xref: /openbmc/linux/block/elevator.c (revision f42b3800)
1 /*
2  *  Block device elevator/IO-scheduler.
3  *
4  *  Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5  *
6  * 30042000 Jens Axboe <axboe@kernel.dk> :
7  *
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
14  *
15  * 20082000 Dave Jones <davej@suse.de> :
16  * Removed tests for max-bomb-segments, which was breaking elvtune
17  *  when run without -bN
18  *
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
23  *
24  */
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 
38 #include <asm/uaccess.h>
39 
40 static DEFINE_SPINLOCK(elv_list_lock);
41 static LIST_HEAD(elv_list);
42 
43 /*
44  * Merge hash stuff.
45  */
46 static const int elv_hash_shift = 6;
47 #define ELV_HASH_BLOCK(sec)	((sec) >> 3)
48 #define ELV_HASH_FN(sec)	\
49 		(hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift))
50 #define ELV_HASH_ENTRIES	(1 << elv_hash_shift)
51 #define rq_hash_key(rq)		((rq)->sector + (rq)->nr_sectors)
52 #define ELV_ON_HASH(rq)		(!hlist_unhashed(&(rq)->hash))
53 
54 /*
55  * Query io scheduler to see if the current process issuing bio may be
56  * merged with rq.
57  */
58 static int elv_iosched_allow_merge(struct request *rq, struct bio *bio)
59 {
60 	struct request_queue *q = rq->q;
61 	elevator_t *e = q->elevator;
62 
63 	if (e->ops->elevator_allow_merge_fn)
64 		return e->ops->elevator_allow_merge_fn(q, rq, bio);
65 
66 	return 1;
67 }
68 
69 /*
70  * can we safely merge with this request?
71  */
72 inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
73 {
74 	if (!rq_mergeable(rq))
75 		return 0;
76 
77 	/*
78 	 * different data direction or already started, don't merge
79 	 */
80 	if (bio_data_dir(bio) != rq_data_dir(rq))
81 		return 0;
82 
83 	/*
84 	 * must be same device and not a special request
85 	 */
86 	if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special)
87 		return 0;
88 
89 	if (!elv_iosched_allow_merge(rq, bio))
90 		return 0;
91 
92 	return 1;
93 }
94 EXPORT_SYMBOL(elv_rq_merge_ok);
95 
96 static inline int elv_try_merge(struct request *__rq, struct bio *bio)
97 {
98 	int ret = ELEVATOR_NO_MERGE;
99 
100 	/*
101 	 * we can merge and sequence is ok, check if it's possible
102 	 */
103 	if (elv_rq_merge_ok(__rq, bio)) {
104 		if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
105 			ret = ELEVATOR_BACK_MERGE;
106 		else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
107 			ret = ELEVATOR_FRONT_MERGE;
108 	}
109 
110 	return ret;
111 }
112 
113 static struct elevator_type *elevator_find(const char *name)
114 {
115 	struct elevator_type *e;
116 
117 	list_for_each_entry(e, &elv_list, list) {
118 		if (!strcmp(e->elevator_name, name))
119 			return e;
120 	}
121 
122 	return NULL;
123 }
124 
125 static void elevator_put(struct elevator_type *e)
126 {
127 	module_put(e->elevator_owner);
128 }
129 
130 static struct elevator_type *elevator_get(const char *name)
131 {
132 	struct elevator_type *e;
133 
134 	spin_lock(&elv_list_lock);
135 
136 	e = elevator_find(name);
137 	if (!e) {
138 		char elv[ELV_NAME_MAX + strlen("-iosched")];
139 
140 		spin_unlock(&elv_list_lock);
141 
142 		if (!strcmp(name, "anticipatory"))
143 			sprintf(elv, "as-iosched");
144 		else
145 			sprintf(elv, "%s-iosched", name);
146 
147 		request_module(elv);
148 		spin_lock(&elv_list_lock);
149 		e = elevator_find(name);
150 	}
151 
152 	if (e && !try_module_get(e->elevator_owner))
153 		e = NULL;
154 
155 	spin_unlock(&elv_list_lock);
156 
157 	return e;
158 }
159 
160 static void *elevator_init_queue(struct request_queue *q,
161 				 struct elevator_queue *eq)
162 {
163 	return eq->ops->elevator_init_fn(q);
164 }
165 
166 static void elevator_attach(struct request_queue *q, struct elevator_queue *eq,
167 			   void *data)
168 {
169 	q->elevator = eq;
170 	eq->elevator_data = data;
171 }
172 
173 static char chosen_elevator[16];
174 
175 static int __init elevator_setup(char *str)
176 {
177 	/*
178 	 * Be backwards-compatible with previous kernels, so users
179 	 * won't get the wrong elevator.
