]>
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> | |
37 | ||
38 | #include <asm/uaccess.h> | |
39 | ||
40 | static DEFINE_SPINLOCK(elv_list_lock); | |
41 | static LIST_HEAD(elv_list); | |
42 | ||
8922e16c TH |
43 | static inline sector_t rq_last_sector(struct request *rq) |
44 | { | |
45 | return rq->sector + rq->nr_sectors; | |
46 | } | |
47 | ||
1da177e4 LT |
48 | /* |
49 | * can we safely merge with this request? | |
50 | */ | |
51 | inline int elv_rq_merge_ok(struct request *rq, struct bio *bio) | |
52 | { | |
53 | if (!rq_mergeable(rq)) | |
54 | return 0; | |
55 | ||
56 | /* | |
57 | * different data direction or already started, don't merge | |
58 | */ | |
59 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
60 | return 0; | |
61 | ||
62 | /* | |
63 | * same device and no special stuff set, merge is ok | |
64 | */ | |
65 | if (rq->rq_disk == bio->bi_bdev->bd_disk && | |
66 | !rq->waiting && !rq->special) | |
67 | return 1; | |
68 | ||
69 | return 0; | |
70 | } | |
71 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
72 | ||
73 | inline int elv_try_merge(struct request *__rq, struct bio *bio) | |
74 | { | |
75 | int ret = ELEVATOR_NO_MERGE; | |
76 | ||
77 | /* | |
78 | * we can merge and sequence is ok, check if it's possible | |
79 | */ | |
80 | if (elv_rq_merge_ok(__rq, bio)) { | |
81 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
82 | ret = ELEVATOR_BACK_MERGE; | |
83 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
84 | ret = ELEVATOR_FRONT_MERGE; | |
85 | } | |
86 | ||
87 | return ret; | |
88 | } | |
89 | EXPORT_SYMBOL(elv_try_merge); | |
90 | ||
91 | inline int elv_try_last_merge(request_queue_t *q, struct bio *bio) | |
92 | { | |
93 | if (q->last_merge) | |
94 | return elv_try_merge(q->last_merge, bio); | |
95 | ||
96 | return ELEVATOR_NO_MERGE; | |
97 | } | |
98 | EXPORT_SYMBOL(elv_try_last_merge); | |
99 | ||
100 | static struct elevator_type *elevator_find(const char *name) | |
101 | { | |
102 | struct elevator_type *e = NULL; | |
103 | struct list_head *entry; | |
104 | ||
1da177e4 LT |
105 | list_for_each(entry, &elv_list) { |
106 | struct elevator_type *__e; | |
107 | ||
108 | __e = list_entry(entry, struct elevator_type, list); | |
109 | ||
110 | if (!strcmp(__e->elevator_name, name)) { | |
111 | e = __e; | |
112 | break; | |
113 | } | |
114 | } | |
1da177e4 LT |
115 | |
116 | return e; | |
117 | } | |
118 | ||
119 | static void elevator_put(struct elevator_type *e) | |
120 | { | |
121 | module_put(e->elevator_owner); | |
122 | } | |
123 | ||
124 | static struct elevator_type *elevator_get(const char *name) | |
125 | { | |
2824bc93 | 126 | struct elevator_type *e; |
1da177e4 | 127 | |
2824bc93 TH |
128 | spin_lock_irq(&elv_list_lock); |
129 | ||
130 | e = elevator_find(name); | |
131 | if (e && !try_module_get(e->elevator_owner)) | |
132 | e = NULL; | |
133 | ||
134 | spin_unlock_irq(&elv_list_lock); | |
1da177e4 LT |
135 | |
136 | return e; | |
137 | } | |
138 | ||
139 | static int elevator_attach(request_queue_t *q, struct elevator_type *e, | |
140 | struct elevator_queue *eq) | |
141 | { | |
142 | int ret = 0; | |
143 | ||
144 | memset(eq, 0, sizeof(*eq)); | |
145 | eq->ops = &e->ops; | |
146 | eq->elevator_type = e; | |
147 | ||
148 | INIT_LIST_HEAD(&q->queue_head); | |
149 | q->last_merge = NULL; | |
150 | q->elevator = eq; | |
8922e16c TH |
151 | q->last_sector = 0; |
152 | q->boundary_rq = NULL; | |
153 | q->max_back_kb = 0; | |
1da177e4 LT |
154 | |
155 | if (eq->ops->elevator_init_fn) | |
