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/delay.h>
35 #include <linux/blktrace_api.h>
36 #include <trace/block.h>
37 #include <linux/hash.h>
38 #include <linux/uaccess.h>
42 static DEFINE_SPINLOCK(elv_list_lock
);
43 static LIST_HEAD(elv_list
);
45 DEFINE_TRACE(block_rq_abort
);
50 static const int elv_hash_shift
= 6;
51 #define ELV_HASH_BLOCK(sec) ((sec) >> 3)
52 #define ELV_HASH_FN(sec) \
53 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
54 #define ELV_HASH_ENTRIES (1 << elv_hash_shift)
55 #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
57 DEFINE_TRACE(block_rq_insert
);
58 DEFINE_TRACE(block_rq_issue
);
61 * Query io scheduler to see if the current process issuing bio may be
64 static int elv_iosched_allow_merge(struct request
*rq
, struct bio
*bio
)
66 struct request_queue
*q
= rq
->q
;
67 struct elevator_queue
*e
= q
->elevator
;
69 if (e
->ops
->elevator_allow_merge_fn
)
70 return e
->ops
->elevator_allow_merge_fn(q
, rq
, bio
);
76 * can we safely merge with this request?
78 int elv_rq_merge_ok(struct request
*rq
, struct bio
*bio
)
80 if (!rq_mergeable(rq
))
84 * Don't merge file system requests and discard requests
86 if (bio_discard(bio
) != bio_discard(rq
->bio
))
90 * different data direction or already started, don't merge
92 if (bio_data_dir(bio
) != rq_data_dir(rq
))
96 * must be same device and not a special request
98 if (rq
->rq_disk
!= bio
->bi_bdev
->bd_disk
|| rq
->special
)
102 * only merge integrity protected bio into ditto rq
104 if (bio_integrity(bio
) != blk_integrity_rq(rq
))
107 if (!elv_iosched_allow_merge(rq
, bio
))
112 EXPORT_SYMBOL(elv_rq_merge_ok
);
114 static inline int elv_try_merge(struct request
*__rq
, struct bio
*bio
)
116 int ret
= ELEVATOR_NO_MERGE
;
119 * we can merge and sequence is ok, check if it's possible
121 if (elv_rq_merge_ok(__rq
, bio
)) {
122 if (blk_rq_pos(__rq
) + blk_rq_sectors(__rq
) == bio
->bi_sector
)
123 ret
= ELEVATOR_BACK_MERGE
;
124 else if (blk_rq_pos(__rq
) - bio_sectors(bio
) == bio
->bi_sector
)
125 ret
= ELEVATOR_FRONT_MERGE
;
131 static struct elevator_type
*elevator_find(const char *name
)
133 struct elevator_type
*e
;
135 list_for_each_entry(e
, &elv_list
, list
) {
136 if (!strcmp(e
->elevator_name
, name
))
143 static void elevator_put(struct elevator_type
*e
)
145 module_put(e
->elevator_owner
);
148 static struct elevator_type
*elevator_get(const char *name
)
150 struct elevator_type
*e
;
152 spin_lock(&elv_list_lock
);
154 e
= elevator_find(name
);
156 char elv
[ELV_NAME_MAX
+ strlen("-iosched")];
158 spin_unlock(&elv_list_lock
);
160 if (!strcmp(name
, "anticipatory"))
161 sprintf(elv
, "as-iosched");
163 sprintf(elv
, "%s-iosched", name
);
165 request_module("%s", elv
);
166 spin_lock(&elv_list_lock
);
167 e
= elevator_find(name
);
170 if (e
&& !try_module_get(e
->elevator_owner
))
173 spin_unlock(&elv_list_lock
);
178 static void *elevator_init_queue(struct request_queue
*q
,
179 struct elevator_queue
*eq
)
181 return eq
->ops
->elevator_init_fn(q
);
184 static void elevator_attach(struct request_queue
*q
, struct elevator_queue
*eq
,
188 eq
->elevator_data
= data
;
191 static char chosen_elevator
[16];
193 static int __init
elevator_setup(char *str
)
196 * Be backwards-compatible with previous kernels, so users
197 * won't get the wrong elevator.
