From 0388aee766376edfd4638f203b6f6260831665f5 Mon Sep 17 00:00:00 2001 From: Leo Liu Date: Fri, 8 Nov 2019 10:23:14 -0500 Subject: drm/amdgpu: use the JPEG structure for general driver support JPEG1.0 will be functional along with VCN1.0 Signed-off-by: Leo Liu Reviewed-by: Alex Deucher Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 6614d8a6f4c8..8f2eea92d67c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -169,10 +169,10 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, } break; case AMDGPU_HW_IP_VCN_JPEG: - for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { + for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) { if (adev->vcn.harvest_config & (1 << j)) continue; - rings[num_rings++] = &adev->vcn.inst[j].ring_jpeg; + rings[num_rings++] = &adev->jpeg.inst[j].ring_dec; } break; } -- cgit From 3504bd45a99eae374ae54327bf7a6ebf1c8de15a Mon Sep 17 00:00:00 2001 From: Leo Liu Date: Mon, 9 Dec 2019 12:58:24 -0500 Subject: drm/amdgpu: fix JPEG instance checking when ctx init Use proper structure. Fixes: 0388aee766376ed ("drm/amdgpu: use the JPEG structure for general driver support") Signed-off-by: Leo Liu Reviewed-by: Alex Deucher Reviewed-by: James Zhu Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 8f2eea92d67c..1d2bbf10614e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -170,7 +170,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, break; case AMDGPU_HW_IP_VCN_JPEG: for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) { - if (adev->vcn.harvest_config & (1 << j)) + if (adev->jpeg.harvest_config & (1 << j)) continue; rings[num_rings++] = &adev->jpeg.inst[j].ring_dec; } -- cgit From b3ac17667f115e64c67ea6101fc814f47134b530 Mon Sep 17 00:00:00 2001 From: Nirmoy Das Date: Thu, 5 Dec 2019 11:38:00 +0100 Subject: drm/scheduler: rework entity creation MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Entity currently keeps a copy of run_queue list and modify it in drm_sched_entity_set_priority(). Entities shouldn't modify run_queue list. Use drm_gpu_scheduler list instead of drm_sched_rq list in drm_sched_entity struct. In this way we can select a runqueue based on entity/ctx's priority for a drm scheduler. Signed-off-by: Nirmoy Das Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 7 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 8 ++-- drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c | 7 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c | 7 +-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 14 ++++-- drivers/gpu/drm/etnaviv/etnaviv_drv.c | 7 +-- drivers/gpu/drm/lima/lima_sched.c | 5 ++- drivers/gpu/drm/panfrost/panfrost_job.c | 8 ++-- drivers/gpu/drm/scheduler/sched_entity.c | 74 +++++++++++++------------------- drivers/gpu/drm/v3d/v3d_drv.c | 8 ++-- include/drm/gpu_scheduler.h | 8 ++-- 11 files changed, 78 insertions(+), 75 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 1d2bbf10614e..38ec5c919bd9 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -122,7 +122,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { struct amdgpu_ring *rings[AMDGPU_MAX_RINGS]; - struct drm_sched_rq *rqs[AMDGPU_MAX_RINGS]; + struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS]; unsigned num_rings = 0; unsigned num_rqs = 0; @@ -181,12 +181,13 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, if (!rings[j]->adev) continue; - rqs[num_rqs++] = &rings[j]->sched.sched_rq[priority]; + sched_list[num_rqs++] = &rings[j]->sched; } for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) r = drm_sched_entity_init(&ctx->entities[i][j].entity, - rqs, num_rqs, &ctx->guilty); + priority, sched_list, + num_rqs, &ctx->guilty); if (r) goto error_cleanup_entities; } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c index 445de594c214..fbf2961202ee 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c @@ -1987,11 +1987,13 @@ void amdgpu_ttm_set_buffer_funcs_status(struct amdgpu_device *adev, bool enable) if (enable) { struct amdgpu_ring *ring; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; ring = adev->mman.buffer_funcs_ring; - rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_KERNEL]; - r = drm_sched_entity_init(&adev->mman.entity, &rq, 1, NULL); + sched = &ring->sched; + r = drm_sched_entity_init(&adev->mman.entity, + DRM_SCHED_PRIORITY_KERNEL, &sched, + 1, NULL); if (r) { DRM_ERROR("Failed setting up TTM BO move entity (%d)\n", r); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c index d587ffe2af8e..a92f3b18e657 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_uvd.c @@ -330,12 +330,13 @@ int amdgpu_uvd_sw_fini(struct amdgpu_device *adev) int amdgpu_uvd_entity_init(struct amdgpu_device *adev) { struct amdgpu_ring *ring; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; int r; ring = &adev->uvd.inst[0].ring; - rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; - r = drm_sched_entity_init(&adev->uvd.entity, &rq, 1, NULL); + sched = &ring->sched; + r = drm_sched_entity_init(&adev->uvd.entity, DRM_SCHED_PRIORITY_NORMAL, + &sched, 1, NULL); if (r) { DRM_ERROR("Failed setting up UVD kernel entity.