xref: /linux/block/mq-deadline.c (revision c6fbb759)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  MQ Deadline i/o scheduler - adaptation of the legacy deadline scheduler,
4  *  for the blk-mq scheduling framework
5  *
6  *  Copyright (C) 2016 Jens Axboe <axboe@kernel.dk>
7  */
8 #include <linux/kernel.h>
9 #include <linux/fs.h>
10 #include <linux/blkdev.h>
11 #include <linux/blk-mq.h>
12 #include <linux/bio.h>
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/init.h>
16 #include <linux/compiler.h>
17 #include <linux/rbtree.h>
18 #include <linux/sbitmap.h>
19 
20 #include <trace/events/block.h>
21 
22 #include "elevator.h"
23 #include "blk.h"
24 #include "blk-mq.h"
25 #include "blk-mq-debugfs.h"
26 #include "blk-mq-tag.h"
27 #include "blk-mq-sched.h"
28 
29 /*
30  * See Documentation/block/deadline-iosched.rst
31  */
32 static const int read_expire = HZ / 2;  /* max time before a read is submitted. */
33 static const int write_expire = 5 * HZ; /* ditto for writes, these limits are SOFT! */
34 /*
35  * Time after which to dispatch lower priority requests even if higher
36  * priority requests are pending.
37  */
38 static const int prio_aging_expire = 10 * HZ;
39 static const int writes_starved = 2;    /* max times reads can starve a write */
40 static const int fifo_batch = 16;       /* # of sequential requests treated as one
41 				     by the above parameters. For throughput. */
42 
43 enum dd_data_dir {
44 	DD_READ		= READ,
45 	DD_WRITE	= WRITE,
46 };
47 
48 enum { DD_DIR_COUNT = 2 };
49 
50 enum dd_prio {
51 	DD_RT_PRIO	= 0,
52 	DD_BE_PRIO	= 1,
53 	DD_IDLE_PRIO	= 2,
54 	DD_PRIO_MAX	= 2,
55 };
56 
57 enum { DD_PRIO_COUNT = 3 };
58 
59 /*
60  * I/O statistics per I/O priority. It is fine if these counters overflow.
61  * What matters is that these counters are at least as wide as
62  * log2(max_outstanding_requests).
63  */
64 struct io_stats_per_prio {
65 	uint32_t inserted;
66 	uint32_t merged;
67 	uint32_t dispatched;
68 	atomic_t completed;
69 };
70 
71 /*
72  * Deadline scheduler data per I/O priority (enum dd_prio). Requests are
73  * present on both sort_list[] and fifo_list[].
74  */
75 struct dd_per_prio {
76 	struct list_head dispatch;
77 	struct rb_root sort_list[DD_DIR_COUNT];
78 	struct list_head fifo_list[DD_DIR_COUNT];
79 	/* Next request in FIFO order. Read, write or both are NULL. */
80 	struct request *next_rq[DD_DIR_COUNT];
81 	struct io_stats_per_prio stats;
82 };
83 
84 struct deadline_data {
85 	/*
86 	 * run time data
87 	 */
88 
89 	struct dd_per_prio per_prio[DD_PRIO_COUNT];
90 
91 	/* Data direction of latest dispatched request. */
92 	enum dd_data_dir last_dir;
93 	unsigned int batching;		/* number of sequential requests made */
94 	unsigned int starved;		/* times reads have starved writes */
95 
96 	/*
97 	 * settings that change how the i/o scheduler behaves
98 	 */
99 	int fifo_expire[DD_DIR_COUNT];
100 	int fifo_batch;
101 	int writes_starved;
102 	int front_merges;
103 	u32 async_depth;
104 	int prio_aging_expire;
105 
106 	spinlock_t lock;
107 	spinlock_t zone_lock;
108 };
109 
110 /* Maps an I/O priority class to a deadline scheduler priority. */
111 static const enum dd_prio ioprio_class_to_prio[] = {
112 	[IOPRIO_CLASS_NONE]	= DD_BE_PRIO,
113 	[IOPRIO_CLASS_RT]	= DD_RT_PRIO,
114 	[IOPRIO_CLASS_BE]	= DD_BE_PRIO,
115 	[IOPRIO_CLASS_IDLE]	= DD_IDLE_PRIO,
116 };
117 
118 static inline struct rb_root *
119 deadline_rb_root(struct dd_per_prio *per_prio, struct request *rq)
120 {
121 	return &per_prio->sort_list[rq_data_dir(rq)];
122 }
123 
124 /*
125  * Returns the I/O priority class (IOPRIO_CLASS_*) that has been assigned to a
126  * request.
127  */
128 static u8 dd_rq_ioclass(struct request *rq)
129 {
130 	return IOPRIO_PRIO_CLASS(req_get_ioprio(rq));
131 }
132 
133 /*
134  * get the request after `rq' in sector-sorted order
135  */
136 static inline struct request *
137 deadline_latter_request(struct request *rq)
138 {
139 	struct rb_node *node = rb_next(&rq->rb_node);
140 
141 	if (node)
142 		return rb_entry_rq(node);
143 
144 	return NULL;
145 }
146 
147 static void
148 deadline_add_rq_rb(struct dd_per_prio *per_prio, struct request *rq)
149 {
150 	struct rb_root *root = deadline_rb_root(per_prio, rq);
151 
152 	elv_rb_add(root, rq);
153 }
154 
155 static inline void
156 deadline_del_rq_rb(struct dd_per_prio *per_prio, struct request *rq)
157 {
158 	const enum dd_data_dir data_dir = rq_data_dir(rq);
159 
160 	if (per_prio->next_rq[data_dir] == rq)
161 		per_prio->next_rq[data_dir] = deadline_latter_request(rq);
162 
163 	elv_rb_del(deadline_rb_root(per_prio, rq), rq);
164 }
165 
166 /*
167  * remove rq from rbtree and fifo.
