]>
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 | ||
43 | /* | |
44 | * can we safely merge with this request? | |
45 | */ | |
46 | inline int elv_rq_merge_ok(struct request *rq, struct bio *bio) | |
47 | { | |
48 | if (!rq_mergeable(rq)) | |
49 | return 0; | |
50 | ||
51 | /* | |
52 | * different data direction or already started, don't merge | |
53 | */ | |
54 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
55 | return 0; | |
56 | ||
57 | /* | |
58 | * same device and no special stuff set, merge is ok | |
59 | */ | |
60 | if (rq->rq_disk == bio->bi_bdev->bd_disk && | |
61 | !rq->waiting && !rq->special) | |
62 | return 1; | |
63 | ||
64 | return 0; | |
65 | } | |
66 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
67 | ||
68 | inline int elv_try_merge(struct request *__rq, struct bio *bio) | |
69 | { | |
70 | int ret = ELEVATOR_NO_MERGE; | |
71 | ||
72 | /* | |
73 | * we can merge and sequence is ok, check if it's possible | |
74 | */ | |
75 | if (elv_rq_merge_ok(__rq, bio)) { | |
76 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
77 | ret = ELEVATOR_BACK_MERGE; | |
78 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
79 | ret = ELEVATOR_FRONT_MERGE; | |
80 | } | |
81 | ||
82 | return ret; | |
83 | } | |
84 | EXPORT_SYMBOL(elv_try_merge); | |
85 | ||
1da177e4 LT |
86 | static struct elevator_type *elevator_find(const char *name) |
87 | { | |
88 | struct elevator_type *e = NULL; | |
89 | struct list_head *entry; | |
90 | ||
1da177e4 LT |
91 | list_for_each(entry, &elv_list) { |
92 | struct elevator_type *__e; | |
93 | ||
94 | __e = list_entry(entry, struct elevator_type, list); | |
95 | ||
96 | if (!strcmp(__e->elevator_name, name)) { | |
97 | e = __e; | |
98 | break; | |
99 | } | |
100 | } | |
1da177e4 LT |
101 | |
102 | return e; | |
103 | } | |
104 | ||
105 | static void elevator_put(struct elevator_type *e) | |
106 | { | |
107 | module_put(e->elevator_owner); | |
108 | } | |
109 | ||
110 | static struct elevator_type *elevator_get(const char *name) | |
111 | { | |
2824bc93 | 112 | struct elevator_type *e; |
1da177e4 | 113 | |
2824bc93 TH |
114 | spin_lock_irq(&elv_list_lock); |
115 | ||
116 | e = elevator_find(name); | |
117 | if (e && !try_module_get(e->elevator_owner)) | |
118 | e = NULL; | |
119 | ||
120 | spin_unlock_irq(&elv_list_lock); | |
1da177e4 LT |
121 | |
122 | return e; | |
123 | } | |
124 | ||
125 | static int elevator_attach(request_queue_t *q, struct elevator_type *e, | |
126 | struct elevator_queue *eq) | |
127 | { | |
128 | int ret = 0; | |
129 | ||
130 | memset(eq, 0, sizeof(*eq)); | |
131 | eq->ops = &e->ops; | |
132 | eq->elevator_type = e; | |
133 | ||
134 | INIT_LIST_HEAD(&q->queue_head); | |
135 | q->last_merge = NULL; | |
136 | q->elevator = eq; | |
1b47f531 | 137 | q->end_sector = 0; |
8922e16c TH |
138 | q->boundary_rq = NULL; |
139 | q->max_back_kb = 0; | |
1da177e4 LT |
140 | |
141 | if (eq->ops->elevator_init_fn) | |
142 | ret = eq->ops->elevator_init_fn(q, eq); | |
143 | ||
144 | return ret; | |
145 | } | |
146 | ||
147 | static char chosen_elevator[16]; | |
148 | ||
149 | static void elevator_setup_default(void) | |
150 | { | |
2824bc93 TH |
151 | struct elevator_type *e; |
152 | ||
1da177e4 LT |
153 | /* |
154 | * check if default is set and exists | |
155 | */ | |
2824bc93 TH |
156 | if (chosen_elevator[0] && (e = elevator_get(chosen_elevator))) { |
157 | elevator_put(e); | |
1da177e4 | 158 | return; |
2824bc93 | 159 | } |
1da177e4 LT |
160 | |
161 | #if defined(CONFIG_IOSCHED_AS) | |
162 | strcpy(chosen_elevator, "anticipatory"); | |
163 | #elif defined(CONFIG_IOSCHED_DEADLINE) | |
164 | strcpy(chosen_elevator, "deadline"); | |
165 | #elif defined(CONFIG_IOSCHED_CFQ) | |
166 | strcpy(chosen_elevator, "cfq"); | |
167 | #elif defined(CONFIG_IOSCHED_NOOP) | |
168 | strcpy(chosen_elevator, "noop"); | |
169 | #else | |
170 | #error "You must build at least 1 IO scheduler into the kernel" | |