\n"); return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c index 46b590af2fd2..ceb0dbf685f1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vce.c @@ -240,12 +240,13 @@ int amdgpu_vce_sw_fini(struct amdgpu_device *adev) int amdgpu_vce_entity_init(struct amdgpu_device *adev) { struct amdgpu_ring *ring; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; int r; ring = &adev->vce.ring[0]; - rq = &ring->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; - r = drm_sched_entity_init(&adev->vce.entity, &rq, 1, NULL); + sched = &ring->sched; + r = drm_sched_entity_init(&adev->vce.entity, DRM_SCHED_PRIORITY_NORMAL, + &sched, 1, NULL); if (r != 0) { DRM_ERROR("Failed setting up VCE run queue.\n"); return r; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 8f26504a59a7..a0be1d2f02dc 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -2740,6 +2740,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, { struct amdgpu_bo_param bp; struct amdgpu_bo *root; + struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS]; int r, i; vm->va = RB_ROOT_CACHED; @@ -2753,14 +2754,19 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, spin_lock_init(&vm->invalidated_lock); INIT_LIST_HEAD(&vm->freed); + for (i = 0; i < adev->vm_manager.vm_pte_num_rqs; i++) + sched_list[i] = adev->vm_manager.vm_pte_rqs[i]->sched; + /* create scheduler entities for page table updates */ - r = drm_sched_entity_init(&vm->direct, adev->vm_manager.vm_pte_rqs, - adev->vm_manager.vm_pte_num_rqs, NULL); + r = drm_sched_entity_init(&vm->direct, DRM_SCHED_PRIORITY_NORMAL, + sched_list, adev->vm_manager.vm_pte_num_rqs, + NULL); if (r) return r; - r = drm_sched_entity_init(&vm->delayed, adev->vm_manager.vm_pte_rqs, - adev->vm_manager.vm_pte_num_rqs, NULL); + r = drm_sched_entity_init(&vm->delayed, DRM_SCHED_PRIORITY_NORMAL, + sched_list, adev->vm_manager.vm_pte_num_rqs, + NULL); if (r) goto error_free_direct; diff --git a/drivers/gpu/drm/etnaviv/etnaviv_drv.c b/drivers/gpu/drm/etnaviv/etnaviv_drv.c index 1f9c01be40d7..76ecdf8bd31c 100644 --- a/drivers/gpu/drm/etnaviv/etnaviv_drv.c +++ b/drivers/gpu/drm/etnaviv/etnaviv_drv.c @@ -65,12 +65,13 @@ static int etnaviv_open(struct drm_device *dev, struct drm_file *file) for (i = 0; i < ETNA_MAX_PIPES; i++) { struct etnaviv_gpu *gpu = priv->gpu[i]; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; if (gpu) { - rq = &gpu->sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; + sched = &gpu->sched; drm_sched_entity_init(&ctx->sched_entity[i], - &rq, 1, NULL); + DRM_SCHED_PRIORITY_NORMAL, &sched, + 1, NULL); } } diff --git a/drivers/gpu/drm/lima/lima_sched.c b/drivers/gpu/drm/lima/lima_sched.c index f522c5f99729..fc8362e4149b 100644 --- a/drivers/gpu/drm/lima/lima_sched.c +++ b/drivers/gpu/drm/lima/lima_sched.c @@ -159,9 +159,10 @@ int lima_sched_context_init(struct lima_sched_pipe *pipe, struct lima_sched_context *context, atomic_t *guilty) { - struct drm_sched_rq *rq = pipe->base.sched_rq + DRM_SCHED_PRIORITY_NORMAL; + struct drm_gpu_scheduler *sched = &pipe->base; - return drm_sched_entity_init(&context->base, &rq, 1, guilty); + return drm_sched_entity_init(&context->base, DRM_SCHED_PRIORITY_NORMAL, + &sched, 1, guilty); } void lima_sched_context_fini(struct lima_sched_pipe *pipe, diff --git a/drivers/gpu/drm/panfrost/panfrost_job.c b/drivers/gpu/drm/panfrost/panfrost_job.c index d411eb6c8eb9..a9ed088ebf08 100644 --- a/drivers/gpu/drm/panfrost/panfrost_job.c +++ b/drivers/gpu/drm/panfrost/panfrost_job.c @@ -542,12 +542,14 @@ int panfrost_job_open(struct panfrost_file_priv *panfrost_priv) { struct panfrost_device *pfdev = panfrost_priv->pfdev; struct panfrost_job_slot *js = pfdev->js; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; int ret, i; for (i = 0; i < NUM_JOB_SLOTS; i++) { - rq = &js->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; - ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], &rq, 1, NULL); + sched = &js->queue[i].sched; + ret = drm_sched_entity_init(&panfrost_priv->sched_entity[i], + DRM_SCHED_PRIORITY_NORMAL, &sched, + 1, NULL); if (WARN_ON(ret)) return ret; } diff --git a/drivers/gpu/drm/scheduler/sched_entity.c b/drivers/gpu/drm/scheduler/sched_entity.c index 461a7a8129f4..f9b6ce29c58f 100644 --- a/drivers/gpu/drm/scheduler/sched_entity.c +++ b/drivers/gpu/drm/scheduler/sched_entity.c @@ -38,9 +38,10 @@ * submit to HW ring. * * @entity: scheduler entity to init - * @rq_list: the list of run queue on which jobs from this + * @priority: priority of the entity + * @sched_list: the list of drm scheds on which jobs from this * entity can be submitted - * @num_rq_list: number of run queue in rq_list + * @num_sched_list: number of drm sched in sched_list * @guilty: atomic_t set to 1 when a job on this queue * is found to be guilty causing a timeout * @@ -50,32 +51,35 @@ * Returns 0 on success or a negative error code on failure. */ int drm_sched_entity_init(struct drm_sched_entity *entity, - struct drm_sched_rq **rq_list, - unsigned int num_rq_list, + enum drm_sched_priority priority, + struct drm_gpu_scheduler **sched_list, + unsigned int num_sched_list, atomic_t *guilty) { int i; - if (!