]>
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 | 149 | /* |
131dda7f | 150 | * If default has not been set, use the compiled-in selection. |
1da177e4 | 151 | */ |
131dda7f ND |
152 | if (!chosen_elevator[0]) |
153 | strcpy(chosen_elevator, CONFIG_DEFAULT_IOSCHED); | |
154 | ||
155 | /* | |
156 | * If the given scheduler is not available, fall back to no-op. | |
157 | */ | |
158 | if (!(e = elevator_find(chosen_elevator))) | |
159 | strcpy(chosen_elevator, "noop"); | |
160 | elevator_put(e); | |
1da177e4 LT |
161 | } |
162 | ||
163 | static int __init elevator_setup(char *str) | |
164 | { | |
165 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
166 | return 0; | |
167 | } | |
168 | ||
169 | __setup("elevator=", elevator_setup); | |
170 | ||
171 | int elevator_init(request_queue_t *q, char *name) | |
172 | { | |
173 | struct elevator_type *e = NULL; | |
174 | struct elevator_queue *eq; | |
175 | int ret = 0; | |
176 | ||
cb98fc8b TH |
177 | INIT_LIST_HEAD(&q->queue_head); |
178 | q->last_merge = NULL; | |
179 | q->end_sector = 0; | |
180 | q->boundary_rq = NULL; | |
cb98fc8b | 181 | |
1da177e4 LT |
182 | elevator_setup_default(); |
183 | ||
184 | if (!name) | |
185 | name = chosen_elevator; | |
186 | ||
187 | e = elevator_get(name); | |
188 | if (!e) | |
189 | return -EINVAL; | |
190 | ||
191 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
192 | if (!eq) { | |
193 | elevator_put(e->elevator_type); | |
194 | return -ENOMEM; | |
195 | } | |
196 | ||
197 | ret = elevator_attach(q, e, eq); | |
198 | if (ret) { | |
199 | kfree(eq); | |
200 | elevator_put(e->elevator_type); | |
201 | } | |
202 | ||
203 | return ret; | |
204 | } | |
205 | ||
206 | void elevator_exit(elevator_t *e) | |
207 | { | |
208 | if (e->ops->elevator_exit_fn) | |
209 | e->ops->elevator_exit_fn(e); | |
210 | ||
211 | elevator_put(e->elevator_type); | |
212 | e->elevator_type = NULL; | |
213 | kfree(e); | |
214 | } | |
215 | ||
8922e16c TH |
216 | /* |
217 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
218 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
219 | * appended to the dispatch queue. To be used by specific elevators. | |
220 | */ | |
1b47f531 | 221 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
222 | { |
223 | sector_t boundary; | |
8922e16c TH |
224 | struct list_head *entry; |
225 | ||
06b86245 TH |
226 | if (q->last_merge == rq) |
227 | q->last_merge = NULL; | |
228 | ||
1b47f531 | 229 | boundary = q->end_sector; |
cb19833d | 230 | |
8922e16c TH |
231 | list_for_each_prev(entry, &q->queue_head) { |
232 | struct request *pos = list_entry_rq(entry); | |
233 | ||
234 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
235 | break; | |
236 | if (rq->sector >= boundary) { | |
237 | if (pos->sector < boundary) | |
238 | continue; | |
239 | } else { | |
240 | if (pos->sector >= boundary) | |
241 | break; | |
242 | } | |
243 | if (rq->sector >= pos->sector) | |
244 | break; | |
245 | } | |
246 | ||
247 | list_add(&rq->queuelist, entry); | |
248 | } | |
249 | ||
1da177e4 LT |
250 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
251 | { | |
252 | elevator_t *e = q->elevator; | |
06b86245 TH |
253 | int ret; |
254 | ||
255 | if (q->last_merge) { | |
256 | ret = elv_try_merge(q->last_merge, bio); | |
257 | if (ret != ELEVATOR_NO_MERGE) { | |
258 | *req = q->last_merge; | |
259 | return ret; | |
260 | } | |
261 | } | |
1da177e4 LT |
262 | |
263 | if (e->ops->elevator_merge_fn) | |
264 | return e->ops->elevator_merge_fn(q, req, bio); | |
265 | ||
266 | return ELEVATOR_NO_MERGE; | |
267 | } | |
268 | ||
269 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
270 | { | |
271 | elevator_t *e = q->elevator; | |
272 | ||
273 | if (e->ops->elevator_merged_fn) | |
274 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
275 | |
276 | q->last_merge = rq; | |
1da177e4 LT |
277 | } |
278 | ||
279 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
280 | struct request *next) | |
281 | { | |
282 | elevator_t *e = q->elevator; | |
283 | ||
1da177e4 LT |
284 | if (e->ops->elevator_merge_req_fn) |
285 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 TH |