168  */
169 static void deadline_remove_request(struct request_queue *q,
170 				    struct dd_per_prio *per_prio,
171 				    struct request *rq)
172 {
173 	list_del_init(&rq->queuelist);
174 
175 	/*
176 	 * We might not be on the rbtree, if we are doing an insert merge
177 	 */
178 	if (!RB_EMPTY_NODE(&rq->rb_node))
179 		deadline_del_rq_rb(per_prio, rq);
180 
181 	elv_rqhash_del(q, rq);
182 	if (q->last_merge == rq)
183 		q->last_merge = NULL;
184 }
185 
186 static void dd_request_merged(struct request_queue *q, struct request *req,
187 			      enum elv_merge type)
188 {
189 	struct deadline_data *dd = q->elevator->elevator_data;
190 	const u8 ioprio_class = dd_rq_ioclass(req);
191 	const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
192 	struct dd_per_prio *per_prio = &dd->per_prio[prio];
193 
194 	/*
195 	 * if the merge was a front merge, we need to reposition request
196 	 */
197 	if (type == ELEVATOR_FRONT_MERGE) {
198 		elv_rb_del(deadline_rb_root(per_prio, req), req);
199 		deadline_add_rq_rb(per_prio, req);
200 	}
201 }
202 
203 /*
204  * Callback function that is invoked after @next has been merged into @req.
205  */
206 static void dd_merged_requests(struct request_queue *q, struct request *req,
207 			       struct request *next)
208 {
209 	struct deadline_data *dd = q->elevator->elevator_data;
210 	const u8 ioprio_class = dd_rq_ioclass(next);
211 	const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
212 
213 	lockdep_assert_held(&dd->lock);
214 
215 	dd->per_prio[prio].stats.merged++;
216 
217 	/*
218 	 * if next expires before rq, assign its expire time to rq
219 	 * and move into next position (next will be deleted) in fifo
220 	 */
221 	if (!list_empty(&req->queuelist) && !list_empty(&next->queuelist)) {
222 		if (time_before((unsigned long)next->fifo_time,
223 				(unsigned long)req->fifo_time)) {
224 			list_move(&req->queuelist, &next->queuelist);
225 			req->fifo_time = next->fifo_time;
226 		}
227 	}
228 
229 	/*
230 	 * kill knowledge of next, this one is a goner
231 	 */
232 	deadline_remove_request(q, &dd->per_prio[prio], next);
233 }
234 
235 /*
236  * move an entry to dispatch queue
237  */
238 static void
239 deadline_move_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
240 		      struct request *rq)
241 {
242 	const enum dd_data_dir data_dir = rq_data_dir(rq);
243 
244 	per_prio->next_rq[data_dir] = deadline_latter_request(rq);
245 
246 	/*
247 	 * take it off the sort and fifo list
248 	 */
249 	deadline_remove_request(rq->q, per_prio, rq);
250 }
251 
252 /* Number of requests queued for a given priority level. */
253 static u32 dd_queued(struct deadline_data *dd, enum dd_prio prio)
254 {
255 	const struct io_stats_per_prio *stats = &dd->per_prio[prio].stats;
256 
257 	lockdep_assert_held(&dd->lock);
258 
259 	return stats->inserted - atomic_read(&stats->completed);
260 }
261 
262 /*
263  * deadline_check_fifo returns 0 if there are no expired requests on the fifo,
264  * 1 otherwise. Requires !list_empty(&dd->fifo_list[data_dir])
265  */
266 static inline int deadline_check_fifo(struct dd_per_prio *per_prio,
267 				      enum dd_data_dir data_dir)
268 {
269 	struct request *rq = rq_entry_fifo(per_prio->fifo_list[data_dir].next);
270 
271 	/*
272 	 * rq is expired!
273 	 */
274 	if (time_after_eq(jiffies, (unsigned long)rq->fifo_time))
275 		return 1;
276 
277 	return 0;
278 }
279 
280 /*
281  * For the specified data direction, return the next request to
282  * dispatch using arrival ordered lists.
283  */
284 static struct request *
285 deadline_fifo_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
286 		      enum dd_data_dir data_dir)
287 {
288 	struct request *rq;
289 	unsigned long flags;
290 
291 	if (list_empty(&per_prio->fifo_list[data_dir]))
292 		return NULL;
293 
294 	rq = rq_entry_fifo(per_prio->fifo_list[data_dir].next);
295 	if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
296 		return rq;
297 
298 	/*
299 	 * Look for a write request that can be dispatched, that is one with
300 	 * an unlocked target zone.
301 	 */
302 	spin_lock_irqsave(&dd->zone_lock, flags);
303 	list_for_each_entry(rq, &per_prio->fifo_list[DD_WRITE], queuelist) {
304 		if (blk_req_can_dispatch_to_zone(rq))
305 			goto out;
306 	}
307 	rq = NULL;
308 out:
309 	spin_unlock_irqrestore(&dd->zone_lock, flags);
310 
311 	return rq;
312 }
313 
314 /*
315  * For the specified data direction, return the next request to
316  * dispatch using sector position sorted lists.
317  */
318 static struct request *
319 deadline_next_request(struct deadline_data *dd, struct dd_per_prio *per_prio,
320 		      enum dd_data_dir data_dir)
321 {
322 	struct request *rq;
323 	unsigned long flags;
324 
325 	rq = per_prio->next_rq[data_dir];
326 	if (!rq)
327 		return NULL;
328 
329 	if (data_dir == DD_READ || !blk_queue_is_zoned(rq->q))
330 		return rq;
331 
332 	/*
333 	 * Look for a write request that can be dispatched, that is one with
334 	 * an unlocked target zone.