(entity && rq_list && (num_rq_list == 0 || rq_list[0]))) + if (!(entity && sched_list && (num_sched_list == 0 || sched_list[0]))) return -EINVAL; memset(entity, 0, sizeof(struct drm_sched_entity)); INIT_LIST_HEAD(&entity->list); entity->rq = NULL; entity->guilty = guilty; - entity->num_rq_list = num_rq_list; - entity->rq_list = kcalloc(num_rq_list, sizeof(struct drm_sched_rq *), - GFP_KERNEL); - if (!entity->rq_list) + entity->num_sched_list = num_sched_list; + entity->priority = priority; + entity->sched_list = kcalloc(num_sched_list, + sizeof(struct drm_gpu_scheduler *), GFP_KERNEL); + + if(!entity->sched_list) return -ENOMEM; init_completion(&entity->entity_idle); - for (i = 0; i < num_rq_list; ++i) - entity->rq_list[i] = rq_list[i]; + for (i = 0; i < num_sched_list; i++) + entity->sched_list[i] = sched_list[i]; - if (num_rq_list) - entity->rq = rq_list[0]; + if (num_sched_list) + entity->rq = &entity->sched_list[0]->sched_rq[entity->priority]; entity->last_scheduled = NULL; @@ -139,10 +143,10 @@ drm_sched_entity_get_free_sched(struct drm_sched_entity *entity) unsigned int min_jobs = UINT_MAX, num_jobs; int i; - for (i = 0; i < entity->num_rq_list; ++i) { - struct drm_gpu_scheduler *sched = entity->rq_list[i]->sched; + for (i = 0; i < entity->num_sched_list; ++i) { + struct drm_gpu_scheduler *sched = entity->sched_list[i]; - if (!entity->rq_list[i]->sched->ready) { + if (!entity->sched_list[i]->ready) { DRM_WARN("sched%s is not ready, skipping", sched->name); continue; } @@ -150,7 +154,7 @@ drm_sched_entity_get_free_sched(struct drm_sched_entity *entity) num_jobs = atomic_read(&sched->num_jobs); if (num_jobs < min_jobs) { min_jobs = num_jobs; - rq = entity->rq_list[i]; + rq = &entity->sched_list[i]->sched_rq[entity->priority]; } } @@ -308,7 +312,7 @@ void drm_sched_entity_fini(struct drm_sched_entity *entity) dma_fence_put(entity->last_scheduled); entity->last_scheduled = NULL; - kfree(entity->rq_list); + kfree(entity->sched_list); } EXPORT_SYMBOL(drm_sched_entity_fini); @@ -353,15 +357,6 @@ static void drm_sched_entity_wakeup(struct dma_fence *f, drm_sched_wakeup(entity->rq->sched); } -/** - * drm_sched_entity_set_rq_priority - helper for drm_sched_entity_set_priority - */ -static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq, - enum drm_sched_priority priority) -{ - *rq = &(*rq)->sched->sched_rq[priority]; -} - /** * drm_sched_entity_set_priority - Sets priority of the entity * @@ -373,19 +368,8 @@ static void drm_sched_entity_set_rq_priority(struct drm_sched_rq **rq, void drm_sched_entity_set_priority(struct drm_sched_entity *entity, enum drm_sched_priority priority) { - unsigned int i; - spin_lock(&entity->rq_lock); - - for (i = 0; i < entity->num_rq_list; ++i) - drm_sched_entity_set_rq_priority(&entity->rq_list[i], priority); - - if (entity->rq) { - drm_sched_rq_remove_entity(entity->rq, entity); - drm_sched_entity_set_rq_priority(&entity->rq, priority); - drm_sched_rq_add_entity(entity->rq, entity); - } - + entity->priority = priority; spin_unlock(&entity->rq_lock); } EXPORT_SYMBOL(drm_sched_entity_set_priority); @@ -490,20 +474,20 @@ void drm_sched_entity_select_rq(struct drm_sched_entity *entity) struct dma_fence *fence; struct drm_sched_rq *rq; - if (spsc_queue_count(&entity->job_queue) || entity->num_rq_list <= 1) + if (spsc_queue_count(&entity->job_queue) || entity->num_sched_list <= 1) return; fence = READ_ONCE(entity->last_scheduled); if (fence && !dma_fence_is_signaled(fence)) return; + spin_lock(&entity->rq_lock); rq = drm_sched_entity_get_free_sched(entity); - if (rq == entity->rq) - return; + if (rq != entity->rq) { + drm_sched_rq_remove_entity(entity->rq, entity); + entity->rq = rq; + } - spin_lock(&entity->rq_lock); - drm_sched_rq_remove_entity(entity->rq, entity); - entity->rq = rq; spin_unlock(&entity->rq_lock); } diff --git a/drivers/gpu/drm/v3d/v3d_drv.c b/drivers/gpu/drm/v3d/v3d_drv.c index 1a07462b4528..eaa8e9682373 100644 --- a/drivers/gpu/drm/v3d/v3d_drv.c +++ b/drivers/gpu/drm/v3d/v3d_drv.c @@ -140,7 +140,7 @@ v3d_open(struct drm_device *dev, struct drm_file *file) { struct v3d_dev *v3d = to_v3d_dev(dev); struct v3d_file_priv *v3d_priv; - struct drm_sched_rq *rq; + struct drm_gpu_scheduler *sched; int i; v3d_priv = kzalloc(sizeof(*v3d_priv), GFP_KERNEL); @@ -150,8 +150,10 @@ v3d_open(struct drm_device *dev, struct drm_file *file) v3d_priv->v3d = v3d; for (i = 0; i < V3D_MAX_QUEUES; i++) { - rq = &v3d->queue[i].sched.sched_rq[DRM_SCHED_PRIORITY_NORMAL]; - drm_sched_entity_init(&v3d_priv->sched_entity[i], &rq, 