286 | |
287 | q->last_merge = rq; | |
1da177e4 LT |
288 | } |
289 | ||
8922e16c | 290 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
291 | { |
292 | elevator_t *e = q->elevator; | |
293 | ||
294 | /* | |
295 | * it already went through dequeue, we need to decrement the | |
296 | * in_flight count again | |
297 | */ | |
8922e16c | 298 | if (blk_account_rq(rq)) { |
1da177e4 | 299 | q->in_flight--; |
8922e16c TH |
300 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
301 | e->ops->elevator_deactivate_req_fn(q, rq); | |
302 | } | |
1da177e4 LT |
303 | |
304 | rq->flags &= ~REQ_STARTED; | |
305 | ||
1da177e4 LT |
306 | /* |
307 | * if this is the flush, requeue the original instead and drop the flush | |
308 | */ | |
309 | if (rq->flags & REQ_BAR_FLUSH) { | |
310 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
311 | rq = rq->end_io_data; | |
312 | } | |
313 | ||
8922e16c | 314 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
315 | } |
316 | ||
317 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
318 | int plug) | |
319 | { | |
8922e16c TH |
320 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
321 | /* | |
322 | * barriers implicitly indicate back insertion | |
323 | */ | |
324 | if (where == ELEVATOR_INSERT_SORT) | |
325 | where = ELEVATOR_INSERT_BACK; | |
326 | ||
327 | /* | |
1b47f531 | 328 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
329 | */ |
330 | if (blk_fs_request(rq)) { | |
1b47f531 | 331 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
332 | q->boundary_rq = rq; |
333 | } | |
cb98fc8b TH |
334 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
335 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
336 | |
337 | if (plug) | |
338 | blk_plug_device(q); | |
339 | ||
340 | rq->q = q; | |
341 | ||
8922e16c TH |
342 | switch (where) { |
343 | case ELEVATOR_INSERT_FRONT: | |
344 | rq->flags |= REQ_SOFTBARRIER; | |
345 | ||
346 | list_add(&rq->queuelist, &q->queue_head); | |
347 | break; | |
348 | ||
349 | case ELEVATOR_INSERT_BACK: | |
350 | rq->flags |= REQ_SOFTBARRIER; | |
351 | ||
352 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
353 | ; | |
354 | list_add_tail(&rq->queuelist, &q->queue_head); | |
355 | /* | |
356 | * We kick the queue here for the following reasons. | |
357 | * - The elevator might have returned NULL previously | |
358 | * to delay requests and returned them now. As the | |
359 | * queue wasn't empty before this request, ll_rw_blk | |
360 | * won't run the queue on return, resulting in hang. | |
361 | * - Usually, back inserted requests won't be merged | |
362 | * with anything. There's no point in delaying queue | |
363 | * processing. | |
364 | */ | |
365 | blk_remove_plug(q); | |
366 | q->request_fn(q); | |
367 | break; | |
368 | ||
369 | case ELEVATOR_INSERT_SORT: | |
370 | BUG_ON(!blk_fs_request(rq)); | |
371 | rq->flags |= REQ_SORTED; | |
372 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
06b86245 TH |
373 | if (q->last_merge == NULL && rq_mergeable(rq)) |
374 | q->last_merge = rq; | |
8922e16c TH |
375 | break; |
376 | ||
377 | default: | |
378 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
379 | __FUNCTION__, where); | |
380 | BUG(); | |
381 | } | |
382 | ||
383 | if (blk_queue_plugged(q)) { | |
384 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
385 | - q->in_flight; | |
386 | ||
387 | if (nrq >= q->unplug_thresh) | |
388 | __generic_unplug_device(q); | |
389 | } | |
1da177e4 LT |
390 | } |
391 | ||
392 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
393 | int plug) | |
394 | { | |
395 | unsigned long flags; | |
396 | ||
397 | spin_lock_irqsave(q->queue_lock, flags); | |
398 | __elv_add_request(q, rq, where, plug); | |
399 | spin_unlock_irqrestore(q->queue_lock, flags); | |
400 | } | |
401 | ||
402 | static inline struct request *__elv_next_request(request_queue_t *q) | |
403 | { | |
8922e16c TH |
404 | struct request *rq; |
405 | ||
406 | if (unlikely(list_empty(&q->queue_head) && | |
407 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
408 | return NULL; | |
409 | ||
410 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
