e1000: save skb counts in TX to avoid cache misses
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / block / elevator.c
blobb0b38ce0dcb6ec4cca1cbf28454e834cf3f1073e
1 /*
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
12 * an existing request
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
19 * Jens:
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>
26 #include <linux/fs.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>
37 #include <linux/uaccess.h>
39 #include <trace/events/block.h>
41 #include "blk.h"
43 static DEFINE_SPINLOCK(elv_list_lock);
44 static LIST_HEAD(elv_list);
47 * Merge hash stuff.
49 static const int elv_hash_shift = 6;
50 #define ELV_HASH_BLOCK(sec) ((sec) >> 3)
51 #define ELV_HASH_FN(sec) \
52 (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
53 #define ELV_HASH_ENTRIES (1 << elv_hash_shift)
54 #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq))
57 * Query io scheduler to see if the current process issuing bio may be
58 * merged with rq.
60 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
62 struct request_queue *q = rq->q;
63 struct elevator_queue *e = q->elevator;
65 if (e->ops->elevator_allow_merge_fn)
66 return e->ops->elevator_allow_merge_fn(q, rq, bio);
68 return 1;
72 * can we safely merge with this request?
74 int elv_rq_merge_ok(struct request *rq, struct bio *bio)
76 if (!rq_mergeable(rq))
77 return 0;
80 * Don't merge file system requests and discard requests
82 if ((bio->bi_rw & REQ_DISCARD) != (rq->bio->bi_rw & REQ_DISCARD))
83 return 0;
86 * Don't merge discard requests and secure discard requests
88 if ((bio->bi_rw & REQ_SECURE) != (rq->bio->bi_rw & REQ_SECURE))
89 return 0;
92 * different data direction or already started, don't merge
94 if (bio_data_dir(bio) != rq_data_dir(rq))
95 return 0;
98 * must be same device and not a special request
100 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
101 return 0;
104 * only merge integrity protected bio into ditto rq
106 if (bio_integrity(bio) != blk_integrity_rq(rq))
107 return 0;
109 if (!elv_iosched_allow_merge(rq, bio))
110 return 0;
112 return 1;
114 EXPORT_SYMBOL(elv_rq_merge_ok);
116 int elv_try_merge(struct request *__rq, struct bio *bio)
118 int ret = ELEVATOR_NO_MERGE;
121 * we can merge and sequence is ok, check if it's possible
123 if (elv_rq_merge_ok(__rq, bio)) {
124 if (blk_rq_pos(__rq) + blk_rq_sectors(__rq) == bio->bi_sector)
125 ret = ELEVATOR_BACK_MERGE;
126 else if (blk_rq_pos(__rq) - bio_sectors(bio) == bio->bi_sector)
127 ret = ELEVATOR_FRONT_MERGE;
130 return ret;
133 static struct elevator_type *elevator_find(const char *name)
135 struct elevator_type *e;
137 list_for_each_entry(e, &elv_list, list) {
138 if (!strcmp(e->elevator_name, name))
139 return e;
142 return NULL;
145 static void elevator_put(struct elevator_type *e)
147 module_put(e->elevator_owner);
150 static struct elevator_type *elevator_get(const char *name)
152 struct elevator_type *e;
154 spin_lock(&elv_list_lock);
156 e = elevator_find(name);
157 if (!e) {
158 spin_unlock(&elv_list_lock);
159 request_module("%s-iosched", name);
160 spin_lock(&elv_list_lock);
161 e = elevator_find(name);
164 if (e && !try_module_get(e->elevator_owner))
165 e = NULL;
167 spin_unlock(&elv_list_lock);
169 return e;
172 static void *elevator_init_queue(struct request_queue *q,
173 struct elevator_queue *eq)
175 return eq->ops->elevator_init_fn(q);
178 static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
179 void *data)
181 q->elevator = eq;
182 eq->elevator_data = data;
185 static char chosen_elevator[16];
187 static int __init elevator_setup(char *str)
190 * Be backwards-compatible with previous kernels, so users
191 * won't get the wrong elevator.
