Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/drivers/block/elevator.c | |
3 | * | |
4 | * Block device elevator/IO-scheduler. | |
5 | * | |
6 | * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
7 | * | |
8 | * 30042000 Jens Axboe <axboe@suse.de> : | |
9 | * | |
10 | * Split the elevator a bit so that it is possible to choose a different | |
11 | * one or even write a new "plug in". There are three pieces: | |
12 | * - elevator_fn, inserts a new request in the queue list | |
13 | * - elevator_merge_fn, decides whether a new buffer can be merged with | |
14 | * an existing request | |
15 | * - elevator_dequeue_fn, called when a request is taken off the active list | |
16 | * | |
17 | * 20082000 Dave Jones <davej@suse.de> : | |
18 | * Removed tests for max-bomb-segments, which was breaking elvtune | |
19 | * when run without -bN | |
20 | * | |
21 | * Jens: | |
22 | * - Rework again to work with bio instead of buffer_heads | |
23 | * - loose bi_dev comparisons, partition handling is right now | |
24 | * - completely modularize elevator setup and teardown | |
25 | * | |
26 | */ | |
27 | #include <linux/kernel.h> | |
28 | #include <linux/fs.h> | |
29 | #include <linux/blkdev.h> | |
30 | #include <linux/elevator.h> | |
31 | #include <linux/bio.h> | |
32 | #include <linux/config.h> | |
33 | #include <linux/module.h> | |
34 | #include <linux/slab.h> | |
35 | #include <linux/init.h> | |
36 | #include <linux/compiler.h> | |
cb98fc8b | 37 | #include <linux/delay.h> |
1da177e4 LT |
38 | |
39 | #include <asm/uaccess.h> | |
40 | ||
41 | static DEFINE_SPINLOCK(elv_list_lock); | |
42 | static LIST_HEAD(elv_list); | |
43 | ||
44 | /* | |
45 | * can we safely merge with this request? | |
46 | */ | |
47 | inline int elv_rq_merge_ok(struct request *rq, struct bio *bio) | |
48 | { | |
49 | if (!rq_mergeable(rq)) | |
50 | return 0; | |
51 | ||
52 | /* | |
53 | * different data direction or already started, don't merge | |
54 | */ | |
55 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
56 | return 0; | |
57 | ||
58 | /* | |
59 | * same device and no special stuff set, merge is ok | |
60 | */ | |
61 | if (rq->rq_disk == bio->bi_bdev->bd_disk && | |
62 | !rq->waiting && !rq->special) | |
63 | return 1; | |
64 | ||
65 | return 0; | |
66 | } | |
67 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
68 | ||
69 | inline int elv_try_merge(struct request *__rq, struct bio *bio) | |
70 | { | |
71 | int ret = ELEVATOR_NO_MERGE; | |
72 | ||
73 | /* | |
74 | * we can merge and sequence is ok, check if it's possible | |
75 | */ | |
76 | if (elv_rq_merge_ok(__rq, bio)) { | |
77 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
78 | ret = ELEVATOR_BACK_MERGE; | |
79 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
80 | ret = ELEVATOR_FRONT_MERGE; | |
81 | } | |
82 | ||
83 | return ret; | |
84 | } | |
85 | EXPORT_SYMBOL(elv_try_merge); | |
86 | ||
1da177e4 LT |
87 | static struct elevator_type *elevator_find(const char *name) |
88 | { | |
89 | struct elevator_type *e = NULL; | |
90 | struct list_head *entry; | |
91 | ||
1da177e4 LT |
92 | list_for_each(entry, &elv_list) { |
93 | struct elevator_type *__e; | |
94 | ||
95 | __e = list_entry(entry, struct elevator_type, list); | |
96 | ||
97 | if (!strcmp(__e->elevator_name, name)) { | |
98 | e = __e; | |
99 | break; | |
100 | } | |
101 | } | |
1da177e4 LT |
102 | |
103 | return e; | |
104 | } | |
105 | ||
106 | static void elevator_put(struct elevator_type *e) | |
107 | { | |
108 | module_put(e->elevator_owner); | |
109 | } | |
