IB/umem: Update on demand page (ODP) support

Currently ODP MR may explicitly register virtual address space area
of limited length.
This change allows MR to cover entire process virtual address space
dynamicaly adding/removing translation entries to device MTT.

Add following changes to support implicit MR:
* Allow umem to be zero size to back-up implicit MR.
* Add new function ib_alloc_odp_umem() to add virtual memory regions
  to implicit MR dynamically on demand.
* Add new function rbt_ib_umem_lookup() to find dynamically added
  virtual memory regions.
* Expose function rbt_ib_umem_for_each_in_range() to other modules and
  make it safe

Signed-off-by: Artemy Kovalyov <artemyko@mellanox.com>
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Signed-off-by: Doug Ledford <dledford@redhat.com>

drivers/infiniband/core/umem.c

@@ -99,9 +99,6 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr,
 	if (dmasync)
 		dma_attrs |= DMA_ATTR_WRITE_BARRIER;
 
-	if (!size)
-		return ERR_PTR(-EINVAL);
-
 	/*
 	 * If the combination of the addr and size requested for this memory
 	 * region causes an integer overflow, return error.

drivers/infiniband/core/umem_odp.c

@@ -239,6 +239,71 @@ static const struct mmu_notifier_ops ib_umem_notifiers = {
 	.invalidate_range_end       = ib_umem_notifier_invalidate_range_end,
 };
 
+struct ib_umem *ib_alloc_odp_umem(struct ib_ucontext *context,
+				  unsigned long addr,
+				  size_t size)
+{
+	struct ib_umem *umem;
+	struct ib_umem_odp *odp_data;
+	int pages = size >> PAGE_SHIFT;
+	int ret;
+
+	umem = kzalloc(sizeof(*umem), GFP_KERNEL);
+	if (!umem)
+		return ERR_PTR(-ENOMEM);
+
+	umem->context   = context;
+	umem->length    = size;
+	umem->address   = addr;
+	umem->page_size = PAGE_SIZE;
+	umem->writable  = 1;
+
+	odp_data = kzalloc(sizeof(*odp_data), GFP_KERNEL);
+	if (!odp_data) {
+		ret = -ENOMEM;
+		goto out_umem;
+	}
+	odp_data->umem = umem;
+
+	mutex_init(&odp_data->umem_mutex);
+	init_completion(&odp_data->notifier_completion);
+
+	odp_data->page_list = vzalloc(pages * sizeof(*odp_data->page_list));
+	if (!odp_data->page_list) {
+		ret = -ENOMEM;
+		goto out_odp_data;
+	}
+
+	odp_data->dma_list = vzalloc(pages * sizeof(*odp_data->dma_list));
+	if (!odp_data->dma_list) {
+		ret = -ENOMEM;
+		goto out_page_list;
+	}
+
+	down_write(&context->umem_rwsem);
+	context->odp_mrs_count++;
+	rbt_ib_umem_insert(&odp_data->interval_tree, &context->umem_tree);
+	if (likely(!atomic_read(&context->notifier_count)))
+		odp_data->mn_counters_active = true;
+	else
+		list_add(&odp_data->no_private_counters,
+			 &context->no_private_counters);
+	up_write(&context->umem_rwsem);
+
+	umem->odp_data = odp_data;
+
+	return umem;
+
+out_page_list:
+	vfree(odp_data->page_list);
+out_odp_data:
+	kfree(odp_data);
+out_umem:
+	kfree(umem);
+	return ERR_PTR(ret);
+}
+EXPORT_SYMBOL(ib_alloc_odp_umem);
+
 int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem)
 {
 	int ret_val;
@@ -270,18 +335,20 @@ int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem)
 
 	init_completion(&umem->odp_data->notifier_completion);
 
-	umem->odp_data->page_list = vzalloc(ib_umem_num_pages(umem) *
+	if (ib_umem_num_pages(umem)) {
+		umem->odp_data->page_list = vzalloc(ib_umem_num_pages(umem) *
 					    sizeof(*umem->odp_data->page_list));
-	if (!umem->odp_data->page_list) {
-		ret_val = -ENOMEM;
-		goto out_odp_data;
-	}
+		if (!umem->odp_data->page_list) {
+			ret_val = -ENOMEM;
+			goto out_odp_data;
+		}
 
-	umem->odp_data->dma_list = vzalloc(ib_umem_num_pages(umem) *
+		umem->odp_data->dma_list = vzalloc(ib_umem_num_pages(umem) *
 					  sizeof(*umem->odp_data->dma_list));
-	if (!umem->odp_data->dma_list) {
-		ret_val = -ENOMEM;
-		goto out_page_list;
+		if (!umem->odp_data->dma_list) {
+			ret_val = -ENOMEM;
+			goto out_page_list;
+		}
 	}
 
