Skip to contentSkip to conversation

drm/amdgpu:cleanup force_completion

PATCH In review Monk Liu 3 participants 15 messages Oct 20, 2017 – Oct 23, 2017 10 messages not in the archive

Subsystems (2)
lore
Share & tools
lore

Monk Liu

author
#0 lore raw

PATCH 1/5 drm/amdgpu:cleanup force_completion

cleanups, now only operate on the given ring

Change-Id I42ee081696ac348660d38be68807f34090ffcce2
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  6 +++---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c  | 29 +++++++----------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h   |  3 +--
 3 files changed, 11 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -3
index 99acf29..dae1e5b 100644
@@ -2871,7 +2871,7 @@ int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, struct amdgpu_job *job)
 		amd_sched_hw_job_reset(&ring->sched);
 
 		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-		amdgpu_fence_driver_force_completion_ring(ring);
+		amdgpu_fence_driver_force_completion(ring);
 	}
 
 	/* request to take full control of GPU before re-initialization  */
@@ -2990,9 +2990,9 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 			continue;
 		kthread_park(ring->sched.thread);
 		amd_sched_hw_job_reset(&ring->sched);
+		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
+		amdgpu_fence_driver_force_completion(ring);
 	}
-	/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-	amdgpu_fence_driver_force_completion(adev);
 
 	need_full_reset = amdgpu_need_full_reset(adev);
 
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +7 -22
index fb9f88ef..2167dac 100644
@@ -499,7 +499,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
 		r = amdgpu_fence_wait_empty(ring);
 		if (r) {
 			/* no need to trigger GPU reset as we are unloading */
-			amdgpu_fence_driver_force_completion(adev);
+			amdgpu_fence_driver_force_completion(ring);
 		}
 		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
 			       ring->fence_drv.irq_type);
@@ -534,7 +534,7 @@ void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
 		r = amdgpu_fence_wait_empty(ring);
 		if (r) {
 			/* delay GPU reset to resume */
-			amdgpu_fence_driver_force_completion(adev);
+			amdgpu_fence_driver_force_completion(ring);
 		}
 
 		/* disable the interrupt */
@@ -571,30 +571,15 @@ void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_fence_driver_force_completion - force all fence waiter to complete
+ * amdgpu_fence_driver_force_completion - force signal latest fence of ring
  *
- * @adev: amdgpu device pointer
+ * @ring: fence of the ring to signal
  *
- * In case of GPU reset failure make sure no process keep waiting on fence
- * that will never complete.
  */
-void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev)
+void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring)
 {
-	int i;
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring || !ring->fence_drv.initialized)
-			continue;
-
-		amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
-	}
-}
-
-void amdgpu_fence_driver_force_completion_ring(struct amdgpu_ring *ring)
-{
-	if (ring)
-		amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+	amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+	amdgpu_fence_process(ring);
 }
 
 /*
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +1 -2
index b18c2b9..a6b89e3 100644
@@ -79,8 +79,7 @@ struct amdgpu_fence_driver {
 
 int amdgpu_fence_driver_init(struct amdgpu_device *adev);
 void amdgpu_fence_driver_fini(struct amdgpu_device *adev);
-void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev);
-void amdgpu_fence_driver_force_completion_ring(struct amdgpu_ring *ring);
+void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring);
 
 int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 				  unsigned num_hw_submission);
Signature · 6 lines
2.7.4

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

A message that is not in the archive (sent off-list or not yet received): 1508470385-16164-1-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org

author
#2 lore raw

PATCH 2/5 drm/amdgpu:add hang_limit for sched

since gpu_scheduler source domain cannot access amdgpu variable so need create the hang_limit membewr for sched, and it can refer it for the upcoming GPU RESET patches