411 | |
412 | /* | |
413 | * if this is a barrier write and the device has to issue a | |
414 | * flush sequence to support it, check how far we are | |
415 | */ | |
8922e16c | 416 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
417 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
418 | ||
419 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
420 | !blk_barrier_preflush(rq)) | |
421 | rq = blk_start_pre_flush(q, rq); | |
422 | } | |
423 | ||
424 | return rq; | |
425 | } | |
426 | ||
427 | struct request *elv_next_request(request_queue_t *q) | |
428 | { | |
429 | struct request *rq; | |
430 | int ret; | |
431 | ||
432 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
433 | if (!(rq->flags & REQ_STARTED)) { |
434 | elevator_t *e = q->elevator; | |
435 | ||
436 | /* | |
437 | * This is the first time the device driver | |
438 | * sees this request (possibly after | |
439 | * requeueing). Notify IO scheduler. | |
440 | */ | |
441 | if (blk_sorted_rq(rq) && | |
442 | e->ops->elevator_activate_req_fn) | |
443 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 444 | |
8922e16c TH |
445 | /* |
446 | * just mark as started even if we don't start | |
447 | * it, a request that has been delayed should | |
448 | * not be passed by new incoming requests | |
449 | */ | |
450 | rq->flags |= REQ_STARTED; | |
451 | } | |
1da177e4 | 452 | |
8922e16c | 453 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 454 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
455 | q->boundary_rq = NULL; |
456 | } | |
1da177e4 LT |
457 | |
458 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
459 | break; | |
460 | ||
461 | ret = q->prep_rq_fn(q, rq); | |
462 | if (ret == BLKPREP_OK) { | |
463 | break; | |
464 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
465 | /* |
466 | * the request may have been (partially) prepped. | |
467 | * we need to keep this request in the front to | |
8922e16c TH |
468 | * avoid resource deadlock. REQ_STARTED will |
469 | * prevent other fs requests from passing this one. | |
2e759cd4 | 470 | */ |
1da177e4 LT |
471 | rq = NULL; |
472 | break; | |
473 | } else if (ret == BLKPREP_KILL) { | |
474 | int nr_bytes = rq->hard_nr_sectors << 9; | |
475 | ||
476 | if (!nr_bytes) | |
477 | nr_bytes = rq->data_len; | |
478 | ||
479 | blkdev_dequeue_request(rq); | |
480 | rq->flags |= REQ_QUIET; | |
481 | end_that_request_chunk(rq, 0, nr_bytes); | |
482 | end_that_request_last(rq); | |
483 | } else { | |
484 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
485 | ret); | |
486 | break; | |
487 | } | |
488 | } | |
489 | ||
490 | return rq; | |
491 | } | |
492 | ||
8922e16c | 493 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 494 | { |
8922e16c TH |
495 | BUG_ON(list_empty(&rq->queuelist)); |
496 | ||
497 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
498 | |
499 | /* | |
500 | * the time frame between a request being removed from the lists | |
501 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 502 | * the driver side. |
1da177e4 LT |
503 | */ |
504 | if (blk_account_rq(rq)) | |
505 | q->in_flight++; | |
1da177e4 LT |
506 | } |
507 | ||
508 | int elv_queue_empty(request_queue_t *q) | |
509 | { | |
510 | elevator_t *e = q->elevator; | |
511 | ||
8922e16c TH |
512 | if (!list_empty(&q->queue_head)) |
513 | return 0; | |
514 | ||
1da177e4 LT |
515 | if (e->ops->elevator_queue_empty_fn) |
516 | return e->ops->elevator_queue_empty_fn(q); | |
517 | ||
8922e16c | 518 | return 1; |
1da177e4 LT |
519 | } |
520 | ||
521 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
522 | { | |
523 | struct list_head *next; | |
524 | ||
525 | elevator_t *e = q->elevator; | |
526 | ||
527 | if (e->ops->elevator_latter_req_fn) | |
528 | return e->ops->elevator_latter_req_fn(q, rq); | |
529 | ||
530 | next = rq->queuelist.next; | |
531 | if (next != &q->queue_head && next != &rq->queuelist) | |
532 | return list_entry_rq(next); | |
533 | ||
534 | return NULL; | |
535 | } | |
536 | ||
537 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
538 | { | |
539 | struct list_head *prev; | |
540 | ||
541 | elevator_t *e = q->elevator; | |