110 | ||
111 | static struct elevator_type *elevator_get(const char *name) | |
112 | { | |
2824bc93 | 113 | struct elevator_type *e; |
1da177e4 | 114 | |
2824bc93 TH |
115 | spin_lock_irq(&elv_list_lock); |
116 | ||
117 | e = elevator_find(name); | |
118 | if (e && !try_module_get(e->elevator_owner)) | |
119 | e = NULL; | |
120 | ||
121 | spin_unlock_irq(&elv_list_lock); | |
1da177e4 LT |
122 | |
123 | return e; | |
124 | } | |
125 | ||
126 | static int elevator_attach(request_queue_t *q, struct elevator_type *e, | |
127 | struct elevator_queue *eq) | |
128 | { | |
129 | int ret = 0; | |
130 | ||
131 | memset(eq, 0, sizeof(*eq)); | |
132 | eq->ops = &e->ops; | |
133 | eq->elevator_type = e; | |
134 | ||
1da177e4 LT |
135 | q->elevator = eq; |
136 | ||
137 | if (eq->ops->elevator_init_fn) | |
138 | ret = eq->ops->elevator_init_fn(q, eq); | |
139 | ||
140 | return ret; | |
141 | } | |
142 | ||
143 | static char chosen_elevator[16]; | |
144 | ||
145 | static void elevator_setup_default(void) | |
146 | { | |
2824bc93 TH |
147 | struct elevator_type *e; |
148 | ||
1da177e4 LT |
149 | /* |
150 | * check if default is set and exists | |
151 | */ | |
2824bc93 TH |
152 | if (chosen_elevator[0] && (e = elevator_get(chosen_elevator))) { |
153 | elevator_put(e); | |
1da177e4 | 154 | return; |
2824bc93 | 155 | } |
1da177e4 LT |
156 | |
157 | #if defined(CONFIG_IOSCHED_AS) | |
158 | strcpy(chosen_elevator, "anticipatory"); | |
159 | #elif defined(CONFIG_IOSCHED_DEADLINE) | |
160 | strcpy(chosen_elevator, "deadline"); | |
161 | #elif defined(CONFIG_IOSCHED_CFQ) | |
162 | strcpy(chosen_elevator, "cfq"); | |
163 | #elif defined(CONFIG_IOSCHED_NOOP) | |
164 | strcpy(chosen_elevator, "noop"); | |
165 | #else | |
166 | #error "You must build at least 1 IO scheduler into the kernel" | |
167 | #endif | |
168 | } | |
169 | ||
170 | static int __init elevator_setup(char *str) | |
171 | { | |
172 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
173 | return 0; | |
174 | } | |
175 | ||
176 | __setup("elevator=", elevator_setup); | |
177 | ||
178 | int elevator_init(request_queue_t *q, char *name) | |
179 | { | |
180 | struct elevator_type *e = NULL; | |
181 | struct elevator_queue *eq; | |
182 | int ret = 0; | |
183 | ||
cb98fc8b TH |
184 | INIT_LIST_HEAD(&q->queue_head); |
185 | q->last_merge = NULL; | |
186 | q->end_sector = 0; | |
187 | q->boundary_rq = NULL; | |
188 | q->max_back_kb = 0; | |
189 | ||
1da177e4 LT |
190 | elevator_setup_default(); |
191 | ||
192 | if (!name) | |
193 | name = chosen_elevator; | |
194 | ||
195 | e = elevator_get(name); | |
196 | if (!e) | |
197 | return -EINVAL; | |
198 | ||
199 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
200 | if (!eq) { | |
201 | elevator_put(e->elevator_type); | |
202 | return -ENOMEM; | |
203 | } | |
204 | ||
205 | ret = elevator_attach(q, e, eq); | |
206 | if (ret) { | |
207 | kfree(eq); | |
208 | elevator_put(e->elevator_type); | |
209 | } | |
210 | ||
211 | return ret; | |
212 | } | |
213 | ||
214 | void elevator_exit(elevator_t *e) | |
215 | { | |
216 | if (e->ops->elevator_exit_fn) | |
217 | e->ops->elevator_exit_fn(e); | |
218 | ||
219 | elevator_put(e->elevator_type); | |
220 | e->elevator_type = NULL; | |
221 | kfree(e); | |
222 | } | |
223 | ||
8922e16c TH |
224 | /* |
225 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
226 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
227 | * appended to the dispatch queue. To be used by specific elevators. | |
228 | */ | |
1b47f531 | 229 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
