1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26#include <linux/kernel.h>
27#include <linux/fs.h>
28#include <linux/blkdev.h>
29#include <linux/elevator.h>
30#include <linux/bio.h>
31#include <linux/module.h>
32#include <linux/slab.h>
33#include <linux/init.h>
34#include <linux/compiler.h>
35#include <linux/blktrace_api.h>
36#include <linux/hash.h>
37#include <linux/uaccess.h>
38#include <linux/pm_runtime.h>
39#include <linux/blk-cgroup.h>
40
41#include <trace/events/block.h>
42
43#include "blk.h"
44#include "blk-mq-sched.h"
45#include "blk-pm.h"
46#include "blk-wbt.h"
47
48static DEFINE_SPINLOCK(elv_list_lock);
49static LIST_HEAD(elv_list);
50
51
52
53
54#define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
55
56
57
58
59
60static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio)
61{
62 struct request_queue *q = rq->q;
63 struct elevator_queue *e = q->elevator;
64
65 if (e->type->ops.allow_merge)
66 return e->type->ops.allow_merge(q, rq, bio);
67
68 return 1;
69}
70
71
72
73
74bool elv_bio_merge_ok(struct request *rq, struct bio *bio)
75{
76 if (!blk_rq_merge_ok(rq, bio))
77 return false;
78
79 if (!elv_iosched_allow_bio_merge(rq, bio))
80 return false;
81
82 return true;
83}
84EXPORT_SYMBOL(elv_bio_merge_ok);
85
86static inline bool elv_support_features(unsigned int elv_features,
87 unsigned int required_features)
88{
89 return (required_features & elv_features) == required_features;
90}
91
92
93
94
95
96
97
98
99
100
101static bool elevator_match(const struct elevator_type *e, const char *name,
102 unsigned int required_features)
103{
104 if (!elv_support_features(e->elevator_features, required_features))
105 return false;
106 if (!strcmp(e->elevator_name, name))
107 return true;
108 if (e->elevator_alias && !strcmp(e->elevator_alias, name))
109 return true;
110
111 return false;
112}
113
114
115
116
117
118
119
120
121
122static struct elevator_type *elevator_find(const char *name,
123 unsigned int required_features)
124{
125 struct elevator_type *e;
126
127 list_for_each_entry(e, &elv_list, list) {
128 if (elevator_match(e, name, required_features))
129 return e;
130 }
131
132 return NULL;
133}
134
135static void elevator_put(struct elevator_type *e)
136{
137 module_put(e->elevator_owner);
138}
139
140static struct elevator_type *elevator_get(struct request_queue *q,
141 const char *name, bool try_loading)
142{
143 struct elevator_type *e;
144
145 spin_lock(&elv_list_lock);
146
147 e = elevator_find(name, q->required_elevator_features);
148 if (!e && try_loading) {
149 spin_unlock(&elv_list_lock);
150 request_module("%s-iosched", name);
151 spin_lock(&elv_list_lock);
152 e = elevator_find(name, q->required_elevator_features);
153 }
154
155 if (e && !try_module_get(e->elevator_owner))
156 e = NULL;
157
158 spin_unlock(&elv_list_lock);
159 return e;
160}
161
162static struct kobj_type elv_ktype;
163
164struct elevator_queue *elevator_alloc(struct request_queue *q,
165 struct elevator_type *e)
166{
167 struct elevator_queue *eq;
168
169 eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node);
170 if (unlikely(!eq))
171 return NULL;
172
173 eq->type = e;
174 kobject_init(&eq->kobj, &elv_ktype);
175 mutex_init(&eq->sysfs_lock);
176 hash_init(eq->hash);
177
178 return eq;
179}
180EXPORT_SYMBOL(elevator_alloc);
181
182static void elevator_release(struct kobject *kobj)
183{
184 struct elevator_queue *e;
185
186 e = container_of(kobj, struct elevator_queue, kobj);
187 elevator_put(e->type);
188 kfree(e);
189}
190
191void __elevator_exit(struct request_queue *q, struct elevator_queue *e)
192{
193 mutex_lock(&e->sysfs_lock);
194 blk_mq_exit_sched(q, e);
195 mutex_unlock(&e->sysfs_lock);
196
197 kobject_put(&e->kobj);
198}
199
200static inline void __elv_rqhash_del(struct request *rq)
201{
202 hash_del(&rq->hash);
203 rq->rq_flags &= ~RQF_HASHED;
204}
205
206void elv_rqhash_del(struct request_queue *q, struct request *rq)
207{
208 if (ELV_ON_HASH(rq))
209 __elv_rqhash_del(rq);
210}
211EXPORT_SYMBOL_GPL(elv_rqhash_del);
212
213void elv_rqhash_add(struct request_queue *q, struct request *rq)
214{
215 struct elevator_queue *e = q->elevator;
216
217 BUG_ON(ELV_ON_HASH(rq));
218 hash_add(e->hash, &rq->hash, rq_hash_key(rq));
219 rq->rq_flags |= RQF_HASHED;
220}
221EXPORT_SYMBOL_GPL(elv_rqhash_add);
222
223void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
224{
225 __elv_rqhash_del(rq);
226 elv_rqhash_add(q, rq);
227}
228
229struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
230{
231 struct elevator_queue *e = q->elevator;
232 struct hlist_node *next;
233 struct request *rq;
234
235 hash_for_each_possible_safe(e->hash, rq, next, hash, offset) {
236 BUG_ON(!ELV_ON_HASH(rq));
237
238 if (unlikely(!rq_mergeable(rq))) {
239 __elv_rqhash_del(rq);
240 continue;
241 }
242
243 if (rq_hash_key(rq) == offset)
244 return rq;
245 }
246
247 return NULL;
248}
249
250
251
252
253
254void elv_rb_add(struct rb_root *root, struct request *rq)
255{
256 struct rb_node **p = &root->rb_node;
257 struct rb_node *parent = NULL;
258 struct request *__rq;
259
260 while (*p) {
261 parent = *p;
262 __rq = rb_entry(parent, struct request, rb_node);
263
264 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
265 p = &(*p)->rb_left;
266 else if (blk_rq_pos(rq) >= blk_rq_pos(__rq))
267 p = &(*p)->rb_right;
268 }
269
270 rb_link_node(&rq->rb_node, parent, p);
271 rb_insert_color(&rq->rb_node, root);
272}
273EXPORT_SYMBOL(elv_rb_add);
274
275void elv_rb_del(struct rb_root *root, struct request *rq)
276{
277 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
278 rb_erase(&rq->rb_node, root);
279 RB_CLEAR_NODE(&rq->rb_node);
280}
281EXPORT_SYMBOL(elv_rb_del);
282
283struct request *elv_rb_find(struct rb_root *root, sector_t sector)
284{
285 struct rb_node *n = root->rb_node;
286 struct request *rq;
287
288 while (n) {
289 rq = rb_entry(n, struct request, rb_node);
290
291 if (sector < blk_rq_pos(rq))
292 n = n->rb_left;
293 else if (sector > blk_rq_pos(rq))
294 n = n->rb_right;
295 else
296 return rq;
297 }
298
299 return NULL;
300}
301EXPORT_SYMBOL(elv_rb_find);
302
303enum elv_merge elv_merge(struct request_queue *q, struct request **req,
304 struct bio *bio)
305{
306 struct elevator_queue *e = q->elevator;
307 struct request *__rq;
308
309
310
311
312
313
314
315 if (blk_queue_nomerges(q) || !bio_mergeable(bio))
316 return ELEVATOR_NO_MERGE;
317
318
319
320
321 if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) {
322 enum elv_merge ret = blk_try_merge(q->last_merge, bio);
323
324 if (ret != ELEVATOR_NO_MERGE) {
325 *req = q->last_merge;
326 return ret;
327 }
328 }
329
330 if (blk_queue_noxmerges(q))
331 return ELEVATOR_NO_MERGE;
332
333
334
335
336 __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector);
337 if (__rq && elv_bio_merge_ok(__rq, bio)) {
338 *req = __rq;
339 return ELEVATOR_BACK_MERGE;
340 }
341
342 if (e->type->ops.request_merge)
343 return e->type->ops.request_merge(q, req, bio);
344
345 return ELEVATOR_NO_MERGE;
346}
347
348
349
350
351
352
353
354
355
356bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq,
