From 16a7133f35b310838ba934b6a29f8af73a88f093 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 09:43:07 +0200 Subject: drm/amdgpu: fix coding style in the scheduler v2 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit v2: fix even more Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 25 +++++++++++++++---------- 1 file changed, 15 insertions(+), 10 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index c16248cee779..f5ac01db287b 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -320,7 +320,9 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job) } static void amd_sched_free_job(struct fence *f, struct fence_cb *cb) { - struct amd_sched_job *job = container_of(cb, struct amd_sched_job, cb_free_job); + struct amd_sched_job *job = container_of(cb, struct amd_sched_job, + cb_free_job); + schedule_work(&job->work_free_job); } @@ -341,7 +343,8 @@ void amd_sched_job_finish(struct amd_sched_job *s_job) struct amd_sched_job, node); if (next) { - INIT_DELAYED_WORK(&next->work_tdr, s_job->timeout_callback); + INIT_DELAYED_WORK(&next->work_tdr, + s_job->timeout_callback); amd_sched_job_get(next); schedule_delayed_work(&next->work_tdr, sched->timeout); } @@ -353,7 +356,8 @@ void amd_sched_job_begin(struct amd_sched_job *s_job) struct amd_gpu_scheduler *sched = s_job->sched; if (sched->timeout != MAX_SCHEDULE_TIMEOUT && - list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node) == s_job) + list_first_entry_or_null(&sched->ring_mirror_list, + struct amd_sched_job, node) == s_job) { INIT_DELAYED_WORK(&s_job->work_tdr, s_job->timeout_callback); amd_sched_job_get(s_job); @@ -374,7 +378,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) sched_job->use_sched = 1; fence_add_callback(&sched_job->s_fence->base, - &sched_job->cb_free_job, amd_sched_free_job); + &sched_job->cb_free_job, amd_sched_free_job); trace_amd_sched_job(sched_job); wait_event(entity->sched->job_scheduled, amd_sched_entity_in(sched_job)); @@ -382,11 +386,11 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) /* init a sched_job with basic field */ int amd_sched_job_init(struct amd_sched_job *job, - struct amd_gpu_scheduler *sched, - struct amd_sched_entity *entity, - void (*timeout_cb)(struct work_struct *work), - void (*free_cb)(struct kref *refcount), - void *owner, struct fence **fence) + struct amd_gpu_scheduler *sched, + struct amd_sched_entity *entity, + void (*timeout_cb)(struct work_struct *work), + void (*free_cb)(struct kref *refcount), + void *owner, struct fence **fence) { INIT_LIST_HEAD(&job->node); kref_init(&job->refcount); @@ -504,7 +508,8 @@ static int amd_sched_main(void *param) if (r == -ENOENT) amd_sched_process_job(fence, &s_fence->cb); else if (r) - DRM_ERROR("fence add callback failed (%d)\n", r); + DRM_ERROR("fence add callback failed (%d)\n", + r); fence_put(fence); } else { DRM_ERROR("Failed to run job!\n"); -- cgit v1.2.3 From 7392c329ee1e49663fc7c7e47e32ab0dcc79b4d2 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 13:00:38 +0200 Subject: drm/amdgpu: remove begin_job/finish_job MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Completely pointless and confusing to use a callback to call into the same code file. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 2 -- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 13 +++++++++---- drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 6 ------ drivers/gpu/drm/amd/scheduler/sched_fence.c | 10 ---------- 4 files changed, 9 insertions(+), 22 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index f0dafa514fe4..001030b6822b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -193,6 +193,4 @@ err: const struct amd_sched_backend_ops amdgpu_sched_ops = { .dependency = amdgpu_job_dependency, .run_job = amdgpu_job_run, - .begin_job = amd_sched_job_begin, - .finish_job = amd_sched_job_finish, }; diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index f5ac01db287b..821bc89bd057 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -329,7 +329,7 @@ static void amd_sched_free_job(struct fence *f, struct fence_cb *cb) { /* job_finish is called after hw fence signaled, and * the job had already been deleted from ring_mirror_list */ -void amd_sched_job_finish(struct amd_sched_job *s_job) +static void amd_sched_job_finish(struct amd_sched_job *s_job) { struct amd_sched_job *next; struct amd_gpu_scheduler *sched = s_job->sched; @@ -351,7 +351,7 @@ void amd_sched_job_finish(struct amd_sched_job *s_job) } } -void amd_sched_job_begin(struct amd_sched_job *s_job) +static void amd_sched_job_begin(struct amd_sched_job *s_job) { struct amd_gpu_scheduler *sched = s_job->sched; @@ -461,7 +461,7 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) /* remove job from ring_mirror_list */ spin_lock_irqsave(&sched->job_list_lock, flags); list_del_init(&s_fence->s_job->node); - sched->ops->finish_job(s_fence->s_job); + amd_sched_job_finish(s_fence->s_job); spin_unlock_irqrestore(&sched->job_list_lock, flags); amd_sched_fence_signal(s_fence); @@ -475,6 +475,7 @@ static int amd_sched_main(void *param) { struct sched_param sparam = {.sched_priority = 1}; struct amd_gpu_scheduler *sched = (struct amd_gpu_scheduler *)param; + unsigned long flags; int r, count; sched_setscheduler(current, SCHED_FIFO, &sparam); @@ -499,7 +500,11 @@ static int amd_sched_main(void *param) s_fence = sched_job->s_fence; atomic_inc(&sched->hw_rq_count); - amd_sched_job_pre_schedule(sched, sched_job); + spin_lock_irqsave(&sched->job_list_lock, flags); + list_add_tail(&sched_job->node, &sched->ring_mirror_list); + amd_sched_job_begin(sched_job); + spin_unlock_irqrestore(&sched->job_list_lock, flags); + fence = sched->ops->run_job(sched_job); amd_sched_fence_scheduled(s_fence); if (fence) { diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 690ae4b0c673..69840d7458e6 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -110,8 +110,6 @@ static inline struct amd_sched_fence *to_amd_sched_fence(struct fence *f) struct amd_sched_backend_ops { struct fence *(*dependency)(struct amd_sched_job *sched_job); struct fence *(*run_job)(struct amd_sched_job *sched_job); - void (*begin_job)(struct amd_sched_job *sched_job); - void (*finish_job)(struct amd_sched_job *sched_job); }; enum amd_sched_priority { @@ -160,10 +158,6 @@ int amd_sched_job_init(struct amd_sched_job *job, void (*timeout_cb)(struct work_struct *work), void (*free_cb)(struct kref* refcount), void *owner, struct fence **fence); -void amd_sched_job_pre_schedule(struct amd_gpu_scheduler *sched , - struct amd_sched_job *s_job); -void amd_sched_job_finish(struct amd_sched_job *s_job); -void amd_sched_job_begin(struct amd_sched_job *s_job); static inline void amd_sched_job_get(struct amd_sched_job *job) { if (job) diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c index 6bdc9b7169d2..71931bc40069 100644 --- a/drivers/gpu/drm/amd/scheduler/sched_fence.c +++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c @@ -58,16 +58,6 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence) FENCE_TRACE(&fence->base, "was already signaled\n"); } -void amd_sched_job_pre_schedule(struct amd_gpu_scheduler *sched , - struct amd_sched_job *s_job) -{ - unsigned long flags; - spin_lock_irqsave(&sched->job_list_lock, flags); - list_add_tail(&s_job->node, &sched->ring_mirror_list); - sched->ops->begin_job(s_job); - spin_unlock_irqrestore(&sched->job_list_lock, flags); -} - void amd_sched_fence_scheduled(struct amd_sched_fence *s_fence) { struct fence_cb *cur, *tmp; -- cgit v1.2.3 From 20df080da27e368eea04e5d69d4c7292f8425281 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 13:04:02 +0200 Subject: drm/amdgpu: remove duplicated timeout callback MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit No need for double housekeeping here. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 5 +---- drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 - 2 files changed, 1 insertion(+), 5 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 821bc89bd057..f3fd80b93a20 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -343,8 +343,6 @@ static void amd_sched_job_finish(struct amd_sched_job *s_job) struct amd_sched_job, node); if (next) { - INIT_DELAYED_WORK(&next->work_tdr, - s_job->timeout_callback); amd_sched_job_get(next); schedule_delayed_work(&next->work_tdr, sched->timeout); } @@ -359,7 +357,6 @@ static void amd_sched_job_begin(struct amd_sched_job *s_job) list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node) == s_job) { - INIT_DELAYED_WORK(&s_job->work_tdr, s_job->timeout_callback); amd_sched_job_get(s_job); schedule_delayed_work(&s_job->work_tdr, sched->timeout); } @@ -401,7 +398,7 @@ int amd_sched_job_init(struct amd_sched_job *job, return -ENOMEM; job->s_fence->s_job = job; - job->timeout_callback = timeout_cb; + INIT_DELAYED_WORK(&job->work_tdr, timeout_cb); job->free_callback = free_cb; if (fence) diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 69840d7458e6..ec55b9f073d1 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -87,7 +87,6 @@ struct amd_sched_job { struct work_struct work_free_job; struct list_head node; struct delayed_work work_tdr; - void (*timeout_callback) (struct work_struct *work); void (*free_callback)(struct kref *refcount); }; -- cgit v1.2.3 From 1e24e31f22a69e0983c3dbc31dc725cd940ad258 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 13:12:12 +0200 Subject: drm/amdgpu: remove use_shed hack in job cleanup MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Remembering the code path in a variable to cleanup differently is usually not a good idea at all. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 13 ++++++++----- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 - drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 - 3 files changed, 8 insertions(+), 7 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index f0fa48511f41..be4698b0b33b 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -86,7 +86,7 @@ int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size, return r; } -void amdgpu_job_free(struct amdgpu_job *job) +static void amdgpu_job_free_resources(struct amdgpu_job *job) { struct fence *f; unsigned i; @@ -100,9 +100,6 @@ void amdgpu_job_free(struct amdgpu_job *job) amdgpu_bo_unref(&job->uf_bo); amdgpu_sync_free(&job->sync); - - if (!job->base.use_sched) - kfree(job); } void amdgpu_job_free_func(struct kref *refcount) @@ -111,6 +108,12 @@ void amdgpu_job_free_func(struct kref *refcount) kfree(job); } +void amdgpu_job_free(struct amdgpu_job *job) +{ + amdgpu_job_free_resources(job); + kfree(job); +} + int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, struct amd_sched_entity *entity, void *owner, struct fence **f) @@ -187,7 +190,7 @@ static struct fence *amdgpu_job_run(struct amd_sched_job *sched_job) err: job->fence = fence; - amdgpu_job_free(job); + amdgpu_job_free_resources(job); return fence; } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index f3fd80b93a20..e8ee90fa6899 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -373,7 +373,6 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) { struct amd_sched_entity *entity = sched_job->s_entity; - sched_job->use_sched = 1; fence_add_callback(&sched_job->s_fence->base, &sched_job->cb_free_job, amd_sched_free_job); trace_amd_sched_job(sched_job); diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index ec55b9f073d1..7e333fa08539 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -82,7 +82,6 @@ struct amd_sched_job { struct amd_gpu_scheduler *sched; struct amd_sched_entity *s_entity; struct amd_sched_fence *s_fence; - bool use_sched; /* true if the job goes to scheduler */ struct fence_cb cb_free_job; struct work_struct work_free_job; struct list_head node; -- cgit v1.2.3 From 0e51a772e2014db55b969c06814e8fe01d167ba2 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 14:19:32 +0200 Subject: drm/amdgpu: properly abstract scheduler timeout handling MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit The driver shouldn't mess with the scheduler internals. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 15 ++++++++------- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 11 +++++++++-- drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 2 +- 5 files changed, 19 insertions(+), 13 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index aeec288b572a..a5d1cfbb1a87 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -759,7 +759,6 @@ void amdgpu_job_free_func(struct kref *refcount); int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, struct amd_sched_entity *entity, void *owner, struct fence **f); -void amdgpu_job_timeout_func(struct work_struct *work); struct amdgpu_ring { struct amdgpu_device *adev; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 9bc8f1d99733..72694d7f11f5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -838,8 +838,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, p->job = NULL; r = amd_sched_job_init(&job->base, &ring->sched, - entity, amdgpu_job_timeout_func, - amdgpu_job_free_func, + entity, amdgpu_job_free_func, p->filp, &fence); if (r) { amdgpu_job_free(job); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index be4698b0b33b..32132f2e236d 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -34,13 +34,14 @@ static void amdgpu_job_free_handler(struct work_struct *ws) amd_sched_job_put(&job->base); } -void amdgpu_job_timeout_func(struct work_struct *work) +static void amdgpu_job_timedout(struct amd_sched_job *s_job) { - struct amdgpu_job *job = container_of(work, struct amdgpu_job, base.work_tdr.work); + struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base); + DRM_ERROR("ring %s timeout, last signaled seq=%u, last emitted seq=%u\n", - job->base.sched->name, - (uint32_t)atomic_read(&job->ring->fence_drv.last_seq), - job->ring->fence_drv.sync_seq); + job->base.sched->name, + atomic_read(&job->ring->fence_drv.last_seq), + job->ring->fence_drv.sync_seq); amd_sched_job_put(&job->base); } @@ -126,8 +127,7 