Skip to contentSkip to conversation

drm/scheduler: only kill entity if last user is killed

PATCH In review Christian König 3 participants 3 messages Jul 26, 2018 – Jul 30, 2018 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/scheduler: only kill entity if last user is killed

Note which task is using the entity and only kill it if the last user of the entity is killed. This should prevent problems when entities are leaked to child processes.

---
 drivers/gpu/drm/scheduler/gpu_scheduler.c | 6 +++++-
 include/drm/gpu_scheduler.h               | 1 +
 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +5 -1
index 3f2fc5e8242a..f563e4fbb4b6 100644
@@ -275,6 +275,7 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout)
 {
 	struct drm_gpu_scheduler *sched;
+	struct task_struct *last_user;
 	long ret = timeout;
 
 	sched = entity->rq->sched;
@@ -295,7 +296,9 @@ long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout)
 
 
 	/* For killed process disable any more IBs enqueue right now */
-	if ((current->flags & PF_EXITING) && (current->exit_code == SIGKILL))
+	last_user = cmpxchg(&entity->last_user, current->group_leader, NULL);
+	if ((!last_user || last_user == current->group_leader) &&
+	    (current->flags & PF_EXITING) && (current->exit_code == SIGKILL))
 		drm_sched_entity_set_rq(entity, NULL);
 
 	return ret;
@@ -541,6 +544,7 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job,
 
 	trace_drm_sched_job(sched_job, entity);
 
+	WRITE_ONCE(entity->last_user, current->group_leader);
 	first = spsc_queue_push(&entity->job_queue, &sched_job->queue_node);
 
 	/* first job wakes up scheduler */
include/drm/gpu_scheduler.h +1 -0
index 091b9afcd184..64016446a103 100644
@@ -85,6 +85,7 @@ struct drm_sched_entity {
 	struct dma_fence_cb		cb;
 	atomic_t			*guilty;
 	struct dma_fence                *last_scheduled;
+	struct task_struct		*last_user;
 };
 
 /**
Signature · 6 lines
2.14.1

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

A message that is not in the archive (sent off-list or not yet received): 20180726114554.3873-1-christian.koenig-5C7GfCeVMHo@public.gmane.org

Reviewed-by
#2 lore raw
Quoted 5 lines · Christian König

Note which task is using the entity and only kill it if the last user of the entity is killed. This should prevent problems when entities are leaked to child processes.

I am not very familiar with scheduler yet. From code logic, it looks fine for me.

Quoted 55 lines
---
 drivers/gpu/drm/scheduler/gpu_scheduler.c | 6 +++++-
 include/drm/gpu_scheduler.h               | 1 +
 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +5 -1
index 3f2fc5e8242a..f563e4fbb4b6 100644
@@ -275,6 +275,7 @@ static void drm_sched_entity_kill_jobs_cb(struct dma_fence *f,
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout)
 {
 	struct drm_gpu_scheduler *sched;
+	struct task_struct *last_user;
 	long ret = timeout;
 
 	sched = entity->rq->sched;
@@ -295,7 +296,9 @@ long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout)
 
 
 	/* For killed process disable any more IBs enqueue right now */
-	if ((current->flags & PF_EXITING) && (current->exit_code == SIGKILL))
+	last_user = cmpxchg(&entity->last_user, current->group_leader, NULL);
+	if ((!last_user || last_user == current->group_leader) &&
+	    (current->flags & PF_EXITING) && (current->exit_code == SIGKILL))
 		drm_sched_entity_set_rq(entity, NULL);
 
 	return ret;
@@ -541,6 +544,7 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job,
 
 	trace_drm_sched_job(sched_job, entity);
 
+	WRITE_ONCE(entity->last_user, current->group_leader);
 	first = spsc_queue_push(&entity->job_queue, &sched_job->queue_node);
 
 	/* first job wakes up scheduler */
include/drm/gpu_scheduler.h +1 -0
index 091b9afcd184..64016446a103 100644
@@ -85,6 +85,7 @@ struct drm_sched_entity {
 	struct dma_fence_cb		cb;
 	atomic_t			*guilty;
 	struct dma_fence                *last_scheduled;
+	struct task_struct		*last_user;
 };
 
 /**
Signature · 6 lines
2.14.1

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

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

#3 lore raw
In reply to a message not in the archive
Quoted 55 lines · Christian König

Note which task is using the entity and only kill it if the last user of the entity is killed. This should prevent problems when entities are leaked to
child processes.

---
 drivers/gpu/drm/scheduler/gpu_scheduler.c | 6 +++++-
 include/drm/gpu_scheduler.h               | 1 +
 2 files changed, 6 insertions(+), 1 deletion(-)
drivers/gpu/drm/scheduler/gpu_scheduler.c +1 -0
index 3f2fc5e8242a..f563e4fbb4b6 100644
@@ -275,6 +275,7 @@ static void drm_sched_entity_kill_jobs_cb(struct
 dma_fence *f,
 long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout)
 {
        struct drm_gpu_scheduler *sched;
+       struct task_struct *last_user;
        long ret = timeout;
 
        sched = entity->rq->sched;
drivers/gpu/drm/scheduler/gpu_scheduler.c +3 -1
@@ -295,7 +296,9 @@ long drm_sched_entity_flush(struct drm_sched_entity
 *entity, long timeout)
 
 
        /* For killed process disable any more IBs enqueue right now */
-       if ((current->flags & PF_EXITING) && (current->exit_code ==
 SIGKILL))
+       last_user = cmpxchg(&entity->last_user, current->group_leader,
 NULL);
+       if ((!last_user || last_user == current->group_leader) &&
+           (current->flags & PF_EXITING) && (current->exit_code ==

SIGKILL))

                drm_sched_entity_set_rq(entity, NULL);

        return ret;
drivers/gpu/drm/scheduler/gpu_scheduler.c +1 -0
@@ -541,6 +544,7 @@ void drm_sched_entity_push_job(struct drm_sched_job
 *sched_job,
 
        trace_drm_sched_job(sched_job, entity);
 
+       WRITE_ONCE(entity->last_user, current->group_leader);
        first = spsc_queue_push(&entity->job_queue,
 &sched_job->queue_node);
        /* first job wakes up scheduler */
include/drm/gpu_scheduler.h +0 -0
index 091b9afcd184..64016446a103 100644

include/drm/gpu_scheduler.h struct drm_sched_entity { · line 89

Quoted 5 lines
include/drm/gpu_scheduler.h +1 -0
@@ -85,6 +85,7 @@ struct drm_sched_entity {
        struct dma_fence_cb             cb;
        atomic_t                        *guilty;
        struct dma_fence                *last_scheduled;
+       struct task_struct              *last_user;
 

Please add a line regarding this field in the kernel doc comments.

Regards,
Nayan

Quoted 11 lines
include/drm/gpu_scheduler.h +0 -0
 };
 

/**
--
2.14.1

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

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