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 <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 * Query io scheduler to see if the current process issuing bio may be
57 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
59 request_queue_t *q = rq->q;
60 elevator_t *e = q->elevator;
62 if (e->ops->elevator_allow_merge_fn)
63 return e->ops->elevator_allow_merge_fn(q, rq, bio);
69 * can we safely merge with this request?
71 inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
73 if (!rq_mergeable(rq))
77 * different data direction or already started, don't merge
79 if (bio_data_dir(bio) != rq_data_dir(rq))
83 * must be same device and not a special request
85 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
88 if (!elv_iosched_allow_merge(rq, bio))
93 EXPORT_SYMBOL(elv_rq_merge_ok);
95 static inline int elv_try_merge(struct request *__rq, struct bio *bio)
97 int ret = ELEVATOR_NO_MERGE;
100 * we can merge and sequence is ok, check if it's possible
102 if (elv_rq_merge_ok(__rq, bio)) {
103 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
104 ret = ELEVATOR_BACK_MERGE;
105 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
106 ret = ELEVATOR_FRONT_MERGE;
112 static struct elevator_type *elevator_find(const char *name)
114 struct elevator_type *e;
116 list_for_each_entry(e, &elv_list, list) {
117 if (!strcmp(e->elevator_name, name))
124 static void elevator_put(struct elevator_type *e)
126 module_put(e->elevator_owner);
129 static struct elevator_type *elevator_get(const char *name)
131 struct elevator_type *e;
133 spin_lock(&elv_list_lock);
135 e = elevator_find(name);
136 if (e && !try_module_get(e->elevator_owner))
139 spin_unlock(&elv_list_lock);
144 static void *elevator_init_queue(request_queue_t *q, struct elevator_queue *eq)
146 return eq->ops->elevator_init_fn(q);
149 static void elevator_attach(request_queue_t *q, struct elevator_queue *eq,
153 eq->elevator_data = data;
156 static char chosen_elevator[16];
158 static int __init elevator_setup(char *str)
161 * Be backwards-compatible with previous kernels, so users
162 * won't get the wrong elevator.
164 if (!strcmp(str, "as"))
165 strcpy(chosen_elevator, "anticipatory");
167 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
171 __setup("elevator=", elevator_setup);
173 static struct kobj_type elv_ktype;
175 static elevator_t *elevator_alloc(request_queue_t *q, struct elevator_type *e)
180 eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL, q->node);
184 memset(eq, 0, sizeof(*eq));
186 eq->elevator_type = e;
187 kobject_init(&eq->kobj);
188 snprintf(eq->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
189 eq->kobj.ktype = &elv_ktype;
190 mutex_init(&eq->sysfs_lock);
192 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
193 GFP_KERNEL, q->node);
197 for (i = 0; i < ELV_HASH_ENTRIES; i++)
198 INIT_HLIST_HEAD(&eq->hash[i]);
207 static void elevator_release(struct kobject *kobj)
209 elevator_t *e = container_of(kobj, elevator_t, kobj);
211 elevator_put(e->elevator_type);
216 int elevator_init(request_queue_t *q, char *name)
218 struct elevator_type *e = NULL;
219 struct elevator_queue *eq;
223 INIT_LIST_HEAD(&q->queue_head);
224 q->last_merge = NULL;
226 q->boundary_rq = NULL;
228 if (name && !(e = elevator_get(name)))
231 if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator)))
232 printk("I/O scheduler %s not found\n", chosen_elevator);
234 if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) {
235 printk("Default I/O scheduler not found, using no-op\n");
236 e = elevator_get("noop");
239 eq = elevator_alloc(q, e);
