parport_gsc: fix printk format error
[linux-2.6/mini2440.git] / block / elevator.c
blob7073a9072577cdf3a0ae6e63c5ca247c2f493a5d
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 <trace/block.h>
37 #include <linux/hash.h>
38 #include <linux/uaccess.h>
40 #include "blk.h"
42 static DEFINE_SPINLOCK(elv_list_lock);
43 static LIST_HEAD(elv_list);
45 DEFINE_TRACE(block_rq_abort);
48 * Merge hash stuff.
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) ((rq)->sector + (rq)->nr_sectors)
56 #define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash))
58 DEFINE_TRACE(block_rq_insert);
59 DEFINE_TRACE(block_rq_issue);
62 * Query io scheduler to see if the current process issuing bio may be
63 * merged with rq.
65 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
67 struct request_queue *q = rq->q;
68 struct elevator_queue *e = q->elevator;
70 if (e->ops->elevator_allow_merge_fn)
71 return e->ops->elevator_allow_merge_fn(q, rq, bio);
73 return 1;
77 * can we safely merge with this request?
79 int elv_rq_merge_ok(struct request *rq, struct bio *bio)
81 if (!rq_mergeable(rq))
82 return 0;
85 * Don't merge file system requests and discard requests
87 if (bio_discard(bio) != bio_discard(rq->bio))
88 return 0;
91 * different data direction or already started, don't merge
93 if (bio_data_dir(bio) != rq_data_dir(rq))
94 return 0;
97 * must be same device and not a special request
99 if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
100 return 0;
103 * only merge integrity protected bio into ditto rq
105 if (bio_integrity(bio) != blk_integrity_rq(rq))
106 return 0;
108 if (!elv_iosched_allow_merge(rq, bio))
109 return 0;
111 return 1;
113 EXPORT_SYMBOL(elv_rq_merge_ok);
115 static inline int elv_try_merge(struct request *__rq, struct bio *bio)
117 int ret = ELEVATOR_NO_MERGE;
120 * we can merge and sequence is ok, check if it's possible
122 if (elv_rq_merge_ok(__rq, bio)) {
123 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
124 ret = ELEVATOR_BACK_MERGE;
125 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
126 ret = ELEVATOR_FRONT_MERGE;
129 return ret;
132 static struct elevator_type *elevator_find(const char *name)
134 struct elevator_type *e;
136 list_for_each_entry(e, &elv_list, list) {
137 if (!strcmp(e->elevator_name, name))
138 return e;
141 return NULL;
144 static void elevator_put(struct elevator_type *e)
146 module_put(e->elevator_owner);
149 static struct elevator_type *elevator_get(const char *name)
151 struct elevator_type *e;
153 spin_lock(&elv_list_lock);
155 e = elevator_find(name);
156 if (!e) {
157 char elv[ELV_NAME_MAX + strlen("-iosched")];
159 spin_unlock(&elv_list_lock);
161 if (!strcmp(name, "anticipatory"))
162 sprintf(elv, "as-iosched");
163 else
164 sprintf(elv, "%s-iosched", name);
166 request_module("%s", elv);
167 spin_lock(&elv_list_lock);
168 e = elevator_find(name);
171 if (e && !try_module_get(e->elevator_owner))
172 e = NULL;
174 spin_unlock(&elv_list_lock);
176 return e;
179 static void *elevator_init_queue(struct request_queue *q,
180 struct elevator_queue *eq)
182 return eq->ops->elevator_init_fn(q);
185 static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
186 void *data)
188 q->elevator = eq;
189 eq->elevator_data = data;
192 static char chosen_elevator[16];
194 static int __init elevator_setup(char *str)
197 * Be backwards-compatible with previous kernels, so users
198 * won't get the wrong elevator.