335 	 */
336 	spin_lock_irqsave(&dd->zone_lock, flags);
337 	while (rq) {
338 		if (blk_req_can_dispatch_to_zone(rq))
339 			break;
340 		rq = deadline_latter_request(rq);
341 	}
342 	spin_unlock_irqrestore(&dd->zone_lock, flags);
343 
344 	return rq;
345 }
346 
347 /*
348  * Returns true if and only if @rq started after @latest_start where
349  * @latest_start is in jiffies.
350  */
351 static bool started_after(struct deadline_data *dd, struct request *rq,
352 			  unsigned long latest_start)
353 {
354 	unsigned long start_time = (unsigned long)rq->fifo_time;
355 
356 	start_time -= dd->fifo_expire[rq_data_dir(rq)];
357 
358 	return time_after(start_time, latest_start);
359 }
360 
361 /*
362  * deadline_dispatch_requests selects the best request according to
363  * read/write expire, fifo_batch, etc and with a start time <= @latest_start.
364  */
365 static struct request *__dd_dispatch_request(struct deadline_data *dd,
366 					     struct dd_per_prio *per_prio,
367 					     unsigned long latest_start)
368 {
369 	struct request *rq, *next_rq;
370 	enum dd_data_dir data_dir;
371 	enum dd_prio prio;
372 	u8 ioprio_class;
373 
374 	lockdep_assert_held(&dd->lock);
375 
376 	if (!list_empty(&per_prio->dispatch)) {
377 		rq = list_first_entry(&per_prio->dispatch, struct request,
378 				      queuelist);
379 		if (started_after(dd, rq, latest_start))
380 			return NULL;
381 		list_del_init(&rq->queuelist);
382 		goto done;
383 	}
384 
385 	/*
386 	 * batches are currently reads XOR writes
387 	 */
388 	rq = deadline_next_request(dd, per_prio, dd->last_dir);
389 	if (rq && dd->batching < dd->fifo_batch)
390 		/* we have a next request are still entitled to batch */
391 		goto dispatch_request;
392 
393 	/*
394 	 * at this point we are not running a batch. select the appropriate
395 	 * data direction (read / write)
396 	 */
397 
398 	if (!list_empty(&per_prio->fifo_list[DD_READ])) {
399 		BUG_ON(RB_EMPTY_ROOT(&per_prio->sort_list[DD_READ]));
400 
401 		if (deadline_fifo_request(dd, per_prio, DD_WRITE) &&
402 		    (dd->starved++ >= dd->writes_starved))
403 			goto dispatch_writes;
404 
405 		data_dir = DD_READ;
406 
407 		goto dispatch_find_request;
408 	}
409 
410 	/*
411 	 * there are either no reads or writes have been starved
412 	 */
413 
414 	if (!list_empty(&per_prio->fifo_list[DD_WRITE])) {
415 dispatch_writes:
416 		BUG_ON(RB_EMPTY_ROOT(&per_prio->sort_list[DD_WRITE]));
417 
418 		dd->starved = 0;
419 
420 		data_dir = DD_WRITE;
421 
422 		goto dispatch_find_request;
423 	}
424 
425 	return NULL;
426 
427 dispatch_find_request:
428 	/*
429 	 * we are not running a batch, find best request for selected data_dir
430 	 */
431 	next_rq = deadline_next_request(dd, per_prio, data_dir);
432 	if (deadline_check_fifo(per_prio, data_dir) || !next_rq) {
433 		/*
434 		 * A deadline has expired, the last request was in the other
435 		 * direction, or we have run out of higher-sectored requests.
436 		 * Start again from the request with the earliest expiry time.
437 		 */
438 		rq = deadline_fifo_request(dd, per_prio, data_dir);
439 	} else {
440 		/*
441 		 * The last req was the same dir and we have a next request in
442 		 * sort order. No expired requests so continue on from here.
443 		 */
444 		rq = next_rq;
445 	}
446 
447 	/*
448 	 * For a zoned block device, if we only have writes queued and none of
449 	 * them can be dispatched, rq will be NULL.
450 	 */
451 	if (!rq)
452 		return NULL;
453 
454 	dd->last_dir = data_dir;
455 	dd->batching = 0;
456 
457 dispatch_request:
458 	if (started_after(dd, rq, latest_start))
459 		return NULL;
460 
461 	/*
462 	 * rq is the selected appropriate request.
463 	 */
464 	dd->batching++;
465 	deadline_move_request(dd, per_prio, rq);
466 done:
467 	ioprio_class = dd_rq_ioclass(rq);
468 	prio = ioprio_class_to_prio[ioprio_class];
469 	dd->per_prio[prio].stats.dispatched++;
470 	/*
471 	 * If the request needs its target zone locked, do it.
472 	 */
473 	blk_req_zone_write_lock(rq);
474 	rq->rq_flags |= RQF_STARTED;
475 	return rq;
476 }
477 
478 /*
479  * Check whether there are any requests with priority other than DD_RT_PRIO
480  * that were inserted more than prio_aging_expire jiffies ago.
481  */
482 static struct request *dd_dispatch_prio_aged_requests(struct deadline_data *dd,
483 						      unsigned long now)
484 {
485 	struct request *rq;
486 	enum dd_prio prio;
487 	int prio_cnt;
488 
489 	lockdep_assert_held(&dd->lock);
490 
491 	prio_cnt = !!dd_queued(dd, DD_RT_PRIO) + !!dd_queued(dd, DD_BE_PRIO) +
492 		   !!dd_queued(dd, DD_IDLE_PRIO);
493 	if (prio_cnt < 2)
494 		return NULL;
495 
496 	for (prio = DD_BE_PRIO; prio <= DD_PRIO_MAX; prio++) {
497 		rq = __dd_dispatch_request(dd, &dd->per_prio[prio],
498 					   now - dd->prio_aging_expire);
499 		if (rq)
500 			return rq;
501 	}
502 
503 	return NULL;
504 }
505 
506 /*
507  * Called from blk_mq_run_hw_queue() -> __blk_mq_sched_dispatch_requests().
