2 * Block device elevator/IO-scheduler.
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
6 * 30042000 Jens Axboe <axboe@suse.de> :
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 <linux/hash.h>
38 #include <asm/uaccess.h>
40 static DEFINE_SPINLOCK(elv_list_lock
);
41 static LIST_HEAD(elv_list
);
46 static const int elv_hash_shift
= 6;
47 #define ELV_HASH_BLOCK(sec) ((sec) >> 3)
48 #define ELV_HASH_FN(sec) (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
49 #define ELV_HASH_ENTRIES (1 << elv_hash_shift)
50 #define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors)
51 #define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash))
54 * can we safely merge with this request?
56 inline int elv_rq_merge_ok(struct request
*rq
, struct bio
*bio
)
58 if (!rq_mergeable(rq
))
62 * different data direction or already started, don't merge
64 if (bio_data_dir(bio
) != rq_data_dir(rq
))
68 * same device and no special stuff set, merge is ok
70 if (rq
->rq_disk
== bio
->bi_bdev
->bd_disk
&&
71 !rq
->waiting
&& !rq
->special
)
76 EXPORT_SYMBOL(elv_rq_merge_ok
);
78 static inline int elv_try_merge(struct request
*__rq
, struct bio
*bio
)
80 int ret
= ELEVATOR_NO_MERGE
;
83 * we can merge and sequence is ok, check if it's possible
85 if (elv_rq_merge_ok(__rq
, bio
)) {
86 if (__rq
->sector
+ __rq
->nr_sectors
== bio
->bi_sector
)
87 ret
= ELEVATOR_BACK_MERGE
;
88 else if (__rq
->sector
- bio_sectors(bio
) == bio
->bi_sector
)
89 ret
= ELEVATOR_FRONT_MERGE
;
95 static struct elevator_type
*elevator_find(const char *name
)
97 struct elevator_type
*e
= NULL
;
98 struct list_head
*entry
;
100 list_for_each(entry
, &elv_list
) {
101 struct elevator_type
*__e
;
103 __e
= list_entry(entry
, struct elevator_type
, list
);
105 if (!strcmp(__e
->elevator_name
, name
)) {
114 static void elevator_put(struct elevator_type
*e
)
116 module_put(e
->elevator_owner
);
119 static struct elevator_type
*elevator_get(const char *name
)
121 struct elevator_type
*e
;
123 spin_lock_irq(&elv_list_lock
);
125 e
= elevator_find(name
);
126 if (e
&& !try_module_get(e
->elevator_owner
))
129 spin_unlock_irq(&elv_list_lock
);
134 static void *elevator_init_queue(request_queue_t
*q
, struct elevator_queue
*eq
)
136 return eq
->ops
->elevator_init_fn(q
, eq
);
139 static void elevator_attach(request_queue_t
*q
, struct elevator_queue
*eq
,
143 eq
->elevator_data
= data
;
146 static char chosen_elevator
[16];
148 static int __init
elevator_setup(char *str
)
151 * Be backwards-compatible with previous kernels, so users
152 * won't get the wrong elevator.
