blk-mq: abstract tag allocation out into sbitmap library

This is a generally useful data structure, so make it available to
anyone else who might want to use it. It's also a nice cleanup
separating the allocation logic from the rest of the tag handling logic.

The code is behind a new Kconfig option, CONFIG_SBITMAP, which is only
selected by CONFIG_BLOCK for now.

This should be a complete noop functionality-wise.
Signed-off-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@fb.com>

MAINTAINERS

@@ -2451,6 +2451,7 @@ T:	git git://git.kernel.org/pub/scm/linux/kernel/git/axboe/linux-block.git
 S:	Maintained
 F:	block/
 F:	kernel/trace/blktrace.c
+F:	lib/sbitmap.c
 
 BLOCK2MTD DRIVER
 M:	Joern Engel <joern@lazybastard.org>

block/Kconfig

@@ -4,6 +4,7 @@
 menuconfig BLOCK
        bool "Enable the block layer" if EXPERT
        default y
+       select SBITMAP
        help
 	 Provide block layer support for the kernel.
 

block/blk-mq-tag.h

@@ -3,31 +3,6 @@
 
 #include "blk-mq.h"
 
-enum {
-	BT_WAIT_QUEUES	= 8,
-	BT_WAIT_BATCH	= 8,
-};
-
-struct bt_wait_state {
-	atomic_t wait_cnt;
-	wait_queue_head_t wait;
-} ____cacheline_aligned_in_smp;
-
-#define TAG_TO_INDEX(bt, tag)	((tag) >> (bt)->bits_per_word)
-#define TAG_TO_BIT(bt, tag)	((tag) & ((1 << (bt)->bits_per_word) - 1))
-
-struct blk_mq_bitmap_tags {
-	unsigned int depth;
-	unsigned int wake_cnt;
-	unsigned int bits_per_word;
-
-	unsigned int map_nr;
-	struct blk_align_bitmap *map;
-
-	atomic_t wake_index;
-	struct bt_wait_state *bs;
-};
-
 /*
  * Tag address space map.
  */
@@ -37,8 +12,8 @@ struct blk_mq_tags {
 
 	atomic_t active_queues;
 
-	struct blk_mq_bitmap_tags bitmap_tags;
-	struct blk_mq_bitmap_tags breserved_tags;
+	struct sbitmap_queue bitmap_tags;
+	struct sbitmap_queue breserved_tags;
 
 	struct request **rqs;
 	struct list_head page_list;
@@ -61,6 +36,14 @@ extern void blk_mq_tag_wakeup_all(struct blk_mq_tags *tags, bool);
 void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_iter_fn *fn,
 		void *priv);
 
+static inline struct sbq_wait_state *bt_wait_ptr(struct sbitmap_queue *bt,
+						 struct blk_mq_hw_ctx *hctx)
+{
+	if (!hctx)
+		return &bt->ws[0];
+	return sbq_wait_ptr(bt, &hctx->wait_index);
+}
+
 enum {
 	BLK_MQ_TAG_CACHE_MIN	= 1,
 	BLK_MQ_TAG_CACHE_MAX	= 64,

block/blk-mq.c

@@ -41,42 +41,23 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx);
  */
 static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx)
 {
-	unsigned int i;
-
-	for (i = 0; i < hctx->ctx_map.size; i++)
-		if (hctx->ctx_map.map[i].word)
-			return true;
-
-	return false;
-}
-
-static inline struct blk_align_bitmap *get_bm(struct blk_mq_hw_ctx *hctx,
-					      struct blk_mq_ctx *ctx)
-{
-	return &hctx->ctx_map.map[ctx->index_hw / hctx->ctx_map.bits_per_word];
+	return sbitmap_any_bit_set(&hctx->ctx_map);
 }
 