357 struct list_head *free)
358{
359 struct request *__rq;
360 bool ret;
361
362 if (blk_queue_nomerges(q))
363 return false;
364
365
366
367
368 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) {
369 list_add(&rq->queuelist, free);
370 return true;
371 }
372
373 if (blk_queue_noxmerges(q))
374 return false;
375
376 ret = false;
377
378
379
380 while (1) {
381 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
382 if (!__rq || !blk_attempt_req_merge(q, __rq, rq))
383 break;
384
385 list_add(&rq->queuelist, free);
386
387 ret = true;
388 rq = __rq;
389 }
390
391 return ret;
392}
393
394void elv_merged_request(struct request_queue *q, struct request *rq,
395 enum elv_merge type)
396{
397 struct elevator_queue *e = q->elevator;
398
399 if (e->type->ops.request_merged)
400 e->type->ops.request_merged(q, rq, type);
401
402 if (type == ELEVATOR_BACK_MERGE)
403 elv_rqhash_reposition(q, rq);
404
405 q->last_merge = rq;
406}
407
408void elv_merge_requests(struct request_queue *q, struct request *rq,
409 struct request *next)
410{
411 struct elevator_queue *e = q->elevator;
412
413 if (e->type->ops.requests_merged)
414 e->type->ops.requests_merged(q, rq, next);
415
416 elv_rqhash_reposition(q, rq);
417 q->last_merge = rq;
418}
419
420struct request *elv_latter_request(struct request_queue *q, struct request *rq)
421{
422 struct elevator_queue *e = q->elevator;
423
424 if (e->type->ops.next_request)
425 return e->type->ops.next_request(q, rq);
426
427 return NULL;
428}
429
430struct request *elv_former_request(struct request_queue *q, struct request *rq)
431{
432 struct elevator_queue *e = q->elevator;
433
434 if (e->type->ops.former_request)
435 return e->type->ops.former_request(q, rq);
436
437 return NULL;
438}
439
440#define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
441
442static ssize_t
443elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
444{
445 struct elv_fs_entry *entry = to_elv(attr);
446 struct elevator_queue *e;
447 ssize_t error;
448
449 if (!entry->show)
450 return -EIO;
451
452 e = container_of(kobj, struct elevator_queue, kobj);
453 mutex_lock(&e->sysfs_lock);
454 error = e->type ? entry->show(e, page) : -ENOENT;
455 mutex_unlock(&e->sysfs_lock);
456 return error;
457}
458
459static ssize_t
460elv_attr_store(struct kobject *kobj, struct attribute *attr,
461 const char *page, size_t length)
462{
463 struct elv_fs_entry *entry = to_elv(attr);
464 struct elevator_queue *e;
465 ssize_t error;
466
467 if (!entry->store)
468 return -EIO;
469
470 e = container_of(kobj, struct elevator_queue, kobj);
471 mutex_lock(&e->sysfs_lock);
472 error = e->type ? entry->store(e, page, length) : -ENOENT;
473 mutex_unlock(&e->sysfs_lock);
474 return error;
475}
476
477static const struct sysfs_ops elv_sysfs_ops = {
478 .show = elv_attr_show,
479 .store = elv_attr_store,
480};
481
482static struct kobj_type elv_ktype = {
483 .sysfs_ops = &elv_sysfs_ops,
484 .release = elevator_release,
485};
486
487int elv_register_queue(struct request_queue *q, bool uevent)
488{
489 struct elevator_queue *e = q->elevator;
490 int error;
491
492 lockdep_assert_held(&q->sysfs_lock);
493
494 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
495 if (!error) {
496 struct elv_fs_entry *attr = e->type->elevator_attrs;
497 if (attr) {
498 while (attr->attr.name) {
499 if (sysfs_create_file(&e->kobj, &attr->attr))
500 break;
501 attr++;
502 }
503 }
504 if (uevent)
505 kobject_uevent(&e->kobj, KOBJ_ADD);
506
507 e->registered = 1;
508 }
509 return error;
