2 * Block device elevator/IO-scheduler.
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * 30042000 Jens Axboe <axboe@kernel.dk> :
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
13 * - elevator_dequeue_fn, called when a request is taken off the active list
15 * 20082000 Dave Jones <davej@suse.de> :
16 * Removed tests for max-bomb-segments, which was breaking elvtune
17 * when run without -bN
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
25 #include <linux/kernel.h>
27 #include <linux/blkdev.h>
28 #include <linux/elevator.h>
29 #include <linux/bio.h>
30 #include <linux/module.h>
31 #include <linux/slab.h>
32 #include <linux/init.h>
33 #include <linux/compiler.h>
34 #include <linux/blktrace_api.h>
35 #include <linux/hash.h>
36 #include <linux/uaccess.h>
37 #include <linux/pm_runtime.h>
38 #include <linux/blk-cgroup.h>
40 #include <trace/events/block.h>
44 static DEFINE_SPINLOCK(elv_list_lock
);
45 static LIST_HEAD(elv_list
);
50 #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
53 * Query io scheduler to see if the current process issuing bio may be
56 static int elv_iosched_allow_bio_merge(struct request
*rq
, struct bio
*bio
)
58 struct request_queue
*q
= rq
->q
;
59 struct elevator_queue
*e
= q
->elevator
;
61 if (e
->type
->ops
.elevator_allow_bio_merge_fn
)
62 return e
->type
->ops
.elevator_allow_bio_merge_fn(q
, rq
, bio
);
68 * can we safely merge with this request?
70 bool elv_bio_merge_ok(struct request
*rq
, struct bio
*bio
)
72 if (!blk_rq_merge_ok(rq
, bio
))
75 if (!elv_iosched_allow_bio_merge(rq
, bio
))
80 EXPORT_SYMBOL(elv_bio_merge_ok
);
82 static struct elevator_type
*elevator_find(const char *name
)
84 struct elevator_type
*e
;
86 list_for_each_entry(e
, &elv_list
, list
) {
87 if (!strcmp(e
->elevator_name
, name
))
94 static void elevator_put(struct elevator_type
*e
)
96 module_put(e
->elevator_owner
);
99 static struct elevator_type
*elevator_get(const char *name
, bool try_loading
)
101 struct elevator_type
*e
;
103 spin_lock(&elv_list_lock
);
105 e
= elevator_find(name
);
106 if (!e
&& try_loading
) {
107 spin_unlock(&elv_list_lock
);
108 request_module("%s-iosched", name
);
109 spin_lock(&elv_list_lock
);
110 e
= elevator_find(name
);
113 if (e
&& !try_module_get(e
->elevator_owner
))
116 spin_unlock(&elv_list_lock
);
121 static char chosen_elevator
[ELV_NAME_MAX
];
123 static int __init
elevator_setup(char *str
)
126 * Be backwards-compatible with previous kernels, so users
127 * won't get the wrong elevator.
129 strncpy(chosen_elevator
, str
, sizeof(chosen_elevator
) - 1);
133 __setup("elevator=", elevator_setup
);
135 /* called during boot to load the elevator chosen by the elevator param */
136 void __init
load_default_elevator_module(void)
138 struct elevator_type
*e
;
140 if (!chosen_elevator
[0])
143 spin_lock(&elv_list_lock
);
144 e
= elevator_find(chosen_elevator
);
145 spin_unlock(&elv_list_lock
);
148 request_module("%s-iosched", chosen_elevator
);
151 static struct kobj_type elv_ktype
;
153 struct elevator_queue
*elevator_alloc(struct request_queue
*q
,
154 struct elevator_type
*e
)
156 struct elevator_queue
*eq
;
158 eq
= kzalloc_node(sizeof(*eq
), GFP_KERNEL
, q
->node
);
163 kobject_init(&eq
->kobj
, &elv_ktype
);
164 mutex_init(&eq
->sysfs_lock
);
169 EXPORT_SYMBOL(elevator_alloc
);
171 static void elevator_release(struct kobject
*kobj
)
173 struct elevator_queue
*e
;
175 e
= container_of(kobj
, struct elevator_queue
, kobj
);
176 elevator_put(e
->type
);
180 int elevator_init(struct request_queue
*q
, char *name
)
182 struct elevator_type
*e
= NULL
;
186 * q->sysfs_lock must be held to provide mutual exclusion between
187 * elevator_switch() and here.
