Skip to contentSkip to conversation

drm/sched: Add bloody spinlocks

PATCH New Philipp Stanner 1 participant 3 messages Oct 30, 2025

Subsystems (4)
Stable branch DRM drivers DRM misc DRM GPU scheduler
lore
Share & tools
lore

Philipp Stanner

author maintainer
#0 lore raw

cover drm/sched: Add bloody spinlocks

The scheduler wouldn't be the scheduler if there weren't functions like drm_sched_job_arm() being called probably a few thousand times per second without taking proper locks, just racing wildly.

This series adds the proper locks to drm_sched_job_init() and drm_sched_job_arm().

Philipp Stanner (2):
drm/sched: Use proper locks in drm_sched_job_arm() drm/sched: Use proper locks for drm_sched_job_init()

 drivers/gpu/drm/scheduler/sched_main.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)
Signature · 1 line
2.49.0
author maintainer
#1 lore raw

PATCH 1/2 drm/sched: Use proper locks in drm_sched_job_arm()

drm_sched_job_arm() is just racing when dereferencing entity and runqueue.

Add the proper spinlocks.

Cc [email protected] # v5.16+
Fixes dbe48d030b28 ("drm/sched: Split drm_sched_job_init")
Signed-off-by Philipp Stanner <[email protected]>
---
 drivers/gpu/drm/scheduler/sched_main.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_main.c +7 -1
index c39f0245e3a9..7f938f491b6f 100644
@@ -859,10 +859,16 @@ void drm_sched_job_arm(struct drm_sched_job *job)
 
 	BUG_ON(!entity);
 	drm_sched_entity_select_rq(entity);
+
+	spin_lock(&entity->lock);
+	spin_lock(&entity->rq->lock);
 	sched = entity->rq->sched;
+	spin_unlock(&entity->rq->lock);
+
+	job->s_priority = entity->priority;
+	spin_unlock(&entity->lock);
 
 	job->sched = sched;
-	job->s_priority = entity->priority;
 
 	drm_sched_fence_init(job->s_fence, job->entity);
 }
Signature · 1 line
2.49.0
author maintainer
#2 lore raw

PATCH 2/2 drm/sched: Use proper locks for drm_sched_job_init()

In reply to Philipp Stanner

drm_sched_job_init() is just racing when checking an entity's runqueue, without taking the proper spinlock.

Add the lock.

Cc [email protected] # 6.7+
Fixes 56e449603f0a ("drm/sched: Convert the GPU scheduler to variable number of run-queues")
Signed-off-by Philipp Stanner <[email protected]>
---
 drivers/gpu/drm/scheduler/sched_main.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/sched_main.c +6 -1
index 7f938f491b6f..30028054385f 100644
@@ -799,7 +799,12 @@ int drm_sched_job_init(struct drm_sched_job *job,
 		       u32 credits, void *owner,
 		       uint64_t drm_client_id)
 {
-	if (!entity->rq) {
+	struct drm_sched_rq *rq;
+
+	spin_lock(&entity->lock);
+	rq = entity->rq;
+	spin_unlock(&entity->lock);
+	if (!rq) {
 		/* This will most likely be followed by missing frames
 		 * or worse--a blank screen--leave a trail in the
 		 * logs, so this can be debugged easier.
Signature · 1 line
2.49.0