[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-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);

int blk_mq_sched_init_hctx_data(struct request_queue *q, size_t size,
				int (*init)(struct blk_mq_hw_ctx *),
				void (*exit)(struct blk_mq_hw_ctx *))
{
	struct blk_mq_hw_ctx *hctx;
	int ret;
	int i;

	queue_for_each_hw_ctx(q, hctx, i) {
		hctx->sched_data = kmalloc_node(size, GFP_KERNEL, hctx->numa_node);
		if (!hctx->sched_data) {
			ret = -ENOMEM;
			goto error;
		}

		if (init) {
			ret = init(hctx);
			if (ret) {
				/*
				 * We don't want to give exit() a partially
				 * initialized sched_data. init() must clean up
				 * if it fails.
				 */
				kfree(hctx->sched_data);
				hctx->sched_data = NULL;
				goto error;
			}
		}
	}

	return 0;
error:
	blk_mq_sched_free_hctx_data(q, exit);
	return ret;
}
EXPORT_SYMBOL_GPL(blk_mq_sched_init_hctx_data);

static void __blk_mq_sched_assign_ioc(struct request_queue *q,
				      struct request *rq,
				      struct bio *bio,
				      struct io_context *ioc)
{
	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;
	}

	rq->elv.icq = icq;
	if (!blk_mq_sched_get_rq_priv(q, rq, bio)) {
		rq->rq_flags |= RQF_ELVPRIV;
		get_io_context(icq->ioc);
		return;
	}

	rq->elv.icq = NULL;
}

static void blk_mq_sched_assign_ioc(struct request_queue *q,
				    struct request *rq, struct bio *bio)
{
	struct io_context *ioc;

	ioc = rq_ioc(bio);
	if (ioc)
		__blk_mq_sched_assign_ioc(q, rq, bio, ioc);
}

struct request *blk_mq_sched_get_request(struct request_queue *q,
					 struct bio *bio,
					 unsigned int op,
					 struct blk_mq_alloc_data *data)
{
	struct elevator_queue *e = q->elevator;
	struct blk_mq_hw_ctx *hctx;
	struct blk_mq_ctx *ctx;
	struct request *rq;

	blk_queue_enter_live(q);
	ctx = blk_mq_get_ctx(q);
	hctx = blk_mq_map_queue(q, ctx->cpu);

	blk_mq_set_alloc_data(data, q, data->flags, ctx, hctx);

	if (e) {
		data->flags |= BLK_MQ_REQ_INTERNAL;

		/*
		 * Flush requests are special and go directly to the
		 * dispatch list.
		 */
		if (!op_is_flush(op) && e->type->ops.mq.get_request) {
			rq = e->type->ops.mq.get_request(q, op, data);
			if (rq)
				rq->rq_flags |= RQF_QUEUED;
		} else
			rq = __blk_mq_alloc_request(data, op);
	} else {
		rq = __blk_mq_alloc_request(data, op);
		if (rq)
			data->hctx->tags->rqs[rq->tag] = rq;
	}

	if (rq) {
		if (!op_is_flush(op)) {
			rq->elv.icq = NULL;
			if (e && e->type->icq_cache)
				blk_mq_sched_assign_ioc(q, rq, bio);
		}
		data->hctx->queued++;
		return rq;
	}

	blk_queue_exit(q);
	return NULL;
}

void blk_mq_sched_put_request(struct request *rq)
{
	struct request_queue *q = rq->q;
	struct elevator_queue *e = q->elevator;

	if (rq->rq_flags & RQF_ELVPRIV) {
		blk_mq_sched_put_rq_priv(rq->q, rq);
		if (rq->elv.icq) {
			put_io_context(rq->elv.icq->ioc);
			rq->elv.icq = NULL;
		}
	}

	if ((rq->rq_flags & RQF_QUEUED) && e && e->type->ops.mq.put_request)
		e->type->ops.mq.put_request(rq);
	else
		blk_mq_finish_request(rq);
}

void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
{
	struct elevator_queue *e = hctx->queue->elevator;
	const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request;
	bool did_work = false;
	LIST_HEAD(rq_list);

	if (unlikely(blk_mq_hctx_stopped(hctx)))
		return;

