[mirror_ubuntu-bionic-kernel.git] / block / blk-mq-sched.c

/*
 * blk-mq scheduling framework
 *
 * Copyright (C) 2016 Jens Axboe
 */
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/blk-mq.h>

#include <trace/events/block.h>

#include "blk.h"
#include "blk-mq.h"
#include "blk-mq-debugfs.h"
#include "blk-mq-sched.h"
#include "blk-mq-tag.h"
#include "blk-wbt.h"

void blk_mq_sched_free_hctx_data(struct request_queue *q,
				 void (*exit)(struct blk_mq_hw_ctx *))
{
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		if (exit && hctx->sched_data)
			exit(hctx);
		kfree(hctx->sched_data);
		hctx->sched_data = NULL;
	}
}
EXPORT_SYMBOL_GPL(blk_mq_sched_free_hctx_data);

void blk_mq_sched_assign_ioc(struct request *rq, struct bio *bio)
{
	struct request_queue *q = rq->q;
	struct io_context *ioc = rq_ioc(bio);
	struct io_cq *icq;

	spin_lock_irq(q->queue_lock);
	icq = ioc_lookup_icq(ioc, q);
	spin_unlock_irq(q->queue_lock);

	if (!icq) {
		icq = ioc_create_icq(ioc, q, GFP_ATOMIC);
		if (!icq)
			return;
	}
	get_io_context(icq->ioc);
	rq->elv.icq = icq;
}

/*
 * Mark a hardware queue as needing a restart. For shared queues, maintain
 * a count of how many hardware queues are marked for restart.
 */
static void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx)
{
	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
		return;

	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
		struct request_queue *q = hctx->queue;

		if (!test_and_set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
			atomic_inc(&q->shared_hctx_restart);
	} else
		set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
}

static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
{
	if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
		return false;

	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
		struct request_queue *q = hctx->queue;

		if (test_and_clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
			atomic_dec(&q->shared_hctx_restart);
	} else
		clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);

	if (blk_mq_hctx_has_pending(hctx)) {
		blk_mq_run_hw_queue(hctx, true);
		return true;
	}

	return false;
}

/* return true if hctx need to run again */
static bool blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
{
	struct request_queue *q = hctx->queue;
	struct elevator_queue *e = q->elevator;
	LIST_HEAD(rq_list);

	do {
		struct request *rq;
		blk_status_t ret;

		if (e->type->ops.mq.has_work &&
				!e->type->ops.mq.has_work(hctx))
			break;

		ret = blk_mq_get_dispatch_budget(hctx);
		if (ret == BLK_STS_RESOURCE)
			return true;

		rq = e->type->ops.mq.dispatch_request(hctx);
		if (!rq) {
			blk_mq_put_dispatch_budget(hctx);
			break;
		} else if (ret != BLK_STS_OK) {
			blk_mq_end_request(rq, ret);
			continue;
		}

		/*
		 * Now this rq owns the budget which has to be released
		 * if this rq won't be queued to driver via .queue_rq()
		 * in blk_mq_dispatch_rq_list().
		 */
		list_add(&rq->queuelist, &rq_list);
	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));

	return false;
}

/* return true if hw queue need to be run again */
bool blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
{
	struct request_queue *q = hctx->queue;
	struct elevator_queue *e = q->elevator;
	const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request;
	LIST_HEAD(rq_list);
	bool run_queue = false;

	/* RCU or SRCU read lock is needed before checking quiesced flag */
	if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q)))
		return false;

	hctx->run++;

	/*
	 * If we have previous entries on our dispatch list, grab them first 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);
	}

	/*
	 * Only ask the scheduler for requests, if we didn't have residual
	 * requests from the dispatch list. This is to avoid the case where
	 * we only ever dispatch a fraction of the requests available because
	 * of low device queue depth. Once we pull requests out of the IO
	 * scheduler, we can no longer merge or sort them. So it's best to
	 * leave them there for as long as we can. Mark the hw queue as
	 * needing a restart in that case.
	 *
	 * We want to dispatch from the scheduler if there was nothing
	 * on the dispatch list or we were able to dispatch from the
	 * dispatch list.
	 */
	if (!list_empty(&rq_list)) {
		blk_mq_sched_mark_restart_hctx(hctx);
		if (blk_mq_dispatch_rq_list(q, &rq_list, false) &&
				has_sched_dispatch)
			run_queue = blk_mq_do_dispatch_sched(hctx);
	} else if (has_sched_dispatch) {
		run_queue = blk_mq_do_dispatch_sched(hctx);
	} else {
		blk_mq_flush_busy_ctxs(hctx, &rq_list);
		blk_mq_dispatch_rq_list(q, &rq_list, false);
	}

	if (run_queue && !blk_mq_sched_needs_restart(hctx) &&
			!test_bit(BLK_MQ_S_TAG_WAITING, &hctx->state)) {
		blk_mq_sched_mark_restart_hctx(hctx);
		return true;
	}

	return false;
}

bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
			    struct request **merged_request)
{
	struct request *rq;

	switch (elv_merge(q, &rq, bio)) {
	case ELEVATOR_BACK_MERGE:
		if (!blk_mq_sched_allow_merge(q, rq, bio))
			return false;
		if (!bio_attempt_back_merge(q, rq, bio))
			return false;
		*merged_request = attempt_back_merge(q, rq);
		if (!*merged_request)
			elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
		return true;
	case ELEVATOR_FRONT_MERGE:
		if (!blk_mq_sched_allow_merge(q, rq, bio))
			return false;
		if (!bio_attempt_front_merge(q, rq, bio))
			return false;
		*merged_request = attempt_front_merge(q, rq);
		if (!*merged_request)
			elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
		return true;
	default:
		return false;
	}
}
EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);