193 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
194 return 1;
197 __setup("elevator=", elevator_setup);
199 static struct kobj_type elv_ktype;
201 static struct elevator_queue *elevator_alloc(struct request_queue *q,
202 struct elevator_type *e)
204 struct elevator_queue *eq;
205 int i;
207 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
208 if (unlikely(!eq))
209 goto err;
211 eq->ops = &e->ops;
212 eq->elevator_type = e;
213 kobject_init(&eq->kobj, &elv_ktype);
214 mutex_init(&eq->sysfs_lock);
216 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
217 GFP_KERNEL, q->node);
218 if (!eq->hash)
219 goto err;
221 for (i = 0; i < ELV_HASH_ENTRIES; i++)
222 INIT_HLIST_HEAD(&eq->hash[i]);
224 return eq;
225 err:
226 kfree(eq);
227 elevator_put(e);
228 return NULL;
231 static void elevator_release(struct kobject *kobj)
233 struct elevator_queue *e;
235 e = container_of(kobj, struct elevator_queue, kobj);
236 elevator_put(e->elevator_type);
237 kfree(e->hash);
238 kfree(e);
241 int elevator_init(struct request_queue *q, char *name)
243 struct elevator_type *e = NULL;
244 struct elevator_queue *eq;
245 void *data;
247 if (unlikely(q->elevator))
248 return 0;
250 INIT_LIST_HEAD(&q->queue_head);
251 q->last_merge = NULL;
252 q->end_sector = 0;
253 q->boundary_rq = NULL;
255 if (name) {
256 e = elevator_get(name);
257 if (!e)
258 return -EINVAL;
261 if (!e && *chosen_elevator) {
262 e = elevator_get(chosen_elevator);
263 if (!e)
264 printk(KERN_ERR "I/O scheduler %s not found\n",
265 chosen_elevator);
268 if (!e) {
269 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
270 if (!e) {
271 printk(KERN_ERR
272 "Default I/O scheduler not found. " \
273 "Using noop.\n");
274 e = elevator_get("noop");
278 eq = elevator_alloc(q, e);
279 if (!eq)
280 return -ENOMEM;
282 data = elevator_init_queue(q, eq);
283 if (!data) {
284 kobject_put(&eq->kobj);
285 return -ENOMEM;
288 elevator_attach(q, eq, data);
289 return 0;
291 EXPORT_SYMBOL(elevator_init);
293 void elevator_exit(struct elevator_queue *e)
295 mutex_lock(&e->sysfs_lock);
296 if (e->ops->elevator_exit_fn)
297 e->ops->elevator_exit_fn(e);
298 e->ops = NULL;
299 mutex_unlock(&e->sysfs_lock);
301 kobject_put(&e->kobj);
303 EXPORT_SYMBOL(elevator_exit);
305 static inline void __elv_rqhash_del(struct request *rq)
307 hlist_del_init(&rq->hash);
310 static void elv_rqhash_del(struct request_queue *q, struct request *rq)
312 if (ELV_ON_HASH(rq))
313 __elv_rqhash_del(rq);
316 static void elv_rqhash_add(struct request_queue *q, struct request *rq)
318 struct elevator_queue *e = q->elevator;
320 BUG_ON(ELV_ON_HASH(rq));
321 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
324 static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
326 __elv_rqhash_del(rq);
327 elv_rqhash_add(q, rq);
330 static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
332 struct elevator_queue *e = q->elevator;
333 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
334 struct hlist_node *entry, *next;
335 struct request *rq;
337 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
338 BUG_ON(!ELV_ON_HASH(rq));
340 if (unlikely(!rq_mergeable(rq))) {
341 __elv_rqhash_del(rq);
342 continue;
345 if (rq_hash_key(rq) == offset)
346 return rq;
349 return NULL;
353 * RB-tree support functions for inserting/lookup/removal of requests
354 * in a sorted RB tree.
