All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-09 21:53 Nirmoy Das
  2019-12-09 21:53 ` [PATCH 2/4] drm/amdgpu: replace vm_pte's run-queue list with drm gpu scheds list Nirmoy Das
                   ` (2 more replies)
  0 siblings, 3 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-09 21:53 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;
 
@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;
 
-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}
 
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 19ffe00d9072..2b6e35893918 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1957,11 +1957,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index e324bfe6c58f..a1a110f5513d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a94c4faa5af1..5e78db30c722 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2687,6 +2687,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;
 
 	vm->va = RB_ROOT_CACHED;
@@ -2700,14 +2701,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);
 
+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;
 
-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)
 
 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}
 
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;
 
-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }
 
 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..4f9ae5a12090 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;
 
 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL, DRM_SCHED_PRIORITY_NORMAL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;
 
 	init_completion(&entity->entity_idle);
 
-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];
 
-	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;
 
@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}
 
@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)
 
 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);
 
@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }
 
-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;
 
+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}
 
-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }
 
diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;
 
 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;
 
 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}
 
 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;
 
 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);
 
 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 2/4] drm/amdgpu: replace vm_pte's run-queue list with drm gpu scheds list
  2019-12-09 21:53 [PATCH 1/4] drm/scheduler: rework entity creation Nirmoy Das
@ 2019-12-09 21:53 ` Nirmoy Das
  2019-12-09 21:53 ` [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues Nirmoy Das
  2019-12-09 21:53 ` [PATCH 4/4] drm/scheduler: do not keep a copy of sched list Nirmoy Das
  2 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-09 21:53 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

drm_sched_entity_init() takes drm gpu scheduler list instead of
drm_sched_rq list. This makes conversion of drm_sched_rq list
to drm gpu scheduler list unnecessary

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c     | 11 ++++-------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h     |  4 ++--
 drivers/gpu/drm/amd/amdgpu/cik_sdma.c      |  8 +++-----
 drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c     |  8 +++-----
 drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c     |  8 +++-----
 drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c     |  5 ++---
 drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c     |  8 +++-----
 drivers/gpu/drm/amd/amdgpu/si_dma.c        |  8 +++-----
 9 files changed, 24 insertions(+), 38 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index f85007382093..cf4953c4e2cf 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2779,7 +2779,7 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 	adev->mman.buffer_funcs = NULL;
 	adev->mman.buffer_funcs_ring = NULL;
 	adev->vm_manager.vm_pte_funcs = NULL;
-	adev->vm_manager.vm_pte_num_rqs = 0;
+	adev->vm_manager.vm_pte_num_scheds = 0;
 	adev->gmc.gmc_funcs = NULL;
 	adev->fence_context = dma_fence_context_alloc(AMDGPU_MAX_RINGS);
 	bitmap_zero(adev->gfx.pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 5e78db30c722..0e1ed8ef2ce7 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2687,7 +2687,6 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
-	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;
 
 	vm->va = RB_ROOT_CACHED;
@@ -2701,19 +2700,17 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);
 
-	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
-		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
 
 	/* create scheduler entities for page table updates */
 	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
-				  sched_list, adev->vm_manager.vm_pte_num_rqs,
-				  NULL);
+				  adev->vm_manager.vm_pte_scheds,
+				  adev->vm_manager.vm_pte_num_scheds, NULL);
 	if (r)
 		return r;
 
 	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
-				  sched_list, adev->vm_manager.vm_pte_num_rqs,
-				  NULL);
+				  adev->vm_manager.vm_pte_scheds,
+				  adev->vm_manager.vm_pte_num_scheds, NULL);
 	if (r)
 		goto error_free_direct;
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index 76fcf853035c..5eaba8645a43 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -322,8 +322,8 @@ struct amdgpu_vm_manager {
 	u64					vram_base_offset;
 	/* vm pte handling */
 	const struct amdgpu_vm_pte_funcs	*vm_pte_funcs;
-	struct drm_sched_rq			*vm_pte_rqs[AMDGPU_MAX_RINGS];
-	unsigned				vm_pte_num_rqs;
+	struct drm_gpu_scheduler		*vm_pte_scheds[AMDGPU_MAX_RINGS];
+	unsigned				vm_pte_num_scheds;
 	struct amdgpu_ring			*page_fault;
 
 	/* partial resident texture handling */
diff --git a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
index 82cdb8f57bfd..1f22a8d0f7f3 100644
--- a/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
+++ b/drivers/gpu/drm/amd/amdgpu/cik_sdma.c
@@ -1373,16 +1373,14 @@ static const struct amdgpu_vm_pte_funcs cik_sdma_vm_pte_funcs = {
 
 static void cik_sdma_set_vm_pte_funcs(struct amdgpu_device *adev)
 {
-	struct drm_gpu_scheduler *sched;
 	unsigned i;
 
 	adev->vm_manager.vm_pte_funcs = &cik_sdma_vm_pte_funcs;
 	for (i = 0; i < adev->sdma.num_instances; i++) {
-		sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
-			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_scheds[i] =
+			&adev->sdma.instance[i].ring.sched;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version cik_sdma_ip_block =
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
index 89e8c74a40f4..606b621145a1 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v2_4.c
@@ -1261,16 +1261,14 @@ static const struct amdgpu_vm_pte_funcs sdma_v2_4_vm_pte_funcs = {
 
 static void sdma_v2_4_set_vm_pte_funcs(struct amdgpu_device *adev)
 {
-	struct drm_gpu_scheduler *sched;
 	unsigned i;
 
 	adev->vm_manager.vm_pte_funcs = &sdma_v2_4_vm_pte_funcs;
 	for (i = 0; i < adev->sdma.num_instances; i++) {
-		sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
-			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_scheds[i] =
+			&adev->sdma.instance[i].ring.sched;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version sdma_v2_4_ip_block =
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
index 011fd12c41fe..a559573ec8fd 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v3_0.c
@@ -1699,16 +1699,14 @@ static const struct amdgpu_vm_pte_funcs sdma_v3_0_vm_pte_funcs = {
 
 static void sdma_v3_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 {
-	struct drm_gpu_scheduler *sched;
 	unsigned i;
 
 	adev->vm_manager.vm_pte_funcs = &sdma_v3_0_vm_pte_funcs;
 	for (i = 0; i < adev->sdma.num_instances; i++) {
-		sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
-			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_scheds[i] =
+			 &adev->sdma.instance[i].ring.sched;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version sdma_v3_0_ip_block =
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
index 350b2c99fefc..bd9ed33bab43 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v4_0.c
@@ -2411,10 +2411,9 @@ static void sdma_v4_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 			sched = &adev->sdma.instance[i].page.sched;
 		else
 			sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
-			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_scheds[i] = sched;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version sdma_v4_0_ip_block = {
diff --git a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
index 64c53eed7fac..63f667cfe3f6 100644
--- a/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/sdma_v5_0.c
@@ -1723,17 +1723,15 @@ static const struct amdgpu_vm_pte_funcs sdma_v5_0_vm_pte_funcs = {
 
 static void sdma_v5_0_set_vm_pte_funcs(struct amdgpu_device *adev)
 {
-	struct drm_gpu_scheduler *sched;
 	unsigned i;
 
 	if (adev->vm_manager.vm_pte_funcs == NULL) {
 		adev->vm_manager.vm_pte_funcs = &sdma_v5_0_vm_pte_funcs;
 		for (i = 0; i < adev->sdma.num_instances; i++) {
-			sched = &adev->sdma.instance[i].ring.sched;
-			adev->vm_manager.vm_pte_rqs[i] =
-				&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+			adev->vm_manager.vm_pte_scheds[i] =
+				&adev->sdma.instance[i].ring.sched;
 		}
-		adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+		adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 	}
 }
 
diff --git a/drivers/gpu/drm/amd/amdgpu/si_dma.c b/drivers/gpu/drm/amd/amdgpu/si_dma.c
index 122df0732f0c..9ad85eddf9c4 100644
--- a/drivers/gpu/drm/amd/amdgpu/si_dma.c
+++ b/drivers/gpu/drm/amd/amdgpu/si_dma.c
@@ -835,16 +835,14 @@ static const struct amdgpu_vm_pte_funcs si_dma_vm_pte_funcs = {
 
 static void si_dma_set_vm_pte_funcs(struct amdgpu_device *adev)
 {
-	struct drm_gpu_scheduler *sched;
 	unsigned i;
 
 	adev->vm_manager.vm_pte_funcs = &si_dma_vm_pte_funcs;
 	for (i = 0; i < adev->sdma.num_instances; i++) {
-		sched = &adev->sdma.instance[i].ring.sched;
-		adev->vm_manager.vm_pte_rqs[i] =
-			&sched->sched_rq[DRM_SCHED_PRIORITY_KERNEL];
+		adev->vm_manager.vm_pte_scheds[i] =
+			&adev->sdma.instance[i].ring.sched;
 	}
-	adev->vm_manager.vm_pte_num_rqs = adev->sdma.num_instances;
+	adev->vm_manager.vm_pte_num_scheds = adev->sdma.num_instances;
 }
 
 const struct amdgpu_ip_block_version si_dma_ip_block =
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues
  2019-12-09 21:53 [PATCH 1/4] drm/scheduler: rework entity creation Nirmoy Das
  2019-12-09 21:53 ` [PATCH 2/4] drm/amdgpu: replace vm_pte's run-queue list with drm gpu scheds list Nirmoy Das
@ 2019-12-09 21:53 ` Nirmoy Das
  2019-12-10 11:28   ` Christian König
  2019-12-09 21:53 ` [PATCH 4/4] drm/scheduler: do not keep a copy of sched list Nirmoy Das
  2 siblings, 1 reply; 12+ messages in thread
From: Nirmoy Das @ 2019-12-09 21:53 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