171 | #endif | |
172 | } | |
173 | ||
174 | static int __init elevator_setup(char *str) | |
175 | { | |
176 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
177 | return 0; | |
178 | } | |
179 | ||
180 | __setup("elevator=", elevator_setup); | |
181 | ||
182 | int elevator_init(request_queue_t *q, char *name) | |
183 | { | |
184 | struct elevator_type *e = NULL; | |
185 | struct elevator_queue *eq; | |
186 | int ret = 0; | |
187 | ||
188 | elevator_setup_default(); | |
189 | ||
190 | if (!name) | |
191 | name = chosen_elevator; | |
192 | ||
193 | e = elevator_get(name); | |
194 | if (!e) | |
195 | return -EINVAL; | |
196 | ||
197 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
198 | if (!eq) { | |
199 | elevator_put(e->elevator_type); | |
200 | return -ENOMEM; | |
201 | } | |
202 | ||
203 | ret = elevator_attach(q, e, eq); | |
204 | if (ret) { | |
205 | kfree(eq); | |
206 | elevator_put(e->elevator_type); | |
207 | } | |
208 | ||
209 | return ret; | |
210 | } | |
211 | ||
212 | void elevator_exit(elevator_t *e) | |
213 | { | |
214 | if (e->ops->elevator_exit_fn) | |
215 | e->ops->elevator_exit_fn(e); | |
216 | ||
217 | elevator_put(e->elevator_type); | |
218 | e->elevator_type = NULL; | |
219 | kfree(e); | |
220 | } | |
221 | ||
8922e16c TH |
222 | /* |
223 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
224 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
225 | * appended to the dispatch queue. To be used by specific elevators. | |
226 | */ | |
1b47f531 | 227 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
228 | { |
229 | sector_t boundary; | |
230 | unsigned max_back; | |
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; |
8922e16c TH |
237 | max_back = q->max_back_kb * 2; |
238 | boundary = boundary > max_back ? boundary - max_back : 0; | |
1b47f531 | 239 | |
8922e16c TH |
240 | list_for_each_prev(entry, &q->queue_head) { |
241 | struct request *pos = list_entry_rq(entry); | |
242 | ||
243 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
244 | break; | |
245 | if (rq->sector >= boundary) { | |
246 | if (pos->sector < boundary) | |
247 | continue; | |
248 | } else { | |
249 | if (pos->sector >= boundary) | |
250 | break; | |
251 | } | |
252 | if (rq->sector >= pos->sector) | |
253 | break; | |
254 | } | |
255 | ||
256 | list_add(&rq->queuelist, entry); | |
257 | } | |
258 | ||
1da177e4 LT |
259 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
260 | { | |
261 | elevator_t *e = q->elevator; | |
06b86245 TH |
262 | int ret; |
263 | ||
264 | if (q->last_merge) { | |
265 | ret = elv_try_merge(q->last_merge, bio); | |
266 | if (ret != ELEVATOR_NO_MERGE) { | |
267 | *req = q->last_merge; | |
268 | return ret; | |
269 | } | |
270 | } | |
1da177e4 LT |
271 | |
272 | if (e->ops->elevator_merge_fn) | |
273 | return e->ops->elevator_merge_fn(q, req, bio); | |
274 | ||
275 | return ELEVATOR_NO_MERGE; | |
276 | } | |
277 | ||
278 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
279 | { | |
280 | elevator_t *e = q->elevator; | |
281 | ||
282 | if (e->ops->elevator_merged_fn) | |
283 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
284 | |
285 | q->last_merge = rq; | |
1da177e4 LT |
286 | } |
287 | ||
288 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
289 | struct request *next) | |
290 | { | |
291 | elevator_t *e = q->elevator; | |
292 | ||
1da177e4 LT |
293 | if (e->ops->elevator_merge_req_fn) |
294 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 TH |
295 | |
296 | q->last_merge = rq; | |
1da177e4 LT |
297 | } |
298 | ||
8922e16c | 299 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
300 | { |
301 | elevator_t *e = q->elevator; | |
302 | ||
303 | /* | |
304 | * it already went through dequeue, we need to decrement the | |
305 | * in_flight count again | |
306 | */ | |
8922e16c | 307 | if (blk_account_rq(rq)) { |
1da177e4 | 308 | q->in_flight--; |
8922e16c TH |
