blob: 39dcccc82ada5b71e3105a8b7e9d8389fc9000b6 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002 * Block device elevator/IO-scheduler.
3 *
4 * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE
5 *
6 * 30042000 Jens Axboe <axboe@suse.de> :
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/config.h>
31#include <linux/module.h>
32#include <linux/slab.h>
33#include <linux/init.h>
34#include <linux/compiler.h>
Tejun Heocb98fc82005-10-28 08:29:39 +020035#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include <asm/uaccess.h>
38
39static DEFINE_SPINLOCK(elv_list_lock);
40static LIST_HEAD(elv_list);
41
42/*
43 * can we safely merge with this request?
44 */
45inline int elv_rq_merge_ok(struct request *rq, struct bio *bio)
46{
47 if (!rq_mergeable(rq))
48 return 0;
49
50 /*
51 * different data direction or already started, don't merge
52 */
53 if (bio_data_dir(bio) != rq_data_dir(rq))
54 return 0;
55
56 /*
57 * same device and no special stuff set, merge is ok
58 */
59 if (rq->rq_disk == bio->bi_bdev->bd_disk &&
60 !rq->waiting && !rq->special)
61 return 1;
62
63 return 0;
64}
65EXPORT_SYMBOL(elv_rq_merge_ok);
66
67inline int elv_try_merge(struct request *__rq, struct bio *bio)
68{
69 int ret = ELEVATOR_NO_MERGE;
70
71 /*
72 * we can merge and sequence is ok, check if it's possible
73 */
74 if (elv_rq_merge_ok(__rq, bio)) {
75 if (__rq->sector + __rq->nr_sectors == bio->bi_sector)
76 ret = ELEVATOR_BACK_MERGE;
77 else if (__rq->sector - bio_sectors(bio) == bio->bi_sector)
78 ret = ELEVATOR_FRONT_MERGE;
79 }
80
81 return ret;
82}
83EXPORT_SYMBOL(elv_try_merge);
84
Linus Torvalds1da177e2005-04-16 15:20:36 -070085static struct elevator_type *elevator_find(const char *name)
86{
87 struct elevator_type *e = NULL;
88 struct list_head *entry;
89
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 list_for_each(entry, &elv_list) {
91 struct elevator_type *__e;
92
93 __e = list_entry(entry, struct elevator_type, list);
94
95 if (!strcmp(__e->elevator_name, name)) {
96 e = __e;
97 break;
98 }
99 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100
101 return e;
102}
103
104static void elevator_put(struct elevator_type *e)
105{
106 module_put(e->elevator_owner);
107}
108
109static struct elevator_type *elevator_get(const char *name)
110{
Tejun Heo2824bc932005-10-20 10:56:41 +0200111 struct elevator_type *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112
Tejun Heo2824bc932005-10-20 10:56:41 +0200113 spin_lock_irq(&elv_list_lock);
114
115 e = elevator_find(name);
116 if (e && !try_module_get(e->elevator_owner))
117 e = NULL;
118
119 spin_unlock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120
121 return e;
122}
123
124static int elevator_attach(request_queue_t *q, struct elevator_type *e,
125 struct elevator_queue *eq)
126{
127 int ret = 0;
128
129 memset(eq, 0, sizeof(*eq));
130 eq->ops = &e->ops;
131 eq->elevator_type = e;
132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 q->elevator = eq;
134
135 if (eq->ops->elevator_init_fn)
136 ret = eq->ops->elevator_init_fn(q, eq);
137
138 return ret;
139}
140
141static char chosen_elevator[16];
142
143static void elevator_setup_default(void)
144{
Tejun Heo2824bc932005-10-20 10:56:41 +0200145 struct elevator_type *e;
146
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 /*
Nate Diller131dda72005-10-30 15:02:19 -0800148 * If default has not been set, use the compiled-in selection.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
Nate Diller131dda72005-10-30 15:02:19 -0800150 if (!chosen_elevator[0])
151 strcpy(chosen_elevator, CONFIG_DEFAULT_IOSCHED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Nate Diller131dda72005-10-30 15:02:19 -0800153 /*
154 * If the given scheduler is not available, fall back to no-op.
