drm/amdgpu: split VM PD and PT handling during CS

This way we avoid the extra allocation for the page directory entry.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>

drivers/gpu/drm/amd/amdgpu/amdgpu.h

@@ -980,9 +980,10 @@ struct amdgpu_vm_manager {
 void amdgpu_vm_manager_fini(struct amdgpu_device *adev);
 int amdgpu_vm_init(struct amdgpu_device *adev, struct amdgpu_vm *vm);
 void amdgpu_vm_fini(struct amdgpu_device *adev, struct amdgpu_vm *vm);
-struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
-					       struct amdgpu_vm *vm,
+void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
 			 struct list_head *validated,
+			 struct amdgpu_bo_list_entry *entry);
+struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
 						  struct list_head *duplicates);
 int amdgpu_vm_grab_id(struct amdgpu_vm *vm, struct amdgpu_ring *ring,
 		      struct amdgpu_sync *sync);
@@ -1253,6 +1254,7 @@ struct amdgpu_cs_parser {
 	unsigned		nchunks;
 	struct amdgpu_cs_chunk	*chunks;
 	/* relocations */
+	struct amdgpu_bo_list_entry	vm_pd;
 	struct amdgpu_bo_list_entry	*vm_bos;
 	struct list_head	validated;
 	struct fence		*fence;

drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c

@@ -387,8 +387,7 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	}
 
 	INIT_LIST_HEAD(&duplicates);
-	p->vm_bos = amdgpu_vm_get_bos(p->adev, &fpriv->vm,
-				      &p->validated, &duplicates);
+	amdgpu_vm_get_pd_bo(&fpriv->vm, &p->validated, &p->vm_pd);
 
 	if (need_mmap_lock)
 		down_read(&current->mm->mmap_sem);
@@ -397,6 +396,12 @@ static int amdgpu_cs_parser_relocs(struct amdgpu_cs_parser *p)
 	if (unlikely(r != 0))
 		goto error_reserve;
 
+	p->vm_bos = amdgpu_vm_get_pt_bos(&fpriv->vm, &duplicates);
+	if (!p->vm_bos) {
+		r = -ENOMEM;
+		goto error_validate;
+	}
+
 	r = amdgpu_cs_list_validate(p->adev, &fpriv->vm, &p->validated);
 	if (r)
 		goto error_validate;

drivers/gpu/drm/amd/amdgpu/amdgpu_gem.c

@@ -448,6 +448,7 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 {
 	struct ttm_validate_buffer tv, *entry;
 	struct amdgpu_bo_list_entry *vm_bos;
+	struct amdgpu_bo_list_entry vm_pd;
 	struct ww_acquire_ctx ticket;
 	struct list_head list, duplicates;
 	unsigned domain;
@@ -460,14 +461,18 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 	tv.shared = true;
 	list_add(&tv.head, &list);
 
-	vm_bos = amdgpu_vm_get_bos(adev, bo_va->vm, &list, &duplicates);
-	if (!vm_bos)
-		return;
+	amdgpu_vm_get_pd_bo(bo_va->vm, &list, &vm_pd);
 
 	/* Provide duplicates to avoid -EALREADY */
 	r = ttm_eu_reserve_buffers(&ticket, &list, true, &duplicates);
 	if (r)
-		goto error_free;
+		goto error_print;
+
+	vm_bos = amdgpu_vm_get_pt_bos(bo_va->vm, &duplicates);
+	if (!vm_bos) {
+		r = -ENOMEM;
+		goto error_unreserve;
+	}
 
 	list_for_each_entry(entry, &list, head) {
 		domain = amdgpu_mem_type_to_domain(entry->bo->mem.mem_type);
@@ -489,10 +494,9 @@ static void amdgpu_gem_va_update_vm(struct amdgpu_device *adev,
 
 error_unreserve:
 	ttm_eu_backoff_reservation(&ticket, &list);
-
-error_free:
 	drm_free_large(vm_bos);
 
+error_print:
 	if (r && r != -ERESTARTSYS)
 		DRM_ERROR("Couldn't update BO_VA (%d)\n", r);
 }

drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c

@@ -75,39 +75,50 @@ static unsigned amdgpu_vm_directory_size(struct amdgpu_device *adev)
 }
 
 /**
- * amdgpu_vm_get_bos - add the vm BOs to a validation list
+ * amdgpu_vm_get_pd_bo - add the VM PD to a validation list
  *
  * @vm: vm providing the BOs
  * @validated: head of validation list
+ * @entry: entry to add
+ *
+ * Add the page directory to the list of BOs to
+ * validate for command submission.
+ */
+void amdgpu_vm_get_pd_bo(struct amdgpu_vm *vm,
+			 struct list_head *validated,
+			 struct amdgpu_bo_list_entry *entry)
+{
+	entry->robj = vm->page_directory;
+	entry->prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
+	entry->allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
+	entry->priority = 0;
+	entry->tv.bo = &vm->page_directory->tbo;
+	entry->tv.shared = true;
+	list_add(&entry->tv.head, validated);
+}
+
+/**
+ * amdgpu_vm_get_bos - add the vm BOs to a validation list
+ *
+ * @vm: vm providing the BOs
  * @duplicates: head of duplicates list
  *
  * Add the page directory to the list of BOs to
  * validate for command submission (cayman+).
  */
-struct amdgpu_bo_list_entry *amdgpu_vm_get_bos(struct amdgpu_device *adev,
-					       struct amdgpu_vm *vm,
-					       struct list_head *validated,
+struct amdgpu_bo_list_entry *amdgpu_vm_get_pt_bos(struct amdgpu_vm *vm,
 						  struct list_head *duplicates)
 {
 	struct amdgpu_bo_list_entry *list;
 	unsigned i, idx;
 
-	list = drm_malloc_ab(vm->max_pde_used + 2,
+	list = drm_malloc_ab(vm->max_pde_used + 1,
 			     sizeof(struct amdgpu_bo_list_entry));
-	if (!list) {
+	if (!list)
 		return NULL;
-	}
 
 	/* add the vm page table to the list */
-	list[0].robj = vm->page_directory;
-	list[0].prefered_domains = AMDGPU_GEM_DOMAIN_VRAM;
-	list[0].allowed_domains = AMDGPU_GEM_DOMAIN_VRAM;
-	list[0].priority = 0;
-	list[0].tv.bo = &vm->page_directory->tbo;
-	list[0].tv.shared = true;
-	list_add(&list[0].tv.head, validated);
-
-	for (i = 0, idx = 1; i <= vm->max_pde_used; i++) {
+	for (i = 0, idx = 0; i <= vm->max_pde_used; i++) {
 		if (!vm->page_tables[i].bo)
 			continue;