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; | |
cb98fc8b | 188 | |
1da177e4 LT |
189 | elevator_setup_default(); |
190 | ||
191 | if (!name) | |
192 | name = chosen_elevator; | |
193 | ||
194 | e = elevator_get(name); | |
195 | if (!e) | |
196 | return -EINVAL; | |
197 | ||
198 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
199 | if (!eq) { | |
200 | elevator_put(e->elevator_type); | |
201 | return -ENOMEM; | |
202 | } | |
203 | ||
204 | ret = elevator_attach(q, e, eq); | |
205 | if (ret) { | |
206 | kfree(eq); | |
207 | elevator_put(e->elevator_type); | |
208 | } | |
209 | ||
210 | return ret; | |
211 | } | |
212 | ||
213 | void elevator_exit(elevator_t *e) | |
214 | { | |
215 | if (e->ops->elevator_exit_fn) | |
216 | e->ops->elevator_exit_fn(e); | |
217 | ||
218 | elevator_put(e->elevator_type); | |
219 | e->elevator_type = NULL; | |
220 | kfree(e); | |
221 | } | |
222 | ||
8922e16c TH |
223 | /* |
224 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
225 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
226 | * appended to the dispatch queue. To be used by specific elevators. | |
227 | */ | |
1b47f531 | 228 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
229 | { |
230 | sector_t boundary; | |
8922e16c TH |
231 | struct list_head *entry; |
232 | ||
06b86245 TH |
233 | if (q->last_merge == rq) |
234 | q->last_merge = NULL; | |
235 | ||
1b47f531 | 236 | boundary = q->end_sector; |
cb19833d | 237 | |
8922e16c TH |
238 | list_for_each_prev(entry, &q->queue_head) { |
239 | struct request *pos = list_entry_rq(entry); | |
240 | ||
241 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
242 | break; | |
243 | if (rq->sector >= boundary) { | |
244 | if (pos->sector < boundary) | |
245 | continue; | |
246 | } else { | |
247 | if (pos->sector >= boundary) | |
248 | break; | |
249 | } | |
250 | if (rq->sector >= pos->sector) | |
251 | break; | |
252 | } | |
253 | ||
254 | list_add(&rq->queuelist, entry); | |
255 | } | |
256 | ||
1da177e4 LT |
257 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
258 | { | |
259 | elevator_t *e = q->elevator; | |
06b86245 TH |
260 | int ret; |
261 | ||
262 | if (q->last_merge) { | |
263 | ret = elv_try_merge(q->last_merge, bio); | |
264 | if (ret != ELEVATOR_NO_MERGE) { | |
265 | *req = q->last_merge; | |
266 | return ret; | |
267 | } | |
268 | } | |
1da177e4 LT |
269 | |
270 | if (e->ops->elevator_merge_fn) | |
271 | return e->ops->elevator_merge_fn(q, req, bio); | |
272 | ||
273 | return ELEVATOR_NO_MERGE; | |
274 | } | |
275 | ||
276 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
277 | { | |
278 | elevator_t *e = q->elevator; | |
279 | ||
280 | if (e->ops->elevator_merged_fn) | |
281 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
282 | |
283 | q->last_merge = rq; | |
1da177e4 LT |
284 | } |
285 | ||
286 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
287 | struct request *next) | |
288 | { | |
289 | elevator_t *e = q->elevator; | |
290 | ||
1da177e4 LT |
291 | if (e->ops->elevator_merge_req_fn) |
292 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 TH |
293 | |
294 | q->last_merge = rq; | |
1da177e4 LT |
295 | } |
296 | ||
8922e16c | 297 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
298 | { |
299 | elevator_t *e = q->elevator; | |
300 | ||
301 | /* | |
302 | * it already went through dequeue, we need to decrement the | |
303 | * in_flight count again | |
304 | */ | |
8922e16c | 305 | if (blk_account_rq(rq)) { |
1da177e4 | 306 | q->in_flight--; |
8922e16c TH |
307 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
308 | e->ops->elevator_deactivate_req_fn(q, rq); | |
309 | } | |
1da177e4 LT |
310 | |
311 | rq->flags &= ~REQ_STARTED; | |
312 | ||
1da177e4 LT |
313 | /* |
