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/Kconfig3
-rw-r--r--drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h1
-rw-r--r--drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_stolen.h2
-rw-r--r--drivers/gpu/drm/xe/display/xe_display.c45
-rw-r--r--drivers/gpu/drm/xe/display/xe_display.h4
-rw-r--r--drivers/gpu/drm/xe/display/xe_dsb_buffer.c18
-rw-r--r--drivers/gpu/drm/xe/display/xe_fb_pin.c17
-rw-r--r--drivers/gpu/drm/xe/instructions/xe_mi_commands.h4
-rw-r--r--drivers/gpu/drm/xe/regs/xe_gt_regs.h1
-rw-r--r--drivers/gpu/drm/xe/regs/xe_reg_defs.h2
-rw-r--r--drivers/gpu/drm/xe/tests/xe_mocs.c4
-rw-r--r--drivers/gpu/drm/xe/xe_bo.c100
-rw-r--r--drivers/gpu/drm/xe/xe_bo.h40
-rw-r--r--drivers/gpu/drm/xe/xe_bo_evict.c14
-rw-r--r--drivers/gpu/drm/xe/xe_bo_types.h10
-rw-r--r--drivers/gpu/drm/xe/xe_debugfs.c3
-rw-r--r--drivers/gpu/drm/xe/xe_devcoredump.c14
-rw-r--r--drivers/gpu/drm/xe/xe_device.c106
-rw-r--r--drivers/gpu/drm/xe/xe_force_wake.h16
-rw-r--r--drivers/gpu/drm/xe/xe_gen_wa_oob.c6
-rw-r--r--drivers/gpu/drm/xe/xe_ggtt.c46
-rw-r--r--drivers/gpu/drm/xe/xe_gpu_scheduler.h10
-rw-r--r--drivers/gpu/drm/xe/xe_gsc.c22
-rw-r--r--drivers/gpu/drm/xe/xe_gsc.h1
-rw-r--r--drivers/gpu/drm/xe/xe_gsc_proxy.c11
-rw-r--r--drivers/gpu/drm/xe/xe_gsc_proxy.h1
-rw-r--r--drivers/gpu/drm/xe/xe_gt.c24
-rw-r--r--drivers/gpu/drm/xe/xe_gt.h1
-rw-r--r--drivers/gpu/drm/xe/xe_gt_freq.c2
-rw-r--r--drivers/gpu/drm/xe/xe_gt_pagefault.c12
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_pf.c114
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_pf.h6
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c37
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c3
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_pf_types.h10
-rw-r--r--drivers/gpu/drm/xe/xe_gt_sriov_vf.c12
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c30
-rw-r--r--drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h2
-rw-r--r--drivers/gpu/drm/xe/xe_guc.c4
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct.c357
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct.h7
-rw-r--r--drivers/gpu/drm/xe/xe_guc_ct_types.h23
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc.c146
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc.h2
-rw-r--r--drivers/gpu/drm/xe/xe_guc_pc_types.h2
-rw-r--r--drivers/gpu/drm/xe/xe_guc_relay.c2
-rw-r--r--drivers/gpu/drm/xe/xe_guc_submit.c23
-rw-r--r--drivers/gpu/drm/xe/xe_guc_types.h5
-rw-r--r--drivers/gpu/drm/xe/xe_hw_engine.c12
-rw-r--r--drivers/gpu/drm/xe/xe_irq.c4
-rw-r--r--drivers/gpu/drm/xe/xe_lmtt.c11
-rw-r--r--drivers/gpu/drm/xe/xe_lrc.c6
-rw-r--r--drivers/gpu/drm/xe/xe_lrc_types.h4
-rw-r--r--drivers/gpu/drm/xe/xe_migrate.c2
-rw-r--r--drivers/gpu/drm/xe/xe_oa.c1
-rw-r--r--drivers/gpu/drm/xe/xe_pci.c2
-rw-r--r--drivers/gpu/drm/xe/xe_pci_sriov.c51
-rw-r--r--drivers/gpu/drm/xe/xe_pm.c8
-rw-r--r--drivers/gpu/drm/xe/xe_pt.c14
-rw-r--r--drivers/gpu/drm/xe/xe_pt.h3
-rw-r--r--drivers/gpu/drm/xe/xe_ring_ops.c7
-rw-r--r--drivers/gpu/drm/xe/xe_sa.c3
-rw-r--r--drivers/gpu/drm/xe/xe_tile.c12
-rw-r--r--drivers/gpu/drm/xe/xe_tile.h1
-rw-r--r--drivers/gpu/drm/xe/xe_trace_bo.h2
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c71
-rw-r--r--drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h2
-rw-r--r--drivers/gpu/drm/xe/xe_uc.c8
-rw-r--r--drivers/gpu/drm/xe/xe_uc.h1
-rw-r--r--drivers/gpu/drm/xe/xe_vm.c40
-rw-r--r--drivers/gpu/drm/xe/xe_wa.c4
-rw-r--r--drivers/gpu/drm/xe/xe_wa_oob.rules2
72 files changed, 1342 insertions, 254 deletions
diff --git a/drivers/gpu/drm/xe/Kconfig b/drivers/gpu/drm/xe/Kconfig
index 7bbe46a98ff1..93e742c1f21e 100644
--- a/drivers/gpu/drm/xe/Kconfig
+++ b/drivers/gpu/drm/xe/Kconfig
@@ -1,7 +1,8 @@
# SPDX-License-Identifier: GPL-2.0-only
config DRM_XE
tristate "Intel Xe Graphics"
- depends on DRM && PCI && MMU && (m || (y && KUNIT=y))
+ depends on DRM && PCI && MMU
+ depends on KUNIT || !KUNIT
select INTERVAL_TREE
# we need shmfs for the swappable backing store, and in particular
# the shmem_readpage() which depends upon tmpfs
diff --git a/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h
index 8f86a16dc577..f58198cf2cf6 100644
--- a/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h
+++ b/drivers/gpu/drm/xe/abi/guc_communication_ctb_abi.h
@@ -52,6 +52,7 @@ struct guc_ct_buffer_desc {
#define GUC_CTB_STATUS_OVERFLOW (1 << 0)
#define GUC_CTB_STATUS_UNDERFLOW (1 << 1)
#define GUC_CTB_STATUS_MISMATCH (1 << 2)
+#define GUC_CTB_STATUS_DISABLED (1 << 3)
u32 reserved[13];
} __packed;
static_assert(sizeof(struct guc_ct_buffer_desc) == 64);
diff --git a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_stolen.h b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_stolen.h
index cb6c7598824b..9c4cf050059a 100644
--- a/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_stolen.h
+++ b/drivers/gpu/drm/xe/compat-i915-headers/gem/i915_gem_stolen.h
@@ -29,7 +29,7 @@ static inline int i915_gem_stolen_insert_node_in_range(struct xe_device *xe,
bo = xe_bo_create_locked_range(xe, xe_device_get_root_tile(xe),
NULL, size, start, end,
- ttm_bo_type_kernel, flags);
+ ttm_bo_type_kernel, flags, 0);
if (IS_ERR(bo)) {
err = PTR_ERR(bo);
bo = NULL;
diff --git a/drivers/gpu/drm/xe/display/xe_display.c b/drivers/gpu/drm/xe/display/xe_display.c
index c6e0c8d77a70..e164e2d71e11 100644
--- a/drivers/gpu/drm/xe/display/xe_display.c
+++ b/drivers/gpu/drm/xe/display/xe_display.c
@@ -96,6 +96,8 @@ int xe_display_create(struct xe_device *xe)
spin_lock_init(&xe->display.fb_tracking.lock);
xe->display.hotplug.dp_wq = alloc_ordered_workqueue("xe-dp", 0);
+ if (!xe->display.hotplug.dp_wq)
+ return -ENOMEM;
return drmm_add_action_or_reset(&xe->drm, display_destroy, NULL);
}
@@ -352,6 +354,36 @@ void xe_display_pm_suspend(struct xe_device *xe)
__xe_display_pm_suspend(xe, false);
}
+void xe_display_pm_shutdown(struct xe_device *xe)
+{
+ struct intel_display *display = &xe->display;
+
+ if (!xe->info.probe_display)
+ return;
+
+ intel_power_domains_disable(xe);
+ intel_fbdev_set_suspend(&xe->drm, FBINFO_STATE_SUSPENDED, true);
+ if (has_display(xe)) {
+ drm_kms_helper_poll_disable(&xe->drm);
+ intel_display_driver_disable_user_access(xe);
+ intel_display_driver_suspend(xe);
+ }
+
+ xe_display_flush_cleanup_work(xe);
+ intel_dp_mst_suspend(xe);
+ intel_hpd_cancel_work(xe);
+
+ if (has_display(xe))
+ intel_display_driver_suspend_access(xe);
+
+ intel_encoder_suspend_all(display);
+ intel_encoder_shutdown_all(display);
+
+ intel_opregion_suspend(display, PCI_D3cold);
+
+ intel_dmc_suspend(xe);
+}
+
void xe_display_pm_runtime_suspend(struct xe_device *xe)
{
if (!xe->info.probe_display)
@@ -376,6 +408,19 @@ void xe_display_pm_suspend_late(struct xe_device *xe)
intel_display_power_suspend_late(xe);
}
+void xe_display_pm_shutdown_late(struct xe_device *xe)
+{
+ if (!xe->info.probe_display)
+ return;
+
+ /*
+ * The only requirement is to reboot with display DC states disabled,
+ * for now leaving all display power wells in the INIT power domain
+ * enabled.
+ */
+ intel_power_domains_driver_remove(xe);
+}
+
void xe_display_pm_resume_early(struct xe_device *xe)
{
if (!xe->info.probe_display)
diff --git a/drivers/gpu/drm/xe/display/xe_display.h b/drivers/gpu/drm/xe/display/xe_display.h
index bed55fd26f30..17afa537aee5 100644
--- a/drivers/gpu/drm/xe/display/xe_display.h
+++ b/drivers/gpu/drm/xe/display/xe_display.h
@@ -35,7 +35,9 @@ 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_shutdown(struct xe_device *xe);
void xe_display_pm_suspend_late(struct xe_device *xe);
+void xe_display_pm_shutdown_late(struct xe_device *xe);
void xe_display_pm_resume_early(struct xe_device *xe);
void xe_display_pm_resume(struct xe_device *xe);
void xe_display_pm_runtime_suspend(struct xe_device *xe);
@@ -66,7 +68,9 @@ 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_shutdown(struct xe_device *xe) {}
static inline void xe_display_pm_suspend_late(struct xe_device *xe) {}
+static inline void xe_display_pm_shutdown_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) {}
static inline void xe_display_pm_runtime_suspend(struct xe_device *xe) {}
diff --git a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
index f99d901a3214..9f941fc2e36b 100644
--- a/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
+++ b/drivers/gpu/drm/xe/display/xe_dsb_buffer.c
@@ -17,10 +17,7 @@ u32 intel_dsb_buffer_ggtt_offset(struct intel_dsb_buffer *dsb_buf)
void intel_dsb_buffer_write(struct intel_dsb_buffer *dsb_buf, u32 idx, u32 val)
{
- struct xe_device *xe = dsb_buf->vma->bo->tile->xe;
-
iosys_map_wr(&dsb_buf->vma->bo->vmap, idx * 4, u32, val);
- xe_device_l2_flush(xe);
}
u32 intel_dsb_buffer_read(struct intel_dsb_buffer *dsb_buf, u32 idx)
@@ -30,12 +27,9 @@ u32 intel_dsb_buffer_read(struct intel_dsb_buffer *dsb_buf, u32 idx)
void intel_dsb_buffer_memset(struct intel_dsb_buffer *dsb_buf, u32 idx, u32 val, size_t size)
{
- struct xe_device *xe = dsb_buf->vma->bo->tile->xe;
-
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);
- xe_device_l2_flush(xe);
}
bool intel_dsb_buffer_create(struct intel_crtc *crtc, struct intel_dsb_buffer *dsb_buf, size_t size)
@@ -48,11 +42,12 @@ bool intel_dsb_buffer_create(struct intel_crtc *crtc, struct intel_dsb_buffer *d
if (!vma)
return false;
+ /* Set scanout flag for WC mapping */
obj = xe_bo_create_pin_map(xe, xe_device_get_root_tile(xe),
NULL, PAGE_ALIGN(size),
ttm_bo_type_kernel,
XE_BO_FLAG_VRAM_IF_DGFX(xe_device_get_root_tile(xe)) |
- XE_BO_FLAG_GGTT);
+ XE_BO_FLAG_SCANOUT | XE_BO_FLAG_GGTT);
if (IS_ERR(obj)) {
kfree(vma);
return false;
@@ -73,5 +68,12 @@ void intel_dsb_buffer_cleanup(struct intel_dsb_buffer *dsb_buf)
void intel_dsb_buffer_flush_map(struct intel_dsb_buffer *dsb_buf)
{
- /* TODO: add xe specific flush_map() for dsb buffer object. */
+ struct xe_device *xe = dsb_buf->vma->bo->tile->xe;
+
+ /*
+ * The memory barrier here is to ensure coherency of DSB vs MMIO,
+ * both for weak ordering archs and discrete cards.
+ */
+ xe_device_wmb(xe);
+ xe_device_l2_flush(xe);
}
diff --git a/drivers/gpu/drm/xe/display/xe_fb_pin.c b/drivers/gpu/drm/xe/display/xe_fb_pin.c
index b58fc4ba2aac..0558b106f8b6 100644
--- a/drivers/gpu/drm/xe/display/xe_fb_pin.c
+++ b/drivers/gpu/drm/xe/display/xe_fb_pin.c
@@ -153,7 +153,10 @@ static int __xe_pin_fb_vma_dpt(const struct intel_framebuffer *fb,
}
vma->dpt = dpt;
- vma->node = dpt->ggtt_node;
+ vma->node = dpt->ggtt_node[tile0->id];
+
+ /* Ensure DPT writes are flushed */
+ xe_device_l2_flush(xe);
return 0;
}
@@ -203,8 +206,8 @@ static int __xe_pin_fb_vma_ggtt(const struct intel_framebuffer *fb,
if (xe_bo_is_vram(bo) && ggtt->flags & XE_GGTT_FLAGS_64K)
align = max_t(u32, align, SZ_64K);
- if (bo->ggtt_node && view->type == I915_GTT_VIEW_NORMAL) {
- vma->node = bo->ggtt_node;
+ if (bo->ggtt_node[ggtt->tile->id] && view->type == I915_GTT_VIEW_NORMAL) {
+ vma->node = bo->ggtt_node[ggtt->tile->id];
} else if (view->type == I915_GTT_VIEW_NORMAL) {
u32 x, size = bo->ttm.base.size;
@@ -318,8 +321,6 @@ static struct i915_vma *__xe_pin_fb_vma(const struct intel_framebuffer *fb,
if (ret)
goto err_unpin;
- /* Ensure DPT writes are flushed */
- xe_device_l2_flush(xe);
return vma;
err_unpin:
@@ -333,10 +334,12 @@ err:
static void __xe_unpin_fb_vma(struct i915_vma *vma)
{
+ u8 tile_id = vma->node->ggtt->tile->id;
+
if (vma->dpt)
xe_bo_unpin_map_no_vm(vma->dpt);
- else if (!xe_ggtt_node_allocated(vma->bo->ggtt_node) ||
- vma->bo->ggtt_node->base.start != vma->node->base.start)
+ else if (!xe_ggtt_node_allocated(vma->bo->ggtt_node[tile_id]) ||
+ vma->bo->ggtt_node[tile_id]->base.start != vma->node->base.start)
xe_ggtt_node_remove(vma->node, false);
ttm_bo_reserve(&vma->bo->ttm, false, false, NULL);
diff --git a/drivers/gpu/drm/xe/instructions/xe_mi_commands.h b/drivers/gpu/drm/xe/instructions/xe_mi_commands.h
index 10ec2920d31b..d4033278be9f 100644
--- a/drivers/gpu/drm/xe/instructions/xe_mi_commands.h
+++ b/drivers/gpu/drm/xe/instructions/xe_mi_commands.h
@@ -47,6 +47,10 @@
#define MI_LRI_FORCE_POSTED REG_BIT(12)
#define MI_LRI_LEN(x) (((x) & 0xff) + 1)
+#define MI_STORE_REGISTER_MEM (__MI_INSTR(0x24) | XE_INSTR_NUM_DW(4))
+#define MI_SRM_USE_GGTT REG_BIT(22)
+#define MI_SRM_ADD_CS_OFFSET REG_BIT(19)
+
#define MI_FLUSH_DW __MI_INSTR(0x26)
#define MI_FLUSH_DW_STORE_INDEX REG_BIT(21)
#define MI_INVALIDATE_TLB REG_BIT(18)
diff --git a/drivers/gpu/drm/xe/regs/xe_gt_regs.h b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
index 5404de2aea54..c160b015d178 100644
--- a/drivers/gpu/drm/xe/regs/xe_gt_regs.h
+++ b/drivers/gpu/drm/xe/regs/xe_gt_regs.h
@@ -157,6 +157,7 @@
#define XEHPG_SC_INSTDONE_EXTRA2 XE_REG_MCR(0x7108)
#define COMMON_SLICE_CHICKEN4 XE_REG(0x7300, XE_REG_OPTION_MASKED)
+#define SBE_PUSH_CONSTANT_BEHIND_FIX_ENABLE REG_BIT(12)
#define DISABLE_TDC_LOAD_BALANCING_CALC REG_BIT(6)
#define COMMON_SLICE_CHICKEN3 XE_REG(0x7304, XE_REG_OPTION_MASKED)
diff --git a/drivers/gpu/drm/xe/regs/xe_reg_defs.h b/drivers/gpu/drm/xe/regs/xe_reg_defs.h
index 23f7dc5bbe99..51fd40ffafcb 100644
--- a/drivers/gpu/drm/xe/regs/xe_reg_defs.h
+++ b/drivers/gpu/drm/xe/regs/xe_reg_defs.h
@@ -128,7 +128,7 @@ struct xe_reg_mcr {
* options.