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, return -EINVAL; r = amd_sched_job_init(&job->base, &ring->sched, - entity, amdgpu_job_timeout_func, - amdgpu_job_free_func, owner, &fence); + entity, amdgpu_job_free_func, owner, &fence); if (r) return r; @@ -197,4 +197,5 @@ err: const struct amd_sched_backend_ops amdgpu_sched_ops = { .dependency = amdgpu_job_dependency, .run_job = amdgpu_job_run, + .timedout_job = amdgpu_job_timedout, }; diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index e8ee90fa6899..f2ed8c56a07e 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -362,6 +362,14 @@ static void amd_sched_job_begin(struct amd_sched_job *s_job) } } +static void amd_sched_job_timedout(struct work_struct *work) +{ + struct amd_sched_job *job = container_of(work, struct amd_sched_job, + work_tdr.work); + + job->sched->ops->timedout_job(job); +} + /** * Submit a job to the job queue * @@ -384,7 +392,6 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void (*timeout_cb)(struct work_struct *work), void (*free_cb)(struct kref *refcount), void *owner, struct fence **fence) { @@ -397,7 +404,7 @@ int amd_sched_job_init(struct amd_sched_job *job, return -ENOMEM; job->s_fence->s_job = job; - INIT_DELAYED_WORK(&job->work_tdr, timeout_cb); + INIT_DELAYED_WORK(&job->work_tdr, amd_sched_job_timedout); job->free_callback = free_cb; if (fence) diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 7e333fa08539..f0de46ce7afe 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -108,6 +108,7 @@ static inline struct amd_sched_fence *to_amd_sched_fence(struct fence *f) struct amd_sched_backend_ops { struct fence *(*dependency)(struct amd_sched_job *sched_job); struct fence *(*run_job)(struct amd_sched_job *sched_job); + void (*timedout_job)(struct amd_sched_job *sched_job); }; enum amd_sched_priority { @@ -153,7 +154,6 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence); int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void (*timeout_cb)(struct work_struct *work), void (*free_cb)(struct kref* refcount), void *owner, struct fence **fence); static inline void amd_sched_job_get(struct amd_sched_job *job) -- cgit v1.2.3 From f42d20a94284149bc6815a66ddb0b449f9f2fe74 Mon Sep 17 00:00:00 2001 From: Christian König Date: Wed, 18 May 2016 15:40:58 +0200 Subject: drm/amdgpu: move locking into the functions who need it MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Otherwise the locking becomes rather confusing. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index f2ed8c56a07e..cb56d9065e43 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -333,7 +333,11 @@ static void amd_sched_job_finish(struct amd_sched_job *s_job) { struct amd_sched_job *next; struct amd_gpu_scheduler *sched = s_job->sched; + unsigned long flags; + /* remove job from ring_mirror_list */ + spin_lock_irqsave(&sched->job_list_lock, flags); + list_del_init(&s_job->node); if (sched->timeout != MAX_SCHEDULE_TIMEOUT) { if (cancel_delayed_work(&s_job->work_tdr)) amd_sched_job_put(s_job); @@ -347,12 +351,16 @@ static void amd_sched_job_finish(struct amd_sched_job *s_job) schedule_delayed_work(&next->work_tdr, sched->timeout); } } + spin_unlock_irqrestore(&sched->job_list_lock, flags); } static void amd_sched_job_begin(struct amd_sched_job *s_job) { struct amd_gpu_scheduler *sched = s_job->sched; + unsigned long flags; + spin_lock_irqsave(&sched->job_list_lock, flags); + list_add_tail(&s_job->node, &sched->ring_mirror_list); if (sched->timeout != MAX_SCHEDULE_TIMEOUT && list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node) == s_job) @@ -360,6 +368,7 @@ static void amd_sched_job_begin(struct amd_sched_job *s_job) amd_sched_job_get(s_job); schedule_delayed_work(&s_job->work_tdr, sched->timeout); } + spin_unlock_irqrestore(&sched->job_list_lock, flags); } static void amd_sched_job_timedout(struct work_struct *work) @@ -457,15 +466,10 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) struct amd_sched_fence *s_fence = container_of(cb, struct amd_sched_fence, cb); struct amd_gpu_scheduler *sched = s_fence->sched; - unsigned long flags; atomic_dec(&sched->hw_rq_count); - /* remove job from ring_mirror_list */ - spin_lock_irqsave(&sched->job_list_lock, flags); - list_del_init(&s_fence->s_job->node); amd_sched_job_finish(s_fence->s_job); - spin_unlock_irqrestore(&sched->job_list_lock, flags); amd_sched_fence_signal(s_fence); @@ -478,7 +482,6 @@ static int amd_sched_main(void *param) { struct sched_param sparam = {.sched_priority = 1}; struct amd_gpu_scheduler *sched = (struct amd_gpu_scheduler *)param; - unsigned long flags; int r, count; sched_setscheduler(current, SCHED_FIFO, &sparam); @@ -503,10 +506,7 @@ static int amd_sched_main(void *param) s_fence = sched_job->s_fence; atomic_inc(&sched->hw_rq_count); - spin_lock_irqsave(&sched->job_list_lock, flags); - list_add_tail(&sched_job->node, &sched->ring_mirror_list); amd_sched_job_begin(sched_job); - spin_unlock_irqrestore(&sched->job_list_lock, flags); fence = sched->ops->run_job(sched_job); amd_sched_fence_scheduled(s_fence); -- cgit v1.2.3 From c5f74f7802775b9ccdb0a4fd90e0c7d0b03da9fa Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 19 May 2016 09:54:15 +0200 Subject: drm/amdgpu: fix and cleanup job destruction MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Remove the job reference counting and just properly destroy it from a work item which blocks on any potential running timeout handler. Signed-off-by: Christian König Reviewed-by: Monk.Liu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 1 - drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 3 +- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 18 +++-------- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 46 +++++++++++---------------- drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 25 +++------------ 5 files changed, 30 insertions(+), 63 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index a5d1cfbb1a87..ea64c6569704 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -755,7 +755,6 @@ int amdgpu_job_alloc_with_ib(struct amdgpu_device *adev, unsigned size, struct amdgpu_job **job); void amdgpu_job_free(struct amdgpu_job *job); -void amdgpu_job_free_func(struct kref *refcount); int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, struct amd_sched_entity *entity, void *owner, struct fence **f); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 72694d7f11f5..523da20e6ea0 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -838,8 +838,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, p->job = NULL; r = amd_sched_job_init(&job->base, &ring->sched, - entity, amdgpu_job_free_func, - p->filp, &fence); + entity, p->filp, &fence); if (r) { amdgpu_job_free(job); return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index 32132f2e236d..34cd971a9afa 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -28,12 +28,6 @@ #include "amdgpu.h" #include "amdgpu_trace.h" -static void amdgpu_job_free_handler(struct work_struct *ws) -{ - struct