508  *
509  * One confusing aspect here is that we get called for a specific
510  * hardware queue, but we may return a request that is for a
511  * different hardware queue. This is because mq-deadline has shared
512  * state for all hardware queues, in terms of sorting, FIFOs, etc.
513  */
514 static struct request *dd_dispatch_request(struct blk_mq_hw_ctx *hctx)
515 {
516 	struct deadline_data *dd = hctx->queue->elevator->elevator_data;
517 	const unsigned long now = jiffies;
518 	struct request *rq;
519 	enum dd_prio prio;
520 
521 	spin_lock(&dd->lock);
522 	rq = dd_dispatch_prio_aged_requests(dd, now);
523 	if (rq)
524 		goto unlock;
525 
526 	/*
527 	 * Next, dispatch requests in priority order. Ignore lower priority
528 	 * requests if any higher priority requests are pending.
529 	 */
530 	for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
531 		rq = __dd_dispatch_request(dd, &dd->per_prio[prio], now);
532 		if (rq || dd_queued(dd, prio))
533 			break;
534 	}
535 
536 unlock:
537 	spin_unlock(&dd->lock);
538 
539 	return rq;
540 }
541 
542 /*
543  * Called by __blk_mq_alloc_request(). The shallow_depth value set by this
544  * function is used by __blk_mq_get_tag().
545  */
546 static void dd_limit_depth(blk_opf_t opf, struct blk_mq_alloc_data *data)
547 {
548 	struct deadline_data *dd = data->q->elevator->elevator_data;
549 
550 	/* Do not throttle synchronous reads. */
551 	if (op_is_sync(opf) && !op_is_write(opf))
552 		return;
553 
554 	/*
555 	 * Throttle asynchronous requests and writes such that these requests
556 	 * do not block the allocation of synchronous requests.
557 	 */
558 	data->shallow_depth = dd->async_depth;
559 }
560 
561 /* Called by blk_mq_update_nr_requests(). */
562 static void dd_depth_updated(struct blk_mq_hw_ctx *hctx)
563 {
564 	struct request_queue *q = hctx->queue;
565 	struct deadline_data *dd = q->elevator->elevator_data;
566 	struct blk_mq_tags *tags = hctx->sched_tags;
567 
568 	dd->async_depth = max(1UL, 3 * q->nr_requests / 4);
569 
570 	sbitmap_queue_min_shallow_depth(&tags->bitmap_tags, dd->async_depth);
571 }
572 
573 /* Called by blk_mq_init_hctx() and blk_mq_init_sched(). */
574 static int dd_init_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx)
575 {
576 	dd_depth_updated(hctx);
577 	return 0;
578 }
579 
580 static void dd_exit_sched(struct elevator_queue *e)
581 {
582 	struct deadline_data *dd = e->elevator_data;
583 	enum dd_prio prio;
584 
585 	for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
586 		struct dd_per_prio *per_prio = &dd->per_prio[prio];
587 		const struct io_stats_per_prio *stats = &per_prio->stats;
588 		uint32_t queued;
589 
590 		WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_READ]));
591 		WARN_ON_ONCE(!list_empty(&per_prio->fifo_list[DD_WRITE]));
592 
593 		spin_lock(&dd->lock);
594 		queued = dd_queued(dd, prio);
595 		spin_unlock(&dd->lock);
596 
597 		WARN_ONCE(queued != 0,
598 			  "statistics for priority %d: i %u m %u d %u c %u\n",
599 			  prio, stats->inserted, stats->merged,
600 			  stats->dispatched, atomic_read(&stats->completed));
601 	}
602 
603 	kfree(dd);
604 }
605 
606 /*
607  * initialize elevator private data (deadline_data).
608  */
609 static int dd_init_sched(struct request_queue *q, struct elevator_type *e)
610 {
611 	struct deadline_data *dd;
612 	struct elevator_queue *eq;
613 	enum dd_prio prio;
614 	int ret = -ENOMEM;
615 
616 	eq = elevator_alloc(q, e);
617 	if (!eq)
618 		return ret;
619 
620 	dd = kzalloc_node(sizeof(*dd), GFP_KERNEL, q->node);
621 	if (!dd)
622 		goto put_eq;
623 
624 	eq->elevator_data = dd;
625 
626 	for (prio = 0; prio <= DD_PRIO_MAX; prio++) {
627 		struct dd_per_prio *per_prio = &dd->per_prio[prio];
628 
629 		INIT_LIST_HEAD(&per_prio->dispatch);
630 		INIT_LIST_HEAD(&per_prio->fifo_list[DD_READ]);
631 		INIT_LIST_HEAD(&per_prio->fifo_list[DD_WRITE]);
632 		per_prio->sort_list[DD_READ] = RB_ROOT;
633 		per_prio->sort_list[DD_WRITE] = RB_ROOT;
634 	}
635 	dd->fifo_expire[DD_READ] = read_expire;
636 	dd->fifo_expire[DD_WRITE] = write_expire;
637 	dd->writes_starved = writes_starved;
638 	dd->front_merges = 1;
639 	dd->last_dir = DD_WRITE;
640 	dd->fifo_batch = fifo_batch;
641 	dd->prio_aging_expire = prio_aging_expire;
642 	spin_lock_init(&dd->lock);
643 	spin_lock_init(&dd->zone_lock);
644 
645 	/* We dispatch from request queue wide instead of hw queue */
646 	blk_queue_flag_set(QUEUE_FLAG_SQ_SCHED, q);
647 
648 	q->elevator = eq;
649 	return 0;
650 
651 put_eq:
652 	kobject_put(&eq->kobj);
653 	return ret;
654 }
655 
656 /*
657  * Try to merge @bio into an existing request. If @bio has been merged into
658  * an existing request, store the pointer to that request into *@rq.