Change-Id I977ae2717e55a8b87c59e58a288bffc3b458b653
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c     | 2 ++
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
 2 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +2 -0
index 2167dac..f6a55cf 100644
@@ -453,6 +453,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
 				  ring->name);
 			return r;
 		}
+
+		ring->sched.hang_limit = amdgpu_job_hang_limit;
 	}
 
 	return 0;
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
index 52c8e54..1dac7bc 100644
@@ -142,6 +142,7 @@ struct amd_gpu_scheduler {
 	struct task_struct		*thread;
 	struct list_head	ring_mirror_list;
 	spinlock_t			job_list_lock;
+	int hang_limit;
 };
 
 int amd_sched_init(struct amd_gpu_scheduler *sched,
Signature · 6 lines
2.7.4

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

A message that is not in the archive (sent off-list or not yet received): 1508470385-16164-2-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org

#4 lore raw

Re: [PATCH 2/5] drm/amdgpu:add hang_limit for sched

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 16 lines

since gpu_scheduler source domain cannot access amdgpu variable so need create the hang_limit membewr for sched, and it can refer it for the upcoming GPU RESET patches

Change-Id I977ae2717e55a8b87c59e58a288bffc3b458b653
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c     | 2 ++
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
  2 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +0 -0
index 2167dac..f6a55cf 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, · line 457

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +2 -0
@@ -453,6 +453,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
  				  ring->name);
  			return r;
  		}
+
+		ring->sched.hang_limit = amdgpu_job_hang_limit;

Don't set it directly, make it a parameter of amd_sched_init().

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_gpu_scheduler { · line 145

Quoted 12 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +0 -0
  	}
 
  	return 0;
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
index 52c8e54..1dac7bc 100644
@@ -142,6 +142,7 @@ struct amd_gpu_scheduler {
  	struct task_struct		*thread;
  	struct list_head	ring_mirror_list;
  	spinlock_t			job_list_lock;
+	int hang_limit;

Please indent the new field the same way as the other one in the structure.

Christian.

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_gpu_scheduler { · line 148

Quoted 3 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
  };
 
  int amd_sched_init(struct amd_gpu_scheduler *sched,

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

A message that is not in the archive (sent off-list or not yet received): 2002cd84-8f78-52c3-d56c-faa43add4050-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

author
#6 lore raw

RE: [PATCH 2/5] drm/amdgpu:add hang_limit for sched

Quoted 1 line

Don't set it directly, make it a parameter of amd_sched_init().

Okay

the indent is somehow always looks incorrect in email, but it already aligned actually

-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: 2017年10月20日 17:03
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 2/5] drm/amdgpu:add hang_limit for sched

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 17 lines

since gpu_scheduler source domain cannot access amdgpu variable so need create the hang_limit membewr for sched, and it can refer it for the upcoming GPU RESET patches

Change-Id I977ae2717e55a8b87c59e58a288bffc3b458b653
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c     | 2 ++
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
  2 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +0 -0
index 2167dac..f6a55cf 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring, · line 457

Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +2 -0
@@ -453,6 +453,8 @@ int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
  				  ring->name);
  			return r;
  		}
+
+		ring->sched.hang_limit = amdgpu_job_hang_limit;

Don't set it directly, make it a parameter of amd_sched_init().

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_gpu_scheduler { · line 145

Quoted 13 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +0 -0
  	}
 
  	return 0;
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
index 52c8e54..1dac7bc 100644
@@ -142,6 +142,7 @@ struct amd_gpu_scheduler {
  	struct task_struct		*thread;
  	struct list_head	ring_mirror_list;
  	spinlock_t			job_list_lock;
+	int hang_limit;

Please indent the new field the same way as the other one in the structure.

Christian.

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_gpu_scheduler { · line 148

Quoted 3 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
  };
 
  int amd_sched_init(struct amd_gpu_scheduler *sched,

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

author
#7 lore raw

PATCH 3/5 drm/amdgpu:implement guilty pointer

In reply to a message not in the archive

for user context there will be a guilty pointer in entity that points to the guilty member of its context, thus we cant track if a given entity is not from kernel ctx and if it is already marked guilty.