155 */
Zachary Amsdencff3ba22005-11-09 13:24:20 +0100156 if ((e = elevator_find(chosen_elevator)))
157 elevator_put(e);
158 else
Nate Diller131dda72005-10-30 15:02:19 -0800159 strcpy(chosen_elevator, "noop");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160}
161
162static int __init elevator_setup(char *str)
163{
164 strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1);
165 return 0;
166}
167
168__setup("elevator=", elevator_setup);
169
170int elevator_init(request_queue_t *q, char *name)
171{
172 struct elevator_type *e = NULL;
173 struct elevator_queue *eq;
174 int ret = 0;
175
Tejun Heocb98fc82005-10-28 08:29:39 +0200176 INIT_LIST_HEAD(&q->queue_head);
177 q->last_merge = NULL;
178 q->end_sector = 0;
179 q->boundary_rq = NULL;
Tejun Heocb98fc82005-10-28 08:29:39 +0200180
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 elevator_setup_default();
182
183 if (!name)
184 name = chosen_elevator;
185
186 e = elevator_get(name);
187 if (!e)
188 return -EINVAL;
189
190 eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL);
191 if (!eq) {
Zachary Amsdenb8ea2cb2005-11-09 13:23:01 +0100192 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 return -ENOMEM;
194 }
195
196 ret = elevator_attach(q, e, eq);
197 if (ret) {
198 kfree(eq);
Zachary Amsdenb8ea2cb2005-11-09 13:23:01 +0100199 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 }
201
202 return ret;
203}
204
205void elevator_exit(elevator_t *e)
206{
207 if (e->ops->elevator_exit_fn)
208 e->ops->elevator_exit_fn(e);
209
210 elevator_put(e->elevator_type);
211 e->elevator_type = NULL;
212 kfree(e);
213}
214
Tejun Heo8922e162005-10-20 16:23:44 +0200215/*
216 * Insert rq into dispatch queue of q. Queue lock must be held on
217 * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be
218 * appended to the dispatch queue. To be used by specific elevators.
219 */
Jens Axboe1b47f532005-10-20 16:37:00 +0200220void elv_dispatch_sort(request_queue_t *q, struct request *rq)
Tejun Heo8922e162005-10-20 16:23:44 +0200221{
222 sector_t boundary;
Tejun Heo8922e162005-10-20 16:23:44 +0200223 struct list_head *entry;
224
Tejun Heo06b86242005-10-20 16:46:23 +0200225 if (q->last_merge == rq)
226 q->last_merge = NULL;
Tejun Heo15853af2005-11-10 08:52:05 +0100227 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200228
Jens Axboe1b47f532005-10-20 16:37:00 +0200229 boundary = q->end_sector;
Tejun Heocb198332005-10-24 08:35:58 +0200230
Tejun Heo8922e162005-10-20 16:23:44 +0200231 list_for_each_prev(entry, &q->queue_head) {
232 struct request *pos = list_entry_rq(entry);
233
234 if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED))
235 break;
236 if (rq->sector >= boundary) {
237 if (pos->sector < boundary)
238 continue;
239 } else {
240 if (pos->sector >= boundary)
241 break;
242 }
243 if (rq->sector >= pos->sector)
244 break;
245 }
246
247 list_add(&rq->queuelist, entry);
248}
249
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250int elv_merge(request_queue_t *q, struct request **req, struct bio *bio)
251{
252 elevator_t *e = q->elevator;
Tejun Heo06b86242005-10-20 16:46:23 +0200253 int ret;
254
255 if (q->last_merge) {
256 ret = elv_try_merge(q->last_merge, bio);
257 if (ret != ELEVATOR_NO_MERGE) {
258 *req = q->last_merge;
259 return ret;
260 }
261 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
263 if (e->ops->elevator_merge_fn)
264 return e->ops->elevator_merge_fn(q, req, bio);
265
266 return ELEVATOR_NO_MERGE;
267}
268
269void elv_merged_request(request_queue_t *q, struct request *rq)
270{
271 elevator_t *e = q->elevator;
272
273 if (e->ops->elevator_merged_fn)
274 e->ops->elevator_merged_fn(q, rq);
Tejun Heo06b86242005-10-20 16:46:23 +0200275
276 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277}