200 if (!strcmp(str, "as"))
201 strcpy(chosen_elevator, "anticipatory");
202 else
203 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
204 return 1;
207 __setup("elevator=", elevator_setup);
209 static struct kobj_type elv_ktype;
211 static struct elevator_queue *elevator_alloc(struct request_queue *q,
212 struct elevator_type *e)
214 struct elevator_queue *eq;
215 int i;
217 eq = kmalloc_node(sizeof(*eq), GFP_KERNEL | __GFP_ZERO, q->node);
218 if (unlikely(!eq))
219 goto err;
221 eq->ops = &e->ops;
222 eq->elevator_type = e;
223 kobject_init(&eq->kobj, &elv_ktype);
224 mutex_init(&eq->sysfs_lock);
226 eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
227 GFP_KERNEL, q->node);
228 if (!eq->hash)
229 goto err;
231 for (i = 0; i < ELV_HASH_ENTRIES; i++)
232 INIT_HLIST_HEAD(&eq->hash[i]);
234 return eq;
235 err:
236 kfree(eq);
237 elevator_put(e);
238 return NULL;
241 static void elevator_release(struct kobject *kobj)
243 struct elevator_queue *e;
245 e = container_of(kobj, struct elevator_queue, kobj);
246 elevator_put(e->elevator_type);
247 kfree(e->hash);
248 kfree(e);
251 int elevator_init(struct request_queue *q, char *name)
253 struct elevator_type *e = NULL;
254 struct elevator_queue *eq;
255 int ret = 0;
256 void *data;
258 INIT_LIST_HEAD(&q->queue_head);
259 q->last_merge = NULL;
260 q->end_sector = 0;
261 q->boundary_rq = NULL;
263 if (name) {
264 e = elevator_get(name);
265 if (!e)
266 return -EINVAL;
269 if (!e && *chosen_elevator) {
270 e = elevator_get(chosen_elevator);
271 if (!e)
272 printk(KERN_ERR "I/O scheduler %s not found\n",
273 chosen_elevator);
276 if (!e) {
277 e = elevator_get(CONFIG_DEFAULT_IOSCHED);
278 if (!e) {
279 printk(KERN_ERR
280 "Default I/O scheduler not found. " \
281 "Using noop.\n");
282 e = elevator_get("noop");
286 eq = elevator_alloc(q, e);
287 if (!eq)
288 return -ENOMEM;
290 data = elevator_init_queue(q, eq);
291 if (!data) {
292 kobject_put(&eq->kobj);
293 return -ENOMEM;
296 elevator_attach(q, eq, data);
297 return ret;
299 EXPORT_SYMBOL(elevator_init);
301 void elevator_exit(struct elevator_queue *e)
303 mutex_lock(&e->sysfs_lock);
304 if (e->ops->elevator_exit_fn)
305 e->ops->elevator_exit_fn(e);
306 e->ops = NULL;
307 mutex_unlock(&e->sysfs_lock);
309 kobject_put(&e->kobj);
311 EXPORT_SYMBOL(elevator_exit);
313 static void elv_activate_rq(struct request_queue *q, struct request *rq)
315 struct elevator_queue *e = q->elevator;
317 if (e->ops->elevator_activate_req_fn)
318 e->ops->elevator_activate_req_fn(q, rq);
321 static void elv_deactivate_rq(struct request_queue *q, struct request *rq)
323 struct elevator_queue *e = q->elevator;
325 if (e->ops->elevator_deactivate_req_fn)
326 e->ops->elevator_deactivate_req_fn(q, rq);
329 static inline void __elv_rqhash_del(struct request *rq)
331 hlist_del_init(&rq->hash);
334 static void elv_rqhash_del(struct request_queue *q, struct request *rq)
336 if (ELV_ON_HASH(rq))
337 __elv_rqhash_del(rq);
340 static void elv_rqhash_add(struct request_queue *q, struct request *rq)
342 struct elevator_queue *e = q->elevator;
344 BUG_ON(ELV_ON_HASH(rq));
345 hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
348 static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
350 __elv_rqhash_del(rq);
351 elv_rqhash_add(q, rq);
354 static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
356 struct elevator_queue *e = q->elevator;
357 struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
358 struct hlist_node *entry, *next;
359 struct request *rq;
361 hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
362 BUG_ON(!ELV_ON_HASH(rq));
364 if (unlikely(!rq_mergeable(rq))) {
365 __elv_rqhash_del(rq);
366 continue;
369 if (rq_hash_key(rq) == offset)
370 return rq;
373 return NULL;
377 * RB-tree support functions for inserting/lookup/removal of requests
378 * in a sorted RB tree.