Change-Id Ie0a30830d89f52a6c4a514e67206140698a46367
---
 drivers/gpu/drm/amd/amdgpu/amdgpu.h           | 1 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 1 +
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
 3 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index 774edc1..6a4178b 100644
@@ -744,6 +744,7 @@ struct amdgpu_ctx {
 	enum amd_sched_priority init_priority;
 	enum amd_sched_priority override_priority;
 	struct mutex            lock;
+	atomic_t	guilty;
 };
 
 struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +1 -0
index c184468..d822e95 100644
@@ -93,6 +93,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 					  rq, amdgpu_sched_jobs);
 		if (r)
 			goto failed;
+		ctx->rings[i].entity.guilty = &ctx->guilty;
 	}
 
 	r = amdgpu_queue_mgr_init(adev, &ctx->queue_mgr);
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
index 1dac7bc..2d59fc5 100644
@@ -50,6 +50,7 @@ struct amd_sched_entity {
 
 	struct dma_fence		*dependency;
 	struct dma_fence_cb		cb;
+	atomic_t	*guilty; /* points to ctx's guilty */
 };
 
 /**
Signature · 6 lines
2.7.4

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

A message that is not in the archive (sent off-list or not yet received): 1508470385-16164-3-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org

#9 lore raw

Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 30 lines

for user context there will be a guilty pointer in entity that points to the guilty member of its context, thus we cant track if a given entity is not from kernel ctx and if it is already marked guilty.

Change-Id Ie0a30830d89f52a6c4a514e67206140698a46367
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h           | 1 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 1 +
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
  3 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index 774edc1..6a4178b 100644
@@ -744,6 +744,7 @@ struct amdgpu_ctx {
  	enum amd_sched_priority init_priority;
  	enum amd_sched_priority override_priority;
  	struct mutex            lock;
+	atomic_t	guilty;
  };
 
  struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
index c184468..d822e95 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static int amdgpu_ctx_init(struct amdgpu_device *adev, · line 96

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +1 -0
@@ -93,6 +93,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
  					  rq, amdgpu_sched_jobs);
  		if (r)
  			goto failed;
+		ctx->rings[i].entity.guilty = &ctx->guilty;

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

Regards,
Christian.

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
  	}
 
  	r = amdgpu_queue_mgr_init(adev, &ctx->queue_mgr);
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
index 1dac7bc..2d59fc5 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_sched_entity { · line 56

Quoted 7 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
@@ -50,6 +50,7 @@ struct amd_sched_entity {
 
  	struct dma_fence		*dependency;
  	struct dma_fence_cb		cb;
+	atomic_t	*guilty; /* points to ctx's guilty */
  };
 
  /**

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

A message that is not in the archive (sent off-list or not yet received): f3a9431e-31fd-e8a5-79e9-d8bb09de366b-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

author
#11 lore raw

RE: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

[ML] that way the first patch will break kernel from compiling because you change the interface of entity_init without changing all place Calling this interface

I suggest just use one patch

-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: 2017年10月20日 17:09
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 31 lines

for user context there will be a guilty pointer in entity that points to the guilty member of its context, thus we cant track if a given entity is not from kernel ctx and if it is already marked guilty.

Change-Id Ie0a30830d89f52a6c4a514e67206140698a46367
---
  drivers/gpu/drm/amd/amdgpu/amdgpu.h           | 1 +
  drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 1 +
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
  3 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index 774edc1..6a4178b 100644
@@ -744,6 +744,7 @@ struct amdgpu_ctx {
  	enum amd_sched_priority init_priority;
  	enum amd_sched_priority override_priority;
  	struct mutex            lock;
+	atomic_t	guilty;
  };
 
  struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
index c184468..d822e95 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static int amdgpu_ctx_init(struct amdgpu_device *adev, · line 96

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +1 -0
@@ -93,6 +93,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
  					  rq, amdgpu_sched_jobs);
  		if (r)
  			goto failed;
+		ctx->rings[i].entity.guilty = &ctx->guilty;

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

Regards,
Christian.

Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
  	}
 
  	r = amdgpu_queue_mgr_init(adev, &ctx->queue_mgr); diff --git 

a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 1dac7bc..2d59fc5 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_sched_entity { · line 56

Quoted 7 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
@@ -50,6 +50,7 @@ struct amd_sched_entity {
 
  	struct dma_fence		*dependency;
  	struct dma_fence_cb		cb;
+	atomic_t	*guilty; /* points to ctx's guilty */
  };
 
  /**

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

A message that is not in the archive (sent off-list or not yet received): BLUPR12MB0449682AC30F4365B0A5068284460-7LeqcoF/[email protected]

#13 lore raw

Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 23.10.2017 um 06:26 schrieb Liu, Monk:

Quoted 11 lines

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

[ML] that way the first patch will break kernel from compiling because you change the interface of entity_init without changing all place Calling this interface

You obviously need to adjust the calls to entity_init and at least provide NULL as value for the new parameter.

Quoted 1 line

I suggest just use one patch

I prefer to separate that, but not a strict must have when you don't have time.

The scheduler and amdgpu using it should essentially be two different components.

Regards,
Christian.

Quoted 64 lines

-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: 2017年10月20日 17:09
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 31 lines

for user context there will be a guilty pointer in entity that points to the guilty member of its context, thus we cant track if a given entity is not from kernel ctx and if it is already marked guilty.

Change-Id Ie0a30830d89f52a6c4a514e67206140698a46367
---
   drivers/gpu/drm/amd/amdgpu/amdgpu.h           | 1 +
   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 1 +
   drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
   3 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index 774edc1..6a4178b 100644
@@ -744,6 +744,7 @@ struct amdgpu_ctx {
   	enum amd_sched_priority init_priority;
   	enum amd_sched_priority override_priority;
   	struct mutex            lock;
+	atomic_t	guilty;
   };
 
   struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
index c184468..d822e95 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static int amdgpu_ctx_init(struct amdgpu_device *adev, · line 96

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +1 -0
@@ -93,6 +93,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
   					  rq, amdgpu_sched_jobs);
   		if (r)
   			goto failed;
+		ctx->rings[i].entity.guilty = &ctx->guilty;

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

Regards,
Christian.

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
   	}
 
   	r = amdgpu_queue_mgr_init(adev, &ctx->queue_mgr); diff --git

a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 1dac7bc..2d59fc5 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
@@ -50,6 +50,7 @@ struct amd_sched_entity {
 
Quoted 7 lines

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_sched_entity { · line 56

Quoted 6 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
   	struct dma_fence		*dependency;
   	struct dma_fence_cb		cb;
+	atomic_t	*guilty; /* points to ctx's guilty */
   };
 
   /**

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

A message that is not in the archive (sent off-list or not yet received): 9ee4a45a-53c5-5a70-449f-714b90a3de08-5C7GfCeVMHo@public.gmane.org

author
#15 lore raw

RE: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Yeah, already did so, please check the amd-gfx

-----Original Message-----
From: Koenig, Christian
Sent: 2017年10月23日 15:17
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 23.10.2017 um 06:26 schrieb Liu, Monk:

Quoted 4 lines

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

Quoted 8 lines

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

[ML] that way the first patch will break kernel from compiling because you change the interface of entity_init without changing all place Calling this interface

You obviously need to adjust the calls to entity_init and at least provide NULL as value for the new parameter.

Quoted 1 line

I suggest just use one patch

I prefer to separate that, but not a strict must have when you don't have time.

The scheduler and amdgpu using it should essentially be two different components.

Regards,
Christian.

Quoted 64 lines

-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: 2017年10月20日 17:09
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 3/5] drm/amdgpu:implement guilty pointer

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 31 lines

for user context there will be a guilty pointer in entity that points to the guilty member of its context, thus we cant track if a given entity is not from kernel ctx and if it is already marked guilty.