180 	 */
181 	if (!strcmp(str, "as"))
182 		strcpy(chosen_elevator, "anticipatory");
183 	else
184 		strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
185 	return 1;
186 }
187 
188 __setup("elevator=", elevator_setup);
189 
190 static struct kobj_type elv_ktype;
191 
192 static elevator_t *elevator_alloc(struct request_queue *q,
193 				  struct elevator_type *e)
194 {
195 	elevator_t *eq;
196 	int i;
197 
198 	eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL | __GFP_ZERO, q->node);
199 	if (unlikely(!eq))
200 		goto err;
201 
202 	eq->ops = &e->ops;
203 	eq->elevator_type = e;
204 	kobject_init(&eq->kobj, &elv_ktype);
205 	mutex_init(&eq->sysfs_lock);
206 
207 	eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES,
208 					GFP_KERNEL, q->node);
209 	if (!eq->hash)
210 		goto err;
211 
212 	for (i = 0; i < ELV_HASH_ENTRIES; i++)
213 		INIT_HLIST_HEAD(&eq->hash[i]);
214 
215 	return eq;
216 err:
217 	kfree(eq);
218 	elevator_put(e);
219 	return NULL;
220 }
221 
222 static void elevator_release(struct kobject *kobj)
223 {
224 	elevator_t *e = container_of(kobj, elevator_t, kobj);
225 
226 	elevator_put(e->elevator_type);
227 	kfree(e->hash);
228 	kfree(e);
229 }
230 
231 int elevator_init(struct request_queue *q, char *name)
232 {
233 	struct elevator_type *e = NULL;
234 	struct elevator_queue *eq;
235 	int ret = 0;
236 	void *data;
237 
238 	INIT_LIST_HEAD(&q->queue_head);
239 	q->last_merge = NULL;
240 	q->end_sector = 0;
241 	q->boundary_rq = NULL;
242 
243 	if (name) {
244 		e = elevator_get(name);
245 		if (!e)
246 			return -EINVAL;
247 	}
248 
249 	if (!e && *chosen_elevator) {
250 		e = elevator_get(chosen_elevator);
251 		if (!e)
252 			printk(KERN_ERR "I/O scheduler %s not found\n",
253 							chosen_elevator);
254 	}
255 
256 	if (!e) {
257 		e = elevator_get(CONFIG_DEFAULT_IOSCHED);
258 		if (!e) {
259 			printk(KERN_ERR
260 				"Default I/O scheduler not found. " \
261 				"Using noop.\n");
262 			e = elevator_get("noop");
263 		}
264 	}
265 
266 	eq = elevator_alloc(q, e);
267 	if (!eq)
268 		return -ENOMEM;
269 
270 	data = elevator_init_queue(q, eq);
271 	if (!data) {
272 		kobject_put(&eq->kobj);
273 		return -ENOMEM;
274 	}
275 
276 	elevator_attach(q, eq, data);
277 	return ret;
278 }
279 EXPORT_SYMBOL(elevator_init);
280 
281 void elevator_exit(elevator_t *e)
282 {
283 	mutex_lock(&e->sysfs_lock);
284 	if (e->ops->elevator_exit_fn)
285 		e->ops->elevator_exit_fn(e);
286 	e->ops = NULL;
287 	mutex_unlock(&e->sysfs_lock);
288 
289 	kobject_put(&e->kobj);
290 }
291 EXPORT_SYMBOL(elevator_exit);
292 
293 static void elv_activate_rq(struct request_queue *q, struct request *rq)
294 {
295 	elevator_t *e = q->elevator;
296 
297 	if (e->ops->elevator_activate_req_fn)
298 		e->ops->elevator_activate_req_fn(q, rq);
299 }
300 
301 static void elv_deactivate_rq(struct request_queue *q, struct request *rq)
302 {
303 	elevator_t *e = q->elevator;
304 
305 	if (e->ops->elevator_deactivate_req_fn)
306 		e->ops->elevator_deactivate_req_fn(q, rq);
307 }
308 
309 static inline void __elv_rqhash_del(struct request *rq)
310 {
311 	hlist_del_init(&rq->hash);
312 }
313 
314 static void elv_rqhash_del(struct request_queue *q, struct request *rq)
315 {
316 	if (ELV_ON_HASH(rq))
317 		__elv_rqhash_del(rq);
318 }
319 
320 static void elv_rqhash_add(struct request_queue *q, struct request *rq)
321 {
322 	elevator_t *e = q->elevator;
323 
324 	BUG_ON(ELV_ON_HASH(rq));
325 	hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]);
326 }
327 
328 static void elv_rqhash_reposition(struct request_queue *q, struct request *rq)
329 {
330 	__elv_rqhash_del(rq);
331 	elv_rqhash_add(q, rq);
332 }
333 
334 static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset)
335 {
336 	elevator_t *e = q->elevator;
337 	struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)];
338 	struct hlist_node *entry, *next;
339 	struct request *rq;
340 
341 	hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) {
342 		BUG_ON(!ELV_ON_HASH(rq));
343 
344 		if (unlikely(!rq_mergeable(rq))) {
345 			__elv_rqhash_del(rq);
346 			continue;
347 		}
348 
349 		if (rq_hash_key(rq) == offset)
350 			return rq;
351 	}
352 
353 	return NULL;
354 }
355 
356 /*
357  * RB-tree support functions for inserting/lookup/removal of requests
358  * in a sorted RB tree.