278
279void elv_merge_requests(request_queue_t *q, struct request *rq,
280 struct request *next)
281{
282 elevator_t *e = q->elevator;
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 if (e->ops->elevator_merge_req_fn)
285 e->ops->elevator_merge_req_fn(q, rq, next);
Tejun Heo15853af2005-11-10 08:52:05 +0100286 q->nr_sorted--;
Tejun Heo06b86242005-10-20 16:46:23 +0200287
288 q->last_merge = rq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289}
290
Tejun Heo8922e162005-10-20 16:23:44 +0200291void elv_requeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292{
293 elevator_t *e = q->elevator;
294
295 /*
296 * it already went through dequeue, we need to decrement the
297 * in_flight count again
298 */
Tejun Heo8922e162005-10-20 16:23:44 +0200299 if (blk_account_rq(rq)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 q->in_flight--;
Tejun Heo8922e162005-10-20 16:23:44 +0200301 if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn)
302 e->ops->elevator_deactivate_req_fn(q, rq);
303 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304
305 rq->flags &= ~REQ_STARTED;
306
Tejun Heo797e7db2006-01-06 09:51:03 +0100307 __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308}
309
Tejun Heo15853af2005-11-10 08:52:05 +0100310static void elv_drain_elevator(request_queue_t *q)
311{
312 static int printed;
313 while (q->elevator->ops->elevator_dispatch_fn(q, 1))
314 ;
315 if (q->nr_sorted == 0)
316 return;
317 if (printed++ < 10) {
318 printk(KERN_ERR "%s: forced dispatching is broken "
319 "(nr_sorted=%u), please report this\n",
320 q->elevator->elevator_type->elevator_name, q->nr_sorted);
321 }
322}
323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324void __elv_add_request(request_queue_t *q, struct request *rq, int where,
325 int plug)
326{
Tejun Heo797e7db2006-01-06 09:51:03 +0100327 struct list_head *pos;
328 unsigned ordseq;
329
330 if (q->ordcolor)
331 rq->flags |= REQ_ORDERED_COLOR;
332
Tejun Heo8922e162005-10-20 16:23:44 +0200333 if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) {
334 /*
Tejun Heo797e7db2006-01-06 09:51:03 +0100335 * toggle ordered color
336 */
337 q->ordcolor ^= 1;
338
339 /*
Tejun Heo8922e162005-10-20 16:23:44 +0200340 * barriers implicitly indicate back insertion
341 */
342 if (where == ELEVATOR_INSERT_SORT)
343 where = ELEVATOR_INSERT_BACK;
344
345 /*
Jens Axboe1b47f532005-10-20 16:37:00 +0200346 * this request is scheduling boundary, update end_sector
Tejun Heo8922e162005-10-20 16:23:44 +0200347 */
348 if (blk_fs_request(rq)) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200349 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200350 q->boundary_rq = rq;
351 }
Tejun Heocb98fc82005-10-28 08:29:39 +0200352 } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT)
353 where = ELEVATOR_INSERT_BACK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 if (plug)
356 blk_plug_device(q);
357
358 rq->q = q;
359
Tejun Heo8922e162005-10-20 16:23:44 +0200360 switch (where) {
361 case ELEVATOR_INSERT_FRONT:
362 rq->flags |= REQ_SOFTBARRIER;
363
364 list_add(&rq->queuelist, &q->queue_head);
365 break;
366
367 case ELEVATOR_INSERT_BACK:
368 rq->flags |= REQ_SOFTBARRIER;
Tejun Heo15853af2005-11-10 08:52:05 +0100369 elv_drain_elevator(q);
Tejun Heo8922e162005-10-20 16:23:44 +0200370 list_add_tail(&rq->queuelist, &q->queue_head);
371 /*
372 * We kick the queue here for the following reasons.
373 * - The elevator might have returned NULL previously
374 * to delay requests and returned them now. As the
375 * queue wasn't empty before this request, ll_rw_blk
376 * won't run the queue on return, resulting in hang.
377 * - Usually, back inserted requests won't be merged
378 * with anything. There's no point in delaying queue
379 * processing.
