Merge branch 'for-3.16/core' into for-3.16/drivers

mtip32xx uses blk_mq_alloc_reserved_request(), so pull in the
core changes so we have a properly merged end result.

Signed-off-by: Jens Axboe <axboe@fb.com>

Documentation/DocBook/filesystems.tmpl

@@ -62,7 +62,7 @@
 !Efs/mpage.c
 !Efs/namei.c
 !Efs/buffer.c
-!Efs/bio.c
+!Eblock/bio.c
 !Efs/seq_file.c
 !Efs/filesystems.c
 !Efs/fs-writeback.c

block/Makefile

@@ -2,13 +2,15 @@
 # Makefile for the kernel block layer
 #
 
-obj-$(CONFIG_BLOCK) := elevator.o blk-core.o blk-tag.o blk-sysfs.o \
+obj-$(CONFIG_BLOCK) := bio.o elevator.o blk-core.o blk-tag.o blk-sysfs.o \
 			blk-flush.o blk-settings.o blk-ioc.o blk-map.o \
 			blk-exec.o blk-merge.o blk-softirq.o blk-timeout.o \
 			blk-iopoll.o blk-lib.o blk-mq.o blk-mq-tag.o \
 			blk-mq-sysfs.o blk-mq-cpu.o blk-mq-cpumap.o ioctl.o \
-			genhd.o scsi_ioctl.o partition-generic.o partitions/
+			genhd.o scsi_ioctl.o partition-generic.o ioprio.o \
+			partitions/
 
+obj-$(CONFIG_BOUNCE)	+= bounce.o
 obj-$(CONFIG_BLK_DEV_BSG)	+= bsg.o
 obj-$(CONFIG_BLK_DEV_BSGLIB)	+= bsg-lib.o
 obj-$(CONFIG_BLK_CGROUP)	+= blk-cgroup.o
@@ -20,3 +22,4 @@ obj-$(CONFIG_IOSCHED_CFQ)	+= cfq-iosched.o
 obj-$(CONFIG_BLOCK_COMPAT)	+= compat_ioctl.o
 obj-$(CONFIG_BLK_DEV_INTEGRITY)	+= blk-integrity.o
 obj-$(CONFIG_BLK_CMDLINE_PARSER)	+= cmdline-parser.o
+obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o

fs/bio-integrity.c → block/bio-integrity.c

@@ -617,7 +617,7 @@ int bioset_integrity_create(struct bio_set *bs, int pool_size)
 	if (!bs->bio_integrity_pool)
 		return -1;
 
-	bs->bvec_integrity_pool = biovec_create_pool(bs, pool_size);
+	bs->bvec_integrity_pool = biovec_create_pool(pool_size);
 	if (!bs->bvec_integrity_pool) {
 		mempool_destroy(bs->bio_integrity_pool);
 		return -1;

fs/bio.c → block/bio.c

@@ -305,6 +305,8 @@ static void bio_chain_endio(struct bio *bio, int error)
 
 /**
  * bio_chain - chain bio completions
+ * @bio: the target bio
+ * @parent: the @bio's parent bio
  *
  * The caller won't have a bi_end_io called when @bio completes - instead,
  * @parent's bi_end_io won't be called until both @parent and @bio have
@@ -1011,8 +1013,7 @@ static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
 	bio->bi_private = bmd;
 }
 
-static struct bio_map_data *bio_alloc_map_data(int nr_segs,
-					       unsigned int iov_count,
+static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
 					       gfp_t gfp_mask)
 {
 	if (iov_count > UIO_MAXIOV)
@@ -1154,7 +1155,7 @@ struct bio *bio_copy_user_iov(struct request_queue *q,
 	if (offset)
 		nr_pages++;
 
-	bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
+	bmd = bio_alloc_map_data(iov_count, gfp_mask);
 	if (!bmd)
 		return ERR_PTR(-ENOMEM);
 
@@ -1859,7 +1860,7 @@ EXPORT_SYMBOL_GPL(bio_trim);
  * create memory pools for biovec's in a bio_set.
  * use the global biovec slabs created for general use.
  */
-mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries)
+mempool_t *biovec_create_pool(int pool_entries)
 {
 	struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
 
@@ -1922,7 +1923,7 @@ struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
 	if (!bs->bio_pool)
 		goto bad;
 
-	bs->bvec_pool = biovec_create_pool(bs, pool_size);
+	bs->bvec_pool = biovec_create_pool(pool_size);
 	if (!bs->bvec_pool)
 		goto bad;
 

block/blk-core.c

@@ -576,12 +576,9 @@ struct request_queue *blk_alloc_queue_node(gfp_t gfp_mask, int node_id)
 	if (!q)
 		return NULL;
 
-	if (percpu_counter_init(&q->mq_usage_counter, 0))
-		goto fail_q;
-
 	q->id = ida_simple_get(&blk_queue_ida, 0, 0, gfp_mask);
 	if (q->id < 0)
-		goto fail_c;
+		goto fail_q;
 
 	q->backing_dev_info.ra_pages =
 			(VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
@@ -639,8 +636,6 @@ fail_bdi:
 	bdi_destroy(&q->backing_dev_info);
 fail_id:
 	ida_simple_remove(&blk_queue_ida, q->id);
-fail_c:
-	percpu_counter_destroy(&q->mq_usage_counter);
 fail_q:
 	kmem_cache_free(blk_requestq_cachep, q);
 	return NULL;
@@ -848,6 +843,47 @@ static void freed_request(struct request_list *rl, unsigned int flags)
 		__freed_request(rl, sync ^ 1);
 }
 
+int blk_update_nr_requests(struct request_queue *q, unsigned int nr)
+{
+	struct request_list *rl;
+
+	spin_lock_irq(q->queue_lock);
+	q->nr_requests = nr;
+	blk_queue_congestion_threshold(q);
+
+	/* congestion isn't cgroup aware and follows root blkcg for now */
+	rl = &q->root_rl;
+
+	if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
+		blk_set_queue_congested(q, BLK_RW_SYNC);
+	else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
+		blk_clear_queue_congested(q, BLK_RW_SYNC);
+
+	if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
+		blk_set_queue_congested(q, BLK_RW_ASYNC);
+	else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
+		blk_clear_queue_congested(q, BLK_RW_ASYNC);
+
+	blk_queue_for_each_rl(rl, q) {
+		if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
+			blk_set_rl_full(rl, BLK_RW_SYNC);
+		} else {
+			blk_clear_rl_full(rl, BLK_RW_SYNC);
+			wake_up(&rl->wait[BLK_RW_SYNC]);
+		}
+
+		if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
+			blk_set_rl_full(rl, BLK_RW_ASYNC);
+		} else {
+			blk_clear_rl_full(rl, BLK_RW_ASYNC);
+			wake_up(&rl->wait[BLK_RW_ASYNC]);
+		}
+	}
+
+	spin_unlock_irq(q->queue_lock);
+	return 0;
+}
+
 /*
  * Determine if elevator data should be initialized when allocating the
  * request associated with @bio.
@@ -1137,7 +1173,7 @@ static struct request *blk_old_get_request(struct request_queue *q, int rw,
 struct request *blk_get_request(struct request_queue *q, int rw, gfp_t gfp_mask)
 {
 	if (q->mq_ops)
-		return blk_mq_alloc_request(q, rw, gfp_mask);
+		return blk_mq_alloc_request(q, rw, gfp_mask, false);
 	else
 		return blk_old_get_request(q, rw, gfp_mask);
 }
@@ -1233,12 +1269,15 @@ static void add_acct_request(struct request_queue *q, struct request *rq,
 static void part_round_stats_single(int cpu, struct hd_struct *part,
 				    unsigned long now)
 {
+	int inflight;
+
 	if (now == part->stamp)
 		return;
 
-	if (part_in_flight(part)) {
+	inflight = part_in_flight(part);
+	if (inflight) {
 		__part_stat_add(cpu, part, time_in_queue,
-				part_in_flight(part) * (now - part->stamp));
+				inflight * (now - part->stamp));
 		__part_stat_add(cpu, part, io_ticks, (now - part->stamp));
 	}
 	part->stamp = now;
@@ -1427,6 +1466,8 @@ bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
  * added on the elevator at this point.  In addition, we don't have
  * reliable access to the elevator outside queue lock.  Only check basic
  * merging parameters without querying the elevator.
+ *
+ * Caller must ensure !blk_queue_nomerges(q) beforehand.
  */
 bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
 			    unsigned int *request_count)
@@ -1436,9 +1477,6 @@ bool blk_attempt_plug_merge(struct request_queue *q, struct bio *bio,
 	bool ret = false;
 	struct list_head *plug_list;
 
-	if (blk_queue_nomerges(q))
-		goto out;
-
 	plug = current->plug;
 	if (!plug)
 		goto out;
@@ -1517,7 +1555,8 @@ void blk_queue_bio(struct request_queue *q, struct bio *bio)
 	 * Check if we can merge with the plugged list before grabbing
 	 * any locks.
 	 */
-	if (blk_attempt_plug_merge(q, bio, &request_count))
+	if (!blk_queue_nomerges(q) &&
+	    blk_attempt_plug_merge(q, bio, &request_count))
 		return;
 
 	spin_lock_irq(q->queue_lock);

block/blk-flush.c

@@ -130,21 +130,13 @@ static void blk_flush_restore_request(struct request *rq)
 	blk_clear_rq_complete(rq);
 }
 
-static void mq_flush_run(struct work_struct *work)
-{
-	struct request *rq;
-
-	rq = container_of(work, struct request, requeue_work);
-
-	memset(&rq->csd, 0, sizeof(rq->csd));
-	blk_mq_insert_request(rq, false, true, false);
-}
-
 static bool blk_flush_queue_rq(struct request *rq, bool add_front)
 {
 	if (rq->q->mq_ops) {
-		INIT_WORK(&rq->requeue_work, mq_flush_run);
-		kblockd_schedule_work(&rq->requeue_work);
+		struct request_queue *q = rq->q;
+
+		blk_mq_add_to_requeue_list(rq, add_front);
+		blk_mq_kick_requeue_list(q);
 		return false;
 	} else {
 		if (add_front)

block/blk-iopoll.c

@@ -64,12 +64,12 @@ EXPORT_SYMBOL(__blk_iopoll_complete);
  *     iopoll handler will not be invoked again before blk_iopoll_sched_prep()
  *     is called.
  **/
-void blk_iopoll_complete(struct blk_iopoll *iopoll)
+void blk_iopoll_complete(struct blk_iopoll *iop)
 {
 	unsigned long flags;
 
 	local_irq_save(flags);
-	__blk_iopoll_complete(iopoll);
+	__blk_iopoll_complete(iop);
 	local_irq_restore(flags);
 }
 EXPORT_SYMBOL(blk_iopoll_complete);

block/blk-lib.c

@@ -226,8 +226,8 @@ EXPORT_SYMBOL(blkdev_issue_write_same);
  *  Generate and issue number of bios with zerofiled pages.
  */
 
-int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
-			sector_t nr_sects, gfp_t gfp_mask)
+static int __blkdev_issue_zeroout(struct block_device *bdev, sector_t sector,
+				  sector_t nr_sects, gfp_t gfp_mask)
 {
 	int ret;
 	struct bio *bio;

block/blk-mq-cpu.c

@@ -18,14 +18,18 @@ static int blk_mq_main_cpu_notify(struct notifier_block *self,
 {
 	unsigned int cpu = (unsigned long) hcpu;
 	struct blk_mq_cpu_notifier *notify;
+	int ret = NOTIFY_OK;
 
 	raw_spin_lock(&blk_mq_cpu_notify_lock);
 
-	list_for_each_entry(notify, &blk_mq_cpu_notify_list, list)
-		notify->notify(notify->data, action, cpu);
+	list_for_each_entry(notify, &blk_mq_cpu_notify_list, list) {
+		ret = notify->notify(notify->data, action, cpu);
+		if (ret != NOTIFY_OK)
+			break;
+	}
 
 	raw_spin_unlock(&blk_mq_cpu_notify_lock);
-	return NOTIFY_OK;
+	return ret;
 }
 
 void blk_mq_register_cpu_notifier(struct blk_mq_cpu_notifier *notifier)
@@ -45,7 +49,7 @@ void blk_mq_unregister_cpu_notifier(struct blk_mq_cpu_notifier *notifier)
 }
 
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      void (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned long, unsigned int),
 			      void *data)
 {
 	notifier->notify = fn;

block/blk-mq-cpumap.c

@@ -96,3 +96,19 @@ unsigned int *blk_mq_make_queue_map(struct blk_mq_tag_set *set)
 	kfree(map);
 	return NULL;
 }
+
+/*
+ * We have no quick way of doing reverse lookups. This is only used at
+ * queue init time, so runtime isn't important.
+ */
+int blk_mq_hw_queue_to_node(unsigned int *mq_map, unsigned int index)
+{
+	int i;
+
+	for_each_possible_cpu(i) {
+		if (index == mq_map[i])
+			return cpu_to_node(i);
+	}
+
+	return NUMA_NO_NODE;
+}

block/blk-mq-sysfs.c

@@ -203,45 +203,14 @@ static ssize_t blk_mq_hw_sysfs_rq_list_show(struct blk_mq_hw_ctx *hctx,
 	return ret;
 }
 
-static ssize_t blk_mq_hw_sysfs_ipi_show(struct blk_mq_hw_ctx *hctx, char *page)
-{
-	ssize_t ret;
-
-	spin_lock(&hctx->lock);
-	ret = sprintf(page, "%u\n", !!(hctx->flags & BLK_MQ_F_SHOULD_IPI));
-	spin_unlock(&hctx->lock);
-
-	return ret;
-}
-
-static ssize_t blk_mq_hw_sysfs_ipi_store(struct blk_mq_hw_ctx *hctx,
-					 const char *page, size_t len)
+static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
 {
-	struct blk_mq_ctx *ctx;
-	unsigned long ret;
-	unsigned int i;
-
-	if (kstrtoul(page, 10, &ret)) {
-		pr_err("blk-mq-sysfs: invalid input '%s'\n", page);
-		return -EINVAL;
-	}
-
-	spin_lock(&hctx->lock);
-	if (ret)
-		hctx->flags |= BLK_MQ_F_SHOULD_IPI;
-	else
-		hctx->flags &= ~BLK_MQ_F_SHOULD_IPI;
-	spin_unlock(&hctx->lock);
-
-	hctx_for_each_ctx(hctx, ctx, i)
-		ctx->ipi_redirect = !!ret;
-
-	return len;
+	return blk_mq_tag_sysfs_show(hctx->tags, page);
 }
 