Change-Id Ie0a30830d89f52a6c4a514e67206140698a46367
---
   drivers/gpu/drm/amd/amdgpu/amdgpu.h           | 1 +
   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c       | 1 +
   drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 +
   3 files changed, 3 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +1 -0
index 774edc1..6a4178b 100644
@@ -744,6 +744,7 @@ struct amdgpu_ctx {
   	enum amd_sched_priority init_priority;
   	enum amd_sched_priority override_priority;
   	struct mutex            lock;
+	atomic_t	guilty;
   };
 
   struct amdgpu_ctx_mgr {
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
index c184468..d822e95 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c static int amdgpu_ctx_init(struct amdgpu_device *adev, · line 96

Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +1 -0
@@ -93,6 +93,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
   					  rq, amdgpu_sched_jobs);
   		if (r)
   			goto failed;
+		ctx->rings[i].entity.guilty = &ctx->guilty;

Don't set this directly, make it a parameter of amd_sched_entity_init().

And please split up the patches differently.

First all changes to the scheduler, including the guilty marking.

Then the changes to amdgpu.

Regards,
Christian.

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +0 -0
   	}
 
   	r = amdgpu_queue_mgr_init(adev, &ctx->queue_mgr); diff --git 

a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h
index 1dac7bc..2d59fc5 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
@@ -50,6 +50,7 @@ struct amd_sched_entity {
 
Quoted 7 lines

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h struct amd_sched_entity { · line 56

Quoted 6 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -0
   	struct dma_fence		*dependency;
   	struct dma_fence_cb		cb;
+	atomic_t	*guilty; /* points to ctx's guilty */
   };
 
   /**

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

author
#16 lore raw

PATCH 4/5 drm/amdgpu:cleanup job reset routine

In reply to a message not in the archive

merge the setting guilty on context into this function to avoid implement extra routine.

Change-Id I7a0063464fdc85d5ac9080046380e745565ff540
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  4 ++--
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 13 ++++++++++++-
 drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  2 +-
 3 files changed, 15 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -2
index dae1e5b..a07544d 100644
@@ -2868,7 +2868,7 @@ int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, struct amdgpu_job *job)
 			amd_sched_job_kickout(&job->base);
 
 		/* only do job_reset on the hang ring if @job not NULL */
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
 
 		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
 		amdgpu_fence_driver_force_completion(ring);
@@ -2989,7 +2989,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
 		if (!ring || !ring->sched.thread)
 			continue;
 		kthread_park(ring->sched.thread);
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
 		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
 		amdgpu_fence_driver_force_completion(ring);
 	}
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +12 -1
index e4d3b4e..322b6c1 100644
@@ -442,7 +442,14 @@ static void amd_sched_job_timedout(struct work_struct *work)
 	job->sched->ops->timedout_job(job);
 }
 
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
+static void amd_sched_set_guilty(struct amd_sched_job *s_job)
+{
+	if (atomic_inc_return(&s_job->karma) > s_job->sched->hang_limit)
+		if (s_job->s_entity->guilty)
+			atomic_set(s_job->s_entity->guilty, 1);
+}
+
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *bad)
 {
 	struct amd_sched_job *s_job;
 
@@ -455,6 +462,10 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
 			s_job->s_fence->parent = NULL;
 			atomic_dec(&sched->hw_rq_count);
 		}
+
+		if (bad && bad == s_job) {
+			amd_sched_set_guilty(s_job);
+		}
 	}
 	spin_unlock(&sched->job_list_lock);
 }
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -1
index 2d59fc5..fff7cc7 100644
@@ -172,7 +172,7 @@ int amd_sched_job_init(struct amd_sched_job *job,
 		       struct amd_gpu_scheduler *sched,
 		       struct amd_sched_entity *entity,
 		       void *owner);
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched);
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *job);
 void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
 bool amd_sched_dependency_optimized(struct dma_fence* fence,
 				    struct amd_sched_entity *entity);
Signature · 6 lines
2.7.4

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

A message that is not in the archive (sent off-list or not yet received): 1508470385-16164-4-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org

#18 lore raw

Re: [PATCH 4/5] drm/amdgpu:cleanup job reset routine

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 38 lines

merge the setting guilty on context into this function to avoid implement extra routine.

