mm: Convert workingset to XArray

We construct an XA_STATE and use it to delete the node with
xas_store() rather than adding a special function for this unique
use case.  Includes a test that simulates this usage for the
test suite.

Signed-off-by: Matthew Wilcox <willy@infradead.org>

include/linux/swap.h

@@ -306,15 +306,6 @@ void workingset_update_node(struct xa_node *node);
 		xas_set_update(xas, workingset_update_node);		\
 } while (0)
 
-/* Returns workingset_update_node() if the mapping has shadow entries. */
-#define workingset_lookup_update(mapping)				\
-({									\
-	radix_tree_update_node_t __helper = workingset_update_node;	\
-	if (dax_mapping(mapping) || shmem_mapping(mapping))		\
-		__helper = NULL;					\
-	__helper;							\
-})
-
 /* linux/mm/page_alloc.c */
 extern unsigned long totalram_pages;
 extern unsigned long totalreserve_pages;

lib/test_xarray.c

@@ -863,6 +863,67 @@ static noinline void check_create_range(struct xarray *xa)
 	check_create_range_3();
 }
 
+static LIST_HEAD(shadow_nodes);
+
+static void test_update_node(struct xa_node *node)
+{
+	if (node->count && node->count == node->nr_values) {
+		if (list_empty(&node->private_list))
+			list_add(&shadow_nodes, &node->private_list);
+	} else {
+		if (!list_empty(&node->private_list))
+			list_del_init(&node->private_list);
+	}
+}
+
+static noinline void shadow_remove(struct xarray *xa)
+{
+	struct xa_node *node;
+
+	xa_lock(xa);
+	while ((node = list_first_entry_or_null(&shadow_nodes,
+					struct xa_node, private_list))) {
+		XA_STATE(xas, node->array, 0);
+		XA_BUG_ON(xa, node->array != xa);
+		list_del_init(&node->private_list);
+		xas.xa_node = xa_parent_locked(node->array, node);
+		xas.xa_offset = node->offset;
+		xas.xa_shift = node->shift + XA_CHUNK_SHIFT;
+		xas_set_update(&xas, test_update_node);
+		xas_store(&xas, NULL);
+	}
+	xa_unlock(xa);
+}
+
+static noinline void check_workingset(struct xarray *xa, unsigned long index)
+{
+	XA_STATE(xas, xa, index);
+	xas_set_update(&xas, test_update_node);
+
+	do {
+		xas_lock(&xas);
+		xas_store(&xas, xa_mk_value(0));
+		xas_next(&xas);
+		xas_store(&xas, xa_mk_value(1));
+		xas_unlock(&xas);
+	} while (xas_nomem(&xas, GFP_KERNEL));
+
+	XA_BUG_ON(xa, list_empty(&shadow_nodes));
+
+	xas_lock(&xas);
+	xas_next(&xas);
+	xas_store(&xas, &xas);
+	XA_BUG_ON(xa, !list_empty(&shadow_nodes));
+
+	xas_store(&xas, xa_mk_value(2));
+	xas_unlock(&xas);
+	XA_BUG_ON(xa, list_empty(&shadow_nodes));
+
+	shadow_remove(xa);
+	XA_BUG_ON(xa, !list_empty(&shadow_nodes));
+	XA_BUG_ON(xa, !xa_empty(xa));
+}
+
 static noinline void check_destroy(struct xarray *xa)
 {
 	unsigned long index;
@@ -916,6 +977,10 @@ static int xarray_checks(void)
 	check_create_range(&array);
 	check_store_iter(&array);
 
+	check_workingset(&array, 0);
+	check_workingset(&array, 64);
+	check_workingset(&array, 4096);
+
 	printk("XArray: %u of %u tests passed\n", tests_passed, tests_run);
 	return (tests_run == tests_passed) ? 0 : -EINVAL;
 }

mm/workingset.c

@@ -148,7 +148,7 @@
  * and activations is maintained (node->inactive_age).
  *
  * On eviction, a snapshot of this counter (along with some bits to
- * identify the node) is stored in the now empty page cache radix tree
+ * identify the node) is stored in the now empty page cache
  * slot of the evicted page.  This is called a shadow entry.
  *
  * On cache misses for which there are shadow entries, an eligible
@@ -162,7 +162,7 @@
 
 /*
  * Eviction timestamps need to be able to cover the full range of
- * actionable refaults. However, bits are tight in the radix tree
+ * actionable refaults. However, bits are tight in the xarray
  * entry, and after storing the identifier for the lruvec there might
  * not be enough left to represent every single actionable refault. In
  * that case, we have to sacrifice granularity for distance, and group
@@ -339,7 +339,7 @@ out:
 
 static struct list_lru shadow_nodes;
 
-void workingset_update_node(struct radix_tree_node *node)
+void workingset_update_node(struct xa_node *node)
 {
 	/*
 	 * Track non-empty nodes that contain only shadow entries;
@@ -368,7 +368,7 @@ static unsigned long count_shadow_nodes(struct shrinker *shrinker,
 	nodes = list_lru_shrink_count(&shadow_nodes, sc);
 