199 if (!strcmp(str
, "as"))
200 strcpy(chosen_elevator
, "anticipatory");
202 strncpy(chosen_elevator
, str
, sizeof(chosen_elevator
) - 1);
206 __setup("elevator=", elevator_setup
);
208 static struct kobj_type elv_ktype
;
210 static struct elevator_queue
*elevator_alloc(struct request_queue
*q
,
211 struct elevator_type
*e
)
213 struct elevator_queue
*eq
;
216 eq
= kmalloc_node(sizeof(*eq
), GFP_KERNEL
| __GFP_ZERO
, q
->node
);
221 eq
->elevator_type
= e
;
222 kobject_init(&eq
->kobj
, &elv_ktype
);
223 mutex_init(&eq
->sysfs_lock
);
225 eq
->hash
= kmalloc_node(sizeof(struct hlist_head
) * ELV_HASH_ENTRIES
,
226 GFP_KERNEL
, q
->node
);
230 for (i
= 0; i
< ELV_HASH_ENTRIES
; i
++)
231 INIT_HLIST_HEAD(&eq
->hash
[i
]);
240 static void elevator_release(struct kobject
*kobj
)
242 struct elevator_queue
*e
;
244 e
= container_of(kobj
, struct elevator_queue
, kobj
);
245 elevator_put(e
->elevator_type
);
250 int elevator_init(struct request_queue
*q
, char *name
)
252 struct elevator_type
*e
= NULL
;
253 struct elevator_queue
*eq
;
257 INIT_LIST_HEAD(&q
->queue_head
);
258 q
->last_merge
= NULL
;
260 q
->boundary_rq
= NULL
;
263 e
= elevator_get(name
);
268 if (!e
&& *chosen_elevator
) {
269 e
= elevator_get(chosen_elevator
);
271 printk(KERN_ERR
"I/O scheduler %s not found\n",
276 e
= elevator_get(CONFIG_DEFAULT_IOSCHED
);
279 "Default I/O scheduler not found. " \
281 e
= elevator_get("noop");
285 eq
= elevator_alloc(q
, e
);
289 data
= elevator_init_queue(q
, eq
);
291 kobject_put(&eq
->kobj
);
295 elevator_attach(q
, eq
, data
);
298 EXPORT_SYMBOL(elevator_init
);
300 void elevator_exit(struct elevator_queue
*e
)
302 mutex_lock(&e
->sysfs_lock
);
303 if (e
->ops
->elevator_exit_fn
)
304 e
->ops
->elevator_exit_fn(e
);
306 mutex_unlock(&e
->sysfs_lock
);
308 kobject_put(&e
->kobj
);
310 EXPORT_SYMBOL(elevator_exit
);
312 static inline void __elv_rqhash_del(struct request
*rq
)
314 hlist_del_init(&rq
->hash
);
317 static void elv_rqhash_del(struct request_queue
*q
, struct request
*rq
)
320 __elv_rqhash_del(rq
);
323 static void elv_rqhash_add(struct request_queue
*q
, struct request
*rq
)
325 struct elevator_queue
*e
= q
->elevator
;
327 BUG_ON(ELV_ON_HASH(rq
));
328 hlist_add_head(&rq
->hash
, &e
->hash
[ELV_HASH_FN(rq_hash_key(rq
))]);
331 static void elv_rqhash_reposition(struct request_queue
*q
, struct request
*rq
)
333 __elv_rqhash_del(rq
);
334 elv_rqhash_add(q
, rq
);
337 static struct request
*elv_rqhash_find(struct request_queue
*q
, sector_t offset
)
339 struct elevator_queue
*e
= q
->elevator
;
340 struct hlist_head
*hash_list
= &e
->hash
[ELV_HASH_FN(offset
)];
341 struct hlist_node
*entry
, *next
;
344 hlist_for_each_entry_safe(rq
, entry
, next
, hash_list
, hash
) {
345 BUG_ON(!ELV_ON_HASH(rq
));
347 if (unlikely(!rq_mergeable(rq
))) {
348 __elv_rqhash_del(rq
);
352 if (rq_hash_key(rq
) == offset
)
360 * RB-tree support functions for inserting/lookup/removal of requests
361 * in a sorted RB tree.