542 | ||
543 | if (e->ops->elevator_former_req_fn) | |
544 | return e->ops->elevator_former_req_fn(q, rq); | |
545 | ||
546 | prev = rq->queuelist.prev; | |
547 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
548 | return list_entry_rq(prev); | |
549 | ||
550 | return NULL; | |
551 | } | |
552 | ||
22e2c507 | 553 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 554 | gfp_t gfp_mask) |
1da177e4 LT |
555 | { |
556 | elevator_t *e = q->elevator; | |
557 | ||
558 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 559 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
560 | |
561 | rq->elevator_private = NULL; | |
562 | return 0; | |
563 | } | |
564 | ||
565 | void elv_put_request(request_queue_t *q, struct request *rq) | |
566 | { | |
567 | elevator_t *e = q->elevator; | |
568 | ||
569 | if (e->ops->elevator_put_req_fn) | |
570 | e->ops->elevator_put_req_fn(q, rq); | |
571 | } | |
572 | ||
22e2c507 | 573 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
574 | { |
575 | elevator_t *e = q->elevator; | |
576 | ||
577 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 578 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
579 | |
580 | return ELV_MQUEUE_MAY; | |
581 | } | |
582 | ||
583 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
584 | { | |
585 | elevator_t *e = q->elevator; | |
586 | ||
587 | /* | |
588 | * request is released from the driver, io must be done | |
589 | */ | |
8922e16c | 590 | if (blk_account_rq(rq)) { |
1da177e4 | 591 | q->in_flight--; |
8922e16c TH |
592 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
593 | e->ops->elevator_completed_req_fn(q, rq); | |
594 | } | |
1da177e4 LT |
595 | } |
596 | ||
597 | int elv_register_queue(struct request_queue *q) | |
598 | { | |
599 | elevator_t *e = q->elevator; | |
600 | ||
601 | e->kobj.parent = kobject_get(&q->kobj); | |
602 | if (!e->kobj.parent) | |
603 | return -EBUSY; | |
604 | ||
605 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
606 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
607 | ||
608 | return kobject_register(&e->kobj); | |
609 | } | |
610 | ||
611 | void elv_unregister_queue(struct request_queue *q) | |
612 | { | |
613 | if (q) { | |
614 | elevator_t *e = q->elevator; | |
615 | kobject_unregister(&e->kobj); | |
616 | kobject_put(&q->kobj); | |
617 | } | |
618 | } | |
619 | ||
620 | int elv_register(struct elevator_type *e) | |
621 | { | |
2824bc93 | 622 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
623 | if (elevator_find(e->elevator_name)) |
624 | BUG(); | |
1da177e4 LT |
625 | list_add_tail(&e->list, &elv_list); |
626 | spin_unlock_irq(&elv_list_lock); | |
627 | ||
628 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
629 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
630 | printk(" (default)"); | |
631 | printk("\n"); | |
632 | return 0; | |
633 | } | |
634 | EXPORT_SYMBOL_GPL(elv_register); | |
635 | ||
636 | void elv_unregister(struct elevator_type *e) | |
637 | { | |
83521d3e CH |
638 | struct task_struct *g, *p; |
639 | ||
640 | /* | |
641 | * Iterate every thread in the process to remove the io contexts. | |
642 | */ | |
643 | read_lock(&tasklist_lock); | |
644 | do_each_thread(g, p) { | |
645 | struct io_context *ioc = p->io_context; | |
646 | if (ioc && ioc->cic) { | |
647 | ioc->cic->exit(ioc->cic); | |
648 | ioc->cic->dtor(ioc->cic); | |
649 | ioc->cic = NULL; | |
650 | } | |
651 | if (ioc && ioc->aic) { | |
652 | ioc->aic->exit(ioc->aic); | |
653 | ioc->aic->dtor(ioc->aic); | |
654 | ioc->aic = NULL; | |
655 | } | |
656 | } while_each_thread(g, p); | |
657 | read_unlock(&tasklist_lock); | |
658 | ||
1da177e4 LT |
659 | spin_lock_irq(&elv_list_lock); |
660 | list_del_init(&e->list); | |
661 | spin_unlock_irq(&elv_list_lock); | |
662 | } | |
663 | EXPORT_SYMBOL_GPL(elv_unregister); | |
664 | ||
665 | /* | |
666 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
667 | * we don't free the old io scheduler, before we have allocated what we | |
668 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 669 | * one, if the new one fails init for some reason. |
1da177e4 LT |
670 | */ |
