAG_MIGRATION)) {
+ for_each_tile(tile, xe, id) {
+ struct xe_gt *gt = tile->primary_gt;
+ struct xe_vm *migrate_vm;
+ struct xe_exec_queue *q;
+ u32 create_flags = EXEC_QUEUE_FLAG_VM;
+
+ if (!vm->pt_root[id])
+ continue;
+
+ migrate_vm = xe_migrate_get_vm(tile->migrate);
+ q = xe_exec_queue_create_class(xe, gt, migrate_vm,
+ XE_ENGINE_CLASS_COPY,
+ create_flags);
+ xe_vm_put(migrate_vm);
+ if (IS_ERR(q)) {
+ err = PTR_ERR(q);
+ goto err_close;
+ }
+ vm->q[id] = q;
+ number_tiles++;
+ }
+ }
+
+ if (number_tiles > 1)
+ vm->composite_fence_ctx = dma_fence_context_alloc(1);
+
+ mutex_lock(&xe->usm.lock);
+ if (flags & XE_VM_FLAG_FAULT_MODE)
+ xe->usm.num_vm_in_fault_mode++;
+ else if (!(flags & XE_VM_FLAG_MIGRATION))
+ xe->usm.num_vm_in_non_fault_mode++;
+ mutex_unlock(&xe->usm.lock);
+
+ trace_xe_vm_create(vm);
+
+ return vm;
+
+err_unlock_close:
+ dma_resv_unlock(xe_vm_resv(vm));
+err_close:
+ xe_vm_close_and_put(vm);
+ return ERR_PTR(err);
+
+err_no_resv:
+ for_each_tile(tile, xe, id)
+ xe_range_fence_tree_fini(&vm->rftree[id]);
+ kfree(vm);
+ if (!(flags & XE_VM_FLAG_MIGRATION))
+ xe_device_mem_access_put(xe);
+ return ERR_PTR(err);
+}
+
+static void xe_vm_close(struct xe_vm *vm)
+{
+ down_write(&vm->lock);
+ vm->size = 0;
+ up_write(&vm->lock);
+}
+
+void xe_vm_close_and_put(struct xe_vm *vm)
+{
+ LIST_HEAD(contested);
+ struct xe_device *xe = vm->xe;
+ struct xe_tile *tile;
+ struct xe_vma *vma, *next_vma;
+ struct drm_gpuva *gpuva, *next;
+ u8 id;
+
+ xe_assert(xe, !vm->preempt.num_exec_queues);
+
+ xe_vm_close(vm);
+ if (xe_vm_in_preempt_fence_mode(vm))
+ flush_work(&vm->preempt.rebind_work);
+
+ down_write(&vm->lock);
+ for_each_tile(tile, xe, id) {
+ if (vm->q[id])
+ xe_exec_queue_last_fence_put(vm->q[id], vm);
+ }
+ up_write(&vm->lock);
+
+ for_each_tile(tile, xe, id) {
+ if (vm->q[id]) {
+ xe_exec_queue_kill(vm->q[id]);
+ xe_exec_queue_put(vm->q[id]);
+ vm->q[id] = NULL;
+ }
+ }
+
+ down_write(&vm->lock);
+ xe_vm_lock(vm, false);
+ drm_gpuvm_for_each_va_safe(gpuva, next, &vm->gpuvm) {
+ vma = gpuva_to_vma(gpuva);
+
+ if (xe_vma_has_no_bo(vma)) {
+ down_read(&vm->userptr.notifier_lock);
+ vma->gpuva.flags |= XE_VMA_DESTROYED;
+ up_read(&vm->userptr.notifier_lock);
+ }
+
+ xe_vm_remove_vma(vm, vma);
+
+ /* easy case, remove from VMA? */
+ if (xe_vma_has_no_bo(vma) || xe_vma_bo(vma)->vm) {
+ list_del_init(&vma->combined_links.rebind);
+ xe_vma_destroy(vma, NULL);
+ continue;
+ }
+
+ list_move_tail(&vma->combined_links.destroy, &contested);
+ vma->gpuva.flags |= XE_VMA_DESTROYED;
+ }
+
+ /*
+ * All vm operations will add shared fences to resv.
+ * The only exception is eviction for a shared object,
+ * but even so, the unbind when evicted would still
+ * install a fence to resv. Hence it's safe to
+ * destroy the pagetables immediately.
+ */
+ xe_vm_free_scratch(vm);
+
+ for_each_tile(tile, xe, id) {
+ if (vm->pt_root[id]) {
+ xe_pt_destroy(vm->pt_root[id], vm->flags, NULL);
+ vm->pt_root[id] = NULL;
+ }
+ }
+ xe_vm_unlock(vm);
+
+ /*
+ * VM is now dead, cannot re-add nodes to vm->vmas if it's NULL
+ * Since we hold a refcount to the bo, we can remove and free
+ * the members safely without locking.
+ */
+ list_for_each_entry_safe(vma, next_vma, &contested,
+ combined_links.destroy) {
+ list_del_init(&vma->combined_links.destroy);
+ xe_vma_destroy_unlocked(vma);
+ }
+
+ up_write(&vm->lock);
+
+ mutex_lock(&xe->usm.lock);
+ if (vm->flags & XE_VM_FLAG_FAULT_MODE)
+ xe->usm.num_vm_in_fault_mode--;
+ else if (!(vm->flags & XE_VM_FLAG_MIGRATION))
+ xe->usm.num_vm_in_non_fault_mode--;
+ mutex_unlock(&xe->usm.lock);
+
+ for_each_tile(tile, xe, id)
+ xe_range_fence_tree_fini(&vm->rftree[id]);
+
+ xe_vm_put(vm);
+}
+
+static void vm_destroy_work_func(struct work_struct *w)
+{
+ struct xe_vm *vm =
+ container_of(w, struct xe_vm, destroy_work);
+ struct xe_device *xe = vm->xe;
+ struct xe_tile *tile;
+ u8 id;
+ void *lookup;
+
+ /* xe_vm_close_and_put was not called? */
+ xe_assert(xe, !vm->size);
+
+ if (!(vm->flags & XE_VM_FLAG_MIGRATION)) {
+ xe_device_mem_access_put(xe);
+
+ if (xe->info.has_asid && vm->usm.asid) {
+ mutex_lock(&xe->usm.lock);
+ lookup = xa_erase(&xe->usm.asid_to_vm, vm->usm.asid);
+ xe_assert(xe, lookup == vm);
+ mutex_unlock(&xe->usm.lock);
+ }
+ }
+
+ for_each_tile(tile, xe, id)
+ XE_WARN_ON(vm->pt_root[id]);
+
+ trace_xe_vm_free(vm);
+ dma_fence_put(vm->rebind_fence);
+ kfree(vm);
+}
+
+static void xe_vm_free(struct drm_gpuvm *gpuvm)
+{
+ struct xe_vm *vm = container_of(gpuvm, struct xe_vm, gpuvm);
+
+ /* To destroy the VM we need to be able to sleep */
+ queue_work(system_unbound_wq, &vm->destroy_work);
+}
+
+struct xe_vm *xe_vm_lookup(struct xe_file *xef, u32 id)
+{
+ struct xe_vm *vm;
+
+ mutex_lock(&xef->vm.lock);
+ vm = xa_load(&xef->vm.xa, id);
+ if (vm)
+ xe_vm_get(vm);
+ mutex_unlock(&xef->vm.lock);
+
+ return vm;
+}
+
+u64 xe_vm_pdp4_descriptor(struct xe_vm *vm, struct xe_tile *tile)
+{
+ return vm->pt_ops->pde_encode_bo(vm->pt_root[tile->id]->bo, 0,
+ tile_to_xe(tile)->pat.idx[XE_CACHE_WB]);
+}
+
+static struct xe_exec_queue *
+to_wait_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q)
+{
+ return q ? q : vm->q[0];
+}
+
+static struct dma_fence *
+xe_vm_unbind_vma(struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool first_op, bool last_op)
+{
+ struct xe_vm *vm = xe_vma_vm(vma);
+ struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q);
+ struct xe_tile *tile;
+ struct dma_fence *fence = NULL;
+ struct dma_fence **fences = NULL;
+ struct dma_fence_array *cf = NULL;
+ int cur_fence = 0, i;
+ int number_tiles = hweight8(vma->tile_present);
+ int err;
+ u8 id;
+
+ trace_xe_vma_unbind(vma);
+
+ if (number_tiles > 1) {
+ fences = kmalloc_array(number_tiles, sizeof(*fences),
+ GFP_KERNEL);
+ if (!fences)
+ return ERR_PTR(-ENOMEM);
+ }
+
+ for_each_tile(tile, vm->xe, id) {
+ if (!(vma->tile_present & BIT(id)))
+ goto next;
+
+ fence = __xe_pt_unbind_vma(tile, vma, q ? q : vm->q[id],
+ first_op ? syncs : NULL,
+ first_op ? num_syncs : 0);
+ if (IS_ERR(fence)) {
+ err = PTR_ERR(fence);
+ goto err_fences;
+ }
+
+ if (fences)
+ fences[cur_fence++] = fence;
+
+next:
+ if (q && vm->pt_root[id] && !list_empty(&q->multi_gt_list))
+ q = list_next_entry(q, multi_gt_list);
+ }
+
+ if (fences) {
+ cf = dma_fence_array_create(number_tiles, fences,
+ vm->composite_fence_ctx,
+ vm->composite_fence_seqno++,
+ false);
+ if (!cf) {
+ --vm->composite_fence_seqno;
+ err = -ENOMEM;
+ goto err_fences;
+ }
+ }
+
+ fence = cf ? &cf->base : !fence ?
+ xe_exec_queue_last_fence_get(wait_exec_queue, vm) : fence;
+ if (last_op) {
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ }
+
+ return fence;
+
+err_fences:
+ if (fences) {
+ while (cur_fence)
+ dma_fence_put(fences[--cur_fence]);
+ kfree(fences);
+ }
+
+ return ERR_PTR(err);
+}
+
+static struct dma_fence *
+xe_vm_bind_vma(struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool first_op, bool last_op)
+{
+ struct xe_tile *tile;
+ struct dma_fence *fence;
+ struct dma_fence **fences = NULL;
+ struct dma_fence_array *cf = NULL;
+ struct xe_vm *vm = xe_vma_vm(vma);
+ int cur_fence = 0, i;
+ int number_tiles = hweight8(vma->tile_mask);
+ int err;
+ u8 id;
+
+ trace_xe_vma_bind(vma);
+
+ if (number_tiles > 1) {
+ fences = kmalloc_array(number_tiles, sizeof(*fences),
+ GFP_KERNEL);
+ if (!fences)
+ return ERR_PTR(-ENOMEM);
+ }
+
+ for_each_tile(tile, vm->xe, id) {
+ if (!(vma->tile_mask & BIT(id)))
+ goto next;
+
+ fence = __xe_pt_bind_vma(tile, vma, q ? q : vm->q[id],
+ first_op ? syncs : NULL,
+ first_op ? num_syncs : 0,
+ vma->tile_present & BIT(id));
+ if (IS_ERR(fence)) {
+ err = PTR_ERR(fence);
+ goto err_fences;
+ }
+
+ if (fences)
+ fences[cur_fence++] = fence;
+
+next:
+ if (q && vm->pt_root[id] && !list_empty(&q->multi_gt_list))
+ q = list_next_entry(q, multi_gt_list);
+ }
+
+ if (fences) {
+ cf = dma_fence_array_create(number_tiles, fences,
+ vm->composite_fence_ctx,
+ vm->composite_fence_seqno++,
+ false);
+ if (!cf) {
+ --vm->composite_fence_seqno;
+ err = -ENOMEM;
+ goto err_fences;
+ }
+ }
+
+ if (last_op) {
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], NULL,
+ cf ? &cf->base : fence);
+ }
+
+ return cf ? &cf->base : fence;
+
+err_fences:
+ if (fences) {
+ while (cur_fence)
+ dma_fence_put(fences[--cur_fence]);
+ kfree(fences);
+ }
+
+ return ERR_PTR(err);
+}
+
+static int __xe_vm_bind(struct xe_vm *vm, struct xe_vma *vma,
+ struct xe_exec_queue *q, struct xe_sync_entry *syncs,
+ u32 num_syncs, bool immediate, bool first_op,
+ bool last_op)
+{
+ struct dma_fence *fence;
+ struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q);
+
+ xe_vm_assert_held(vm);
+
+ if (immediate) {
+ fence = xe_vm_bind_vma(vma, q, syncs, num_syncs, first_op,
+ last_op);
+ if (IS_ERR(fence))
+ return PTR_ERR(fence);
+ } else {
+ int i;
+
+ xe_assert(vm->xe, xe_vm_in_fault_mode(vm));
+
+ fence = xe_exec_queue_last_fence_get(wait_exec_queue, vm);
+ if (last_op) {
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ }
+ }
+
+ if (last_op)
+ xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence);
+ dma_fence_put(fence);
+
+ return 0;
+}
+
+static int xe_vm_bind(struct xe_vm *vm, struct xe_vma *vma, struct xe_exec_queue *q,
+ struct xe_bo *bo, struct xe_sync_entry *syncs,
+ u32 num_syncs, bool immediate, bool first_op,
+ bool last_op)
+{
+ int err;
+
+ xe_vm_assert_held(vm);
+ xe_bo_assert_held(bo);
+
+ if (bo && immediate) {
+ err = xe_bo_validate(bo, vm, true);
+ if (err)
+ return err;
+ }
+
+ return __xe_vm_bind(vm, vma, q, syncs, num_syncs, immediate, first_op,
+ last_op);
+}
+
+static int xe_vm_unbind(struct xe_vm *vm, struct xe_vma *vma,
+ struct xe_exec_queue *q, struct xe_sync_entry *syncs,
+ u32 num_syncs, bool first_op, bool last_op)
+{
+ struct dma_fence *fence;
+ struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q);
+
+ xe_vm_assert_held(vm);
+ xe_bo_assert_held(xe_vma_bo(vma));
+
+ fence = xe_vm_unbind_vma(vma, q, syncs, num_syncs, first_op, last_op);
+ if (IS_ERR(fence))
+ return PTR_ERR(fence);
+
+ xe_vma_destroy(vma, fence);
+ if (last_op)
+ xe_exec_queue_last_fence_set(wait_exec_queue, vm, fence);
+ dma_fence_put(fence);
+
+ return 0;
+}
+
+#define ALL_DRM_XE_VM_CREATE_FLAGS (DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE | \
+ DRM_XE_VM_CREATE_FLAG_LR_MODE | \
+ DRM_XE_VM_CREATE_FLAG_FAULT_MODE)
+
+int xe_vm_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_vm_create *args = data;
+ struct xe_tile *tile;
+ struct xe_vm *vm;
+ u32 id, asid;
+ int err;
+ u32 flags = 0;
+
+ if (XE_IOCTL_DBG(xe, args->extensions))
+ return -EINVAL;
+
+ if (XE_WA(xe_root_mmio_gt(xe), 14016763929))
+ args->flags |= DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE;
+
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE &&
+ !xe->info.has_usm))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags & ~ALL_DRM_XE_VM_CREATE_FLAGS))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE &&
+ args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !(args->flags & DRM_XE_VM_CREATE_FLAG_LR_MODE) &&
+ args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE &&
+ xe_device_in_non_fault_mode(xe)))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, !(args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE) &&
+ xe_device_in_fault_mode(xe)))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->extensions))
+ return -EINVAL;
+
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE)
+ flags |= XE_VM_FLAG_SCRATCH_PAGE;
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_LR_MODE)
+ flags |= XE_VM_FLAG_LR_MODE;
+ if (args->flags & DRM_XE_VM_CREATE_FLAG_FAULT_MODE)
+ flags |= XE_VM_FLAG_FAULT_MODE;
+
+ vm = xe_vm_create(xe, flags);
+ if (IS_ERR(vm))
+ return PTR_ERR(vm);
+
+ mutex_lock(&xef->vm.lock);
+ err = xa_alloc(&xef->vm.xa, &id, vm, xa_limit_32b, GFP_KERNEL);
+ mutex_unlock(&xef->vm.lock);
+ if (err) {
+ xe_vm_close_and_put(vm);
+ return err;
+ }
+
+ if (xe->info.has_asid) {
+ mutex_lock(&xe->usm.lock);
+ err = xa_alloc_cyclic(&xe->usm.asid_to_vm, &asid, vm,
+ XA_LIMIT(1, XE_MAX_ASID - 1),
+ &xe->usm.next_asid, GFP_KERNEL);
+ mutex_unlock(&xe->usm.lock);
+ if (err < 0) {
+ xe_vm_close_and_put(vm);
+ return err;
+ }
+ err = 0;
+ vm->usm.asid = asid;
+ }
+
+ args->vm_id = id;
+ vm->xef = xef;
+
+ /* Record BO memory for VM pagetable created against client */
+ for_each_tile(tile, xe, id)
+ if (vm->pt_root[id])
+ xe_drm_client_add_bo(vm->xef->client, vm->pt_root[id]->bo);
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_MEM)
+ /* Warning: Security issue - never enable by default */
+ args->reserved[0] = xe_bo_main_addr(vm->pt_root[0]->bo, XE_PAGE_SIZE);
+#endif
+
+ return 0;
+}
+
+int xe_vm_destroy_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_vm_destroy *args = data;
+ struct xe_vm *vm;
+ int err = 0;
+
+ if (XE_IOCTL_DBG(xe, args->pad) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ mutex_lock(&xef->vm.lock);
+ vm = xa_load(&xef->vm.xa, args->vm_id);
+ if (XE_IOCTL_DBG(xe, !vm))
+ err = -ENOENT;
+ else if (XE_IOCTL_DBG(xe, vm->preempt.num_exec_queues))
+ err = -EBUSY;
+ else
+ xa_erase(&xef->vm.xa, args->vm_id);
+ mutex_unlock(&xef->vm.lock);
+
+ if (!err)
+ xe_vm_close_and_put(vm);
+
+ return err;
+}
+
+static const u32 region_to_mem_type[] = {
+ XE_PL_TT,
+ XE_PL_VRAM0,
+ XE_PL_VRAM1,
+};
+
+static int xe_vm_prefetch(struct xe_vm *vm, struct xe_vma *vma,
+ struct xe_exec_queue *q, u32 region,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ bool first_op, bool last_op)
+{
+ struct xe_exec_queue *wait_exec_queue = to_wait_exec_queue(vm, q);
+ int err;
+
+ xe_assert(vm->xe, region <= ARRAY_SIZE(region_to_mem_type));
+
+ if (!xe_vma_has_no_bo(vma)) {
+ err = xe_bo_migrate(xe_vma_bo(vma), region_to_mem_type[region]);
+ if (err)
+ return err;
+ }
+
+ if (vma->tile_mask != (vma->tile_present & ~vma->usm.tile_invalidated)) {
+ return xe_vm_bind(vm, vma, q, xe_vma_bo(vma), syncs, num_syncs,
+ true, first_op, last_op);
+ } else {
+ int i;
+
+ /* Nothing to do, signal fences now */
+ if (last_op) {
+ for (i = 0; i < num_syncs; i++) {
+ struct dma_fence *fence =
+ xe_exec_queue_last_fence_get(wait_exec_queue, vm);
+
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+ }
+ }
+
+ return 0;
+ }
+}
+
+static void prep_vma_destroy(struct xe_vm *vm, struct xe_vma *vma,
+ bool post_commit)
+{
+ down_read(&vm->userptr.notifier_lock);
+ vma->gpuva.flags |= XE_VMA_DESTROYED;
+ up_read(&vm->userptr.notifier_lock);
+ if (post_commit)
+ xe_vm_remove_vma(vm, vma);
+}
+
+#undef ULL
+#define ULL unsigned long long
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
+static void print_op(struct xe_device *xe, struct drm_gpuva_op *op)
+{
+ struct xe_vma *vma;
+
+ switch (op->op) {
+ case DRM_GPUVA_OP_MAP:
+ vm_dbg(&xe->drm, "MAP: addr=0x%016llx, range=0x%016llx",
+ (ULL)op->map.va.addr, (ULL)op->map.va.range);
+ break;
+ case DRM_GPUVA_OP_REMAP:
+ vma = gpuva_to_vma(op->remap.unmap->va);
+ vm_dbg(&xe->drm, "REMAP:UNMAP: addr=0x%016llx, range=0x%016llx, keep=%d",
+ (ULL)xe_vma_start(vma), (ULL)xe_vma_size(vma),
+ op->remap.unmap->keep ? 1 : 0);
+ if (op->remap.prev)
+ vm_dbg(&xe->drm,
+ "REMAP:PREV: addr=0x%016llx, range=0x%016llx",
+ (ULL)op->remap.prev->va.addr,
+ (ULL)op->remap.prev->va.range);
+ if (op->remap.next)
+ vm_dbg(&xe->drm,
+ "REMAP:NEXT: addr=0x%016llx, range=0x%016llx",
+ (ULL)op->remap.next->va.addr,
+ (ULL)op->remap.next->va.range);
+ break;
+ case DRM_GPUVA_OP_UNMAP:
+ vma = gpuva_to_vma(op->unmap.va);
+ vm_dbg(&xe->drm, "UNMAP: addr=0x%016llx, range=0x%016llx, keep=%d",
+ (ULL)xe_vma_start(vma), (ULL)xe_vma_size(vma),
+ op->unmap.keep ? 1 : 0);
+ break;
+ case DRM_GPUVA_OP_PREFETCH:
+ vma = gpuva_to_vma(op->prefetch.va);
+ vm_dbg(&xe->drm, "PREFETCH: addr=0x%016llx, range=0x%016llx",
+ (ULL)xe_vma_start(vma), (ULL)xe_vma_size(vma));
+ break;
+ default:
+ drm_warn(&xe->drm, "NOT POSSIBLE");
+ }
+}
+#else
+static void print_op(struct xe_device *xe, struct drm_gpuva_op *op)
+{
+}
+#endif
+
+/*
+ * Create operations list from IOCTL arguments, setup operations fields so parse
+ * and commit steps are decoupled from IOCTL arguments. This step can fail.
