]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - block/elevator.c
kernel/async.c: switch to pr_foo()
[mirror_ubuntu-artful-kernel.git] / block / elevator.c
CommitLineData
1da177e4 1/*
1da177e4
LT
2 * Block device elevator/IO-scheduler.
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 *
0fe23479 6 * 30042000 Jens Axboe <axboe@kernel.dk> :
1da177e4
LT
7 *
8 * Split the elevator a bit so that it is possible to choose a different
9 * one or even write a new "plug in". There are three pieces:
10 * - elevator_fn, inserts a new request in the queue list
11 * - elevator_merge_fn, decides whether a new buffer can be merged with
12 * an existing request
13 * - elevator_dequeue_fn, called when a request is taken off the active list
14 *
15 * 20082000 Dave Jones <davej@suse.de> :
16 * Removed tests for max-bomb-segments, which was breaking elvtune
17 * when run without -bN
18 *
19 * Jens:
20 * - Rework again to work with bio instead of buffer_heads
21 * - loose bi_dev comparisons, partition handling is right now
22 * - completely modularize elevator setup and teardown
23 *
24 */
25#include <linux/kernel.h>
26#include <linux/fs.h>
27#include <linux/blkdev.h>
28#include <linux/elevator.h>
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
242f9dcb 41#include "blk.h"
72e06c25 42#include "blk-cgroup.h"
242f9dcb 43
1da177e4
LT
44static DEFINE_SPINLOCK(elv_list_lock);
45static LIST_HEAD(elv_list);
46
9817064b
JA
47/*
48 * Merge hash stuff.
49 */
83096ebf 50#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
9817064b 51
da775265
JA
52/*
53 * Query io scheduler to see if the current process issuing bio may be
54 * merged with rq.
55 */
56static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
57{
165125e1 58 struct request_queue *q = rq->q;
b374d18a 59 struct elevator_queue *e = q->elevator;
da775265 60
22f746e2
TH
61 if (e->type->ops.elevator_allow_merge_fn)
62 return e->type->ops.elevator_allow_merge_fn(q, rq, bio);
da775265
JA
63
64 return 1;
65}
66
1da177e4
LT
67/*
68 * can we safely merge with this request?
69 */
050c8ea8 70bool elv_rq_merge_ok(struct request *rq, struct bio *bio)
1da177e4 71{
050c8ea8 72 if (!blk_rq_merge_ok(rq, bio))
7ba1ba12
MP
73 return 0;
74
da775265
JA
75 if (!elv_iosched_allow_merge(rq, bio))
76 return 0;
1da177e4 77
da775265 78 return 1;
1da177e4
LT
79}
80EXPORT_SYMBOL(elv_rq_merge_ok);
81
1da177e4
LT
82static struct elevator_type *elevator_find(const char *name)
83{
a22b169d 84 struct elevator_type *e;
1da177e4 85
70cee26e 86 list_for_each_entry(e, &elv_list, list) {
a22b169d
VT
87 if (!strcmp(e->elevator_name, name))
88 return e;
1da177e4 89 }
1da177e4 90
a22b169d 91 return NULL;
1da177e4
LT
92}
93
94static void elevator_put(struct elevator_type *e)
95{
96 module_put(e->elevator_owner);
97}
98
21c3c5d2 99static struct elevator_type *elevator_get(const char *name, bool try_loading)
1da177e4 100{
2824bc93 101 struct elevator_type *e;
1da177e4 102
2a12dcd7 103 spin_lock(&elv_list_lock);
2824bc93
TH
104
105 e = elevator_find(name);
21c3c5d2 106 if (!e && try_loading) {
e1640949 107 spin_unlock(&elv_list_lock);
490b94be 108 request_module("%s-iosched", name);
e1640949
JA
109 spin_lock(&elv_list_lock);
110 e = elevator_find(name);
111 }
112
2824bc93
TH
113 if (e && !try_module_get(e->elevator_owner))
114 e = NULL;
115
2a12dcd7 116 spin_unlock(&elv_list_lock);
1da177e4
LT
117
118 return e;
119}
120
484fc254 121static char chosen_elevator[ELV_NAME_MAX];
1da177e4 122
5f003976 123static int __init elevator_setup(char *str)
1da177e4 124{
752a3b79
CE
125 /*
126 * Be backwards-compatible with previous kernels, so users
127 * won't get the wrong elevator.