-#define CTX_TO_BIT(hctx, ctx)	\
-	((ctx)->index_hw & ((hctx)->ctx_map.bits_per_word - 1))
-
 /*
  * Mark this ctx as having pending work in this hardware queue
  */
 static void blk_mq_hctx_mark_pending(struct blk_mq_hw_ctx *hctx,
 				     struct blk_mq_ctx *ctx)
 {
-	struct blk_align_bitmap *bm = get_bm(hctx, ctx);
-
-	if (!test_bit(CTX_TO_BIT(hctx, ctx), &bm->word))
-		set_bit(CTX_TO_BIT(hctx, ctx), &bm->word);
+	if (!sbitmap_test_bit(&hctx->ctx_map, ctx->index_hw))
+		sbitmap_set_bit(&hctx->ctx_map, ctx->index_hw);
 }
 
 static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx,
 				      struct blk_mq_ctx *ctx)
 {
-	struct blk_align_bitmap *bm = get_bm(hctx, ctx);
-
-	clear_bit(CTX_TO_BIT(hctx, ctx), &bm->word);
+	sbitmap_clear_bit(&hctx->ctx_map, ctx->index_hw);
 }
 
 void blk_mq_freeze_queue_start(struct request_queue *q)
@@ -755,38 +736,36 @@ static bool blk_mq_attempt_merge(struct request_queue *q,
 	return false;
 }
 
+struct flush_busy_ctx_data {
+	struct blk_mq_hw_ctx *hctx;
+	struct list_head *list;
+};
+
+static bool flush_busy_ctx(struct sbitmap *sb, unsigned int bitnr, void *data)
+{
+	struct flush_busy_ctx_data *flush_data = data;
+	struct blk_mq_hw_ctx *hctx = flush_data->hctx;
+	struct blk_mq_ctx *ctx = hctx->ctxs[bitnr];
+
+	sbitmap_clear_bit(sb, bitnr);
+	spin_lock(&ctx->lock);
+	list_splice_tail_init(&ctx->rq_list, flush_data->list);
+	spin_unlock(&ctx->lock);
+	return true;
+}
+
 /*
  * Process software queues that have been marked busy, splicing them
  * to the for-dispatch
  */
 static void flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list)
 {
-	struct blk_mq_ctx *ctx;
-	int i;
-
-	for (i = 0; i < hctx->ctx_map.size; i++) {
-		struct blk_align_bitmap *bm = &hctx->ctx_map.map[i];
-		unsigned int off, bit;
-
-		if (!bm->word)
-			continue;
-
-		bit = 0;
-		off = i * hctx->ctx_map.bits_per_word;
-		do {
-			bit = find_next_bit(&bm->word, bm->depth, bit);
-			if (bit >= bm->depth)
-				break;
-
-			ctx = hctx->ctxs[bit + off];
-			clear_bit(bit, &bm->word);
-			spin_lock(&ctx->lock);
-			list_splice_tail_init(&ctx->rq_list, list);
-			spin_unlock(&ctx->lock);
+	struct flush_busy_ctx_data data = {
+		.hctx = hctx,
+		.list = list,
+	};
 
-			bit++;
-		} while (1);
-	}
+	sbitmap_for_each_set(&hctx->ctx_map, flush_busy_ctx, &data);
 }
 
 static inline unsigned int queued_to_index(unsigned int queued)
@@ -1609,32 +1588,6 @@ static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
 	return NULL;
 }
 
-static void blk_mq_free_bitmap(struct blk_mq_ctxmap *bitmap)
-{
-	kfree(bitmap->map);
-}
-
-static int blk_mq_alloc_bitmap(struct blk_mq_ctxmap *bitmap, int node)
-{
-	unsigned int bpw = 8, total, num_maps, i;
-
-	bitmap->bits_per_word = bpw;
-
-	num_maps = ALIGN(nr_cpu_ids, bpw) / bpw;
-	bitmap->map = kzalloc_node(num_maps * sizeof(struct blk_align_bitmap),
-					GFP_KERNEL, node);
-	if (!bitmap->map)
-		return -ENOMEM;
-
-	total = nr_cpu_ids;
-	for (i = 0; i < num_maps; i++) {
-		bitmap->map[i].depth = min(total, bitmap->bits_per_word);
-		total -= bitmap->map[i].depth;
-	}
-
-	return 0;
-}
-
 /*
  * 'cpu' is going away. splice any existing rq_list entries from this
  * software queue to the hw queue dispatch list, and ensure that it
@@ -1700,7 +1653,7 @@ static void blk_mq_exit_hctx(struct request_queue *q,
 
