Skip to contentSkip to conversation

drm/sched: Add warning for removing hack in drm_sched_fini()

PATCH Applied Philipp Stanner 3 participants 4 messages Oct 23, 2025 – Oct 31, 2025

Subsystems (4)
lore
Share & tools
lore

Philipp Stanner

author maintainer
#0 lore raw

PATCH v3 1/1 drm/sched: Add warning for removing hack in drm_sched_fini()

The assembled developers agreed at the X.Org Developers Conference 2025 that the hack added for amdgpu in drm_sched_fini() shall be removed. It shouldn't be needed by amdgpu anymore.

As it's unclear whether all drivers really follow the life time rule of entities having to be torn down before their scheduler, it is reasonable to warn for a while before removing the hack.

Add a warning in drm_sched_fini() that fires if an entity is still active.

---
Changes in v3:
- Add a READ_ONCE() + comment to make the warning slightly less horrible.

Changes in v2:
- Fix broken brackets.

---
 drivers/gpu/drm/scheduler/sched_main.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_main.c +8 -1
index 46119aacb809..31039b08c7b9 100644
@@ -1419,7 +1419,7 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
 		struct drm_sched_rq *rq = sched->sched_rq[i];
 
 		spin_lock(&rq->lock);
-		list_for_each_entry(s_entity, &rq->entities, list)
+		list_for_each_entry(s_entity, &rq->entities, list) {
 			/*
 			 * Prevents reinsertion and marks job_queue as idle,
 			 * it will be removed from the rq in drm_sched_entity_fini()
@@ -1440,8 +1440,15 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
 			 * For now, this remains a potential race in all
 			 * drivers that keep entities alive for longer than
 			 * the scheduler.
+			 *
+			 * The READ_ONCE() is there to make the lockless read
+			 * (warning about the lockless write below) slightly
+			 * less broken...
 			 */
+			if (!READ_ONCE(s_entity->stopped))
+				dev_warn(sched->dev, "Tearing down scheduler with active entities!\n");
 			s_entity->stopped = true;
+		}
 		spin_unlock(&rq->lock);
 		kfree(sched->sched_rq[i]);
 	}
Signature · 1 line
2.49.0
#1 lore raw
Quoted 4 lines · Philipp Stanner

The assembled developers agreed at the X.Org Developers Conference 2025 that the hack added for amdgpu in drm_sched_fini() shall be removed. It shouldn't be needed by amdgpu anymore.

Quoted 8 lines

As it's unclear whether all drivers really follow the life time rule of entities having to be torn down before their scheduler, it is reasonable to warn for a while before removing the hack.

Add a warning in drm_sched_fini() that fires if an entity is still active.

Can someone review this?

At XDC we agreed on removing the hack, but wanted to add a warning print first for a few releases, to really catch if there are no users anymore.

Thx
P.

Quoted 40 lines

---
Changes in v3:
  - Add a READ_ONCE() + comment to make the warning slightly less horrible.

Changes in v2:
  - Fix broken brackets.
---
 drivers/gpu/drm/scheduler/sched_main.c | 9 ++++++++-

 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_main.c +0 -0
index 46119aacb809..31039b08c7b9 100644
@@ -1419,7 +1419,7 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
  		struct drm_sched_rq *rq = sched->sched_rq[i];


  spin_lock(&rq->lock);

drivers/gpu/drm/scheduler/sched_main.c +1 -1
-		list_for_each_entry(s_entity, &rq->entities, list)
+		list_for_each_entry(s_entity, &rq->entities, list) {

  /*
  * Prevents reinsertion and marks job_queue as idle,
  * it will be removed from the rq in drm_sched_entity_fini()

drivers/gpu/drm/scheduler/sched_main.c +7 -0
@@ -1440,8 +1440,15 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
  			 * For now, this remains a potential race in all
  			 * drivers that keep entities alive for longer than
  			 * the scheduler.
+			 *
+			 * The READ_ONCE() is there to make the lockless read
+			 * (warning about the lockless write below) slightly
+			 * less broken...
  			 */
+			if (!READ_ONCE(s_entity->stopped))
+				dev_warn(sched->dev, "Tearing down scheduler with active entities!\n");
  			s_entity->stopped = true;
+		}
  		spin_unlock(&rq->lock);
  		kfree(sched->sched_rq[i]);
  	}
#2 lore raw
In reply to Philipp Stanner