380 struct request *elv_rb_add(struct rb_root *root, struct request *rq)
382 struct rb_node **p = &root->rb_node;
383 struct rb_node *parent = NULL;
384 struct request *__rq;
386 while (*p) {
387 parent = *p;
388 __rq = rb_entry(parent, struct request, rb_node);
390 if (rq->sector < __rq->sector)
391 p = &(*p)->rb_left;
392 else if (rq->sector > __rq->sector)
393 p = &(*p)->rb_right;
394 else
395 return __rq;
398 rb_link_node(&rq->rb_node, parent, p);
399 rb_insert_color(&rq->rb_node, root);
400 return NULL;
402 EXPORT_SYMBOL(elv_rb_add);
404 void elv_rb_del(struct rb_root *root, struct request *rq)
406 BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
407 rb_erase(&rq->rb_node, root);
408 RB_CLEAR_NODE(&rq->rb_node);
410 EXPORT_SYMBOL(elv_rb_del);
412 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
414 struct rb_node *n = root->rb_node;
415 struct request *rq;
417 while (n) {
418 rq = rb_entry(n, struct request, rb_node);
420 if (sector < rq->sector)
421 n = n->rb_left;
422 else if (sector > rq->sector)
423 n = n->rb_right;
424 else
425 return rq;
428 return NULL;
430 EXPORT_SYMBOL(elv_rb_find);
433 * Insert rq into dispatch queue of q. Queue lock must be held on
434 * entry. rq is sort instead into the dispatch queue. To be used by
435 * specific elevators.
437 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
439 sector_t boundary;
440 struct list_head *entry;
441 int stop_flags;
443 if (q->last_merge == rq)
444 q->last_merge = NULL;
446 elv_rqhash_del(q, rq);
448 q->nr_sorted--;
450 boundary = q->end_sector;
451 stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
452 list_for_each_prev(entry, &q->queue_head) {
453 struct request *pos = list_entry_rq(entry);
455 if (blk_discard_rq(rq) != blk_discard_rq(pos))
456 break;
457 if (rq_data_dir(rq) != rq_data_dir(pos))
458 break;
459 if (pos->cmd_flags & stop_flags)
460 break;
461 if (rq->sector >= boundary) {
462 if (pos->sector < boundary)
463 continue;
464 } else {
465 if (pos->sector >= boundary)
466 break;
468 if (rq->sector >= pos->sector)
469 break;
472 list_add(&rq->queuelist, entry);
474 EXPORT_SYMBOL(elv_dispatch_sort);
477 * Insert rq into dispatch queue of q. Queue lock must be held on
478 * entry. rq is added to the back of the dispatch queue. To be used by
479 * specific elevators.
481 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
483 if (q->last_merge == rq)
484 q->last_merge = NULL;
486 elv_rqhash_del(q, rq);
488 q->nr_sorted--;
490 q->end_sector = rq_end_sector(rq);
491 q->boundary_rq = rq;
492 list_add_tail(&rq->queuelist, &q->queue_head);
494 EXPORT_SYMBOL(elv_dispatch_add_tail);
496 int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
498 struct elevator_queue *e = q->elevator;
499 struct request *__rq;
500 int ret;
503 * First try one-hit cache.
505 if (q->last_merge) {
506 ret = elv_try_merge(q->last_merge, bio);
507 if (ret != ELEVATOR_NO_MERGE) {
508 *req = q->last_merge;
509 return ret;
513 if (blk_queue_nomerges(q))
514 return ELEVATOR_NO_MERGE;
517 * See if our hash lookup can find a potential backmerge.