	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.
	 */
	if (!list_empty(&rq_list)) {
		blk_mq_sched_mark_restart(hctx);
		did_work = blk_mq_dispatch_rq_list(hctx, &rq_list);
	} else if (!has_sched_dispatch) {
		blk_mq_flush_busy_ctxs(hctx, &rq_list);
		blk_mq_dispatch_rq_list(hctx, &rq_list);
	}

	/*
	 * We want to dispatch from the scheduler if we had no work left
	 * on the dispatch list, OR if we did have work but weren't able
	 * to make progress.
	 */
	if (!did_work && has_sched_dispatch) {
		do {
			struct request *rq;

			rq = e->type->ops.mq.dispatch_request(hctx);
			if (!rq)
				break;
			list_add(&rq->queuelist, &rq_list);
		} while (blk_mq_dispatch_rq_list(hctx, &rq_list));
	}
}

void blk_mq_sched_move_to_dispatch(struct blk_mq_hw_ctx *hctx,
				   struct list_head *rq_list,
				   struct request *(*get_rq)(struct blk_mq_hw_ctx *))
{
	do {
		struct request *rq;

		rq = get_rq(hctx);
		if (!rq)
			break;

		list_add_tail(&rq->queuelist, rq_list);
	} while (1);
}
EXPORT_SYMBOL_GPL(blk_mq_sched_move_to_dispatch);

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);

bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio)
{
	struct elevator_queue *e = q->elevator;

	if (e->type->ops.mq.bio_merge) {
		struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
		struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);

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

	return false;
}

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;
}

static void blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
{
	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) {
		clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
		if (blk_mq_hctx_has_pending(hctx))
			blk_mq_run_hw_queue(hctx, true);
	}
}

void blk_mq_sched_restart_queues(struct blk_mq_hw_ctx *hctx)
{
	unsigned int i;

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

		if (!test_bit(QUEUE_FLAG_RESTART, &q->queue_flags))
			return;

		clear_bit(QUEUE_FLAG_RESTART, &q->queue_flags);

		queue_for_each_hw_ctx(q, hctx, i)
			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;
	}
}

int blk_mq_sched_setup(struct request_queue *q)
{
	struct blk_mq_tag_set *set = q->tag_set;
	struct blk_mq_hw_ctx *hctx;
	int ret, i;

	/*
	 * Default to 256, 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 * BLKDEV_MAX_RQ;

	/*
	 * We're switching to using an IO scheduler, so setup the hctx
	 * scheduler tags and switch the request map from the regular
	 * tags to scheduler tags. First allocate what we need, so we
	 * can safely fail and fallback, if needed.
	 */
	ret = 0;
	queue_for_each_hw_ctx(q, hctx, i) {
		hctx->sched_tags = blk_mq_alloc_rq_map(set, i, q->nr_requests, 0);
		if (!hctx->sched_tags) {
			ret = -ENOMEM;
			break;
		}
		ret = blk_mq_alloc_rqs(set, hctx->sched_tags, i, q->nr_requests);
		if (ret)
			break;
	}

	/*
	 * If we failed, free what we did allocate
	 */
	if (ret) {
		queue_for_each_hw_ctx(q, hctx, i) {
			if (!hctx->sched_tags)
				continue;
			blk_mq_sched_free_tags(set, hctx, i);
		}