359  */
360 struct request *elv_rb_add(struct rb_root *root, struct request *rq)
361 {
362 	struct rb_node **p = &root->rb_node;
363 	struct rb_node *parent = NULL;
364 	struct request *__rq;
365 
366 	while (*p) {
367 		parent = *p;
368 		__rq = rb_entry(parent, struct request, rb_node);
369 
370 		if (rq->sector < __rq->sector)
371 			p = &(*p)->rb_left;
372 		else if (rq->sector > __rq->sector)
373 			p = &(*p)->rb_right;
374 		else
375 			return __rq;
376 	}
377 
378 	rb_link_node(&rq->rb_node, parent, p);
379 	rb_insert_color(&rq->rb_node, root);
380 	return NULL;
381 }
382 EXPORT_SYMBOL(elv_rb_add);
383 
384 void elv_rb_del(struct rb_root *root, struct request *rq)
385 {
386 	BUG_ON(RB_EMPTY_NODE(&rq->rb_node));
387 	rb_erase(&rq->rb_node, root);
388 	RB_CLEAR_NODE(&rq->rb_node);
389 }
390 EXPORT_SYMBOL(elv_rb_del);
391 
392 struct request *elv_rb_find(struct rb_root *root, sector_t sector)
393 {
394 	struct rb_node *n = root->rb_node;
395 	struct request *rq;
396 
397 	while (n) {
398 		rq = rb_entry(n, struct request, rb_node);
399 
400 		if (sector < rq->sector)
401 			n = n->rb_left;
402 		else if (sector > rq->sector)
403 			n = n->rb_right;
404 		else
405 			return rq;
406 	}
407 
408 	return NULL;
409 }
410 EXPORT_SYMBOL(elv_rb_find);
411 
412 /*
413  * Insert rq into dispatch queue of q.  Queue lock must be held on
414  * entry.  rq is sort instead into the dispatch queue. To be used by
415  * specific elevators.
416  */
417 void elv_dispatch_sort(struct request_queue *q, struct request *rq)
418 {
419 	sector_t boundary;
420 	struct list_head *entry;
421 	int stop_flags;
422 
423 	if (q->last_merge == rq)
424 		q->last_merge = NULL;
425 
426 	elv_rqhash_del(q, rq);
427 
428 	q->nr_sorted--;
429 
430 	boundary = q->end_sector;
431 	stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED;
432 	list_for_each_prev(entry, &q->queue_head) {
433 		struct request *pos = list_entry_rq(entry);
434 
435 		if (rq_data_dir(rq) != rq_data_dir(pos))
436 			break;
437 		if (pos->cmd_flags & stop_flags)
438 			break;
439 		if (rq->sector >= boundary) {
440 			if (pos->sector < boundary)
441 				continue;
442 		} else {
443 			if (pos->sector >= boundary)
444 				break;
445 		}
446 		if (rq->sector >= pos->sector)
447 			break;
448 	}
449 
450 	list_add(&rq->queuelist, entry);
451 }
452 EXPORT_SYMBOL(elv_dispatch_sort);
453 
454 /*
455  * Insert rq into dispatch queue of q.  Queue lock must be held on
456  * entry.  rq is added to the back of the dispatch queue. To be used by
457  * specific elevators.
458  */
459 void elv_dispatch_add_tail(struct request_queue *q, struct request *rq)
460 {
461 	if (q->last_merge == rq)
462 		q->last_merge = NULL;
463 
464 	elv_rqhash_del(q, rq);
465 
466 	q->nr_sorted--;
467 
468 	q->end_sector = rq_end_sector(rq);
469 	q->boundary_rq = rq;
470 	list_add_tail(&rq->queuelist, &q->queue_head);
471 }
472 EXPORT_SYMBOL(elv_dispatch_add_tail);
473 
474 int elv_merge(struct request_queue *q, struct request **req, struct bio *bio)
475 {
476 	elevator_t *e = q->elevator;
477 	struct request *__rq;
478 	int ret;
479 
480 	/*
481 	 * First try one-hit cache.