519 __rq = elv_rqhash_find(q, bio->bi_sector);
520 if (__rq && elv_rq_merge_ok(__rq, bio)) {
521 *req = __rq;
522 return ELEVATOR_BACK_MERGE;
525 if (e->ops->elevator_merge_fn)
526 return e->ops->elevator_merge_fn(q, req, bio);
528 return ELEVATOR_NO_MERGE;
531 void elv_merged_request(struct request_queue *q, struct request *rq, int type)
533 struct elevator_queue *e = q->elevator;
535 if (e->ops->elevator_merged_fn)
536 e->ops->elevator_merged_fn(q, rq, type);
538 if (type == ELEVATOR_BACK_MERGE)
539 elv_rqhash_reposition(q, rq);
541 q->last_merge = rq;
544 void elv_merge_requests(struct request_queue *q, struct request *rq,
545 struct request *next)
547 struct elevator_queue *e = q->elevator;
549 if (e->ops->elevator_merge_req_fn)
550 e->ops->elevator_merge_req_fn(q, rq, next);
552 elv_rqhash_reposition(q, rq);
553 elv_rqhash_del(q, next);
555 q->nr_sorted--;
556 q->last_merge = rq;
559 void elv_requeue_request(struct request_queue *q, struct request *rq)
562 * it already went through dequeue, we need to decrement the
563 * in_flight count again
565 if (blk_account_rq(rq)) {
566 q->in_flight--;
567 if (blk_sorted_rq(rq))
568 elv_deactivate_rq(q, rq);
571 rq->cmd_flags &= ~REQ_STARTED;
573 elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
576 void elv_drain_elevator(struct request_queue *q)
578 static int printed;
579 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
581 if (q->nr_sorted == 0)
582 return;
583 if (printed++ < 10) {
584 printk(KERN_ERR "%s: forced dispatching is broken "
585 "(nr_sorted=%u), please report this\n",
586 q->elevator->elevator_type->elevator_name, q->nr_sorted);
591 * Call with queue lock held, interrupts disabled
593 void elv_quiesce_start(struct request_queue *q)
595 queue_flag_set(QUEUE_FLAG_ELVSWITCH, q);
598 * make sure we don't have any requests in flight
600 elv_drain_elevator(q);
601 while (q->rq.elvpriv) {
602 blk_start_queueing(q);
603 spin_unlock_irq(q->queue_lock);
604 msleep(10);
605 spin_lock_irq(q->queue_lock);
606 elv_drain_elevator(q);
610 void elv_quiesce_end(struct request_queue *q)
612 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
615 void elv_insert(struct request_queue *q, struct request *rq, int where)
617 struct list_head *pos;
618 unsigned ordseq;
619 int unplug_it = 1;
621 trace_block_rq_insert(q, rq);
623 rq->q = q;
625 switch (where) {
626 case ELEVATOR_INSERT_FRONT:
627 rq->cmd_flags |= REQ_SOFTBARRIER;
629 list_add(&rq->queuelist, &q->queue_head);
630 break;
632 case ELEVATOR_INSERT_BACK:
633 rq->cmd_flags |= REQ_SOFTBARRIER;
634 elv_drain_elevator(q);
635 list_add_tail(&rq->queuelist, &q->queue_head);
637 * We kick the queue here for the following reasons.
638 * - The elevator might have returned NULL previously
639 * to delay requests and returned them now. As the
640 * queue wasn't empty before this request, ll_rw_blk
641 * won't run the queue on return, resulting in hang.
642 * - Usually, back inserted requests won't be merged
643 * with anything. There's no point in delaying queue
644 * processing.