		return ret;
	}

	return 0;
}

void blk_mq_sched_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(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"
	14	#include "blk-mq-sched.h"
	15	#include "blk-mq-tag.h"
	16	#include "blk-wbt.h"
	17
	18	void blk_mq_sched_free_hctx_data(struct request_queue *q,
	19	void (exit)(struct blk_mq_hw_ctx ))
	20	{
	21	struct blk_mq_hw_ctx *hctx;
	22	int i;
	23
	24	queue_for_each_hw_ctx(q, hctx, i) {
	25	if (exit && hctx->sched_data)
	26	exit(hctx);
	27	kfree(hctx->sched_data);
	28	hctx->sched_data = NULL;
	29	}
	30	}
	31	EXPORT_SYMBOL_GPL(blk_mq_sched_free_hctx_data);
	32
	33	int blk_mq_sched_init_hctx_data(struct request_queue *q, size_t size,
	34	int (init)(struct blk_mq_hw_ctx ),
	35	void (exit)(struct blk_mq_hw_ctx ))
	36	{
	37	struct blk_mq_hw_ctx *hctx;
	38	int ret;
	39	int i;
	40
	41	queue_for_each_hw_ctx(q, hctx, i) {
	42	hctx->sched_data = kmalloc_node(size, GFP_KERNEL, hctx->numa_node);
	43	if (!hctx->sched_data) {
	44	ret = -ENOMEM;
	45	goto error;
	46	}
	47
	48	if (init) {
	49	ret = init(hctx);
	50	if (ret) {
	51	/*
	52	* We don't want to give exit() a partially
	53	* initialized sched_data. init() must clean up
	54	* if it fails.
	55	*/
	56	kfree(hctx->sched_data);
	57	hctx->sched_data = NULL;
	58	goto error;
	59	}
	60	}
	61	}
	62
	63	return 0;
	64	error:
65	blk_mq_sched_free_hctx_data(q, exit);
66	return ret;
67	}
68	EXPORT_SYMBOL_GPL(blk_mq_sched_init_hctx_data);
69
70	static void __blk_mq_sched_assign_ioc(struct request_queue *q,
f1ba8261 PV	71	struct request *rq,
	72	struct bio *bio,
	73	struct io_context *ioc)
bd166ef1 JA	74	{
	75	struct io_cq *icq;
	76
	77	spin_lock_irq(q->queue_lock);
	78	icq = ioc_lookup_icq(ioc, q);
	79	spin_unlock_irq(q->queue_lock);
	80
	81	if (!icq) {
	82	icq = ioc_create_icq(ioc, q, GFP_ATOMIC);
	83	if (!icq)
	84	return;
	85	}
	86
	87	rq->elv.icq = icq;
f1ba8261	88	if (!blk_mq_sched_get_rq_priv(q, rq, bio)) {
bd166ef1 JA	89	rq->rq_flags \|= RQF_ELVPRIV;
	90	get_io_context(icq->ioc);
	91	return;
	92	}
	93
	94	rq->elv.icq = NULL;
	95	}
	96
	97	static void blk_mq_sched_assign_ioc(struct request_queue *q,
	98	struct request rq, struct bio bio)
	99	{
	100	struct io_context *ioc;
	101
	102	ioc = rq_ioc(bio);
	103	if (ioc)
f1ba8261	104	__blk_mq_sched_assign_ioc(q, rq, bio, ioc);
bd166ef1 JA	105	}
	106
	107	struct request blk_mq_sched_get_request(struct request_queue q,
	108	struct bio *bio,
	109	unsigned int op,
	110	struct blk_mq_alloc_data *data)
	111	{
	112	struct elevator_queue *e = q->elevator;
	113	struct blk_mq_hw_ctx *hctx;
	114	struct blk_mq_ctx *ctx;
	115	struct request *rq;
bd166ef1 JA	116
	117	blk_queue_enter_live(q);
	118	ctx = blk_mq_get_ctx(q);
	119	hctx = blk_mq_map_queue(q, ctx->cpu);
	120
5a797e00	121	blk_mq_set_alloc_data(data, q, data->flags, ctx, hctx);
bd166ef1 JA	122
	123	if (e) {
	124	data->flags \|= BLK_MQ_REQ_INTERNAL;
	125
	126	/*
	127	* Flush requests are special and go directly to the
	128	* dispatch list.
	129	*/
f73f44eb	130	if (!op_is_flush(op) && e->type->ops.mq.get_request) {
bd166ef1 JA	131	rq = e->type->ops.mq.get_request(q, op, data);
	132	if (rq)
	133	rq->rq_flags \|= RQF_QUEUED;
	134	} else
	135	rq = __blk_mq_alloc_request(data, op);
	136	} else {
	137	rq = __blk_mq_alloc_request(data, op);
b48fda09 JA	138	if (rq)
b48fda09 JA	139	data->hctx->tags->rqs[rq->tag] = rq;
bd166ef1 JA	140	}
	141
	142	if (rq) {
f73f44eb	143	if (!op_is_flush(op)) {
bd166ef1 JA	144	rq->elv.icq = NULL;
	145	if (e && e->type->icq_cache)
	146	blk_mq_sched_assign_ioc(q, rq, bio);
	147	}
	148	data->hctx->queued++;
	149	return rq;