*/
#define XE_REG_MCR(r_, ...) ((const struct xe_reg_mcr){ \
- .__reg = XE_REG_INITIALIZER(r_, ##__VA_ARGS__, .mcr = 1) \
+ .__reg = XE_REG_INITIALIZER(r_, ##__VA_ARGS__, .mcr = 1) \
})
static inline bool xe_reg_is_valid(struct xe_reg r)
diff --git a/drivers/gpu/drm/xe/tests/xe_mocs.c b/drivers/gpu/drm/xe/tests/xe_mocs.c
index 79be73b4a02b..bf3f97d0c9c7 100644
--- a/drivers/gpu/drm/xe/tests/xe_mocs.c
+++ b/drivers/gpu/drm/xe/tests/xe_mocs.c
@@ -47,7 +47,7 @@ static void read_l3cc_table(struct xe_gt *gt,
u32 reg_val;
u32 ret;
- ret = xe_force_wake_get(gt_to_fw(gt), XE_FW_GT);
+ ret = xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
KUNIT_ASSERT_EQ_MSG(test, ret, 0, "Forcewake Failed.\n");
for (i = 0; i < info->num_mocs_regs; i++) {
@@ -72,7 +72,7 @@ static void read_l3cc_table(struct xe_gt *gt,
KUNIT_EXPECT_EQ_MSG(test, l3cc_expected, l3cc,
"l3cc idx=%u has incorrect val.\n", i);
}
- xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
}
static void read_mocs_table(struct xe_gt *gt,
diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
index 84e327b56925..5f745d9ed6cc 100644
--- a/drivers/gpu/drm/xe/xe_bo.c
+++ b/drivers/gpu/drm/xe/xe_bo.c
@@ -702,6 +702,21 @@ static int xe_bo_move(struct ttm_buffer_object *ttm_bo, bool evict,
goto out;
}
+ /* Reject BO eviction if BO is bound to current VM. */
+ if (evict && ctx->resv) {
+ struct drm_gpuvm_bo *vm_bo;
+
+ drm_gem_for_each_gpuvm_bo(vm_bo, &bo->ttm.base) {
+ struct xe_vm *vm = gpuvm_to_vm(vm_bo->vm);
+
+ if (xe_vm_resv(vm) == ctx->resv &&
+ xe_vm_in_preempt_fence_mode(vm)) {
+ ret = -EBUSY;
+ goto out;
+ }
+ }
+ }
+
/*
* Failed multi-hop where the old_mem is still marked as
* TTM_PL_FLAG_TEMPORARY, should just be a dummy move.
@@ -1115,6 +1130,8 @@ 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);
+ struct xe_tile *tile;
+ u8 id;
if (bo->ttm.base.import_attach)
drm_prime_gem_destroy(&bo->ttm.base, NULL);
@@ -1122,8 +1139,9 @@ static void xe_ttm_bo_destroy(struct ttm_buffer_object *ttm_bo)
xe_assert(xe, list_empty(&ttm_bo->base.gpuva.list));
- if (bo->ggtt_node && bo->ggtt_node->base.size)
- xe_ggtt_remove_bo(bo->tile->mem.ggtt, bo);
+ for_each_tile(tile, xe, id)
+ if (bo->ggtt_node[id] && bo->ggtt_node[id]->base.size)
+ xe_ggtt_remove_bo(tile->mem.ggtt, bo);
#ifdef CONFIG_PROC_FS
if (bo->client)
@@ -1293,6 +1311,10 @@ struct xe_bo *___xe_bo_create_locked(struct xe_device *xe, struct xe_bo *bo,
return ERR_PTR(-EINVAL);
}
+ /* XE_BO_FLAG_GGTTx requires XE_BO_FLAG_GGTT also be set */
+ if ((flags & XE_BO_FLAG_GGTT_ALL) && !(flags & XE_BO_FLAG_GGTT))
+ return ERR_PTR(-EINVAL);
+
if (flags & (XE_BO_FLAG_VRAM_MASK | XE_BO_FLAG_STOLEN) &&
!(flags & XE_BO_FLAG_IGNORE_MIN_PAGE_SIZE) &&
((xe->info.vram_flags & XE_VRAM_FLAGS_NEED64K) ||
@@ -1439,7 +1461,8 @@ 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)
+ u16 cpu_caching, enum ttm_bo_type type, u32 flags,
+ u64 alignment)
{
struct xe_bo *bo = NULL;
int err;
@@ -1468,6 +1491,8 @@ __xe_bo_create_locked(struct xe_device *xe,
if (IS_ERR(bo))
return bo;
+ bo->min_align = alignment;
+
/*
* 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
@@ -1480,19 +1505,29 @@ __xe_bo_create_locked(struct xe_device *xe,
bo->vm = vm;
if (bo->flags & XE_BO_FLAG_GGTT) {
- if (!tile && flags & XE_BO_FLAG_STOLEN)
- tile = xe_device_get_root_tile(xe);
+ struct xe_tile *t;
+ u8 id;
- xe_assert(xe, tile);
+ if (!(bo->flags & XE_BO_FLAG_GGTT_ALL)) {
+ if (!tile && flags & XE_BO_FLAG_STOLEN)
+ tile = xe_device_get_root_tile(xe);
- if (flags & XE_BO_FLAG_FIXED_PLACEMENT) {
- 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);
+ xe_assert(xe, tile);
+ }
+
+ for_each_tile(t, xe, id) {
+ if (t != tile && !(bo->flags & XE_BO_FLAG_GGTTx(t)))
+ continue;
+
+ if (flags & XE_BO_FLAG_FIXED_PLACEMENT) {
+ err = xe_ggtt_insert_bo_at(t->mem.ggtt, bo,
+ start + bo->size, U64_MAX);
+ } else {
+ err = xe_ggtt_insert_bo(t->mem.ggtt, bo);
+ }
+ if (err)
+ goto err_unlock_put_bo;
}
- if (err)
- goto err_unlock_put_bo;
}
return bo;
@@ -1508,16 +1543,18 @@ 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)
+ enum ttm_bo_type type, u32 flags, u64 alignment)
{
- return __xe_bo_create_locked(xe, tile, vm, size, start, end, 0, type, flags);
+ return __xe_bo_create_locked(xe, tile, vm, size, start, end, 0, type,
+ flags, alignment);
}
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);
+ return __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL, 0, type,
+ flags, 0);
}
struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_tile *tile,
@@ -1527,7 +1564,7 @@ struct xe_bo *xe_bo_create_user(struct xe_device *xe, struct xe_tile *tile,
{
struct xe_bo *bo = __xe_bo_create_locked(xe, tile, vm, size, 0, ~0ULL,
cpu_caching, ttm_bo_type_device,
- flags | XE_BO_FLAG_USER);
+ flags | XE_BO_FLAG_USER, 0);
if (!IS_ERR(bo))
xe_bo_unlock_vm_held(bo);
@@ -1551,6 +1588,17 @@ struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile
size_t size, u64 offset,
enum ttm_bo_type type, u32 flags)
{
+ return xe_bo_create_pin_map_at_aligned(xe, tile, vm, size, offset,
+ type, flags, 0);
+}
+
+struct xe_bo *xe_bo_create_pin_map_at_aligned(struct xe_device *xe,
+ struct xe_tile *tile,
+ struct xe_vm *vm,
+ size_t size, u64 offset,
+ enum ttm_bo_type type, u32 flags,
+ u64 alignment)
+{
struct xe_bo *bo;
int err;
u64 start = offset == ~0ull ? 0 : offset;
@@ -1561,7 +1609,8 @@ struct xe_bo *xe_bo_create_pin_map_at(struct xe_device *xe, struct xe_tile *tile
flags |= XE_BO_FLAG_GGTT;
bo = xe_bo_create_locked_range(xe, tile, vm, size, start, end, type,
- flags | XE_BO_FLAG_NEEDS_CPU_ACCESS);
+ flags | XE_BO_FLAG_NEEDS_CPU_ACCESS,
+ alignment);
if (IS_ERR(bo))
return bo;
@@ -1975,6 +2024,7 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data,
struct xe_file *xef = to_xe_file(file);
struct drm_xe_gem_create *args = data;
struct xe_vm *vm = NULL;
+ ktime_t end = 0;
struct xe_bo *bo;
unsigned int bo_flags;
u32 handle;
@@ -2047,6 +2097,10 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data,
vm = xe_vm_lookup(xef, args->vm_id);
if (XE_IOCTL_DBG(xe, !vm))
return -ENOENT;
+ }
+
+retry:
+ if (vm) {
err = xe_vm_lock(vm, true);
if (err)
goto out_vm;
@@ -2060,6 +2114,8 @@ int xe_gem_create_ioctl(struct drm_device *dev, void *data,
if (IS_ERR(bo)) {
err = PTR_ERR(bo);
+ if (xe_vm_validate_should_retry(NULL, err, &end))
+ goto retry;
goto out_vm;
}
@@ -2333,14 +2389,18 @@ void xe_bo_put_commit(struct llist_head *deferred)
void xe_bo_put(struct xe_bo *bo)
{
+ struct xe_tile *tile;
+ u8 id;
+
might_sleep();
if (bo) {
#ifdef CONFIG_PROC_FS
if (bo->client)
might_lock(&bo->client->bos_lock);
#endif
- if (bo->ggtt_node && bo->ggtt_node->ggtt)
- might_lock(&bo->ggtt_node->ggtt->lock);
+ for_each_tile(tile, xe_bo_device(bo), id)
+ if (bo->ggtt_node[id] && bo->ggtt_node[id]->ggtt)
+ might_lock(&bo->ggtt_node[id]->ggtt->lock);
drm_gem_object_put(&bo->ttm.base);
}
}
diff --git a/drivers/gpu/drm/xe/xe_bo.h b/drivers/gpu/drm/xe/xe_bo.h
index d22269a230aa..d04159c59846 100644
--- a/drivers/gpu/drm/xe/xe_bo.h
+++ b/drivers/gpu/drm/xe/xe_bo.h
@@ -39,10 +39,22 @@
#define XE_BO_FLAG_NEEDS_64K BIT(15)
#define XE_BO_FLAG_NEEDS_2M BIT(16)
#define XE_BO_FLAG_GGTT_INVALIDATE BIT(17)
+#define XE_BO_FLAG_GGTT0 BIT(18)
+#define XE_BO_FLAG_GGTT1 BIT(19)
+#define XE_BO_FLAG_GGTT2 BIT(20)
+#define XE_BO_FLAG_GGTT3 BIT(21)
+#define XE_BO_FLAG_GGTT_ALL (XE_BO_FLAG_GGTT0 | \
+ XE_BO_FLAG_GGTT1 | \
+ XE_BO_FLAG_GGTT2 | \
+ XE_BO_FLAG_GGTT3)
+
/* this one is trigger internally only */
#define XE_BO_FLAG_INTERNAL_TEST BIT(30)
#define XE_BO_FLAG_INTERNAL_64K BIT(31)
+#define XE_BO_FLAG_GGTTx(tile) \
+ (XE_BO_FLAG_GGTT0 << (tile)->id)
+
#define XE_PTE_SHIFT 12
#define XE_PAGE_SIZE (1 << XE_PTE_SHIFT)
#define XE_PTE_MASK (XE_PAGE_SIZE - 1)
@@ -77,7 +89,7 @@ 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);
+ enum ttm_bo_type type, u32 flags, u64 alignment);
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);
@@ -94,6 +106,12 @@ struct xe_bo *xe_bo_create_pin_map(struct xe_device *xe, struct xe_tile *tile,
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_pin_map_at_aligned(struct xe_device *xe,
+ struct xe_tile *tile,
+ struct xe_vm *vm,
+ size_t size, u64 offset,
+ enum ttm_bo_type type, u32 flags,
+ u64 alignment);
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);
@@ -188,14 +206,24 @@ xe_bo_main_addr(struct xe_bo *bo, size_t page_size)
}
static inline u32
-xe_bo_ggtt_addr(struct xe_bo *bo)
+__xe_bo_ggtt_addr(struct xe_bo *bo, u8 tile_id)
{
- if (XE_WARN_ON(!bo->ggtt_node))
+ struct xe_ggtt_node *ggtt_node = bo->ggtt_node[tile_id];
+
+ if (XE_WARN_ON(!ggtt_node))
return 0;
- XE_WARN_ON(bo->ggtt_node->base.size > bo->size);
- XE_WARN_ON(bo->ggtt_node->base.start + bo->ggtt_node->base.size > (1ull << 32));
- return bo->ggtt_node->base.start;
+ XE_WARN_ON(ggtt_node->base.size > bo->size);
+ XE_WARN_ON(ggtt_node->base.start + ggtt_node->base.size > (1ull << 32));
+ return ggtt_node->base.start;
+}
+
+static inline u32
+xe_bo_ggtt_addr(struct xe_bo *bo)
+{
+ xe_assert(xe_bo_device(bo), bo->tile);
+
+ return __xe_bo_ggtt_addr(bo, bo->tile->id);
}
int xe_bo_vmap(struct xe_bo *bo);
diff --git a/drivers/gpu/drm/xe/xe_bo_evict.c b/drivers/gpu/drm/xe/xe_bo_evict.c
index 8fb2be061003..6a40eedd9db1 100644
--- a/drivers/gpu/drm/xe/xe_bo_evict.c
+++ b/drivers/gpu/drm/xe/xe_bo_evict.c
@@ -152,11 +152,17 @@ int xe_bo_restore_kernel(struct xe_device *xe)
}
if (bo->flags & XE_BO_FLAG_GGTT) {
- struct xe_tile *tile = bo->tile;
+ struct xe_tile *tile;
+ u8 id;
- mutex_lock(&tile->mem.ggtt->lock);
- xe_ggtt_map_bo(tile->mem.ggtt, bo);
- mutex_unlock(&tile->mem.ggtt->lock);
+ for_each_tile(tile, xe, id) {
+ if (tile != bo->tile && !(bo->flags & XE_BO_FLAG_GGTTx(tile)))
+ continue;
+
+ mutex_lock(&tile->mem.ggtt->lock);
+ xe_ggtt_map_bo(tile->mem.ggtt, bo);
+ mutex_unlock(&tile->mem.ggtt->lock);
+ }
}
/*
diff --git a/drivers/gpu/drm/xe/xe_bo_types.h b/drivers/gpu/drm/xe/xe_bo_types.h
index 2ed558ac2264..aa298d33c250 100644
--- a/drivers/gpu/drm/xe/xe_bo_types.h
+++ b/drivers/gpu/drm/xe/xe_bo_types.h
@@ -13,6 +13,7 @@
#include <drm/ttm/ttm_execbuf_util.h>
#include <drm/ttm/ttm_placement.h>
+#include "xe_device_types.h"
#include "xe_ggtt_types.h"
struct xe_device;
@@ -39,8 +40,8 @@ struct xe_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 xe_ggtt_node *ggtt_node;
+ /** @ggtt_node: Array of GGTT nodes if this BO is mapped in the GGTTs */
+ struct xe_ggtt_node *ggtt_node[XE_MAX_TILES_PER_DEVICE];
/** @vmap: iosys map of this buffer */
struct iosys_map vmap;
/** @ttm_kmap: TTM bo kmap object for internal use only. Keep off. */
@@ -76,6 +77,11 @@ struct xe_bo {
/** @vram_userfault_link: Link into @mem_access.vram_userfault.list */
struct list_head vram_userfault_link;
+
+ /** @min_align: minimum alignment needed for this BO if different
+ * from default
+ */
+ u64 min_align;
};
#define intel_bo_to_drm_bo(bo) (&(bo)->ttm.base)
diff --git a/drivers/gpu/drm/xe/xe_debugfs.c b/drivers/gpu/drm/xe/xe_debugfs.c
index fe4319eb13fd..278ce1c37b39 100644
--- a/drivers/gpu/drm/xe/xe_debugfs.c
+++ b/drivers/gpu/drm/xe/xe_debugfs.c
@@ -147,7 +147,7 @@ static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf,
return -EINVAL;
if (xe->wedged.mode == wedged_mode)
- return 0;
+ return size;
xe->wedged.mode = wedged_mode;
@@ -156,6 +156,7 @@ static ssize_t wedged_mode_set(struct file *f, const char __user *ubuf,
ret = xe_guc_ads_scheduler_policy_toggle_reset(&gt->uc.guc.ads);
if (ret) {
xe_gt_err(gt, "Failed to update GuC ADS scheduler policy. GuC may still cause engine reset even with wedged_mode=2\n");
+ xe_pm_runtime_put(xe);
return -EIO;
}
}
diff --git a/drivers/gpu/drm/xe/xe_devcoredump.c b/drivers/gpu/drm/xe/xe_devcoredump.c
index 8050938389b6..e412a70323cc 100644
--- a/drivers/gpu/drm/xe/xe_devcoredump.c
+++ b/drivers/gpu/drm/xe/xe_devcoredump.c
@@ -197,7 +197,6 @@ static void xe_devcoredump_deferred_snap_work(struct work_struct *work)
struct xe_devcoredump_snapshot *ss = container_of(work, typeof(*ss), work);
struct xe_devcoredump *coredump = container_of(ss, typeof(*coredump), snapshot);
struct xe_device *xe = coredump_to_xe(coredump);
- unsigned int fw_ref;
/*
* NB: Despite passing a GFP_ flags parameter here, more allocations are done