156 | ret = eq->ops->elevator_init_fn(q, eq); | |
157 | ||
158 | return ret; | |
159 | } | |
160 | ||
161 | static char chosen_elevator[16]; | |
162 | ||
163 | static void elevator_setup_default(void) | |
164 | { | |
2824bc93 TH |
165 | struct elevator_type *e; |
166 | ||
1da177e4 LT |
167 | /* |
168 | * check if default is set and exists | |
169 | */ | |
2824bc93 TH |
170 | if (chosen_elevator[0] && (e = elevator_get(chosen_elevator))) { |
171 | elevator_put(e); | |
1da177e4 | 172 | return; |
2824bc93 | 173 | } |
1da177e4 LT |
174 | |
175 | #if defined(CONFIG_IOSCHED_AS) | |
176 | strcpy(chosen_elevator, "anticipatory"); | |
177 | #elif defined(CONFIG_IOSCHED_DEADLINE) | |
178 | strcpy(chosen_elevator, "deadline"); | |
179 | #elif defined(CONFIG_IOSCHED_CFQ) | |
180 | strcpy(chosen_elevator, "cfq"); | |
181 | #elif defined(CONFIG_IOSCHED_NOOP) | |
182 | strcpy(chosen_elevator, "noop"); | |
183 | #else | |
184 | #error "You must build at least 1 IO scheduler into the kernel" | |
185 | #endif | |
186 | } | |
187 | ||
188 | static int __init elevator_setup(char *str) | |
189 | { | |
190 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
191 | return 0; | |
192 | } | |
193 | ||
194 | __setup("elevator=", elevator_setup); | |
195 | ||
196 | int elevator_init(request_queue_t *q, char *name) | |
197 | { | |
198 | struct elevator_type *e = NULL; | |
199 | struct elevator_queue *eq; | |
200 | int ret = 0; | |
201 | ||
202 | elevator_setup_default(); | |
203 | ||
204 | if (!name) | |
205 | name = chosen_elevator; | |
206 | ||
207 | e = elevator_get(name); | |
208 | if (!e) | |
209 | return -EINVAL; | |
210 | ||
211 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
212 | if (!eq) { | |
213 | elevator_put(e->elevator_type); | |
214 | return -ENOMEM; | |
215 | } | |
216 | ||
217 | ret = elevator_attach(q, e, eq); | |
218 | if (ret) { | |
219 | kfree(eq); | |
220 | elevator_put(e->elevator_type); | |
221 | } | |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
226 | void elevator_exit(elevator_t *e) | |
227 | { | |
228 | if (e->ops->elevator_exit_fn) | |
229 | e->ops->elevator_exit_fn(e); | |
230 | ||
231 | elevator_put(e->elevator_type); | |
232 | e->elevator_type = NULL; | |
233 | kfree(e); | |
234 | } | |
235 | ||
8922e16c TH |
236 | /* |
237 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
238 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
239 | * appended to the dispatch queue. To be used by specific elevators. | |
240 | */ | |
241 | void elv_dispatch_insert(request_queue_t *q, struct request *rq, int sort) | |
242 | { | |
243 | sector_t boundary; | |
244 | unsigned max_back; | |
245 | struct list_head *entry; | |
246 | ||
247 | if (!sort) { | |
248 | /* Specific elevator is performing sort. Step away. */ | |
249 | q->last_sector = rq_last_sector(rq); | |
250 | q->boundary_rq = rq; | |
251 | list_add_tail(&rq->queuelist, &q->queue_head); | |
252 | return; | |
253 | } | |
254 | ||
255 | boundary = q->last_sector; | |
256 | max_back = q->max_back_kb * 2; | |
257 | boundary = boundary > max_back ? boundary - max_back : 0; | |
258 | ||
259 | list_for_each_prev(entry, &q->queue_head) { | |
260 | struct request *pos = list_entry_rq(entry); | |
261 | ||
262 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
263 | break; | |
264 | if (rq->sector >= boundary) { | |
265 | if (pos->sector < boundary) | |
266 | continue; | |
267 | } else { | |
268 | if (pos->sector >= boundary) | |
269 | break; | |
270 | } | |
271 | if (rq->sector >= pos->sector) | |