363 struct request
*elv_rb_add(struct rb_root
*root
, struct request
*rq
)
365 struct rb_node
**p
= &root
->rb_node
;
366 struct rb_node
*parent
= NULL
;
367 struct request
*__rq
;
371 __rq
= rb_entry(parent
, struct request
, rb_node
);
373 if (blk_rq_pos(rq
) < blk_rq_pos(__rq
))
375 else if (blk_rq_pos(rq
) > blk_rq_pos(__rq
))
381 rb_link_node(&rq
->rb_node
, parent
, p
);
382 rb_insert_color(&rq
->rb_node
, root
);
385 EXPORT_SYMBOL(elv_rb_add
);
387 void elv_rb_del(struct rb_root
*root
, struct request
*rq
)
389 BUG_ON(RB_EMPTY_NODE(&rq
->rb_node
));
390 rb_erase(&rq
->rb_node
, root
);
391 RB_CLEAR_NODE(&rq
->rb_node
);
393 EXPORT_SYMBOL(elv_rb_del
);
395 struct request
*elv_rb_find(struct rb_root
*root
, sector_t sector
)
397 struct rb_node
*n
= root
->rb_node
;
401 rq
= rb_entry(n
, struct request
, rb_node
);
403 if (sector
< blk_rq_pos(rq
))
405 else if (sector
> blk_rq_pos(rq
))
413 EXPORT_SYMBOL(elv_rb_find
);
416 * Insert rq into dispatch queue of q. Queue lock must be held on
417 * entry. rq is sort instead into the dispatch queue. To be used by
418 * specific elevators.
420 void elv_dispatch_sort(struct request_queue
*q
, struct request
*rq
)
423 struct list_head
*entry
;
426 if (q
->last_merge
== rq
)
427 q
->last_merge
= NULL
;
429 elv_rqhash_del(q
, rq
);
433 boundary
= q
->end_sector
;
434 stop_flags
= REQ_SOFTBARRIER
| REQ_HARDBARRIER
| REQ_STARTED
;
435 list_for_each_prev(entry
, &q
->queue_head
) {
436 struct request
*pos
= list_entry_rq(entry
);
438 if (blk_discard_rq(rq
) != blk_discard_rq(pos
))
440 if (rq_data_dir(rq
) != rq_data_dir(pos
))
442 if (pos
->cmd_flags
& stop_flags
)
444 if (blk_rq_pos(rq
) >= boundary
) {
445 if (blk_rq_pos(pos
) < boundary
)
448 if (blk_rq_pos(pos
) >= boundary
)
451 if (blk_rq_pos(rq
) >= blk_rq_pos(pos
))
455 list_add(&rq
->queuelist
, entry
);
457 EXPORT_SYMBOL(elv_dispatch_sort
);
460 * Insert rq into dispatch queue of q. Queue lock must be held on
461 * entry. rq is added to the back of the dispatch queue. To be used by
462 * specific elevators.
464 void elv_dispatch_add_tail(struct request_queue
*q
, struct request
*rq
)
466 if (q
->last_merge
== rq
)
467 q
->last_merge
= NULL
;
469 elv_rqhash_del(q
, rq
);
473 q
->end_sector
= rq_end_sector(rq
);
475 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
477 EXPORT_SYMBOL(elv_dispatch_add_tail
);
479 int elv_merge(struct request_queue
*q
, struct request
**req
, struct bio
*bio
)
481 struct elevator_queue
*e
= q
->elevator
;
482 struct request
*__rq
;
486 * First try one-hit cache.
489 ret
= elv_try_merge(q
->last_merge
, bio
);
490 if (ret
!= ELEVATOR_NO_MERGE
) {
491 *req
= q
->last_merge
;
496 if (blk_queue_nomerges(q
))
497 return ELEVATOR_NO_MERGE
;
500 * See if our hash lookup can find a potential backmerge.
