XArray: Add support for 1s-based allocation

A lot of places want to allocate IDs starting at 1 instead of 0.
While the xa_alloc() API supports this, it's not very efficient if lots
of IDs are allocated, due to having to walk down to the bottom of the
tree to see if ID 1 is available, then all the way over to the next
non-allocated ID.  This method marks ID 0 as being occupied which wastes
one slot in the XArray, but preserves xa_empty() as working.

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

2 files changed, 66 insertions, 33 deletions

diff --git a/lib/test_xarray.c b/lib/test_xarray.c
index 9d894e93456c..cd74f8f32abe 100644
--- a/lib/test_xarray.c
+++ b/lib/test_xarray.c
@@ -589,64 +589,86 @@ static noinline void check_multi_store(struct xarray *xa)
 #endif
 }
 
-static DEFINE_XARRAY_ALLOC(xa0);
-
-static noinline void check_xa_alloc(void)
+static noinline void check_xa_alloc_1(struct xarray *xa, unsigned int base)
 {
 	int i;
 	u32 id;
 
-	/* An empty array should assign 0 to the first alloc */
-	xa_alloc_index(&xa0, 0, GFP_KERNEL);
+	XA_BUG_ON(xa, !xa_empty(xa));
+	/* An empty array should assign %base to the first alloc */
+	xa_alloc_index(xa, base, GFP_KERNEL);
 
 	/* Erasing it should make the array empty again */
-	xa_erase_index(&xa0, 0);
-	XA_BUG_ON(&xa0, !xa_empty(&xa0));
+	xa_erase_index(xa, base);
+	XA_BUG_ON(xa, !xa_empty(xa));
+
+	/* And it should assign %base again */
+	xa_alloc_index(xa, base, GFP_KERNEL);
+
+	/* Allocating and then erasing a lot should not lose base */
+	for (i = base + 1; i < 2 * XA_CHUNK_SIZE; i++)
+		xa_alloc_index(xa, i, GFP_KERNEL);
+	for (i = base; i < 2 * XA_CHUNK_SIZE; i++)
+		xa_erase_index(xa, i);
+	xa_alloc_index(xa, base, GFP_KERNEL);
+
+	/* Destroying the array should do the same as erasing */
+	xa_destroy(xa);
 
-	/* And it should assign 0 again */
-	xa_alloc_index(&xa0, 0, GFP_KERNEL);
+	/* And it should assign %base again */
+	xa_alloc_index(xa, base, GFP_KERNEL);
 
-	/* The next assigned ID should be 1 */
-	xa_alloc_index(&xa0, 1, GFP_KERNEL);
-	xa_erase_index(&xa0, 1);
+	/* The next assigned ID should be base+1 */
+	xa_alloc_index(xa, base + 1, GFP_KERNEL);
+	xa_erase_index(xa, base + 1);
 
 	/* Storing a value should mark it used */
-	xa_store_index(&xa0, 1, GFP_KERNEL);
-	xa_alloc_index(&xa0, 2, GFP_KERNEL);
+	xa_store_index(xa, base + 1, GFP_KERNEL);
+	xa_alloc_index(xa, base + 2, GFP_KERNEL);
 
-	/* If we then erase 0, it should be free */
-	xa_erase_index(&xa0, 0);
-	xa_alloc_index(&xa0, 0, GFP_KERNEL);
+	/* If we then erase base, it should be free */
+	xa_erase_index(xa, base);
+	xa_alloc_index(xa, base, GFP_KERNEL);
 
-	xa_erase_index(&xa0, 1);
-	xa_erase_index(&xa0, 2);
+	xa_erase_index(xa, base + 1);
+	xa_erase_index(xa, base + 2);
 
 	for (i = 1; i < 5000; i++) {
-		xa_alloc_index(&xa0, i, GFP_KERNEL);
+		xa_alloc_index(xa, base + i, GFP_KERNEL);
 	}
 
-	xa_destroy(&xa0);
+	xa_destroy(xa);
 