646 blk_remove_plug(q);
647 blk_start_queueing(q);
648 break;
650 case ELEVATOR_INSERT_SORT:
651 BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq));
652 rq->cmd_flags |= REQ_SORTED;
653 q->nr_sorted++;
654 if (rq_mergeable(rq)) {
655 elv_rqhash_add(q, rq);
656 if (!q->last_merge)
657 q->last_merge = rq;
661 * Some ioscheds (cfq) run q->request_fn directly, so
662 * rq cannot be accessed after calling
663 * elevator_add_req_fn.
665 q->elevator->ops->elevator_add_req_fn(q, rq);
666 break;
668 case ELEVATOR_INSERT_REQUEUE:
670 * If ordered flush isn't in progress, we do front
671 * insertion; otherwise, requests should be requeued
672 * in ordseq order.
674 rq->cmd_flags |= REQ_SOFTBARRIER;
677 * Most requeues happen because of a busy condition,
678 * don't force unplug of the queue for that case.
680 unplug_it = 0;
682 if (q->ordseq == 0) {
683 list_add(&rq->queuelist, &q->queue_head);
684 break;
687 ordseq = blk_ordered_req_seq(rq);
689 list_for_each(pos, &q->queue_head) {
690 struct request *pos_rq = list_entry_rq(pos);
691 if (ordseq <= blk_ordered_req_seq(pos_rq))
692 break;
695 list_add_tail(&rq->queuelist, pos);
696 break;
698 default:
699 printk(KERN_ERR "%s: bad insertion point %d\n",
700 __func__, where);
701 BUG();
704 if (unplug_it && blk_queue_plugged(q)) {
705 int nrq = q->rq.count[BLK_RW_SYNC] + q->rq.count[BLK_RW_ASYNC]
706 - q->in_flight;
708 if (nrq >= q->unplug_thresh)
709 __generic_unplug_device(q);
713 void __elv_add_request(struct request_queue *q, struct request *rq, int where,
714 int plug)
716 if (q->ordcolor)
717 rq->cmd_flags |= REQ_ORDERED_COLOR;
719 if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
721 * toggle ordered color
723 if (blk_barrier_rq(rq))
724 q->ordcolor ^= 1;
727 * barriers implicitly indicate back insertion
729 if (where == ELEVATOR_INSERT_SORT)
730 where = ELEVATOR_INSERT_BACK;
733 * this request is scheduling boundary, update
734 * end_sector
736 if (blk_fs_request(rq) || blk_discard_rq(rq)) {
737 q->end_sector = rq_end_sector(rq);
738 q->boundary_rq = rq;
740 } else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
741 where == ELEVATOR_INSERT_SORT)
742 where = ELEVATOR_INSERT_BACK;
744 if (plug)
745 blk_plug_device(q);
747 elv_insert(q, rq, where);
749 EXPORT_SYMBOL(__elv_add_request);
751 void elv_add_request(struct request_queue *q, struct request *rq, int where,
752 int plug)
754 unsigned long flags;
756 spin_lock_irqsave(q->queue_lock, flags);
757 __elv_add_request(q, rq, where, plug);
758 spin_unlock_irqrestore(q->queue_lock, flags);
760 EXPORT_SYMBOL(elv_add_request);
762 static inline struct request *__elv_next_request(struct request_queue *q)
764 struct request *rq;
766 while (1) {
767 while (!list_empty(&q->queue_head)) {
768 rq = list_entry_rq(q->queue_head.next);
769 if (blk_do_ordered(q, &rq))
770 return rq;
773 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
774 return NULL;
778 struct request *elv_next_request(struct request_queue *q)
780 struct request *rq;
781 int ret;
783 while ((rq = __elv_next_request(q)) != NULL) {
784 if (!(rq->cmd_flags & REQ_STARTED)) {
786 * This is the first time the device driver
787 * sees this request (possibly after
788 * requeueing). Notify IO scheduler.