154 if (!strcmp(str
, "as"))
155 strcpy(chosen_elevator
, "anticipatory");
157 strncpy(chosen_elevator
, str
, sizeof(chosen_elevator
) - 1);
161 __setup("elevator=", elevator_setup
);
163 static struct kobj_type elv_ktype
;
165 static elevator_t
*elevator_alloc(struct elevator_type
*e
)
170 eq
= kmalloc(sizeof(elevator_t
), GFP_KERNEL
);
174 memset(eq
, 0, sizeof(*eq
));
176 eq
->elevator_type
= e
;
177 kobject_init(&eq
->kobj
);
178 snprintf(eq
->kobj
.name
, KOBJ_NAME_LEN
, "%s", "iosched");
179 eq
->kobj
.ktype
= &elv_ktype
;
180 mutex_init(&eq
->sysfs_lock
);
182 eq
->hash
= kmalloc(sizeof(struct hlist_head
) * ELV_HASH_ENTRIES
, GFP_KERNEL
);
186 for (i
= 0; i
< ELV_HASH_ENTRIES
; i
++)
187 INIT_HLIST_HEAD(&eq
->hash
[i
]);
196 static void elevator_release(struct kobject
*kobj
)
198 elevator_t
*e
= container_of(kobj
, elevator_t
, kobj
);
200 elevator_put(e
->elevator_type
);
205 int elevator_init(request_queue_t
*q
, char *name
)
207 struct elevator_type
*e
= NULL
;
208 struct elevator_queue
*eq
;
212 INIT_LIST_HEAD(&q
->queue_head
);
213 q
->last_merge
= NULL
;
215 q
->boundary_rq
= NULL
;
217 if (name
&& !(e
= elevator_get(name
)))
220 if (!e
&& *chosen_elevator
&& !(e
= elevator_get(chosen_elevator
)))
221 printk("I/O scheduler %s not found\n", chosen_elevator
);
223 if (!e
&& !(e
= elevator_get(CONFIG_DEFAULT_IOSCHED
))) {
224 printk("Default I/O scheduler not found, using no-op\n");
225 e
= elevator_get("noop");
228 eq
= elevator_alloc(e
);
232 data
= elevator_init_queue(q
, eq
);
234 kobject_put(&eq
->kobj
);
238 elevator_attach(q
, eq
, data
);
242 EXPORT_SYMBOL(elevator_init
);
244 void elevator_exit(elevator_t
*e
)
246 mutex_lock(&e
->sysfs_lock
);
247 if (e
->ops
->elevator_exit_fn
)
248 e
->ops
->elevator_exit_fn(e
);
250 mutex_unlock(&e
->sysfs_lock
);
252 kobject_put(&e
->kobj
);
255 EXPORT_SYMBOL(elevator_exit
);
257 static inline void __elv_rqhash_del(struct request
*rq
)
259 hlist_del_init(&rq
->hash
);
262 static void elv_rqhash_del(request_queue_t
*q
, struct request
*rq
)
265 __elv_rqhash_del(rq
);
268 static void elv_rqhash_add(request_queue_t
*q
, struct request
*rq
)
270 elevator_t
*e
= q
->elevator
;
272 BUG_ON(ELV_ON_HASH(rq
));
273 hlist_add_head(&rq
->hash
, &e
->hash
[ELV_HASH_FN(rq_hash_key(rq
))]);
276 static void elv_rqhash_reposition(request_queue_t
*q
, struct request
*rq
)
278 __elv_rqhash_del(rq
);
279 elv_rqhash_add(q
, rq
);
282 static struct request
*elv_rqhash_find(request_queue_t
*q
, sector_t offset
)
284 elevator_t
*e
= q
->elevator
;
285 struct hlist_head
*hash_list
= &e
->hash
[ELV_HASH_FN(offset
)];
286 struct hlist_node
*entry
, *next
;
289 hlist_for_each_entry_safe(rq
, entry
, next
, hash_list
, hash
) {
290 BUG_ON(!ELV_ON_HASH(rq
));
292 if (unlikely(!rq_mergeable(rq
))) {
293 __elv_rqhash_del(rq
);
297 if (rq_hash_key(rq
) == offset
)
305 * RB-tree support functions for inserting/lookup/removal of requests
306 * in a sorted RB tree.
308 struct request
*elv_rb_add(struct rb_root
*root
, struct request
*rq
)
310 struct rb_node
**p
= &root
->rb_node
;
311 struct rb_node
*parent
= NULL
;
312 struct request
*__rq
;
316 __rq
= rb_entry(parent
, struct request
, rb_node
);
318 if (rq
->sector
< __rq
->sector
)
320 else if (rq
->sector
> __rq
->sector
)
326 rb_link_node(&rq
->rb_node
, parent
, p
);
327 rb_insert_color(&rq
->rb_node
, root
);
331 EXPORT_SYMBOL(elv_rb_add
);
333 void elv_rb_del(struct rb_root
*root
, struct request
*rq
)
335 BUG_ON(RB_EMPTY_NODE(&rq
->rb_node
));
336 rb_erase(&rq
->rb_node
, root
);
337 RB_CLEAR_NODE(&rq
->rb_node
);
340 EXPORT_SYMBOL(elv_rb_del
);
342 struct request
*elv_rb_find(struct rb_root
*root
, sector_t sector
)
344 struct rb_node
*n
= root
->rb_node
;
348 rq
= rb_entry(n
, struct request
, rb_node
);
350 if (sector
< rq
->sector
)
352 else if (sector
> rq
->sector
)
361 EXPORT_SYMBOL(elv_rb_find
);
364 * Insert rq into dispatch queue of q. Queue lock must be held on
365 * entry. rq is sort insted into the dispatch queue. To be used by
366 * specific elevators.