243 data = elevator_init_queue(q, eq);
245 kobject_put(&eq->kobj);
249 elevator_attach(q, eq, data);
253 EXPORT_SYMBOL(elevator_init);
255 void elevator_exit(elevator_t *e)
257 mutex_lock(&e->sysfs_lock);
258 if (e->ops->elevator_exit_fn)
259 e->ops->elevator_exit_fn(e);
261 mutex_unlock(&e->sysfs_lock);
263 kobject_put(&e->kobj);
266 EXPORT_SYMBOL(elevator_exit);
268 static void elv_activate_rq(request_queue_t *q, struct request *rq)
270 elevator_t *e = q->elevator;
272 if (e->ops->elevator_activate_req_fn)
273 e->ops->elevator_activate_req_fn(q, rq);
276 static void elv_deactivate_rq(request_queue_t *q, struct request *rq)
278 elevator_t *e = q->elevator;
280 if (e->ops->elevator_deactivate_req_fn)
281 e->ops->elevator_deactivate_req_fn(q, rq);
284 static inline void __elv_rqhash_del(struct request *rq)
286 hlist_del_init(&rq->hash);
289 static void elv_rqhash_del(request_queue_t *q, struct request *rq)
292 __elv_rqhash_del(rq);
295 static void elv_rqhash_add(request_queue_t *q, struct request *rq)
297 elevator_t *e = q->elevator;
299 BUG_ON(ELV_ON_HASH(rq));
300 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
303 static void elv_rqhash_reposition(request_queue_t *q, struct request *rq)
305 __elv_rqhash_del(rq);
306 elv_rqhash_add(q, rq);
309 static struct request *elv_rqhash_find(request_queue_t *q, sector_t offset)
311 elevator_t *e = q->elevator;
312 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
313 struct hlist_node *entry, *next;
316 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
317 BUG_ON(!ELV_ON_HASH(rq));
319 if (unlikely(!rq_mergeable(rq))) {
320 __elv_rqhash_del(rq);
324 if (rq_hash_key(rq) == offset)
332 * RB-tree support functions for inserting/lookup/removal of requests
333 * in a sorted RB tree.
335 struct request *elv_rb_add(struct rb_root *root, struct request *rq)
337 struct rb_node **p = &root->rb_node;
338 struct rb_node *parent = NULL;
339 struct request *__rq;
343 __rq = rb_entry(parent, struct request, rb_node);
345 if (rq->sector < __rq->sector)
347 else if (rq->sector > __rq->sector)
353 rb_link_node(&rq->rb_node, parent, p);
354 rb_insert_color(&rq->rb_node, root);
358 EXPORT_SYMBOL(elv_rb_add);
360 void elv_rb_del(struct rb_root *root, struct request *rq)
362 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
363 rb_erase(&rq->rb_node, root);
364 RB_CLEAR_NODE(&rq->rb_node);
367 EXPORT_SYMBOL(elv_rb_del);
369 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
371 struct rb_node *n = root->rb_node;
375 rq = rb_entry(n, struct request, rb_node);
377 if (sector < rq->sector)
379 else if (sector > rq->sector)
388 EXPORT_SYMBOL(elv_rb_find);
391 * Insert rq into dispatch queue of q. Queue lock must be held on
392 * entry. rq is sort insted into the dispatch queue. To be used by
393 * specific elevators.
395 void elv_dispatch_sort(request_queue_t *q, struct request *rq)
398 struct list_head *entry;
400 if (q->last_merge == rq)
401 q->last_merge = NULL;
403 elv_rqhash_del(q, rq);
407 boundary = q->end_sector;
409 list_for_each_prev(entry, &q->queue_head) {
410 struct request *pos = list_entry_rq(entry);
412 if (rq_data_dir(rq) != rq_data_dir(pos))
414 if (pos->cmd_flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
416 if (rq->sector >= boundary) {
417 if (pos->sector < boundary)
420 if (pos->sector >= boundary)
423 if (rq->sector >= pos->sector)
427 list_add(&rq->queuelist, entry);
430 EXPORT_SYMBOL(elv_dispatch_sort);
433 * Insert rq into dispatch queue of q. Queue lock must be held on
434 * entry. rq is added to the back of the dispatch queue. To be used by
435 * specific elevators.
