blk-mq.c

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/backing-dev.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
#include <linux/mm.h>
#include <linux/init.h>
#include <linux/slab.h>
#include <linux/workqueue.h>
#include <linux/smp.h>
#include <linux/llist.h>
#include <linux/list_sort.h>
#include <linux/cpu.h>
#include <linux/cache.h>
#include <linux/sched/sysctl.h>
#include <linux/delay.h>

#include <trace/events/block.h>

#include <linux/blk-mq.h>
#include "blk.h"
#include "blk-mq.h"
#include "blk-mq-tag.h"

static DEFINE_MUTEX(all_q_mutex);
static LIST_HEAD(all_q_list);

static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx);

static struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
					   unsigned int cpu)
{
	return per_cpu_ptr(q->queue_ctx, cpu);
}

/*
 * This assumes per-cpu software queueing queues. They could be per-node
 * as well, for instance. For now this is hardcoded as-is. Note that we don't
 * care about preemption, since we know the ctx's are persistent. This does
 * mean that we can't rely on ctx always matching the currently running CPU.
 */
static struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
{
	return __blk_mq_get_ctx(q, get_cpu());
}

static void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
{
	put_cpu();
}

/*
 * Check if any of the ctx's have pending work in this hardware queue
 */
static bool blk_mq_hctx_has_pending(struct blk_mq_hw_ctx *hctx)
{
	unsigned int 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)
{
	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 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);
}

static struct request *__blk_mq_alloc_request(struct blk_mq_hw_ctx *hctx,
					      struct blk_mq_ctx *ctx,
					      gfp_t gfp, bool reserved)
{
	struct request *rq;
	unsigned int tag;

	tag = blk_mq_get_tag(hctx, &ctx->last_tag, gfp, reserved);
	if (tag != BLK_MQ_TAG_FAIL) {
		rq = hctx->tags->rqs[tag];

		rq->cmd_flags = 0;
		if (blk_mq_tag_busy(hctx)) {
			rq->cmd_flags = REQ_MQ_INFLIGHT;
			atomic_inc(&hctx->nr_active);
		}

		rq->tag = tag;
		return rq;
	}

	return NULL;
}

static int blk_mq_queue_enter(struct request_queue *q)
{
	int ret;

	__percpu_counter_add(&q->mq_usage_counter, 1, 1000000);
	smp_wmb();
	/* we have problems to freeze the queue if it's initializing */
	if (!blk_queue_bypass(q) || !blk_queue_init_done(q))
		return 0;

	__percpu_counter_add(&q->mq_usage_counter, -1, 1000000);

	spin_lock_irq(q->queue_lock);
	ret = wait_event_interruptible_lock_irq(q->mq_freeze_wq,
		!blk_queue_bypass(q) || blk_queue_dying(q),
		*q->queue_lock);
	/* inc usage with lock hold to avoid freeze_queue runs here */
	if (!ret && !blk_queue_dying(q))
		__percpu_counter_add(&q->mq_usage_counter, 1, 1000000);
	else if (blk_queue_dying(q))
		ret = -ENODEV;
	spin_unlock_irq(q->queue_lock);

	return ret;
}

static void blk_mq_queue_exit(struct request_queue *q)
{
	__percpu_counter_add(&q->mq_usage_counter, -1, 1000000);
}

static void __blk_mq_drain_queue(struct request_queue *q)
{
	while (true) {
		s64 count;

		spin_lock_irq(q->queue_lock);
		count = percpu_counter_sum(&q->mq_usage_counter);
		spin_unlock_irq(q->queue_lock);

		if (count == 0)
			break;
		blk_mq_run_queues(q, false);
		msleep(10);
	}
}

/*
 * Guarantee no request is in use, so we can change any data structure of
 * the queue afterward.
 */
static void blk_mq_freeze_queue(struct request_queue *q)
{
	bool drain;