amdgpu_job *job = container_of(ws, struct amdgpu_job, base.work_free_job); - amd_sched_job_put(&job->base); -} - static void amdgpu_job_timedout(struct amd_sched_job *s_job) { struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base); @@ -42,8 +36,6 @@ static void amdgpu_job_timedout(struct amd_sched_job *s_job) job->base.sched->name, atomic_read(&job->ring->fence_drv.last_seq), job->ring->fence_drv.sync_seq); - - amd_sched_job_put(&job->base); } int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, @@ -64,7 +56,6 @@ int amdgpu_job_alloc(struct amdgpu_device *adev, unsigned num_ibs, (*job)->vm = vm; (*job)->ibs = (void *)&(*job)[1]; (*job)->num_ibs = num_ibs; - INIT_WORK(&(*job)->base.work_free_job, amdgpu_job_free_handler); amdgpu_sync_create(&(*job)->sync); @@ -103,9 +94,10 @@ static void amdgpu_job_free_resources(struct amdgpu_job *job) amdgpu_sync_free(&job->sync); } -void amdgpu_job_free_func(struct kref *refcount) +void amdgpu_job_free_cb(struct amd_sched_job *s_job) { - struct amdgpu_job *job = container_of(refcount, struct amdgpu_job, base.refcount); + struct amdgpu_job *job = container_of(s_job, struct amdgpu_job, base); + kfree(job); } @@ -126,8 +118,7 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, if (!f) return -EINVAL; - r = amd_sched_job_init(&job->base, &ring->sched, - entity, amdgpu_job_free_func, owner, &fence); + r = amd_sched_job_init(&job->base, &ring->sched, entity, owner, &fence); if (r) return r; @@ -198,4 +189,5 @@ const struct amd_sched_backend_ops amdgpu_sched_ops = { .dependency = amdgpu_job_dependency, .run_job = amdgpu_job_run, .timedout_job = amdgpu_job_timedout, + .free_job = amdgpu_job_free_cb }; diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index cb56d9065e43..2425172e612e 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -319,19 +319,13 @@ static bool amd_sched_entity_in(struct amd_sched_job *sched_job) return added; } -static void amd_sched_free_job(struct fence *f, struct fence_cb *cb) { - struct amd_sched_job *job = container_of(cb, struct amd_sched_job, - cb_free_job); - - schedule_work(&job->work_free_job); -} - /* job_finish is called after hw fence signaled, and * the job had already been deleted from ring_mirror_list */ -static void amd_sched_job_finish(struct amd_sched_job *s_job) +static void amd_sched_job_finish(struct work_struct *work) { - struct amd_sched_job *next; + struct amd_sched_job *s_job = container_of(work, struct amd_sched_job, + finish_work); struct amd_gpu_scheduler *sched = s_job->sched; unsigned long flags; @@ -339,19 +333,26 @@ static void amd_sched_job_finish(struct amd_sched_job *s_job) spin_lock_irqsave(&sched->job_list_lock, flags); list_del_init(&s_job->node); if (sched->timeout != MAX_SCHEDULE_TIMEOUT) { - if (cancel_delayed_work(&s_job->work_tdr)) - amd_sched_job_put(s_job); + struct amd_sched_job *next; + + cancel_delayed_work_sync(&s_job->work_tdr); /* queue TDR for next job */ next = list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node); - if (next) { - amd_sched_job_get(next); + if (next) schedule_delayed_work(&next->work_tdr, sched->timeout); - } } spin_unlock_irqrestore(&sched->job_list_lock, flags); + sched->ops->free_job(s_job); +} + +static void amd_sched_job_finish_cb(struct fence *f, struct fence_cb *cb) +{ + struct amd_sched_job *job = container_of(cb, struct amd_sched_job, + finish_cb); + schedule_work(&job->finish_work); } static void amd_sched_job_begin(struct amd_sched_job *s_job) @@ -364,10 +365,7 @@ static void amd_sched_job_begin(struct amd_sched_job *s_job) if (sched->timeout != MAX_SCHEDULE_TIMEOUT && list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node) == s_job) - { - amd_sched_job_get(s_job); schedule_delayed_work(&s_job->work_tdr, sched->timeout); - } spin_unlock_irqrestore(&sched->job_list_lock, flags); } @@ -390,9 +388,9 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) { struct amd_sched_entity *entity = sched_job->s_entity; - fence_add_callback(&sched_job->s_fence->base, - &sched_job->cb_free_job, amd_sched_free_job); trace_amd_sched_job(sched_job); + fence_add_callback(&sched_job->s_fence->base, &sched_job->finish_cb, + amd_sched_job_finish_cb); wait_event(entity->sched->job_scheduled, amd_sched_entity_in(sched_job)); } @@ -401,20 +399,17 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void (*free_cb)(struct kref *refcount), void *owner, struct fence **fence) { - INIT_LIST_HEAD(&job->node); - kref_init(&job->refcount); job->sched = sched; job->s_entity = entity; job->s_fence = amd_sched_fence_create(entity, owner); if (!job->s_fence) return -ENOMEM; - job->s_fence->s_job = job; + INIT_WORK(&job->finish_work, amd_sched_job_finish); + INIT_LIST_HEAD(&job->node); INIT_DELAYED_WORK(&job->work_tdr, amd_sched_job_timedout); - job->free_callback = free_cb; if (fence) *fence = &job->s_fence->base; @@ -468,9 +463,6 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) struct amd_gpu_scheduler *sched = s_fence->sched; atomic_dec(&sched->hw_rq_count); - - amd_sched_job_finish(s_fence->s_job); - amd_sched_fence_signal(s_fence); trace_amd_sched_process_job(s_fence); diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index f0de46ce7afe..e63034ec7fa4 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -74,19 +74,16 @@ struct amd_sched_fence { struct amd_gpu_scheduler *sched; spinlock_t lock; void *owner; - struct amd_sched_job *s_job; }; struct amd_sched_job { - struct kref refcount; struct amd_gpu_scheduler *sched; struct amd_sched_entity *s_entity; struct amd_sched_fence *s_fence; - struct fence_cb cb_free_job; - struct work_struct work_free_job; - struct list_head node; - struct delayed_work work_tdr; - void (*free_callback)(struct kref *refcount); + struct fence_cb finish_cb; + struct work_struct finish_work; + struct list_head node; + struct delayed_work work_tdr; }; extern const struct fence_ops amd_sched_fence_ops; @@ -109,6 +106,7 @@ struct amd_sched_backend_ops { struct fence *(*dependency)(struct amd_sched_job *sched_job); struct fence *(*run_job)(struct amd_sched_job *sched_job); void (*timedout_job)(struct amd_sched_job *sched_job); + void (*free_job)(struct amd_sched_job *sched_job); }; enum amd_sched_priority { @@ -154,18 +152,5 @@ void amd_sched_fence_signal(struct amd_sched_fence *fence); int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void (*free_cb)(struct kref* refcount), void *owner, struct fence **fence); -static inline void amd_sched_job_get(struct amd_sched_job *job) -{ - if (job) - kref_get(&job->refcount); -} - -static inline void amd_sched_job_put(struct amd_sched_job *job) -{ - if (job) - kref_put(&job->refcount, job->free_callback); -} - #endif -- cgit v1.2.3 From 6fc1367582534a54476c701326f00e7d2ec81f22 Mon Sep 17 00:00:00 2001 From: Christian König Date: Fri, 20 May 2016 12:53:52 +0200 Subject: drm/amdgpu: generalize the scheduler fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Make it two events, one for the job being scheduled and one when it is finished. Acked-by: Alex