128 */
492af635 129 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
9b41046c 130 return 1;
1da177e4
LT
131}
132
133__setup("elevator=", elevator_setup);
134
bb813f4c
TH
135/* called during boot to load the elevator chosen by the elevator param */
136void __init load_default_elevator_module(void)
137{
138 struct elevator_type *e;
139
140 if (!chosen_elevator[0])
141 return;
142
143 spin_lock(&elv_list_lock);
144 e = elevator_find(chosen_elevator);
145 spin_unlock(&elv_list_lock);
146
147 if (!e)
148 request_module("%s-iosched", chosen_elevator);
149}
150
3d1ab40f
AV
151static struct kobj_type elv_ktype;
152
d50235b7 153struct elevator_queue *elevator_alloc(struct request_queue *q,
165125e1 154 struct elevator_type *e)
3d1ab40f 155{
b374d18a 156 struct elevator_queue *eq;
9817064b 157
c1b511eb 158 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
9817064b
JA
159 if (unlikely(!eq))
160 goto err;
161
22f746e2 162 eq->type = e;
f9cb074b 163 kobject_init(&eq->kobj, &elv_ktype);
9817064b 164 mutex_init(&eq->sysfs_lock);
242d98f0 165 hash_init(eq->hash);
9817064b 166
3d1ab40f 167 return eq;
9817064b
JA
168err:
169 kfree(eq);
170 elevator_put(e);
171 return NULL;
3d1ab40f 172}
d50235b7 173EXPORT_SYMBOL(elevator_alloc);
3d1ab40f
AV
174
175static void elevator_release(struct kobject *kobj)
176{
b374d18a 177 struct elevator_queue *e;
9817064b 178
b374d18a 179 e = container_of(kobj, struct elevator_queue, kobj);
22f746e2 180 elevator_put(e->type);
3d1ab40f
AV
181 kfree(e);
182}
183
165125e1 184int elevator_init(struct request_queue *q, char *name)
1da177e4
LT
185{
186 struct elevator_type *e = NULL;
f8fc877d 187 int err;
1da177e4 188
eb1c160b
TS
189 /*
190 * q->sysfs_lock must be held to provide mutual exclusion between
191 * elevator_switch() and here.
192 */
193 lockdep_assert_held(&q->sysfs_lock);
194
1abec4fd
MS
195 if (unlikely(q->elevator))
196 return 0;
197
cb98fc8b
TH
198 INIT_LIST_HEAD(&q->queue_head);
199 q->last_merge = NULL;
200 q->end_sector = 0;
201 q->boundary_rq = NULL;
cb98fc8b 202
4eb166d9 203 if (name) {
21c3c5d2 204 e = elevator_get(name, true);
4eb166d9
JA
205 if (!e)
206 return -EINVAL;
207 }
1da177e4 208
21c3c5d2
TH
209 /*
210 * Use the default elevator specified by config boot param or
211 * config option. Don't try to load modules as we could be running
212 * off async and request_module() isn't allowed from async.
213 */
4eb166d9 214 if (!e && *chosen_elevator) {
21c3c5d2 215 e = elevator_get(chosen_elevator, false);
4eb166d9
JA
216 if (!e)
217 printk(KERN_ERR "I/O scheduler %s not found\n",
218 chosen_elevator);
219 }
248d5ca5 220
4eb166d9 221 if (!e) {
21c3c5d2 222 e = elevator_get(CONFIG_DEFAULT_IOSCHED, false);
4eb166d9
JA
223 if (!e) {
224 printk(KERN_ERR
225 "Default I/O scheduler not found. " \
226 "Using noop.\n");
21c3c5d2 227 e = elevator_get("noop", false);
4eb166d9 228 }
5f003976
ND
229 }
230
d50235b7 231 err = e->ops.elevator_init_fn(q, e);
1abec4fd 232 return 0;
1da177e4 233}
2e662b65
JA
234EXPORT_SYMBOL(elevator_init);
235
b374d18a 236void elevator_exit(struct elevator_queue *e)
1da177e4 237{
3d1ab40f 238 mutex_lock(&e->sysfs_lock);
22f746e2
TH
239 if (e->type->ops.elevator_exit_fn)
240 e->type->ops.elevator_exit_fn(e);
3d1ab40f 241 mutex_unlock(&e->sysfs_lock);
1da177e4 242
3d1ab40f 243 kobject_put(&e->kobj);
1da177e4 244}
2e662b65
JA
245EXPORT_SYMBOL(elevator_exit);
246
9817064b
JA
247static inline void __elv_rqhash_del(struct request *rq)
248{
242d98f0 249 hash_del(&rq->hash);
360f92c2 250 rq->cmd_flags &= ~REQ_HASHED;
9817064b
JA
251}
252
165125e1 253static void elv_rqhash_del(struct request_queue *q, struct request *rq)
9817064b
JA
254{
255 if (ELV_ON_HASH(rq))
256 __elv_rqhash_del(rq);
257}
258
165125e1 259static void elv_rqhash_add(struct request_queue *q, struct request *rq)
9817064b 260{
b374d18a 261 struct elevator_queue *e = q->elevator;
9817064b
JA
262
263 BUG_ON(ELV_ON_HASH(rq));
242d98f0 264 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
360f92c2 265 rq->cmd_flags |= REQ_HASHED;
9817064b
JA
266}
267
165125e1 268static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
9817064b
JA
269{
270 __elv_rqhash_del(rq);
271 elv_rqhash_add(q, rq);
272}
273
165125e1 274static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
9817064b 275{
b374d18a 276 struct elevator_queue *e = q->elevator;
b67bfe0d 277 struct hlist_node *next;
9817064b
JA
278 struct request *rq;
279
ee89f812 280 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
9817064b
JA
281 BUG_ON(!ELV_ON_HASH(rq));
282
283 if (unlikely(!rq_mergeable(rq))) {
284 __elv_rqhash_del(rq);
285 continue;
286 }
287
288 if (rq_hash_key(rq) == offset)
289 return rq;
290 }
291
292 return NULL;
293}
294
2e662b65
JA
295/*
296 * RB-tree support functions for inserting/lookup/removal of requests
297 * in a sorted RB tree.
