kernel/linux.git/drivers/gpu/drm/msm/msm_ringbuffer.c, branch v6.12.80

Revert "drm/msm/gpu: Push gpu lock down past runpm"

2024-02-01T23:24:10+00:00

This reverts commit abe2023b4cea192ab266b351fd38dc9dbd846df0. Changing the locking order means that scheduler/msm_job_run() can race with the recovery kthread worker, with the result that the GPU gets an extra runpm get when we are trying to power it off. Leaving the GPU in an unrecovered state. I'll need to come up with a different scheme for appeasing lockdep. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/573835/

drm/msm/gem: Split out submit_unpin_objects() helper

2023-12-10T18:23:13+00:00

Untangle unpinning from unlock/unref loop. The unpin only happens in error paths so it is easier to decouple from the normal unlock path. Since we never have an intermediate state where a subset of buffers are pinned (ie. we never bail out of the pin or unpin loops) we can replace the bo state flag bit with a global flag in the submit. Signed-off-by: Rob Clark Reviewed-by: Dmitry Baryshkov Patchwork: https://patchwork.freedesktop.org/patch/568335/

drm/sched: Convert drm scheduler to use a work queue rather than kthread

2023-11-01T21:29:21+00:00

In Xe, the new Intel GPU driver, a choice has made to have a 1 to 1 mapping between a drm_gpu_scheduler and drm_sched_entity. At first this seems a bit odd but let us explain the reasoning below. 1. In Xe the submission order from multiple drm_sched_entity is not guaranteed to be the same completion even if targeting the same hardware engine. This is because in Xe we have a firmware scheduler, the GuC, which allowed to reorder, timeslice, and preempt submissions. If a using shared drm_gpu_scheduler across multiple drm_sched_entity, the TDR falls apart as the TDR expects submission order == completion order. Using a dedicated drm_gpu_scheduler per drm_sched_entity solve this problem. 2. In Xe submissions are done via programming a ring buffer (circular buffer), a drm_gpu_scheduler provides a limit on number of jobs, if the limit of number jobs is set to RING_SIZE / MAX_SIZE_PER_JOB we get flow control on the ring for free. A problem with this design is currently a drm_gpu_scheduler uses a kthread for submission / job cleanup. This doesn't scale if a large number of drm_gpu_scheduler are used. To work around the scaling issue, use a worker rather than kthread for submission / job cleanup. v2: - (Rob Clark) Fix msm build - Pass in run work queue v3: - (Boris) don't have loop in worker v4: - (Tvrtko) break out submit ready, stop, start helpers into own patch v5: - (Boris) default to ordered work queue v6: - (Luben / checkpatch) fix alignment in msm_ringbuffer.c - (Luben) s/drm_sched_submit_queue/drm_sched_wqueue_enqueue - (Luben) Update comment for drm_sched_wqueue_enqueue - (Luben) Positive check for submit_wq in drm_sched_init - (Luben) s/alloc_submit_wq/own_submit_wq v7: - (Luben) s/drm_sched_wqueue_enqueue/drm_sched_run_job_queue v8: - (Luben) Adjust var names / comments Signed-off-by: Matthew Brost Reviewed-by: Luben Tuikov Link: https://lore.kernel.org/r/20231031032439.1558703-3-matthew.brost@intel.com Signed-off-by: Luben Tuikov

drm/sched: Convert the GPU scheduler to variable number of run-queues

2023-10-26T16:03:47+00:00

The GPU scheduler has now a variable number of run-queues, which are set up at drm_sched_init() time. This way, each driver announces how many run-queues it requires (supports) per each GPU scheduler it creates. Note, that run-queues correspond to scheduler "priorities", thus if the number of run-queues is set to 1 at drm_sched_init(), then that scheduler supports a single run-queue, i.e. single "priority". If a driver further sets a single entity per run-queue, then this creates a 1-to-1 correspondence between a scheduler and a scheduled entity. Cc: Lucas Stach Cc: Russell King Cc: Qiang Yu Cc: Rob Clark Cc: Abhinav Kumar Cc: Dmitry Baryshkov Cc: Danilo Krummrich Cc: Matthew Brost Cc: Boris Brezillon Cc: Alex Deucher Cc: Christian König Cc: Emma Anholt Cc: etnaviv@lists.freedesktop.org Cc: lima@lists.freedesktop.org Cc: linux-arm-msm@vger.kernel.org Cc: freedreno@lists.freedesktop.org Cc: nouveau@lists.freedesktop.org Cc: dri-devel@lists.freedesktop.org Signed-off-by: Luben Tuikov Acked-by: Christian König Link: https://lore.kernel.org/r/20231023032251.164775-1-luben.tuikov@amd.com