@@ -211,12 +210,11 @@ static void xe_devcoredump_deferred_snap_work(struct work_struct *work)
xe_pm_runtime_get(xe);
/* keep going if fw fails as we still want to save the memory and SW data */
- fw_ref = xe_force_wake_get(gt_to_fw(ss->gt), XE_FORCEWAKE_ALL);
- if (!xe_force_wake_ref_has_domain(fw_ref, XE_FORCEWAKE_ALL))
+ if (xe_force_wake_get(gt_to_fw(ss->gt), XE_FORCEWAKE_ALL))
xe_gt_info(ss->gt, "failed to get forcewake for coredump capture\n");
xe_vm_snapshot_capture_delayed(ss->vm);
xe_guc_exec_queue_snapshot_capture_delayed(ss->ge);
- xe_force_wake_put(gt_to_fw(ss->gt), fw_ref);
+ xe_force_wake_put(gt_to_fw(ss->gt), XE_FORCEWAKE_ALL);
xe_pm_runtime_put(xe);
@@ -243,9 +241,8 @@ static void devcoredump_snapshot(struct xe_devcoredump *coredump,
u32 width_mask = (0x1 << q->width) - 1;
const char *process_name = "no process";
- unsigned int fw_ref;
- bool cookie;
int i;
+ bool cookie;
ss->snapshot_time = ktime_get_real();
ss->boot_time = ktime_get_boottime();
@@ -268,7 +265,8 @@ static void devcoredump_snapshot(struct xe_devcoredump *coredump,
}
/* keep going if fw fails as we still want to save the memory and SW data */
- fw_ref = xe_force_wake_get(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
+ if (xe_force_wake_get(gt_to_fw(q->gt), XE_FORCEWAKE_ALL))
+ xe_gt_info(ss->gt, "failed to get forcewake for coredump capture\n");
ss->ct = xe_guc_ct_snapshot_capture(&guc->ct, true);
ss->ge = xe_guc_exec_queue_snapshot_capture(q);
@@ -286,7 +284,7 @@ static void devcoredump_snapshot(struct xe_devcoredump *coredump,
queue_work(system_unbound_wq, &ss->work);
- xe_force_wake_put(gt_to_fw(q->gt), fw_ref);
+ xe_force_wake_put(gt_to_fw(q->gt), XE_FORCEWAKE_ALL);
dma_fence_end_signalling(cookie);
}
diff --git a/drivers/gpu/drm/xe/xe_device.c b/drivers/gpu/drm/xe/xe_device.c
index bb85208cf1a9..2e1d6d248d2e 100644
--- a/drivers/gpu/drm/xe/xe_device.c
+++ b/drivers/gpu/drm/xe/xe_device.c
@@ -37,6 +37,7 @@
#include "xe_gt_printk.h"
#include "xe_gt_sriov_vf.h"
#include "xe_guc.h"
+#include "xe_guc_pc.h"
#include "xe_hw_engine_group.h"
#include "xe_hwmon.h"
#include "xe_irq.h"
@@ -374,6 +375,11 @@ err:
return ERR_PTR(err);
}
+static bool xe_driver_flr_disabled(struct xe_device *xe)
+{
+ return xe_mmio_read32(xe_root_mmio_gt(xe), GU_CNTL_PROTECTED) & DRIVERINT_FLR_DIS;
+}
+
/*
* 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
@@ -387,17 +393,12 @@ err:
* 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)
+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");
/*
@@ -438,6 +439,16 @@ static void xe_driver_flr(struct xe_device *xe)
xe_mmio_write32(gt, GU_DEBUG, DRIVERFLR_STATUS);
}
+static void xe_driver_flr(struct xe_device *xe)
+{
+ if (xe_driver_flr_disabled(xe)) {
+ drm_info_once(&xe->drm, "BIOS Disabled Driver-FLR\n");
+ return;
+ }
+
+ __xe_driver_flr(xe);
+}
+
static void xe_driver_flr_fini(void *arg)
{
struct xe_device *xe = arg;
@@ -538,6 +549,7 @@ static void update_device_info(struct xe_device *xe)
/* disable features that are not available/applicable to VFs */
if (IS_SRIOV_VF(xe)) {
xe->info.probe_display = 0;
+ xe->info.has_heci_cscfi = 0;
xe->info.has_heci_gscfi = 0;
xe->info.skip_guc_pc = 1;
xe->info.skip_pcode = 1;
@@ -694,7 +706,9 @@ int xe_device_probe(struct xe_device *xe)
}
/* Allocate and map stolen after potential VRAM resize */
- xe_ttm_stolen_mgr_init(xe);
+ err = xe_ttm_stolen_mgr_init(xe);
+ if (err)
+ return err;
/*
* Now that GT is initialized (TTM in particular),
@@ -706,6 +720,12 @@ int xe_device_probe(struct xe_device *xe)
if (err)
goto err;
+ for_each_tile(tile, xe, id) {
+ err = xe_tile_init(tile);
+ if (err)
+ goto err;
+ }
+
for_each_gt(gt, xe, id) {
last_gt = id;
@@ -789,6 +809,24 @@ void xe_device_remove(struct xe_device *xe)
void xe_device_shutdown(struct xe_device *xe)
{
+ struct xe_gt *gt;
+ u8 id;
+
+ drm_dbg(&xe->drm, "Shutting down device\n");
+
+ if (xe_driver_flr_disabled(xe)) {
+ xe_display_pm_shutdown(xe);
+
+ xe_irq_suspend(xe);
+
+ for_each_gt(gt, xe, id)
+ xe_gt_shutdown(gt);
+
+ xe_display_pm_shutdown_late(xe);
+ } else {
+ /* BOOM! */
+ __xe_driver_flr(xe);
+ }
}
/**
@@ -835,31 +873,37 @@ void xe_device_td_flush(struct xe_device *xe)
if (!IS_DGFX(xe) || GRAPHICS_VER(xe) < 20)
return;
- if (XE_WA(xe_root_mmio_gt(xe), 16023588340)) {
+ gt = xe_root_mmio_gt(xe);
+ if (XE_WA(gt, 16023588340)) {
+ /* A transient flush is not sufficient: flush the L2 */
xe_device_l2_flush(xe);
- return;
- }
-
- for_each_gt(gt, xe, id) {
- if (xe_gt_is_media_type(gt))
- continue;
-
- if (xe_force_wake_get(gt_to_fw(gt), XE_FW_GT))
- return;
-
- xe_mmio_write32(gt, XE2_TDF_CTRL, TRANSIENT_FLUSH_REQUEST);
- /*
- * FIXME: We can likely do better here with our choice of
- * timeout. Currently we just assume the worst case, i.e. 150us,
- * which is believed to be sufficient to cover the worst case
- * scenario on current platforms if all cache entries are
- * transient and need to be flushed..
- */
- if (xe_mmio_wait32(gt, XE2_TDF_CTRL, TRANSIENT_FLUSH_REQUEST, 0,
- 150, NULL, false))
- xe_gt_err_once(gt, "TD flush timeout\n");
-
- xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ } else {
+ xe_guc_pc_apply_flush_freq_limit(&gt->uc.guc.pc);
+
+ /* Execute TDF flush on all graphics GTs */
+ for_each_gt(gt, xe, id) {
+ if (xe_gt_is_media_type(gt))
+ continue;
+
+ if (xe_force_wake_get(gt_to_fw(gt), XE_FW_GT))
+ return;
+
+ xe_mmio_write32(gt, XE2_TDF_CTRL, TRANSIENT_FLUSH_REQUEST);
+ /*
+ * FIXME: We can likely do better here with our choice of
+ * timeout. Currently we just assume the worst case, i.e. 150us,
+ * which is believed to be sufficient to cover the worst case
+ * scenario on current platforms if all cache entries are
+ * transient and need to be flushed..
+ */
+ if (xe_mmio_wait32(gt, XE2_TDF_CTRL, TRANSIENT_FLUSH_REQUEST, 0,
+ 150, NULL, false))
+ xe_gt_err_once(gt, "TD flush timeout\n");
+
+ xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
+ }
+
+ xe_guc_pc_remove_flush_freq_limit(&xe_root_mmio_gt(xe)->uc.guc.pc);
}
}
diff --git a/drivers/gpu/drm/xe/xe_force_wake.h b/drivers/gpu/drm/xe/xe_force_wake.h
index 1608a55edc84..a2577672f4e3 100644
--- a/drivers/gpu/drm/xe/xe_force_wake.h
+++ b/drivers/gpu/drm/xe/xe_force_wake.h
@@ -46,20 +46,4 @@ xe_force_wake_assert_held(struct xe_force_wake *fw,
xe_gt_assert(fw->gt, fw->awake_domains & domain);
}
-/**
- * xe_force_wake_ref_has_domain - verifies if the domains are in fw_ref
- * @fw_ref : the force_wake reference
- * @domain : forcewake domain to verify
- *
- * This function confirms whether the @fw_ref includes a reference to the
- * specified @domain.
- *
- * Return: true if domain is refcounted.
- */
-static inline bool
-xe_force_wake_ref_has_domain(unsigned int fw_ref, enum xe_force_wake_domains domain)
-{
- return fw_ref & domain;
-}
-
#endif
diff --git a/drivers/gpu/drm/xe/xe_gen_wa_oob.c b/drivers/gpu/drm/xe/xe_gen_wa_oob.c
index 904cf47925aa..ed9183599e31 100644
--- a/drivers/gpu/drm/xe/xe_gen_wa_oob.c
+++ b/drivers/gpu/drm/xe/xe_gen_wa_oob.c
@@ -28,10 +28,10 @@
"\n" \
"#endif\n"
-static void print_usage(FILE *f)
+static void print_usage(FILE *f, const char *progname)
{
fprintf(f, "usage: %s <input-rule-file> <generated-c-source-file> <generated-c-header-file>\n",
- program_invocation_short_name);
+ progname);
}
static void print_parse_error(const char *err_msg, const char *line,
@@ -144,7 +144,7 @@ int main(int argc, const char *argv[])
if (argc < 3) {
fprintf(stderr, "ERROR: wrong arguments\n");
- print_usage(stderr);
+ print_usage(stderr, argv[0]);
return 1;
}
diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c
index ff19eca5d358..76e1092f51d9 100644
--- a/drivers/gpu/drm/xe/xe_ggtt.c
+++ b/drivers/gpu/drm/xe/xe_ggtt.c
@@ -198,6 +198,13 @@ static const struct xe_ggtt_pt_ops xelpg_pt_wa_ops = {
.ggtt_set_pte = xe_ggtt_set_pte_and_flush,
};
+static void dev_fini_ggtt(void *arg)
+{
+ struct xe_ggtt *ggtt = arg;
+
+ drain_workqueue(ggtt->wq);
+}
+
/**
* xe_ggtt_init_early - Early GGTT initialization
* @ggtt: the &xe_ggtt to be initialized
@@ -254,6 +261,10 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt)
if (err)
return err;
+ err = devm_add_action_or_reset(xe->drm.dev, dev_fini_ggtt, ggtt);
+ if (err)
+ return err;
+
if (IS_SRIOV_VF(xe)) {
err = xe_gt_sriov_vf_prepare_ggtt(xe_tile_get_gt(ggtt->tile, 0));
if (err)
@@ -594,10 +605,10 @@ void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
u64 start;
u64 offset, pte;
- if (XE_WARN_ON(!bo->ggtt_node))
+ if (XE_WARN_ON(!bo->ggtt_node[ggtt->tile->id]))
return;
- start = bo->ggtt_node->base.start;
+ start = bo->ggtt_node[ggtt->tile->id]->base.start;
for (offset = 0; offset < bo->size; offset += XE_PAGE_SIZE) {
pte = ggtt->pt_ops->pte_encode_bo(bo, offset, pat_index);
@@ -608,15 +619,16 @@ void xe_ggtt_map_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
u64 start, u64 end)
{
+ u64 alignment = bo->min_align > 0 ? bo->min_align : XE_PAGE_SIZE;
+ u8 tile_id = ggtt->tile->id;
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)) {
+ if (XE_WARN_ON(bo->ggtt_node[tile_id])) {
/* Someone's already inserted this BO in the GGTT */
- xe_tile_assert(ggtt->tile, bo->ggtt_node->base.size == bo->size);
+ xe_tile_assert(ggtt->tile, bo->ggtt_node[tile_id]->base.size == bo->size);
return 0;
}
@@ -626,19 +638,19 @@ static int __xe_ggtt_insert_bo_at(struct xe_ggtt *ggtt, struct xe_bo *bo,
xe_pm_runtime_get_noresume(tile_to_xe(ggtt->tile));
- bo->ggtt_node = xe_ggtt_node_init(ggtt);
- if (IS_ERR(bo->ggtt_node)) {
- err = PTR_ERR(bo->ggtt_node);
- bo->ggtt_node = NULL;
+ bo->ggtt_node[tile_id] = xe_ggtt_node_init(ggtt);
+ if (IS_ERR(bo->ggtt_node[tile_id])) {
+ err = PTR_ERR(bo->ggtt_node[tile_id]);
+ bo->ggtt_node[tile_id] = NULL;
goto out;
}
mutex_lock(&ggtt->lock);
- err = drm_mm_insert_node_in_range(&ggtt->mm, &bo->ggtt_node->base, bo->size,
- alignment, 0, start, end, 0);
+ err = drm_mm_insert_node_in_range(&ggtt->mm, &bo->ggtt_node[tile_id]->base,
+ bo->size, alignment, 0, start, end, 0);
if (err) {
- xe_ggtt_node_fini(bo->ggtt_node);
- bo->ggtt_node = NULL;
+ xe_ggtt_node_fini(bo->ggtt_node[tile_id]);
+ bo->ggtt_node[tile_id] = NULL;
} else {
xe_ggtt_map_bo(ggtt, bo);
}
@@ -687,13 +699,15 @@ int xe_ggtt_insert_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
*/
void xe_ggtt_remove_bo(struct xe_ggtt *ggtt, struct xe_bo *bo)
{
- if (XE_WARN_ON(!bo->ggtt_node))
+ u8 tile_id = ggtt->tile->id;
+
+ if (XE_WARN_ON(!bo->ggtt_node[tile_id]))
return;
/* This BO is not currently in the GGTT */
- xe_tile_assert(ggtt->tile, bo->ggtt_node->base.size == bo->size);
+ xe_tile_assert(ggtt->tile, bo->ggtt_node[tile_id]->base.size == bo->size);
- xe_ggtt_node_remove(bo->ggtt_node,
+ xe_ggtt_node_remove(bo->ggtt_node[tile_id],
bo->flags & XE_BO_FLAG_GGTT_INVALIDATE);
}
diff --git a/drivers/gpu/drm/xe/xe_gpu_scheduler.h b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
index 64b2ae6839db..400b2e9e89ab 100644
--- a/drivers/gpu/drm/xe/xe_gpu_scheduler.h
+++ b/drivers/gpu/drm/xe/xe_gpu_scheduler.h
@@ -51,7 +51,15 @@ static inline void xe_sched_tdr_queue_imm(struct xe_gpu_scheduler *sched)
static inline void xe_sched_resubmit_jobs(struct xe_gpu_scheduler *sched)
{
- drm_sched_resubmit_jobs(&sched->base);
+ struct drm_sched_job *s_job;
+
+ list_for_each_entry(s_job, &sched->base.pending_list, list) {
+ struct drm_sched_fence *s_fence = s_job->s_fence;
+ struct dma_fence *hw_fence = s_fence->parent;
+
+ if (hw_fence && !dma_fence_is_signaled(hw_fence))
+ sched->base.ops->run_job(s_job);
+ }
}
static inline bool
diff --git a/drivers/gpu/drm/xe/xe_gsc.c b/drivers/gpu/drm/xe/xe_gsc.c
index 6fbea70d3d36..feb680d127e6 100644
--- a/drivers/gpu/drm/xe/xe_gsc.c
+++ b/drivers/gpu/drm/xe/xe_gsc.c
@@ -564,6 +564,28 @@ void xe_gsc_remove(struct xe_gsc *gsc)
xe_gsc_proxy_remove(gsc);
}
+void xe_gsc_stop_prepare(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+ int ret;
+
+ if (!xe_uc_fw_is_loadable(&gsc->fw) || xe_uc_fw_is_in_error_state(&gsc->fw))
+ return;
+
+ xe_force_wake_assert_held(gt_to_fw(gt), XE_FW_GSC);
+
+ /*
+ * If the GSC FW load or the proxy init are interrupted, the only way
+ * to recover it is to do an FLR and reload the GSC from scratch.