314 | * if this is the flush, requeue the original instead and drop the flush | |
315 | */ | |
316 | if (rq->flags & REQ_BAR_FLUSH) { | |
317 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
318 | rq = rq->end_io_data; | |
319 | } | |
320 | ||
8922e16c | 321 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
322 | } |
323 | ||
324 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
325 | int plug) | |
326 | { | |
8922e16c TH |
327 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
328 | /* | |
329 | * barriers implicitly indicate back insertion | |
330 | */ | |
331 | if (where == ELEVATOR_INSERT_SORT) | |
332 | where = ELEVATOR_INSERT_BACK; | |
333 | ||
334 | /* | |
1b47f531 | 335 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
336 | */ |
337 | if (blk_fs_request(rq)) { | |
1b47f531 | 338 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
339 | q->boundary_rq = rq; |
340 | } | |
cb98fc8b TH |
341 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
342 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
343 | |
344 | if (plug) | |
345 | blk_plug_device(q); | |
346 | ||
347 | rq->q = q; | |
348 | ||
8922e16c TH |
349 | switch (where) { |
350 | case ELEVATOR_INSERT_FRONT: | |
351 | rq->flags |= REQ_SOFTBARRIER; | |
352 | ||
353 | list_add(&rq->queuelist, &q->queue_head); | |
354 | break; | |
355 | ||
356 | case ELEVATOR_INSERT_BACK: | |
357 | rq->flags |= REQ_SOFTBARRIER; | |
358 | ||
359 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
360 | ; | |
361 | list_add_tail(&rq->queuelist, &q->queue_head); | |
362 | /* | |
363 | * We kick the queue here for the following reasons. | |
364 | * - The elevator might have returned NULL previously | |
365 | * to delay requests and returned them now. As the | |
366 | * queue wasn't empty before this request, ll_rw_blk | |
367 | * won't run the queue on return, resulting in hang. | |
368 | * - Usually, back inserted requests won't be merged | |
369 | * with anything. There's no point in delaying queue | |
370 | * processing. | |
371 | */ | |
372 | blk_remove_plug(q); | |
373 | q->request_fn(q); | |
374 | break; | |
375 | ||
376 | case ELEVATOR_INSERT_SORT: | |
377 | BUG_ON(!blk_fs_request(rq)); | |
378 | rq->flags |= REQ_SORTED; | |
379 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
06b86245 TH |
380 | if (q->last_merge == NULL && rq_mergeable(rq)) |
381 | q->last_merge = rq; | |
8922e16c TH |
382 | break; |
383 | ||
384 | default: | |
385 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
386 | __FUNCTION__, where); | |
387 | BUG(); | |
388 | } | |
389 | ||
390 | if (blk_queue_plugged(q)) { | |
391 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
392 | - q->in_flight; | |
393 | ||
394 | if (nrq >= q->unplug_thresh) | |
395 | __generic_unplug_device(q); | |
396 | } | |
1da177e4 LT |
397 | } |
398 | ||
399 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
400 | int plug) | |
401 | { | |
402 | unsigned long flags; | |
403 | ||
404 | spin_lock_irqsave(q->queue_lock, flags); | |
405 | __elv_add_request(q, rq, where, plug); | |
406 | spin_unlock_irqrestore(q->queue_lock, flags); | |
407 | } | |
408 | ||
409 | static inline struct request *__elv_next_request(request_queue_t *q) | |
410 | { | |
8922e16c TH |
411 | struct request *rq; |
412 | ||
413 | if (unlikely(list_empty(&q->queue_head) && | |
414 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
415 | return NULL; | |
416 | ||
417 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
418 | |
419 | /* | |
420 | * if this is a barrier write and the device has to issue a | |
421 | * flush sequence to support it, check how far we are | |
422 | */ | |
8922e16c | 423 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
424 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
425 | ||
