1 files changed, 378 insertions, 161 deletions

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
index dfa025d694f8..928e8d57cd08 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gpuvm.c
@@ -72,16 +72,16 @@ static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
 	return (struct amdgpu_device *)kgd;
 }
 
-static bool check_if_add_bo_to_vm(struct amdgpu_vm *avm,
+static bool kfd_mem_is_attached(struct amdgpu_vm *avm,
 		struct kgd_mem *mem)
 {
-	struct kfd_bo_va_list *entry;
+	struct kfd_mem_attachment *entry;
 
-	list_for_each_entry(entry, &mem->bo_va_list, bo_list)
+	list_for_each_entry(entry, &mem->attachments, list)
 		if (entry->bo_va->base.vm == avm)
-			return false;
+			return true;
 
-	return true;
+	return false;
 }
 
 /* Set memory usage limits. Current, limits are
@@ -433,7 +433,8 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
 				mapping_flags |= coherent ?
 					AMDGPU_VM_MTYPE_CC : AMDGPU_VM_MTYPE_RW;
 			else
-				mapping_flags |= AMDGPU_VM_MTYPE_UC;
+				mapping_flags |= coherent ?
+					AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
 		} else {
 			mapping_flags |= coherent ?
 				AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
@@ -452,7 +453,8 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
 				if (adev->gmc.xgmi.connected_to_cpu)
 					snoop = true;
 			} else {
-				mapping_flags |= AMDGPU_VM_MTYPE_UC;
+				mapping_flags |= coherent ?
+					AMDGPU_VM_MTYPE_UC : AMDGPU_VM_MTYPE_NC;
 				if (amdgpu_xgmi_same_hive(adev, bo_adev))
 					snoop = true;
 			}
@@ -473,87 +475,318 @@ static uint64_t get_pte_flags(struct amdgpu_device *adev, struct kgd_mem *mem)
 	return pte_flags;
 }
 