1, NULL); + sched = &v3d->queue[i].sched; + drm_sched_entity_init(&v3d_priv->sched_entity[i], + DRM_SCHED_PRIORITY_NORMAL, &sched, + 1, NULL); } file->driver_priv = v3d_priv; diff --git a/include/drm/gpu_scheduler.h b/include/drm/gpu_scheduler.h index 684692a8ed76..96a1a1b7526e 100644 --- a/include/drm/gpu_scheduler.h +++ b/include/drm/gpu_scheduler.h @@ -81,8 +81,9 @@ enum drm_sched_priority { struct drm_sched_entity { struct list_head list; struct drm_sched_rq *rq; - struct drm_sched_rq **rq_list; - unsigned int num_rq_list; + unsigned int num_sched_list; + struct drm_gpu_scheduler **sched_list; + enum drm_sched_priority priority; spinlock_t rq_lock; struct spsc_queue job_queue; @@ -312,7 +313,8 @@ void drm_sched_rq_remove_entity(struct drm_sched_rq *rq, struct drm_sched_entity *entity); int drm_sched_entity_init(struct drm_sched_entity *entity, - struct drm_sched_rq **rq_list, + enum drm_sched_priority priority, + struct drm_gpu_scheduler **sched_list, unsigned int num_rq_list, atomic_t *guilty); long drm_sched_entity_flush(struct drm_sched_entity *entity, long timeout); -- cgit From f880799d7fcf0a63ca2295d950cd12f5520251d9 Mon Sep 17 00:00:00 2001 From: Nirmoy Das Date: Mon, 16 Dec 2019 14:43:34 +0100 Subject: amd/amdgpu: add sched array to IPs with multiple run-queues MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit This sched array can be passed on to entity creation routine instead of manually creating such sched array on every context creation. v2: squash in missing break fix Signed-off-by: Nirmoy Das Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 113 ++++++++++++++++++----------- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h | 3 + drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 4 + drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h | 9 ++- 7 files changed, 89 insertions(+), 46 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 38ec5c919bd9..63f6365312d5 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -74,7 +74,7 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, struct amdgpu_ctx *ctx) { unsigned num_entities = amdgpu_ctx_total_num_entities(); - unsigned i, j, k; + unsigned i, j; int r; if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX) @@ -121,73 +121,56 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, ctx->override_priority = DRM_SCHED_PRIORITY_UNSET; for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { - struct amdgpu_ring *rings[AMDGPU_MAX_RINGS]; - struct drm_gpu_scheduler *sched_list[AMDGPU_MAX_RINGS]; - unsigned num_rings = 0; - unsigned num_rqs = 0; + struct drm_gpu_scheduler **scheds; + struct drm_gpu_scheduler *sched; + unsigned num_scheds = 0; switch (i) { case AMDGPU_HW_IP_GFX: - rings[0] = &adev->gfx.gfx_ring[0]; - num_rings = 1; + scheds = adev->gfx.gfx_sched; + num_scheds = 1; break; case AMDGPU_HW_IP_COMPUTE: - for (j = 0; j < adev->gfx.num_compute_rings; ++j) - rings[j] = &adev->gfx.compute_ring[j]; - num_rings = adev->gfx.num_compute_rings; + scheds = adev->gfx.compute_sched; + num_scheds = adev->gfx.num_compute_sched; break; case AMDGPU_HW_IP_DMA: - for (j = 0; j < adev->sdma.num_instances; ++j) - rings[j] = &adev->sdma.instance[j].ring; - num_rings = adev->sdma.num_instances; + scheds = adev->sdma.sdma_sched; + num_scheds = adev->sdma.num_sdma_sched; break; case AMDGPU_HW_IP_UVD: - rings[0] = &adev->uvd.inst[0].ring; - num_rings = 1; + sched = &adev->uvd.inst[0].ring.sched; + scheds = &sched; + num_scheds = 1; break; case AMDGPU_HW_IP_VCE: - rings[0] = &adev->vce.ring[0]; - num_rings = 1; + sched = &adev->vce.ring[0].sched; + scheds = &sched; + num_scheds = 1; break; case AMDGPU_HW_IP_UVD_ENC: - rings[0] = &adev->uvd.inst[0].ring_enc[0]; - num_rings = 1; + sched = &adev->uvd.inst[0].ring_enc[0].sched; + scheds = &sched; + num_scheds = 1; break; case AMDGPU_HW_IP_VCN_DEC: - for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { - if (adev->vcn.harvest_config & (1 << j)) - continue; - rings[num_rings++] = &adev->vcn.inst[j].ring_dec; - } + scheds = adev->vcn.vcn_dec_sched; + num_scheds = adev->vcn.num_vcn_dec_sched; break; case AMDGPU_HW_IP_VCN_ENC: - for (j = 0; j < adev->vcn.num_vcn_inst; ++j) { - if (adev->vcn.harvest_config & (1 << j)) - continue; - for (k = 0; k < adev->vcn.num_enc_rings; ++k) - rings[num_rings++] = &adev->vcn.inst[j].ring_enc[k]; - } + scheds = adev->vcn.vcn_enc_sched; + num_scheds = adev->vcn.num_vcn_enc_sched; break; case AMDGPU_HW_IP_VCN_JPEG: - for (j = 0; j < adev->jpeg.num_jpeg_inst; ++j) { - if (adev->jpeg.harvest_config & (1 << j)) - continue; - rings[num_rings++] = &adev->jpeg.inst[j].ring_dec; - } + scheds = adev->jpeg.jpeg_sched; + num_scheds = adev->jpeg.num_jpeg_sched; break; } - for (j = 0; j < num_rings; ++j) { - if (!rings[j]->adev) - continue; - - sched_list[num_rqs++] = &rings[j]->sched; - } - for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) r = drm_sched_entity_init(&ctx->entities[i][j].entity, - priority, sched_list, - num_rqs, &ctx->guilty); + priority, scheds, + num_scheds, &ctx->guilty); if (r) goto error_cleanup_entities; } @@ -628,3 +611,45 @@ void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr) idr_destroy(&mgr->ctx_handles); mutex_destroy(&mgr->lock); } + +void amdgpu_ctx_init_sched(struct amdgpu_device *adev) +{ + int i, j; + + for (i = 0; i < adev->gfx.num_gfx_rings; i++) { + adev->gfx.gfx_sched[i] = &adev->gfx.gfx_ring[i].sched; + adev->gfx.num_gfx_sched++; + } + + for (i = 0; i < adev->gfx.num_compute_rings; i++) { + adev->gfx.compute_sched[i] = &adev->gfx.compute_ring[i].sched; + adev->gfx.num_compute_sched++; + } + + for (i = 0; i < adev->sdma.num_instances; i++) { + adev->sdma.sdma_sched[i] = &adev->sdma.instance[i].ring.sched; + adev->sdma.num_sdma_sched++; + } + + for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { + if (adev->vcn.harvest_config & (1 << i)) + continue; + adev->vcn.vcn_dec_sched[adev->vcn.num_vcn_dec_sched++] = + &adev->vcn.inst[i].ring_dec.sched; + } + + for (i = 0; i < adev->vcn.num_vcn_inst; ++i) { + if (adev->vcn.harvest_config & (1 << i)) + continue; + for (j = 0; j < adev->vcn.num_enc_rings; ++j) + adev->vcn.vcn_enc_sched[adev->vcn.num_vcn_enc_sched++] = + &adev->vcn.inst[i].ring_enc[j].sched; + } + + for (i = 0; i < adev->jpeg.num_jpeg_inst; ++i) { + if (adev->jpeg.harvest_config & (1 << i)) + continue; + adev->jpeg.jpeg_sched[adev->jpeg.num_jpeg_sched++] = + &adev->jpeg.inst[i].ring_dec.sched; + } +} diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h index da808633732b..4ad90a44dc3c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h @@ -87,4 +87,7 @@ void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr); long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout); void amdgpu_ctx_mgr_fini(struct amdgpu_ctx_mgr *mgr); +void amdgpu_ctx_init_sched(struct amdgpu_device *adev); + + #endif diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c index db91663df4f6..e1f8b715301a 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c @@ -3036,6 +3036,8 @@ fence_driver_init: adev->gfx.config.max_cu_per_sh, adev->gfx.cu_info.number); + amdgpu_ctx_init_sched(adev); + adev->accel_working = true; amdgpu_vm_check_compute_bug(adev); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h index 0ae0a2715b0d..8e88e0411662 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h @@ -269,8 +269,12 @@ struct amdgpu_gfx { bool me_fw_write_wait; bool cp_fw_write_wait; struct amdgpu_ring gfx_ring[AMDGPU_MAX_GFX_RINGS]; + struct drm_gpu_scheduler *gfx_sched[AMDGPU_MAX_GFX_RINGS]; + uint32_t num_gfx_sched; unsigned num_gfx_rings; struct amdgpu_ring compute_ring[AMDGPU_MAX_COMPUTE_RINGS]; + struct drm_gpu_scheduler *compute_sched[AMDGPU_MAX_COMPUTE_RINGS]; + uint32_t num_compute_sched; unsigned num_compute_rings; struct amdgpu_irq_src eop_irq; struct amdgpu_irq_src priv_reg_irq; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h index 5131a0a1bc8a..bd9ef9cc86de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_jpeg.h @@ -43,6 +43,8 @@ struct amdgpu_jpeg { uint8_t num_jpeg_inst; struct amdgpu_jpeg_inst inst[AMDGPU_MAX_JPEG_INSTANCES]; struct amdgpu_jpeg_reg internal; + struct drm_gpu_scheduler *jpeg_sched[AMDGPU_MAX_JPEG_INSTANCES]; + uint32_t num_jpeg_sched; unsigned harvest_config; struct delayed_work idle_work; enum amd_powergating_state cur_state; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h index 761ff8be6314..346dcb1f7146 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sdma.h @@ -52,6 +52,8 @@ struct amdgpu_sdma_instance { struct amdgpu_sdma { struct amdgpu_sdma_instance instance[AMDGPU_MAX_SDMA_INSTANCES]; + struct drm_gpu_scheduler *sdma_sched[AMDGPU_MAX_SDMA_INSTANCES]; + uint32_t num_sdma_sched; struct amdgpu_irq_src trap_irq; struct amdgpu_irq_src illegal_inst_irq; struct amdgpu_irq_src ecc_irq; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h index 3484ead62046..e6dee8224d33 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vcn.h @@ -31,6 +31,7 @@ #define AMDGPU_VCN_MAX_ENC_RINGS 3 #define AMDGPU_MAX_VCN_INSTANCES 2 +#define AMDGPU_MAX_VCN_ENC_RINGS AMDGPU_VCN_MAX_ENC_RINGS * AMDGPU_MAX_VCN_INSTANCES #define AMDGPU_VCN_HARVEST_VCN0 (1 << 0) #define AMDGPU_VCN_HARVEST_VCN1 (1 << 1) @@ -189,8 +190,12 @@ struct amdgpu_vcn { uint32_t *dpg_sram_curr_addr; uint8_t num_vcn_inst; - struct amdgpu_vcn_inst inst[AMDGPU_MAX_VCN_INSTANCES]; - struct amdgpu_vcn_reg internal; + struct amdgpu_vcn_inst inst[AMDGPU_MAX_VCN_INSTANCES]; + struct amdgpu_vcn_reg internal; + struct drm_gpu_scheduler *vcn_enc_sched[AMDGPU_MAX_VCN_ENC_RINGS]; + struct drm_gpu_scheduler *vcn_dec_sched[AMDGPU_MAX_VCN_INSTANCES]; + uint32_t num_vcn_enc_sched; + uint32_t num_vcn_dec_sched; unsigned harvest_config; int (*pause_dpg_mode)(struct amdgpu_device *adev, -- cgit From 0a96afc7c56cbac075426a6a5f82a88cadb648f7 Mon Sep 17 00:00:00 2001 From: Le Ma Date: Thu, 19 Dec 2019 19:26:02 +0800 Subject: drm/amdgpu: fix ctx init failure for asics without gfx ring This workaround does not affect other asics because amdgpu only need expose one gfx sched to user for now. Signed-off-by: Le Ma Reviewed-by: Nirmoy Das Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 63f6365312d5..64e2babbc36e 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -127,7 +127,8 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, switch (i) { case AMDGPU_HW_IP_GFX: - scheds = adev->gfx.gfx_sched; + sched = &adev->gfx.gfx_ring[0].sched; + scheds = &sched; num_scheds = 1; break; case AMDGPU_HW_IP_COMPUTE: -- cgit From 63e3ab9a8226c0271e823f32fc1c3de67d142c78 Mon Sep 17 00:00:00 2001 From: Nirmoy Das Date: Tue, 21 Jan 2020 13:29:20 +0100 Subject: drm/amdgpu: individualize fence allocation per entity MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Allocate fences for each entity and remove ctx->fences reference as fences should be bound to amdgpu_ctx_entity instead amdgpu_ctx. Signed-off-by: Nirmoy Das Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 44 ++++++++++++++++++++------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h | 1 - 2 files changed, 27 insertions(+), 18 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 64e2babbc36e..05c2af61e7de 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -87,24 +87,24 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, memset(ctx, 0, sizeof(*ctx)); ctx->adev = adev; - ctx->fences = kcalloc(amdgpu_sched_jobs * num_entities, - sizeof(struct dma_fence*), GFP_KERNEL); - if (!ctx->fences) - return -ENOMEM; ctx->entities[0] = kcalloc(num_entities, sizeof(struct amdgpu_ctx_entity), GFP_KERNEL); - if (!ctx->entities[0]) { - r = -ENOMEM; - goto error_free_fences; - } + if (!ctx->entities[0]) + return -ENOMEM; + for (i = 0; i < num_entities; ++i) { struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; entity->sequence = 1; - entity->fences = &ctx->fences[amdgpu_sched_jobs * i]; + entity->fences = kcalloc(amdgpu_sched_jobs, + sizeof(struct dma_fence*), GFP_KERNEL); + if (!entity->fences) { + r = -ENOMEM; + goto error_cleanup_memory; + } } for (i = 1; i < AMDGPU_HW_IP_NUM; ++i) ctx->entities[i] = ctx->entities[i - 1] + @@ -181,11 +181,17 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, error_cleanup_entities: for (i = 0; i < num_entities; ++i) drm_sched_entity_destroy(&ctx->entities[0][i].entity); - kfree(ctx->entities[0]); -error_free_fences: - kfree(ctx->fences); - ctx->fences = NULL; +error_cleanup_memory: + for (i = 0; i < num_entities; ++i) { + struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; + + kfree(entity->fences); + entity->fences = NULL; + } + + kfree(ctx->entities[0]); + ctx->entities[0] = NULL; return r; } @@ -199,12 +205,16 @@ static void amdgpu_ctx_fini(struct kref *ref) if (!adev) return; - for (i = 0; i < num_entities; ++i) + for (i = 0; i < num_entities; ++i) { + struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; + for (j = 0; j < amdgpu_sched_jobs; ++j) - dma_fence_put(ctx->entities[0][i].fences[j]); - kfree(ctx->fences); - kfree(ctx->entities[0]); + dma_fence_put(entity->fences[j]); + kfree(entity->fences); + } + + kfree(ctx->entities[0]); mutex_destroy(&ctx->lock); kfree(ctx); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h index 4ad90a44dc3c..a6cd9d4b078c 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h @@ -42,7 +42,6 @@ struct amdgpu_ctx { unsigned reset_counter_query; uint32_t vram_lost_counter; spinlock_t ring_lock; - struct dma_fence **fences; struct amdgpu_ctx_entity *entities[AMDGPU_HW_IP_NUM]; bool preamble_presented; enum drm_sched_priority init_priority; -- cgit From 977f7e1068be60f3cac7d80ab18692341f3b7fc5 Mon Sep 17 00:00:00 2001 From: Nirmoy Das Date: Tue, 21 Jan 2020 15:53:53 +0100 Subject: drm/amdgpu: allocate entities on demand MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Currently we pre-allocate entities and fences for all the HW IPs on context creation and some of which are might never be used. This patch tries to resolve entity/fences wastage by creating entity only when needed. v2: allocate memory for entity and fences together Signed-off-by: Nirmoy Das Reviewed-by: Christian König Signed-off-by: Alex Deucher --- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c | 235 ++++++++++++++++---------------- drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h | 6 +- 2 files changed, 124 insertions(+), 117 deletions(-) (limited to 'drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c') diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c index 05c2af61e7de..94a6c42f29ea 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.c @@ -42,19 +42,12 @@ const unsigned int amdgpu_ctx_num_entities[AMDGPU_HW_IP_NUM] = { [AMDGPU_HW_IP_VCN_JPEG] = 1, }; -static int amdgpu_ctx_total_num_entities(void) -{ - unsigned i, num_entities = 0; - - for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) - num_entities += amdgpu_ctx_num_entities[i]; - - return num_entities; -} - static int amdgpu_ctx_priority_permit(struct drm_file *filp, enum drm_sched_priority priority) { + if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX) + return -EINVAL; + /* NORMAL and below are accessible by everyone */ if (priority <= DRM_SCHED_PRIORITY_NORMAL) return 0; @@ -68,64 +61,24 @@ static int amdgpu_ctx_priority_permit(struct drm_file *filp, return -EACCES; } -static int amdgpu_ctx_init(struct amdgpu_device *adev, - enum drm_sched_priority priority, - struct drm_file *filp, - struct amdgpu_ctx *ctx) +static int amdgpu_ctx_init_entity(struct amdgpu_ctx *ctx, const u32 hw_ip, const u32 ring) { - unsigned num_entities = amdgpu_ctx_total_num_entities(); - unsigned i, j; + struct amdgpu_device *adev = ctx->adev; + struct amdgpu_ctx_entity *entity; + struct drm_gpu_scheduler **scheds = NULL, *sched = NULL; + unsigned num_scheds = 0; + enum drm_sched_priority priority; int r; - if (priority < 0 || priority >= DRM_SCHED_PRIORITY_MAX) - return -EINVAL; - - r = amdgpu_ctx_priority_permit(filp, priority); - if (r) - return r; - - memset(ctx, 0, sizeof(*ctx)); - ctx->adev = adev; - - - ctx->entities[0] = kcalloc(num_entities, - sizeof(struct amdgpu_ctx_entity), - GFP_KERNEL); - if (!ctx->entities[0]) - return -ENOMEM; - + entity = kcalloc(1, offsetof(typeof(*entity), fences[amdgpu_sched_jobs]), + GFP_KERNEL); + if (!entity) + return -ENOMEM; - for (i = 0; i < num_entities; ++i) { - struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; - - entity->sequence = 1; - entity->fences = kcalloc(amdgpu_sched_jobs, - sizeof(struct dma_fence*), GFP_KERNEL); - if (!entity->fences) { - r = -ENOMEM; - goto error_cleanup_memory; - } - } - for (i = 1; i < AMDGPU_HW_IP_NUM; ++i) - ctx->entities[i] = ctx->entities[i - 1] + - amdgpu_ctx_num_entities[i - 1]; - - kref_init(&ctx->refcount); - spin_lock_init(&ctx->ring_lock); - mutex_init(&ctx->lock); - - ctx->reset_counter = atomic_read(&adev->gpu_reset_counter); - ctx->reset_counter_query = ctx->reset_counter; - ctx->vram_lost_counter = atomic_read(&adev->vram_lost_counter); - ctx->init_priority = priority; - ctx->override_priority = DRM_SCHED_PRIORITY_UNSET; - - for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { - struct drm_gpu_scheduler **scheds; - struct drm_gpu_scheduler *sched; - unsigned num_scheds = 0; - - switch (i) { + entity->sequence = 1; + priority = (ctx->override_priority == DRM_SCHED_PRIORITY_UNSET) ? + ctx->init_priority : ctx->override_priority; + switch (hw_ip) { case AMDGPU_HW_IP_GFX: sched = &adev->gfx.gfx_ring[0].sched; scheds = &sched; @@ -166,63 +119,90 @@ static int amdgpu_ctx_init(struct amdgpu_device *adev, scheds = adev->jpeg.jpeg_sched; num_scheds = adev->jpeg.num_jpeg_sched; break; - } - - for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) - r = drm_sched_entity_init(&ctx->entities[i][j].entity, - priority, scheds, - num_scheds, &ctx->guilty); - if (r) - goto error_cleanup_entities; } + r = drm_sched_entity_init(&entity->entity, priority, scheds, num_scheds, + &ctx->guilty); + if (r) + goto error_free_entity; + + ctx->entities[hw_ip][ring] = entity; return 0; -error_cleanup_entities: - for (i = 0; i < num_entities; ++i) - drm_sched_entity_destroy(&ctx->entities[0][i].entity); +error_free_entity: + kfree(entity); -error_cleanup_memory: - for (i = 0; i < num_entities; ++i) { - struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; + return r; +} - kfree(entity->fences); - entity->fences = NULL; - } +static int amdgpu_ctx_init(struct amdgpu_device *adev, + enum drm_sched_priority priority, + struct drm_file *filp, + struct amdgpu_ctx *ctx) +{ + int r; - kfree(ctx->entities[0]); - ctx->entities[0] = NULL; - return r; + r = amdgpu_ctx_priority_permit(filp, priority); + if (r) + return r; + + memset(ctx, 0, sizeof(*ctx)); + + ctx->adev = adev; + + kref_init(&ctx->refcount); + spin_lock_init(&ctx->ring_lock); + mutex_init(&ctx->lock); + + ctx->reset_counter = atomic_read(&adev->gpu_reset_counter); + ctx->reset_counter_query = ctx->reset_counter; + ctx->vram_lost_counter = atomic_read(&adev->vram_lost_counter); + ctx->init_priority = priority; + ctx->override_priority = DRM_SCHED_PRIORITY_UNSET; + + return 0; + +} + +static void amdgpu_ctx_fini_entity(struct amdgpu_ctx_entity *entity) +{ + + int i; + + if (!entity) + return; + + for (i = 0; i < amdgpu_sched_jobs; ++i) + dma_fence_put(entity->fences[i]); + + kfree(entity); } static void amdgpu_ctx_fini(struct kref *ref) { struct amdgpu_ctx *ctx = container_of(ref, struct amdgpu_ctx, refcount); - unsigned num_entities = amdgpu_ctx_total_num_entities(); struct amdgpu_device *adev = ctx->adev; unsigned i, j; if (!adev) return; - for (i = 0; i < num_entities; ++i) { - struct amdgpu_ctx_entity *entity = &ctx->entities[0][i]; - - for (j = 0; j < amdgpu_sched_jobs; ++j) - dma_fence_put(entity->fences[j]); - - kfree(entity->fences); + for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { + for (j = 0; j < AMDGPU_MAX_ENTITY_NUM; ++j) { + amdgpu_ctx_fini_entity(ctx->entities[i][j]); + ctx->entities[i][j] = NULL; + } } - kfree(ctx->entities[0]); mutex_destroy(&ctx->lock); - kfree(ctx); } int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance, u32 ring, struct drm_sched_entity **entity) { + int r; + if (hw_ip >= AMDGPU_HW_IP_NUM) { DRM_ERROR("unknown HW IP type: %d\n", hw_ip); return -EINVAL; @@ -239,7 +219,13 @@ int amdgpu_ctx_get_entity(struct amdgpu_ctx *ctx, u32 hw_ip, u32 instance, return -EINVAL; } - *entity = &ctx->entities[hw_ip][ring].entity; + if (ctx->entities[hw_ip][ring] == NULL) { + r = amdgpu_ctx_init_entity(ctx, hw_ip, ring); + if (r) + return r; + } + + *entity = &ctx->entities[hw_ip][ring]->entity; return 0; } @@ -279,14 +265,17 @@ static int amdgpu_ctx_alloc(struct amdgpu_device *adev, static void amdgpu_ctx_do_release(struct kref *ref) { struct amdgpu_ctx *ctx; - unsigned num_entities; - u32 i; + u32 i, j; ctx = container_of(ref, struct amdgpu_ctx, refcount); + for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { + for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) { + if (!ctx->entities[i][j]) + continue; - num_entities = amdgpu_ctx_total_num_entities(); - for (i = 0; i < num_entities; i++) - drm_sched_entity_destroy(&ctx->entities[0][i].entity); + drm_sched_entity_destroy(&ctx->entities[i][j]->entity); + } + } amdgpu_ctx_fini(ref); } @@ -516,19 +505,23 @@ struct dma_fence *amdgpu_ctx_get_fence(struct amdgpu_ctx *ctx, void amdgpu_ctx_priority_override(struct amdgpu_ctx *ctx, enum drm_sched_priority priority) { - unsigned num_entities = amdgpu_ctx_total_num_entities(); enum drm_sched_priority ctx_prio; - unsigned i; + unsigned i, j; ctx->override_priority = priority; ctx_prio = (ctx->override_priority == DRM_SCHED_PRIORITY_UNSET) ? ctx->init_priority : ctx->override_priority; + for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { + for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) { + struct drm_sched_entity *entity; - for (i = 0; i < num_entities; i++) { - struct drm_sched_entity *entity = &ctx->entities[0][i].entity; + if (!ctx->entities[i][j]) + continue; - drm_sched_entity_set_priority(entity, ctx_prio); + entity = &ctx->entities[i][j]->entity; + drm_sched_entity_set_priority(entity, ctx_prio); + } } } @@ -564,20 +557,24 @@ void amdgpu_ctx_mgr_init(struct amdgpu_ctx_mgr *mgr) long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout) { - unsigned num_entities = amdgpu_ctx_total_num_entities(); struct amdgpu_ctx *ctx; struct idr *idp; - uint32_t id, i; + uint32_t id, i, j; idp = &mgr->ctx_handles; mutex_lock(&mgr->lock); idr_for_each_entry(idp, ctx, id) { - for (i = 0; i < num_entities; i++) { - struct drm_sched_entity *entity; + for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { + for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) { + struct drm_sched_entity *entity; + + if (!ctx->entities[i][j]) + continue; - entity = &ctx->entities[0][i].entity; - timeout = drm_sched_entity_flush(entity, timeout); + entity = &ctx->entities[i][j]->entity; + timeout = drm_sched_entity_flush(entity, timeout); + } } } mutex_unlock(&mgr->lock); @@ -586,10 +583,9 @@ long amdgpu_ctx_mgr_entity_flush(struct amdgpu_ctx_mgr *mgr, long timeout) void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr) { - unsigned num_entities = amdgpu_ctx_total_num_entities(); struct amdgpu_ctx *ctx; struct idr *idp; - uint32_t id, i; + uint32_t id, i, j; idp = &mgr->ctx_handles; @@ -599,8 +595,17 @@ void amdgpu_ctx_mgr_entity_fini(struct amdgpu_ctx_mgr *mgr) continue; } - for (i = 0; i < num_entities; i++) - drm_sched_entity_fini(&ctx->entities[0][i].entity); + for (i = 0; i < AMDGPU_HW_IP_NUM; ++i) { + for (j = 0; j < amdgpu_ctx_num_entities[i]; ++j) { + struct drm_sched_entity *entity; + + if (!ctx->entities[i][j]) + continue; + + entity = &ctx->entities[i][j]->entity; + drm_sched_entity_fini(entity); + } + } } } diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h index a6cd9d4b078c..de490f183af2 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ctx.h @@ -29,10 +29,12 @@ struct drm_device; struct drm_file; struct amdgpu_fpriv; +#define AMDGPU_MAX_ENTITY_NUM 4 + struct amdgpu_ctx_entity { uint64_t sequence; - struct dma_fence **fences; struct drm_sched_entity entity; + struct dma_fence *fences[]; }; struct amdgpu_ctx { @@ -42,7 +44,7 @@ struct amdgpu_ctx { unsigned reset_counter_query; uint32_t vram_lost_counter; spinlock_t ring_lock; - struct amdgpu_ctx_entity *entities[AMDGPU_HW_IP_NUM]; + struct amdgpu_ctx_entity *entities[AMDGPU_HW_IP_NUM][AMDGPU_MAX_ENTITY_NUM]; bool preamble_presented; enum drm_sched_priority init_priority; enum drm_sched_priority override_priority; -- cgit