 	/*
-	 * Approximate a reasonable limit for the radix tree nodes
+	 * Approximate a reasonable limit for the nodes
 	 * containing shadow entries. We don't need to keep more
 	 * shadow entries than possible pages on the active list,
 	 * since refault distances bigger than that are dismissed.
@@ -383,11 +383,11 @@ static unsigned long count_shadow_nodes(struct shrinker *shrinker,
 	 * worst-case density of 1/8th. Below that, not all eligible
 	 * refaults can be detected anymore.
 	 *
-	 * On 64-bit with 7 radix_tree_nodes per page and 64 slots
+	 * On 64-bit with 7 xa_nodes per page and 64 slots
 	 * each, this will reclaim shadow entries when they consume
 	 * ~1.8% of available memory:
 	 *
-	 * PAGE_SIZE / radix_tree_nodes / node_entries * 8 / PAGE_SIZE
+	 * PAGE_SIZE / xa_nodes / node_entries * 8 / PAGE_SIZE
 	 */
 	if (sc->memcg) {
 		cache = mem_cgroup_node_nr_lru_pages(sc->memcg, sc->nid,
@@ -396,7 +396,7 @@ static unsigned long count_shadow_nodes(struct shrinker *shrinker,
 		cache = node_page_state(NODE_DATA(sc->nid), NR_ACTIVE_FILE) +
 			node_page_state(NODE_DATA(sc->nid), NR_INACTIVE_FILE);
 	}
-	max_nodes = cache >> (RADIX_TREE_MAP_SHIFT - 3);
+	max_nodes = cache >> (XA_CHUNK_SHIFT - 3);
 
 	if (!nodes)
 		return SHRINK_EMPTY;
@@ -409,11 +409,11 @@ static unsigned long count_shadow_nodes(struct shrinker *shrinker,
 static enum lru_status shadow_lru_isolate(struct list_head *item,
 					  struct list_lru_one *lru,
 					  spinlock_t *lru_lock,
-					  void *arg)
+					  void *arg) __must_hold(lru_lock)
 {
+	struct xa_node *node = container_of(item, struct xa_node, private_list);
+	XA_STATE(xas, node->array, 0);
 	struct address_space *mapping;
-	struct radix_tree_node *node;
-	unsigned int i;
 	int ret;
 
 	/*
@@ -421,14 +421,13 @@ static enum lru_status shadow_lru_isolate(struct list_head *item,
 	 * the shadow node LRU under the i_pages lock and the
 	 * lru_lock.  Because the page cache tree is emptied before
 	 * the inode can be destroyed, holding the lru_lock pins any
-	 * address_space that has radix tree nodes on the LRU.
+	 * address_space that has nodes on the LRU.
 	 *
 	 * We can then safely transition to the i_pages lock to
 	 * pin only the address_space of the particular node we want
 	 * to reclaim, take the node off-LRU, and drop the lru_lock.
 	 */
 
-	node = container_of(item, struct xa_node, private_list);
 	mapping = container_of(node->array, struct address_space, i_pages);
 
 	/* Coming from the list, invert the lock order */
@@ -450,25 +449,17 @@ static enum lru_status shadow_lru_isolate(struct list_head *item,
 		goto out_invalid;
 	if (WARN_ON_ONCE(node->count != node->nr_values))
 		goto out_invalid;
-	for (i = 0; i < RADIX_TREE_MAP_SIZE; i++) {
+	mapping->nrexceptional -= node->nr_values;
-		if (node->slots[i]) {
+	xas.xa_node = xa_parent_locked(&mapping->i_pages, node);
-			if (WARN_ON_ONCE(!xa_is_value(node->slots[i])))
+	xas.xa_offset = node->offset;
-				goto out_invalid;
+	xas.xa_shift = node->shift + XA_CHUNK_SHIFT;
-			if (WARN_ON_ONCE(!node->nr_values))
+	xas_set_update(&xas, workingset_update_node);
-				goto out_invalid;
+	/*
-			if (WARN_ON_ONCE(!mapping->nrexceptional))
+	 * We could store a shadow entry here which was the minimum of the
-				goto out_invalid;
+	 * shadow entries we were tracking ...
-			node->slots[i] = NULL;
+	 */
-			node->nr_values--;
+	xas_store(&xas, NULL);
-			node->count--;
-			mapping->nrexceptional--;
-		}
-	}
-	if (WARN_ON_ONCE(node->nr_values))
-		goto out_invalid;
 	inc_lruvec_page_state(virt_to_page(node), WORKINGSET_NODERECLAIM);
-	__radix_tree_delete_node(&mapping->i_pages, node,
-				 workingset_lookup_update(mapping));
 
 out_invalid:
 	xa_unlock_irq(&mapping->i_pages);