-/* add_bo_to_vm - Add a BO to a VM
+static int
+kfd_mem_dmamap_userptr(struct kgd_mem *mem,
+		       struct kfd_mem_attachment *attachment)
+{
+	enum dma_data_direction direction =
+		mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *src_ttm = mem->bo->tbo.ttm;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+	int ret;
+
+	ttm->sg = kmalloc(sizeof(*ttm->sg), GFP_KERNEL);
+	if (unlikely(!ttm->sg))
+		return -ENOMEM;
+
+	if (WARN_ON(ttm->num_pages != src_ttm->num_pages))
+		return -EINVAL;
+
+	/* Same sequence as in amdgpu_ttm_tt_pin_userptr */
+	ret = sg_alloc_table_from_pages(ttm->sg, src_ttm->pages,
+					ttm->num_pages, 0,
+					(u64)ttm->num_pages << PAGE_SHIFT,
+					GFP_KERNEL);
+	if (unlikely(ret))
+		goto free_sg;
+
+	ret = dma_map_sgtable(adev->dev, ttm->sg, direction, 0);
+	if (unlikely(ret))
+		goto release_sg;
+
+	drm_prime_sg_to_dma_addr_array(ttm->sg, ttm->dma_address,
+				       ttm->num_pages);
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+	ret = ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+	if (ret)
+		goto unmap_sg;
+
+	return 0;
+
+unmap_sg:
+	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
+release_sg:
+	pr_err("DMA map userptr failed: %d\n", ret);
+	sg_free_table(ttm->sg);
+free_sg:
+	kfree(ttm->sg);
+	ttm->sg = NULL;
+	return ret;
+}
+
+static int
+kfd_mem_dmamap_dmabuf(struct kfd_mem_attachment *attachment)
+{
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_GTT);
+	return ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+}
+
+static int
+kfd_mem_dmamap_attachment(struct kgd_mem *mem,
+			  struct kfd_mem_attachment *attachment)
+{
+	switch (attachment->type) {
+	case KFD_MEM_ATT_SHARED:
+		return 0;
+	case KFD_MEM_ATT_USERPTR:
+		return kfd_mem_dmamap_userptr(mem, attachment);
+	case KFD_MEM_ATT_DMABUF:
+		return kfd_mem_dmamap_dmabuf(attachment);
+	default:
+		WARN_ON_ONCE(1);
+	}
+	return -EINVAL;
+}
+
+static void
+kfd_mem_dmaunmap_userptr(struct kgd_mem *mem,
+			 struct kfd_mem_attachment *attachment)
+{
+	enum dma_data_direction direction =
+		mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+		DMA_BIDIRECTIONAL : DMA_TO_DEVICE;
+	struct ttm_operation_ctx ctx = {.interruptible = false};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+	struct amdgpu_device *adev = attachment->adev;
+	struct ttm_tt *ttm = bo->tbo.ttm;
+
+	if (unlikely(!ttm->sg))
+		return;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+	ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+
+	dma_unmap_sgtable(adev->dev, ttm->sg, direction, 0);
+	sg_free_table(ttm->sg);
+	ttm->sg = NULL;
+}
+
+static void
+kfd_mem_dmaunmap_dmabuf(struct kfd_mem_attachment *attachment)
+{
+	struct ttm_operation_ctx ctx = {.interruptible = true};
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+
+	amdgpu_bo_placement_from_domain(bo, AMDGPU_GEM_DOMAIN_CPU);
+	ttm_bo_validate(&bo->tbo, &bo->placement, &ctx);
+}
+
+static void
+kfd_mem_dmaunmap_attachment(struct kgd_mem *mem,
+			    struct kfd_mem_attachment *attachment)
+{
+	switch (attachment->type) {
+	case KFD_MEM_ATT_SHARED:
+		break;
+	case KFD_MEM_ATT_USERPTR:
+		kfd_mem_dmaunmap_userptr(mem, attachment);
+		break;
+	case KFD_MEM_ATT_DMABUF:
+		kfd_mem_dmaunmap_dmabuf(attachment);
+		break;
+	default:
+		WARN_ON_ONCE(1);
+	}
+}
+
+static int
+kfd_mem_attach_userptr(struct amdgpu_device *adev, struct kgd_mem *mem,
+		       struct amdgpu_bo **bo)
+{
+	unsigned long bo_size = mem->bo->tbo.base.size;
+	struct drm_gem_object *gobj;
+	int ret;
+
+	ret = amdgpu_bo_reserve(mem->bo, false);
+	if (ret)
+		return ret;
+
+	ret = amdgpu_gem_object_create(adev, bo_size, 1,
+				       AMDGPU_GEM_DOMAIN_CPU,
+				       0, ttm_bo_type_sg,
+				       mem->bo->tbo.base.resv,
+				       &gobj);
+	if (ret)
+		return ret;
+
+	amdgpu_bo_unreserve(mem->bo);
+
+	*bo = gem_to_amdgpu_bo(gobj);
+	(*bo)->parent = amdgpu_bo_ref(mem->bo);
+
+	return 0;
+}
+
+static int
+kfd_mem_attach_dmabuf(struct amdgpu_device *adev, struct kgd_mem *mem,
+		      struct amdgpu_bo **bo)
+{
+	struct drm_gem_object *gobj;
+
+	if (!mem->dmabuf) {
+		mem->dmabuf = amdgpu_gem_prime_export(&mem->bo->tbo.base,
+			mem->alloc_flags & KFD_IOC_ALLOC_MEM_FLAGS_WRITABLE ?
+				DRM_RDWR : 0);
+		if (IS_ERR(mem->dmabuf)) {
+			mem->dmabuf = NULL;
+			return PTR_ERR(mem->dmabuf);
+		}
+	}
+
+	gobj = amdgpu_gem_prime_import(&adev->ddev, mem->dmabuf);
+	if (IS_ERR(gobj))
+		return PTR_ERR(gobj);
+
+	/* Import takes an extra reference on the dmabuf. Drop it now to
+	 * avoid leaking it. We only need the one reference in
+	 * kgd_mem->dmabuf.
+	 */
+	dma_buf_put(mem->dmabuf);
+
+	*bo = gem_to_amdgpu_bo(gobj);
+	(*bo)->parent = amdgpu_bo_ref(mem->bo);
+
+	return 0;
+}
+
+/* kfd_mem_attach - Add a BO to a VM
  *
  * Everything that needs to bo done only once when a BO is first added
  * to a VM. It can later be mapped and unmapped many times without
  * repeating these steps.
  *
+ * 0. Create BO for DMA mapping, if needed
  * 1. Allocate and initialize BO VA entry data structure
  * 2. Add BO to the VM
  * 3. Determine ASIC-specific PTE flags
  * 4. Alloc page tables and directories if needed
  * 4a.  Validate new page tables and directories
  */
-static int add_bo_to_vm(struct amdgpu_device *adev, struct kgd_mem *mem,
-		struct amdgpu_vm *vm, bool is_aql,
-		struct kfd_bo_va_list **p_bo_va_entry)
+static int kfd_mem_attach(struct amdgpu_device *adev, struct kgd_mem *mem,
+		struct amdgpu_vm *vm, bool is_aql)
 {
-	int ret;
-	struct kfd_bo_va_list *bo_va_entry;
-	struct amdgpu_bo *bo = mem->bo;
+	struct amdgpu_device *bo_adev = amdgpu_ttm_adev(mem->bo->tbo.bdev);
+	unsigned long bo_size = mem->bo->tbo.base.size;
 	uint64_t va = mem->va;
-	struct list_head *list_bo_va = &mem->bo_va_list;
-	unsigned long bo_size = bo->tbo.base.size;
+	struct kfd_mem_attachment *attachment[2] = {NULL, NULL};
+	struct amdgpu_bo *bo[2] = {NULL, NULL};
+	int i, ret;
 