	150	}
	151
	152	blk_queue_exit(q);
	153	return NULL;
	154	}
	155
	156	void blk_mq_sched_put_request(struct request *rq)
	157	{
	158	struct request_queue *q = rq->q;
	159	struct elevator_queue *e = q->elevator;
	160
	161	if (rq->rq_flags & RQF_ELVPRIV) {
	162	blk_mq_sched_put_rq_priv(rq->q, rq);
	163	if (rq->elv.icq) {
	164	put_io_context(rq->elv.icq->ioc);
	165	rq->elv.icq = NULL;
	166	}
	167	}
	168
	169	if ((rq->rq_flags & RQF_QUEUED) && e && e->type->ops.mq.put_request)
	170	e->type->ops.mq.put_request(rq);
	171	else
	172	blk_mq_finish_request(rq);
	173	}
	174
	175	void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
	176	{
	177	struct elevator_queue *e = hctx->queue->elevator;
64765a75 JA	178	const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request;
64765a75 JA	179	bool did_work = false;
bd166ef1 JA	180	LIST_HEAD(rq_list);
	181
	182	if (unlikely(blk_mq_hctx_stopped(hctx)))
	183	return;
	184
	185	hctx->run++;
	186
	187	/*
	188	* If we have previous entries on our dispatch list, grab them first for
	189	* more fair dispatch.
	190	*/
	191	if (!list_empty_careful(&hctx->dispatch)) {
	192	spin_lock(&hctx->lock);
	193	if (!list_empty(&hctx->dispatch))
	194	list_splice_init(&hctx->dispatch, &rq_list);
	195	spin_unlock(&hctx->lock);
	196	}
	197
	198	/*
	199	* Only ask the scheduler for requests, if we didn't have residual
	200	* requests from the dispatch list. This is to avoid the case where
	201	* we only ever dispatch a fraction of the requests available because
	202	* of low device queue depth. Once we pull requests out of the IO
	203	* scheduler, we can no longer merge or sort them. So it's best to
	204	* leave them there for as long as we can. Mark the hw queue as
	205	* needing a restart in that case.
	206	*/
c13660a0	207	if (!list_empty(&rq_list)) {
bd166ef1	208	blk_mq_sched_mark_restart(hctx);
64765a75 JA	209	did_work = blk_mq_dispatch_rq_list(hctx, &rq_list);
64765a75 JA	210	} else if (!has_sched_dispatch) {
c13660a0 JA	211	blk_mq_flush_busy_ctxs(hctx, &rq_list);
c13660a0 JA	212	blk_mq_dispatch_rq_list(hctx, &rq_list);
64765a75 JA	213	}
	214
	215	/*
	216	* We want to dispatch from the scheduler if we had no work left
	217	* on the dispatch list, OR if we did have work but weren't able
	218	* to make progress.
	219	*/
	220	if (!did_work && has_sched_dispatch) {
c13660a0 JA	221	do {
	222	struct request *rq;
	223
	224	rq = e->type->ops.mq.dispatch_request(hctx);
	225	if (!rq)
	226	break;
	227	list_add(&rq->queuelist, &rq_list);
	228	} while (blk_mq_dispatch_rq_list(hctx, &rq_list));
	229	}
bd166ef1 JA	230	}
	231
	232	void blk_mq_sched_move_to_dispatch(struct blk_mq_hw_ctx *hctx,
	233	struct list_head *rq_list,
	234	struct request (get_rq)(struct blk_mq_hw_ctx *))
	235	{
	236	do {
	237	struct request *rq;
	238
	239	rq = get_rq(hctx);
	240	if (!rq)
	241	break;
	242
	243	list_add_tail(&rq->queuelist, rq_list);
	244	} while (1);
	245	}
	246	EXPORT_SYMBOL_GPL(blk_mq_sched_move_to_dispatch);
	247
e4d750c9 JA	248	bool blk_mq_sched_try_merge(struct request_queue q, struct bio bio,
e4d750c9 JA	249	struct request **merged_request)
bd166ef1 JA	250	{
bd166ef1 JA	251	struct request *rq;
bd166ef1	252
34fe7c05 CH	253	switch (elv_merge(q, &rq, bio)) {
34fe7c05 CH	254	case ELEVATOR_BACK_MERGE:
bd166ef1 JA	255	if (!blk_mq_sched_allow_merge(q, rq, bio))
bd166ef1 JA	256	return false;
34fe7c05 CH	257	if (!bio_attempt_back_merge(q, rq, bio))
	258	return false;
	259	*merged_request = attempt_back_merge(q, rq);
	260	if (!*merged_request)
	261	elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
	262	return true;