 	blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
 	blk_free_flush_queue(hctx->fq);
-	blk_mq_free_bitmap(&hctx->ctx_map);
+	sbitmap_free(&hctx->ctx_map);
 }
 
 static void blk_mq_exit_hw_queues(struct request_queue *q,
@@ -1760,7 +1713,8 @@ static int blk_mq_init_hctx(struct request_queue *q,
 	if (!hctx->ctxs)
 		goto unregister_cpu_notifier;
 
-	if (blk_mq_alloc_bitmap(&hctx->ctx_map, node))
+	if (sbitmap_init_node(&hctx->ctx_map, nr_cpu_ids, ilog2(8), GFP_KERNEL,
+			      node))
 		goto free_ctxs;
 
 	hctx->nr_ctx = 0;
@@ -1787,7 +1741,7 @@ static int blk_mq_init_hctx(struct request_queue *q,
 	if (set->ops->exit_hctx)
 		set->ops->exit_hctx(hctx, hctx_idx);
  free_bitmap:
-	blk_mq_free_bitmap(&hctx->ctx_map);
+	sbitmap_free(&hctx->ctx_map);
  free_ctxs:
 	kfree(hctx->ctxs);
  unregister_cpu_notifier:
@@ -1863,8 +1817,6 @@ static void blk_mq_map_swqueue(struct request_queue *q,
 	mutex_unlock(&q->sysfs_lock);
 
 	queue_for_each_hw_ctx(q, hctx, i) {
-		struct blk_mq_ctxmap *map = &hctx->ctx_map;
-
 		/*
 		 * If no software queues are mapped to this hardware queue,
 		 * disable it and free the request entries.
@@ -1890,7 +1842,7 @@ static void blk_mq_map_swqueue(struct request_queue *q,
 		 * This is more accurate and more efficient than looping
 		 * over all possibly mapped software queues.
 		 */
-		map->size = DIV_ROUND_UP(hctx->nr_ctx, map->bits_per_word);
+		sbitmap_resize(&hctx->ctx_map, hctx->nr_ctx);
 
 		/*
 		 * Initialize batch roundrobin counts

block/blk-mq.h

@@ -63,15 +63,6 @@ extern void blk_mq_rq_timed_out(struct request *req, bool reserved);
 
 void blk_mq_release(struct request_queue *q);
 
-/*
- * Basic implementation of sparser bitmap, allowing the user to spread
- * the bits over more cachelines.
- */
-struct blk_align_bitmap {
-	unsigned long word;
-	unsigned long depth;
-} ____cacheline_aligned_in_smp;
-
 static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
 					   unsigned int cpu)
 {

include/linux/blk-mq.h

@@ -2,6 +2,7 @@
 #define BLK_MQ_H
 
 #include <linux/blkdev.h>
+#include <linux/sbitmap.h>
 
 struct blk_mq_tags;
 struct blk_flush_queue;
@@ -12,12 +13,6 @@ struct blk_mq_cpu_notifier {
 	int (*notify)(void *data, unsigned long action, unsigned int cpu);
 };
 
-struct blk_mq_ctxmap {
-	unsigned int size;
-	unsigned int bits_per_word;
-	struct blk_align_bitmap *map;
-};
-
 struct blk_mq_hw_ctx {
 	struct {
 		spinlock_t		lock;
@@ -37,7 +32,7 @@ struct blk_mq_hw_ctx {
 
 	void			*driver_data;
 