502 __rq
= elv_rqhash_find(q
, bio
->bi_sector
);
503 if (__rq
&& elv_rq_merge_ok(__rq
, bio
)) {
505 return ELEVATOR_BACK_MERGE
;
508 if (e
->ops
->elevator_merge_fn
)
509 return e
->ops
->elevator_merge_fn(q
, req
, bio
);
511 return ELEVATOR_NO_MERGE
;
514 void elv_merged_request(struct request_queue
*q
, struct request
*rq
, int type
)
516 struct elevator_queue
*e
= q
->elevator
;
518 if (e
->ops
->elevator_merged_fn
)
519 e
->ops
->elevator_merged_fn(q
, rq
, type
);
521 if (type
== ELEVATOR_BACK_MERGE
)
522 elv_rqhash_reposition(q
, rq
);
527 void elv_merge_requests(struct request_queue
*q
, struct request
*rq
,
528 struct request
*next
)
530 struct elevator_queue
*e
= q
->elevator
;
532 if (e
->ops
->elevator_merge_req_fn
)
533 e
->ops
->elevator_merge_req_fn(q
, rq
, next
);
535 elv_rqhash_reposition(q
, rq
);
536 elv_rqhash_del(q
, next
);
542 void elv_requeue_request(struct request_queue
*q
, struct request
*rq
)
545 * it already went through dequeue, we need to decrement the
546 * in_flight count again
548 if (blk_account_rq(rq
)) {
549 q
->in_flight
[rq_is_sync(rq
)]--;
550 if (blk_sorted_rq(rq
))
551 elv_deactivate_rq(q
, rq
);
554 rq
->cmd_flags
&= ~REQ_STARTED
;
556 elv_insert(q
, rq
, ELEVATOR_INSERT_REQUEUE
);
559 void elv_drain_elevator(struct request_queue
*q
)
562 while (q
->elevator
->ops
->elevator_dispatch_fn(q
, 1))
564 if (q
->nr_sorted
== 0)
566 if (printed
++ < 10) {
567 printk(KERN_ERR
"%s: forced dispatching is broken "
568 "(nr_sorted=%u), please report this\n",
569 q
->elevator
->elevator_type
->elevator_name
, q
->nr_sorted
);
574 * Call with queue lock held, interrupts disabled
576 void elv_quiesce_start(struct request_queue
*q
)
578 queue_flag_set(QUEUE_FLAG_ELVSWITCH
, q
);
581 * make sure we don't have any requests in flight
583 elv_drain_elevator(q
);
584 while (q
->rq
.elvpriv
) {
586 spin_unlock_irq(q
->queue_lock
);
588 spin_lock_irq(q
->queue_lock
);
589 elv_drain_elevator(q
);
593 void elv_quiesce_end(struct request_queue
*q
)
595 queue_flag_clear(QUEUE_FLAG_ELVSWITCH
, q
);
598 void elv_insert(struct request_queue
*q
, struct request
*rq
, int where
)
600 struct list_head
*pos
;
604 trace_block_rq_insert(q
, rq
);
609 case ELEVATOR_INSERT_FRONT
:
610 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
612 list_add(&rq
->queuelist
, &q
->queue_head
);
615 case ELEVATOR_INSERT_BACK
:
616 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
617 elv_drain_elevator(q
);
618 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
620 * We kick the queue here for the following reasons.
621 * - The elevator might have returned NULL previously
622 * to delay requests and returned them now. As the
623 * queue wasn't empty before this request, ll_rw_blk
624 * won't run the queue on return, resulting in hang.
625 * - Usually, back inserted requests won't be merged
626 * with anything. There's no point in delaying queue
632 case ELEVATOR_INSERT_SORT
:
633 BUG_ON(!blk_fs_request(rq
) && !blk_discard_rq(rq
));
634 rq
->cmd_flags
|= REQ_SORTED
;
636 if (rq_mergeable(rq
)) {
637 elv_rqhash_add(q
, rq
);
643 * Some ioscheds (cfq) run q->request_fn directly, so
644 * rq cannot be accessed after calling
645 * elevator_add_req_fn.
647 q
->elevator
->ops
->elevator_add_req_fn(q
, rq
);
650 case ELEVATOR_INSERT_REQUEUE
:
652 * If ordered flush isn't in progress, we do front
653 * insertion; otherwise, requests should be requeued
656 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
659 * Most requeues happen because of a busy condition,
660 * don't force unplug of the queue for that case.
