blk-mq: fix two misuses on RQF_USE_SCHED
[linux-block.git] / block / elevator.c
CommitLineData
3dcf60bc 1// SPDX-License-Identifier: GPL-2.0
1da177e4 2/*
1da177e4
LT
3 * Block device elevator/IO-scheduler.
4 *
5 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 *
0fe23479 7 * 30042000 Jens Axboe <axboe@kernel.dk> :
1da177e4
LT
8 *
9 * Split the elevator a bit so that it is possible to choose a different
10 * one or even write a new "plug in". There are three pieces:
11 * - elevator_fn, inserts a new request in the queue list
12 * - elevator_merge_fn, decides whether a new buffer can be merged with
13 * an existing request
14 * - elevator_dequeue_fn, called when a request is taken off the active list
15 *
16 * 20082000 Dave Jones <davej@suse.de> :
17 * Removed tests for max-bomb-segments, which was breaking elvtune
18 * when run without -bN
19 *
20 * Jens:
21 * - Rework again to work with bio instead of buffer_heads
22 * - loose bi_dev comparisons, partition handling is right now
23 * - completely modularize elevator setup and teardown
24 *
25 */
26#include <linux/kernel.h>
27#include <linux/fs.h>
28#include <linux/blkdev.h>
1da177e4 29#include <linux/bio.h>
1da177e4
LT
30#include <linux/module.h>
31#include <linux/slab.h>
32#include <linux/init.h>
33#include <linux/compiler.h>
2056a782 34#include <linux/blktrace_api.h>
9817064b 35#include <linux/hash.h>
0835da67 36#include <linux/uaccess.h>
c8158819 37#include <linux/pm_runtime.h>
1da177e4 38
55782138
LZ
39#include <trace/events/block.h>
40
2e9bc346 41#include "elevator.h"
242f9dcb 42#include "blk.h"
bd166ef1 43#include "blk-mq-sched.h"
bca6b067 44#include "blk-pm.h"
8330cdb0 45#include "blk-wbt.h"
672fdcf0 46#include "blk-cgroup.h"
242f9dcb 47
1da177e4
LT
48static DEFINE_SPINLOCK(elv_list_lock);
49static LIST_HEAD(elv_list);
50
9817064b
JA
51/*
52 * Merge hash stuff.
53 */
83096ebf 54#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
9817064b 55
da775265
JA
56/*
57 * Query io scheduler to see if the current process issuing bio may be
58 * merged with rq.
59 */
8d283ee6 60static bool elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio)
da775265 61{
165125e1 62 struct request_queue *q = rq->q;
b374d18a 63 struct elevator_queue *e = q->elevator;
da775265 64
f9cd4bfe
JA
65 if (e->type->ops.allow_merge)
66 return e->type->ops.allow_merge(q, rq, bio);
da775265 67
8d283ee6 68 return true;
da775265
JA
69}
70
1da177e4
LT
71/*
72 * can we safely merge with this request?
73 */
72ef799b 74bool elv_bio_merge_ok(struct request *rq, struct bio *bio)
1da177e4 75{
050c8ea8 76 if (!blk_rq_merge_ok(rq, bio))
72ef799b 77 return false;
7ba1ba12 78
72ef799b
TE
79 if (!elv_iosched_allow_bio_merge(rq, bio))
80 return false;
1da177e4 81
72ef799b 82 return true;
1da177e4 83}
72ef799b 84EXPORT_SYMBOL(elv_bio_merge_ok);
1da177e4 85
ffb86425
CH
86static inline bool elv_support_features(struct request_queue *q,
87 const struct elevator_type *e)
8ac0d9a8 88{
ffb86425
CH
89 return (q->required_elevator_features & e->elevator_features) ==
90 q->required_elevator_features;
68c43f13
DLM
91}
92
93/**
f69b5e8f 94 * elevator_match - Check whether @e's name or alias matches @name
68c43f13
DLM
95 * @e: Scheduler to test
96 * @name: Elevator name to test
68c43f13 97 *
f69b5e8f 98 * Return true if the elevator @e's name or alias matches @name.