drm/msm/gpu: Push gpu lock down past runpm

2023-08-15T17:09:30+00:00

Avoid holding gpu lock when calling runpm, to avoid this lockdep splat: ====================================================== WARNING: possible circular locking dependency detected 6.4.3-debug+ #14 Not tainted ------------------------------------------------------ ring0/373 is trying to acquire lock: ffffffead86efb98 (prepare_lock){+.+.}-{3:3}, at: clk_prepare_lock+0x70/0x98 but task is already holding lock: ffffff809cd19170 (&gpu->lock){+.+.}-{3:3}, at: msm_job_run+0x7c/0x128 [msm] which lock already depends on the new lock. the existing dependency chain (in reverse order) is: -> #4 (&gpu->lock){+.+.}-{3:3}: __mutex_lock+0xc8/0x388 mutex_lock_nested+0x2c/0x38 msm_job_run+0x7c/0x128 [msm] drm_sched_main+0x264/0x354 [gpu_sched] kthread+0xf0/0x100 ret_from_fork+0x10/0x20 -> #3 (dma_fence_map){++++}-{0:0}: __dma_fence_might_wait+0x74/0xc0 dma_resv_lockdep+0x1f0/0x2e8 do_one_initcall+0xb4/0x214 kernel_init_freeable+0x338/0x33c kernel_init+0x30/0x134 ret_from_fork+0x10/0x20 -> #2 (mmu_notifier_invalidate_range_start){+.+.}-{0:0}: fs_reclaim_acquire+0x7c/0x9c slab_pre_alloc_hook.constprop.0+0x40/0x250 __kmem_cache_alloc_node+0x60/0x18c kmalloc_node_trace+0x40/0x84 alloc_worker+0x2c/0x64 init_rescuer+0x34/0xe0 workqueue_init+0x168/0x1fc kernel_init_freeable+0x15c/0x33c kernel_init+0x30/0x134 ret_from_fork+0x10/0x20 -> #1 (fs_reclaim){+.+.}-{0:0}: __fs_reclaim_acquire+0x3c/0x48 fs_reclaim_acquire+0x50/0x9c slab_pre_alloc_hook.constprop.0+0x40/0x250 __kmem_cache_alloc_node+0x60/0x18c kmalloc_trace+0x44/0x88 clk_rcg2_dfs_determine_rate+0x60/0x214 clk_core_determine_round_nolock+0xb8/0xf0 clk_core_round_rate_nolock+0x84/0x118 clk_core_round_rate_nolock+0xd8/0x118 clk_round_rate+0x6c/0xd0 geni_se_clk_tbl_get+0x78/0xc0 geni_se_clk_freq_match+0x44/0xe4 get_spi_clk_cfg+0x50/0xf4 geni_spi_set_clock_and_bw+0x54/0x104 spi_geni_prepare_message+0x130/0x174 __spi_pump_transfer_message+0x200/0x4d8 __spi_sync+0x13c/0x23c spi_sync_locked+0x18/0x24 do_cros_ec_pkt_xfer_spi+0x124/0x3f0 cros_ec_xfer_high_pri_work+0x28/0x3c kthread_worker_fn+0x14c/0x27c kthread+0xf0/0x100 ret_from_fork+0x10/0x20 -> #0 (prepare_lock){+.+.}-{3:3}: __lock_acquire+0xdf8/0x109c lock_acquire+0x234/0x284 __mutex_lock+0xc8/0x388 mutex_lock_nested+0x2c/0x38 clk_prepare_lock+0x70/0x98 clk_prepare+0x24/0x50 clk_bulk_prepare+0x50/0x9c a6xx_gmu_resume+0x94/0x800 [msm] a6xx_gmu_pm_resume+0x38/0x158 [msm] adreno_runtime_resume+0x2c/0x38 [msm] pm_generic_runtime_resume+0x30/0x44 __rpm_callback+0x4c/0x134 rpm_callback+0x78/0x7c rpm_resume+0x3a4/0x46c __pm_runtime_resume+0x78/0xbc pm_runtime_get_sync.isra.0+0x14/0x20 [msm] msm_gpu_submit+0x4c/0x12c [msm] msm_job_run+0x88/0x128 [msm] drm_sched_main+0x264/0x354 [gpu_sched] kthread+0xf0/0x100 ret_from_fork+0x10/0x20 other info that might help us debug this: Chain exists of: prepare_lock --> dma_fence_map --> &gpu->lock Possible unsafe locking scenario: CPU0 CPU1 ---- ---- lock(&gpu->lock); lock(dma_fence_map); lock(&gpu->lock); lock(prepare_lock); *** DEADLOCK *** 2 locks held by ring0/373: #0: ffffffead875ae50 (dma_fence_map){++++}-{0:0}, at: drm_sched_main+0x54/0x354 [gpu_sched] #1: ffffff809cd19170 (&gpu->lock){+.+.}-{3:3}, at: msm_job_run+0x7c/0x128 [msm] stack backtrace: CPU: 2 PID: 373 Comm: ring0 Not tainted 6.4.3-debug+ #14 Hardware name: Google Villager (rev1+) with LTE (DT) Call trace: dump_backtrace+0xb4/0xf0 show_stack+0x20/0x30 dump_stack_lvl+0x60/0x84 dump_stack+0x18/0x24 print_circular_bug+0x1cc/0x234 check_noncircular+0x78/0xac __lock_acquire+0xdf8/0x109c lock_acquire+0x234/0x284 __mutex_lock+0xc8/0x388 mutex_lock_nested+0x2c/0x38 clk_prepare_lock+0x70/0x98 clk_prepare+0x24/0x50 clk_bulk_prepare+0x50/0x9c a6xx_gmu_resume+0x94/0x800 [msm] a6xx_gmu_pm_resume+0x38/0x158 [msm] adreno_runtime_resume+0x2c/0x38 [msm] pm_generic_runtime_resume+0x30/0x44 __rpm_callback+0x4c/0x134 rpm_callback+0x78/0x7c rpm_resume+0x3a4/0x46c __pm_runtime_resume+0x78/0xbc pm_runtime_get_sync.isra.0+0x14/0x20 [msm] msm_gpu_submit+0x4c/0x12c [msm] msm_job_run+0x88/0x128 [msm] drm_sched_main+0x264/0x354 [gpu_sched] kthread+0xf0/0x100 ret_from_fork+0x10/0x20 Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/552298/