	263	case ELEVATOR_FRONT_MERGE:
bd166ef1 JA	264	if (!blk_mq_sched_allow_merge(q, rq, bio))
bd166ef1 JA	265	return false;
34fe7c05 CH	266	if (!bio_attempt_front_merge(q, rq, bio))
	267	return false;
	268	*merged_request = attempt_front_merge(q, rq);
	269	if (!*merged_request)
	270	elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
	271	return true;
	272	default:
	273	return false;
bd166ef1	274	}
bd166ef1 JA	275	}
	276	EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
	277
	278	bool __blk_mq_sched_bio_merge(struct request_queue q, struct bio bio)
	279	{
	280	struct elevator_queue *e = q->elevator;
	281
	282	if (e->type->ops.mq.bio_merge) {
	283	struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
	284	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	285
	286	blk_mq_put_ctx(ctx);
	287	return e->type->ops.mq.bio_merge(hctx, bio);
	288	}
	289
	290	return false;
	291	}
	292
	293	bool blk_mq_sched_try_insert_merge(struct request_queue q, struct request rq)
	294	{
	295	return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq);
	296	}
	297	EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge);
	298
	299	void blk_mq_sched_request_inserted(struct request *rq)
	300	{
	301	trace_block_rq_insert(rq->q, rq);
	302	}
	303	EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted);
	304
0cacba6c OS	305	static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx,
0cacba6c OS	306	struct request *rq)
bd166ef1 JA	307	{
	308	if (rq->tag == -1) {
	309	rq->rq_flags \|= RQF_SORTED;
	310	return false;
	311	}
	312
	313	/*
	314	* If we already have a real request tag, send directly to
	315	* the dispatch list.
	316	*/
	317	spin_lock(&hctx->lock);
	318	list_add(&rq->queuelist, &hctx->dispatch);
	319	spin_unlock(&hctx->lock);
	320	return true;
	321	}
bd166ef1	322
50e1dab8 JA	323	static void blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx)
	324	{
	325	if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) {
	326	clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state);
	327	if (blk_mq_hctx_has_pending(hctx))
	328	blk_mq_run_hw_queue(hctx, true);
	329	}
	330	}
	331
	332	void blk_mq_sched_restart_queues(struct blk_mq_hw_ctx *hctx)
	333	{
	334	unsigned int i;
	335
	336	if (!(hctx->flags & BLK_MQ_F_TAG_SHARED))
	337	blk_mq_sched_restart_hctx(hctx);
	338	else {
	339	struct request_queue *q = hctx->queue;
	340
	341	if (!test_bit(QUEUE_FLAG_RESTART, &q->queue_flags))
	342	return;
	343
	344	clear_bit(QUEUE_FLAG_RESTART, &q->queue_flags);
	345
	346	queue_for_each_hw_ctx(q, hctx, i)
	347	blk_mq_sched_restart_hctx(hctx);
	348	}
	349	}
	350
bd6737f1 JA	351	/*
	352	* Add flush/fua to the queue. If we fail getting a driver tag, then
	353	* punt to the requeue list. Requeue will re-invoke us from a context
	354	* that's safe to block from.
	355	*/
	356	static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx,
	357	struct request *rq, bool can_block)
	358	{
	359	if (blk_mq_get_driver_tag(rq, &hctx, can_block)) {
	360	blk_insert_flush(rq);
	361	blk_mq_run_hw_queue(hctx, true);
	362	} else
c7a571b4	363	blk_mq_add_to_requeue_list(rq, false, true);
bd6737f1 JA	364	}
	365
	366	void blk_mq_sched_insert_request(struct request *rq, bool at_head,
	367	bool run_queue, bool async, bool can_block)
	368	{
	369	struct request_queue *q = rq->q;
	370	struct elevator_queue *e = q->elevator;
	371	struct blk_mq_ctx *ctx = rq->mq_ctx;
	372	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	373
f3a8ab7d	374	if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) {
bd6737f1 JA	375	blk_mq_sched_insert_flush(hctx, rq, can_block);
