Re: [PATCH 4/5] drm/sched: Re-group and rename the entity run-queue lock

Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxx> · Tue, 15 Oct 2024 14:16:19 +0100

On 15/10/2024 12:56, Philipp Stanner wrote:
On Mon, 2024-10-14 at 11:46 +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxx>

Christian suggested to rename the lock and improve the documentation

Let's move it to Annotators:
Suggested-by: Christian König <christian.koenig@xxxxxxx>

Ack.

(Otherwise some time in the future a Christian Kaiser might start
working on the scheduler on steal the praise ^^)

of
what it protects.

So without Christian's name here I'd phrase it as:
"When writing to a drm_sched_entity's run-queue, writers are protected
through the lock drm_sched_entity.rq_lock. This naming, however,
frequently collides with the separate internal lock of struct
drm_sched_rq, resulting in uses like this:

	spin_lock(&entity->rq_lock);
	spin_lock(&entity->rq->lock);

Rename drm_sched_entity.rq_lock to improve readability. While at it,
re-order that struct's members to make it more obvious what the lock
protects.

Will copy&paste - thanks for typing it out.

And to also re-order the structure members so all
protected by the lock are together in a block.



Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxx>
Cc: Christian König <christian.koenig@xxxxxxx>
Cc: Alex Deucher <alexander.deucher@xxxxxxx>
Cc: Luben Tuikov <ltuikov89@xxxxxxxxx>
Cc: Matthew Brost <matthew.brost@xxxxxxxxx>
Cc: Philipp Stanner <pstanner@xxxxxxxxxx>
Reviewed-by: Christian König <christian.koenig@xxxxxxx>
---
  drivers/gpu/drm/scheduler/sched_entity.c | 28 ++++++++++++----------
--
  drivers/gpu/drm/scheduler/sched_main.c   |  2 +-
  include/drm/gpu_scheduler.h              | 15 +++++++------
  3 files changed, 23 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/scheduler/sched_entity.c
b/drivers/gpu/drm/scheduler/sched_entity.c
index b72cba292839..c013c2b49aa5 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -105,7 +105,7 @@ int drm_sched_entity_init(struct drm_sched_entity
*entity,
  	/* We start in an idle state. */
  	complete_all(&entity->entity_idle);
  
-	spin_lock_init(&entity->rq_lock);
+	spin_lock_init(&entity->lock);
  	spsc_queue_init(&entity->job_queue);
  
  	atomic_set(&entity->fence_seq, 0);
@@ -133,10 +133,10 @@ void drm_sched_entity_modify_sched(struct
drm_sched_entity *entity,
  {
  	WARN_ON(!num_sched_list || !sched_list);
  
-	spin_lock(&entity->rq_lock);
+	spin_lock(&entity->lock);
  	entity->sched_list = sched_list;
  	entity->num_sched_list = num_sched_list;
-	spin_unlock(&entity->rq_lock);
+	spin_unlock(&entity->lock);
  }
  EXPORT_SYMBOL(drm_sched_entity_modify_sched);
  
@@ -244,10 +244,10 @@ static void drm_sched_entity_kill(struct
drm_sched_entity *entity)
  	if (!entity->rq)
  		return;
  
-	spin_lock(&entity->rq_lock);
+	spin_lock(&entity->lock);
  	entity->stopped = true;
  	drm_sched_rq_remove_entity(entity->rq, entity);
-	spin_unlock(&entity->rq_lock);
+	spin_unlock(&entity->lock);
  
  	/* Make sure this entity is not used by the scheduler at the
moment */
  	wait_for_completion(&entity->entity_idle);
@@ -396,9 +396,9 @@ static void drm_sched_entity_wakeup(struct
dma_fence *f,
  void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
  				   enum drm_sched_priority priority)
  {
-	spin_lock(&entity->rq_lock);
+	spin_lock(&entity->lock);
  	entity->priority = priority;
-	spin_unlock(&entity->rq_lock);
+	spin_unlock(&entity->lock);
  }
  EXPORT_SYMBOL(drm_sched_entity_set_priority);
  
@@ -515,10 +515,10 @@ struct drm_sched_job
*drm_sched_entity_pop_job(struct drm_sched_entity *entity)
  
  		next = to_drm_sched_job(spsc_queue_peek(&entity-
job_queue));
  		if (next) {
-			spin_lock(&entity->rq_lock);
+			spin_lock(&entity->lock);
  			drm_sched_rq_update_fifo_locked(entity,
  							next-
submit_ts);
-			spin_unlock(&entity->rq_lock);
+			spin_unlock(&entity->lock);
  		}
  	}
  