Deucher Signed-off-by: Christian König Acked-by: Chunming Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h | 4 +- drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h | 4 +- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 34 +++++++------ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 19 ++++---- drivers/gpu/drm/amd/scheduler/sched_fence.c | 63 ++++++++++++++++++------- 6 files changed, 79 insertions(+), 49 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index d4791a7f8075..ddfed939c893 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -85,7 +85,7 @@ static void amdgpu_job_free_resources(struct amdgpu_job *job) unsigned i; /* use sched fence if available */ - f = job->base.s_fence ? &job->base.s_fence->base : job->fence; + f = job->base.s_fence ? &job->base.s_fence->finished : job->fence; for (i = 0; i < job->num_ibs; ++i) amdgpu_ib_free(job->adev, &job->ibs[i], f); @@ -143,7 +143,7 @@ static struct fence *amdgpu_job_dependency(struct amd_sched_job *sched_job) int r; r = amdgpu_vm_grab_id(vm, ring, &job->sync, - &job->base.s_fence->base, + &job->base.s_fence->finished, &job->vm_id, &job->vm_pd_addr); if (r) DRM_ERROR("Error getting VM ID (%d)\n", r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h index 38e56890043c..ecd08f8914a3 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_trace.h @@ -102,7 +102,7 @@ TRACE_EVENT(amdgpu_cs_ioctl, __entry->adev = job->adev; __entry->sched_job = &job->base; __entry->ib = job->ibs; - __entry->fence = &job->base.s_fence->base; + __entry->fence = &job->base.s_fence->finished; __entry->ring_name = job->ring->name; __entry->num_ibs = job->num_ibs; ), @@ -127,7 +127,7 @@ TRACE_EVENT(amdgpu_sched_run_job, __entry->adev = job->adev; __entry->sched_job = &job->base; __entry->ib = job->ibs; - __entry->fence = &job->base.s_fence->base; + __entry->fence = &job->base.s_fence->finished; __entry->ring_name = job->ring->name; __entry->num_ibs = job->num_ibs; ), diff --git a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h index c89dc777768f..b961a1c6caf3 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_sched_trace.h @@ -26,7 +26,7 @@ TRACE_EVENT(amd_sched_job, TP_fast_assign( __entry->entity = sched_job->s_entity; __entry->sched_job = sched_job; - __entry->fence = &sched_job->s_fence->base; + __entry->fence = &sched_job->s_fence->finished; __entry->name = sched_job->sched->name; __entry->job_count = kfifo_len( &sched_job->s_entity->job_queue) / sizeof(sched_job); @@ -46,7 +46,7 @@ TRACE_EVENT(amd_sched_process_job, ), TP_fast_assign( - __entry->fence = &fence->base; + __entry->fence = &fence->finished; ), TP_printk("fence=%p signaled", __entry->fence) ); diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 2425172e612e..74aa0b32f3e3 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -140,7 +140,7 @@ int amd_sched_entity_init(struct amd_gpu_scheduler *sched, return r; atomic_set(&entity->fence_seq, 0); - entity->fence_context = fence_context_alloc(1); + entity->fence_context = fence_context_alloc(2); return 0; } @@ -251,17 +251,21 @@ static bool amd_sched_entity_add_dependency_cb(struct amd_sched_entity *entity) s_fence = to_amd_sched_fence(fence); if (s_fence && s_fence->sched == sched) { - /* Fence is from the same scheduler */ - if (test_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &fence->flags)) { - /* Ignore it when it is already scheduled */ - fence_put(entity->dependency); - return false; - } - /* Wait for fence to be scheduled */ - entity->cb.func = amd_sched_entity_clear_dep; - list_add_tail(&entity->cb.node, &s_fence->scheduled_cb); - return true; + /* + * Fence is from the same scheduler, only need to wait for + * it to be scheduled + */ + fence = fence_get(&s_fence->scheduled); + fence_put(entity->dependency); + entity->dependency = fence; + if (!fence_add_callback(fence, &entity->cb, + amd_sched_entity_clear_dep)) + return true; + + /* Ignore it when it is already scheduled */ + fence_put(fence); + return false; } if (!fence_add_callback(entity->dependency, &entity->cb, @@ -389,7 +393,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) struct amd_sched_entity *entity = sched_job->s_entity; trace_amd_sched_job(sched_job); - fence_add_callback(&sched_job->s_fence->base, &sched_job->finish_cb, + fence_add_callback(&sched_job->s_fence->finished, &sched_job->finish_cb, amd_sched_job_finish_cb); wait_event(entity->sched->job_scheduled, amd_sched_entity_in(sched_job)); @@ -412,7 +416,7 @@ int amd_sched_job_init(struct amd_sched_job *job, INIT_DELAYED_WORK(&job->work_tdr, amd_sched_job_timedout); if (fence) - *fence = &job->s_fence->base; + *fence = &job->s_fence->finished; return 0; } @@ -463,10 +467,10 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) struct amd_gpu_scheduler *sched = s_fence->sched; atomic_dec(&sched->hw_rq_count); - amd_sched_fence_signal(s_fence); + amd_sched_fence_finished(s_fence); trace_amd_sched_process_job(s_fence); - fence_put(&s_fence->base); + fence_put(&s_fence->finished); wake_up_interruptible(&sched->wake_up_worker); } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index e63034ec7fa4..3e989b153282 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -27,8 +27,6 @@ #include #include -#define AMD_SCHED_FENCE_SCHEDULED_BIT FENCE_FLAG_USER_BITS - struct amd_gpu_scheduler; struct amd_sched_rq; @@ -68,9 +66,9 @@ struct amd_sched_rq { }; struct amd_sched_fence { - struct fence base; + struct fence scheduled; + struct fence finished; struct fence_cb cb; - struct list_head scheduled_cb; struct amd_gpu_scheduler *sched; spinlock_t lock; void *owner; @@ -86,14 +84,15 @@ struct amd_sched_job { struct delayed_work work_tdr; }; -extern const struct fence_ops amd_sched_fence_ops; +extern const struct fence_ops amd_sched_fence_ops_scheduled; +extern const struct fence_ops amd_sched_fence_ops_finished; static inline struct amd_sched_fence *to_amd_sched_fence(struct fence *f) { - struct amd_sched_fence *__f = container_of(f, struct amd_sched_fence, - base); + if (f->ops == &amd_sched_fence_ops_scheduled) + return container_of(f, struct amd_sched_fence, scheduled); - if (__f->base.ops == &amd_sched_fence_ops) - return __f; + if (f->ops == &amd_sched_fence_ops_finished) + return container_of(f, struct amd_sched_fence, finished); return NULL; } @@ -148,7 +147,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job); struct amd_sched_fence *amd_sched_fence_create( struct amd_sched_entity *s_entity, void *owner); void amd_sched_fence_scheduled(struct amd_sched_fence *fence); -void amd_sched_fence_signal(struct amd_sched_fence *fence); +void amd_sched_fence_finished(struct amd_sched_fence *fence); int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c index 71931bc40069..a5e3fef622fb 100644 --- a/drivers/gpu/drm/amd/scheduler/sched_fence.c +++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c @@ -37,36 +37,37 @@ struct amd_sched_fence *amd_sched_fence_create(struct amd_sched_entity *entity, if (fence == NULL) return