510}
511
512void elv_unregister_queue(struct request_queue *q)
513{
514 lockdep_assert_held(&q->sysfs_lock);
515
516 if (q) {
517 struct elevator_queue *e = q->elevator;
518
519 kobject_uevent(&e->kobj, KOBJ_REMOVE);
520 kobject_del(&e->kobj);
521
522 e->registered = 0;
523
524 wbt_enable_default(q);
525 }
526}
527
528int elv_register(struct elevator_type *e)
529{
530
531 if (WARN_ON_ONCE(!e->ops.insert_requests || !e->ops.dispatch_request))
532 return -EINVAL;
533
534
535 if (e->icq_size) {
536 if (WARN_ON(e->icq_size < sizeof(struct io_cq)) ||
537 WARN_ON(e->icq_align < __alignof__(struct io_cq)))
538 return -EINVAL;
539
540 snprintf(e->icq_cache_name, sizeof(e->icq_cache_name),
541 "%s_io_cq", e->elevator_name);
542 e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size,
543 e->icq_align, 0, NULL);
544 if (!e->icq_cache)
545 return -ENOMEM;
546 }
547
548
549 spin_lock(&elv_list_lock);
550 if (elevator_find(e->elevator_name, 0)) {
551 spin_unlock(&elv_list_lock);
552 kmem_cache_destroy(e->icq_cache);
553 return -EBUSY;
554 }
555 list_add_tail(&e->list, &elv_list);
556 spin_unlock(&elv_list_lock);
557
558 printk(KERN_INFO "io scheduler %s registered\n", e->elevator_name);
559
560 return 0;
561}
562EXPORT_SYMBOL_GPL(elv_register);
563
564void elv_unregister(struct elevator_type *e)
565{
566
567 spin_lock(&elv_list_lock);
568 list_del_init(&e->list);
569 spin_unlock(&elv_list_lock);
570
571
572
573
574
575 if (e->icq_cache) {
576 rcu_barrier();
577 kmem_cache_destroy(e->icq_cache);
578 e->icq_cache = NULL;
579 }
580}
581EXPORT_SYMBOL_GPL(elv_unregister);
582
583int elevator_switch_mq(struct request_queue *q,
584 struct elevator_type *new_e)
585{
586 int ret;
587
588 lockdep_assert_held(&q->sysfs_lock);
589
590 if (q->elevator) {
591 if (q->elevator->registered)
592 elv_unregister_queue(q);
593
594 ioc_clear_queue(q);
595 elevator_exit(q, q->elevator);
596 }
597
598 ret = blk_mq_init_sched(q, new_e);
599 if (ret)
600 goto out;
601
602 if (new_e) {
603 ret = elv_register_queue(q, true);
604 if (ret) {
605 elevator_exit(q, q->elevator);
606 goto out;
607 }
608 }
609
610 if (new_e)
611 blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name);
612 else
613 blk_add_trace_msg(q, "elv switch: none");
614
615out:
616 return ret;
617}
618
619static inline bool elv_support_iosched(struct request_queue *q)
620{
621 if (!queue_is_mq(q) ||
622 (q->tag_set && (q->tag_set->flags & BLK_MQ_F_NO_SCHED)))
623 return false;
624 return true;
625}
626
627
628
629
630
631static struct elevator_type *elevator_get_default(struct request_queue *q)
632{
633 if (q->nr_hw_queues != 1 &&
634 !blk_mq_is_sbitmap_shared(q->tag_set->flags))
635 return NULL;
636
637 return elevator_get(q, "mq-deadline", false);
638}
639
640
641
642
643
644static struct elevator_type *elevator_get_by_features(struct request_queue *q)
645{
646 struct elevator_type *e, *found = NULL;
647
648 spin_lock(&elv_list_lock);
649
650 list_for_each_entry(e, &elv_list, list) {
651 if (elv_support_features(e->elevator_features,
652 q->required_elevator_features)) {
653 found = e;
654 break;
655 }
656 }
657
658 if (found && !try_module_get(found->elevator_owner))
659 found = NULL;
660
661 spin_unlock(&elv_list_lock);
662 return found;
663}
664
665
666
667
668
669
670
671void elevator_init_mq(struct request_queue *q)
672{
673 struct elevator_type *e;
674 int err;
675
676 if (!elv_support_iosched(q))
677 return;
678
679 WARN_ON_ONCE(blk_queue_registered(q));
680
681 if (unlikely(q->elevator))
682 return;
683