356 struct request *elv_rb_add(struct rb_root *root, struct request *rq)
358 struct rb_node **p = &root->rb_node;
359 struct rb_node *parent = NULL;
360 struct request *__rq;
362 while (*p) {
363 parent = *p;
364 __rq = rb_entry(parent, struct request, rb_node);
366 if (blk_rq_pos(rq) < blk_rq_pos(__rq))
367 p = &(*p)->rb_left;
368 else if (blk_rq_pos(rq) > blk_rq_pos(__rq))
369 p = &(*p)->rb_right;
370 else
371 return __rq;
374 rb_link_node(&rq->rb_node, parent, p);
375 rb_insert_color(&rq->rb_node, root);
376 return NULL;
378 EXPORT_SYMBOL(elv_rb_add);
380 void elv_rb_del(struct rb_root *root, struct request *rq)
382 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
383 rb_erase(&rq->rb_node, root);
384 RB_CLEAR_NODE(&rq->rb_node);
386 EXPORT_SYMBOL(elv_rb_del);
388 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
390 struct rb_node *n = root->rb_node;
391 struct request *rq;
393 while (n) {
394 rq = rb_entry(n, struct request, rb_node);
396 if (sector < blk_rq_pos(rq))
397 n = n->rb_left;
398 else if (sector > blk_rq_pos(rq))
399 n = n->rb_right;
400 else
401 return rq;
404 return NULL;
406 EXPORT_SYMBOL(elv_rb_find);
409 * Insert rq into dispatch queue of q. Queue lock must be held on
410 * entry. rq is sort instead into the dispatch queue. To be used by
411 * specific elevators.
413 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
415 sector_t boundary;
416 struct list_head *entry;
417 int stop_flags;
419 if (q->last_merge == rq)
420 q->last_merge = NULL;
422 elv_rqhash_del(q, rq);
424 q->nr_sorted--;
426 boundary = q->end_sector;
427 stop_flags = REQ_SOFTBARRIER | REQ_STARTED;
428 list_for_each_prev(entry, &q->queue_head) {
429 struct request *pos = list_entry_rq(entry);
431 if ((rq->cmd_flags & REQ_DISCARD) !=
432 (pos->cmd_flags & REQ_DISCARD))
433 break;
434 if (rq_data_dir(rq) != rq_data_dir(pos))
435 break;
436 if (pos->cmd_flags & stop_flags)
437 break;
438 if (blk_rq_pos(rq) >= boundary) {
439 if (blk_rq_pos(pos) < boundary)
440 continue;
441 } else {
442 if (blk_rq_pos(pos) >= boundary)
443 break;
445 if (blk_rq_pos(rq) >= blk_rq_pos(pos))
446 break;
449 list_add(&rq->queuelist, entry);
451 EXPORT_SYMBOL(elv_dispatch_sort);
454 * Insert rq into dispatch queue of q. Queue lock must be held on
455 * entry. rq is added to the back of the dispatch queue. To be used by
456 * specific elevators.
458 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
460 if (q->last_merge == rq)
461 q->last_merge = NULL;
463 elv_rqhash_del(q, rq);
465 q->nr_sorted--;
467 q->end_sector = rq_end_sector(rq);
468 q->boundary_rq = rq;
469 list_add_tail(&rq->queuelist, &q->queue_head);
471 EXPORT_SYMBOL(elv_dispatch_add_tail);
473 int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
475 struct elevator_queue *e = q->elevator;
476 struct request *__rq;
477 int ret;
480 * Levels of merges:
481 * nomerges: No merges at all attempted
482 * noxmerges: Only simple one-hit cache try
483 * merges: All merge tries attempted
485 if (blk_queue_nomerges(q))
486 return ELEVATOR_NO_MERGE;
489 * First try one-hit cache.
491 if (q->last_merge) {
492 ret = elv_try_merge(q->last_merge, bio);
493 if (ret != ELEVATOR_NO_MERGE) {
494 *req = q->last_merge;
495 return ret;
499 if (blk_queue_noxmerges(q))
500 return ELEVATOR_NO_MERGE;
503 * See if our hash lookup can find a potential backmerge.