drm/msm: Remove vma use tracking

2023-08-10T20:08:03+00:00

This was not strictly necessary, as page unpinning (ie. shrinker) only cares about the resv. It did give us some extra sanity checking for userspace controlled iova, and was useful to catch issues on kernel and userspace side when enabling userspace iova. But if userspace screws this up, it just corrupts it's own gpu buffers and/or gets iova faults. So we can just let userspace shoot it's own foot and drop the extra per- buffer SUBMIT overhead. Signed-off-by: Rob Clark Acked-by: Daniel Vetter Patchwork: https://patchwork.freedesktop.org/patch/551023/

drm/msm: Use drm_gem_object in submit bos table

2023-08-10T17:44:02+00:00

Basically everywhere wants the base ptr type. So store that instead of msm_gem_object. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/551021/

drm/msm: Take lru lock once per job_run

2023-08-10T17:44:01+00:00

Rather than acquiring it and dropping it for each individual obj. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/551019/

drm/msm/gem: Avoid obj lock in job_run()

2023-03-25T23:31:44+00:00

Now that everything that controls which LRU an obj lives in *except* the backing pages is protected by the LRU lock, add a special path to unpin in the job_run() path, where we are assured that we already have backing pages and will not be racing against eviction (because the GEM object's dma_resv contains the fence that will be signaled when the submit/job completes). Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527845/ Link: https://lore.kernel.org/r/20230320144356.803762-10-robdclark@gmail.com

drm/msm: Decouple vma tracking from obj lock

2023-03-25T23:31:44+00:00

We need to use the inuse count to track that a BO is pinned until we have the hw_fence. But we want to remove the obj lock from the job_run() path as this could deadlock against reclaim/shrinker (because it is blocking the hw_fence from eventually being signaled). So split that tracking out into a per-vma lock with narrower scope. Signed-off-by: Rob Clark Patchwork: https://patchwork.freedesktop.org/patch/527839/ Link: https://lore.kernel.org/r/20230320144356.803762-5-robdclark@gmail.com