368 void elv_dispatch_sort(request_queue_t
*q
, struct request
*rq
)
371 struct list_head
*entry
;
373 if (q
->last_merge
== rq
)
374 q
->last_merge
= NULL
;
376 elv_rqhash_del(q
, rq
);
380 boundary
= q
->end_sector
;
382 list_for_each_prev(entry
, &q
->queue_head
) {
383 struct request
*pos
= list_entry_rq(entry
);
385 if (pos
->cmd_flags
& (REQ_SOFTBARRIER
|REQ_HARDBARRIER
|REQ_STARTED
))
387 if (rq
->sector
>= boundary
) {
388 if (pos
->sector
< boundary
)
391 if (pos
->sector
>= boundary
)
394 if (rq
->sector
>= pos
->sector
)
398 list_add(&rq
->queuelist
, entry
);
401 EXPORT_SYMBOL(elv_dispatch_sort
);
404 * Insert rq into dispatch queue of q. Queue lock must be held on
405 * entry. rq is added to the back of the dispatch queue. To be used by
406 * specific elevators.
408 void elv_dispatch_add_tail(struct request_queue
*q
, struct request
*rq
)
410 if (q
->last_merge
== rq
)
411 q
->last_merge
= NULL
;
413 elv_rqhash_del(q
, rq
);
417 q
->end_sector
= rq_end_sector(rq
);
419 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
422 EXPORT_SYMBOL(elv_dispatch_add_tail
);
424 int elv_merge(request_queue_t
*q
, struct request
**req
, struct bio
*bio
)
426 elevator_t
*e
= q
->elevator
;
427 struct request
*__rq
;
431 * First try one-hit cache.
434 ret
= elv_try_merge(q
->last_merge
, bio
);
435 if (ret
!= ELEVATOR_NO_MERGE
) {
436 *req
= q
->last_merge
;
442 * See if our hash lookup can find a potential backmerge.
444 __rq
= elv_rqhash_find(q
, bio
->bi_sector
);
445 if (__rq
&& elv_rq_merge_ok(__rq
, bio
)) {
447 return ELEVATOR_BACK_MERGE
;
450 if (e
->ops
->elevator_merge_fn
)
451 return e
->ops
->elevator_merge_fn(q
, req
, bio
);
453 return ELEVATOR_NO_MERGE
;
456 void elv_merged_request(request_queue_t
*q
, struct request
*rq
, int type
)
458 elevator_t
*e
= q
->elevator
;
460 if (e
->ops
->elevator_merged_fn
)
461 e
->ops
->elevator_merged_fn(q
, rq
, type
);
463 if (type
== ELEVATOR_BACK_MERGE
)
464 elv_rqhash_reposition(q
, rq
);
469 void elv_merge_requests(request_queue_t
*q
, struct request
*rq
,
470 struct request
*next
)
472 elevator_t
*e
= q
->elevator
;
474 if (e
->ops
->elevator_merge_req_fn
)
475 e
->ops
->elevator_merge_req_fn(q
, rq
, next
);
477 elv_rqhash_reposition(q
, rq
);
478 elv_rqhash_del(q
, next
);
484 void elv_requeue_request(request_queue_t
*q
, struct request
*rq
)
486 elevator_t
*e
= q
->elevator
;
489 * it already went through dequeue, we need to decrement the
490 * in_flight count again
492 if (blk_account_rq(rq
)) {
494 if (blk_sorted_rq(rq
) && e
->ops
->elevator_deactivate_req_fn
)
495 e
->ops
->elevator_deactivate_req_fn(q
, rq
);
498 rq
->cmd_flags
&= ~REQ_STARTED
;
500 elv_insert(q
, rq
, ELEVATOR_INSERT_REQUEUE
);
503 static void elv_drain_elevator(request_queue_t
*q
)
506 while (q
->elevator
->ops
->elevator_dispatch_fn(q
, 1))
508 if (q
->nr_sorted
== 0)
510 if (printed
++ < 10) {
511 printk(KERN_ERR
"%s: forced dispatching is broken "
512 "(nr_sorted=%u), please report this\n",
513 q
->elevator
->elevator_type
->elevator_name
, q
->nr_sorted
);
517 void elv_insert(request_queue_t
*q
, struct request
*rq
, int where
)
519 struct list_head
*pos
;
523 blk_add_trace_rq(q
, rq
, BLK_TA_INSERT
);
528 case ELEVATOR_INSERT_FRONT
:
529 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
531 list_add(&rq
->queuelist
, &q
->queue_head
);
534 case ELEVATOR_INSERT_BACK
:
535 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
536 elv_drain_elevator(q
);
537 list_add_tail(&rq
->queuelist
, &q
->queue_head
);
539 * We kick the queue here for the following reasons.
