drm/sched: Barriers are needed for entity->last_scheduled

Commit Message

Daniel Vetter July 8, 2021, 9:54 p.m. UTC

It might be good enough on x86 with just READ_ONCE, but the write side
should then at least be WRITE_ONCE because x86 has total store order.

It's definitely not enough on arm.

Fix this proplery, which means
- explain the need for the barrier in both places
- point at the other side in each comment

Also pull out the !sched_list case as the first check, so that the
code flow is clearer.

While at it sprinkle some comments around because it was very
non-obvious to me what's actually going on here and why.

Note that we really need full barriers here, at first I thought
store-release and load-acquire on ->last_scheduled would be enough,
but we actually requiring ordering between that and the queue state.

v2: Put smp_rmp() in the right place and fix up comment (Andrey)

Signed-off-by: Daniel Vetter <[e-mail]>
Cc: "Christian König" <[e-mail]>
Cc: Steven Price <[e-mail]>
Cc: Daniel Vetter <[e-mail]>
Cc: Andrey Grodzovsky <[e-mail]>
Cc: Lee Jones <[e-mail]>
Cc: Boris Brezillon <[e-mail]>
---
 drivers/gpu/drm/scheduler/sched_entity.c | 27 ++++++++++++++++++++++--
 1 file changed, 25 insertions(+), 2 deletions(-)

Comments

Christian König July 9, 2021, 6:57 a.m. UTC | #1

Am 08.07.21 um 23:54 schrieb Daniel Vetter:
> It might be good enough on x86 with just READ_ONCE, but the write side
> should then at least be WRITE_ONCE because x86 has total store order.
>
> It's definitely not enough on arm.
>
> Fix this proplery, which means
> - explain the need for the barrier in both places
> - point at the other side in each comment
>
> Also pull out the !sched_list case as the first check, so that the
> code flow is clearer.
>
> While at it sprinkle some comments around because it was very
> non-obvious to me what's actually going on here and why.
>
> Note that we really need full barriers here, at first I thought
> store-release and load-acquire on ->last_scheduled would be enough,
> but we actually requiring ordering between that and the queue state.
>
> v2: Put smp_rmp() in the right place and fix up comment (Andrey)
>
> Signed-off-by: Daniel Vetter <[e-mail]>
> Cc: "Christian König" <[e-mail]>
> Cc: Steven Price <[e-mail]>
> Cc: Daniel Vetter <[e-mail]>
> Cc: Andrey Grodzovsky <[e-mail]>
> Cc: Lee Jones <[e-mail]>
> Cc: Boris Brezillon <[e-mail]>
> ---
>   drivers/gpu/drm/scheduler/sched_entity.c | 27 ++++++++++++++++++++++--
>   1 file changed, 25 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
> index 64d398166644..6366006c0fcf 100644
> --- a/drivers/gpu/drm/scheduler/sched_entity.c
> +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> @@ -439,8 +439,16 @@ struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity)
>   		dma_fence_set_error(&sched_job->s_fence->finished, -ECANCELED);
>   
>   	dma_fence_put(entity->last_scheduled);
> +
>   	entity->last_scheduled = dma_fence_get(&sched_job->s_fence->finished);
>   
> +	/*
> +	 * If the queue is empty we allow drm_sched_entity_select_rq() to
> +	 * locklessly access ->last_scheduled. This only works if we set the
> +	 * pointer before we dequeue and if we a write barrier here.
> +	 */
> +	smp_wmb();
> +

That whole stuff needs to be inside the spsc queue, not outside.

Otherwise drm_sched_entity_is_idle() won't work either and cause a lot 
of trouble during process tear down.

Christian.

>   	spsc_queue_pop(&entity->job_queue);
>   	return sched_job;
>   }
> @@ -459,10 +467,25 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
>   	struct drm_gpu_scheduler *sched;
>   	struct drm_sched_rq *rq;
>   
> -	if (spsc_queue_count(&entity->job_queue) || !entity->sched_list)
> +	/* single possible engine and already selected */
> +	if (!entity->sched_list)
> +		return;
> +
> +	/* queue non-empty, stay on the same engine */
> +	if (spsc_queue_count(&entity->job_queue))
>   		return;
>   
> -	fence = READ_ONCE(entity->last_scheduled);
> +	/*
> +	 * Only when the queue is empty are we guaranteed that the scheduler
> +	 * thread cannot change ->last_scheduled. To enforce ordering we need
> +	 * a read barrier here. See drm_sched_entity_pop_job() for the other
> +	 * side.
> +	 */
> +	smp_rmb();
> +
> +	fence = entity->last_scheduled;
> +
> +	/* stay on the same engine if the previous job hasn't finished */
>   	if (fence && !dma_fence_is_signaled(fence))
>   		return;
>

