For page fault handling we need to use a direct update which can't be blocked by ongoing user CS.
Signed-off-by: Christian König christian.koenig@amd.com Reviewed-by: Felix Kuehling Felix.Kuehling@amd.com --- drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 6 +++--- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 21 +++++++++++++++------ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 5 +++-- drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 5 +++-- 4 files changed, 24 insertions(+), 13 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c index cd15540c5622..dfe155566571 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c @@ -282,7 +282,7 @@ static int amdgpu_vmid_grab_reserved(struct amdgpu_vm *vm, !dma_fence_is_later(updates, (*id)->flushed_updates)) updates = NULL;
- if ((*id)->owner != vm->entity.fence_context || + if ((*id)->owner != vm->direct.fence_context || job->vm_pd_addr != (*id)->pd_gpu_addr || updates || !(*id)->last_flush || ((*id)->last_flush->context != fence_context && @@ -349,7 +349,7 @@ static int amdgpu_vmid_grab_used(struct amdgpu_vm *vm, struct dma_fence *flushed;
/* Check all the prerequisites to using this VMID */ - if ((*id)->owner != vm->entity.fence_context) + if ((*id)->owner != vm->direct.fence_context) continue;
if ((*id)->pd_gpu_addr != job->vm_pd_addr) @@ -449,7 +449,7 @@ int amdgpu_vmid_grab(struct amdgpu_vm *vm, struct amdgpu_ring *ring, }
id->pd_gpu_addr = job->vm_pd_addr; - id->owner = vm->entity.fence_context; + id->owner = vm->direct.fence_context;
if (job->vm_needs_flush) { dma_fence_put(id->last_flush); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index e11764164cbf..8a9ba66c8383 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -2670,12 +2670,17 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, spin_lock_init(&vm->invalidated_lock); INIT_LIST_HEAD(&vm->freed);
- /* create scheduler entity for page table updates */ - r = drm_sched_entity_init(&vm->entity, adev->vm_manager.vm_pte_rqs, + /* 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); 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); + if (r) + goto error_free_direct; + vm->pte_support_ats = false;
if (vm_context == AMDGPU_VM_CONTEXT_COMPUTE) { @@ -2704,7 +2709,7 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, bp.flags &= ~AMDGPU_GEM_CREATE_SHADOW; r = amdgpu_bo_create(adev, &bp, &root); if (r) - goto error_free_sched_entity; + goto error_free_delayed;
r = amdgpu_bo_reserve(root, true); if (r) @@ -2747,8 +2752,11 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, amdgpu_bo_unref(&vm->root.base.bo); vm->root.base.bo = NULL;
-error_free_sched_entity: - drm_sched_entity_destroy(&vm->entity); +error_free_delayed: + drm_sched_entity_destroy(&vm->delayed); + +error_free_direct: + drm_sched_entity_destroy(&vm->direct);
return r; } @@ -2937,7 +2945,8 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) spin_unlock_irqrestore(&adev->vm_manager.pasid_lock, flags); }
- drm_sched_entity_destroy(&vm->entity); + drm_sched_entity_destroy(&vm->direct); + drm_sched_entity_destroy(&vm->delayed);
if (!RB_EMPTY_ROOT(&vm->va.rb_root)) { dev_err(adev->dev, "still active bo inside vm\n"); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h index 3352a87b822e..7138722ee55f 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h @@ -257,8 +257,9 @@ struct amdgpu_vm { struct amdgpu_vm_pt root; struct dma_fence *last_update;
- /* Scheduler entity for page table updates */ - struct drm_sched_entity entity; + /* Scheduler entities for page table updates */ + struct drm_sched_entity direct; + struct drm_sched_entity delayed;
unsigned int pasid; /* dedicated to vm */ diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c index ddd181f5ed37..d087d6650d79 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c @@ -99,12 +99,13 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, struct dma_fence *f; int r;
- ring = container_of(p->vm->entity.rq->sched, struct amdgpu_ring, sched); + ring = container_of(p->vm->delayed.rq->sched, struct amdgpu_ring, + sched);
WARN_ON(ib->length_dw == 0); amdgpu_ring_pad_ib(ring, ib); WARN_ON(ib->length_dw > p->num_dw_left); - r = amdgpu_job_submit(p->job, &p->vm->entity, + r = amdgpu_job_submit(p->job, &p->vm->delayed, AMDGPU_FENCE_OWNER_VM, &f); if (r) goto error;