426 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
427 | !blk_barrier_preflush(rq)) | |
428 | rq = blk_start_pre_flush(q, rq); | |
429 | } | |
430 | ||
431 | return rq; | |
432 | } | |
433 | ||
434 | struct request *elv_next_request(request_queue_t *q) | |
435 | { | |
436 | struct request *rq; | |
437 | int ret; | |
438 | ||
439 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
440 | if (!(rq->flags & REQ_STARTED)) { |
441 | elevator_t *e = q->elevator; | |
442 | ||
443 | /* | |
444 | * This is the first time the device driver | |
445 | * sees this request (possibly after | |
446 | * requeueing). Notify IO scheduler. | |
447 | */ | |
448 | if (blk_sorted_rq(rq) && | |
449 | e->ops->elevator_activate_req_fn) | |
450 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 451 | |
8922e16c TH |
452 | /* |
453 | * just mark as started even if we don't start | |
454 | * it, a request that has been delayed should | |
455 | * not be passed by new incoming requests | |
456 | */ | |
457 | rq->flags |= REQ_STARTED; | |
458 | } | |
1da177e4 | 459 | |
8922e16c | 460 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 461 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
462 | q->boundary_rq = NULL; |
463 | } | |
1da177e4 LT |
464 | |
465 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
466 | break; | |
467 | ||
468 | ret = q->prep_rq_fn(q, rq); | |
469 | if (ret == BLKPREP_OK) { | |
470 | break; | |
471 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
472 | /* |
473 | * the request may have been (partially) prepped. | |
474 | * we need to keep this request in the front to | |
8922e16c TH |
475 | * avoid resource deadlock. REQ_STARTED will |
476 | * prevent other fs requests from passing this one. | |
2e759cd4 | 477 | */ |
1da177e4 LT |
478 | rq = NULL; |
479 | break; | |
480 | } else if (ret == BLKPREP_KILL) { | |
481 | int nr_bytes = rq->hard_nr_sectors << 9; | |
482 | ||
483 | if (!nr_bytes) | |
484 | nr_bytes = rq->data_len; | |
485 | ||
486 | blkdev_dequeue_request(rq); | |
487 | rq->flags |= REQ_QUIET; | |
488 | end_that_request_chunk(rq, 0, nr_bytes); | |
489 | end_that_request_last(rq); | |
490 | } else { | |
491 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
492 | ret); | |
493 | break; | |
494 | } | |
495 | } | |
496 | ||
497 | return rq; | |
498 | } | |
499 | ||
8922e16c | 500 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 501 | { |
8922e16c TH |
502 | BUG_ON(list_empty(&rq->queuelist)); |
503 | ||
504 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
505 | |
506 | /* | |
507 | * the time frame between a request being removed from the lists | |
508 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 509 | * the driver side. |
1da177e4 LT |
510 | */ |
511 | if (blk_account_rq(rq)) | |
512 | q->in_flight++; | |
1da177e4 LT |
513 | } |
514 | ||
515 | int elv_queue_empty(request_queue_t *q) | |
516 | { | |
517 | elevator_t *e = q->elevator; | |
518 | ||
8922e16c TH |
519 | if (!list_empty(&q->queue_head)) |
520 | return 0; | |
521 | ||
1da177e4 LT |
522 | if (e->ops->elevator_queue_empty_fn) |
523 | return e->ops->elevator_queue_empty_fn(q); | |
524 | ||
8922e16c | 525 | return 1; |
1da177e4 LT |
526 | } |
527 | ||
528 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
529 | { | |
530 | struct list_head *next; | |
531 | ||
532 | elevator_t *e = q->elevator; | |
533 | ||
534 | if (e->ops->elevator_latter_req_fn) | |
535 | return e->ops->elevator_latter_req_fn(q, rq); | |
536 | ||
537 | next = rq->queuelist.next; | |
538 | if (next != &q->queue_head && next != &rq->queuelist) | |
539 | return list_entry_rq(next); | |
540 | ||
541 | return NULL; | |
542 | } | |
543 | ||
544 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
545 | { | |
546 | struct list_head *prev; | |