664 if (q
->ordseq
== 0) {
665 list_add(&rq
->queuelist
, &q
->queue_head
);
669 ordseq
= blk_ordered_req_seq(rq
);
671 list_for_each(pos
, &q
->queue_head
) {
672 struct request
*pos_rq
= list_entry_rq(pos
);
673 if (ordseq
<= blk_ordered_req_seq(pos_rq
))
677 list_add_tail(&rq
->queuelist
, pos
);
681 printk(KERN_ERR
"%s: bad insertion point %d\n",
686 if (unplug_it
&& blk_queue_plugged(q
)) {
687 int nrq
= q
->rq
.count
[BLK_RW_SYNC
] + q
->rq
.count
[BLK_RW_ASYNC
]
688 - queue_in_flight(q
);
690 if (nrq
>= q
->unplug_thresh
)
691 __generic_unplug_device(q
);
695 void __elv_add_request(struct request_queue
*q
, struct request
*rq
, int where
,
699 rq
->cmd_flags
|= REQ_ORDERED_COLOR
;
701 if (rq
->cmd_flags
& (REQ_SOFTBARRIER
| REQ_HARDBARRIER
)) {
703 * toggle ordered color
705 if (blk_barrier_rq(rq
))
709 * barriers implicitly indicate back insertion
711 if (where
== ELEVATOR_INSERT_SORT
)
712 where
= ELEVATOR_INSERT_BACK
;
715 * this request is scheduling boundary, update
718 if (blk_fs_request(rq
) || blk_discard_rq(rq
)) {
719 q
->end_sector
= rq_end_sector(rq
);
722 } else if (!(rq
->cmd_flags
& REQ_ELVPRIV
) &&
723 where
== ELEVATOR_INSERT_SORT
)
724 where
= ELEVATOR_INSERT_BACK
;
729 elv_insert(q
, rq
, where
);
731 EXPORT_SYMBOL(__elv_add_request
);
733 void elv_add_request(struct request_queue
*q
, struct request
*rq
, int where
,
738 spin_lock_irqsave(q
->queue_lock
, flags
);
739 __elv_add_request(q
, rq
, where
, plug
);
740 spin_unlock_irqrestore(q
->queue_lock
, flags
);
742 EXPORT_SYMBOL(elv_add_request
);
744 int elv_queue_empty(struct request_queue
*q
)
746 struct elevator_queue
*e
= q
->elevator
;
748 if (!list_empty(&q
->queue_head
))
751 if (e
->ops
->elevator_queue_empty_fn
)
752 return e
->ops
->elevator_queue_empty_fn(q
);
756 EXPORT_SYMBOL(elv_queue_empty
);
758 struct request
*elv_latter_request(struct request_queue
*q
, struct request
*rq
)
760 struct elevator_queue
*e
= q
->elevator
;
762 if (e
->ops
->elevator_latter_req_fn
)
763 return e
->ops
->elevator_latter_req_fn(q
, rq
);
767 struct request
*elv_former_request(struct request_queue
*q
, struct request
*rq
)
769 struct elevator_queue
*e
= q
->elevator
;
771 if (e
->ops
->elevator_former_req_fn
)
772 return e
->ops
->elevator_former_req_fn(q
, rq
);
776 int elv_set_request(struct request_queue
*q
, struct request
*rq
, gfp_t gfp_mask
)
778 struct elevator_queue
*e
= q
->elevator
;
780 if (e
->ops
->elevator_set_req_fn
)
781 return e
->ops
->elevator_set_req_fn(q
, rq
, gfp_mask
);
783 rq
->elevator_private
= NULL
;
787 void elv_put_request(struct request_queue
*q
, struct request
*rq
)
789 struct elevator_queue
*e
= q
->elevator
;
791 if (e
->ops
->elevator_put_req_fn
)
792 e
->ops
->elevator_put_req_fn(rq
);
795 int elv_may_queue(struct request_queue
*q
, int rw
)
797 struct elevator_queue
*e
= q
->elevator
;
799 if (e
->ops
->elevator_may_queue_fn
)
800 return e
->ops
->elevator_may_queue_fn(q
, rw
);
802 return ELV_MQUEUE_MAY
;
805 void elv_abort_queue(struct request_queue
*q
)
809 while (!list_empty(&q
->queue_head
)) {
810 rq
= list_entry_rq(q
->queue_head
.next
);
811 rq
->cmd_flags
|= REQ_QUIET
;
812 trace_block_rq_abort(q
, rq
);
813 __blk_end_request_all(rq
, -EIO
);
816 EXPORT_SYMBOL(elv_abort_queue
);
818 void elv_completed_request(struct request_queue
*q
, struct request
*rq
)
820 struct elevator_queue
*e
= q
->elevator
;
823 * request is released from the driver, io must be done
825 if (blk_account_rq(rq
)) {
826 q
->in_flight
[rq_is_sync(rq
)]--;
827 if (blk_sorted_rq(rq
) && e
->ops
->elevator_completed_req_fn
)
828 e
->ops
->elevator_completed_req_fn(q
, rq
);
832 * Check if the queue is waiting for fs requests to be
833 * drained for flush sequence.