-static ssize_t blk_mq_hw_sysfs_tags_show(struct blk_mq_hw_ctx *hctx, char *page)
+static ssize_t blk_mq_hw_sysfs_active_show(struct blk_mq_hw_ctx *hctx, char *page)
 {
-	return blk_mq_tag_sysfs_show(hctx->tags, page);
+	return sprintf(page, "%u\n", atomic_read(&hctx->nr_active));
 }
 
 static ssize_t blk_mq_hw_sysfs_cpus_show(struct blk_mq_hw_ctx *hctx, char *page)
@@ -303,15 +272,14 @@ static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_dispatched = {
 	.attr = {.name = "dispatched", .mode = S_IRUGO },
 	.show = blk_mq_hw_sysfs_dispatched_show,
 };
+static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_active = {
+	.attr = {.name = "active", .mode = S_IRUGO },
+	.show = blk_mq_hw_sysfs_active_show,
+};
 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_pending = {
 	.attr = {.name = "pending", .mode = S_IRUGO },
 	.show = blk_mq_hw_sysfs_rq_list_show,
 };
-static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_ipi = {
-	.attr = {.name = "ipi_redirect", .mode = S_IRUGO | S_IWUSR},
-	.show = blk_mq_hw_sysfs_ipi_show,
-	.store = blk_mq_hw_sysfs_ipi_store,
-};
 static struct blk_mq_hw_ctx_sysfs_entry blk_mq_hw_sysfs_tags = {
 	.attr = {.name = "tags", .mode = S_IRUGO },
 	.show = blk_mq_hw_sysfs_tags_show,
@@ -326,9 +294,9 @@ static struct attribute *default_hw_ctx_attrs[] = {
 	&blk_mq_hw_sysfs_run.attr,
 	&blk_mq_hw_sysfs_dispatched.attr,
 	&blk_mq_hw_sysfs_pending.attr,
-	&blk_mq_hw_sysfs_ipi.attr,
 	&blk_mq_hw_sysfs_tags.attr,
 	&blk_mq_hw_sysfs_cpus.attr,
+	&blk_mq_hw_sysfs_active.attr,
 	NULL,
 };
 

block/blk-mq-tag.c

@@ -1,64 +1,333 @@
 #include <linux/kernel.h>
 #include <linux/module.h>
+#include <linux/random.h>
 
 #include <linux/blk-mq.h>
 #include "blk.h"
 #include "blk-mq.h"
 #include "blk-mq-tag.h"
 
-void blk_mq_wait_for_tags(struct blk_mq_tags *tags)
+static bool bt_has_free_tags(struct blk_mq_bitmap_tags *bt)
 {
-	int tag = blk_mq_get_tag(tags, __GFP_WAIT, false);
-	blk_mq_put_tag(tags, tag);
+	int i;
+
+	for (i = 0; i < bt->map_nr; i++) {
+		struct blk_align_bitmap *bm = &bt->map[i];
+		int ret;
+
+		ret = find_first_zero_bit(&bm->word, bm->depth);
+		if (ret < bm->depth)
+			return true;
+	}
+
+	return false;
 }
 
 bool blk_mq_has_free_tags(struct blk_mq_tags *tags)
 {
-	return !tags ||
-		percpu_ida_free_tags(&tags->free_tags, nr_cpu_ids) != 0;
+	if (!tags)
+		return true;
+
+	return bt_has_free_tags(&tags->bitmap_tags);
+}
+
+static inline void bt_index_inc(unsigned int *index)
+{
+	*index = (*index + 1) & (BT_WAIT_QUEUES - 1);
+}
+
+/*
+ * If a previously inactive queue goes active, bump the active user count.
+ */
+bool __blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)
+{
+	if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state) &&
+	    !test_and_set_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
+		atomic_inc(&hctx->tags->active_queues);
+
+	return true;
+}
+
+/*
+ * Wakeup all potentially sleeping on normal (non-reserved) tags
+ */
+static void blk_mq_tag_wakeup_all(struct blk_mq_tags *tags)
+{
+	struct blk_mq_bitmap_tags *bt;
+	int i, wake_index;
+
+	bt = &tags->bitmap_tags;
+	wake_index = bt->wake_index;
+	for (i = 0; i < BT_WAIT_QUEUES; i++) {
+		struct bt_wait_state *bs = &bt->bs[wake_index];
+
+		if (waitqueue_active(&bs->wait))
+			wake_up(&bs->wait);
+
+		bt_index_inc(&wake_index);
+	}
 }
 
-static unsigned int __blk_mq_get_tag(struct blk_mq_tags *tags, gfp_t gfp)
+/*
+ * If a previously busy queue goes inactive, potential waiters could now
+ * be allowed to queue. Wake them up and check.
+ */
+void __blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)
 {
+	struct blk_mq_tags *tags = hctx->tags;
+
+	if (!test_and_clear_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
+		return;
+
+	atomic_dec(&tags->active_queues);
+
+	blk_mq_tag_wakeup_all(tags);
+}
+
+/*
+ * For shared tag users, we track the number of currently active users
+ * and attempt to provide a fair share of the tag depth for each of them.
+ */
+static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
+				  struct blk_mq_bitmap_tags *bt)
+{
+	unsigned int depth, users;
+
+	if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_SHARED))
+		return true;
+	if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
+		return true;
+
+	/*
+	 * Don't try dividing an ant
+	 */
+	if (bt->depth == 1)
+		return true;
+
+	users = atomic_read(&hctx->tags->active_queues);
+	if (!users)
+		return true;
+
+	/*
+	 * Allow at least some tags
+	 */
+	depth = max((bt->depth + users - 1) / users, 4U);
+	return atomic_read(&hctx->nr_active) < depth;
+}
+
+static int __bt_get_word(struct blk_align_bitmap *bm, unsigned int last_tag)
+{
+	int tag, org_last_tag, end;
+
+	org_last_tag = last_tag;
+	end = bm->depth;
+	do {
+restart:
+		tag = find_next_zero_bit(&bm->word, end, last_tag);
+		if (unlikely(tag >= end)) {
+			/*
+			 * We started with an offset, start from 0 to
+			 * exhaust the map.
+			 */
+			if (org_last_tag && last_tag) {
+				end = last_tag;
+				last_tag = 0;
+				goto restart;
+			}
+			return -1;
+		}
+		last_tag = tag + 1;
+	} while (test_and_set_bit_lock(tag, &bm->word));
+
+	return tag;
+}
+
+/*
+ * Straight forward bitmap tag implementation, where each bit is a tag
+ * (cleared == free, and set == busy). The small twist is using per-cpu
+ * last_tag caches, which blk-mq stores in the blk_mq_ctx software queue
+ * contexts. This enables us to drastically limit the space searched,
+ * without dirtying an extra shared cacheline like we would if we stored
+ * the cache value inside the shared blk_mq_bitmap_tags structure. On top
+ * of that, each word of tags is in a separate cacheline. This means that
+ * multiple users will tend to stick to different cachelines, at least
+ * until the map is exhausted.
+ */
+static int __bt_get(struct blk_mq_hw_ctx *hctx, struct blk_mq_bitmap_tags *bt,
+		    unsigned int *tag_cache)
+{
+	unsigned int last_tag, org_last_tag;
+	int index, i, tag;
+
+	if (!hctx_may_queue(hctx, bt))
+		return -1;
+
+	last_tag = org_last_tag = *tag_cache;
+	index = TAG_TO_INDEX(bt, last_tag);
+
+	for (i = 0; i < bt->map_nr; i++) {
+		tag = __bt_get_word(&bt->map[index], TAG_TO_BIT(bt, last_tag));
+		if (tag != -1) {
+			tag += (index << bt->bits_per_word);
+			goto done;
+		}
+
+		last_tag = 0;
+		if (++index >= bt->map_nr)
+			index = 0;
+	}
+
+	*tag_cache = 0;
+	return -1;
+
+	/*
+	 * Only update the cache from the allocation path, if we ended
+	 * up using the specific cached tag.
+	 */
+done:
+	if (tag == org_last_tag) {
+		last_tag = tag + 1;
+		if (last_tag >= bt->depth - 1)
+			last_tag = 0;
+
+		*tag_cache = last_tag;
+	}
+
+	return tag;
+}
+
+static struct bt_wait_state *bt_wait_ptr(struct blk_mq_bitmap_tags *bt,
+					 struct blk_mq_hw_ctx *hctx)
+{
+	struct bt_wait_state *bs;
+
+	if (!hctx)
+		return &bt->bs[0];
+
+	bs = &bt->bs[hctx->wait_index];
+	bt_index_inc(&hctx->wait_index);
+	return bs;
+}
+
+static int bt_get(struct blk_mq_bitmap_tags *bt, struct blk_mq_hw_ctx *hctx,
+		  unsigned int *last_tag, gfp_t gfp)
+{
+	struct bt_wait_state *bs;
+	DEFINE_WAIT(wait);
 	int tag;
 
-	tag = percpu_ida_alloc(&tags->free_tags, (gfp & __GFP_WAIT) ?
-			       TASK_UNINTERRUPTIBLE : TASK_RUNNING);
-	if (tag < 0)
-		return BLK_MQ_TAG_FAIL;
-	return tag + tags->nr_reserved_tags;
+	tag = __bt_get(hctx, bt, last_tag);
+	if (tag != -1)
+		return tag;
+
+	if (!(gfp & __GFP_WAIT))
+		return -1;
+
+	bs = bt_wait_ptr(bt, hctx);
+	do {
+		bool was_empty;
+
+		was_empty = list_empty(&wait.task_list);
+		prepare_to_wait(&bs->wait, &wait, TASK_UNINTERRUPTIBLE);
+
+		tag = __bt_get(hctx, bt, last_tag);
+		if (tag != -1)
+			break;
+
+		if (was_empty)
+			atomic_set(&bs->wait_cnt, bt->wake_cnt);
+
+		io_schedule();
+	} while (1);
+
+	finish_wait(&bs->wait, &wait);
+	return tag;
+}
+
+static unsigned int __blk_mq_get_tag(struct blk_mq_tags *tags,
+				     struct blk_mq_hw_ctx *hctx,
+				     unsigned int *last_tag, gfp_t gfp)
+{
+	int tag;
+
+	tag = bt_get(&tags->bitmap_tags, hctx, last_tag, gfp);
+	if (tag >= 0)
+		return tag + tags->nr_reserved_tags;
+
+	return BLK_MQ_TAG_FAIL;
 }
 
 static unsigned int __blk_mq_get_reserved_tag(struct blk_mq_tags *tags,
 					      gfp_t gfp)
 {
-	int tag;
+	int tag, zero = 0;
 
 	if (unlikely(!tags->nr_reserved_tags)) {
 		WARN_ON_ONCE(1);
 		return BLK_MQ_TAG_FAIL;
 	}
 
-	tag = percpu_ida_alloc(&tags->reserved_tags, (gfp & __GFP_WAIT) ?
-			       TASK_UNINTERRUPTIBLE : TASK_RUNNING);
+	tag = bt_get(&tags->breserved_tags, NULL, &zero, gfp);
 	if (tag < 0)
 		return BLK_MQ_TAG_FAIL;
+
 	return tag;
 }
 
-unsigned int blk_mq_get_tag(struct blk_mq_tags *tags, gfp_t gfp, bool reserved)
+unsigned int blk_mq_get_tag(struct blk_mq_hw_ctx *hctx, unsigned int *last_tag,
+			    gfp_t gfp, bool reserved)
 {
 	if (!reserved)
-		return __blk_mq_get_tag(tags, gfp);
+		return __blk_mq_get_tag(hctx->tags, hctx, last_tag, gfp);
 
-	return __blk_mq_get_reserved_tag(tags, gfp);
+	return __blk_mq_get_reserved_tag(hctx->tags, gfp);
+}
+
+static struct bt_wait_state *bt_wake_ptr(struct blk_mq_bitmap_tags *bt)
+{
+	int i, wake_index;
+
+	wake_index = bt->wake_index;
+	for (i = 0; i < BT_WAIT_QUEUES; i++) {
+		struct bt_wait_state *bs = &bt->bs[wake_index];
+
+		if (waitqueue_active(&bs->wait)) {
+			if (wake_index != bt->wake_index)
+				bt->wake_index = wake_index;
+
+			return bs;
+		}
+
+		bt_index_inc(&wake_index);
+	}
+
+	return NULL;
+}
+
+static void bt_clear_tag(struct blk_mq_bitmap_tags *bt, unsigned int tag)
+{
+	const int index = TAG_TO_INDEX(bt, tag);
+	struct bt_wait_state *bs;
+
+	/*
+	 * The unlock memory barrier need to order access to req in free
+	 * path and clearing tag bit
+	 */
+	clear_bit_unlock(TAG_TO_BIT(bt, tag), &bt->map[index].word);
+
+	bs = bt_wake_ptr(bt);
+	if (bs && atomic_dec_and_test(&bs->wait_cnt)) {
+		atomic_set(&bs->wait_cnt, bt->wake_cnt);
+		bt_index_inc(&bt->wake_index);
+		wake_up(&bs->wait);
+	}
 }
 
 static void __blk_mq_put_tag(struct blk_mq_tags *tags, unsigned int tag)
 {
 	BUG_ON(tag >= tags->nr_tags);
 
-	percpu_ida_free(&tags->free_tags, tag - tags->nr_reserved_tags);
+	bt_clear_tag(&tags->bitmap_tags, tag);
 }
 
 static void __blk_mq_put_reserved_tag(struct blk_mq_tags *tags,
@@ -66,22 +335,43 @@ static void __blk_mq_put_reserved_tag(struct blk_mq_tags *tags,
 {
 	BUG_ON(tag >= tags->nr_reserved_tags);
 