+ * Therefore, let's wait for the init to complete before stopping
+ * operations. The proxy init is the last step, so we can just wait on
+ * that
+ */
+ ret = xe_gsc_wait_for_proxy_init_done(gsc);
+ if (ret)
+ xe_gt_err(gt, "failed to wait for GSC init completion before uc stop\n");
+}
+
/*
* 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
diff --git a/drivers/gpu/drm/xe/xe_gsc.h b/drivers/gpu/drm/xe/xe_gsc.h
index e282b9ef6ec4..c31fe24c4b66 100644
--- a/drivers/gpu/drm/xe/xe_gsc.h
+++ b/drivers/gpu/drm/xe/xe_gsc.h
@@ -16,6 +16,7 @@ struct xe_hw_engine;
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_stop_prepare(struct xe_gsc *gsc);
void xe_gsc_load_start(struct xe_gsc *gsc);
void xe_gsc_remove(struct xe_gsc *gsc);
void xe_gsc_hwe_irq_handler(struct xe_hw_engine *hwe, u16 intr_vec);
diff --git a/drivers/gpu/drm/xe/xe_gsc_proxy.c b/drivers/gpu/drm/xe/xe_gsc_proxy.c
index 2d6ea8c01445..85801de4fab1 100644
--- a/drivers/gpu/drm/xe/xe_gsc_proxy.c
+++ b/drivers/gpu/drm/xe/xe_gsc_proxy.c
@@ -71,6 +71,17 @@ bool xe_gsc_proxy_init_done(struct xe_gsc *gsc)
HECI1_FWSTS1_PROXY_STATE_NORMAL;
}
+int xe_gsc_wait_for_proxy_init_done(struct xe_gsc *gsc)
+{
+ struct xe_gt *gt = gsc_to_gt(gsc);
+
+ /* Proxy init can take up to 500ms, so wait double that for safety */
+ return xe_mmio_wait32(gt, HECI_FWSTS1(MTL_GSC_HECI1_BASE),
+ HECI1_FWSTS1_CURRENT_STATE,
+ HECI1_FWSTS1_PROXY_STATE_NORMAL,
+ USEC_PER_SEC, NULL, false);
+}
+
static void __gsc_proxy_irq_rmw(struct xe_gsc *gsc, u32 clr, u32 set)
{
struct xe_gt *gt = gsc_to_gt(gsc);
diff --git a/drivers/gpu/drm/xe/xe_gsc_proxy.h b/drivers/gpu/drm/xe/xe_gsc_proxy.h
index c511ade6b863..e2498aa6de18 100644
--- a/drivers/gpu/drm/xe/xe_gsc_proxy.h
+++ b/drivers/gpu/drm/xe/xe_gsc_proxy.h
@@ -13,6 +13,7 @@ struct xe_gsc;
int xe_gsc_proxy_init(struct xe_gsc *gsc);
bool xe_gsc_proxy_init_done(struct xe_gsc *gsc);
void xe_gsc_proxy_remove(struct xe_gsc *gsc);
+int xe_gsc_wait_for_proxy_init_done(struct xe_gsc *gsc);
int xe_gsc_proxy_start(struct xe_gsc *gsc);
int xe_gsc_proxy_request_handler(struct xe_gsc *gsc);
diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
index 17ba15132a98..3b53d46aad54 100644
--- a/drivers/gpu/drm/xe/xe_gt.c
+++ b/drivers/gpu/drm/xe/xe_gt.c
@@ -114,7 +114,7 @@ static void xe_gt_enable_host_l2_vram(struct xe_gt *gt)
xe_gt_mcr_multicast_write(gt, XE2_GAMREQSTRM_CTRL, reg);
}
- xe_gt_mcr_multicast_write(gt, XEHPC_L3CLOS_MASK(3), 0x3);
+ xe_gt_mcr_multicast_write(gt, XEHPC_L3CLOS_MASK(3), 0xF);
xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
}
@@ -389,6 +389,8 @@ int xe_gt_init_early(struct xe_gt *gt)
if (err)
return err;
+ xe_mocs_init_early(gt);
+
return 0;
}
@@ -590,17 +592,15 @@ int xe_gt_init(struct xe_gt *gt)
xe_hw_fence_irq_init(&gt->fence_irq[i]);
}
- err = xe_gt_pagefault_init(gt);
+ err = xe_gt_sysfs_init(gt);
if (err)
return err;
- xe_mocs_init_early(gt);
-
- err = xe_gt_sysfs_init(gt);
+ err = gt_fw_domain_init(gt);
if (err)
return err;
- err = gt_fw_domain_init(gt);
+ err = xe_gt_pagefault_init(gt);
if (err)
return err;
@@ -768,6 +768,9 @@ static int gt_reset(struct xe_gt *gt)
if (err)
goto err_msg;
+ if (IS_SRIOV_PF(gt_to_xe(gt)))
+ xe_gt_sriov_pf_stop_prepare(gt);
+
xe_uc_gucrc_disable(&gt->uc);
xe_uc_stop_prepare(&gt->uc);
xe_gt_pagefault_reset(gt);
@@ -828,7 +831,7 @@ void xe_gt_suspend_prepare(struct xe_gt *gt)
{
XE_WARN_ON(xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL));
- xe_uc_stop_prepare(&gt->uc);
+ xe_uc_suspend_prepare(&gt->uc);
XE_WARN_ON(xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL));
}
@@ -865,6 +868,13 @@ err_msg:
return err;
}
+void xe_gt_shutdown(struct xe_gt *gt)
+{
+ xe_force_wake_get(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+ do_gt_reset(gt);
+ xe_force_wake_put(gt_to_fw(gt), XE_FORCEWAKE_ALL);
+}
+
/**
* xe_gt_sanitize_freq() - Restore saved frequencies if necessary.
* @gt: the GT object
diff --git a/drivers/gpu/drm/xe/xe_gt.h b/drivers/gpu/drm/xe/xe_gt.h
index ee138e9768a2..881f1cbc2c49 100644
--- a/drivers/gpu/drm/xe/xe_gt.h
+++ b/drivers/gpu/drm/xe/xe_gt.h
@@ -48,6 +48,7 @@ void xe_gt_record_user_engines(struct xe_gt *gt);
void xe_gt_suspend_prepare(struct xe_gt *gt);
int xe_gt_suspend(struct xe_gt *gt);
+void xe_gt_shutdown(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);
diff --git a/drivers/gpu/drm/xe/xe_gt_freq.c b/drivers/gpu/drm/xe/xe_gt_freq.c
index ab76973f3e1e..a05fde2c7b12 100644
--- a/drivers/gpu/drm/xe/xe_gt_freq.c
+++ b/drivers/gpu/drm/xe/xe_gt_freq.c
@@ -32,6 +32,7 @@
* 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.
@@ -39,6 +40,7 @@
* - 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.
diff --git a/drivers/gpu/drm/xe/xe_gt_pagefault.c b/drivers/gpu/drm/xe/xe_gt_pagefault.c
index 79c426dc2505..656c2ab6ca9f 100644
--- a/drivers/gpu/drm/xe/xe_gt_pagefault.c
+++ b/drivers/gpu/drm/xe/xe_gt_pagefault.c
@@ -423,9 +423,17 @@ static int xe_alloc_pf_queue(struct xe_gt *gt, struct pf_queue *pf_queue)
num_eus = bitmap_weight(gt->fuse_topo.eu_mask_per_dss,
XE_MAX_EU_FUSE_BITS) * num_dss;
- /* user can issue separate page faults per EU and per CS */
+ /*
+ * user can issue separate page faults per EU and per CS
+ *
+ * XXX: Multiplier required as compute UMD are getting PF queue errors
+ * without it. Follow on why this multiplier is required.
+ */
+#define PF_MULTIPLIER 8
pf_queue->num_dw =
- (num_eus + XE_NUM_HW_ENGINES) * PF_MSG_LEN_DW;
+ (num_eus + XE_NUM_HW_ENGINES) * PF_MSG_LEN_DW * PF_MULTIPLIER;
+ pf_queue->num_dw = roundup_pow_of_two(pf_queue->num_dw);
+#undef PF_MULTIPLIER
pf_queue->gt = gt;
pf_queue->data = devm_kcalloc(xe->drm.dev, pf_queue->num_dw,
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
index 905f409db74b..57e9eddc092e 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf.c
@@ -5,14 +5,20 @@
#include <drm/drm_managed.h>
+#include "regs/xe_guc_regs.h"
#include "regs/xe_regs.h"
+#include "xe_gt.h"
#include "xe_gt_sriov_pf.h"
#include "xe_gt_sriov_pf_config.h"
#include "xe_gt_sriov_pf_control.h"
#include "xe_gt_sriov_pf_helpers.h"
#include "xe_gt_sriov_pf_service.h"
+#include "xe_gt_sriov_printk.h"
#include "xe_mmio.h"
+#include "xe_pm.h"
+
+static void pf_worker_restart_func(struct work_struct *w);
/*
* VF's metadata is maintained in the flexible array where:
@@ -38,6 +44,11 @@ static int pf_alloc_metadata(struct xe_gt *gt)
return 0;
}
+static void pf_init_workers(struct xe_gt *gt)
+{
+ INIT_WORK(&gt->sriov.pf.workers.restart, pf_worker_restart_func);
+}
+
/**
* xe_gt_sriov_pf_init_early - Prepare SR-IOV PF data structures on PF.
* @gt: the &xe_gt to initialize
@@ -62,6 +73,8 @@ int xe_gt_sriov_pf_init_early(struct xe_gt *gt)
if (err)
return err;
+ pf_init_workers(gt);
+
return 0;
}
@@ -89,14 +102,111 @@ void xe_gt_sriov_pf_init_hw(struct xe_gt *gt)
xe_gt_sriov_pf_service_update(gt);
}
+static u32 pf_get_vf_regs_stride(struct xe_device *xe)
+{
+ return GRAPHICS_VERx100(xe) > 1200 ? 0x400 : 0x1000;
+}
+
+static struct xe_reg xe_reg_vf_to_pf(struct xe_reg vf_reg, unsigned int vfid, u32 stride)
+{
+ struct xe_reg pf_reg = vf_reg;
+
+ pf_reg.vf = 0;
+ pf_reg.addr += stride * vfid;
+
+ return pf_reg;
+}
+
+static void pf_clear_vf_scratch_regs(struct xe_gt *gt, unsigned int vfid)
+{
+ u32 stride = pf_get_vf_regs_stride(gt_to_xe(gt));
+ struct xe_reg scratch;
+ int n, count;
+
+ if (xe_gt_is_media_type(gt)) {
+ count = MED_VF_SW_FLAG_COUNT;
+ for (n = 0; n < count; n++) {
+ scratch = xe_reg_vf_to_pf(MED_VF_SW_FLAG(n), vfid, stride);
+ xe_mmio_write32(gt, scratch, 0);
+ }
+ } else {
+ count = VF_SW_FLAG_COUNT;
+ for (n = 0; n < count; n++) {
+ scratch = xe_reg_vf_to_pf(VF_SW_FLAG(n), vfid, stride);
+ xe_mmio_write32(gt, scratch, 0);
+ }
+ }
+}
+
/**
- * xe_gt_sriov_pf_restart - Restart SR-IOV support after a GT reset.
+ * xe_gt_sriov_pf_sanitize_hw() - Reset hardware state related to a VF.
* @gt: the &xe_gt
+ * @vfid: the VF identifier
*
* This function can only be called on PF.
*/
-void xe_gt_sriov_pf_restart(struct xe_gt *gt)
+void xe_gt_sriov_pf_sanitize_hw(struct xe_gt *gt, unsigned int vfid)
+{
+ xe_gt_assert(gt, IS_SRIOV_PF(gt_to_xe(gt)));
+
+ pf_clear_vf_scratch_regs(gt, vfid);
+}
+
+static void pf_cancel_restart(struct xe_gt *gt)
+{
+ xe_gt_assert(gt, IS_SRIOV_PF(gt_to_xe(gt)));
+
+ if (cancel_work_sync(&gt->sriov.pf.workers.restart))
+ xe_gt_sriov_dbg_verbose(gt, "pending restart canceled!\n");
+}
+
+/**
+ * xe_gt_sriov_pf_stop_prepare() - Prepare to stop SR-IOV support.
+ * @gt: the &xe_gt
+ *
+ * This function can only be called on the PF.
+ */
+void xe_gt_sriov_pf_stop_prepare(struct xe_gt *gt)
{
+ pf_cancel_restart(gt);
+}
+
+static void pf_restart(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ xe_pm_runtime_get(xe);
xe_gt_sriov_pf_config_restart(gt);
xe_gt_sriov_pf_control_restart(gt);
+ xe_pm_runtime_put(xe);
+
+ xe_gt_sriov_dbg(gt, "restart completed\n");
+}
+
+static void pf_worker_restart_func(struct work_struct *w)
+{
+ struct xe_gt *gt = container_of(w, typeof(*gt), sriov.pf.workers.restart);
+
+ pf_restart(gt);
+}
+
+static void pf_queue_restart(struct xe_gt *gt)
+{
+ struct xe_device *xe = gt_to_xe(gt);
+
+ xe_gt_assert(gt, IS_SRIOV_PF(xe));
+
+ if (!queue_work(xe->sriov.wq, &gt->sriov.pf.workers.restart))
+ xe_gt_sriov_dbg(gt, "restart already in queue!\n");
+}
+
+/**
+ * xe_gt_sriov_pf_restart - Restart SR-IOV support after a GT reset.
+ * @gt: the &xe_gt
+ *
+ * This function can only be called on PF.
+ */
+void xe_gt_sriov_pf_restart(struct xe_gt *gt)
+{
+ pf_queue_restart(gt);
}
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf.h b/drivers/gpu/drm/xe/xe_gt_sriov_pf.h
index f0cb726a6919..165ba31d0391 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf.h
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf.h
@@ -11,6 +11,8 @@ struct xe_gt;
#ifdef CONFIG_PCI_IOV
int xe_gt_sriov_pf_init_early(struct xe_gt *gt);
void xe_gt_sriov_pf_init_hw(struct xe_gt *gt);
+void xe_gt_sriov_pf_sanitize_hw(struct xe_gt *gt, unsigned int vfid);
+void xe_gt_sriov_pf_stop_prepare(struct xe_gt *gt);
void xe_gt_sriov_pf_restart(struct xe_gt *gt);
#else
static inline int xe_gt_sriov_pf_init_early(struct xe_gt *gt)
@@ -22,6 +24,10 @@ static inline void xe_gt_sriov_pf_init_hw(struct xe_gt *gt)
{
}
+static inline void xe_gt_sriov_pf_stop_prepare(struct xe_gt *gt)
+{
+}
+
static inline void xe_gt_sriov_pf_restart(struct xe_gt *gt)
{
}
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
index c9ed996b9cb0..786f0dba4143 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
@@ -323,6 +323,26 @@ static int pf_push_full_vf_config(struct xe_gt *gt, unsigned int vfid)
return err;
}
+static int pf_push_vf_cfg(struct xe_gt *gt, unsigned int vfid, bool reset)
+{
+ int err = 0;
+
+ xe_gt_assert(gt, vfid);
+ lockdep_assert_held(xe_gt_sriov_pf_master_mutex(gt));
+
+ if (reset)
+ err = pf_send_vf_cfg_reset(gt, vfid);
+ if (!err)
+ err = pf_push_full_vf_config(gt, vfid);
+
+ return err;
+}
+
+static int pf_refresh_vf_cfg(struct xe_gt *gt, unsigned int vfid)
+{
+ return pf_push_vf_cfg(gt, vfid, true);
+}
+
static u64 pf_get_ggtt_alignment(struct xe_gt *gt)
{
struct xe_device *xe = gt_to_xe(gt);
@@ -419,6 +439,10 @@ static int pf_provision_vf_ggtt(struct xe_gt *gt, unsigned int vfid, u64 size)
return err;
pf_release_vf_config_ggtt(gt, config);
+
+ err = pf_refresh_vf_cfg(gt, vfid);
+ if (unlikely(err))
+ return err;
}
xe_gt_assert(gt, !xe_ggtt_node_allocated(config->ggtt_region));
@@ -744,6 +768,10 @@ static int pf_provision_vf_ctxs(struct xe_gt *gt, unsigned int vfid, u32 num_ctx
return ret;
pf_release_config_ctxs(gt, config);
+
+ ret = pf_refresh_vf_cfg(gt, vfid);
+ if (unlikely(ret))
+ return ret;
}
if (!num_ctxs)
@@ -1041,6 +1069,10 @@ static int pf_provision_vf_dbs(struct xe_gt *gt, unsigned int vfid, u32 num_dbs)
return ret;
pf_release_config_dbs(gt, config);
+
+ ret = pf_refresh_vf_cfg(gt, vfid);
+ if (unlikely(ret))
+ return ret;
}
if (!num_dbs)
@@ -2003,10 +2035,7 @@ int xe_gt_sriov_pf_config_push(struct xe_gt *gt, unsigned int vfid, bool refresh
xe_gt_assert(gt, vfid);
mutex_lock(xe_gt_sriov_pf_master_mutex(gt));
- if (refresh)
- err = pf_send_vf_cfg_reset(gt, vfid);
- if (!err)
- err = pf_push_full_vf_config(gt, vfid);
+ err = pf_push_vf_cfg(gt, vfid, refresh);
mutex_unlock(xe_gt_sriov_pf_master_mutex(gt));
if (unlikely(err)) {
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c b/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
index 02f7328bd6ce..b4fd5a81aff1 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_control.c
@@ -9,6 +9,7 @@
#include "xe_device.h"
#include "xe_gt.h"
+#include "xe_gt_sriov_pf.h"
#include "xe_gt_sriov_pf_config.h"
#include "xe_gt_sriov_pf_control.h"
#include "xe_gt_sriov_pf_helpers.h"
@@ -1008,7 +1009,7 @@ static bool pf_exit_vf_flr_reset_mmio(struct xe_gt *gt, unsigned int vfid)
if (!pf_exit_vf_state(gt, vfid, XE_GT_SRIOV_STATE_FLR_RESET_MMIO))
return false;
- /* XXX: placeholder */
+ xe_gt_sriov_pf_sanitize_hw(gt, vfid);
pf_enter_vf_flr_send_finish(gt, vfid);
return true;
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_pf_types.h b/drivers/gpu/drm/xe/xe_gt_sriov_pf_types.h
index 28e1b130bf87..a69d128c4f45 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_pf_types.h
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_pf_types.h
@@ -32,7 +32,16 @@ struct xe_gt_sriov_metadata {
};
/**
+ * struct xe_gt_sriov_pf_workers - GT level workers used by the PF.
+ */
+struct xe_gt_sriov_pf_workers {
+ /** @restart: worker that executes actions post GT reset */
+ struct work_struct restart;
+};
+
+/**
* struct xe_gt_sriov_pf - GT level PF virtualization data.
+ * @workers: workers data.
* @service: service data.
* @control: control data.
* @policy: policy data.
@@ -40,6 +49,7 @@ struct xe_gt_sriov_metadata {
* @vfs: metadata for all VFs.