-	struct blk_mq_ctxmap	ctx_map;
+	struct sbitmap		ctx_map;
 
 	struct blk_mq_ctx	**ctxs;
 	unsigned int		nr_ctx;

include/linux/sbitmap.h

+/*
+ * Fast and scalable bitmaps.
+ *
+ * Copyright (C) 2016 Facebook
+ * Copyright (C) 2013-2014 Jens Axboe
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#ifndef __LINUX_SCALE_BITMAP_H
+#define __LINUX_SCALE_BITMAP_H
+
+#include <linux/kernel.h>
+#include <linux/slab.h>
+
+/**
+ * struct sbitmap_word - Word in a &struct sbitmap.
+ */
+struct sbitmap_word {
+	/**
+	 * @word: The bitmap word itself.
+	 */
+	unsigned long word;
+
+	/**
+	 * @depth: Number of bits being used in @word.
+	 */
+	unsigned long depth;
+} ____cacheline_aligned_in_smp;
+
+/**
+ * struct sbitmap - Scalable bitmap.
+ *
+ * A &struct sbitmap is spread over multiple cachelines to avoid ping-pong. This
+ * trades off higher memory usage for better scalability.
+ */
+struct sbitmap {
+	/**
+	 * @depth: Number of bits used in the whole bitmap.
+	 */
+	unsigned int depth;
+
+	/**
+	 * @shift: log2(number of bits used per word)
+	 */
+	unsigned int shift;
+
+	/**
+	 * @map_nr: Number of words (cachelines) being used for the bitmap.
+	 */
+	unsigned int map_nr;
+
+	/**
+	 * @map: Allocated bitmap.
+	 */
+	struct sbitmap_word *map;
+};
+
+#define SBQ_WAIT_QUEUES 8
+#define SBQ_WAKE_BATCH 8
+
+/**
+ * struct sbq_wait_state - Wait queue in a &struct sbitmap_queue.
+ */
+struct sbq_wait_state {
+	/**
+	 * @wait_cnt: Number of frees remaining before we wake up.
+	 */
+	atomic_t wait_cnt;
+
+	/**
+	 * @wait: Wait queue.
+	 */
+	wait_queue_head_t wait;
+} ____cacheline_aligned_in_smp;
+
+/**
+ * struct sbitmap_queue - Scalable bitmap with the added ability to wait on free
+ * bits.
+ *
+ * A &struct sbitmap_queue uses multiple wait queues and rolling wakeups to
+ * avoid contention on the wait queue spinlock. This ensures that we don't hit a
+ * scalability wall when we run out of free bits and have to start putting tasks
+ * to sleep.
+ */
+struct sbitmap_queue {
+	/**
+	 * @sb: Scalable bitmap.
+	 */
+	struct sbitmap sb;
+
+	/**
+	 * @wake_batch: Number of bits which must be freed before we wake up any
+	 * waiters.
+	 */
+	unsigned int wake_batch;
+
+	/**
+	 * @wake_index: Next wait queue in @ws to wake up.
+	 */
+	atomic_t wake_index;
+
+	/**
+	 * @ws: Wait queues.
+	 */
+	struct sbq_wait_state *ws;
+};
+
+/**
+ * sbitmap_init_node() - Initialize a &struct sbitmap on a specific memory node.
+ * @sb: Bitmap to initialize.
+ * @depth: Number of bits to allocate.
+ * @shift: Use 2^@shift bits per word in the bitmap; if a negative number if
+ *         given, a good default is chosen.
+ * @flags: Allocation flags.
+ * @node: Memory node to allocate on.
+ *
+ * Return: Zero on success or negative errno on failure.
+ */
+int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
