RDMA/odp: Make it clearer when a umem is an implicit ODP umem

Implicit ODP umems are special, they don't have any page lists, they don't exist in the interval tree and they are never DMA mapped. Instead of trying to guess this based on a zero length use an explicit flag. Further, do not allow non-implicit umems to be 0 size. Link: https://lore.kernel.org/r/[email protected] Signed-off-by: Leon Romanovsky <[email protected]> Signed-off-by: Jason Gunthorpe <[email protected]>
jakajancar · Aug 21, 2019 · fd7dbf0 · fd7dbf0
1 parent f993de8
commit fd7dbf0
Show file tree

Hide file tree

Showing 4 changed files with 40 additions and 26 deletions.
diff --git a/drivers/infiniband/core/umem_odp.c b/drivers/infiniband/core/umem_odp.c
@@ -176,18 +176,15 @@ static void add_umem_to_per_mm(struct ib_umem_odp *umem_odp)
 	struct ib_ucontext_per_mm *per_mm = umem_odp->per_mm;
 
 	down_write(&per_mm->umem_rwsem);
-	if (likely(ib_umem_start(umem_odp) != ib_umem_end(umem_odp))) {
-		/*
-		 * Note that the representation of the intervals in the
-		 * interval tree considers the ending point as contained in
-		 * the interval, while the function ib_umem_end returns the
-		 * first address which is not contained in the umem.
-		 */
-		umem_odp->interval_tree.start = ib_umem_start(umem_odp);
-		umem_odp->interval_tree.last = ib_umem_end(umem_odp) - 1;
-		interval_tree_insert(&umem_odp->interval_tree,
-				     &per_mm->umem_tree);
-	}
+	/*
+	 * Note that the representation of the intervals in the interval tree
+	 * considers the ending point as contained in the interval, while the
+	 * function ib_umem_end returns the first address which is not
+	 * contained in the umem.
+	 */
+	umem_odp->interval_tree.start = ib_umem_start(umem_odp);
+	umem_odp->interval_tree.last = ib_umem_end(umem_odp) - 1;
+	interval_tree_insert(&umem_odp->interval_tree, &per_mm->umem_tree);
 	up_write(&per_mm->umem_rwsem);
 }
 
@@ -196,11 +193,8 @@ static void remove_umem_from_per_mm(struct ib_umem_odp *umem_odp)
 	struct ib_ucontext_per_mm *per_mm = umem_odp->per_mm;
 
 	down_write(&per_mm->umem_rwsem);
-	if (likely(ib_umem_start(umem_odp) != ib_umem_end(umem_odp)))
-		interval_tree_remove(&umem_odp->interval_tree,
-				     &per_mm->umem_tree);
+	interval_tree_remove(&umem_odp->interval_tree, &per_mm->umem_tree);
 	complete_all(&umem_odp->notifier_completion);
-
 	up_write(&per_mm->umem_rwsem);
 }
 
@@ -320,6 +314,9 @@ struct ib_umem_odp *ib_alloc_odp_umem(struct ib_umem_odp *root,
 	int pages = size >> PAGE_SHIFT;
 	int ret;
 
+	if (!size)
+		return ERR_PTR(-EINVAL);
+
 	odp_data = kzalloc(sizeof(*odp_data), GFP_KERNEL);
 	if (!odp_data)
 		return ERR_PTR(-ENOMEM);
@@ -381,6 +378,9 @@ int ib_umem_odp_get(struct ib_umem_odp *umem_odp, int access)
 	struct mm_struct *mm = umem->owning_mm;
 	int ret_val;
 
+	if (umem_odp->umem.address == 0 && umem_odp->umem.length == 0)
+		umem_odp->is_implicit_odp = 1;
+
 	umem_odp->page_shift = PAGE_SHIFT;
 	if (access & IB_ACCESS_HUGETLB) {
 		struct vm_area_struct *vma;
@@ -401,7 +401,10 @@ int ib_umem_odp_get(struct ib_umem_odp *umem_odp, int access)
 
 	init_completion(&umem_odp->notifier_completion);
 
-	if (ib_umem_odp_num_pages(umem_odp)) {
+	if (!umem_odp->is_implicit_odp) {
+		if (!ib_umem_odp_num_pages(umem_odp))
+			return -EINVAL;
+
 		umem_odp->page_list =
 			vzalloc(array_size(sizeof(*umem_odp->page_list),
 					   ib_umem_odp_num_pages(umem_odp)));
@@ -420,7 +423,9 @@ int ib_umem_odp_get(struct ib_umem_odp *umem_odp, int access)
 	ret_val = get_per_mm(umem_odp);
 	if (ret_val)
 		goto out_dma_list;
-	add_umem_to_per_mm(umem_odp);
+
+	if (!umem_odp->is_implicit_odp)
+		add_umem_to_per_mm(umem_odp);
 
 	return 0;
 
@@ -439,13 +444,14 @@ void ib_umem_odp_release(struct ib_umem_odp *umem_odp)
 	 * It is the driver's responsibility to ensure, before calling us,
 	 * that the hardware will not attempt to access the MR any more.
 	 */
-	ib_umem_odp_unmap_dma_pages(umem_odp, ib_umem_start(umem_odp),
-				    ib_umem_end(umem_odp));
-
-	remove_umem_from_per_mm(umem_odp);
+	if (!umem_odp->is_implicit_odp) {
+		ib_umem_odp_unmap_dma_pages(umem_odp, ib_umem_start(umem_odp),
+					    ib_umem_end(umem_odp));
+		remove_umem_from_per_mm(umem_odp);
+		vfree(umem_odp->dma_list);
+		vfree(umem_odp->page_list);
+	}
 	put_per_mm(umem_odp);
-	vfree(umem_odp->dma_list);
-	vfree(umem_odp->page_list);
 }
 
 /*

diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
@@ -1600,7 +1600,7 @@ static void dereg_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
 		/* Wait for all running page-fault handlers to finish. */
 		synchronize_srcu(&dev->mr_srcu);
 		/* Destroy all page mappings */
-		if (umem_odp->page_list)
+		if (!umem_odp->is_implicit_odp)
 			mlx5_ib_invalidate_range(umem_odp,
 						 ib_umem_start(umem_odp),
 						 ib_umem_end(umem_odp));

diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
@@ -584,7 +584,7 @@ static int pagefault_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr,
 	struct ib_umem_odp *odp;
 	size_t size;
 
-	if (!odp_mr->page_list) {
+	if (odp_mr->is_implicit_odp) {
 		odp = implicit_mr_get_data(mr, io_virt, bcnt);
 
 		if (IS_ERR(odp))

diff --git a/include/rdma/ib_umem_odp.h b/include/rdma/ib_umem_odp.h
@@ -69,6 +69,14 @@ struct ib_umem_odp {
 	/* Tree tracking */
 	struct interval_tree_node interval_tree;
 
+	/*
+	 * An implicit odp umem cannot be DMA mapped, has 0 length, and serves
+	 * only as an anchor for the driver to hold onto the per_mm. FIXME:
+	 * This should be removed and drivers should work with the per_mm
+	 * directly.
+	 */
+	bool is_implicit_odp;
+
 	struct completion	notifier_completion;
 	int			dying;
 	unsigned int		page_shift;