*/
struct xe_gt_sriov_pf {
+ struct xe_gt_sriov_pf_workers workers;
struct xe_gt_sriov_pf_service service;
struct xe_gt_sriov_pf_control control;
struct xe_gt_sriov_pf_policy policy;
diff --git a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
index f982d6f9f218..29badbd829ab 100644
--- a/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
+++ b/drivers/gpu/drm/xe/xe_gt_sriov_vf.c
@@ -46,12 +46,19 @@ static int guc_action_vf_reset(struct xe_guc *guc)
return ret > 0 ? -EPROTO : ret;
}
+#define GUC_RESET_VF_STATE_RETRY_MAX 10
static int vf_reset_guc_state(struct xe_gt *gt)
{
+ unsigned int retry = GUC_RESET_VF_STATE_RETRY_MAX;
struct xe_guc *guc = &gt->uc.guc;
int err;
- err = guc_action_vf_reset(guc);
+ do {
+ err = guc_action_vf_reset(guc);
+ if (!err || err != -ETIMEDOUT)
+ break;
+ } while (--retry);
+
if (unlikely(err))
xe_gt_sriov_err(gt, "Failed to reset GuC state (%pe)\n", ERR_PTR(err));
return err;
@@ -228,6 +235,9 @@ int xe_gt_sriov_vf_bootstrap(struct xe_gt *gt)
{
int err;
+ if (!xe_device_uc_enabled(gt_to_xe(gt)))
+ return -ENODEV;
+
err = vf_reset_guc_state(gt);
if (unlikely(err))
return err;
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
index 98a450271f5c..9deb9b44c3c3 100644
--- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.c
@@ -138,6 +138,14 @@ void xe_gt_tlb_invalidation_reset(struct xe_gt *gt)
int pending_seqno;
/*
+ * we can get here before the CTs are even initialized if we're wedging
+ * very early, in which case there are not going to be any pending
+ * fences so we can bail immediately.
+ */
+ if (!xe_guc_ct_initialized(&gt->uc.guc.ct))
+ return;
+
+ /*
* CT channel is already disabled at this point. No new TLB requests can
* appear.
*/
@@ -407,6 +415,28 @@ int xe_gt_tlb_invalidation_range(struct xe_gt *gt,
}
/**
+ * xe_gt_tlb_invalidation_vm - Issue a TLB invalidation on this GT for a VM
+ * @gt: graphics tile
+ * @vm: VM to invalidate
+ *
+ * Invalidate entire VM's address space
+ */
+void xe_gt_tlb_invalidation_vm(struct xe_gt *gt, struct xe_vm *vm)
+{
+ struct xe_gt_tlb_invalidation_fence fence;
+ u64 range = 1ull << vm->xe->info.va_bits;
+ int ret;
+
+ xe_gt_tlb_invalidation_fence_init(gt, &fence, true);
+
+ ret = xe_gt_tlb_invalidation_range(gt, &fence, 0, range, vm->usm.asid);
+ if (ret < 0)
+ return;
+
+ xe_gt_tlb_invalidation_fence_wait(&fence);
+}
+
+/**
* 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
diff --git a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h
index 672acfcdf0d7..abe9b03d543e 100644
--- a/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h
+++ b/drivers/gpu/drm/xe/xe_gt_tlb_invalidation.h
@@ -12,6 +12,7 @@
struct xe_gt;
struct xe_guc;
+struct xe_vm;
struct xe_vma;
int xe_gt_tlb_invalidation_init_early(struct xe_gt *gt);
@@ -21,6 +22,7 @@ int xe_gt_tlb_invalidation_ggtt(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);
+void xe_gt_tlb_invalidation_vm(struct xe_gt *gt, struct xe_vm *vm);
int xe_gt_tlb_invalidation_range(struct xe_gt *gt,
struct xe_gt_tlb_invalidation_fence *fence,
u64 start, u64 end, u32 asid);
diff --git a/drivers/gpu/drm/xe/xe_guc.c b/drivers/gpu/drm/xe/xe_guc.c
index 52df28032a6f..96373cdb366b 100644
--- a/drivers/gpu/drm/xe/xe_guc.c
+++ b/drivers/gpu/drm/xe/xe_guc.c
@@ -985,7 +985,7 @@ timeout:
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,
+ ret = xe_mmio_wait32(gt, reply_reg, resp_mask, resp_mask,
1000000, &header, false);
if (unlikely(FIELD_GET(GUC_HXG_MSG_0_ORIGIN, header) !=
@@ -1175,7 +1175,7 @@ void xe_guc_print_info(struct xe_guc *guc, struct drm_printer *p)
xe_force_wake_put(gt_to_fw(gt), XE_FW_GT);
- xe_guc_ct_print(&guc->ct, p, false);
+ xe_guc_ct_print(&guc->ct, p);
xe_guc_submit_print(guc, p);
}
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.c b/drivers/gpu/drm/xe/xe_guc_ct.c
index cd6a5f09d631..f1ce4e14dcb5 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct.c
+++ b/drivers/gpu/drm/xe/xe_guc_ct.c
@@ -25,12 +25,53 @@
#include "xe_gt_sriov_pf_monitor.h"
#include "xe_gt_tlb_invalidation.h"
#include "xe_guc.h"
+#include "xe_guc_log.h"
#include "xe_guc_relay.h"
#include "xe_guc_submit.h"
#include "xe_map.h"
#include "xe_pm.h"
#include "xe_trace_guc.h"
+static void receive_g2h(struct xe_guc_ct *ct);
+static void g2h_worker_func(struct work_struct *w);
+static void safe_mode_worker_func(struct work_struct *w);
+static void ct_exit_safe_mode(struct xe_guc_ct *ct);
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+enum {
+ /* Internal states, not error conditions */
+ CT_DEAD_STATE_REARM, /* 0x0001 */
+ CT_DEAD_STATE_CAPTURE, /* 0x0002 */
+
+ /* Error conditions */
+ CT_DEAD_SETUP, /* 0x0004 */
+ CT_DEAD_H2G_WRITE, /* 0x0008 */
+ CT_DEAD_H2G_HAS_ROOM, /* 0x0010 */
+ CT_DEAD_G2H_READ, /* 0x0020 */
+ CT_DEAD_G2H_RECV, /* 0x0040 */
+ CT_DEAD_G2H_RELEASE, /* 0x0080 */
+ CT_DEAD_DEADLOCK, /* 0x0100 */
+ CT_DEAD_PROCESS_FAILED, /* 0x0200 */
+ CT_DEAD_FAST_G2H, /* 0x0400 */
+ CT_DEAD_PARSE_G2H_RESPONSE, /* 0x0800 */
+ CT_DEAD_PARSE_G2H_UNKNOWN, /* 0x1000 */
+ CT_DEAD_PARSE_G2H_ORIGIN, /* 0x2000 */
+ CT_DEAD_PARSE_G2H_TYPE, /* 0x4000 */
+};
+
+static void ct_dead_worker_func(struct work_struct *w);
+static void ct_dead_capture(struct xe_guc_ct *ct, struct guc_ctb *ctb, u32 reason_code);
+
+#define CT_DEAD(ct, ctb, reason_code) ct_dead_capture((ct), (ctb), CT_DEAD_##reason_code)
+#else
+#define CT_DEAD(ct, ctb, reason) \
+ do { \
+ struct guc_ctb *_ctb = (ctb); \
+ if (_ctb) \
+ _ctb->info.broken = true; \
+ } while (0)
+#endif
+
/* Used when a CT send wants to block and / or receive data */
struct g2h_fence {
u32 *response_buffer;
@@ -147,14 +188,11 @@ static void guc_ct_fini(struct drm_device *drm, void *arg)
{
struct xe_guc_ct *ct = arg;
+ ct_exit_safe_mode(ct);
destroy_workqueue(ct->g2h_wq);
xa_destroy(&ct->fence_lookup);
}
-static void receive_g2h(struct xe_guc_ct *ct);
-static void g2h_worker_func(struct work_struct *w);
-static void safe_mode_worker_func(struct work_struct *w);
-
static void primelockdep(struct xe_guc_ct *ct)
{
if (!IS_ENABLED(CONFIG_LOCKDEP))
@@ -182,7 +220,11 @@ int xe_guc_ct_init(struct xe_guc_ct *ct)
spin_lock_init(&ct->fast_lock);
xa_init(&ct->fence_lookup);
INIT_WORK(&ct->g2h_worker, g2h_worker_func);
- INIT_DELAYED_WORK(&ct->safe_mode_worker, safe_mode_worker_func);
+ INIT_DELAYED_WORK(&ct->safe_mode_worker, safe_mode_worker_func);
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+ spin_lock_init(&ct->dead.lock);
+ INIT_WORK(&ct->dead.worker, ct_dead_worker_func);
+#endif
init_waitqueue_head(&ct->wq);
init_waitqueue_head(&ct->g2h_fence_wq);
@@ -419,10 +461,22 @@ int xe_guc_ct_enable(struct xe_guc_ct *ct)
if (ct_needs_safe_mode(ct))
ct_enter_safe_mode(ct);
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+ /*
+ * The CT has now been reset so the dumper can be re-armed
+ * after any existing dead state has been dumped.
+ */
+ spin_lock_irq(&ct->dead.lock);
+ if (ct->dead.reason)
+ ct->dead.reason |= (1 << CT_DEAD_STATE_REARM);
+ spin_unlock_irq(&ct->dead.lock);
+#endif
+
return 0;
err_out:
xe_gt_err(gt, "Failed to enable GuC CT (%pe)\n", ERR_PTR(err));
+ CT_DEAD(ct, NULL, SETUP);
return err;
}
@@ -454,6 +508,9 @@ void xe_guc_ct_disable(struct xe_guc_ct *ct)
*/
void xe_guc_ct_stop(struct xe_guc_ct *ct)
{
+ if (!xe_guc_ct_initialized(ct))
+ return;
+
xe_guc_ct_set_state(ct, XE_GUC_CT_STATE_STOPPED);
stop_g2h_handler(ct);
}
@@ -466,6 +523,19 @@ static bool h2g_has_room(struct xe_guc_ct *ct, u32 cmd_len)
if (cmd_len > h2g->info.space) {
h2g->info.head = desc_read(ct_to_xe(ct), h2g, head);
+
+ if (h2g->info.head > h2g->info.size) {
+ struct xe_device *xe = ct_to_xe(ct);
+ u32 desc_status = desc_read(xe, h2g, status);
+
+ desc_write(xe, h2g, status, desc_status | GUC_CTB_STATUS_OVERFLOW);
+
+ xe_gt_err(ct_to_gt(ct), "CT: invalid head offset %u >= %u)\n",
+ h2g->info.head, h2g->info.size);
+ CT_DEAD(ct, h2g, H2G_HAS_ROOM);
+ return false;
+ }
+
h2g->info.space = CIRC_SPACE(h2g->info.tail, h2g->info.head,
h2g->info.size) -
h2g->info.resv_space;
@@ -521,10 +591,24 @@ static void __g2h_reserve_space(struct xe_guc_ct *ct, u32 g2h_len, u32 num_g2h)
static void __g2h_release_space(struct xe_guc_ct *ct, u32 g2h_len)
{
+ bool bad = false;
+
lockdep_assert_held(&ct->fast_lock);
- xe_gt_assert(ct_to_gt(ct), ct->ctbs.g2h.info.space + g2h_len <=
- ct->ctbs.g2h.info.size - ct->ctbs.g2h.info.resv_space);
- xe_gt_assert(ct_to_gt(ct), ct->g2h_outstanding);
+
+ bad = ct->ctbs.g2h.info.space + g2h_len >
+ ct->ctbs.g2h.info.size - ct->ctbs.g2h.info.resv_space;
+ bad |= !ct->g2h_outstanding;
+
+ if (bad) {
+ xe_gt_err(ct_to_gt(ct), "Invalid G2H release: %d + %d vs %d - %d -> %d vs %d, outstanding = %d!\n",
+ 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->ctbs.g2h.info.size - ct->ctbs.g2h.info.resv_space,
+ ct->g2h_outstanding);
+ CT_DEAD(ct, &ct->ctbs.g2h, G2H_RELEASE);
+ return;
+ }
ct->ctbs.g2h.info.space += g2h_len;
if (!--ct->g2h_outstanding)
@@ -551,12 +635,43 @@ static int h2g_write(struct xe_guc_ct *ct, const u32 *action, u32 len,
u32 full_len;
struct iosys_map map = IOSYS_MAP_INIT_OFFSET(&h2g->cmds,
tail * sizeof(u32));
+ u32 desc_status;
full_len = len + GUC_CTB_HDR_LEN;
lockdep_assert_held(&ct->lock);
xe_gt_assert(gt, full_len <= GUC_CTB_MSG_MAX_LEN);
- xe_gt_assert(gt, tail <= h2g->info.size);
+
+ desc_status = desc_read(xe, h2g, status);
+ if (desc_status) {
+ xe_gt_err(gt, "CT write: non-zero status: %u\n", desc_status);
+ goto corrupted;
+ }
+
+ if (IS_ENABLED(CONFIG_DRM_XE_DEBUG)) {
+ u32 desc_tail = desc_read(xe, h2g, tail);
+ u32 desc_head = desc_read(xe, h2g, head);
+
+ if (tail != desc_tail) {
+ desc_write(xe, h2g, status, desc_status | GUC_CTB_STATUS_MISMATCH);
+ xe_gt_err(gt, "CT write: tail was modified %u != %u\n", desc_tail, tail);
+ goto corrupted;
+ }
+
+ if (tail > h2g->info.size) {
+ desc_write(xe, h2g, status, desc_status | GUC_CTB_STATUS_OVERFLOW);
+ xe_gt_err(gt, "CT write: tail out of range: %u vs %u\n",
+ tail, h2g->info.size);
+ goto corrupted;
+ }
+
+ if (desc_head >= h2g->info.size) {
+ desc_write(xe, h2g, status, desc_status | GUC_CTB_STATUS_OVERFLOW);
+ xe_gt_err(gt, "CT write: invalid head offset %u >= %u)\n",
+ desc_head, h2g->info.size);
+ goto corrupted;
+ }
+ }
/* Command will wrap, zero fill (NOPs), return and check credits again */
if (tail + full_len > h2g->info.size) {
@@ -609,6 +724,10 @@ static int h2g_write(struct xe_guc_ct *ct, const u32 *action, u32 len,
desc_read(xe, h2g, head), h2g->info.tail);
return 0;
+
+corrupted:
+ CT_DEAD(ct, &ct->ctbs.h2g, H2G_WRITE);
+ return -EPIPE;
}
/*
@@ -638,7 +757,7 @@ static int __guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action,
u16 seqno;
int ret;
- xe_gt_assert(gt, ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED);
+ xe_gt_assert(gt, xe_guc_ct_initialized(ct));
xe_gt_assert(gt, !g2h_len || !g2h_fence);
xe_gt_assert(gt, !num_g2h || !g2h_fence);
xe_gt_assert(gt, !g2h_len || num_g2h);
@@ -716,7 +835,6 @@ static int guc_ct_send_locked(struct xe_guc_ct *ct, const u32 *action, u32 len,
{
struct xe_device *xe = ct_to_xe(ct);
struct xe_gt *gt = ct_to_gt(ct);
- struct drm_printer p = xe_gt_info_printer(gt);
unsigned int sleep_period_ms = 1;
int ret;
@@ -769,8 +887,13 @@ try_again:
goto broken;
#undef g2h_avail
- if (dequeue_one_g2h(ct) < 0)
+ ret = dequeue_one_g2h(ct);
+ if (ret < 0) {
+ if (ret != -ECANCELED)
+ xe_gt_err(ct_to_gt(ct), "CTB receive failed (%pe)",
+ ERR_PTR(ret));
goto broken;
+ }
goto try_again;
}
@@ -779,8 +902,7 @@ try_again:
broken:
xe_gt_err(gt, "No forward process on H2G, reset required\n");
- xe_guc_ct_print(ct, &p, true);
- ct->ctbs.h2g.info.broken = true;
+ CT_DEAD(ct, &ct->ctbs.h2g, DEADLOCK);
return -EDEADLK;
}
@@ -848,7 +970,7 @@ static bool retry_failure(struct xe_guc_ct *ct, int ret)
#define ct_alive(ct) \
(xe_guc_ct_enabled(ct) && !ct->ctbs.h2g.info.broken && \
!ct->ctbs.g2h.info.broken)
- if (!wait_event_interruptible_timeout(ct->wq, ct_alive(ct), HZ * 5))
+ if (!wait_event_interruptible_timeout(ct->wq, ct_alive(ct), HZ * 5))
return false;
#undef ct_alive
@@ -1046,6 +1168,7 @@ static int parse_g2h_response(struct xe_guc_ct *ct, u32 *msg, u32 len)
else
xe_gt_err(gt, "unexpected response %u for FAST_REQ H2G fence 0x%x!\n",
type, fence);
+ CT_DEAD(ct, NULL, PARSE_G2H_RESPONSE);
return -EPROTO;
}
@@ -1053,6 +1176,7 @@ static int parse_g2h_response(struct xe_guc_ct *ct, u32 *msg, u32 len)
g2h_fence = xa_erase(&ct->fence_lookup, fence);
if (unlikely(!g2h_fence)) {
/* Don't tear down channel, as send could've timed out */
+ /* CT_DEAD(ct, NULL, PARSE_G2H_UNKNOWN); */
xe_gt_warn(gt, "G2H fence (%u) not found!\n", fence);
g2h_release_space(ct, GUC_CTB_HXG_MSG_MAX_LEN);
return 0;
@@ -1097,7 +1221,7 @@ static int parse_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len)
if (unlikely(origin != GUC_HXG_ORIGIN_GUC)) {
xe_gt_err(gt, "G2H channel broken on read, origin=%u, reset required\n",
origin);
- ct->ctbs.g2h.info.broken = true;
+ CT_DEAD(ct, &ct->ctbs.g2h, PARSE_G2H_ORIGIN);
return -EPROTO;
}
@@ -1115,7 +1239,7 @@ static int parse_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len)
default:
xe_gt_err(gt, "G2H channel broken on read, type=%u, reset required\n",
type);
- ct->ctbs.g2h.info.broken = true;
+ CT_DEAD(ct, &ct->ctbs.g2h, PARSE_G2H_TYPE);
ret = -EOPNOTSUPP;
}
@@ -1192,9 +1316,11 @@ static int process_g2h_msg(struct xe_guc_ct *ct, u32 *msg, u32 len)
xe_gt_err(gt, "unexpected G2H action 0x%04x\n", action);
}
- if (ret)
+ if (ret) {
xe_gt_err(gt, "G2H action 0x%04x failed (%pe)\n",
action, ERR_PTR(ret));
+ CT_DEAD(ct, NULL, PROCESS_FAILED);
+ }
return 0;
}
@@ -1204,12 +1330,12 @@ static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
struct xe_device *xe = ct_to_xe(ct);
struct xe_gt *gt = ct_to_gt(ct);
struct guc_ctb *g2h = &ct->ctbs.g2h;
- u32 tail, head, len;
+ u32 tail, head, len, desc_status;
s32 avail;
u32 action;
u32 *hxg;
- xe_gt_assert(gt, ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED);
+ xe_gt_assert(gt, xe_guc_ct_initialized(ct));
lockdep_assert_held(&ct->fast_lock);
if (ct->state == XE_GUC_CT_STATE_DISABLED)
@@ -1223,6 +1349,63 @@ static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
xe_gt_assert(gt, xe_guc_ct_enabled(ct));
+ desc_status = desc_read(xe, g2h, status);
+ if (desc_status) {
+ if (desc_status & GUC_CTB_STATUS_DISABLED) {
+ /*
+ * Potentially valid if a CLIENT_RESET request resulted in
+ * contexts/engines being reset. But should never happen as
+ * no contexts should be active when CLIENT_RESET is sent.