540 * - The elevator might have returned NULL previously
541 * to delay requests and returned them now. As the
542 * queue wasn't empty before this request, ll_rw_blk
543 * won't run the queue on return, resulting in hang.
544 * - Usually, back inserted requests won't be merged
545 * with anything. There's no point in delaying queue
552 case ELEVATOR_INSERT_SORT
:
553 BUG_ON(!blk_fs_request(rq
));
554 rq
->cmd_flags
|= REQ_SORTED
;
556 if (rq_mergeable(rq
)) {
557 elv_rqhash_add(q
, rq
);
563 * Some ioscheds (cfq) run q->request_fn directly, so
564 * rq cannot be accessed after calling
565 * elevator_add_req_fn.
567 q
->elevator
->ops
->elevator_add_req_fn(q
, rq
);
570 case ELEVATOR_INSERT_REQUEUE
:
572 * If ordered flush isn't in progress, we do front
573 * insertion; otherwise, requests should be requeued
576 rq
->cmd_flags
|= REQ_SOFTBARRIER
;
578 if (q
->ordseq
== 0) {
579 list_add(&rq
->queuelist
, &q
->queue_head
);
583 ordseq
= blk_ordered_req_seq(rq
);
585 list_for_each(pos
, &q
->queue_head
) {
586 struct request
*pos_rq
= list_entry_rq(pos
);
587 if (ordseq
<= blk_ordered_req_seq(pos_rq
))
591 list_add_tail(&rq
->queuelist
, pos
);
593 * most requeues happen because of a busy condition, don't
594 * force unplug of the queue for that case.
600 printk(KERN_ERR
"%s: bad insertion point %d\n",
601 __FUNCTION__
, where
);
605 if (unplug_it
&& blk_queue_plugged(q
)) {
606 int nrq
= q
->rq
.count
[READ
] + q
->rq
.count
[WRITE
]
609 if (nrq
>= q
->unplug_thresh
)
610 __generic_unplug_device(q
);
614 void __elv_add_request(request_queue_t
*q
, struct request
*rq
, int where
,
618 rq
->cmd_flags
|= REQ_ORDERED_COLOR
;
620 if (rq
->cmd_flags
& (REQ_SOFTBARRIER
| REQ_HARDBARRIER
)) {
622 * toggle ordered color
624 if (blk_barrier_rq(rq
))
628 * barriers implicitly indicate back insertion
630 if (where
== ELEVATOR_INSERT_SORT
)
631 where
= ELEVATOR_INSERT_BACK
;
634 * this request is scheduling boundary, update
637 if (blk_fs_request(rq
)) {
638 q
->end_sector
= rq_end_sector(rq
);
641 } else if (!(rq
->cmd_flags
& REQ_ELVPRIV
) && where
== ELEVATOR_INSERT_SORT
)
642 where
= ELEVATOR_INSERT_BACK
;
647 elv_insert(q
, rq
, where
);
650 EXPORT_SYMBOL(__elv_add_request
);
652 void elv_add_request(request_queue_t
*q
, struct request
*rq
, int where
,
657 spin_lock_irqsave(q
->queue_lock
, flags
);
658 __elv_add_request(q
, rq
, where
, plug
);
659 spin_unlock_irqrestore(q
->queue_lock
, flags
);
662 EXPORT_SYMBOL(elv_add_request
);
664 static inline struct request
*__elv_next_request(request_queue_t
*q
)
669 while (!list_empty(&q
->queue_head
)) {
670 rq
= list_entry_rq(q
->queue_head
.next
);
671 if (blk_do_ordered(q
, &rq
))
675 if (!q
->elevator
->ops
->elevator_dispatch_fn(q
, 0))
680 struct request
*elv_next_request(request_queue_t
*q
)
685 while ((rq
= __elv_next_request(q
)) != NULL
) {
686 if (!(rq
->cmd_flags
& REQ_STARTED
)) {
687 elevator_t
*e
= q
->elevator
;
690 * This is the first time the device driver
691 * sees this request (possibly after
692 * requeueing). Notify IO scheduler.