272 | break; | |
273 | } | |
274 | ||
275 | list_add(&rq->queuelist, entry); | |
276 | } | |
277 | ||
1da177e4 LT |
278 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
279 | { | |
280 | elevator_t *e = q->elevator; | |
281 | ||
282 | if (e->ops->elevator_merge_fn) | |
283 | return e->ops->elevator_merge_fn(q, req, bio); | |
284 | ||
285 | return ELEVATOR_NO_MERGE; | |
286 | } | |
287 | ||
288 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
289 | { | |
290 | elevator_t *e = q->elevator; | |
291 | ||
292 | if (e->ops->elevator_merged_fn) | |
293 | e->ops->elevator_merged_fn(q, rq); | |
294 | } | |
295 | ||
296 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
297 | struct request *next) | |
298 | { | |
299 | elevator_t *e = q->elevator; | |
300 | ||
301 | if (q->last_merge == next) | |
302 | q->last_merge = NULL; | |
303 | ||
304 | if (e->ops->elevator_merge_req_fn) | |
305 | e->ops->elevator_merge_req_fn(q, rq, next); | |
306 | } | |
307 | ||
8922e16c | 308 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
309 | { |
310 | elevator_t *e = q->elevator; | |
311 | ||
312 | /* | |
313 | * it already went through dequeue, we need to decrement the | |
314 | * in_flight count again | |
315 | */ | |
8922e16c | 316 | if (blk_account_rq(rq)) { |
1da177e4 | 317 | q->in_flight--; |
8922e16c TH |
318 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
319 | e->ops->elevator_deactivate_req_fn(q, rq); | |
320 | } | |
1da177e4 LT |
321 | |
322 | rq->flags &= ~REQ_STARTED; | |
323 | ||
1da177e4 LT |
324 | /* |
325 | * if this is the flush, requeue the original instead and drop the flush | |
326 | */ | |
327 | if (rq->flags & REQ_BAR_FLUSH) { | |
328 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
329 | rq = rq->end_io_data; | |
330 | } | |
331 | ||
8922e16c | 332 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
333 | } |
334 | ||
335 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
336 | int plug) | |
337 | { | |
8922e16c TH |
338 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
339 | /* | |
340 | * barriers implicitly indicate back insertion | |
341 | */ | |
342 | if (where == ELEVATOR_INSERT_SORT) | |
343 | where = ELEVATOR_INSERT_BACK; | |
344 | ||
345 | /* | |
346 | * this request is scheduling boundary, update last_sector | |
347 | */ | |
348 | if (blk_fs_request(rq)) { | |
349 | q->last_sector = rq_last_sector(rq); | |
350 | q->boundary_rq = rq; | |
351 | } | |
352 | } | |
1da177e4 LT |
353 | |
354 | if (plug) | |
355 | blk_plug_device(q); | |
356 | ||
357 | rq->q = q; | |
358 | ||
8922e16c | 359 | if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) { |
1da177e4 LT |
360 | /* |
361 | * if drain is set, store the request "locally". when the drain | |
362 | * is finished, the requests will be handed ordered to the io | |
363 | * scheduler | |
364 | */ | |
365 | list_add_tail(&rq->queuelist, &q->drain_list); | |
8922e16c TH |
366 | return; |
367 | } | |
368 | ||
369 | switch (where) { | |
370 | case ELEVATOR_INSERT_FRONT: | |
371 | rq->flags |= REQ_SOFTBARRIER; | |
372 | ||
373 | list_add(&rq->queuelist, &q->queue_head); | |
374 | break; | |
375 | ||
376 | case ELEVATOR_INSERT_BACK: | |
377 | rq->flags |= REQ_SOFTBARRIER; | |
378 | ||
379 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
380 | ; | |
381 | list_add_tail(&rq->queuelist, &q->queue_head); | |
382 | /* | |
383 | * We kick the queue here for the following reasons. | |
384 | * - The elevator might have returned NULL previously | |
385 | * to delay requests and returned them now. As the | |
