Skip to contentSkip to conversation

drm/amdgpu: fix sched fence slab teardown

PATCH Applied Grazvydas Ignotas 4 participants 4 messages Oct 23, 2016 – Oct 24, 2016 1 message not in the archive

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/amdgpu: fix sched fence slab teardown

To free fences, call_rcu() is used, which calls amd_sched_fence_free() after a grace period. During teardown, there is no guarantee all callbacks have finished, so sched_fence_slab may be destroyed before all fences have been freed. If we are lucky, this results in some slab warnings, if not, we get a crash in one of rcu threads because callback is called after amdgpu has already been unloaded.

Fix it with a rcu_barrier().

Fixes 189e0fb76304 ("drm/amdgpu: RCU protected amd_sched_fence_release")
---
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +1 -0
index 963a24d..910b8d5 100644
@@ -645,6 +645,7 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
 {
 	if (sched->thread)
 		kthread_stop(sched->thread);
+	rcu_barrier();
 	if (atomic_dec_and_test(&sched_fence_slab_ref))
 		kmem_cache_destroy(sched_fence_slab);
 }
Signature · 6 lines
2.7.4

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel
Acked-by on v1 1/1
#1 lore raw
Quoted 20 lines · Grazvydas Ignotas

To free fences, call_rcu() is used, which calls amd_sched_fence_free() after a grace period. During teardown, there is no guarantee all callbacks have finished, so sched_fence_slab may be destroyed before all fences have been freed. If we are lucky, this results in some slab warnings, if not, we get a crash in one of rcu threads because callback is called after amdgpu has already been unloaded.

Fix it with a rcu_barrier().

Fixes 189e0fb76304 ("drm/amdgpu: RCU protected amd_sched_fence_release")
---
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 +
  1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +0 -0
index 963a24d..910b8d5 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c void amd_sched_fini(struct amd_gpu_scheduler *sched) · line 651

Quoted 7 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +1 -0
@@ -645,6 +645,7 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
  {
  	if (sched->thread)
  		kthread_stop(sched->thread);
+	rcu_barrier();
  	if (atomic_dec_and_test(&sched_fence_slab_ref))
  		kmem_cache_destroy(sched_fence_slab);
  }

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Reviewed-by
#2 lore raw

Am 24.10.2016 um 04:34 schrieb zhoucm1:

Quoted 27 lines
Quoted 24 lines · Grazvydas Ignotas

To free fences, call_rcu() is used, which calls amd_sched_fence_free() after a grace period. During teardown, there is no guarantee all callbacks have finished, so sched_fence_slab may be destroyed before all fences have been freed. If we are lucky, this results in some slab warnings, if not, we get a crash in one of rcu threads because callback is called after amdgpu has already been unloaded.

Fix it with a rcu_barrier().

Fixes 189e0fb76304 ("drm/amdgpu: RCU protected amd_sched_fence_release")
---
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 +
  1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +1 -0
index 963a24d..910b8d5 100644
@@ -645,6 +645,7 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
  {
      if (sched->thread)
          kthread_stop(sched->thread);
+    rcu_barrier();
Quoted 8 lines

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c void amd_sched_fini(struct amd_gpu_scheduler *sched) · line 651

Quoted 3 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +0 -0
      if (atomic_dec_and_test(&sched_fence_slab_ref))
          kmem_cache_destroy(sched_fence_slab);
  }

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Applied
#4 lore raw
In reply to a message not in the archive
Quoted 38 lines · Christian König

Am 24.10.2016 um 04:34 schrieb zhoucm1:

Quoted 34 lines
Quoted 29 lines · Grazvydas Ignotas

To free fences, call_rcu() is used, which calls amd_sched_fence_free() after a grace period. During teardown, there is no guarantee all callbacks have finished, so sched_fence_slab may be destroyed before all fences have been freed. If we are lucky, this results in some slab warnings, if not, we get a crash in one of rcu threads because callback is called after amdgpu has already been unloaded.

Fix it with a rcu_barrier().

Fixes 189e0fb76304 ("drm/amdgpu: RCU protected amd_sched_fence_release")
---
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 +
  1 file changed, 1 insertion(+)
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +1 -0
index 963a24d..910b8d5 100644
@@ -645,6 +645,7 @@ void amd_sched_fini(struct amd_gpu_scheduler *sched)
  {
      if (sched->thread)
          kthread_stop(sched->thread);
+    rcu_barrier();
      if (atomic_dec_and_test(&sched_fence_slab_ref))
          kmem_cache_destroy(sched_fence_slab);
  }

Applied. thanks!

Alex
_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx