Skip to contentSkip to conversation

drm/scheduler: Add drm_sched_job_cleanup

RFC PATCH In review Sharat Masetty 3 participants 3 messages Oct 24, 2018 – Oct 25, 2018 1 message not in the archive

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 drm/scheduler: Add drm_sched_job_cleanup

Hi,

Can you please review this and share your thoughts on this approach. We primarily need a clean way to release the finished fence in case the job was not pushed to the queue. So, this patch adds the new API to clean up the resources allocated in sched_job_init()

Additionally I also move the fence_put(finished_fence) from the scheduler to the drivers handler of free_job(). The drivers get to use this new API. This is done so that the layer creating the sched object is the one freeing up the resources as well.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c   |  2 ++
 drivers/gpu/drm/etnaviv/etnaviv_sched.c   |  2 ++
 drivers/gpu/drm/msm/msm_sched.c           |  8 +++-----
 drivers/gpu/drm/scheduler/gpu_scheduler.c | 11 +++++++++--
 drivers/gpu/drm/v3d/v3d_sched.c           |  2 ++
 include/drm/gpu_scheduler.h               |  1 +
 7 files changed, 20 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +1 -1
index 9c85a90..9a83152 100644
@@ -1198,7 +1198,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	r = amdgpu_ctx_add_fence(p->ctx, ring, p->fence, &seq);
 	if (r) {
 		dma_fence_put(p->fence);
-		dma_fence_put(&job->base.s_fence->finished);
+		drm_sched_job_cleanup(&job->base);
 		amdgpu_job_free(job);
 		amdgpu_mn_unlock(p->mn);
 		return r;
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +2 -0
index 2bd5676..5d252d4 100644
@@ -100,6 +100,8 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
 {
 	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
 
+	drm_sched_job_cleanup(s_job);
+
 	amdgpu_ring_priority_put(job->ring, s_job->s_priority);
 	dma_fence_put(job->fence);
 	amdgpu_sync_free(&job->sync);
drivers/gpu/drm/etnaviv/etnaviv_sched.c +2 -0
index 50d6b88..30398c5 100644
@@ -127,6 +127,8 @@ static void etnaviv_sched_free_job(struct drm_sched_job *sched_job)
 {
 	struct etnaviv_gem_submit *submit = to_etnaviv_submit(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
 	etnaviv_submit_put(submit);
 }
 
drivers/gpu/drm/msm/msm_sched.c +3 -5
index 70b7713..748acc6 100644
@@ -235,6 +235,8 @@ static void msm_sched_free_job(struct drm_sched_job *sched_job)
 	struct msm_gpu *gpu = submit->gpu;
 	int i;
 
+	drm_sched_job_cleanup(sched_job);
+
 	mutex_lock(&gpu->dev->struct_mutex);
 
 	for (i = 0; i < submit->nr_bos; i++) {
@@ -300,11 +302,7 @@ int msm_sched_job_init(struct drm_sched_job *sched_job)
 	mutex_unlock(&ring->fence_idr_lock);
 
 	if (submit->out_fence_id < 0) {
-		/*
-		 * TODO: The scheduler's finished fence leaks here since the
-		 * job will not be pushed to the queue. Need to update scheduler
-		 * to fix this cleanly(?)
-		 */
+		drm_sched_job_cleanup(sched_job);
 		dma_fence_put(submit->out_fence);
 		submit->out_fence = NULL;
 		return -ENOMEM;
drivers/gpu/drm/scheduler/gpu_scheduler.c +9 -2
index f5534ff..bbf9315 100644
@@ -214,7 +214,6 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
 						 finish_cb);
 	drm_sched_fence_finished(job->s_fence);
 	WARN_ON(job->s_fence->parent);
-	dma_fence_put(&job->s_fence->finished);
 	job->sched->ops->free_job(job);
 }
 
@@ -480,7 +479,6 @@ static void drm_sched_job_finish(struct work_struct *work)
 			drm_sched_queue_delayed_work(next);
 	}
 	spin_unlock(&sched->job_list_lock);
-	dma_fence_put(&s_job->s_fence->finished);
 	sched->ops->free_job(s_job);
 }
 
@@ -636,6 +634,15 @@ int drm_sched_job_init(struct drm_sched_job *job,
 EXPORT_SYMBOL(drm_sched_job_init);
 
 /**
+ * Cleanup sched_job resources
+ */
+void drm_sched_job_cleanup(struct drm_sched_job *job)
+{
+	dma_fence_put(&job->s_fence->finished);
+}
+EXPORT_SYMBOL(drm_sched_job_cleanup);
+
+/**
  * Return ture if we can push more jobs to the hw.
  */
 static bool drm_sched_ready(struct drm_gpu_scheduler *sched)
drivers/gpu/drm/v3d/v3d_sched.c +2 -0
index b07bece..4abd73c 100644
@@ -35,6 +35,8 @@
 {
 	struct v3d_job *job = to_v3d_job(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
 	v3d_exec_put(job->exec);
 }
 
include/drm/gpu_scheduler.h +1 -0
index 5c59c38..2ec3ddf 100644
@@ -213,6 +213,7 @@ int drm_sched_job_init(struct drm_sched_job *job,
 		       struct drm_gpu_scheduler *sched,
 		       struct drm_sched_entity *entity,
 		       void *owner);
+void drm_sched_job_cleanup(struct drm_sched_job *job);
 void drm_sched_hw_job_reset(struct drm_gpu_scheduler *sched,
 			    struct drm_sched_job *job);
 void drm_sched_job_recovery(struct drm_gpu_scheduler *sched);
Signature · 6 lines
1.9.1

_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno

A message that is not in the archive (sent off-list or not yet received): 1540388423-30236-1-git-send-email-smasetty-sgV2jX0FEOL9JmXXK+q4OQ@public.gmane.org

#2 lore raw
Quoted 12 lines · Sharat Masetty

Hi,

Can you please review this and share your thoughts on this approach. We primarily need a clean way to release the finished fence in case the job was not pushed to the queue. So, this patch adds the new API to clean up the resources allocated in sched_job_init()

Additionally I also move the fence_put(finished_fence) from the scheduler to the drivers handler of free_job(). The drivers get to use this new API. This is done so that the layer creating the sched object is the one freeing up the resources as well.

Quoted 6 lines
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    |  2 +-
 drivers/gpu/drm/amd/amdgpu/amdgpu_job.c   |  2 ++
 drivers/gpu/drm/etnaviv/etnaviv_sched.c   |  2 ++
 drivers/gpu/drm/msm/msm_sched.c           |  8 +++-----

Just a nit - msm_sched.c isn't a thing yet. Can you re-generate a patch against drm-next so that if folks like this they can get it merged sooner?

Jordan

Quoted 134 lines
 drivers/gpu/drm/scheduler/gpu_scheduler.c | 11 +++++++++--
 drivers/gpu/drm/v3d/v3d_sched.c           |  2 ++
 include/drm/gpu_scheduler.h               |  1 +
 7 files changed, 20 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +1 -1
index 9c85a90..9a83152 100644
@@ -1198,7 +1198,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	r = amdgpu_ctx_add_fence(p->ctx, ring, p->fence, &seq);
 	if (r) {
 		dma_fence_put(p->fence);
-		dma_fence_put(&job->base.s_fence->finished);
+		drm_sched_job_cleanup(&job->base);
 		amdgpu_job_free(job);
 		amdgpu_mn_unlock(p->mn);
 		return r;
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +2 -0
index 2bd5676..5d252d4 100644
@@ -100,6 +100,8 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
 {
 	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
 
+	drm_sched_job_cleanup(s_job);
+
 	amdgpu_ring_priority_put(job->ring, s_job->s_priority);
 	dma_fence_put(job->fence);
 	amdgpu_sync_free(&job->sync);
drivers/gpu/drm/etnaviv/etnaviv_sched.c +2 -0
index 50d6b88..30398c5 100644
@@ -127,6 +127,8 @@ static void etnaviv_sched_free_job(struct drm_sched_job *sched_job)
 {
 	struct etnaviv_gem_submit *submit = to_etnaviv_submit(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
 	etnaviv_submit_put(submit);
 }
 
drivers/gpu/drm/msm/msm_sched.c +3 -5
index 70b7713..748acc6 100644
@@ -235,6 +235,8 @@ static void msm_sched_free_job(struct drm_sched_job *sched_job)
 	struct msm_gpu *gpu = submit->gpu;
 	int i;
 
+	drm_sched_job_cleanup(sched_job);
+
 	mutex_lock(&gpu->dev->struct_mutex);
 
 	for (i = 0; i < submit->nr_bos; i++) {
@@ -300,11 +302,7 @@ int msm_sched_job_init(struct drm_sched_job *sched_job)
 	mutex_unlock(&ring->fence_idr_lock);
 
 	if (submit->out_fence_id < 0) {
-		/*
-		 * TODO: The scheduler's finished fence leaks here since the
-		 * job will not be pushed to the queue. Need to update scheduler
-		 * to fix this cleanly(?)
-		 */
+		drm_sched_job_cleanup(sched_job);
 		dma_fence_put(submit->out_fence);
 		submit->out_fence = NULL;
 		return -ENOMEM;
drivers/gpu/drm/scheduler/gpu_scheduler.c +9 -2
index f5534ff..bbf9315 100644
@@ -214,7 +214,6 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
 						 finish_cb);
 	drm_sched_fence_finished(job->s_fence);
 	WARN_ON(job->s_fence->parent);
-	dma_fence_put(&job->s_fence->finished);
 	job->sched->ops->free_job(job);
 }
 
@@ -480,7 +479,6 @@ static void drm_sched_job_finish(struct work_struct *work)
 			drm_sched_queue_delayed_work(next);
 	}
 	spin_unlock(&sched->job_list_lock);
-	dma_fence_put(&s_job->s_fence->finished);
 	sched->ops->free_job(s_job);
 }
 
@@ -636,6 +634,15 @@ int drm_sched_job_init(struct drm_sched_job *job,
 EXPORT_SYMBOL(drm_sched_job_init);
 
 /**
+ * Cleanup sched_job resources
+ */
+void drm_sched_job_cleanup(struct drm_sched_job *job)
+{
+	dma_fence_put(&job->s_fence->finished);
+}
+EXPORT_SYMBOL(drm_sched_job_cleanup);
+
+/**
  * Return ture if we can push more jobs to the hw.
  */
 static bool drm_sched_ready(struct drm_gpu_scheduler *sched)
drivers/gpu/drm/v3d/v3d_sched.c +2 -0
index b07bece..4abd73c 100644
@@ -35,6 +35,8 @@
 {
 	struct v3d_job *job = to_v3d_job(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
 	v3d_exec_put(job->exec);
 }
 
include/drm/gpu_scheduler.h +1 -0
index 5c59c38..2ec3ddf 100644
@@ -213,6 +213,7 @@ int drm_sched_job_init(struct drm_sched_job *job,
 		       struct drm_gpu_scheduler *sched,
 		       struct drm_sched_entity *entity,
 		       void *owner);
+void drm_sched_job_cleanup(struct drm_sched_job *job);
 void drm_sched_hw_job_reset(struct drm_gpu_scheduler *sched,
 			    struct drm_sched_job *job);
 void drm_sched_job_recovery(struct drm_gpu_scheduler *sched);
Signature · 1 line
1.9.1
Signature · 6 lines
The Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
_______________________________________________
Freedreno mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/freedreno
#3 lore raw

Am 24.10.18 um 16:55 schrieb Jordan Crouse:

Quoted 13 lines · Sharat Masetty
Quoted 12 lines

Hi,

Can you please review this and share your thoughts on this approach. We primarily need a clean way to release the finished fence in case the job was not pushed to the queue. So, this patch adds the new API to clean up the resources allocated in sched_job_init()

Additionally I also move the fence_put(finished_fence) from the scheduler to the drivers handler of free_job(). The drivers get to use this new API. This is done so that the layer creating the sched object is the one freeing up the resources as well.

Quoted 8 lines
Quoted 6 lines
---
  drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c    |  2 +-
  drivers/gpu/drm/amd/amdgpu/amdgpu_job.c   |  2 ++
  drivers/gpu/drm/etnaviv/etnaviv_sched.c   |  2 ++
  drivers/gpu/drm/msm/msm_sched.c           |  8 +++-----

Just a nit - msm_sched.c isn't a thing yet. Can you re-generate a patch against drm-next so that if folks like this they can get it merged sooner?

Not a bad idea, but first of all as Jordan commented please rebase this on drm-next maybe even Alex amd-staging-drm-next branch.

Additional to that you missed one more occurrence where this needs to be applied in amdgpu_cs_submit().

Regards,
Christian.

Quoted 121 lines

Jordan

Quoted 119 lines
  drivers/gpu/drm/scheduler/gpu_scheduler.c | 11 +++++++++--
  drivers/gpu/drm/v3d/v3d_sched.c           |  2 ++
  include/drm/gpu_scheduler.h               |  1 +
  7 files changed, 20 insertions(+), 8 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +1 -1
index 9c85a90..9a83152 100644
@@ -1198,7 +1198,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
  	r = amdgpu_ctx_add_fence(p->ctx, ring, p->fence, &seq);
  	if (r) {
  		dma_fence_put(p->fence);
-		dma_fence_put(&job->base.s_fence->finished);
+		drm_sched_job_cleanup(&job->base);
  		amdgpu_job_free(job);
  		amdgpu_mn_unlock(p->mn);
  		return r;
drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +2 -0
index 2bd5676..5d252d4 100644
@@ -100,6 +100,8 @@ static void amdgpu_job_free_cb(struct drm_sched_job *s_job)
  {
  	struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base);
 
+	drm_sched_job_cleanup(s_job);
+
  	amdgpu_ring_priority_put(job->ring, s_job->s_priority);
  	dma_fence_put(job->fence);
  	amdgpu_sync_free(&job->sync);
drivers/gpu/drm/etnaviv/etnaviv_sched.c +2 -0
index 50d6b88..30398c5 100644
@@ -127,6 +127,8 @@ static void etnaviv_sched_free_job(struct drm_sched_job *sched_job)
  {
  	struct etnaviv_gem_submit *submit = to_etnaviv_submit(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
  	etnaviv_submit_put(submit);
  }
 
drivers/gpu/drm/msm/msm_sched.c +3 -5
index 70b7713..748acc6 100644
@@ -235,6 +235,8 @@ static void msm_sched_free_job(struct drm_sched_job *sched_job)
  	struct msm_gpu *gpu = submit->gpu;
  	int i;
 
+	drm_sched_job_cleanup(sched_job);
+
  	mutex_lock(&gpu->dev->struct_mutex);
 
  	for (i = 0; i < submit->nr_bos; i++) {
@@ -300,11 +302,7 @@ int msm_sched_job_init(struct drm_sched_job *sched_job)
  	mutex_unlock(&ring->fence_idr_lock);
 
  	if (submit->out_fence_id < 0) {
-		/*
-		 * TODO: The scheduler's finished fence leaks here since the
-		 * job will not be pushed to the queue. Need to update scheduler
-		 * to fix this cleanly(?)
-		 */
+		drm_sched_job_cleanup(sched_job);
  		dma_fence_put(submit->out_fence);
  		submit->out_fence = NULL;
  		return -ENOMEM;
drivers/gpu/drm/scheduler/gpu_scheduler.c +9 -2
index f5534ff..bbf9315 100644
@@ -214,7 +214,6 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
  						 finish_cb);
  	drm_sched_fence_finished(job->s_fence);
  	WARN_ON(job->s_fence->parent);
-	dma_fence_put(&job->s_fence->finished);
  	job->sched->ops->free_job(job);
  }
 
@@ -480,7 +479,6 @@ static void drm_sched_job_finish(struct work_struct *work)
  			drm_sched_queue_delayed_work(next);
  	}
  	spin_unlock(&sched->job_list_lock);
-	dma_fence_put(&s_job->s_fence->finished);
  	sched->ops->free_job(s_job);
  }
 
@@ -636,6 +634,15 @@ int drm_sched_job_init(struct drm_sched_job *job,
  EXPORT_SYMBOL(drm_sched_job_init);
 
  /**
+ * Cleanup sched_job resources
+ */
+void drm_sched_job_cleanup(struct drm_sched_job *job)
+{
+	dma_fence_put(&job->s_fence->finished);
+}
+EXPORT_SYMBOL(drm_sched_job_cleanup);
+
+/**
   * Return ture if we can push more jobs to the hw.
   */
  static bool drm_sched_ready(struct drm_gpu_scheduler *sched)
drivers/gpu/drm/v3d/v3d_sched.c +2 -0
index b07bece..4abd73c 100644
@@ -35,6 +35,8 @@
  {
  	struct v3d_job *job = to_v3d_job(sched_job);
 
+	drm_sched_job_cleanup(sched_job);
+
  	v3d_exec_put(job->exec);
  }
 
Quoted 15 lines
Quoted 15 lines
include/drm/gpu_scheduler.h +1 -0
index 5c59c38..2ec3ddf 100644
@@ -213,6 +213,7 @@ int drm_sched_job_init(struct drm_sched_job *job,
  		       struct drm_gpu_scheduler *sched,
  		       struct drm_sched_entity *entity,
  		       void *owner);
+void drm_sched_job_cleanup(struct drm_sched_job *job);
  void drm_sched_hw_job_reset(struct drm_gpu_scheduler *sched,
  			    struct drm_sched_job *job);
  void drm_sched_job_recovery(struct drm_gpu_scheduler *sched);
Signature · 1 line
1.9.1

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel