diff options
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c | 10 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 30 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 | ||||
| -rw-r--r-- | drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c | 11 |
4 files changed, 42 insertions, 13 deletions
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c index f1e5fbef54d8..a09b6b9c27d1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_sync.c @@ -240,13 +240,11 @@ int amdgpu_sync_resv(struct amdgpu_device *adev, continue; if (amdgpu_sync_same_dev(adev, f)) { - /* VM updates are only interesting - * for other VM updates and moves. + /* VM updates only sync with moves but not with user + * command submissions or KFD evictions fences */ - if ((owner != AMDGPU_FENCE_OWNER_UNDEFINED) && - (fence_owner != AMDGPU_FENCE_OWNER_UNDEFINED) && - ((owner == AMDGPU_FENCE_OWNER_VM) != - (fence_owner == AMDGPU_FENCE_OWNER_VM))) + if (owner == AMDGPU_FENCE_OWNER_VM && + fence_owner != AMDGPU_FENCE_OWNER_UNDEFINED) continue; /* Ignore fence from the same owner and explicit one as diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c index 145ea95d977a..81cbb8105a68 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c @@ -562,8 +562,8 @@ void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm, { entry->priority = 0; entry->tv.bo = &vm->root.base.bo->tbo; - /* One for the VM updates, one for TTM and one for the CS job */ - entry->tv.num_shared = 3; + /* One for TTM and one for the CS job */ + entry->tv.num_shared = 2; entry->user_pages = NULL; list_add(&entry->tv.head, validated); } @@ -2518,6 +2518,11 @@ bool amdgpu_vm_evictable(struct amdgpu_bo *bo) if (!dma_resv_test_signaled_rcu(bo->tbo.base.resv, true)) return false; + /* Don't evict VM page tables while they are updated */ + if (!dma_fence_is_signaled(bo_base->vm->last_direct) || + !dma_fence_is_signaled(bo_base->vm->last_delayed)) + return false; + return true; } @@ -2683,8 +2688,16 @@ void amdgpu_vm_adjust_size(struct amdgpu_device *adev, uint32_t min_vm_size, */ long amdgpu_vm_wait_idle(struct amdgpu_vm *vm, long timeout) { - return dma_resv_wait_timeout_rcu(vm->root.base.bo->tbo.base.resv, - true, true, timeout); + timeout = dma_resv_wait_timeout_rcu(vm->root.base.bo->tbo.base.resv, + true, true, timeout); + if (timeout <= 0) + return timeout; + + timeout = dma_fence_wait_timeout(vm->last_direct, true, timeout); + if (timeout <= 0) + return timeout; + + return dma_fence_wait_timeout(vm->last_delayed, true, timeout); } /** @@ -2753,6 +2766,8 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm, else vm->update_funcs = &amdgpu_vm_sdma_funcs; vm->last_update = NULL; + vm->last_direct = dma_fence_get_stub(); + vm->last_delayed = dma_fence_get_stub(); amdgpu_vm_bo_param(adev, vm, adev->vm_manager.root_level, false, &bp); if (vm_context == AMDGPU_VM_CONTEXT_COMPUTE) @@ -2803,6 +2818,8 @@ error_free_root: vm->root.base.bo = NULL; error_free_delayed: + dma_fence_put(vm->last_direct); + dma_fence_put(vm->last_delayed); drm_sched_entity_destroy(&vm->delayed); error_free_direct: @@ -3003,6 +3020,11 @@ void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm) vm->pasid = 0; } + dma_fence_wait(vm->last_direct, false); + dma_fence_put(vm->last_direct); + dma_fence_wait(vm->last_delayed, false); + dma_fence_put(vm->last_delayed); + list_for_each_entry_safe(mapping, tmp, &vm->freed, list) { if (mapping->flags & AMDGPU_PTE_PRT && prt_fini_needed) { amdgpu_vm_prt_fini(adev, vm); diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h index d3b7a5cff56c..6c2124b10820 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h @@ -266,6 +266,10 @@ struct amdgpu_vm { struct drm_sched_entity direct; struct drm_sched_entity delayed; + /* Last submission to the scheduler entities */ + struct dma_fence *last_direct; + struct dma_fence *last_delayed; + unsigned int pasid; /* dedicated to vm */ struct amdgpu_vmid *reserved_vmid[AMDGPU_MAX_VMHUBS]; diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c index 50f487666977..19b7f80758f1 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm_sdma.c @@ -95,11 +95,10 @@ static int amdgpu_vm_sdma_prepare(struct amdgpu_vm_update_params *p, static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, struct dma_fence **fence) { - struct amdgpu_bo *root = p->vm->root.base.bo; struct amdgpu_ib *ib = p->job->ibs; struct drm_sched_entity *entity; + struct dma_fence *f, *tmp; struct amdgpu_ring *ring; - struct dma_fence *f; int r; entity = p->direct ? &p->vm->direct : &p->vm->delayed; @@ -112,7 +111,13 @@ static int amdgpu_vm_sdma_commit(struct amdgpu_vm_update_params *p, if (r) goto error; - amdgpu_bo_fence(root, f, true); + tmp = dma_fence_get(f); + if (p->direct) + swap(p->vm->last_direct, tmp); + else + swap(p->vm->last_delayed, tmp); + dma_fence_put(tmp); + if (fence && !p->direct) swap(*fence, f); dma_fence_put(f); |