+ */
+static struct drm_gpuva_ops *
+vm_bind_ioctl_ops_create(struct xe_vm *vm, struct xe_bo *bo,
+ u64 bo_offset_or_userptr, u64 addr, u64 range,
+ u32 operation, u32 flags,
+ u32 prefetch_region, u16 pat_index)
+{
+ struct drm_gem_object *obj = bo ? &bo->ttm.base : NULL;
+ struct drm_gpuva_ops *ops;
+ struct drm_gpuva_op *__op;
+ struct xe_vma_op *op;
+ struct drm_gpuvm_bo *vm_bo;
+ int err;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ vm_dbg(&vm->xe->drm,
+ "op=%d, addr=0x%016llx, range=0x%016llx, bo_offset_or_userptr=0x%016llx",
+ operation, (ULL)addr, (ULL)range,
+ (ULL)bo_offset_or_userptr);
+
+ switch (operation) {
+ case DRM_XE_VM_BIND_OP_MAP:
+ case DRM_XE_VM_BIND_OP_MAP_USERPTR:
+ ops = drm_gpuvm_sm_map_ops_create(&vm->gpuvm, addr, range,
+ obj, bo_offset_or_userptr);
+ break;
+ case DRM_XE_VM_BIND_OP_UNMAP:
+ ops = drm_gpuvm_sm_unmap_ops_create(&vm->gpuvm, addr, range);
+ break;
+ case DRM_XE_VM_BIND_OP_PREFETCH:
+ ops = drm_gpuvm_prefetch_ops_create(&vm->gpuvm, addr, range);
+ break;
+ case DRM_XE_VM_BIND_OP_UNMAP_ALL:
+ xe_assert(vm->xe, bo);
+
+ err = xe_bo_lock(bo, true);
+ if (err)
+ return ERR_PTR(err);
+
+ vm_bo = drm_gpuvm_bo_find(&vm->gpuvm, obj);
+ if (!vm_bo)
+ break;
+
+ ops = drm_gpuvm_bo_unmap_ops_create(vm_bo);
+ drm_gpuvm_bo_put(vm_bo);
+ xe_bo_unlock(bo);
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ ops = ERR_PTR(-EINVAL);
+ }
+ if (IS_ERR(ops))
+ return ops;
+
+#ifdef TEST_VM_ASYNC_OPS_ERROR
+ if (operation & FORCE_ASYNC_OP_ERROR) {
+ op = list_first_entry_or_null(&ops->list, struct xe_vma_op,
+ base.entry);
+ if (op)
+ op->inject_error = true;
+ }
+#endif
+
+ drm_gpuva_for_each_op(__op, ops) {
+ struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
+
+ if (__op->op == DRM_GPUVA_OP_MAP) {
+ op->map.immediate =
+ flags & DRM_XE_VM_BIND_FLAG_IMMEDIATE;
+ op->map.read_only =
+ flags & DRM_XE_VM_BIND_FLAG_READONLY;
+ op->map.is_null = flags & DRM_XE_VM_BIND_FLAG_NULL;
+ op->map.pat_index = pat_index;
+ } else if (__op->op == DRM_GPUVA_OP_PREFETCH) {
+ op->prefetch.region = prefetch_region;
+ }
+
+ print_op(vm->xe, __op);
+ }
+
+ return ops;
+}
+
+static struct xe_vma *new_vma(struct xe_vm *vm, struct drm_gpuva_op_map *op,
+ u16 pat_index, unsigned int flags)
+{
+ struct xe_bo *bo = op->gem.obj ? gem_to_xe_bo(op->gem.obj) : NULL;
+ struct drm_exec exec;
+ struct xe_vma *vma;
+ int err;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ if (bo) {
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+ drm_exec_until_all_locked(&exec) {
+ err = 0;
+ if (!bo->vm) {
+ err = drm_exec_lock_obj(&exec, xe_vm_obj(vm));
+ drm_exec_retry_on_contention(&exec);
+ }
+ if (!err) {
+ err = drm_exec_lock_obj(&exec, &bo->ttm.base);
+ drm_exec_retry_on_contention(&exec);
+ }
+ if (err) {
+ drm_exec_fini(&exec);
+ return ERR_PTR(err);
+ }
+ }
+ }
+ vma = xe_vma_create(vm, bo, op->gem.offset,
+ op->va.addr, op->va.addr +
+ op->va.range - 1, pat_index, flags);
+ if (bo)
+ drm_exec_fini(&exec);
+
+ if (xe_vma_is_userptr(vma)) {
+ err = xe_vma_userptr_pin_pages(vma);
+ if (err) {
+ prep_vma_destroy(vm, vma, false);
+ xe_vma_destroy_unlocked(vma);
+ return ERR_PTR(err);
+ }
+ } else if (!xe_vma_has_no_bo(vma) && !bo->vm) {
+ err = add_preempt_fences(vm, bo);
+ if (err) {
+ prep_vma_destroy(vm, vma, false);
+ xe_vma_destroy_unlocked(vma);
+ return ERR_PTR(err);
+ }
+ }
+
+ return vma;
+}
+
+static u64 xe_vma_max_pte_size(struct xe_vma *vma)
+{
+ if (vma->gpuva.flags & XE_VMA_PTE_1G)
+ return SZ_1G;
+ else if (vma->gpuva.flags & XE_VMA_PTE_2M)
+ return SZ_2M;
+
+ return SZ_4K;
+}
+
+static u64 xe_vma_set_pte_size(struct xe_vma *vma, u64 size)
+{
+ switch (size) {
+ case SZ_1G:
+ vma->gpuva.flags |= XE_VMA_PTE_1G;
+ break;
+ case SZ_2M:
+ vma->gpuva.flags |= XE_VMA_PTE_2M;
+ break;
+ }
+
+ return SZ_4K;
+}
+
+static int xe_vma_op_commit(struct xe_vm *vm, struct xe_vma_op *op)
+{
+ int err = 0;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ switch (op->base.op) {
+ case DRM_GPUVA_OP_MAP:
+ err |= xe_vm_insert_vma(vm, op->map.vma);
+ if (!err)
+ op->flags |= XE_VMA_OP_COMMITTED;
+ break;
+ case DRM_GPUVA_OP_REMAP:
+ {
+ u8 tile_present =
+ gpuva_to_vma(op->base.remap.unmap->va)->tile_present;
+
+ prep_vma_destroy(vm, gpuva_to_vma(op->base.remap.unmap->va),
+ true);
+ op->flags |= XE_VMA_OP_COMMITTED;
+
+ if (op->remap.prev) {
+ err |= xe_vm_insert_vma(vm, op->remap.prev);
+ if (!err)
+ op->flags |= XE_VMA_OP_PREV_COMMITTED;
+ if (!err && op->remap.skip_prev) {
+ op->remap.prev->tile_present =
+ tile_present;
+ op->remap.prev = NULL;
+ }
+ }
+ if (op->remap.next) {
+ err |= xe_vm_insert_vma(vm, op->remap.next);
+ if (!err)
+ op->flags |= XE_VMA_OP_NEXT_COMMITTED;
+ if (!err && op->remap.skip_next) {
+ op->remap.next->tile_present =
+ tile_present;
+ op->remap.next = NULL;
+ }
+ }
+
+ /* Adjust for partial unbind after removin VMA from VM */
+ if (!err) {
+ op->base.remap.unmap->va->va.addr = op->remap.start;
+ op->base.remap.unmap->va->va.range = op->remap.range;
+ }
+ break;
+ }
+ case DRM_GPUVA_OP_UNMAP:
+ prep_vma_destroy(vm, gpuva_to_vma(op->base.unmap.va), true);
+ op->flags |= XE_VMA_OP_COMMITTED;
+ break;
+ case DRM_GPUVA_OP_PREFETCH:
+ op->flags |= XE_VMA_OP_COMMITTED;
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ }
+
+ return err;
+}
+
+
+static int vm_bind_ioctl_ops_parse(struct xe_vm *vm, struct xe_exec_queue *q,
+ struct drm_gpuva_ops *ops,
+ struct xe_sync_entry *syncs, u32 num_syncs,
+ struct list_head *ops_list, bool last)
+{
+ struct xe_vma_op *last_op = NULL;
+ struct drm_gpuva_op *__op;
+ int err = 0;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ drm_gpuva_for_each_op(__op, ops) {
+ struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
+ struct xe_vma *vma;
+ bool first = list_empty(ops_list);
+ unsigned int flags = 0;
+
+ INIT_LIST_HEAD(&op->link);
+ list_add_tail(&op->link, ops_list);
+
+ if (first) {
+ op->flags |= XE_VMA_OP_FIRST;
+ op->num_syncs = num_syncs;
+ op->syncs = syncs;
+ }
+
+ op->q = q;
+
+ switch (op->base.op) {
+ case DRM_GPUVA_OP_MAP:
+ {
+ flags |= op->map.read_only ?
+ VMA_CREATE_FLAG_READ_ONLY : 0;
+ flags |= op->map.is_null ?
+ VMA_CREATE_FLAG_IS_NULL : 0;
+
+ vma = new_vma(vm, &op->base.map, op->map.pat_index,
+ flags);
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
+
+ op->map.vma = vma;
+ break;
+ }
+ case DRM_GPUVA_OP_REMAP:
+ {
+ struct xe_vma *old =
+ gpuva_to_vma(op->base.remap.unmap->va);
+
+ op->remap.start = xe_vma_start(old);
+ op->remap.range = xe_vma_size(old);
+
+ if (op->base.remap.prev) {
+ flags |= op->base.remap.unmap->va->flags &
+ XE_VMA_READ_ONLY ?
+ VMA_CREATE_FLAG_READ_ONLY : 0;
+ flags |= op->base.remap.unmap->va->flags &
+ DRM_GPUVA_SPARSE ?
+ VMA_CREATE_FLAG_IS_NULL : 0;
+
+ vma = new_vma(vm, op->base.remap.prev,
+ old->pat_index, flags);
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
+
+ op->remap.prev = vma;
+
+ /*
+ * Userptr creates a new SG mapping so
+ * we must also rebind.
+ */
+ op->remap.skip_prev = !xe_vma_is_userptr(old) &&
+ IS_ALIGNED(xe_vma_end(vma),
+ xe_vma_max_pte_size(old));
+ if (op->remap.skip_prev) {
+ xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old));
+ op->remap.range -=
+ xe_vma_end(vma) -
+ xe_vma_start(old);
+ op->remap.start = xe_vma_end(vma);
+ }
+ }
+
+ if (op->base.remap.next) {
+ flags |= op->base.remap.unmap->va->flags &
+ XE_VMA_READ_ONLY ?
+ VMA_CREATE_FLAG_READ_ONLY : 0;
+ flags |= op->base.remap.unmap->va->flags &
+ DRM_GPUVA_SPARSE ?
+ VMA_CREATE_FLAG_IS_NULL : 0;
+
+ vma = new_vma(vm, op->base.remap.next,
+ old->pat_index, flags);
+ if (IS_ERR(vma))
+ return PTR_ERR(vma);
+
+ op->remap.next = vma;
+
+ /*
+ * Userptr creates a new SG mapping so
+ * we must also rebind.
+ */
+ op->remap.skip_next = !xe_vma_is_userptr(old) &&
+ IS_ALIGNED(xe_vma_start(vma),
+ xe_vma_max_pte_size(old));
+ if (op->remap.skip_next) {
+ xe_vma_set_pte_size(vma, xe_vma_max_pte_size(old));
+ op->remap.range -=
+ xe_vma_end(old) -
+ xe_vma_start(vma);
+ }
+ }
+ break;
+ }
+ case DRM_GPUVA_OP_UNMAP:
+ case DRM_GPUVA_OP_PREFETCH:
+ /* Nothing to do */
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ }
+
+ last_op = op;
+
+ err = xe_vma_op_commit(vm, op);
+ if (err)
+ return err;
+ }
+
+ /* FIXME: Unhandled corner case */
+ XE_WARN_ON(!last_op && last && !list_empty(ops_list));
+
+ if (!last_op)
+ return 0;
+
+ last_op->ops = ops;
+ if (last) {
+ last_op->flags |= XE_VMA_OP_LAST;
+ last_op->num_syncs = num_syncs;
+ last_op->syncs = syncs;
+ }
+
+ return 0;
+}
+
+static int op_execute(struct drm_exec *exec, struct xe_vm *vm,
+ struct xe_vma *vma, struct xe_vma_op *op)
+{
+ int err;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ err = xe_vm_prepare_vma(exec, vma, 1);
+ if (err)
+ return err;
+
+ xe_vm_assert_held(vm);
+ xe_bo_assert_held(xe_vma_bo(vma));
+
+ switch (op->base.op) {
+ case DRM_GPUVA_OP_MAP:
+ err = xe_vm_bind(vm, vma, op->q, xe_vma_bo(vma),
+ op->syncs, op->num_syncs,
+ op->map.immediate || !xe_vm_in_fault_mode(vm),
+ op->flags & XE_VMA_OP_FIRST,
+ op->flags & XE_VMA_OP_LAST);
+ break;
+ case DRM_GPUVA_OP_REMAP:
+ {
+ bool prev = !!op->remap.prev;
+ bool next = !!op->remap.next;
+
+ if (!op->remap.unmap_done) {
+ if (prev || next)
+ vma->gpuva.flags |= XE_VMA_FIRST_REBIND;
+ err = xe_vm_unbind(vm, vma, op->q, op->syncs,
+ op->num_syncs,
+ op->flags & XE_VMA_OP_FIRST,
+ op->flags & XE_VMA_OP_LAST &&
+ !prev && !next);
+ if (err)
+ break;
+ op->remap.unmap_done = true;
+ }
+
+ if (prev) {
+ op->remap.prev->gpuva.flags |= XE_VMA_LAST_REBIND;
+ err = xe_vm_bind(vm, op->remap.prev, op->q,
+ xe_vma_bo(op->remap.prev), op->syncs,
+ op->num_syncs, true, false,
+ op->flags & XE_VMA_OP_LAST && !next);
+ op->remap.prev->gpuva.flags &= ~XE_VMA_LAST_REBIND;
+ if (err)
+ break;
+ op->remap.prev = NULL;
+ }
+
+ if (next) {
+ op->remap.next->gpuva.flags |= XE_VMA_LAST_REBIND;
+ err = xe_vm_bind(vm, op->remap.next, op->q,
+ xe_vma_bo(op->remap.next),
+ op->syncs, op->num_syncs,
+ true, false,
+ op->flags & XE_VMA_OP_LAST);
+ op->remap.next->gpuva.flags &= ~XE_VMA_LAST_REBIND;
+ if (err)
+ break;
+ op->remap.next = NULL;
+ }
+
+ break;
+ }
+ case DRM_GPUVA_OP_UNMAP:
+ err = xe_vm_unbind(vm, vma, op->q, op->syncs,
+ op->num_syncs, op->flags & XE_VMA_OP_FIRST,
+ op->flags & XE_VMA_OP_LAST);
+ break;
+ case DRM_GPUVA_OP_PREFETCH:
+ err = xe_vm_prefetch(vm, vma, op->q, op->prefetch.region,
+ op->syncs, op->num_syncs,
+ op->flags & XE_VMA_OP_FIRST,
+ op->flags & XE_VMA_OP_LAST);
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ }
+
+ if (err)
+ trace_xe_vma_fail(vma);
+
+ return err;
+}
+
+static int __xe_vma_op_execute(struct xe_vm *vm, struct xe_vma *vma,
+ struct xe_vma_op *op)
+{
+ struct drm_exec exec;
+ int err;
+
+retry_userptr:
+ drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT, 0);
+ drm_exec_until_all_locked(&exec) {
+ err = op_execute(&exec, vm, vma, op);
+ drm_exec_retry_on_contention(&exec);
+ if (err)
+ break;
+ }
+ drm_exec_fini(&exec);
+
+ if (err == -EAGAIN && xe_vma_is_userptr(vma)) {
+ lockdep_assert_held_write(&vm->lock);
+ err = xe_vma_userptr_pin_pages(vma);
+ if (!err)
+ goto retry_userptr;
+
+ trace_xe_vma_fail(vma);
+ }
+
+ return err;
+}
+
+static int xe_vma_op_execute(struct xe_vm *vm, struct xe_vma_op *op)
+{
+ int ret = 0;
+
+ lockdep_assert_held_write(&vm->lock);
+
+#ifdef TEST_VM_ASYNC_OPS_ERROR
+ if (op->inject_error) {
+ op->inject_error = false;
+ return -ENOMEM;
+ }
+#endif
+
+ switch (op->base.op) {
+ case DRM_GPUVA_OP_MAP:
+ ret = __xe_vma_op_execute(vm, op->map.vma, op);
+ break;
+ case DRM_GPUVA_OP_REMAP:
+ {
+ struct xe_vma *vma;
+
+ if (!op->remap.unmap_done)
+ vma = gpuva_to_vma(op->base.remap.unmap->va);
+ else if (op->remap.prev)
+ vma = op->remap.prev;
+ else
+ vma = op->remap.next;
+
+ ret = __xe_vma_op_execute(vm, vma, op);
+ break;
+ }
+ case DRM_GPUVA_OP_UNMAP:
+ ret = __xe_vma_op_execute(vm, gpuva_to_vma(op->base.unmap.va),
+ op);
+ break;
+ case DRM_GPUVA_OP_PREFETCH:
+ ret = __xe_vma_op_execute(vm,
+ gpuva_to_vma(op->base.prefetch.va),
+ op);
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ }
+
+ return ret;
+}
+
+static void xe_vma_op_cleanup(struct xe_vm *vm, struct xe_vma_op *op)
+{
+ bool last = op->flags & XE_VMA_OP_LAST;
+
+ if (last) {
+ while (op->num_syncs--)
+ xe_sync_entry_cleanup(&op->syncs[op->num_syncs]);
+ kfree(op->syncs);
+ if (op->q)
+ xe_exec_queue_put(op->q);
+ }
+ if (!list_empty(&op->link))
+ list_del(&op->link);
+ if (op->ops)
+ drm_gpuva_ops_free(&vm->gpuvm, op->ops);
+ if (last)
+ xe_vm_put(vm);
+}
+
+static void xe_vma_op_unwind(struct xe_vm *vm, struct xe_vma_op *op,
+ bool post_commit, bool prev_post_commit,
+ bool next_post_commit)
+{
+ lockdep_assert_held_write(&vm->lock);
+
+ switch (op->base.op) {
+ case DRM_GPUVA_OP_MAP:
+ if (op->map.vma) {
+ prep_vma_destroy(vm, op->map.vma, post_commit);
+ xe_vma_destroy_unlocked(op->map.vma);
+ }
+ break;
+ case DRM_GPUVA_OP_UNMAP:
+ {
+ struct xe_vma *vma = gpuva_to_vma(op->base.unmap.va);
+
+ if (vma) {
+ down_read(&vm->userptr.notifier_lock);
+ vma->gpuva.flags &= ~XE_VMA_DESTROYED;
+ up_read(&vm->userptr.notifier_lock);
+ if (post_commit)
+ xe_vm_insert_vma(vm, vma);
+ }
+ break;
+ }
+ case DRM_GPUVA_OP_REMAP:
+ {
+ struct xe_vma *vma = gpuva_to_vma(op->base.remap.unmap->va);
+
+ if (op->remap.prev) {
+ prep_vma_destroy(vm, op->remap.prev, prev_post_commit);
+ xe_vma_destroy_unlocked(op->remap.prev);
+ }
+ if (op->remap.next) {
+ prep_vma_destroy(vm, op->remap.next, next_post_commit);
+ xe_vma_destroy_unlocked(op->remap.next);
+ }
+ if (vma) {
+ down_read(&vm->userptr.notifier_lock);
+ vma->gpuva.flags &= ~XE_VMA_DESTROYED;
+ up_read(&vm->userptr.notifier_lock);
+ if (post_commit)
+ xe_vm_insert_vma(vm, vma);
+ }
+ break;
+ }
+ case DRM_GPUVA_OP_PREFETCH:
+ /* Nothing to do */
+ break;
+ default:
+ drm_warn(&vm->xe->drm, "NOT POSSIBLE");
+ }
+}
+
+static void vm_bind_ioctl_ops_unwind(struct xe_vm *vm,
+ struct drm_gpuva_ops **ops,
+ int num_ops_list)
+{
+ int i;
+
+ for (i = num_ops_list - 1; i; ++i) {
+ struct drm_gpuva_ops *__ops = ops[i];
+ struct drm_gpuva_op *__op;
+
+ if (!__ops)
+ continue;
+
+ drm_gpuva_for_each_op_reverse(__op, __ops) {
+ struct xe_vma_op *op = gpuva_op_to_vma_op(__op);
+
+ xe_vma_op_unwind(vm, op,
+ op->flags & XE_VMA_OP_COMMITTED,
+ op->flags & XE_VMA_OP_PREV_COMMITTED,
+ op->flags & XE_VMA_OP_NEXT_COMMITTED);
+ }
+
+ drm_gpuva_ops_free(&vm->gpuvm, __ops);
+ }
+}
+
+static int vm_bind_ioctl_ops_execute(struct xe_vm *vm,
+ struct list_head *ops_list)
+{
+ struct xe_vma_op *op, *next;
+ int err;
+
+ lockdep_assert_held_write(&vm->lock);
+
+ list_for_each_entry_safe(op, next, ops_list, link) {
+ err = xe_vma_op_execute(vm, op);
+ if (err) {
+ drm_warn(&vm->xe->drm, "VM op(%d) failed with %d",
+ op->base.op, err);
+ /*
+ * FIXME: Killing VM rather than proper error handling
+ */
+ xe_vm_kill(vm);
+ return -ENOSPC;
+ }
+ xe_vma_op_cleanup(vm, op);
+ }
+
+ return 0;
+}
+
+#ifdef TEST_VM_ASYNC_OPS_ERROR
+#define SUPPORTED_FLAGS \