309 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
310 | e->ops->elevator_deactivate_req_fn(q, rq); | |
311 | } | |
1da177e4 LT |
312 | |
313 | rq->flags &= ~REQ_STARTED; | |
314 | ||
1da177e4 LT |
315 | /* |
316 | * if this is the flush, requeue the original instead and drop the flush | |
317 | */ | |
318 | if (rq->flags & REQ_BAR_FLUSH) { | |
319 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
320 | rq = rq->end_io_data; | |
321 | } | |
322 | ||
8922e16c | 323 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
324 | } |
325 | ||
326 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
327 | int plug) | |
328 | { | |
8922e16c TH |
329 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
330 | /* | |
331 | * barriers implicitly indicate back insertion | |
332 | */ | |
333 | if (where == ELEVATOR_INSERT_SORT) | |
334 | where = ELEVATOR_INSERT_BACK; | |
335 | ||
336 | /* | |
1b47f531 | 337 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
338 | */ |
339 | if (blk_fs_request(rq)) { | |
1b47f531 | 340 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
341 | q->boundary_rq = rq; |
342 | } | |
343 | } | |
1da177e4 LT |
344 | |
345 | if (plug) | |
346 | blk_plug_device(q); | |
347 | ||
348 | rq->q = q; | |
349 | ||
8922e16c | 350 | if (unlikely(test_bit(QUEUE_FLAG_DRAIN, &q->queue_flags))) { |
1da177e4 LT |
351 | /* |
352 | * if drain is set, store the request "locally". when the drain | |
353 | * is finished, the requests will be handed ordered to the io | |
354 | * scheduler | |
355 | */ | |
356 | list_add_tail(&rq->queuelist, &q->drain_list); | |
8922e16c TH |
357 | return; |
358 | } | |
359 | ||
360 | 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; | |
369 | ||
370 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
371 | ; | |
372 | list_add_tail(&rq->queuelist, &q->queue_head); | |
373 | /* | |
374 | * We kick the queue here for the following reasons. | |
375 | * - The elevator might have returned NULL previously | |
376 | * to delay requests and returned them now. As the | |
377 | * queue wasn't empty before this request, ll_rw_blk | |
378 | * won't run the queue on return, resulting in hang. | |
379 | * - Usually, back inserted requests won't be merged | |
380 | * with anything. There's no point in delaying queue | |
381 | * processing. | |
382 | */ | |
383 | blk_remove_plug(q); | |
384 | q->request_fn(q); | |
385 | break; | |
386 | ||
387 | case ELEVATOR_INSERT_SORT: | |
388 | BUG_ON(!blk_fs_request(rq)); | |
389 | rq->flags |= REQ_SORTED; | |
390 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
06b86245 TH |
391 | if (q->last_merge == NULL && rq_mergeable(rq)) |
392 | q->last_merge = rq; | |
8922e16c TH |
393 | break; |
394 | ||
395 | default: | |
396 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
397 | __FUNCTION__, where); | |
398 | BUG(); | |
399 | } | |
400 | ||
401 | if (blk_queue_plugged(q)) { | |
402 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
403 | - q->in_flight; | |
404 | ||
405 | if (nrq >= q->unplug_thresh) | |
406 | __generic_unplug_device(q); | |
407 | } | |
1da177e4 LT |
408 | } |
409 | ||
410 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
411 | int plug) | |
412 | { | |
413 | unsigned long flags; | |
414 | ||
415 | spin_lock_irqsave(q->queue_lock, flags); | |
416 | __elv_add_request(q, rq, where, plug); | |
417 | spin_unlock_irqrestore(q->queue_lock, flags); | |
418 | } | |
419 | ||
420 | static inline struct request *__elv_next_request(request_queue_t *q) | |
421 | { | |
8922e16c TH |
422 | struct request *rq; |
423 | ||
424 | if (unlikely(list_empty(&q->queue_head) && | |
425 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
426 | return NULL; | |
427 | ||
428 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
429 | |
430 | /* | |
431 | * if this is a barrier write and the device has to issue a | |
432 | * flush sequence to support it, check how far we are | |
433 | */ | |