230 | { |
231 | sector_t boundary; | |
8922e16c TH |
232 | struct list_head *entry; |
233 | ||
06b86245 TH |
234 | if (q->last_merge == rq) |
235 | q->last_merge = NULL; | |
236 | ||
1b47f531 | 237 | boundary = q->end_sector; |
cb19833d | 238 | |
8922e16c TH |
239 | list_for_each_prev(entry, &q->queue_head) { |
240 | struct request *pos = list_entry_rq(entry); | |
241 | ||
242 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
243 | break; | |
244 | if (rq->sector >= boundary) { | |
245 | if (pos->sector < boundary) | |
246 | continue; | |
247 | } else { | |
248 | if (pos->sector >= boundary) | |
249 | break; | |
250 | } | |
251 | if (rq->sector >= pos->sector) | |
252 | break; | |
253 | } | |
254 | ||
255 | list_add(&rq->queuelist, entry); | |
256 | } | |
257 | ||
1da177e4 LT |
258 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
259 | { | |
260 | elevator_t *e = q->elevator; | |
06b86245 TH |
261 | int ret; |
262 | ||
263 | if (q->last_merge) { | |
264 | ret = elv_try_merge(q->last_merge, bio); | |
265 | if (ret != ELEVATOR_NO_MERGE) { | |
266 | *req = q->last_merge; | |
267 | return ret; | |
268 | } | |
269 | } | |
1da177e4 LT |
270 | |
271 | if (e->ops->elevator_merge_fn) | |
272 | return e->ops->elevator_merge_fn(q, req, bio); | |
273 | ||
274 | return ELEVATOR_NO_MERGE; | |
275 | } | |
276 | ||
277 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
278 | { | |
279 | elevator_t *e = q->elevator; | |
280 | ||
281 | if (e->ops->elevator_merged_fn) | |
282 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
283 | |
284 | q->last_merge = rq; | |
1da177e4 LT |
285 | } |
286 | ||
287 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
288 | struct request *next) | |
289 | { | |
290 | elevator_t *e = q->elevator; | |
291 | ||
1da177e4 LT |
292 | if (e->ops->elevator_merge_req_fn) |
293 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 TH |
294 | |
295 | q->last_merge = rq; | |
1da177e4 LT |
296 | } |
297 | ||
8922e16c | 298 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
299 | { |
300 | elevator_t *e = q->elevator; | |
301 | ||
302 | /* | |
303 | * it already went through dequeue, we need to decrement the | |
304 | * in_flight count again | |
305 | */ | |
8922e16c | 306 | if (blk_account_rq(rq)) { |
1da177e4 | 307 | q->in_flight--; |
8922e16c TH |
308 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
309 | e->ops->elevator_deactivate_req_fn(q, rq); | |
310 | } | |
1da177e4 LT |
311 | |
312 | rq->flags &= ~REQ_STARTED; | |
313 | ||
1da177e4 LT |
314 | /* |
315 | * if this is the flush, requeue the original instead and drop the flush | |
316 | */ | |
317 | if (rq->flags & REQ_BAR_FLUSH) { | |
318 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
319 | rq = rq->end_io_data; | |
320 | } | |
321 | ||
8922e16c | 322 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
323 | } |
324 | ||
325 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
326 | int plug) | |
327 | { | |
8922e16c TH |
328 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
329 | /* | |
330 | * barriers implicitly indicate back insertion | |
331 | */ | |
332 | if (where == ELEVATOR_INSERT_SORT) | |
333 | where = ELEVATOR_INSERT_BACK; | |
334 | ||
335 | /* | |
1b47f531 | 336 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
337 | */ |
338 | if (blk_fs_request(rq)) { | |
1b47f531 | 339 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
340 | q->boundary_rq = rq; |
341 | } | |
cb98fc8b TH |
342 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