This sched list can be passed on to entity creation routine
instead of manually creating such sched list on every context creation.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    | 69 ++++++++--------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 44 ++++++++++++++
 drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h    |  4 ++
 drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h   |  2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h   |  2 +
 drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h    |  9 ++-
 6 files changed, 85 insertions(+), 45 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index 1d6850af9908..c1fc75299b7d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			   struct amdgpu_ctx *ctx)
 {
 	unsigned num_entities = amdgpu_ctx_total_num_entities();
-	unsigned i, j, k;
+	unsigned i, j;
 	int r;
 
 	if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
@@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 	ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
-		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
-		unsigned num_rings = 0;
-		unsigned num_rqs = 0;
+		struct drm_gpu_scheduler **sched_list;
+		struct drm_gpu_scheduler *sched;
+		unsigned num_scheds = 0;
 
 		switch (i) {
 		case AMDGPU_HW_IP_GFX:
-			rings[0] = &adev->gfx.gfx_ring[0];
-			num_rings = 1;
+			sched_list = adev->gfx.gfx_sched_list;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_COMPUTE:
-			for (j = 0; j < adev->gfx.num_compute_rings; ++j)
-				rings[j] = &adev->gfx.compute_ring[j];
-			num_rings = adev->gfx.num_compute_rings;
+			sched_list = adev->gfx.compute_sched_list;
+			num_scheds = adev->gfx.num_compute_rings;
 			break;
 		case AMDGPU_HW_IP_DMA:
-			for (j = 0; j < adev->sdma.num_instances; ++j)
-				rings[j] = &adev->sdma.instance[j].ring;
-			num_rings = adev->sdma.num_instances;
+			sched_list = adev->sdma.sdma_sched_list;
+			num_scheds = adev->sdma.num_instances;
 			break;
 		case AMDGPU_HW_IP_UVD:
-			rings[0] = &adev->uvd.inst[0].ring;
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring.sched;
+			sched_list = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCE:
-			rings[0] = &adev->vce.ring[0];
-			num_rings = 1;
+			sched = &adev->vce.ring[0].sched;
+			sched_list = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_UVD_ENC:
-			rings[0] = &adev->uvd.inst[0].ring_enc[0];
-			num_rings = 1;
+			sched = &adev->uvd.inst[0].ring_enc[0].sched;
+			sched_list = &sched;
+			num_scheds = 1;
 			break;
 		case AMDGPU_HW_IP_VCN_DEC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
-			}
+			sched_list = adev->vcn.vcn_dec_sched_list;
+			num_scheds =  adev->vcn.num_vcn_dec_sched_list;
 			break;
 		case AMDGPU_HW_IP_VCN_ENC:
-			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				for (k = 0; k < adev->vcn.num_enc_rings; ++k)
-					rings[num_rings++] = &adev->vcn.inst[j].ring_enc[k];
-			}
+			sched_list = adev->vcn.vcn_enc_sched_list;
+			num_scheds =  adev->vcn.num_vcn_enc_sched_list;
 			break;
 		case AMDGPU_HW_IP_VCN_JPEG:
-			for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
-				if (adev->vcn.harvest_config & (1 << j))
-					continue;
-				rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
-			}
+			sched_list = adev->jpeg.jpeg_sched_list;
+			num_scheds =  adev->jpeg.num_jpeg_sched_list;
 			break;
 		}
 
-		for (j = 0; j < num_rings; ++j) {
-			if (!rings[j]->adev)
-				continue;
-
-			sched_list[num_rqs++] = &rings[j]->sched;
-		}
-
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
 						  priority, sched_list,
-						  num_rqs, &ctx->guilty);
+						  num_scheds, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
index cf4953c4e2cf..f973b61a26da 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
@@ -2738,6 +2738,48 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
 	return ret;
 }
 
+static void amdgpu_device_init_sched_list(struct amdgpu_device *adev)
+{
+	int i, j;
+
+	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
+		adev->gfx.gfx_sched_list[i] = &adev->gfx.gfx_ring[i].sched;
+		adev->gfx.num_gfx_sched_list++;
+	}
+
+	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
+		adev->gfx.compute_sched_list[i] = &adev->gfx.compute_ring[i].sched;
+		adev->gfx.num_compute_sched_list++;
+	}
+
+	for (i = 0; i < adev->sdma.num_instances; i++) {
+		adev->sdma.sdma_sched_list[i] = &adev->sdma.instance[i].ring.sched;
+		adev->sdma.num_sdma_sched_list++;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		adev->vcn.vcn_dec_sched_list[adev->vcn.num_vcn_dec_sched_list++] =
+			&adev->vcn.inst[i].ring_dec.sched;
+	}
+
+	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		for (j = 0; j < adev->vcn.num_enc_rings; ++j)
+			adev->vcn.vcn_enc_sched_list[adev->vcn.num_vcn_enc_sched_list++] =
+				&adev->vcn.inst[i].ring_enc[j].sched;
+	}
+
+	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
+		if (adev->vcn.harvest_config & (1 << i))
+			continue;
+		adev->jpeg.jpeg_sched_list[adev->jpeg.num_jpeg_sched_list++] =
+			&adev->jpeg.inst[i].ring_dec.sched;
+	}
+}
+
 /**
  * amdgpu_device_init - initialize the driver
  *
@@ -3024,6 +3066,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
 		goto failed;
 	}
 
+	amdgpu_device_init_sched_list(adev);
+
 	adev->accel_working = true;
 
 	amdgpu_vm_check_compute_bug(adev);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
index 0ae0a2715b0d..b2264d4cd510 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
@@ -269,8 +269,12 @@ struct amdgpu_gfx {
 	bool				me_fw_write_wait;
 	bool				cp_fw_write_wait;
 	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
+	struct drm_gpu_scheduler	*gfx_sched_list[AMDGPU_MAX_GFX_RINGS];
+	uint32_t			num_gfx_sched_list;
 	unsigned			num_gfx_rings;
 	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
+	uint32_t			num_compute_sched_list;
+	struct drm_gpu_scheduler	*compute_sched_list[AMDGPU_MAX_COMPUTE_RINGS];
 	unsigned			num_compute_rings;
 	struct amdgpu_irq_src		eop_irq;
 	struct amdgpu_irq_src		priv_reg_irq;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
index 5131a0a1bc8a..5ce6ab05eeac 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
@@ -43,6 +43,8 @@ struct amdgpu_jpeg {
 	uint8_t	num_jpeg_inst;
 	struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
 	struct amdgpu_jpeg_reg internal;
+	struct drm_gpu_scheduler *jpeg_sched_list[AMDGPU_MAX_JPEG_INSTANCES];
+	uint32_t num_jpeg_sched_list;
 	unsigned harvest_config;
 	struct delayed_work idle_work;
 	enum amd_powergating_state cur_state;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
index b3134655789f..0f6cb6768398 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
@@ -52,6 +52,8 @@ struct amdgpu_sdma_instance {
 
 struct amdgpu_sdma {
 	struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
+	struct drm_gpu_scheduler    *sdma_sched_list[AMDGPU_MAX_SDMA_INSTANCES];
+	uint32_t		    num_sdma_sched_list;
 	struct amdgpu_irq_src	trap_irq;
 	struct amdgpu_irq_src	illegal_inst_irq;
 	struct amdgpu_irq_src	ecc_irq;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
index 402a5046b985..ae65cb8f07e8 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
@@ -31,6 +31,7 @@
 #define AMDGPU_VCN_MAX_ENC_RINGS	3
 
 #define AMDGPU_MAX_VCN_INSTANCES	2
+#define AMDGPU_MAX_VCN_ENC_RINGS  AMDGPU_VCN_MAX_ENC_RINGS * AMDGPU_MAX_VCN_INSTANCES
 
 #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0)
 #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1)
@@ -186,8 +187,12 @@ struct amdgpu_vcn {
 	uint32_t		*dpg_sram_curr_addr;
 
 	uint8_t	num_vcn_inst;
-	struct amdgpu_vcn_inst	inst[AMDGPU_MAX_VCN_INSTANCES];
-	struct amdgpu_vcn_reg	internal;
+	struct amdgpu_vcn_inst	 inst[AMDGPU_MAX_VCN_INSTANCES];
+	struct amdgpu_vcn_reg	 internal;
+	struct drm_gpu_scheduler *vcn_enc_sched_list[AMDGPU_MAX_VCN_ENC_RINGS];
+	struct drm_gpu_scheduler *vcn_dec_sched_list[AMDGPU_MAX_VCN_INSTANCES];
+	uint32_t		 num_vcn_enc_sched_list;
+	uint32_t		 num_vcn_dec_sched_list;
 
 	unsigned	harvest_config;
 	int (*pause_dpg_mode)(struct amdgpu_device *adev,
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 4/4] drm/scheduler: do not keep a copy of sched list
  2019-12-09 21:53 [PATCH 1/4] drm/scheduler: rework entity creation Nirmoy Das
  2019-12-09 21:53 ` [PATCH 2/4] drm/amdgpu: replace vm_pte's run-queue list with drm gpu scheds list Nirmoy Das
  2019-12-09 21:53 ` [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues Nirmoy Das
@ 2019-12-09 21:53 ` Nirmoy Das
  2 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-09 21:53 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

entity should not keep copy and maintain sched list for
itself.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
---
 drivers/gpu/drm/scheduler/sched_entity.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)

diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index f9b6ce29c58f..a5f729f421f8 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -67,17 +67,10 @@ int drm_sched_entity_init(struct drm_sched_entity *entity,
 	entity->guilty = guilty;
 	entity->num_sched_list = num_sched_list;
 	entity->priority = priority;
-	entity->sched_list =  kcalloc(num_sched_list,
-				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+	entity->sched_list =  sched_list;
 
-	if(!entity->sched_list)
-		return -ENOMEM;
 
 	init_completion(&entity->entity_idle);
-
-	for (i = 0; i < num_sched_list; i++)
-		entity->sched_list[i] = sched_list[i];
-
 	if (num_sched_list)
 		entity->rq = &entity->sched_list[0]->sched_rq[entity->priority];
 
@@ -312,7 +305,6 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)
 
 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);
 
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* Re: [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues
  2019-12-09 21:53 ` [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues Nirmoy Das
@ 2019-12-10 11:28   ` Christian König
  2019-12-10 12:55     ` Nirmoy
  0 siblings, 1 reply; 12+ messages in thread
From: Christian König @ 2019-12-10 11:28 UTC (permalink / raw)
  To: Nirmoy Das, alexander.deucher, kenny.ho; +Cc: nirmoy.das, amd-gfx

Am 09.12.19 um 22:53 schrieb Nirmoy Das:
> This sched list can be passed on to entity creation routine
> instead of manually creating such sched list on every context creation.

Please drop the "_list" from the names here. A list usually means a 
linked list and those are actually arrays.

Additional to that amdgpu_device_init_sched_list() should probably go 
into amdgpu_ctx.c instead. That is actually not really device related, 
but more UAPI/ctx stuff.

Apart from that looks good to me,
Christian.

>
> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    | 69 ++++++++--------------
>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 44 ++++++++++++++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h    |  4 ++
>   drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h   |  2 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h   |  2 +
>   drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h    |  9 ++-
>   6 files changed, 85 insertions(+), 45 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> index 1d6850af9908..c1fc75299b7d 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
> @@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
>   			   struct amdgpu_ctx *ctx)
>   {
>   	unsigned num_entities = amdgpu_ctx_total_num_entities();
> -	unsigned i, j, k;
> +	unsigned i, j;
>   	int r;
>   
>   	if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
> @@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
>   	ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
>   
>   	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
> -		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
> -		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
> -		unsigned num_rings = 0;
> -		unsigned num_rqs = 0;
> +		struct drm_gpu_scheduler **sched_list;
> +		struct drm_gpu_scheduler *sched;
> +		unsigned num_scheds = 0;
>   
>   		switch (i) {
>   		case AMDGPU_HW_IP_GFX:
> -			rings[0] = &adev->gfx.gfx_ring[0];
> -			num_rings = 1;
> +			sched_list = adev->gfx.gfx_sched_list;
> +			num_scheds = 1;
>   			break;
>   		case AMDGPU_HW_IP_COMPUTE:
> -			for (j = 0; j < adev->gfx.num_compute_rings; ++j)
> -				rings[j] = &adev->gfx.compute_ring[j];
> -			num_rings = adev->gfx.num_compute_rings;
> +			sched_list = adev->gfx.compute_sched_list;
> +			num_scheds = adev->gfx.num_compute_rings;
>   			break;
>   		case AMDGPU_HW_IP_DMA:
> -			for (j = 0; j < adev->sdma.num_instances; ++j)
> -				rings[j] = &adev->sdma.instance[j].ring;
> -			num_rings = adev->sdma.num_instances;
> +			sched_list = adev->sdma.sdma_sched_list;
> +			num_scheds = adev->sdma.num_instances;
>   			break;
>   		case AMDGPU_HW_IP_UVD:
> -			rings[0] = &adev->uvd.inst[0].ring;
> -			num_rings = 1;
> +			sched = &adev->uvd.inst[0].ring.sched;
> +			sched_list = &sched;
> +			num_scheds = 1;
>   			break;
>   		case AMDGPU_HW_IP_VCE:
> -			rings[0] = &adev->vce.ring[0];
> -			num_rings = 1;
> +			sched = &adev->vce.ring[0].sched;
> +			sched_list = &sched;
> +			num_scheds = 1;
>   			break;
>   		case AMDGPU_HW_IP_UVD_ENC:
> -			rings[0] = &adev->uvd.inst[0].ring_enc[0];
> -			num_rings = 1;
> +			sched = &adev->uvd.inst[0].ring_enc[0].sched;
> +			sched_list = &sched;
> +			num_scheds = 1;
>   			break;
>   		case AMDGPU_HW_IP_VCN_DEC:
> -			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
> -				if (adev->vcn.harvest_config & (1 << j))
> -					continue;