482 	 */
483 	if (q->last_merge) {
484 		ret = elv_try_merge(q->last_merge, bio);
485 		if (ret != ELEVATOR_NO_MERGE) {
486 			*req = q->last_merge;
487 			return ret;
488 		}
489 	}
490 
491 	/*
492 	 * See if our hash lookup can find a potential backmerge.
493 	 */
494 	__rq = elv_rqhash_find(q, bio->bi_sector);
495 	if (__rq && elv_rq_merge_ok(__rq, bio)) {
496 		*req = __rq;
497 		return ELEVATOR_BACK_MERGE;
498 	}
499 
500 	if (e->ops->elevator_merge_fn)
501 		return e->ops->elevator_merge_fn(q, req, bio);
502 
503 	return ELEVATOR_NO_MERGE;
504 }
505 
506 void elv_merged_request(struct request_queue *q, struct request *rq, int type)
507 {
508 	elevator_t *e = q->elevator;
509 
510 	if (e->ops->elevator_merged_fn)
511 		e->ops->elevator_merged_fn(q, rq, type);
512 
513 	if (type == ELEVATOR_BACK_MERGE)
514 		elv_rqhash_reposition(q, rq);
515 
516 	q->last_merge = rq;
517 }
518 
519 void elv_merge_requests(struct request_queue *q, struct request *rq,
520 			     struct request *next)
521 {
522 	elevator_t *e = q->elevator;
523 
524 	if (e->ops->elevator_merge_req_fn)
525 		e->ops->elevator_merge_req_fn(q, rq, next);
526 
527 	elv_rqhash_reposition(q, rq);
528 	elv_rqhash_del(q, next);
529 
530 	q->nr_sorted--;
531 	q->last_merge = rq;
532 }
533 
534 void elv_requeue_request(struct request_queue *q, struct request *rq)
535 {
536 	/*
537 	 * it already went through dequeue, we need to decrement the
538 	 * in_flight count again
539 	 */
540 	if (blk_account_rq(rq)) {
541 		q->in_flight--;
542 		if (blk_sorted_rq(rq))
543 			elv_deactivate_rq(q, rq);
544 	}
545 
546 	rq->cmd_flags &= ~REQ_STARTED;
547 
548 	elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE);
549 }
550 
551 static void elv_drain_elevator(struct request_queue *q)
552 {
553 	static int printed;
554 	while (q->elevator->ops->elevator_dispatch_fn(q, 1))
555 		;
556 	if (q->nr_sorted == 0)
557 		return;
558 	if (printed++ < 10) {
559 		printk(KERN_ERR "%s: forced dispatching is broken "
560 		       "(nr_sorted=%u), please report this\n",
561 		       q->elevator->elevator_type->elevator_name, q->nr_sorted);
562 	}
563 }
564 
565 void elv_insert(struct request_queue *q, struct request *rq, int where)
566 {
567 	struct list_head *pos;
568 	unsigned ordseq;
569 	int unplug_it = 1;
570 
571 	blk_add_trace_rq(q, rq, BLK_TA_INSERT);
572 
573 	rq->q = q;
574 
575 	switch (where) {
576 	case ELEVATOR_INSERT_FRONT:
577 		rq->cmd_flags |= REQ_SOFTBARRIER;
578 
579 		list_add(&rq->queuelist, &q->queue_head);
580 		break;
581 
582 	case ELEVATOR_INSERT_BACK:
583 		rq->cmd_flags |= REQ_SOFTBARRIER;
584 		elv_drain_elevator(q);
585 		list_add_tail(&rq->queuelist, &q->queue_head);
586 		/*
587 		 * We kick the queue here for the following reasons.
588 		 * - The elevator might have returned NULL previously
589 		 *   to delay requests and returned them now.  As the
590 		 *   queue wasn't empty before this request, ll_rw_blk
591 		 *   won't run the queue on return, resulting in hang.
592 		 * - Usually, back inserted requests won't be merged
593 		 *   with anything.  There's no point in delaying queue
594 		 *   processing.
595 		 */
596 		blk_remove_plug(q);
597 		q->request_fn(q);
598 		break;
599 
600 	case ELEVATOR_INSERT_SORT:
601 		BUG_ON(!blk_fs_request(rq));
602 		rq->cmd_flags |= REQ_SORTED;
603 		q->nr_sorted++;
604 		if (rq_mergeable(rq)) {
605 			elv_rqhash_add(q, rq);
606 			if (!q->last_merge)
607 				q->last_merge = rq;
608 		}
609 
610 		/*
611 		 * Some ioscheds (cfq) run q->request_fn directly, so
612 		 * rq cannot be accessed after calling
613 		 * elevator_add_req_fn.