 	/*
@@ -466,6 +533,7 @@ static int ib_umem_odp_map_dma_single_page(
 		}
 		umem->odp_data->dma_list[page_index] = dma_addr | access_mask;
 		umem->odp_data->page_list[page_index] = page;
+		umem->npages++;
 		stored_page = 1;
 	} else if (umem->odp_data->page_list[page_index] == page) {
 		umem->odp_data->dma_list[page_index] |= access_mask;
@@ -665,6 +733,7 @@ void ib_umem_odp_unmap_dma_pages(struct ib_umem *umem, u64 virt,
 				put_page(page);
 			umem->odp_data->page_list[idx] = NULL;
 			umem->odp_data->dma_list[idx] = 0;
+			umem->npages--;
 		}
 	}
 	mutex_unlock(&umem->odp_data->umem_mutex);

drivers/infiniband/core/umem_rbtree.c

@@ -78,17 +78,32 @@ int rbt_ib_umem_for_each_in_range(struct rb_root *root,
 				  void *cookie)
 {
 	int ret_val = 0;
-	struct umem_odp_node *node;
+	struct umem_odp_node *node, *next;
 	struct ib_umem_odp *umem;
 
 	if (unlikely(start == last))
 		return ret_val;
 
-	for (node = rbt_ib_umem_iter_first(root, start, last - 1); node;
-			node = rbt_ib_umem_iter_next(node, start, last - 1)) {
+	for (node = rbt_ib_umem_iter_first(root, start, last - 1);
+			node; node = next) {
+		next = rbt_ib_umem_iter_next(node, start, last - 1);
 		umem = container_of(node, struct ib_umem_odp, interval_tree);
 		ret_val = cb(umem->umem, start, last, cookie) || ret_val;
 	}
 
 	return ret_val;
 }
+EXPORT_SYMBOL(rbt_ib_umem_for_each_in_range);
+
+struct ib_umem_odp *rbt_ib_umem_lookup(struct rb_root *root,
+				       u64 addr, u64 length)
+{
+	struct umem_odp_node *node;
+
+	node = rbt_ib_umem_iter_first(root, addr, addr + length - 1);
+	if (node)
+		return container_of(node, struct ib_umem_odp, interval_tree);
+	return NULL;
+
+}
+EXPORT_SYMBOL(rbt_ib_umem_lookup);

include/rdma/ib_umem_odp.h

@@ -79,11 +79,15 @@ struct ib_umem_odp {
 
 	struct completion	notifier_completion;
 	int			dying;
+	struct work_struct	work;
 };
 
 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
 
 int ib_umem_odp_get(struct ib_ucontext *context, struct ib_umem *umem);
+struct ib_umem *ib_alloc_odp_umem(struct ib_ucontext *context,
+				  unsigned long addr,
+				  size_t size);
 
 void ib_umem_odp_release(struct ib_umem *umem);
 
@@ -117,10 +121,12 @@ typedef int (*umem_call_back)(struct ib_umem *item, u64 start, u64 end,
 int rbt_ib_umem_for_each_in_range(struct rb_root *root, u64 start, u64 end,
 				  umem_call_back cb, void *cookie);
 
-struct umem_odp_node *rbt_ib_umem_iter_first(struct rb_root *root,
-					     u64 start, u64 last);
-struct umem_odp_node *rbt_ib_umem_iter_next(struct umem_odp_node *node,
-					    u64 start, u64 last);
+/*
+ * Find first region intersecting with address range.
+ * Return NULL if not found
+ */
+struct ib_umem_odp *rbt_ib_umem_lookup(struct rb_root *root,
+				       u64 addr, u64 length);
 
 static inline int ib_umem_mmu_notifier_retry(struct ib_umem *item,
 					     unsigned long mmu_seq)
@@ -153,6 +159,13 @@ static inline int ib_umem_odp_get(struct ib_ucontext *context,
 	return -EINVAL;
 }
 
+static inline struct ib_umem *ib_alloc_odp_umem(struct ib_ucontext *context,
+						unsigned long addr,
+						size_t size)
+{
+	return ERR_PTR(-EINVAL);
+}
+
 static inline void ib_umem_odp_release(struct ib_umem *umem) {}
 
 #endif /* CONFIG_INFINIBAND_ON_DEMAND_PAGING */