8922e16c | 434 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
435 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
436 | ||
437 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
438 | !blk_barrier_preflush(rq)) | |
439 | rq = blk_start_pre_flush(q, rq); | |
440 | } | |
441 | ||
442 | return rq; | |
443 | } | |
444 | ||
445 | struct request *elv_next_request(request_queue_t *q) | |
446 | { | |
447 | struct request *rq; | |
448 | int ret; | |
449 | ||
450 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
451 | if (!(rq->flags & REQ_STARTED)) { |
452 | elevator_t *e = q->elevator; | |
453 | ||
454 | /* | |
455 | * This is the first time the device driver | |
456 | * sees this request (possibly after | |
457 | * requeueing). Notify IO scheduler. | |
458 | */ | |
459 | if (blk_sorted_rq(rq) && | |
460 | e->ops->elevator_activate_req_fn) | |
461 | e->ops->elevator_activate_req_fn(q, rq); | |
462 | ||
463 | /* | |
464 | * just mark as started even if we don't start | |
465 | * it, a request that has been delayed should | |
466 | * not be passed by new incoming requests | |
467 | */ | |
468 | rq->flags |= REQ_STARTED; | |
469 | } | |
1da177e4 | 470 | |
8922e16c | 471 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 472 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
473 | q->boundary_rq = NULL; |
474 | } | |
475 | ||
1da177e4 LT |
476 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) |
477 | break; | |
478 | ||
479 | ret = q->prep_rq_fn(q, rq); | |
480 | if (ret == BLKPREP_OK) { | |
481 | break; | |
482 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
483 | /* |
484 | * the request may have been (partially) prepped. | |
485 | * we need to keep this request in the front to | |
8922e16c TH |
486 | * avoid resource deadlock. REQ_STARTED will |
487 | * prevent other fs requests from passing this one. | |
2e759cd4 | 488 | */ |
1da177e4 LT |
489 | rq = NULL; |
490 | break; | |
491 | } else if (ret == BLKPREP_KILL) { | |
492 | int nr_bytes = rq->hard_nr_sectors << 9; | |
493 | ||
494 | if (!nr_bytes) | |
495 | nr_bytes = rq->data_len; | |
496 | ||
497 | blkdev_dequeue_request(rq); | |
498 | rq->flags |= REQ_QUIET; | |
499 | end_that_request_chunk(rq, 0, nr_bytes); | |
500 | end_that_request_last(rq); | |
501 | } else { | |
502 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
503 | ret); | |
504 | break; | |
505 | } | |
506 | } | |
507 | ||
508 | return rq; | |
509 | } | |
510 | ||
8922e16c | 511 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 512 | { |
8922e16c TH |
513 | BUG_ON(list_empty(&rq->queuelist)); |
514 | ||
515 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
516 | |
517 | /* | |
518 | * the time frame between a request being removed from the lists | |
519 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 520 | * the driver side. |
1da177e4 LT |
521 | */ |
522 | if (blk_account_rq(rq)) | |
523 | q->in_flight++; | |
1da177e4 LT |
524 | } |
525 | ||
526 | int elv_queue_empty(request_queue_t *q) | |
527 | { | |
528 | elevator_t *e = q->elevator; | |
529 | ||
8922e16c TH |
530 | if (!list_empty(&q->queue_head)) |
531 | return 0; | |
532 | ||
1da177e4 LT |
533 | if (e->ops->elevator_queue_empty_fn) |
534 | return e->ops->elevator_queue_empty_fn(q); | |
535 | ||
8922e16c | 536 | return 1; |
1da177e4 LT |
537 | } |
538 | ||
539 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
540 | { | |
541 | struct list_head *next; | |
542 | ||
543 | elevator_t *e = q->elevator; | |
544 | ||
545 | if (e->ops->elevator_latter_req_fn) | |
546 | return e->ops->elevator_latter_req_fn(q, rq); | |
547 | ||
548 | next = rq->queuelist.next; | |
549 | if (next != &q->queue_head && next != &rq->queuelist) | |
550 | return list_entry_rq(next); | |
551 | ||
552 | return NULL; | |
553 | } | |
554 | ||
555 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
556 | { | |
557 | struct list_head *prev; | |
558 | ||
559 | elevator_t *e = q->elevator; | |