+ (FORCE_ASYNC_OP_ERROR | DRM_XE_VM_BIND_FLAG_READONLY | \
+ DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL | 0xffff)
+#else
+#define SUPPORTED_FLAGS \
+ (DRM_XE_VM_BIND_FLAG_READONLY | \
+ DRM_XE_VM_BIND_FLAG_IMMEDIATE | DRM_XE_VM_BIND_FLAG_NULL | \
+ 0xffff)
+#endif
+#define XE_64K_PAGE_MASK 0xffffull
+#define ALL_DRM_XE_SYNCS_FLAGS (DRM_XE_SYNCS_FLAG_WAIT_FOR_OP)
+
+#define MAX_BINDS 512 /* FIXME: Picking random upper limit */
+
+static int vm_bind_ioctl_check_args(struct xe_device *xe,
+ struct drm_xe_vm_bind *args,
+ struct drm_xe_vm_bind_op **bind_ops)
+{
+ int err;
+ int i;
+
+ if (XE_IOCTL_DBG(xe, args->pad || args->pad2) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->extensions) ||
+ XE_IOCTL_DBG(xe, args->num_binds > MAX_BINDS))
+ return -EINVAL;
+
+ if (args->num_binds > 1) {
+ u64 __user *bind_user =
+ u64_to_user_ptr(args->vector_of_binds);
+
+ *bind_ops = kmalloc(sizeof(struct drm_xe_vm_bind_op) *
+ args->num_binds, GFP_KERNEL);
+ if (!*bind_ops)
+ return -ENOMEM;
+
+ err = __copy_from_user(*bind_ops, bind_user,
+ sizeof(struct drm_xe_vm_bind_op) *
+ args->num_binds);
+ if (XE_IOCTL_DBG(xe, err)) {
+ err = -EFAULT;
+ goto free_bind_ops;
+ }
+ } else {
+ *bind_ops = &args->bind;
+ }
+
+ for (i = 0; i < args->num_binds; ++i) {
+ u64 range = (*bind_ops)[i].range;
+ u64 addr = (*bind_ops)[i].addr;
+ u32 op = (*bind_ops)[i].op;
+ u32 flags = (*bind_ops)[i].flags;
+ u32 obj = (*bind_ops)[i].obj;
+ u64 obj_offset = (*bind_ops)[i].obj_offset;
+ u32 prefetch_region = (*bind_ops)[i].prefetch_mem_region_instance;
+ bool is_null = flags & DRM_XE_VM_BIND_FLAG_NULL;
+ u16 pat_index = (*bind_ops)[i].pat_index;
+ u16 coh_mode;
+
+ if (XE_IOCTL_DBG(xe, pat_index >= xe->pat.n_entries)) {
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+
+ pat_index = array_index_nospec(pat_index, xe->pat.n_entries);
+ (*bind_ops)[i].pat_index = pat_index;
+ coh_mode = xe_pat_index_get_coh_mode(xe, pat_index);
+ if (XE_IOCTL_DBG(xe, !coh_mode)) { /* hw reserved */
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+
+ if (XE_WARN_ON(coh_mode > XE_COH_AT_LEAST_1WAY)) {
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+
+ if (XE_IOCTL_DBG(xe, op > DRM_XE_VM_BIND_OP_PREFETCH) ||
+ XE_IOCTL_DBG(xe, flags & ~SUPPORTED_FLAGS) ||
+ XE_IOCTL_DBG(xe, obj && is_null) ||
+ XE_IOCTL_DBG(xe, obj_offset && is_null) ||
+ XE_IOCTL_DBG(xe, op != DRM_XE_VM_BIND_OP_MAP &&
+ is_null) ||
+ XE_IOCTL_DBG(xe, !obj &&
+ op == DRM_XE_VM_BIND_OP_MAP &&
+ !is_null) ||
+ XE_IOCTL_DBG(xe, !obj &&
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
+ XE_IOCTL_DBG(xe, addr &&
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
+ XE_IOCTL_DBG(xe, range &&
+ op == DRM_XE_VM_BIND_OP_UNMAP_ALL) ||
+ XE_IOCTL_DBG(xe, obj &&
+ op == DRM_XE_VM_BIND_OP_MAP_USERPTR) ||
+ XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE &&
+ op == DRM_XE_VM_BIND_OP_MAP_USERPTR) ||
+ XE_IOCTL_DBG(xe, obj &&
+ op == DRM_XE_VM_BIND_OP_PREFETCH) ||
+ XE_IOCTL_DBG(xe, prefetch_region &&
+ op != DRM_XE_VM_BIND_OP_PREFETCH) ||
+ XE_IOCTL_DBG(xe, !(BIT(prefetch_region) &
+ xe->info.mem_region_mask)) ||
+ XE_IOCTL_DBG(xe, obj &&
+ op == DRM_XE_VM_BIND_OP_UNMAP)) {
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+
+ if (XE_IOCTL_DBG(xe, obj_offset & ~PAGE_MASK) ||
+ XE_IOCTL_DBG(xe, addr & ~PAGE_MASK) ||
+ XE_IOCTL_DBG(xe, range & ~PAGE_MASK) ||
+ XE_IOCTL_DBG(xe, !range &&
+ op != DRM_XE_VM_BIND_OP_UNMAP_ALL)) {
+ err = -EINVAL;
+ goto free_bind_ops;
+ }
+ }
+
+ return 0;
+
+free_bind_ops:
+ if (args->num_binds > 1)
+ kfree(*bind_ops);
+ return err;
+}
+
+static int vm_bind_ioctl_signal_fences(struct xe_vm *vm,
+ struct xe_exec_queue *q,
+ struct xe_sync_entry *syncs,
+ int num_syncs)
+{
+ struct dma_fence *fence;
+ int i, err = 0;
+
+ fence = xe_sync_in_fence_get(syncs, num_syncs,
+ to_wait_exec_queue(vm, q), vm);
+ if (IS_ERR(fence))
+ return PTR_ERR(fence);
+
+ for (i = 0; i < num_syncs; i++)
+ xe_sync_entry_signal(&syncs[i], NULL, fence);
+
+ xe_exec_queue_last_fence_set(to_wait_exec_queue(vm, q), vm,
+ fence);
+ dma_fence_put(fence);
+
+ return err;
+}
+
+int xe_vm_bind_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ struct drm_xe_vm_bind *args = data;
+ struct drm_xe_sync __user *syncs_user;
+ struct xe_bo **bos = NULL;
+ struct drm_gpuva_ops **ops = NULL;
+ struct xe_vm *vm;
+ struct xe_exec_queue *q = NULL;
+ u32 num_syncs;
+ struct xe_sync_entry *syncs = NULL;
+ struct drm_xe_vm_bind_op *bind_ops;
+ LIST_HEAD(ops_list);
+ int err;
+ int i;
+
+ err = vm_bind_ioctl_check_args(xe, args, &bind_ops);
+ if (err)
+ return err;
+
+ if (args->exec_queue_id) {
+ q = xe_exec_queue_lookup(xef, args->exec_queue_id);
+ if (XE_IOCTL_DBG(xe, !q)) {
+ err = -ENOENT;
+ goto free_objs;
+ }
+
+ if (XE_IOCTL_DBG(xe, !(q->flags & EXEC_QUEUE_FLAG_VM))) {
+ err = -EINVAL;
+ goto put_exec_queue;
+ }
+ }
+
+ vm = xe_vm_lookup(xef, args->vm_id);
+ if (XE_IOCTL_DBG(xe, !vm)) {
+ err = -EINVAL;
+ goto put_exec_queue;
+ }
+
+ err = down_write_killable(&vm->lock);
+ if (err)
+ goto put_vm;
+
+ if (XE_IOCTL_DBG(xe, xe_vm_is_closed_or_banned(vm))) {
+ err = -ENOENT;
+ goto release_vm_lock;
+ }
+
+ for (i = 0; i < args->num_binds; ++i) {
+ u64 range = bind_ops[i].range;
+ u64 addr = bind_ops[i].addr;
+
+ if (XE_IOCTL_DBG(xe, range > vm->size) ||
+ XE_IOCTL_DBG(xe, addr > vm->size - range)) {
+ err = -EINVAL;
+ goto release_vm_lock;
+ }
+ }
+
+ if (args->num_binds) {
+ bos = kcalloc(args->num_binds, sizeof(*bos), GFP_KERNEL);
+ if (!bos) {
+ err = -ENOMEM;
+ goto release_vm_lock;
+ }
+
+ ops = kcalloc(args->num_binds, sizeof(*ops), GFP_KERNEL);
+ if (!ops) {
+ err = -ENOMEM;
+ goto release_vm_lock;
+ }
+ }
+
+ for (i = 0; i < args->num_binds; ++i) {
+ struct drm_gem_object *gem_obj;
+ u64 range = bind_ops[i].range;
+ u64 addr = bind_ops[i].addr;
+ u32 obj = bind_ops[i].obj;
+ u64 obj_offset = bind_ops[i].obj_offset;
+ u16 pat_index = bind_ops[i].pat_index;
+ u16 coh_mode;
+
+ if (!obj)
+ continue;
+
+ gem_obj = drm_gem_object_lookup(file, obj);
+ if (XE_IOCTL_DBG(xe, !gem_obj)) {
+ err = -ENOENT;
+ goto put_obj;
+ }
+ bos[i] = gem_to_xe_bo(gem_obj);
+
+ if (XE_IOCTL_DBG(xe, range > bos[i]->size) ||
+ XE_IOCTL_DBG(xe, obj_offset >
+ bos[i]->size - range)) {
+ err = -EINVAL;
+ goto put_obj;
+ }
+
+ if (bos[i]->flags & XE_BO_INTERNAL_64K) {
+ if (XE_IOCTL_DBG(xe, obj_offset &
+ XE_64K_PAGE_MASK) ||
+ XE_IOCTL_DBG(xe, addr & XE_64K_PAGE_MASK) ||
+ XE_IOCTL_DBG(xe, range & XE_64K_PAGE_MASK)) {
+ err = -EINVAL;
+ goto put_obj;
+ }
+ }
+
+ coh_mode = xe_pat_index_get_coh_mode(xe, pat_index);
+ if (bos[i]->cpu_caching) {
+ if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE &&
+ bos[i]->cpu_caching == DRM_XE_GEM_CPU_CACHING_WB)) {
+ err = -EINVAL;
+ goto put_obj;
+ }
+ } else if (XE_IOCTL_DBG(xe, coh_mode == XE_COH_NONE)) {
+ /*
+ * Imported dma-buf from a different device should
+ * require 1way or 2way coherency since we don't know
+ * how it was mapped on the CPU. Just assume is it
+ * potentially cached on CPU side.
+ */
+ err = -EINVAL;
+ goto put_obj;
+ }
+ }
+
+ if (args->num_syncs) {
+ syncs = kcalloc(args->num_syncs, sizeof(*syncs), GFP_KERNEL);
+ if (!syncs) {
+ err = -ENOMEM;
+ goto put_obj;
+ }
+ }
+
+ syncs_user = u64_to_user_ptr(args->syncs);
+ for (num_syncs = 0; num_syncs < args->num_syncs; num_syncs++) {
+ err = xe_sync_entry_parse(xe, xef, &syncs[num_syncs],
+ &syncs_user[num_syncs],
+ (xe_vm_in_lr_mode(vm) ?
+ SYNC_PARSE_FLAG_LR_MODE : 0) |
+ (!args->num_binds ?
+ SYNC_PARSE_FLAG_DISALLOW_USER_FENCE : 0));
+ if (err)
+ goto free_syncs;
+ }
+
+ if (!args->num_binds) {
+ err = -ENODATA;
+ goto free_syncs;
+ }
+
+ for (i = 0; i < args->num_binds; ++i) {
+ u64 range = bind_ops[i].range;
+ u64 addr = bind_ops[i].addr;
+ u32 op = bind_ops[i].op;
+ u32 flags = bind_ops[i].flags;
+ u64 obj_offset = bind_ops[i].obj_offset;
+ u32 prefetch_region = bind_ops[i].prefetch_mem_region_instance;
+ u16 pat_index = bind_ops[i].pat_index;
+
+ ops[i] = vm_bind_ioctl_ops_create(vm, bos[i], obj_offset,
+ addr, range, op, flags,
+ prefetch_region, pat_index);
+ if (IS_ERR(ops[i])) {
+ err = PTR_ERR(ops[i]);
+ ops[i] = NULL;
+ goto unwind_ops;
+ }
+
+ err = vm_bind_ioctl_ops_parse(vm, q, ops[i], syncs, num_syncs,
+ &ops_list,
+ i == args->num_binds - 1);
+ if (err)
+ goto unwind_ops;
+ }
+
+ /* Nothing to do */
+ if (list_empty(&ops_list)) {
+ err = -ENODATA;
+ goto unwind_ops;
+ }
+
+ xe_vm_get(vm);
+ if (q)
+ xe_exec_queue_get(q);
+
+ err = vm_bind_ioctl_ops_execute(vm, &ops_list);
+
+ up_write(&vm->lock);
+
+ if (q)
+ xe_exec_queue_put(q);
+ xe_vm_put(vm);
+
+ for (i = 0; bos && i < args->num_binds; ++i)
+ xe_bo_put(bos[i]);
+
+ kfree(bos);
+ kfree(ops);
+ if (args->num_binds > 1)
+ kfree(bind_ops);
+
+ return err;
+
+unwind_ops:
+ vm_bind_ioctl_ops_unwind(vm, ops, args->num_binds);
+free_syncs:
+ if (err == -ENODATA)
+ err = vm_bind_ioctl_signal_fences(vm, q, syncs, num_syncs);
+ while (num_syncs--)
+ xe_sync_entry_cleanup(&syncs[num_syncs]);
+
+ kfree(syncs);
+put_obj:
+ for (i = 0; i < args->num_binds; ++i)
+ xe_bo_put(bos[i]);
+release_vm_lock:
+ up_write(&vm->lock);
+put_vm:
+ xe_vm_put(vm);
+put_exec_queue:
+ if (q)
+ xe_exec_queue_put(q);
+free_objs:
+ kfree(bos);
+ kfree(ops);
+ if (args->num_binds > 1)
+ kfree(bind_ops);
+ return err;
+}
+
+/**
+ * xe_vm_lock() - Lock the vm's dma_resv object
+ * @vm: The struct xe_vm whose lock is to be locked
+ * @intr: Whether to perform any wait interruptible
+ *
+ * Return: 0 on success, -EINTR if @intr is true and the wait for a
+ * contended lock was interrupted. If @intr is false, the function
+ * always returns 0.
+ */
+int xe_vm_lock(struct xe_vm *vm, bool intr)
+{
+ if (intr)
+ return dma_resv_lock_interruptible(xe_vm_resv(vm), NULL);
+
+ return dma_resv_lock(xe_vm_resv(vm), NULL);
+}
+
+/**
+ * xe_vm_unlock() - Unlock the vm's dma_resv object
+ * @vm: The struct xe_vm whose lock is to be released.
+ *
+ * Unlock a buffer object lock that was locked by xe_vm_lock().
+ */
+void xe_vm_unlock(struct xe_vm *vm)
+{
+ dma_resv_unlock(xe_vm_resv(vm));
+}
+
+/**
+ * xe_vm_invalidate_vma - invalidate GPU mappings for VMA without a lock
+ * @vma: VMA to invalidate
+ *
+ * Walks a list of page tables leaves which it memset the entries owned by this
+ * VMA to zero, invalidates the TLBs, and block until TLBs invalidation is
+ * complete.
+ *
+ * Returns 0 for success, negative error code otherwise.
+ */
+int xe_vm_invalidate_vma(struct xe_vma *vma)
+{
+ struct xe_device *xe = xe_vma_vm(vma)->xe;
+ struct xe_tile *tile;
+ u32 tile_needs_invalidate = 0;
+ int seqno[XE_MAX_TILES_PER_DEVICE];
+ u8 id;
+ int ret;
+
+ xe_assert(xe, xe_vm_in_fault_mode(xe_vma_vm(vma)));
+ xe_assert(xe, !xe_vma_is_null(vma));
+ trace_xe_vma_usm_invalidate(vma);
+
+ /* Check that we don't race with page-table updates */
+ if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
+ if (xe_vma_is_userptr(vma)) {
+ WARN_ON_ONCE(!mmu_interval_check_retry
+ (&vma->userptr.notifier,
+ vma->userptr.notifier_seq));
+ WARN_ON_ONCE(!dma_resv_test_signaled(xe_vm_resv(xe_vma_vm(vma)),
+ DMA_RESV_USAGE_BOOKKEEP));
+
+ } else {
+ xe_bo_assert_held(xe_vma_bo(vma));
+ }
+ }
+
+ for_each_tile(tile, xe, id) {
+ if (xe_pt_zap_ptes(tile, vma)) {
+ tile_needs_invalidate |= BIT(id);
+ xe_device_wmb(xe);
+ /*
+ * FIXME: We potentially need to invalidate multiple
+ * GTs within the tile
+ */
+ seqno[id] = xe_gt_tlb_invalidation_vma(tile->primary_gt, NULL, vma);
+ if (seqno[id] < 0)
+ return seqno[id];
+ }
+ }
+
+ for_each_tile(tile, xe, id) {
+ if (tile_needs_invalidate & BIT(id)) {
+ ret = xe_gt_tlb_invalidation_wait(tile->primary_gt, seqno[id]);
+ if (ret < 0)
+ return ret;
+ }
+ }
+
+ vma->usm.tile_invalidated = vma->tile_mask;
+
+ return 0;
+}
+
+int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id)
+{
+ struct drm_gpuva *gpuva;
+ bool is_vram;
+ uint64_t addr;
+
+ if (!down_read_trylock(&vm->lock)) {
+ drm_printf(p, " Failed to acquire VM lock to dump capture");
+ return 0;
+ }
+ if (vm->pt_root[gt_id]) {
+ addr = xe_bo_addr(vm->pt_root[gt_id]->bo, 0, XE_PAGE_SIZE);
+ is_vram = xe_bo_is_vram(vm->pt_root[gt_id]->bo);
+ drm_printf(p, " VM root: A:0x%llx %s\n", addr,
+ is_vram ? "VRAM" : "SYS");
+ }
+
+ drm_gpuvm_for_each_va(gpuva, &vm->gpuvm) {
+ struct xe_vma *vma = gpuva_to_vma(gpuva);
+ bool is_userptr = xe_vma_is_userptr(vma);
+ bool is_null = xe_vma_is_null(vma);
+
+ if (is_null) {
+ addr = 0;
+ } else if (is_userptr) {
+ struct xe_res_cursor cur;
+
+ if (vma->userptr.sg) {
+ xe_res_first_sg(vma->userptr.sg, 0, XE_PAGE_SIZE,
+ &cur);
+ addr = xe_res_dma(&cur);
+ } else {
+ addr = 0;
+ }
+ } else {
+ addr = __xe_bo_addr(xe_vma_bo(vma), 0, XE_PAGE_SIZE);
+ is_vram = xe_bo_is_vram(xe_vma_bo(vma));
+ }
+ drm_printf(p, " [%016llx-%016llx] S:0x%016llx A:%016llx %s\n",
+ xe_vma_start(vma), xe_vma_end(vma) - 1,
+ xe_vma_size(vma),
+ addr, is_null ? "NULL" : is_userptr ? "USR" :
+ is_vram ? "VRAM" : "SYS");
+ }
+ up_read(&vm->lock);
+
+ return 0;
+}
diff --git a/drivers/gpu/drm/xe/xe_vm.h b/drivers/gpu/drm/xe/xe_vm.h
new file mode 100644
index 000000000000..cf2f96e8c1ab
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_vm.h
@@ -0,0 +1,263 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2021 Intel Corporation
+ */
+
+#ifndef _XE_VM_H_
+#define _XE_VM_H_
+
+#include "xe_bo_types.h"
+#include "xe_macros.h"
+#include "xe_map.h"
+#include "xe_vm_types.h"
+
+struct drm_device;
+struct drm_printer;
+struct drm_file;
+
+struct ttm_buffer_object;
+struct ttm_validate_buffer;
+
+struct xe_exec_queue;
+struct xe_file;
+struct xe_sync_entry;
+struct drm_exec;
+
+struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags);
+
+struct xe_vm *xe_vm_lookup(struct xe_file *xef, u32 id);
+int xe_vma_cmp_vma_cb(const void *key, const struct rb_node *node);
+
+static inline struct xe_vm *xe_vm_get(struct xe_vm *vm)
+{
+ drm_gpuvm_get(&vm->gpuvm);
+ return vm;
+}
+
+static inline void xe_vm_put(struct xe_vm *vm)
+{
+ drm_gpuvm_put(&vm->gpuvm);
+}
+
+int xe_vm_lock(struct xe_vm *vm, bool intr);
+
+void xe_vm_unlock(struct xe_vm *vm);
+
+static inline bool xe_vm_is_closed(struct xe_vm *vm)
+{
+ /* Only guaranteed not to change when vm->lock is held */
+ return !vm->size;
+}
+
+static inline bool xe_vm_is_banned(struct xe_vm *vm)
+{
+ return vm->flags & XE_VM_FLAG_BANNED;
+}
+
+static inline bool xe_vm_is_closed_or_banned(struct xe_vm *vm)
+{
+ lockdep_assert_held(&vm->lock);
+ return xe_vm_is_closed(vm) || xe_vm_is_banned(vm);
+}
+
+struct xe_vma *
+xe_vm_find_overlapping_vma(struct xe_vm *vm, u64 start, u64 range);
+
+/**
+ * xe_vm_has_scratch() - Whether the vm is configured for scratch PTEs
+ * @vm: The vm
+ *
+ * Return: whether the vm populates unmapped areas with scratch PTEs
+ */
+static inline bool xe_vm_has_scratch(const struct xe_vm *vm)
+{
+ return vm->flags & XE_VM_FLAG_SCRATCH_PAGE;
+}
+
+/**
+ * gpuvm_to_vm() - Return the embedding xe_vm from a struct drm_gpuvm pointer
+ * @gpuvm: The struct drm_gpuvm pointer
+ *
+ * Return: Pointer to the embedding struct xe_vm.