343 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
344 | |
345 | if (plug) | |
346 | blk_plug_device(q); | |
347 | ||
348 | rq->q = q; | |
349 | ||
8922e16c TH |
350 | switch (where) { |
351 | case ELEVATOR_INSERT_FRONT: | |
352 | rq->flags |= REQ_SOFTBARRIER; | |
353 | ||
354 | list_add(&rq->queuelist, &q->queue_head); | |
355 | break; | |
356 | ||
357 | case ELEVATOR_INSERT_BACK: | |
358 | rq->flags |= REQ_SOFTBARRIER; | |
359 | ||
360 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
361 | ; | |
362 | list_add_tail(&rq->queuelist, &q->queue_head); | |
363 | /* | |
364 | * We kick the queue here for the following reasons. | |
365 | * - The elevator might have returned NULL previously | |
366 | * to delay requests and returned them now. As the | |
367 | * queue wasn't empty before this request, ll_rw_blk | |
368 | * won't run the queue on return, resulting in hang. | |
369 | * - Usually, back inserted requests won't be merged | |
370 | * with anything. There's no point in delaying queue | |
371 | * processing. | |
372 | */ | |
373 | blk_remove_plug(q); | |
374 | q->request_fn(q); | |
375 | break; | |
376 | ||
377 | case ELEVATOR_INSERT_SORT: | |
378 | BUG_ON(!blk_fs_request(rq)); | |
379 | rq->flags |= REQ_SORTED; | |
380 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
06b86245 TH |
381 | if (q->last_merge == NULL && rq_mergeable(rq)) |
382 | q->last_merge = rq; | |
8922e16c TH |
383 | break; |
384 | ||
385 | default: | |
386 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
387 | __FUNCTION__, where); | |
388 | BUG(); | |
389 | } | |
390 | ||
391 | if (blk_queue_plugged(q)) { | |
392 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
393 | - q->in_flight; | |
394 | ||
395 | if (nrq >= q->unplug_thresh) | |
396 | __generic_unplug_device(q); | |
397 | } | |
1da177e4 LT |
398 | } |
399 | ||
400 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
401 | int plug) | |
402 | { | |
403 | unsigned long flags; | |
404 | ||
405 | spin_lock_irqsave(q->queue_lock, flags); | |
406 | __elv_add_request(q, rq, where, plug); | |
407 | spin_unlock_irqrestore(q->queue_lock, flags); | |
408 | } | |
409 | ||
410 | static inline struct request *__elv_next_request(request_queue_t *q) | |
411 | { | |
8922e16c TH |
412 | struct request *rq; |
413 | ||
414 | if (unlikely(list_empty(&q->queue_head) && | |
415 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
416 | return NULL; | |
417 | ||
418 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
419 | |
420 | /* | |
421 | * if this is a barrier write and the device has to issue a | |
422 | * flush sequence to support it, check how far we are | |
423 | */ | |
8922e16c | 424 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
425 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
426 | ||
427 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
428 | !blk_barrier_preflush(rq)) | |
429 | rq = blk_start_pre_flush(q, rq); | |
430 | } | |
431 | ||
432 | return rq; | |
433 | } | |
434 | ||
435 | struct request *elv_next_request(request_queue_t *q) | |
436 | { | |
437 | struct request *rq; | |
438 | int ret; | |
439 | ||
440 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
441 | if (!(rq->flags & REQ_STARTED)) { |
442 | elevator_t *e = q->elevator; | |
443 | ||
444 | /* | |
445 | * This is the first time the device driver | |
446 | * sees this request (possibly after | |
447 | * requeueing). Notify IO scheduler. | |
448 | */ | |
449 | if (blk_sorted_rq(rq) && | |
450 | e->ops->elevator_activate_req_fn) | |