-	percpu_ida_free(&tags->reserved_tags, tag);
+	bt_clear_tag(&tags->breserved_tags, tag);
 }
 
-void blk_mq_put_tag(struct blk_mq_tags *tags, unsigned int tag)
+void blk_mq_put_tag(struct blk_mq_hw_ctx *hctx, unsigned int tag,
+		    unsigned int *last_tag)
 {
-	if (tag >= tags->nr_reserved_tags)
-		__blk_mq_put_tag(tags, tag);
-	else
+	struct blk_mq_tags *tags = hctx->tags;
+
+	if (tag >= tags->nr_reserved_tags) {
+		const int real_tag = tag - tags->nr_reserved_tags;
+
+		__blk_mq_put_tag(tags, real_tag);
+		*last_tag = real_tag;
+	} else
 		__blk_mq_put_reserved_tag(tags, tag);
 }
 
-static int __blk_mq_tag_iter(unsigned id, void *data)
+static void bt_for_each_free(struct blk_mq_bitmap_tags *bt,
+			     unsigned long *free_map, unsigned int off)
 {
-	unsigned long *tag_map = data;
-	__set_bit(id, tag_map);
-	return 0;
+	int i;
+
+	for (i = 0; i < bt->map_nr; i++) {
+		struct blk_align_bitmap *bm = &bt->map[i];
+		int bit = 0;
+
+		do {
+			bit = find_next_zero_bit(&bm->word, bm->depth, bit);
+			if (bit >= bm->depth)
+				break;
+
+			__set_bit(bit + off, free_map);
+			bit++;
+		} while (1);
+
+		off += (1 << bt->bits_per_word);
+	}
 }
 
 void blk_mq_tag_busy_iter(struct blk_mq_tags *tags,
@@ -95,21 +385,128 @@ void blk_mq_tag_busy_iter(struct blk_mq_tags *tags,
 	if (!tag_map)
 		return;
 
-	percpu_ida_for_each_free(&tags->free_tags, __blk_mq_tag_iter, tag_map);
+	bt_for_each_free(&tags->bitmap_tags, tag_map, tags->nr_reserved_tags);
 	if (tags->nr_reserved_tags)
-		percpu_ida_for_each_free(&tags->reserved_tags, __blk_mq_tag_iter,
-			tag_map);
+		bt_for_each_free(&tags->breserved_tags, tag_map, 0);
 
 	fn(data, tag_map);
 	kfree(tag_map);
 }
+EXPORT_SYMBOL(blk_mq_tag_busy_iter);
+
+static unsigned int bt_unused_tags(struct blk_mq_bitmap_tags *bt)
+{
+	unsigned int i, used;
+
+	for (i = 0, used = 0; i < bt->map_nr; i++) {
+		struct blk_align_bitmap *bm = &bt->map[i];
+
+		used += bitmap_weight(&bm->word, bm->depth);
+	}
+
+	return bt->depth - used;
+}
+
+static void bt_update_count(struct blk_mq_bitmap_tags *bt,
+			    unsigned int depth)
+{
+	unsigned int tags_per_word = 1U << bt->bits_per_word;
+	unsigned int map_depth = depth;
+
+	if (depth) {
+		int i;
+
+		for (i = 0; i < bt->map_nr; i++) {
+			bt->map[i].depth = min(map_depth, tags_per_word);
+			map_depth -= bt->map[i].depth;
+		}
+	}
+
+	bt->wake_cnt = BT_WAIT_BATCH;
+	if (bt->wake_cnt > depth / 4)
+		bt->wake_cnt = max(1U, depth / 4);
+
+	bt->depth = depth;
+}
+
+static int bt_alloc(struct blk_mq_bitmap_tags *bt, unsigned int depth,
+			int node, bool reserved)
+{
+	int i;
+
+	bt->bits_per_word = ilog2(BITS_PER_LONG);
+
+	/*
+	 * Depth can be zero for reserved tags, that's not a failure
+	 * condition.
+	 */
+	if (depth) {
+		unsigned int nr, tags_per_word;
+
+		tags_per_word = (1 << bt->bits_per_word);
+
+		/*
+		 * If the tag space is small, shrink the number of tags
+		 * per word so we spread over a few cachelines, at least.
+		 * If less than 4 tags, just forget about it, it's not
+		 * going to work optimally anyway.
+		 */
+		if (depth >= 4) {
+			while (tags_per_word * 4 > depth) {
+				bt->bits_per_word--;
+				tags_per_word = (1 << bt->bits_per_word);
+			}
+		}
+
+		nr = ALIGN(depth, tags_per_word) / tags_per_word;
+		bt->map = kzalloc_node(nr * sizeof(struct blk_align_bitmap),
+						GFP_KERNEL, node);
+		if (!bt->map)
+			return -ENOMEM;
+
+		bt->map_nr = nr;
+	}
+
+	bt->bs = kzalloc(BT_WAIT_QUEUES * sizeof(*bt->bs), GFP_KERNEL);
+	if (!bt->bs) {
+		kfree(bt->map);
+		return -ENOMEM;
+	}
+
+	for (i = 0; i < BT_WAIT_QUEUES; i++)
+		init_waitqueue_head(&bt->bs[i].wait);
+
+	bt_update_count(bt, depth);
+	return 0;
+}
+
+static void bt_free(struct blk_mq_bitmap_tags *bt)
+{
+	kfree(bt->map);
+	kfree(bt->bs);
+}
+
+static struct blk_mq_tags *blk_mq_init_bitmap_tags(struct blk_mq_tags *tags,
+						   int node)
+{
+	unsigned int depth = tags->nr_tags - tags->nr_reserved_tags;
+
+	if (bt_alloc(&tags->bitmap_tags, depth, node, false))
+		goto enomem;
+	if (bt_alloc(&tags->breserved_tags, tags->nr_reserved_tags, node, true))
+		goto enomem;
+
+	return tags;
+enomem:
+	bt_free(&tags->bitmap_tags);
+	kfree(tags);
+	return NULL;
+}
 
 struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags,
 				     unsigned int reserved_tags, int node)
 {
-	unsigned int nr_tags, nr_cache;
 	struct blk_mq_tags *tags;
-	int ret;
 
 	if (total_tags > BLK_MQ_TAG_MAX) {
 		pr_err("blk-mq: tag depth too large\n");
@@ -120,73 +517,59 @@ struct blk_mq_tags *blk_mq_init_tags(unsigned int total_tags,
 	if (!tags)
 		return NULL;
 
-	nr_tags = total_tags - reserved_tags;
-	nr_cache = nr_tags / num_possible_cpus();
-
-	if (nr_cache < BLK_MQ_TAG_CACHE_MIN)
-		nr_cache = BLK_MQ_TAG_CACHE_MIN;
-	else if (nr_cache > BLK_MQ_TAG_CACHE_MAX)
-		nr_cache = BLK_MQ_TAG_CACHE_MAX;
-
 	tags->nr_tags = total_tags;
 	tags->nr_reserved_tags = reserved_tags;
-	tags->nr_max_cache = nr_cache;
-	tags->nr_batch_move = max(1u, nr_cache / 2);
 
-	ret = __percpu_ida_init(&tags->free_tags, tags->nr_tags -
-				tags->nr_reserved_tags,
-				tags->nr_max_cache,
-				tags->nr_batch_move);
-	if (ret)
-		goto err_free_tags;
+	return blk_mq_init_bitmap_tags(tags, node);
+}
 
-	if (reserved_tags) {
-		/*
-		 * With max_cahe and batch set to 1, the allocator fallbacks to
-		 * no cached. It's fine reserved tags allocation is slow.
-		 */
-		ret = __percpu_ida_init(&tags->reserved_tags, reserved_tags,
-				1, 1);
-		if (ret)
-			goto err_reserved_tags;
-	}
+void blk_mq_free_tags(struct blk_mq_tags *tags)
+{
+	bt_free(&tags->bitmap_tags);
+	bt_free(&tags->breserved_tags);
+	kfree(tags);
+}
 
-	return tags;
+void blk_mq_tag_init_last_tag(struct blk_mq_tags *tags, unsigned int *tag)
+{
+	unsigned int depth = tags->nr_tags - tags->nr_reserved_tags;
 
-err_reserved_tags:
-	percpu_ida_destroy(&tags->free_tags);
-err_free_tags:
-	kfree(tags);
-	return NULL;
+	*tag = prandom_u32() % depth;
 }
 
-void blk_mq_free_tags(struct blk_mq_tags *tags)
+int blk_mq_tag_update_depth(struct blk_mq_tags *tags, unsigned int tdepth)
 {
-	percpu_ida_destroy(&tags->free_tags);
-	percpu_ida_destroy(&tags->reserved_tags);
-	kfree(tags);
+	tdepth -= tags->nr_reserved_tags;
+	if (tdepth > tags->nr_tags)
+		return -EINVAL;
+
+	/*
+	 * Don't need (or can't) update reserved tags here, they remain
+	 * static and should never need resizing.
+	 */
+	bt_update_count(&tags->bitmap_tags, tdepth);
+	blk_mq_tag_wakeup_all(tags);
+	return 0;
 }
 
 ssize_t blk_mq_tag_sysfs_show(struct blk_mq_tags *tags, char *page)
 {
 	char *orig_page = page;
-	unsigned int cpu;
+	unsigned int free, res;
 
 	if (!tags)
 		return 0;
 
-	page += sprintf(page, "nr_tags=%u, reserved_tags=%u, batch_move=%u,"
-			" max_cache=%u\n", tags->nr_tags, tags->nr_reserved_tags,
-			tags->nr_batch_move, tags->nr_max_cache);
+	page += sprintf(page, "nr_tags=%u, reserved_tags=%u, "
+			"bits_per_word=%u\n",
+			tags->nr_tags, tags->nr_reserved_tags,
+			tags->bitmap_tags.bits_per_word);
 
-	page += sprintf(page, "nr_free=%u, nr_reserved=%u\n",
-			percpu_ida_free_tags(&tags->free_tags, nr_cpu_ids),
-			percpu_ida_free_tags(&tags->reserved_tags, nr_cpu_ids));
+	free = bt_unused_tags(&tags->bitmap_tags);
+	res = bt_unused_tags(&tags->breserved_tags);
 
-	for_each_possible_cpu(cpu) {
-		page += sprintf(page, "  cpu%02u: nr_free=%u\n", cpu,
-				percpu_ida_free_tags(&tags->free_tags, cpu));
-	}
+	page += sprintf(page, "nr_free=%u, nr_reserved=%u\n", free, res);
+	page += sprintf(page, "active_queues=%u\n", atomic_read(&tags->active_queues));
 
 	return page - orig_page;
 }

block/blk-mq-tag.h

@@ -1,7 +1,32 @@
 #ifndef INT_BLK_MQ_TAG_H
 #define INT_BLK_MQ_TAG_H
 
-#include <linux/percpu_ida.h>
+#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;
+
+	unsigned int wake_index;
+	struct bt_wait_state *bs;
+};
 
 /*
  * Tag address space map.
@@ -9,11 +34,11 @@
 struct blk_mq_tags {
 	unsigned int nr_tags;
 	unsigned int nr_reserved_tags;
-	unsigned int nr_batch_move;
-	unsigned int nr_max_cache;
 
-	struct percpu_ida free_tags;
-	struct percpu_ida reserved_tags;
+	atomic_t active_queues;
+
+	struct blk_mq_bitmap_tags bitmap_tags;
+	struct blk_mq_bitmap_tags breserved_tags;
 
 	struct request **rqs;
 	struct list_head page_list;
@@ -23,12 +48,12 @@ struct blk_mq_tags {
 extern struct blk_mq_tags *blk_mq_init_tags(unsigned int nr_tags, unsigned int reserved_tags, int node);
 extern void blk_mq_free_tags(struct blk_mq_tags *tags);
 
-extern unsigned int blk_mq_get_tag(struct blk_mq_tags *tags, gfp_t gfp, bool reserved);
-extern void blk_mq_wait_for_tags(struct blk_mq_tags *tags);
-extern void blk_mq_put_tag(struct blk_mq_tags *tags, unsigned int tag);
-extern void blk_mq_tag_busy_iter(struct blk_mq_tags *tags, void (*fn)(void *data, unsigned long *), void *data);
+extern unsigned int blk_mq_get_tag(struct blk_mq_hw_ctx *hctx, unsigned int *last_tag, gfp_t gfp, bool reserved);
+extern void blk_mq_put_tag(struct blk_mq_hw_ctx *hctx, unsigned int tag, unsigned int *last_tag);
 extern bool blk_mq_has_free_tags(struct blk_mq_tags *tags);
 extern ssize_t blk_mq_tag_sysfs_show(struct blk_mq_tags *tags, char *page);
+extern void blk_mq_tag_init_last_tag(struct blk_mq_tags *tags, unsigned int *last_tag);
+extern int blk_mq_tag_update_depth(struct blk_mq_tags *tags, unsigned int depth);
 
 enum {
 	BLK_MQ_TAG_CACHE_MIN	= 1,
@@ -41,4 +66,23 @@ enum {
 	BLK_MQ_TAG_MAX		= BLK_MQ_TAG_FAIL - 1,
 };
 
+extern bool __blk_mq_tag_busy(struct blk_mq_hw_ctx *);
+extern void __blk_mq_tag_idle(struct blk_mq_hw_ctx *);
+
+static inline bool blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)
+{
+	if (!(hctx->flags & BLK_MQ_F_TAG_SHARED))
+		return false;
+
+	return __blk_mq_tag_busy(hctx);
+}
+
+static inline void blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)
+{
+	if (!(hctx->flags & BLK_MQ_F_TAG_SHARED))
+		return;
+
+	__blk_mq_tag_idle(hctx);
+}
+
 #endif

block/blk-mq.c

@@ -56,39 +56,40 @@ static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx)
 {
 	unsigned int i;
 