437 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
439 if (q->last_merge == rq)
440 q->last_merge = NULL;
442 elv_rqhash_del(q, rq);
446 q->end_sector = rq_end_sector(rq);
448 list_add_tail(&rq->queuelist, &q->queue_head);
451 EXPORT_SYMBOL(elv_dispatch_add_tail);
453 int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
455 elevator_t *e = q->elevator;
456 struct request *__rq;
460 * First try one-hit cache.
463 ret = elv_try_merge(q->last_merge, bio);
464 if (ret != ELEVATOR_NO_MERGE) {
465 *req = q->last_merge;
471 * See if our hash lookup can find a potential backmerge.
473 __rq = elv_rqhash_find(q, bio->bi_sector);
474 if (__rq && elv_rq_merge_ok(__rq, bio)) {
476 return ELEVATOR_BACK_MERGE;
479 if (e->ops->elevator_merge_fn)
480 return e->ops->elevator_merge_fn(q, req, bio);
482 return ELEVATOR_NO_MERGE;
485 void elv_merged_request(request_queue_t *q, struct request *rq, int type)
487 elevator_t *e = q->elevator;
489 if (e->ops->elevator_merged_fn)
490 e->ops->elevator_merged_fn(q, rq, type);
492 if (type == ELEVATOR_BACK_MERGE)
493 elv_rqhash_reposition(q, rq);
498 void elv_merge_requests(request_queue_t *q, struct request *rq,
499 struct request *next)
501 elevator_t *e = q->elevator;
503 if (e->ops->elevator_merge_req_fn)
504 e->ops->elevator_merge_req_fn(q, rq, next);
506 elv_rqhash_reposition(q, rq);
507 elv_rqhash_del(q, next);
513 void elv_requeue_request(request_queue_t *q, struct request *rq)
516 * it already went through dequeue, we need to decrement the
517 * in_flight count again
519 if (blk_account_rq(rq)) {
521 if (blk_sorted_rq(rq))
522 elv_deactivate_rq(q, rq);
525 rq->cmd_flags &= ~REQ_STARTED;
527 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
530 static void elv_drain_elevator(request_queue_t *q)
533 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
535 if (q->nr_sorted == 0)
537 if (printed++ < 10) {
538 printk(KERN_ERR "%s: forced dispatching is broken "
539 "(nr_sorted=%u), please report this\n",
540 q->elevator->elevator_type->elevator_name, q->nr_sorted);
544 void elv_insert(request_queue_t *q, struct request *rq, int where)
546 struct list_head *pos;
550 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
555 case ELEVATOR_INSERT_FRONT:
556 rq->cmd_flags |= REQ_SOFTBARRIER;
558 list_add(&rq->queuelist, &q->queue_head);
561 case ELEVATOR_INSERT_BACK:
562 rq->cmd_flags |= REQ_SOFTBARRIER;
563 elv_drain_elevator(q);
564 list_add_tail(&rq->queuelist, &q->queue_head);
566 * We kick the queue here for the following reasons.
567 * - The elevator might have returned NULL previously
568 * to delay requests and returned them now. As the
569 * queue wasn't empty before this request, ll_rw_blk
570 * won't run the queue on return, resulting in hang.
571 * - Usually, back inserted requests won't be merged
572 * with anything. There's no point in delaying queue
579 case ELEVATOR_INSERT_SORT:
580 BUG_ON(!blk_fs_request(rq));
581 rq->cmd_flags |= REQ_SORTED;
583 if (rq_mergeable(rq)) {
584 elv_rqhash_add(q, rq);
590 * Some ioscheds (cfq) run q->request_fn directly, so
591 * rq cannot be accessed after calling
592 * elevator_add_req_fn.
594 q->elevator->ops->elevator_add_req_fn(q, rq);
597 case ELEVATOR_INSERT_REQUEUE:
599 * If ordered flush isn't in progress, we do front
600 * insertion; otherwise, requests should be requeued
603 rq->cmd_flags |= REQ_SOFTBARRIER;
606 * Most requeues happen because of a busy condition,
607 * don't force unplug of the queue for that case.
