[RFC PATCH] drm/amdgpu: allocate entities on demand

Christian König christian.koenig at amd.com
Tue Dec 3 17:33:39 UTC 2019 

Am 03.12.19 um 16:02 schrieb Nirmoy:
> Hi Christian,
>
> On 12/2/19 3:59 PM, Christian König wrote:
>> Am 02.12.19 um 15:43 schrieb Nirmoy:
>>>
>>> Do you mean something like
>>>
>>> diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
>>> index 684692a8ed76..ac67f8f098fa 100644
>>> --- a/include/drm/gpu_scheduler.h
>>> +++ b/include/drm/gpu_scheduler.h
>>> @@ -81,7 +81,7 @@ enum drm_sched_priority {
>>>  struct drm_sched_entity {
>>>         struct list_head                list;
>>>         struct drm_sched_rq             *rq;
>>> -       struct drm_sched_rq             **rq_list;
>>> +      struct drm_gpu_scheduler        **sched;
>>>         unsigned int                    num_rq_list;
>>>         spinlock_t                      rq_lock;
>>
>> Yes, exactly. Problem is that I'm not 100% sure if that really works 
>> with all users of the rq_list.
>
> currently rq_list users does two main tasks.
>
> 1  change rq priority for a context on user requests
>
> 2  helps drm scheduler to find rq  with least load.
>
> Can you please check the bellow diff it doesn't really work because I 
> get some kernel panic. But do you think
>
> it is matching your idea ?

Yes, that looks exactly like what I had in mind.

Christian.

>
> test at install:~/linux> git diff 
> drivers/gpu/drm/scheduler/sched_entity.c |tee
> diff --git a/drivers/gpu/drm/scheduler/sched_entity.c 
> b/drivers/gpu/drm/scheduler/sched_entity.c
> index 1a5153197fe9..0bbd8ddd6c83 100644
> --- a/drivers/gpu/drm/scheduler/sched_entity.c
> +++ b/drivers/gpu/drm/scheduler/sched_entity.c
> @@ -37,9 +37,9 @@
>   * submit to HW ring.
>   *
>   * @entity: scheduler entity to init
> - * @rq_list: the list of run queue on which jobs from this
> + * @sched_list: the list of drm scheds on which jobs from this
>   *           entity can be submitted
> - * @num_rq_list: number of run queue in rq_list
> + * @num_sched_list: number of drm sched in sched_list
>   * @guilty: atomic_t set to 1 when a job on this queue
>   *          is found to be guilty causing a timeout
>   *
> @@ -49,30 +49,24 @@
>   * Returns 0 on success or a negative error code on failure.
>   */
>  int drm_sched_entity_init(struct drm_sched_entity *entity,
> -              struct drm_sched_rq **rq_list,
> -              unsigned int num_rq_list,
> -              atomic_t *guilty)
> +              struct drm_gpu_scheduler **sched_list,
> +              unsigned int num_sched_list,
> +              atomic_t *guilty, enum drm_sched_priority priority)
>  {
> -    int i;
>
> -    if (!(entity && rq_list && (num_rq_list == 0 || rq_list[0])))
> +    if (!(entity && sched_list && (num_sched_list == 0 || 
> sched_list[0])))
>          return -EINVAL;
>
>      memset(entity, 0, sizeof(struct drm_sched_entity));
>      INIT_LIST_HEAD(&entity->list);
>      entity->rq = NULL;
>      entity->guilty = guilty;
> -    entity->num_rq_list = num_rq_list;
> -    entity->rq_list = kcalloc(num_rq_list, sizeof(struct drm_sched_rq 
> *),
> -                GFP_KERNEL);
> -    if (!entity->rq_list)
> -        return -ENOMEM;
> -
> -    for (i = 0; i < num_rq_list; ++i)
> -        entity->rq_list[i] = rq_list[i];
> +    entity->num_sched_list = num_sched_list;
> +    entity->sched_list =  sched_list
> +    entity->priority = priority;
>
> -    if (num_rq_list)
> -        entity->rq = rq_list[0];
> +    if (num_sched_list)
> +        entity->rq = &entity->sched_list[0]->sched_rq[entity->priority];
>
>      entity->last_scheduled = NULL;
>
> @@ -136,10 +130,10 @@ drm_sched_entity_get_free_sched(struct 
> drm_sched_entity *entity)
>      unsigned int min_jobs = UINT_MAX, num_jobs;
>      int i;
>
> -    for (i = 0; i < entity->num_rq_list; ++i) {
> -        struct drm_gpu_scheduler *sched = entity->rq_list[i]->sched;
> +    for (i = 0; i < entity->num_sched_list; ++i) {
> +        struct drm_gpu_scheduler *sched = entity->sched_list[i];
>
> -        if (!entity->rq_list[i]->sched->ready) {
> +        if (!entity->sched_list[i]->ready) {
>              DRM_WARN("sched%s is not ready, skipping", sched->name);
>              continue;
>          }
> @@ -147,7 +141,7 @@ drm_sched_entity_get_free_sched(struct 
> drm_sched_entity *entity)
>          num_jobs = atomic_read(&sched->num_jobs);
>          if (num_jobs < min_jobs) {
>              min_jobs = num_jobs;
> -            rq = entity->rq_list[i];
> +            rq = &entity->sched_list[i]->sched_rq[entity->priority];
>          }
>      }
>
> @@ -304,7 +298,6 @@ void drm_sched_entity_fini(struct drm_sched_entity 
> *entity)
>
>      dma_fence_put(entity->last_scheduled);
>      entity->last_scheduled = NULL;
> -    kfree(entity->rq_list);
>  }
>  EXPORT_SYMBOL(drm_sched_entity_fini);
>
> @@ -372,8 +365,9 @@ void drm_sched_entity_set_priority(struct 
> drm_sched_entity *entity,
>      unsigned int i;
>
>      spin_lock(&entity->rq_lock);
> -
> -    for (i = 0; i < entity->num_rq_list; ++i)
> +//TODO
> +/*
> +    for (i = 0; i < entity->num_sched_list; ++i)
>  drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
>
>      if (entity->rq) {
> @@ -381,7 +375,7 @@ void drm_sched_entity_set_priority(struct 
> drm_sched_entity *entity,
>          drm_sched_entity_set_rq_priority(&entity->rq, priority);
>          drm_sched_rq_add_entity(entity->rq, entity);
>      }
> -
> +*/
>      spin_unlock(&entity->rq_lock);
>  }
>  EXPORT_SYMBOL(drm_sched_entity_set_priority);
> @@ -486,7 +480,7 @@ 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) || entity->num_rq_list 
> <= 1)
> +    if (spsc_queue_count(&entity->job_queue) || 
> entity->num_sched_list <= 1)
>          return;
>
>      fence = READ_ONCE(entity->last_scheduled);

More information about the amd-gfx mailing list