+	/* Check that we fail properly at the limit of allocation */
 	id = 0xfffffffeU;
-	XA_BUG_ON(&xa0, xa_alloc(&xa0, &id, UINT_MAX, xa_mk_index(id),
+	XA_BUG_ON(xa, xa_alloc(xa, &id, UINT_MAX, xa_mk_index(id),
 				GFP_KERNEL) != 0);
-	XA_BUG_ON(&xa0, id != 0xfffffffeU);
-	XA_BUG_ON(&xa0, xa_alloc(&xa0, &id, UINT_MAX, xa_mk_index(id),
+	XA_BUG_ON(xa, id != 0xfffffffeU);
+	XA_BUG_ON(xa, xa_alloc(xa, &id, UINT_MAX, xa_mk_index(id),
 				GFP_KERNEL) != 0);
-	XA_BUG_ON(&xa0, id != 0xffffffffU);
-	XA_BUG_ON(&xa0, xa_alloc(&xa0, &id, UINT_MAX, xa_mk_index(id),
+	XA_BUG_ON(xa, id != 0xffffffffU);
+	XA_BUG_ON(xa, xa_alloc(xa, &id, UINT_MAX, xa_mk_index(id),
 				GFP_KERNEL) != -ENOSPC);
-	XA_BUG_ON(&xa0, id != 0xffffffffU);
-	xa_destroy(&xa0);
+	XA_BUG_ON(xa, id != 0xffffffffU);
+	xa_destroy(xa);
 
 	id = 10;
-	XA_BUG_ON(&xa0, xa_alloc(&xa0, &id, 5, xa_mk_index(id),
+	XA_BUG_ON(xa, xa_alloc(xa, &id, 5, xa_mk_index(id),
 				GFP_KERNEL) != -ENOSPC);
-	XA_BUG_ON(&xa0, xa_store_index(&xa0, 3, GFP_KERNEL) != 0);
-	XA_BUG_ON(&xa0, xa_alloc(&xa0, &id, 5, xa_mk_index(id),
+	XA_BUG_ON(xa, xa_store_index(xa, 3, GFP_KERNEL) != 0);
+	XA_BUG_ON(xa, xa_alloc(xa, &id, 5, xa_mk_index(id),
 				GFP_KERNEL) != -ENOSPC);
-	xa_erase_index(&xa0, 3);
-	XA_BUG_ON(&xa0, !xa_empty(&xa0));
+	xa_erase_index(xa, 3);
+	XA_BUG_ON(xa, !xa_empty(xa));
+}
+
+static DEFINE_XARRAY_ALLOC(xa0);
+static DEFINE_XARRAY_ALLOC1(xa1);
+
+static noinline void check_xa_alloc(void)
+{
+	check_xa_alloc_1(&xa0, 0);
+	check_xa_alloc_1(&xa1, 1);
 }
 
 static noinline void __check_store_iter(struct xarray *xa, unsigned long start,
diff --git a/lib/xarray.c b/lib/xarray.c
index 1b97ca58bd15..468fb7b7963f 100644
--- a/lib/xarray.c
+++ b/lib/xarray.c
@@ -57,6 +57,11 @@ static inline bool xa_track_free(const struct xarray *xa)
 	return xa->xa_flags & XA_FLAGS_TRACK_FREE;
 }
 
+static inline bool xa_zero_busy(const struct xarray *xa)
+{
+	return xa->xa_flags & XA_FLAGS_ZERO_BUSY;
+}
+
 static inline void xa_mark_set(struct xarray *xa, xa_mark_t mark)
 {
 	if (!(xa->xa_flags & XA_FLAGS_MARK(mark)))
@@ -432,6 +437,8 @@ static void xas_shrink(struct xa_state *xas)
 			break;
 		if (!xa_is_node(entry) && node->shift)
 			break;
+		if (xa_is_zero(entry) && xa_zero_busy(xa))
+			entry = NULL;
 		xas->xa_node = XAS_BOUNDS;
 
 		RCU_INIT_POINTER(xa->xa_head, entry);
@@ -628,6 +635,8 @@ static void *xas_create(struct xa_state *xas, bool allow_root)
 	if (xas_top(node)) {
 		entry = xa_head_locked(xa);
 		xas->xa_node = NULL;
+		if (!entry && xa_zero_busy(xa))
+			entry = XA_ZERO_ENTRY;
 		shift = xas_expand(xas, entry);
 		if (shift < 0)
 			return NULL;
@@ -1942,6 +1951,8 @@ void xa_destroy(struct xarray *xa)
 	entry = xa_head_locked(xa);
 	RCU_INIT_POINTER(xa->xa_head, NULL);
 	xas_init_marks(&xas);
+	if (xa_zero_busy(xa))
+		xa_mark_clear(xa, XA_FREE_MARK);
 	/* lockdep checks we're still holding the lock in xas_free_nodes() */
 	if (xa_is_node(entry))
 		xas_free_nodes(&xas, xa_to_node(entry));