/*
 * 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;

	lockdep_assert_held(&ctx->lock);

	list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) {
		bool merged = false;

		if (!checked--)
			break;

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

		switch (blk_try_merge(rq, bio)) {
		case ELEVATOR_BACK_MERGE:
			if (blk_mq_sched_allow_merge(q, rq, bio))
				merged = bio_attempt_back_merge(q, rq, bio);
			break;
		case ELEVATOR_FRONT_MERGE:
			if (blk_mq_sched_allow_merge(q, rq, bio))
				merged = bio_attempt_front_merge(q, rq, bio);
			break;
		case ELEVATOR_DISCARD_MERGE:
			merged = bio_attempt_discard_merge(q, rq, bio);
			break;
		default:
			continue;
		}

		if (merged)
			ctx->rq_merged++;
		return merged;
	}

	return false;
}

bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio)
{
	struct elevator_queue *e = q->elevator;
	struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	bool ret = false;

	if (e && e->type->ops.mq.bio_merge) {
		blk_mq_put_ctx(ctx);
		return e->type->ops.mq.bio_merge(hctx, bio);
	}

	if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) {
		/* default per sw-queue merge */
		spin_lock(&ctx->lock);
		ret = blk_mq_attempt_merge(q, ctx, bio);
		spin_unlock(&ctx->lock);
	}

	blk_mq_put_ctx(ctx);
	return ret;
}

bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq)
{
	return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq);
}
EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge);

void blk_mq_sched_request_inserted(struct request *rq)
{
	trace_block_rq_insert(rq->q, rq);
}
EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted);

static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx,
				       struct request *rq)
{
	if (rq->tag == -1) {
		rq->rq_flags |= RQF_SORTED;
		return false;
	}

	/*
	 * If we already have a real request tag, send directly to
	 * the dispatch list.
	 */
	spin_lock(&hctx->lock);
	list_add(&rq->queuelist, &hctx->dispatch);
	spin_unlock(&hctx->lock);
	return true;
}

/**
 * list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list
 * @pos:    loop cursor.
 * @skip:   the list element that will not be examined. Iteration starts at
 *          @skip->next.
 * @head:   head of the list to examine. This list must have at least one
 *          element, namely @skip.
 * @member: name of the list_head structure within typeof(*pos).
 */
#define list_for_each_entry_rcu_rr(pos, skip, head, member)		\
	for ((pos) = (skip);						\
	     (pos = (pos)->member.next != (head) ? list_entry_rcu(	\
			(pos)->member.next, typeof(*pos), member) :	\
	      list_entry_rcu((pos)->member.next->next, typeof(*pos), member)), \
	     (pos) != (skip); )

/*
 * Called after a driver tag has been freed to check whether a hctx needs to
 * be restarted. Restarts @hctx if its tag set is not shared. Restarts hardware
 * queues in a round-robin fashion if the tag set of @hctx is shared with other
 * hardware queues.
 */
void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx)
{
	struct blk_mq_tags *const tags = hctx->tags;
	struct blk_mq_tag_set *const set = hctx->queue->tag_set;
	struct request_queue *const queue = hctx->queue, *q;
	struct blk_mq_hw_ctx *hctx2;
	unsigned int i, j;

	if (set->flags & BLK_MQ_F_TAG_SHARED) {
		/*
		 * If this is 0, then we know that no hardware queues
		 * have RESTART marked. We're done.
		 */
		if (!atomic_read(&queue->shared_hctx_restart))
			return;

		rcu_read_lock();
		list_for_each_entry_rcu_rr(q, queue, &set->tag_list,
					   tag_set_list) {
			queue_for_each_hw_ctx(q, hctx2, i)
				if (hctx2->tags == tags &&
				    blk_mq_sched_restart_hctx(hctx2))
					goto done;
		}
		j = hctx->queue_num + 1;
		for (i = 0; i < queue->nr_hw_queues; i++, j++) {
			if (j == queue->nr_hw_queues)
				j = 0;
			hctx2 = queue->queue_hw_ctx[j];
			if (hctx2->tags == tags &&
			    blk_mq_sched_restart_hctx(hctx2))
				break;
		}
done:
		rcu_read_unlock();
	} else {
		blk_mq_sched_restart_hctx(hctx);
	}
}