386 | * queue wasn't empty before this request, ll_rw_blk | |
387 | * won't run the queue on return, resulting in hang. | |
388 | * - Usually, back inserted requests won't be merged | |
389 | * with anything. There's no point in delaying queue | |
390 | * processing. | |
391 | */ | |
392 | blk_remove_plug(q); | |
393 | q->request_fn(q); | |
394 | break; | |
395 | ||
396 | case ELEVATOR_INSERT_SORT: | |
397 | BUG_ON(!blk_fs_request(rq)); | |
398 | rq->flags |= REQ_SORTED; | |
399 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
400 | break; | |
401 | ||
402 | default: | |
403 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
404 | __FUNCTION__, where); | |
405 | BUG(); | |
406 | } | |
407 | ||
408 | if (blk_queue_plugged(q)) { | |
409 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
410 | - q->in_flight; | |
411 | ||
412 | if (nrq >= q->unplug_thresh) | |
413 | __generic_unplug_device(q); | |
414 | } | |
1da177e4 LT |
415 | } |
416 | ||
417 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
418 | int plug) | |
419 | { | |
420 | unsigned long flags; | |
421 | ||
422 | spin_lock_irqsave(q->queue_lock, flags); | |
423 | __elv_add_request(q, rq, where, plug); | |
424 | spin_unlock_irqrestore(q->queue_lock, flags); | |
425 | } | |
426 | ||
427 | static inline struct request *__elv_next_request(request_queue_t *q) | |
428 | { | |
8922e16c TH |
429 | struct request *rq; |
430 | ||
431 | if (unlikely(list_empty(&q->queue_head) && | |
432 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
433 | return NULL; | |
434 | ||
435 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
436 | |
437 | /* | |
438 | * if this is a barrier write and the device has to issue a | |
439 | * flush sequence to support it, check how far we are | |
440 | */ | |
8922e16c | 441 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
442 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
443 | ||
444 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
445 | !blk_barrier_preflush(rq)) | |
446 | rq = blk_start_pre_flush(q, rq); | |
447 | } | |
448 | ||
449 | return rq; | |
450 | } | |
451 | ||
452 | struct request *elv_next_request(request_queue_t *q) | |
453 | { | |
454 | struct request *rq; | |
455 | int ret; | |
456 | ||
457 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
458 | if (!(rq->flags & REQ_STARTED)) { |
459 | elevator_t *e = q->elevator; | |
460 | ||
461 | /* | |
462 | * This is the first time the device driver | |
463 | * sees this request (possibly after | |
464 | * requeueing). Notify IO scheduler. | |
465 | */ | |
466 | if (blk_sorted_rq(rq) && | |
467 | e->ops->elevator_activate_req_fn) | |
468 | e->ops->elevator_activate_req_fn(q, rq); | |
469 | ||
470 | /* | |
471 | * just mark as started even if we don't start | |
472 | * it, a request that has been delayed should | |
473 | * not be passed by new incoming requests | |
474 | */ | |
475 | rq->flags |= REQ_STARTED; | |
476 | } | |
1da177e4 LT |
477 | |
478 | if (rq == q->last_merge) | |
479 | q->last_merge = NULL; | |
480 | ||
8922e16c TH |
481 | if (!q->boundary_rq || q->boundary_rq == rq) { |
482 | q->last_sector = rq_last_sector(rq); | |
483 | q->boundary_rq = NULL; | |
484 | } | |
485 | ||
1da177e4 LT |
486 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) |
487 | break; | |
488 | ||
489 | ret = q->prep_rq_fn(q, rq); | |
490 | if (ret == BLKPREP_OK) { | |
491 | break; | |
492 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
493 | /* |