835 if (unlikely(q
->ordseq
)) {
836 struct request
*next
= NULL
;
838 if (!list_empty(&q
->queue_head
))
839 next
= list_entry_rq(q
->queue_head
.next
);
841 if (!queue_in_flight(q
) &&
842 blk_ordered_cur_seq(q
) == QUEUE_ORDSEQ_DRAIN
&&
843 (!next
|| blk_ordered_req_seq(next
) > QUEUE_ORDSEQ_DRAIN
)) {
844 blk_ordered_complete_seq(q
, QUEUE_ORDSEQ_DRAIN
, 0);
850 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
853 elv_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
855 struct elv_fs_entry
*entry
= to_elv(attr
);
856 struct elevator_queue
*e
;
862 e
= container_of(kobj
, struct elevator_queue
, kobj
);
863 mutex_lock(&e
->sysfs_lock
);
864 error
= e
->ops
? entry
->show(e
, page
) : -ENOENT
;
865 mutex_unlock(&e
->sysfs_lock
);
870 elv_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
871 const char *page
, size_t length
)
873 struct elv_fs_entry
*entry
= to_elv(attr
);
874 struct elevator_queue
*e
;
880 e
= container_of(kobj
, struct elevator_queue
, kobj
);
881 mutex_lock(&e
->sysfs_lock
);
882 error
= e
->ops
? entry
->store(e
, page
, length
) : -ENOENT
;
883 mutex_unlock(&e
->sysfs_lock
);
887 static struct sysfs_ops elv_sysfs_ops
= {
888 .show
= elv_attr_show
,
889 .store
= elv_attr_store
,
892 static struct kobj_type elv_ktype
= {
893 .sysfs_ops
= &elv_sysfs_ops
,
894 .release
= elevator_release
,
897 int elv_register_queue(struct request_queue
*q
)
899 struct elevator_queue
*e
= q
->elevator
;
902 error
= kobject_add(&e
->kobj
, &q
->kobj
, "%s", "iosched");
904 struct elv_fs_entry
*attr
= e
->elevator_type
->elevator_attrs
;
906 while (attr
->attr
.name
) {
907 if (sysfs_create_file(&e
->kobj
, &attr
->attr
))
912 kobject_uevent(&e
->kobj
, KOBJ_ADD
);
917 static void __elv_unregister_queue(struct elevator_queue
*e
)
919 kobject_uevent(&e
->kobj
, KOBJ_REMOVE
);
920 kobject_del(&e
->kobj
);
923 void elv_unregister_queue(struct request_queue
*q
)
926 __elv_unregister_queue(q
->elevator
);
929 void elv_register(struct elevator_type
*e
)
933 spin_lock(&elv_list_lock
);
934 BUG_ON(elevator_find(e
->elevator_name
));
935 list_add_tail(&e
->list
, &elv_list
);
936 spin_unlock(&elv_list_lock
);
938 if (!strcmp(e
->elevator_name
, chosen_elevator
) ||
939 (!*chosen_elevator
&&
940 !strcmp(e
->elevator_name
, CONFIG_DEFAULT_IOSCHED
)))
943 printk(KERN_INFO
"io scheduler %s registered%s\n", e
->elevator_name
,
946 EXPORT_SYMBOL_GPL(elv_register
);
948 void elv_unregister(struct elevator_type
*e
)
950 struct task_struct
*g
, *p
;
953 * Iterate every thread in the process to remove the io contexts.