> -				rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
> -			}
> +			sched_list = adev->vcn.vcn_dec_sched_list;
> +			num_scheds =  adev->vcn.num_vcn_dec_sched_list;
>   			break;
>   		case AMDGPU_HW_IP_VCN_ENC:
> -			for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
> -				if (adev->vcn.harvest_config & (1 << j))
> -					continue;
> -				for (k = 0; k < adev->vcn.num_enc_rings; ++k)
> -					rings[num_rings++] = &adev->vcn.inst[j].ring_enc[k];
> -			}
> +			sched_list = adev->vcn.vcn_enc_sched_list;
> +			num_scheds =  adev->vcn.num_vcn_enc_sched_list;
>   			break;
>   		case AMDGPU_HW_IP_VCN_JPEG:
> -			for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
> -				if (adev->vcn.harvest_config & (1 << j))
> -					continue;
> -				rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
> -			}
> +			sched_list = adev->jpeg.jpeg_sched_list;
> +			num_scheds =  adev->jpeg.num_jpeg_sched_list;
>   			break;
>   		}
>   
> -		for (j = 0; j < num_rings; ++j) {
> -			if (!rings[j]->adev)
> -				continue;
> -
> -			sched_list[num_rqs++] = &rings[j]->sched;
> -		}
> -
>   		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
>   			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
>   						  priority, sched_list,
> -						  num_rqs, &ctx->guilty);
> +						  num_scheds, &ctx->guilty);
>   		if (r)
>   			goto error_cleanup_entities;
>   	}
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index cf4953c4e2cf..f973b61a26da 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -2738,6 +2738,48 @@ static int amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
>   	return ret;
>   }
>   
> +static void amdgpu_device_init_sched_list(struct amdgpu_device *adev)
> +{
> +	int i, j;
> +
> +	for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
> +		adev->gfx.gfx_sched_list[i] = &adev->gfx.gfx_ring[i].sched;
> +		adev->gfx.num_gfx_sched_list++;
> +	}
> +
> +	for (i = 0; i < adev->gfx.num_compute_rings; i++) {
> +		adev->gfx.compute_sched_list[i] = &adev->gfx.compute_ring[i].sched;
> +		adev->gfx.num_compute_sched_list++;
> +	}
> +
> +	for (i = 0; i < adev->sdma.num_instances; i++) {
> +		adev->sdma.sdma_sched_list[i] = &adev->sdma.instance[i].ring.sched;
> +		adev->sdma.num_sdma_sched_list++;
> +	}
> +
> +	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
> +		if (adev->vcn.harvest_config & (1 << i))
> +			continue;
> +		adev->vcn.vcn_dec_sched_list[adev->vcn.num_vcn_dec_sched_list++] =
> +			&adev->vcn.inst[i].ring_dec.sched;
> +	}
> +
> +	for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
> +		if (adev->vcn.harvest_config & (1 << i))
> +			continue;
> +		for (j = 0; j < adev->vcn.num_enc_rings; ++j)
> +			adev->vcn.vcn_enc_sched_list[adev->vcn.num_vcn_enc_sched_list++] =
> +				&adev->vcn.inst[i].ring_enc[j].sched;
> +	}
> +
> +	for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
> +		if (adev->vcn.harvest_config & (1 << i))
> +			continue;
> +		adev->jpeg.jpeg_sched_list[adev->jpeg.num_jpeg_sched_list++] =
> +			&adev->jpeg.inst[i].ring_dec.sched;
> +	}
> +}
> +
>   /**
>    * amdgpu_device_init - initialize the driver
>    *
> @@ -3024,6 +3066,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>   		goto failed;
>   	}
>   
> +	amdgpu_device_init_sched_list(adev);
> +
>   	adev->accel_working = true;
>   
>   	amdgpu_vm_check_compute_bug(adev);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> index 0ae0a2715b0d..b2264d4cd510 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
> @@ -269,8 +269,12 @@ struct amdgpu_gfx {
>   	bool				me_fw_write_wait;
>   	bool				cp_fw_write_wait;
>   	struct amdgpu_ring		gfx_ring[AMDGPU_MAX_GFX_RINGS];
> +	struct drm_gpu_scheduler	*gfx_sched_list[AMDGPU_MAX_GFX_RINGS];
> +	uint32_t			num_gfx_sched_list;
>   	unsigned			num_gfx_rings;
>   	struct amdgpu_ring		compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
> +	uint32_t			num_compute_sched_list;
> +	struct drm_gpu_scheduler	*compute_sched_list[AMDGPU_MAX_COMPUTE_RINGS];
>   	unsigned			num_compute_rings;
>   	struct amdgpu_irq_src		eop_irq;
>   	struct amdgpu_irq_src		priv_reg_irq;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
> index 5131a0a1bc8a..5ce6ab05eeac 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
> @@ -43,6 +43,8 @@ struct amdgpu_jpeg {
>   	uint8_t	num_jpeg_inst;
>   	struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
>   	struct amdgpu_jpeg_reg internal;
> +	struct drm_gpu_scheduler *jpeg_sched_list[AMDGPU_MAX_JPEG_INSTANCES];
> +	uint32_t num_jpeg_sched_list;
>   	unsigned harvest_config;
>   	struct delayed_work idle_work;
>   	enum amd_powergating_state cur_state;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
> index b3134655789f..0f6cb6768398 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
> @@ -52,6 +52,8 @@ struct amdgpu_sdma_instance {
>   
>   struct amdgpu_sdma {
>   	struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
> +	struct drm_gpu_scheduler    *sdma_sched_list[AMDGPU_MAX_SDMA_INSTANCES];
> +	uint32_t		    num_sdma_sched_list;
>   	struct amdgpu_irq_src	trap_irq;
>   	struct amdgpu_irq_src	illegal_inst_irq;
>   	struct amdgpu_irq_src	ecc_irq;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> index 402a5046b985..ae65cb8f07e8 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
> @@ -31,6 +31,7 @@
>   #define AMDGPU_VCN_MAX_ENC_RINGS	3
>   
>   #define AMDGPU_MAX_VCN_INSTANCES	2
> +#define AMDGPU_MAX_VCN_ENC_RINGS  AMDGPU_VCN_MAX_ENC_RINGS * AMDGPU_MAX_VCN_INSTANCES
>   
>   #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0)
>   #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1)
> @@ -186,8 +187,12 @@ struct amdgpu_vcn {
>   	uint32_t		*dpg_sram_curr_addr;
>   
>   	uint8_t	num_vcn_inst;
> -	struct amdgpu_vcn_inst	inst[AMDGPU_MAX_VCN_INSTANCES];
> -	struct amdgpu_vcn_reg	internal;
> +	struct amdgpu_vcn_inst	 inst[AMDGPU_MAX_VCN_INSTANCES];
> +	struct amdgpu_vcn_reg	 internal;
> +	struct drm_gpu_scheduler *vcn_enc_sched_list[AMDGPU_MAX_VCN_ENC_RINGS];
> +	struct drm_gpu_scheduler *vcn_dec_sched_list[AMDGPU_MAX_VCN_INSTANCES];
> +	uint32_t		 num_vcn_enc_sched_list;
> +	uint32_t		 num_vcn_dec_sched_list;
>   
>   	unsigned	harvest_config;
>   	int (*pause_dpg_mode)(struct amdgpu_device *adev,

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 12+ messages in thread
* Re: [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues
  2019-12-10 11:28   ` Christian König
@ 2019-12-10 12:55     ` Nirmoy
  2019-12-10 12:56       ` Christian König
  0 siblings, 1 reply; 12+ messages in thread
From: Nirmoy @ 2019-12-10 12:55 UTC (permalink / raw)
  To: Christian König, Nirmoy Das, alexander.deucher, kenny.ho
  Cc: nirmoy.das, amd-gfx

Thanks Christian. That make sense, resent modified patches.

On 12/10/19 12:28 PM, Christian König wrote:
> Am 09.12.19 um 22:53 schrieb Nirmoy Das:
>> This sched list can be passed on to entity creation routine
>> instead of manually creating such sched list on every context creation.
>
> Please drop the "_list" from the names here. A list usually means a 
> linked list and those are actually arrays.
>
> Additional to that amdgpu_device_init_sched_list() should probably go 
> into amdgpu_ctx.c instead. That is actually not really device related, 
> but more UAPI/ctx stuff.
>
> Apart from that looks good to me,
> Christian.
>
>>
>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>> ---
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    | 69 ++++++++--------------
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 44 ++++++++++++++
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h    |  4 ++
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h   |  2 +
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h   |  2 +
>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h    |  9 ++-
>>   6 files changed, 85 insertions(+), 45 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> index 1d6850af9908..c1fc75299b7d 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>> @@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
>>                  struct amdgpu_ctx *ctx)
>>   {
>>       unsigned num_entities = amdgpu_ctx_total_num_entities();
>> -    unsigned i, j, k;
>> +    unsigned i, j;
>>       int r;
>>         if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
>> @@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device 
>> *adev,
>>       ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
>>         for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
>> -        struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
>> -        struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
>> -        unsigned num_rings = 0;
>> -        unsigned num_rqs = 0;
>> +        struct drm_gpu_scheduler **sched_list;
>> +        struct drm_gpu_scheduler *sched;
>> +        unsigned num_scheds = 0;
>>             switch (i) {
>>           case AMDGPU_HW_IP_GFX:
>> -            rings[0] = &adev->gfx.gfx_ring[0];
>> -            num_rings = 1;
>> +            sched_list = adev->gfx.gfx_sched_list;
>> +            num_scheds = 1;
>>               break;
>>           case AMDGPU_HW_IP_COMPUTE:
>> -            for (j = 0; j < adev->gfx.num_compute_rings; ++j)
>> -                rings[j] = &adev->gfx.compute_ring[j];
>> -            num_rings = adev->gfx.num_compute_rings;
>> +            sched_list = adev->gfx.compute_sched_list;
>> +            num_scheds = adev->gfx.num_compute_rings;
>>               break;
>>           case AMDGPU_HW_IP_DMA:
>> -            for (j = 0; j < adev->sdma.num_instances; ++j)
>> -                rings[j] = &adev->sdma.instance[j].ring;
>> -            num_rings = adev->sdma.num_instances;
>> +            sched_list = adev->sdma.sdma_sched_list;
>> +            num_scheds = adev->sdma.num_instances;
>>               break;
>>           case AMDGPU_HW_IP_UVD:
>> -            rings[0] = &adev->uvd.inst[0].ring;
>> -            num_rings = 1;
>> +            sched = &adev->uvd.inst[0].ring.sched;
>> +            sched_list = &sched;
>> +            num_scheds = 1;
>>               break;
>>           case AMDGPU_HW_IP_VCE:
>> -            rings[0] = &adev->vce.ring[0];
>> -            num_rings = 1;
>> +            sched = &adev->vce.ring[0].sched;
>> +            sched_list = &sched;
>> +            num_scheds = 1;
>>               break;
>>           case AMDGPU_HW_IP_UVD_ENC:
>> -            rings[0] = &adev->uvd.inst[0].ring_enc[0];
>> -            num_rings = 1;
>> +            sched = &adev->uvd.inst[0].ring_enc[0].sched;
>> +            sched_list = &sched;
>> +            num_scheds = 1;
>>               break;
>>           case AMDGPU_HW_IP_VCN_DEC:
>> -            for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
>> -                if (adev->vcn.harvest_config & (1 << j))
>> -                    continue;
>> -                rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
>> -            }
>> +            sched_list = adev->vcn.vcn_dec_sched_list;
>> +            num_scheds =  adev->vcn.num_vcn_dec_sched_list;
>>               break;
>>           case AMDGPU_HW_IP_VCN_ENC:
>> -            for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
>> -                if (adev->vcn.harvest_config & (1 << j))
>> -                    continue;
>> -                for (k = 0; k < adev->vcn.num_enc_rings; ++k)
>> -                    rings[num_rings++] = 
>> &adev->vcn.inst[j].ring_enc[k];
>> -            }
>> +            sched_list = adev->vcn.vcn_enc_sched_list;
>> +            num_scheds =  adev->vcn.num_vcn_enc_sched_list;
>>               break;
>>           case AMDGPU_HW_IP_VCN_JPEG:
>> -            for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
>> -                if (adev->vcn.harvest_config & (1 << j))
>> -                    continue;
>> -                rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
>> -            }
>> +            sched_list = adev->jpeg.jpeg_sched_list;
>> +            num_scheds =  adev->jpeg.num_jpeg_sched_list;
>>               break;
>>           }
>>   -        for (j = 0; j < num_rings; ++j) {
>> -            if (!rings[j]->adev)
>> -                continue;
>> -
>> -            sched_list[num_rqs++] = &rings[j]->sched;
>> -        }
>> -
>>           for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
>>               r = drm_sched_entity_init(&ctx->entities[i][j].entity,
>>                             priority, sched_list,
>> -                          num_rqs, &ctx->guilty);
>> +                          num_scheds, &ctx->guilty);
>>           if (r)
>>               goto error_cleanup_entities;
>>       }
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> index cf4953c4e2cf..f973b61a26da 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>> @@ -2738,6 +2738,48 @@ static int 
>> amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
>>       return ret;
>>   }
>>   +static void amdgpu_device_init_sched_list(struct amdgpu_device *adev)
>> +{
>> +    int i, j;
>> +
>> +    for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
>> +        adev->gfx.gfx_sched_list[i] = &adev->gfx.gfx_ring[i].sched;
>> +        adev->gfx.num_gfx_sched_list++;
>> +    }
>> +
>> +    for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>> +        adev->gfx.compute_sched_list[i] = 
>> &adev->gfx.compute_ring[i].sched;
>> +        adev->gfx.num_compute_sched_list++;
>> +    }
>> +
>> +    for (i = 0; i < adev->sdma.num_instances; i++) {
>> +        adev->sdma.sdma_sched_list[i] = 
>> &adev->sdma.instance[i].ring.sched;
>> +        adev->sdma.num_sdma_sched_list++;
>> +    }
>> +
>> +    for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
>> +        if (adev->vcn.harvest_config & (1 << i))
>> +            continue;
>> + adev->vcn.vcn_dec_sched_list[adev->vcn.num_vcn_dec_sched_list++] =
>> +            &adev->vcn.inst[i].ring_dec.sched;
>> +    }
>> +
>> +    for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
>> +        if (adev->vcn.harvest_config & (1 << i))
>> +            continue;
>> +        for (j = 0; j < adev->vcn.num_enc_rings; ++j)
>> + adev->vcn.vcn_enc_sched_list[adev->vcn.num_vcn_enc_sched_list++] =
>> +                &adev->vcn.inst[i].ring_enc[j].sched;
>> +    }
>> +
>> +    for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
>> +        if (adev->vcn.harvest_config & (1 << i))
>> +            continue;
>> + adev->jpeg.jpeg_sched_list[adev->jpeg.num_jpeg_sched_list++] =
>> +            &adev->jpeg.inst[i].ring_dec.sched;
>> +    }
>> +}
>> +
>>   /**
>>    * amdgpu_device_init - initialize the driver
>>    *
>> @@ -3024,6 +3066,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>>           goto failed;
>>       }
>>   +    amdgpu_device_init_sched_list(adev);
>> +
>>       adev->accel_working = true;
>>         amdgpu_vm_check_compute_bug(adev);
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> index 0ae0a2715b0d..b2264d4cd510 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>> @@ -269,8 +269,12 @@ struct amdgpu_gfx {
>>       bool                me_fw_write_wait;
>>       bool                cp_fw_write_wait;
>>       struct amdgpu_ring        gfx_ring[AMDGPU_MAX_GFX_RINGS];
>> +    struct drm_gpu_scheduler *gfx_sched_list[AMDGPU_MAX_GFX_RINGS];
>> +    uint32_t            num_gfx_sched_list;
>>       unsigned            num_gfx_rings;
>>       struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
>> +    uint32_t            num_compute_sched_list;
>> +    struct drm_gpu_scheduler 
>> *compute_sched_list[AMDGPU_MAX_COMPUTE_RINGS];
>>       unsigned            num_compute_rings;
>>       struct amdgpu_irq_src        eop_irq;
>>       struct amdgpu_irq_src        priv_reg_irq;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>> index 5131a0a1bc8a..5ce6ab05eeac 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>> @@ -43,6 +43,8 @@ struct amdgpu_jpeg {
>>       uint8_t    num_jpeg_inst;
>>       struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
>>       struct amdgpu_jpeg_reg internal;
>> +    struct drm_gpu_scheduler 
>> *jpeg_sched_list[AMDGPU_MAX_JPEG_INSTANCES];
>> +    uint32_t num_jpeg_sched_list;
>>       unsigned harvest_config;
>>       struct delayed_work idle_work;
>>       enum amd_powergating_state cur_state;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>> index b3134655789f..0f6cb6768398 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>> @@ -52,6 +52,8 @@ struct amdgpu_sdma_instance {
>>     struct amdgpu_sdma {
>>       struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
>> +    struct drm_gpu_scheduler 
>> *sdma_sched_list[AMDGPU_MAX_SDMA_INSTANCES];
>> +    uint32_t            num_sdma_sched_list;
>>       struct amdgpu_irq_src    trap_irq;
>>       struct amdgpu_irq_src    illegal_inst_irq;
>>       struct amdgpu_irq_src    ecc_irq;
>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h 
>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>> index 402a5046b985..ae65cb8f07e8 100644
>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>> @@ -31,6 +31,7 @@
>>   #define AMDGPU_VCN_MAX_ENC_RINGS    3
>>     #define AMDGPU_MAX_VCN_INSTANCES    2
>> +#define AMDGPU_MAX_VCN_ENC_RINGS  AMDGPU_VCN_MAX_ENC_RINGS * 
>> AMDGPU_MAX_VCN_INSTANCES
>>     #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0)
>>   #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1)
>> @@ -186,8 +187,12 @@ struct amdgpu_vcn {
>>       uint32_t        *dpg_sram_curr_addr;
>>         uint8_t    num_vcn_inst;
>> -    struct amdgpu_vcn_inst    inst[AMDGPU_MAX_VCN_INSTANCES];
>> -    struct amdgpu_vcn_reg    internal;
>> +    struct amdgpu_vcn_inst     inst[AMDGPU_MAX_VCN_INSTANCES];
>> +    struct amdgpu_vcn_reg     internal;
>> +    struct drm_gpu_scheduler 
>> *vcn_enc_sched_list[AMDGPU_MAX_VCN_ENC_RINGS];
>> +    struct drm_gpu_scheduler 
>> *vcn_dec_sched_list[AMDGPU_MAX_VCN_INSTANCES];
>> +    uint32_t         num_vcn_enc_sched_list;
>> +    uint32_t         num_vcn_dec_sched_list;
>>         unsigned    harvest_config;
>>       int (*pause_dpg_mode)(struct amdgpu_device *adev,
>
_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 12+ messages in thread
* Re: [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues
  2019-12-10 12:55     ` Nirmoy
@ 2019-12-10 12:56       ` Christian König
  0 siblings, 0 replies; 12+ messages in thread
From: Christian König @ 2019-12-10 12:56 UTC (permalink / raw)
  To: Nirmoy, Nirmoy Das, alexander.deucher, kenny.ho; +Cc: nirmoy.das, amd-gfx

Yeah, but you are to fast for me. I was still looking into comments for 
patch #4 :)

Christian.