189 lockdep_assert_held(&q
->sysfs_lock
);
191 if (unlikely(q
->elevator
))
194 INIT_LIST_HEAD(&q
->queue_head
);
195 q
->last_merge
= NULL
;
197 q
->boundary_rq
= NULL
;
200 e
= elevator_get(name
, true);
206 * Use the default elevator specified by config boot param or
207 * config option. Don't try to load modules as we could be running
208 * off async and request_module() isn't allowed from async.
210 if (!e
&& *chosen_elevator
) {
211 e
= elevator_get(chosen_elevator
, false);
213 printk(KERN_ERR
"I/O scheduler %s not found\n",
218 e
= elevator_get(CONFIG_DEFAULT_IOSCHED
, false);
221 "Default I/O scheduler not found. " \
223 e
= elevator_get("noop", false);
227 err
= e
->ops
.elevator_init_fn(q
, e
);
232 EXPORT_SYMBOL(elevator_init
);
234 void elevator_exit(struct elevator_queue
*e
)
236 mutex_lock(&e
->sysfs_lock
);
237 if (e
->type
->ops
.elevator_exit_fn
)
238 e
->type
->ops
.elevator_exit_fn(e
);
239 mutex_unlock(&e
->sysfs_lock
);
241 kobject_put(&e
->kobj
);
243 EXPORT_SYMBOL(elevator_exit
);
245 static inline void __elv_rqhash_del(struct request
*rq
)
248 rq
->rq_flags
&= ~RQF_HASHED
;
251 void elv_rqhash_del(struct request_queue
*q
, struct request
*rq
)
254 __elv_rqhash_del(rq
);
257 void elv_rqhash_add(struct request_queue
*q
, struct request
*rq
)
259 struct elevator_queue
*e
= q
->elevator
;
261 BUG_ON(ELV_ON_HASH(rq
));
262 hash_add(e
->hash
, &rq
->hash
, rq_hash_key(rq
));
263 rq
->rq_flags
|= RQF_HASHED
;
266 void elv_rqhash_reposition(struct request_queue
*q
, struct request
*rq
)
268 __elv_rqhash_del(rq
);
269 elv_rqhash_add(q
, rq
);
272 struct request
*elv_rqhash_find(struct request_queue
*q
, sector_t offset
)
274 struct elevator_queue
*e
= q
->elevator
;
275 struct hlist_node
*next
;
278 hash_for_each_possible_safe(e
->hash
, rq
, next
, hash
, offset
) {
279 BUG_ON(!ELV_ON_HASH(rq
));
281 if (unlikely(!rq_mergeable(rq
))) {
282 __elv_rqhash_del(rq
);
286 if (rq_hash_key(rq
) == offset
)
294 * RB-tree support functions for inserting/lookup/removal of requests
295 * in a sorted RB tree.
297 void elv_rb_add(struct rb_root
*root
, struct request
*rq
)
299 struct rb_node
**p
= &root
->rb_node
;
300 struct rb_node
*parent
= NULL
;
301 struct request
*__rq
;
305 __rq
= rb_entry(parent
, struct request
, rb_node
);
307 if (blk_rq_pos(rq
) < blk_rq_pos(__rq
))
309 else if (blk_rq_pos(rq
) >= blk_rq_pos(__rq
))
313 rb_link_node(&rq
->rb_node
, parent
, p
);
314 rb_insert_color(&rq
->rb_node
, root
);
316 EXPORT_SYMBOL(elv_rb_add
);
318 void elv_rb_del(struct rb_root
*root
, struct request
*rq
)
320 BUG_ON(RB_EMPTY_NODE(&rq
->rb_node
));
321 rb_erase(&rq
->rb_node
, root
);
322 RB_CLEAR_NODE(&rq
->rb_node
);
324 EXPORT_SYMBOL(elv_rb_del
);
326 struct request
*elv_rb_find(struct rb_root
*root
, sector_t sector
)
328 struct rb_node
*n
= root
->rb_node
;
332 rq
= rb_entry(n
, struct request
, rb_node
);
334 if (sector
< blk_rq_pos(rq
))
336 else if (sector
> blk_rq_pos(rq
))
344 EXPORT_SYMBOL(elv_rb_find
);
347 * Insert rq into dispatch queue of q. Queue lock must be held on
348 * entry. rq is sort instead into the dispatch queue. To be used by
349 * specific elevators.