Change-Id I7a0063464fdc85d5ac9080046380e745565ff540
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  4 ++--
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 13 ++++++++++++-
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  2 +-
  3 files changed, 15 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -2
index dae1e5b..a07544d 100644
@@ -2868,7 +2868,7 @@ int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, struct amdgpu_job *job)
  			amd_sched_job_kickout(&job->base);
 
  		/* only do job_reset on the hang ring if @job not NULL */
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
 
  		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
  		amdgpu_fence_driver_force_completion(ring);
@@ -2989,7 +2989,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  		if (!ring || !ring->sched.thread)
  			continue;
  		kthread_park(ring->sched.thread);
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
  		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
  		amdgpu_fence_driver_force_completion(ring);
  	}
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +0 -0
index e4d3b4e..322b6c1 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c static void amd_sched_job_timedout(struct work_struct *work) · line 449

Quoted 9 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +5 -1
@@ -442,7 +442,14 @@ static void amd_sched_job_timedout(struct work_struct *work)
  	job->sched->ops->timedout_job(job);
  }
 
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
+static void amd_sched_set_guilty(struct amd_sched_job *s_job)
+{
+	if (atomic_inc_return(&s_job->karma) > s_job->sched->hang_limit)
+		if (s_job->s_entity->guilty)
+			atomic_set(s_job->s_entity->guilty, 1);

The entity might be long freed at this point.

You need to walk the list of entities (with lock held) and check if entity->fence_context == job->s_fence->scheduled.context and only if you found the right one set the guilty flag there.

Christian.

Quoted 22 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +7 -0
+}
+
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *bad)
  {
  	struct amd_sched_job *s_job;
 
@@ -455,6 +462,10 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
  			s_job->s_fence->parent = NULL;
  			atomic_dec(&sched->hw_rq_count);
  		}
+
+		if (bad && bad == s_job) {
+			amd_sched_set_guilty(s_job);
+		}
  	}
  	spin_unlock(&sched->job_list_lock);
  }
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
index 2d59fc5..fff7cc7 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h int amd_sched_job_init(struct amd_sched_job *job, · line 178

Quoted 8 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -1
@@ -172,7 +172,7 @@ int amd_sched_job_init(struct amd_sched_job *job,
  		       struct amd_gpu_scheduler *sched,
  		       struct amd_sched_entity *entity,
  		       void *owner);
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched);
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *job);
  void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
  bool amd_sched_dependency_optimized(struct dma_fence* fence,
  				    struct amd_sched_entity *entity);

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

A message that is not in the archive (sent off-list or not yet received): 85ee460e-1f65-69c7-c17b-329d9b31d205-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org

author
#20 lore raw

RE: [PATCH 4/5] drm/amdgpu:cleanup job reset routine

Quoted 1 line

You need to walk the list of entities (with lock held) and check if

entity->fence_context == job->s_fence->scheduled.context and only if you found the right one set the guilty flag there.

Yeah you remind me, I did this in the old gpu reset patch, will add them back !

BR Monk
-----Original Message-----
From: Christian König [mailto:[email protected]]
Sent: 2017年10月20日 17:07
To: Liu, Monk <[email protected]>; [email protected]
Subject: Re: [PATCH 4/5] drm/amdgpu:cleanup job reset routine

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 40 lines

merge the setting guilty on context into this function to avoid implement extra routine.

Change-Id I7a0063464fdc85d5ac9080046380e745565ff540
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c    |  4 ++--
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 13 ++++++++++++-
  drivers/gpu/drm/amd/scheduler/gpu_scheduler.h |  2 +-
  3 files changed, 15 insertions(+), 4 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +2 -2
index dae1e5b..a07544d 100644
@@ -2868,7 +2868,7 @@ int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, struct amdgpu_job *job)
  			amd_sched_job_kickout(&job->base);
 
  		/* only do job_reset on the hang ring if @job not NULL */
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
 
  		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
  		amdgpu_fence_driver_force_completion(ring);
@@ -2989,7 +2989,7 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  		if (!ring || !ring->sched.thread)
  			continue;
  		kthread_park(ring->sched.thread);
-		amd_sched_hw_job_reset(&ring->sched);
+		amd_sched_hw_job_reset(&ring->sched, NULL);
  		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
  		amdgpu_fence_driver_force_completion(ring);
  	}
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +0 -0
index e4d3b4e..322b6c1 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c static void amd_sched_job_timedout(struct work_struct *work) · line 448