494 | * the request may have been (partially) prepped. | |
495 | * we need to keep this request in the front to | |
8922e16c TH |
496 | * avoid resource deadlock. REQ_STARTED will |
497 | * prevent other fs requests from passing this one. | |
2e759cd4 | 498 | */ |
1da177e4 LT |
499 | rq = NULL; |
500 | break; | |
501 | } else if (ret == BLKPREP_KILL) { | |
502 | int nr_bytes = rq->hard_nr_sectors << 9; | |
503 | ||
504 | if (!nr_bytes) | |
505 | nr_bytes = rq->data_len; | |
506 | ||
507 | blkdev_dequeue_request(rq); | |
508 | rq->flags |= REQ_QUIET; | |
509 | end_that_request_chunk(rq, 0, nr_bytes); | |
510 | end_that_request_last(rq); | |
511 | } else { | |
512 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
513 | ret); | |
514 | break; | |
515 | } | |
516 | } | |
517 | ||
518 | return rq; | |
519 | } | |
520 | ||
8922e16c | 521 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 522 | { |
8922e16c TH |
523 | BUG_ON(list_empty(&rq->queuelist)); |
524 | ||
525 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
526 | |
527 | /* | |
528 | * the time frame between a request being removed from the lists | |
529 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 530 | * the driver side. |
1da177e4 LT |
531 | */ |
532 | if (blk_account_rq(rq)) | |
533 | q->in_flight++; | |
534 | ||
535 | /* | |
536 | * the main clearing point for q->last_merge is on retrieval of | |
537 | * request by driver (it calls elv_next_request()), but it _can_ | |
538 | * also happen here if a request is added to the queue but later | |
539 | * deleted without ever being given to driver (merged with another | |
540 | * request). | |
541 | */ | |
542 | if (rq == q->last_merge) | |
543 | q->last_merge = NULL; | |
1da177e4 LT |
544 | } |
545 | ||
546 | int elv_queue_empty(request_queue_t *q) | |
547 | { | |
548 | elevator_t *e = q->elevator; | |
549 | ||
8922e16c TH |
550 | if (!list_empty(&q->queue_head)) |
551 | return 0; | |
552 | ||
1da177e4 LT |
553 | if (e->ops->elevator_queue_empty_fn) |
554 | return e->ops->elevator_queue_empty_fn(q); | |
555 | ||
8922e16c | 556 | return 1; |
1da177e4 LT |
557 | } |
558 | ||
559 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
560 | { | |
561 | struct list_head *next; | |
562 | ||
563 | elevator_t *e = q->elevator; | |
564 | ||
565 | if (e->ops->elevator_latter_req_fn) | |
566 | return e->ops->elevator_latter_req_fn(q, rq); | |
567 | ||
568 | next = rq->queuelist.next; | |
569 | if (next != &q->queue_head && next != &rq->queuelist) | |
570 | return list_entry_rq(next); | |
571 | ||
572 | return NULL; | |
573 | } | |
574 | ||
575 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
576 | { | |
577 | struct list_head *prev; | |
578 | ||
579 | elevator_t *e = q->elevator; | |
580 | ||
581 | if (e->ops->elevator_former_req_fn) | |
582 | return e->ops->elevator_former_req_fn(q, rq); | |
583 | ||
584 | prev = rq->queuelist.prev; | |
585 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
586 | return list_entry_rq(prev); | |
587 | ||
588 | return NULL; | |
589 | } | |
590 | ||
22e2c507 JA |
591 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
592 | int gfp_mask) | |
1da177e4 LT |
593 | { |
594 | elevator_t *e = q->elevator; | |
595 | ||
596 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 597 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
598 | |
599 | rq->elevator_private = NULL; | |
600 | return 0; | |
601 | } | |
602 | ||
603 | void elv_put_request(request_queue_t *q, struct request *rq) | |
604 | { | |
605 | elevator_t *e = q->elevator; | |