351 void elv_dispatch_sort(struct request_queue
*q
, struct request
*rq
)
354 struct list_head
*entry
;
356 if (q
->last_merge
== rq
)
357 q
->last_merge
= NULL
;
359 elv_rqhash_del(q
, rq
);
363 boundary
= q
->end_sector
;
364 list_for_each_prev(entry
, &q
->queue_head
) {
365 struct request
*pos
= list_entry_rq(entry
);
367 if (req_op(rq
) != req_op(pos
))
369 if (rq_data_dir(rq
) != rq_data_dir(pos
))
371 if (pos
->rq_flags
& (RQF_STARTED
| RQF_SOFTBARRIER
))
373 if (blk_rq_pos(rq
) >= boundary
) {
374 if (blk_rq_pos(pos
) < boundary
)
377 if (blk_rq_pos(pos
) >= boundary
)
380 if (blk_rq_pos(rq
) >= blk_rq_pos(pos
))
384 list_add(&rq
->queuelist
, entry
);
386 EXPORT_SYMBOL(elv_dispatch_sort
);
389 * Insert rq into dispatch queue of q. Queue lock must be held on
390 * entry. rq is added to the back of the dispatch queue. To be used by
391 * specific elevators.
393 void elv_dispatch_add_tail(struct request_queue
*q
, struct request
*rq
)
395 if (q
->last_merge
== rq
)
396 q
->last_merge
= NULL
;
398 elv_rqhash_del(q
, rq
);
402 q
->end_sector
= rq_end_sector(rq
);
404 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
406 EXPORT_SYMBOL(elv_dispatch_add_tail
);
408 int elv_merge(struct request_queue
*q
, struct request
**req
, struct bio
*bio
)
410 struct elevator_queue
*e
= q
->elevator
;
411 struct request
*__rq
;
416 * nomerges: No merges at all attempted
417 * noxmerges: Only simple one-hit cache try
418 * merges: All merge tries attempted
420 if (blk_queue_nomerges(q
) || !bio_mergeable(bio
))
421 return ELEVATOR_NO_MERGE
;
424 * First try one-hit cache.
426 if (q
->last_merge
&& elv_bio_merge_ok(q
->last_merge
, bio
)) {
427 ret
= blk_try_merge(q
->last_merge
, bio
);
428 if (ret
!= ELEVATOR_NO_MERGE
) {
429 *req
= q
->last_merge
;
434 if (blk_queue_noxmerges(q
))
435 return ELEVATOR_NO_MERGE
;
438 * See if our hash lookup can find a potential backmerge.
440 __rq
= elv_rqhash_find(q
, bio
->bi_iter
.bi_sector
);
441 if (__rq
&& elv_bio_merge_ok(__rq
, bio
)) {
443 return ELEVATOR_BACK_MERGE
;
446 if (e
->type
->ops
.elevator_merge_fn
)
447 return e
->type
->ops
.elevator_merge_fn(q
, req
, bio
);
449 return ELEVATOR_NO_MERGE
;
453 * Attempt to do an insertion back merge. Only check for the case where
454 * we can append 'rq' to an existing request, so we can throw 'rq' away
457 * Returns true if we merged, false otherwise
459 static bool elv_attempt_insert_merge(struct request_queue
*q
,
462 struct request
*__rq
;
465 if (blk_queue_nomerges(q
))
469 * First try one-hit cache.
471 if (q
->last_merge
&& blk_attempt_req_merge(q
, q
->last_merge
, rq
))
474 if (blk_queue_noxmerges(q
))
479 * See if our hash lookup can find a potential backmerge.