956 read_lock(&tasklist_lock
);
957 do_each_thread(g
, p
) {
960 e
->ops
.trim(p
->io_context
);
962 } while_each_thread(g
, p
);
963 read_unlock(&tasklist_lock
);
966 spin_lock(&elv_list_lock
);
967 list_del_init(&e
->list
);
968 spin_unlock(&elv_list_lock
);
970 EXPORT_SYMBOL_GPL(elv_unregister
);
973 * switch to new_e io scheduler. be careful not to introduce deadlocks -
974 * we don't free the old io scheduler, before we have allocated what we
975 * need for the new one. this way we have a chance of going back to the old
976 * one, if the new one fails init for some reason.
978 static int elevator_switch(struct request_queue
*q
, struct elevator_type
*new_e
)
980 struct elevator_queue
*old_elevator
, *e
;
984 * Allocate new elevator
986 e
= elevator_alloc(q
, new_e
);
990 data
= elevator_init_queue(q
, e
);
992 kobject_put(&e
->kobj
);
997 * Turn on BYPASS and drain all requests w/ elevator private data
999 spin_lock_irq(q
->queue_lock
);
1000 elv_quiesce_start(q
);
1003 * Remember old elevator.
1005 old_elevator
= q
->elevator
;
1008 * attach and start new elevator
1010 elevator_attach(q
, e
, data
);
1012 spin_unlock_irq(q
->queue_lock
);
1014 __elv_unregister_queue(old_elevator
);
1016 if (elv_register_queue(q
))
1020 * finally exit old elevator and turn off BYPASS.
1022 elevator_exit(old_elevator
);
1023 spin_lock_irq(q
->queue_lock
);
1025 spin_unlock_irq(q
->queue_lock
);
1027 blk_add_trace_msg(q
, "elv switch: %s", e
->elevator_type
->elevator_name
);
1033 * switch failed, exit the new io scheduler and reattach the old
1034 * one again (along with re-adding the sysfs dir)
1037 q
->elevator
= old_elevator
;
1038 elv_register_queue(q
);
1040 spin_lock_irq(q
->queue_lock
);
1041 queue_flag_clear(QUEUE_FLAG_ELVSWITCH
, q
);
1042 spin_unlock_irq(q
->queue_lock
);
1047 ssize_t
elv_iosched_store(struct request_queue
*q
, const char *name
,
1050 char elevator_name
[ELV_NAME_MAX
];
1051 struct elevator_type
*e
;
1053 strlcpy(elevator_name
, name
, sizeof(elevator_name
));
1054 strstrip(elevator_name
);
1056 e
= elevator_get(elevator_name
);
1058 printk(KERN_ERR
"elevator: type %s not found\n", elevator_name
);
1062 if (!strcmp(elevator_name
, q
->elevator
->elevator_type
->elevator_name
)) {
1067 if (!elevator_switch(q
, e
))
1068 printk(KERN_ERR
"elevator: switch to %s failed\n",
1073 ssize_t
elv_iosched_show(struct request_queue
*q
, char *name
)
1075 struct elevator_queue
*e
= q
->elevator
;
1076 struct elevator_type
*elv
= e
->elevator_type
;
1077 struct elevator_type
*__e
;
1080 spin_lock(&elv_list_lock
);
1081 list_for_each_entry(__e
, &elv_list
, list
) {
1082 if (!strcmp(elv
->elevator_name
, __e
->elevator_name
))
1083 len
+= sprintf(name
+len
, "[%s] ", elv
->elevator_name
);
1085 len
+= sprintf(name
+len
, "%s ", __e
->elevator_name
);
1087 spin_unlock(&elv_list_lock
);
1089 len
+= sprintf(len
+name
, "\n");
1093 struct request
*elv_rb_former_request(struct request_queue
*q
,
1096 struct rb_node
*rbprev
= rb_prev(&rq
->rb_node
);
1099 return rb_entry_rq(rbprev
);
1103 EXPORT_SYMBOL(elv_rb_former_request
);
1105 struct request
*elv_rb_latter_request(struct request_queue
*q
,
1108 struct rb_node
*rbnext
= rb_next(&rq
->rb_node
);
1111 return rb_entry_rq(rbnext
);
1115 EXPORT_SYMBOL(elv_rb_latter_request
);