Am 10.12.19 um 13:55 schrieb Nirmoy:
> Thanks Christian. That make sense, resent modified patches.
>
> On 12/10/19 12:28 PM, Christian König wrote:
>> Am 09.12.19 um 22:53 schrieb Nirmoy Das:
>>> This sched list can be passed on to entity creation routine
>>> instead of manually creating such sched list on every context creation.
>>
>> Please drop the "_list" from the names here. A list usually means a 
>> linked list and those are actually arrays.
>>
>> Additional to that amdgpu_device_init_sched_list() should probably go 
>> into amdgpu_ctx.c instead. That is actually not really device 
>> related, but more UAPI/ctx stuff.
>>
>> Apart from that looks good to me,
>> Christian.
>>
>>>
>>> Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
>>> ---
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c    | 69 
>>> ++++++++--------------
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 44 ++++++++++++++
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h    |  4 ++
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h   |  2 +
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h   |  2 +
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h    |  9 ++-
>>>   6 files changed, 85 insertions(+), 45 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> index 1d6850af9908..c1fc75299b7d 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
>>> @@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device 
>>> *adev,
>>>                  struct amdgpu_ctx *ctx)
>>>   {
>>>       unsigned num_entities = amdgpu_ctx_total_num_entities();
>>> -    unsigned i, j, k;
>>> +    unsigned i, j;
>>>       int r;
>>>         if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX)
>>> @@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct 
>>> amdgpu_device *adev,
>>>       ctx->override_priority = DRM_SCHED_PRIORITY_UNSET;
>>>         for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
>>> -        struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
>>> -        struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
>>> -        unsigned num_rings = 0;
>>> -        unsigned num_rqs = 0;
>>> +        struct drm_gpu_scheduler **sched_list;
>>> +        struct drm_gpu_scheduler *sched;
>>> +        unsigned num_scheds = 0;
>>>             switch (i) {
>>>           case AMDGPU_HW_IP_GFX:
>>> -            rings[0] = &adev->gfx.gfx_ring[0];
>>> -            num_rings = 1;
>>> +            sched_list = adev->gfx.gfx_sched_list;
>>> +            num_scheds = 1;
>>>               break;
>>>           case AMDGPU_HW_IP_COMPUTE:
>>> -            for (j = 0; j < adev->gfx.num_compute_rings; ++j)
>>> -                rings[j] = &adev->gfx.compute_ring[j];
>>> -            num_rings = adev->gfx.num_compute_rings;
>>> +            sched_list = adev->gfx.compute_sched_list;
>>> +            num_scheds = adev->gfx.num_compute_rings;
>>>               break;
>>>           case AMDGPU_HW_IP_DMA:
>>> -            for (j = 0; j < adev->sdma.num_instances; ++j)
>>> -                rings[j] = &adev->sdma.instance[j].ring;
>>> -            num_rings = adev->sdma.num_instances;
>>> +            sched_list = adev->sdma.sdma_sched_list;
>>> +            num_scheds = adev->sdma.num_instances;
>>>               break;
>>>           case AMDGPU_HW_IP_UVD:
>>> -            rings[0] = &adev->uvd.inst[0].ring;
>>> -            num_rings = 1;
>>> +            sched = &adev->uvd.inst[0].ring.sched;
>>> +            sched_list = &sched;
>>> +            num_scheds = 1;
>>>               break;
>>>           case AMDGPU_HW_IP_VCE:
>>> -            rings[0] = &adev->vce.ring[0];
>>> -            num_rings = 1;
>>> +            sched = &adev->vce.ring[0].sched;
>>> +            sched_list = &sched;
>>> +            num_scheds = 1;
>>>               break;
>>>           case AMDGPU_HW_IP_UVD_ENC:
>>> -            rings[0] = &adev->uvd.inst[0].ring_enc[0];
>>> -            num_rings = 1;
>>> +            sched = &adev->uvd.inst[0].ring_enc[0].sched;
>>> +            sched_list = &sched;
>>> +            num_scheds = 1;
>>>               break;
>>>           case AMDGPU_HW_IP_VCN_DEC:
>>> -            for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
>>> -                if (adev->vcn.harvest_config & (1 << j))
>>> -                    continue;
>>> -                rings[num_rings++] = &adev->vcn.inst[j].ring_dec;
>>> -            }
>>> +            sched_list = adev->vcn.vcn_dec_sched_list;
>>> +            num_scheds = adev->vcn.num_vcn_dec_sched_list;
>>>               break;
>>>           case AMDGPU_HW_IP_VCN_ENC:
>>> -            for (j = 0; j < adev->vcn.num_vcn_inst; ++j) {
>>> -                if (adev->vcn.harvest_config & (1 << j))
>>> -                    continue;
>>> -                for (k = 0; k < adev->vcn.num_enc_rings; ++k)
>>> -                    rings[num_rings++] = 
>>> &adev->vcn.inst[j].ring_enc[k];
>>> -            }
>>> +            sched_list = adev->vcn.vcn_enc_sched_list;
>>> +            num_scheds = adev->vcn.num_vcn_enc_sched_list;
>>>               break;
>>>           case AMDGPU_HW_IP_VCN_JPEG:
>>> -            for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) {
>>> -                if (adev->vcn.harvest_config & (1 << j))
>>> -                    continue;
>>> -                rings[num_rings++] = &adev->jpeg.inst[j].ring_dec;
>>> -            }
>>> +            sched_list = adev->jpeg.jpeg_sched_list;
>>> +            num_scheds =  adev->jpeg.num_jpeg_sched_list;
>>>               break;
>>>           }
>>>   -        for (j = 0; j < num_rings; ++j) {
>>> -            if (!rings[j]->adev)
>>> -                continue;
>>> -
>>> -            sched_list[num_rqs++] = &rings[j]->sched;
>>> -        }
>>> -
>>>           for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
>>>               r = drm_sched_entity_init(&ctx->entities[i][j].entity,
>>>                             priority, sched_list,
>>> -                          num_rqs, &ctx->guilty);
>>> +                          num_scheds, &ctx->guilty);
>>>           if (r)
>>>               goto error_cleanup_entities;
>>>       }
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> index cf4953c4e2cf..f973b61a26da 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> @@ -2738,6 +2738,48 @@ static int 
>>> amdgpu_device_get_job_timeout_settings(struct amdgpu_device *adev)
>>>       return ret;
>>>   }
>>>   +static void amdgpu_device_init_sched_list(struct amdgpu_device 
>>> *adev)
>>> +{
>>> +    int i, j;
>>> +
>>> +    for (i = 0; i < adev->gfx.num_gfx_rings; i++) {
>>> +        adev->gfx.gfx_sched_list[i] = &adev->gfx.gfx_ring[i].sched;
>>> +        adev->gfx.num_gfx_sched_list++;
>>> +    }
>>> +
>>> +    for (i = 0; i < adev->gfx.num_compute_rings; i++) {
>>> +        adev->gfx.compute_sched_list[i] = 
>>> &adev->gfx.compute_ring[i].sched;
>>> +        adev->gfx.num_compute_sched_list++;
>>> +    }
>>> +
>>> +    for (i = 0; i < adev->sdma.num_instances; i++) {
>>> +        adev->sdma.sdma_sched_list[i] = 
>>> &adev->sdma.instance[i].ring.sched;
>>> +        adev->sdma.num_sdma_sched_list++;
>>> +    }
>>> +
>>> +    for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
>>> +        if (adev->vcn.harvest_config & (1 << i))
>>> +            continue;
>>> + adev->vcn.vcn_dec_sched_list[adev->vcn.num_vcn_dec_sched_list++] =
>>> +            &adev->vcn.inst[i].ring_dec.sched;
>>> +    }
>>> +
>>> +    for (i = 0; i < adev->vcn.num_vcn_inst; ++i) {
>>> +        if (adev->vcn.harvest_config & (1 << i))
>>> +            continue;
>>> +        for (j = 0; j < adev->vcn.num_enc_rings; ++j)
>>> + adev->vcn.vcn_enc_sched_list[adev->vcn.num_vcn_enc_sched_list++] =
>>> +                &adev->vcn.inst[i].ring_enc[j].sched;
>>> +    }
>>> +
>>> +    for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) {
>>> +        if (adev->vcn.harvest_config & (1 << i))
>>> +            continue;
>>> + adev->jpeg.jpeg_sched_list[adev->jpeg.num_jpeg_sched_list++] =
>>> +            &adev->jpeg.inst[i].ring_dec.sched;
>>> +    }
>>> +}
>>> +
>>>   /**
>>>    * amdgpu_device_init - initialize the driver
>>>    *
>>> @@ -3024,6 +3066,8 @@ int amdgpu_device_init(struct amdgpu_device 
>>> *adev,
>>>           goto failed;
>>>       }
>>>   +    amdgpu_device_init_sched_list(adev);
>>> +
>>>       adev->accel_working = true;
>>>         amdgpu_vm_check_compute_bug(adev);
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>>> index 0ae0a2715b0d..b2264d4cd510 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h
>>> @@ -269,8 +269,12 @@ struct amdgpu_gfx {
>>>       bool                me_fw_write_wait;
>>>       bool                cp_fw_write_wait;
>>>       struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS];
>>> +    struct drm_gpu_scheduler *gfx_sched_list[AMDGPU_MAX_GFX_RINGS];
>>> +    uint32_t            num_gfx_sched_list;
>>>       unsigned            num_gfx_rings;
>>>       struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS];
>>> +    uint32_t            num_compute_sched_list;
>>> +    struct drm_gpu_scheduler 
>>> *compute_sched_list[AMDGPU_MAX_COMPUTE_RINGS];
>>>       unsigned            num_compute_rings;
>>>       struct amdgpu_irq_src        eop_irq;
>>>       struct amdgpu_irq_src        priv_reg_irq;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>>> index 5131a0a1bc8a..5ce6ab05eeac 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h
>>> @@ -43,6 +43,8 @@ struct amdgpu_jpeg {
>>>       uint8_t    num_jpeg_inst;
>>>       struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES];
>>>       struct amdgpu_jpeg_reg internal;
>>> +    struct drm_gpu_scheduler 
>>> *jpeg_sched_list[AMDGPU_MAX_JPEG_INSTANCES];
>>> +    uint32_t num_jpeg_sched_list;
>>>       unsigned harvest_config;
>>>       struct delayed_work idle_work;
>>>       enum amd_powergating_state cur_state;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>>> index b3134655789f..0f6cb6768398 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h
>>> @@ -52,6 +52,8 @@ struct amdgpu_sdma_instance {
>>>     struct amdgpu_sdma {
>>>       struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES];
>>> +    struct drm_gpu_scheduler 
>>> *sdma_sched_list[AMDGPU_MAX_SDMA_INSTANCES];
>>> +    uint32_t            num_sdma_sched_list;
>>>       struct amdgpu_irq_src    trap_irq;
>>>       struct amdgpu_irq_src    illegal_inst_irq;
>>>       struct amdgpu_irq_src    ecc_irq;
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>>> index 402a5046b985..ae65cb8f07e8 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h
>>> @@ -31,6 +31,7 @@
>>>   #define AMDGPU_VCN_MAX_ENC_RINGS    3
>>>     #define AMDGPU_MAX_VCN_INSTANCES    2
>>> +#define AMDGPU_MAX_VCN_ENC_RINGS  AMDGPU_VCN_MAX_ENC_RINGS * 
>>> AMDGPU_MAX_VCN_INSTANCES
>>>     #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0)
>>>   #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1)
>>> @@ -186,8 +187,12 @@ struct amdgpu_vcn {
>>>       uint32_t        *dpg_sram_curr_addr;
>>>         uint8_t    num_vcn_inst;
>>> -    struct amdgpu_vcn_inst    inst[AMDGPU_MAX_VCN_INSTANCES];
>>> -    struct amdgpu_vcn_reg    internal;
>>> +    struct amdgpu_vcn_inst inst[AMDGPU_MAX_VCN_INSTANCES];
>>> +    struct amdgpu_vcn_reg     internal;
>>> +    struct drm_gpu_scheduler 
>>> *vcn_enc_sched_list[AMDGPU_MAX_VCN_ENC_RINGS];
>>> +    struct drm_gpu_scheduler 
>>> *vcn_dec_sched_list[AMDGPU_MAX_VCN_INSTANCES];
>>> +    uint32_t         num_vcn_enc_sched_list;
>>> +    uint32_t         num_vcn_dec_sched_list;
>>>         unsigned    harvest_config;
>>>       int (*pause_dpg_mode)(struct amdgpu_device *adev,
>>

