Re: [PATCH] drm/scheduler: change entities rq even earlier

On Wed, Aug 8, 2018 at 4:44 PM Christian König < ckoenig.leichtzumerken@gmail.com> wrote:

Looks like for correct debugging we need to know the scheduler even earlier. So move picking a rq for an entity into job creation.

Signed-off-by: Christian König christian.koenig@amd.com

drivers/gpu/drm/scheduler/gpu_scheduler.c | 50 ++++++++++++++++++++----------- drivers/gpu/drm/scheduler/sched_fence.c | 2 +- 2 files changed, 33 insertions(+), 19 deletions(-)

diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c b/drivers/gpu/drm/scheduler/gpu_scheduler.c index bd7883d1b964..bb2bd4c07e85 100644 --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c @@ -552,6 +552,34 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity) return sched_job; }

+/**

• • drm_sched_entity_select_rq - select a new rq for the entity
• • @entity: scheduler entity
• • Check all prerequisites and select a new rq for the entity for load
• • balancing.
• */

+static void drm_sched_entity_select_rq(struct drm_sched_entity *entity) +{

•   struct dma_fence *fence;
  

•   struct drm_sched_rq *rq;
  

•   if (!spsc_queue_count(&entity->job_queue) == 0 ||
  

•       entity->num_rq_list <= 1)
  

•           return;
  

•   fence = READ_ONCE(entity->last_scheduled);
  

•   if (fence && !dma_fence_is_signaled(fence))
  

•           return;
  

•   rq = drm_sched_entity_get_free_sched(entity);
  
  

We can add something like this here: if (rq == entity->rq) return;

to avoid redundant reschedules. But then again this is a slight improvement over the original case so we might as well move it to a different patch.

With or without this change the patch is Reviewed-by: Nayan Deshmukh < nayan26deshmukh@gmail.com>

•   spin_lock(&entity->rq_lock);
  

•   drm_sched_rq_remove_entity(entity->rq, entity);
  

•   entity->rq = rq;
  

•   spin_unlock(&entity->rq_lock);
  

+}

/**

• drm_sched_entity_push_job - Submit a job to the entity's job queue

@@ -567,25 +595,8 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity) void drm_sched_entity_push_job(struct drm_sched_job *sched_job, struct drm_sched_entity *entity) {

•   struct drm_sched_rq *rq = entity->rq;
    bool first;
  
  

•   first = spsc_queue_count(&entity->job_queue) == 0;
  

•   if (first && (entity->num_rq_list > 1)) {
  

•           struct dma_fence *fence;
  

•           fence = READ_ONCE(entity->last_scheduled);
  

•           if (fence == NULL || dma_fence_is_signaled(fence)) {
  

•                   rq = drm_sched_entity_get_free_sched(entity);
  

•                   spin_lock(&entity->rq_lock);
  

•                   drm_sched_rq_remove_entity(entity->rq, entity);
  

•                   entity->rq = rq;
  

•                   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);
  

@@ -790,7 +801,10 @@ int drm_sched_job_init(struct drm_sched_job *job, struct drm_sched_entity *entity, void *owner) {

•   struct drm_gpu_scheduler *sched = entity->rq->sched;
  

•   struct drm_gpu_scheduler *sched;
  

•   drm_sched_entity_select_rq(entity);
  

•   sched = entity->rq->sched;
  
    job->sched = sched;
    job->entity = entity;
  

diff --git a/drivers/gpu/drm/scheduler/sched_fence.c b/drivers/gpu/drm/scheduler/sched_fence.c index 6dab18d288d7..4029312fdd81 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 = NULL;
  

•   fence->sched = entity->rq->sched;
    spin_lock_init(&fence->lock);
  
    seq = atomic_inc_return(&entity->fence_seq);
  

-- 2.14.1