/*
 * Add flush/fua to the queue. If we fail getting a driver tag, then
 * punt to the requeue list. Requeue will re-invoke us from a context
 * that's safe to block from.
 */
static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx,
				      struct request *rq, bool can_block)
{
	if (blk_mq_get_driver_tag(rq, &hctx, can_block)) {
		blk_insert_flush(rq);
		blk_mq_run_hw_queue(hctx, true);
	} else
		blk_mq_add_to_requeue_list(rq, false, true);
}

void blk_mq_sched_insert_request(struct request *rq, bool at_head,
				 bool run_queue, bool async, bool can_block)
{
	struct request_queue *q = rq->q;
	struct elevator_queue *e = q->elevator;
	struct blk_mq_ctx *ctx = rq->mq_ctx;
	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);

	if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) {
		blk_mq_sched_insert_flush(hctx, rq, can_block);
		return;
	}

	if (e && blk_mq_sched_bypass_insert(hctx, rq))
		goto run;

	if (e && e->type->ops.mq.insert_requests) {
		LIST_HEAD(list);

		list_add(&rq->queuelist, &list);
		e->type->ops.mq.insert_requests(hctx, &list, at_head);
	} else {
		spin_lock(&ctx->lock);
		__blk_mq_insert_request(hctx, rq, at_head);
		spin_unlock(&ctx->lock);
	}

run:
	if (run_queue)
		blk_mq_run_hw_queue(hctx, async);
}

void blk_mq_sched_insert_requests(struct request_queue *q,
				  struct blk_mq_ctx *ctx,
				  struct list_head *list, bool run_queue_async)
{
	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	struct elevator_queue *e = hctx->queue->elevator;

	if (e) {
		struct request *rq, *next;

		/*
		 * We bypass requests that already have a driver tag assigned,
		 * which should only be flushes. Flushes are only ever inserted
		 * as single requests, so we shouldn't ever hit the
		 * WARN_ON_ONCE() below (but let's handle it just in case).
		 */
		list_for_each_entry_safe(rq, next, list, queuelist) {
			if (WARN_ON_ONCE(rq->tag != -1)) {
				list_del_init(&rq->queuelist);
				blk_mq_sched_bypass_insert(hctx, rq);
			}
		}
	}

	if (e && e->type->ops.mq.insert_requests)
		e->type->ops.mq.insert_requests(hctx, list, false);
	else
		blk_mq_insert_requests(hctx, ctx, list);

	blk_mq_run_hw_queue(hctx, run_queue_async);
}

static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set,
				   struct blk_mq_hw_ctx *hctx,
				   unsigned int hctx_idx)
{
	if (hctx->sched_tags) {
		blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx);
		blk_mq_free_rq_map(hctx->sched_tags);
		hctx->sched_tags = NULL;
	}
}

static int blk_mq_sched_alloc_tags(struct request_queue *q,
				   struct blk_mq_hw_ctx *hctx,
				   unsigned int hctx_idx)
{
	struct blk_mq_tag_set *set = q->tag_set;
	int ret;

	hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests,
					       set->reserved_tags);
	if (!hctx->sched_tags)
		return -ENOMEM;

	ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests);
	if (ret)
		blk_mq_sched_free_tags(set, hctx, hctx_idx);

	return ret;
}

static void blk_mq_sched_tags_teardown(struct request_queue *q)
{
	struct blk_mq_tag_set *set = q->tag_set;
	struct blk_mq_hw_ctx *hctx;
	int i;

	queue_for_each_hw_ctx(q, hctx, i)
		blk_mq_sched_free_tags(set, hctx, i);
}

int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
			   unsigned int hctx_idx)
{
	struct elevator_queue *e = q->elevator;
	int ret;

	if (!e)
		return 0;

	ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx);
	if (ret)
		return ret;

	if (e->type->ops.mq.init_hctx) {
		ret = e->type->ops.mq.init_hctx(hctx, hctx_idx);
		if (ret) {
			blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
			return ret;
		}
	}

	blk_mq_debugfs_register_sched_hctx(q, hctx);

	return 0;
}

void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
			    unsigned int hctx_idx)
{
	struct elevator_queue *e = q->elevator;

	if (!e)
		return;

	blk_mq_debugfs_unregister_sched_hctx(hctx);

	if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
		e->type->ops.mq.exit_hctx(hctx, hctx_idx);
		hctx->sched_data = NULL;
	}

	blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
}

int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
{
	struct blk_mq_hw_ctx *hctx;
	struct elevator_queue *eq;
	unsigned int i;
	int ret;

	if (!e) {
		q->elevator = NULL;
		return 0;
	}

	/*
	 * Default to double of smaller one between hw queue_depth and 128,
	 * since we don't split into sync/async like the old code did.
	 * Additionally, this is a per-hw queue depth.
	 */
	q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,
				   BLKDEV_MAX_RQ);

	queue_for_each_hw_ctx(q, hctx, i) {
		ret = blk_mq_sched_alloc_tags(q, hctx, i);
		if (ret)
			goto err;
	}

	ret = e->ops.mq.init_sched(q, e);
	if (ret)
		goto err;

	blk_mq_debugfs_register_sched(q);

	queue_for_each_hw_ctx(q, hctx, i) {
		if (e->ops.mq.init_hctx) {
			ret = e->ops.mq.init_hctx(hctx, i);
			if (ret) {
				eq = q->elevator;
				blk_mq_exit_sched(q, eq);
				kobject_put(&eq->kobj);
				return ret;
			}
		}
		blk_mq_debugfs_register_sched_hctx(q, hctx);
	}