505 __rq = elv_rqhash_find(q, bio->bi_sector);
506 if (__rq && elv_rq_merge_ok(__rq, bio)) {
507 *req = __rq;
508 return ELEVATOR_BACK_MERGE;
511 if (e->ops->elevator_merge_fn)
512 return e->ops->elevator_merge_fn(q, req, bio);
514 return ELEVATOR_NO_MERGE;
518 * Attempt to do an insertion back merge. Only check for the case where
519 * we can append 'rq' to an existing request, so we can throw 'rq' away
520 * afterwards.
522 * Returns true if we merged, false otherwise
524 static bool elv_attempt_insert_merge(struct request_queue *q,
525 struct request *rq)
527 struct request *__rq;
529 if (blk_queue_nomerges(q))
530 return false;
533 * First try one-hit cache.
535 if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq))
536 return true;
538 if (blk_queue_noxmerges(q))
539 return false;
542 * See if our hash lookup can find a potential backmerge.
544 __rq = elv_rqhash_find(q, blk_rq_pos(rq));
545 if (__rq && blk_attempt_req_merge(q, __rq, rq))
546 return true;
548 return false;
551 void elv_merged_request(struct request_queue *q, struct request *rq, int type)
553 struct elevator_queue *e = q->elevator;
555 if (e->ops->elevator_merged_fn)
556 e->ops->elevator_merged_fn(q, rq, type);
558 if (type == ELEVATOR_BACK_MERGE)
559 elv_rqhash_reposition(q, rq);
561 q->last_merge = rq;
564 void elv_merge_requests(struct request_queue *q, struct request *rq,
565 struct request *next)
567 struct elevator_queue *e = q->elevator;
568 const int next_sorted = next->cmd_flags & REQ_SORTED;
570 if (next_sorted && e->ops->elevator_merge_req_fn)
571 e->ops->elevator_merge_req_fn(q, rq, next);
573 elv_rqhash_reposition(q, rq);
575 if (next_sorted) {
576 elv_rqhash_del(q, next);
577 q->nr_sorted--;
580 q->last_merge = rq;
583 void elv_bio_merged(struct request_queue *q, struct request *rq,
584 struct bio *bio)
586 struct elevator_queue *e = q->elevator;
588 if (e->ops->elevator_bio_merged_fn)
589 e->ops->elevator_bio_merged_fn(q, rq, bio);
592 void elv_requeue_request(struct request_queue *q, struct request *rq)
595 * it already went through dequeue, we need to decrement the
596 * in_flight count again
598 if (blk_account_rq(rq)) {
599 q->in_flight[rq_is_sync(rq)]--;
600 if (rq->cmd_flags & REQ_SORTED)
601 elv_deactivate_rq(q, rq);
604 rq->cmd_flags &= ~REQ_STARTED;
606 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE);
609 void elv_drain_elevator(struct request_queue *q)
611 static int printed;
612 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
614 if (q->nr_sorted == 0)
615 return;
616 if (printed++ < 10) {
617 printk(KERN_ERR "%s: forced dispatching is broken "
618 "(nr_sorted=%u), please report this\n",
619 q->elevator->elevator_type->elevator_name, q->nr_sorted);
624 * Call with queue lock held, interrupts disabled
626 void elv_quiesce_start(struct request_queue *q)
628 if (!q->elevator)
629 return;
631 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
634 * make sure we don't have any requests in flight
636 elv_drain_elevator(q);
637 while (q->rq.elvpriv) {
638 __blk_run_queue(q);
639 spin_unlock_irq(q->queue_lock);
640 msleep(10);
641 spin_lock_irq(q->queue_lock);
642 elv_drain_elevator(q);
646 void elv_quiesce_end(struct request_queue *q)
648 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
651 void __elv_add_request(struct request_queue *q, struct request *rq, int where)
653 trace_block_rq_insert(q, rq);
655 rq->q = q;
657 if (rq->cmd_flags & REQ_SOFTBARRIER) {
658 /* barriers are scheduling boundary, update end_sector */
659 if (rq->cmd_type == REQ_TYPE_FS ||
660 (rq->cmd_flags & REQ_DISCARD)) {
661 q->end_sector = rq_end_sector(rq);
662 q->boundary_rq = rq;
664 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
665 (where == ELEVATOR_INSERT_SORT ||
666 where == ELEVATOR_INSERT_SORT_MERGE))
667 where = ELEVATOR_INSERT_BACK;
669 switch (where) {
670 case ELEVATOR_INSERT_REQUEUE:
671 case ELEVATOR_INSERT_FRONT:
672 rq->cmd_flags |= REQ_SOFTBARRIER;
673 list_add(&rq->queuelist, &q->queue_head);
674 break;
676 case ELEVATOR_INSERT_BACK:
677 rq->cmd_flags |= REQ_SOFTBARRIER;
678 elv_drain_elevator(q);
679 list_add_tail(&rq->queuelist, &q->queue_head);
681 * We kick the queue here for the following reasons.