790 if (blk_sorted_rq(rq))
791 elv_activate_rq(q, rq);
794 * just mark as started even if we don't start
795 * it, a request that has been delayed should
796 * not be passed by new incoming requests
798 rq->cmd_flags |= REQ_STARTED;
799 trace_block_rq_issue(q, rq);
802 if (!q->boundary_rq || q->boundary_rq == rq) {
803 q->end_sector = rq_end_sector(rq);
804 q->boundary_rq = NULL;
807 if (rq->cmd_flags & REQ_DONTPREP)
808 break;
810 if (q->dma_drain_size && rq->data_len) {
812 * make sure space for the drain appears we
813 * know we can do this because max_hw_segments
814 * has been adjusted to be one fewer than the
815 * device can handle
817 rq->nr_phys_segments++;
820 if (!q->prep_rq_fn)
821 break;
823 ret = q->prep_rq_fn(q, rq);
824 if (ret == BLKPREP_OK) {
825 break;
826 } else if (ret == BLKPREP_DEFER) {
828 * the request may have been (partially) prepped.
829 * we need to keep this request in the front to
830 * avoid resource deadlock. REQ_STARTED will
831 * prevent other fs requests from passing this one.
833 if (q->dma_drain_size && rq->data_len &&
834 !(rq->cmd_flags & REQ_DONTPREP)) {
836 * remove the space for the drain we added
837 * so that we don't add it again
839 --rq->nr_phys_segments;
842 rq = NULL;
843 break;
844 } else if (ret == BLKPREP_KILL) {
845 rq->cmd_flags |= REQ_QUIET;
846 __blk_end_request(rq, -EIO, blk_rq_bytes(rq));
847 } else {
848 printk(KERN_ERR "%s: bad return=%d\n", __func__, ret);
849 break;
853 return rq;
855 EXPORT_SYMBOL(elv_next_request);
857 void elv_dequeue_request(struct request_queue *q, struct request *rq)
859 BUG_ON(list_empty(&rq->queuelist));
860 BUG_ON(ELV_ON_HASH(rq));
862 list_del_init(&rq->queuelist);
865 * the time frame between a request being removed from the lists
866 * and to it is freed is accounted as io that is in progress at
867 * the driver side.
869 if (blk_account_rq(rq))
870 q->in_flight++;
873 int elv_queue_empty(struct request_queue *q)
875 struct elevator_queue *e = q->elevator;
877 if (!list_empty(&q->queue_head))
878 return 0;
880 if (e->ops->elevator_queue_empty_fn)
881 return e->ops->elevator_queue_empty_fn(q);
883 return 1;
885 EXPORT_SYMBOL(elv_queue_empty);
887 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
889 struct elevator_queue *e = q->elevator;
891 if (e->ops->elevator_latter_req_fn)
892 return e->ops->elevator_latter_req_fn(q, rq);
893 return NULL;
896 struct request *elv_former_request(struct request_queue *q, struct request *rq)
898 struct elevator_queue *e = q->elevator;
900 if (e->ops->elevator_former_req_fn)
901 return e->ops->elevator_former_req_fn(q, rq);
902 return NULL;
905 int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
907 struct elevator_queue *e = q->elevator;
909 if (e->ops->elevator_set_req_fn)
910 return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
912 rq->elevator_private = NULL;
913 return 0;
916 void elv_put_request(struct request_queue *q, struct request *rq)
918 struct elevator_queue *e = q->elevator;
920 if (e->ops->elevator_put_req_fn)
921 e->ops->elevator_put_req_fn(rq);
924 int elv_may_queue(struct request_queue *q, int rw)
926 struct elevator_queue *e = q->elevator;
928 if (e->ops->elevator_may_queue_fn)
929 return e->ops->elevator_may_queue_fn(q, rw);
931 return ELV_MQUEUE_MAY;
934 void elv_abort_queue(struct request_queue *q)
936 struct request *rq;
938 while (!list_empty(&q->queue_head)) {
939 rq = list_entry_rq(q->queue_head.next);
940 rq->cmd_flags |= REQ_QUIET;
941 trace_block_rq_abort(q, rq);
942 __blk_end_request(rq, -EIO, blk_rq_bytes(rq));
945 EXPORT_SYMBOL(elv_abort_queue);
947 void elv_completed_request(struct request_queue *q, struct request *rq)
949 struct elevator_queue *e = q->elevator;
952 * request is released from the driver, io must be done
954 if (blk_account_rq(rq)) {
955 q->in_flight--;
956 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
957 e->ops->elevator_completed_req_fn(q, rq);
961 * Check if the queue is waiting for fs requests to be
962 * drained for flush sequence.