	return 0;

err:
	blk_mq_sched_tags_teardown(q);
	q->elevator = NULL;
	return ret;
}

void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e)
{
	struct blk_mq_hw_ctx *hctx;
	unsigned int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		blk_mq_debugfs_unregister_sched_hctx(hctx);
		if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
			e->type->ops.mq.exit_hctx(hctx, i);
			hctx->sched_data = NULL;
		}
	}
	blk_mq_debugfs_unregister_sched(q);
	if (e->type->ops.mq.exit_sched)
		e->type->ops.mq.exit_sched(e);
	blk_mq_sched_tags_teardown(q);
	q->elevator = NULL;
}

int blk_mq_sched_init(struct request_queue *q)
{
	int ret;

	mutex_lock(&q->sysfs_lock);
	ret = elevator_init(q, NULL);
	mutex_unlock(&q->sysfs_lock);

	return ret;
}
Commit	Line	Data
bd166ef1 JA	1	/*
	2	* blk-mq scheduling framework
	3	*
	4	* Copyright (C) 2016 Jens Axboe
	5	*/
	6	#include <linux/kernel.h>
	7	#include <linux/module.h>
	8	#include <linux/blk-mq.h>
	9
	10	#include <trace/events/block.h>
	11
	12	#include "blk.h"
	13	#include "blk-mq.h"
d332ce09	14	#include "blk-mq-debugfs.h"
bd166ef1 JA	15	#include "blk-mq-sched.h"
	16	#include "blk-mq-tag.h"
	17	#include "blk-wbt.h"
	18
	19	void blk_mq_sched_free_hctx_data(struct request_queue *q,
	20	void (exit)(struct blk_mq_hw_ctx ))
	21	{
	22	struct blk_mq_hw_ctx *hctx;
	23	int i;
	24
	25	queue_for_each_hw_ctx(q, hctx, i) {
	26	if (exit && hctx->sched_data)
	27	exit(hctx);
	28	kfree(hctx->sched_data);
	29	hctx->sched_data = NULL;
	30	}
	31	}
	32	EXPORT_SYMBOL_GPL(blk_mq_sched_free_hctx_data);
	33
44e8c2bf	34	void blk_mq_sched_assign_ioc(struct request rq, struct bio bio)
bd166ef1	35	{
44e8c2bf CH	36	struct request_queue *q = rq->q;
44e8c2bf CH	37	struct io_context *ioc = rq_ioc(bio);
bd166ef1 JA	38	struct io_cq *icq;
	39
	40	spin_lock_irq(q->queue_lock);
	41	icq = ioc_lookup_icq(ioc, q);
	42	spin_unlock_irq(q->queue_lock);
	43
	44	if (!icq) {
	45	icq = ioc_create_icq(ioc, q, GFP_ATOMIC);
	46	if (!icq)
	47	return;
	48	}
ea511e3c	49	get_io_context(icq->ioc);
44e8c2bf	50	rq->elv.icq = icq;
bd166ef1 JA	51	}
bd166ef1 JA	52
8e8320c9 JA	53	/*
	54	* Mark a hardware queue as needing a restart. For shared queues, maintain
	55	* a count of how many hardware queues are marked for restart.
	56	*/
	57	static void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx)
	58	{
	59	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
	60	return;
	61
	62	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
	63	struct request_queue *q = hctx->queue;
	64
	65	if (!test_and_set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
	66	atomic_inc(&q->shared_hctx_restart);
	67	} else
	68	set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
	69	}
	70
	71	static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
	72	{
	73	if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
	74	return false;
	75
	76	if (hctx->flags & BLK_MQ_F_TAG_SHARED) {
	77	struct request_queue *q = hctx->queue;
	78
	79	if (test_and_clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state))
	80	atomic_dec(&q->shared_hctx_restart);
	81	} else
	82	clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
	83
	84	if (blk_mq_hctx_has_pending(hctx)) {
	85	blk_mq_run_hw_queue(hctx, true);
	86	return true;
	87	}
	88
	89	return false;
	90	}
	91
de148297 ML	92	/* return true if hctx need to run again */
de148297 ML	93	static bool blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
caf8eb0d ML	94	{
	95	struct request_queue *q = hctx->queue;
	96	struct elevator_queue *e = q->elevator;
	97	LIST_HEAD(rq_list);
	98
	99	do {
de148297 ML	100	struct request *rq;
de148297 ML	101	blk_status_t ret;
caf8eb0d	102
de148297 ML	103	if (e->type->ops.mq.has_work &&
de148297 ML	104	!e->type->ops.mq.has_work(hctx))
caf8eb0d	105	break;
de148297 ML	106
	107	ret = blk_mq_get_dispatch_budget(hctx);
	108	if (ret == BLK_STS_RESOURCE)
	109	return true;
	110
	111	rq = e->type->ops.mq.dispatch_request(hctx);
	112	if (!rq) {
	113	blk_mq_put_dispatch_budget(hctx);
	114	break;
	115	} else if (ret != BLK_STS_OK) {
	116	blk_mq_end_request(rq, ret);