684 if (!q->required_elevator_features)
685 e = elevator_get_default(q);
686 else
687 e = elevator_get_by_features(q);
688 if (!e)
689 return;
690
691 blk_mq_freeze_queue(q);
692 blk_mq_quiesce_queue(q);
693
694 err = blk_mq_init_sched(q, e);
695
696 blk_mq_unquiesce_queue(q);
697 blk_mq_unfreeze_queue(q);
698
699 if (err) {
700 pr_warn("\"%s\" elevator initialization failed, "
701 "falling back to \"none\"\n", e->elevator_name);
702 elevator_put(e);
703 }
704}
705EXPORT_SYMBOL_GPL(elevator_init_mq);
706
707
708
709
710
711
712
713static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
714{
715 int err;
716
717 lockdep_assert_held(&q->sysfs_lock);
718
719 blk_mq_freeze_queue(q);
720 blk_mq_quiesce_queue(q);
721
722 err = elevator_switch_mq(q, new_e);
723
724 blk_mq_unquiesce_queue(q);
725 blk_mq_unfreeze_queue(q);
726
727 return err;
728}
729
730
731
732
733static int __elevator_change(struct request_queue *q, const char *name)
734{
735 char elevator_name[ELV_NAME_MAX];
736 struct elevator_type *e;
737
738
739 if (!blk_queue_registered(q))
740 return -ENOENT;
741
742
743
744
745 if (!strncmp(name, "none", 4)) {
746 if (!q->elevator)
747 return 0;
748 return elevator_switch(q, NULL);
749 }
750
751 strlcpy(elevator_name, name, sizeof(elevator_name));
752 e = elevator_get(q, strstrip(elevator_name), true);
753 if (!e)
754 return -EINVAL;
755
756 if (q->elevator &&
757 elevator_match(q->elevator->type, elevator_name, 0)) {
758 elevator_put(e);
759 return 0;
760 }
761
762 return elevator_switch(q, e);
763}
764
765ssize_t elv_iosched_store(struct request_queue *q, const char *name,
766 size_t count)
767{
768 int ret;
769
770 if (!elv_support_iosched(q))
771 return count;
772
773 ret = __elevator_change(q, name);
774 if (!ret)
775 return count;
776
777 return ret;
778}
779
780ssize_t elv_iosched_show(struct request_queue *q, char *name)
781{
782 struct elevator_queue *e = q->elevator;
783 struct elevator_type *elv = NULL;
784 struct elevator_type *__e;
785 int len = 0;
786
787 if (!queue_is_mq(q))
788 return sprintf(name, "none\n");
789
790 if (!q->elevator)
791 len += sprintf(name+len, "[none] ");
792 else
793 elv = e->type;
794
795 spin_lock(&elv_list_lock);
796 list_for_each_entry(__e, &elv_list, list) {
797 if (elv && elevator_match(elv, __e->elevator_name, 0)) {
798 len += sprintf(name+len, "[%s] ", elv->elevator_name);
799 continue;
800 }
801 if (elv_support_iosched(q) &&
802 elevator_match(__e, __e->elevator_name,
803 q->required_elevator_features))
804 len += sprintf(name+len, "%s ", __e->elevator_name);
805 }
806 spin_unlock(&elv_list_lock);
807
808 if (q->elevator)
809 len += sprintf(name+len, "none");
810
811 len += sprintf(len+name, "\n");
812 return len;
813}
814
815struct request *elv_rb_former_request(struct request_queue *q,
816 struct request *rq)
817{
818 struct rb_node *rbprev = rb_prev(&rq->rb_node);
819
820 if (rbprev)
821 return rb_entry_rq(rbprev);
822
823 return NULL;
824}
825EXPORT_SYMBOL(elv_rb_former_request);
826
827struct request *elv_rb_latter_request(struct request_queue *q,
828 struct request *rq)
829{
830 struct rb_node *rbnext = rb_next(&rq->rb_node);
831
832 if (rbnext)
833 return rb_entry_rq(rbnext);
834
835 return NULL;
836}
837EXPORT_SYMBOL(elv_rb_latter_request);
838
839static int __init elevator_setup(char *str)
840{
841 pr_warn("Kernel parameter elevator= does not have any effect anymore.\n"
842 "Please use sysfs to set IO scheduler for individual devices.\n");
843 return 1;
844}
845
846__setup("elevator=", elevator_setup);
847