682 * - The elevator might have returned NULL previously
683 * to delay requests and returned them now. As the
684 * queue wasn't empty before this request, ll_rw_blk
685 * won't run the queue on return, resulting in hang.
686 * - Usually, back inserted requests won't be merged
687 * with anything. There's no point in delaying queue
688 * processing.
690 __blk_run_queue(q);
691 break;
693 case ELEVATOR_INSERT_SORT_MERGE:
695 * If we succeed in merging this request with one in the
696 * queue already, we are done - rq has now been freed,
697 * so no need to do anything further.
699 if (elv_attempt_insert_merge(q, rq))
700 break;
701 case ELEVATOR_INSERT_SORT:
702 BUG_ON(rq->cmd_type != REQ_TYPE_FS &&
703 !(rq->cmd_flags & REQ_DISCARD));
704 rq->cmd_flags |= REQ_SORTED;
705 q->nr_sorted++;
706 if (rq_mergeable(rq)) {
707 elv_rqhash_add(q, rq);
708 if (!q->last_merge)
709 q->last_merge = rq;
713 * Some ioscheds (cfq) run q->request_fn directly, so
714 * rq cannot be accessed after calling
715 * elevator_add_req_fn.
717 q->elevator->ops->elevator_add_req_fn(q, rq);
718 break;
720 case ELEVATOR_INSERT_FLUSH:
721 rq->cmd_flags |= REQ_SOFTBARRIER;
722 blk_insert_flush(rq);
723 break;
724 default:
725 printk(KERN_ERR "%s: bad insertion point %d\n",
726 __func__, where);
727 BUG();
730 EXPORT_SYMBOL(__elv_add_request);
732 void elv_add_request(struct request_queue *q, struct request *rq, int where)
734 unsigned long flags;
736 spin_lock_irqsave(q->queue_lock, flags);
737 __elv_add_request(q, rq, where);
738 spin_unlock_irqrestore(q->queue_lock, flags);
740 EXPORT_SYMBOL(elv_add_request);
742 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
744 struct elevator_queue *e = q->elevator;
746 if (e->ops->elevator_latter_req_fn)
747 return e->ops->elevator_latter_req_fn(q, rq);
748 return NULL;
751 struct request *elv_former_request(struct request_queue *q, struct request *rq)
753 struct elevator_queue *e = q->elevator;
755 if (e->ops->elevator_former_req_fn)
756 return e->ops->elevator_former_req_fn(q, rq);
757 return NULL;
760 int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
762 struct elevator_queue *e = q->elevator;
764 if (e->ops->elevator_set_req_fn)
765 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
767 rq->elevator_private[0] = NULL;
768 return 0;
771 void elv_put_request(struct request_queue *q, struct request *rq)
773 struct elevator_queue *e = q->elevator;
775 if (e->ops->elevator_put_req_fn)
776 e->ops->elevator_put_req_fn(rq);
779 int elv_may_queue(struct request_queue *q, int rw)
781 struct elevator_queue *e = q->elevator;
783 if (e->ops->elevator_may_queue_fn)
784 return e->ops->elevator_may_queue_fn(q, rw);
786 return ELV_MQUEUE_MAY;
789 void elv_abort_queue(struct request_queue *q)
791 struct request *rq;
793 blk_abort_flushes(q);
795 while (!list_empty(&q->queue_head)) {
796 rq = list_entry_rq(q->queue_head.next);
797 rq->cmd_flags |= REQ_QUIET;
798 trace_block_rq_abort(q, rq);
800 * Mark this request as started so we don't trigger
801 * any debug logic in the end I/O path.