	117	continue;
	118	}
	119
	120	/*
	121	* Now this rq owns the budget which has to be released
	122	* if this rq won't be queued to driver via .queue_rq()
	123	* in blk_mq_dispatch_rq_list().
	124	*/
caf8eb0d	125	list_add(&rq->queuelist, &rq_list);
de148297 ML	126	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));
	127
	128	return false;
caf8eb0d ML	129	}
caf8eb0d ML	130
de148297 ML	131	/* return true if hw queue need to be run again */
de148297 ML	132	bool blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
bd166ef1	133	{
81380ca1 OS	134	struct request_queue *q = hctx->queue;
81380ca1 OS	135	struct elevator_queue *e = q->elevator;
64765a75	136	const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request;
bd166ef1	137	LIST_HEAD(rq_list);
de148297	138	bool run_queue = false;
bd166ef1	139
f4560ffe ML	140	/* RCU or SRCU read lock is needed before checking quiesced flag */
f4560ffe ML	141	if (unlikely(blk_mq_hctx_stopped(hctx) \|\| blk_queue_quiesced(q)))
de148297	142	return false;
bd166ef1 JA	143
	144	hctx->run++;
	145
	146	/*
	147	* If we have previous entries on our dispatch list, grab them first for
	148	* more fair dispatch.
	149	*/
	150	if (!list_empty_careful(&hctx->dispatch)) {
	151	spin_lock(&hctx->lock);
	152	if (!list_empty(&hctx->dispatch))
	153	list_splice_init(&hctx->dispatch, &rq_list);
	154	spin_unlock(&hctx->lock);
	155	}
	156
	157	/*
	158	* Only ask the scheduler for requests, if we didn't have residual
	159	* requests from the dispatch list. This is to avoid the case where
	160	* we only ever dispatch a fraction of the requests available because
	161	* of low device queue depth. Once we pull requests out of the IO
	162	* scheduler, we can no longer merge or sort them. So it's best to
	163	* leave them there for as long as we can. Mark the hw queue as
	164	* needing a restart in that case.
caf8eb0d ML	165	*
	166	* We want to dispatch from the scheduler if there was nothing
	167	* on the dispatch list or we were able to dispatch from the
	168	* dispatch list.
bd166ef1	169	*/
c13660a0	170	if (!list_empty(&rq_list)) {
d38d3515	171	blk_mq_sched_mark_restart_hctx(hctx);
de148297 ML	172	if (blk_mq_dispatch_rq_list(q, &rq_list, false) &&
	173	has_sched_dispatch)
	174	run_queue = blk_mq_do_dispatch_sched(hctx);
caf8eb0d	175	} else if (has_sched_dispatch) {
de148297	176	run_queue = blk_mq_do_dispatch_sched(hctx);
caf8eb0d	177	} else {
c13660a0	178	blk_mq_flush_busy_ctxs(hctx, &rq_list);
de148297	179	blk_mq_dispatch_rq_list(q, &rq_list, false);
64765a75	180	}
de148297 ML	181
	182	if (run_queue && !blk_mq_sched_needs_restart(hctx) &&
	183	!test_bit(BLK_MQ_S_TAG_WAITING, &hctx->state)) {
	184	blk_mq_sched_mark_restart_hctx(hctx);
	185	return true;
	186	}
	187
	188	return false;
bd166ef1 JA	189	}
bd166ef1 JA	190
e4d750c9 JA	191	bool blk_mq_sched_try_merge(struct request_queue q, struct bio bio,
e4d750c9 JA	192	struct request **merged_request)
bd166ef1 JA	193	{
bd166ef1 JA	194	struct request *rq;
bd166ef1	195
34fe7c05 CH	196	switch (elv_merge(q, &rq, bio)) {
34fe7c05 CH	197	case ELEVATOR_BACK_MERGE:
bd166ef1 JA	198	if (!blk_mq_sched_allow_merge(q, rq, bio))
bd166ef1 JA	199	return false;
34fe7c05 CH	200	if (!bio_attempt_back_merge(q, rq, bio))
	201	return false;
	202	*merged_request = attempt_back_merge(q, rq);
	203	if (!*merged_request)
	204	elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
	205	return true;
	206	case ELEVATOR_FRONT_MERGE:
bd166ef1 JA	207	if (!blk_mq_sched_allow_merge(q, rq, bio))
bd166ef1 JA	208	return false;
34fe7c05 CH	209	if (!bio_attempt_front_merge(q, rq, bio))
	210	return false;
	211	*merged_request = attempt_front_merge(q, rq);
	212	if (!*merged_request)
	213	elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
	214	return true;
	215	default:
	216	return false;
bd166ef1	217	}
bd166ef1 JA	218	}
	219	EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
	220
9bddeb2a ML	221	/*
	222	* Reverse check our software queue for entries that we could potentially
	223	* merge with. Currently includes a hand-wavy stop count of 8, to not spend