	spin_lock_irq(q->queue_lock);
	drain = !q->bypass_depth++;
	queue_flag_set(QUEUE_FLAG_BYPASS, q);
	spin_unlock_irq(q->queue_lock);

	if (drain)
		__blk_mq_drain_queue(q);
}

void blk_mq_drain_queue(struct request_queue *q)
{
	__blk_mq_drain_queue(q);
}

static void blk_mq_unfreeze_queue(struct request_queue *q)
{
	bool wake = false;

	spin_lock_irq(q->queue_lock);
	if (!--q->bypass_depth) {
		queue_flag_clear(QUEUE_FLAG_BYPASS, q);
		wake = true;
	}
	WARN_ON_ONCE(q->bypass_depth < 0);
	spin_unlock_irq(q->queue_lock);
	if (wake)
		wake_up_all(&q->mq_freeze_wq);
}

bool blk_mq_can_queue(struct blk_mq_hw_ctx *hctx)
{
	return blk_mq_has_free_tags(hctx->tags);
}
EXPORT_SYMBOL(blk_mq_can_queue);

static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
			       struct request *rq, unsigned int rw_flags)
{
	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_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)
{
	struct request *rq;

	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, ctx, gfp & ~__GFP_WAIT,
						reserved);
		if (rq) {
			blk_mq_rq_ctx_init(q, ctx, rq, rw);
			break;
		}

		if (gfp & __GFP_WAIT) {
			__blk_mq_run_hw_queue(hctx);
			blk_mq_put_ctx(ctx);
		} else {
			blk_mq_put_ctx(ctx);
			break;
		}

		blk_mq_wait_for_tags(hctx, reserved);
	} while (1);

	return rq;
}

struct request *blk_mq_alloc_request(struct request_queue *q, int rw, gfp_t gfp,
		bool reserved)
{
	struct request *rq;

	if (blk_mq_queue_enter(q))
		return NULL;

	rq = blk_mq_alloc_request_pinned(q, rw, gfp, reserved);
	if (rq)
		blk_mq_put_ctx(rq->mq_ctx);
	return rq;
}
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)
{
	const int tag = rq->tag;
	struct request_queue *q = rq->q;

	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);
}

void blk_mq_free_request(struct request *rq)
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;
	struct blk_mq_hw_ctx *hctx;
	struct request_queue *q = rq->q;

	ctx->rq_completed[rq_is_sync(rq)]++;

	hctx = q->mq_ops->map_queue(q, ctx->cpu);
	__blk_mq_free_request(hctx, ctx, rq);
}

/*
 * Clone all relevant state from a request that has been put on hold in
 * the flush state machine into the preallocated flush request that hangs
 * off the request queue.
 *
 * For a driver the flush request should be invisible, that's why we are
 * impersonating the original request here.
 */
void blk_mq_clone_flush_request(struct request *flush_rq,
		struct request *orig_rq)
{
	struct blk_mq_hw_ctx *hctx =
		orig_rq->q->mq_ops->map_queue(orig_rq->q, orig_rq->mq_ctx->cpu);

	flush_rq->mq_ctx = orig_rq->mq_ctx;
	flush_rq->tag = orig_rq->tag;
	memcpy(blk_mq_rq_to_pdu(flush_rq), blk_mq_rq_to_pdu(orig_rq),
		hctx->cmd_size);
}

inline void __blk_mq_end_io(struct request *rq, int error)
{
	blk_account_io_done(rq);

	if (rq->end_io) {
		rq->end_io(rq, error);
	} else {
		if (unlikely(blk_bidi_rq(rq)))
			blk_mq_free_request(rq->next_rq);
		blk_mq_free_request(rq);
	}
}
EXPORT_SYMBOL(__blk_mq_end_io);

void blk_mq_end_io(struct request *rq, int error)
{
	if (blk_update_request(rq, error, blk_rq_bytes(rq)))
		BUG();
	__blk_mq_end_io(rq, error);
}
EXPORT_SYMBOL(blk_mq_end_io);