964 if (unlikely(q->ordseq)) {
965 struct request *next = NULL;
967 if (!list_empty(&q->queue_head))
968 next = list_entry_rq(q->queue_head.next);
970 if (!q->in_flight &&
971 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
972 (!next || blk_ordered_req_seq(next) > QUEUE_ORDSEQ_DRAIN)) {
973 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
974 blk_start_queueing(q);
979 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
981 static ssize_t
982 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
984 struct elv_fs_entry *entry = to_elv(attr);
985 struct elevator_queue *e;
986 ssize_t error;
988 if (!entry->show)
989 return -EIO;
991 e = container_of(kobj, struct elevator_queue, kobj);
992 mutex_lock(&e->sysfs_lock);
993 error = e->ops ? entry->show(e, page) : -ENOENT;
994 mutex_unlock(&e->sysfs_lock);
995 return error;
998 static ssize_t
999 elv_attr_store(struct kobject *kobj, struct attribute *attr,
1000 const char *page, size_t length)
1002 struct elv_fs_entry *entry = to_elv(attr);
1003 struct elevator_queue *e;
1004 ssize_t error;
1006 if (!entry->store)
1007 return -EIO;
1009 e = container_of(kobj, struct elevator_queue, kobj);
1010 mutex_lock(&e->sysfs_lock);
1011 error = e->ops ? entry->store(e, page, length) : -ENOENT;
1012 mutex_unlock(&e->sysfs_lock);
1013 return error;
1016 static struct sysfs_ops elv_sysfs_ops = {
1017 .show = elv_attr_show,
1018 .store = elv_attr_store,
1021 static struct kobj_type elv_ktype = {
1022 .sysfs_ops = &elv_sysfs_ops,
1023 .release = elevator_release,
1026 int elv_register_queue(struct request_queue *q)
1028 struct elevator_queue *e = q->elevator;
1029 int error;
1031 error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
1032 if (!error) {
1033 struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
1034 if (attr) {
1035 while (attr->attr.name) {
1036 if (sysfs_create_file(&e->kobj, &attr->attr))
1037 break;
1038 attr++;
1041 kobject_uevent(&e->kobj, KOBJ_ADD);
1043 return error;
1046 static void __elv_unregister_queue(struct elevator_queue *e)
1048 kobject_uevent(&e->kobj, KOBJ_REMOVE);
1049 kobject_del(&e->kobj);
1052 void elv_unregister_queue(struct request_queue *q)
1054 if (q)
1055 __elv_unregister_queue(q->elevator);
1058 void elv_register(struct elevator_type *e)
1060 char *def = "";
1062 spin_lock(&elv_list_lock);
1063 BUG_ON(elevator_find(e->elevator_name));
1064 list_add_tail(&e->list, &elv_list);
1065 spin_unlock(&elv_list_lock);
1067 if (!strcmp(e->elevator_name, chosen_elevator) ||
1068 (!*chosen_elevator &&
1069 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
1070 def = " (default)";
1072 printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
1073 def);
1075 EXPORT_SYMBOL_GPL(elv_register);
1077 void elv_unregister(struct elevator_type *e)
1079 struct task_struct *g, *p;
1082 * Iterate every thread in the process to remove the io contexts.
1084 if (e->ops.trim) {
1085 read_lock(&tasklist_lock);
1086 do_each_thread(g, p) {
1087 task_lock(p);
1088 if (p->io_context)
1089 e->ops.trim(p->io_context);
1090 task_unlock(p);
1091 } while_each_thread(g, p);
1092 read_unlock(&tasklist_lock);
1095 spin_lock(&elv_list_lock);
1096 list_del_init(&e->list);
1097 spin_unlock(&elv_list_lock);
1099 EXPORT_SYMBOL_GPL(elv_unregister);
1102 * switch to new_e io scheduler. be careful not to introduce deadlocks -
1103 * we don't free the old io scheduler, before we have allocated what we
1104 * need for the new one. this way we have a chance of going back to the old
1105 * one, if the new one fails init for some reason.
1107 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
1109 struct elevator_queue *old_elevator, *e;
1110 void *data;
1113 * Allocate new elevator
1115 e = elevator_alloc(q, new_e);
1116 if (!e)
1117 return 0;
1119 data = elevator_init_queue(q, e);
1120 if (!data) {
1121 kobject_put(&e->kobj);
1122 return 0;
1126 * Turn on BYPASS and drain all requests w/ elevator private data
1128 spin_lock_irq(q->queue_lock);
1129 elv_quiesce_start(q);
1132 * Remember old elevator.
1134 old_elevator = q->elevator;
1137 * attach and start new elevator
1139 elevator_attach(q, e, data);
1141 spin_unlock_irq(q->queue_lock);
1143 __elv_unregister_queue(old_elevator);
1145 if (elv_register_queue(q))
1146 goto fail_register;
1149 * finally exit old elevator and turn off BYPASS.
1151 elevator_exit(old_elevator);
1152 spin_lock_irq(q->queue_lock);
1153 elv_quiesce_end(q);
1154 spin_unlock_irq(q->queue_lock);
1156 blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name);
1158 return 1;
1160 fail_register:
1162 * switch failed, exit the new io scheduler and reattach the old
1163 * one again (along with re-adding the sysfs dir)
1165 elevator_exit(e);
1166 q->elevator = old_elevator;
1167 elv_register_queue(q);
1169 spin_lock_irq(q->queue_lock);
1170 queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q);
1171 spin_unlock_irq(q->queue_lock);
1173 return 0;
1176 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1177 size_t count)
1179 char elevator_name[ELV_NAME_MAX];
1180 struct elevator_type *e;
1182 strlcpy(elevator_name, name, sizeof(elevator_name));
1183 strstrip(elevator_name);
1185 e = elevator_get(elevator_name);
1186 if (!e) {
1187 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1188 return -EINVAL;
1191 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1192 elevator_put(e);
1193 return count;
1196 if (!elevator_switch(q, e))
1197 printk(KERN_ERR "elevator: switch to %s failed\n",
1198 elevator_name);
1199 return count;
1202 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1204 struct elevator_queue *e = q->elevator;
1205 struct elevator_type *elv = e->elevator_type;
1206 struct elevator_type *__e;
1207 int len = 0;
1209 spin_lock(&elv_list_lock);
1210 list_for_each_entry(__e, &elv_list, list) {
1211 if (!strcmp(elv->elevator_name, __e->elevator_name))
1212 len += sprintf(name+len, "[%s] ", elv->elevator_name);
1213 else
1214 len += sprintf(name+len, "%s ", __e->elevator_name);
1216 spin_unlock(&elv_list_lock);
1218 len += sprintf(len+name, "\n");
1219 return len;
1222 struct request *elv_rb_former_request(struct request_queue *q,
1223 struct request *rq)
1225 struct rb_node *rbprev = rb_prev(&rq->rb_node);
1227 if (rbprev)
1228 return rb_entry_rq(rbprev);
1230 return NULL;
1232 EXPORT_SYMBOL(elv_rb_former_request);
1234 struct request *elv_rb_latter_request(struct request_queue *q,
1235 struct request *rq)
1237 struct rb_node *rbnext = rb_next(&rq->rb_node);
1239 if (rbnext)
1240 return rb_entry_rq(rbnext);
1242 return NULL;
1244 EXPORT_SYMBOL(elv_rb_latter_request);