659  */
660 static int dd_request_merge(struct request_queue *q, struct request **rq,
661 			    struct bio *bio)
662 {
663 	struct deadline_data *dd = q->elevator->elevator_data;
664 	const u8 ioprio_class = IOPRIO_PRIO_CLASS(bio->bi_ioprio);
665 	const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
666 	struct dd_per_prio *per_prio = &dd->per_prio[prio];
667 	sector_t sector = bio_end_sector(bio);
668 	struct request *__rq;
669 
670 	if (!dd->front_merges)
671 		return ELEVATOR_NO_MERGE;
672 
673 	__rq = elv_rb_find(&per_prio->sort_list[bio_data_dir(bio)], sector);
674 	if (__rq) {
675 		BUG_ON(sector != blk_rq_pos(__rq));
676 
677 		if (elv_bio_merge_ok(__rq, bio)) {
678 			*rq = __rq;
679 			if (blk_discard_mergable(__rq))
680 				return ELEVATOR_DISCARD_MERGE;
681 			return ELEVATOR_FRONT_MERGE;
682 		}
683 	}
684 
685 	return ELEVATOR_NO_MERGE;
686 }
687 
688 /*
689  * Attempt to merge a bio into an existing request. This function is called
690  * before @bio is associated with a request.
691  */
692 static bool dd_bio_merge(struct request_queue *q, struct bio *bio,
693 		unsigned int nr_segs)
694 {
695 	struct deadline_data *dd = q->elevator->elevator_data;
696 	struct request *free = NULL;
697 	bool ret;
698 
699 	spin_lock(&dd->lock);
700 	ret = blk_mq_sched_try_merge(q, bio, nr_segs, &free);
701 	spin_unlock(&dd->lock);
702 
703 	if (free)
704 		blk_mq_free_request(free);
705 
706 	return ret;
707 }
708 
709 /*
710  * add rq to rbtree and fifo
711  */
712 static void dd_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
713 			      bool at_head)
714 {
715 	struct request_queue *q = hctx->queue;
716 	struct deadline_data *dd = q->elevator->elevator_data;
717 	const enum dd_data_dir data_dir = rq_data_dir(rq);
718 	u16 ioprio = req_get_ioprio(rq);
719 	u8 ioprio_class = IOPRIO_PRIO_CLASS(ioprio);
720 	struct dd_per_prio *per_prio;
721 	enum dd_prio prio;
722 	LIST_HEAD(free);
723 
724 	lockdep_assert_held(&dd->lock);
725 
726 	/*
727 	 * This may be a requeue of a write request that has locked its
728 	 * target zone. If it is the case, this releases the zone lock.
729 	 */
730 	blk_req_zone_write_unlock(rq);
731 
732 	prio = ioprio_class_to_prio[ioprio_class];
733 	per_prio = &dd->per_prio[prio];
734 	if (!rq->elv.priv[0]) {
735 		per_prio->stats.inserted++;
736 		rq->elv.priv[0] = (void *)(uintptr_t)1;
737 	}
738 
739 	if (blk_mq_sched_try_insert_merge(q, rq, &free)) {
740 		blk_mq_free_requests(&free);
741 		return;
742 	}
743 
744 	trace_block_rq_insert(rq);
745 
746 	if (at_head) {
747 		list_add(&rq->queuelist, &per_prio->dispatch);
748 		rq->fifo_time = jiffies;
749 	} else {
750 		deadline_add_rq_rb(per_prio, rq);
751 
752 		if (rq_mergeable(rq)) {
753 			elv_rqhash_add(q, rq);
754 			if (!q->last_merge)
755 				q->last_merge = rq;
756 		}
757 
758 		/*
759 		 * set expire time and add to fifo list
760 		 */
761 		rq->fifo_time = jiffies + dd->fifo_expire[data_dir];
762 		list_add_tail(&rq->queuelist, &per_prio->fifo_list[data_dir]);
763 	}
764 }
765 
766 /*
767  * Called from blk_mq_sched_insert_request() or blk_mq_sched_insert_requests().
768  */
769 static void dd_insert_requests(struct blk_mq_hw_ctx *hctx,
770 			       struct list_head *list, bool at_head)
771 {
772 	struct request_queue *q = hctx->queue;
773 	struct deadline_data *dd = q->elevator->elevator_data;
774 
775 	spin_lock(&dd->lock);
776 	while (!list_empty(list)) {
777 		struct request *rq;
778 
779 		rq = list_first_entry(list, struct request, queuelist);
780 		list_del_init(&rq->queuelist);
781 		dd_insert_request(hctx, rq, at_head);
782 	}
783 	spin_unlock(&dd->lock);
784 }
785 
786 /* Callback from inside blk_mq_rq_ctx_init(). */
787 static void dd_prepare_request(struct request *rq)
788 {
789 	rq->elv.priv[0] = NULL;
790 }
791 
792 /*
793  * Callback from inside blk_mq_free_request().
794  *
795  * For zoned block devices, write unlock the target zone of
796  * completed write requests. Do this while holding the zone lock
797  * spinlock so that the zone is never unlocked while deadline_fifo_request()
798  * or deadline_next_request() are executing. This function is called for
799  * all requests, whether or not these requests complete successfully.
800  *
801  * For a zoned block device, __dd_dispatch_request() may have stopped
802  * dispatching requests if all the queued requests are write requests directed
803  * at zones that are already locked due to on-going write requests. To ensure
804  * write request dispatch progress in this case, mark the queue as needing a
805  * restart to ensure that the queue is run again after completion of the
806  * request and zones being unlocked.
807  */
808 static void dd_finish_request(struct request *rq)
809 {
810 	struct request_queue *q = rq->q;
811 	struct deadline_data *dd = q->elevator->elevator_data;
812 	const u8 ioprio_class = dd_rq_ioclass(rq);
813 	const enum dd_prio prio = ioprio_class_to_prio[ioprio_class];
814 	struct dd_per_prio *per_prio = &dd->per_prio[prio];
815 
816 	/*
817 	 * The block layer core may call dd_finish_request() without having
818 	 * called dd_insert_requests(). Skip requests that bypassed I/O
819 	 * scheduling. See also blk_mq_request_bypass_insert().
820 	 */
821 	if (!rq->elv.priv[0])
822 		return;
823 
824 	atomic_inc(&per_prio->stats.completed);
825 
826 	if (blk_queue_is_zoned(q)) {
827 		unsigned long flags;
828 
829 		spin_lock_irqsave(&dd->zone_lock, flags);
830 		blk_req_zone_write_unlock(rq);
831 		if (!list_empty(&per_prio->fifo_list[DD_WRITE]))
832 			blk_mq_sched_mark_restart_hctx(rq->mq_hctx);
833 		spin_unlock_irqrestore(&dd->zone_lock, flags);
834 	}
835 }
836 
837 static bool dd_has_work_for_prio(struct dd_per_prio *per_prio)
838 {
839 	return !list_empty_careful(&per_prio->dispatch) ||
840 		!list_empty_careful(&per_prio->fifo_list[DD_READ]) ||
841 		!list_empty_careful(&per_prio->fifo_list[DD_WRITE]);
842 }
843 
844 static bool dd_has_work(struct blk_mq_hw_ctx *hctx)
845 {
846 	struct deadline_data *dd = hctx->queue->elevator->elevator_data;
847 	enum dd_prio prio;
848 
849 	for (prio = 0; prio <= DD_PRIO_MAX; prio++)
850 		if (dd_has_work_for_prio(&dd->per_prio[prio]))
851 			return true;
852 
853 	return false;
854 }
855 
856 /*
857  * sysfs parts below
858  */
859 #define SHOW_INT(__FUNC, __VAR)						\
860 static ssize_t __FUNC(struct elevator_queue *e, char *page)		\
861 {									\
862 	struct deadline_data *dd = e->elevator_data;			\
863 									\
864 	return sysfs_emit(page, "%d\n", __VAR);				\
865 }
866 #define SHOW_JIFFIES(__FUNC, __VAR) SHOW_INT(__FUNC, jiffies_to_msecs(__VAR))
867 SHOW_JIFFIES(deadline_read_expire_show, dd->fifo_expire[DD_READ]);
868 SHOW_JIFFIES(deadline_write_expire_show, dd->fifo_expire[DD_WRITE]);
869 SHOW_JIFFIES(deadline_prio_aging_expire_show, dd->prio_aging_expire);
870 SHOW_INT(deadline_writes_starved_show, dd->writes_starved);
871 SHOW_INT(deadline_front_merges_show, dd->front_merges);
872 SHOW_INT(deadline_async_depth_show, dd->async_depth);
873 SHOW_INT(deadline_fifo_batch_show, dd->fifo_batch);
874 #undef SHOW_INT
875 #undef SHOW_JIFFIES
876 
877 #define STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, __CONV)			\
878 static ssize_t __FUNC(struct elevator_queue *e, const char *page, size_t count)	\
879 {									\
880 	struct deadline_data *dd = e->elevator_data;			\
881 	int __data, __ret;						\
882 									\
883 	__ret = kstrtoint(page, 0, &__data);				\
884 	if (__ret < 0)							\
885 		return __ret;						\
886 	if (__data < (MIN))						\
887 		__data = (MIN);						\
888 	else if (__data > (MAX))					\
889 		__data = (MAX);						\
890 	*(__PTR) = __CONV(__data);					\
891 	return count;							\
892 }
893 #define STORE_INT(__FUNC, __PTR, MIN, MAX)				\
894 	STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, )
895 #define STORE_JIFFIES(__FUNC, __PTR, MIN, MAX)				\
896 	STORE_FUNCTION(__FUNC, __PTR, MIN, MAX, msecs_to_jiffies)
897 STORE_JIFFIES(deadline_read_expire_store, &dd->fifo_expire[DD_READ], 0, INT_MAX);
898 STORE_JIFFIES(deadline_write_expire_store, &dd->fifo_expire[DD_WRITE], 0, INT_MAX);
899 STORE_JIFFIES(deadline_prio_aging_expire_store, &dd->prio_aging_expire, 0, INT_MAX);
900 STORE_INT(deadline_writes_starved_store, &dd->writes_starved, INT_MIN, INT_MAX);
901 STORE_INT(deadline_front_merges_store, &dd->front_merges, 0, 1);
902 STORE_INT(deadline_async_depth_store, &dd->async_depth, 1, INT_MAX);
903 STORE_INT(deadline_fifo_batch_store, &dd->fifo_batch, 0, INT_MAX);
904 #undef STORE_FUNCTION
905 #undef STORE_INT
906 #undef STORE_JIFFIES
907 
908 #define DD_ATTR(name) \
909 	__ATTR(name, 0644, deadline_##name##_show, deadline_##name##_store)
910 
911 static struct elv_fs_entry deadline_attrs[] = {
912 	DD_ATTR(read_expire),
913 	DD_ATTR(write_expire),
914 	DD_ATTR(writes_starved),
915 	DD_ATTR(front_merges),
916 	DD_ATTR(async_depth),
917 	DD_ATTR(fifo_batch),
918 	DD_ATTR(prio_aging_expire),
919 	__ATTR_NULL
920 };
921 
922 #ifdef CONFIG_BLK_DEBUG_FS
923 #define DEADLINE_DEBUGFS_DDIR_ATTRS(prio, data_dir, name)		\
924 static void *deadline_##name##_fifo_start(struct seq_file *m,		\
925 					  loff_t *pos)			\
926 	__acquires(&dd->lock)						\
927 {									\
928 	struct request_queue *q = m->private;				\
929 	struct deadline_data *dd = q->elevator->elevator_data;		\
930 	struct dd_per_prio *per_prio = &dd->per_prio[prio];		\
931 									\
932 	spin_lock(&dd->lock);						\
933 	return seq_list_start(&per_prio->fifo_list[data_dir], *pos);	\
934 }									\
935 									\
936 static void *deadline_##name##_fifo_next(struct seq_file *m, void *v,	\
937 					 loff_t *pos)			\
938 {									\
939 	struct request_queue *q = m->private;				\
940 	struct deadline_data *dd = q->elevator->elevator_data;		\
941 	struct dd_per_prio *per_prio = &dd->per_prio[prio];		\
942 									\
943 	return seq_list_next(v, &per_prio->fifo_list[data_dir], pos);	\
944 }									\
945 									\
946 static void deadline_##name##_fifo_stop(struct seq_file *m, void *v)	\
947 	__releases(&dd->lock)						\
948 {									\
949 	struct request_queue *q = m->private;				\
950 	struct deadline_data *dd = q->elevator->elevator_data;		\
951 									\
952 	spin_unlock(&dd->lock);						\
953 }									\
954 									\
955 static const struct seq_operations deadline_##name##_fifo_seq_ops = {	\
956 	.start	= deadline_##name##_fifo_start,				\
957 	.next	= deadline_##name##_fifo_next,				\
958 	.stop	= deadline_##name##_fifo_stop,				\
959 	.show	= blk_mq_debugfs_rq_show,				\
960 };									\
961 									\
962 static int deadline_##name##_next_rq_show(void *data,			\
963 					  struct seq_file *m)		\
964 {									\
965 	struct request_queue *q = data;					\
966 	struct deadline_data *dd = q->elevator->elevator_data;		\
967 	struct dd_per_prio *per_prio = &dd->per_prio[prio];		\
968 	struct request *rq = per_prio->next_rq[data_dir];		\
969 									\
970 	if (rq)								\
971 		__blk_mq_debugfs_rq_show(m, rq);			\
972 	return 0;							\
973 }
974 
975 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_RT_PRIO, DD_READ, read0);
976 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_RT_PRIO, DD_WRITE, write0);
977 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_BE_PRIO, DD_READ, read1);
978 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_BE_PRIO, DD_WRITE, write1);
979 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_IDLE_PRIO, DD_READ, read2);
980 DEADLINE_DEBUGFS_DDIR_ATTRS(DD_IDLE_PRIO, DD_WRITE, write2);
981 #undef DEADLINE_DEBUGFS_DDIR_ATTRS
982 
983 static int deadline_batching_show(void *data, struct seq_file *m)
984 {
985 	struct request_queue *q = data;
986 	struct deadline_data *dd = q->elevator->elevator_data;
987 
988 	seq_printf(m, "%u\n", dd->batching);
989 	return 0;
990 }
991 
992 static int deadline_starved_show(void *data, struct seq_file *m)
993 {
994 	struct request_queue *q = data;
995 	struct deadline_data *dd = q->elevator->elevator_data;
996 
997 	seq_printf(m, "%u\n", dd->starved);
998 	return 0;
999 }
1000 
1001 static int dd_async_depth_show(void *data, struct seq_file *m)
1002 {
1003 	struct request_queue *q = data;
1004 	struct deadline_data *dd = q->elevator->elevator_data;
1005 
1006 	seq_printf(m, "%u\n", dd->async_depth);
1007 	return 0;
1008 }
1009 
1010 static int dd_queued_show(void *data, struct seq_file *m)
1011 {
1012 	struct request_queue *q = data;
1013 	struct deadline_data *dd = q->elevator->elevator_data;
1014 	u32 rt, be, idle;
1015 
1016 	spin_lock(&dd->lock);
1017 	rt = dd_queued(dd, DD_RT_PRIO);
1018 	be = dd_queued(dd, DD_BE_PRIO);
1019 	idle = dd_queued(dd, DD_IDLE_PRIO);
1020 	spin_unlock(&dd->lock);
1021 
1022 	seq_printf(m, "%u %u %u\n", rt, be, idle);
1023 
1024 	return 0;
1025 }
1026 
1027 /* Number of requests owned by the block driver for a given priority. */
1028 static u32 dd_owned_by_driver(struct deadline_data *dd, enum dd_prio prio)
1029 {
1030 	const struct io_stats_per_prio *stats = &dd->per_prio[prio].stats;
1031 
1032 	lockdep_assert_held(&dd->lock);
1033 
1034 	return stats->dispatched + stats->merged -
1035 		atomic_read(&stats->completed);
1036 }
1037 
1038 static int dd_owned_by_driver_show(void *data, struct seq_file *m)
1039 {
1040 	struct request_queue *q = data;
1041 	struct deadline_data *dd = q->elevator->elevator_data;
1042 	u32 rt, be, idle;
1043 
1044 	spin_lock(&dd->lock);
1045 	rt = dd_owned_by_driver(dd, DD_RT_PRIO);
1046 	be = dd_owned_by_driver(dd, DD_BE_PRIO);
1047 	idle = dd_owned_by_driver(dd, DD_IDLE_PRIO);
1048 	spin_unlock(&dd->lock);
1049 
1050 	seq_printf(m, "%u %u %u\n", rt, be, idle);
1051 
1052 	return 0;
1053 }
1054 
1055 #define DEADLINE_DISPATCH_ATTR(prio)					\