-	for (i = 0; i < hctx->nr_ctx_map; i++)
-		if (hctx->ctx_map[i])
+	for (i = 0; i < hctx->ctx_map.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];
+}
+
+#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)
 {
-	if (!test_bit(ctx->index_hw, hctx->ctx_map))
-		set_bit(ctx->index_hw, hctx->ctx_map);
+	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);
 }
 
-static struct request *__blk_mq_alloc_request(struct blk_mq_hw_ctx *hctx,
-					      gfp_t gfp, bool reserved)
+static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx,
+				      struct blk_mq_ctx *ctx)
 {
-	struct request *rq;
-	unsigned int tag;
-
-	tag = blk_mq_get_tag(hctx->tags, gfp, reserved);
-	if (tag != BLK_MQ_TAG_FAIL) {
-		rq = hctx->tags->rqs[tag];
-		blk_rq_init(hctx->queue, rq);
-		rq->tag = tag;
-
-		return rq;
-	}
+	struct blk_align_bitmap *bm = get_bm(hctx, ctx);
 
-	return NULL;
+	clear_bit(CTX_TO_BIT(hctx, ctx), &bm->word);
 }
 
 static int blk_mq_queue_enter(struct request_queue *q)
@@ -187,70 +188,109 @@ static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
 	if (blk_queue_io_stat(q))
 		rw_flags |= REQ_IO_STAT;
 
+	INIT_LIST_HEAD(&rq->queuelist);
+	/* csd/requeue_work/fifo_time is initialized before use */
+	rq->q = q;
 	rq->mq_ctx = ctx;
-	rq->cmd_flags = rw_flags;
+	rq->cmd_flags |= rw_flags;
+	rq->cmd_type = 0;
+	/* do not touch atomic flags, it needs atomic ops against the timer */
+	rq->cpu = -1;
+	rq->__data_len = 0;
+	rq->__sector = (sector_t) -1;
+	rq->bio = NULL;
+	rq->biotail = NULL;
+	INIT_HLIST_NODE(&rq->hash);
+	RB_CLEAR_NODE(&rq->rb_node);
+	memset(&rq->flush, 0, max(sizeof(rq->flush), sizeof(rq->elv)));
+	rq->rq_disk = NULL;
+	rq->part = NULL;
 	rq->start_time = jiffies;
+#ifdef CONFIG_BLK_CGROUP
+	rq->rl = NULL;
 	set_start_time_ns(rq);
+	rq->io_start_time_ns = 0;
+#endif
+	rq->nr_phys_segments = 0;
+#if defined(CONFIG_BLK_DEV_INTEGRITY)
+	rq->nr_integrity_segments = 0;
+#endif
+	rq->ioprio = 0;
+	rq->special = NULL;
+	/* tag was already set */
+	rq->errors = 0;
+	memset(rq->__cmd, 0, sizeof(rq->__cmd));
+	rq->cmd = rq->__cmd;
+	rq->cmd_len = BLK_MAX_CDB;
+
+	rq->extra_len = 0;
+	rq->sense_len = 0;
+	rq->resid_len = 0;
+	rq->sense = NULL;
+
+	rq->deadline = 0;
+	INIT_LIST_HEAD(&rq->timeout_list);
+	rq->timeout = 0;
+	rq->retries = 0;
+	rq->end_io = NULL;
+	rq->end_io_data = NULL;
+	rq->next_rq = NULL;
+
 	ctx->rq_dispatched[rw_is_sync(rw_flags)]++;
 }
 
-static struct request *blk_mq_alloc_request_pinned(struct request_queue *q,
-						   int rw, gfp_t gfp,
-						   bool reserved)
+static struct request *
+__blk_mq_alloc_request(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
+		struct blk_mq_ctx *ctx, int rw, gfp_t gfp, bool reserved)
 {
 	struct request *rq;
+	unsigned int tag;
 
-	do {
-		struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
-		struct blk_mq_hw_ctx *hctx = q->mq_ops->map_queue(q, ctx->cpu);
-
-		rq = __blk_mq_alloc_request(hctx, gfp & ~__GFP_WAIT, reserved);
-		if (rq) {
-			blk_mq_rq_ctx_init(q, ctx, rq, rw);
-			break;
-		}
+	tag = blk_mq_get_tag(hctx, &ctx->last_tag, gfp, reserved);
+	if (tag != BLK_MQ_TAG_FAIL) {
+		rq = hctx->tags->rqs[tag];
 
-		if (gfp & __GFP_WAIT) {
-			__blk_mq_run_hw_queue(hctx);
-			blk_mq_put_ctx(ctx);
-		} else {
-			blk_mq_put_ctx(ctx);
-			break;
+		rq->cmd_flags = 0;
+		if (blk_mq_tag_busy(hctx)) {
+			rq->cmd_flags = REQ_MQ_INFLIGHT;
+			atomic_inc(&hctx->nr_active);
 		}
 
-		blk_mq_wait_for_tags(hctx->tags);
-	} while (1);
+		rq->tag = tag;
+		blk_mq_rq_ctx_init(q, ctx, rq, rw);
+		return rq;
+	}
 
-	return rq;
+	return NULL;
 }
 
-struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp)
+struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp,
+		bool reserved)
 {
+	struct blk_mq_ctx *ctx;
+	struct blk_mq_hw_ctx *hctx;
 	struct request *rq;
 
 	if (blk_mq_queue_enter(q))
 		return NULL;
 
-	rq = blk_mq_alloc_request_pinned(q, rw, gfp, false);
-	if (rq)
-		blk_mq_put_ctx(rq->mq_ctx);
-	return rq;
-}
-
-struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw,
-					      gfp_t gfp)
-{
-	struct request *rq;
+	ctx = blk_mq_get_ctx(q);
+	hctx = q->mq_ops->map_queue(q, ctx->cpu);
 
-	if (blk_mq_queue_enter(q))
-		return NULL;
+	rq = __blk_mq_alloc_request(q, hctx, ctx, rw, gfp & ~__GFP_WAIT,
+				    reserved);
+	if (!rq && (gfp & __GFP_WAIT)) {
+		__blk_mq_run_hw_queue(hctx);
+		blk_mq_put_ctx(ctx);
 
-	rq = blk_mq_alloc_request_pinned(q, rw, gfp, true);
-	if (rq)
-		blk_mq_put_ctx(rq->mq_ctx);
+		ctx = blk_mq_get_ctx(q);
+		hctx = q->mq_ops->map_queue(q, ctx->cpu);
+		rq =  __blk_mq_alloc_request(q, hctx, ctx, rw, gfp, reserved);
+	}
+	blk_mq_put_ctx(ctx);
 	return rq;
 }
-EXPORT_SYMBOL(blk_mq_alloc_reserved_request);
+EXPORT_SYMBOL(blk_mq_alloc_request);
 
 static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx, struct request *rq)
@@ -258,7 +298,11 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 	const int tag = rq->tag;
 	struct request_queue *q = rq->q;
 
-	blk_mq_put_tag(hctx->tags, tag);
+	if (rq->cmd_flags & REQ_MQ_INFLIGHT)
+		atomic_dec(&hctx->nr_active);
+
+	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
+	blk_mq_put_tag(hctx, tag, &ctx->last_tag);
 	blk_mq_queue_exit(q);
 }
 
@@ -326,15 +370,19 @@ static void __blk_mq_complete_request_remote(void *data)
 void __blk_mq_complete_request(struct request *rq)
 {
 	struct blk_mq_ctx *ctx = rq->mq_ctx;
+	bool shared = false;
 	int cpu;
 
-	if (!ctx->ipi_redirect) {
+	if (!test_bit(QUEUE_FLAG_SAME_COMP, &rq->q->queue_flags)) {
 		rq->q->softirq_done_fn(rq);
 		return;
 	}
 
 	cpu = get_cpu();
-	if (cpu != ctx->cpu && cpu_online(ctx->cpu)) {
+	if (!test_bit(QUEUE_FLAG_SAME_FORCE, &rq->q->queue_flags))
+		shared = cpus_share_cache(cpu, ctx->cpu);
+
+	if (cpu != ctx->cpu && !shared && cpu_online(ctx->cpu)) {
 		rq->csd.func = __blk_mq_complete_request_remote;
 		rq->csd.info = rq;
 		rq->csd.flags = 0;
@@ -355,10 +403,16 @@ void __blk_mq_complete_request(struct request *rq)
  **/
 void blk_mq_complete_request(struct request *rq)
 {
-	if (unlikely(blk_should_fake_timeout(rq->q)))
+	struct request_queue *q = rq->q;
+
+	if (unlikely(blk_should_fake_timeout(q)))
 		return;
-	if (!blk_mark_rq_complete(rq))
-		__blk_mq_complete_request(rq);
+	if (!blk_mark_rq_complete(rq)) {
+		if (q->softirq_done_fn)
+			__blk_mq_complete_request(rq);
+		else
+			blk_mq_end_io(rq, rq->errors);
+	}
 }
 EXPORT_SYMBOL(blk_mq_complete_request);
 
@@ -375,10 +429,22 @@ static void blk_mq_start_request(struct request *rq, bool last)
 	/*
 	 * Just mark start time and set the started bit. Due to memory
 	 * ordering, we know we'll see the correct deadline as long as
-	 * REQ_ATOMIC_STARTED is seen.
+	 * REQ_ATOMIC_STARTED is seen. Use the default queue timeout,
+	 * unless one has been set in the request.
+	 */
+	if (!rq->timeout)
+		rq->deadline = jiffies + q->rq_timeout;
+	else
+		rq->deadline = jiffies + rq->timeout;
+
+	/*
+	 * Mark us as started and clear complete. Complete might have been
+	 * set if requeue raced with timeout, which then marked it as
+	 * complete. So be sure to clear complete again when we start
+	 * the request, otherwise we'll ignore the completion event.
 	 */
-	rq->deadline = jiffies + q->rq_timeout;
 	set_bit(REQ_ATOM_STARTED, &rq->atomic_flags);
+	clear_bit(REQ_ATOM_COMPLETE, &rq->atomic_flags);
 
 	if (q->dma_drain_size && blk_rq_bytes(rq)) {
 		/*
@@ -415,18 +481,72 @@ static void __blk_mq_requeue_request(struct request *rq)
 
 void blk_mq_requeue_request(struct request *rq)
 {
-	struct request_queue *q = rq->q;
-
 	__blk_mq_requeue_request(rq);
 	blk_clear_rq_complete(rq);
 
-	trace_block_rq_requeue(q, rq);
-
 	BUG_ON(blk_queued_rq(rq));
-	blk_mq_insert_request(rq, true, true, false);
+	blk_mq_add_to_requeue_list(rq, true);
 }
 EXPORT_SYMBOL(blk_mq_requeue_request);
 
+static void blk_mq_requeue_work(struct work_struct *work)
+{
+	struct request_queue *q =
+		container_of(work, struct request_queue, requeue_work);
+	LIST_HEAD(rq_list);
+	struct request *rq, *next;
+	unsigned long flags;
+
+	spin_lock_irqsave(&q->requeue_lock, flags);
+	list_splice_init(&q->requeue_list, &rq_list);
+	spin_unlock_irqrestore(&q->requeue_lock, flags);
+
+	list_for_each_entry_safe(rq, next, &rq_list, queuelist) {
+		if (!(rq->cmd_flags & REQ_SOFTBARRIER))
+			continue;
+
+		rq->cmd_flags &= ~REQ_SOFTBARRIER;
+		list_del_init(&rq->queuelist);
+		blk_mq_insert_request(rq, true, false, false);
+	}
+
+	while (!list_empty(&rq_list)) {
+		rq = list_entry(rq_list.next, struct request, queuelist);
+		list_del_init(&rq->queuelist);
+		blk_mq_insert_request(rq, false, false, false);
+	}
+
+	blk_mq_run_queues(q, false);
+}
+
+void blk_mq_add_to_requeue_list(struct request *rq, bool at_head)
+{
+	struct request_queue *q = rq->q;
+	unsigned long flags;
+
+	/*
+	 * We abuse this flag that is otherwise used by the I/O scheduler to
+	 * request head insertation from the workqueue.
+	 */
+	BUG_ON(rq->cmd_flags & REQ_SOFTBARRIER);
+
+	spin_lock_irqsave(&q->requeue_lock, flags);
+	if (at_head) {
+		rq->cmd_flags |= REQ_SOFTBARRIER;
+		list_add(&rq->queuelist, &q->requeue_list);
+	} else {
+		list_add_tail(&rq->queuelist, &q->requeue_list);
+	}
+	spin_unlock_irqrestore(&q->requeue_lock, flags);
+}
+EXPORT_SYMBOL(blk_mq_add_to_requeue_list);
+
+void blk_mq_kick_requeue_list(struct request_queue *q)
+{
+	kblockd_schedule_work(&q->requeue_work);
+}
+EXPORT_SYMBOL(blk_mq_kick_requeue_list);
+
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag)
 {
 	return tags->rqs[tag];
@@ -485,6 +605,28 @@ static void blk_mq_hw_ctx_check_timeout(struct blk_mq_hw_ctx *hctx,
 	blk_mq_tag_busy_iter(hctx->tags, blk_mq_timeout_check, &data);
 }
 
+static enum blk_eh_timer_return blk_mq_rq_timed_out(struct request *rq)
+{
+	struct request_queue *q = rq->q;
+
+	/*
+	 * We know that complete is set at this point. If STARTED isn't set
+	 * anymore, then the request isn't active and the "timeout" should
+	 * just be ignored. This can happen due to the bitflag ordering.
+	 * Timeout first checks if STARTED is set, and if it is, assumes
+	 * the request is active. But if we race with completion, then
+	 * we both flags will get cleared. So check here again, and ignore
+	 * a timeout event with a request that isn't active.
+	 */
+	if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
+		return BLK_EH_NOT_HANDLED;
+
+	if (!q->mq_ops->timeout)
+		return BLK_EH_RESET_TIMER;
+
+	return q->mq_ops->timeout(rq);
+}
+
 static void blk_mq_rq_timer(unsigned long data)
 {
 	struct request_queue *q = (struct request_queue *) data;
@@ -492,11 +634,24 @@ static void blk_mq_rq_timer(unsigned long data)
 	unsigned long next = 0;
 	int i, next_set = 0;
 
