diff options
author | Linus Torvalds <torvalds@linux-foundation.org> | 2015-11-29 17:38:08 -0800 |
---|---|---|
committer | Linus Torvalds <torvalds@linux-foundation.org> | 2015-11-29 17:38:08 -0800 |
commit | c5bc1c9305e87d84c20415cc5ed2d3944d5bad8d (patch) | |
tree | 91b35672d7419548d0c532e72eb09a1f2c82b9f4 /drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | |
parent | 818aba30b3a178b350b1d7bd12c3430fbf9c212f (diff) | |
parent | 8c14f72b57570d71afda03c5c51b06f0b4af4367 (diff) |
Merge branch 'drm-fixes' of git://people.freedesktop.org/~airlied/linux
Pull nouveau and radeon fixes from Dave Airlie:
"Just some nouveau and radeon/amdgpu fixes.
The nouveau fixes look large as the firmware context files are
regenerated, but the actual change is quite small"
* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
drm/radeon: make some dpm errors debug only
drm/nouveau/volt/pwm/gk104: fix an off-by-one resulting in the voltage not being set
drm/nouveau/nvif: allow userspace access to its own client object
drm/nouveau/gr/gf100-: fix oops when calling zbc methods
drm/nouveau/gr/gf117-: assume no PPC if NV_PGRAPH_GPC_GPM_PD_PES_TPC_ID_MASK is zero
drm/nouveau/gr/gf117-: read NV_PGRAPH_GPC_GPM_PD_PES_TPC_ID_MASK from correct GPC
drm/nouveau/gr/gf100-: split out per-gpc address calculation macro
drm/nouveau/bios: return actual size of the buffer retrieved via _ROM
drm/nouveau/instmem: protect instobj list with a spinlock
drm/nouveau/pci: enable c800 magic for some unknown Samsung laptop
drm/nouveau/pci: enable c800 magic for Clevo P157SM
drm/radeon: make rv770_set_sw_state failures non-fatal
drm/amdgpu: move dependency handling out of atomic section v2
drm/amdgpu: optimize scheduler fence handling
drm/amdgpu: remove vm->mutex
drm/amdgpu: add mutex for ba_va->valids/invalids
drm/amdgpu: adapt vce session create interface changes
drm/amdgpu: vce use multiple cache surface starting from stoney
drm/amdgpu: reset vce trap interrupt flag
Diffstat (limited to 'drivers/gpu/drm/amd/scheduler/gpu_scheduler.c')
-rw-r--r-- | drivers/gpu/drm/amd/scheduler/gpu_scheduler.c | 122 |
1 files changed, 82 insertions, 40 deletions
diff --git a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c index ea30d6ad4c13..651129f2ec1d 100644 --- a/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c +++ b/drivers/gpu/drm/amd/scheduler/gpu_scheduler.c @@ -30,8 +30,7 @@ #define CREATE_TRACE_POINTS #include "gpu_sched_trace.h" -static struct amd_sched_job * -amd_sched_entity_pop_job(struct amd_sched_entity *entity); +static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity); static void amd_sched_wakeup(struct amd_gpu_scheduler *sched); struct kmem_cache *sched_fence_slab; @@ -64,36 +63,36 @@ static void amd_sched_rq_remove_entity(struct amd_sched_rq *rq, } /** - * Select next job from a specified run queue with round robin policy. - * Return NULL if nothing available. + * Select an entity which could provide a job to run + * + * @rq The run queue to check. + * + * Try to find a ready entity, returns NULL if none found. */ -static struct amd_sched_job * -amd_sched_rq_select_job(struct amd_sched_rq *rq) +static struct amd_sched_entity * +amd_sched_rq_select_entity(struct amd_sched_rq *rq) { struct amd_sched_entity *entity; - struct amd_sched_job *sched_job; spin_lock(&rq->lock); entity = rq->current_entity; if (entity) { list_for_each_entry_continue(entity, &rq->entities, list) { - sched_job = amd_sched_entity_pop_job(entity); - if (sched_job) { + if (amd_sched_entity_is_ready(entity)) { rq->current_entity = entity; spin_unlock(&rq->lock); - return sched_job; + return entity; } } } list_for_each_entry(entity, &rq->entities, list) { - sched_job = amd_sched_entity_pop_job(entity); - if (sched_job) { + if (amd_sched_entity_is_ready(entity)) { rq->current_entity = entity; spin_unlock(&rq->lock); - return