 	if (!va) {
 		pr_err("Invalid VA when adding BO to VM\n");
 		return -EINVAL;
 	}
 
-	if (is_aql)
-		va += bo_size;
-
-	bo_va_entry = kzalloc(sizeof(*bo_va_entry), GFP_KERNEL);
-	if (!bo_va_entry)
-		return -ENOMEM;
+	for (i = 0; i <= is_aql; i++) {
+		attachment[i] = kzalloc(sizeof(*attachment[i]), GFP_KERNEL);
+		if (unlikely(!attachment[i])) {
+			ret = -ENOMEM;
+			goto unwind;
+		}
 
-	pr_debug("\t add VA 0x%llx - 0x%llx to vm %p\n", va,
-			va + bo_size, vm);
+		pr_debug("\t add VA 0x%llx - 0x%llx to vm %p\n", va,
+			 va + bo_size, vm);
 
-	/* Add BO to VM internal data structures*/
-	bo_va_entry->bo_va = amdgpu_vm_bo_add(adev, vm, bo);
-	if (!bo_va_entry->bo_va) {
-		ret = -EINVAL;
-		pr_err("Failed to add BO object to VM. ret == %d\n",
-				ret);
-		goto err_vmadd;
-	}
+		if (adev == bo_adev || (mem->domain == AMDGPU_GEM_DOMAIN_VRAM &&
+					amdgpu_xgmi_same_hive(adev, bo_adev))) {
+			/* Mappings on the local GPU and VRAM mappings in the
+			 * local hive share the original BO
+			 */
+			attachment[i]->type = KFD_MEM_ATT_SHARED;
+			bo[i] = mem->bo;
+			drm_gem_object_get(&bo[i]->tbo.base);
+		} else if (i > 0) {
+			/* Multiple mappings on the same GPU share the BO */
+			attachment[i]->type = KFD_MEM_ATT_SHARED;
+			bo[i] = bo[0];
+			drm_gem_object_get(&bo[i]->tbo.base);
+		} else if (amdgpu_ttm_tt_get_usermm(mem->bo->tbo.ttm)) {
+			/* Create an SG BO to DMA-map userptrs on other GPUs */
+			attachment[i]->type = KFD_MEM_ATT_USERPTR;
+			ret = kfd_mem_attach_userptr(adev, mem, &bo[i]);
+			if (ret)
+				goto unwind;
+		} else if (mem->domain == AMDGPU_GEM_DOMAIN_GTT &&
+			   mem->bo->tbo.type != ttm_bo_type_sg) {
+			/* GTT BOs use DMA-mapping ability of dynamic-attach
+			 * DMA bufs. TODO: The same should work for VRAM on
+			 * large-BAR GPUs.
+			 */
+			attachment[i]->type = KFD_MEM_ATT_DMABUF;
+			ret = kfd_mem_attach_dmabuf(adev, mem, &bo[i]);
+			if (ret)
+				goto unwind;
+		} else {
+			/* FIXME: Need to DMA-map other BO types:
+			 * large-BAR VRAM, doorbells, MMIO remap
+			 */
+			attachment[i]->type = KFD_MEM_ATT_SHARED;
+			bo[i] = mem->bo;
+			drm_gem_object_get(&bo[i]->tbo.base);
+		}
 
-	bo_va_entry->va = va;
-	bo_va_entry->pte_flags = get_pte_flags(adev, mem);
-	bo_va_entry->kgd_dev = (void *)adev;
-	list_add(&bo_va_entry->bo_list, list_bo_va);
+		/* Add BO to VM internal data structures */
+		attachment[i]->bo_va = amdgpu_vm_bo_add(adev, vm, bo[i]);
+		if (unlikely(!attachment[i]->bo_va)) {
+			ret = -ENOMEM;
+			pr_err("Failed to add BO object to VM. ret == %d\n",
+			       ret);
+			goto unwind;
+		}
 
-	if (p_bo_va_entry)
-		*p_bo_va_entry = bo_va_entry;
+		attachment[i]->va = va;
+		attachment[i]->pte_flags = get_pte_flags(adev, mem);
+		attachment[i]->adev = adev;
+		list_add(&attachment[i]->list, &mem->attachments);
 
-	/* Allocate validate page tables if needed */
-	ret = vm_validate_pt_pd_bos(vm);
-	if (ret) {
-		pr_err("validate_pt_pd_bos() failed\n");
-		goto err_alloc_pts;
+		va += bo_size;
 	}
 
 	return 0;
 
-err_alloc_pts:
-	amdgpu_vm_bo_rmv(adev, bo_va_entry->bo_va);
-	list_del(&bo_va_entry->bo_list);
-err_vmadd:
-	kfree(bo_va_entry);
+unwind:
+	for (; i >= 0; i--) {
+		if (!attachment[i])
+			continue;
+		if (attachment[i]->bo_va) {
+			amdgpu_vm_bo_rmv(adev, attachment[i]->bo_va);
+			list_del(&attachment[i]->list);
+		}
+		if (bo[i])
+			drm_gem_object_put(&bo[i]->tbo.base);
+		kfree(attachment[i]);
+	}
 	return ret;
 }
 
-static void remove_bo_from_vm(struct amdgpu_device *adev,
-		struct kfd_bo_va_list *entry, unsigned long size)
+static void kfd_mem_detach(struct kfd_mem_attachment *attachment)
 {
-	pr_debug("\t remove VA 0x%llx - 0x%llx in entry %p\n",
-			entry->va,
-			entry->va + size, entry);
-	amdgpu_vm_bo_rmv(adev, entry->bo_va);
-	list_del(&entry->bo_list);
-	kfree(entry);
+	struct amdgpu_bo *bo = attachment->bo_va->base.bo;
+
+	pr_debug("\t remove VA 0x%llx in entry %p\n",
+			attachment->va, attachment);
+	amdgpu_vm_bo_rmv(attachment->adev, attachment->bo_va);
+	drm_gem_object_put(&bo->tbo.base);
+	list_del(&attachment->list);
+	kfree(attachment);
 }
 
 static void add_kgd_mem_to_kfd_bo_list(struct kgd_mem *mem,
@@ -728,7 +961,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
 				struct bo_vm_reservation_context *ctx)
 {
 	struct amdgpu_bo *bo = mem->bo;
-	struct kfd_bo_va_list *entry;
+	struct kfd_mem_attachment *entry;
 	unsigned int i;
 	int ret;
 
@@ -740,7 +973,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
 	INIT_LIST_HEAD(&ctx->list);
 	INIT_LIST_HEAD(&ctx->duplicates);
 
-	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
+	list_for_each_entry(entry, &mem->attachments, list) {
 		if ((vm && vm != entry->bo_va->base.vm) ||
 			(entry->is_mapped != map_type
 			&& map_type != BO_VM_ALL))
@@ -762,7 +995,7 @@ static int reserve_bo_and_cond_vms(struct kgd_mem *mem,
 	list_add(&ctx->kfd_bo.tv.head, &ctx->list);
 
 	i = 0;
-	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
+	list_for_each_entry(entry, &mem->attachments, list) {
 		if ((vm && vm != entry->bo_va->base.vm) ||
 			(entry->is_mapped != map_type
 			&& map_type != BO_VM_ALL))
@@ -816,11 +1049,12 @@ static int unreserve_bo_and_vms(struct bo_vm_reservation_context *ctx,
 	return ret;
 }
 
-static int unmap_bo_from_gpuvm(struct amdgpu_device *adev,
-				struct kfd_bo_va_list *entry,
+static void unmap_bo_from_gpuvm(struct kgd_mem *mem,
+				struct kfd_mem_attachment *entry,
 				struct amdgpu_sync *sync)
 {
 	struct amdgpu_bo_va *bo_va = entry->bo_va;
+	struct amdgpu_device *adev = entry->adev;
 	struct amdgpu_vm *vm = bo_va->base.vm;
 
 	amdgpu_vm_bo_unmap(adev, bo_va, entry->va);
@@ -829,15 +1063,20 @@ static int unmap_bo_from_gpuvm(struct amdgpu_device *adev,
 
 	amdgpu_sync_fence(sync, bo_va->last_pt_update);
 
-	return 0;
+	kfd_mem_dmaunmap_attachment(mem, entry);
 }
 
-static int update_gpuvm_pte(struct amdgpu_device *adev,
-		struct kfd_bo_va_list *entry,
-		struct amdgpu_sync *sync)
+static int update_gpuvm_pte(struct kgd_mem *mem,
+			    struct kfd_mem_attachment *entry,
+			    struct amdgpu_sync *sync)
 {
-	int ret;
 	struct amdgpu_bo_va *bo_va = entry->bo_va;
+	struct amdgpu_device *adev = entry->adev;
+	int ret;
+
+	ret = kfd_mem_dmamap_attachment(mem, entry);
+	if (ret)
+		return ret;
 
 	/* Update the page tables  */
 	ret = amdgpu_vm_bo_update(adev, bo_va, false);
@@ -849,14 +1088,15 @@ static int update_gpuvm_pte(struct amdgpu_device *adev,
 	return amdgpu_sync_fence(sync, bo_va->last_pt_update);
 }
 
-static int map_bo_to_gpuvm(struct amdgpu_device *adev,
-		struct kfd_bo_va_list *entry, struct amdgpu_sync *sync,
-		bool no_update_pte)
+static int map_bo_to_gpuvm(struct kgd_mem *mem,
+			   struct kfd_mem_attachment *entry,
+			   struct amdgpu_sync *sync,
+			   bool no_update_pte)
 {
 	int ret;
 
 	/* Set virtual address for the allocation */
-	ret = amdgpu_vm_bo_map(adev, entry->bo_va, entry->va, 0,
+	ret = amdgpu_vm_bo_map(entry->adev, entry->bo_va, entry->va, 0,
 			       amdgpu_bo_size(entry->bo_va->base.bo),
 			       entry->pte_flags);
 	if (ret) {
@@ -868,7 +1108,7 @@ static int map_bo_to_gpuvm(struct amdgpu_device *adev,
 	if (no_update_pte)
 		return 0;
 
-	ret = update_gpuvm_pte(adev, entry, sync);
+	ret = update_gpuvm_pte(mem, entry, sync);
 	if (ret) {
 		pr_err("update_gpuvm_pte() failed\n");
 		goto update_gpuvm_pte_failed;
@@ -877,7 +1117,7 @@ static int map_bo_to_gpuvm(struct amdgpu_device *adev,
 	return 0;
 
 update_gpuvm_pte_failed:
-	unmap_bo_from_gpuvm(adev, entry, sync);
+	unmap_bo_from_gpuvm(mem, entry, sync);
 	return ret;
 }
 
@@ -1194,7 +1434,7 @@ int amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu(
 		ret = -ENOMEM;
 		goto err;
 	}
-	INIT_LIST_HEAD(&(*mem)->bo_va_list);
+	INIT_LIST_HEAD(&(*mem)->attachments);
 	mutex_init(&(*mem)->lock);
 	(*mem)->aql_queue = !!(flags & KFD_IOC_ALLOC_MEM_FLAGS_AQL_QUEUE_MEM);
 
@@ -1283,7 +1523,7 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 {
 	struct amdkfd_process_info *process_info = mem->process_info;
 	unsigned long bo_size = mem->bo->tbo.base.size;
-	struct kfd_bo_va_list *entry, *tmp;
+	struct kfd_mem_attachment *entry, *tmp;
 	struct bo_vm_reservation_context ctx;
 	struct ttm_validate_buffer *bo_list_entry;
 	unsigned int mapped_to_gpu_memory;
@@ -1326,13 +1566,12 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 	pr_debug("Release VA 0x%llx - 0x%llx\n", mem->va,
 		mem->va + bo_size * (1 + mem->aql_queue));
 
-	/* Remove from VM internal data structures */
-	list_for_each_entry_safe(entry, tmp, &mem->bo_va_list, bo_list)
-		remove_bo_from_vm((struct amdgpu_device *)entry->kgd_dev,
-				entry, bo_size);
-
 	ret = unreserve_bo_and_vms(&ctx, false, false);
 
+	/* Remove from VM internal data structures */
+	list_for_each_entry_safe(entry, tmp, &mem->attachments, list)
+		kfd_mem_detach(entry);
+
 	/* Free the sync object */
 	amdgpu_sync_free(&mem->sync);
 
@@ -1357,6 +1596,8 @@ int amdgpu_amdkfd_gpuvm_free_memory_of_gpu(
 
 	/* Free the BO*/
 	drm_vma_node_revoke(&mem->bo->tbo.base.vma_node, drm_priv);
+	if (mem->dmabuf)
+		dma_buf_put(mem->dmabuf);
 	drm_gem_object_put(&mem->bo->tbo.base);
 	mutex_destroy(&mem->lock);
 	kfree(mem);
@@ -1372,10 +1613,8 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	int ret;
 	struct amdgpu_bo *bo;
 	uint32_t domain;
-	struct kfd_bo_va_list *entry;
+	struct kfd_mem_attachment *entry;
 	struct bo_vm_reservation_context ctx;
-	struct kfd_bo_va_list *bo_va_entry = NULL;
-	struct kfd_bo_va_list *bo_va_entry_aql = NULL;
 	unsigned long bo_size;
 	bool is_invalid_userptr = false;
 
@@ -1411,6 +1650,12 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 			mem->va + bo_size * (1 + mem->aql_queue),
 			avm, domain_string(domain));
 
+	if (!kfd_mem_is_attached(avm, mem)) {
+		ret = kfd_mem_attach(adev, mem, avm, mem->aql_queue);
+		if (ret)
+			goto out;
+	}
+
 	ret = reserve_bo_and_vm(mem, avm, &ctx);
 	if (unlikely(ret))
 		goto out;
@@ -1424,22 +1669,9 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 	    bo->tbo.mem.mem_type == TTM_PL_SYSTEM)
 		is_invalid_userptr = true;
 
-	if (check_if_add_bo_to_vm(avm, mem)) {
-		ret = add_bo_to_vm(adev, mem, avm, false,
-				&bo_va_entry);
-		if (ret)
-			goto add_bo_to_vm_failed;
-		if (mem->aql_queue) {
-			ret = add_bo_to_vm(adev, mem, avm,
-					true, &bo_va_entry_aql);
-			if (ret)
-				goto add_bo_to_vm_failed_aql;
-		}
-	} else {
-		ret = vm_validate_pt_pd_bos(avm);
-		if (unlikely(ret))
-			goto add_bo_to_vm_failed;
-	}
+	ret = vm_validate_pt_pd_bos(avm);
+	if (unlikely(ret))
+		goto out_unreserve;
 
 	if (mem->mapped_to_gpu_memory == 0 &&
 	    !amdgpu_ttm_tt_get_usermm(bo->tbo.ttm)) {
@@ -1450,34 +1682,34 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 		ret = amdgpu_amdkfd_bo_validate(bo, domain, true);
 		if (ret) {
 			pr_debug("Validate failed\n");
-			goto map_bo_to_gpuvm_failed;
+			goto out_unreserve;
 		}
 	}
 
-	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
-		if (entry->bo_va->base.vm == avm && !entry->is_mapped) {
-			pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n",
-					entry->va, entry->va + bo_size,
-					entry);
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->bo_va->base.vm != avm || entry->is_mapped)
+			continue;
 
-			ret = map_bo_to_gpuvm(adev, entry, ctx.sync,
-					      is_invalid_userptr);
-			if (ret) {
-				pr_err("Failed to map bo to gpuvm\n");
-				goto map_bo_to_gpuvm_failed;
-			}
+		pr_debug("\t map VA 0x%llx - 0x%llx in entry %p\n",
+			 entry->va, entry->va + bo_size, entry);
 
-			ret = vm_update_pds(avm, ctx.sync);
-			if (ret) {
-				pr_err("Failed to update page directories\n");
-				goto map_bo_to_gpuvm_failed;
-			}
+		ret = map_bo_to_gpuvm(mem, entry, ctx.sync,
+				      is_invalid_userptr);
+		if (ret) {
+			pr_err("Failed to map bo to gpuvm\n");
+			goto out_unreserve;
+		}
 
-			entry->is_mapped = true;
-			mem->mapped_to_gpu_memory++;
-			pr_debug("\t INC mapping count %d\n",
-					mem->mapped_to_gpu_memory);
+		ret = vm_update_pds(avm, ctx.sync);
+		if (ret) {
+			pr_err("Failed to update page directories\n");
+			goto out_unreserve;
 		}
+
+		entry->is_mapped = true;
+		mem->mapped_to_gpu_memory++;
+		pr_debug("\t INC mapping count %d\n",
+			 mem->mapped_to_gpu_memory);
 	}
 
 	if (!amdgpu_ttm_tt_get_usermm(bo->tbo.ttm) && !bo->tbo.pin_count)
@@ -1488,13 +1720,7 @@ int amdgpu_amdkfd_gpuvm_map_memory_to_gpu(
 
 	goto out;
 
-map_bo_to_gpuvm_failed:
-	if (bo_va_entry_aql)
-		remove_bo_from_vm(adev, bo_va_entry_aql, bo_size);
-add_bo_to_vm_failed_aql:
-	if (bo_va_entry)
-		remove_bo_from_vm(adev, bo_va_entry, bo_size);
-add_bo_to_vm_failed:
+out_unreserve:
 	unreserve_bo_and_vms(&ctx, false, false);
 out:
 	mutex_unlock(&mem->process_info->lock);
@@ -1505,11 +1731,10 @@ out:
 int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 		struct kgd_dev *kgd, struct kgd_mem *mem, void *drm_priv)
 {
-	struct amdgpu_device *adev = get_amdgpu_device(kgd);
 	struct amdgpu_vm *avm = drm_priv_to_vm(drm_priv);
 	struct amdkfd_process_info *process_info = avm->process_info;
 	unsigned long bo_size = mem->bo->tbo.base.size;
-	struct kfd_bo_va_list *entry;
+	struct kfd_mem_attachment *entry;
 	struct bo_vm_reservation_context ctx;
 	int ret;
 
@@ -1533,26 +1758,19 @@ int amdgpu_amdkfd_gpuvm_unmap_memory_from_gpu(
 		mem->va + bo_size * (1 + mem->aql_queue),
 		avm);
 
-	list_for_each_entry(entry, &mem->bo_va_list, bo_list) {
-		if (entry->bo_va->base.vm == avm && entry->is_mapped) {
-			pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n",
-					entry->va,
-					entry->va + bo_size,
-					entry);
+	list_for_each_entry(entry, &mem->attachments, list) {
+		if (entry->bo_va->base.vm != avm || !entry->is_mapped)
+			continue;
 
-			ret = unmap_bo_from_gpuvm(adev, entry, ctx.sync);
-			if (ret == 0) {
-				entry->is_mapped = false;
-			} else {
-				pr_err("failed to unmap VA 0x%llx\n",
-						mem->va);
-				goto unreserve_out;
-			}
+		pr_debug("\t unmap VA 0x%llx - 0x%llx from entry %p\n",
+			 entry->va, entry->va + bo_size, entry);
 