611 if (q->ordseq == 0) {
612 list_add(&rq->queuelist, &q->queue_head);
616 ordseq = blk_ordered_req_seq(rq);
618 list_for_each(pos, &q->queue_head) {
619 struct request *pos_rq = list_entry_rq(pos);
620 if (ordseq <= blk_ordered_req_seq(pos_rq))
624 list_add_tail(&rq->queuelist, pos);
628 printk(KERN_ERR "%s: bad insertion point %d\n",
629 __FUNCTION__, where);
633 if (unplug_it && blk_queue_plugged(q)) {
634 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
637 if (nrq >= q->unplug_thresh)
638 __generic_unplug_device(q);
642 void __elv_add_request(request_queue_t *q, struct request *rq, int where,
646 rq->cmd_flags |= REQ_ORDERED_COLOR;
648 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
650 * toggle ordered color
652 if (blk_barrier_rq(rq))
656 * barriers implicitly indicate back insertion
658 if (where == ELEVATOR_INSERT_SORT)
659 where = ELEVATOR_INSERT_BACK;
662 * this request is scheduling boundary, update
665 if (blk_fs_request(rq)) {
666 q->end_sector = rq_end_sector(rq);
669 } else if (!(rq->cmd_flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
670 where = ELEVATOR_INSERT_BACK;
675 elv_insert(q, rq, where);
678 EXPORT_SYMBOL(__elv_add_request);
680 void elv_add_request(request_queue_t *q, struct request *rq, int where,
685 spin_lock_irqsave(q->queue_lock, flags);
686 __elv_add_request(q, rq, where, plug);
687 spin_unlock_irqrestore(q->queue_lock, flags);
690 EXPORT_SYMBOL(elv_add_request);
692 static inline struct request *__elv_next_request(request_queue_t *q)
697 while (!list_empty(&q->queue_head)) {
698 rq = list_entry_rq(q->queue_head.next);
699 if (blk_do_ordered(q, &rq))
703 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
708 struct request *elv_next_request(request_queue_t *q)
713 while ((rq = __elv_next_request(q)) != NULL) {
714 if (!(rq->cmd_flags & REQ_STARTED)) {
716 * This is the first time the device driver
717 * sees this request (possibly after
718 * requeueing). Notify IO scheduler.
720 if (blk_sorted_rq(rq))
721 elv_activate_rq(q, rq);
724 * just mark as started even if we don't start
725 * it, a request that has been delayed should
726 * not be passed by new incoming requests
728 rq->cmd_flags |= REQ_STARTED;
729 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
732 if (!q->boundary_rq || q->boundary_rq == rq) {
733 q->end_sector = rq_end_sector(rq);
734 q->boundary_rq = NULL;
737 if ((rq->cmd_flags & REQ_DONTPREP) || !q->prep_rq_fn)
740 ret = q->prep_rq_fn(q, rq);
741 if (ret == BLKPREP_OK) {
743 } else if (ret == BLKPREP_DEFER) {
745 * the request may have been (partially) prepped.
746 * we need to keep this request in the front to
747 * avoid resource deadlock. REQ_STARTED will
748 * prevent other fs requests from passing this one.