sched_job; + return entity; } if (entity == rq->current_entity) @@ -177,6 +176,24 @@ static bool amd_sched_entity_is_idle(struct amd_sched_entity *entity) } /** + * Check if entity is ready + * + * @entity The pointer to a valid scheduler entity + * + * Return true if entity could provide a job. + */ +static bool amd_sched_entity_is_ready(struct amd_sched_entity *entity) +{ + if (kfifo_is_empty(&entity->job_queue)) + return false; + + if (ACCESS_ONCE(entity->dependency)) + return false; + + return true; +} + +/** * Destroy a context entity * * @sched Pointer to scheduler instance @@ -211,32 +228,53 @@ static void amd_sched_entity_wakeup(struct fence *f, struct fence_cb *cb) amd_sched_wakeup(entity->sched); } +static bool amd_sched_entity_add_dependency_cb(struct amd_sched_entity *entity) +{ + struct amd_gpu_scheduler *sched = entity->sched; + struct fence * fence = entity->dependency; + struct amd_sched_fence *s_fence; + + if (fence->context == entity->fence_context) { + /* We can ignore fences from ourself */ + fence_put(entity->dependency); + return false; + } + + 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_wakeup; + list_add_tail(&entity->cb.node, &s_fence->scheduled_cb); + return true; + } + + if (!fence_add_callback(entity->dependency, &entity->cb, + amd_sched_entity_wakeup)) + return true; + + fence_put(entity->dependency); + return false; +} + static struct amd_sched_job * amd_sched_entity_pop_job(struct amd_sched_entity *entity) { struct amd_gpu_scheduler *sched = entity->sched; struct amd_sched_job *sched_job; - if (ACCESS_ONCE(entity->dependency)) - return NULL; - if (!kfifo_out_peek(&entity->job_queue, &sched_job, sizeof(sched_job))) return NULL; - while ((entity->dependency = sched->ops->dependency(sched_job))) { - - if (entity->dependency->context == entity->fence_context) { - /* We can ignore fences from ourself */ - fence_put(entity->dependency); - continue; - } - - if (fence_add_callback(entity->dependency, &entity->cb, - amd_sched_entity_wakeup)) - fence_put(entity->dependency); - else + while ((entity->dependency = sched->ops->dependency(sched_job))) + if (amd_sched_entity_add_dependency_cb(entity)) return NULL; - } return sched_job; } @@ -304,22 +342,22 @@ static void amd_sched_wakeup(struct amd_gpu_scheduler *sched) } /** - * Select next to run + * Select next entity to process */ -static struct amd_sched_job * -amd_sched_select_job(struct amd_gpu_scheduler *sched) +static struct amd_sched_entity * +amd_sched_select_entity(struct amd_gpu_scheduler *sched) { - struct amd_sched_job *sched_job; + struct amd_sched_entity *entity; if (!amd_sched_ready(sched)) return NULL; /* Kernel run queue has higher priority than normal run queue*/ - sched_job = amd_sched_rq_select_job(&sched->kernel_rq); - if (sched_job == NULL) - sched_job = amd_sched_rq_select_job(&sched->sched_rq); + entity = amd_sched_rq_select_entity(&sched->kernel_rq); + if (entity == NULL) + entity = amd_sched_rq_select_entity(&sched->sched_rq); - return sched_job; + return entity; } static void amd_sched_process_job(struct fence *f, struct fence_cb *cb) @@ -381,13 +419,16 @@ static int amd_sched_main(void *param) unsigned long flags; wait_event_interruptible(sched->wake_up_worker, - kthread_should_stop() || - (sched_job = amd_sched_select_job(sched))); + (entity = amd_sched_select_entity(sched)) || + kthread_should_stop()); + if (!entity) + continue; + + sched_job = amd_sched_entity_pop_job(entity); if (!sched_job) continue; - entity = sched_job->s_entity; s_fence = sched_job->s_fence; if (sched->timeout != MAX_SCHEDULE_TIMEOUT) { @@ -400,6 +441,7 @@ static int amd_sched_main(void *param) atomic_inc(&sched->hw_rq_count); fence = sched->ops->run_job(sched_job); + amd_sched_fence_scheduled(s_fence); if (fence) { r = fence_add_callback(fence, &s_fence->cb, amd_sched_process_job); |