-			mem->mapped_to_gpu_memory--;
-			pr_debug("\t DEC mapping count %d\n",
-					mem->mapped_to_gpu_memory);
-		}
+		unmap_bo_from_gpuvm(mem, entry, ctx.sync);
+		entry->is_mapped = false;
+
+		mem->mapped_to_gpu_memory--;
+		pr_debug("\t DEC mapping count %d\n",
+			 mem->mapped_to_gpu_memory);
 	}
 
 	/* If BO is unmapped from all VMs, unfence it. It can be evicted if
@@ -1701,7 +1919,7 @@ int amdgpu_amdkfd_gpuvm_import_dmabuf(struct kgd_dev *kgd,
 	if (mmap_offset)
 		*mmap_offset = amdgpu_bo_mmap_offset(bo);
 
-	INIT_LIST_HEAD(&(*mem)->bo_va_list);
+	INIT_LIST_HEAD(&(*mem)->attachments);
 	mutex_init(&(*mem)->lock);
 
 	(*mem)->alloc_flags =
@@ -1898,7 +2116,7 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
 	list_for_each_entry_safe(mem, tmp_mem,
 				 &process_info->userptr_inval_list,
 				 validate_list.head) {
-		struct kfd_bo_va_list *bo_va_entry;
+		struct kfd_mem_attachment *attachment;
 
 		bo = mem->bo;
 
@@ -1921,13 +2139,12 @@ static int validate_invalid_user_pages(struct amdkfd_process_info *process_info)
 		 * VM faults if the GPU tries to access the invalid
 		 * memory.
 		 */
-		list_for_each_entry(bo_va_entry, &mem->bo_va_list, bo_list) {
-			if (!bo_va_entry->is_mapped)
+		list_for_each_entry(attachment, &mem->attachments, list) {
+			if (!attachment->is_mapped)
 				continue;
 
-			ret = update_gpuvm_pte((struct amdgpu_device *)
-					       bo_va_entry->kgd_dev,
-					       bo_va_entry, &sync);
+			kfd_mem_dmaunmap_attachment(mem, attachment);
+			ret = update_gpuvm_pte(mem, attachment, &sync);
 			if (ret) {
 				pr_err("%s: update PTE failed\n", __func__);
 				/* make sure this gets validated again */
@@ -2108,7 +2325,7 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
 
 		struct amdgpu_bo *bo = mem->bo;
 		uint32_t domain = mem->domain;
-		struct kfd_bo_va_list *bo_va_entry;
+		struct kfd_mem_attachment *attachment;
 
 		total_size += amdgpu_bo_size(bo);
 
@@ -2128,12 +2345,12 @@ int amdgpu_amdkfd_gpuvm_restore_process_bos(void *info, struct dma_fence **ef)
 			pr_debug("Memory eviction: Sync BO fence failed. Try again\n");
 			goto validate_map_fail;
 		}
-		list_for_each_entry(bo_va_entry, &mem->bo_va_list,
-				    bo_list) {
-			ret = update_gpuvm_pte((struct amdgpu_device *)
-					      bo_va_entry->kgd_dev,
-					      bo_va_entry,
-					      &sync_obj);
+		list_for_each_entry(attachment, &mem->attachments, list) {
+			if (!attachment->is_mapped)
+				continue;
+
+			kfd_mem_dmaunmap_attachment(mem, attachment);
+			ret = update_gpuvm_pte(mem, attachment, &sync_obj);
 			if (ret) {
 				pr_debug("Memory eviction: update PTE failed. Try again\n");
 				goto validate_map_fail;
@@ -2208,7 +2425,7 @@ int amdgpu_amdkfd_add_gws_to_process(void *info, void *gws, struct kgd_mem **mem
 		return -ENOMEM;
 
 	mutex_init(&(*mem)->lock);
-	INIT_LIST_HEAD(&(*mem)->bo_va_list);
+	INIT_LIST_HEAD(&(*mem)->attachments);
 	(*mem)->bo = amdgpu_bo_ref(gws_bo);
 	(*mem)->domain = AMDGPU_GEM_DOMAIN_GWS;
 	(*mem)->process_info = process_info;