560 | ||
561 | if (e->ops->elevator_former_req_fn) | |
562 | return e->ops->elevator_former_req_fn(q, rq); | |
563 | ||
564 | prev = rq->queuelist.prev; | |
565 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
566 | return list_entry_rq(prev); | |
567 | ||
568 | return NULL; | |
569 | } | |
570 | ||
22e2c507 JA |
571 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
572 | int gfp_mask) | |
1da177e4 LT |
573 | { |
574 | elevator_t *e = q->elevator; | |
575 | ||
576 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 577 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
578 | |
579 | rq->elevator_private = NULL; | |
580 | return 0; | |
581 | } | |
582 | ||
583 | void elv_put_request(request_queue_t *q, struct request *rq) | |
584 | { | |
585 | elevator_t *e = q->elevator; | |
586 | ||
587 | if (e->ops->elevator_put_req_fn) | |
588 | e->ops->elevator_put_req_fn(q, rq); | |
589 | } | |
590 | ||
22e2c507 | 591 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
592 | { |
593 | elevator_t *e = q->elevator; | |
594 | ||
595 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 596 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
597 | |
598 | return ELV_MQUEUE_MAY; | |
599 | } | |
600 | ||
601 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
602 | { | |
603 | elevator_t *e = q->elevator; | |
604 | ||
605 | /* | |
606 | * request is released from the driver, io must be done | |
607 | */ | |
8922e16c | 608 | if (blk_account_rq(rq)) { |
1da177e4 | 609 | q->in_flight--; |
8922e16c TH |
610 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
611 | e->ops->elevator_completed_req_fn(q, rq); | |
612 | } | |
1da177e4 LT |
613 | } |
614 | ||
615 | int elv_register_queue(struct request_queue *q) | |
616 | { | |
617 | elevator_t *e = q->elevator; | |
618 | ||
619 | e->kobj.parent = kobject_get(&q->kobj); | |
620 | if (!e->kobj.parent) | |
621 | return -EBUSY; | |
622 | ||
623 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
624 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
625 | ||
626 | return kobject_register(&e->kobj); | |
627 | } | |
628 | ||
629 | void elv_unregister_queue(struct request_queue *q) | |
630 | { | |
631 | if (q) { | |
632 | elevator_t *e = q->elevator; | |
633 | kobject_unregister(&e->kobj); | |
634 | kobject_put(&q->kobj); | |
635 | } | |
636 | } | |
637 | ||
638 | int elv_register(struct elevator_type *e) | |
639 | { | |
2824bc93 | 640 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
641 | if (elevator_find(e->elevator_name)) |
642 | BUG(); | |
1da177e4 LT |
643 | list_add_tail(&e->list, &elv_list); |
644 | spin_unlock_irq(&elv_list_lock); | |
645 | ||
646 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
647 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
648 | printk(" (default)"); | |
649 | printk("\n"); | |
650 | return 0; | |
651 | } | |
652 | EXPORT_SYMBOL_GPL(elv_register); | |
653 | ||
654 | void elv_unregister(struct elevator_type *e) | |
655 | { | |
656 | spin_lock_irq(&elv_list_lock); | |
657 | list_del_init(&e->list); | |
658 | spin_unlock_irq(&elv_list_lock); | |
659 | } | |
660 | EXPORT_SYMBOL_GPL(elv_unregister); | |
661 | ||
662 | /* | |
663 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
664 | * we don't free the old io scheduler, before we have allocated what we | |
665 | * need for the new one. this way we have a chance of going back to the old | |
666 | * one, if the new one fails init for some reason. we also do an intermediate | |
667 | * switch to noop to ensure safety with stack-allocated requests, since they | |
668 | * don't originate from the block layer allocator. noop is safe here, because | |
669 | * it never needs to touch the elevator itself for completion events. DRAIN | |
670 | * flags will make sure we don't touch it for additions either. | |
671 | */ | |
672 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
673 | { | |