-	queue_for_each_hw_ctx(q, hctx, i)
+	queue_for_each_hw_ctx(q, hctx, i) {
+		/*
+		 * If not software queues are currently mapped to this
+		 * hardware queue, there's nothing to check
+		 */
+		if (!hctx->nr_ctx || !hctx->tags)
+			continue;
+
 		blk_mq_hw_ctx_check_timeout(hctx, &next, &next_set);
+	}
 
-	if (next_set)
-		mod_timer(&q->timeout, round_jiffies_up(next));
+	if (next_set) {
+		next = blk_rq_timeout(round_jiffies_up(next));
+		mod_timer(&q->timeout, next);
+	} else {
+		queue_for_each_hw_ctx(q, hctx, i)
+			blk_mq_tag_idle(hctx);
+	}
 }
 
 /*
@@ -538,9 +693,38 @@ static bool blk_mq_attempt_merge(struct request_queue *q,
 	return false;
 }
 
-void blk_mq_add_timer(struct request *rq)
+/*
+ * 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)
 {
-	__blk_add_timer(rq, NULL);
+	struct blk_mq_ctx *ctx;
+	int i;
+
+	for (i = 0; i < hctx->ctx_map.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);
+
+			bit++;
+		} while (1);
+	}
 }
 
 /*
@@ -552,10 +736,9 @@ void blk_mq_add_timer(struct request *rq)
 static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
 {
 	struct request_queue *q = hctx->queue;
-	struct blk_mq_ctx *ctx;
 	struct request *rq;
 	LIST_HEAD(rq_list);
-	int bit, queued;
+	int queued;
 
 	WARN_ON(!cpumask_test_cpu(raw_smp_processor_id(), hctx->cpumask));
 
@@ -567,15 +750,7 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
 	/*
 	 * Touch any software queue that has pending entries.
 	 */
-	for_each_set_bit(bit, hctx->ctx_map, hctx->nr_ctx) {
-		clear_bit(bit, hctx->ctx_map);
-		ctx = hctx->ctxs[bit];
-		BUG_ON(bit != ctx->index_hw);
-
-		spin_lock(&ctx->lock);
-		list_splice_tail_init(&ctx->rq_list, &rq_list);
-		spin_unlock(&ctx->lock);
-	}
+	flush_busy_ctxs(hctx, &rq_list);
 
 	/*
 	 * If we have previous entries on our dispatch list, grab them
@@ -588,14 +763,10 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
 		spin_unlock(&hctx->lock);
 	}
 
-	/*
-	 * Delete and return all entries from our dispatch list
-	 */
-	queued = 0;
-
 	/*
 	 * Now process all the entries, sending them to the driver.
 	 */
+	queued = 0;
 	while (!list_empty(&rq_list)) {
 		int ret;
 
@@ -610,11 +781,6 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
 			queued++;
 			continue;
 		case BLK_MQ_RQ_QUEUE_BUSY:
-			/*
-			 * FIXME: we should have a mechanism to stop the queue
-			 * like blk_stop_queue, otherwise we will waste cpu
-			 * time
-			 */
 			list_add(&rq->queuelist, &rq_list);
 			__blk_mq_requeue_request(rq);
 			break;
@@ -646,6 +812,30 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
 	}
 }
 
+/*
+ * It'd be great if the workqueue API had a way to pass
+ * in a mask and had some smarts for more clever placement.
+ * For now we just round-robin here, switching for every
+ * BLK_MQ_CPU_WORK_BATCH queued items.
+ */
+static int blk_mq_hctx_next_cpu(struct blk_mq_hw_ctx *hctx)
+{
+	int cpu = hctx->next_cpu;
+
+	if (--hctx->next_cpu_batch <= 0) {
+		int next_cpu;
+
+		next_cpu = cpumask_next(hctx->next_cpu, hctx->cpumask);
+		if (next_cpu >= nr_cpu_ids)
+			next_cpu = cpumask_first(hctx->cpumask);
+
+		hctx->next_cpu = next_cpu;
+		hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH;
+	}
+
+	return cpu;
+}
+
 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async)
 {
 	if (unlikely(test_bit(BLK_MQ_S_STOPPED, &hctx->state)))
@@ -658,13 +848,7 @@ void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async)
 	else {
 		unsigned int cpu;
 
-		/*
-		 * It'd be great if the workqueue API had a way to pass
-		 * in a mask and had some smarts for more clever placement
-		 * than the first CPU. Or we could round-robin here. For now,
-		 * just queue on the first CPU.
-		 */
-		cpu = cpumask_first(hctx->cpumask);
+		cpu = blk_mq_hctx_next_cpu(hctx);
 		kblockd_schedule_delayed_work_on(cpu, &hctx->run_work, 0);
 	}
 }
@@ -771,13 +955,7 @@ void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs)
 	else {
 		unsigned int cpu;
 
-		/*
-		 * It'd be great if the workqueue API had a way to pass
-		 * in a mask and had some smarts for more clever placement
-		 * than the first CPU. Or we could round-robin here. For now,
-		 * just queue on the first CPU.
-		 */
-		cpu = cpumask_first(hctx->cpumask);
+		cpu = blk_mq_hctx_next_cpu(hctx);
 		kblockd_schedule_delayed_work_on(cpu, &hctx->delay_work, tmo);
 	}
 }
@@ -794,12 +972,13 @@ static void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx,
 		list_add(&rq->queuelist, &ctx->rq_list);
 	else
 		list_add_tail(&rq->queuelist, &ctx->rq_list);
+
 	blk_mq_hctx_mark_pending(hctx, ctx);
 
 	/*
 	 * We do this early, to ensure we are on the right CPU.
 	 */
-	blk_mq_add_timer(rq);
+	blk_add_timer(rq);
 }
 
 void blk_mq_insert_request(struct request *rq, bool at_head, bool run_queue,
@@ -930,21 +1109,161 @@ static void blk_mq_bio_to_request(struct request *rq, struct bio *bio)
 	blk_account_io_start(rq, 1);
 }
 
-static void blk_mq_make_request(struct request_queue *q, struct bio *bio)
+static inline bool blk_mq_merge_queue_io(struct blk_mq_hw_ctx *hctx,
+					 struct blk_mq_ctx *ctx,
+					 struct request *rq, struct bio *bio)
+{
+	struct request_queue *q = hctx->queue;
+
+	if (!(hctx->flags & BLK_MQ_F_SHOULD_MERGE)) {
+		blk_mq_bio_to_request(rq, bio);
+		spin_lock(&ctx->lock);
+insert_rq:
+		__blk_mq_insert_request(hctx, rq, false);
+		spin_unlock(&ctx->lock);
+		return false;
+	} else {
+		spin_lock(&ctx->lock);
+		if (!blk_mq_attempt_merge(q, ctx, bio)) {
+			blk_mq_bio_to_request(rq, bio);
+			goto insert_rq;
+		}
+
+		spin_unlock(&ctx->lock);
+		__blk_mq_free_request(hctx, ctx, rq);
+		return true;
+	}
+}
+
+struct blk_map_ctx {
+	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_ctx *ctx;
+};
+
+static struct request *blk_mq_map_request(struct request_queue *q,
+					  struct bio *bio,
+					  struct blk_map_ctx *data)
 {
 	struct blk_mq_hw_ctx *hctx;
 	struct blk_mq_ctx *ctx;
+	struct request *rq;
+	int rw = bio_data_dir(bio);
+
+	if (unlikely(blk_mq_queue_enter(q))) {
+		bio_endio(bio, -EIO);
+		return NULL;
+	}
+
+	ctx = blk_mq_get_ctx(q);
+	hctx = q->mq_ops->map_queue(q, ctx->cpu);
+
+	if (rw_is_sync(bio->bi_rw))
+		rw |= REQ_SYNC;
+
+	trace_block_getrq(q, bio, rw);
+	rq = __blk_mq_alloc_request(q, hctx, ctx, rw, GFP_ATOMIC, false);
+	if (unlikely(!rq)) {
+		__blk_mq_run_hw_queue(hctx);
+		blk_mq_put_ctx(ctx);
+		trace_block_sleeprq(q, bio, rw);
+
+		ctx = blk_mq_get_ctx(q);
+		hctx = q->mq_ops->map_queue(q, ctx->cpu);
+		rq = __blk_mq_alloc_request(q, hctx, ctx, rw,
+					    __GFP_WAIT|GFP_ATOMIC, false);
+	}
+
+	hctx->queued++;
+	data->hctx = hctx;
+	data->ctx = ctx;
+	return rq;
+}
+
+/*
+ * Multiple hardware queue variant. This will not use per-process plugs,
+ * but will attempt to bypass the hctx queueing if we can go straight to
+ * hardware for SYNC IO.
+ */
+static void blk_mq_make_request(struct request_queue *q, struct bio *bio)
+{
 	const int is_sync = rw_is_sync(bio->bi_rw);
 	const int is_flush_fua = bio->bi_rw & (REQ_FLUSH | REQ_FUA);
-	int rw = bio_data_dir(bio);
+	struct blk_map_ctx data;
 	struct request *rq;
+
+	blk_queue_bounce(q, &bio);
+
+	if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
+		bio_endio(bio, -EIO);
+		return;
+	}
+
+	rq = blk_mq_map_request(q, bio, &data);
+	if (unlikely(!rq))
+		return;
+
+	if (unlikely(is_flush_fua)) {
+		blk_mq_bio_to_request(rq, bio);
+		blk_insert_flush(rq);
+		goto run_queue;
+	}
+
+	if (is_sync) {
+		int ret;
+
+		blk_mq_bio_to_request(rq, bio);
+		blk_mq_start_request(rq, true);
+
+		/*
+		 * For OK queue, we are done. For error, kill it. Any other
+		 * error (busy), just add it to our list as we previously
+		 * would have done
+		 */
+		ret = q->mq_ops->queue_rq(data.hctx, rq);
+		if (ret == BLK_MQ_RQ_QUEUE_OK)
+			goto done;
+		else {
+			__blk_mq_requeue_request(rq);
+
+			if (ret == BLK_MQ_RQ_QUEUE_ERROR) {
+				rq->errors = -EIO;
+				blk_mq_end_io(rq, rq->errors);
+				goto done;
+			}
+		}
+	}
+
+	if (!blk_mq_merge_queue_io(data.hctx, data.ctx, rq, bio)) {
+		/*
+		 * For a SYNC request, send it to the hardware immediately. For
+		 * an ASYNC request, just ensure that we run it later on. The
+		 * latter allows for merging opportunities and more efficient
+		 * dispatching.
+		 */
+run_queue:
+		blk_mq_run_hw_queue(data.hctx, !is_sync || is_flush_fua);
+	}
+done:
+	blk_mq_put_ctx(data.ctx);
+}
+
+/*
+ * Single hardware queue variant. This will attempt to use any per-process
+ * plug for merging and IO deferral.
+ */
+static void blk_sq_make_request(struct request_queue *q, struct bio *bio)
+{
+	const int is_sync = rw_is_sync(bio->bi_rw);
+	const int is_flush_fua = bio->bi_rw & (REQ_FLUSH | REQ_FUA);
 	unsigned int use_plug, request_count = 0;
+	struct blk_map_ctx data;
+	struct request *rq;
 
 	/*
 	 * If we have multiple hardware queues, just go directly to
 	 * one of those for sync IO.
 	 */
-	use_plug = !is_flush_fua && ((q->nr_hw_queues == 1) || !is_sync);
+	use_plug = !is_flush_fua && !is_sync;
 
 	blk_queue_bounce(q, &bio);
 
@@ -953,33 +1272,11 @@ static void blk_mq_make_request(struct request_queue *q, struct bio *bio)
 		return;
 	}
 
-	if (use_plug && blk_attempt_plug_merge(q, bio, &request_count))
+	if (use_plug && !blk_queue_nomerges(q) &&
+	    blk_attempt_plug_merge(q, bio, &request_count))
 		return;
 
-	if (blk_mq_queue_enter(q)) {
-		bio_endio(bio, -EIO);
-		return;
-	}
-
-	ctx = blk_mq_get_ctx(q);
-	hctx = q->mq_ops->map_queue(q, ctx->cpu);
-
-	if (is_sync)
-		rw |= REQ_SYNC;
-	trace_block_getrq(q, bio, rw);
-	rq = __blk_mq_alloc_request(hctx, GFP_ATOMIC, false);
-	if (likely(rq))
-		blk_mq_rq_ctx_init(q, ctx, rq, rw);
-	else {
-		blk_mq_put_ctx(ctx);
-		trace_block_sleeprq(q, bio, rw);
-		rq = blk_mq_alloc_request_pinned(q, rw, __GFP_WAIT|GFP_ATOMIC,
-							false);
-		ctx = rq->mq_ctx;
-		hctx = q->mq_ops->map_queue(q, ctx->cpu);
-	}
-
-	hctx->queued++;
+	rq = blk_mq_map_request(q, bio, &data);
 
 	if (unlikely(is_flush_fua)) {
 		blk_mq_bio_to_request(rq, bio);
@@ -1004,31 +1301,23 @@ static void blk_mq_make_request(struct request_queue *q, struct bio *bio)
 				trace_block_plug(q);
 			}
 			list_add_tail(&rq->queuelist, &plug->mq_list);
-			blk_mq_put_ctx(ctx);
+			blk_mq_put_ctx(data.ctx);
 			return;
 		}
 	}
 
-	spin_lock(&ctx->lock);
-
-	if ((hctx->flags & BLK_MQ_F_SHOULD_MERGE) &&
-	    blk_mq_attempt_merge(q, ctx, bio))
-		__blk_mq_free_request(hctx, ctx, rq);
-	else {
-		blk_mq_bio_to_request(rq, bio);
-		__blk_mq_insert_request(hctx, rq, false);
+	if (!blk_mq_merge_queue_io(data.hctx, data.ctx, rq, bio)) {
+		/*
+		 * For a SYNC request, send it to the hardware immediately. For
+		 * an ASYNC request, just ensure that we run it later on. The
+		 * latter allows for merging opportunities and more efficient
+		 * dispatching.
+		 */
+run_queue:
+		blk_mq_run_hw_queue(data.hctx, !is_sync || is_flush_fua);
 	}
 