694 if (blk_sorted_rq(rq
) &&
695 e
->ops
->elevator_activate_req_fn
)
696 e
->ops
->elevator_activate_req_fn(q
, rq
);
699 * just mark as started even if we don't start
700 * it, a request that has been delayed should
701 * not be passed by new incoming requests
703 rq
->cmd_flags
|= REQ_STARTED
;
704 blk_add_trace_rq(q
, rq
, BLK_TA_ISSUE
);
707 if (!q
->boundary_rq
|| q
->boundary_rq
== rq
) {
708 q
->end_sector
= rq_end_sector(rq
);
709 q
->boundary_rq
= NULL
;
712 if ((rq
->cmd_flags
& REQ_DONTPREP
) || !q
->prep_rq_fn
)
715 ret
= q
->prep_rq_fn(q
, rq
);
716 if (ret
== BLKPREP_OK
) {
718 } else if (ret
== BLKPREP_DEFER
) {
720 * the request may have been (partially) prepped.
721 * we need to keep this request in the front to
722 * avoid resource deadlock. REQ_STARTED will
723 * prevent other fs requests from passing this one.
727 } else if (ret
== BLKPREP_KILL
) {
728 int nr_bytes
= rq
->hard_nr_sectors
<< 9;
731 nr_bytes
= rq
->data_len
;
733 blkdev_dequeue_request(rq
);
734 rq
->cmd_flags
|= REQ_QUIET
;
735 end_that_request_chunk(rq
, 0, nr_bytes
);
736 end_that_request_last(rq
, 0);
738 printk(KERN_ERR
"%s: bad return=%d\n", __FUNCTION__
,
747 EXPORT_SYMBOL(elv_next_request
);
749 void elv_dequeue_request(request_queue_t
*q
, struct request
*rq
)
751 BUG_ON(list_empty(&rq
->queuelist
));
752 BUG_ON(ELV_ON_HASH(rq
));
754 list_del_init(&rq
->queuelist
);
757 * the time frame between a request being removed from the lists
758 * and to it is freed is accounted as io that is in progress at
761 if (blk_account_rq(rq
))
765 EXPORT_SYMBOL(elv_dequeue_request
);
767 int elv_queue_empty(request_queue_t
*q
)
769 elevator_t
*e
= q
->elevator
;
771 if (!list_empty(&q
->queue_head
))
774 if (e
->ops
->elevator_queue_empty_fn
)
775 return e
->ops
->elevator_queue_empty_fn(q
);
780 EXPORT_SYMBOL(elv_queue_empty
);
782 struct request
*elv_latter_request(request_queue_t
*q
, struct request
*rq
)
784 elevator_t
*e
= q
->elevator
;
786 if (e
->ops
->elevator_latter_req_fn
)
787 return e
->ops
->elevator_latter_req_fn(q
, rq
);
791 struct request
*elv_former_request(request_queue_t
*q
, struct request
*rq
)
793 elevator_t
*e
= q
->elevator
;
795 if (e
->ops
->elevator_former_req_fn
)
796 return e
->ops
->elevator_former_req_fn(q
, rq
);
800 int elv_set_request(request_queue_t
*q
, struct request
*rq
, struct bio
*bio
,
803 elevator_t
*e
= q
->elevator
;
805 if (e
->ops
->elevator_set_req_fn
)
806 return e
->ops
->elevator_set_req_fn(q
, rq
, bio
, gfp_mask
);
808 rq
->elevator_private
= NULL
;
812 void elv_put_request(request_queue_t
*q
, struct request
*rq
)
814 elevator_t
*e
= q
->elevator
;
816 if (e
->ops
->elevator_put_req_fn
)
817 e
->ops
->elevator_put_req_fn(q
, rq
);
820 int elv_may_queue(request_queue_t
*q
, int rw
, struct bio
*bio
)
822 elevator_t
*e
= q
->elevator
;
824 if (e
->ops
->elevator_may_queue_fn
)
825 return e
->ops
->elevator_may_queue_fn(q
, rw
, bio
);
827 return ELV_MQUEUE_MAY
;
830 void elv_completed_request(request_queue_t
*q
, struct request
*rq
)
832 elevator_t
*e
= q
->elevator
;
835 * request is released from the driver, io must be done
837 if (blk_account_rq(rq
)) {
839 if (blk_sorted_rq(rq
) && e
->ops
->elevator_completed_req_fn
)
840 e
->ops
->elevator_completed_req_fn(q
, rq
);
844 * Check if the queue is waiting for fs requests to be
845 * drained for flush sequence.