+ */
+static inline struct xe_vm *gpuvm_to_vm(struct drm_gpuvm *gpuvm)
+{
+ return container_of(gpuvm, struct xe_vm, gpuvm);
+}
+
+static inline struct xe_vm *gpuva_to_vm(struct drm_gpuva *gpuva)
+{
+ return gpuvm_to_vm(gpuva->vm);
+}
+
+static inline struct xe_vma *gpuva_to_vma(struct drm_gpuva *gpuva)
+{
+ return container_of(gpuva, struct xe_vma, gpuva);
+}
+
+static inline struct xe_vma_op *gpuva_op_to_vma_op(struct drm_gpuva_op *op)
+{
+ return container_of(op, struct xe_vma_op, base);
+}
+
+/**
+ * DOC: Provide accessors for vma members to facilitate easy change of
+ * implementation.
+ */
+static inline u64 xe_vma_start(struct xe_vma *vma)
+{
+ return vma->gpuva.va.addr;
+}
+
+static inline u64 xe_vma_size(struct xe_vma *vma)
+{
+ return vma->gpuva.va.range;
+}
+
+static inline u64 xe_vma_end(struct xe_vma *vma)
+{
+ return xe_vma_start(vma) + xe_vma_size(vma);
+}
+
+static inline u64 xe_vma_bo_offset(struct xe_vma *vma)
+{
+ return vma->gpuva.gem.offset;
+}
+
+static inline struct xe_bo *xe_vma_bo(struct xe_vma *vma)
+{
+ return !vma->gpuva.gem.obj ? NULL :
+ container_of(vma->gpuva.gem.obj, struct xe_bo, ttm.base);
+}
+
+static inline struct xe_vm *xe_vma_vm(struct xe_vma *vma)
+{
+ return container_of(vma->gpuva.vm, struct xe_vm, gpuvm);
+}
+
+static inline bool xe_vma_read_only(struct xe_vma *vma)
+{
+ return vma->gpuva.flags & XE_VMA_READ_ONLY;
+}
+
+static inline u64 xe_vma_userptr(struct xe_vma *vma)
+{
+ return vma->gpuva.gem.offset;
+}
+
+static inline bool xe_vma_is_null(struct xe_vma *vma)
+{
+ return vma->gpuva.flags & DRM_GPUVA_SPARSE;
+}
+
+static inline bool xe_vma_has_no_bo(struct xe_vma *vma)
+{
+ return !xe_vma_bo(vma);
+}
+
+static inline bool xe_vma_is_userptr(struct xe_vma *vma)
+{
+ return xe_vma_has_no_bo(vma) && !xe_vma_is_null(vma);
+}
+
+u64 xe_vm_pdp4_descriptor(struct xe_vm *vm, struct xe_tile *tile);
+
+int xe_vm_create_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_vm_destroy_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+int xe_vm_bind_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+
+void xe_vm_close_and_put(struct xe_vm *vm);
+
+static inline bool xe_vm_in_fault_mode(struct xe_vm *vm)
+{
+ return vm->flags & XE_VM_FLAG_FAULT_MODE;
+}
+
+static inline bool xe_vm_in_lr_mode(struct xe_vm *vm)
+{
+ return vm->flags & XE_VM_FLAG_LR_MODE;
+}
+
+static inline bool xe_vm_in_preempt_fence_mode(struct xe_vm *vm)
+{
+ return xe_vm_in_lr_mode(vm) && !xe_vm_in_fault_mode(vm);
+}
+
+int xe_vm_add_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q);
+void xe_vm_remove_compute_exec_queue(struct xe_vm *vm, struct xe_exec_queue *q);
+
+int xe_vm_userptr_pin(struct xe_vm *vm);
+
+int __xe_vm_userptr_needs_repin(struct xe_vm *vm);
+
+int xe_vm_userptr_check_repin(struct xe_vm *vm);
+
+struct dma_fence *xe_vm_rebind(struct xe_vm *vm, bool rebind_worker);
+
+int xe_vm_invalidate_vma(struct xe_vma *vma);
+
+extern struct ttm_device_funcs xe_ttm_funcs;
+
+static inline void xe_vm_queue_rebind_worker(struct xe_vm *vm)
+{
+ xe_assert(vm->xe, xe_vm_in_preempt_fence_mode(vm));
+ queue_work(vm->xe->ordered_wq, &vm->preempt.rebind_work);
+}
+
+/**
+ * xe_vm_reactivate_rebind() - Reactivate the rebind functionality on compute
+ * vms.
+ * @vm: The vm.
+ *
+ * If the rebind functionality on a compute vm was disabled due
+ * to nothing to execute. Reactivate it and run the rebind worker.
+ * This function should be called after submitting a batch to a compute vm.
+ */
+static inline void xe_vm_reactivate_rebind(struct xe_vm *vm)
+{
+ if (xe_vm_in_preempt_fence_mode(vm) && vm->preempt.rebind_deactivated) {
+ vm->preempt.rebind_deactivated = false;
+ xe_vm_queue_rebind_worker(vm);
+ }
+}
+
+int xe_vma_userptr_pin_pages(struct xe_vma *vma);
+
+int xe_vma_userptr_check_repin(struct xe_vma *vma);
+
+bool xe_vm_validate_should_retry(struct drm_exec *exec, int err, ktime_t *end);
+
+int xe_analyze_vm(struct drm_printer *p, struct xe_vm *vm, int gt_id);
+
+int xe_vm_prepare_vma(struct drm_exec *exec, struct xe_vma *vma,
+ unsigned int num_shared);
+
+/**
+ * xe_vm_resv() - Return's the vm's reservation object
+ * @vm: The vm
+ *
+ * Return: Pointer to the vm's reservation object.
+ */
+static inline struct dma_resv *xe_vm_resv(struct xe_vm *vm)
+{
+ return drm_gpuvm_resv(&vm->gpuvm);
+}
+
+/**
+ * xe_vm_assert_held(vm) - Assert that the vm's reservation object is held.
+ * @vm: The vm
+ */
+#define xe_vm_assert_held(vm) dma_resv_assert_held(xe_vm_resv(vm))
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG_VM)
+#define vm_dbg drm_dbg
+#else
+__printf(2, 3)
+static inline void vm_dbg(const struct drm_device *dev,
+ const char *format, ...)
+{ /* noop */ }
+#endif
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vm_doc.h b/drivers/gpu/drm/xe/xe_vm_doc.h
new file mode 100644
index 000000000000..bdc6659891a5
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_vm_doc.h
@@ -0,0 +1,555 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_VM_DOC_H_
+#define _XE_VM_DOC_H_
+
+/**
+ * DOC: XE VM (user address space)
+ *
+ * VM creation
+ * ===========
+ *
+ * Allocate a physical page for root of the page table structure, create default
+ * bind engine, and return a handle to the user.
+ *
+ * Scratch page
+ * ------------
+ *
+ * If the VM is created with the flag, DRM_XE_VM_CREATE_FLAG_SCRATCH_PAGE, set the
+ * entire page table structure defaults pointing to blank page allocated by the
+ * VM. Invalid memory access rather than fault just read / write to this page.
+ *
+ * VM bind (create GPU mapping for a BO or userptr)
+ * ================================================
+ *
+ * Creates GPU mapings for a BO or userptr within a VM. VM binds uses the same
+ * in / out fence interface (struct drm_xe_sync) as execs which allows users to
+ * think of binds and execs as more or less the same operation.
+ *
+ * Operations
+ * ----------
+ *
+ * DRM_XE_VM_BIND_OP_MAP - Create mapping for a BO
+ * DRM_XE_VM_BIND_OP_UNMAP - Destroy mapping for a BO / userptr
+ * DRM_XE_VM_BIND_OP_MAP_USERPTR - Create mapping for userptr
+ *
+ * Implementation details
+ * ~~~~~~~~~~~~~~~~~~~~~~
+ *
+ * All bind operations are implemented via a hybrid approach of using the CPU
+ * and GPU to modify page tables. If a new physical page is allocated in the
+ * page table structure we populate that page via the CPU and insert that new
+ * page into the existing page table structure via a GPU job. Also any existing
+ * pages in the page table structure that need to be modified also are updated
+ * via the GPU job. As the root physical page is prealloced on VM creation our
+ * GPU job will always have at least 1 update. The in / out fences are passed to
+ * this job so again this is conceptually the same as an exec.
+ *
+ * Very simple example of few binds on an empty VM with 48 bits of address space
+ * and the resulting operations:
+ *
+ * .. code-block::
+ *
+ * bind BO0 0x0-0x1000
+ * alloc page level 3a, program PTE[0] to BO0 phys address (CPU)
+ * alloc page level 2, program PDE[0] page level 3a phys address (CPU)
+ * alloc page level 1, program PDE[0] page level 2 phys address (CPU)
+ * update root PDE[0] to page level 1 phys address (GPU)
+ *
+ * bind BO1 0x201000-0x202000
+ * alloc page level 3b, program PTE[1] to BO1 phys address (CPU)
+ * update page level 2 PDE[1] to page level 3b phys address (GPU)
+ *
+ * bind BO2 0x1ff000-0x201000
+ * update page level 3a PTE[511] to BO2 phys addres (GPU)
+ * update page level 3b PTE[0] to BO2 phys addres + 0x1000 (GPU)
+ *
+ * GPU bypass
+ * ~~~~~~~~~~
+ *
+ * In the above example the steps using the GPU can be converted to CPU if the
+ * bind can be done immediately (all in-fences satisfied, VM dma-resv kernel
+ * slot is idle).
+ *
+ * Address space
+ * -------------
+ *
+ * Depending on platform either 48 or 57 bits of address space is supported.
+ *
+ * Page sizes
+ * ----------
+ *
+ * The minimum page size is either 4k or 64k depending on platform and memory
+ * placement (sysmem vs. VRAM). We enforce that binds must be aligned to the
+ * minimum page size.
+ *
+ * Larger pages (2M or 1GB) can be used for BOs in VRAM, the BO physical address
+ * is aligned to the larger pages size, and VA is aligned to the larger page
+ * size. Larger pages for userptrs / BOs in sysmem should be possible but is not
+ * yet implemented.
+ *
+ * Sync error handling mode
+ * ------------------------
+ *
+ * In both modes during the bind IOCTL the user input is validated. In sync
+ * error handling mode the newly bound BO is validated (potentially moved back
+ * to a region of memory where is can be used), page tables are updated by the
+ * CPU and the job to do the GPU binds is created in the IOCTL itself. This step
+ * can fail due to memory pressure. The user can recover by freeing memory and
+ * trying this operation again.
+ *
+ * Async error handling mode
+ * -------------------------
+ *
+ * In async error handling the step of validating the BO, updating page tables,
+ * and generating a job are deferred to an async worker. As this step can now
+ * fail after the IOCTL has reported success we need an error handling flow for
+ * which the user can recover from.
+ *
+ * The solution is for a user to register a user address with the VM which the
+ * VM uses to report errors to. The ufence wait interface can be used to wait on
+ * a VM going into an error state. Once an error is reported the VM's async
+ * worker is paused. While the VM's async worker is paused sync,
+ * DRM_XE_VM_BIND_OP_UNMAP operations are allowed (this can free memory). Once the
+ * uses believe the error state is fixed, the async worker can be resumed via
+ * XE_VM_BIND_OP_RESTART operation. When VM async bind work is restarted, the
+ * first operation processed is the operation that caused the original error.
+ *
+ * Bind queues / engines
+ * ---------------------
+ *
+ * Think of the case where we have two bind operations A + B and are submitted
+ * in that order. A has in fences while B has none. If using a single bind
+ * queue, B is now blocked on A's in fences even though it is ready to run. This
+ * example is a real use case for VK sparse binding. We work around this
+ * limitation by implementing bind engines.
+ *
+ * In the bind IOCTL the user can optionally pass in an engine ID which must map
+ * to an engine which is of the special class DRM_XE_ENGINE_CLASS_VM_BIND.
+ * Underneath this is a really virtual engine that can run on any of the copy
+ * hardware engines. The job(s) created each IOCTL are inserted into this
+ * engine's ring. In the example above if A and B have different bind engines B
+ * is free to pass A. If the engine ID field is omitted, the default bind queue
+ * for the VM is used.
+ *
+ * TODO: Explain race in issue 41 and how we solve it
+ *
+ * Array of bind operations
+ * ------------------------
+ *
+ * The uAPI allows multiple binds operations to be passed in via a user array,
+ * of struct drm_xe_vm_bind_op, in a single VM bind IOCTL. This interface
+ * matches the VK sparse binding API. The implementation is rather simple, parse
+ * the array into a list of operations, pass the in fences to the first operation,
+ * and pass the out fences to the last operation. The ordered nature of a bind
+ * engine makes this possible.
+ *
+ * Munmap semantics for unbinds
+ * ----------------------------
+ *
+ * Munmap allows things like:
+ *
+ * .. code-block::
+ *
+ * 0x0000-0x2000 and 0x3000-0x5000 have mappings
+ * Munmap 0x1000-0x4000, results in mappings 0x0000-0x1000 and 0x4000-0x5000
+ *
+ * To support this semantic in the above example we decompose the above example
+ * into 4 operations:
+ *
+ * .. code-block::
+ *
+ * unbind 0x0000-0x2000
+ * unbind 0x3000-0x5000
+ * rebind 0x0000-0x1000
+ * rebind 0x4000-0x5000
+ *
+ * Why not just do a partial unbind of 0x1000-0x2000 and 0x3000-0x4000? This
+ * falls apart when using large pages at the edges and the unbind forces us to
+ * use a smaller page size. For simplity we always issue a set of unbinds
+ * unmapping anything in the range and at most 2 rebinds on the edges.
+ *
+ * Similar to an array of binds, in fences are passed to the first operation and
+ * out fences are signaled on the last operation.
+ *
+ * In this example there is a window of time where 0x0000-0x1000 and
+ * 0x4000-0x5000 are invalid but the user didn't ask for these addresses to be
+ * removed from the mapping. To work around this we treat any munmap style
+ * unbinds which require a rebind as a kernel operations (BO eviction or userptr
+ * invalidation). The first operation waits on the VM's
+ * DMA_RESV_USAGE_PREEMPT_FENCE slots (waits for all pending jobs on VM to
+ * complete / triggers preempt fences) and the last operation is installed in
+ * the VM's DMA_RESV_USAGE_KERNEL slot (blocks future jobs / resume compute mode
+ * VM). The caveat is all dma-resv slots must be updated atomically with respect
+ * to execs and compute mode rebind worker. To accomplish this, hold the
+ * vm->lock in write mode from the first operation until the last.
+ *
+ * Deferred binds in fault mode
+ * ----------------------------
+ *
+ * In a VM is in fault mode (TODO: link to fault mode), new bind operations that
+ * create mappings are by default are deferred to the page fault handler (first
+ * use). This behavior can be overriden by setting the flag
+ * DRM_XE_VM_BIND_FLAG_IMMEDIATE which indicates to creating the mapping
+ * immediately.
+ *
+ * User pointer
+ * ============
+ *
+ * User pointers are user allocated memory (malloc'd, mmap'd, etc..) for which the
+ * user wants to create a GPU mapping. Typically in other DRM drivers a dummy BO
+ * was created and then a binding was created. We bypass creating a dummy BO in
+ * XE and simply create a binding directly from the userptr.
+ *
+ * Invalidation
+ * ------------
+ *
+ * Since this a core kernel managed memory the kernel can move this memory
+ * whenever it wants. We register an invalidation MMU notifier to alert XE when
+ * a user poiter is about to move. The invalidation notifier needs to block
+ * until all pending users (jobs or compute mode engines) of the userptr are
+ * idle to ensure no faults. This done by waiting on all of VM's dma-resv slots.
+ *
+ * Rebinds
+ * -------
+ *
+ * Either the next exec (non-compute) or rebind worker (compute mode) will
+ * rebind the userptr. The invalidation MMU notifier kicks the rebind worker
+ * after the VM dma-resv wait if the VM is in compute mode.
+ *
+ * Compute mode
+ * ============
+ *
+ * A VM in compute mode enables long running workloads and ultra low latency
+ * submission (ULLS). ULLS is implemented via a continuously running batch +
+ * semaphores. This enables to the user to insert jump to new batch commands
+ * into the continuously running batch. In both cases these batches exceed the
+ * time a dma fence is allowed to exist for before signaling, as such dma fences
+ * are not used when a VM is in compute mode. User fences (TODO: link user fence
+ * doc) are used instead to signal operation's completion.
+ *
+ * Preempt fences
+ * --------------
+ *
+ * If the kernel decides to move memory around (either userptr invalidate, BO
+ * eviction, or mumap style unbind which results in a rebind) and a batch is
+ * running on an engine, that batch can fault or cause a memory corruption as
+ * page tables for the moved memory are no longer valid. To work around this we
+ * introduce the concept of preempt fences. When sw signaling is enabled on a
+ * preempt fence it tells the submission backend to kick that engine off the
+ * hardware and the preempt fence signals when the engine is off the hardware.
+ * Once all preempt fences are signaled for a VM the kernel can safely move the
+ * memory and kick the rebind worker which resumes all the engines execution.
+ *
+ * A preempt fence, for every engine using the VM, is installed the VM's
+ * dma-resv DMA_RESV_USAGE_PREEMPT_FENCE slot. The same preempt fence, for every
+ * engine using the VM, is also installed into the same dma-resv slot of every
+ * external BO mapped in the VM.
+ *
+ * Rebind worker
+ * -------------
+ *
+ * The rebind worker is very similar to an exec. It is resposible for rebinding
+ * evicted BOs or userptrs, waiting on those operations, installing new preempt
+ * fences, and finally resuming executing of engines in the VM.
+ *
+ * Flow
+ * ~~~~
+ *
+ * .. code-block::
+ *
+ * <----------------------------------------------------------------------|
+ * Check if VM is closed, if so bail out |
+ * Lock VM global lock in read mode |
+ * Pin userptrs (also finds userptr invalidated since last rebind worker) |
+ * Lock VM dma-resv and external BOs dma-resv |
+ * Validate BOs that have been evicted |
+ * Wait on and allocate new preempt fences for every engine using the VM |
+ * Rebind invalidated userptrs + evicted BOs |
+ * Wait on last rebind fence |
+ * Wait VM's DMA_RESV_USAGE_KERNEL dma-resv slot |
+ * Install preeempt fences and issue resume for every engine using the VM |
+ * Check if any userptrs invalidated since pin |
+ * Squash resume for all engines |
+ * Unlock all |
+ * Wait all VM's dma-resv slots |
+ * Retry ----------------------------------------------------------
+ * Release all engines waiting to resume
+ * Unlock all
+ *
+ * Timeslicing
+ * -----------
+ *
+ * In order to prevent an engine from continuously being kicked off the hardware
+ * and making no forward progress an engine has a period of time it allowed to
+ * run after resume before it can be kicked off again. This effectively gives
+ * each engine a timeslice.