298 */
796d5116 299void elv_rb_add(struct rb_root *root, struct request *rq)
2e662b65
JA
300{
301 struct rb_node **p = &root->rb_node;
302 struct rb_node *parent = NULL;
303 struct request *__rq;
304
305 while (*p) {
306 parent = *p;
307 __rq = rb_entry(parent, struct request, rb_node);
308
83096ebf 309 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
2e662b65 310 p = &(*p)->rb_left;
796d5116 311 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
2e662b65 312 p = &(*p)->rb_right;
2e662b65
JA
313 }
314
315 rb_link_node(&rq->rb_node, parent, p);
316 rb_insert_color(&rq->rb_node, root);
2e662b65 317}
2e662b65
JA
318EXPORT_SYMBOL(elv_rb_add);
319
320void elv_rb_del(struct rb_root *root, struct request *rq)
321{
322 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
323 rb_erase(&rq->rb_node, root);
324 RB_CLEAR_NODE(&rq->rb_node);
325}
2e662b65
JA
326EXPORT_SYMBOL(elv_rb_del);
327
328struct request *elv_rb_find(struct rb_root *root, sector_t sector)
329{
330 struct rb_node *n = root->rb_node;
331 struct request *rq;
332
333 while (n) {
334 rq = rb_entry(n, struct request, rb_node);
335
83096ebf 336 if (sector < blk_rq_pos(rq))
2e662b65 337 n = n->rb_left;
83096ebf 338 else if (sector > blk_rq_pos(rq))
2e662b65
JA
339 n = n->rb_right;
340 else
341 return rq;
342 }
343
344 return NULL;
345}
2e662b65
JA
346EXPORT_SYMBOL(elv_rb_find);
347
8922e16c
TH
348/*
349 * Insert rq into dispatch queue of q. Queue lock must be held on
dbe7f76d 350 * entry. rq is sort instead into the dispatch queue. To be used by
2e662b65 351 * specific elevators.
8922e16c 352 */
165125e1 353void elv_dispatch_sort(struct request_queue *q, struct request *rq)
8922e16c
TH
354{
355 sector_t boundary;
8922e16c 356 struct list_head *entry;
4eb166d9 357 int stop_flags;
8922e16c 358
06b86245
TH
359 if (q->last_merge == rq)
360 q->last_merge = NULL;
9817064b
JA
361
362 elv_rqhash_del(q, rq);
363
15853af9 364 q->nr_sorted--;
06b86245 365
1b47f531 366 boundary = q->end_sector;
02e031cb 367 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
8922e16c
TH
368 list_for_each_prev(entry, &q->queue_head) {
369 struct request *pos = list_entry_rq(entry);
370
33659ebb
CH
371 if ((rq->cmd_flags & REQ_DISCARD) !=
372 (pos->cmd_flags & REQ_DISCARD))
e17fc0a1 373 break;
783660b2
JA
374 if (rq_data_dir(rq) != rq_data_dir(pos))
375 break;
4eb166d9 376 if (pos->cmd_flags & stop_flags)
8922e16c 377 break;
83096ebf
TH
378 if (blk_rq_pos(rq) >= boundary) {
379 if (blk_rq_pos(pos) < boundary)
8922e16c
TH
380 continue;
381 } else {
83096ebf 382 if (blk_rq_pos(pos) >= boundary)
8922e16c
TH
383 break;
384 }
83096ebf 385 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
8922e16c
TH
386 break;
387 }
388
389 list_add(&rq->queuelist, entry);
390}
2e662b65
JA
391EXPORT_SYMBOL(elv_dispatch_sort);
392
9817064b 393/*
2e662b65
JA
394 * Insert rq into dispatch queue of q. Queue lock must be held on
395 * entry. rq is added to the back of the dispatch queue. To be used by
396 * specific elevators.
9817064b
JA
397 */
398void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
399{
400 if (q->last_merge == rq)
401 q->last_merge = NULL;
402
403 elv_rqhash_del(q, rq);
404
405 q->nr_sorted--;
406
407 q->end_sector = rq_end_sector(rq);
408 q->boundary_rq = rq;
409 list_add_tail(&rq->queuelist, &q->queue_head);
410}
2e662b65
JA
411EXPORT_SYMBOL(elv_dispatch_add_tail);
412
165125e1 413int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
1da177e4 414{
b374d18a 415 struct elevator_queue *e = q->elevator;
9817064b 416 struct request *__rq;
06b86245
TH
417 int ret;
418
488991e2
AB
419 /*
420 * Levels of merges:
421 * nomerges: No merges at all attempted
422 * noxmerges: Only simple one-hit cache try
423 * merges: All merge tries attempted
424 */
425 if (blk_queue_nomerges(q))
426 return ELEVATOR_NO_MERGE;
427
9817064b
JA
428 /*
429 * First try one-hit cache.