482 __rq
= elv_rqhash_find(q
, blk_rq_pos(rq
));
483 if (!__rq
|| !blk_attempt_req_merge(q
, __rq
, rq
))
486 /* The merged request could be merged with others, try again */
494 void elv_merged_request(struct request_queue
*q
, struct request
*rq
, int type
)
496 struct elevator_queue
*e
= q
->elevator
;
498 if (e
->type
->ops
.elevator_merged_fn
)
499 e
->type
->ops
.elevator_merged_fn(q
, rq
, type
);
501 if (type
== ELEVATOR_BACK_MERGE
)
502 elv_rqhash_reposition(q
, rq
);
507 void elv_merge_requests(struct request_queue
*q
, struct request
*rq
,
508 struct request
*next
)
510 struct elevator_queue
*e
= q
->elevator
;
511 const int next_sorted
= next
->rq_flags
& RQF_SORTED
;
513 if (next_sorted
&& e
->type
->ops
.elevator_merge_req_fn
)
514 e
->type
->ops
.elevator_merge_req_fn(q
, rq
, next
);
516 elv_rqhash_reposition(q
, rq
);
519 elv_rqhash_del(q
, next
);
526 void elv_bio_merged(struct request_queue
*q
, struct request
*rq
,
529 struct elevator_queue
*e
= q
->elevator
;
531 if (e
->type
->ops
.elevator_bio_merged_fn
)
532 e
->type
->ops
.elevator_bio_merged_fn(q
, rq
, bio
);
536 static void blk_pm_requeue_request(struct request
*rq
)
538 if (rq
->q
->dev
&& !(rq
->rq_flags
& RQF_PM
))
542 static void blk_pm_add_request(struct request_queue
*q
, struct request
*rq
)
544 if (q
->dev
&& !(rq
->rq_flags
& RQF_PM
) && q
->nr_pending
++ == 0 &&
545 (q
->rpm_status
== RPM_SUSPENDED
|| q
->rpm_status
== RPM_SUSPENDING
))
546 pm_request_resume(q
->dev
);
549 static inline void blk_pm_requeue_request(struct request
*rq
) {}
550 static inline void blk_pm_add_request(struct request_queue
*q
,
556 void elv_requeue_request(struct request_queue
*q
, struct request
*rq
)
559 * it already went through dequeue, we need to decrement the
560 * in_flight count again
562 if (blk_account_rq(rq
)) {
563 q
->in_flight
[rq_is_sync(rq
)]--;
564 if (rq
->rq_flags
& RQF_SORTED
)
565 elv_deactivate_rq(q
, rq
);
568 rq
->rq_flags
&= ~RQF_STARTED
;
570 blk_pm_requeue_request(rq
);
572 __elv_add_request(q
, rq
, ELEVATOR_INSERT_REQUEUE
);
575 void elv_drain_elevator(struct request_queue
*q
)
579 lockdep_assert_held(q
->queue_lock
);
581 while (q
->elevator
->type
->ops
.elevator_dispatch_fn(q
, 1))
583 if (q
->nr_sorted
&& printed
++ < 10) {
584 printk(KERN_ERR
"%s: forced dispatching is broken "
585 "(nr_sorted=%u), please report this\n",
586 q
->elevator
->type
->elevator_name
, q
->nr_sorted
);
590 void __elv_add_request(struct request_queue
*q
, struct request
*rq
, int where
)
592 trace_block_rq_insert(q
, rq
);
594 blk_pm_add_request(q
, rq
);
598 if (rq
->rq_flags
& RQF_SOFTBARRIER
) {
599 /* barriers are scheduling boundary, update end_sector */
600 if (rq
->cmd_type
== REQ_TYPE_FS
) {
601 q
->end_sector
= rq_end_sector(rq
);
604 } else if (!(rq
->rq_flags
& RQF_ELVPRIV
) &&
605 (where
== ELEVATOR_INSERT_SORT
||
606 where
== ELEVATOR_INSERT_SORT_MERGE
))
607 where
= ELEVATOR_INSERT_BACK
;
610 case ELEVATOR_INSERT_REQUEUE
:
611 case ELEVATOR_INSERT_FRONT
:
612 rq
->rq_flags
|= RQF_SOFTBARRIER
;
613 list_add(&rq
->queuelist
, &q
->queue_head
);
616 case ELEVATOR_INSERT_BACK
:
617 rq
->rq_flags
|= RQF_SOFTBARRIER
;
618 elv_drain_elevator(q
);
619 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
621 * We kick the queue here for the following reasons.