451 | e->ops->elevator_activate_req_fn(q, rq); | |
452 | ||
453 | /* | |
454 | * just mark as started even if we don't start | |
455 | * it, a request that has been delayed should | |
456 | * not be passed by new incoming requests | |
457 | */ | |
458 | rq->flags |= REQ_STARTED; | |
459 | } | |
1da177e4 | 460 | |
8922e16c | 461 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 462 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
463 | q->boundary_rq = NULL; |
464 | } | |
465 | ||
1da177e4 LT |
466 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) |
467 | break; | |
468 | ||
469 | ret = q->prep_rq_fn(q, rq); | |
470 | if (ret == BLKPREP_OK) { | |
471 | break; | |
472 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
473 | /* |
474 | * the request may have been (partially) prepped. | |
475 | * we need to keep this request in the front to | |
8922e16c TH |
476 | * avoid resource deadlock. REQ_STARTED will |
477 | * prevent other fs requests from passing this one. | |
2e759cd4 | 478 | */ |
1da177e4 LT |
479 | rq = NULL; |
480 | break; | |
481 | } else if (ret == BLKPREP_KILL) { | |
482 | int nr_bytes = rq->hard_nr_sectors << 9; | |
483 | ||
484 | if (!nr_bytes) | |
485 | nr_bytes = rq->data_len; | |
486 | ||
487 | blkdev_dequeue_request(rq); | |
488 | rq->flags |= REQ_QUIET; | |
489 | end_that_request_chunk(rq, 0, nr_bytes); | |
490 | end_that_request_last(rq); | |
491 | } else { | |
492 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
493 | ret); | |
494 | break; | |
495 | } | |
496 | } | |
497 | ||
498 | return rq; | |
499 | } | |
500 | ||
8922e16c | 501 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 502 | { |
8922e16c TH |
503 | BUG_ON(list_empty(&rq->queuelist)); |
504 | ||
505 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
506 | |
507 | /* | |
508 | * the time frame between a request being removed from the lists | |
509 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 510 | * the driver side. |
1da177e4 LT |
511 | */ |
512 | if (blk_account_rq(rq)) | |
513 | q->in_flight++; | |
1da177e4 LT |
514 | } |
515 | ||
516 | int elv_queue_empty(request_queue_t *q) | |
517 | { | |
518 | elevator_t *e = q->elevator; | |
519 | ||
8922e16c TH |
520 | if (!list_empty(&q->queue_head)) |
521 | return 0; | |
522 | ||
1da177e4 LT |
523 | if (e->ops->elevator_queue_empty_fn) |
524 | return e->ops->elevator_queue_empty_fn(q); | |
525 | ||
8922e16c | 526 | return 1; |
1da177e4 LT |
527 | } |
528 | ||
529 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
530 | { | |
531 | struct list_head *next; | |
532 | ||
533 | elevator_t *e = q->elevator; | |
534 | ||
535 | if (e->ops->elevator_latter_req_fn) | |
536 | return e->ops->elevator_latter_req_fn(q, rq); | |
537 | ||
538 | next = rq->queuelist.next; | |
539 | if (next != &q->queue_head && next != &rq->queuelist) | |
540 | return list_entry_rq(next); | |
541 | ||
542 | return NULL; | |
543 | } | |
544 | ||
545 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
546 | { | |
547 | struct list_head *prev; | |
548 | ||
549 | elevator_t *e = q->elevator; | |
550 | ||
551 | if (e->ops->elevator_former_req_fn) | |
552 | return e->ops->elevator_former_req_fn(q, rq); | |
553 | ||
554 | prev = rq->queuelist.prev; | |
555 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
556 | return list_entry_rq(prev); | |
557 | ||
558 | return NULL; | |
559 | } | |
560 | ||
22e2c507 JA |
561 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
562 | int gfp_mask) | |
1da177e4 LT |
563 | { |
