drm/sched: Fix dynamic job-flow control race
Subsystems (4)
PATCH v2 1/1 drm/sched: Fix dynamic job-flow control race
From: Rob Clark <[email protected]>
Fixes a race condition reported here: https://github.com/AsahiLinux/linux/issues/309#issuecomment-2238968609
The whole premise of lockless access to a single-producer-single- consumer queue is that there is just a single producer and single consumer. That means we can't call drm_sched_can_queue() (which is about queueing more work to the hw, not to the spsc queue) from anywhere other than the consumer (wq).
This call in the producer is just an optimization to avoid scheduling the consuming worker if it cannot yet queue more work to the hw. It is safe to drop this optimization to avoid the race condition.
--- drivers/gpu/drm/scheduler/sched_entity.c | 4 ++-- drivers/gpu/drm/scheduler/sched_main.c | 7 ++----- include/drm/gpu_scheduler.h | 2 +- 3 files changed, 5 insertions(+), 8 deletions(-)
drivers/gpu/drm/scheduler/sched_entity.c +2 -2
@@ -380,7 +380,7 @@ static void drm_sched_entity_wakeup(struct dma_fence *f, container_of(cb, struct drm_sched_entity, cb);drm_sched_entity_clear_dep (f, cb); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); }/**
@@ -612,7 +612,7 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job)if (drm_sched_policy ==DRM_SCHED_POLICY_FIFO )drm_sched_rq_update_fifo (entity, submit_ts); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); } }EXPORT_SYMBOL (drm_sched_entity_push_job);
drivers/gpu/drm/scheduler/sched_main.c +2 -5
@@ -1013,15 +1013,12 @@ EXPORT_SYMBOL(drm_sched_job_cleanup);/** * drm_sched_wakeup - Wake up the scheduler if it is ready to queue * @sched: scheduler instance -* @entity: the scheduler entity * * Wake up the scheduler if we can queue jobs. */ -void drm_sched_wakeup (struct drm_gpu_scheduler *sched, -struct drm_sched_entity *entity) +void drm_sched_wakeup (struct drm_gpu_scheduler *sched) { -if (drm_sched_can_queue (sched, entity)) -drm_sched_run_job_queue (sched); +drm_sched_run_job_queue (sched); }/**
include/drm/gpu_scheduler.h +1 -1
@@ -574,7 +574,7 @@ void drm_sched_entity_modify_sched(struct drm_sched_entity *entity,void drm_sched_tdr_queue_imm (struct drm_gpu_scheduler *sched);void drm_sched_job_cleanup (struct drm_sched_job *job); -void drm_sched_wakeup (struct drm_gpu_scheduler *sched,struct drm_sched_entity *entity); +void drm_sched_wakeup (struct drm_gpu_scheduler *sched);bool drm_sched_wqueue_ready (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_stop (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_start (struct drm_gpu_scheduler *sched);
Signature · 1 line
2.46.0
Quoted 14 lines · Rob Clark
From: Rob Clark <[email protected]>
Fixes a race condition reported here: https://github.com/AsahiLinux/linux/issues/309#issuecomment-2238968609
The whole premise of lockless access to a single-producer-single- consumer queue is that there is just a single producer and single consumer. That means we can't call
drm_sched_can_queue()(which is about queueing more work to the hw, not to the spsc queue) from anywhere other than the consumer (wq).This call in the producer is just an optimization to avoid scheduling the consuming worker if it cannot yet queue more work to the hw. It is safe to drop this optimization to avoid the race condition.
Quoted 5 lines
Suggested-by Asahi Lina <[email protected]>Fixes a78422e9dff3 ("drm/sched: implement dynamic job-flow control")Signed-off-by Rob Clark <[email protected]>
Quoted 66 lines
--- drivers/gpu/drm/scheduler/sched_entity.c | 4 ++-- drivers/gpu/drm/scheduler/sched_main.c | 7 ++----- include/drm/gpu_scheduler.h | 2 +- 3 files changed, 5 insertions(+), 8 deletions(-)drivers/gpu/drm/scheduler/sched_entity.c +2
-2@@ -380,7 +380,7 @@ static void drm_sched_entity_wakeup(struct dma_fence *f, container_of(cb, struct drm_sched_entity, cb);drm_sched_entity_clear_dep (f, cb); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); }/** @@ -612,7 +612,7 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job)if (drm_sched_policy ==DRM_SCHED_POLICY_FIFO )drm_sched_rq_update_fifo (entity, submit_ts); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); } }EXPORT_SYMBOL (drm_sched_entity_push_job);drivers/gpu/drm/scheduler/sched_main.c +2
-5@@ -1013,15 +1013,12 @@ EXPORT_SYMBOL(drm_sched_job_cleanup);/** * drm_sched_wakeup - Wake up the scheduler if it is ready to queue * @sched: scheduler instance -* @entity: the scheduler entity * * Wake up the scheduler if we can queue jobs. */ -void drm_sched_wakeup (struct drm_gpu_scheduler *sched, -struct drm_sched_entity *entity) +void drm_sched_wakeup (struct drm_gpu_scheduler *sched) { -if (drm_sched_can_queue (sched, entity)) -drm_sched_run_job_queue (sched); +drm_sched_run_job_queue (sched); }/** include/drm/gpu_scheduler.h +1
-1@@ -574,7 +574,7 @@ void drm_sched_entity_modify_sched(struct drm_sched_entity *entity,void drm_sched_tdr_queue_imm (struct drm_gpu_scheduler *sched);void drm_sched_job_cleanup (struct drm_sched_job *job); -void drm_sched_wakeup (struct drm_gpu_scheduler *sched,struct drm_sched_entity *entity); +void drm_sched_wakeup (struct drm_gpu_scheduler *sched);bool drm_sched_wqueue_ready (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_stop (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_start (struct drm_gpu_scheduler *sched);Signature · 1 line
2.46.0
Quoted 19 lines · Rob Clark
From: Rob Clark <[email protected]>
Fixes a race condition reported here: https://github.com/AsahiLinux/linux/issues/309#issuecomment-2238968609
The whole premise of lockless access to a single-producer-single- consumer queue is that there is just a single producer and single consumer. That means we can't call
drm_sched_can_queue()(which is about queueing more work to the hw, not to the spsc queue) from anywhere other than the consumer (wq).This call in the producer is just an optimization to avoid scheduling the consuming worker if it cannot yet queue more work to the hw. It is safe to drop this optimization to avoid the race condition.