622 * - The elevator might have returned NULL previously
623 * to delay requests and returned them now. As the
624 * queue wasn't empty before this request, ll_rw_blk
625 * won't run the queue on return, resulting in hang.
626 * - Usually, back inserted requests won't be merged
627 * with anything. There's no point in delaying queue
633 case ELEVATOR_INSERT_SORT_MERGE
:
635 * If we succeed in merging this request with one in the
636 * queue already, we are done - rq has now been freed,
637 * so no need to do anything further.
639 if (elv_attempt_insert_merge(q
, rq
))
641 case ELEVATOR_INSERT_SORT
:
642 BUG_ON(rq
->cmd_type
!= REQ_TYPE_FS
);
643 rq
->rq_flags
|= RQF_SORTED
;
645 if (rq_mergeable(rq
)) {
646 elv_rqhash_add(q
, rq
);
652 * Some ioscheds (cfq) run q->request_fn directly, so
653 * rq cannot be accessed after calling
654 * elevator_add_req_fn.
656 q
->elevator
->type
->ops
.elevator_add_req_fn(q
, rq
);
659 case ELEVATOR_INSERT_FLUSH
:
660 rq
->rq_flags
|= RQF_SOFTBARRIER
;
661 blk_insert_flush(rq
);
664 printk(KERN_ERR
"%s: bad insertion point %d\n",
669 EXPORT_SYMBOL(__elv_add_request
);
671 void elv_add_request(struct request_queue
*q
, struct request
*rq
, int where
)
675 spin_lock_irqsave(q
->queue_lock
, flags
);
676 __elv_add_request(q
, rq
, where
);
677 spin_unlock_irqrestore(q
->queue_lock
, flags
);
679 EXPORT_SYMBOL(elv_add_request
);
681 struct request
*elv_latter_request(struct request_queue
*q
, struct request
*rq
)
683 struct elevator_queue
*e
= q
->elevator
;
685 if (e
->type
->ops
.elevator_latter_req_fn
)
686 return e
->type
->ops
.elevator_latter_req_fn(q
, rq
);
690 struct request
*elv_former_request(struct request_queue
*q
, struct request
*rq
)
692 struct elevator_queue
*e
= q
->elevator
;
694 if (e
->type
->ops
.elevator_former_req_fn
)
695 return e
->type
->ops
.elevator_former_req_fn(q
, rq
);
699 int elv_set_request(struct request_queue
*q
, struct request
*rq
,
700 struct bio
*bio
, gfp_t gfp_mask
)
702 struct elevator_queue
*e
= q
->elevator
;
704 if (e
->type
->ops
.elevator_set_req_fn
)
705 return e
->type
->ops
.elevator_set_req_fn(q
, rq
, bio
, gfp_mask
);
709 void elv_put_request(struct request_queue
*q
, struct request
*rq
)
711 struct elevator_queue
*e
= q
->elevator
;
713 if (e
->type
->ops
.elevator_put_req_fn
)
714 e
->type
->ops
.elevator_put_req_fn(rq
);
717 int elv_may_queue(struct request_queue
*q
, unsigned int op
)
719 struct elevator_queue
*e
= q
->elevator
;
721 if (e
->type
->ops
.elevator_may_queue_fn
)
722 return e
->type
->ops
.elevator_may_queue_fn(q
, op
);
724 return ELV_MQUEUE_MAY
;
727 void elv_completed_request(struct request_queue
*q
, struct request
*rq
)
729 struct elevator_queue
*e
= q
->elevator
;
732 * request is released from the driver, io must be done
734 if (blk_account_rq(rq
)) {
735 q
->in_flight
[rq_is_sync(rq
)]--;
736 if ((rq
->rq_flags
& RQF_SORTED
) &&
737 e
->type
->ops
.elevator_completed_req_fn
)
738 e
->type
->ops