752 } else if (ret == BLKPREP_KILL) {
753 int nr_bytes = rq->hard_nr_sectors << 9;
756 nr_bytes = rq->data_len;
758 blkdev_dequeue_request(rq);
759 rq->cmd_flags |= REQ_QUIET;
760 end_that_request_chunk(rq, 0, nr_bytes);
761 end_that_request_last(rq, 0);
763 printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
772 EXPORT_SYMBOL(elv_next_request);
774 void elv_dequeue_request(request_queue_t *q, struct request *rq)
776 BUG_ON(list_empty(&rq->queuelist));
777 BUG_ON(ELV_ON_HASH(rq));
779 list_del_init(&rq->queuelist);
782 * the time frame between a request being removed from the lists
783 * and to it is freed is accounted as io that is in progress at
786 if (blk_account_rq(rq))
790 EXPORT_SYMBOL(elv_dequeue_request);
792 int elv_queue_empty(request_queue_t *q)
794 elevator_t *e = q->elevator;
796 if (!list_empty(&q->queue_head))
799 if (e->ops->elevator_queue_empty_fn)
800 return e->ops->elevator_queue_empty_fn(q);
805 EXPORT_SYMBOL(elv_queue_empty);
807 struct request *elv_latter_request(request_queue_t *q, struct request *rq)
809 elevator_t *e = q->elevator;
811 if (e->ops->elevator_latter_req_fn)
812 return e->ops->elevator_latter_req_fn(q, rq);
816 struct request *elv_former_request(request_queue_t *q, struct request *rq)
818 elevator_t *e = q->elevator;
820 if (e->ops->elevator_former_req_fn)
821 return e->ops->elevator_former_req_fn(q, rq);
825 int elv_set_request(request_queue_t *q, struct request *rq, gfp_t gfp_mask)
827 elevator_t *e = q->elevator;
829 if (e->ops->elevator_set_req_fn)
830 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
832 rq->elevator_private = NULL;
836 void elv_put_request(request_queue_t *q, struct request *rq)
838 elevator_t *e = q->elevator;
840 if (e->ops->elevator_put_req_fn)
841 e->ops->elevator_put_req_fn(rq);
844 int elv_may_queue(request_queue_t *q, int rw)
846 elevator_t *e = q->elevator;
848 if (e->ops->elevator_may_queue_fn)
849 return e->ops->elevator_may_queue_fn(q, rw);
851 return ELV_MQUEUE_MAY;
854 void elv_completed_request(request_queue_t *q, struct request *rq)
856 elevator_t *e = q->elevator;
859 * request is released from the driver, io must be done
861 if (blk_account_rq(rq)) {
863 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
864 e->ops->elevator_completed_req_fn(q, rq);
868 * Check if the queue is waiting for fs requests to be
869 * drained for flush sequence.
871 if (unlikely(q->ordseq)) {
872 struct request *first_rq = list_entry_rq(q->queue_head.next);
873 if (q->in_flight == 0 &&
874 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
875 blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
876 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
882 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
885 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
887 elevator_t *e = container_of(kobj, elevator_t, kobj);
888 struct elv_fs_entry *entry = to_elv(attr);
894 mutex_lock(&e->sysfs_lock);
895 error = e->ops ? entry->show(e, page) : -ENOENT;
896 mutex_unlock(&e->sysfs_lock);
901 elv_attr_store(struct kobject *kobj, struct attribute *attr,
902 const char *page, size_t length)
904 elevator_t *e = container_of(kobj, elevator_t, kobj);
905 struct elv_fs_entry *entry = to_elv(attr);
911 mutex_lock(&e->sysfs_lock);
912 error = e->ops ? entry->store(e, page, length) : -ENOENT;
913 mutex_unlock(&e->sysfs_lock);
917 static struct sysfs_ops elv_sysfs_ops = {
918 .show = elv_attr_show,
919 .store = elv_attr_store,
922 static struct kobj_type elv_ktype = {
923 .sysfs_ops = &elv_sysfs_ops,
924 .release = elevator_release,
927 int elv_register_queue(struct request_queue *q)
929 elevator_t *e = q->elevator;
932 e->kobj.parent = &q->kobj;
934 error = kobject_add(&e->kobj);
936 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
938 while (attr->attr.name) {
939 if (sysfs_create_file(&e->kobj, &attr->attr))
944 kobject_uevent(&e->kobj, KOBJ_ADD);
949 static void __elv_unregister_queue(elevator_t *e)
951 kobject_uevent(&e->kobj, KOBJ_REMOVE);
952 kobject_del(&e->kobj);
955 void elv_unregister_queue(struct request_queue *q)
958 __elv_unregister_queue(q->elevator);
961 int elv_register(struct elevator_type *e)
965 spin_lock(&elv_list_lock);
966 BUG_ON(elevator_find(e->elevator_name));
967 list_add_tail(&e->list, &elv_list);
968 spin_unlock(&elv_list_lock);
970 if (!strcmp(e->elevator_name, chosen_elevator) ||
971 (!*chosen_elevator &&
972 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
975 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name, def);
978 EXPORT_SYMBOL_GPL(elv_register);
980 void elv_unregister(struct elevator_type *e)
982 struct task_struct *g, *p;
985 * Iterate every thread in the process to remove the io contexts.