380 */
381 blk_remove_plug(q);
382 q->request_fn(q);
383 break;
384
385 case ELEVATOR_INSERT_SORT:
386 BUG_ON(!blk_fs_request(rq));
387 rq->flags |= REQ_SORTED;
Tejun Heo15853af2005-11-10 08:52:05 +0100388 q->nr_sorted++;
Tejun Heo06b86242005-10-20 16:46:23 +0200389 if (q->last_merge == NULL && rq_mergeable(rq))
390 q->last_merge = rq;
Tejun Heoca235092005-11-01 17:23:49 +0900391 /*
392 * Some ioscheds (cfq) run q->request_fn directly, so
393 * rq cannot be accessed after calling
394 * elevator_add_req_fn.
395 */
396 q->elevator->ops->elevator_add_req_fn(q, rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200397 break;
398
Tejun Heo797e7db2006-01-06 09:51:03 +0100399 case ELEVATOR_INSERT_REQUEUE:
400 /*
401 * If ordered flush isn't in progress, we do front
402 * insertion; otherwise, requests should be requeued
403 * in ordseq order.
404 */
405 rq->flags |= REQ_SOFTBARRIER;
406
407 if (q->ordseq == 0) {
408 list_add(&rq->queuelist, &q->queue_head);
409 break;
410 }
411
412 ordseq = blk_ordered_req_seq(rq);
413
414 list_for_each(pos, &q->queue_head) {
415 struct request *pos_rq = list_entry_rq(pos);
416 if (ordseq <= blk_ordered_req_seq(pos_rq))
417 break;
418 }
419
420 list_add_tail(&rq->queuelist, pos);
421 break;
422
Tejun Heo8922e162005-10-20 16:23:44 +0200423 default:
424 printk(KERN_ERR "%s: bad insertion point %d\n",
425 __FUNCTION__, where);
426 BUG();
427 }
428
429 if (blk_queue_plugged(q)) {
430 int nrq = q->rq.count[READ] + q->rq.count[WRITE]
431 - q->in_flight;
432
433 if (nrq >= q->unplug_thresh)
434 __generic_unplug_device(q);
435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
438void elv_add_request(request_queue_t *q, struct request *rq, int where,
439 int plug)
440{
441 unsigned long flags;
442
443 spin_lock_irqsave(q->queue_lock, flags);
444 __elv_add_request(q, rq, where, plug);
445 spin_unlock_irqrestore(q->queue_lock, flags);
446}
447
448static inline struct request *__elv_next_request(request_queue_t *q)
449{
Tejun Heo8922e162005-10-20 16:23:44 +0200450 struct request *rq;
451
Tejun Heo797e7db2006-01-06 09:51:03 +0100452 while (1) {
453 while (!list_empty(&q->queue_head)) {
454 rq = list_entry_rq(q->queue_head.next);
455 if (blk_do_ordered(q, &rq))
456 return rq;
457 }
Tejun Heo8922e162005-10-20 16:23:44 +0200458
Tejun Heo797e7db2006-01-06 09:51:03 +0100459 if (!q->elevator->ops->elevator_dispatch_fn(q, 0))
460 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462}
463
464struct request *elv_next_request(request_queue_t *q)
465{
466 struct request *rq;
467 int ret;
468
469 while ((rq = __elv_next_request(q)) != NULL) {
Tejun Heo8922e162005-10-20 16:23:44 +0200470 if (!(rq->flags & REQ_STARTED)) {
471 elevator_t *e = q->elevator;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472
Tejun Heo8922e162005-10-20 16:23:44 +0200473 /*
474 * This is the first time the device driver
475 * sees this request (possibly after
476 * requeueing). Notify IO scheduler.
477 */
478 if (blk_sorted_rq(rq) &&
479 e->ops->elevator_activate_req_fn)
480 e->ops->elevator_activate_req_fn(q, rq);
481
482 /*
483 * just mark as started even if we don't start
484 * it, a request that has been delayed should
485 * not be passed by new incoming requests
486 */
487 rq->flags |= REQ_STARTED;
488 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Tejun Heo8922e162005-10-20 16:23:44 +0200490 if (!q->boundary_rq || q->boundary_rq == rq) {
Jens Axboe1b47f532005-10-20 16:37:00 +0200491 q->end_sector = rq_end_sector(rq);
Tejun Heo8922e162005-10-20 16:23:44 +0200492 q->boundary_rq = NULL;
493 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
495 if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn)
496 break;
497
498 ret = q->prep_rq_fn(q, rq);
499 if (ret == BLKPREP_OK) {
500 break;
501 } else if (ret == BLKPREP_DEFER) {
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500502 /*
503 * the request may have been (partially) prepped.