@@ -559,14 +559,14 @@ void drm_sched_entity_select_rq(struct
drm_sched_entity *entity)
  	if (fence && !dma_fence_is_signaled(fence))
  		return;
  
-	spin_lock(&entity->rq_lock);
+	spin_lock(&entity->lock);
  	sched = drm_sched_pick_best(entity->sched_list, entity-
num_sched_list);
  	rq = sched ? sched->sched_rq[entity->priority] : NULL;
  	if (rq != entity->rq) {
  		drm_sched_rq_remove_entity(entity->rq, entity);
  		entity->rq = rq;
  	}
-	spin_unlock(&entity->rq_lock);
+	spin_unlock(&entity->lock);
  
  	if (entity->num_sched_list == 1)
  		entity->sched_list = NULL;
@@ -605,9 +605,9 @@ void drm_sched_entity_push_job(struct
drm_sched_job *sched_job)
  		struct drm_sched_rq *rq;
  
  		/* Add the entity to the run queue */
-		spin_lock(&entity->rq_lock);
+		spin_lock(&entity->lock);
  		if (entity->stopped) {
-			spin_unlock(&entity->rq_lock);
+			spin_unlock(&entity->lock);
  
  			DRM_ERROR("Trying to push to a killed
entity\n");
  			return;
@@ -621,7 +621,7 @@ void drm_sched_entity_push_job(struct
drm_sched_job *sched_job)
  		if (drm_sched_policy == DRM_SCHED_POLICY_FIFO)
  			drm_sched_rq_update_fifo_locked(entity,
submit_ts);
  
-		spin_unlock(&entity->rq_lock);
+		spin_unlock(&entity->lock);
  
  		drm_sched_wakeup(sched);
  	}
diff --git a/drivers/gpu/drm/scheduler/sched_main.c
b/drivers/gpu/drm/scheduler/sched_main.c
index 07ee386b8e4b..2670bf9f34b2 100644
--- a/drivers/gpu/drm/scheduler/sched_main.c
+++ b/drivers/gpu/drm/scheduler/sched_main.c
@@ -176,7 +176,7 @@ void drm_sched_rq_update_fifo_locked(struct
drm_sched_entity *entity, ktime_t ts
  	 * for entity from within concurrent
drm_sched_entity_select_rq and the
  	 * other to update the rb tree structure.
  	 */
-	lockdep_assert_held(&entity->rq_lock);
+	lockdep_assert_held(&entity->lock);
  
  	spin_lock(&entity->rq->lock);
  
diff --git a/include/drm/gpu_scheduler.h
b/include/drm/gpu_scheduler.h
index b6d095074c19..683fff8939e4 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -96,6 +96,14 @@ struct drm_sched_entity {
  	 */
  	struct list_head		list;
  

Uh, btw, while reviewing, I just saw that we still have that FIXME
further up:

	/**
	 * @rq:
	 *
	 * Runqueue on which this entity is currently scheduled.
	 *
	 * FIXME: Locking is very unclear for this. Writers are protected by
	 * @rq_lock, but readers are generally lockless and seem to just race
	 * with not even a READ_ONCE.
	 */
	struct drm_sched_rq		*rq;

At the very least, rq_lock should be renamed here, too. AFAICS the

Good catch!

series doesn't solve the FIXME, so we keep it, agreed?

Yep.

Regards,

Tvrtko

+	/**
+	 * @lock:
+	 *
+	 * Lock protecting the run-queue (@rq) to which this entity
belongs,
+	 * @priority and the list of schedulers (@sched_list,
@num_sched_list).
+	 */
+	spinlock_t			lock;
+
  	/**
  	 * @rq:
  	 *
@@ -140,13 +148,6 @@ struct drm_sched_entity {
  	 */

I think this comment here above also uses the term "rq_lock". While
you're fixing it, maybe also do a quick grep for "rq_lock" in case I
overlooked it somewhere else. I stopped drinking coffee today, so...


Thx,
P.

  	enum drm_sched_priority         priority;
  
-	/**
-	 * @rq_lock:
-	 *
-	 * Lock to modify the runqueue to which this entity belongs.
-	 */
-	spinlock_t			rq_lock;
-
  	/**
  	 * @job_queue: the list of jobs of this entity.
  	 */