614 		 */
615 		q->elevator->ops->elevator_add_req_fn(q, rq);
616 		break;
617 
618 	case ELEVATOR_INSERT_REQUEUE:
619 		/*
620 		 * If ordered flush isn't in progress, we do front
621 		 * insertion; otherwise, requests should be requeued
622 		 * in ordseq order.
623 		 */
624 		rq->cmd_flags |= REQ_SOFTBARRIER;
625 
626 		/*
627 		 * Most requeues happen because of a busy condition,
628 		 * don't force unplug of the queue for that case.
629 		 */
630 		unplug_it = 0;
631 
632 		if (q->ordseq == 0) {
633 			list_add(&rq->queuelist, &q->queue_head);
634 			break;
635 		}
636 
637 		ordseq = blk_ordered_req_seq(rq);
638 
639 		list_for_each(pos, &q->queue_head) {
640 			struct request *pos_rq = list_entry_rq(pos);
641 			if (ordseq <= blk_ordered_req_seq(pos_rq))
642 				break;
643 		}
644 
645 		list_add_tail(&rq->queuelist, pos);
646 		break;
647 
648 	default:
649 		printk(KERN_ERR "%s: bad insertion point %d\n",
650 		       __FUNCTION__, where);
651 		BUG();
652 	}
653 
654 	if (unplug_it && blk_queue_plugged(q)) {
655 		int nrq = q->rq.count[READ] + q->rq.count[WRITE]
656 			- q->in_flight;
657 
658 		if (nrq >= q->unplug_thresh)
659 			__generic_unplug_device(q);
660 	}
661 }
662 
663 void __elv_add_request(struct request_queue *q, struct request *rq, int where,
664 		       int plug)
665 {
666 	if (q->ordcolor)
667 		rq->cmd_flags |= REQ_ORDERED_COLOR;
668 
669 	if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
670 		/*
671 		 * toggle ordered color
672 		 */
673 		if (blk_barrier_rq(rq))
674 			q->ordcolor ^= 1;
675 
676 		/*
677 		 * barriers implicitly indicate back insertion
678 		 */
679 		if (where == ELEVATOR_INSERT_SORT)
680 			where = ELEVATOR_INSERT_BACK;
681 
682 		/*
683 		 * this request is scheduling boundary, update
684 		 * end_sector
685 		 */
686 		if (blk_fs_request(rq)) {
687 			q->end_sector = rq_end_sector(rq);
688 			q->boundary_rq = rq;
689 		}
690 	} else if (!(rq->cmd_flags & REQ_ELVPRIV) &&
691 		    where == ELEVATOR_INSERT_SORT)
692 		where = ELEVATOR_INSERT_BACK;
693 
694 	if (plug)
695 		blk_plug_device(q);
696 
697 	elv_insert(q, rq, where);
698 }
699 EXPORT_SYMBOL(__elv_add_request);
700 
701 void elv_add_request(struct request_queue *q, struct request *rq, int where,
702 		     int plug)
703 {
704 	unsigned long flags;
705 
706 	spin_lock_irqsave(q->queue_lock, flags);
707 	__elv_add_request(q, rq, where, plug);
708 	spin_unlock_irqrestore(q->queue_lock, flags);
709 }
710 EXPORT_SYMBOL(elv_add_request);
711 
712 static inline struct request *__elv_next_request(struct request_queue *q)
713 {
714 	struct request *rq;
715 
716 	while (1) {
717 		while (!list_empty(&q->queue_head)) {
718 			rq = list_entry_rq(q->queue_head.next);
719 			if (blk_do_ordered(q, &rq))
720 				return rq;
721 		}
722 
723 		if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
724 			return NULL;
725 	}
726 }
727 
728 struct request *elv_next_request(struct request_queue *q)
729 {
730 	struct request *rq;
731 	int ret;
732 
733 	while ((rq = __elv_next_request(q)) != NULL) {
734 		/*
735 		 * Kill the empty barrier place holder, the driver must
736 		 * not ever see it.
737 		 */
738 		if (blk_empty_barrier(rq)) {
739 			end_queued_request(rq, 1);
740 			continue;
741 		}
742 		if (!(rq->cmd_flags & REQ_STARTED)) {
743 			/*
744 			 * This is the first time the device driver
745 			 * sees this request (possibly after
746 			 * requeueing).  Notify IO scheduler.