-	spin_unlock(&ctx->lock);
-
-	/*
-	 * For a SYNC request, send it to the hardware immediately. For an
-	 * ASYNC request, just ensure that we run it later on. The latter
-	 * allows for merging opportunities and more efficient dispatching.
-	 */
-run_queue:
-	blk_mq_run_hw_queue(hctx, !is_sync || is_flush_fua);
-	blk_mq_put_ctx(ctx);
+	blk_mq_put_ctx(data.ctx);
 }
 
 /*
@@ -1041,10 +1330,10 @@ struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q, const int cpu)
 EXPORT_SYMBOL(blk_mq_map_queue);
 
 struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *set,
-						   unsigned int hctx_index)
+						   unsigned int hctx_index,
+						   int node)
 {
-	return kmalloc_node(sizeof(struct blk_mq_hw_ctx),
-				GFP_KERNEL | __GFP_ZERO, set->numa_node);
+	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, node);
 }
 EXPORT_SYMBOL(blk_mq_alloc_single_hw_queue);
 
@@ -1055,52 +1344,6 @@ void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *hctx,
 }
 EXPORT_SYMBOL(blk_mq_free_single_hw_queue);
 
-static void blk_mq_hctx_notify(void *data, unsigned long action,
-			       unsigned int cpu)
-{
-	struct blk_mq_hw_ctx *hctx = data;
-	struct request_queue *q = hctx->queue;
-	struct blk_mq_ctx *ctx;
-	LIST_HEAD(tmp);
-
-	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN)
-		return;
-
-	/*
-	 * Move ctx entries to new CPU, if this one is going away.
-	 */
-	ctx = __blk_mq_get_ctx(q, cpu);
-
-	spin_lock(&ctx->lock);
-	if (!list_empty(&ctx->rq_list)) {
-		list_splice_init(&ctx->rq_list, &tmp);
-		clear_bit(ctx->index_hw, hctx->ctx_map);
-	}
-	spin_unlock(&ctx->lock);
-
-	if (list_empty(&tmp))
-		return;
-
-	ctx = blk_mq_get_ctx(q);
-	spin_lock(&ctx->lock);
-
-	while (!list_empty(&tmp)) {
-		struct request *rq;
-
-		rq = list_first_entry(&tmp, struct request, queuelist);
-		rq->mq_ctx = ctx;
-		list_move_tail(&rq->queuelist, &ctx->rq_list);
-	}
-
-	hctx = q->mq_ops->map_queue(q, ctx->cpu);
-	blk_mq_hctx_mark_pending(hctx, ctx);
-
-	spin_unlock(&ctx->lock);
-
-	blk_mq_run_hw_queue(hctx, true);
-	blk_mq_put_ctx(ctx);
-}
-
 static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 		struct blk_mq_tags *tags, unsigned int hctx_idx)
 {
@@ -1130,12 +1373,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-	size_t ret = PAGE_SIZE;
-
-	while (order--)
-		ret *= 2;
-
-	return ret;
+	return (size_t)PAGE_SIZE << order;
 }
 
 static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
@@ -1219,17 +1457,147 @@ fail:
 	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;
+
+	bitmap->map_size = num_maps;
+
+	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;
+}
+
+static int blk_mq_hctx_cpu_offline(struct blk_mq_hw_ctx *hctx, int cpu)
+{
+	struct request_queue *q = hctx->queue;
+	struct blk_mq_ctx *ctx;
+	LIST_HEAD(tmp);
+
+	/*
+	 * Move ctx entries to new CPU, if this one is going away.
+	 */
+	ctx = __blk_mq_get_ctx(q, cpu);
+
+	spin_lock(&ctx->lock);
+	if (!list_empty(&ctx->rq_list)) {
+		list_splice_init(&ctx->rq_list, &tmp);
+		blk_mq_hctx_clear_pending(hctx, ctx);
+	}
+	spin_unlock(&ctx->lock);
+
+	if (list_empty(&tmp))
+		return NOTIFY_OK;
+
+	ctx = blk_mq_get_ctx(q);
+	spin_lock(&ctx->lock);
+
+	while (!list_empty(&tmp)) {
+		struct request *rq;
+
+		rq = list_first_entry(&tmp, struct request, queuelist);
+		rq->mq_ctx = ctx;
+		list_move_tail(&rq->queuelist, &ctx->rq_list);
+	}
+
+	hctx = q->mq_ops->map_queue(q, ctx->cpu);
+	blk_mq_hctx_mark_pending(hctx, ctx);
+
+	spin_unlock(&ctx->lock);
+
+	blk_mq_run_hw_queue(hctx, true);
+	blk_mq_put_ctx(ctx);
+	return NOTIFY_OK;
+}
+
+static int blk_mq_hctx_cpu_online(struct blk_mq_hw_ctx *hctx, int cpu)
+{
+	struct request_queue *q = hctx->queue;
+	struct blk_mq_tag_set *set = q->tag_set;
+
+	if (set->tags[hctx->queue_num])
+		return NOTIFY_OK;
+
+	set->tags[hctx->queue_num] = blk_mq_init_rq_map(set, hctx->queue_num);
+	if (!set->tags[hctx->queue_num])
+		return NOTIFY_STOP;
+
+	hctx->tags = set->tags[hctx->queue_num];
+	return NOTIFY_OK;
+}
+
+static int blk_mq_hctx_notify(void *data, unsigned long action,
+			      unsigned int cpu)
+{
+	struct blk_mq_hw_ctx *hctx = data;
+
+	if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
+		return blk_mq_hctx_cpu_offline(hctx, cpu);
+	else if (action == CPU_ONLINE || action == CPU_ONLINE_FROZEN)
+		return blk_mq_hctx_cpu_online(hctx, cpu);
+
+	return NOTIFY_OK;
+}
+
+static void blk_mq_exit_hw_queues(struct request_queue *q,
+		struct blk_mq_tag_set *set, int nr_queue)
+{
+	struct blk_mq_hw_ctx *hctx;
+	unsigned int i;
+
+	queue_for_each_hw_ctx(q, hctx, i) {
+		if (i == nr_queue)
+			break;
+
+		if (set->ops->exit_hctx)
+			set->ops->exit_hctx(hctx, i);
+
+		blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
+		kfree(hctx->ctxs);
+		blk_mq_free_bitmap(&hctx->ctx_map);
+	}
+
+}
+
+static void blk_mq_free_hw_queues(struct request_queue *q,
+		struct blk_mq_tag_set *set)
+{
+	struct blk_mq_hw_ctx *hctx;
+	unsigned int i;
+
+	queue_for_each_hw_ctx(q, hctx, i) {
+		free_cpumask_var(hctx->cpumask);
+		set->ops->free_hctx(hctx, i);
+	}
+}
+
 static int blk_mq_init_hw_queues(struct request_queue *q,
 		struct blk_mq_tag_set *set)
 {
 	struct blk_mq_hw_ctx *hctx;
-	unsigned int i, j;
+	unsigned int i;
 
 	/*
 	 * Initialize hardware queues
 	 */
 	queue_for_each_hw_ctx(q, hctx, i) {
-		unsigned int num_maps;
 		int node;
 
 		node = hctx->numa_node;
@@ -1260,13 +1628,9 @@ static int blk_mq_init_hw_queues(struct request_queue *q,
 		if (!hctx->ctxs)
 			break;
 
-		num_maps = ALIGN(nr_cpu_ids, BITS_PER_LONG) / BITS_PER_LONG;
-		hctx->ctx_map = kzalloc_node(num_maps * sizeof(unsigned long),
-						GFP_KERNEL, node);
-		if (!hctx->ctx_map)
+		if (blk_mq_alloc_bitmap(&hctx->ctx_map, node))
 			break;
 
-		hctx->nr_ctx_map = num_maps;
 		hctx->nr_ctx = 0;
 
 		if (set->ops->init_hctx &&
@@ -1280,16 +1644,7 @@ static int blk_mq_init_hw_queues(struct request_queue *q,
 	/*
 	 * Init failed
 	 */
-	queue_for_each_hw_ctx(q, hctx, j) {
-		if (i == j)
-			break;
-
-		if (set->ops->exit_hctx)
-			set->ops->exit_hctx(hctx, j);
-
-		blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
-		kfree(hctx->ctxs);
-	}
+	blk_mq_exit_hw_queues(q, set, i);
 
 	return 1;
 }
@@ -1350,6 +1705,79 @@ static void blk_mq_map_swqueue(struct request_queue *q)
 		ctx->index_hw = hctx->nr_ctx;
 		hctx->ctxs[hctx->nr_ctx++] = ctx;
 	}
+
+	queue_for_each_hw_ctx(q, hctx, i) {
+		/*
+		 * If not software queues are mapped to this hardware queue,
+		 * disable it and free the request entries
+		 */
+		if (!hctx->nr_ctx) {
+			struct blk_mq_tag_set *set = q->tag_set;
+
+			if (set->tags[i]) {
+				blk_mq_free_rq_map(set, set->tags[i], i);
+				set->tags[i] = NULL;
+				hctx->tags = NULL;
+			}
+			continue;
+		}
+
+		/*
+		 * Initialize batch roundrobin counts
+		 */
+		hctx->next_cpu = cpumask_first(hctx->cpumask);
+		hctx->next_cpu_batch = BLK_MQ_CPU_WORK_BATCH;
+	}
+}
+
+static void blk_mq_update_tag_set_depth(struct blk_mq_tag_set *set)
+{
+	struct blk_mq_hw_ctx *hctx;
+	struct request_queue *q;
+	bool shared;
+	int i;
+
+	if (set->tag_list.next == set->tag_list.prev)
+		shared = false;
+	else
+		shared = true;
+
+	list_for_each_entry(q, &set->tag_list, tag_set_list) {
+		blk_mq_freeze_queue(q);
+
+		queue_for_each_hw_ctx(q, hctx, i) {
+			if (shared)
+				hctx->flags |= BLK_MQ_F_TAG_SHARED;
+			else
+				hctx->flags &= ~BLK_MQ_F_TAG_SHARED;
+		}
+		blk_mq_unfreeze_queue(q);
+	}
+}
+
+static void blk_mq_del_queue_tag_set(struct request_queue *q)
+{
+	struct blk_mq_tag_set *set = q->tag_set;
+
+	blk_mq_freeze_queue(q);
+
+	mutex_lock(&set->tag_list_lock);
+	list_del_init(&q->tag_set_list);
+	blk_mq_update_tag_set_depth(set);
+	mutex_unlock(&set->tag_list_lock);
+
+	blk_mq_unfreeze_queue(q);
+}
+
+static void blk_mq_add_queue_tag_set(struct blk_mq_tag_set *set,
+				     struct request_queue *q)
+{
+	q->tag_set = set;
+
+	mutex_lock(&set->tag_list_lock);
+	list_add_tail(&q->tag_set_list, &set->tag_list);
+	blk_mq_update_tag_set_depth(set);
+	mutex_unlock(&set->tag_list_lock);
 }
 
 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
@@ -1357,6 +1785,7 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	struct blk_mq_hw_ctx **hctxs;
 	struct blk_mq_ctx *ctx;
 	struct request_queue *q;
+	unsigned int *map;
 	int i;
 
 	ctx = alloc_percpu(struct blk_mq_ctx);
@@ -1369,15 +1798,22 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	if (!hctxs)
 		goto err_percpu;
 
+	map = blk_mq_make_queue_map(set);
+	if (!map)
+		goto err_map;
+
 	for (i = 0; i < set->nr_hw_queues; i++) {
-		hctxs[i] = set->ops->alloc_hctx(set, i);
+		int node = blk_mq_hw_queue_to_node(map, i);
+
+		hctxs[i] = set->ops->alloc_hctx(set, i, node);
 		if (!hctxs[i])
 			goto err_hctxs;
 
 		if (!zalloc_cpumask_var(&hctxs[i]->cpumask, GFP_KERNEL))
 			goto err_hctxs;
 
-		hctxs[i]->numa_node = NUMA_NO_NODE;
+		atomic_set(&hctxs[i]->nr_active, 0);
+		hctxs[i]->numa_node = node;
 		hctxs[i]->queue_num = i;
 	}
 
@@ -1385,8 +1821,7 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	if (!q)
 		goto err_hctxs;
 
-	q->mq_map = blk_mq_make_queue_map(set);
-	if (!q->mq_map)
+	if (percpu_counter_init(&q->mq_usage_counter, 0))
 		goto err_map;
 
 	setup_timer(&q->timeout, blk_mq_rq_timer, (unsigned long) q);
@@ -1394,6 +1829,7 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 
 	q->nr_queues = nr_cpu_ids;
 	q->nr_hw_queues = set->nr_hw_queues;
+	q->mq_map = map;
 
 	q->queue_ctx = ctx;
 	q->queue_hw_ctx = hctxs;
@@ -1403,11 +1839,24 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 
 	q->sg_reserved_size = INT_MAX;
 
-	blk_queue_make_request(q, blk_mq_make_request);
-	blk_queue_rq_timed_out(q, set->ops->timeout);
+	INIT_WORK(&q->requeue_work, blk_mq_requeue_work);
+	INIT_LIST_HEAD(&q->requeue_list);
+	spin_lock_init(&q->requeue_lock);
+
+	if (q->nr_hw_queues > 1)
+		blk_queue_make_request(q, blk_mq_make_request);
+	else
+		blk_queue_make_request(q, blk_sq_make_request);
+
+	blk_queue_rq_timed_out(q, blk_mq_rq_timed_out);
 	if (set->timeout)
 		blk_queue_rq_timeout(q, set->timeout);
 
+	/*
+	 * Do this after blk_queue_make_request() overrides it...
+	 */
+	q->nr_requests = set->queue_depth;
+
 	if (set->ops->complete)
 		blk_queue_softirq_done(q, set->ops->complete);
 
@@ -1423,27 +1872,29 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
 	if (blk_mq_init_hw_queues(q, set))
 		goto err_flush_rq;
 