	224	* too much time checking for merges.
	225	*/
	226	static bool blk_mq_attempt_merge(struct request_queue *q,
	227	struct blk_mq_ctx ctx, struct bio bio)
	228	{
	229	struct request *rq;
	230	int checked = 8;
	231
7b607814 BVA	232	lockdep_assert_held(&ctx->lock);
7b607814 BVA	233
9bddeb2a ML	234	list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) {
	235	bool merged = false;
	236
	237	if (!checked--)
	238	break;
	239
	240	if (!blk_rq_merge_ok(rq, bio))
	241	continue;
	242
	243	switch (blk_try_merge(rq, bio)) {
	244	case ELEVATOR_BACK_MERGE:
	245	if (blk_mq_sched_allow_merge(q, rq, bio))
	246	merged = bio_attempt_back_merge(q, rq, bio);
	247	break;
	248	case ELEVATOR_FRONT_MERGE:
	249	if (blk_mq_sched_allow_merge(q, rq, bio))
	250	merged = bio_attempt_front_merge(q, rq, bio);
	251	break;
	252	case ELEVATOR_DISCARD_MERGE:
	253	merged = bio_attempt_discard_merge(q, rq, bio);
	254	break;
	255	default:
	256	continue;
	257	}
	258
	259	if (merged)
	260	ctx->rq_merged++;
	261	return merged;
	262	}
	263
	264	return false;
	265	}
	266
bd166ef1 JA	267	bool __blk_mq_sched_bio_merge(struct request_queue q, struct bio bio)
	268	{
	269	struct elevator_queue *e = q->elevator;
9bddeb2a ML	270	struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
	271	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	272	bool ret = false;
bd166ef1	273
9bddeb2a	274	if (e && e->type->ops.mq.bio_merge) {
bd166ef1 JA	275	blk_mq_put_ctx(ctx);
	276	return e->type->ops.mq.bio_merge(hctx, bio);
	277	}
	278
9bddeb2a ML	279	if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) {
	280	/* default per sw-queue merge */
	281	spin_lock(&ctx->lock);
	282	ret = blk_mq_attempt_merge(q, ctx, bio);
	283	spin_unlock(&ctx->lock);
	284	}
	285
	286	blk_mq_put_ctx(ctx);
	287	return ret;
bd166ef1 JA	288	}
	289
	290	bool blk_mq_sched_try_insert_merge(struct request_queue q, struct request rq)
	291	{
	292	return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq);
	293	}
	294	EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge);
	295
	296	void blk_mq_sched_request_inserted(struct request *rq)
	297	{
	298	trace_block_rq_insert(rq->q, rq);
	299	}
	300	EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted);
	301
0cacba6c OS	302	static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx,
0cacba6c OS	303	struct request *rq)
bd166ef1 JA	304	{
	305	if (rq->tag == -1) {
	306	rq->rq_flags \|= RQF_SORTED;
	307	return false;
	308	}
	309
	310	/*
	311	* If we already have a real request tag, send directly to
	312	* the dispatch list.
	313	*/
	314	spin_lock(&hctx->lock);
	315	list_add(&rq->queuelist, &hctx->dispatch);
	316	spin_unlock(&hctx->lock);
	317	return true;
	318	}
bd166ef1	319
6d8c6c0f BVA	320	/**
	321	* list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list
	322	* @pos: loop cursor.
	323	* @skip: the list element that will not be examined. Iteration starts at
	324	* @skip->next.
	325	* @head: head of the list to examine. This list must have at least one
	326	* element, namely @skip.
	327	* @member: name of the list_head structure within typeof(*pos).
	328	*/
	329	#define list_for_each_entry_rcu_rr(pos, skip, head, member) \
	330	for ((pos) = (skip); \
	331	(pos = (pos)->member.next != (head) ? list_entry_rcu( \
	332	(pos)->member.next, typeof(*pos), member) : \
	333	list_entry_rcu((pos)->member.next->next, typeof(*pos), member)), \
	334	(pos) != (skip); )
50e1dab8	335
6d8c6c0f BVA	336	/*
	337	* Called after a driver tag has been freed to check whether a hctx needs to
	338	* be restarted. Restarts @hctx if its tag set is not shared. Restarts hardware
	339	* queues in a round-robin fashion if the tag set of @hctx is shared with other
	340	* hardware queues.
	341	*/
	342	void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx)
	343	{
	344	struct blk_mq_tags *const tags = hctx->tags;
	345	struct blk_mq_tag_set *const set = hctx->queue->tag_set;
	346	struct request_queue const queue = hctx->queue, q;
	347	struct blk_mq_hw_ctx *hctx2;
	348	unsigned int i, j;
	349
	350	if (set->flags & BLK_MQ_F_TAG_SHARED) {
8e8320c9 JA	351	/*