Quoted 8 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +4 -1
@@ -442,7 +442,14 @@ static void amd_sched_job_timedout(struct work_struct *work)
  	job->sched->ops->timedout_job(job);
  }
 
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
+static void amd_sched_set_guilty(struct amd_sched_job *s_job) {
+	if (atomic_inc_return(&s_job->karma) > s_job->sched->hang_limit)
+		if (s_job->s_entity->guilty)
+			atomic_set(s_job->s_entity->guilty, 1);

The entity might be long freed at this point.

You need to walk the list of entities (with lock held) and check if entity->fence_context == job->s_fence->scheduled.context and only if you found the right one set the guilty flag there.

Christian.

Quoted 22 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +7 -0
+}
+
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *bad)
  {
  	struct amd_sched_job *s_job;
 
@@ -455,6 +462,10 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched)
  			s_job->s_fence->parent = NULL;
  			atomic_dec(&sched->hw_rq_count);
  		}
+
+		if (bad && bad == s_job) {
+			amd_sched_set_guilty(s_job);
+		}
  	}
  	spin_unlock(&sched->job_list_lock);
  }
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +0 -0
index 2d59fc5..fff7cc7 100644

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h int amd_sched_job_init(struct amd_sched_job *job, · line 178

Quoted 8 lines
drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +1 -1
@@ -172,7 +172,7 @@ int amd_sched_job_init(struct amd_sched_job *job,
  		       struct amd_gpu_scheduler *sched,
  		       struct amd_sched_entity *entity,
  		       void *owner);
-void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched);
+void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched, struct amd_sched_job *job);
  void amd_sched_job_recovery(struct amd_gpu_scheduler *sched);
  bool amd_sched_dependency_optimized(struct dma_fence* fence,
  				    struct amd_sched_entity *entity);

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

author
#21 lore raw

PATCH 5/5 drm/amdgpu:skip job for guilty ctx in parser_init

In reply to a message not in the archive
Change-Id I44019f6475b1eaaba55633cf5f8bb84284f19a2c
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 ++++++
 1 file changed, 6 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +6 -0
index f7fceb6..1d5cbe7 100644
@@ -90,6 +90,12 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
 		goto free_chunk;
 	}
 
+	/* skip guilty context job */
+	if (atomic_read(&p->ctx->guilty) == 1) {
+		ret = -ECANCELED;
+		goto free_chunk;
+	}
+
 	mutex_lock(&p->ctx->lock);
 
 	/* get chunks */
Signature · 6 lines
2.7.4

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

A message that is not in the archive (sent off-list or not yet received): 1508470385-16164-5-git-send-email-Monk.Liu-5C7GfCeVMHo@public.gmane.org

Reviewed-by
#23 lore raw

Re: [PATCH 5/5] drm/amdgpu:skip job for guilty ctx in parser_init

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 2 lines
Change-Id I44019f6475b1eaaba55633cf5f8bb84284f19a2c
Quoted 11 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 6 ++++++
  1 file changed, 6 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +0 -0
index f7fceb6..1d5cbe7 100644
@@ -90,6 +90,12 @@ static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data)
  		goto free_chunk;
  	}

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c static int amdgpu_cs_parser_init(struct amdgpu_cs_parser *p, void *data) · line 101

Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +6 -0
 
+	/* skip guilty context job */
+	if (atomic_read(&p->ctx->guilty) == 1) {
+		ret = -ECANCELED;
+		goto free_chunk;
+	}
+
  	mutex_lock(&p->ctx->lock);
 
  	/* get chunks */

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

Reviewed-by
#24 lore raw
In reply to a message not in the archive

Am 20.10.2017 um 05:33 schrieb Monk Liu:

