Re: [PATCH 1/2] drm/scheduler: bind job earlier to scheduler

Christian König <ckoenig.leichtzumerken@xxxxxxxxx> · Mon, 6 Aug 2018 15:15:01 +0200



    Am 06.08.2018 um 15:11 schrieb Nayan
      Deshmukh:

    
        Hi Christian,

          
              On Mon, Aug 6, 2018 at 5:49 PM Christian
                König <ckoenig.leichtzumerken@xxxxxxxxx>
                wrote:

              
              Update
                job earlier with the scheduler it is supposed to be
                scheduled on.

                
                Otherwise we could incorrectly optimize dependencies
                when moving an

                entity from one scheduler to another.

              
              I don't think change is really required, the old code
                should work fine. Read below for more comments.

                 
                Signed-off-by: Christian König <christian.koenig@xxxxxxx>

                ---

                 drivers/gpu/drm/scheduler/gpu_scheduler.c | 4 ++--

                 drivers/gpu/drm/scheduler/sched_fence.c   | 2 +-

                 2 files changed, 3 insertions(+), 3 deletions(-)

                
                diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c
                b/drivers/gpu/drm/scheduler/gpu_scheduler.c

                index 65078dd3c82c..029863726c99 100644

                --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c

                +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c

                @@ -533,8 +533,6 @@ drm_sched_entity_pop_job(struct
                drm_sched_entity *entity)

                        if (!sched_job)

                                return NULL;

                
                -       sched_job->sched = sched;

                -       sched_job->s_fence->sched = sched;

                        while ((entity->dependency =
                sched->ops->dependency(sched_job, entity))) 

              
              The job should have the correct scheduler before the
                sched->oops->dependency() function is called for
                the first time. Hence I placed the assignment here.

                
              It might also be that I am missing some case so
                please let me know if such a case exists where this
                might lead to wrong optimization.

              
    The problem is that the scheduler fence of job A could be the
    dependency of another job B

    
    Job B then makes an incorrect optimization because it sees the wrong
    scheduler in the scheduler fence of job A.

    
    Regards,

    Christian.

    
              Regards,

              
              Nayan

              
                                if
                (drm_sched_entity_add_dependency_cb(entity))

                                        return NULL;

                @@ -581,6 +579,8 @@ void
                drm_sched_entity_push_job(struct drm_sched_job
                *sched_job,

                                spin_unlock(&entity->rq_lock);

                        }

                
                +       sched_job->sched = entity->rq->sched;

                +       sched_job->s_fence->sched =
                entity->rq->sched;

                        trace_drm_sched_job(sched_job, entity);

                       
                atomic_inc(&entity->rq->sched->num_jobs);

                        WRITE_ONCE(entity->last_user,
                current->group_leader);

                diff --git a/drivers/gpu/drm/scheduler/sched_fence.c
                b/drivers/gpu/drm/scheduler/sched_fence.c

                index 4029312fdd81..6dab18d288d7 100644

                --- a/drivers/gpu/drm/scheduler/sched_fence.c

                +++ b/drivers/gpu/drm/scheduler/sched_fence.c

                @@ -172,7 +172,7 @@ struct drm_sched_fence
                *drm_sched_fence_create(struct drm_sched_entity *entity,

                                return NULL;

                
                        fence->owner = owner;

                -       fence->sched = entity->rq->sched;

                +       fence->sched = NULL;

                        spin_lock_init(&fence->lock);

                
                        seq =
                atomic_inc_return(&entity->fence_seq);

                -- 

                2.14.1

                
_______________________________________________
dri-devel mailing list
dri-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/dri-devel