-	blk_mq_map_swqueue(q);
-
 	mutex_lock(&all_q_mutex);
 	list_add_tail(&q->all_q_node, &all_q_list);
 	mutex_unlock(&all_q_mutex);
 
+	blk_mq_add_queue_tag_set(set, q);
+
+	blk_mq_map_swqueue(q);
+
 	return q;
 
 err_flush_rq:
 	kfree(q->flush_rq);
 err_hw:
-	kfree(q->mq_map);
-err_map:
 	blk_cleanup_queue(q);
 err_hctxs:
+	kfree(map);
 	for (i = 0; i < set->nr_hw_queues; i++) {
 		if (!hctxs[i])
 			break;
 		free_cpumask_var(hctxs[i]->cpumask);
 		set->ops->free_hctx(hctxs[i], i);
 	}
+err_map:
 	kfree(hctxs);
 err_percpu:
 	free_percpu(ctx);
@@ -1453,18 +1904,14 @@ EXPORT_SYMBOL(blk_mq_init_queue);
 
 void blk_mq_free_queue(struct request_queue *q)
 {
-	struct blk_mq_hw_ctx *hctx;
-	int i;
+	struct blk_mq_tag_set	*set = q->tag_set;
 
-	queue_for_each_hw_ctx(q, hctx, i) {
-		kfree(hctx->ctx_map);
-		kfree(hctx->ctxs);
-		blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
-		if (q->mq_ops->exit_hctx)
-			q->mq_ops->exit_hctx(hctx, i);
-		free_cpumask_var(hctx->cpumask);
-		q->mq_ops->free_hctx(hctx, i);
-	}
+	blk_mq_del_queue_tag_set(q);
+
+	blk_mq_exit_hw_queues(q, set, set->nr_hw_queues);
+	blk_mq_free_hw_queues(q, set);
+
+	percpu_counter_destroy(&q->mq_usage_counter);
 
 	free_percpu(q->queue_ctx);
 	kfree(q->queue_hw_ctx);
@@ -1503,10 +1950,10 @@ static int blk_mq_queue_reinit_notify(struct notifier_block *nb,
 	struct request_queue *q;
 
 	/*
-	 * Before new mapping is established, hotadded cpu might already start
-	 * handling requests. This doesn't break anything as we map offline
-	 * CPUs to first hardware queue. We will re-init queue below to get
-	 * optimal settings.
+	 * Before new mappings are established, hotadded cpu might already
+	 * start handling requests. This doesn't break anything as we map
+	 * offline CPUs to first hardware queue. We will re-init the queue
+	 * below to get optimal settings.
 	 */
 	if (action != CPU_DEAD && action != CPU_DEAD_FROZEN &&
 	    action != CPU_ONLINE && action != CPU_ONLINE_FROZEN)
@@ -1536,7 +1983,8 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		return -EINVAL;
 
 
-	set->tags = kmalloc_node(set->nr_hw_queues * sizeof(struct blk_mq_tags),
+	set->tags = kmalloc_node(set->nr_hw_queues *
+				 sizeof(struct blk_mq_tags *),
 				 GFP_KERNEL, set->numa_node);
 	if (!set->tags)
 		goto out;
@@ -1547,6 +1995,9 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 			goto out_unwind;
 	}
 
+	mutex_init(&set->tag_list_lock);
+	INIT_LIST_HEAD(&set->tag_list);
+
 	return 0;
 
 out_unwind:
@@ -1561,11 +2012,37 @@ void blk_mq_free_tag_set(struct blk_mq_tag_set *set)
 {
 	int i;
 
-	for (i = 0; i < set->nr_hw_queues; i++)
-		blk_mq_free_rq_map(set, set->tags[i], i);
+	for (i = 0; i < set->nr_hw_queues; i++) {
+		if (set->tags[i])
+			blk_mq_free_rq_map(set, set->tags[i], i);
+	}
+
+	kfree(set->tags);
 }
 EXPORT_SYMBOL(blk_mq_free_tag_set);
 
+int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr)
+{
+	struct blk_mq_tag_set *set = q->tag_set;
+	struct blk_mq_hw_ctx *hctx;
+	int i, ret;
+
+	if (!set || nr > set->queue_depth)
+		return -EINVAL;
+
+	ret = 0;
+	queue_for_each_hw_ctx(q, hctx, i) {
+		ret = blk_mq_tag_update_depth(hctx->tags, nr);
+		if (ret)
+			break;
+	}
+
+	if (!ret)
+		q->nr_requests = nr;
+
+	return ret;
+}
+
 void blk_mq_disable_hotplug(void)
 {
 	mutex_lock(&all_q_mutex);

block/blk-mq.h

@@ -11,7 +11,8 @@ struct blk_mq_ctx {
 
 	unsigned int		cpu;
 	unsigned int		index_hw;
-	unsigned int		ipi_redirect;
+
+	unsigned int		last_tag ____cacheline_aligned_in_smp;
 
 	/* incremented at dispatch time */
 	unsigned long		rq_dispatched[2];
@@ -22,7 +23,7 @@ struct blk_mq_ctx {
 
 	struct request_queue	*queue;
 	struct kobject		kobj;
-};
+} ____cacheline_aligned_in_smp;
 
 void __blk_mq_complete_request(struct request *rq);
 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
@@ -31,13 +32,14 @@ void blk_mq_drain_queue(struct request_queue *q);
 void blk_mq_free_queue(struct request_queue *q);
 void blk_mq_clone_flush_request(struct request *flush_rq,
 		struct request *orig_rq);
+int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr);
 
 /*
  * CPU hotplug helpers
  */
 struct blk_mq_cpu_notifier;
 void blk_mq_init_cpu_notifier(struct blk_mq_cpu_notifier *notifier,
-			      void (*fn)(void *, unsigned long, unsigned int),
+			      int (*fn)(void *, unsigned long, unsigned int),
 			      void *data);
 void blk_mq_register_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
 void blk_mq_unregister_cpu_notifier(struct blk_mq_cpu_notifier *notifier);
@@ -50,7 +52,15 @@ void blk_mq_disable_hotplug(void);
  */
 extern unsigned int *blk_mq_make_queue_map(struct blk_mq_tag_set *set);
 extern int blk_mq_update_queue_map(unsigned int *map, unsigned int nr_queues);
+extern int blk_mq_hw_queue_to_node(unsigned int *map, unsigned int);
 
-void blk_mq_add_timer(struct request *rq);
+/*
+ * 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;
 
 #endif

block/blk-sysfs.c

@@ -48,11 +48,10 @@ static ssize_t queue_requests_show(struct request_queue *q, char *page)
 static ssize_t
 queue_requests_store(struct request_queue *q, const char *page, size_t count)
 {
-	struct request_list *rl;
 	unsigned long nr;
-	int ret;
+	int ret, err;
 
-	if (!q->request_fn)
+	if (!q->request_fn && !q->mq_ops)
 		return -EINVAL;
 
 	ret = queue_var_store(&nr, page, count);
@@ -62,40 +61,14 @@ queue_requests_store(struct request_queue *q, const char *page, size_t count)
 	if (nr < BLKDEV_MIN_RQ)
 		nr = BLKDEV_MIN_RQ;
 
-	spin_lock_irq(q->queue_lock);
-	q->nr_requests = nr;
-	blk_queue_congestion_threshold(q);
-
-	/* congestion isn't cgroup aware and follows root blkcg for now */
-	rl = &q->root_rl;
-
-	if (rl->count[BLK_RW_SYNC] >= queue_congestion_on_threshold(q))
-		blk_set_queue_congested(q, BLK_RW_SYNC);
-	else if (rl->count[BLK_RW_SYNC] < queue_congestion_off_threshold(q))
-		blk_clear_queue_congested(q, BLK_RW_SYNC);
-
-	if (rl->count[BLK_RW_ASYNC] >= queue_congestion_on_threshold(q))
-		blk_set_queue_congested(q, BLK_RW_ASYNC);
-	else if (rl->count[BLK_RW_ASYNC] < queue_congestion_off_threshold(q))
-		blk_clear_queue_congested(q, BLK_RW_ASYNC);
-
-	blk_queue_for_each_rl(rl, q) {
-		if (rl->count[BLK_RW_SYNC] >= q->nr_requests) {
-			blk_set_rl_full(rl, BLK_RW_SYNC);
-		} else {
-			blk_clear_rl_full(rl, BLK_RW_SYNC);
-			wake_up(&rl->wait[BLK_RW_SYNC]);
-		}
-
-		if (rl->count[BLK_RW_ASYNC] >= q->nr_requests) {
-			blk_set_rl_full(rl, BLK_RW_ASYNC);
-		} else {
-			blk_clear_rl_full(rl, BLK_RW_ASYNC);
-			wake_up(&rl->wait[BLK_RW_ASYNC]);
-		}
-	}
+	if (q->request_fn)
+		err = blk_update_nr_requests(q, nr);
+	else
+		err = blk_mq_update_nr_requests(q, nr);
+
+	if (err)
+		return err;
 
-	spin_unlock_irq(q->queue_lock);
 	return ret;
 }
 
@@ -544,8 +517,6 @@ static void blk_release_queue(struct kobject *kobj)
 	if (q->queue_tags)
 		__blk_queue_free_tags(q);
 
-	percpu_counter_destroy(&q->mq_usage_counter);
-
 	if (q->mq_ops)
 		blk_mq_free_queue(q);
 

block/blk-throttle.c

@@ -744,7 +744,7 @@ static inline void throtl_extend_slice(struct throtl_grp *tg, bool rw,
 static bool throtl_slice_used(struct throtl_grp *tg, bool rw)
 {
 	if (time_in_range(jiffies, tg->slice_start[rw], tg->slice_end[rw]))
-		return 0;
+		return false;
 
 	return 1;
 }
@@ -842,7 +842,7 @@ static bool tg_with_in_iops_limit(struct throtl_grp *tg, struct bio *bio,
 	if (tg->io_disp[rw] + 1 <= io_allowed) {
 		if (wait)
 			*wait = 0;
-		return 1;
+		return true;
 	}
 
 	/* Calc approx time to dispatch */
@@ -880,7 +880,7 @@ static bool tg_with_in_bps_limit(struct throtl_grp *tg, struct bio *bio,
 	if (tg->bytes_disp[rw] + bio->bi_iter.bi_size <= bytes_allowed) {
 		if (wait)
 			*wait = 0;
-		return 1;
+		return true;
 	}
 
 	/* Calc approx time to dispatch */
@@ -923,7 +923,7 @@ static bool tg_may_dispatch(struct throtl_grp *tg, struct bio *bio,
 	if (tg->bps[rw] == -1 && tg->iops[rw] == -1) {
 		if (wait)
 			*wait = 0;
-		return 1;
+		return true;
 	}
 
 	/*
@@ -1258,7 +1258,7 @@ out_unlock:
  * of throtl_data->service_queue.  Those bio's are ready and issued by this
  * function.
  */
-void blk_throtl_dispatch_work_fn(struct work_struct *work)
+static void blk_throtl_dispatch_work_fn(struct work_struct *work)
 {
 	struct throtl_data *td = container_of(work, struct throtl_data,
 					      dispatch_work);

block/blk-timeout.c

@@ -96,11 +96,7 @@ static void blk_rq_timed_out(struct request *req)
 			__blk_complete_request(req);
 		break;
 	case BLK_EH_RESET_TIMER:
-		if (q->mq_ops)
-			blk_mq_add_timer(req);
-		else
-			blk_add_timer(req);
-
+		blk_add_timer(req);
 		blk_clear_rq_complete(req);
 		break;
 	case BLK_EH_NOT_HANDLED:
@@ -170,7 +166,26 @@ void blk_abort_request(struct request *req)
 }
 EXPORT_SYMBOL_GPL(blk_abort_request);
 
-void __blk_add_timer(struct request *req, struct list_head *timeout_list)
+unsigned long blk_rq_timeout(unsigned long timeout)
+{
+	unsigned long maxt;
+
+	maxt = round_jiffies_up(jiffies + BLK_MAX_TIMEOUT);
+	if (time_after(timeout, maxt))
+		timeout = maxt;
+
+	return timeout;
+}
+
+/**
+ * blk_add_timer - Start timeout timer for a single request
+ * @req:	request that is about to start running.
+ *
+ * Notes:
+ *    Each request has its own timer, and as it is added to the queue, we
+ *    set up the timer. When the request completes, we cancel the timer.
+ */
+void blk_add_timer(struct request *req)
 {
 	struct request_queue *q = req->q;
 	unsigned long expiry;
@@ -188,15 +203,15 @@ void __blk_add_timer(struct request *req, struct list_head *timeout_list)
 		req->timeout = q->rq_timeout;
 
 	req->deadline = jiffies + req->timeout;
-	if (timeout_list)
-		list_add_tail(&req->timeout_list, timeout_list);
+	if (!q->mq_ops)
+		list_add_tail(&req->timeout_list, &req->q->timeout_list);
 
 	/*
 	 * If the timer isn't already pending or this timeout is earlier
 	 * than an existing one, modify the timer. Round up to next nearest
 	 * second.
 	 */
-	expiry = round_jiffies_up(req->deadline);
+	expiry = blk_rq_timeout(round_jiffies_up(req->deadline));
 
 	if (!timer_pending(&q->timeout) ||
 	    time_before(expiry, q->timeout.expires)) {
@@ -214,17 +229,3 @@ void __blk_add_timer(struct request *req, struct list_head *timeout_list)
 	}
 
 }
-
-/**
- * blk_add_timer - Start timeout timer for a single request
- * @req:	request that is about to start running.
- *
- * Notes:
- *    Each request has its own timer, and as it is added to the queue, we
- *    set up the timer. When the request completes, we cancel the timer.
- */
-void blk_add_timer(struct request *req)
-{
-	__blk_add_timer(req, &req->q->timeout_list);
-}
-

block/blk.h

@@ -9,6 +9,9 @@
 /* Number of requests a "batching" process may submit */
 #define BLK_BATCH_REQ	32
 