68c43f13 99 */
ffb86425 100static bool elevator_match(const struct elevator_type *e, const char *name)
68c43f13 101{
ffb86425
CH
102 return !strcmp(e->elevator_name, name) ||
103 (e->elevator_alias && !strcmp(e->elevator_alias, name));
8ac0d9a8
JA
104}
105
ffb86425 106static struct elevator_type *__elevator_find(const char *name)
1da177e4 107{
a22b169d 108 struct elevator_type *e;
1da177e4 109
ffb86425
CH
110 list_for_each_entry(e, &elv_list, list)
111 if (elevator_match(e, name))
a22b169d 112 return e;
a22b169d 113 return NULL;
1da177e4
LT
114}
115
81eaca44
CH
116static struct elevator_type *elevator_find_get(struct request_queue *q,
117 const char *name)
1da177e4 118{
2824bc93 119 struct elevator_type *e;
1da177e4 120
2a12dcd7 121 spin_lock(&elv_list_lock);
ffb86425
CH
122 e = __elevator_find(name);
123 if (e && (!elv_support_features(q, e) || !elevator_tryget(e)))
2824bc93 124 e = NULL;
2a12dcd7 125 spin_unlock(&elv_list_lock);
1da177e4
LT
126 return e;
127}
128
5f622417 129static const struct kobj_type elv_ktype;
3d1ab40f 130
d50235b7 131struct elevator_queue *elevator_alloc(struct request_queue *q,
165125e1 132 struct elevator_type *e)
3d1ab40f 133{
b374d18a 134 struct elevator_queue *eq;
9817064b 135
c1b511eb 136 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
9817064b 137 if (unlikely(!eq))
8406a4d5 138 return NULL;
9817064b 139
8ed40ee3 140 __elevator_get(e);
22f746e2 141 eq->type = e;
f9cb074b 142 kobject_init(&eq->kobj, &elv_ktype);
9817064b 143 mutex_init(&eq->sysfs_lock);
242d98f0 144 hash_init(eq->hash);
9817064b 145
3d1ab40f
AV
146 return eq;
147}
d50235b7 148EXPORT_SYMBOL(elevator_alloc);
3d1ab40f
AV
149
150static void elevator_release(struct kobject *kobj)
151{
b374d18a 152 struct elevator_queue *e;
9817064b 153
b374d18a 154 e = container_of(kobj, struct elevator_queue, kobj);
22f746e2 155 elevator_put(e->type);
3d1ab40f
AV
156 kfree(e);
157}
158
0c6cb3a2 159void elevator_exit(struct request_queue *q)
1da177e4 160{
0c6cb3a2
CH
161 struct elevator_queue *e = q->elevator;
162
28883074
CH
163 ioc_clear_queue(q);
164 blk_mq_sched_free_rqs(q);
165
3d1ab40f 166 mutex_lock(&e->sysfs_lock);
dd1c372d 167 blk_mq_exit_sched(q, e);
3d1ab40f 168 mutex_unlock(&e->sysfs_lock);
1da177e4 169
3d1ab40f 170 kobject_put(&e->kobj);
1da177e4 171}
2e662b65 172
9817064b
JA
173static inline void __elv_rqhash_del(struct request *rq)
174{
242d98f0 175 hash_del(&rq->hash);
e8064021 176 rq->rq_flags &= ~RQF_HASHED;
9817064b
JA
177}
178
70b3ea05 179void elv_rqhash_del(struct request_queue *q, struct request *rq)
9817064b
JA
180{
181 if (ELV_ON_HASH(rq))
182 __elv_rqhash_del(rq);
183}
bd166ef1 184EXPORT_SYMBOL_GPL(elv_rqhash_del);
9817064b 185
70b3ea05 186void elv_rqhash_add(struct request_queue *q, struct request *rq)
9817064b 187{
b374d18a 188 struct elevator_queue *e = q->elevator;
9817064b
JA
189
190 BUG_ON(ELV_ON_HASH(rq));
242d98f0 191 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
e8064021 192 rq->rq_flags |= RQF_HASHED;
9817064b 193}
bd166ef1 194EXPORT_SYMBOL_GPL(elv_rqhash_add);
9817064b 195
70b3ea05 196void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
9817064b
JA
197{
198 __elv_rqhash_del(rq);
199 elv_rqhash_add(q, rq);
200}
201
70b3ea05 202struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
9817064b 203{
b374d18a 204 struct elevator_queue *e = q->elevator;
b67bfe0d 205 struct hlist_node *next;
9817064b
JA
206 struct request *rq;
207
ee89f812 208 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
9817064b
JA
209 BUG_ON(!ELV_ON_HASH(rq));
210
211 if (unlikely(!rq_mergeable(rq))) {
212 __elv_rqhash_del(rq);
213 continue;
214 }
215
216 if (rq_hash_key(rq) == offset)
217 return rq;
218 }
219
220 return NULL;
221}
222
2e662b65
JA
223/*
224 * RB-tree support functions for inserting/lookup/removal of requests
225 * in a sorted RB tree.