.elevator_completed_req_fn(q
, rq
);
742 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
745 elv_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
747 struct elv_fs_entry
*entry
= to_elv(attr
);
748 struct elevator_queue
*e
;
754 e
= container_of(kobj
, struct elevator_queue
, kobj
);
755 mutex_lock(&e
->sysfs_lock
);
756 error
= e
->type
? entry
->show(e
, page
) : -ENOENT
;
757 mutex_unlock(&e
->sysfs_lock
);
762 elv_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
763 const char *page
, size_t length
)
765 struct elv_fs_entry
*entry
= to_elv(attr
);
766 struct elevator_queue
*e
;
772 e
= container_of(kobj
, struct elevator_queue
, kobj
);
773 mutex_lock(&e
->sysfs_lock
);
774 error
= e
->type
? entry
->store(e
, page
, length
) : -ENOENT
;
775 mutex_unlock(&e
->sysfs_lock
);
779 static const struct sysfs_ops elv_sysfs_ops
= {
780 .show
= elv_attr_show
,
781 .store
= elv_attr_store
,
784 static struct kobj_type elv_ktype
= {
785 .sysfs_ops
= &elv_sysfs_ops
,
786 .release
= elevator_release
,
789 int elv_register_queue(struct request_queue
*q
)
791 struct elevator_queue
*e
= q
->elevator
;
794 error
= kobject_add(&e
->kobj
, &q
->kobj
, "%s", "iosched");
796 struct elv_fs_entry
*attr
= e
->type
->elevator_attrs
;
798 while (attr
->attr
.name
) {
799 if (sysfs_create_file(&e
->kobj
, &attr
->attr
))
804 kobject_uevent(&e
->kobj
, KOBJ_ADD
);
806 if (e
->type
->ops
.elevator_registered_fn
)
807 e
->type
->ops
.elevator_registered_fn(q
);
811 EXPORT_SYMBOL(elv_register_queue
);
813 void elv_unregister_queue(struct request_queue
*q
)
816 struct elevator_queue
*e
= q
->elevator
;
818 kobject_uevent(&e
->kobj
, KOBJ_REMOVE
);
819 kobject_del(&e
->kobj
);
823 EXPORT_SYMBOL(elv_unregister_queue
);
825 int elv_register(struct elevator_type
*e
)
829 /* create icq_cache if requested */
831 if (WARN_ON(e
->icq_size
< sizeof(struct io_cq
)) ||
832 WARN_ON(e
->icq_align
< __alignof__(struct io_cq
)))
835 snprintf(e
->icq_cache_name
, sizeof(e
->icq_cache_name
),
836 "%s_io_cq", e
->elevator_name
);
837 e
->icq_cache
= kmem_cache_create(e
->icq_cache_name
, e
->icq_size
,
838 e
->icq_align
, 0, NULL
);
843 /* register, don't allow duplicate names */
844 spin_lock(&elv_list_lock
);
845 if (elevator_find(e
->elevator_name
)) {
846 spin_unlock(&elv_list_lock
);
848 kmem_cache_destroy(e
->icq_cache
);
851 list_add_tail(&e
->list
, &elv_list
);
852 spin_unlock(&elv_list_lock
);
854 /* print pretty message */
855 if (!strcmp(e
->elevator_name
, chosen_elevator
) ||
856 (!*chosen_elevator
&&
857 !strcmp(e
->elevator_name
, CONFIG_DEFAULT_IOSCHED
)))
860 printk(KERN_INFO
"io scheduler %s registered%s\n", e
->elevator_name
,
864 EXPORT_SYMBOL_GPL(elv_register
);
866 void elv_unregister(struct elevator_type
*e
)
869 spin_lock(&elv_list_lock
);
870 list_del_init(&e
->list
);
871 spin_unlock(&elv_list_lock
);
874 * Destroy icq_cache if it exists. icq's are RCU managed. Make
875 * sure all RCU operations are complete before proceeding.