	352	* If this is 0, then we know that no hardware queues
	353	* have RESTART marked. We're done.
	354	*/
	355	if (!atomic_read(&queue->shared_hctx_restart))
	356	return;
	357
6d8c6c0f BVA	358	rcu_read_lock();
	359	list_for_each_entry_rcu_rr(q, queue, &set->tag_list,
	360	tag_set_list) {
	361	queue_for_each_hw_ctx(q, hctx2, i)
	362	if (hctx2->tags == tags &&
	363	blk_mq_sched_restart_hctx(hctx2))
	364	goto done;
	365	}
	366	j = hctx->queue_num + 1;
	367	for (i = 0; i < queue->nr_hw_queues; i++, j++) {
	368	if (j == queue->nr_hw_queues)
	369	j = 0;
	370	hctx2 = queue->queue_hw_ctx[j];
	371	if (hctx2->tags == tags &&
	372	blk_mq_sched_restart_hctx(hctx2))
	373	break;
d38d3515	374	}
6d8c6c0f BVA	375	done:
6d8c6c0f BVA	376	rcu_read_unlock();
d38d3515	377	} else {
50e1dab8	378	blk_mq_sched_restart_hctx(hctx);
50e1dab8 JA	379	}
	380	}
	381
bd6737f1 JA	382	/*
	383	* Add flush/fua to the queue. If we fail getting a driver tag, then
	384	* punt to the requeue list. Requeue will re-invoke us from a context
	385	* that's safe to block from.
	386	*/
	387	static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx,
	388	struct request *rq, bool can_block)
	389	{
	390	if (blk_mq_get_driver_tag(rq, &hctx, can_block)) {
	391	blk_insert_flush(rq);
	392	blk_mq_run_hw_queue(hctx, true);
	393	} else
c7a571b4	394	blk_mq_add_to_requeue_list(rq, false, true);
bd6737f1 JA	395	}
	396
	397	void blk_mq_sched_insert_request(struct request *rq, bool at_head,
	398	bool run_queue, bool async, bool can_block)
	399	{
	400	struct request_queue *q = rq->q;
	401	struct elevator_queue *e = q->elevator;
	402	struct blk_mq_ctx *ctx = rq->mq_ctx;
	403	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	404
f3a8ab7d	405	if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) {
bd6737f1 JA	406	blk_mq_sched_insert_flush(hctx, rq, can_block);
	407	return;
	408	}
	409
0cacba6c OS	410	if (e && blk_mq_sched_bypass_insert(hctx, rq))
	411	goto run;
	412
bd6737f1 JA	413	if (e && e->type->ops.mq.insert_requests) {
	414	LIST_HEAD(list);
	415
	416	list_add(&rq->queuelist, &list);
	417	e->type->ops.mq.insert_requests(hctx, &list, at_head);
	418	} else {
	419	spin_lock(&ctx->lock);
	420	__blk_mq_insert_request(hctx, rq, at_head);
	421	spin_unlock(&ctx->lock);
	422	}
	423
0cacba6c	424	run:
bd6737f1 JA	425	if (run_queue)
	426	blk_mq_run_hw_queue(hctx, async);
	427	}
	428
	429	void blk_mq_sched_insert_requests(struct request_queue *q,
	430	struct blk_mq_ctx *ctx,
	431	struct list_head *list, bool run_queue_async)
	432	{
	433	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	434	struct elevator_queue *e = hctx->queue->elevator;
	435
0cacba6c OS	436	if (e) {
	437	struct request rq, next;
	438
	439	/*
	440	* We bypass requests that already have a driver tag assigned,
	441	* which should only be flushes. Flushes are only ever inserted
	442	* as single requests, so we shouldn't ever hit the
	443	* WARN_ON_ONCE() below (but let's handle it just in case).
	444	*/
	445	list_for_each_entry_safe(rq, next, list, queuelist) {
	446	if (WARN_ON_ONCE(rq->tag != -1)) {
	447	list_del_init(&rq->queuelist);
	448	blk_mq_sched_bypass_insert(hctx, rq);
	449	}
	450	}
	451	}
	452
bd6737f1 JA	453	if (e && e->type->ops.mq.insert_requests)
	454	e->type->ops.mq.insert_requests(hctx, list, false);
	455	else
	456	blk_mq_insert_requests(hctx, ctx, list);
	457
	458	blk_mq_run_hw_queue(hctx, run_queue_async);
	459	}
	460
bd166ef1 JA	461	static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set,
	462	struct blk_mq_hw_ctx *hctx,
	463	unsigned int hctx_idx)
	464	{
	465	if (hctx->sched_tags) {
	466	blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx);
	467	blk_mq_free_rq_map(hctx->sched_tags);
	468	hctx->sched_tags = NULL;
	469	}
	470	}
	471
6917ff0b OS	472	static int blk_mq_sched_alloc_tags(struct request_queue *q,
	473	struct blk_mq_hw_ctx *hctx,
	474	unsigned int hctx_idx)
	475	{
	476	struct blk_mq_tag_set *set = q->tag_set;
	477	int ret;
	478
	479	hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests,
	480	set->reserved_tags);
	481	if (!hctx->sched_tags)
	482	return -ENOMEM;