+ */
+ xe_gt_err(gt, "CT read: unexpected G2H after GuC has stopped!\n");
+ desc_status &= ~GUC_CTB_STATUS_DISABLED;
+ }
+
+ if (desc_status) {
+ xe_gt_err(gt, "CT read: non-zero status: %u\n", desc_status);
+ goto corrupted;
+ }
+ }
+
+ if (IS_ENABLED(CONFIG_DRM_XE_DEBUG)) {
+ u32 desc_tail = desc_read(xe, g2h, tail);
+ /*
+ u32 desc_head = desc_read(xe, g2h, head);
+
+ * info.head and desc_head are updated back-to-back at the end of
+ * this function and nowhere else. Hence, they cannot be different
+ * unless two g2h_read calls are running concurrently. Which is not
+ * possible because it is guarded by ct->fast_lock. And yet, some
+ * discrete platforms are reguarly hitting this error :(.
+ *
+ * desc_head rolling backwards shouldn't cause any noticeable
+ * problems - just a delay in GuC being allowed to proceed past that
+ * point in the queue. So for now, just disable the error until it
+ * can be root caused.
+ *
+ if (g2h->info.head != desc_head) {
+ desc_write(xe, g2h, status, desc_status | GUC_CTB_STATUS_MISMATCH);
+ xe_gt_err(gt, "CT read: head was modified %u != %u\n",
+ desc_head, g2h->info.head);
+ goto corrupted;
+ }
+ */
+
+ if (g2h->info.head > g2h->info.size) {
+ desc_write(xe, g2h, status, desc_status | GUC_CTB_STATUS_OVERFLOW);
+ xe_gt_err(gt, "CT read: head out of range: %u vs %u\n",
+ g2h->info.head, g2h->info.size);
+ goto corrupted;
+ }
+
+ if (desc_tail >= g2h->info.size) {
+ desc_write(xe, g2h, status, desc_status | GUC_CTB_STATUS_OVERFLOW);
+ xe_gt_err(gt, "CT read: invalid tail offset %u >= %u)\n",
+ desc_tail, g2h->info.size);
+ goto corrupted;
+ }
+ }
+
/* Calculate DW available to read */
tail = desc_read(xe, g2h, tail);
avail = tail - g2h->info.head;
@@ -1239,9 +1422,7 @@ static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
if (len > avail) {
xe_gt_err(gt, "G2H channel broken on read, avail=%d, len=%d, reset required\n",
avail, len);
- g2h->info.broken = true;
-
- return -EPROTO;
+ goto corrupted;
}
head = (g2h->info.head + 1) % g2h->info.size;
@@ -1287,6 +1468,10 @@ static int g2h_read(struct xe_guc_ct *ct, u32 *msg, bool fast_path)
action, len, g2h->info.head, tail);
return len;
+
+corrupted:
+ CT_DEAD(ct, &ct->ctbs.g2h, G2H_READ);
+ return -EPROTO;
}
static void g2h_fast_path(struct xe_guc_ct *ct, u32 *msg, u32 len)
@@ -1313,9 +1498,11 @@ static void g2h_fast_path(struct xe_guc_ct *ct, u32 *msg, u32 len)
xe_gt_warn(gt, "NOT_POSSIBLE");
}
- if (ret)
+ if (ret) {
xe_gt_err(gt, "G2H action 0x%04x failed (%pe)\n",
action, ERR_PTR(ret));
+ CT_DEAD(ct, NULL, FAST_G2H);
+ }
}
/**
@@ -1375,7 +1562,6 @@ static int dequeue_one_g2h(struct xe_guc_ct *ct)
static void receive_g2h(struct xe_guc_ct *ct)
{
- struct xe_gt *gt = ct_to_gt(ct);
bool ongoing;
int ret;
@@ -1412,9 +1598,8 @@ static void receive_g2h(struct xe_guc_ct *ct)
mutex_unlock(&ct->lock);
if (unlikely(ret == -EPROTO || ret == -EOPNOTSUPP)) {
- struct drm_printer p = xe_gt_info_printer(gt);
-
- xe_guc_ct_print(ct, &p, false);
+ xe_gt_err(ct_to_gt(ct), "CT dequeue failed: %d", ret);
+ CT_DEAD(ct, NULL, G2H_RECV);
kick_reset(ct);
}
} while (ret == 1);
@@ -1442,9 +1627,8 @@ static void guc_ctb_snapshot_capture(struct xe_device *xe, struct guc_ctb *ctb,
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");
+ drm_err(&xe->drm, "Skipping CTB commands snapshot. Only CT info will be available.\n");
return;
}
@@ -1525,7 +1709,7 @@ struct xe_guc_ct_snapshot *xe_guc_ct_snapshot_capture(struct xe_guc_ct *ct,
atomic ? GFP_ATOMIC : GFP_KERNEL);
if (!snapshot) {
- drm_err(&xe->drm, "Skipping CTB snapshot entirely.\n");
+ xe_gt_err(ct_to_gt(ct), "Skipping CTB snapshot entirely.\n");
return NULL;
}
@@ -1589,16 +1773,119 @@ void xe_guc_ct_snapshot_free(struct xe_guc_ct_snapshot *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)
+void xe_guc_ct_print(struct xe_guc_ct *ct, struct drm_printer *p)
{
struct xe_guc_ct_snapshot *snapshot;
- snapshot = xe_guc_ct_snapshot_capture(ct, atomic);
+ snapshot = xe_guc_ct_snapshot_capture(ct, false);
xe_guc_ct_snapshot_print(snapshot, p);
xe_guc_ct_snapshot_free(snapshot);
}
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+static void ct_dead_capture(struct xe_guc_ct *ct, struct guc_ctb *ctb, u32 reason_code)
+{
+ struct xe_guc_log_snapshot *snapshot_log;
+ struct xe_guc_ct_snapshot *snapshot_ct;
+ struct xe_guc *guc = ct_to_guc(ct);
+ unsigned long flags;
+ bool have_capture;
+
+ if (ctb)
+ ctb->info.broken = true;
+
+ /* Ignore further errors after the first dump until a reset */
+ if (ct->dead.reported)
+ return;
+
+ spin_lock_irqsave(&ct->dead.lock, flags);
+
+ /* And only capture one dump at a time */
+ have_capture = ct->dead.reason & (1 << CT_DEAD_STATE_CAPTURE);
+ ct->dead.reason |= (1 << reason_code) |
+ (1 << CT_DEAD_STATE_CAPTURE);
+
+ spin_unlock_irqrestore(&ct->dead.lock, flags);
+
+ if (have_capture)
+ return;
+
+ snapshot_log = xe_guc_log_snapshot_capture(&guc->log, true);
+ snapshot_ct = xe_guc_ct_snapshot_capture((ct), true);
+
+ spin_lock_irqsave(&ct->dead.lock, flags);
+
+ if (ct->dead.snapshot_log || ct->dead.snapshot_ct) {
+ xe_gt_err(ct_to_gt(ct), "Got unexpected dead CT capture!\n");
+ xe_guc_log_snapshot_free(snapshot_log);
+ xe_guc_ct_snapshot_free(snapshot_ct);
+ } else {
+ ct->dead.snapshot_log = snapshot_log;
+ ct->dead.snapshot_ct = snapshot_ct;
+ }
+
+ spin_unlock_irqrestore(&ct->dead.lock, flags);
+
+ queue_work(system_unbound_wq, &(ct)->dead.worker);
+}
+
+static void ct_dead_print(struct xe_dead_ct *dead)
+{
+ struct xe_guc_ct *ct = container_of(dead, struct xe_guc_ct, dead);
+ struct xe_device *xe = ct_to_xe(ct);
+ struct xe_gt *gt = ct_to_gt(ct);
+ static int g_count;
+ struct drm_printer ip = xe_gt_info_printer(gt);
+ struct drm_printer lp = drm_line_printer(&ip, "Capture", ++g_count);
+
+ if (!dead->reason) {
+ xe_gt_err(gt, "CTB is dead for no reason!?\n");
+ return;
+ }
+
+ drm_printf(&lp, "CTB is dead - reason=0x%X\n", dead->reason);
+
+ /* Can't generate a genuine core dump at this point, so just do the good bits */
+ drm_puts(&lp, "**** Xe Device Coredump ****\n");
+ xe_device_snapshot_print(xe, &lp);
+
+ drm_printf(&lp, "**** GT #%d ****\n", gt->info.id);
+ drm_printf(&lp, "\tTile: %d\n", gt->tile->id);
+
+ drm_puts(&lp, "**** GuC Log ****\n");
+ xe_guc_log_snapshot_print(dead->snapshot_log, &lp);
+
+ drm_puts(&lp, "**** GuC CT ****\n");
+ xe_guc_ct_snapshot_print(dead->snapshot_ct, &lp);
+
+ drm_puts(&lp, "Done.\n");
+}
+
+static void ct_dead_worker_func(struct work_struct *w)
+{
+ struct xe_guc_ct *ct = container_of(w, struct xe_guc_ct, dead.worker);
+
+ if (!ct->dead.reported) {
+ ct->dead.reported = true;
+ ct_dead_print(&ct->dead);
+ }
+
+ spin_lock_irq(&ct->dead.lock);
+
+ xe_guc_log_snapshot_free(ct->dead.snapshot_log);
+ ct->dead.snapshot_log = NULL;
+ xe_guc_ct_snapshot_free(ct->dead.snapshot_ct);
+ ct->dead.snapshot_ct = NULL;
+
+ if (ct->dead.reason & (1 << CT_DEAD_STATE_REARM)) {
+ /* A reset has occurred so re-arm the error reporting */
+ ct->dead.reason = 0;
+ ct->dead.reported = false;
+ }
+
+ spin_unlock_irq(&ct->dead.lock);
+}
+#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_ct.h b/drivers/gpu/drm/xe/xe_guc_ct.h
index 190202fce2d0..c7ac9407b861 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct.h
+++ b/drivers/gpu/drm/xe/xe_guc_ct.h
@@ -21,7 +21,12 @@ 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);
+void xe_guc_ct_print(struct xe_guc_ct *ct, struct drm_printer *p);
+
+static inline bool xe_guc_ct_initialized(struct xe_guc_ct *ct)
+{
+ return ct->state != XE_GUC_CT_STATE_NOT_INITIALIZED;
+}
static inline bool xe_guc_ct_enabled(struct xe_guc_ct *ct)
{
diff --git a/drivers/gpu/drm/xe/xe_guc_ct_types.h b/drivers/gpu/drm/xe/xe_guc_ct_types.h
index 761cb9031298..85e127ec91d7 100644
--- a/drivers/gpu/drm/xe/xe_guc_ct_types.h
+++ b/drivers/gpu/drm/xe/xe_guc_ct_types.h
@@ -86,6 +86,24 @@ enum xe_guc_ct_state {
XE_GUC_CT_STATE_ENABLED,
};
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+/** struct xe_dead_ct - Information for debugging a dead CT */
+struct xe_dead_ct {
+ /** @lock: protects memory allocation/free operations, and @reason updates */
+ spinlock_t lock;
+ /** @reason: bit mask of CT_DEAD_* reason codes */
+ unsigned int reason;
+ /** @reported: for preventing multiple dumps per error sequence */
+ bool reported;
+ /** @worker: worker thread to get out of interrupt context before dumping */
+ struct work_struct worker;
+ /** snapshot_ct: copy of CT state and CTB content at point of error */
+ struct xe_guc_ct_snapshot *snapshot_ct;
+ /** snapshot_log: copy of GuC log at point of error */
+ struct xe_guc_log_snapshot *snapshot_log;
+};
+#endif
+
/**
* struct xe_guc_ct - GuC command transport (CT) layer
*
@@ -128,6 +146,11 @@ struct xe_guc_ct {
u32 msg[GUC_CTB_MSG_MAX_LEN];
/** @fast_msg: Message buffer */
u32 fast_msg[GUC_CTB_MSG_MAX_LEN];
+
+#if IS_ENABLED(CONFIG_DRM_XE_DEBUG)
+ /** @dead: information for debugging dead CTs */
+ struct xe_dead_ct dead;
+#endif
};
#endif
diff --git a/drivers/gpu/drm/xe/xe_guc_pc.c b/drivers/gpu/drm/xe/xe_guc_pc.c
index 034b29984d5e..af02803c145b 100644
--- a/drivers/gpu/drm/xe/xe_guc_pc.c
+++ b/drivers/gpu/drm/xe/xe_guc_pc.c
@@ -6,6 +6,9 @@
#include "xe_guc_pc.h"
#include <linux/delay.h>
+#include <linux/jiffies.h>
+#include <linux/ktime.h>
+#include <linux/wait_bit.h>
#include <drm/drm_managed.h>
#include <generated/xe_wa_oob.h>
@@ -47,6 +50,12 @@
#define LNL_MERT_FREQ_CAP 800
#define BMG_MERT_FREQ_CAP 2133
+#define BMG_MIN_FREQ 1200
+#define BMG_MERT_FLUSH_FREQ_CAP 2600
+
+#define SLPC_RESET_TIMEOUT_MS 5 /* roughly 5ms, but no need for precision */
+#define SLPC_RESET_EXTENDED_TIMEOUT_MS 1000 /* To be used only at pc_start */
+#define SLPC_ACT_FREQ_TIMEOUT_MS 100
/**
* DOC: GuC Power Conservation (PC)
@@ -133,6 +142,36 @@ static int wait_for_pc_state(struct xe_guc_pc *pc,
return -ETIMEDOUT;
}
+static int wait_for_flush_complete(struct xe_guc_pc *pc)
+{
+ const unsigned long timeout = msecs_to_jiffies(30);
+
+ if (!wait_var_event_timeout(&pc->flush_freq_limit,
+ !atomic_read(&pc->flush_freq_limit),
+ timeout))
+ return -ETIMEDOUT;
+
+ return 0;
+}
+
+static int wait_for_act_freq_limit(struct xe_guc_pc *pc, u32 freq)
+{
+ int timeout_us = SLPC_ACT_FREQ_TIMEOUT_MS * USEC_PER_MSEC;
+ int slept, wait = 10;
+
+ for (slept = 0; slept < timeout_us;) {
+ if (xe_guc_pc_get_act_freq(pc) <= freq)
+ 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_ct(pc);
@@ -584,6 +623,11 @@ int xe_guc_pc_set_max_freq(struct xe_guc_pc *pc, u32 freq)
{
int ret;
+ if (XE_WA(pc_to_gt(pc), 22019338487)) {
+ if (wait_for_flush_complete(pc) != 0)
+ return -EAGAIN;
+ }
+
mutex_lock(&pc->freq_lock);
if (!pc->freq_ready) {
/* Might be in the middle of a gt reset */
@@ -793,6 +837,106 @@ static int pc_adjust_requested_freq(struct xe_guc_pc *pc)
return ret;
}
+static bool needs_flush_freq_limit(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+
+ return XE_WA(gt, 22019338487) &&
+ pc->rp0_freq > BMG_MERT_FLUSH_FREQ_CAP;
+}
+
+/**
+ * xe_guc_pc_apply_flush_freq_limit() - Limit max GT freq during L2 flush
+ * @pc: the xe_guc_pc object
+ *
+ * As per the WA, reduce max GT frequency during L2 cache flush
+ */
+void xe_guc_pc_apply_flush_freq_limit(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ u32 max_freq;
+ int ret;
+
+ if (!needs_flush_freq_limit(pc))
+ return;
+
+ mutex_lock(&pc->freq_lock);
+
+ if (!pc->freq_ready) {
+ mutex_unlock(&pc->freq_lock);
+ return;
+ }
+
+ ret = pc_action_query_task_state(pc);
+ if (ret) {
+ mutex_unlock(&pc->freq_lock);
+ return;
+ }
+
+ max_freq = pc_get_max_freq(pc);
+ if (max_freq > BMG_MERT_FLUSH_FREQ_CAP) {
+ ret = pc_set_max_freq(pc, BMG_MERT_FLUSH_FREQ_CAP);
+ if (ret) {
+ xe_gt_err_once(gt, "Failed to cap max freq on flush to %u, %pe\n",
+ BMG_MERT_FLUSH_FREQ_CAP, ERR_PTR(ret));
+ mutex_unlock(&pc->freq_lock);
+ return;
+ }
+
+ atomic_set(&pc->flush_freq_limit, 1);
+
+ /*
+ * If user has previously changed max freq, stash that value to
+ * restore later, otherwise use the current max. New user
+ * requests wait on flush.