988 read_lock(&tasklist_lock);
989 do_each_thread(g, p) {
992 e->ops.trim(p->io_context);
994 } while_each_thread(g, p);
995 read_unlock(&tasklist_lock);
998 spin_lock(&elv_list_lock);
999 list_del_init(&e->list);
1000 spin_unlock(&elv_list_lock);
1002 EXPORT_SYMBOL_GPL(elv_unregister);
1005 * switch to new_e io scheduler. be careful not to introduce deadlocks -
1006 * we don't free the old io scheduler, before we have allocated what we
1007 * need for the new one. this way we have a chance of going back to the old
1008 * one, if the new one fails init for some reason.
1010 static int elevator_switch(request_queue_t *q, struct elevator_type *new_e)
1012 elevator_t *old_elevator, *e;
1016 * Allocate new elevator
1018 e = elevator_alloc(q, new_e);
1022 data = elevator_init_queue(q, e);
1024 kobject_put(&e->kobj);
1029 * Turn on BYPASS and drain all requests w/ elevator private data
1031 spin_lock_irq(q->queue_lock);
1033 set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1035 elv_drain_elevator(q);
1037 while (q->rq.elvpriv) {
1040 spin_unlock_irq(q->queue_lock);
1042 spin_lock_irq(q->queue_lock);
1043 elv_drain_elevator(q);
1047 * Remember old elevator.
1049 old_elevator = q->elevator;
1052 * attach and start new elevator
1054 elevator_attach(q, e, data);
1056 spin_unlock_irq(q->queue_lock);
1058 __elv_unregister_queue(old_elevator);
1060 if (elv_register_queue(q))
1064 * finally exit old elevator and turn off BYPASS.
1066 elevator_exit(old_elevator);
1067 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1072 * switch failed, exit the new io scheduler and reattach the old
1073 * one again (along with re-adding the sysfs dir)
1076 q->elevator = old_elevator;
1077 elv_register_queue(q);
1078 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1082 ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
1084 char elevator_name[ELV_NAME_MAX];
1086 struct elevator_type *e;
1088 elevator_name[sizeof(elevator_name) - 1] = '\0';
1089 strncpy(elevator_name, name, sizeof(elevator_name) - 1);
1090 len = strlen(elevator_name);
1092 if (len && elevator_name[len - 1] == '\n')
1093 elevator_name[len - 1] = '\0';
1095 e = elevator_get(elevator_name);
1097 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1101 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1106 if (!elevator_switch(q, e))
1107 printk(KERN_ERR "elevator: switch to %s failed\n",elevator_name);
1111 ssize_t elv_iosched_show(request_queue_t *q, char *name)
1113 elevator_t *e = q->elevator;
1114 struct elevator_type *elv = e->elevator_type;
1115 struct elevator_type *__e;
1118 spin_lock(&elv_list_lock);
1119 list_for_each_entry(__e, &elv_list, list) {
1120 if (!strcmp(elv->elevator_name, __e->elevator_name))
1121 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1123 len += sprintf(name+len, "%s ", __e->elevator_name);
1125 spin_unlock(&elv_list_lock);
1127 len += sprintf(len+name, "\n");
1131 struct request *elv_rb_former_request(request_queue_t *q, struct request *rq)
1133 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1136 return rb_entry_rq(rbprev);
1141 EXPORT_SYMBOL(elv_rb_former_request);
1143 struct request *elv_rb_latter_request(request_queue_t *q, struct request *rq)
1145 struct rb_node *rbnext = rb_next(&rq->rb_node);
1148 return rb_entry_rq(rbnext);
1153 EXPORT_SYMBOL(elv_rb_latter_request);