drm/sched: Avoid lockdep spalt on killing a processes
authorAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Thu, 28 Oct 2021 16:24:03 +0000 (12:24 -0400)
committerAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Mon, 1 Nov 2021 15:08:21 +0000 (11:08 -0400)
commit542cff7893a37445f98ece26aeb3c9c1055e9ea4
tree99236b129149a7482c2ae3c247c1af2d9411bc7f
parentf99413e4e1ceb23976207a8ad33a4fb65055092f
drm/sched: Avoid lockdep spalt on killing a processes

Probelm:
Singlaning one sched fence from within another's sched
fence singal callback generates lockdep splat because
the both have same lockdep class of their fence->lock

Fix:
Fix bellow stack by rescheduling to irq work of
signaling and killing of jobs that left when entity is killed.

[11176.741181]  dump_stack+0x10/0x12
[11176.741186] __lock_acquire.cold+0x208/0x2df
[11176.741197]  lock_acquire+0xc6/0x2d0
[11176.741204]  ? dma_fence_signal+0x28/0x80
[11176.741212] _raw_spin_lock_irqsave+0x4d/0x70
[11176.741219]  ? dma_fence_signal+0x28/0x80
[11176.741225]  dma_fence_signal+0x28/0x80
[11176.741230] drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[11176.741240] drm_sched_entity_kill_jobs_cb+0x1c/0x50 [gpu_sched]
[11176.741248] dma_fence_signal_timestamp_locked+0xac/0x1a0
[11176.741254]  dma_fence_signal+0x3b/0x80
[11176.741260] drm_sched_fence_finished+0x12/0x20 [gpu_sched]
[11176.741268] drm_sched_job_done.isra.0+0x7f/0x1a0 [gpu_sched]
[11176.741277] drm_sched_job_done_cb+0x12/0x20 [gpu_sched]
[11176.741284] dma_fence_signal_timestamp_locked+0xac/0x1a0
[11176.741290]  dma_fence_signal+0x3b/0x80
[11176.741296] amdgpu_fence_process+0xd1/0x140 [amdgpu]
[11176.741504] sdma_v4_0_process_trap_irq+0x8c/0xb0 [amdgpu]
[11176.741731]  amdgpu_irq_dispatch+0xce/0x250 [amdgpu]
[11176.741954]  amdgpu_ih_process+0x81/0x100 [amdgpu]
[11176.742174]  amdgpu_irq_handler+0x26/0xa0 [amdgpu]
[11176.742393] __handle_irq_event_percpu+0x4f/0x2c0
[11176.742402] handle_irq_event_percpu+0x33/0x80
[11176.742408]  handle_irq_event+0x39/0x60
[11176.742414]  handle_edge_irq+0x93/0x1d0
[11176.742419]  __common_interrupt+0x50/0xe0
[11176.742426]  common_interrupt+0x80/0x90

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Suggested-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Suggested-by: Christian König <christian.koenig@amd.com>
Tested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://www.spinics.net/lists/dri-devel/msg321250.html
drivers/gpu/drm/scheduler/sched_entity.c
include/drm/gpu_scheduler.h