+ *
+ * Handling multiple GTs
+ * =====================
+ *
+ * If a GT has slower access to some regions and the page table structure are in
+ * the slow region, the performance on that GT could adversely be affected. To
+ * work around this we allow a VM page tables to be shadowed in multiple GTs.
+ * When VM is created, a default bind engine and PT table structure are created
+ * on each GT.
+ *
+ * Binds can optionally pass in a mask of GTs where a mapping should be created,
+ * if this mask is zero then default to all the GTs where the VM has page
+ * tables.
+ *
+ * The implementation for this breaks down into a bunch for_each_gt loops in
+ * various places plus exporting a composite fence for multi-GT binds to the
+ * user.
+ *
+ * Fault mode (unified shared memory)
+ * ==================================
+ *
+ * A VM in fault mode can be enabled on devices that support page faults. If
+ * page faults are enabled, using dma fences can potentially induce a deadlock:
+ * A pending page fault can hold up the GPU work which holds up the dma fence
+ * signaling, and memory allocation is usually required to resolve a page
+ * fault, but memory allocation is not allowed to gate dma fence signaling. As
+ * such, dma fences are not allowed when VM is in fault mode. Because dma-fences
+ * are not allowed, long running workloads and ULLS are enabled on a faulting
+ * VM.
+ *
+ * Defered VM binds
+ * ----------------
+ *
+ * By default, on a faulting VM binds just allocate the VMA and the actual
+ * updating of the page tables is defered to the page fault handler. This
+ * behavior can be overridden by setting the flag DRM_XE_VM_BIND_FLAG_IMMEDIATE in
+ * the VM bind which will then do the bind immediately.
+ *
+ * Page fault handler
+ * ------------------
+ *
+ * Page faults are received in the G2H worker under the CT lock which is in the
+ * path of dma fences (no memory allocations are allowed, faults require memory
+ * allocations) thus we cannot process faults under the CT lock. Another issue
+ * is faults issue TLB invalidations which require G2H credits and we cannot
+ * allocate G2H credits in the G2H handlers without deadlocking. Lastly, we do
+ * not want the CT lock to be an outer lock of the VM global lock (VM global
+ * lock required to fault processing).
+ *
+ * To work around the above issue with processing faults in the G2H worker, we
+ * sink faults to a buffer which is large enough to sink all possible faults on
+ * the GT (1 per hardware engine) and kick a worker to process the faults. Since
+ * the page faults G2H are already received in a worker, kicking another worker
+ * adds more latency to a critical performance path. We add a fast path in the
+ * G2H irq handler which looks at first G2H and if it is a page fault we sink
+ * the fault to the buffer and kick the worker to process the fault. TLB
+ * invalidation responses are also in the critical path so these can also be
+ * processed in this fast path.
+ *
+ * Multiple buffers and workers are used and hashed over based on the ASID so
+ * faults from different VMs can be processed in parallel.
+ *
+ * The page fault handler itself is rather simple, flow is below.
+ *
+ * .. code-block::
+ *
+ * Lookup VM from ASID in page fault G2H
+ * Lock VM global lock in read mode
+ * Lookup VMA from address in page fault G2H
+ * Check if VMA is valid, if not bail
+ * Check if VMA's BO has backing store, if not allocate
+ * <----------------------------------------------------------------------|
+ * If userptr, pin pages |
+ * Lock VM & BO dma-resv locks |
+ * If atomic fault, migrate to VRAM, else validate BO location |
+ * Issue rebind |
+ * Wait on rebind to complete |
+ * Check if userptr invalidated since pin |
+ * Drop VM & BO dma-resv locks |
+ * Retry ----------------------------------------------------------
+ * Unlock all
+ * Issue blocking TLB invalidation |
+ * Send page fault response to GuC
+ *
+ * Access counters
+ * ---------------
+ *
+ * Access counters can be configured to trigger a G2H indicating the device is
+ * accessing VMAs in system memory frequently as hint to migrate those VMAs to
+ * VRAM.
+ *
+ * Same as the page fault handler, access counters G2H cannot be processed the
+ * G2H worker under the CT lock. Again we use a buffer to sink access counter
+ * G2H. Unlike page faults there is no upper bound so if the buffer is full we
+ * simply drop the G2H. Access counters are a best case optimization and it is
+ * safe to drop these unlike page faults.
+ *
+ * The access counter handler itself is rather simple flow is below.
+ *
+ * .. code-block::
+ *
+ * Lookup VM from ASID in access counter G2H
+ * Lock VM global lock in read mode
+ * Lookup VMA from address in access counter G2H
+ * If userptr, bail nothing to do
+ * Lock VM & BO dma-resv locks
+ * Issue migration to VRAM
+ * Unlock all
+ *
+ * Notice no rebind is issued in the access counter handler as the rebind will
+ * be issued on next page fault.
+ *
+ * Cavets with eviction / user pointer invalidation
+ * ------------------------------------------------
+ *
+ * In the case of eviction and user pointer invalidation on a faulting VM, there
+ * is no need to issue a rebind rather we just need to blow away the page tables
+ * for the VMAs and the page fault handler will rebind the VMAs when they fault.
+ * The cavet is to update / read the page table structure the VM global lock is
+ * neeeed. In both the case of eviction and user pointer invalidation locks are
+ * held which make acquiring the VM global lock impossible. To work around this
+ * every VMA maintains a list of leaf page table entries which should be written
+ * to zero to blow away the VMA's page tables. After writing zero to these
+ * entries a blocking TLB invalidate is issued. At this point it is safe for the
+ * kernel to move the VMA's memory around. This is a necessary lockless
+ * algorithm and is safe as leafs cannot be changed while either an eviction or
+ * userptr invalidation is occurring.
+ *
+ * Locking
+ * =======
+ *
+ * VM locking protects all of the core data paths (bind operations, execs,
+ * evictions, and compute mode rebind worker) in XE.
+ *
+ * Locks
+ * -----
+ *
+ * VM global lock (vm->lock) - rw semaphore lock. Outer most lock which protects
+ * the list of userptrs mapped in the VM, the list of engines using this VM, and
+ * the array of external BOs mapped in the VM. When adding or removing any of the
+ * aforemented state from the VM should acquire this lock in write mode. The VM
+ * bind path also acquires this lock in write while the exec / compute mode
+ * rebind worker acquire this lock in read mode.
+ *
+ * VM dma-resv lock (vm->ttm.base.resv->lock) - WW lock. Protects VM dma-resv
+ * slots which is shared with any private BO in the VM. Expected to be acquired
+ * during VM binds, execs, and compute mode rebind worker. This lock is also
+ * held when private BOs are being evicted.
+ *
+ * external BO dma-resv lock (bo->ttm.base.resv->lock) - WW lock. Protects
+ * external BO dma-resv slots. Expected to be acquired during VM binds (in
+ * addition to the VM dma-resv lock). All external BO dma-locks within a VM are
+ * expected to be acquired (in addition to the VM dma-resv lock) during execs
+ * and the compute mode rebind worker. This lock is also held when an external
+ * BO is being evicted.
+ *
+ * Putting it all together
+ * -----------------------
+ *
+ * 1. An exec and bind operation with the same VM can't be executing at the same
+ * time (vm->lock).
+ *
+ * 2. A compute mode rebind worker and bind operation with the same VM can't be
+ * executing at the same time (vm->lock).
+ *
+ * 3. We can't add / remove userptrs or external BOs to a VM while an exec with
+ * the same VM is executing (vm->lock).
+ *
+ * 4. We can't add / remove userptrs, external BOs, or engines to a VM while a
+ * compute mode rebind worker with the same VM is executing (vm->lock).
+ *
+ * 5. Evictions within a VM can't be happen while an exec with the same VM is
+ * executing (dma-resv locks).
+ *
+ * 6. Evictions within a VM can't be happen while a compute mode rebind worker
+ * with the same VM is executing (dma-resv locks).
+ *
+ * dma-resv usage
+ * ==============
+ *
+ * As previously stated to enforce the ordering of kernel ops (eviction, userptr
+ * invalidation, munmap style unbinds which result in a rebind), rebinds during
+ * execs, execs, and resumes in the rebind worker we use both the VMs and
+ * external BOs dma-resv slots. Let try to make this as clear as possible.
+ *
+ * Slot installation
+ * -----------------
+ *
+ * 1. Jobs from kernel ops install themselves into the DMA_RESV_USAGE_KERNEL
+ * slot of either an external BO or VM (depends on if kernel op is operating on
+ * an external or private BO)
+ *
+ * 2. In non-compute mode, jobs from execs install themselves into the
+ * DMA_RESV_USAGE_BOOKKEEP slot of the VM
+ *
+ * 3. In non-compute mode, jobs from execs install themselves into the
+ * DMA_RESV_USAGE_WRITE slot of all external BOs in the VM
+ *
+ * 4. Jobs from binds install themselves into the DMA_RESV_USAGE_BOOKKEEP slot
+ * of the VM
+ *
+ * 5. Jobs from binds install themselves into the DMA_RESV_USAGE_BOOKKEEP slot
+ * of the external BO (if the bind is to an external BO, this is addition to #4)
+ *
+ * 6. Every engine using a compute mode VM has a preempt fence in installed into
+ * the DMA_RESV_USAGE_PREEMPT_FENCE slot of the VM
+ *
+ * 7. Every engine using a compute mode VM has a preempt fence in installed into
+ * the DMA_RESV_USAGE_PREEMPT_FENCE slot of all the external BOs in the VM
+ *
+ * Slot waiting
+ * ------------
+ *
+ * 1. The exection of all jobs from kernel ops shall wait on all slots
+ * (DMA_RESV_USAGE_PREEMPT_FENCE) of either an external BO or VM (depends on if
+ * kernel op is operating on external or private BO)
+ *
+ * 2. In non-compute mode, the exection of all jobs from rebinds in execs shall
+ * wait on the DMA_RESV_USAGE_KERNEL slot of either an external BO or VM
+ * (depends on if the rebind is operatiing on an external or private BO)
+ *
+ * 3. In non-compute mode, the exection of all jobs from execs shall wait on the
+ * last rebind job
+ *
+ * 4. In compute mode, the exection of all jobs from rebinds in the rebind
+ * worker shall wait on the DMA_RESV_USAGE_KERNEL slot of either an external BO
+ * or VM (depends on if rebind is operating on external or private BO)
+ *
+ * 5. In compute mode, resumes in rebind worker shall wait on last rebind fence
+ *
+ * 6. In compute mode, resumes in rebind worker shall wait on the
+ * DMA_RESV_USAGE_KERNEL slot of the VM
+ *
+ * Putting it all together
+ * -----------------------
+ *
+ * 1. New jobs from kernel ops are blocked behind any existing jobs from
+ * non-compute mode execs
+ *
+ * 2. New jobs from non-compute mode execs are blocked behind any existing jobs
+ * from kernel ops and rebinds
+ *
+ * 3. New jobs from kernel ops are blocked behind all preempt fences signaling in
+ * compute mode
+ *
+ * 4. Compute mode engine resumes are blocked behind any existing jobs from
+ * kernel ops and rebinds
+ *
+ * Future work
+ * ===========
+ *
+ * Support large pages for sysmem and userptr.
+ *
+ * Update page faults to handle BOs are page level grainularity (e.g. part of BO
+ * could be in system memory while another part could be in VRAM).
+ *
+ * Page fault handler likely we be optimized a bit more (e.g. Rebinds always
+ * wait on the dma-resv kernel slots of VM or BO, technically we only have to
+ * wait the BO moving. If using a job to do the rebind, we could not block in
+ * the page fault handler rather attach a callback to fence of the rebind job to
+ * signal page fault complete. Our handling of short circuting for atomic faults
+ * for bound VMAs could be better. etc...). We can tune all of this once we have
+ * benchmarks / performance number from workloads up and running.
+ */
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_vm_types.h b/drivers/gpu/drm/xe/xe_vm_types.h
new file mode 100644
index 000000000000..63e8a50b88e9
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_vm_types.h
@@ -0,0 +1,373 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_VM_TYPES_H_
+#define _XE_VM_TYPES_H_
+
+#include <drm/drm_gpuvm.h>
+
+#include <linux/dma-resv.h>
+#include <linux/kref.h>
+#include <linux/mmu_notifier.h>
+#include <linux/scatterlist.h>
+
+#include "xe_device_types.h"
+#include "xe_pt_types.h"
+#include "xe_range_fence.h"
+
+struct xe_bo;
+struct xe_sync_entry;
+struct xe_vm;
+
+#define TEST_VM_ASYNC_OPS_ERROR
+#define FORCE_ASYNC_OP_ERROR BIT(31)
+
+#define XE_VMA_READ_ONLY DRM_GPUVA_USERBITS
+#define XE_VMA_DESTROYED (DRM_GPUVA_USERBITS << 1)
+#define XE_VMA_ATOMIC_PTE_BIT (DRM_GPUVA_USERBITS << 2)
+#define XE_VMA_FIRST_REBIND (DRM_GPUVA_USERBITS << 3)
+#define XE_VMA_LAST_REBIND (DRM_GPUVA_USERBITS << 4)
+#define XE_VMA_PTE_4K (DRM_GPUVA_USERBITS << 5)
+#define XE_VMA_PTE_2M (DRM_GPUVA_USERBITS << 6)
+#define XE_VMA_PTE_1G (DRM_GPUVA_USERBITS << 7)
+
+/** struct xe_userptr - User pointer */
+struct xe_userptr {
+ /** @invalidate_link: Link for the vm::userptr.invalidated list */
+ struct list_head invalidate_link;
+ /**
+ * @notifier: MMU notifier for user pointer (invalidation call back)
+ */
+ struct mmu_interval_notifier notifier;
+ /** @sgt: storage for a scatter gather table */
+ struct sg_table sgt;
+ /** @sg: allocated scatter gather table */
+ struct sg_table *sg;
+ /** @notifier_seq: notifier sequence number */
+ unsigned long notifier_seq;
+ /**
+ * @initial_bind: user pointer has been bound at least once.
+ * write: vm->userptr.notifier_lock in read mode and vm->resv held.
+ * read: vm->userptr.notifier_lock in write mode or vm->resv held.
+ */
+ bool initial_bind;
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+ u32 divisor;
+#endif
+};
+
+struct xe_vma {
+ /** @gpuva: Base GPUVA object */
+ struct drm_gpuva gpuva;
+
+ /**
+ * @combined_links: links into lists which are mutually exclusive.
+ * Locking: vm lock in write mode OR vm lock in read mode and the vm's
+ * resv.
+ */
+ union {
+ /** @userptr: link into VM repin list if userptr. */
+ struct list_head userptr;
+ /** @rebind: link into VM if this VMA needs rebinding. */
+ struct list_head rebind;
+ /** @destroy: link to contested list when VM is being closed. */
+ struct list_head destroy;
+ } combined_links;
+
+ union {
+ /** @destroy_cb: callback to destroy VMA when unbind job is done */
+ struct dma_fence_cb destroy_cb;
+ /** @destroy_work: worker to destroy this BO */
+ struct work_struct destroy_work;
+ };
+
+ /** @usm: unified shared memory state */
+ struct {
+ /** @tile_invalidated: VMA has been invalidated */
+ u8 tile_invalidated;
+ } usm;
+
+ /** @tile_mask: Tile mask of where to create binding for this VMA */
+ u8 tile_mask;
+
+ /**
+ * @tile_present: GT mask of binding are present for this VMA.
+ * protected by vm->lock, vm->resv and for userptrs,
+ * vm->userptr.notifier_lock for writing. Needs either for reading,
+ * but if reading is done under the vm->lock only, it needs to be held
+ * in write mode.
+ */
+ u8 tile_present;
+
+ /**
+ * @pat_index: The pat index to use when encoding the PTEs for this vma.
+ */
+ u16 pat_index;
+
+ /**
+ * @userptr: user pointer state, only allocated for VMAs that are
+ * user pointers
+ */
+ struct xe_userptr userptr;
+};
+
+struct xe_device;
+
+struct xe_vm {
+ /** @gpuvm: base GPUVM used to track VMAs */
+ struct drm_gpuvm gpuvm;
+
+ struct xe_device *xe;
+
+ /* exec queue used for (un)binding vma's */
+ struct xe_exec_queue *q[XE_MAX_TILES_PER_DEVICE];
+
+ /** @lru_bulk_move: Bulk LRU move list for this VM's BOs */
+ struct ttm_lru_bulk_move lru_bulk_move;
+
+ u64 size;
+
+ struct xe_pt *pt_root[XE_MAX_TILES_PER_DEVICE];
+ struct xe_pt *scratch_pt[XE_MAX_TILES_PER_DEVICE][XE_VM_MAX_LEVEL];
+
+ /**
+ * @flags: flags for this VM, statically setup a creation time aside
+ * from XE_VM_FLAG_BANNED which requires vm->lock to set / read safely
+ */
+#define XE_VM_FLAG_64K BIT(0)
+#define XE_VM_FLAG_LR_MODE BIT(1)
+#define XE_VM_FLAG_MIGRATION BIT(2)
+#define XE_VM_FLAG_SCRATCH_PAGE BIT(3)
+#define XE_VM_FLAG_FAULT_MODE BIT(4)
+#define XE_VM_FLAG_BANNED BIT(5)
+#define XE_VM_FLAG_TILE_ID(flags) FIELD_GET(GENMASK(7, 6), flags)
+#define XE_VM_FLAG_SET_TILE_ID(tile) FIELD_PREP(GENMASK(7, 6), (tile)->id)
+ unsigned long flags;
+
+ /** @composite_fence_ctx: context composite fence */
+ u64 composite_fence_ctx;
+ /** @composite_fence_seqno: seqno for composite fence */
+ u32 composite_fence_seqno;
+
+ /**
+ * @lock: outer most lock, protects objects of anything attached to this
+ * VM
+ */
+ struct rw_semaphore lock;
+
+ /**
+ * @rebind_list: list of VMAs that need rebinding. Protected by the
+ * vm->lock in write mode, OR (the vm->lock in read mode and the
+ * vm resv).
+ */
+ struct list_head rebind_list;
+
+ /** @rebind_fence: rebind fence from execbuf */
+ struct dma_fence *rebind_fence;
+
+ /**
+ * @destroy_work: worker to destroy VM, needed as a dma_fence signaling
+ * from an irq context can be last put and the destroy needs to be able
+ * to sleep.
+ */
+ struct work_struct destroy_work;
+
+ /**
+ * @rftree: range fence tree to track updates to page table structure.
+ * Used to implement conflict tracking between independent bind engines.
+ */
+ struct xe_range_fence_tree rftree[XE_MAX_TILES_PER_DEVICE];
+
+ /** @async_ops: async VM operations (bind / unbinds) */
+ struct {
+ /** @list: list of pending async VM ops */
+ struct list_head pending;
+ /** @work: worker to execute async VM ops */
+ struct work_struct work;
+ /** @lock: protects list of pending async VM ops and fences */
+ spinlock_t lock;
+ /** @fence: fence state */
+ struct {
+ /** @context: context of async fence */
+ u64 context;
+ /** @seqno: seqno of async fence */
+ u32 seqno;
+ } fence;
+ /** @error: error state for async VM ops */
+ int error;
+ /**
+ * @munmap_rebind_inflight: an munmap style VM bind is in the
+ * middle of a set of ops which requires a rebind at the end.
+ */
+ bool munmap_rebind_inflight;
+ } async_ops;
+
+ const struct xe_pt_ops *pt_ops;
+
+ /** @userptr: user pointer state */
+ struct {
+ /**
+ * @userptr.repin_list: list of VMAs which are user pointers,
+ * and needs repinning. Protected by @lock.
+ */
+ struct list_head repin_list;
+ /**
+ * @notifier_lock: protects notifier in write mode and
+ * submission in read mode.
+ */
+ struct rw_semaphore notifier_lock;
+ /**
+ * @userptr.invalidated_lock: Protects the
+ * @userptr.invalidated list.
+ */
+ spinlock_t invalidated_lock;
+ /**
+ * @userptr.invalidated: List of invalidated userptrs, not yet
+ * picked
+ * up for revalidation. Protected from access with the
+ * @invalidated_lock. Removing items from the list
+ * additionally requires @lock in write mode, and adding
+ * items to the list requires the @userptr.notifer_lock in
+ * write mode.
+ */
+ struct list_head invalidated;
+ } userptr;
+
+ /** @preempt: preempt state */
+ struct {
+ /**
+ * @min_run_period_ms: The minimum run period before preempting
+ * an engine again
+ */
+ s64 min_run_period_ms;
+ /** @exec_queues: list of exec queues attached to this VM */
+ struct list_head exec_queues;
+ /** @num_exec_queues: number exec queues attached to this VM */
+ int num_exec_queues;
+ /**
+ * @rebind_deactivated: Whether rebind has been temporarily deactivated
+ * due to no work available. Protected by the vm resv.
