xref: /linux/block/blk-mq-sched.c (revision e60e1ee60630cafef5e430c2ae364877e061d980)
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-debugfs.h"
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 
34 void blk_mq_sched_assign_ioc(struct request *rq, struct bio *bio)
35 {
36 	struct request_queue *q = rq->q;
37 	struct io_context *ioc = rq_ioc(bio);
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 	}
49 	get_io_context(icq->ioc);
50 	rq->elv.icq = icq;
51 }
52 
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 	return blk_mq_run_hw_queue(hctx, true);
85 }
86 
87 /*
88  * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
89  * its queue by itself in its completion handler, so we don't need to
90  * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
91  */
92 static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx)
93 {
94 	struct request_queue *q = hctx->queue;
95 	struct elevator_queue *e = q->elevator;
96 	LIST_HEAD(rq_list);
97 
98 	do {
99 		struct request *rq;
100 
101 		if (e->type->ops.mq.has_work &&
102 				!e->type->ops.mq.has_work(hctx))
103 			break;
104 
105 		if (!blk_mq_get_dispatch_budget(hctx))
106 			break;
107 
108 		rq = e->type->ops.mq.dispatch_request(hctx);
109 		if (!rq) {
110 			blk_mq_put_dispatch_budget(hctx);
111 			break;
112 		}
113 
114 		/*
115 		 * Now this rq owns the budget which has to be released
116 		 * if this rq won't be queued to driver via .queue_rq()
117 		 * in blk_mq_dispatch_rq_list().
118 		 */
119 		list_add(&rq->queuelist, &rq_list);
120 	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));
121 }
122 
123 static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx,
124 					  struct blk_mq_ctx *ctx)
125 {
126 	unsigned idx = ctx->index_hw;
127 
128 	if (++idx == hctx->nr_ctx)
129 		idx = 0;
130 
131 	return hctx->ctxs[idx];
132 }
133 
134 /*
135  * Only SCSI implements .get_budget and .put_budget, and SCSI restarts
136  * its queue by itself in its completion handler, so we don't need to
137  * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE.
138  */
139 static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx)
140 {
141 	struct request_queue *q = hctx->queue;
142 	LIST_HEAD(rq_list);
143 	struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from);
144 
145 	do {
146 		struct request *rq;
147 
148 		if (!sbitmap_any_bit_set(&hctx->ctx_map))
149 			break;
150 
151 		if (!blk_mq_get_dispatch_budget(hctx))
152 			break;
153 
154 		rq = blk_mq_dequeue_from_ctx(hctx, ctx);
155 		if (!rq) {
156 			blk_mq_put_dispatch_budget(hctx);
157 			break;
158 		}
159 
160 		/*
161 		 * Now this rq owns the budget which has to be released
162 		 * if this rq won't be queued to driver via .queue_rq()
163 		 * in blk_mq_dispatch_rq_list().
164 		 */
165 		list_add(&rq->queuelist, &rq_list);
166 
167 		/* round robin for fair dispatch */
168 		ctx = blk_mq_next_ctx(hctx, rq->mq_ctx);
169 
170 	} while (blk_mq_dispatch_rq_list(q, &rq_list, true));
171 
172 	WRITE_ONCE(hctx->dispatch_from, ctx);
173 }
174 
175 /* return true if hw queue need to be run again */
176 void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx)
177 {
178 	struct request_queue *q = hctx->queue;
179 	struct elevator_queue *e = q->elevator;
180 	const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request;
181 	LIST_HEAD(rq_list);
182 
183 	/* RCU or SRCU read lock is needed before checking quiesced flag */
184 	if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q)))
185 		return;
186 
187 	hctx->run++;
188 
189 	/*
190 	 * If we have previous entries on our dispatch list, grab them first for
191 	 * more fair dispatch.
192 	 */
193 	if (!list_empty_careful(&hctx->dispatch)) {
194 		spin_lock(&hctx->lock);
195 		if (!list_empty(&hctx->dispatch))
196 			list_splice_init(&hctx->dispatch, &rq_list);
197 		spin_unlock(&hctx->lock);
198 	}
199 
200 	/*
201 	 * Only ask the scheduler for requests, if we didn't have residual
202 	 * requests from the dispatch list. This is to avoid the case where
203 	 * we only ever dispatch a fraction of the requests available because
204 	 * of low device queue depth. Once we pull requests out of the IO
205 	 * scheduler, we can no longer merge or sort them. So it's best to
206 	 * leave them there for as long as we can. Mark the hw queue as
207 	 * needing a restart in that case.
208 	 *
209 	 * We want to dispatch from the scheduler if there was nothing
210 	 * on the dispatch list or we were able to dispatch from the
211 	 * dispatch list.
212 	 */
213 	if (!list_empty(&rq_list)) {
214 		blk_mq_sched_mark_restart_hctx(hctx);
215 		if (blk_mq_dispatch_rq_list(q, &rq_list, false)) {
216 			if (has_sched_dispatch)
217 				blk_mq_do_dispatch_sched(hctx);
218 			else
219 				blk_mq_do_dispatch_ctx(hctx);
220 		}
221 	} else if (has_sched_dispatch) {
222 		blk_mq_do_dispatch_sched(hctx);
223 	} else if (q->mq_ops->get_budget) {
224 		/*
225 		 * If we need to get budget before queuing request, we
226 		 * dequeue request one by one from sw queue for avoiding
227 		 * to mess up I/O merge when dispatch runs out of resource.
228 		 *
229 		 * TODO: get more budgets, and dequeue more requests in
230 		 * one time.
231 		 */
232 		blk_mq_do_dispatch_ctx(hctx);
233 	} else {
234 		blk_mq_flush_busy_ctxs(hctx, &rq_list);
235 		blk_mq_dispatch_rq_list(q, &rq_list, false);
236 	}
237 }
238 
239 bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio,
240 			    struct request **merged_request)
241 {
242 	struct request *rq;
243 
244 	switch (elv_merge(q, &rq, bio)) {
245 	case ELEVATOR_BACK_MERGE:
246 		if (!blk_mq_sched_allow_merge(q, rq, bio))
247 			return false;
248 		if (!bio_attempt_back_merge(q, rq, bio))
249 			return false;
250 		*merged_request = attempt_back_merge(q, rq);
251 		if (!*merged_request)
252 			elv_merged_request(q, rq, ELEVATOR_BACK_MERGE);
253 		return true;
254 	case ELEVATOR_FRONT_MERGE:
255 		if (!blk_mq_sched_allow_merge(q, rq, bio))
256 			return false;
257 		if (!bio_attempt_front_merge(q, rq, bio))
258 			return false;
259 		*merged_request = attempt_front_merge(q, rq);
260 		if (!*merged_request)
261 			elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE);
262 		return true;
263 	default:
264 		return false;
265 	}
266 }
267 EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge);
268 
269 /*
270  * Reverse check our software queue for entries that we could potentially
271  * merge with. Currently includes a hand-wavy stop count of 8, to not spend
272  * too much time checking for merges.
273  */
274 static bool blk_mq_attempt_merge(struct request_queue *q,
275 				 struct blk_mq_ctx *ctx, struct bio *bio)
276 {
277 	struct request *rq;
278 	int checked = 8;
279 
280 	lockdep_assert_held(&ctx->lock);
281 
282 	list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) {
283 		bool merged = false;
284 
285 		if (!checked--)
286 			break;
287 
288 		if (!blk_rq_merge_ok(rq, bio))
289 			continue;
290 
291 		switch (blk_try_merge(rq, bio)) {
292 		case ELEVATOR_BACK_MERGE:
293 			if (blk_mq_sched_allow_merge(q, rq, bio))
294 				merged = bio_attempt_back_merge(q, rq, bio);
295 			break;
296 		case ELEVATOR_FRONT_MERGE:
297 			if (blk_mq_sched_allow_merge(q, rq, bio))
298 				merged = bio_attempt_front_merge(q, rq, bio);
299 			break;
300 		case ELEVATOR_DISCARD_MERGE:
301 			merged = bio_attempt_discard_merge(q, rq, bio);
302 			break;
303 		default:
304 			continue;
305 		}
306 
307 		if (merged)
308 			ctx->rq_merged++;
309 		return merged;
310 	}
311 
312 	return false;
313 }
314 
315 bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio)
316 {
317 	struct elevator_queue *e = q->elevator;
318 	struct blk_mq_ctx *ctx = blk_mq_get_ctx(q);
319 	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
320 	bool ret = false;
321 
322 	if (e && e->type->ops.mq.bio_merge) {
323 		blk_mq_put_ctx(ctx);
324 		return e->type->ops.mq.bio_merge(hctx, bio);
325 	}
326 
327 	if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) {
328 		/* default per sw-queue merge */
329 		spin_lock(&ctx->lock);
330 		ret = blk_mq_attempt_merge(q, ctx, bio);
331 		spin_unlock(&ctx->lock);
332 	}
333 
334 	blk_mq_put_ctx(ctx);
335 	return ret;
336 }
337 
338 bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq)
339 {
340 	return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq);
341 }
342 EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge);
343 
344 void blk_mq_sched_request_inserted(struct request *rq)
345 {
346 	trace_block_rq_insert(rq->q, rq);
347 }
348 EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted);
349 
350 static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx,
351 				       bool has_sched,
352 				       struct request *rq)
353 {
354 	/* dispatch flush rq directly */
355 	if (rq->rq_flags & RQF_FLUSH_SEQ) {
356 		spin_lock(&hctx->lock);
357 		list_add(&rq->queuelist, &hctx->dispatch);
358 		spin_unlock(&hctx->lock);
359 		return true;
360 	}
361 
362 	if (has_sched)
363 		rq->rq_flags |= RQF_SORTED;
364 
365 	return false;
366 }
367 
368 /**
369  * list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list
370  * @pos:    loop cursor.
371  * @skip:   the list element that will not be examined. Iteration starts at
372  *          @skip->next.
373  * @head:   head of the list to examine. This list must have at least one
374  *          element, namely @skip.
375  * @member: name of the list_head structure within typeof(*pos).
376  */
377 #define list_for_each_entry_rcu_rr(pos, skip, head, member)		\
378 	for ((pos) = (skip);						\
379 	     (pos = (pos)->member.next != (head) ? list_entry_rcu(	\
380 			(pos)->member.next, typeof(*pos), member) :	\
381 	      list_entry_rcu((pos)->member.next->next, typeof(*pos), member)), \
382 	     (pos) != (skip); )
383 
384 /*
385  * Called after a driver tag has been freed to check whether a hctx needs to
386  * be restarted. Restarts @hctx if its tag set is not shared. Restarts hardware
387  * queues in a round-robin fashion if the tag set of @hctx is shared with other
388  * hardware queues.
389  */
390 void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx)
391 {
392 	struct blk_mq_tags *const tags = hctx->tags;
393 	struct blk_mq_tag_set *const set = hctx->queue->tag_set;
394 	struct request_queue *const queue = hctx->queue, *q;
395 	struct blk_mq_hw_ctx *hctx2;
396 	unsigned int i, j;
397 
398 	if (set->flags & BLK_MQ_F_TAG_SHARED) {
399 		/*
400 		 * If this is 0, then we know that no hardware queues
401 		 * have RESTART marked. We're done.