747 			 */
748 			if (blk_sorted_rq(rq))
749 				elv_activate_rq(q, rq);
750 
751 			/*
752 			 * just mark as started even if we don't start
753 			 * it, a request that has been delayed should
754 			 * not be passed by new incoming requests
755 			 */
756 			rq->cmd_flags |= REQ_STARTED;
757 			blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
758 		}
759 
760 		if (!q->boundary_rq || q->boundary_rq == rq) {
761 			q->end_sector = rq_end_sector(rq);
762 			q->boundary_rq = NULL;
763 		}
764 
765 		if (rq->cmd_flags & REQ_DONTPREP)
766 			break;
767 
768 		if (q->dma_drain_size && rq->data_len) {
769 			/*
770 			 * make sure space for the drain appears we
771 			 * know we can do this because max_hw_segments
772 			 * has been adjusted to be one fewer than the
773 			 * device can handle
774 			 */
775 			rq->nr_phys_segments++;
776 			rq->nr_hw_segments++;
777 		}
778 
779 		if (!q->prep_rq_fn)
780 			break;
781 
782 		ret = q->prep_rq_fn(q, rq);
783 		if (ret == BLKPREP_OK) {
784 			break;
785 		} else if (ret == BLKPREP_DEFER) {
786 			/*
787 			 * the request may have been (partially) prepped.
788 			 * we need to keep this request in the front to
789 			 * avoid resource deadlock.  REQ_STARTED will
790 			 * prevent other fs requests from passing this one.
791 			 */
792 			if (q->dma_drain_size && rq->data_len &&
793 			    !(rq->cmd_flags & REQ_DONTPREP)) {
794 				/*
795 				 * remove the space for the drain we added
796 				 * so that we don't add it again
797 				 */
798 				--rq->nr_phys_segments;
799 				--rq->nr_hw_segments;
800 			}
801 
802 			rq = NULL;
803 			break;
804 		} else if (ret == BLKPREP_KILL) {
805 			rq->cmd_flags |= REQ_QUIET;
806 			end_queued_request(rq, 0);
807 		} else {
808 			printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
809 								ret);
810 			break;
811 		}
812 	}
813 
814 	return rq;
815 }
816 EXPORT_SYMBOL(elv_next_request);
817 
818 void elv_dequeue_request(struct request_queue *q, struct request *rq)
819 {
820 	BUG_ON(list_empty(&rq->queuelist));
821 	BUG_ON(ELV_ON_HASH(rq));
822 
823 	list_del_init(&rq->queuelist);
824 
825 	/*
826 	 * the time frame between a request being removed from the lists
827 	 * and to it is freed is accounted as io that is in progress at
828 	 * the driver side.
829 	 */
830 	if (blk_account_rq(rq))
831 		q->in_flight++;
832 }
833 EXPORT_SYMBOL(elv_dequeue_request);
834 
835 int elv_queue_empty(struct request_queue *q)
836 {
837 	elevator_t *e = q->elevator;
838 
839 	if (!list_empty(&q->queue_head))
840 		return 0;
841 
842 	if (e->ops->elevator_queue_empty_fn)
843 		return e->ops->elevator_queue_empty_fn(q);
844 
845 	return 1;
846 }
847 EXPORT_SYMBOL(elv_queue_empty);
848 
849 struct request *elv_latter_request(struct request_queue *q, struct request *rq)
850 {
851 	elevator_t *e = q->elevator;
852 
853 	if (e->ops->elevator_latter_req_fn)
854 		return e->ops->elevator_latter_req_fn(q, rq);
855 	return NULL;
856 }
857 
858 struct request *elv_former_request(struct request_queue *q, struct request *rq)
859 {
860 	elevator_t *e = q->elevator;
861 
862 	if (e->ops->elevator_former_req_fn)
863 		return e->ops->elevator_former_req_fn(q, rq);
864 	return NULL;
865 }
866 
867 int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask)
868 {
869 	elevator_t *e = q->elevator;
870 
871 	if (e->ops->elevator_set_req_fn)
872 		return e->ops->elevator_set_req_fn(q, rq, gfp_mask);
873 
874 	rq->elevator_private = NULL;
875 	return 0;
876 }
877 
878 void elv_put_request(struct request_queue *q, struct request *rq)
879 {
880 	elevator_t *e = q->elevator;
881 
882 	if (e->ops->elevator_put_req_fn)
883 		e->ops->elevator_put_req_fn(rq);
884 }
885 
886 int elv_may_queue(struct request_queue *q, int rw)
887 {
888 	elevator_t *e = q->elevator;
889 
890 	if (e->ops->elevator_may_queue_fn)
891 		return e->ops->elevator_may_queue_fn(q, rw);
892 
893 	return ELV_MQUEUE_MAY;
894 }
895 
896 void elv_completed_request(struct request_queue *q, struct request *rq)
897 {
898 	elevator_t *e = q->elevator;
899 
900 	/*
901 	 * request is released from the driver, io must be done
902 	 */
903 	if (blk_account_rq(rq)) {
904 		q->in_flight--;
905 		if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
906 			e->ops->elevator_completed_req_fn(q, rq);
907 	}
908 
909 	/*
910 	 * Check if the queue is waiting for fs requests to be
911 	 * drained for flush sequence.