+		      gfp_t flags, int node);
+
+/**
+ * sbitmap_free() - Free memory used by a &struct sbitmap.
+ * @sb: Bitmap to free.
+ */
+static inline void sbitmap_free(struct sbitmap *sb)
+{
+	kfree(sb->map);
+	sb->map = NULL;
+}
+
+/**
+ * sbitmap_resize() - Resize a &struct sbitmap.
+ * @sb: Bitmap to resize.
+ * @depth: New number of bits to resize to.
+ *
+ * Doesn't reallocate anything. It's up to the caller to ensure that the new
+ * depth doesn't exceed the depth that the sb was initialized with.
+ */
+void sbitmap_resize(struct sbitmap *sb, unsigned int depth);
+
+/**
+ * sbitmap_get() - Try to allocate a free bit from a &struct sbitmap.
+ * @sb: Bitmap to allocate from.
+ * @alloc_hint: Hint for where to start searching for a free bit.
+ * @round_robin: If true, be stricter about allocation order; always allocate
+ *               starting from the last allocated bit. This is less efficient
+ *               than the default behavior (false).
+ *
+ * Return: Non-negative allocated bit number if successful, -1 otherwise.
+ */
+int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin);
+
+/**
+ * sbitmap_any_bit_set() - Check for a set bit in a &struct sbitmap.
+ * @sb: Bitmap to check.
+ *
+ * Return: true if any bit in the bitmap is set, false otherwise.
+ */
+bool sbitmap_any_bit_set(const struct sbitmap *sb);
+
+/**
+ * sbitmap_any_bit_clear() - Check for an unset bit in a &struct
+ * sbitmap.
+ * @sb: Bitmap to check.
+ *
+ * Return: true if any bit in the bitmap is clear, false otherwise.
+ */
+bool sbitmap_any_bit_clear(const struct sbitmap *sb);
+
+typedef bool (*sb_for_each_fn)(struct sbitmap *, unsigned int, void *);
+
+/**
+ * sbitmap_for_each_set() - Iterate over each set bit in a &struct sbitmap.
+ * @sb: Bitmap to iterate over.
+ * @fn: Callback. Should return true to continue or false to break early.
+ * @data: Pointer to pass to callback.
+ *
+ * This is inline even though it's non-trivial so that the function calls to the
+ * callback will hopefully get optimized away.
+ */
+static inline void sbitmap_for_each_set(struct sbitmap *sb, sb_for_each_fn fn,
+					void *data)
+{
+	unsigned int i;
+
+	for (i = 0; i < sb->map_nr; i++) {
+		struct sbitmap_word *word = &sb->map[i];
+		unsigned int off, nr;
+
+		if (!word->word)
+			continue;
+
+		nr = 0;
+		off = i << sb->shift;
+		while (1) {
+			nr = find_next_bit(&word->word, word->depth, nr);
+			if (nr >= word->depth)
+				break;
+
+			if (!fn(sb, off + nr, data))
+				return;
+
+			nr++;
+		}
+	}
+}
+
+#define SB_NR_TO_INDEX(sb, bitnr) ((bitnr) >> (sb)->shift)
+#define SB_NR_TO_BIT(sb, bitnr) ((bitnr) & ((1U << (sb)->shift) - 1U))
+
+static inline unsigned long *__sbitmap_word(struct sbitmap *sb,
+					    unsigned int bitnr)
+{
+	return &sb->map[SB_NR_TO_INDEX(sb, bitnr)].word;
+}
+