803 blk_start_request(rq);
804 __blk_end_request_all(rq, -EIO);
807 EXPORT_SYMBOL(elv_abort_queue);
809 void elv_completed_request(struct request_queue *q, struct request *rq)
811 struct elevator_queue *e = q->elevator;
814 * request is released from the driver, io must be done
816 if (blk_account_rq(rq)) {
817 q->in_flight[rq_is_sync(rq)]--;
818 if ((rq->cmd_flags & REQ_SORTED) &&
819 e->ops->elevator_completed_req_fn)
820 e->ops->elevator_completed_req_fn(q, rq);
824 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
826 static ssize_t
827 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
829 struct elv_fs_entry *entry = to_elv(attr);
830 struct elevator_queue *e;
831 ssize_t error;
833 if (!entry->show)
834 return -EIO;
836 e = container_of(kobj, struct elevator_queue, kobj);
837 mutex_lock(&e->sysfs_lock);
838 error = e->ops ? entry->show(e, page) : -ENOENT;
839 mutex_unlock(&e->sysfs_lock);
840 return error;
843 static ssize_t
844 elv_attr_store(struct kobject *kobj, struct attribute *attr,
845 const char *page, size_t length)
847 struct elv_fs_entry *entry = to_elv(attr);
848 struct elevator_queue *e;
849 ssize_t error;
851 if (!entry->store)
852 return -EIO;
854 e = container_of(kobj, struct elevator_queue, kobj);
855 mutex_lock(&e->sysfs_lock);
856 error = e->ops ? entry->store(e, page, length) : -ENOENT;
857 mutex_unlock(&e->sysfs_lock);
858 return error;
861 static const struct sysfs_ops elv_sysfs_ops = {
862 .show = elv_attr_show,
863 .store = elv_attr_store,
866 static struct kobj_type elv_ktype = {
867 .sysfs_ops = &elv_sysfs_ops,
868 .release = elevator_release,
871 int elv_register_queue(struct request_queue *q)
873 struct elevator_queue *e = q->elevator;
874 int error;
876 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
877 if (!error) {
878 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
879 if (attr) {
880 while (attr->attr.name) {
881 if (sysfs_create_file(&e->kobj, &attr->attr))
882 break;
883 attr++;
886 kobject_uevent(&e->kobj, KOBJ_ADD);
887 e->registered = 1;
889 return error;
891 EXPORT_SYMBOL(elv_register_queue);
893 static void __elv_unregister_queue(struct elevator_queue *e)
895 kobject_uevent(&e->kobj, KOBJ_REMOVE);
896 kobject_del(&e->kobj);
897 e->registered = 0;
900 void elv_unregister_queue(struct request_queue *q)
902 if (q)
903 __elv_unregister_queue(q->elevator);
905 EXPORT_SYMBOL(elv_unregister_queue);
907 void elv_register(struct elevator_type *e)
909 char *def = "";
911 spin_lock(&elv_list_lock);
912 BUG_ON(elevator_find(e->elevator_name));
913 list_add_tail(&e->list, &elv_list);
914 spin_unlock(&elv_list_lock);
916 if (!strcmp(e->elevator_name, chosen_elevator) ||
917 (!*chosen_elevator &&
918 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
919 def = " (default)";
921 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
922 def);
924 EXPORT_SYMBOL_GPL(elv_register);
926 void elv_unregister(struct elevator_type *e)
928 struct task_struct *g, *p;
931 * Iterate every thread in the process to remove the io contexts.
933 if (e->ops.trim) {
934 read_lock(&tasklist_lock);
935 do_each_thread(g, p) {
936 task_lock(p);
937 if (p->io_context)
938 e->ops.trim(p->io_context);
939 task_unlock(p);
940 } while_each_thread(g, p);
941 read_unlock(&tasklist_lock);
944 spin_lock(&elv_list_lock);
945 list_del_init(&e->list);
946 spin_unlock(&elv_list_lock);
948 EXPORT_SYMBOL_GPL(elv_unregister);
951 * switch to new_e io scheduler. be careful not to introduce deadlocks -
952 * we don't free the old io scheduler, before we have allocated what we
953 * need for the new one. this way we have a chance of going back to the old
954 * one, if the new one fails init for some reason.