402 		 */
403 		if (!atomic_read(&queue->shared_hctx_restart))
404 			return;
405 
406 		rcu_read_lock();
407 		list_for_each_entry_rcu_rr(q, queue, &set->tag_list,
408 					   tag_set_list) {
409 			queue_for_each_hw_ctx(q, hctx2, i)
410 				if (hctx2->tags == tags &&
411 				    blk_mq_sched_restart_hctx(hctx2))
412 					goto done;
413 		}
414 		j = hctx->queue_num + 1;
415 		for (i = 0; i < queue->nr_hw_queues; i++, j++) {
416 			if (j == queue->nr_hw_queues)
417 				j = 0;
418 			hctx2 = queue->queue_hw_ctx[j];
419 			if (hctx2->tags == tags &&
420 			    blk_mq_sched_restart_hctx(hctx2))
421 				break;
422 		}
423 done:
424 		rcu_read_unlock();
425 	} else {
426 		blk_mq_sched_restart_hctx(hctx);
427 	}
428 }
429 
430 void blk_mq_sched_insert_request(struct request *rq, bool at_head,
431 				 bool run_queue, bool async, bool can_block)
432 {
433 	struct request_queue *q = rq->q;
434 	struct elevator_queue *e = q->elevator;
435 	struct blk_mq_ctx *ctx = rq->mq_ctx;
436 	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
437 
438 	/* flush rq in flush machinery need to be dispatched directly */
439 	if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) {
440 		blk_insert_flush(rq);
441 		goto run;
442 	}
443 
444 	WARN_ON(e && (rq->tag != -1));
445 
446 	if (blk_mq_sched_bypass_insert(hctx, !!e, rq))
447 		goto run;
448 
449 	if (e && e->type->ops.mq.insert_requests) {
450 		LIST_HEAD(list);
451 
452 		list_add(&rq->queuelist, &list);
453 		e->type->ops.mq.insert_requests(hctx, &list, at_head);
454 	} else {
455 		spin_lock(&ctx->lock);
456 		__blk_mq_insert_request(hctx, rq, at_head);
457 		spin_unlock(&ctx->lock);
458 	}
459 
460 run:
461 	if (run_queue)
462 		blk_mq_run_hw_queue(hctx, async);
463 }
464 
465 void blk_mq_sched_insert_requests(struct request_queue *q,
466 				  struct blk_mq_ctx *ctx,
467 				  struct list_head *list, bool run_queue_async)
468 {
469 	struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu);
470 	struct elevator_queue *e = hctx->queue->elevator;
471 
472 	if (e && e->type->ops.mq.insert_requests)
473 		e->type->ops.mq.insert_requests(hctx, list, false);
474 	else
475 		blk_mq_insert_requests(hctx, ctx, list);
476 
477 	blk_mq_run_hw_queue(hctx, run_queue_async);
478 }
479 
480 static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set,
481 				   struct blk_mq_hw_ctx *hctx,
482 				   unsigned int hctx_idx)
483 {
484 	if (hctx->sched_tags) {
485 		blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx);
486 		blk_mq_free_rq_map(hctx->sched_tags);
487 		hctx->sched_tags = NULL;
488 	}
489 }
490 
491 static int blk_mq_sched_alloc_tags(struct request_queue *q,
492 				   struct blk_mq_hw_ctx *hctx,
493 				   unsigned int hctx_idx)
494 {
495 	struct blk_mq_tag_set *set = q->tag_set;
496 	int ret;
497 
498 	hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests,
499 					       set->reserved_tags);
500 	if (!hctx->sched_tags)
501 		return -ENOMEM;
502 
503 	ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests);
504 	if (ret)
505 		blk_mq_sched_free_tags(set, hctx, hctx_idx);
506 
507 	return ret;
508 }
509 
510 static void blk_mq_sched_tags_teardown(struct request_queue *q)
511 {
512 	struct blk_mq_tag_set *set = q->tag_set;
513 	struct blk_mq_hw_ctx *hctx;
514 	int i;
515 
516 	queue_for_each_hw_ctx(q, hctx, i)
517 		blk_mq_sched_free_tags(set, hctx, i);
518 }
519 
520 int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
521 			   unsigned int hctx_idx)
522 {
523 	struct elevator_queue *e = q->elevator;
524 	int ret;
525 
526 	if (!e)
527 		return 0;
528 
529 	ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx);
530 	if (ret)
531 		return ret;
532 
533 	if (e->type->ops.mq.init_hctx) {
534 		ret = e->type->ops.mq.init_hctx(hctx, hctx_idx);
535 		if (ret) {
536 			blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
537 			return ret;
538 		}
539 	}
540 
541 	blk_mq_debugfs_register_sched_hctx(q, hctx);
542 
543 	return 0;
544 }
545 
546 void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx,
547 			    unsigned int hctx_idx)
548 {
549 	struct elevator_queue *e = q->elevator;
550 
551 	if (!e)
552 		return;
553 
554 	blk_mq_debugfs_unregister_sched_hctx(hctx);
555 
556 	if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
557 		e->type->ops.mq.exit_hctx(hctx, hctx_idx);
558 		hctx->sched_data = NULL;
559 	}
560 
561 	blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx);
562 }
563 
564 int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e)
565 {
566 	struct blk_mq_hw_ctx *hctx;
567 	struct elevator_queue *eq;
568 	unsigned int i;
569 	int ret;
570 
571 	if (!e) {
572 		q->elevator = NULL;
573 		return 0;
574 	}
575 
576 	/*
577 	 * Default to double of smaller one between hw queue_depth and 128,
578 	 * since we don't split into sync/async like the old code did.
579 	 * Additionally, this is a per-hw queue depth.
580 	 */
581 	q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth,
582 				   BLKDEV_MAX_RQ);
583 
584 	queue_for_each_hw_ctx(q, hctx, i) {
585 		ret = blk_mq_sched_alloc_tags(q, hctx, i);
586 		if (ret)
587 			goto err;
588 	}
589 
590 	ret = e->ops.mq.init_sched(q, e);
591 	if (ret)
592 		goto err;
593 
594 	blk_mq_debugfs_register_sched(q);
595 
596 	queue_for_each_hw_ctx(q, hctx, i) {
597 		if (e->ops.mq.init_hctx) {
598 			ret = e->ops.mq.init_hctx(hctx, i);
599 			if (ret) {
600 				eq = q->elevator;
601 				blk_mq_exit_sched(q, eq);
602 				kobject_put(&eq->kobj);
603 				return ret;
604 			}
605 		}
606 		blk_mq_debugfs_register_sched_hctx(q, hctx);
607 	}
608 
609 	return 0;
610 
611 err:
612 	blk_mq_sched_tags_teardown(q);
613 	q->elevator = NULL;
614 	return ret;
615 }
616 
617 void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e)
618 {
619 	struct blk_mq_hw_ctx *hctx;
620 	unsigned int i;
621 
622 	queue_for_each_hw_ctx(q, hctx, i) {
623 		blk_mq_debugfs_unregister_sched_hctx(hctx);
624 		if (e->type->ops.mq.exit_hctx && hctx->sched_data) {
625 			e->type->ops.mq.exit_hctx(hctx, i);
626 			hctx->sched_data = NULL;
627 		}
628 	}
629 	blk_mq_debugfs_unregister_sched(q);
630 	if (e->type->ops.mq.exit_sched)
631 		e->type->ops.mq.exit_sched(e);
632 	blk_mq_sched_tags_teardown(q);
633 	q->elevator = NULL;
634 }
635 
636 int blk_mq_sched_init(struct request_queue *q)
637 {
638 	int ret;
639 
640 	mutex_lock(&q->sysfs_lock);
641 	ret = elevator_init(q, NULL);
642 	mutex_unlock(&q->sysfs_lock);
643 
644 	return ret;
645 }
646