static void __blk_mq_complete_request_remote(void *data)
{
	struct request *rq = data;

	rq->q->softirq_done_fn(rq);
}

void __blk_mq_complete_request(struct request *rq)
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;
	bool shared = false;
	int cpu;

	if (!test_bit(QUEUE_FLAG_SAME_COMP, &rq->q->queue_flags)) {
		rq->q->softirq_done_fn(rq);
		return;
	}

	cpu = get_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;
		smp_call_function_single_async(ctx->cpu, &rq->csd);
	} else {
		rq->q->softirq_done_fn(rq);
	}
	put_cpu();
}

/**
 * blk_mq_complete_request - end I/O on a request
 * @rq:		the request being processed
 *
 * Description:
 *	Ends all I/O on a request. It does not handle partial completions.
 *	The actual completion happens out-of-order, through a IPI handler.
 **/
void blk_mq_complete_request(struct request *rq)
{
	struct request_queue *q = rq->q;

	if (unlikely(blk_should_fake_timeout(q)))
		return;
	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);

static void blk_mq_start_request(struct request *rq, bool last)
{
	struct request_queue *q = rq->q;

	trace_block_rq_issue(q, rq);

	rq->resid_len = blk_rq_bytes(rq);
	if (unlikely(blk_bidi_rq(rq)))
		rq->next_rq->resid_len = blk_rq_bytes(rq->next_rq);

	/*
	 * 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. 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.
	 */
	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)) {
		/*
		 * Make sure space for the drain appears.  We know we can do
		 * this because max_hw_segments has been adjusted to be one
		 * fewer than the device can handle.
		 */
		rq->nr_phys_segments++;
	}

	/*
	 * Flag the last request in the series so that drivers know when IO
	 * should be kicked off, if they don't do it on a per-request basis.
	 *
	 * Note: the flag isn't the only condition drivers should do kick off.
	 * If drive is busy, the last request might not have the bit set.
	 */
	if (last)
		rq->cmd_flags |= REQ_END;
}

static void __blk_mq_requeue_request(struct request *rq)
{
	struct request_queue *q = rq->q;

	trace_block_rq_requeue(q, rq);
	clear_bit(REQ_ATOM_STARTED, &rq->atomic_flags);

	rq->cmd_flags &= ~REQ_END;

	if (q->dma_drain_size && blk_rq_bytes(rq))
		rq->nr_phys_segments--;
}

void blk_mq_requeue_request(struct request *rq)
{
	__blk_mq_requeue_request(rq);
	blk_clear_rq_complete(rq);

	BUG_ON(blk_queued_rq(rq));
	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];
}
EXPORT_SYMBOL(blk_mq_tag_to_rq);

struct blk_mq_timeout_data {
	struct blk_mq_hw_ctx *hctx;
	unsigned long *next;
	unsigned int *next_set;
};

static void blk_mq_timeout_check(void *__data, unsigned long *free_tags)
{
	struct blk_mq_timeout_data *data = __data;
	struct blk_mq_hw_ctx *hctx = data->hctx;
	unsigned int tag;

	 /* It may not be in flight yet (this is where
	 * the REQ_ATOMIC_STARTED flag comes in). The requests are
	 * statically allocated, so we know it's always safe to access the
	 * memory associated with a bit offset into ->rqs[].
	 */
	tag = 0;
	do {
		struct request *rq;

		tag = find_next_zero_bit(free_tags, hctx->tags->nr_tags, tag);
		if (tag >= hctx->tags->nr_tags)
			break;

		rq = blk_mq_tag_to_rq(hctx->tags, tag++);
		if (rq->q != hctx->queue)
			continue;
		if (!test_bit(REQ_ATOM_STARTED, &rq->atomic_flags))
			continue;

		blk_rq_check_expired(rq, data->next, data->next_set);
	} while (1);
}

static void blk_mq_hw_ctx_check_timeout(struct blk_mq_hw_ctx *hctx,
					unsigned long *next,
					unsigned int *next_set)
{
	struct blk_mq_timeout_data data = {
		.hctx		= hctx,
		.next		= next,
		.next_set	= next_set,
	};