956 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
958 struct elevator_queue *old_elevator, *e;
959 void *data;
960 int err;
963 * Allocate new elevator
965 e = elevator_alloc(q, new_e);
966 if (!e)
967 return -ENOMEM;
969 data = elevator_init_queue(q, e);
970 if (!data) {
971 kobject_put(&e->kobj);
972 return -ENOMEM;
976 * Turn on BYPASS and drain all requests w/ elevator private data
978 spin_lock_irq(q->queue_lock);
979 elv_quiesce_start(q);
982 * Remember old elevator.
984 old_elevator = q->elevator;
987 * attach and start new elevator
989 elevator_attach(q, e, data);
991 spin_unlock_irq(q->queue_lock);
993 if (old_elevator->registered) {
994 __elv_unregister_queue(old_elevator);
996 err = elv_register_queue(q);
997 if (err)
998 goto fail_register;
1002 * finally exit old elevator and turn off BYPASS.
1004 elevator_exit(old_elevator);
1005 spin_lock_irq(q->queue_lock);
1006 elv_quiesce_end(q);
1007 spin_unlock_irq(q->queue_lock);
1009 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1011 return 0;
1013 fail_register:
1015 * switch failed, exit the new io scheduler and reattach the old
1016 * one again (along with re-adding the sysfs dir)
1018 elevator_exit(e);
1019 q->elevator = old_elevator;
1020 elv_register_queue(q);
1022 spin_lock_irq(q->queue_lock);
1023 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1024 spin_unlock_irq(q->queue_lock);
1026 return err;
1030 * Switch this queue to the given IO scheduler.
1032 int elevator_change(struct request_queue *q, const char *name)
1034 char elevator_name[ELV_NAME_MAX];
1035 struct elevator_type *e;
1037 if (!q->elevator)
1038 return -ENXIO;
1040 strlcpy(elevator_name, name, sizeof(elevator_name));
1041 e = elevator_get(strstrip(elevator_name));
1042 if (!e) {
1043 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1044 return -EINVAL;
1047 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1048 elevator_put(e);
1049 return 0;
1052 return elevator_switch(q, e);
1054 EXPORT_SYMBOL(elevator_change);
1056 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1057 size_t count)
1059 int ret;
1061 if (!q->elevator)
1062 return count;
1064 ret = elevator_change(q, name);
1065 if (!ret)
1066 return count;
1068 printk(KERN_ERR "elevator: switch to %s failed\n", name);
1069 return ret;
1072 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1074 struct elevator_queue *e = q->elevator;
1075 struct elevator_type *elv;
1076 struct elevator_type *__e;
1077 int len = 0;
1079 if (!q->elevator || !blk_queue_stackable(q))
1080 return sprintf(name, "none\n");
1082 elv = e->elevator_type;
1084 spin_lock(&elv_list_lock);
1085 list_for_each_entry(__e, &elv_list, list) {
1086 if (!strcmp(elv->elevator_name, __e->elevator_name))
1087 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1088 else
1089 len += sprintf(name+len, "%s ", __e->elevator_name);
1091 spin_unlock(&elv_list_lock);
1093 len += sprintf(len+name, "\n");
1094 return len;
1097 struct request *elv_rb_former_request(struct request_queue *q,
1098 struct request *rq)
1100 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1102 if (rbprev)
1103 return rb_entry_rq(rbprev);
1105 return NULL;
1107 EXPORT_SYMBOL(elv_rb_former_request);
1109 struct request *elv_rb_latter_request(struct request_queue *q,
1110 struct request *rq)
1112 struct rb_node *rbnext = rb_next(&rq->rb_node);
1114 if (rbnext)
1115 return rb_entry_rq(rbnext);
1117 return NULL;
1119 EXPORT_SYMBOL(elv_rb_latter_request);