564 | elevator_t *e = q->elevator; | |
565 | ||
566 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 567 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
568 | |
569 | rq->elevator_private = NULL; | |
570 | return 0; | |
571 | } | |
572 | ||
573 | void elv_put_request(request_queue_t *q, struct request *rq) | |
574 | { | |
575 | elevator_t *e = q->elevator; | |
576 | ||
577 | if (e->ops->elevator_put_req_fn) | |
578 | e->ops->elevator_put_req_fn(q, rq); | |
579 | } | |
580 | ||
22e2c507 | 581 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
582 | { |
583 | elevator_t *e = q->elevator; | |
584 | ||
585 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 586 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
587 | |
588 | return ELV_MQUEUE_MAY; | |
589 | } | |
590 | ||
591 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
592 | { | |
593 | elevator_t *e = q->elevator; | |
594 | ||
595 | /* | |
596 | * request is released from the driver, io must be done | |
597 | */ | |
8922e16c | 598 | if (blk_account_rq(rq)) { |
1da177e4 | 599 | q->in_flight--; |
8922e16c TH |
600 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
601 | e->ops->elevator_completed_req_fn(q, rq); | |
602 | } | |
1da177e4 LT |
603 | } |
604 | ||
605 | int elv_register_queue(struct request_queue *q) | |
606 | { | |
607 | elevator_t *e = q->elevator; | |
608 | ||
609 | e->kobj.parent = kobject_get(&q->kobj); | |
610 | if (!e->kobj.parent) | |
611 | return -EBUSY; | |
612 | ||
613 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
614 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
615 | ||
616 | return kobject_register(&e->kobj); | |
617 | } | |
618 | ||
619 | void elv_unregister_queue(struct request_queue *q) | |
620 | { | |
621 | if (q) { | |
622 | elevator_t *e = q->elevator; | |
623 | kobject_unregister(&e->kobj); | |
624 | kobject_put(&q->kobj); | |
625 | } | |
626 | } | |
627 | ||
628 | int elv_register(struct elevator_type *e) | |
629 | { | |
2824bc93 | 630 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
631 | if (elevator_find(e->elevator_name)) |
632 | BUG(); | |
1da177e4 LT |
633 | list_add_tail(&e->list, &elv_list); |
634 | spin_unlock_irq(&elv_list_lock); | |
635 | ||
636 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
637 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
638 | printk(" (default)"); | |
639 | printk("\n"); | |
640 | return 0; | |
641 | } | |
642 | EXPORT_SYMBOL_GPL(elv_register); | |
643 | ||
644 | void elv_unregister(struct elevator_type *e) | |
645 | { | |
646 | spin_lock_irq(&elv_list_lock); | |
647 | list_del_init(&e->list); | |
648 | spin_unlock_irq(&elv_list_lock); | |
649 | } | |
650 | EXPORT_SYMBOL_GPL(elv_unregister); | |
651 | ||
652 | /* | |
653 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
654 | * we don't free the old io scheduler, before we have allocated what we | |
655 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 656 | * one, if the new one fails init for some reason. |
1da177e4 LT |
657 | */ |
658 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
659 | { | |
cb98fc8b | 660 | elevator_t *old_elevator, *e; |
1da177e4 | 661 | |
cb98fc8b TH |
662 | /* |
663 | * Allocate new elevator | |
664 | */ | |
665 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
666 | if (!e) |
667 | goto error; | |
668 | ||
669 | /* | |
cb98fc8b | 670 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 671 | */ |
cb98fc8b TH |
672 | spin_lock_irq(q->queue_lock); |
673 | ||
674 | set_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); | |