	/*
	 * Ask the tagging code to iterate busy requests, so we can
	 * check them for timeout.
	 */
	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;
	struct blk_mq_hw_ctx *hctx;
	unsigned long next = 0;
	int i, next_set = 0;

	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) {
		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);
	}
}

/*
 * Reverse check our software queue for entries that we could potentially
 * merge with. Currently includes a hand-wavy stop count of 8, to not spend
 * too much time checking for merges.
 */
static bool blk_mq_attempt_merge(struct request_queue *q,
				 struct blk_mq_ctx *ctx, struct bio *bio)
{
	struct request *rq;
	int checked = 8;

	list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) {
		int el_ret;

		if (!checked--)
			break;

		if (!blk_rq_merge_ok(rq, bio))
			continue;

		el_ret = blk_try_merge(rq, bio);
		if (el_ret == ELEVATOR_BACK_MERGE) {
			if (bio_attempt_back_merge(q, rq, bio)) {
				ctx->rq_merged++;
				return true;
			}
			break;
		} else if (el_ret == ELEVATOR_FRONT_MERGE) {
			if (bio_attempt_front_merge(q, rq, bio)) {
				ctx->rq_merged++;
				return true;
			}
			break;
		}
	}

	return false;
}

/*
 * 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.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);
	}
}

/*
 * Run this hardware queue, pulling any software queues mapped to it in.
 * Note that this function currently has various problems around ordering
 * of IO. In particular, we'd like FIFO behaviour on handling existing
 * items on the hctx->dispatch list. Ignore that for now.
 */
static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
{
	struct request_queue *q = hctx->queue;
	struct request *rq;
	LIST_HEAD(rq_list);
	int queued;

	WARN_ON(!cpumask_test_cpu(raw_smp_processor_id(), hctx->cpumask));

	if (unlikely(test_bit(BLK_MQ_S_STOPPED, &hctx->state)))
		return;

	hctx->run++;

	/*
	 * Touch any software queue that has pending entries.
	 */
	flush_busy_ctxs(hctx, &rq_list);

	/*
	 * If we have previous entries on our dispatch list, grab them
	 * and stuff them at the front for more fair dispatch.
	 */
	if (!list_empty_careful(&hctx->dispatch)) {
		spin_lock(&hctx->lock);
		if (!list_empty(&hctx->dispatch))
			list_splice_init(&hctx->dispatch, &rq_list);
		spin_unlock(&hctx->lock);
	}

	/*
	 * Now process all the entries, sending them to the driver.
	 */
	queued = 0;
	while (!list_empty(&rq_list)) {
		int ret;

		rq = list_first_entry(&rq_list, struct request, queuelist);
		list_del_init(&rq->queuelist);

		blk_mq_start_request(rq, list_empty(&rq_list));

		ret = q->mq_ops->queue_rq(hctx, rq);
		switch (ret) {
		case BLK_MQ_RQ_QUEUE_OK:
			queued++;
			continue;
		case BLK_MQ_RQ_QUEUE_BUSY:
			list_add(&rq->queuelist, &rq_list);
			__blk_mq_requeue_request(rq);
			break;
		default:
			pr_err("blk-mq: bad return on queue: %d\n", ret);
		case BLK_MQ_RQ_QUEUE_ERROR:
			rq->errors = -EIO;
			blk_mq_end_io(rq, rq->errors);
			break;
		}

		if (ret == BLK_MQ_RQ_QUEUE_BUSY)
			break;
	}

	if (!queued)
		hctx->dispatched[0]++;
	else if (queued < (1 << (BLK_MQ_MAX_DISPATCH_ORDER - 1)))
		hctx->dispatched[ilog2(queued) + 1]++;