226 */
796d5116 227void elv_rb_add(struct rb_root *root, struct request *rq)
2e662b65
JA
228{
229 struct rb_node **p = &root->rb_node;
230 struct rb_node *parent = NULL;
231 struct request *__rq;
232
233 while (*p) {
234 parent = *p;
235 __rq = rb_entry(parent, struct request, rb_node);
236
83096ebf 237 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
2e662b65 238 p = &(*p)->rb_left;
796d5116 239 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
2e662b65 240 p = &(*p)->rb_right;
2e662b65
JA
241 }
242
243 rb_link_node(&rq->rb_node, parent, p);
244 rb_insert_color(&rq->rb_node, root);
2e662b65 245}
2e662b65
JA
246EXPORT_SYMBOL(elv_rb_add);
247
248void elv_rb_del(struct rb_root *root, struct request *rq)
249{
250 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
251 rb_erase(&rq->rb_node, root);
252 RB_CLEAR_NODE(&rq->rb_node);
253}
2e662b65
JA
254EXPORT_SYMBOL(elv_rb_del);
255
256struct request *elv_rb_find(struct rb_root *root, sector_t sector)
257{
258 struct rb_node *n = root->rb_node;
259 struct request *rq;
260
261 while (n) {
262 rq = rb_entry(n, struct request, rb_node);
263
83096ebf 264 if (sector < blk_rq_pos(rq))
2e662b65 265 n = n->rb_left;
83096ebf 266 else if (sector > blk_rq_pos(rq))
2e662b65
JA
267 n = n->rb_right;
268 else
269 return rq;
270 }
271
272 return NULL;
273}
2e662b65
JA
274EXPORT_SYMBOL(elv_rb_find);
275
34fe7c05
CH
276enum elv_merge elv_merge(struct request_queue *q, struct request **req,
277 struct bio *bio)
1da177e4 278{
b374d18a 279 struct elevator_queue *e = q->elevator;
9817064b 280 struct request *__rq;
06b86245 281
488991e2
AB
282 /*
283 * Levels of merges:
284 * nomerges: No merges at all attempted
285 * noxmerges: Only simple one-hit cache try
286 * merges: All merge tries attempted
287 */
7460d389 288 if (blk_queue_nomerges(q) || !bio_mergeable(bio))
488991e2
AB
289 return ELEVATOR_NO_MERGE;
290
9817064b
JA
291 /*
292 * First try one-hit cache.
293 */
72ef799b 294 if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) {
34fe7c05
CH
295 enum elv_merge ret = blk_try_merge(q->last_merge, bio);
296
06b86245
TH
297 if (ret != ELEVATOR_NO_MERGE) {
298 *req = q->last_merge;
299 return ret;
300 }
301 }
1da177e4 302
488991e2 303 if (blk_queue_noxmerges(q))
ac9fafa1
AB
304 return ELEVATOR_NO_MERGE;
305
9817064b
JA
306 /*
307 * See if our hash lookup can find a potential backmerge.
308 */
4f024f37 309 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
72ef799b 310 if (__rq && elv_bio_merge_ok(__rq, bio)) {
9817064b 311 *req = __rq;
866663b7
ML
312
313 if (blk_discard_mergable(__rq))
314 return ELEVATOR_DISCARD_MERGE;
9817064b
JA
315 return ELEVATOR_BACK_MERGE;
316 }
317
f9cd4bfe
JA
318 if (e->type->ops.request_merge)
319 return e->type->ops.request_merge(q, req, bio);
1da177e4
LT
320
321 return ELEVATOR_NO_MERGE;
322}
323
5e84ea3a
JA
324/*
325 * Attempt to do an insertion back merge. Only check for the case where
326 * we can append 'rq' to an existing request, so we can throw 'rq' away
327 * afterwards.
328 *
fd2ef39c
JK
329 * Returns true if we merged, false otherwise. 'free' will contain all
330 * requests that need to be freed.
5e84ea3a 331 */
fd2ef39c
JK
332bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq,
333 struct list_head *free)
5e84ea3a
JA
334{
335 struct request *__rq;
bee0393c 336 bool ret;
5e84ea3a
JA
337
338 if (blk_queue_nomerges(q))
339 return false;
340
341 /*
342 * First try one-hit cache.
343 */
fd2ef39c
JK
344 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) {
345 list_add(&rq->queuelist, free);
5e84ea3a 346 return true;
fd2ef39c 347 }
5e84ea3a
JA
348
349 if (blk_queue_noxmerges(q))
350 return false;
351
bee0393c 352 ret = false;
5e84ea3a
JA
353 /*
354 * See if our hash lookup can find a potential backmerge.
355 */
bee0393c
SL
356 while (1) {
357 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
358 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
359 break;
360
fd2ef39c 361 list_add(&rq->queuelist, free);
bee0393c
SL
362 /* The merged request could be merged with others, try again */
363 ret = true;
364 rq = __rq;
365 }
27419322 366
bee0393c 367 return ret;
5e84ea3a
JA
368}
369
34fe7c05
CH
370void elv_merged_request(struct request_queue *q, struct request *rq,
371 enum elv_merge type)
1da177e4 372{
b374d18a 373 struct elevator_queue *e = q->elevator;
1da177e4 374
f9cd4bfe
JA
375 if (e->type->ops.request_merged)
376 e->type->ops.request_merged(q, rq, type);
06b86245 377
2e662b65
JA
378 if (type == ELEVATOR_BACK_MERGE)
379 elv_rqhash_reposition(q, rq);
9817064b 380
06b86245 381 q->last_merge = rq;
1da177e4
LT
382}
383
165125e1 384void elv_merge_requests(struct request_queue *q, struct request *rq,
1da177e4
LT
385 struct request *next)
386{
b374d18a 387 struct elevator_queue *e = q->elevator;
bd166ef1 388
f9cd4bfe
JA
389 if (e->type->ops.requests_merged)
390 e->type->ops.requests_merged(q, rq, next);
06b86245 391
9817064b 392 elv_rqhash_reposition(q, rq);
06b86245 393 q->last_merge = rq;
1da177e4
LT
394}
395
165125e1 396struct request *elv_latter_request(struct request_queue *q, struct request *rq)
1da177e4 397{
b374d18a 398 struct elevator_queue *e = q->elevator;
1da177e4 399
f9cd4bfe
JA
400 if (e->type->ops.next_request)
401 return e->type->ops.next_request(q, rq);
bd166ef1 402
1da177e4
LT
403 return NULL;
404}
405
165125e1 406struct request *elv_former_request(struct request_queue *q, struct request *rq)
1da177e4 407{
b374d18a 408 struct elevator_queue *e = q->elevator;
1da177e4 409
f9cd4bfe
JA
410 if (e->type->ops.former_request)
411 return e->type->ops.former_request(q, rq);
bd166ef1 412
a1ce35fa 413 return NULL;
1da177e4
LT
414}
415
3d1ab40f
AV
416#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
417
418static ssize_t
419elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1da177e4 420{
3d1ab40f 421 struct elv_fs_entry *entry = to_elv(attr);
b374d18a 422 struct elevator_queue *e;
3d1ab40f
AV
423 ssize_t error;
424
425 if (!entry->show)
426 return -EIO;
427
b374d18a 428 e = container_of(kobj, struct elevator_queue, kobj);
3d1ab40f 429 mutex_lock(&e->sysfs_lock);
22f746e2 430 error = e->type ? entry->show(e, page) : -ENOENT;
3d1ab40f
AV
431 mutex_unlock(&e->sysfs_lock);
432 return error;
433}
1da177e4 434
3d1ab40f
AV
435static ssize_t
436elv_attr_store(struct kobject *kobj, struct attribute *attr,
437 const char *page, size_t length)
438{
3d1ab40f 439 struct elv_fs_entry *entry = to_elv(attr);
b374d18a 440 struct elevator_queue *e;
3d1ab40f 441 ssize_t error;
1da177e4 442
3d1ab40f
AV
443 if (!entry->store)
444 return -EIO;
1da177e4 445
b374d18a 446 e = container_of(kobj, struct elevator_queue, kobj);
3d1ab40f 447 mutex_lock(&e->sysfs_lock);
22f746e2 448 error = e->type ? entry->store(e, page, length) : -ENOENT;
3d1ab40f
AV
449 mutex_unlock(&e->sysfs_lock);
450 return error;
451}
452
52cf25d0 453static const struct sysfs_ops elv_sysfs_ops = {
3d1ab40f
AV
454 .show = elv_attr_show,
455 .store = elv_attr_store,
456};
457
5f622417 458static const struct kobj_type elv_ktype = {
3d1ab40f
AV
459 .sysfs_ops = &elv_sysfs_ops,
460 .release = elevator_release,
461};
462
cecf5d87 463int elv_register_queue(struct request_queue *q, bool uevent)
3d1ab40f 464{
5a5bafdc 465 struct elevator_queue *e = q->elevator;
3d1ab40f
AV
466 int error;
467
f0c6ae09
YY
468 lockdep_assert_held(&q->sysfs_lock);
469
2bd85221 470 error = kobject_add(&e->kobj, &q->disk->queue_kobj, "iosched");
3d1ab40f 471 if (!error) {
22f746e2 472 struct elv_fs_entry *attr = e->type->elevator_attrs;
3d1ab40f 473 if (attr) {
e572ec7e
AV
474 while (attr->attr.name) {
475 if (sysfs_create_file(&e->kobj, &attr->attr))
3d1ab40f 476 break;
e572ec7e 477 attr++;
3d1ab40f
AV
478 }
479 }
cecf5d87
ML
480 if (uevent)
481 kobject_uevent(&e->kobj, KOBJ_ADD);
482
181d0663 483 set_bit(ELEVATOR_FLAG_REGISTERED, &e->flags);
3d1ab40f
AV
484 }
485 return error;
1da177e4 486}
bc1c1169 487
1da177e4
LT
488void elv_unregister_queue(struct request_queue *q)
489{
f5ec592d
EB
490 struct elevator_queue *e = q->elevator;
491
f0c6ae09
YY
492 lockdep_assert_held(&q->sysfs_lock);
493
181d0663 494 if (e && test_and_clear_bit(ELEVATOR_FLAG_REGISTERED, &e->flags)) {
f8fc877d
TH
495 kobject_uevent(&e->kobj, KOBJ_REMOVE);
496 kobject_del(&e->kobj);
f8fc877d 497 }
1da177e4
LT
498}
499
e567bf71 500int elv_register(struct elevator_type *e)
1da177e4 501{
e42cfb1d
DLM
502 /* insert_requests and dispatch_request are mandatory */
503 if (WARN_ON_ONCE(!e->ops.insert_requests || !e->ops.dispatch_request))
504 return -EINVAL;
505
3d3c2379
TH
506 /* create icq_cache if requested */
507 if (e->icq_size) {
508 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
509 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
510 return -EINVAL;
511
512 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
513 "%s_io_cq", e->elevator_name);
514 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
515 e->icq_align, 0, NULL);
516 if (!e->icq_cache)
517 return -ENOMEM;
518 }
519
520 /* register, don't allow duplicate names */
2a12dcd7 521 spin_lock(&elv_list_lock);
ffb86425 522 if (__elevator_find(e->elevator_name)) {
3d3c2379 523 spin_unlock(&elv_list_lock);
62d2a194 524 kmem_cache_destroy(e->icq_cache);
3d3c2379
TH
525 return -EBUSY;
526 }
1da177e4 527 list_add_tail(&e->list, &elv_list);
2a12dcd7 528 spin_unlock(&elv_list_lock);
1da177e4 529
d0b0a81a
HT
530 printk(KERN_INFO "io scheduler %s registered\n", e->elevator_name);
531
3d3c2379 532 return 0;
1da177e4
LT
533}
534EXPORT_SYMBOL_GPL(elv_register);
535
536void elv_unregister(struct elevator_type *e)
537{
3d3c2379 538 /* unregister */
2a12dcd7 539 spin_lock(&elv_list_lock);
1da177e4 540 list_del_init(&e->list);
2a12dcd7 541 spin_unlock(&elv_list_lock);
3d3c2379
TH
542
543 /*
544 * Destroy icq_cache if it exists. icq's are RCU managed. Make
545 * sure all RCU operations are complete before proceeding.
546 */
547 if (e->icq_cache) {
548 rcu_barrier();
549 kmem_cache_destroy(e->icq_cache);
550 e->icq_cache = NULL;
551 }
1da177e4
LT
552}
553EXPORT_SYMBOL_GPL(elv_unregister);
554
61db437d
DLM
555static inline bool elv_support_iosched(struct request_queue *q)
556{
6251b754 557 if (!queue_is_mq(q) ||
7a7c5e71 558 (q->tag_set && (q->tag_set->flags & BLK_MQ_F_NO_SCHED)))
61db437d
DLM
559 return false;
560 return true;
561}
562
131d08e1 563/*
a0958ba7
DLM
564 * For single queue devices, default to using mq-deadline. If we have multiple
565 * queues or mq-deadline is not available, default to "none".
566 */
567static struct elevator_type *elevator_get_default(struct request_queue *q)
568{
90b71980
BVA
569 if (q->tag_set && q->tag_set->flags & BLK_MQ_F_NO_SCHED_BY_DEFAULT)
570 return NULL;
571
580dca81 572 if (q->nr_hw_queues != 1 &&
079a2e3e 573 !blk_mq_is_shared_tags(q->tag_set->flags))
a0958ba7
DLM
574 return NULL;
575
81eaca44 576 return elevator_find_get(q, "mq-deadline");
a0958ba7
DLM
577}
578
579/*
580 * Get the first elevator providing the features required by the request queue.
581 * Default to "none" if no matching elevator is found.
582 */
583static struct elevator_type *elevator_get_by_features(struct request_queue *q)
584{
a2614255 585 struct elevator_type *e, *found = NULL;
a0958ba7
DLM
586
587 spin_lock(&elv_list_lock);
588
589 list_for_each_entry(e, &elv_list, list) {
ffb86425 590 if (elv_support_features(q, e)) {
a2614255 591 found = e;
a0958ba7 592 break;
a2614255 593 }
a0958ba7
DLM
594 }
595
dd6f7f17 596 if (found && !elevator_tryget(found))
a2614255 597 found = NULL;
a0958ba7
DLM
598
599 spin_unlock(&elv_list_lock);
a2614255 600 return found;
a0958ba7
DLM
601}
602
603/*
604 * For a device queue that has no required features, use the default elevator
605 * settings. Otherwise, use the first elevator available matching the required
606 * features. If no suitable elevator is find or if the chosen elevator
607 * initialization fails, fall back to the "none" elevator (no elevator).
131d08e1 608 */
954b4a5c 609void elevator_init_mq(struct request_queue *q)
131d08e1
CH
610{
611 struct elevator_type *e;
954b4a5c 612 int err;
131d08e1 613
61db437d 614 if (!elv_support_iosched(q))
954b4a5c 615 return;
61db437d 616
75e6c00f 617 WARN_ON_ONCE(blk_queue_registered(q));
c48dac13 618
131d08e1 619 if (unlikely(q->elevator))
954b4a5c 620 return;
131d08e1 621
a0958ba7
DLM
622 if (!q->required_elevator_features)
623 e = elevator_get_default(q);
624 else
625 e = elevator_get_by_features(q);
131d08e1 626 if (!e)
954b4a5c 627 return;
131d08e1 628
245a489e
ML
629 /*
630 * We are called before adding disk, when there isn't any FS I/O,
631 * so freezing queue plus canceling dispatch work is enough to
632 * drain any dispatch activities originated from passthrough
633 * requests, then no need to quiesce queue which may add long boot
634 * latency, especially when lots of disks are involved.
635 */
737eb78e 636 blk_mq_freeze_queue(q);
245a489e 637 blk_mq_cancel_work_sync(q);
737eb78e 638
131d08e1 639 err = blk_mq_init_sched(q, e);
737eb78e 640
737eb78e
DLM
641 blk_mq_unfreeze_queue(q);
642
954b4a5c
DLM
643 if (err) {
644 pr_warn("\"%s\" elevator initialization failed, "
645 "falling back to \"none\"\n", e->elevator_name);
954b4a5c 646 }
8ed40ee3
JC
647
648 elevator_put(e);
131d08e1 649}
131d08e1 650
1da177e4 651/*
ac1171bd
JC
652 * Switch to new_e io scheduler.
653 *
654 * If switching fails, we are most likely running out of memory and not able
655 * to restore the old io scheduler, so leaving the io scheduler being none.
1da177e4 656 */
8237c01f 657int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
1da177e4 658{
64b36075 659 int ret;
1da177e4 660
14a23498
BVA
661 lockdep_assert_held(&q->sysfs_lock);
662
a1ce35fa
JA
663 blk_mq_freeze_queue(q);
664 blk_mq_quiesce_queue(q);
4722dc52 665
64b36075
CH
666 if (q->elevator) {
667 elv_unregister_queue(q);
668 elevator_exit(q);
669 }
670
671 ret = blk_mq_init_sched(q, new_e);
672 if (ret)
673 goto out_unfreeze;
674
675 ret = elv_register_queue(q, true);
676 if (ret) {
677 elevator_exit(q);
678 goto out_unfreeze;
679 }
680 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
1da177e4 681
64b36075 682out_unfreeze:
a1ce35fa
JA
683 blk_mq_unquiesce_queue(q);
684 blk_mq_unfreeze_queue(q);
e0cca8bc
JC
685
686 if (ret) {
687 pr_warn("elv: switch to \"%s\" failed, falling back to \"none\"\n",
688 new_e->elevator_name);
689 }
690
64b36075
CH
691 return ret;
692}
693
694void elevator_disable(struct request_queue *q)
695{
696 lockdep_assert_held(&q->sysfs_lock);
75ad23bc 697
64b36075
CH
698 blk_mq_freeze_queue(q);
699 blk_mq_quiesce_queue(q);
700
701 elv_unregister_queue(q);
702 elevator_exit(q);
703 blk_queue_flag_clear(QUEUE_FLAG_SQ_SCHED, q);
704 q->elevator = NULL;
705 q->nr_requests = q->tag_set->queue_depth;
706 blk_add_trace_msg(q, "elv switch: none");
707
708 blk_mq_unquiesce_queue(q);
709 blk_mq_unfreeze_queue(q);
1da177e4
LT
710}
711
5dd531a0
JA
712/*
713 * Switch this queue to the given IO scheduler.
