diff options
Diffstat (limited to 'drivers/gpu/drm/scheduler/gpu_scheduler.c')
| -rw-r--r-- | drivers/gpu/drm/scheduler/gpu_scheduler.c | 91 |
1 files changed, 71 insertions, 20 deletions
diff --git a/drivers/gpu/drm/scheduler/gpu_scheduler.c b/drivers/gpu/drm/scheduler/gpu_scheduler.c index 0d95888ccc3e..df1578d6f42e 100644 --- a/drivers/gpu/drm/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/scheduler/gpu_scheduler.c @@ -30,7 +30,7 @@ #include <drm/spsc_queue.h> #define CREATE_TRACE_POINTS -#include <drm/gpu_scheduler_trace.h> +#include "gpu_scheduler_trace.h" #define to_drm_sched_job(sched_job) \ container_of((sched_job), struct drm_sched_job, queue_node) @@ -117,15 +117,15 @@ drm_sched_rq_select_entity(struct drm_sched_rq *rq) * @sched The pointer to the scheduler * @entity The pointer to a valid drm_sched_entity * @rq The run queue this entity belongs - * @kernel If this is an entity for the kernel - * @jobs The max number of jobs in the job queue + * @guilty atomic_t set to 1 when a job on this queue + * is found to be guilty causing a timeout * * return 0 if succeed. negative error code on failure */ int drm_sched_entity_init(struct drm_gpu_scheduler *sched, struct drm_sched_entity *entity, struct drm_sched_rq *rq, - uint32_t jobs, atomic_t *guilty) + atomic_t *guilty) { if (!(sched && entity && rq)) return -EINVAL; @@ -135,9 +135,10 @@ int drm_sched_entity_init(struct drm_gpu_scheduler *sched, entity->rq = rq; entity->sched = sched; entity->guilty = guilty; + entity->fini_status = 0; + entity->last_scheduled = NULL; spin_lock_init(&entity->rq_lock); - spin_lock_init(&entity->queue_lock); spsc_queue_init(&entity->job_queue); atomic_set(&entity->fence_seq, 0); @@ -196,19 +197,30 @@ static bool drm_sched_entity_is_ready(struct drm_sched_entity *entity) return true; } +static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f, + struct dma_fence_cb *cb) +{ + struct drm_sched_job *job = container_of(cb, struct drm_sched_job, + finish_cb); + drm_sched_fence_finished(job->s_fence); + WARN_ON(job->s_fence->parent); + dma_fence_put(&job->s_fence->finished); + job->sched->ops->free_job(job); +} + + /** * Destroy a context entity * * @sched Pointer to scheduler instance * @entity The pointer to a valid scheduler entity * - * Cleanup and free the allocated resources. + * Splitting drm_sched_entity_fini() into two functions, The first one is does the waiting, + * removes the entity from the runqueue and returns an error when the process was killed. */ -void drm_sched_entity_fini(struct drm_gpu_scheduler *sched, +void drm_sched_entity_do_release(struct drm_gpu_scheduler *sched, struct drm_sched_entity *entity) { - int r; - if (!drm_sched_entity_is_initialized(sched, entity)) return; /** @@ -216,13 +228,28 @@ void drm_sched_entity_fini(struct drm_gpu_scheduler *sched, * queued IBs or discard them on SIGKILL */ if ((current->flags & PF_SIGNALED) && current->exit_code == SIGKILL) - r = -ERESTARTSYS; + entity->fini_status = -ERESTARTSYS; else - r = wait_event_killable(sched->job_scheduled, + entity->fini_status = wait_event_killable(sched->job_scheduled, drm_sched_entity_is_idle(entity)); drm_sched_entity_set_rq(entity, NULL); - if (r) { +} +EXPORT_SYMBOL(drm_sched_entity_do_release); + +/** + * Destroy a context entity + * + * @sched Pointer to scheduler instance + * @entity The pointer to a valid scheduler entity + * + * The second one then goes over the entity and signals all jobs with an error code. + */ +void drm_sched_entity_cleanup(struct drm_gpu_scheduler *sched, + struct drm_sched_entity *entity) +{ + if (entity->fini_status) { struct drm_sched_job *job; + int r; /* Park the kernel for a moment to make sure it isn't processing * our enity. @@ -240,12 +267,25 @@ void drm_sched_entity_fini(struct drm_gpu_scheduler *sched, struct drm_sched_fence *s_fence = job->s_fence; drm_sched_fence_scheduled(s_fence); dma_fence_set_error(&s_fence->finished, -ESRCH); - drm_sched_fence_finished(s_fence); - WARN_ON(s_fence->parent); - dma_fence_put(&s_fence->finished); - sched->ops->free_job(job); + r = dma_fence_add_callback(entity->last_scheduled, &job->finish_cb, + drm_sched_entity_kill_jobs_cb); + if (r == -ENOENT) + drm_sched_entity_kill_jobs_cb(NULL, &job->finish_cb); + else if (r) + DRM_ERROR("fence add callback failed (%d)\n", r); } } + + dma_fence_put(entity->last_scheduled); + entity->last_scheduled = NULL; +} +EXPORT_SYMBOL(drm_sched_entity_cleanup); + +void drm_sched_entity_fini(struct drm_gpu_scheduler *sched, + struct drm_sched_entity *entity) +{ + drm_sched_entity_do_release(sched, entity); + drm_sched_entity_cleanup(sched, entity); } EXPORT_SYMBOL(drm_sched_entity_fini); @@ -360,6 +400,9 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity) if (entity->guilty && atomic_read(entity->guilty)) 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); + spsc_queue_pop(&entity->job_queue); return sched_job; } @@ -369,6 +412,10 @@ drm_sched_entity_pop_job(struct drm_sched_entity *entity) * * @sched_job The pointer to job required to submit * + * Note: To guarantee that the order of insertion to queue matches + * the job's fence sequence number this function should be + * called with drm_sched_job_init under common lock. + * * Returns 0 for success, negative error code otherwise. */ void drm_sched_entity_push_job(struct drm_sched_job *sched_job, @@ -379,11 +426,8 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job, trace_drm_sched_job(sched_job, entity); - spin_lock(&entity->queue_lock); first = spsc_queue_push(&entity->job_queue, &sched_job->queue_node); - spin_unlock(&entity->queue_lock); - /* first job wakes up scheduler */ if (first) { /* Add the entity to the run queue */ @@ -529,6 +573,7 @@ void drm_sched_job_recovery(struct drm_gpu_scheduler *sched) spin_unlock(&sched->job_list_lock); fence = sched->ops->run_job(s_job); atomic_inc(&sched->hw_rq_count); + if (fence) { s_fence->parent = dma_fence_get(fence); r = dma_fence_add_callback(fence, &s_fence->cb, @@ -548,13 +593,19 @@ void drm_sched_job_recovery(struct drm_gpu_scheduler *sched) } EXPORT_SYMBOL(drm_sched_job_recovery); -/* init a sched_job with basic field */ +/** + * Init a sched_job with basic field + * + * Note: Refer to drm_sched_entity_push_job documentation + * for locking considerations. + */ int drm_sched_job_init(struct drm_sched_job *job, struct drm_gpu_scheduler *sched, struct drm_sched_entity *entity, void *owner) { job->sched = sched; + job->entity = entity; job->s_priority = entity->rq - sched->sched_rq; job->s_fence = drm_sched_fence_create(entity, owner); if (!job->s_fence) |