	/*
	 * Any items that need requeuing? Stuff them into hctx->dispatch,
	 * that is where we will continue on next queue run.
	 */
	if (!list_empty(&rq_list)) {
		spin_lock(&hctx->lock);
		list_splice(&rq_list, &hctx->dispatch);
		spin_unlock(&hctx->lock);
	}
}

/*
 * 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)))
		return;

	if (!async && cpumask_test_cpu(smp_processor_id(), hctx->cpumask))
		__blk_mq_run_hw_queue(hctx);
	else if (hctx->queue->nr_hw_queues == 1)
		kblockd_schedule_delayed_work(&hctx->run_work, 0);
	else {
		unsigned int cpu;

		cpu = blk_mq_hctx_next_cpu(hctx);
		kblockd_schedule_delayed_work_on(cpu, &hctx->run_work, 0);
	}
}

void blk_mq_run_queues(struct request_queue *q, bool async)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if ((!blk_mq_hctx_has_pending(hctx) &&
		    list_empty_careful(&hctx->dispatch)) ||
		    test_bit(BLK_MQ_S_STOPPED, &hctx->state))
			continue;

		preempt_disable();
		blk_mq_run_hw_queue(hctx, async);
		preempt_enable();
	}
}
EXPORT_SYMBOL(blk_mq_run_queues);

void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx)
{
	cancel_delayed_work(&hctx->run_work);
	cancel_delayed_work(&hctx->delay_work);
	set_bit(BLK_MQ_S_STOPPED, &hctx->state);
}
EXPORT_SYMBOL(blk_mq_stop_hw_queue);

void blk_mq_stop_hw_queues(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i)
		blk_mq_stop_hw_queue(hctx);
}
EXPORT_SYMBOL(blk_mq_stop_hw_queues);

void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx)
{
	clear_bit(BLK_MQ_S_STOPPED, &hctx->state);

	preempt_disable();
	__blk_mq_run_hw_queue(hctx);
	preempt_enable();
}
EXPORT_SYMBOL(blk_mq_start_hw_queue);

void blk_mq_start_hw_queues(struct request_queue *q)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i)
		blk_mq_start_hw_queue(hctx);
}
EXPORT_SYMBOL(blk_mq_start_hw_queues);


void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async)
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if (!test_bit(BLK_MQ_S_STOPPED, &hctx->state))
			continue;

		clear_bit(BLK_MQ_S_STOPPED, &hctx->state);
		preempt_disable();
		blk_mq_run_hw_queue(hctx, async);
		preempt_enable();
	}
}
EXPORT_SYMBOL(blk_mq_start_stopped_hw_queues);

static void blk_mq_run_work_fn(struct work_struct *work)
{
	struct blk_mq_hw_ctx *hctx;

	hctx = container_of(work, struct blk_mq_hw_ctx, run_work.work);

	__blk_mq_run_hw_queue(hctx);
}

static void blk_mq_delay_work_fn(struct work_struct *work)
{
	struct blk_mq_hw_ctx *hctx;

	hctx = container_of(work, struct blk_mq_hw_ctx, delay_work.work);

	if (test_and_clear_bit(BLK_MQ_S_STOPPED, &hctx->state))
		__blk_mq_run_hw_queue(hctx);
}

void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs)
{
	unsigned long tmo = msecs_to_jiffies(msecs);

	if (hctx->queue->nr_hw_queues == 1)
		kblockd_schedule_delayed_work(&hctx->delay_work, tmo);
	else {
		unsigned int cpu;

		cpu = blk_mq_hctx_next_cpu(hctx);
		kblockd_schedule_delayed_work_on(cpu, &hctx->delay_work, tmo);
	}
}
EXPORT_SYMBOL(blk_mq_delay_queue);

static void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx,
				    struct request *rq, bool at_head)
{
	struct blk_mq_ctx *ctx = rq->mq_ctx;

	trace_block_rq_insert(hctx->queue, rq);

	if (at_head)
		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_add_timer(rq);
}