NULL; - INIT_LIST_HEAD(&fence->scheduled_cb); fence->owner = owner; fence->sched = entity->sched; spin_lock_init(&fence->lock); seq = atomic_inc_return(&entity->fence_seq); - fence_init(&fence->base, &amd_sched_fence_ops, &fence->lock, - entity->fence_context, seq); + fence_init(&fence->scheduled, &amd_sched_fence_ops_scheduled, + &fence->lock, entity->fence_context, seq); + fence_init(&fence->finished, &amd_sched_fence_ops_finished, + &fence->lock, entity->fence_context + 1, seq); return fence; } -void amd_sched_fence_signal(struct amd_sched_fence *fence) +void amd_sched_fence_scheduled(struct amd_sched_fence *fence) { - int ret = fence_signal(&fence->base); + int ret = fence_signal(&fence->scheduled); + if (!ret) - FENCE_TRACE(&fence->base, "signaled from irq context\n"); + FENCE_TRACE(&fence->scheduled, "signaled from irq context\n"); else - FENCE_TRACE(&fence->base, "was already signaled\n"); + FENCE_TRACE(&fence->scheduled, "was already signaled\n"); } -void amd_sched_fence_scheduled(struct amd_sched_fence *s_fence) +void amd_sched_fence_finished(struct amd_sched_fence *fence) { - struct fence_cb *cur, *tmp; + int ret = fence_signal(&fence->finished); - set_bit(AMD_SCHED_FENCE_SCHEDULED_BIT, &s_fence->base.flags); - list_for_each_entry_safe(cur, tmp, &s_fence->scheduled_cb, node) { - list_del_init(&cur->node); - cur->func(&s_fence->base, cur); - } + if (!ret) + FENCE_TRACE(&fence->finished, "signaled from irq context\n"); + else + FENCE_TRACE(&fence->finished, "was already signaled\n"); } static const char *amd_sched_fence_get_driver_name(struct fence *fence) @@ -96,6 +97,7 @@ static void amd_sched_fence_free(struct rcu_head *rcu) { struct fence *f = container_of(rcu, struct fence, rcu); struct amd_sched_fence *fence = to_amd_sched_fence(f); + kmem_cache_free(sched_fence_slab, fence); } @@ -107,16 +109,41 @@ static void amd_sched_fence_free(struct rcu_head *rcu) * This function is called when the reference count becomes zero. * It just RCU schedules freeing up the fence. */ -static void amd_sched_fence_release(struct fence *f) +static void amd_sched_fence_release_scheduled(struct fence *f) +{ + struct amd_sched_fence *fence = to_amd_sched_fence(f); + + call_rcu(&fence->finished.rcu, amd_sched_fence_free); +} + +/** + * amd_sched_fence_release_scheduled - drop extra reference + * + * @f: fence + * + * Drop the extra reference from the scheduled fence to the base fence. + */ +static void amd_sched_fence_release_finished(struct fence *f) { - call_rcu(&f->rcu, amd_sched_fence_free); + struct amd_sched_fence *fence = to_amd_sched_fence(f); + + fence_put(&fence->scheduled); } -const struct fence_ops amd_sched_fence_ops = { +const struct fence_ops amd_sched_fence_ops_scheduled = { + .get_driver_name = amd_sched_fence_get_driver_name, + .get_timeline_name = amd_sched_fence_get_timeline_name, + .enable_signaling = amd_sched_fence_enable_signaling, + .signaled = NULL, + .wait = fence_default_wait, + .release = amd_sched_fence_release_scheduled, +}; + +const struct fence_ops amd_sched_fence_ops_finished = { .get_driver_name = amd_sched_fence_get_driver_name, .get_timeline_name = amd_sched_fence_get_timeline_name, .enable_signaling = amd_sched_fence_enable_signaling, .signaled = NULL, .wait = fence_default_wait, - .release = amd_sched_fence_release, + .release = amd_sched_fence_release_finished, }; -- cgit v1.2.3 From a8bd3e1c71c342782b6597ac6bf4680d1f13254d Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 13 Jun 2016 16:12:42 +0200 Subject: drm/amdgpu: stop trying to schedule() with a spin held MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Drop the lock before calling cancel_delayed_work_sync(). Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96445 Reviewed-by: Alex Deucher Signed-off-by: Christian König Tested-by: Nicolai Hähnle Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 74aa0b32f3e3..b1d49c5d8e44 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -339,7 +339,9 @@ static void amd_sched_job_finish(struct work_struct *work) if (sched->timeout != MAX_SCHEDULE_TIMEOUT) { struct amd_sched_job *next; + spin_unlock_irqrestore(&sched->job_list_lock, flags); cancel_delayed_work_sync(&s_job->work_tdr); + spin_lock_irqsave(&sched->job_list_lock, flags); /* queue TDR for next job */ next = list_first_entry_or_null(&sched->ring_mirror_list, -- cgit v1.2.3 From 0875dc9e80eb3b01406b21947c140265527488ba Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Sun, 12 Jun 2016 15:41:58 +0800 Subject: drm/amdgpu: block scheduler when gpu reset MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Signed-off-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 15 +++++++++++++-- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 17 ++++++++++++++--- 2 files changed, 27 insertions(+), 5 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index ae801e9cec94..a7a84286a06f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -25,6 +25,7 @@ * Alex Deucher * Jerome Glisse */ +#include #include #include #include @@ -1895,6 +1896,14 @@ int amdgpu_gpu_reset(struct amdgpu_device *adev) atomic_inc(&adev->gpu_reset_counter); + /* block scheduler */ + for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { + struct amdgpu_ring *ring = adev->rings[i]; + + if (!ring) + continue; + kthread_park(ring->sched.thread); + } /* block TTM */ resched = ttm_bo_lock_delayed_workqueue(&adev->mman.bdev); @@ -1928,7 +1937,7 @@ retry: struct amdgpu_ring *ring = adev->rings[i]; if (!ring) continue; - + kthread_unpark(ring->sched.thread); amdgpu_ring_restore(ring, ring_sizes[i], ring_data[i]); ring_sizes[i] = 0; ring_data[i] = NULL; @@ -1946,8 +1955,10 @@ retry: } else { amdgpu_fence_driver_force_completion(adev); for (i = 0; i < AMDGPU_MAX_RINGS; ++i) { - if (adev->rings[i]) + if (adev->rings[i]) { + kthread_unpark(adev->rings[i]->sched.thread); kfree(ring_data[i]); + } } } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index b1d49c5d8e44..60f58f76a1eb 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -476,6 +476,16 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) wake_up_interruptible(&sched->wake_up_worker); } +static bool amd_sched_blocked(struct amd_gpu_scheduler *sched) +{ + if (kthread_should_park()) { + kthread_parkme(); + return true; + } + + return false; +} + static int amd_sched_main(void *param) { struct sched_param sparam = {.sched_priority = 1}; @@ -485,14 +495,15 @@ static int amd_sched_main(void *param) sched_setscheduler(current, SCHED_FIFO, &sparam); while (!kthread_should_stop()) { - struct amd_sched_entity *entity; + struct amd_sched_entity *entity = NULL; struct amd_sched_fence *s_fence; struct amd_sched_job *sched_job; struct fence *fence; wait_event_interruptible(sched->wake_up_worker, - (entity = amd_sched_select_entity(sched)) || - kthread_should_stop()); + (!amd_sched_blocked(sched) && + (entity = amd_sched_select_entity(sched))) || + kthread_should_stop()); if (!entity) continue; -- cgit v1.2.3 From 1059e117cce1b876c419db9ead2edf03ac2ddbc4 Mon Sep 17 00:00:00 2001 