606 | ||
607 | if (e->ops->elevator_put_req_fn) | |
608 | e->ops->elevator_put_req_fn(q, rq); | |
609 | } | |
610 | ||
22e2c507 | 611 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
612 | { |
613 | elevator_t *e = q->elevator; | |
614 | ||
615 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 616 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
617 | |
618 | return ELV_MQUEUE_MAY; | |
619 | } | |
620 | ||
621 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
622 | { | |
623 | elevator_t *e = q->elevator; | |
624 | ||
625 | /* | |
626 | * request is released from the driver, io must be done | |
627 | */ | |
8922e16c | 628 | if (blk_account_rq(rq)) { |
1da177e4 | 629 | q->in_flight--; |
8922e16c TH |
630 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
631 | e->ops->elevator_completed_req_fn(q, rq); | |
632 | } | |
1da177e4 LT |
633 | } |
634 | ||
635 | int elv_register_queue(struct request_queue *q) | |
636 | { | |
637 | elevator_t *e = q->elevator; | |
638 | ||
639 | e->kobj.parent = kobject_get(&q->kobj); | |
640 | if (!e->kobj.parent) | |
641 | return -EBUSY; | |
642 | ||
643 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
644 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
645 | ||
646 | return kobject_register(&e->kobj); | |
647 | } | |
648 | ||
649 | void elv_unregister_queue(struct request_queue *q) | |
650 | { | |
651 | if (q) { | |
652 | elevator_t *e = q->elevator; | |
653 | kobject_unregister(&e->kobj); | |
654 | kobject_put(&q->kobj); | |
655 | } | |
656 | } | |
657 | ||
658 | int elv_register(struct elevator_type *e) | |
659 | { | |
2824bc93 | 660 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
661 | if (elevator_find(e->elevator_name)) |
662 | BUG(); | |
1da177e4 LT |
663 | list_add_tail(&e->list, &elv_list); |
664 | spin_unlock_irq(&elv_list_lock); | |
665 | ||
666 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
667 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
668 | printk(" (default)"); | |
669 | printk("\n"); | |
670 | return 0; | |
671 | } | |
672 | EXPORT_SYMBOL_GPL(elv_register); | |
673 | ||
674 | void elv_unregister(struct elevator_type *e) | |
675 | { | |
676 | spin_lock_irq(&elv_list_lock); | |
677 | list_del_init(&e->list); | |
678 | spin_unlock_irq(&elv_list_lock); | |
679 | } | |
680 | EXPORT_SYMBOL_GPL(elv_unregister); | |
681 | ||
682 | /* | |
683 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
684 | * we don't free the old io scheduler, before we have allocated what we | |
685 | * need for the new one. this way we have a chance of going back to the old | |
686 | * one, if the new one fails init for some reason. we also do an intermediate | |
687 | * switch to noop to ensure safety with stack-allocated requests, since they | |
688 | * don't originate from the block layer allocator. noop is safe here, because | |
689 | * it never needs to touch the elevator itself for completion events. DRAIN | |
690 | * flags will make sure we don't touch it for additions either. | |
691 | */ | |
692 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
693 | { | |
694 | elevator_t *e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
695 | struct elevator_type *noop_elevator = NULL; | |
696 | elevator_t *old_elevator; | |
697 | ||
698 | if (!e) | |
699 | goto error; | |
700 | ||
701 | /* | |
702 | * first step, drain requests from the block freelist | |
703 | */ | |
704 | blk_wait_queue_drained(q, 0); | |
705 | ||
706 | /* | |
707 | * unregister old elevator data | |