547 | ||
548 | elevator_t *e = q->elevator; | |
549 | ||
550 | if (e->ops->elevator_former_req_fn) | |
551 | return e->ops->elevator_former_req_fn(q, rq); | |
552 | ||
553 | prev = rq->queuelist.prev; | |
554 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
555 | return list_entry_rq(prev); | |
556 | ||
557 | return NULL; | |
558 | } | |
559 | ||
22e2c507 | 560 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 561 | gfp_t gfp_mask) |
1da177e4 LT |
562 | { |
563 | elevator_t *e = q->elevator; | |
564 | ||
565 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 566 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
567 | |
568 | rq->elevator_private = NULL; | |
569 | return 0; | |
570 | } | |
571 | ||
572 | void elv_put_request(request_queue_t *q, struct request *rq) | |
573 | { | |
574 | elevator_t *e = q->elevator; | |
575 | ||
576 | if (e->ops->elevator_put_req_fn) | |
577 | e->ops->elevator_put_req_fn(q, rq); | |
578 | } | |
579 | ||
22e2c507 | 580 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
581 | { |
582 | elevator_t *e = q->elevator; | |
583 | ||
584 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 585 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
586 | |
587 | return ELV_MQUEUE_MAY; | |
588 | } | |
589 | ||
590 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
591 | { | |
592 | elevator_t *e = q->elevator; | |
593 | ||
594 | /* | |
595 | * request is released from the driver, io must be done | |
596 | */ | |
8922e16c | 597 | if (blk_account_rq(rq)) { |
1da177e4 | 598 | q->in_flight--; |
8922e16c TH |
599 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
600 | e->ops->elevator_completed_req_fn(q, rq); | |
601 | } | |
1da177e4 LT |
602 | } |
603 | ||
604 | int elv_register_queue(struct request_queue *q) | |
605 | { | |
606 | elevator_t *e = q->elevator; | |
607 | ||
608 | e->kobj.parent = kobject_get(&q->kobj); | |
609 | if (!e->kobj.parent) | |
610 | return -EBUSY; | |
611 | ||
612 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
613 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
614 | ||
615 | return kobject_register(&e->kobj); | |
616 | } | |
617 | ||
618 | void elv_unregister_queue(struct request_queue *q) | |
619 | { | |
620 | if (q) { | |
621 | elevator_t *e = q->elevator; | |
622 | kobject_unregister(&e->kobj); | |
623 | kobject_put(&q->kobj); | |
624 | } | |
625 | } | |
626 | ||
627 | int elv_register(struct elevator_type *e) | |
628 | { | |
2824bc93 | 629 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
630 | if (elevator_find(e->elevator_name)) |
631 | BUG(); | |
1da177e4 LT |
632 | list_add_tail(&e->list, &elv_list); |
633 | spin_unlock_irq(&elv_list_lock); | |
634 | ||
635 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
636 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
637 | printk(" (default)"); | |
638 | printk("\n"); | |
639 | return 0; | |
640 | } | |
641 | EXPORT_SYMBOL_GPL(elv_register); | |
642 | ||
643 | void elv_unregister(struct elevator_type *e) | |
644 | { | |
645 | spin_lock_irq(&elv_list_lock); | |
646 | list_del_init(&e->list); | |
647 | spin_unlock_irq(&elv_list_lock); | |
648 | } | |
649 | EXPORT_SYMBOL_GPL(elv_unregister); | |
650 | ||
651 | /* | |
652 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
653 | * we don't free the old io scheduler, before we have allocated what we | |
654 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 655 | * one, if the new one fails init for some reason. |
1da177e4 LT |
656 | */ |
657 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
658 | { | |
cb98fc8b | 659 | elevator_t *old_elevator, *e; |
1da177e4 | 660 | |
cb98fc8b TH |
661 | /* |
662 | * Allocate new elevator | |
663 | */ | |
664 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
665 | if (!e) |