879 kmem_cache_destroy(e
->icq_cache
);
883 EXPORT_SYMBOL_GPL(elv_unregister
);
886 * switch to new_e io scheduler. be careful not to introduce deadlocks -
887 * we don't free the old io scheduler, before we have allocated what we
888 * need for the new one. this way we have a chance of going back to the old
889 * one, if the new one fails init for some reason.
891 static int elevator_switch(struct request_queue
*q
, struct elevator_type
*new_e
)
893 struct elevator_queue
*old
= q
->elevator
;
894 bool registered
= old
->registered
;
898 * Turn on BYPASS and drain all requests w/ elevator private data.
899 * Block layer doesn't call into a quiesced elevator - all requests
900 * are directly put on the dispatch list without elevator data
901 * using INSERT_BACK. All requests have SOFTBARRIER set and no
902 * merge happens either.
904 blk_queue_bypass_start(q
);
906 /* unregister and clear all auxiliary data of the old elevator */
908 elv_unregister_queue(q
);
910 spin_lock_irq(q
->queue_lock
);
912 spin_unlock_irq(q
->queue_lock
);
914 /* allocate, init and register new elevator */
915 err
= new_e
->ops
.elevator_init_fn(q
, new_e
);
920 err
= elv_register_queue(q
);
925 /* done, kill the old one and finish */
927 blk_queue_bypass_end(q
);
929 blk_add_trace_msg(q
, "elv switch: %s", new_e
->elevator_name
);
934 elevator_exit(q
->elevator
);
936 /* switch failed, restore and re-register old elevator */
938 elv_register_queue(q
);
939 blk_queue_bypass_end(q
);
945 * Switch this queue to the given IO scheduler.
947 static int __elevator_change(struct request_queue
*q
, const char *name
)
949 char elevator_name
[ELV_NAME_MAX
];
950 struct elevator_type
*e
;
955 strlcpy(elevator_name
, name
, sizeof(elevator_name
));
956 e
= elevator_get(strstrip(elevator_name
), true);
958 printk(KERN_ERR
"elevator: type %s not found\n", elevator_name
);
962 if (!strcmp(elevator_name
, q
->elevator
->type
->elevator_name
)) {
967 return elevator_switch(q
, e
);
970 int elevator_change(struct request_queue
*q
, const char *name
)
974 /* Protect q->elevator from elevator_init() */
975 mutex_lock(&q
->sysfs_lock
);
976 ret
= __elevator_change(q
, name
);
977 mutex_unlock(&q
->sysfs_lock
);
981 EXPORT_SYMBOL(elevator_change
);
983 ssize_t
elv_iosched_store(struct request_queue
*q
, const char *name
,
991 ret
= __elevator_change(q
, name
);
995 printk(KERN_ERR
"elevator: switch to %s failed\n", name
);
999 ssize_t
elv_iosched_show(struct request_queue
*q
, char *name
)
1001 struct elevator_queue
*e
= q
->elevator
;
1002 struct elevator_type
*elv
;
1003 struct elevator_type
*__e
;
1006 if (!q
->elevator
|| !blk_queue_stackable(q
))
1007 return sprintf(name
, "none\n");
1011 spin_lock(&elv_list_lock
);
1012 list_for_each_entry(__e
, &elv_list
, list
) {
1013 if (!strcmp(elv
->elevator_name
, __e
->elevator_name
))
1014 len
+= sprintf(name
+len
, "[%s] ", elv
->elevator_name
);
1016 len
+= sprintf(name
+len
, "%s ", __e
->elevator_name
);
1018 spin_unlock(&elv_list_lock
);
1020 len
+= sprintf(len
+name
, "\n");
1024 struct request
*elv_rb_former_request(struct request_queue
*q
,
1027 struct rb_node
*rbprev
= rb_prev(&rq
->rb_node
);
1030 return rb_entry_rq(rbprev
);
1034 EXPORT_SYMBOL(elv_rb_former_request
);
1036 struct request
*elv_rb_latter_request(struct request_queue
*q
,
1039 struct rb_node
*rbnext
= rb_next(&rq
->rb_node
);
1042 return rb_entry_rq(rbnext
);
1046 EXPORT_SYMBOL(elv_rb_latter_request
);