708 | */ | |
709 | elv_unregister_queue(q); | |
710 | old_elevator = q->elevator; | |
711 | ||
712 | /* | |
713 | * next step, switch to noop since it uses no private rq structures | |
714 | * and doesn't allocate any memory for anything. then wait for any | |
715 | * non-fs requests in-flight | |
716 | */ | |
717 | noop_elevator = elevator_get("noop"); | |
718 | spin_lock_irq(q->queue_lock); | |
719 | elevator_attach(q, noop_elevator, e); | |
720 | spin_unlock_irq(q->queue_lock); | |
721 | ||
722 | blk_wait_queue_drained(q, 1); | |
723 | ||
724 | /* | |
725 | * attach and start new elevator | |
726 | */ | |
727 | if (elevator_attach(q, new_e, e)) | |
728 | goto fail; | |
729 | ||
730 | if (elv_register_queue(q)) | |
731 | goto fail_register; | |
732 | ||
733 | /* | |
734 | * finally exit old elevator and start queue again | |
735 | */ | |
736 | elevator_exit(old_elevator); | |
737 | blk_finish_queue_drain(q); | |
738 | elevator_put(noop_elevator); | |
739 | return; | |
740 | ||
741 | fail_register: | |
742 | /* | |
743 | * switch failed, exit the new io scheduler and reattach the old | |
744 | * one again (along with re-adding the sysfs dir) | |
745 | */ | |
746 | elevator_exit(e); | |
747 | fail: | |
748 | q->elevator = old_elevator; | |
749 | elv_register_queue(q); | |
750 | blk_finish_queue_drain(q); | |
751 | error: | |
752 | if (noop_elevator) | |
753 | elevator_put(noop_elevator); | |
754 | elevator_put(new_e); | |
755 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
756 | } | |
757 | ||
758 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
759 | { | |
760 | char elevator_name[ELV_NAME_MAX]; | |
761 | struct elevator_type *e; | |
762 | ||
763 | memset(elevator_name, 0, sizeof(elevator_name)); | |
764 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
765 | ||
766 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
767 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
768 | ||
769 | e = elevator_get(elevator_name); | |
770 | if (!e) { | |
771 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
772 | return -EINVAL; | |
773 | } | |
774 | ||
775 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) | |
776 | return count; | |
777 | ||
778 | elevator_switch(q, e); | |
779 | return count; | |
780 | } | |
781 | ||
782 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
783 | { | |
784 | elevator_t *e = q->elevator; | |
785 | struct elevator_type *elv = e->elevator_type; | |
786 | struct list_head *entry; | |
787 | int len = 0; | |
788 | ||
789 | spin_lock_irq(q->queue_lock); | |
790 | list_for_each(entry, &elv_list) { | |
791 | struct elevator_type *__e; | |
792 | ||
793 | __e = list_entry(entry, struct elevator_type, list); | |
794 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
795 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
796 | else | |
797 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
798 | } | |
799 | spin_unlock_irq(q->queue_lock); | |
800 | ||
801 | len += sprintf(len+name, "\n"); | |
802 | return len; | |
803 | } | |
804 | ||
8922e16c | 805 | EXPORT_SYMBOL(elv_dispatch_insert); |
1da177e4 LT |
806 | EXPORT_SYMBOL(elv_add_request); |
807 | EXPORT_SYMBOL(__elv_add_request); | |
808 | EXPORT_SYMBOL(elv_requeue_request); | |
809 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 810 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
811 | EXPORT_SYMBOL(elv_queue_empty); |
812 | EXPORT_SYMBOL(elv_completed_request); | |
813 | EXPORT_SYMBOL(elevator_exit); | |
814 | EXPORT_SYMBOL(elevator_init); |