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply	[flat|nested] 12+ messages in thread
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-11 15:42 ` Nirmoy Das
  0 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-11 15:42 UTC (permalink / raw)
  To: dri-devel
  Cc: alexander.deucher, kenny.ho, nirmoy.das, christian.koenig, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,

 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;

@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;

-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}

 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 81f6764f1ba6..2ff63d0414c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1954,11 +1954,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)

 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;

 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index d587ffe2af8e..a92f3b18e657 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;

 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;

 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 54430c9f7a27..1c65b5bffa6b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2744,6 +2744,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;

 	vm->va = RB_ROOT_CACHED;
@@ -2757,14 +2758,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);

+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;

-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)

 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;

 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}

diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;

-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }

 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..a9ed088ebf08 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;

 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;

 	init_completion(&entity->entity_idle);

-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];

-	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;

@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}

@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)

 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);

@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }

-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;

+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}

-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }

diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;

 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;

 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}

 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;

 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);

 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
--
2.24.0

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

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-11 15:42 ` Nirmoy Das
  0 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-11 15:42 UTC (permalink / raw)
  To: dri-devel
  Cc: alexander.deucher, kenny.ho, nirmoy.das, christian.koenig, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,

 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;

@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;

-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}

 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 81f6764f1ba6..2ff63d0414c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1954,11 +1954,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)

 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;

 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index d587ffe2af8e..a92f3b18e657 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;

 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;

 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 54430c9f7a27..1c65b5bffa6b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2744,6 +2744,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;

 	vm->va = RB_ROOT_CACHED;
@@ -2757,14 +2758,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);

+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;

-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;

diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)

 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;

 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}

diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;

-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }

 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..a9ed088ebf08 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;

 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;

 	init_completion(&entity->entity_idle);

-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];

-	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;

@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}

@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)

 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);

@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }

-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;

+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}

-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }

diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;

 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;

 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}

 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;

 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);

 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
--
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-10 18:17 Nirmoy Das
  0 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-10 18:17 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;
 
@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;
 
-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}
 
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 81f6764f1ba6..2ff63d0414c9 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1954,11 +1954,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index d587ffe2af8e..a92f3b18e657 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 54430c9f7a27..1c65b5bffa6b 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2744,6 +2744,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;
 
 	vm->va = RB_ROOT_CACHED;
@@ -2757,14 +2758,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);
 
+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;
 
-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)
 
 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}
 
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;
 
-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }
 
 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..4f9ae5a12090 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;
 
 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL, DRM_SCHED_PRIORITY_NORMAL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;
 
 	init_completion(&entity->entity_idle);
 
-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];
 
-	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;
 
@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}
 
@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)
 
 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);
 
@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }
 
-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;
 
+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}
 
-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }
 
diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;
 
 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;
 
 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}
 
 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;
 
 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);
 
 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-10 12:52 Nirmoy Das
  0 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-10 12:52 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;
 
@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;
 
-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}
 
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 19ffe00d9072..2b6e35893918 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1957,11 +1957,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index e324bfe6c58f..a1a110f5513d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a94c4faa5af1..5e78db30c722 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2687,6 +2687,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;
 
 	vm->va = RB_ROOT_CACHED;
@@ -2700,14 +2701,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);
 
+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;
 
-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)
 
 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}
 
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;
 
-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }
 
 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..4f9ae5a12090 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;
 
 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL, DRM_SCHED_PRIORITY_NORMAL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;
 
 	init_completion(&entity->entity_idle);
 
-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];
 
-	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;
 
@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}
 
@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)
 
 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);
 
@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }
 
-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;
 
+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}
 
-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }
 
diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;
 
 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;
 
 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}
 
 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;
 
 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);
 
 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
* [PATCH 1/4] drm/scheduler: rework entity creation
@ 2019-12-06 17:33 Nirmoy Das
  0 siblings, 0 replies; 12+ messages in thread
From: Nirmoy Das @ 2019-12-06 17:33 UTC (permalink / raw)
  To: alexander.deucher, kenny.ho, christian.koenig; +Cc: nirmoy.das, amd-gfx

Entity currently keeps a copy of run_queue list and modify it in
drm_sched_entity_set_priority(). Entities shouldn't modify run_queue
list. Use drm_gpu_scheduler list instead of drm_sched_rq list
in drm_sched_entity struct. In this way we can select a runqueue based
on entity/ctx's priority for a  drm scheduler.

Signed-off-by: Nirmoy Das <nirmoy.das@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c  |  8 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c  |  7 ++-
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c   | 14 +++--
 drivers/gpu/drm/etnaviv/etnaviv_drv.c    |  7 ++-
 drivers/gpu/drm/lima/lima_sched.c        |  5 +-
 drivers/gpu/drm/panfrost/panfrost_job.c  |  8 ++-
 drivers/gpu/drm/scheduler/sched_entity.c | 74 ++++++++++--------------
 drivers/gpu/drm/v3d/v3d_drv.c            |  8 ++-
 include/drm/gpu_scheduler.h              |  8 ++-
 11 files changed, 78 insertions(+), 75 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
index a0d3d7b756eb..1d6850af9908 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c
@@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 
 	for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) {
 		struct amdgpu_ring *rings[AMDGPU_MAX_RINGS];
-		struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS];
+		struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 		unsigned num_rings = 0;
 		unsigned num_rqs = 0;
 
@@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev,
 			if (!rings[j]->adev)
 				continue;
 
-			rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority];
+			sched_list[num_rqs++] = &rings[j]->sched;
 		}
 
 		for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j)
 			r = drm_sched_entity_init(&ctx->entities[i][j].entity,
-						  rqs, num_rqs, &ctx->guilty);
+						  priority, sched_list,
+						  num_rqs, &ctx->guilty);
 		if (r)
 			goto error_cleanup_entities;
 	}
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
index 19ffe00d9072..2b6e35893918 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c
@@ -1957,11 +1957,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable)
 
 	if (enable) {
 		struct amdgpu_ring *ring;
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		ring = adev->mman.buffer_funcs_ring;
-		rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL];
-		r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL);
+		sched = &ring->sched;
+		r = drm_sched_entity_init(&adev->mman.entity,
+				          DRM_SCHED_PRIORITY_KERNEL, &sched,
+					  1, NULL);
 		if (r) {
 			DRM_ERROR("Failed setting up TTM BO move entity (%d)\n",
 				  r);
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
index e324bfe6c58f..a1a110f5513d 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c
@@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev)
 int amdgpu_uvd_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->uvd.inst[0].ring;
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r) {
 		DRM_ERROR("Failed setting up UVD kernel entity.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
index 46b590af2fd2..ceb0dbf685f1 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c
@@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev)
 int amdgpu_vce_entity_init(struct amdgpu_device *adev)
 {
 	struct amdgpu_ring *ring;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int r;
 
 	ring = &adev->vce.ring[0];
-	rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-	r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL);
+	sched = &ring->sched;
+	r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL,
+				  &sched, 1, NULL);
 	if (r != 0) {
 		DRM_ERROR("Failed setting up VCE run queue.\n");
 		return r;
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index a94c4faa5af1..5e78db30c722 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -2687,6 +2687,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 {
 	struct amdgpu_bo_param bp;
 	struct amdgpu_bo *root;
+	struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS];
 	int r, i;
 
 	vm->va = RB_ROOT_CACHED;
@@ -2700,14 +2701,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	spin_lock_init(&vm->invalidated_lock);
 	INIT_LIST_HEAD(&vm->freed);
 