430 */
050c8ea8
TH
431 if (q->last_merge && elv_rq_merge_ok(q->last_merge, bio)) {
432 ret = blk_try_merge(q->last_merge, bio);
06b86245
TH
433 if (ret != ELEVATOR_NO_MERGE) {
434 *req = q->last_merge;
435 return ret;
436 }
437 }
1da177e4 438
488991e2 439 if (blk_queue_noxmerges(q))
ac9fafa1
AB
440 return ELEVATOR_NO_MERGE;
441
9817064b
JA
442 /*
443 * See if our hash lookup can find a potential backmerge.
444 */
4f024f37 445 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
9817064b
JA
446 if (__rq && elv_rq_merge_ok(__rq, bio)) {
447 *req = __rq;
448 return ELEVATOR_BACK_MERGE;
449 }
450
22f746e2
TH
451 if (e->type->ops.elevator_merge_fn)
452 return e->type->ops.elevator_merge_fn(q, req, bio);
1da177e4
LT
453
454 return ELEVATOR_NO_MERGE;
455}
456
5e84ea3a
JA
457/*
458 * Attempt to do an insertion back merge. Only check for the case where
459 * we can append 'rq' to an existing request, so we can throw 'rq' away
460 * afterwards.
461 *
462 * Returns true if we merged, false otherwise
463 */
464static bool elv_attempt_insert_merge(struct request_queue *q,
465 struct request *rq)
466{
467 struct request *__rq;
bee0393c 468 bool ret;
5e84ea3a
JA
469
470 if (blk_queue_nomerges(q))
471 return false;
472
473 /*
474 * First try one-hit cache.
475 */
476 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
477 return true;
478
479 if (blk_queue_noxmerges(q))
480 return false;
481
bee0393c 482 ret = false;
5e84ea3a
JA
483 /*
484 * See if our hash lookup can find a potential backmerge.
485 */
bee0393c
SL
486 while (1) {
487 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
488 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
489 break;
490
491 /* The merged request could be merged with others, try again */
492 ret = true;
493 rq = __rq;
494 }
27419322 495
bee0393c 496 return ret;
5e84ea3a
JA
497}
498
165125e1 499void elv_merged_request(struct request_queue *q, struct request *rq, int type)
1da177e4 500{
b374d18a 501 struct elevator_queue *e = q->elevator;
1da177e4 502
22f746e2
TH
503 if (e->type->ops.elevator_merged_fn)
504 e->type->ops.elevator_merged_fn(q, rq, type);
06b86245 505
2e662b65
JA
506 if (type == ELEVATOR_BACK_MERGE)
507 elv_rqhash_reposition(q, rq);
9817064b 508
06b86245 509 q->last_merge = rq;
1da177e4
LT
510}
511
165125e1 512void elv_merge_requests(struct request_queue *q, struct request *rq,
1da177e4
LT
513 struct request *next)
514{
b374d18a 515 struct elevator_queue *e = q->elevator;
5e84ea3a 516 const int next_sorted = next->cmd_flags & REQ_SORTED;
1da177e4 517
22f746e2
TH
518 if (next_sorted && e->type->ops.elevator_merge_req_fn)
519 e->type->ops.elevator_merge_req_fn(q, rq, next);
06b86245 520
9817064b 521 elv_rqhash_reposition(q, rq);
9817064b 522
5e84ea3a
JA
523 if (next_sorted) {
524 elv_rqhash_del(q, next);
525 q->nr_sorted--;
526 }
527
06b86245 528 q->last_merge = rq;
1da177e4
LT
529}
530
812d4026
DS
531void elv_bio_merged(struct request_queue *q, struct request *rq,
532 struct bio *bio)
533{
534 struct elevator_queue *e = q->elevator;
535
22f746e2
TH
536 if (e->type->ops.elevator_bio_merged_fn)
537 e->type->ops.elevator_bio_merged_fn(q, rq, bio);
812d4026
DS
538}
539
c8158819
LM
540#ifdef CONFIG_PM_RUNTIME
541static void blk_pm_requeue_request(struct request *rq)
542{
543 if (rq->q->dev && !(rq->cmd_flags & REQ_PM))
544 rq->q->nr_pending--;
545}
546
547static void blk_pm_add_request(struct request_queue *q, struct request *rq)
548{
549 if (q->dev && !(rq->cmd_flags & REQ_PM) && q->nr_pending++ == 0 &&
550 (q->rpm_status == RPM_SUSPENDED || q->rpm_status == RPM_SUSPENDING))
551 pm_request_resume(q->dev);