+ */
+ bool rebind_deactivated;
+ /**
+ * @rebind_work: worker to rebind invalidated userptrs / evicted
+ * BOs
+ */
+ struct work_struct rebind_work;
+ } preempt;
+
+ /** @um: unified memory state */
+ struct {
+ /** @asid: address space ID, unique to each VM */
+ u32 asid;
+ /**
+ * @last_fault_vma: Last fault VMA, used for fast lookup when we
+ * get a flood of faults to the same VMA
+ */
+ struct xe_vma *last_fault_vma;
+ } usm;
+
+ /** @error_capture: allow to track errors */
+ struct {
+ /** @capture_once: capture only one error per VM */
+ bool capture_once;
+ } error_capture;
+
+ /** @batch_invalidate_tlb: Always invalidate TLB before batch start */
+ bool batch_invalidate_tlb;
+ /** @xef: XE file handle for tracking this VM's drm client */
+ struct xe_file *xef;
+};
+
+/** struct xe_vma_op_map - VMA map operation */
+struct xe_vma_op_map {
+ /** @vma: VMA to map */
+ struct xe_vma *vma;
+ /** @immediate: Immediate bind */
+ bool immediate;
+ /** @read_only: Read only */
+ bool read_only;
+ /** @is_null: is NULL binding */
+ bool is_null;
+ /** @pat_index: The pat index to use for this operation. */
+ u16 pat_index;
+};
+
+/** struct xe_vma_op_remap - VMA remap operation */
+struct xe_vma_op_remap {
+ /** @prev: VMA preceding part of a split mapping */
+ struct xe_vma *prev;
+ /** @next: VMA subsequent part of a split mapping */
+ struct xe_vma *next;
+ /** @start: start of the VMA unmap */
+ u64 start;
+ /** @range: range of the VMA unmap */
+ u64 range;
+ /** @skip_prev: skip prev rebind */
+ bool skip_prev;
+ /** @skip_next: skip next rebind */
+ bool skip_next;
+ /** @unmap_done: unmap operation in done */
+ bool unmap_done;
+};
+
+/** struct xe_vma_op_prefetch - VMA prefetch operation */
+struct xe_vma_op_prefetch {
+ /** @region: memory region to prefetch to */
+ u32 region;
+};
+
+/** enum xe_vma_op_flags - flags for VMA operation */
+enum xe_vma_op_flags {
+ /** @XE_VMA_OP_FIRST: first VMA operation for a set of syncs */
+ XE_VMA_OP_FIRST = BIT(0),
+ /** @XE_VMA_OP_LAST: last VMA operation for a set of syncs */
+ XE_VMA_OP_LAST = BIT(1),
+ /** @XE_VMA_OP_COMMITTED: VMA operation committed */
+ XE_VMA_OP_COMMITTED = BIT(2),
+ /** @XE_VMA_OP_PREV_COMMITTED: Previous VMA operation committed */
+ XE_VMA_OP_PREV_COMMITTED = BIT(3),
+ /** @XE_VMA_OP_NEXT_COMMITTED: Next VMA operation committed */
+ XE_VMA_OP_NEXT_COMMITTED = BIT(4),
+};
+
+/** struct xe_vma_op - VMA operation */
+struct xe_vma_op {
+ /** @base: GPUVA base operation */
+ struct drm_gpuva_op base;
+ /**
+ * @ops: GPUVA ops, when set call drm_gpuva_ops_free after this
+ * operations is processed
+ */
+ struct drm_gpuva_ops *ops;
+ /** @q: exec queue for this operation */
+ struct xe_exec_queue *q;
+ /**
+ * @syncs: syncs for this operation, only used on first and last
+ * operation
+ */
+ struct xe_sync_entry *syncs;
+ /** @num_syncs: number of syncs */
+ u32 num_syncs;
+ /** @link: async operation link */
+ struct list_head link;
+ /** @flags: operation flags */
+ enum xe_vma_op_flags flags;
+
+#ifdef TEST_VM_ASYNC_OPS_ERROR
+ /** @inject_error: inject error to test async op error handling */
+ bool inject_error;
+#endif
+
+ union {
+ /** @map: VMA map operation specific data */
+ struct xe_vma_op_map map;
+ /** @remap: VMA remap operation specific data */
+ struct xe_vma_op_remap remap;
+ /** @prefetch: VMA prefetch operation specific data */
+ struct xe_vma_op_prefetch prefetch;
+ };
+};
+#endif
diff --git a/drivers/gpu/drm/xe/xe_wa.c b/drivers/gpu/drm/xe/xe_wa.c
new file mode 100644
index 000000000000..5f61dd87c586
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wa.c
@@ -0,0 +1,895 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_wa.h"
+
+#include <drm/drm_managed.h>
+#include <kunit/visibility.h>
+#include <linux/compiler_types.h>
+
+#include "generated/xe_wa_oob.h"
+#include "regs/xe_engine_regs.h"
+#include "regs/xe_gt_regs.h"
+#include "regs/xe_regs.h"
+#include "xe_device_types.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_hw_engine_types.h"
+#include "xe_mmio.h"
+#include "xe_platform_types.h"
+#include "xe_rtp.h"
+#include "xe_step.h"
+
+/**
+ * DOC: Hardware workarounds
+ *
+ * Hardware workarounds are register programming documented to be executed in
+ * the driver that fall outside of the normal programming sequences for a
+ * platform. There are some basic categories of workarounds, depending on
+ * how/when they are applied:
+ *
+ * - LRC workarounds: workarounds that touch registers that are
+ * saved/restored to/from the HW context image. The list is emitted (via Load
+ * Register Immediate commands) once when initializing the device and saved in
+ * the default context. That default context is then used on every context
+ * creation to have a "primed golden context", i.e. a context image that
+ * already contains the changes needed to all the registers.
+ *
+ * - Engine workarounds: the list of these WAs is applied whenever the specific
+ * engine is reset. It's also possible that a set of engine classes share a
+ * common power domain and they are reset together. This happens on some
+ * platforms with render and compute engines. In this case (at least) one of
+ * them need to keeep the workaround programming: the approach taken in the
+ * driver is to tie those workarounds to the first compute/render engine that
+ * is registered. When executing with GuC submission, engine resets are
+ * outside of kernel driver control, hence the list of registers involved in
+ * written once, on engine initialization, and then passed to GuC, that
+ * saves/restores their values before/after the reset takes place. See
+ * ``drivers/gpu/drm/xe/xe_guc_ads.c`` for reference.
+ *
+ * - GT workarounds: the list of these WAs is applied whenever these registers
+ * revert to their default values: on GPU reset, suspend/resume [1]_, etc.
+ *
+ * - Register whitelist: some workarounds need to be implemented in userspace,
+ * but need to touch privileged registers. The whitelist in the kernel
+ * instructs the hardware to allow the access to happen. From the kernel side,
+ * this is just a special case of a MMIO workaround (as we write the list of
+ * these to/be-whitelisted registers to some special HW registers).
+ *
+ * - Workaround batchbuffers: buffers that get executed automatically by the
+ * hardware on every HW context restore. These buffers are created and
+ * programmed in the default context so the hardware always go through those
+ * programming sequences when switching contexts. The support for workaround
+ * batchbuffers is enabled these hardware mechanisms:
+ *
+ * #. INDIRECT_CTX: A batchbuffer and an offset are provided in the default
+ * context, pointing the hardware to jump to that location when that offset
+ * is reached in the context restore. Workaround batchbuffer in the driver
+ * currently uses this mechanism for all platforms.
+ *
+ * #. BB_PER_CTX_PTR: A batchbuffer is provided in the default context,
+ * pointing the hardware to a buffer to continue executing after the
+ * engine registers are restored in a context restore sequence. This is
+ * currently not used in the driver.
+ *
+ * - Other/OOB: There are WAs that, due to their nature, cannot be applied from
+ * a central place. Those are peppered around the rest of the code, as needed.
+ * Workarounds related to the display IP are the main example.
+ *
+ * .. [1] Technically, some registers are powercontext saved & restored, so they
+ * survive a suspend/resume. In practice, writing them again is not too
+ * costly and simplifies things, so it's the approach taken in the driver.
+ *
+ * .. note::
+ * Hardware workarounds in xe work the same way as in i915, with the
+ * difference of how they are maintained in the code. In xe it uses the
+ * xe_rtp infrastructure so the workarounds can be kept in tables, following
+ * a more declarative approach rather than procedural.
+ */
+
+#undef XE_REG_MCR
+#define XE_REG_MCR(...) XE_REG(__VA_ARGS__, .mcr = 1)
+
+__diag_push();
+__diag_ignore_all("-Woverride-init", "Allow field overrides in table");
+
+static const struct xe_rtp_entry_sr gt_was[] = {
+ { XE_RTP_NAME("14011060649"),
+ XE_RTP_RULES(MEDIA_VERSION_RANGE(1200, 1255),
+ ENGINE_CLASS(VIDEO_DECODE),
+ FUNC(xe_rtp_match_even_instance)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F10(0), IECPUNIT_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+ { XE_RTP_NAME("14011059788"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210)),
+ XE_RTP_ACTIONS(SET(DFR_RATIO_EN_AND_CHICKEN, DFR_DISABLE))
+ },
+ { XE_RTP_NAME("14015795083"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1260)),
+ XE_RTP_ACTIONS(CLR(MISCCPCTL, DOP_CLOCK_GATE_RENDER_ENABLE))
+ },
+
+ /* DG1 */
+
+ { XE_RTP_NAME("1409420604"),
+ XE_RTP_RULES(PLATFORM(DG1)),
+ XE_RTP_ACTIONS(SET(SUBSLICE_UNIT_LEVEL_CLKGATE2, CPSSUNIT_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("1408615072"),
+ XE_RTP_RULES(PLATFORM(DG1)),
+ XE_RTP_ACTIONS(SET(UNSLICE_UNIT_LEVEL_CLKGATE2, VSUNIT_CLKGATE2_DIS))
+ },
+
+ /* DG2 */
+
+ { XE_RTP_NAME("16010515920"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10),
+ GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(VIDEO_DECODE)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F18(0), ALNUNIT_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+ { XE_RTP_NAME("22010523718"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10)),
+ XE_RTP_ACTIONS(SET(UNSLICE_UNIT_LEVEL_CLKGATE, CG3DDISCFEG_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14011006942"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10)),
+ XE_RTP_ACTIONS(SET(SUBSLICE_UNIT_LEVEL_CLKGATE, DSS_ROUTER_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14012362059"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_MERT_MOD_CTRL, FORCE_MISS_FTLB))
+ },
+ { XE_RTP_NAME("14012362059"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_MERT_MOD_CTRL, FORCE_MISS_FTLB))
+ },
+ { XE_RTP_NAME("14010948348"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(UNSLCGCTL9430, MSQDUNIT_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14011037102"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(UNSLCGCTL9444, LTCDD_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14011371254"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_SLICE_UNIT_LEVEL_CLKGATE, NODEDSS_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14011431319"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(UNSLCGCTL9440,
+ GAMTLBOACS_CLKGATE_DIS |
+ GAMTLBVDBOX7_CLKGATE_DIS | GAMTLBVDBOX6_CLKGATE_DIS |
+ GAMTLBVDBOX5_CLKGATE_DIS | GAMTLBVDBOX4_CLKGATE_DIS |
+ GAMTLBVDBOX3_CLKGATE_DIS | GAMTLBVDBOX2_CLKGATE_DIS |
+ GAMTLBVDBOX1_CLKGATE_DIS | GAMTLBVDBOX0_CLKGATE_DIS |
+ GAMTLBKCR_CLKGATE_DIS | GAMTLBGUC_CLKGATE_DIS |
+ GAMTLBBLT_CLKGATE_DIS),
+ SET(UNSLCGCTL9444,
+ GAMTLBGFXA0_CLKGATE_DIS | GAMTLBGFXA1_CLKGATE_DIS |
+ GAMTLBCOMPA0_CLKGATE_DIS | GAMTLBCOMPA1_CLKGATE_DIS |
+ GAMTLBCOMPB0_CLKGATE_DIS | GAMTLBCOMPB1_CLKGATE_DIS |
+ GAMTLBCOMPC0_CLKGATE_DIS | GAMTLBCOMPC1_CLKGATE_DIS |
+ GAMTLBCOMPD0_CLKGATE_DIS | GAMTLBCOMPD1_CLKGATE_DIS |
+ GAMTLBMERT_CLKGATE_DIS |
+ GAMTLBVEBOX3_CLKGATE_DIS | GAMTLBVEBOX2_CLKGATE_DIS |
+ GAMTLBVEBOX1_CLKGATE_DIS | GAMTLBVEBOX0_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14010569222"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(UNSLICE_UNIT_LEVEL_CLKGATE, GAMEDIA_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14011028019"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(SSMCGCTL9530, RTFUNIT_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14010680813"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_GAMSTLB_CTRL,
+ CONTROL_BLOCK_CLKGATE_DIS |
+ EGRESS_BLOCK_CLKGATE_DIS |
+ TAG_BLOCK_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14014830051"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(CLR(SARB_CHICKEN1, COMP_CKN_IN))
+ },
+ { XE_RTP_NAME("18018781329"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(RENDER_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(COMP_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(XEHP_VDBX_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(XEHP_VEBX_MOD_CTRL, FORCE_MISS_FTLB))
+ },
+ { XE_RTP_NAME("1509235366"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_GAMCNTRL_CTRL,
+ INVALIDATION_BROADCAST_MODE_DIS |
+ GLOBAL_INVALIDATION_MODE))
+ },
+ { XE_RTP_NAME("14010648519"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_L3NODEARBCFG, XEHP_LNESPARE))
+ },
+
+ /* PVC */
+
+ { XE_RTP_NAME("18018781329"),
+ XE_RTP_RULES(PLATFORM(PVC)),
+ XE_RTP_ACTIONS(SET(RENDER_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(COMP_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(XEHP_VDBX_MOD_CTRL, FORCE_MISS_FTLB),
+ SET(XEHP_VEBX_MOD_CTRL, FORCE_MISS_FTLB))
+ },
+ { XE_RTP_NAME("16016694945"),
+ XE_RTP_RULES(PLATFORM(PVC)),
+ XE_RTP_ACTIONS(SET(XEHPC_LNCFMISCCFGREG0, XEHPC_OVRLSCCC))
+ },
+
+ /* Xe_LPG */
+
+ { XE_RTP_NAME("14015795083"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(CLR(MISCCPCTL, DOP_CLOCK_GATE_RENDER_ENABLE))
+ },
+ { XE_RTP_NAME("14018575942"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271)),
+ XE_RTP_ACTIONS(SET(COMP_MOD_CTRL, FORCE_MISS_FTLB))
+ },
+ { XE_RTP_NAME("22016670082"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271)),
+ XE_RTP_ACTIONS(SET(SQCNT1, ENFORCE_RAR))
+ },
+
+ /* Xe_LPM+ */
+
+ { XE_RTP_NAME("16021867713"),
+ XE_RTP_RULES(MEDIA_VERSION(1300),
+ ENGINE_CLASS(VIDEO_DECODE)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F1C(0), MFXPIPE_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+ { XE_RTP_NAME("22016670082"),
+ XE_RTP_RULES(MEDIA_VERSION(1300)),
+ XE_RTP_ACTIONS(SET(XELPMP_SQCNT1, ENFORCE_RAR))
+ },
+
+ /* Xe2_LPG */
+
+ { XE_RTP_NAME("16020975621"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_SLICE_UNIT_LEVEL_CLKGATE, SBEUNIT_CLKGATE_DIS))
+ },
+ { XE_RTP_NAME("14018157293"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHPC_L3CLOS_MASK(0), ~0),
+ SET(XEHPC_L3CLOS_MASK(1), ~0),
+ SET(XEHPC_L3CLOS_MASK(2), ~0),
+ SET(XEHPC_L3CLOS_MASK(3), ~0))
+ },
+
+ /* Xe2_LPM */
+
+ { XE_RTP_NAME("14017421178"),
+ XE_RTP_RULES(MEDIA_VERSION(2000),
+ ENGINE_CLASS(VIDEO_DECODE)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F10(0), IECPUNIT_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+ { XE_RTP_NAME("16021867713"),
+ XE_RTP_RULES(MEDIA_VERSION(2000),
+ ENGINE_CLASS(VIDEO_DECODE)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F1C(0), MFXPIPE_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+ { XE_RTP_NAME("14019449301"),
+ XE_RTP_RULES(MEDIA_VERSION(2000), ENGINE_CLASS(VIDEO_DECODE)),
+ XE_RTP_ACTIONS(SET(VDBOX_CGCTL3F08(0), CG3DDISHRS_CLKGATE_DIS)),
+ XE_RTP_ENTRY_FLAG(FOREACH_ENGINE),
+ },
+
+ {}
+};
+
+static const struct xe_rtp_entry_sr engine_was[] = {
+ { XE_RTP_NAME("22010931296, 18011464164, 14010919138"),
+ XE_RTP_RULES(GRAPHICS_VERSION(1200), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(FF_THREAD_MODE(RENDER_RING_BASE),
+ FF_TESSELATION_DOP_GATE_DISABLE))
+ },
+ { XE_RTP_NAME("1409804808"),
+ XE_RTP_RULES(GRAPHICS_VERSION(1200),
+ ENGINE_CLASS(RENDER),
+ IS_INTEGRATED),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN2, PUSH_CONST_DEREF_HOLD_DIS))
+ },
+ { XE_RTP_NAME("14010229206, 1409085225"),
+ XE_RTP_RULES(GRAPHICS_VERSION(1200),
+ ENGINE_CLASS(RENDER),
+ IS_INTEGRATED),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4, DISABLE_TDL_PUSH))
+ },
+ { XE_RTP_NAME("1606931601"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN2, DISABLE_EARLY_READ))
+ },
+ { XE_RTP_NAME("14010826681, 1606700617, 22010271021, 18019627453"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1255), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(CS_DEBUG_MODE1(RENDER_RING_BASE),
+ FF_DOP_CLOCK_GATE_DISABLE))
+ },
+ { XE_RTP_NAME("1406941453"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(SAMPLER_MODE, ENABLE_SMALLPL))
+ },
+ { XE_RTP_NAME("FtrPerCtxtPreemptionGranularityControl"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1250), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(FF_SLICE_CS_CHICKEN1(RENDER_RING_BASE),
+ FFSC_PERCTX_PREEMPT_CTRL))
+ },
+
+ /* TGL */
+
+ { XE_RTP_NAME("1607297627, 1607030317, 1607186500"),
+ XE_RTP_RULES(PLATFORM(TIGERLAKE), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(RING_PSMI_CTL(RENDER_RING_BASE),