671 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
672 | { | |
cb98fc8b | 673 | elevator_t *old_elevator, *e; |
1da177e4 | 674 | |
cb98fc8b TH |
675 | /* |
676 | * Allocate new elevator | |
677 | */ | |
678 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
679 | if (!e) |
680 | goto error; | |
681 | ||
682 | /* | |
cb98fc8b | 683 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 684 | */ |
cb98fc8b TH |
685 | spin_lock_irq(q->queue_lock); |
686 | ||
64521d1a | 687 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
688 | |
689 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
690 | ; | |
691 | ||
692 | while (q->rq.elvpriv) { | |
693 | spin_unlock_irq(q->queue_lock); | |
64521d1a | 694 | msleep(10); |
cb98fc8b TH |
695 | spin_lock_irq(q->queue_lock); |
696 | } | |
697 | ||
698 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
699 | |
700 | /* | |
701 | * unregister old elevator data | |
702 | */ | |
703 | elv_unregister_queue(q); | |
704 | old_elevator = q->elevator; | |
705 | ||
1da177e4 LT |
706 | /* |
707 | * attach and start new elevator | |
708 | */ | |
709 | if (elevator_attach(q, new_e, e)) | |
710 | goto fail; | |
711 | ||
712 | if (elv_register_queue(q)) | |
713 | goto fail_register; | |
714 | ||
715 | /* | |
cb98fc8b | 716 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
717 | */ |
718 | elevator_exit(old_elevator); | |
64521d1a | 719 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
720 | return; |
721 | ||
722 | fail_register: | |
723 | /* | |
724 | * switch failed, exit the new io scheduler and reattach the old | |
725 | * one again (along with re-adding the sysfs dir) | |
726 | */ | |
727 | elevator_exit(e); | |
cb98fc8b | 728 | e = NULL; |
1da177e4 LT |
729 | fail: |
730 | q->elevator = old_elevator; | |
731 | elv_register_queue(q); | |
64521d1a | 732 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 733 | kfree(e); |
1da177e4 | 734 | error: |
1da177e4 LT |
735 | elevator_put(new_e); |
736 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
737 | } | |
738 | ||
739 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
740 | { | |
741 | char elevator_name[ELV_NAME_MAX]; | |
742 | struct elevator_type *e; | |
743 | ||
744 | memset(elevator_name, 0, sizeof(elevator_name)); | |
745 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
746 | ||
747 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
748 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
749 | ||
750 | e = elevator_get(elevator_name); | |
751 | if (!e) { | |
752 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
753 | return -EINVAL; | |
754 | } | |
755 | ||
2ca7d93b ND |
756 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
757 | elevator_put(e); | |
1da177e4 | 758 | return count; |
2ca7d93b | 759 | } |
1da177e4 LT |
760 | |
761 | elevator_switch(q, e); | |
762 | return count; | |
763 | } | |
764 | ||
765 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
766 | { | |
767 | elevator_t *e = q->elevator; | |
768 | struct elevator_type *elv = e->elevator_type; | |
769 | struct list_head *entry; | |
770 | int len = 0; | |
771 | ||
772 | spin_lock_irq(q->queue_lock); | |
773 | list_for_each(entry, &elv_list) { | |
774 | struct elevator_type *__e; | |
775 | ||
776 | __e = list_entry(entry, struct elevator_type, list); | |
777 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
778 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
779 | else | |
780 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
781 | } | |
782 | spin_unlock_irq(q->queue_lock); | |
783 | ||
784 | len += sprintf(len+name, "\n"); | |
785 | return len; | |
786 | } | |
787 | ||
1b47f531 | 788 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
789 | EXPORT_SYMBOL(elv_add_request); |
790 | EXPORT_SYMBOL(__elv_add_request); | |
791 | EXPORT_SYMBOL(elv_requeue_request); | |
792 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 793 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
794 | EXPORT_SYMBOL(elv_queue_empty); |
795 | EXPORT_SYMBOL(elv_completed_request); | |
796 | EXPORT_SYMBOL(elevator_exit); | |
797 | EXPORT_SYMBOL(elevator_init); |