552}
553#else
554static inline void blk_pm_requeue_request(struct request *rq) {}
555static inline void blk_pm_add_request(struct request_queue *q,
556 struct request *rq)
557{
558}
559#endif
560
165125e1 561void elv_requeue_request(struct request_queue *q, struct request *rq)
1da177e4 562{
1da177e4
LT
563 /*
564 * it already went through dequeue, we need to decrement the
565 * in_flight count again
566 */
8922e16c 567 if (blk_account_rq(rq)) {
0a7ae2ff 568 q->in_flight[rq_is_sync(rq)]--;
33659ebb 569 if (rq->cmd_flags & REQ_SORTED)
cad97516 570 elv_deactivate_rq(q, rq);
8922e16c 571 }
1da177e4 572
4aff5e23 573 rq->cmd_flags &= ~REQ_STARTED;
1da177e4 574
c8158819
LM
575 blk_pm_requeue_request(rq);
576
b710a480 577 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
1da177e4
LT
578}
579
26308eab 580void elv_drain_elevator(struct request_queue *q)
15853af9
TH
581{
582 static int printed;
e3c78ca5
TH
583
584 lockdep_assert_held(q->queue_lock);
585
22f746e2 586 while (q->elevator->type->ops.elevator_dispatch_fn(q, 1))
15853af9 587 ;
e3c78ca5 588 if (q->nr_sorted && printed++ < 10) {
15853af9
TH
589 printk(KERN_ERR "%s: forced dispatching is broken "
590 "(nr_sorted=%u), please report this\n",
22f746e2 591 q->elevator->type->elevator_name, q->nr_sorted);
15853af9
TH
592 }
593}
594
b710a480 595void __elv_add_request(struct request_queue *q, struct request *rq, int where)
1da177e4 596{
5f3ea37c 597 trace_block_rq_insert(q, rq);
2056a782 598
c8158819
LM
599 blk_pm_add_request(q, rq);
600
1da177e4
LT
601 rq->q = q;
602
b710a480
JA
603 if (rq->cmd_flags & REQ_SOFTBARRIER) {
604 /* barriers are scheduling boundary, update end_sector */
e2a60da7 605 if (rq->cmd_type == REQ_TYPE_FS) {
b710a480
JA
606 q->end_sector = rq_end_sector(rq);
607 q->boundary_rq = rq;
608 }
609 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
3aa72873
JA
610 (where == ELEVATOR_INSERT_SORT ||
611 where == ELEVATOR_INSERT_SORT_MERGE))
b710a480
JA
612 where = ELEVATOR_INSERT_BACK;
613
8922e16c 614 switch (where) {
28e7d184 615 case ELEVATOR_INSERT_REQUEUE:
8922e16c 616 case ELEVATOR_INSERT_FRONT:
4aff5e23 617 rq->cmd_flags |= REQ_SOFTBARRIER;
8922e16c
TH
618 list_add(&rq->queuelist, &q->queue_head);
619 break;
620
621 case ELEVATOR_INSERT_BACK:
4aff5e23 622 rq->cmd_flags |= REQ_SOFTBARRIER;
15853af9 623 elv_drain_elevator(q);
8922e16c
TH
624 list_add_tail(&rq->queuelist, &q->queue_head);
625 /*
626 * We kick the queue here for the following reasons.
627 * - The elevator might have returned NULL previously
628 * to delay requests and returned them now. As the
629 * queue wasn't empty before this request, ll_rw_blk
630 * won't run the queue on return, resulting in hang.
631 * - Usually, back inserted requests won't be merged
632 * with anything. There's no point in delaying queue
633 * processing.
634 */
24ecfbe2 635 __blk_run_queue(q);
8922e16c
TH
636 break;
637
5e84ea3a
JA
638 case ELEVATOR_INSERT_SORT_MERGE:
639 /*
640 * If we succeed in merging this request with one in the
641 * queue already, we are done - rq has now been freed,
642 * so no need to do anything further.
643 */
644 if (elv_attempt_insert_merge(q, rq))
645 break;
8922e16c 646 case ELEVATOR_INSERT_SORT:
e2a60da7 647 BUG_ON(rq->cmd_type != REQ_TYPE_FS);
4aff5e23 648 rq->cmd_flags |= REQ_SORTED;
15853af9 649 q->nr_sorted++;
9817064b
JA
650 if (rq_mergeable(rq)) {
651 elv_rqhash_add(q, rq);
652 if (!q->last_merge)
653 q->last_merge = rq;
654 }
655
ca23509f
TH
656 /*
657 * Some ioscheds (cfq) run q->request_fn directly, so
658 * rq cannot be accessed after calling
659 * elevator_add_req_fn.