912 	 */
913 	if (unlikely(q->ordseq)) {
914 		struct request *first_rq = list_entry_rq(q->queue_head.next);
915 		if (q->in_flight == 0 &&
916 		    blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
917 		    blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
918 			blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
919 			q->request_fn(q);
920 		}
921 	}
922 }
923 
924 #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr)
925 
926 static ssize_t
927 elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
928 {
929 	elevator_t *e = container_of(kobj, elevator_t, kobj);
930 	struct elv_fs_entry *entry = to_elv(attr);
931 	ssize_t error;
932 
933 	if (!entry->show)
934 		return -EIO;
935 
936 	mutex_lock(&e->sysfs_lock);
937 	error = e->ops ? entry->show(e, page) : -ENOENT;
938 	mutex_unlock(&e->sysfs_lock);
939 	return error;
940 }
941 
942 static ssize_t
943 elv_attr_store(struct kobject *kobj, struct attribute *attr,
944 	       const char *page, size_t length)
945 {
946 	elevator_t *e = container_of(kobj, elevator_t, kobj);
947 	struct elv_fs_entry *entry = to_elv(attr);
948 	ssize_t error;
949 
950 	if (!entry->store)
951 		return -EIO;
952 
953 	mutex_lock(&e->sysfs_lock);
954 	error = e->ops ? entry->store(e, page, length) : -ENOENT;
955 	mutex_unlock(&e->sysfs_lock);
956 	return error;
957 }
958 
959 static struct sysfs_ops elv_sysfs_ops = {
960 	.show	= elv_attr_show,
961 	.store	= elv_attr_store,
962 };
963 
964 static struct kobj_type elv_ktype = {
965 	.sysfs_ops	= &elv_sysfs_ops,
966 	.release	= elevator_release,
967 };
968 
969 int elv_register_queue(struct request_queue *q)
970 {
971 	elevator_t *e = q->elevator;
972 	int error;
973 
974 	error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched");
975 	if (!error) {
976 		struct elv_fs_entry *attr = e->elevator_type->elevator_attrs;
977 		if (attr) {
978 			while (attr->attr.name) {
979 				if (sysfs_create_file(&e->kobj, &attr->attr))
980 					break;
981 				attr++;
982 			}
983 		}
984 		kobject_uevent(&e->kobj, KOBJ_ADD);
985 	}
986 	return error;
987 }
988 
989 static void __elv_unregister_queue(elevator_t *e)
990 {
991 	kobject_uevent(&e->kobj, KOBJ_REMOVE);
992 	kobject_del(&e->kobj);
993 }
994 
995 void elv_unregister_queue(struct request_queue *q)
996 {
997 	if (q)
998 		__elv_unregister_queue(q->elevator);
999 }
1000 
1001 void elv_register(struct elevator_type *e)
1002 {
1003 	char *def = "";
1004 
1005 	spin_lock(&elv_list_lock);
1006 	BUG_ON(elevator_find(e->elevator_name));
1007 	list_add_tail(&e->list, &elv_list);
1008 	spin_unlock(&elv_list_lock);
1009 
1010 	if (!strcmp(e->elevator_name, chosen_elevator) ||
1011 			(!*chosen_elevator &&
1012 			 !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED)))
1013 				def = " (default)";
1014 
1015 	printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name,
1016 								def);
1017 }
1018 EXPORT_SYMBOL_GPL(elv_register);
1019 
1020 void elv_unregister(struct elevator_type *e)
1021 {
1022 	struct task_struct *g, *p;
1023 
1024 	/*
1025 	 * Iterate every thread in the process to remove the io contexts.
1026 	 */
1027 	if (e->ops.trim) {
1028 		read_lock(&tasklist_lock);
1029 		do_each_thread(g, p) {
1030 			task_lock(p);
1031 			if (p->io_context)
1032 				e->ops.trim(p->io_context);
1033 			task_unlock(p);
1034 		} while_each_thread(g, p);
1035 		read_unlock(&tasklist_lock);
1036 	}
1037 
1038 	spin_lock(&elv_list_lock);
1039 	list_del_init(&e->list);
1040 	spin_unlock(&elv_list_lock);
1041 }
1042 EXPORT_SYMBOL_GPL(elv_unregister);
1043 
1044 /*
1045  * switch to new_e io scheduler. be careful not to introduce deadlocks -
1046  * we don't free the old io scheduler, before we have allocated what we
1047  * need for the new one. this way we have a chance of going back to the old
1048  * one, if the new one fails init for some reason.