847 if (unlikely(q
->ordseq
)) {
848 struct request
*first_rq
= list_entry_rq(q
->queue_head
.next
);
849 if (q
->in_flight
== 0 &&
850 blk_ordered_cur_seq(q
) == QUEUE_ORDSEQ_DRAIN
&&
851 blk_ordered_req_seq(first_rq
) > QUEUE_ORDSEQ_DRAIN
) {
852 blk_ordered_complete_seq(q
, QUEUE_ORDSEQ_DRAIN
, 0);
858 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
861 elv_attr_show(struct kobject
*kobj
, struct attribute
*attr
, char *page
)
863 elevator_t
*e
= container_of(kobj
, elevator_t
, kobj
);
864 struct elv_fs_entry
*entry
= to_elv(attr
);
870 mutex_lock(&e
->sysfs_lock
);
871 error
= e
->ops
? entry
->show(e
, page
) : -ENOENT
;
872 mutex_unlock(&e
->sysfs_lock
);
877 elv_attr_store(struct kobject
*kobj
, struct attribute
*attr
,
878 const char *page
, size_t length
)
880 elevator_t
*e
= container_of(kobj
, elevator_t
, kobj
);
881 struct elv_fs_entry
*entry
= to_elv(attr
);
887 mutex_lock(&e
->sysfs_lock
);
888 error
= e
->ops
? entry
->store(e
, page
, length
) : -ENOENT
;
889 mutex_unlock(&e
->sysfs_lock
);
893 static struct sysfs_ops elv_sysfs_ops
= {
894 .show
= elv_attr_show
,
895 .store
= elv_attr_store
,
898 static struct kobj_type elv_ktype
= {
899 .sysfs_ops
= &elv_sysfs_ops
,
900 .release
= elevator_release
,
903 int elv_register_queue(struct request_queue
*q
)
905 elevator_t
*e
= q
->elevator
;
908 e
->kobj
.parent
= &q
->kobj
;
910 error
= kobject_add(&e
->kobj
);
912 struct elv_fs_entry
*attr
= e
->elevator_type
->elevator_attrs
;
914 while (attr
->attr
.name
) {
915 if (sysfs_create_file(&e
->kobj
, &attr
->attr
))
920 kobject_uevent(&e
->kobj
, KOBJ_ADD
);
925 static void __elv_unregister_queue(elevator_t
*e
)
927 kobject_uevent(&e
->kobj
, KOBJ_REMOVE
);
928 kobject_del(&e
->kobj
);
931 void elv_unregister_queue(struct request_queue
*q
)
934 __elv_unregister_queue(q
->elevator
);
937 int elv_register(struct elevator_type
*e
)
939 spin_lock_irq(&elv_list_lock
);
940 BUG_ON(elevator_find(e
->elevator_name
));
941 list_add_tail(&e
->list
, &elv_list
);
942 spin_unlock_irq(&elv_list_lock
);
944 printk(KERN_INFO
"io scheduler %s registered", e
->elevator_name
);
945 if (!strcmp(e
->elevator_name
, chosen_elevator
) ||
946 (!*chosen_elevator
&&
947 !strcmp(e
->elevator_name
, CONFIG_DEFAULT_IOSCHED
)))
948 printk(" (default)");
952 EXPORT_SYMBOL_GPL(elv_register
);
954 void elv_unregister(struct elevator_type
*e
)
956 struct task_struct
*g
, *p
;
959 * Iterate every thread in the process to remove the io contexts.