+/* Max future timer expiry for timeouts */
+#define BLK_MAX_TIMEOUT		(5 * HZ)
+
 extern struct kmem_cache *blk_requestq_cachep;
 extern struct kmem_cache *request_cachep;
 extern struct kobj_type blk_queue_ktype;
@@ -37,9 +40,9 @@ bool __blk_end_bidi_request(struct request *rq, int error,
 void blk_rq_timed_out_timer(unsigned long data);
 void blk_rq_check_expired(struct request *rq, unsigned long *next_timeout,
 			  unsigned int *next_set);
-void __blk_add_timer(struct request *req, struct list_head *timeout_list);
+unsigned long blk_rq_timeout(unsigned long timeout);
+void blk_add_timer(struct request *req);
 void blk_delete_timer(struct request *);
-void blk_add_timer(struct request *);
 
 
 bool bio_attempt_front_merge(struct request_queue *q, struct request *req,
@@ -185,6 +188,8 @@ static inline int queue_congestion_off_threshold(struct request_queue *q)
 	return q->nr_congestion_off;
 }
 
+extern int blk_update_nr_requests(struct request_queue *, unsigned int);
+
 /*
  * Contribute to IO statistics IFF:
  *

block/cfq-iosched.c

@@ -4460,7 +4460,7 @@ out_free:
 static ssize_t
 cfq_var_show(unsigned int var, char *page)
 {
-	return sprintf(page, "%d\n", var);
+	return sprintf(page, "%u\n", var);
 }
 
 static ssize_t

drivers/block/mtip32xx/mtip32xx.c

@@ -178,7 +178,7 @@ static struct mtip_cmd *mtip_get_int_command(struct driver_data *dd)
 {
 	struct request *rq;
 
-	rq = blk_mq_alloc_reserved_request(dd->queue, 0, __GFP_WAIT);
+	rq = blk_mq_alloc_request(dd->queue, 0, __GFP_WAIT, true);
 	return blk_mq_rq_to_pdu(rq);
 }
 

drivers/block/null_blk.c

@@ -322,39 +322,10 @@ static int null_queue_rq(struct blk_mq_hw_ctx *hctx, struct request *rq)
 }
 
 static struct blk_mq_hw_ctx *null_alloc_hctx(struct blk_mq_tag_set *set,
-		unsigned int hctx_index)
+					     unsigned int hctx_index,
+					     int node)
 {
-	int b_size = DIV_ROUND_UP(set->nr_hw_queues, nr_online_nodes);
-	int tip = (set->nr_hw_queues % nr_online_nodes);
-	int node = 0, i, n;
-
-	/*
-	 * Split submit queues evenly wrt to the number of nodes. If uneven,
-	 * fill the first buckets with one extra, until the rest is filled with
-	 * no extra.
-	 */
-	for (i = 0, n = 1; i < hctx_index; i++, n++) {
-		if (n % b_size == 0) {
-			n = 0;
-			node++;
-
-			tip--;
-			if (!tip)
-				b_size = set->nr_hw_queues / nr_online_nodes;
-		}
-	}
-
-	/*
-	 * A node might not be online, therefore map the relative node id to the
-	 * real node id.
-	 */
-	for_each_online_node(n) {
-		if (!node)
-			break;
-		node--;
-	}
-
-	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, n);
+	return kzalloc_node(sizeof(struct blk_mq_hw_ctx), GFP_KERNEL, node);
 }
 
 static void null_free_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_index)

drivers/block/skd_main.c

@@ -743,6 +743,7 @@ static void skd_request_fn(struct request_queue *q)
 				break;
 			}
 			skreq->discard_page = 1;
+			req->completion_data = page;
 			skd_prep_discard_cdb(scsi_req, skreq, page, lba, count);
 
 		} else if (flush == SKD_FLUSH_ZERO_SIZE_FIRST) {
@@ -855,10 +856,9 @@ static void skd_end_request(struct skd_device *skdev,
 
 	if ((io_flags & REQ_DISCARD) &&
 		(skreq->discard_page == 1)) {
-		struct bio *bio = req->bio;
 		pr_debug("%s:%s:%d, free the page!",
 			 skdev->name, __func__, __LINE__);
-		__free_page(bio->bi_io_vec->bv_page);
+		__free_page(req->completion_data);
 	}
 
 	if (unlikely(error)) {

drivers/char/random.c

@@ -902,6 +902,7 @@ void add_disk_randomness(struct gendisk *disk)
 	add_timer_randomness(disk->random, 0x100 + disk_devt(disk));
 	trace_add_disk_randomness(disk_devt(disk), ENTROPY_BITS(&input_pool));
 }
+EXPORT_SYMBOL_GPL(add_disk_randomness);
 #endif
 
 /*********************************************************************

drivers/scsi/sd.c

@@ -737,6 +737,7 @@ static int sd_setup_discard_cmnd(struct scsi_device *sdp, struct request *rq)
 		goto out;
 	}
 
+	rq->completion_data = page;
 	blk_add_request_payload(rq, page, len);
 	ret = scsi_setup_blk_pc_cmnd(sdp, rq);
 	rq->__data_len = nr_bytes;
@@ -839,11 +840,9 @@ static void sd_unprep_fn(struct request_queue *q, struct request *rq)
 {
 	struct scsi_cmnd *SCpnt = rq->special;
 
-	if (rq->cmd_flags & REQ_DISCARD) {
-		struct bio *bio = rq->bio;
+	if (rq->cmd_flags & REQ_DISCARD)
+		__free_page(rq->completion_data);
 
-		__free_page(bio->bi_io_vec->bv_page);
-	}
 	if (SCpnt->cmnd != rq->cmd) {
 		mempool_free(SCpnt->cmnd, sd_cdb_pool);
 		SCpnt->cmnd = NULL;

fs/Makefile

@@ -14,14 +14,13 @@ obj-y :=	open.o read_write.o file_table.o super.o \
 		stack.o fs_struct.o statfs.o
 
 ifeq ($(CONFIG_BLOCK),y)
-obj-y +=	buffer.o bio.o block_dev.o direct-io.o mpage.o ioprio.o
+obj-y +=	buffer.o block_dev.o direct-io.o mpage.o
 else
 obj-y +=	no-block.o
 endif
 
 obj-$(CONFIG_PROC_FS) += proc_namespace.o
 
-obj-$(CONFIG_BLK_DEV_INTEGRITY) += bio-integrity.o
 obj-y				+= notify/
 obj-$(CONFIG_EPOLL)		+= eventpoll.o
 obj-$(CONFIG_ANON_INODES)	+= anon_inodes.o

include/linux/bio.h

@@ -333,7 +333,7 @@ static inline struct bio *bio_next_split(struct bio *bio, int sectors,
 
 extern struct bio_set *bioset_create(unsigned int, unsigned int);
 extern void bioset_free(struct bio_set *);
-extern mempool_t *biovec_create_pool(struct bio_set *bs, int pool_entries);
+extern mempool_t *biovec_create_pool(int pool_entries);
 
 extern struct bio *bio_alloc_bioset(gfp_t, int, struct bio_set *);
 extern void bio_put(struct bio *);

include/linux/blk-mq.h

@@ -8,7 +8,13 @@ struct blk_mq_tags;
 struct blk_mq_cpu_notifier {
 	struct list_head list;
 	void *data;
-	void (*notify)(void *data, unsigned long action, unsigned int cpu);
+	int (*notify)(void *data, unsigned long action, unsigned int cpu);
+};
+
+struct blk_mq_ctxmap {
+	unsigned int map_size;
+	unsigned int bits_per_word;
+	struct blk_align_bitmap *map;
 };
 
 struct blk_mq_hw_ctx {
@@ -21,6 +27,8 @@ struct blk_mq_hw_ctx {
 	struct delayed_work	run_work;
 	struct delayed_work	delay_work;
 	cpumask_var_t		cpumask;
+	int			next_cpu;
+	int			next_cpu_batch;
 
 	unsigned long		flags;		/* BLK_MQ_F_* flags */
 
@@ -29,10 +37,12 @@ struct blk_mq_hw_ctx {
 
 	void			*driver_data;
 
+	struct blk_mq_ctxmap	ctx_map;
+
 	unsigned int		nr_ctx;
 	struct blk_mq_ctx	**ctxs;
-	unsigned int 		nr_ctx_map;
-	unsigned long		*ctx_map;
+
+	unsigned int		wait_index;
 
 	struct blk_mq_tags	*tags;
 
@@ -44,6 +54,8 @@ struct blk_mq_hw_ctx {
 	unsigned int		numa_node;
 	unsigned int		cmd_size;	/* per-request extra data */
 
+	atomic_t		nr_active;
+
 	struct blk_mq_cpu_notifier	cpu_notifier;
 	struct kobject		kobj;
 };
@@ -51,7 +63,7 @@ struct blk_mq_hw_ctx {
 struct blk_mq_tag_set {
 	struct blk_mq_ops	*ops;
 	unsigned int		nr_hw_queues;
-	unsigned int		queue_depth;
+	unsigned int		queue_depth;	/* max hw supported */
 	unsigned int		reserved_tags;
 	unsigned int		cmd_size;	/* per-request extra data */
 	int			numa_node;
@@ -60,12 +72,15 @@ struct blk_mq_tag_set {
 	void			*driver_data;
 
 	struct blk_mq_tags	**tags;
+
+	struct mutex		tag_list_lock;
+	struct list_head	tag_list;
 };
 
 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, struct request *);
 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
 typedef struct blk_mq_hw_ctx *(alloc_hctx_fn)(struct blk_mq_tag_set *,
-		unsigned int);
+		unsigned int, int);
 typedef void (free_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
@@ -122,11 +137,14 @@ enum {
 
 	BLK_MQ_F_SHOULD_MERGE	= 1 << 0,
 	BLK_MQ_F_SHOULD_SORT	= 1 << 1,
-	BLK_MQ_F_SHOULD_IPI	= 1 << 2,
+	BLK_MQ_F_TAG_SHARED	= 1 << 2,
 
 	BLK_MQ_S_STOPPED	= 0,
+	BLK_MQ_S_TAG_ACTIVE	= 1,
 
 	BLK_MQ_MAX_DEPTH	= 2048,
+
+	BLK_MQ_CPU_WORK_BATCH	= 8,
 };
 
 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
@@ -142,19 +160,20 @@ void blk_mq_insert_request(struct request *, bool, bool, bool);
 void blk_mq_run_queues(struct request_queue *q, bool async);
 void blk_mq_free_request(struct request *rq);
 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
-struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp);
-struct request *blk_mq_alloc_reserved_request(struct request_queue *q, int rw, gfp_t gfp);
+struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
+		gfp_t gfp, bool reserved);
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 
 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
-struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int);
+struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
 void blk_mq_free_single_hw_queue(struct blk_mq_hw_ctx *, unsigned int);
 
 void blk_mq_end_io(struct request *rq, int error);
 void __blk_mq_end_io(struct request *rq, int error);
 
 void blk_mq_requeue_request(struct request *rq);
-
+void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
+void blk_mq_kick_requeue_list(struct request_queue *q);
 void blk_mq_complete_request(struct request *rq);
 
 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
@@ -163,6 +182,7 @@ void blk_mq_stop_hw_queues(struct request_queue *q);
 void blk_mq_start_hw_queues(struct request_queue *q);
 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
+void blk_mq_tag_busy_iter(struct blk_mq_tags *tags, void (*fn)(void *data, unsigned long *), void *data);
 
 /*
  * Driver command data is immediately after the request. So subtract request

include/linux/blk_types.h

@@ -190,6 +190,7 @@ enum rq_flag_bits {
 	__REQ_PM,		/* runtime pm request */
 	__REQ_END,		/* last of chain of requests */
 	__REQ_HASHED,		/* on IO scheduler merge hash */
+	__REQ_MQ_INFLIGHT,	/* track inflight for MQ */
 	__REQ_NR_BITS,		/* stops here */
 };
 
@@ -243,5 +244,6 @@ enum rq_flag_bits {
 #define REQ_PM			(1ULL << __REQ_PM)
 #define REQ_END			(1ULL << __REQ_END)
 #define REQ_HASHED		(1ULL << __REQ_HASHED)
+#define REQ_MQ_INFLIGHT		(1ULL << __REQ_MQ_INFLIGHT)
 
 #endif /* __LINUX_BLK_TYPES_H */

include/linux/blkdev.h

@@ -90,15 +90,15 @@ enum rq_cmd_type_bits {
 #define BLK_MAX_CDB	16
 
 /*
- * try to put the fields that are referenced together in the same cacheline.
- * if you modify this structure, be sure to check block/blk-core.c:blk_rq_init()
- * as well!
+ * Try to put the fields that are referenced together in the same cacheline.
+ *
+ * If you modify this structure, make sure to update blk_rq_init() and
+ * especially blk_mq_rq_ctx_init() to take care of the added fields.
  */
 struct request {
 	struct list_head queuelist;
 	union {
 		struct call_single_data csd;
-		struct work_struct requeue_work;
 		unsigned long fifo_time;
 	};
 
@@ -462,6 +462,10 @@ struct request_queue {
 	struct request		*flush_rq;
 	spinlock_t		mq_flush_lock;
 
+	struct list_head	requeue_list;
+	spinlock_t		requeue_lock;
+	struct work_struct	requeue_work;
+
 	struct mutex		sysfs_lock;
 
 	int			bypass_depth;
@@ -480,6 +484,9 @@ struct request_queue {
 	wait_queue_head_t	mq_freeze_wq;
 	struct percpu_counter	mq_usage_counter;
 	struct list_head	all_q_node;
+
+	struct blk_mq_tag_set	*tag_set;
+	struct list_head	tag_set_list;
 };
 
 #define QUEUE_FLAG_QUEUED	1	/* uses generic tag queueing */

mm/Makefile

@@ -30,7 +30,6 @@ endif
 
 obj-$(CONFIG_HAVE_MEMBLOCK) += memblock.o
 
-obj-$(CONFIG_BOUNCE)	+= bounce.o
 obj-$(CONFIG_SWAP)	+= page_io.o swap_state.o swapfile.o
 obj-$(CONFIG_FRONTSWAP)	+= frontswap.o
 obj-$(CONFIG_ZSWAP)	+= zswap.o