+/* Helpers equivalent to the operations in asm/bitops.h and linux/bitmap.h */
+
+static inline void sbitmap_set_bit(struct sbitmap *sb, unsigned int bitnr)
+{
+	set_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr));
+}
+
+static inline void sbitmap_clear_bit(struct sbitmap *sb, unsigned int bitnr)
+{
+	clear_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr));
+}
+
+static inline int sbitmap_test_bit(struct sbitmap *sb, unsigned int bitnr)
+{
+	return test_bit(SB_NR_TO_BIT(sb, bitnr), __sbitmap_word(sb, bitnr));
+}
+
+unsigned int sbitmap_weight(const struct sbitmap *sb);
+
+/**
+ * sbitmap_queue_init_node() - Initialize a &struct sbitmap_queue on a specific
+ * memory node.
+ * @sbq: Bitmap queue to initialize.
+ * @depth: See sbitmap_init_node().
+ * @shift: See sbitmap_init_node().
+ * @flags: Allocation flags.
+ * @node: Memory node to allocate on.
+ *
+ * Return: Zero on success or negative errno on failure.
+ */
+int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth,
+			    int shift, gfp_t flags, int node);
+
+/**
+ * sbitmap_queue_free() - Free memory used by a &struct sbitmap_queue.
+ *
+ * @sbq: Bitmap queue to free.
+ */
+static inline void sbitmap_queue_free(struct sbitmap_queue *sbq)
+{
+	kfree(sbq->ws);
+	sbitmap_free(&sbq->sb);
+}
+
+/**
+ * sbitmap_queue_resize() - Resize a &struct sbitmap_queue.
+ * @sbq: Bitmap queue to resize.
+ * @depth: New number of bits to resize to.
+ *
+ * Like sbitmap_resize(), this doesn't reallocate anything. It has to do
+ * some extra work on the &struct sbitmap_queue, so it's not safe to just
+ * resize the underlying &struct sbitmap.
+ */
+void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth);
+
+/**
+ * sbitmap_queue_clear() - Free an allocated bit and wake up waiters on a
+ * &struct sbitmap_queue.
+ * @sbq: Bitmap to free from.
+ * @nr: Bit number to free.
+ */
+void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr);
+
+static inline int sbq_index_inc(int index)
+{
+	return (index + 1) & (SBQ_WAIT_QUEUES - 1);
+}
+
+static inline void sbq_index_atomic_inc(atomic_t *index)
+{
+	int old = atomic_read(index);
+	int new = sbq_index_inc(old);
+	atomic_cmpxchg(index, old, new);
+}
+
+/**
+ * sbq_wait_ptr() - Get the next wait queue to use for a &struct
+ * sbitmap_queue.
+ * @sbq: Bitmap queue to wait on.
+ * @wait_index: A counter per "user" of @sbq.
+ */
+static inline struct sbq_wait_state *sbq_wait_ptr(struct sbitmap_queue *sbq,
+						  atomic_t *wait_index)
+{
+	struct sbq_wait_state *ws;
+
+	ws = &sbq->ws[atomic_read(wait_index)];
+	sbq_index_atomic_inc(wait_index);
+	return ws;
+}
+
+/**
+ * sbitmap_queue_wake_all() - Wake up everything waiting on a &struct
+ * sbitmap_queue.
+ * @sbq: Bitmap queue to wake up.
+ */
+void sbitmap_queue_wake_all(struct sbitmap_queue *sbq);
+
+#endif /* __LINUX_SCALE_BITMAP_H */