1049  */
1050 static int elevator_switch(struct request_queue *q, struct elevator_type *new_e)
1051 {
1052 	elevator_t *old_elevator, *e;
1053 	void *data;
1054 
1055 	/*
1056 	 * Allocate new elevator
1057 	 */
1058 	e = elevator_alloc(q, new_e);
1059 	if (!e)
1060 		return 0;
1061 
1062 	data = elevator_init_queue(q, e);
1063 	if (!data) {
1064 		kobject_put(&e->kobj);
1065 		return 0;
1066 	}
1067 
1068 	/*
1069 	 * Turn on BYPASS and drain all requests w/ elevator private data
1070 	 */
1071 	spin_lock_irq(q->queue_lock);
1072 
1073 	set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1074 
1075 	elv_drain_elevator(q);
1076 
1077 	while (q->rq.elvpriv) {
1078 		blk_remove_plug(q);
1079 		q->request_fn(q);
1080 		spin_unlock_irq(q->queue_lock);
1081 		msleep(10);
1082 		spin_lock_irq(q->queue_lock);
1083 		elv_drain_elevator(q);
1084 	}
1085 
1086 	/*
1087 	 * Remember old elevator.
1088 	 */
1089 	old_elevator = q->elevator;
1090 
1091 	/*
1092 	 * attach and start new elevator
1093 	 */
1094 	elevator_attach(q, e, data);
1095 
1096 	spin_unlock_irq(q->queue_lock);
1097 
1098 	__elv_unregister_queue(old_elevator);
1099 
1100 	if (elv_register_queue(q))
1101 		goto fail_register;
1102 
1103 	/*
1104 	 * finally exit old elevator and turn off BYPASS.
1105 	 */
1106 	elevator_exit(old_elevator);
1107 	clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1108 	return 1;
1109 
1110 fail_register:
1111 	/*
1112 	 * switch failed, exit the new io scheduler and reattach the old
1113 	 * one again (along with re-adding the sysfs dir)
1114 	 */
1115 	elevator_exit(e);
1116 	q->elevator = old_elevator;
1117 	elv_register_queue(q);
1118 	clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
1119 	return 0;
1120 }
1121 
1122 ssize_t elv_iosched_store(struct request_queue *q, const char *name,
1123 			  size_t count)
1124 {
1125 	char elevator_name[ELV_NAME_MAX];
1126 	size_t len;
1127 	struct elevator_type *e;
1128 
1129 	elevator_name[sizeof(elevator_name) - 1] = '\0';
1130 	strncpy(elevator_name, name, sizeof(elevator_name) - 1);
1131 	len = strlen(elevator_name);
1132 
1133 	if (len && elevator_name[len - 1] == '\n')
1134 		elevator_name[len - 1] = '\0';
1135 
1136 	e = elevator_get(elevator_name);
1137 	if (!e) {
1138 		printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
1139 		return -EINVAL;
1140 	}
1141 
1142 	if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
1143 		elevator_put(e);
1144 		return count;
1145 	}
1146 
1147 	if (!elevator_switch(q, e))
1148 		printk(KERN_ERR "elevator: switch to %s failed\n",
1149 							elevator_name);
1150 	return count;
1151 }
1152 
1153 ssize_t elv_iosched_show(struct request_queue *q, char *name)
1154 {
1155 	elevator_t *e = q->elevator;
1156 	struct elevator_type *elv = e->elevator_type;
1157 	struct elevator_type *__e;
1158 	int len = 0;
1159 
1160 	spin_lock(&elv_list_lock);
1161 	list_for_each_entry(__e, &elv_list, list) {
1162 		if (!strcmp(elv->elevator_name, __e->elevator_name))
1163 			len += sprintf(name+len, "[%s] ", elv->elevator_name);
1164 		else
1165 			len += sprintf(name+len, "%s ", __e->elevator_name);
1166 	}
1167 	spin_unlock(&elv_list_lock);
1168 
1169 	len += sprintf(len+name, "\n");
1170 	return len;
1171 }
1172 
1173 struct request *elv_rb_former_request(struct request_queue *q,
1174 				      struct request *rq)
1175 {
1176 	struct rb_node *rbprev = rb_prev(&rq->rb_node);
1177 
1178 	if (rbprev)
1179 		return rb_entry_rq(rbprev);
1180 
1181 	return NULL;
1182 }
1183 EXPORT_SYMBOL(elv_rb_former_request);
1184 
1185 struct request *elv_rb_latter_request(struct request_queue *q,
1186 				      struct request *rq)
1187 {
1188 	struct rb_node *rbnext = rb_next(&rq->rb_node);
1189 
1190 	if (rbnext)
1191 		return rb_entry_rq(rbnext);
1192 
1193 	return NULL;
1194 }
1195 EXPORT_SYMBOL(elv_rb_latter_request);
1196