+ WAIT_FOR_EVENT_POWER_DOWN_DISABLE |
+ RC_SEMA_IDLE_MSG_DISABLE))
+ },
+
+ /* RKL */
+
+ { XE_RTP_NAME("1607297627, 1607030317, 1607186500"),
+ XE_RTP_RULES(PLATFORM(ROCKETLAKE), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(RING_PSMI_CTL(RENDER_RING_BASE),
+ WAIT_FOR_EVENT_POWER_DOWN_DISABLE |
+ RC_SEMA_IDLE_MSG_DISABLE))
+ },
+
+ /* ADL-P */
+
+ { XE_RTP_NAME("1607297627, 1607030317, 1607186500"),
+ XE_RTP_RULES(PLATFORM(ALDERLAKE_P), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(RING_PSMI_CTL(RENDER_RING_BASE),
+ WAIT_FOR_EVENT_POWER_DOWN_DISABLE |
+ RC_SEMA_IDLE_MSG_DISABLE))
+ },
+
+ /* DG2 */
+
+ { XE_RTP_NAME("22013037850"),
+ XE_RTP_RULES(PLATFORM(DG2), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW,
+ DISABLE_128B_EVICTION_COMMAND_UDW))
+ },
+ { XE_RTP_NAME("22014226127"),
+ XE_RTP_RULES(PLATFORM(DG2), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, DISABLE_D8_D16_COASLESCE))
+ },
+ { XE_RTP_NAME("18017747507"),
+ XE_RTP_RULES(PLATFORM(DG2), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(VFG_PREEMPTION_CHICKEN,
+ POLYGON_TRIFAN_LINELOOP_DISABLE))
+ },
+ { XE_RTP_NAME("22012826095, 22013059131"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(B0, C0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(FIELD_SET(LSC_CHICKEN_BIT_0_UDW,
+ MAXREQS_PER_BANK,
+ REG_FIELD_PREP(MAXREQS_PER_BANK, 2)))
+ },
+ { XE_RTP_NAME("22012826095, 22013059131"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(FIELD_SET(LSC_CHICKEN_BIT_0_UDW,
+ MAXREQS_PER_BANK,
+ REG_FIELD_PREP(MAXREQS_PER_BANK, 2)))
+ },
+ { XE_RTP_NAME("22013059131"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(B0, C0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, FORCE_1_SUB_MESSAGE_PER_FRAGMENT))
+ },
+ { XE_RTP_NAME("22013059131"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, FORCE_1_SUB_MESSAGE_PER_FRAGMENT))
+ },
+ { XE_RTP_NAME("14010918519"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW,
+ FORCE_SLM_FENCE_SCOPE_TO_TILE |
+ FORCE_UGM_FENCE_SCOPE_TO_TILE,
+ /*
+ * Ignore read back as it always returns 0 in these
+ * steps
+ */
+ .read_mask = 0))
+ },
+ { XE_RTP_NAME("14015227452"),
+ XE_RTP_RULES(PLATFORM(DG2),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4, XEHP_DIS_BBL_SYSPIPE))
+ },
+ { XE_RTP_NAME("16015675438"),
+ XE_RTP_RULES(PLATFORM(DG2),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(FF_SLICE_CS_CHICKEN2(RENDER_RING_BASE),
+ PERF_FIX_BALANCING_CFE_DISABLE))
+ },
+ { XE_RTP_NAME("18028616096"),
+ XE_RTP_RULES(PLATFORM(DG2),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW, UGM_FRAGMENT_THRESHOLD_TO_3))
+ },
+ { XE_RTP_NAME("16011620976, 22015475538"),
+ XE_RTP_RULES(PLATFORM(DG2),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW, DIS_CHAIN_2XSIMD8))
+ },
+ { XE_RTP_NAME("22012654132"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, C0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(CACHE_MODE_SS, ENABLE_PREFETCH_INTO_IC,
+ /*
+ * Register can't be read back for verification on
+ * DG2 due to Wa_14012342262
+ */
+ .read_mask = 0))
+ },
+ { XE_RTP_NAME("22012654132"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(CACHE_MODE_SS, ENABLE_PREFETCH_INTO_IC,
+ /*
+ * Register can't be read back for verification on
+ * DG2 due to Wa_14012342262
+ */
+ .read_mask = 0))
+ },
+ { XE_RTP_NAME("1509727124"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(SAMPLER_MODE, SC_DISABLE_POWER_OPTIMIZATION_EBB))
+ },
+ { XE_RTP_NAME("22012856258"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN2, DISABLE_READ_SUPPRESSION))
+ },
+ { XE_RTP_NAME("14013392000"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN2, ENABLE_LARGE_GRF_MODE))
+ },
+ { XE_RTP_NAME("14012419201"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4,
+ DISABLE_HDR_PAST_PAYLOAD_HOLD_FIX))
+ },
+ { XE_RTP_NAME("14012419201"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4,
+ DISABLE_HDR_PAST_PAYLOAD_HOLD_FIX))
+ },
+ { XE_RTP_NAME("1308578152"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(B0, C0),
+ ENGINE_CLASS(RENDER),
+ FUNC(xe_rtp_match_first_gslice_fused_off)),
+ XE_RTP_ACTIONS(CLR(CS_DEBUG_MODE1(RENDER_RING_BASE),
+ REPLAY_MODE_GRANULARITY))
+ },
+ { XE_RTP_NAME("22010960976, 14013347512"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(CLR(XEHP_HDC_CHICKEN0,
+ LSC_L1_FLUSH_CTL_3D_DATAPORT_FLUSH_EVENTS_MASK))
+ },
+ { XE_RTP_NAME("1608949956, 14010198302"),
+ XE_RTP_RULES(PLATFORM(DG2), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN,
+ MDQ_ARBITRATION_MODE | UGM_BACKUP_MODE))
+ },
+ { XE_RTP_NAME("22010430635"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4,
+ DISABLE_GRF_CLEAR))
+ },
+ { XE_RTP_NAME("14013202645"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(B0, C0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(RT_CTRL, DIS_NULL_QUERY))
+ },
+ { XE_RTP_NAME("14013202645"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(RT_CTRL, DIS_NULL_QUERY))
+ },
+ { XE_RTP_NAME("22012532006"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, C0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(HALF_SLICE_CHICKEN7,
+ DG2_DISABLE_ROUND_ENABLE_ALLOW_FOR_SSLA))
+ },
+ { XE_RTP_NAME("22012532006"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(HALF_SLICE_CHICKEN7,
+ DG2_DISABLE_ROUND_ENABLE_ALLOW_FOR_SSLA))
+ },
+ { XE_RTP_NAME("14015150844"),
+ XE_RTP_RULES(PLATFORM(DG2), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(XEHP_HDC_CHICKEN0, DIS_ATOMIC_CHAINING_TYPED_WRITES,
+ XE_RTP_NOCHECK))
+ },
+
+ /* PVC */
+
+ { XE_RTP_NAME("22014226127"),
+ XE_RTP_RULES(PLATFORM(PVC), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, DISABLE_D8_D16_COASLESCE))
+ },
+ { XE_RTP_NAME("14015227452"),
+ XE_RTP_RULES(PLATFORM(PVC), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4, XEHP_DIS_BBL_SYSPIPE))
+ },
+ { XE_RTP_NAME("16015675438"),
+ XE_RTP_RULES(PLATFORM(PVC), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(FF_SLICE_CS_CHICKEN2(RENDER_RING_BASE),
+ PERF_FIX_BALANCING_CFE_DISABLE))
+ },
+ { XE_RTP_NAME("14014999345"),
+ XE_RTP_RULES(PLATFORM(PVC), ENGINE_CLASS(COMPUTE),
+ GRAPHICS_STEP(B0, C0)),
+ XE_RTP_ACTIONS(SET(CACHE_MODE_SS, DISABLE_ECC))
+ },
+
+ /* Xe_LPG */
+
+ { XE_RTP_NAME("14017856879"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN3, DIS_FIX_EOT1_FLUSH))
+ },
+ { XE_RTP_NAME("14015150844"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(XEHP_HDC_CHICKEN0, DIS_ATOMIC_CHAINING_TYPED_WRITES,
+ XE_RTP_NOCHECK))
+ },
+
+ /* Xe2_LPG */
+
+ { XE_RTP_NAME("18032247524"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, SEQUENTIAL_ACCESS_UPGRADE_DISABLE))
+ },
+ { XE_RTP_NAME("16018712365"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW, XE2_ALLOC_DPA_STARVE_FIX_DIS))
+ },
+ { XE_RTP_NAME("14018957109"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(HALF_SLICE_CHICKEN5, DISABLE_SAMPLE_G_PERFORMANCE))
+ },
+ { XE_RTP_NAME("16021540221"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN4, DISABLE_TDL_PUSH))
+ },
+ { XE_RTP_NAME("14019322943"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0),
+ FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0, TGM_WRITE_EOM_FORCE))
+ },
+ { XE_RTP_NAME("14018471104"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(LSC_CHICKEN_BIT_0_UDW, ENABLE_SMP_LD_RENDER_SURFACE_CONTROL))
+ },
+ { XE_RTP_NAME("16018737384"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), FUNC(xe_rtp_match_first_render_or_compute)),
+ XE_RTP_ACTIONS(SET(ROW_CHICKEN, EARLY_EOT_DIS))
+ },
+ /*
+ * These two workarounds are the same, just applying to different
+ * engines. Although Wa_18032095049 (for the RCS) isn't required on
+ * all steppings, disabling these reports has no impact for our
+ * driver or the GuC, so we go ahead and treat it the same as
+ * Wa_16021639441 which does apply to all steppings.
+ */
+ { XE_RTP_NAME("18032095049, 16021639441"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004)),
+ XE_RTP_ACTIONS(SET(CSFE_CHICKEN1(0),
+ GHWSP_CSB_REPORT_DIS |
+ PPHWSP_CSB_AND_TIMESTAMP_REPORT_DIS,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE)))
+ },
+
+ {}
+};
+
+static const struct xe_rtp_entry_sr lrc_was[] = {
+ { XE_RTP_NAME("1409342910, 14010698770, 14010443199, 1408979724, 1409178076, 1409207793, 1409217633, 1409252684, 1409347922, 1409142259"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210)),
+ XE_RTP_ACTIONS(SET(COMMON_SLICE_CHICKEN3,
+ DISABLE_CPS_AWARE_COLOR_PIPE))
+ },
+ { XE_RTP_NAME("WaDisableGPGPUMidThreadPreemption"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1200, 1210)),
+ XE_RTP_ACTIONS(FIELD_SET(CS_CHICKEN1(RENDER_RING_BASE),
+ PREEMPT_GPGPU_LEVEL_MASK,
+ PREEMPT_GPGPU_THREAD_GROUP_LEVEL))
+ },
+ { XE_RTP_NAME("1806527549"),
+ XE_RTP_RULES(GRAPHICS_VERSION(1200)),
+ XE_RTP_ACTIONS(SET(HIZ_CHICKEN, HZ_DEPTH_TEST_LE_GE_OPT_DISABLE))
+ },
+ { XE_RTP_NAME("1606376872"),
+ XE_RTP_RULES(GRAPHICS_VERSION(1200)),
+ XE_RTP_ACTIONS(SET(COMMON_SLICE_CHICKEN4, DISABLE_TDC_LOAD_BALANCING_CALC))
+ },
+
+ /* DG1 */
+
+ { XE_RTP_NAME("1409044764"),
+ XE_RTP_RULES(PLATFORM(DG1)),
+ XE_RTP_ACTIONS(CLR(COMMON_SLICE_CHICKEN3,
+ DG1_FLOAT_POINT_BLEND_OPT_STRICT_MODE_EN))
+ },
+ { XE_RTP_NAME("22010493298"),
+ XE_RTP_RULES(PLATFORM(DG1)),
+ XE_RTP_ACTIONS(SET(HIZ_CHICKEN,
+ DG1_HZ_READ_SUPPRESSION_OPTIMIZATION_DISABLE))
+ },
+
+ /* DG2 */
+
+ { XE_RTP_NAME("16011186671"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(CLR(VFLSKPD, DIS_MULT_MISS_RD_SQUASH),
+ SET(VFLSKPD, DIS_OVER_FETCH_CACHE))
+ },
+ { XE_RTP_NAME("14010469329"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_COMMON_SLICE_CHICKEN3,
+ XEHP_DUAL_SIMD8_SEQ_MERGE_DISABLE))
+ },
+ { XE_RTP_NAME("14010698770, 22010613112, 22010465075"),
+ XE_RTP_RULES(SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)),
+ XE_RTP_ACTIONS(SET(XEHP_COMMON_SLICE_CHICKEN3,
+ DISABLE_CPS_AWARE_COLOR_PIPE))
+ },
+ { XE_RTP_NAME("16013271637"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_SLICE_COMMON_ECO_CHICKEN1,
+ MSC_MSAA_REODER_BUF_BYPASS_DISABLE))
+ },
+ { XE_RTP_NAME("14014947963"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(FIELD_SET(VF_PREEMPTION,
+ PREEMPTION_VERTEX_COUNT,
+ 0x4000))
+ },
+ { XE_RTP_NAME("18018764978"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_PSS_MODE2,
+ SCOREBOARD_STALL_FLUSH_CONTROL))
+ },
+ { XE_RTP_NAME("18019271663"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(CACHE_MODE_1, MSAA_OPTIMIZATION_REDUC_DISABLE))
+ },
+ { XE_RTP_NAME("14019877138"),
+ XE_RTP_RULES(PLATFORM(DG2)),
+ XE_RTP_ACTIONS(SET(XEHP_PSS_CHICKEN, FD_END_COLLECT))
+ },
+
+ /* PVC */
+
+ { XE_RTP_NAME("16017236439"),
+ XE_RTP_RULES(PLATFORM(PVC), ENGINE_CLASS(COPY),
+ FUNC(xe_rtp_match_even_instance)),
+ XE_RTP_ACTIONS(SET(BCS_SWCTRL(0),
+ BCS_SWCTRL_DISABLE_256B,
+ XE_RTP_ACTION_FLAG(ENGINE_BASE))),
+ },
+
+ /* Xe_LPG */
+
+ { XE_RTP_NAME("18019271663"),
+ XE_RTP_RULES(GRAPHICS_VERSION_RANGE(1270, 1271)),
+ XE_RTP_ACTIONS(SET(CACHE_MODE_1, MSAA_OPTIMIZATION_REDUC_DISABLE))
+ },
+
+ /* Xe2_LPG */
+
+ { XE_RTP_NAME("16020518922"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(FF_MODE,
+ DIS_TE_AUTOSTRIP |
+ DIS_MESH_PARTIAL_AUTOSTRIP |
+ DIS_MESH_AUTOSTRIP),
+ SET(VFLSKPD,
+ DIS_PARTIAL_AUTOSTRIP |
+ DIS_AUTOSTRIP))
+ },
+ { XE_RTP_NAME("14019386621"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(VF_SCRATCHPAD, XE2_VFG_TED_CREDIT_INTERFACE_DISABLE))
+ },
+ { XE_RTP_NAME("14019877138"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(XEHP_PSS_CHICKEN, FD_END_COLLECT))
+ },
+ { XE_RTP_NAME("14020013138"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0),
+ ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(WM_CHICKEN3, HIZ_PLANE_COMPRESSION_DIS))
+ },
+ { XE_RTP_NAME("14019988906"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2004), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(XEHP_PSS_CHICKEN, FLSH_IGNORES_PSD))
+ },
+
+ {}
+};
+
+static __maybe_unused const struct xe_rtp_entry oob_was[] = {
+#include <generated/xe_wa_oob.c>
+ {}
+};
+
+static_assert(ARRAY_SIZE(oob_was) - 1 == _XE_WA_OOB_COUNT);
+
+__diag_pop();
+
+/**
+ * xe_wa_process_oob - process OOB workaround table
+ * @gt: GT instance to process workarounds for
+ *
+ * Process OOB workaround table for this platform, marking in @gt the
+ * workarounds that are active.
+ */
+void xe_wa_process_oob(struct xe_gt *gt)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(gt);
+
+ xe_rtp_process_ctx_enable_active_tracking(&ctx, gt->wa_active.oob,
+ ARRAY_SIZE(oob_was));
+ xe_rtp_process(&ctx, oob_was);
+}
+
+/**
+ * xe_wa_process_gt - process GT workaround table
+ * @gt: GT instance to process workarounds for
+ *
+ * Process GT workaround table for this platform, saving in @gt all the
+ * workarounds that need to be applied at the GT level.
+ */
+void xe_wa_process_gt(struct xe_gt *gt)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(gt);
+
+ xe_rtp_process_ctx_enable_active_tracking(&ctx, gt->wa_active.gt,
+ ARRAY_SIZE(gt_was));
+ xe_rtp_process_to_sr(&ctx, gt_was, &gt->reg_sr);
+}
+EXPORT_SYMBOL_IF_KUNIT(xe_wa_process_gt);
+
+/**
+ * xe_wa_process_engine - process engine workaround table
+ * @hwe: engine instance to process workarounds for
+ *
+ * Process engine workaround table for this platform, saving in @hwe all the
+ * workarounds that need to be applied at the engine level that match this
+ * engine.
+ */
+void xe_wa_process_engine(struct xe_hw_engine *hwe)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+
+ xe_rtp_process_ctx_enable_active_tracking(&ctx, hwe->gt->wa_active.engine,
+ ARRAY_SIZE(engine_was));
+ xe_rtp_process_to_sr(&ctx, engine_was, &hwe->reg_sr);
+}
+
+/**
+ * xe_wa_process_lrc - process context workaround table
+ * @hwe: engine instance to process workarounds for
+ *
+ * Process context workaround table for this platform, saving in @hwe all the
+ * workarounds that need to be applied on context restore. These are workarounds
+ * touching registers that are part of the HW context image.
+ */
+void xe_wa_process_lrc(struct xe_hw_engine *hwe)
+{
+ struct xe_rtp_process_ctx ctx = XE_RTP_PROCESS_CTX_INITIALIZER(hwe);
+
+ xe_rtp_process_ctx_enable_active_tracking(&ctx, hwe->gt->wa_active.lrc,
+ ARRAY_SIZE(lrc_was));
+ xe_rtp_process_to_sr(&ctx, lrc_was, &hwe->reg_lrc);
+}
+
+/**
+ * xe_wa_init - initialize gt with workaround bookkeeping
+ * @gt: GT instance to initialize
+ *
+ * Returns 0 for success, negative error code otherwise.
+ */
+int xe_wa_init(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+ size_t n_oob, n_lrc, n_engine, n_gt, total;
+ unsigned long *p;
+
+ n_gt = BITS_TO_LONGS(ARRAY_SIZE(gt_was));
+ n_engine = BITS_TO_LONGS(ARRAY_SIZE(engine_was));
+ n_lrc = BITS_TO_LONGS(ARRAY_SIZE(lrc_was));
+ n_oob = BITS_TO_LONGS(ARRAY_SIZE(oob_was));
+ total = n_gt + n_engine + n_lrc + n_oob;
+
+ p = drmm_kzalloc(&xe->drm, sizeof(*p) * total, GFP_KERNEL);
+ if (!p)
+ return -ENOMEM;
+
+ gt->wa_active.gt = p;
+ p += n_gt;
+ gt->wa_active.engine = p;
+ p += n_engine;
+ gt->wa_active.lrc = p;
+ p += n_lrc;
+ gt->wa_active.oob = p;
+
+ return 0;
+}
+
+void xe_wa_dump(struct xe_gt *gt, struct drm_printer *p)
+{
+ size_t idx;
+
+ drm_printf(p, "GT Workarounds\n");
+ for_each_set_bit(idx, gt->wa_active.gt, ARRAY_SIZE(gt_was))
+ drm_printf_indent(p, 1, "%s\n", gt_was[idx].name);
+
+ drm_printf(p, "\nEngine Workarounds\n");
+ for_each_set_bit(idx, gt->wa_active.engine, ARRAY_SIZE(engine_was))
+ drm_printf_indent(p, 1, "%s\n", engine_was[idx].name);
+
+ drm_printf(p, "\nLRC Workarounds\n");
+ for_each_set_bit(idx, gt->wa_active.lrc, ARRAY_SIZE(lrc_was))
+ drm_printf_indent(p, 1, "%s\n", lrc_was[idx].name);
+
+ drm_printf(p, "\nOOB Workarounds\n");
+ for_each_set_bit(idx, gt->wa_active.oob, ARRAY_SIZE(oob_was))
+ if (oob_was[idx].name)
+ drm_printf_indent(p, 1, "%s\n", oob_was[idx].name);
+}
+
+/*
+ * Apply tile (non-GT, non-display) workarounds. Think very carefully before
+ * adding anything to this function; most workarounds should be implemented
+ * elsewhere. The programming here is primarily for sgunit/soc workarounds,
+ * which are relatively rare. Since the registers these workarounds target are
+ * outside the GT, they should only need to be applied once at device
+ * probe/resume; they will not lose their values on any kind of GT or engine
+ * reset.
+ *
+ * TODO: We may want to move this over to xe_rtp in the future once we have
+ * enough workarounds to justify the work.