Daniel Vetter July 9, 2021, 7:40 a.m. UTC | #2

On Fri, Jul 9, 2021 at 8:58 AM Christian König <[e-mail]> wrote:
>
>
>
> Am 08.07.21 um 23:54 schrieb Daniel Vetter:
> > It might be good enough on x86 with just READ_ONCE, but the write side
> > should then at least be WRITE_ONCE because x86 has total store order.
> >
> > It's definitely not enough on arm.
> >
> > Fix this proplery, which means
> > - explain the need for the barrier in both places
> > - point at the other side in each comment
> >
> > Also pull out the !sched_list case as the first check, so that the
> > code flow is clearer.
> >
> > While at it sprinkle some comments around because it was very
> > non-obvious to me what's actually going on here and why.
> >
> > Note that we really need full barriers here, at first I thought
> > store-release and load-acquire on ->last_scheduled would be enough,
> > but we actually requiring ordering between that and the queue state.
> >
> > v2: Put smp_rmp() in the right place and fix up comment (Andrey)
> >
> > Signed-off-by: Daniel Vetter <[e-mail]>
> > Cc: "Christian König" <[e-mail]>
> > Cc: Steven Price <[e-mail]>
> > Cc: Daniel Vetter <[e-mail]>
> > Cc: Andrey Grodzovsky <[e-mail]>
> > Cc: Lee Jones <[e-mail]>
> > Cc: Boris Brezillon <[e-mail]>
> > ---
> >   drivers/gpu/drm/scheduler/sched_entity.c | 27 ++++++++++++++++++++++--
> >   1 file changed, 25 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
> > index 64d398166644..6366006c0fcf 100644
> > --- a/drivers/gpu/drm/scheduler/sched_entity.c
> > +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> > @@ -439,8 +439,16 @@ struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity)
> >               dma_fence_set_error(&sched_job->s_fence->finished, -ECANCELED);
> >
> >       dma_fence_put(entity->last_scheduled);
> > +
> >       entity->last_scheduled = dma_fence_get(&sched_job->s_fence->finished);
> >
> > +     /*
> > +      * If the queue is empty we allow drm_sched_entity_select_rq() to
> > +      * locklessly access ->last_scheduled. This only works if we set the
> > +      * pointer before we dequeue and if we a write barrier here.
> > +      */
> > +     smp_wmb();
> > +
>
> That whole stuff needs to be inside the spsc queue, not outside.
>
> Otherwise drm_sched_entity_is_idle() won't work either and cause a lot
> of trouble during process tear down.

Nah, that just means you need another 2 comments with their barrier to
explain how things are serialized there against entity->stopped. The
queue only needs to provide store-release and load-acquire barriers
from a functional pov, if you assumie more then that's very strange.
We need barriers in the other direction here (I haven't looked at what
entity_is_idle) needs.

This is why the first rule of lockless algorithms is "don't", and the
second one is to very painstaikingly document every barrier necessary
with
- explanation what it synchronizes and why
- and a pointer to where the other side of the barrier is in the code
(there always has to be one, barrier on one side does nothing)
-Daniel

> Christian.
>
> >       spsc_queue_pop(&entity->job_queue);
> >       return sched_job;
> >   }
> > @@ -459,10 +467,25 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
> >       struct drm_gpu_scheduler *sched;
> >       struct drm_sched_rq *rq;
> >
> > -     if (spsc_queue_count(&entity->job_queue) || !entity->sched_list)
> > +     /* single possible engine and already selected */
> > +     if (!entity->sched_list)
> > +             return;
> > +
> > +     /* queue non-empty, stay on the same engine */
> > +     if (spsc_queue_count(&entity->job_queue))
> >               return;
> >
> > -     fence = READ_ONCE(entity->last_scheduled);
> > +     /*
> > +      * Only when the queue is empty are we guaranteed that the scheduler
> > +      * thread cannot change ->last_scheduled. To enforce ordering we need
> > +      * a read barrier here. See drm_sched_entity_pop_job() for the other
> > +      * side.
> > +      */
> > +     smp_rmb();
> > +
> > +     fence = entity->last_scheduled;
> > +
> > +     /* stay on the same engine if the previous job hasn't finished */
> >       if (fence && !dma_fence_is_signaled(fence))
> >               return;
> >
>

Patch

diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 64d398166644..6366006c0fcf 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -439,8 +439,16 @@  struct drm_sched_job *drm_sched_entity_pop_job(struct drm_sched_entity *entity)
 		dma_fence_set_error(&sched_job->s_fence->finished, -ECANCELED);
 
 	dma_fence_put(entity->last_scheduled);
+
 	entity->last_scheduled = dma_fence_get(&sched_job->s_fence->finished);
 
+	/*
+	 * If the queue is empty we allow drm_sched_entity_select_rq() to
+	 * locklessly access ->last_scheduled. This only works if we set the
+	 * pointer before we dequeue and if we a write barrier here.
+	 */
+	smp_wmb();
+
 	spsc_queue_pop(&entity->job_queue);
 	return sched_job;
 }
@@ -459,10 +467,25 @@  void drm_sched_entity_select_rq(struct drm_sched_entity *entity)
 	struct drm_gpu_scheduler *sched;
 	struct drm_sched_rq *rq;
 
-	if (spsc_queue_count(&entity->job_queue) || !entity->sched_list)
+	/* single possible engine and already selected */
+	if (!entity->sched_list)
+		return;
+
+	/* queue non-empty, stay on the same engine */
+	if (spsc_queue_count(&entity->job_queue))
 		return;
 
-	fence = READ_ONCE(entity->last_scheduled);
+	/*
+	 * Only when the queue is empty are we guaranteed that the scheduler
+	 * thread cannot change ->last_scheduled. To enforce ordering we need
+	 * a read barrier here. See drm_sched_entity_pop_job() for the other
+	 * side.
+	 */
+	smp_rmb();
+
+	fence = entity->last_scheduled;
+
+	/* stay on the same engine if the previous job hasn't finished */
 	if (fence && !dma_fence_is_signaled(fence))
 		return;