714 */
58367c8a 715static int elevator_change(struct request_queue *q, const char *elevator_name)
1da177e4 716{
1da177e4 717 struct elevator_type *e;
8ed40ee3 718 int ret;
1da177e4 719
e9a823fb 720 /* Make sure queue is not in the middle of being removed */
58c898ba 721 if (!blk_queue_registered(q))
e9a823fb
DJ
722 return -ENOENT;
723
58367c8a 724 if (!strncmp(elevator_name, "none", 4)) {
64b36075
CH
725 if (q->elevator)
726 elevator_disable(q);
727 return 0;
fbd72127 728 }
cd43e26f 729
ffb86425 730 if (q->elevator && elevator_match(q->elevator->type, elevator_name))
b54c2ad9
JC
731 return 0;
732
81eaca44
CH
733 e = elevator_find_get(q, elevator_name);
734 if (!e) {
735 request_module("%s-iosched", elevator_name);
736 e = elevator_find_get(q, elevator_name);
737 if (!e)
738 return -EINVAL;
739 }
8ed40ee3
JC
740 ret = elevator_switch(q, e);
741 elevator_put(e);
742 return ret;
5dd531a0 743}
7c8a3679 744
58367c8a 745ssize_t elv_iosched_store(struct request_queue *q, const char *buf,
5dd531a0
JA
746 size_t count)
747{
58367c8a 748 char elevator_name[ELV_NAME_MAX];
5dd531a0
JA
749 int ret;
750
6251b754 751 if (!elv_support_iosched(q))
5dd531a0
JA
752 return count;
753
20d09975 754 strscpy(elevator_name, buf, sizeof(elevator_name));
58367c8a 755 ret = elevator_change(q, strstrip(elevator_name));
5dd531a0
JA
756 if (!ret)
757 return count;
5dd531a0 758 return ret;
1da177e4
LT
759}
760
165125e1 761ssize_t elv_iosched_show(struct request_queue *q, char *name)
1da177e4 762{
16095af2
CH
763 struct elevator_queue *eq = q->elevator;
764 struct elevator_type *cur = NULL, *e;
1da177e4
LT
765 int len = 0;
766
aae2a643 767 if (!elv_support_iosched(q))
cd43e26f
MP
768 return sprintf(name, "none\n");
769
7919d679 770 if (!q->elevator) {
bd166ef1 771 len += sprintf(name+len, "[none] ");
7919d679
JC
772 } else {
773 len += sprintf(name+len, "none ");
16095af2 774 cur = eq->type;
7919d679 775 }
cd43e26f 776
2a12dcd7 777 spin_lock(&elv_list_lock);
16095af2 778 list_for_each_entry(e, &elv_list, list) {
5998249e 779 if (e == cur)
7a3b3660 780 len += sprintf(name+len, "[%s] ", e->elevator_name);
5998249e 781 else if (elv_support_features(q, e))
16095af2 782 len += sprintf(name+len, "%s ", e->elevator_name);
1da177e4 783 }
2a12dcd7 784 spin_unlock(&elv_list_lock);
1da177e4 785
c6451ede 786 len += sprintf(name+len, "\n");
1da177e4
LT
787 return len;
788}
789
165125e1
JA
790struct request *elv_rb_former_request(struct request_queue *q,
791 struct request *rq)
2e662b65
JA
792{
793 struct rb_node *rbprev = rb_prev(&rq->rb_node);
794
795 if (rbprev)
796 return rb_entry_rq(rbprev);
797
798 return NULL;
799}
2e662b65
JA
800EXPORT_SYMBOL(elv_rb_former_request);
801
165125e1
JA
802struct request *elv_rb_latter_request(struct request_queue *q,
803 struct request *rq)
2e662b65
JA
804{
805 struct rb_node *rbnext = rb_next(&rq->rb_node);
806
807 if (rbnext)
808 return rb_entry_rq(rbnext);
809
810 return NULL;
811}
2e662b65 812EXPORT_SYMBOL(elv_rb_latter_request);
f8db3835
JK
813
814static int __init elevator_setup(char *str)
815{
816 pr_warn("Kernel parameter elevator= does not have any effect anymore.\n"
817 "Please use sysfs to set IO scheduler for individual devices.\n");
818 return 1;
819}
820
821__setup("elevator=", elevator_setup);