Quoted 4 lines

cleanups, now only operate on the given ring

Change-Id I42ee081696ac348660d38be68807f34090ffcce2
Quoted 94 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  6 +++---
  drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c  | 29 +++++++----------------------
  drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h   |  3 +--
  3 files changed, 11 insertions(+), 27 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +3 -3
index 99acf29..dae1e5b 100644
@@ -2871,7 +2871,7 @@ int amdgpu_sriov_gpu_reset(struct amdgpu_device *adev, struct amdgpu_job *job)
  		amd_sched_hw_job_reset(&ring->sched);
 
  		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-		amdgpu_fence_driver_force_completion_ring(ring);
+		amdgpu_fence_driver_force_completion(ring);
  	}
 
  	/* request to take full control of GPU before re-initialization  */
@@ -2990,9 +2990,9 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev)
  			continue;
  		kthread_park(ring->sched.thread);
  		amd_sched_hw_job_reset(&ring->sched);
+		/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
+		amdgpu_fence_driver_force_completion(ring);
  	}
-	/* after all hw jobs are reset, hw fence is meaningless, so force_completion */
-	amdgpu_fence_driver_force_completion(adev);
 
  	need_full_reset = amdgpu_need_full_reset(adev);
 
drivers/gpu/drm/amd/amdgpu/amdgpu_fence.c +7 -22
index fb9f88ef..2167dac 100644
@@ -499,7 +499,7 @@ void amdgpu_fence_driver_fini(struct amdgpu_device *adev)
  		r = amdgpu_fence_wait_empty(ring);
  		if (r) {
  			/* no need to trigger GPU reset as we are unloading */
-			amdgpu_fence_driver_force_completion(adev);
+			amdgpu_fence_driver_force_completion(ring);
  		}
  		amdgpu_irq_put(adev, ring->fence_drv.irq_src,
  			       ring->fence_drv.irq_type);
@@ -534,7 +534,7 @@ void amdgpu_fence_driver_suspend(struct amdgpu_device *adev)
  		r = amdgpu_fence_wait_empty(ring);
  		if (r) {
  			/* delay GPU reset to resume */
-			amdgpu_fence_driver_force_completion(adev);
+			amdgpu_fence_driver_force_completion(ring);
  		}
 
  		/* disable the interrupt */
@@ -571,30 +571,15 @@ void amdgpu_fence_driver_resume(struct amdgpu_device *adev)
  }
 
  /**
- * amdgpu_fence_driver_force_completion - force all fence waiter to complete
+ * amdgpu_fence_driver_force_completion - force signal latest fence of ring
   *
- * @adev: amdgpu device pointer
+ * @ring: fence of the ring to signal
   *
- * In case of GPU reset failure make sure no process keep waiting on fence
- * that will never complete.
   */
-void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev)
+void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring)
  {
-	int i;
-
-	for (i = 0; i < AMDGPU_MAX_RINGS; i++) {
-		struct amdgpu_ring *ring = adev->rings[i];
-		if (!ring || !ring->fence_drv.initialized)
-			continue;
-
-		amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
-	}
-}
-
-void amdgpu_fence_driver_force_completion_ring(struct amdgpu_ring *ring)
-{
-	if (ring)
-		amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+	amdgpu_fence_write(ring, ring->fence_drv.sync_seq);
+	amdgpu_fence_process(ring);
  }
 
  /*
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +0 -0
index b18c2b9..a6b89e3 100644

drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h struct amdgpu_fence_driver { · line 85

Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ring.h +1 -2
@@ -79,8 +79,7 @@ struct amdgpu_fence_driver {
 
  int amdgpu_fence_driver_init(struct amdgpu_device *adev);
  void amdgpu_fence_driver_fini(struct amdgpu_device *adev);
-void amdgpu_fence_driver_force_completion(struct amdgpu_device *adev);
-void amdgpu_fence_driver_force_completion_ring(struct amdgpu_ring *ring);
+void amdgpu_fence_driver_force_completion(struct amdgpu_ring *ring);
 
  int amdgpu_fence_driver_init_ring(struct amdgpu_ring *ring,
  				  unsigned num_hw_submission);

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