All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] drm/sched: Extend locking in drm_sched_entity_select_rq()
@ 2026-11-17  9:37 Philipp Stanner
  2026-11-17  9:46 ` sashiko-bot
  0 siblings, 1 reply; 2+ messages in thread
From: Philipp Stanner @ 2026-11-17  9:37 UTC (permalink / raw)
  To: Matthew Brost, Danilo Krummrich, Philipp Stanner,
	Christian König, Maarten Lankhorst, Maxime Ripard,
	Thomas Zimmermann, David Airlie, Simona Vetter, Tvrtko Ursulin
  Cc: dri-devel, linux-kernel

drm_sched_entity_select_rq() evaluates entity->sched_list locklessly,
which is elsewhere always lock-protected. Another lockless check regards
spsc_queue_count(). spsc_queue is evaluated under lock protection at
many places already and the plan is to replace it with fully locked
infrastructure anyways.

Move the spin_lock() in drm_sched_entity_select_rq() upwards.

Signed-off-by: Philipp Stanner <phasta@kernel.org>
---
 drivers/gpu/drm/scheduler/sched_entity.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)

diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 03e68b4102d5..3856d5e507d7 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -577,22 +577,21 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
 	struct drm_gpu_scheduler *sched;
 	struct drm_sched_rq *rq;
 
+	spin_lock(&entity->lock);
+
 	/* single possible engine and already selected */
 	if (!entity->sched_list)
-		return;
+		goto unlock;
 
 	/* queue non-empty, stay on the same engine */
 	if (spsc_queue_count(&entity->job_queue))
-		return;
+		goto unlock;
 
-	spin_lock(&entity->lock);
 	fence = entity->last_scheduled;
 
 	/* stay on the same engine if the previous job hasn't finished */
-	if (fence && !dma_fence_is_signaled(fence)) {
-		spin_unlock(&entity->lock);
-		return;
-	}
+	if (fence && !dma_fence_is_signaled(fence))
+		goto unlock;
 
 	sched = drm_sched_pick_best(entity->sched_list, entity->num_sched_list);
 	rq = sched ? sched->sched_rq[entity->rq_priority] : NULL;
@@ -604,6 +603,7 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
 	if (entity->num_sched_list == 1)
 		entity->sched_list = NULL;
 
+unlock:
 	spin_unlock(&entity->lock);
 }
 
-- 
2.56.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

* Re: [PATCH] drm/sched: Extend locking in drm_sched_entity_select_rq()
  2026-11-17  9:37 [PATCH] drm/sched: Extend locking in drm_sched_entity_select_rq() Philipp Stanner
@ 2026-11-17  9:46 ` sashiko-bot
  0 siblings, 0 replies; 2+ messages in thread
From: sashiko-bot @ 2026-11-17  9:46 UTC (permalink / raw)
  To: Philipp Stanner; +Cc: dri-devel

Thank you for your contribution! Sashiko AI review found no new issues
in this patch, but noticed 1 related pre-existing issue it does not
introduce:
- [Critical] drm/sched: out-of-bounds array read in
  drm_sched_entity_select_rq() [1]
--

commit fd8a3dbada3c5189d04991dce47c3b6554b565d5
Author: Philipp Stanner <phasta@kernel.org>

drm/sched: Extend locking in drm_sched_entity_select_rq()

This commit moves spin_lock(&entity->lock) to the start of
drm_sched_entity_select_rq() so that entity->sched_list and
spsc_queue_count(&entity->job_queue) are evaluated under lock protection.

> diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
> index 03e68b4102d50..3856d5e507d7f 100644
> --- a/drivers/gpu/drm/scheduler/sched_entity.c
> +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> @@ -577,22 +577,21 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
>  	struct drm_gpu_scheduler *sched;
>  	struct drm_sched_rq *rq;
>  
> +	spin_lock(&entity->lock);
> +
>  	/* single possible engine and already selected */
>  	if (!entity->sched_list)
> -		return;
> +		goto unlock;
>  
>  	/* queue non-empty, stay on the same engine */
>  	if (spsc_queue_count(&entity->job_queue))
> -		return;
> +		goto unlock;
>  
> -	spin_lock(&entity->lock);
>  	fence = entity->last_scheduled;
>  
>  	/* stay on the same engine if the previous job hasn't finished */
> -	if (fence && !dma_fence_is_signaled(fence)) {
> -		spin_unlock(&entity->lock);
> -		return;
> -	}
> +	if (fence && !dma_fence_is_signaled(fence))
> +		goto unlock;
>  
>  	sched = drm_sched_pick_best(entity->sched_list, entity->num_sched_list);
>  	rq = sched ? sched->sched_rq[entity->rq_priority] : NULL;

[Severity: Critical]
This problem wasn't introduced by this patch, but can
sched->sched_rq[entity->rq_priority] read out of bounds here [1]?

drivers/gpu/drm/scheduler/sched_entity.c:drm_sched_entity_init() {
	...
	if (!sched_list[0]->sched_rq) {
		...
		pr_warn("%s: called with uninitialized scheduler\n", __func__);
	} else if (priority >= sched_list[0]->num_user_rqs) {
		...
		priority = max_t(s32,
				 (s32)sched_list[0]->num_user_rqs - 1,
				 (s32)DRM_SCHED_PRIORITY_KERNEL);
	}

	entity->priority = priority;
	entity->rq_priority = drm_sched_policy == DRM_SCHED_POLICY_FAIR ?
			      DRM_SCHED_PRIORITY_KERNEL : priority;
	...
}

When drm_sched_entity_init() is called with an uninitialized scheduler
(!sched_list[0]->sched_rq), it warns and skips clamping priority against
num_user_rqs, storing the unvalidated value into entity->priority and
entity->rq_priority.  Later, drm_sched_entity_select_rq() indexes
sched->sched_rq[entity->rq_priority] and drm_sched_entity_update_vruntime()
indexes vruntime_shift[entity->priority] out of bounds, which can crash the
kernel.

[1] https://sashiko.dev/#/bug/linux-588f726e-4d3a-4204-b39b-8ff2cd63a19c

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261117093708.647488-2-phasta@kernel.org?part=1

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-17  9:46 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-17  9:37 [PATCH] drm/sched: Extend locking in drm_sched_entity_select_rq() Philipp Stanner
2026-11-17  9:46 ` sashiko-bot

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.