660 */
22f746e2 661 q->elevator->type->ops.elevator_add_req_fn(q, rq);
8922e16c
TH
662 break;
663
ae1b1539
TH
664 case ELEVATOR_INSERT_FLUSH:
665 rq->cmd_flags |= REQ_SOFTBARRIER;
666 blk_insert_flush(rq);
667 break;
8922e16c
TH
668 default:
669 printk(KERN_ERR "%s: bad insertion point %d\n",
24c03d47 670 __func__, where);
8922e16c
TH
671 BUG();
672 }
1da177e4 673}
2e662b65
JA
674EXPORT_SYMBOL(__elv_add_request);
675
7eaceacc 676void elv_add_request(struct request_queue *q, struct request *rq, int where)
1da177e4
LT
677{
678 unsigned long flags;
679
680 spin_lock_irqsave(q->queue_lock, flags);
7eaceacc 681 __elv_add_request(q, rq, where);
1da177e4
LT
682 spin_unlock_irqrestore(q->queue_lock, flags);
683}
2e662b65
JA
684EXPORT_SYMBOL(elv_add_request);
685
165125e1 686struct request *elv_latter_request(struct request_queue *q, struct request *rq)
1da177e4 687{
b374d18a 688 struct elevator_queue *e = q->elevator;
1da177e4 689
22f746e2
TH
690 if (e->type->ops.elevator_latter_req_fn)
691 return e->type->ops.elevator_latter_req_fn(q, rq);
1da177e4
LT
692 return NULL;
693}
694
165125e1 695struct request *elv_former_request(struct request_queue *q, struct request *rq)
1da177e4 696{
b374d18a 697 struct elevator_queue *e = q->elevator;
1da177e4 698
22f746e2
TH
699 if (e->type->ops.elevator_former_req_fn)
700 return e->type->ops.elevator_former_req_fn(q, rq);
1da177e4
LT
701 return NULL;
702}
703
852c788f
TH
704int elv_set_request(struct request_queue *q, struct request *rq,
705 struct bio *bio, gfp_t gfp_mask)
1da177e4 706{
b374d18a 707 struct elevator_queue *e = q->elevator;
1da177e4 708
22f746e2 709 if (e->type->ops.elevator_set_req_fn)
852c788f 710 return e->type->ops.elevator_set_req_fn(q, rq, bio, gfp_mask);
1da177e4
LT
711 return 0;
712}
713
165125e1 714void elv_put_request(struct request_queue *q, struct request *rq)
1da177e4 715{
b374d18a 716 struct elevator_queue *e = q->elevator;
1da177e4 717
22f746e2
TH
718 if (e->type->ops.elevator_put_req_fn)
719 e->type->ops.elevator_put_req_fn(rq);
1da177e4
LT
720}
721
165125e1 722int elv_may_queue(struct request_queue *q, int rw)
1da177e4 723{
b374d18a 724 struct elevator_queue *e = q->elevator;
1da177e4 725
22f746e2
TH
726 if (e->type->ops.elevator_may_queue_fn)
727 return e->type->ops.elevator_may_queue_fn(q, rw);
1da177e4
LT
728
729 return ELV_MQUEUE_MAY;
730}
731
165125e1 732void elv_completed_request(struct request_queue *q, struct request *rq)
1da177e4 733{
b374d18a 734 struct elevator_queue *e = q->elevator;
1da177e4
LT
735
736 /*
737 * request is released from the driver, io must be done
738 */
8922e16c 739 if (blk_account_rq(rq)) {
0a7ae2ff 740 q->in_flight[rq_is_sync(rq)]--;
33659ebb 741 if ((rq->cmd_flags & REQ_SORTED) &&
22f746e2
TH
742 e->type->ops.elevator_completed_req_fn)
743 e->type->ops.elevator_completed_req_fn(q, rq);
1bc691d3 744 }
1da177e4
LT
745}
746
3d1ab40f
AV
747#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
748
749static ssize_t
750elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1da177e4 751{
3d1ab40f 752 struct elv_fs_entry *entry = to_elv(attr);
b374d18a 753 struct elevator_queue *e;
3d1ab40f
AV
754 ssize_t error;
755
756 if (!entry->show)
757 return -EIO;
758
b374d18a 759 e = container_of(kobj, struct elevator_queue, kobj);
3d1ab40f 760 mutex_lock(&e->sysfs_lock);
22f746e2 761 error = e->type ? entry->show(e, page) : -ENOENT;
3d1ab40f
AV
762 mutex_unlock(&e->sysfs_lock);
763 return error;
764}
1da177e4 765
3d1ab40f
AV
766static ssize_t
767elv_attr_store(struct kobject *kobj, struct attribute *attr,
768 const char *page, size_t length)
769{
3d1ab40f 770 struct elv_fs_entry *entry = to_elv(attr);
b374d18a 771 struct elevator_queue *e;
3d1ab40f 772 ssize_t error;
1da177e4 773
3d1ab40f
AV
774 if (!entry->store)
775 return -EIO;
1da177e4 776
b374d18a 777 e = container_of(kobj, struct elevator_queue, kobj);
3d1ab40f 778 mutex_lock(&e->sysfs_lock);
22f746e2 779 error = e->type ? entry->store(e, page, length) : -ENOENT;
3d1ab40f
AV