From: Christian König Date: Mon, 13 Jun 2016 16:12:43 +0200 Subject: drm/amdgpu: stop disabling irqs when it isn't neccessary MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit A regular spin_lock/unlock should do here as well. Signed-off-by: Christian König Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 14 ++++++-------- 1 file changed, 6 insertions(+), 8 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 60f58f76a1eb..2baedc4e667d 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -331,17 +331,16 @@ static void amd_sched_job_finish(struct work_struct *work) struct amd_sched_job *s_job = container_of(work, struct amd_sched_job, finish_work); struct amd_gpu_scheduler *sched = s_job->sched; - unsigned long flags; /* remove job from ring_mirror_list */ - spin_lock_irqsave(&sched->job_list_lock, flags); + spin_lock(&sched->job_list_lock); list_del_init(&s_job->node); if (sched->timeout != MAX_SCHEDULE_TIMEOUT) { struct amd_sched_job *next; - spin_unlock_irqrestore(&sched->job_list_lock, flags); + spin_unlock(&sched->job_list_lock); cancel_delayed_work_sync(&s_job->work_tdr); - spin_lock_irqsave(&sched->job_list_lock, flags); + spin_lock(&sched->job_list_lock); /* queue TDR for next job */ next = list_first_entry_or_null(&sched->ring_mirror_list, @@ -350,7 +349,7 @@ static void amd_sched_job_finish(struct work_struct *work) if (next) schedule_delayed_work(&next->work_tdr, sched->timeout); } - spin_unlock_irqrestore(&sched->job_list_lock, flags); + spin_unlock(&sched->job_list_lock); sched->ops->free_job(s_job); } @@ -364,15 +363,14 @@ static void amd_sched_job_finish_cb(struct fence *f, struct fence_cb *cb) static void amd_sched_job_begin(struct amd_sched_job *s_job) { struct amd_gpu_scheduler *sched = s_job->sched; - unsigned long flags; - spin_lock_irqsave(&sched->job_list_lock, flags); + spin_lock(&sched->job_list_lock); list_add_tail(&s_job->node, &sched->ring_mirror_list); if (sched->timeout != MAX_SCHEDULE_TIMEOUT && list_first_entry_or_null(&sched->ring_mirror_list, struct amd_sched_job, node) == s_job) schedule_delayed_work(&s_job->work_tdr, sched->timeout); - spin_unlock_irqrestore(&sched->job_list_lock, flags); + spin_unlock(&sched->job_list_lock); } static void amd_sched_job_timedout(struct work_struct *work) -- cgit v1.2.3 From 595a9cd68c132e474ee5daf97067d4d15c618739 Mon Sep 17 00:00:00 2001 From: Christian König Date: Thu, 30 Jun 2016 10:52:03 +0200 Subject: drm/amdgpu: remove fence parameter from amd_sched_job_init MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit We return the fence as part of the job structur anyway, no need to do this twice. Signed-off-by: Christian König Reviewed-by: Chunming Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c | 8 +++----- drivers/gpu/drm/amd/amdgpu/amdgpu_job.c | 5 ++--- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 4 +--- drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 2 +- 4 files changed, 7 insertions(+), 12 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c index 0db32e3b13c0..0307ff5887c5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c @@ -836,15 +836,13 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, { struct amdgpu_ring *ring = p->job->ring; struct amd_sched_entity *entity = &p->ctx->rings[ring->idx].entity; - struct fence *fence; struct amdgpu_job *job; int r; job = p->job; p->job = NULL; - r = amd_sched_job_init(&job->base, &ring->sched, - entity, p->filp, &fence); + r = amd_sched_job_init(&job->base, &ring->sched, entity, p->filp); if (r) { amdgpu_job_free(job); return r; @@ -852,8 +850,8 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p, job->owner = p->filp; job->ctx = entity->fence_context; - p->fence = fence_get(fence); - cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, fence); + p->fence = fence_get(&job->base.s_fence->finished); + cs->out.handle = amdgpu_ctx_add_fence(p->ctx, ring, p->fence); job->uf_sequence = cs->out.handle; amdgpu_job_free_resources(job); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c index cdcfda6618d0..327f4df55c86 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_job.c @@ -113,20 +113,19 @@ int amdgpu_job_submit(struct amdgpu_job *job, struct amdgpu_ring *ring, struct amd_sched_entity *entity, void *owner, struct fence **f) { - struct fence *fence; int r; job->ring = ring; if (!f) return -EINVAL; - r = amd_sched_job_init(&job->base, &ring->sched, entity, owner, &fence); + r = amd_sched_job_init(&job->base, &ring->sched, entity, owner); if (r) return r; job->owner = owner; job->ctx = entity->fence_context; - *f = fence_get(fence); + *f = fence_get(&job->base.s_fence->finished); amdgpu_job_free_resources(job); amd_sched_entity_push_job(&job->base); diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 2baedc4e667d..3405372cda0c 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -403,7 +403,7 @@ void amd_sched_entity_push_job(struct amd_sched_job *sched_job) int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void *owner, struct fence **fence) + void *owner) { job->sched = sched; job->s_entity = entity; @@ -415,8 +415,6 @@ int amd_sched_job_init(struct amd_sched_job *job, INIT_LIST_HEAD(&job->node); INIT_DELAYED_WORK(&job->work_tdr, amd_sched_job_timedout); - if (fence) - *fence = &job->s_fence->finished; return 0; } diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 3e989b153282..9024dc5592e3 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -151,5 +151,5 @@ void amd_sched_fence_finished(struct amd_sched_fence *fence); int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, - void *owner, struct fence **fence); + void *owner); #endif -- cgit v1.2.3 From 754ce0fa55c4e464626e961ddb98f935ab85aa19 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Thu, 30 Jun 2016 11:23:31 +0800 Subject: drm/amd: add parent for sched fence MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Parent of sched fence is hw fence which is to signal sched fence. Signed-off-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 1 + drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 + drivers/gpu/drm/amd/scheduler/sched_fence.c | 1 + 3 files changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 3405372cda0c..717655d0c642 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -516,6 +516,7 @@ static int amd_sched_main(void *param) fence = sched->ops->run_job(sched_job); amd_sched_fence_scheduled(s_fence); if (fence) { + s_fence->parent = fence_get(fence); r = fence_add_callback(fence, &s_fence->cb, amd_sched_process_job); if (r == -ENOENT) diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 9024dc5592e3..545fda73a05f 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -69,6 +69,7 @@ struct amd_sched_fence { struct fence scheduled; struct fence finished; struct fence_cb cb; + struct fence *parent; struct amd_gpu_scheduler *sched; spinlock_t lock; void *owner; diff --git a/drivers/gpu/drm/amd/scheduler/sched_fence.c b/drivers/gpu/drm/amd/scheduler/sched_fence.c index a5e3fef622fb..6b63beaf7574 