[PATCH 2/4] drm/amdgpu: rework VM state machine lock handling

Christian König ckoenig.leichtzumerken at gmail.com
Thu Apr 19 09:29:51 UTC 2018


Only the moved state needs a separate spin lock protection. All other
states are protected by reserving the VM anyway.

Signed-off-by: Christian König <christian.koenig at amd.com>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 70 ++++++++++++----------------------
 drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h |  4 +-
 2 files changed, 25 insertions(+), 49 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
index 6a372ca11ee3..dbf24b628835 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c
@@ -198,7 +198,6 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	struct ttm_bo_global *glob = adev->mman.bdev.glob;
 	int r;
 
-	spin_lock(&vm->status_lock);
 	while (!list_empty(&vm->evicted)) {
 		struct amdgpu_vm_bo_base *bo_base;
 		struct amdgpu_bo *bo;
@@ -206,10 +205,8 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 		bo_base = list_first_entry(&vm->evicted,
 					   struct amdgpu_vm_bo_base,
 					   vm_status);
-		spin_unlock(&vm->status_lock);
 
 		bo = bo_base->bo;
-		BUG_ON(!bo);
 		if (bo->parent) {
 			r = validate(param, bo);
 			if (r)
@@ -229,13 +226,14 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 				return r;
 		}
 
-		spin_lock(&vm->status_lock);
-		if (bo->tbo.type != ttm_bo_type_kernel)
+		if (bo->tbo.type != ttm_bo_type_kernel) {
+			spin_lock(&vm->moved_lock);
 			list_move(&bo_base->vm_status, &vm->moved);
-		else
+			spin_unlock(&vm->moved_lock);
+		} else {
 			list_move(&bo_base->vm_status, &vm->relocated);
+		}
 	}
-	spin_unlock(&vm->status_lock);
 
 	return 0;
 }
@@ -249,13 +247,7 @@ int amdgpu_vm_validate_pt_bos(struct amdgpu_device *adev, struct amdgpu_vm *vm,
  */
 bool amdgpu_vm_ready(struct amdgpu_vm *vm)
 {
-	bool ready;
-
-	spin_lock(&vm->status_lock);
-	ready = list_empty(&vm->evicted);
-	spin_unlock(&vm->status_lock);
-
-	return ready;
+	return list_empty(&vm->evicted);
 }
 
 /**
@@ -449,9 +441,7 @@ static int amdgpu_vm_alloc_levels(struct amdgpu_device *adev,
 			entry->base.vm = vm;
 			entry->base.bo = pt;
 			list_add_tail(&entry->base.bo_list, &pt->va);
-			spin_lock(&vm->status_lock);
 			list_add(&entry->base.vm_status, &vm->relocated);
-			spin_unlock(&vm->status_lock);
 		}
 
 		if (level < AMDGPU_VM_PTB) {
@@ -898,10 +888,8 @@ static void amdgpu_vm_invalidate_level(struct amdgpu_device *adev,
 		if (!entry->base.bo)
 			continue;
 
-		spin_lock(&vm->status_lock);
 		if (list_empty(&entry->base.vm_status))
 			list_add(&entry->base.vm_status, &vm->relocated);
-		spin_unlock(&vm->status_lock);
 		amdgpu_vm_invalidate_level(adev, vm, entry, level + 1);
 	}
 }
@@ -946,7 +934,6 @@ int amdgpu_vm_update_directories(struct amdgpu_device *adev,
 		params.func = amdgpu_vm_do_set_ptes;
 	}
 
-	spin_lock(&vm->status_lock);
 	while (!list_empty(&vm->relocated)) {
 		struct amdgpu_vm_bo_base *bo_base, *parent;
 		struct amdgpu_vm_pt *pt, *entry;
@@ -956,13 +943,10 @@ int amdgpu_vm_update_directories(struct amdgpu_device *adev,
 					   struct amdgpu_vm_bo_base,
 					   vm_status);
 		list_del_init(&bo_base->vm_status);
-		spin_unlock(&vm->status_lock);
 
 		bo = bo_base->bo->parent;
-		if (!bo) {
-			spin_lock(&vm->status_lock);
+		if (!bo)
 			continue;
-		}
 
 		parent = list_first_entry(&bo->va, struct amdgpu_vm_bo_base,
 					  bo_list);
@@ -971,12 +955,10 @@ int amdgpu_vm_update_directories(struct amdgpu_device *adev,
 
 		amdgpu_vm_update_pde(&params, vm, pt, entry);
 
-		spin_lock(&vm->status_lock);
 		if (!vm->use_cpu_for_update &&
 		    (ndw - params.ib->length_dw) < 32)
 			break;
 	}
-	spin_unlock(&vm->status_lock);
 
 	if (vm->use_cpu_for_update) {
 		/* Flush HDP */
@@ -1079,9 +1061,9 @@ static void amdgpu_vm_handle_huge_pages(struct amdgpu_pte_update_params *p,
 		if (entry->huge) {
 			/* Add the entry to the relocated list to update it. */
 			entry->huge = false;
-			spin_lock(&p->vm->status_lock);
+			spin_lock(&p->vm->moved_lock);
 			list_move(&entry->base.vm_status, &p->vm->relocated);
-			spin_unlock(&p->vm->status_lock);
+			spin_unlock(&p->vm->moved_lock);
 		}
 		return;
 	}
@@ -1561,8 +1543,9 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 		amdgpu_asic_flush_hdp(adev, NULL);
 	}
 
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	list_del_init(&bo_va->base.vm_status);
+	spin_unlock(&vm->moved_lock);
 
 	/* If the BO is not in its preferred location add it back to
 	 * the evicted list so that it gets validated again on the
@@ -1572,7 +1555,6 @@ int amdgpu_vm_bo_update(struct amdgpu_device *adev,
 	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv &&
 	    !(bo->preferred_domains & amdgpu_mem_type_to_domain(mem_type)))
 		list_add_tail(&bo_va->base.vm_status, &vm->evicted);
-	spin_unlock(&vm->status_lock);
 
 	list_splice_init(&bo_va->invalids, &bo_va->valids);
 	bo_va->cleared = clear;
@@ -1784,14 +1766,14 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 	bool clear;
 	int r = 0;
 
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	while (!list_empty(&vm->moved)) {
 		struct amdgpu_bo_va *bo_va;
 		struct reservation_object *resv;
 
 		bo_va = list_first_entry(&vm->moved,
 			struct amdgpu_bo_va, base.vm_status);
-		spin_unlock(&vm->status_lock);
+		spin_unlock(&vm->moved_lock);
 
 		resv = bo_va->base.bo->tbo.resv;
 
@@ -1812,9 +1794,9 @@ int amdgpu_vm_handle_moved(struct amdgpu_device *adev,
 		if (!clear && resv != vm->root.base.bo->tbo.resv)
 			reservation_object_unlock(resv);
 
-		spin_lock(&vm->status_lock);
+		spin_lock(&vm->moved_lock);
 	}
-	spin_unlock(&vm->status_lock);
+	spin_unlock(&vm->moved_lock);
 
 	return r;
 }
@@ -1868,9 +1850,9 @@ struct amdgpu_bo_va *amdgpu_vm_bo_add(struct amdgpu_device *adev,
 	 * is currently evicted. add the BO to the evicted list to make sure it
 	 * is validated on next VM use to avoid fault.
 	 * */
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	list_move_tail(&bo_va->base.vm_status, &vm->evicted);
-	spin_unlock(&vm->status_lock);
+	spin_unlock(&vm->moved_lock);
 
 	return bo_va;
 }
@@ -1900,10 +1882,10 @@ static void amdgpu_vm_bo_insert_map(struct amdgpu_device *adev,
 		amdgpu_vm_prt_get(adev);
 
 	if (bo && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
-		spin_lock(&vm->status_lock);
+		spin_lock(&vm->moved_lock);
 		if (list_empty(&bo_va->base.vm_status))
 			list_add(&bo_va->base.vm_status, &vm->moved);
-		spin_unlock(&vm->status_lock);
+		spin_unlock(&vm->moved_lock);
 	}
 	trace_amdgpu_vm_bo_map(bo_va, mapping);
 }
@@ -2213,9 +2195,9 @@ void amdgpu_vm_bo_rmv(struct amdgpu_device *adev,
 
 	list_del(&bo_va->base.bo_list);
 
-	spin_lock(&vm->status_lock);
+	spin_lock(&vm->moved_lock);
 	list_del(&bo_va->base.vm_status);
-	spin_unlock(&vm->status_lock);
+	spin_unlock(&vm->moved_lock);
 
 	list_for_each_entry_safe(mapping, next, &bo_va->valids, list) {
 		list_del(&mapping->list);
@@ -2254,28 +2236,24 @@ void amdgpu_vm_bo_invalidate(struct amdgpu_device *adev,
 
 		bo_base->moved = true;
 		if (evicted && bo->tbo.resv == vm->root.base.bo->tbo.resv) {
-			spin_lock(&bo_base->vm->status_lock);
 			if (bo->tbo.type == ttm_bo_type_kernel)
 				list_move(&bo_base->vm_status, &vm->evicted);
 			else
 				list_move_tail(&bo_base->vm_status,
 					       &vm->evicted);
-			spin_unlock(&bo_base->vm->status_lock);
 			continue;
 		}
 
 		if (bo->tbo.type == ttm_bo_type_kernel) {
-			spin_lock(&bo_base->vm->status_lock);
 			if (list_empty(&bo_base->vm_status))
 				list_add(&bo_base->vm_status, &vm->relocated);
-			spin_unlock(&bo_base->vm->status_lock);
 			continue;
 		}
 
-		spin_lock(&bo_base->vm->status_lock);
+		spin_lock(&bo_base->vm->moved_lock);
 		if (list_empty(&bo_base->vm_status))
 			list_add(&bo_base->vm_status, &vm->moved);
-		spin_unlock(&bo_base->vm->status_lock);
+		spin_unlock(&bo_base->vm->moved_lock);
 	}
 }
 
@@ -2383,9 +2361,9 @@ int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm,
 	vm->va = RB_ROOT_CACHED;
 	for (i = 0; i < AMDGPU_MAX_VMHUBS; i++)
 		vm->reserved_vmid[i] = NULL;
-	spin_lock_init(&vm->status_lock);
 	INIT_LIST_HEAD(&vm->evicted);
 	INIT_LIST_HEAD(&vm->relocated);
+	spin_lock_init(&vm->moved_lock);
 	INIT_LIST_HEAD(&vm->moved);
 	INIT_LIST_HEAD(&vm->freed);
 
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
index aef897b93866..4e75b6959446 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h
@@ -167,9 +167,6 @@ struct amdgpu_vm {
 	/* tree of virtual addresses mapped */
 	struct rb_root_cached	va;
 
-	/* protecting invalidated */
-	spinlock_t		status_lock;
-
 	/* BOs who needs a validation */
 	struct list_head	evicted;
 
@@ -178,6 +175,7 @@ struct amdgpu_vm {
 
 	/* BOs moved, but not yet updated in the PT */
 	struct list_head	moved;
+	spinlock_t		moved_lock;
 
 	/* BO mappings freed, but not yet updated in the PT */
 	struct list_head	freed;
-- 
2.14.1



More information about the amd-gfx mailing list