+ */
+ if (pc->user_requested_max != 0)
+ pc->stashed_max_freq = pc->user_requested_max;
+ else
+ pc->stashed_max_freq = max_freq;
+ }
+
+ mutex_unlock(&pc->freq_lock);
+
+ /*
+ * Wait for actual freq to go below the flush cap: even if the previous
+ * max was below cap, the current one might still be above it
+ */
+ ret = wait_for_act_freq_limit(pc, BMG_MERT_FLUSH_FREQ_CAP);
+ if (ret)
+ xe_gt_err_once(gt, "Actual freq did not reduce to %u, %pe\n",
+ BMG_MERT_FLUSH_FREQ_CAP, ERR_PTR(ret));
+}
+
+/**
+ * xe_guc_pc_remove_flush_freq_limit() - Remove max GT freq limit after L2 flush completes.
+ * @pc: the xe_guc_pc object
+ *
+ * Retrieve the previous GT max frequency value.
+ */
+void xe_guc_pc_remove_flush_freq_limit(struct xe_guc_pc *pc)
+{
+ struct xe_gt *gt = pc_to_gt(pc);
+ int ret = 0;
+
+ if (!needs_flush_freq_limit(pc))
+ return;
+
+ if (!atomic_read(&pc->flush_freq_limit))
+ return;
+
+ mutex_lock(&pc->freq_lock);
+
+ ret = pc_set_max_freq(&gt->uc.guc.pc, pc->stashed_max_freq);
+ if (ret)
+ xe_gt_err_once(gt, "Failed to restore max freq %u:%d",
+ pc->stashed_max_freq, ret);
+
+ atomic_set(&pc->flush_freq_limit, 0);
+ mutex_unlock(&pc->freq_lock);
+ wake_up_var(&pc->flush_freq_limit);
+}
+
static int pc_set_mert_freq_cap(struct xe_guc_pc *pc)
{
int ret = 0;
@@ -975,7 +1119,7 @@ int xe_guc_pc_start(struct xe_guc_pc *pc)
goto out;
}
- memset(pc->bo->vmap.vaddr, 0, size);
+ xe_map_memset(xe, &pc->bo->vmap, 0, 0, size);
slpc_shared_data_write(pc, header.size, size);
ret = pc_action_reset(pc);
diff --git a/drivers/gpu/drm/xe/xe_guc_pc.h b/drivers/gpu/drm/xe/xe_guc_pc.h
index efda432fadfc..7154b3aab0d8 100644
--- a/drivers/gpu/drm/xe/xe_guc_pc.h
+++ b/drivers/gpu/drm/xe/xe_guc_pc.h
@@ -34,5 +34,7 @@ u64 xe_guc_pc_mc6_residency(struct xe_guc_pc *pc);
void xe_guc_pc_init_early(struct xe_guc_pc *pc);
int xe_guc_pc_restore_stashed_freq(struct xe_guc_pc *pc);
void xe_guc_pc_raise_unslice(struct xe_guc_pc *pc);
+void xe_guc_pc_apply_flush_freq_limit(struct xe_guc_pc *pc);
+void xe_guc_pc_remove_flush_freq_limit(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
index 13810be015db..5b86d91296cb 100644
--- a/drivers/gpu/drm/xe/xe_guc_pc_types.h
+++ b/drivers/gpu/drm/xe/xe_guc_pc_types.h
@@ -15,6 +15,8 @@
struct xe_guc_pc {
/** @bo: GGTT buffer object that is shared with GuC PC */
struct xe_bo *bo;
+ /** @flush_freq_limit: 1 when max freq changes are limited by driver */
+ atomic_t flush_freq_limit;
/** @rp0_freq: HW RP0 frequency - The Maximum one */
u32 rp0_freq;
/** @rpe_freq: HW RPe frequency - The Efficient one */
diff --git a/drivers/gpu/drm/xe/xe_guc_relay.c b/drivers/gpu/drm/xe/xe_guc_relay.c
index ade6162dc259..0e5b43e1518e 100644
--- a/drivers/gpu/drm/xe/xe_guc_relay.c
+++ b/drivers/gpu/drm/xe/xe_guc_relay.c
@@ -224,7 +224,7 @@ __relay_get_transaction(struct xe_guc_relay *relay, bool incoming, u32 remote, u
* with CTB lock held which is marked as used in the reclaim path.
* Btw, that's one of the reason why we use mempool here!
*/
- txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_KERNEL);
+ txn = mempool_alloc(&relay->pool, incoming ? GFP_ATOMIC : GFP_NOWAIT);
if (!txn)
return ERR_PTR(-ENOMEM);
diff --git a/drivers/gpu/drm/xe/xe_guc_submit.c b/drivers/gpu/drm/xe/xe_guc_submit.c
index 20d05efdd406..0e17820a35e2 100644
--- a/drivers/gpu/drm/xe/xe_guc_submit.c
+++ b/drivers/gpu/drm/xe/xe_guc_submit.c
@@ -227,6 +227,17 @@ static bool exec_queue_killed_or_banned_or_wedged(struct xe_exec_queue *q)
static void guc_submit_fini(struct drm_device *drm, void *arg)
{
struct xe_guc *guc = arg;
+ struct xe_device *xe = guc_to_xe(guc);
+ struct xe_gt *gt = guc_to_gt(guc);
+ int ret;
+
+ ret = wait_event_timeout(guc->submission_state.fini_wq,
+ xa_empty(&guc->submission_state.exec_queue_lookup),
+ HZ * 5);
+
+ drain_workqueue(xe->destroy_wq);
+
+ xe_gt_assert(gt, ret);
xa_destroy(&guc->submission_state.exec_queue_lookup);
}
@@ -298,6 +309,8 @@ int xe_guc_submit_init(struct xe_guc *guc, unsigned int num_ids)
primelockdep(guc);
+ guc->submission_state.initialized = true;
+
return drmm_add_action_or_reset(&xe->drm, guc_submit_fini, guc);
}
@@ -826,6 +839,13 @@ void xe_guc_submit_wedge(struct xe_guc *guc)
xe_gt_assert(guc_to_gt(guc), guc_to_xe(guc)->wedged.mode);
+ /*
+ * If device is being wedged even before submission_state is
+ * initialized, there's nothing to do here.
+ */
+ if (!guc->submission_state.initialized)
+ return;
+
err = devm_add_action_or_reset(guc_to_xe(guc)->drm.dev,
guc_submit_wedged_fini, guc);
if (err) {
@@ -1702,6 +1722,9 @@ int xe_guc_submit_reset_prepare(struct xe_guc *guc)
{
int ret;
+ if (!guc->submission_state.initialized)
+ return 0;
+
/*
* Using an atomic here rather than submission_state.lock as this
* function can be called while holding the CT lock (engine reset
diff --git a/drivers/gpu/drm/xe/xe_guc_types.h b/drivers/gpu/drm/xe/xe_guc_types.h
index ed150fc09ad0..7842b71e68be 100644
--- a/drivers/gpu/drm/xe/xe_guc_types.h
+++ b/drivers/gpu/drm/xe/xe_guc_types.h
@@ -74,6 +74,11 @@ struct xe_guc {
struct mutex lock;
/** @submission_state.enabled: submission is enabled */
bool enabled;
+ /**
+ * @submission_state.initialized: mark when submission state is
+ * even initialized - before that not even the lock is valid
+ */
+ bool initialized;
/** @submission_state.fini_wq: submit fini wait queue */
wait_queue_head_t fini_wq;
} submission_state;
diff --git a/drivers/gpu/drm/xe/xe_hw_engine.c b/drivers/gpu/drm/xe/xe_hw_engine.c
index b11bc0f00dfd..0248597c6269 100644
--- a/drivers/gpu/drm/xe/xe_hw_engine.c
+++ b/drivers/gpu/drm/xe/xe_hw_engine.c
@@ -381,12 +381,6 @@ xe_hw_engine_setup_default_lrc_state(struct xe_hw_engine *hwe)
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))
- },
/* Disable WMTP if HW doesn't support it */
{ XE_RTP_NAME("DISABLE_WMTP_ON_UNSUPPORTED_HW"),
XE_RTP_RULES(FUNC(xe_rtp_cfeg_wmtp_disabled)),
@@ -454,6 +448,12 @@ hw_engine_setup_default_state(struct xe_hw_engine *hwe)
XE_RTP_ACTIONS(SET(CSFE_CHICKEN1(0), CS_PRIORITY_MEM_READ,
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))
+ },
{}
};
diff --git a/drivers/gpu/drm/xe/xe_irq.c b/drivers/gpu/drm/xe/xe_irq.c
index 5f2c368c35ad..14c3a476597a 100644
--- a/drivers/gpu/drm/xe/xe_irq.c
+++ b/drivers/gpu/drm/xe/xe_irq.c
@@ -173,7 +173,7 @@ void xe_irq_enable_hwe(struct xe_gt *gt)
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);
+ xe_mmio_write32(gt, CCS2_CCS3_INTR_MASK, ~dmask);
}
if (xe_gt_is_media_type(gt) || MEDIA_VER(xe) < 13) {
@@ -504,7 +504,7 @@ static void gt_irq_reset(struct xe_tile *tile)
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);
+ 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)) ||
diff --git a/drivers/gpu/drm/xe/xe_lmtt.c b/drivers/gpu/drm/xe/xe_lmtt.c
index 8999ac511555..485658f69fba 100644
--- a/drivers/gpu/drm/xe/xe_lmtt.c
+++ b/drivers/gpu/drm/xe/xe_lmtt.c
@@ -78,6 +78,9 @@ static struct xe_lmtt_pt *lmtt_pt_alloc(struct xe_lmtt *lmtt, unsigned int level
}
lmtt_assert(lmtt, xe_bo_is_vram(bo));
+ lmtt_debug(lmtt, "level=%u addr=%#llx\n", level, (u64)xe_bo_main_addr(bo, XE_PAGE_SIZE));
+
+ xe_map_memset(lmtt_to_xe(lmtt), &bo->vmap, 0, 0, bo->size);
pt->level = level;
pt->bo = bo;
@@ -91,6 +94,9 @@ out:
static void lmtt_pt_free(struct xe_lmtt_pt *pt)
{
+ lmtt_debug(&pt->bo->tile->sriov.pf.lmtt, "level=%u addr=%llx\n",
+ pt->level, (u64)xe_bo_main_addr(pt->bo, XE_PAGE_SIZE));
+
xe_bo_unpin_map_no_vm(pt->bo);
kfree(pt);
}
@@ -226,9 +232,14 @@ static void lmtt_write_pte(struct xe_lmtt *lmtt, struct xe_lmtt_pt *pt,
switch (lmtt->ops->lmtt_pte_size(level)) {
case sizeof(u32):
+ lmtt_assert(lmtt, !overflows_type(pte, u32));
+ lmtt_assert(lmtt, !pte || !iosys_map_rd(&pt->bo->vmap, idx * sizeof(u32), u32));
+
xe_map_wr(lmtt_to_xe(lmtt), &pt->bo->vmap, idx * sizeof(u32), u32, pte);
break;
case sizeof(u64):
+ lmtt_assert(lmtt, !pte || !iosys_map_rd(&pt->bo->vmap, idx * sizeof(u64), u64));
+
xe_map_wr(lmtt_to_xe(lmtt), &pt->bo->vmap, idx * sizeof(u64), u64, pte);
break;
default:
diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
index aec7db39c061..ce6d2167b94a 100644
--- a/drivers/gpu/drm/xe/xe_lrc.c
+++ b/drivers/gpu/drm/xe/xe_lrc.c
@@ -694,7 +694,7 @@ static inline u32 __xe_lrc_start_seqno_offset(struct xe_lrc *lrc)
static u32 __xe_lrc_ctx_job_timestamp_offset(struct xe_lrc *lrc)
{
- /* The start seqno is stored in the driver-defined portion of PPHWSP */
+ /* This is stored in the driver-defined portion of PPHWSP */
return xe_lrc_pphwsp_offset(lrc) + LRC_CTX_JOB_TIMESTAMP_OFFSET;
}
@@ -874,7 +874,7 @@ static void *empty_lrc_data(struct xe_hw_engine *hwe)
static void xe_lrc_set_ppgtt(struct xe_lrc *lrc, struct xe_vm *vm)
{
- u64 desc = xe_vm_pdp4_descriptor(vm, lrc->tile);
+ u64 desc = xe_vm_pdp4_descriptor(vm, gt_to_tile(lrc->gt));
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));
@@ -905,6 +905,7 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
int err;
kref_init(&lrc->refcount);
+ lrc->gt = gt;
lrc->flags = 0;
lrc_size = ring_size + xe_gt_lrc_size(gt, hwe->class);
if (xe_gt_has_indirect_ring_state(gt))
@@ -923,7 +924,6 @@ static int xe_lrc_init(struct xe_lrc *lrc, struct xe_hw_engine *hwe,
return PTR_ERR(lrc->bo);
lrc->size = lrc_size;
- lrc->tile = gt_to_tile(hwe->gt);
lrc->ring.size = ring_size;
lrc->ring.tail = 0;
lrc->ctx_timestamp = 0;
diff --git a/drivers/gpu/drm/xe/xe_lrc_types.h b/drivers/gpu/drm/xe/xe_lrc_types.h
index 71ecb453f811..cd38586ae989 100644
--- a/drivers/gpu/drm/xe/xe_lrc_types.h
+++ b/drivers/gpu/drm/xe/xe_lrc_types.h
@@ -25,8 +25,8 @@ struct xe_lrc {
/** @size: size of lrc including any indirect ring state page */
u32 size;
- /** @tile: tile which this LRC belongs to */
- struct xe_tile *tile;
+ /** @gt: gt which this LRC belongs to */
+ struct xe_gt *gt;
/** @flags: LRC flags */
#define XE_LRC_FLAG_INDIRECT_RING_STATE 0x1
diff --git a/drivers/gpu/drm/xe/xe_migrate.c b/drivers/gpu/drm/xe/xe_migrate.c
index 6431697c6169..c2da2691fd2b 100644
--- a/drivers/gpu/drm/xe/xe_migrate.c
+++ b/drivers/gpu/drm/xe/xe_migrate.c
@@ -860,7 +860,7 @@ struct dma_fence *xe_migrate_copy(struct xe_migrate *m,
if (src_is_vram && xe_migrate_allow_identity(src_L0, &src_it))
xe_res_next(&src_it, src_L0);
else
- emit_pte(m, bb, src_L0_pt, src_is_vram, copy_system_ccs,
+ emit_pte(m, bb, src_L0_pt, src_is_vram, copy_system_ccs || use_comp_pat,
&src_it, src_L0, src);
if (dst_is_vram && xe_migrate_allow_identity(src_L0, &dst_it))
diff --git a/drivers/gpu/drm/xe/xe_oa.c b/drivers/gpu/drm/xe/xe_oa.c
index 448766033690..d306ed0a0443 100644
--- a/drivers/gpu/drm/xe/xe_oa.c
+++ b/drivers/gpu/drm/xe/xe_oa.c
@@ -535,6 +535,7 @@ static ssize_t xe_oa_read(struct file *file, char __user *buf,
mutex_unlock(&stream->stream_lock);
} while (!offset && !ret);
} else {
+ xe_oa_buffer_check_unlocked(stream);
mutex_lock(&stream->stream_lock);
ret = __xe_oa_read(stream, buf, count, &offset);
mutex_unlock(&stream->stream_lock);
diff --git a/drivers/gpu/drm/xe/xe_pci.c b/drivers/gpu/drm/xe/xe_pci.c
index 025d64943467..da09c26249f5 100644
--- a/drivers/gpu/drm/xe/xe_pci.c
+++ b/drivers/gpu/drm/xe/xe_pci.c
@@ -164,7 +164,6 @@ static const struct xe_graphics_desc graphics_xelpg = {
.has_asid = 1, \
.has_atomic_enable_pte_bit = 1, \
.has_flat_ccs = 1, \
- .has_indirect_ring_state = 1, \
.has_range_tlb_invalidation = 1, \
.has_usm = 1, \
.va_bits = 48, \
@@ -910,6 +909,7 @@ static int xe_pci_suspend(struct device *dev)
pci_save_state(pdev);
pci_disable_device(pdev);
+ pci_set_power_state(pdev, PCI_D3cold);
return 0;
}
diff --git a/drivers/gpu/drm/xe/xe_pci_sriov.c b/drivers/gpu/drm/xe/xe_pci_sriov.c
index aaceee748287..09ee8a06fe2e 100644
--- a/drivers/gpu/drm/xe/xe_pci_sriov.c
+++ b/drivers/gpu/drm/xe/xe_pci_sriov.c
@@ -62,6 +62,55 @@ static void pf_reset_vfs(struct xe_device *xe, unsigned int num_vfs)
xe_gt_sriov_pf_control_trigger_flr(gt, n);
}
+static struct pci_dev *xe_pci_pf_get_vf_dev(struct xe_device *xe, unsigned int vf_id)
+{
+ struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
+
+ xe_assert(xe, IS_SRIOV_PF(xe));
+
+ /* caller must use pci_dev_put() */
+ return pci_get_domain_bus_and_slot(pci_domain_nr(pdev->bus),
+ pdev->bus->number,
+ pci_iov_virtfn_devfn(pdev, vf_id));
+}
+
+static void pf_link_vfs(struct xe_device *xe, int num_vfs)
+{
+ struct pci_dev *pdev_pf = to_pci_dev(xe->drm.dev);
+ struct device_link *link;
+ struct pci_dev *pdev_vf;
+ unsigned int n;
+
+ /*
+ * When both PF and VF devices are enabled on the host, during system
+ * resume they are resuming in parallel.
+ *
+ * But PF has to complete the provision of VF first to allow any VFs to
+ * successfully resume.
+ *
+ * Create a parent-child device link between PF and VF devices that will
+ * enforce correct resume order.