lib/Kconfig

@@ -550,4 +550,7 @@ config STACKDEPOT
 	bool
 	select STACKTRACE
 
+config SBITMAP
+	bool
+
 endmenu

lib/Makefile

@@ -228,3 +228,5 @@ obj-$(CONFIG_UCS2_STRING) += ucs2_string.o
 obj-$(CONFIG_UBSAN) += ubsan.o
 
 UBSAN_SANITIZE_ubsan.o := n
+
+obj-$(CONFIG_SBITMAP) += sbitmap.o

lib/sbitmap.c

+/*
+ * Copyright (C) 2016 Facebook
+ * Copyright (C) 2013-2014 Jens Axboe
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <linux/sbitmap.h>
+
+int sbitmap_init_node(struct sbitmap *sb, unsigned int depth, int shift,
+		      gfp_t flags, int node)
+{
+	unsigned int bits_per_word;
+	unsigned int i;
+
+	if (shift < 0) {
+		shift = ilog2(BITS_PER_LONG);
+		/*
+		 * If the bitmap is small, shrink the number of bits per word so
+		 * we spread over a few cachelines, at least. If less than 4
+		 * bits, just forget about it, it's not going to work optimally
+		 * anyway.
+		 */
+		if (depth >= 4) {
+			while ((4U << shift) > depth)
+				shift--;
+		}
+	}
+	bits_per_word = 1U << shift;
+	if (bits_per_word > BITS_PER_LONG)
+		return -EINVAL;
+
+	sb->shift = shift;
+	sb->depth = depth;
+	sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
+
+	if (depth == 0) {
+		sb->map = NULL;
+		return 0;
+	}
+
+	sb->map = kzalloc_node(sb->map_nr * sizeof(*sb->map), flags, node);
+	if (!sb->map)
+		return -ENOMEM;
+
+	for (i = 0; i < sb->map_nr; i++) {
+		sb->map[i].depth = min(depth, bits_per_word);
+		depth -= sb->map[i].depth;
+	}
+	return 0;
+}
+EXPORT_SYMBOL_GPL(sbitmap_init_node);
+
+void sbitmap_resize(struct sbitmap *sb, unsigned int depth)
+{
+	unsigned int bits_per_word = 1U << sb->shift;
+	unsigned int i;
+
+	sb->depth = depth;
+	sb->map_nr = DIV_ROUND_UP(sb->depth, bits_per_word);
+
+	for (i = 0; i < sb->map_nr; i++) {
+		sb->map[i].depth = min(depth, bits_per_word);
+		depth -= sb->map[i].depth;
+	}
+}
+EXPORT_SYMBOL_GPL(sbitmap_resize);
+
+static int __sbitmap_get_word(struct sbitmap_word *word, unsigned int hint,
+			      bool wrap)
+{
+	unsigned int orig_hint = hint;
+	int nr;
+
+	while (1) {
+		nr = find_next_zero_bit(&word->word, word->depth, hint);
+		if (unlikely(nr >= word->depth)) {
+			/*
+			 * We started with an offset, and we didn't reset the
+			 * offset to 0 in a failure case, so start from 0 to
+			 * exhaust the map.
+			 */
+			if (orig_hint && hint && wrap) {
+				hint = orig_hint = 0;
+				continue;
+			}
+			return -1;
+		}
+
+		if (!test_and_set_bit(nr, &word->word))
+			break;
+
+		hint = nr + 1;
+		if (hint >= word->depth - 1)
+			hint = 0;
+	}
+
+	return nr;
+}
+
+int sbitmap_get(struct sbitmap *sb, unsigned int alloc_hint, bool round_robin)
+{
+	unsigned int i, index;
+	int nr = -1;
+
+	index = SB_NR_TO_INDEX(sb, alloc_hint);
+
+	for (i = 0; i < sb->map_nr; i++) {
+		nr = __sbitmap_get_word(&sb->map[index],
+					SB_NR_TO_BIT(sb, alloc_hint),
+					!round_robin);
+		if (nr != -1) {
+			nr += index << sb->shift;
+			break;
+		}
+
+		/* Jump to next index. */
+		index++;
+		alloc_hint = index << sb->shift;
+
+		if (index >= sb->map_nr) {
+			index = 0;
+			alloc_hint = 0;
+		}
+	}
+
+	return nr;
+}
+EXPORT_SYMBOL_GPL(sbitmap_get);
+
+bool sbitmap_any_bit_set(const struct sbitmap *sb)
+{
+	unsigned int i;
+
+	for (i = 0; i < sb->map_nr; i++) {
+		if (sb->map[i].word)
+			return true;
+	}
+	return false;
+}
+EXPORT_SYMBOL_GPL(sbitmap_any_bit_set);
+
+bool sbitmap_any_bit_clear(const struct sbitmap *sb)
+{
+	unsigned int i;
+
+	for (i = 0; i < sb->map_nr; i++) {