675 | ||
676 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
677 | ; | |
678 | ||
679 | while (q->rq.elvpriv) { | |
680 | spin_unlock_irq(q->queue_lock); | |
681 | msleep(100); | |
682 | spin_lock_irq(q->queue_lock); | |
683 | } | |
684 | ||
685 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
686 | |
687 | /* | |
688 | * unregister old elevator data | |
689 | */ | |
690 | elv_unregister_queue(q); | |
691 | old_elevator = q->elevator; | |
692 | ||
1da177e4 LT |
693 | /* |
694 | * attach and start new elevator | |
695 | */ | |
696 | if (elevator_attach(q, new_e, e)) | |
697 | goto fail; | |
698 | ||
699 | if (elv_register_queue(q)) | |
700 | goto fail_register; | |
701 | ||
702 | /* | |
cb98fc8b | 703 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
704 | */ |
705 | elevator_exit(old_elevator); | |
cb98fc8b | 706 | clear_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); |
1da177e4 LT |
707 | return; |
708 | ||
709 | fail_register: | |
710 | /* | |
711 | * switch failed, exit the new io scheduler and reattach the old | |
712 | * one again (along with re-adding the sysfs dir) | |
713 | */ | |
714 | elevator_exit(e); | |
cb98fc8b | 715 | e = NULL; |
1da177e4 LT |
716 | fail: |
717 | q->elevator = old_elevator; | |
718 | elv_register_queue(q); | |
cb98fc8b TH |
719 | clear_bit(QUEUE_FLAG_BYPASS, &q->queue_flags); |
720 | kfree(e); | |
1da177e4 | 721 | error: |
1da177e4 LT |
722 | elevator_put(new_e); |
723 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
724 | } | |
725 | ||
726 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
727 | { | |
728 | char elevator_name[ELV_NAME_MAX]; | |
729 | struct elevator_type *e; | |
730 | ||
731 | memset(elevator_name, 0, sizeof(elevator_name)); | |
732 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
733 | ||
734 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
735 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
736 | ||
737 | e = elevator_get(elevator_name); | |
738 | if (!e) { | |
739 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
740 | return -EINVAL; | |
741 | } | |
742 | ||
743 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) | |
744 | return count; | |
745 | ||
746 | elevator_switch(q, e); | |
747 | return count; | |
748 | } | |
749 | ||
750 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
751 | { | |
752 | elevator_t *e = q->elevator; | |
753 | struct elevator_type *elv = e->elevator_type; | |
754 | struct list_head *entry; | |
755 | int len = 0; | |
756 | ||
757 | spin_lock_irq(q->queue_lock); | |
758 | list_for_each(entry, &elv_list) { | |
759 | struct elevator_type *__e; | |
760 | ||
761 | __e = list_entry(entry, struct elevator_type, list); | |
762 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
763 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
764 | else | |
765 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
766 | } | |
767 | spin_unlock_irq(q->queue_lock); | |
768 | ||
769 | len += sprintf(len+name, "\n"); | |
770 | return len; | |
771 | } | |
772 | ||
1b47f531 | 773 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
774 | EXPORT_SYMBOL(elv_add_request); |
775 | EXPORT_SYMBOL(__elv_add_request); | |
776 | EXPORT_SYMBOL(elv_requeue_request); | |
777 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 778 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
779 | EXPORT_SYMBOL(elv_queue_empty); |
780 | EXPORT_SYMBOL(elv_completed_request); | |
781 | EXPORT_SYMBOL(elevator_exit); | |
782 | EXPORT_SYMBOL(elevator_init); |