	483
	484	ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests);
	485	if (ret)
	486	blk_mq_sched_free_tags(set, hctx, hctx_idx);
	487
	488	return ret;
	489	}
	490
54d5329d	491	static void blk_mq_sched_tags_teardown(struct request_queue *q)
bd166ef1 JA	492	{
	493	struct blk_mq_tag_set *set = q->tag_set;
	494	struct blk_mq_hw_ctx *hctx;
6917ff0b OS	495	int i;
	496
	497	queue_for_each_hw_ctx(q, hctx, i)
	498	blk_mq_sched_free_tags(set, hctx, i);
	499	}
	500
93252632 OS	501	int blk_mq_sched_init_hctx(struct request_queue q, struct blk_mq_hw_ctx hctx,
	502	unsigned int hctx_idx)
	503	{
	504	struct elevator_queue *e = q->elevator;
ee056f98	505	int ret;
93252632 OS	506
	507	if (!e)
	508	return 0;
	509
ee056f98 OS	510	ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx);
	511	if (ret)
	512	return ret;
	513
	514	if (e->type->ops.mq.init_hctx) {
	515	ret = e->type->ops.mq.init_hctx(hctx, hctx_idx);
	516	if (ret) {
	517	blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
	518	return ret;
	519	}
	520	}
	521
d332ce09 OS	522	blk_mq_debugfs_register_sched_hctx(q, hctx);
d332ce09 OS	523
ee056f98	524	return 0;
93252632 OS	525	}
	526
	527	void blk_mq_sched_exit_hctx(struct request_queue q, struct blk_mq_hw_ctx hctx,
	528	unsigned int hctx_idx)
	529	{
	530	struct elevator_queue *e = q->elevator;
	531
	532	if (!e)
	533	return;
	534
d332ce09 OS	535	blk_mq_debugfs_unregister_sched_hctx(hctx);
d332ce09 OS	536
ee056f98 OS	537	if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
	538	e->type->ops.mq.exit_hctx(hctx, hctx_idx);
	539	hctx->sched_data = NULL;
	540	}
	541
93252632 OS	542	blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
	543	}
	544
6917ff0b OS	545	int blk_mq_init_sched(struct request_queue q, struct elevator_type e)
	546	{
	547	struct blk_mq_hw_ctx *hctx;
ee056f98	548	struct elevator_queue *eq;
6917ff0b OS	549	unsigned int i;
	550	int ret;
	551
	552	if (!e) {
	553	q->elevator = NULL;
	554	return 0;
	555	}
bd166ef1 JA	556
bd166ef1 JA	557	/*
32825c45 ML	558	* Default to double of smaller one between hw queue_depth and 128,
	559	* since we don't split into sync/async like the old code did.
	560	* Additionally, this is a per-hw queue depth.
bd166ef1	561	*/
32825c45 ML	562	q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,
32825c45 ML	563	BLKDEV_MAX_RQ);
bd166ef1	564
bd166ef1	565	queue_for_each_hw_ctx(q, hctx, i) {
6917ff0b	566	ret = blk_mq_sched_alloc_tags(q, hctx, i);
bd166ef1	567	if (ret)
6917ff0b	568	goto err;
bd166ef1 JA	569	}
bd166ef1 JA	570
6917ff0b OS	571	ret = e->ops.mq.init_sched(q, e);
	572	if (ret)
	573	goto err;
bd166ef1	574
d332ce09 OS	575	blk_mq_debugfs_register_sched(q);
	576
	577	queue_for_each_hw_ctx(q, hctx, i) {
	578	if (e->ops.mq.init_hctx) {
ee056f98 OS	579	ret = e->ops.mq.init_hctx(hctx, i);
	580	if (ret) {
	581	eq = q->elevator;
	582	blk_mq_exit_sched(q, eq);
	583	kobject_put(&eq->kobj);
	584	return ret;
	585	}
	586	}
d332ce09	587	blk_mq_debugfs_register_sched_hctx(q, hctx);
ee056f98 OS	588	}
ee056f98 OS	589
bd166ef1	590	return 0;
bd166ef1	591
6917ff0b	592	err:
54d5329d OS	593	blk_mq_sched_tags_teardown(q);
54d5329d OS	594	q->elevator = NULL;
6917ff0b	595	return ret;
bd166ef1	596	}
d3484991	597
54d5329d OS	598	void blk_mq_exit_sched(struct request_queue q, struct elevator_queue e)
54d5329d OS	599	{
ee056f98 OS	600	struct blk_mq_hw_ctx *hctx;
	601	unsigned int i;
	602
d332ce09 OS	603	queue_for_each_hw_ctx(q, hctx, i) {
	604	blk_mq_debugfs_unregister_sched_hctx(hctx);
	605	if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
	606	e->type->ops.mq.exit_hctx(hctx, i);
	607	hctx->sched_data = NULL;
ee056f98 OS	608	}
ee056f98 OS	609	}
d332ce09	610	blk_mq_debugfs_unregister_sched(q);
54d5329d OS	611	if (e->type->ops.mq.exit_sched)
	612	e->type->ops.mq.exit_sched(e);
	613	blk_mq_sched_tags_teardown(q);
	614	q->elevator = NULL;
	615	}
	616
d3484991 JA	617	int blk_mq_sched_init(struct request_queue *q)
	618	{
	619	int ret;
	620
d3484991 JA	621	mutex_lock(&q->sysfs_lock);
	622	ret = elevator_init(q, NULL);
	623	mutex_unlock(&q->sysfs_lock);
	624
	625	return ret;
	626	}