100644 --- a/drivers/gpu/drm/amd/scheduler/sched_fence.c +++ b/drivers/gpu/drm/amd/scheduler/sched_fence.c @@ -98,6 +98,7 @@ static void amd_sched_fence_free(struct rcu_head *rcu) struct fence *f = container_of(rcu, struct fence, rcu); struct amd_sched_fence *fence = to_amd_sched_fence(f); + fence_put(fence->parent); kmem_cache_free(sched_fence_slab, fence); } -- cgit v1.2.3 From e686e75dacb262a7d5f6673e1b9bfe0e62cd287a Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Thu, 30 Jun 2016 11:30:37 +0800 Subject: drm/amd: add amd_sched_hw_job_reset MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit amd_sched_hw_job_reset will remove callback from hw fence. Signed-off-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 14 ++++++++++++++ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 + 2 files changed, 15 insertions(+) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 717655d0c642..cf2d64bef0df 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -381,6 +381,20 @@ static void amd_sched_job_timedout(struct work_struct *work) job->sched->ops->timedout_job(job); } +void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched) +{ + struct amd_sched_job *s_job; + + spin_lock(&sched->job_list_lock); + list_for_each_entry_reverse(s_job, &sched->ring_mirror_list, node) { + if (fence_remove_callback(s_job->s_fence->parent, &s_job->s_fence->cb)) { + fence_put(s_job->s_fence->parent); + s_job->s_fence->parent = NULL; + } + } + spin_unlock(&sched->job_list_lock); +} + /** * Submit a job to the job queue * diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index 545fda73a05f..fdcd8fbf5e26 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -153,4 +153,5 @@ int amd_sched_job_init(struct amd_sched_job *job, struct amd_gpu_scheduler *sched, struct amd_sched_entity *entity, void *owner); +void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched); #endif -- cgit v1.2.3 From ec75f573c358bf4b36148659534b3674b80f7a09 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Wed, 29 Jun 2016 15:23:55 +0800 Subject: drm/amdgpu: add amd_sched_job_recovery MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Which is to recover hw jobs when gpu reset. Signed-off-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 33 +++++++++++++++++++++++++++ drivers/gpu/drm/amd/scheduler/gpu_scheduler.h | 1 + 2 files changed, 34 insertions(+) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index cf2d64bef0df..70ff09d10885 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -32,6 +32,7 @@ static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity); static void amd_sched_wakeup(struct amd_gpu_scheduler *sched); +static void amd_sched_process_job(struct fence *f, struct fence_cb *cb); struct kmem_cache *sched_fence_slab; atomic_t sched_fence_slab_ref = ATOMIC_INIT(0); @@ -395,6 +396,38 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched) spin_unlock(&sched->job_list_lock); } +void amd_sched_job_recovery(struct amd_gpu_scheduler *sched) +{ + struct amd_sched_job *s_job; + int r; + + spin_lock(&sched->job_list_lock); + s_job = list_first_entry_or_null(&sched->ring_mirror_list, + struct amd_sched_job, node); + if (s_job) + schedule_delayed_work(&s_job->work_tdr, sched->timeout); + + list_for_each_entry(s_job, &sched->ring_mirror_list, node) { + struct amd_sched_fence *s_fence = s_job->s_fence; + struct fence *fence = sched->ops->run_job(s_job); + if (fence) { + s_fence->parent = fence_get(fence); + r = fence_add_callback(fence, &s_fence->cb, + amd_sched_process_job); + if (r == -ENOENT) + amd_sched_process_job(fence, &s_fence->cb); + else if (r) + DRM_ERROR("fence add callback failed (%d)\n", + r); + fence_put(fence); + } else { + DRM_ERROR("Failed to run job!\n"); + amd_sched_process_job(NULL, &s_fence->cb); + } + } + spin_unlock(&sched->job_list_lock); +} + /** * Submit a job to the job queue * diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h index fdcd8fbf5e26..7cbbbfb502ef 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.h @@ -154,4 +154,5 @@ int amd_sched_job_init(struct amd_sched_job *job, struct amd_sched_entity *entity, void *owner); void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched); +void amd_sched_job_recovery(struct amd_gpu_scheduler *sched); #endif -- cgit v1.2.3 From bdc2eea4722b96f54edb7588078540505b5ac096 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Fri, 22 Jul 2016 13:01:02 +0800 Subject: drm/amd: reset hw count when reset job MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Means the hw ring is empty after gpu reset. Signed-off-by: Chunming Zhou Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 3 +++ 1 file changed, 3 insertions(+) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 70ff09d10885..21c49d38fcb3 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -393,6 +393,7 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched) s_job->s_fence->parent = NULL; } } + atomic_set(&sched->hw_rq_count, 0); spin_unlock(&sched->job_list_lock); } @@ -410,6 +411,8 @@ void amd_sched_job_recovery(struct amd_gpu_scheduler *sched) list_for_each_entry(s_job, &sched->ring_mirror_list, node) { struct amd_sched_fence *s_fence = s_job->s_fence; struct fence *fence = sched->ops->run_job(s_job); + + atomic_inc(&sched->hw_rq_count); if (fence) { s_fence->parent = fence_get(fence); r = fence_add_callback(fence, &s_fence->cb, -- cgit v1.2.3 From 1c62cf91e6ac32346b2dd3bb9d739219a9cc39d0 Mon Sep 17 00:00:00 2001 From: Chunming Zhou Date: Mon, 25 Jul 2016 13:55:35 +0800 Subject: drm/amd: fix deadlock of job_list_lock V2 run_job involves mutex, which could sleep. V2: use list_for_each_entry_safe, since the job might complete while we dropped the lock. Signed-off-by: Chunming Zhou Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c') diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index 21c49d38fcb3..ef312bb75fda 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -399,7 +399,7 @@ void amd_sched_hw_job_reset(struct amd_gpu_scheduler *sched) void amd_sched_job_recovery(struct amd_gpu_scheduler *sched) { - struct amd_sched_job *s_job; + struct amd_sched_job *s_job, *tmp; int r; spin_lock(&sched->job_list_lock); @@ -408,10 +408,12 @@ void amd_sched_job_recovery(struct amd_gpu_scheduler *sched) if (s_job) schedule_delayed_work(&s_job->work_tdr, sched->timeout); - list_for_each_entry(s_job, &sched->ring_mirror_list, node) { + list_for_each_entry_safe(s_job, tmp, &sched->ring_mirror_list, node) { struct amd_sched_fence *s_fence = s_job->s_fence; - struct fence *fence = sched->ops->run_job(s_job); + struct fence *fence; + spin_unlock(&sched->job_list_lock); + fence = sched->ops->run_job(s_job); atomic_inc(&sched->hw_rq_count); if (fence) { s_fence->parent = fence_get(fence); @@ -427,6 +429,7 @@ void amd_sched_job_recovery(struct amd_gpu_scheduler *sched) DRM_ERROR("Failed to run job!\n"); amd_sched_process_job(NULL, &s_fence->cb); } + spin_lock(&sched->job_list_lock); } spin_unlock(&sched->job_list_lock); } -- cgit v1.2.3