666 | goto error; | |
667 | ||
668 | /* | |
cb98fc8b | 669 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 670 | */ |
cb98fc8b TH |
671 | spin_lock_irq(q->queue_lock); |
672 | ||
64521d1a | 673 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
674 | |
675 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
676 | ; | |
677 | ||
678 | while (q->rq.elvpriv) { | |
679 | spin_unlock_irq(q->queue_lock); | |
64521d1a | 680 | msleep(10); |
cb98fc8b TH |
681 | spin_lock_irq(q->queue_lock); |
682 | } | |
683 | ||
684 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
685 | |
686 | /* | |
687 | * unregister old elevator data | |
688 | */ | |
689 | elv_unregister_queue(q); | |
690 | old_elevator = q->elevator; | |
691 | ||
1da177e4 LT |
692 | /* |
693 | * attach and start new elevator | |
694 | */ | |
695 | if (elevator_attach(q, new_e, e)) | |
696 | goto fail; | |
697 | ||
698 | if (elv_register_queue(q)) | |
699 | goto fail_register; | |
700 | ||
701 | /* | |
cb98fc8b | 702 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
703 | */ |
704 | elevator_exit(old_elevator); | |
64521d1a | 705 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
706 | return; |
707 | ||
708 | fail_register: | |
709 | /* | |
710 | * switch failed, exit the new io scheduler and reattach the old | |
711 | * one again (along with re-adding the sysfs dir) | |
712 | */ | |
713 | elevator_exit(e); | |
cb98fc8b | 714 | e = NULL; |
1da177e4 LT |
715 | fail: |
716 | q->elevator = old_elevator; | |
717 | elv_register_queue(q); | |
64521d1a | 718 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 719 | kfree(e); |
1da177e4 | 720 | error: |
1da177e4 LT |
721 | elevator_put(new_e); |
722 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
723 | } | |
724 | ||
725 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
726 | { | |
727 | char elevator_name[ELV_NAME_MAX]; | |
728 | struct elevator_type *e; | |
729 | ||
730 | memset(elevator_name, 0, sizeof(elevator_name)); | |
731 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
732 | ||
733 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
734 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
735 | ||
736 | e = elevator_get(elevator_name); | |
737 | if (!e) { | |
738 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
739 | return -EINVAL; | |
740 | } | |
741 | ||
742 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) | |
743 | return count; | |
744 | ||
745 | elevator_switch(q, e); | |
746 | return count; | |
747 | } | |
748 | ||
749 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
750 | { | |
751 | elevator_t *e = q->elevator; | |
752 | struct elevator_type *elv = e->elevator_type; | |
753 | struct list_head *entry; | |
754 | int len = 0; | |
755 | ||
756 | spin_lock_irq(q->queue_lock); | |
757 | list_for_each(entry, &elv_list) { | |
758 | struct elevator_type *__e; | |
759 | ||
760 | __e = list_entry(entry, struct elevator_type, list); | |
761 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
762 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
763 | else | |
764 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
765 | } | |
766 | spin_unlock_irq(q->queue_lock); | |
767 | ||
768 | len += sprintf(len+name, "\n"); | |
769 | return len; | |
770 | } | |
771 | ||
1b47f531 | 772 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
773 | EXPORT_SYMBOL(elv_add_request); |
774 | EXPORT_SYMBOL(__elv_add_request); | |
775 | EXPORT_SYMBOL(elv_requeue_request); | |
776 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 777 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
778 | EXPORT_SYMBOL(elv_queue_empty); |
779 | EXPORT_SYMBOL(elv_completed_request); | |
780 | EXPORT_SYMBOL(elevator_exit); | |
781 | EXPORT_SYMBOL(elevator_init); |