962 read_lock(&tasklist_lock
);
963 do_each_thread(g
, p
) {
966 e
->ops
.trim(p
->io_context
);
968 } while_each_thread(g
, p
);
969 read_unlock(&tasklist_lock
);
972 spin_lock_irq(&elv_list_lock
);
973 list_del_init(&e
->list
);
974 spin_unlock_irq(&elv_list_lock
);
976 EXPORT_SYMBOL_GPL(elv_unregister
);
979 * switch to new_e io scheduler. be careful not to introduce deadlocks -
980 * we don't free the old io scheduler, before we have allocated what we
981 * need for the new one. this way we have a chance of going back to the old
982 * one, if the new one fails init for some reason.
984 static int elevator_switch(request_queue_t
*q
, struct elevator_type
*new_e
)
986 elevator_t
*old_elevator
, *e
;
990 * Allocate new elevator
992 e
= elevator_alloc(new_e
);
996 data
= elevator_init_queue(q
, e
);
998 kobject_put(&e
->kobj
);
1003 * Turn on BYPASS and drain all requests w/ elevator private data
1005 spin_lock_irq(q
->queue_lock
);
1007 set_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
1009 elv_drain_elevator(q
);
1011 while (q
->rq
.elvpriv
) {
1014 spin_unlock_irq(q
->queue_lock
);
1016 spin_lock_irq(q
->queue_lock
);
1017 elv_drain_elevator(q
);
1021 * Remember old elevator.
1023 old_elevator
= q
->elevator
;
1026 * attach and start new elevator
1028 elevator_attach(q
, e
, data
);
1030 spin_unlock_irq(q
->queue_lock
);
1032 __elv_unregister_queue(old_elevator
);
1034 if (elv_register_queue(q
))
1038 * finally exit old elevator and turn off BYPASS.
1040 elevator_exit(old_elevator
);
1041 clear_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
1046 * switch failed, exit the new io scheduler and reattach the old
1047 * one again (along with re-adding the sysfs dir)
1050 q
->elevator
= old_elevator
;
1051 elv_register_queue(q
);
1052 clear_bit(QUEUE_FLAG_ELVSWITCH
, &q
->queue_flags
);
1056 ssize_t
elv_iosched_store(request_queue_t
*q
, const char *name
, size_t count
)
1058 char elevator_name
[ELV_NAME_MAX
];
1060 struct elevator_type
*e
;
1062 elevator_name
[sizeof(elevator_name
) - 1] = '\0';
1063 strncpy(elevator_name
, name
, sizeof(elevator_name
) - 1);
1064 len
= strlen(elevator_name
);
1066 if (len
&& elevator_name
[len
- 1] == '\n')
1067 elevator_name
[len
- 1] = '\0';
1069 e
= elevator_get(elevator_name
);
1071 printk(KERN_ERR
"elevator: type %s not found\n", elevator_name
);
1075 if (!strcmp(elevator_name
, q
->elevator
->elevator_type
->elevator_name
)) {
1080 if (!elevator_switch(q
, e
))
1081 printk(KERN_ERR
"elevator: switch to %s failed\n",elevator_name
);
1085 ssize_t
elv_iosched_show(request_queue_t
*q
, char *name
)
1087 elevator_t
*e
= q
->elevator
;
1088 struct elevator_type
*elv
= e
->elevator_type
;
1089 struct list_head
*entry
;
1092 spin_lock_irq(q
->queue_lock
);
1093 list_for_each(entry
, &elv_list
) {
1094 struct elevator_type
*__e
;
1096 __e
= list_entry(entry
, struct elevator_type
, list
);
1097 if (!strcmp(elv
->elevator_name
, __e
->elevator_name
))
1098 len
+= sprintf(name
+len
, "[%s] ", elv
->elevator_name
);
1100 len
+= sprintf(name
+len
, "%s ", __e
->elevator_name
);
1102 spin_unlock_irq(q
->queue_lock
);
1104 len
+= sprintf(len
+name
, "\n");
1108 struct request
*elv_rb_former_request(request_queue_t
*q
, struct request
*rq
)
1110 struct rb_node
*rbprev
= rb_prev(&rq
->rb_node
);
1113 return rb_entry_rq(rbprev
);
1118 EXPORT_SYMBOL(elv_rb_former_request
);
1120 struct request
*elv_rb_latter_request(request_queue_t
*q
, struct request
*rq
)
1122 struct rb_node
*rbnext
= rb_next(&rq
->rb_node
);
1125 return rb_entry_rq(rbnext
);
1130 EXPORT_SYMBOL(elv_rb_latter_request
);