+ */
+void xe_wa_apply_tile_workarounds(struct xe_tile *tile)
+{
+ struct xe_gt *mmio = tile->primary_gt;
+
+ if (XE_WA(mmio, 22010954014))
+ xe_mmio_rmw32(mmio, XEHP_CLOCK_GATE_DIS, 0, SGSI_SIDECLK_DIS);
+}
diff --git a/drivers/gpu/drm/xe/xe_wa.h b/drivers/gpu/drm/xe/xe_wa.h
new file mode 100644
index 000000000000..1b24d66f9d80
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wa.h
@@ -0,0 +1,32 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_WA_
+#define _XE_WA_
+
+struct drm_printer;
+struct xe_gt;
+struct xe_hw_engine;
+struct xe_tile;
+
+int xe_wa_init(struct xe_gt *gt);
+void xe_wa_process_oob(struct xe_gt *gt);
+void xe_wa_process_gt(struct xe_gt *gt);
+void xe_wa_process_engine(struct xe_hw_engine *hwe);
+void xe_wa_process_lrc(struct xe_hw_engine *hwe);
+void xe_wa_apply_tile_workarounds(struct xe_tile *tile);
+
+void xe_reg_whitelist_process_engine(struct xe_hw_engine *hwe);
+void xe_wa_dump(struct xe_gt *gt, struct drm_printer *p);
+
+/**
+ * XE_WA - Out-of-band workarounds, that don't fit the lifecycle any
+ * other more specific type
+ * @gt__: gt instance
+ * @id__: XE_OOB_<id__>, as generated by build system in generated/xe_wa_oob.h
+ */
+#define XE_WA(gt__, id__) test_bit(XE_WA_OOB_ ## id__, (gt__)->wa_active.oob)
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
new file mode 100644
index 000000000000..727bdc429212
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -0,0 +1,24 @@
+22012773006 GRAPHICS_VERSION_RANGE(1200, 1250)
+16011759253 SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, B0)
+14014475959 GRAPHICS_VERSION_RANGE(1270, 1271), GRAPHICS_STEP(A0, B0)
+ PLATFORM(DG2)
+22011391025 PLATFORM(DG2)
+14012197797 PLATFORM(DG2), GRAPHICS_STEP(A0, B0)
+16011777198 SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, C0)
+ SUBPLATFORM(DG2, G11), GRAPHICS_STEP(A0, B0)
+22012727170 SUBPLATFORM(DG2, G10), GRAPHICS_STEP(A0, C0)
+ SUBPLATFORM(DG2, G11)
+22012727685 SUBPLATFORM(DG2, G11)
+16015675438 PLATFORM(PVC)
+ SUBPLATFORM(DG2, G10)
+ SUBPLATFORM(DG2, G12)
+18020744125 PLATFORM(PVC)
+1509372804 PLATFORM(PVC), GRAPHICS_STEP(A0, C0)
+1409600907 GRAPHICS_VERSION_RANGE(1200, 1250)
+14016763929 SUBPLATFORM(DG2, G10)
+ SUBPLATFORM(DG2, G12)
+16017236439 PLATFORM(PVC)
+22010954014 PLATFORM(DG2)
+14019821291 MEDIA_VERSION_RANGE(1300, 2000)
+14015076503 MEDIA_VERSION(1300)
+16020292621 GRAPHICS_VERSION(2004), GRAPHICS_STEP(A0, B0)
diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.c b/drivers/gpu/drm/xe/xe_wait_user_fence.c
new file mode 100644
index 000000000000..a75eeba7bfe5
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wait_user_fence.c
@@ -0,0 +1,179 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_wait_user_fence.h"
+
+#include <drm/drm_device.h>
+#include <drm/drm_file.h>
+#include <drm/drm_utils.h>
+#include <drm/xe_drm.h>
+
+#include "xe_device.h"
+#include "xe_gt.h"
+#include "xe_macros.h"
+#include "xe_exec_queue.h"
+
+static int do_compare(u64 addr, u64 value, u64 mask, u16 op)
+{
+ u64 rvalue;
+ int err;
+ bool passed;
+
+ err = copy_from_user(&rvalue, u64_to_user_ptr(addr), sizeof(rvalue));
+ if (err)
+ return -EFAULT;
+
+ switch (op) {
+ case DRM_XE_UFENCE_WAIT_OP_EQ:
+ passed = (rvalue & mask) == (value & mask);
+ break;
+ case DRM_XE_UFENCE_WAIT_OP_NEQ:
+ passed = (rvalue & mask) != (value & mask);
+ break;
+ case DRM_XE_UFENCE_WAIT_OP_GT:
+ passed = (rvalue & mask) > (value & mask);
+ break;
+ case DRM_XE_UFENCE_WAIT_OP_GTE:
+ passed = (rvalue & mask) >= (value & mask);
+ break;
+ case DRM_XE_UFENCE_WAIT_OP_LT:
+ passed = (rvalue & mask) < (value & mask);
+ break;
+ case DRM_XE_UFENCE_WAIT_OP_LTE:
+ passed = (rvalue & mask) <= (value & mask);
+ break;
+ default:
+ XE_WARN_ON("Not possible");
+ return -EINVAL;
+ }
+
+ return passed ? 0 : 1;
+}
+
+#define VALID_FLAGS DRM_XE_UFENCE_WAIT_FLAG_ABSTIME
+#define MAX_OP DRM_XE_UFENCE_WAIT_OP_LTE
+
+static long to_jiffies_timeout(struct xe_device *xe,
+ struct drm_xe_wait_user_fence *args)
+{
+ unsigned long long t;
+ long timeout;
+
+ /*
+ * For negative timeout we want to wait "forever" by setting
+ * MAX_SCHEDULE_TIMEOUT. But we have to assign this value also
+ * to args->timeout to avoid being zeroed on the signal delivery
+ * (see arithmetics after wait).
+ */
+ if (args->timeout < 0) {
+ args->timeout = MAX_SCHEDULE_TIMEOUT;
+ return MAX_SCHEDULE_TIMEOUT;
+ }
+
+ if (args->timeout == 0)
+ return 0;
+
+ /*
+ * Save the timeout to an u64 variable because nsecs_to_jiffies
+ * might return a value that overflows s32 variable.
+ */
+ if (args->flags & DRM_XE_UFENCE_WAIT_FLAG_ABSTIME)
+ t = drm_timeout_abs_to_jiffies(args->timeout);
+ else
+ t = nsecs_to_jiffies(args->timeout);
+
+ /*
+ * Anything greater then MAX_SCHEDULE_TIMEOUT is meaningless,
+ * also we don't want to cap it at MAX_SCHEDULE_TIMEOUT because
+ * apparently user doesn't mean to wait forever, otherwise the
+ * args->timeout should have been set to a negative value.
+ */
+ if (t > MAX_SCHEDULE_TIMEOUT)
+ timeout = MAX_SCHEDULE_TIMEOUT - 1;
+ else
+ timeout = t;
+
+ return timeout ?: 1;
+}
+
+int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file)
+{
+ struct xe_device *xe = to_xe_device(dev);
+ struct xe_file *xef = to_xe_file(file);
+ DEFINE_WAIT_FUNC(w_wait, woken_wake_function);
+ struct drm_xe_wait_user_fence *args = data;
+ struct xe_exec_queue *q = NULL;
+ u64 addr = args->addr;
+ int err = 0;
+ long timeout;
+ ktime_t start;
+
+ if (XE_IOCTL_DBG(xe, args->extensions) || XE_IOCTL_DBG(xe, args->pad) ||
+ XE_IOCTL_DBG(xe, args->pad2) ||
+ XE_IOCTL_DBG(xe, args->reserved[0] || args->reserved[1]))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->flags & ~VALID_FLAGS))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, args->op > MAX_OP))
+ return -EINVAL;
+
+ if (XE_IOCTL_DBG(xe, addr & 0x7))
+ return -EINVAL;
+
+ if (args->exec_queue_id) {
+ q = xe_exec_queue_lookup(xef, args->exec_queue_id);
+ if (XE_IOCTL_DBG(xe, !q))
+ return -ENOENT;
+ }
+
+ timeout = to_jiffies_timeout(xe, args);
+
+ start = ktime_get();
+
+ add_wait_queue(&xe->ufence_wq, &w_wait);
+ for (;;) {
+ err = do_compare(addr, args->value, args->mask, args->op);
+ if (err <= 0)
+ break;
+
+ if (signal_pending(current)) {
+ err = -ERESTARTSYS;
+ break;
+ }
+
+ if (q) {
+ if (q->ops->reset_status(q)) {
+ drm_info(&xe->drm, "exec gueue reset detected\n");
+ err = -EIO;
+ break;
+ }
+ }
+
+ if (!timeout) {
+ err = -ETIME;
+ break;
+ }
+
+ timeout = wait_woken(&w_wait, TASK_INTERRUPTIBLE, timeout);
+ }
+ remove_wait_queue(&xe->ufence_wq, &w_wait);
+
+ if (!(args->flags & DRM_XE_UFENCE_WAIT_FLAG_ABSTIME)) {
+ args->timeout -= ktime_to_ns(ktime_sub(ktime_get(), start));
+ if (args->timeout < 0)
+ args->timeout = 0;
+ }
+
+ if (!timeout && !(err < 0))
+ err = -ETIME;
+
+ if (q)
+ xe_exec_queue_put(q);
+
+ return err;
+}
diff --git a/drivers/gpu/drm/xe/xe_wait_user_fence.h b/drivers/gpu/drm/xe/xe_wait_user_fence.h
new file mode 100644
index 000000000000..0e268978f9e6
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wait_user_fence.h
@@ -0,0 +1,15 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_WAIT_USER_FENCE_H_
+#define _XE_WAIT_USER_FENCE_H_
+
+struct drm_device;
+struct drm_file;
+
+int xe_wait_user_fence_ioctl(struct drm_device *dev, void *data,
+ struct drm_file *file);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_wopcm.c b/drivers/gpu/drm/xe/xe_wopcm.c
new file mode 100644
index 000000000000..d3a99157e523
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wopcm.c
@@ -0,0 +1,270 @@
+// SPDX-License-Identifier: MIT
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#include "xe_wopcm.h"
+
+#include "regs/xe_guc_regs.h"
+#include "xe_device.h"
+#include "xe_force_wake.h"
+#include "xe_gt.h"
+#include "xe_mmio.h"
+#include "xe_uc_fw.h"
+
+/**
+ * DOC: Write Once Protected Content Memory (WOPCM) Layout
+ *
+ * The layout of the WOPCM will be fixed after writing to GuC WOPCM size and
+ * offset registers whose values are calculated and determined by HuC/GuC
+ * firmware size and set of hardware requirements/restrictions as shown below:
+ *
+ * ::
+ *
+ * +=========> +====================+ <== WOPCM Top
+ * ^ | HW contexts RSVD |
+ * | +===> +====================+ <== GuC WOPCM Top
+ * | ^ | |
+ * | | | |
+ * | | | |
+ * | GuC | |
+ * | WOPCM | |
+ * | Size +--------------------+
+ * WOPCM | | GuC FW RSVD |
+ * | | +--------------------+
+ * | | | GuC Stack RSVD |
+ * | | +------------------- +
+ * | v | GuC WOPCM RSVD |
+ * | +===> +====================+ <== GuC WOPCM base
+ * | | WOPCM RSVD |
+ * | +------------------- + <== HuC Firmware Top
+ * v | HuC FW |
+ * +=========> +====================+ <== WOPCM Base
+ *
+ * GuC accessible WOPCM starts at GuC WOPCM base and ends at GuC WOPCM top.
+ * The top part of the WOPCM is reserved for hardware contexts (e.g. RC6
+ * context).
+ */
+
+/* Default WOPCM size is 2MB from Gen11, 1MB on previous platforms */
+/* FIXME: Larger size require for 2 tile PVC, do a proper probe sooner or later */
+#define DGFX_WOPCM_SIZE SZ_4M
+/* FIXME: Larger size require for MTL, do a proper probe sooner or later */
+#define MTL_WOPCM_SIZE SZ_4M
+#define WOPCM_SIZE SZ_2M
+
+#define MAX_WOPCM_SIZE SZ_8M
+
+/* 16KB WOPCM (RSVD WOPCM) is reserved from HuC firmware top. */
+#define WOPCM_RESERVED_SIZE SZ_16K
+
+/* 16KB reserved at the beginning of GuC WOPCM. */
+#define GUC_WOPCM_RESERVED SZ_16K
+/* 8KB from GUC_WOPCM_RESERVED is reserved for GuC stack. */
+#define GUC_WOPCM_STACK_RESERVED SZ_8K
+
+/* GuC WOPCM Offset value needs to be aligned to 16KB. */
+#define GUC_WOPCM_OFFSET_ALIGNMENT (1UL << GUC_WOPCM_OFFSET_SHIFT)
+
+/* 36KB WOPCM reserved at the end of WOPCM */
+#define WOPCM_HW_CTX_RESERVED (SZ_32K + SZ_4K)
+
+static inline struct xe_gt *wopcm_to_gt(struct xe_wopcm *wopcm)
+{
+ return container_of(wopcm, struct xe_gt, uc.wopcm);
+}
+
+static inline struct xe_device *wopcm_to_xe(struct xe_wopcm *wopcm)
+{
+ return gt_to_xe(wopcm_to_gt(wopcm));
+}
+
+static u32 context_reserved_size(void)
+{
+ return WOPCM_HW_CTX_RESERVED;
+}
+
+static bool __check_layout(struct xe_device *xe, u32 wopcm_size,
+ u32 guc_wopcm_base, u32 guc_wopcm_size,
+ u32 guc_fw_size, u32 huc_fw_size)
+{
+ const u32 ctx_rsvd = context_reserved_size();
+ u32 size;
+
+ size = wopcm_size - ctx_rsvd;
+ if (unlikely(guc_wopcm_base >= size ||
+ guc_wopcm_size > size - guc_wopcm_base)) {
+ drm_err(&xe->drm,
+ "WOPCM: invalid GuC region layout: %uK + %uK > %uK\n",
+ guc_wopcm_base / SZ_1K, guc_wopcm_size / SZ_1K,
+ size / SZ_1K);
+ return false;
+ }
+
+ size = guc_fw_size + GUC_WOPCM_RESERVED + GUC_WOPCM_STACK_RESERVED;
+ if (unlikely(guc_wopcm_size < size)) {
+ drm_err(&xe->drm, "WOPCM: no space for %s: %uK < %uK\n",
+ xe_uc_fw_type_repr(XE_UC_FW_TYPE_GUC),
+ guc_wopcm_size / SZ_1K, size / SZ_1K);
+ return false;
+ }
+
+ size = huc_fw_size + WOPCM_RESERVED_SIZE;
+ if (unlikely(guc_wopcm_base < size)) {
+ drm_err(&xe->drm, "WOPCM: no space for %s: %uK < %uK\n",
+ xe_uc_fw_type_repr(XE_UC_FW_TYPE_HUC),
+ guc_wopcm_base / SZ_1K, size / SZ_1K);
+ return false;
+ }
+
+ return true;
+}
+
+static bool __wopcm_regs_locked(struct xe_gt *gt,
+ u32 *guc_wopcm_base, u32 *guc_wopcm_size)
+{
+ u32 reg_base = xe_mmio_read32(gt, DMA_GUC_WOPCM_OFFSET);
+ u32 reg_size = xe_mmio_read32(gt, GUC_WOPCM_SIZE);
+
+ if (!(reg_size & GUC_WOPCM_SIZE_LOCKED) ||
+ !(reg_base & GUC_WOPCM_OFFSET_VALID))
+ return false;
+
+ *guc_wopcm_base = reg_base & GUC_WOPCM_OFFSET_MASK;
+ *guc_wopcm_size = reg_size & GUC_WOPCM_SIZE_MASK;
+ return true;
+}
+
+static int __wopcm_init_regs(struct xe_device *xe, struct xe_gt *gt,
+ struct xe_wopcm *wopcm)
+{
+ u32 base = wopcm->guc.base;
+ u32 size = wopcm->guc.size;
+ u32 huc_agent = xe_uc_fw_is_available(&gt->uc.huc.fw) ? HUC_LOADING_AGENT_GUC : 0;
+ u32 mask;
+ int err;
+
+ XE_WARN_ON(!(base & GUC_WOPCM_OFFSET_MASK));
+ XE_WARN_ON(base & ~GUC_WOPCM_OFFSET_MASK);
+ XE_WARN_ON(!(size & GUC_WOPCM_SIZE_MASK));
+ XE_WARN_ON(size & ~GUC_WOPCM_SIZE_MASK);
+
+ mask = GUC_WOPCM_SIZE_MASK | GUC_WOPCM_SIZE_LOCKED;
+ err = xe_mmio_write32_and_verify(gt, GUC_WOPCM_SIZE, size, mask,
+ size | GUC_WOPCM_SIZE_LOCKED);
+ if (err)
+ goto err_out;
+
+ mask = GUC_WOPCM_OFFSET_MASK | GUC_WOPCM_OFFSET_VALID | huc_agent;
+ err = xe_mmio_write32_and_verify(gt, DMA_GUC_WOPCM_OFFSET,
+ base | huc_agent, mask,
+ base | huc_agent |
+ GUC_WOPCM_OFFSET_VALID);
+ if (err)
+ goto err_out;
+
+ return 0;
+
+err_out:
+ drm_notice(&xe->drm, "Failed to init uC WOPCM registers!\n");
+ drm_notice(&xe->drm, "%s(%#x)=%#x\n", "DMA_GUC_WOPCM_OFFSET",
+ DMA_GUC_WOPCM_OFFSET.addr,
+ xe_mmio_read32(gt, DMA_GUC_WOPCM_OFFSET));
+ drm_notice(&xe->drm, "%s(%#x)=%#x\n", "GUC_WOPCM_SIZE",
+ GUC_WOPCM_SIZE.addr,
+ xe_mmio_read32(gt, GUC_WOPCM_SIZE));
+
+ return err;
+}
+
+u32 xe_wopcm_size(struct xe_device *xe)
+{
+ return IS_DGFX(xe) ? DGFX_WOPCM_SIZE :
+ xe->info.platform == XE_METEORLAKE ? MTL_WOPCM_SIZE :
+ WOPCM_SIZE;
+}
+
+/**
+ * xe_wopcm_init() - Initialize the WOPCM structure.
+ * @wopcm: pointer to xe_wopcm.
+ *
+ * This function will partition WOPCM space based on GuC and HuC firmware sizes
+ * and will allocate max remaining for use by GuC. This function will also
+ * enforce platform dependent hardware restrictions on GuC WOPCM offset and
+ * size. It will fail the WOPCM init if any of these checks fail, so that the
+ * following WOPCM registers setup and GuC firmware uploading would be aborted.
+ */
+int xe_wopcm_init(struct xe_wopcm *wopcm)
+{
+ struct xe_device *xe = wopcm_to_xe(wopcm);
+ struct xe_gt *gt = wopcm_to_gt(wopcm);
+ u32 guc_fw_size = xe_uc_fw_get_upload_size(&gt->uc.guc.fw);
+ u32 huc_fw_size = xe_uc_fw_get_upload_size(&gt->uc.huc.fw);
+ u32 ctx_rsvd = context_reserved_size();
+ u32 guc_wopcm_base;
+ u32 guc_wopcm_size;
+ bool locked;
+ int ret = 0;
+
+ if (!guc_fw_size)
+ return -EINVAL;
+
+ wopcm->size = xe_wopcm_size(xe);
+ drm_dbg(&xe->drm, "WOPCM: %uK\n", wopcm->size / SZ_1K);
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GT);
+ XE_WARN_ON(guc_fw_size >= wopcm->size);
+ XE_WARN_ON(huc_fw_size >= wopcm->size);
+ XE_WARN_ON(ctx_rsvd + WOPCM_RESERVED_SIZE >= wopcm->size);
+
+ locked = __wopcm_regs_locked(gt, &guc_wopcm_base, &guc_wopcm_size);
+ if (locked) {
+ drm_dbg(&xe->drm, "GuC WOPCM is already locked [%uK, %uK)\n",
+ guc_wopcm_base / SZ_1K, guc_wopcm_size / SZ_1K);
+ /*
+ * When the GuC wopcm base and size are preprogrammed by
+ * BIOS/IFWI, check against the max allowed wopcm size to
+ * validate if the programmed values align to the wopcm layout.
+ */
+ wopcm->size = MAX_WOPCM_SIZE;
+
+ goto check;
+ }
+
+ /*
+ * Aligned value of guc_wopcm_base will determine available WOPCM space
+ * for HuC firmware and mandatory reserved area.
+ */
+ guc_wopcm_base = huc_fw_size + WOPCM_RESERVED_SIZE;
+ guc_wopcm_base = ALIGN(guc_wopcm_base, GUC_WOPCM_OFFSET_ALIGNMENT);
+
+ /*
+ * Need to clamp guc_wopcm_base now to make sure the following math is
+ * correct. Formal check of whole WOPCM layout will be done below.
+ */
+ guc_wopcm_base = min(guc_wopcm_base, wopcm->size - ctx_rsvd);
+
+ /* Aligned remainings of usable WOPCM space can be assigned to GuC. */
+ guc_wopcm_size = wopcm->size - ctx_rsvd - guc_wopcm_base;
+ guc_wopcm_size &= GUC_WOPCM_SIZE_MASK;
+
+ drm_dbg(&xe->drm, "Calculated GuC WOPCM [%uK, %uK)\n",
+ guc_wopcm_base / SZ_1K, guc_wopcm_size / SZ_1K);
+
+check:
+ if (__check_layout(xe, wopcm->size, guc_wopcm_base, guc_wopcm_size,
+ guc_fw_size, huc_fw_size)) {
+ wopcm->guc.base = guc_wopcm_base;
+ wopcm->guc.size = guc_wopcm_size;
+ XE_WARN_ON(!wopcm->guc.base);
+ XE_WARN_ON(!wopcm->guc.size);
+ } else {
+ drm_notice(&xe->drm, "Unsuccessful WOPCM partitioning\n");
+ return -E2BIG;
+ }
+
+ if (!locked)
+ ret = __wopcm_init_regs(xe, gt, wopcm);
+
+ return ret;
+}
diff --git a/drivers/gpu/drm/xe/xe_wopcm.h b/drivers/gpu/drm/xe/xe_wopcm.h
new file mode 100644
index 000000000000..0197a282460b
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wopcm.h
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_WOPCM_H_
+#define _XE_WOPCM_H_
+
+#include "xe_wopcm_types.h"
+
+struct xe_device;
+
+int xe_wopcm_init(struct xe_wopcm *wopcm);
+u32 xe_wopcm_size(struct xe_device *xe);
+
+#endif
diff --git a/drivers/gpu/drm/xe/xe_wopcm_types.h b/drivers/gpu/drm/xe/xe_wopcm_types.h
new file mode 100644
index 000000000000..486d850c4084
--- /dev/null
+++ b/drivers/gpu/drm/xe/xe_wopcm_types.h
@@ -0,0 +1,26 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * Copyright © 2022 Intel Corporation
+ */
+
+#ifndef _XE_WOPCM_TYPES_H_
+#define _XE_WOPCM_TYPES_H_
+
+#include <linux/types.h>
+
+/**
+ * struct xe_wopcm - Overall WOPCM info and WOPCM regions.
+ */
+struct xe_wopcm {
+ /** @size: Size of overall WOPCM */
+ u32 size;
+ /** @guc: GuC WOPCM Region info */
+ struct {
+ /** @base: GuC WOPCM base which is offset from WOPCM base */
+ u32 base;
+ /** @size: Size of the GuC WOPCM region */
+ u32 size;
+ } guc;
+};
+
+#endif