504 * we need to keep this request in the front to
Tejun Heo8922e162005-10-20 16:23:44 +0200505 * avoid resource deadlock. REQ_STARTED will
506 * prevent other fs requests from passing this one.
Tejun Heo 2e759cd2005-04-24 02:04:21 -0500507 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 rq = NULL;
509 break;
510 } else if (ret == BLKPREP_KILL) {
511 int nr_bytes = rq->hard_nr_sectors << 9;
512
513 if (!nr_bytes)
514 nr_bytes = rq->data_len;
515
516 blkdev_dequeue_request(rq);
517 rq->flags |= REQ_QUIET;
518 end_that_request_chunk(rq, 0, nr_bytes);
Tejun Heo8ffdc652006-01-06 09:49:03 +0100519 end_that_request_last(rq, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520 } else {
521 printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__,
522 ret);
523 break;
524 }
525 }
526
527 return rq;
528}
529
Tejun Heo8922e162005-10-20 16:23:44 +0200530void elv_dequeue_request(request_queue_t *q, struct request *rq)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531{
Tejun Heo8922e162005-10-20 16:23:44 +0200532 BUG_ON(list_empty(&rq->queuelist));
533
534 list_del_init(&rq->queuelist);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 /*
537 * the time frame between a request being removed from the lists
538 * and to it is freed is accounted as io that is in progress at
Tejun Heo8922e162005-10-20 16:23:44 +0200539 * the driver side.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 */
541 if (blk_account_rq(rq))
542 q->in_flight++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543}
544
545int elv_queue_empty(request_queue_t *q)
546{
547 elevator_t *e = q->elevator;
548
Tejun Heo8922e162005-10-20 16:23:44 +0200549 if (!list_empty(&q->queue_head))
550 return 0;
551
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 if (e->ops->elevator_queue_empty_fn)
553 return e->ops->elevator_queue_empty_fn(q);
554
Tejun Heo8922e162005-10-20 16:23:44 +0200555 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556}
557
558struct request *elv_latter_request(request_queue_t *q, struct request *rq)
559{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 elevator_t *e = q->elevator;
561
562 if (e->ops->elevator_latter_req_fn)
563 return e->ops->elevator_latter_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 return NULL;
565}
566
567struct request *elv_former_request(request_queue_t *q, struct request *rq)
568{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 elevator_t *e = q->elevator;
570
571 if (e->ops->elevator_former_req_fn)
572 return e->ops->elevator_former_req_fn(q, rq);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 return NULL;
574}
575
Jens Axboe22e2c502005-06-27 10:55:12 +0200576int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio,
Al Viro8267e262005-10-21 03:20:53 -0400577 gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578{
579 elevator_t *e = q->elevator;
580
581 if (e->ops->elevator_set_req_fn)
Jens Axboe22e2c502005-06-27 10:55:12 +0200582 return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 rq->elevator_private = NULL;
585 return 0;
586}
587
588void elv_put_request(request_queue_t *q, struct request *rq)
589{
590 elevator_t *e = q->elevator;
591
592 if (e->ops->elevator_put_req_fn)
593 e->ops->elevator_put_req_fn(q, rq);
594}
595
Jens Axboe22e2c502005-06-27 10:55:12 +0200596int elv_may_queue(request_queue_t *q, int rw, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 elevator_t *e = q->elevator;
599
600 if (e->ops->elevator_may_queue_fn)
Jens Axboe22e2c502005-06-27 10:55:12 +0200601 return e->ops->elevator_may_queue_fn(q, rw, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
603 return ELV_MQUEUE_MAY;
604}
605
606void elv_completed_request(request_queue_t *q, struct request *rq)
607{
608 elevator_t *e = q->elevator;
609
610 /*
611 * request is released from the driver, io must be done
612 */
Tejun Heo8922e162005-10-20 16:23:44 +0200613 if (blk_account_rq(rq)) {
Tejun Heo797e7db2006-01-06 09:51:03 +0100614 struct request *first_rq = list_entry_rq(q->queue_head.next);
615
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 q->in_flight--;
Tejun Heo797e7db2006-01-06 09:51:03 +0100617
618 /*
619 * Check if the queue is waiting for fs requests to be
620 * drained for flush sequence.