+		const struct sbitmap_word *word = &sb->map[i];
+		unsigned long ret;
+
+		ret = find_first_zero_bit(&word->word, word->depth);
+		if (ret < word->depth)
+			return true;
+	}
+	return false;
+}
+EXPORT_SYMBOL_GPL(sbitmap_any_bit_clear);
+
+unsigned int sbitmap_weight(const struct sbitmap *sb)
+{
+	unsigned int i, weight;
+
+	for (i = 0; i < sb->map_nr; i++) {
+		const struct sbitmap_word *word = &sb->map[i];
+
+		weight += bitmap_weight(&word->word, word->depth);
+	}
+	return weight;
+}
+EXPORT_SYMBOL_GPL(sbitmap_weight);
+
+static unsigned int sbq_calc_wake_batch(unsigned int depth)
+{
+	unsigned int wake_batch;
+
+	/*
+	 * For each batch, we wake up one queue. We need to make sure that our
+	 * batch size is small enough that the full depth of the bitmap is
+	 * enough to wake up all of the queues.
+	 */
+	wake_batch = SBQ_WAKE_BATCH;
+	if (wake_batch > depth / SBQ_WAIT_QUEUES)
+		wake_batch = max(1U, depth / SBQ_WAIT_QUEUES);
+
+	return wake_batch;
+}
+
+int sbitmap_queue_init_node(struct sbitmap_queue *sbq, unsigned int depth,
+			    int shift, gfp_t flags, int node)
+{
+	int ret;
+	int i;
+
+	ret = sbitmap_init_node(&sbq->sb, depth, shift, flags, node);
+	if (ret)
+		return ret;
+
+	sbq->wake_batch = sbq_calc_wake_batch(depth);
+	atomic_set(&sbq->wake_index, 0);
+
+	sbq->ws = kzalloc(SBQ_WAIT_QUEUES * sizeof(*sbq->ws), flags);
+	if (!sbq->ws) {
+		sbitmap_free(&sbq->sb);
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
+		init_waitqueue_head(&sbq->ws[i].wait);
+		atomic_set(&sbq->ws[i].wait_cnt, sbq->wake_batch);
+	}
+	return 0;
+}
+EXPORT_SYMBOL_GPL(sbitmap_queue_init_node);
+
+void sbitmap_queue_resize(struct sbitmap_queue *sbq, unsigned int depth)
+{
+	sbq->wake_batch = sbq_calc_wake_batch(depth);
+	sbitmap_resize(&sbq->sb, depth);
+}
+EXPORT_SYMBOL_GPL(sbitmap_queue_resize);
+
+static struct sbq_wait_state *sbq_wake_ptr(struct sbitmap_queue *sbq)
+{
+	int i, wake_index;
+
+	wake_index = atomic_read(&sbq->wake_index);
+	for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
+		struct sbq_wait_state *ws = &sbq->ws[wake_index];
+
+		if (waitqueue_active(&ws->wait)) {
+			int o = atomic_read(&sbq->wake_index);
+
+			if (wake_index != o)
+				atomic_cmpxchg(&sbq->wake_index, o, wake_index);
+			return ws;
+		}
+
+		wake_index = sbq_index_inc(wake_index);
+	}
+
+	return NULL;
+}
+
+static void sbq_wake_up(struct sbitmap_queue *sbq)
+{
+	struct sbq_wait_state *ws;
+	int wait_cnt;
+
+	/* Ensure that the wait list checks occur after clear_bit(). */
+	smp_mb();
+
+	ws = sbq_wake_ptr(sbq);
+	if (!ws)
+		return;
+
+	wait_cnt = atomic_dec_return(&ws->wait_cnt);
+	if (unlikely(wait_cnt < 0))
+		wait_cnt = atomic_inc_return(&ws->wait_cnt);
+	if (wait_cnt == 0) {
+		atomic_add(sbq->wake_batch, &ws->wait_cnt);
+		sbq_index_atomic_inc(&sbq->wake_index);
+		wake_up(&ws->wait);
+	}
+}
+
+void sbitmap_queue_clear(struct sbitmap_queue *sbq, unsigned int nr)
+{
+	sbitmap_clear_bit(&sbq->sb, nr);
+	sbq_wake_up(sbq);
+}
+EXPORT_SYMBOL_GPL(sbitmap_queue_clear);
+
+void sbitmap_queue_wake_all(struct sbitmap_queue *sbq)
+{
+	int i, wake_index;
+
+	/*
+	 * Make sure all changes prior to this are visible from other CPUs.
+	 */
+	smp_mb();
+	wake_index = atomic_read(&sbq->wake_index);
+	for (i = 0; i < SBQ_WAIT_QUEUES; i++) {
+		struct sbq_wait_state *ws = &sbq->ws[wake_index];
+
+		if (waitqueue_active(&ws->wait))
+			wake_up(&ws->wait);
+
+		wake_index = sbq_index_inc(wake_index);
+	}
+}
+EXPORT_SYMBOL_GPL(sbitmap_queue_wake_all);