Suggested-by Asahi Lina <[email protected]>Fixes a78422e9dff3 ("drm/sched: implement dynamic job-flow control")Signed-off-by Rob Clark <[email protected]>
Quoted 5 lines
--- drivers/gpu/drm/scheduler/sched_entity.c | 4 ++-- drivers/gpu/drm/scheduler/sched_main.c | 7 ++----- include/drm/gpu_scheduler.h | 2 +- 3 files changed, 5 insertions(+), 8 deletions(-)
Tested for several hours with CONFIG_PREMPT=y and kasan with a similar workload as in the github issue without reports or oopses.
Feel free to add
thanks,
Janne
Quoted 14 lines · Rob Clark
From: Rob Clark <[email protected]>
Fixes a race condition reported here: https://github.com/AsahiLinux/linux/issues/309#issuecomment-2238968609
The whole premise of lockless access to a single-producer-single- consumer queue is that there is just a single producer and single consumer. That means we can't call
drm_sched_can_queue()(which is about queueing more work to the hw, not to the spsc queue) from anywhere other than the consumer (wq).This call in the producer is just an optimization to avoid scheduling the consuming worker if it cannot yet queue more work to the hw. It is safe to drop this optimization to avoid the race condition.
Quoted 5 lines
Suggested-by Asahi Lina <[email protected]>Fixes a78422e9dff3 ("drm/sched: implement dynamic job-flow control")Signed-off-by Rob Clark <[email protected]>
Applied to drm-misc-fixes, thanks!
Quoted 66 lines
--- drivers/gpu/drm/scheduler/sched_entity.c | 4 ++-- drivers/gpu/drm/scheduler/sched_main.c | 7 ++----- include/drm/gpu_scheduler.h | 2 +- 3 files changed, 5 insertions(+), 8 deletions(-)drivers/gpu/drm/scheduler/sched_entity.c +2
-2@@ -380,7 +380,7 @@ static void drm_sched_entity_wakeup(struct dma_fence *f, container_of(cb, struct drm_sched_entity, cb);drm_sched_entity_clear_dep (f, cb); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); }/** @@ -612,7 +612,7 @@ void drm_sched_entity_push_job(struct drm_sched_job *sched_job)if (drm_sched_policy ==DRM_SCHED_POLICY_FIFO )drm_sched_rq_update_fifo (entity, submit_ts); -drm_sched_wakeup (entity->rq->sched, entity); +drm_sched_wakeup (entity->rq->sched); } }EXPORT_SYMBOL (drm_sched_entity_push_job);drivers/gpu/drm/scheduler/sched_main.c +2
-5@@ -1013,15 +1013,12 @@ EXPORT_SYMBOL(drm_sched_job_cleanup);/** * drm_sched_wakeup - Wake up the scheduler if it is ready to queue * @sched: scheduler instance -* @entity: the scheduler entity * * Wake up the scheduler if we can queue jobs. */ -void drm_sched_wakeup (struct drm_gpu_scheduler *sched, -struct drm_sched_entity *entity) +void drm_sched_wakeup (struct drm_gpu_scheduler *sched) { -if (drm_sched_can_queue (sched, entity)) -drm_sched_run_job_queue (sched); +drm_sched_run_job_queue (sched); }/** include/drm/gpu_scheduler.h +1
-1@@ -574,7 +574,7 @@ void drm_sched_entity_modify_sched(struct drm_sched_entity *entity,void drm_sched_tdr_queue_imm (struct drm_gpu_scheduler *sched);void drm_sched_job_cleanup (struct drm_sched_job *job); -void drm_sched_wakeup (struct drm_gpu_scheduler *sched,struct drm_sched_entity *entity); +void drm_sched_wakeup (struct drm_gpu_scheduler *sched);bool drm_sched_wqueue_ready (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_stop (struct drm_gpu_scheduler *sched);void drm_sched_wqueue_start (struct drm_gpu_scheduler *sched);Signature · 1 line
2.46.0