780 mutex_unlock(&e->sysfs_lock);
781 return error;
782}
783
52cf25d0 784static const struct sysfs_ops elv_sysfs_ops = {
3d1ab40f
AV
785 .show = elv_attr_show,
786 .store = elv_attr_store,
787};
788
789static struct kobj_type elv_ktype = {
790 .sysfs_ops = &elv_sysfs_ops,
791 .release = elevator_release,
792};
793
5a5bafdc 794int elv_register_queue(struct request_queue *q)
3d1ab40f 795{
5a5bafdc 796 struct elevator_queue *e = q->elevator;
3d1ab40f
AV
797 int error;
798
b2d6db58 799 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
3d1ab40f 800 if (!error) {
22f746e2 801 struct elv_fs_entry *attr = e->type->elevator_attrs;
3d1ab40f 802 if (attr) {
e572ec7e
AV
803 while (attr->attr.name) {
804 if (sysfs_create_file(&e->kobj, &attr->attr))
3d1ab40f 805 break;
e572ec7e 806 attr++;
3d1ab40f
AV
807 }
808 }
809 kobject_uevent(&e->kobj, KOBJ_ADD);
430c62fb 810 e->registered = 1;
3d1ab40f
AV
811 }
812 return error;
1da177e4 813}
f8fc877d 814EXPORT_SYMBOL(elv_register_queue);
bc1c1169 815
1da177e4
LT
816void elv_unregister_queue(struct request_queue *q)
817{
f8fc877d
TH
818 if (q) {
819 struct elevator_queue *e = q->elevator;
820
821 kobject_uevent(&e->kobj, KOBJ_REMOVE);
822 kobject_del(&e->kobj);
823 e->registered = 0;
824 }
1da177e4 825}
01effb0d 826EXPORT_SYMBOL(elv_unregister_queue);
1da177e4 827
e567bf71 828int elv_register(struct elevator_type *e)
1da177e4 829{
1ffb96c5 830 char *def = "";
2a12dcd7 831
3d3c2379
TH
832 /* create icq_cache if requested */
833 if (e->icq_size) {
834 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
835 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
836 return -EINVAL;
837
838 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
839 "%s_io_cq", e->elevator_name);
840 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
841 e->icq_align, 0, NULL);
842 if (!e->icq_cache)
843 return -ENOMEM;
844 }
845
846 /* register, don't allow duplicate names */
2a12dcd7 847 spin_lock(&elv_list_lock);
3d3c2379
TH
848 if (elevator_find(e->elevator_name)) {
849 spin_unlock(&elv_list_lock);
850 if (e->icq_cache)
851 kmem_cache_destroy(e->icq_cache);
852 return -EBUSY;
853 }
1da177e4 854 list_add_tail(&e->list, &elv_list);
2a12dcd7 855 spin_unlock(&elv_list_lock);
1da177e4 856
3d3c2379 857 /* print pretty message */
5f003976
ND
858 if (!strcmp(e->elevator_name, chosen_elevator) ||
859 (!*chosen_elevator &&
860 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
1ffb96c5
TV
861 def = " (default)";
862
4eb166d9
JA
863 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
864 def);
3d3c2379 865 return 0;
1da177e4
LT
866}
867EXPORT_SYMBOL_GPL(elv_register);
868
869void elv_unregister(struct elevator_type *e)
870{
3d3c2379 871 /* unregister */
2a12dcd7 872 spin_lock(&elv_list_lock);
1da177e4 873 list_del_init(&e->list);
2a12dcd7 874 spin_unlock(&elv_list_lock);
3d3c2379
TH
875
876 /*
877 * Destroy icq_cache if it exists. icq's are RCU managed. Make
878 * sure all RCU operations are complete before proceeding.
879 */
880 if (e->icq_cache) {
881 rcu_barrier();
882 kmem_cache_destroy(e->icq_cache);
883 e->icq_cache = NULL;
884 }
1da177e4
LT
885}
886EXPORT_SYMBOL_GPL(elv_unregister);
887
888/*
889 * switch to new_e io scheduler. be careful not to introduce deadlocks -
890 * we don't free the old io scheduler, before we have allocated what we
891 * need for the new one. this way we have a chance of going back to the old
cb98fc8b 892 * one, if the new one fails init for some reason.
1da177e4 893 */
165125e1 894static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
1da177e4 895{
5a5bafdc
TH
896 struct elevator_queue *old = q->elevator;
897 bool registered = old->registered;
e8989fae 898 int err;
1da177e4 899
5a5bafdc
TH
900 /*
901 * Turn on BYPASS and drain all requests w/ elevator private data.
902 * Block layer doesn't call into a quiesced elevator - all requests
903 * are directly put on the dispatch list without elevator data
904 * using INSERT_BACK. All requests have SOFTBARRIER set and no
905 * merge happens either.