674 | elevator_t *e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
675 | struct elevator_type *noop_elevator = NULL; | |
676 | elevator_t *old_elevator; | |
677 | ||
678 | if (!e) | |
679 | goto error; | |
680 | ||
681 | /* | |
682 | * first step, drain requests from the block freelist | |
683 | */ | |
684 | blk_wait_queue_drained(q, 0); | |
685 | ||
686 | /* | |
687 | * unregister old elevator data | |
688 | */ | |
689 | elv_unregister_queue(q); | |
690 | old_elevator = q->elevator; | |
691 | ||
692 | /* | |
693 | * next step, switch to noop since it uses no private rq structures | |
694 | * and doesn't allocate any memory for anything. then wait for any | |
695 | * non-fs requests in-flight | |
696 | */ | |
697 | noop_elevator = elevator_get("noop"); | |
698 | spin_lock_irq(q->queue_lock); | |
699 | elevator_attach(q, noop_elevator, e); | |
700 | spin_unlock_irq(q->queue_lock); | |
701 | ||
702 | blk_wait_queue_drained(q, 1); | |
703 | ||
704 | /* | |
705 | * attach and start new elevator | |
706 | */ | |
707 | if (elevator_attach(q, new_e, e)) | |
708 | goto fail; | |
709 | ||
710 | if (elv_register_queue(q)) | |
711 | goto fail_register; | |
712 | ||
713 | /* | |
714 | * finally exit old elevator and start queue again | |
715 | */ | |
716 | elevator_exit(old_elevator); | |
717 | blk_finish_queue_drain(q); | |
718 | elevator_put(noop_elevator); | |
719 | return; | |
720 | ||
721 | fail_register: | |
722 | /* | |
723 | * switch failed, exit the new io scheduler and reattach the old | |
724 | * one again (along with re-adding the sysfs dir) | |
725 | */ | |
726 | elevator_exit(e); | |
727 | fail: | |
728 | q->elevator = old_elevator; | |
729 | elv_register_queue(q); | |
730 | blk_finish_queue_drain(q); | |
731 | error: | |
732 | if (noop_elevator) | |
733 | elevator_put(noop_elevator); | |
734 | elevator_put(new_e); | |
735 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
736 | } | |
737 | ||
738 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
739 | { | |
740 | char elevator_name[ELV_NAME_MAX]; | |
741 | struct elevator_type *e; | |
742 | ||
743 | memset(elevator_name, 0, sizeof(elevator_name)); | |
744 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
745 | ||
746 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
747 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
748 | ||
749 | e = elevator_get(elevator_name); | |
750 | if (!e) { | |
751 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
752 | return -EINVAL; | |
753 | } | |
754 | ||
755 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) | |
756 | return count; | |
757 | ||
758 | elevator_switch(q, e); | |
759 | return count; | |
760 | } | |
761 | ||
762 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
763 | { | |
764 | elevator_t *e = q->elevator; | |
765 | struct elevator_type *elv = e->elevator_type; | |
766 | struct list_head *entry; | |
767 | int len = 0; | |
768 | ||
769 | spin_lock_irq(q->queue_lock); | |
770 | list_for_each(entry, &elv_list) { | |
771 | struct elevator_type *__e; | |
772 | ||
773 | __e = list_entry(entry, struct elevator_type, list); | |
774 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
775 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
776 | else | |
777 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
778 | } | |
779 | spin_unlock_irq(q->queue_lock); | |
780 | ||
781 | len += sprintf(len+name, "\n"); | |
782 | return len; | |
783 | } | |
784 | ||
1b47f531 | 785 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
786 | EXPORT_SYMBOL(elv_add_request); |
787 | EXPORT_SYMBOL(__elv_add_request); | |
788 | EXPORT_SYMBOL(elv_requeue_request); | |
789 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 790 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
791 | EXPORT_SYMBOL(elv_queue_empty); |
792 | EXPORT_SYMBOL(elv_completed_request); | |
793 | EXPORT_SYMBOL(elevator_exit); | |
794 | EXPORT_SYMBOL(elevator_init); |