	376	return;
	377	}
	378
0cacba6c OS	379	if (e && blk_mq_sched_bypass_insert(hctx, rq))
	380	goto run;
	381
bd6737f1 JA	382	if (e && e->type->ops.mq.insert_requests) {
	383	LIST_HEAD(list);
	384
	385	list_add(&rq->queuelist, &list);
	386	e->type->ops.mq.insert_requests(hctx, &list, at_head);
	387	} else {
	388	spin_lock(&ctx->lock);
	389	__blk_mq_insert_request(hctx, rq, at_head);
	390	spin_unlock(&ctx->lock);
	391	}
	392
0cacba6c	393	run:
bd6737f1 JA	394	if (run_queue)
	395	blk_mq_run_hw_queue(hctx, async);
	396	}
	397
	398	void blk_mq_sched_insert_requests(struct request_queue *q,
	399	struct blk_mq_ctx *ctx,
	400	struct list_head *list, bool run_queue_async)
	401	{
	402	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
	403	struct elevator_queue *e = hctx->queue->elevator;
	404
0cacba6c OS	405	if (e) {
	406	struct request rq, next;
	407
	408	/*
	409	* We bypass requests that already have a driver tag assigned,
	410	* which should only be flushes. Flushes are only ever inserted
	411	* as single requests, so we shouldn't ever hit the
	412	* WARN_ON_ONCE() below (but let's handle it just in case).
	413	*/
	414	list_for_each_entry_safe(rq, next, list, queuelist) {
	415	if (WARN_ON_ONCE(rq->tag != -1)) {
	416	list_del_init(&rq->queuelist);
	417	blk_mq_sched_bypass_insert(hctx, rq);
	418	}
	419	}
	420	}
	421
bd6737f1 JA	422	if (e && e->type->ops.mq.insert_requests)
	423	e->type->ops.mq.insert_requests(hctx, list, false);
	424	else
	425	blk_mq_insert_requests(hctx, ctx, list);
	426
	427	blk_mq_run_hw_queue(hctx, run_queue_async);
	428	}
	429
bd166ef1 JA	430	static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set,
	431	struct blk_mq_hw_ctx *hctx,
	432	unsigned int hctx_idx)
	433	{
	434	if (hctx->sched_tags) {
	435	blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx);
	436	blk_mq_free_rq_map(hctx->sched_tags);
	437	hctx->sched_tags = NULL;
	438	}
	439	}
	440
	441	int blk_mq_sched_setup(struct request_queue *q)
	442	{
	443	struct blk_mq_tag_set *set = q->tag_set;
	444	struct blk_mq_hw_ctx *hctx;
	445	int ret, i;
	446
	447	/*
	448	* Default to 256, since we don't split into sync/async like the
	449	* old code did. Additionally, this is a per-hw queue depth.
	450	*/
	451	q->nr_requests = 2 * BLKDEV_MAX_RQ;
	452
	453	/*
	454	* We're switching to using an IO scheduler, so setup the hctx
	455	* scheduler tags and switch the request map from the regular
	456	* tags to scheduler tags. First allocate what we need, so we
	457	* can safely fail and fallback, if needed.
	458	*/
	459	ret = 0;
	460	queue_for_each_hw_ctx(q, hctx, i) {
	461	hctx->sched_tags = blk_mq_alloc_rq_map(set, i, q->nr_requests, 0);
	462	if (!hctx->sched_tags) {
	463	ret = -ENOMEM;
	464	break;
	465	}
	466	ret = blk_mq_alloc_rqs(set, hctx->sched_tags, i, q->nr_requests);
	467	if (ret)
	468	break;
	469	}
	470
	471	/*
	472	* If we failed, free what we did allocate
	473	*/
	474	if (ret) {
	475	queue_for_each_hw_ctx(q, hctx, i) {
	476	if (!hctx->sched_tags)
	477	continue;
	478	blk_mq_sched_free_tags(set, hctx, i);
	479	}
	480
	481	return ret;
	482	}
	483
	484	return 0;
	485	}
	486
	487	void blk_mq_sched_teardown(struct request_queue *q)
	488	{
	489	struct blk_mq_tag_set *set = q->tag_set;
	490	struct blk_mq_hw_ctx *hctx;
	491	int i;
	492
	493	queue_for_each_hw_ctx(q, hctx, i)
494	blk_mq_sched_free_tags(set, hctx, i);
495	}
d3484991 JA	496
	497	int blk_mq_sched_init(struct request_queue *q)
	498	{
	499	int ret;
	500
d3484991 JA	501	mutex_lock(&q->sysfs_lock);
	502	ret = elevator_init(q, NULL);
	503	mutex_unlock(&q->sysfs_lock);
	504
	505	return ret;
	506	}