906 */
d732580b 907 blk_queue_bypass_start(q);
cb98fc8b 908
5a5bafdc
TH
909 /* unregister and clear all auxiliary data of the old elevator */
910 if (registered)
f8fc877d 911 elv_unregister_queue(q);
1da177e4 912
f8fc877d 913 spin_lock_irq(q->queue_lock);
7e5a8794 914 ioc_clear_queue(q);
f8fc877d
TH
915 spin_unlock_irq(q->queue_lock);
916
5a5bafdc 917 /* allocate, init and register new elevator */
d50235b7
JM
918 err = new_e->ops.elevator_init_fn(q, new_e);
919 if (err)
5a5bafdc
TH
920 goto fail_init;
921
5a5bafdc
TH
922 if (registered) {
923 err = elv_register_queue(q);
924 if (err)
925 goto fail_register;
926 }
927
928 /* done, kill the old one and finish */
929 elevator_exit(old);
d732580b 930 blk_queue_bypass_end(q);
75ad23bc 931
5a5bafdc 932 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
4722dc52 933
5dd531a0 934 return 0;
1da177e4
LT
935
936fail_register:
5a5bafdc
TH
937 elevator_exit(q->elevator);
938fail_init:
939 /* switch failed, restore and re-register old elevator */
940 q->elevator = old;
1da177e4 941 elv_register_queue(q);
d732580b 942 blk_queue_bypass_end(q);
75ad23bc 943
5dd531a0 944 return err;
1da177e4
LT
945}
946
5dd531a0
JA
947/*
948 * Switch this queue to the given IO scheduler.
949 */
7c8a3679 950static int __elevator_change(struct request_queue *q, const char *name)
1da177e4
LT
951{
952 char elevator_name[ELV_NAME_MAX];
953 struct elevator_type *e;
954
cd43e26f 955 if (!q->elevator)
5dd531a0 956 return -ENXIO;
cd43e26f 957
ee2e992c 958 strlcpy(elevator_name, name, sizeof(elevator_name));
21c3c5d2 959 e = elevator_get(strstrip(elevator_name), true);
1da177e4
LT
960 if (!e) {
961 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
962 return -EINVAL;
963 }
964
22f746e2 965 if (!strcmp(elevator_name, q->elevator->type->elevator_name)) {
2ca7d93b 966 elevator_put(e);
5dd531a0 967 return 0;
2ca7d93b 968 }
1da177e4 969
5dd531a0
JA
970 return elevator_switch(q, e);
971}
7c8a3679
TS
972
973int elevator_change(struct request_queue *q, const char *name)
974{
975 int ret;
976
977 /* Protect q->elevator from elevator_init() */
978 mutex_lock(&q->sysfs_lock);
979 ret = __elevator_change(q, name);
980 mutex_unlock(&q->sysfs_lock);
981
982 return ret;
983}
5dd531a0
JA
984EXPORT_SYMBOL(elevator_change);
985
986ssize_t elv_iosched_store(struct request_queue *q, const char *name,
987 size_t count)
988{
989 int ret;
990
991 if (!q->elevator)
992 return count;
993
7c8a3679 994 ret = __elevator_change(q, name);
5dd531a0
JA
995 if (!ret)
996 return count;
997
998 printk(KERN_ERR "elevator: switch to %s failed\n", name);
999 return ret;
1da177e4
LT
1000}
1001
165125e1 1002ssize_t elv_iosched_show(struct request_queue *q, char *name)
1da177e4 1003{
b374d18a 1004 struct elevator_queue *e = q->elevator;
cd43e26f 1005 struct elevator_type *elv;
70cee26e 1006 struct elevator_type *__e;
1da177e4
LT
1007 int len = 0;
1008
e36f724b 1009 if (!q->elevator || !blk_queue_stackable(q))
cd43e26f
MP
1010 return sprintf(name, "none\n");
1011
22f746e2 1012 elv = e->type;
cd43e26f 1013
2a12dcd7 1014 spin_lock(&elv_list_lock);
70cee26e 1015 list_for_each_entry(__e, &elv_list, list) {
1da177e4
LT
1016 if (!strcmp(elv->elevator_name, __e->elevator_name))
1017 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1018 else
1019 len += sprintf(name+len, "%s ", __e->elevator_name);
1020 }
2a12dcd7 1021 spin_unlock(&elv_list_lock);
1da177e4
LT
1022
1023 len += sprintf(len+name, "\n");
1024 return len;
1025}
1026
165125e1
JA
1027struct request *elv_rb_former_request(struct request_queue *q,
1028 struct request *rq)
2e662b65
JA
1029{
1030 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1031
1032 if (rbprev)
1033 return rb_entry_rq(rbprev);
1034
1035 return NULL;
1036}
2e662b65
JA
1037EXPORT_SYMBOL(elv_rb_former_request);
1038
165125e1
JA
1039struct request *elv_rb_latter_request(struct request_queue *q,
1040 struct request *rq)
2e662b65
JA
1041{
1042 struct rb_node *rbnext = rb_next(&rq->rb_node);
1043
1044 if (rbnext)
1045 return rb_entry_rq(rbnext);
1046
1047 return NULL;
1048}
2e662b65 1049EXPORT_SYMBOL(elv_rb_latter_request);