+ */
+ for (n = 1; n <= num_vfs; n++) {
+ pdev_vf = xe_pci_pf_get_vf_dev(xe, n - 1);
+
+ /* unlikely, something weird is happening, abort */
+ if (!pdev_vf) {
+ xe_sriov_err(xe, "Cannot find VF%u device, aborting link%s creation!\n",
+ n, str_plural(num_vfs));
+ break;
+ }
+
+ link = device_link_add(&pdev_vf->dev, &pdev_pf->dev,
+ DL_FLAG_AUTOREMOVE_CONSUMER);
+ /* unlikely and harmless, continue with other VFs */
+ if (!link)
+ xe_sriov_notice(xe, "Failed linking VF%u\n", n);
+
+ pci_dev_put(pdev_vf);
+ }
+}
+
static int pf_enable_vfs(struct xe_device *xe, int num_vfs)
{
struct pci_dev *pdev = to_pci_dev(xe->drm.dev);
@@ -92,6 +141,8 @@ static int pf_enable_vfs(struct xe_device *xe, int num_vfs)
if (err < 0)
goto failed;
+ pf_link_vfs(xe, num_vfs);
+
xe_sriov_info(xe, "Enabled %u of %u VF%s\n",
num_vfs, total_vfs, str_plural(total_vfs));
return num_vfs;
diff --git a/drivers/gpu/drm/xe/xe_pm.c b/drivers/gpu/drm/xe/xe_pm.c
index 06f50aa31326..46c73ff10c74 100644
--- a/drivers/gpu/drm/xe/xe_pm.c
+++ b/drivers/gpu/drm/xe/xe_pm.c
@@ -682,11 +682,13 @@ void xe_pm_assert_unbounded_bridge(struct xe_device *xe)
}
/**
- * xe_pm_set_vram_threshold - Set a vram threshold for allowing/blocking D3Cold
+ * xe_pm_set_vram_threshold - Set a VRAM threshold for allowing/blocking D3Cold
* @xe: xe device instance
- * @threshold: VRAM size in bites for the D3cold threshold
+ * @threshold: VRAM size in MiB for the D3cold threshold
*
- * Returns 0 for success, negative error code otherwise.
+ * Return:
+ * * 0 - success
+ * * -EINVAL - invalid argument
*/
int xe_pm_set_vram_threshold(struct xe_device *xe, u32 threshold)
{
diff --git a/drivers/gpu/drm/xe/xe_pt.c b/drivers/gpu/drm/xe/xe_pt.c
index 230cf47fb9c5..fb94ff55c736 100644
--- a/drivers/gpu/drm/xe/xe_pt.c
+++ b/drivers/gpu/drm/xe/xe_pt.c
@@ -218,6 +218,20 @@ void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred)
}
/**
+ * xe_pt_clear() - Clear a page-table.
+ * @xe: xe device.
+ * @pt: The page-table.
+ *
+ * Clears page-table by setting to zero.
+ */
+void xe_pt_clear(struct xe_device *xe, struct xe_pt *pt)
+{
+ struct iosys_map *map = &pt->bo->vmap;
+
+ xe_map_memset(xe, map, 0, 0, SZ_4K);
+}
+
+/**
* DOC: Pagetable building
*
* Below we use the term "page-table" for both page-directories, containing
diff --git a/drivers/gpu/drm/xe/xe_pt.h b/drivers/gpu/drm/xe/xe_pt.h
index 9ab386431cad..8e43912ae8e9 100644
--- a/drivers/gpu/drm/xe/xe_pt.h
+++ b/drivers/gpu/drm/xe/xe_pt.h
@@ -13,6 +13,7 @@ struct dma_fence;
struct xe_bo;
struct xe_device;
struct xe_exec_queue;
+struct xe_svm_range;
struct xe_sync_entry;
struct xe_tile;
struct xe_vm;
@@ -35,6 +36,8 @@ void xe_pt_populate_empty(struct xe_tile *tile, struct xe_vm *vm,
void xe_pt_destroy(struct xe_pt *pt, u32 flags, struct llist_head *deferred);
+void xe_pt_clear(struct xe_device *xe, struct xe_pt *pt);
+
int xe_pt_update_ops_prepare(struct xe_tile *tile, struct xe_vma_ops *vops);
struct dma_fence *xe_pt_update_ops_run(struct xe_tile *tile,
struct xe_vma_ops *vops);
diff --git a/drivers/gpu/drm/xe/xe_ring_ops.c b/drivers/gpu/drm/xe/xe_ring_ops.c
index 9f327f27c072..fb31e09acb51 100644
--- a/drivers/gpu/drm/xe/xe_ring_ops.c
+++ b/drivers/gpu/drm/xe/xe_ring_ops.c
@@ -229,13 +229,10 @@ static u32 get_ppgtt_flag(struct xe_sched_job *job)
static int emit_copy_timestamp(struct xe_lrc *lrc, u32 *dw, int i)
{
- dw[i++] = MI_COPY_MEM_MEM | MI_COPY_MEM_MEM_SRC_GGTT |
- MI_COPY_MEM_MEM_DST_GGTT;
+ dw[i++] = MI_STORE_REGISTER_MEM | MI_SRM_USE_GGTT | MI_SRM_ADD_CS_OFFSET;
+ dw[i++] = RING_CTX_TIMESTAMP(0).addr;
dw[i++] = xe_lrc_ctx_job_timestamp_ggtt_addr(lrc);
dw[i++] = 0;
- dw[i++] = xe_lrc_ctx_timestamp_ggtt_addr(lrc);
- dw[i++] = 0;
- dw[i++] = MI_NOOP;
return i;
}
diff --git a/drivers/gpu/drm/xe/xe_sa.c b/drivers/gpu/drm/xe/xe_sa.c
index fe2cb2a96f78..1d4254759006 100644
--- a/drivers/gpu/drm/xe/xe_sa.c
+++ b/drivers/gpu/drm/xe/xe_sa.c
@@ -57,8 +57,6 @@ struct xe_sa_manager *xe_sa_bo_manager_init(struct xe_tile *tile, u32 size, u32
}
sa_manager->bo = bo;
sa_manager->is_iomem = bo->vmap.is_iomem;
-
- drm_suballoc_manager_init(&sa_manager->base, managed_size, align);
sa_manager->gpu_addr = xe_bo_ggtt_addr(bo);
if (bo->vmap.is_iomem) {
@@ -72,6 +70,7 @@ struct xe_sa_manager *xe_sa_bo_manager_init(struct xe_tile *tile, u32 size, u32
memset(sa_manager->cpu_ptr, 0, bo->ttm.base.size);
}
+ drm_suballoc_manager_init(&sa_manager->base, managed_size, align);
ret = drmm_add_action_or_reset(&xe->drm, xe_sa_bo_manager_fini,
sa_manager);
if (ret)
diff --git a/drivers/gpu/drm/xe/xe_tile.c b/drivers/gpu/drm/xe/xe_tile.c
index dda5268507d8..36c87d7c72fb 100644
--- a/drivers/gpu/drm/xe/xe_tile.c
+++ b/drivers/gpu/drm/xe/xe_tile.c
@@ -167,17 +167,19 @@ int xe_tile_init_noalloc(struct xe_tile *tile)
if (err)
return err;
+ xe_wa_apply_tile_workarounds(tile);
+
+ return xe_tile_sysfs_init(tile);
+}
+
+int xe_tile_init(struct xe_tile *tile)
+{
tile->mem.kernel_bb_pool = xe_sa_bo_manager_init(tile, SZ_1M, 16);
if (IS_ERR(tile->mem.kernel_bb_pool))
return PTR_ERR(tile->mem.kernel_bb_pool);
- xe_wa_apply_tile_workarounds(tile);
-
- err = xe_tile_sysfs_init(tile);
-
return 0;
}
-
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
index 1c9e42ade6b0..eb939316d55b 100644
--- a/drivers/gpu/drm/xe/xe_tile.h
+++ b/drivers/gpu/drm/xe/xe_tile.h
@@ -12,6 +12,7 @@ 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);
+int xe_tile_init(struct xe_tile *tile);
void xe_tile_migrate_wait(struct xe_tile *tile);
diff --git a/drivers/gpu/drm/xe/xe_trace_bo.h b/drivers/gpu/drm/xe/xe_trace_bo.h
index ba0f61e7d2d6..4ff023b5d040 100644
--- a/drivers/gpu/drm/xe/xe_trace_bo.h
+++ b/drivers/gpu/drm/xe/xe_trace_bo.h
@@ -189,7 +189,7 @@ DECLARE_EVENT_CLASS(xe_vm,
),
TP_printk("dev=%s, vm=%p, asid=0x%05x", __get_str(dev),
- __entry->vm, __entry->asid)
+ __entry->vm, __entry->asid)
);
DEFINE_EVENT(xe_vm, xe_vm_kill,
diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
index f7113cf6109d..34e38bb167ba 100644
--- a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
+++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.c
@@ -57,12 +57,35 @@ bool xe_ttm_stolen_cpu_access_needs_ggtt(struct xe_device *xe)
return GRAPHICS_VERx100(xe) < 1270 && !IS_DGFX(xe);
}
+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 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 stolen_size, wopcm_size;
u64 tile_offset;
u64 tile_size;
@@ -74,7 +97,13 @@ static s64 detect_bar2_dgfx(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
if (drm_WARN_ON(&xe->drm, tile_size < mgr->stolen_base))
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 = tile_size - mgr->stolen_base;
+ stolen_size -= wopcm_size;
/* Verify usage fits in the actual resource available */
if (mgr->stolen_base + stolen_size <= pci_resource_len(pdev, LMEM_BAR))
@@ -89,29 +118,6 @@ static s64 detect_bar2_dgfx(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
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);
@@ -201,17 +207,16 @@ static u64 detect_stolen(struct xe_device *xe, struct xe_ttm_stolen_mgr *mgr)
#endif
}
-void xe_ttm_stolen_mgr_init(struct xe_device *xe)
+int 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);
+ struct xe_ttm_stolen_mgr *mgr;
u64 stolen_size, io_size;
int err;
- if (!mgr) {
- drm_dbg_kms(&xe->drm, "Stolen mgr init failed\n");
- return;
- }
+ mgr = drmm_kzalloc(&xe->drm, sizeof(*mgr), GFP_KERNEL);
+ if (!mgr)
+ return -ENOMEM;
if (IS_SRIOV_VF(xe))
stolen_size = 0;
@@ -224,7 +229,7 @@ void xe_ttm_stolen_mgr_init(struct xe_device *xe)
if (!stolen_size) {
drm_dbg_kms(&xe->drm, "No stolen memory support\n");
- return;
+ return 0;
}
/*
@@ -240,7 +245,7 @@ void xe_ttm_stolen_mgr_init(struct xe_device *xe)
io_size, PAGE_SIZE);
if (err) {
drm_dbg_kms(&xe->drm, "Stolen mgr init failed: %i\n", err);
- return;
+ return err;
}
drm_dbg_kms(&xe->drm, "Initialized stolen memory support with %llu bytes\n",
@@ -248,6 +253,8 @@ void xe_ttm_stolen_mgr_init(struct xe_device *xe)
if (io_size)
mgr->mapping = devm_ioremap_wc(&pdev->dev, mgr->io_base, io_size);
+
+ return 0;
}
u64 xe_ttm_stolen_io_offset(struct xe_bo *bo, u32 offset)
diff --git a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h
index 1777245ff810..8e877d1e839b 100644
--- a/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h
+++ b/drivers/gpu/drm/xe/xe_ttm_stolen_mgr.h
@@ -12,7 +12,7 @@ struct ttm_resource;
struct xe_bo;
struct xe_device;
-void xe_ttm_stolen_mgr_init(struct xe_device *xe);
+int 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);
diff --git a/drivers/gpu/drm/xe/xe_uc.c b/drivers/gpu/drm/xe/xe_uc.c
index 0d073a9987c2..bb03c524613f 100644
--- a/drivers/gpu/drm/xe/xe_uc.c
+++ b/drivers/gpu/drm/xe/xe_uc.c
@@ -241,7 +241,7 @@ void xe_uc_gucrc_disable(struct xe_uc *uc)
void xe_uc_stop_prepare(struct xe_uc *uc)
{
- xe_gsc_wait_for_worker_completion(&uc->gsc);
+ xe_gsc_stop_prepare(&uc->gsc);
xe_guc_stop_prepare(&uc->guc);
}
@@ -275,6 +275,12 @@ again:
goto again;
}
+void xe_uc_suspend_prepare(struct xe_uc *uc)
+{
+ xe_gsc_wait_for_worker_completion(&uc->gsc);
+ xe_guc_stop_prepare(&uc->guc);
+}
+
int xe_uc_suspend(struct xe_uc *uc)
{
/* GuC submission not enabled, nothing to do */
diff --git a/drivers/gpu/drm/xe/xe_uc.h b/drivers/gpu/drm/xe/xe_uc.h
index 506517c11333..ba2937ab94cf 100644
--- a/drivers/gpu/drm/xe/xe_uc.h
+++ b/drivers/gpu/drm/xe/xe_uc.h
@@ -18,6 +18,7 @@ int xe_uc_reset_prepare(struct xe_uc *uc);
void xe_uc_stop_prepare(struct xe_uc *uc);
void xe_uc_stop(struct xe_uc *uc);
int xe_uc_start(struct xe_uc *uc);
+void xe_uc_suspend_prepare(struct xe_uc *uc);
int xe_uc_suspend(struct xe_uc *uc);
int xe_uc_sanitize_reset(struct xe_uc *uc);
void xe_uc_remove(struct xe_uc *uc);
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 872de052d670..15fd497c920c 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -8,6 +8,7 @@
#include <linux/dma-fence-array.h>
#include <linux/nospec.h>
+#include <drm/drm_drv.h>
#include <drm/drm_exec.h>
#include <drm/drm_print.h>
#include <drm/ttm/ttm_execbuf_util.h>
@@ -1476,8 +1477,10 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
* scheduler drops all the references of it, hence protecting the VM
* for this case is necessary.
*/
- if (flags & XE_VM_FLAG_LR_MODE)
+ if (flags & XE_VM_FLAG_LR_MODE) {
+ INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func);
xe_pm_runtime_get_noresume(xe);
+ }
vm_resv_obj = drm_gpuvm_resv_object_alloc(&xe->drm);
if (!vm_resv_obj) {
@@ -1522,10 +1525,8 @@ struct xe_vm *xe_vm_create(struct xe_device *xe, u32 flags)
vm->batch_invalidate_tlb = true;
}
- if (vm->flags & XE_VM_FLAG_LR_MODE) {
- INIT_WORK(&vm->preempt.rebind_work, preempt_rebind_work_func);
+ if (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) {
@@ -1581,9 +1582,40 @@ err_no_resv:
static void xe_vm_close(struct xe_vm *vm)
{
+ struct xe_device *xe = vm->xe;
+ bool bound;
+ int idx;
+
+ bound = drm_dev_enter(&xe->drm, &idx);
+
down_write(&vm->lock);
+
vm->size = 0;
+
+ if (!((vm->flags & XE_VM_FLAG_MIGRATION))) {
+ struct xe_tile *tile;
+ struct xe_gt *gt;
+ u8 id;
+
+ /* Wait for pending binds */
+ dma_resv_wait_timeout(xe_vm_resv(vm),
+ DMA_RESV_USAGE_BOOKKEEP,
+ false, MAX_SCHEDULE_TIMEOUT);
+
+ if (bound) {
+ for_each_tile(tile, xe, id)
+ if (vm->pt_root[id])
+ xe_pt_clear(xe, vm->pt_root[id]);
+
+ for_each_gt(gt, xe, id)
+ xe_gt_tlb_invalidation_vm(gt, vm);
+ }
+ }
+
up_write(&vm->lock);
+
+ if (bound)
+ drm_dev_exit(idx);
}
void xe_vm_close_and_put(struct xe_vm *vm)
diff --git a/drivers/gpu/drm/xe/xe_wa.c b/drivers/gpu/drm/xe/xe_wa.c
index 0a1905f8d380..aea6034a8107 100644
--- a/drivers/gpu/drm/xe/xe_wa.c
+++ b/drivers/gpu/drm/xe/xe_wa.c
@@ -783,6 +783,10 @@ static const struct xe_rtp_entry_sr lrc_was[] = {
XE_RTP_RULES(GRAPHICS_VERSION(2001), ENGINE_CLASS(RENDER)),
XE_RTP_ACTIONS(SET(CHICKEN_RASTER_1, DIS_CLIP_NEGATIVE_BOUNDING_BOX))
},
+ { XE_RTP_NAME("22021007897"),
+ XE_RTP_RULES(GRAPHICS_VERSION(2001), ENGINE_CLASS(RENDER)),
+ XE_RTP_ACTIONS(SET(COMMON_SLICE_CHICKEN4, SBE_PUSH_CONSTANT_BEHIND_FIX_ENABLE))
+ },
/* Xe3_LPG */
{ XE_RTP_NAME("14021490052"),
diff --git a/drivers/gpu/drm/xe/xe_wa_oob.rules b/drivers/gpu/drm/xe/xe_wa_oob.rules
index 264d6e116499..93fa2708ee37 100644
--- a/drivers/gpu/drm/xe/xe_wa_oob.rules
+++ b/drivers/gpu/drm/xe/xe_wa_oob.rules
@@ -29,8 +29,10 @@
13011645652 GRAPHICS_VERSION(2004)
14022293748 GRAPHICS_VERSION(2001)
GRAPHICS_VERSION(2004)
+ GRAPHICS_VERSION_RANGE(3000, 3001)
22019794406 GRAPHICS_VERSION(2001)
GRAPHICS_VERSION(2004)
+ GRAPHICS_VERSION_RANGE(3000, 3001)
22019338487 MEDIA_VERSION(2000)
GRAPHICS_VERSION(2001)
MEDIA_VERSION(3000), MEDIA_STEP(A0, B0)