1056 static void *deadline_dispatch##prio##_start(struct seq_file *m,	\
1057 					     loff_t *pos)		\
1058 	__acquires(&dd->lock)						\
1059 {									\
1060 	struct request_queue *q = m->private;				\
1061 	struct deadline_data *dd = q->elevator->elevator_data;		\
1062 	struct dd_per_prio *per_prio = &dd->per_prio[prio];		\
1063 									\
1064 	spin_lock(&dd->lock);						\
1065 	return seq_list_start(&per_prio->dispatch, *pos);		\
1066 }									\
1067 									\
1068 static void *deadline_dispatch##prio##_next(struct seq_file *m,		\
1069 					    void *v, loff_t *pos)	\
1070 {									\
1071 	struct request_queue *q = m->private;				\
1072 	struct deadline_data *dd = q->elevator->elevator_data;		\
1073 	struct dd_per_prio *per_prio = &dd->per_prio[prio];		\
1074 									\
1075 	return seq_list_next(v, &per_prio->dispatch, pos);		\
1076 }									\
1077 									\
1078 static void deadline_dispatch##prio##_stop(struct seq_file *m, void *v)	\
1079 	__releases(&dd->lock)						\
1080 {									\
1081 	struct request_queue *q = m->private;				\
1082 	struct deadline_data *dd = q->elevator->elevator_data;		\
1083 									\
1084 	spin_unlock(&dd->lock);						\
1085 }									\
1086 									\
1087 static const struct seq_operations deadline_dispatch##prio##_seq_ops = { \
1088 	.start	= deadline_dispatch##prio##_start,			\
1089 	.next	= deadline_dispatch##prio##_next,			\
1090 	.stop	= deadline_dispatch##prio##_stop,			\
1091 	.show	= blk_mq_debugfs_rq_show,				\
1092 }
1093 
1094 DEADLINE_DISPATCH_ATTR(0);
1095 DEADLINE_DISPATCH_ATTR(1);
1096 DEADLINE_DISPATCH_ATTR(2);
1097 #undef DEADLINE_DISPATCH_ATTR
1098 
1099 #define DEADLINE_QUEUE_DDIR_ATTRS(name)					\
1100 	{#name "_fifo_list", 0400,					\
1101 			.seq_ops = &deadline_##name##_fifo_seq_ops}
1102 #define DEADLINE_NEXT_RQ_ATTR(name)					\
1103 	{#name "_next_rq", 0400, deadline_##name##_next_rq_show}
1104 static const struct blk_mq_debugfs_attr deadline_queue_debugfs_attrs[] = {
1105 	DEADLINE_QUEUE_DDIR_ATTRS(read0),
1106 	DEADLINE_QUEUE_DDIR_ATTRS(write0),
1107 	DEADLINE_QUEUE_DDIR_ATTRS(read1),
1108 	DEADLINE_QUEUE_DDIR_ATTRS(write1),
1109 	DEADLINE_QUEUE_DDIR_ATTRS(read2),
1110 	DEADLINE_QUEUE_DDIR_ATTRS(write2),
1111 	DEADLINE_NEXT_RQ_ATTR(read0),
1112 	DEADLINE_NEXT_RQ_ATTR(write0),
1113 	DEADLINE_NEXT_RQ_ATTR(read1),
1114 	DEADLINE_NEXT_RQ_ATTR(write1),
1115 	DEADLINE_NEXT_RQ_ATTR(read2),
1116 	DEADLINE_NEXT_RQ_ATTR(write2),
1117 	{"batching", 0400, deadline_batching_show},
1118 	{"starved", 0400, deadline_starved_show},
1119 	{"async_depth", 0400, dd_async_depth_show},
1120 	{"dispatch0", 0400, .seq_ops = &deadline_dispatch0_seq_ops},
1121 	{"dispatch1", 0400, .seq_ops = &deadline_dispatch1_seq_ops},
1122 	{"dispatch2", 0400, .seq_ops = &deadline_dispatch2_seq_ops},
1123 	{"owned_by_driver", 0400, dd_owned_by_driver_show},
1124 	{"queued", 0400, dd_queued_show},
1125 	{},
1126 };
1127 #undef DEADLINE_QUEUE_DDIR_ATTRS
1128 #endif
1129 
1130 static struct elevator_type mq_deadline = {
1131 	.ops = {
1132 		.depth_updated		= dd_depth_updated,
1133 		.limit_depth		= dd_limit_depth,
1134 		.insert_requests	= dd_insert_requests,
1135 		.dispatch_request	= dd_dispatch_request,
1136 		.prepare_request	= dd_prepare_request,
1137 		.finish_request		= dd_finish_request,
1138 		.next_request		= elv_rb_latter_request,
1139 		.former_request		= elv_rb_former_request,
1140 		.bio_merge		= dd_bio_merge,
1141 		.request_merge		= dd_request_merge,
1142 		.requests_merged	= dd_merged_requests,
1143 		.request_merged		= dd_request_merged,
1144 		.has_work		= dd_has_work,
1145 		.init_sched		= dd_init_sched,
1146 		.exit_sched		= dd_exit_sched,
1147 		.init_hctx		= dd_init_hctx,
1148 	},
1149 
1150 #ifdef CONFIG_BLK_DEBUG_FS
1151 	.queue_debugfs_attrs = deadline_queue_debugfs_attrs,
1152 #endif
1153 	.elevator_attrs = deadline_attrs,
1154 	.elevator_name = "mq-deadline",
1155 	.elevator_alias = "deadline",
1156 	.elevator_features = ELEVATOR_F_ZBD_SEQ_WRITE,
1157 	.elevator_owner = THIS_MODULE,
1158 };
1159 MODULE_ALIAS("mq-deadline-iosched");
1160 
1161 static int __init deadline_init(void)
1162 {
1163 	return elv_register(&mq_deadline);
1164 }
1165 
1166 static void __exit deadline_exit(void)
1167 {
1168 	elv_unregister(&mq_deadline);
1169 }
1170 
1171 module_init(deadline_init);
1172 module_exit(deadline_exit);
1173 
1174 MODULE_AUTHOR("Jens Axboe, Damien Le Moal and Bart Van Assche");
1175 MODULE_LICENSE("GPL");
1176 MODULE_DESCRIPTION("MQ deadline IO scheduler");
1177