Skip to contentSkip to conversation

drm/amdgpu: Do gpu recovery when no job is running

PATCH In review Liu ChengZhe 3 participants 3 messages Sep 9, 2020

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

PATCH 1/1 drm/amdgpu: Do gpu recovery when no job is running

In function flr_work, do gpu recovery when no job is running instead of when some job is running. Because if there is job in list, amdgpu_job_timedout will do the gpu recovery.

---
 drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 3 ++-
 drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +2 -1
index 9c07014d9bd6..f5ce9a9f4cf5 100644
@@ -262,7 +262,8 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 
 	/* Trigger recovery for world switch failure if no TDR */
 	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) || adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
+		&& (!amdgpu_device_has_job_running(adev) ||
+		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
 		amdgpu_device_gpu_recover(adev, NULL);
 }
 
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +1 -1
index 9c23abf9b140..666ed99cc14b 100644
@@ -283,7 +283,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)
 
 	/* Trigger recovery for world switch failure if no TDR */
 	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) ||
+		&& (!amdgpu_device_has_job_running(adev) ||
 		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT ||
 		adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||
 		adev->compute_timeout == MAX_SCHEDULE_TIMEOUT ||
Signature · 6 lines
2.25.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx
maintainer
#1 lore raw

Am 09.09.20 um 10:56 schrieb Liu ChengZhe:

Quoted 3 lines

In function flr_work, do gpu recovery when no job is running instead of when some job is running. Because if there is job in list, amdgpu_job_timedout will do the gpu recovery.

Maybe write something like "Fix the logic by inverting it".

With that the patch is Acked-by: Christian König <[email protected]>

Quoted 26 lines
---
  drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 3 ++-
  drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +-
  2 files changed, 3 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +2 -1
index 9c07014d9bd6..f5ce9a9f4cf5 100644
@@ -262,7 +262,8 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 
  	/* Trigger recovery for world switch failure if no TDR */
  	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) || adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
+		&& (!amdgpu_device_has_job_running(adev) ||
+		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
  		amdgpu_device_gpu_recover(adev, NULL);
  }
 
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +0 -0
index 9c23abf9b140..666ed99cc14b 100644

drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c static void xgpu_nv_mailbox_flr_work(struct work_struct *work) · line 289

Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +1 -1
@@ -283,7 +283,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)
 
  	/* Trigger recovery for world switch failure if no TDR */
  	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) ||
+		&& (!amdgpu_device_has_job_running(adev) ||
  		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT ||
  		adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||
  		adev->compute_timeout == MAX_SCHEDULE_TIMEOUT ||

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

#2 lore raw
Quoted 3 lines · Christian König

Am 09.09.20 um 10:56 schrieb Liu ChengZhe:

Quoted 2 lines

In function flr_work, do gpu recovery when no job is running instead of when some job is running. Because if there is job

                                                           ^a
Quoted 1 line
Quoted 1 line

in list, amdgpu_job_timedout will do the gpu recovery.

^the ^remove "the".

To become: "is a job in the list", and "will do GPU recovery."

Quoted 2 lines

Maybe write something like "Fix the logic by inverting it".

Your commit message is fine, but as Christian pointed out, _add_ a sentence to the commit message, possibly in a new paragraph, simply stating that you're inverting the logic, possibly just cutting and pasting what he wrote above.

Regards,
Luben

Quoted 30 lines

With that the patch is Acked-by: Christian König <[email protected]>

Quoted 27 lines
---
  drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c | 3 ++-
  drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c | 2 +-
  2 files changed, 3 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/mxgpu_ai.c +2 -1
index 9c07014d9bd6..f5ce9a9f4cf5 100644
@@ -262,7 +262,8 @@ static void xgpu_ai_mailbox_flr_work(struct work_struct *work)
 
  	/* Trigger recovery for world switch failure if no TDR */
  	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) || adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
+		&& (!amdgpu_device_has_job_running(adev) ||
+		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT))
  		amdgpu_device_gpu_recover(adev, NULL);
  }
 
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +0 -0
index 9c23abf9b140..666ed99cc14b 100644
@@ -283,7 +283,7 @@ static void xgpu_nv_mailbox_flr_work(struct work_struct *work)
 
Quoted 8 lines

drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c static void xgpu_nv_mailbox_flr_work(struct work_struct *work) · line 289

Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/mxgpu_nv.c +1 -1
  	/* Trigger recovery for world switch failure if no TDR */
  	if (amdgpu_device_should_recover_gpu(adev)
-		&& (amdgpu_device_has_job_running(adev) ||
+		&& (!amdgpu_device_has_job_running(adev) ||
  		adev->sdma_timeout == MAX_SCHEDULE_TIMEOUT ||
  		adev->gfx_timeout == MAX_SCHEDULE_TIMEOUT ||
  		adev->compute_timeout == MAX_SCHEDULE_TIMEOUT ||

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