621 */
622 if (q->ordseq && q->in_flight == 0 &&
623 blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN &&
624 blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) {
625 blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0);
626 q->request_fn(q);
627 }
628
Tejun Heo8922e162005-10-20 16:23:44 +0200629 if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn)
630 e->ops->elevator_completed_req_fn(q, rq);
631 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
634int elv_register_queue(struct request_queue *q)
635{
636 elevator_t *e = q->elevator;
637
638 e->kobj.parent = kobject_get(&q->kobj);
639 if (!e->kobj.parent)
640 return -EBUSY;
641
642 snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched");
643 e->kobj.ktype = e->elevator_type->elevator_ktype;
644
645 return kobject_register(&e->kobj);
646}
647
648void elv_unregister_queue(struct request_queue *q)
649{
650 if (q) {
651 elevator_t *e = q->elevator;
652 kobject_unregister(&e->kobj);
653 kobject_put(&q->kobj);
654 }
655}
656
657int elv_register(struct elevator_type *e)
658{
Tejun Heo2824bc932005-10-20 10:56:41 +0200659 spin_lock_irq(&elv_list_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660 if (elevator_find(e->elevator_name))
661 BUG();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 list_add_tail(&e->list, &elv_list);
663 spin_unlock_irq(&elv_list_lock);
664
665 printk(KERN_INFO "io scheduler %s registered", e->elevator_name);
666 if (!strcmp(e->elevator_name, chosen_elevator))
667 printk(" (default)");
668 printk("\n");
669 return 0;
670}
671EXPORT_SYMBOL_GPL(elv_register);
672
673void elv_unregister(struct elevator_type *e)
674{
Christoph Hellwig83521d32005-10-30 15:01:39 -0800675 struct task_struct *g, *p;
676
677 /*
678 * Iterate every thread in the process to remove the io contexts.
679 */
680 read_lock(&tasklist_lock);
681 do_each_thread(g, p) {
682 struct io_context *ioc = p->io_context;
683 if (ioc && ioc->cic) {
684 ioc->cic->exit(ioc->cic);
685 ioc->cic->dtor(ioc->cic);
686 ioc->cic = NULL;
687 }
688 if (ioc && ioc->aic) {
689 ioc->aic->exit(ioc->aic);
690 ioc->aic->dtor(ioc->aic);
691 ioc->aic = NULL;
692 }
693 } while_each_thread(g, p);
694 read_unlock(&tasklist_lock);
695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 spin_lock_irq(&elv_list_lock);
697 list_del_init(&e->list);
698 spin_unlock_irq(&elv_list_lock);
699}
700EXPORT_SYMBOL_GPL(elv_unregister);
701
702/*
703 * switch to new_e io scheduler. be careful not to introduce deadlocks -
704 * we don't free the old io scheduler, before we have allocated what we
705 * need for the new one. this way we have a chance of going back to the old
Tejun Heocb98fc82005-10-28 08:29:39 +0200706 * one, if the new one fails init for some reason.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 */
708static void elevator_switch(request_queue_t *q, struct elevator_type *new_e)
709{
Tejun Heocb98fc82005-10-28 08:29:39 +0200710 elevator_t *old_elevator, *e;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711
Tejun Heocb98fc82005-10-28 08:29:39 +0200712 /*
713 * Allocate new elevator
714 */
715 e = kmalloc(sizeof(elevator_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (!e)
717 goto error;
718
719 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200720 * Turn on BYPASS and drain all requests w/ elevator private data
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 */
Tejun Heocb98fc82005-10-28 08:29:39 +0200722 spin_lock_irq(q->queue_lock);
723
Jens Axboe64521d12005-10-28 08:30:39 +0200724 set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200725
Tejun Heo15853af2005-11-10 08:52:05 +0100726 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200727
728 while (q->rq.elvpriv) {
Tejun Heo407df2a2005-11-10 08:48:21 +0100729 blk_remove_plug(q);
730 q->request_fn(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200731 spin_unlock_irq(q->queue_lock);
Jens Axboe64521d12005-10-28 08:30:39 +0200732 msleep(10);