+	for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++)
+		sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched;
+
 	/* create scheduler entities for page table updates */
-	r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		return r;
 
-	r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs,
-				  adev->vm_manager.vm_pte_num_rqs, NULL);
+	r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL,
+				  sched_list, adev->vm_manager.vm_pte_num_rqs,
+				  NULL);
 	if (r)
 		goto error_free_direct;
 
diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
index 1f9c01be40d7..76ecdf8bd31c 100644
--- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c
+++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c
@@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file)
 
 	for (i = 0; i < ETNA_MAX_PIPES; i++) {
 		struct etnaviv_gpu *gpu = priv->gpu[i];
-		struct drm_sched_rq *rq;
+		struct drm_gpu_scheduler *sched;
 
 		if (gpu) {
-			rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
+			sched = &gpu->sched;
 			drm_sched_entity_init(&ctx->sched_entity[i],
-					      &rq, 1, NULL);
+					      DRM_SCHED_PRIORITY_NORMAL, &sched,
+					      1, NULL);
 			}
 	}
 
diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c
index f522c5f99729..fc8362e4149b 100644
--- a/drivers/gpu/drm/lima/lima_sched.c
+++ b/drivers/gpu/drm/lima/lima_sched.c
@@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe,
 			    struct lima_sched_context *context,
 			    atomic_t *guilty)
 {
-	struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL;
+	struct drm_gpu_scheduler *sched = &pipe->base;
 
-	return drm_sched_entity_init(&context->base, &rq, 1, guilty);
+	return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL,
+				     &sched, 1, guilty);
 }
 
 void lima_sched_context_fini(struct lima_sched_pipe *pipe,
diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c
index d411eb6c8eb9..4f9ae5a12090 100644
--- a/drivers/gpu/drm/panfrost/panfrost_job.c
+++ b/drivers/gpu/drm/panfrost/panfrost_job.c
@@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv)
 {
 	struct panfrost_device *pfdev = panfrost_priv->pfdev;
 	struct panfrost_job_slot *js = pfdev->js;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int ret, i;
 
 	for (i = 0; i < NUM_JOB_SLOTS; i++) {
-		rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &js->queue[i].sched;
+		ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i],
+					    DRM_SCHED_PRIORITY_NORMAL, &sched,
+					    1, NULL, DRM_SCHED_PRIORITY_NORMAL);
 		if (WARN_ON(ret))
 			return ret;
 	}
diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c
index 461a7a8129f4..f9b6ce29c58f 100644
--- a/drivers/gpu/drm/scheduler/sched_entity.c
+++ b/drivers/gpu/drm/scheduler/sched_entity.c
@@ -38,9 +38,10 @@
  * submit to HW ring.
  *
  * @entity: scheduler entity to init
- * @rq_list: the list of run queue on which jobs from this
+ * @priority: priority of the entity
+ * @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
  *
@@ -50,32 +51,35 @@
  * 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,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
+			  unsigned int num_sched_list,
 			  atomic_t *guilty)
 {
 	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)
+	entity->num_sched_list = num_sched_list;
+	entity->priority = priority;
+	entity->sched_list =  kcalloc(num_sched_list,
+				      sizeof(struct drm_gpu_scheduler *), GFP_KERNEL);
+
+	if(!entity->sched_list)
 		return -ENOMEM;
 
 	init_completion(&entity->entity_idle);
 
-	for (i = 0; i < num_rq_list; ++i)
-		entity->rq_list[i] = rq_list[i];
+	for (i = 0; i < num_sched_list; i++)
+		entity->sched_list[i] = sched_list[i];
 
-	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;
 
@@ -139,10 +143,10 @@ 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;
 		}
@@ -150,7 +154,7 @@ 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];
 		}
 	}
 
@@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity)
 
 	dma_fence_put(entity->last_scheduled);
 	entity->last_scheduled = NULL;
-	kfree(entity->rq_list);
+	kfree(entity->sched_list);
 }
 EXPORT_SYMBOL(drm_sched_entity_fini);
 
@@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f,
 	drm_sched_wakeup(entity->rq->sched);
 }
 
-/**
- * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority
- */
-static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
-					     enum drm_sched_priority priority)
-{
-	*rq = &(*rq)->sched->sched_rq[priority];
-}
-
 /**
  * drm_sched_entity_set_priority - Sets priority of the entity
  *
@@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq,
 void drm_sched_entity_set_priority(struct drm_sched_entity *entity,
 				   enum drm_sched_priority priority)
 {
-	unsigned int i;
-
 	spin_lock(&entity->rq_lock);
-
-	for (i = 0; i < entity->num_rq_list; ++i)
-		drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority);
-
-	if (entity->rq) {
-		drm_sched_rq_remove_entity(entity->rq, entity);
-		drm_sched_entity_set_rq_priority(&entity->rq, priority);
-		drm_sched_rq_add_entity(entity->rq, entity);
-	}
-
+	entity->priority = priority;
 	spin_unlock(&entity->rq_lock);
 }
 EXPORT_SYMBOL(drm_sched_entity_set_priority);
@@ -490,20 +474,20 @@ 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);
 	if (fence && !dma_fence_is_signaled(fence))
 		return;
 
+	spin_lock(&entity->rq_lock);
 	rq = drm_sched_entity_get_free_sched(entity);
-	if (rq == entity->rq)
-		return;
+	if (rq != entity->rq) {
+		drm_sched_rq_remove_entity(entity->rq, entity);
+		entity->rq = rq;
+	}
 
-	spin_lock(&entity->rq_lock);
-	drm_sched_rq_remove_entity(entity->rq, entity);
-	entity->rq = rq;
 	spin_unlock(&entity->rq_lock);
 }
 
diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c
index 1a07462b4528..eaa8e9682373 100644
--- a/drivers/gpu/drm/v3d/v3d_drv.c
+++ b/drivers/gpu/drm/v3d/v3d_drv.c
@@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 {
 	struct v3d_dev *v3d = to_v3d_dev(dev);
 	struct v3d_file_priv *v3d_priv;
-	struct drm_sched_rq *rq;
+	struct drm_gpu_scheduler *sched;
 	int i;
 
 	v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL);
@@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file)
 	v3d_priv->v3d = v3d;
 
 	for (i = 0; i < V3D_MAX_QUEUES; i++) {
-		rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL];
-		drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL);
+		sched = &v3d->queue[i].sched;
+		drm_sched_entity_init(&v3d_priv->sched_entity[i],
+				      DRM_SCHED_PRIORITY_NORMAL, &sched,
+				      1, NULL);
 	}
 
 	file->driver_priv = v3d_priv;
diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h
index 684692a8ed76..96a1a1b7526e 100644
--- a/include/drm/gpu_scheduler.h
+++ b/include/drm/gpu_scheduler.h
@@ -81,8 +81,9 @@ enum drm_sched_priority {
 struct drm_sched_entity {
 	struct list_head		list;
 	struct drm_sched_rq		*rq;
-	struct drm_sched_rq		**rq_list;
-	unsigned int                    num_rq_list;
+	unsigned int                    num_sched_list;
+	struct drm_gpu_scheduler        **sched_list;
+	enum drm_sched_priority         priority;
 	spinlock_t			rq_lock;
 
 	struct spsc_queue		job_queue;
@@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq,
 				struct drm_sched_entity *entity);
 
 int drm_sched_entity_init(struct drm_sched_entity *entity,
-			  struct drm_sched_rq **rq_list,
+			  enum drm_sched_priority priority,
+			  struct drm_gpu_scheduler **sched_list,
 			  unsigned int num_rq_list,
 			  atomic_t *guilty);
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout);
-- 
2.24.0

_______________________________________________
amd-gfx mailing list
amd-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

^ permalink raw reply related	[flat|nested] 12+ messages in thread
end of thread, other threads:[~2019-12-12  8:55 UTC | newest]

Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-12-09 21:53 [PATCH 1/4] drm/scheduler: rework entity creation Nirmoy Das
2019-12-09 21:53 ` [PATCH 2/4] drm/amdgpu: replace vm_pte's run-queue list with drm gpu scheds list Nirmoy Das
2019-12-09 21:53 ` [PATCH 3/4] amd/amdgpu: add sched list to IPs with multiple run-queues Nirmoy Das
2019-12-10 11:28   ` Christian König
2019-12-10 12:55     ` Nirmoy
2019-12-10 12:56       ` Christian König
2019-12-09 21:53 ` [PATCH 4/4] drm/scheduler: do not keep a copy of sched list Nirmoy Das
  -- strict thread matches above, loose matches on Subject: below --
2019-12-11 15:42 [PATCH 1/4] drm/scheduler: rework entity creation Nirmoy Das
2019-12-11 15:42 ` Nirmoy Das
2019-12-10 18:17 Nirmoy Das
2019-12-10 12:52 Nirmoy Das
2019-12-06 17:33 Nirmoy Das

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.