Hi Philipp,

Le 23/10/2025 à 14:34, Philipp Stanner a écrit :

Quoted 39 lines

The assembled developers agreed at the X.Org Developers Conference 2025 that the hack added for amdgpu in drm_sched_fini() shall be removed. It shouldn't be needed by amdgpu anymore.

As it's unclear whether all drivers really follow the life time rule of entities having to be torn down before their scheduler, it is reasonable to warn for a while before removing the hack.

Add a warning in drm_sched_fini() that fires if an entity is still active.

---
Changes in v3:
- Add a READ_ONCE() + comment to make the warning slightly less horrible.

Changes in v2:
- Fix broken brackets.

---
  drivers/gpu/drm/scheduler/sched_main.c | 9 ++++++++-
  1 file changed, 8 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_main.c +1 -1
index 46119aacb809..31039b08c7b9 100644
@@ -1419,7 +1419,7 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
  		struct drm_sched_rq *rq = sched->sched_rq[i];
 
  		spin_lock(&rq->lock);
-		list_for_each_entry(s_entity, &rq->entities, list)
+		list_for_each_entry(s_entity, &rq->entities, list) {
  			/*
  			 * Prevents reinsertion and marks job_queue as idle,
  			 * it will be removed from the rq in drm_sched_entity_fini()
@@ -1440,8 +1440,15 @@ void drm_sched_fini(struct drm_gpu_scheduler *sched)
  			 * For now, this remains a potential race in all
  			 * drivers that keep entities alive for longer than

drivers/gpu/drm/scheduler/sched_main.c void drm_sched_fini(struct drm_gpu_scheduler *sched) · line 1451

Quoted 10 lines
drivers/gpu/drm/scheduler/sched_main.c +7 -0
  			 * the scheduler.
+			 *
+			 * The READ_ONCE() is there to make the lockless read
+			 * (warning about the lockless write below) slightly
+			 * less broken...
  			 */
+			if (!READ_ONCE(s_entity->stopped))
+				dev_warn(sched->dev, "Tearing down scheduler with active entities!\n");
  			s_entity->stopped = true;
+		}

The patch is Acked-by: Pierre-Eric Pelloux-Prayer <[email protected]>

Thanks.

Quoted 3 lines
drivers/gpu/drm/scheduler/sched_main.c +0 -0
  		spin_unlock(&rq->lock);
  		kfree(sched->sched_rq[i]);
  	}
#3 lore raw
Quoted 15 lines · Pierre-Eric Pelloux-Prayer

Hi Philipp,

Le 23/10/2025 à 14:34, Philipp Stanner a écrit :

Quoted 12 lines

The assembled developers agreed at the X.Org Developers Conference 2025 that the hack added for amdgpu in drm_sched_fini() shall be removed. It shouldn't be needed by amdgpu anymore.

As it's unclear whether all drivers really follow the life time rule of entities having to be torn down before their scheduler, it is reasonable to warn for a while before removing the hack.

Add a warning in drm_sched_fini() that fires if an entity is still active.

[…]

Quoted 3 lines

The patch is Acked-by: Pierre-Eric Pelloux-Prayer
<[email protected]>

Pushed to drm-misc-next, thanks.

For the future: b4 / maintainer-tools wasn't able to automatically harvest your Acked-by. Would be helpful if you have the A-b on a single line without line break and without other content in the future

Have a nice weekend,
P.

Quoted 7 lines

Thanks.

Quoted 3 lines

   spin_unlock(&rq->lock);
   kfree(sched->sched_rq[i]);
   }