Tejun Heocb98fc82005-10-28 08:29:39 +0200733 spin_lock_irq(q->queue_lock);
Tejun Heo15853af2005-11-10 08:52:05 +0100734 elv_drain_elevator(q);
Tejun Heocb98fc82005-10-28 08:29:39 +0200735 }
736
737 spin_unlock_irq(q->queue_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 /*
740 * unregister old elevator data
741 */
742 elv_unregister_queue(q);
743 old_elevator = q->elevator;
744
745 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 * attach and start new elevator
747 */
748 if (elevator_attach(q, new_e, e))
749 goto fail;
750
751 if (elv_register_queue(q))
752 goto fail_register;
753
754 /*
Tejun Heocb98fc82005-10-28 08:29:39 +0200755 * finally exit old elevator and turn off BYPASS.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
757 elevator_exit(old_elevator);
Jens Axboe64521d12005-10-28 08:30:39 +0200758 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 return;
760
761fail_register:
762 /*
763 * switch failed, exit the new io scheduler and reattach the old
764 * one again (along with re-adding the sysfs dir)
765 */
766 elevator_exit(e);
Tejun Heocb98fc82005-10-28 08:29:39 +0200767 e = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768fail:
769 q->elevator = old_elevator;
770 elv_register_queue(q);
Jens Axboe64521d12005-10-28 08:30:39 +0200771 clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags);
Tejun Heocb98fc82005-10-28 08:29:39 +0200772 kfree(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 elevator_put(new_e);
775 printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name);
776}
777
778ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count)
779{
780 char elevator_name[ELV_NAME_MAX];
Tejun Heobe561232005-11-10 08:55:01 +0100781 size_t len;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 struct elevator_type *e;
783
Tejun Heobe561232005-11-10 08:55:01 +0100784 elevator_name[sizeof(elevator_name) - 1] = '\0';
785 strncpy(elevator_name, name, sizeof(elevator_name) - 1);
786 len = strlen(elevator_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
Tejun Heobe561232005-11-10 08:55:01 +0100788 if (len && elevator_name[len - 1] == '\n')
789 elevator_name[len - 1] = '\0';
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790
791 e = elevator_get(elevator_name);
792 if (!e) {
793 printk(KERN_ERR "elevator: type %s not found\n", elevator_name);
794 return -EINVAL;
795 }
796
Nate Diller2ca7d932005-10-30 15:02:24 -0800797 if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) {
798 elevator_put(e);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 return count;
Nate Diller2ca7d932005-10-30 15:02:24 -0800800 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
802 elevator_switch(q, e);
803 return count;
804}
805
806ssize_t elv_iosched_show(request_queue_t *q, char *name)
807{
808 elevator_t *e = q->elevator;
809 struct elevator_type *elv = e->elevator_type;
810 struct list_head *entry;
811 int len = 0;
812
813 spin_lock_irq(q->queue_lock);
814 list_for_each(entry, &elv_list) {
815 struct elevator_type *__e;
816
817 __e = list_entry(entry, struct elevator_type, list);
818 if (!strcmp(elv->elevator_name, __e->elevator_name))
819 len += sprintf(name+len, "[%s] ", elv->elevator_name);
820 else
821 len += sprintf(name+len, "%s ", __e->elevator_name);
822 }
823 spin_unlock_irq(q->queue_lock);
824
825 len += sprintf(len+name, "\n");
826 return len;
827}
828
Jens Axboe1b47f532005-10-20 16:37:00 +0200829EXPORT_SYMBOL(elv_dispatch_sort);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830EXPORT_SYMBOL(elv_add_request);
831EXPORT_SYMBOL(__elv_add_request);
832EXPORT_SYMBOL(elv_requeue_request);
833EXPORT_SYMBOL(elv_next_request);
Tejun Heo8922e162005-10-20 16:23:44 +0200834EXPORT_SYMBOL(elv_dequeue_request);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835EXPORT_SYMBOL(elv_queue_empty);
836EXPORT_SYMBOL(elv_completed_request);
837EXPORT_SYMBOL(elevator_exit);
838EXPORT_SYMBOL(elevator_init);