]>
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 | */ | |
cff3ba22 ZA |
158 | if ((e = elevator_find(chosen_elevator))) |
159 | elevator_put(e); | |
160 | else | |
131dda7f | 161 | strcpy(chosen_elevator, "noop"); |
1da177e4 LT |
162 | } |
163 | ||
164 | static int __init elevator_setup(char *str) | |
165 | { | |
166 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
167 | return 0; | |
168 | } | |
169 | ||
170 | __setup("elevator=", elevator_setup); | |
171 | ||
172 | int elevator_init(request_queue_t *q, char *name) | |
173 | { | |
174 | struct elevator_type *e = NULL; | |
175 | struct elevator_queue *eq; | |
176 | int ret = 0; | |
177 | ||
cb98fc8b TH |
178 | INIT_LIST_HEAD(&q->queue_head); |
179 | q->last_merge = NULL; | |
180 | q->end_sector = 0; | |
181 | q->boundary_rq = NULL; | |
cb98fc8b | 182 | |
1da177e4 LT |
183 | elevator_setup_default(); |
184 | ||
185 | if (!name) | |
186 | name = chosen_elevator; | |
187 | ||
188 | e = elevator_get(name); | |
189 | if (!e) | |
190 | return -EINVAL; | |
191 | ||
192 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
193 | if (!eq) { | |
b8ea2cb5 | 194 | elevator_put(e); |
1da177e4 LT |
195 | return -ENOMEM; |
196 | } | |
197 | ||
198 | ret = elevator_attach(q, e, eq); | |
199 | if (ret) { | |
200 | kfree(eq); | |
b8ea2cb5 | 201 | elevator_put(e); |
1da177e4 LT |
202 | } |
203 | ||
204 | return ret; | |
205 | } | |
206 | ||
207 | void elevator_exit(elevator_t *e) | |
208 | { | |
209 | if (e->ops->elevator_exit_fn) | |
210 | e->ops->elevator_exit_fn(e); | |
211 | ||
212 | elevator_put(e->elevator_type); | |
213 | e->elevator_type = NULL; | |
214 | kfree(e); | |
215 | } | |
216 | ||
8922e16c TH |
217 | /* |
218 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
219 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
220 | * appended to the dispatch queue. To be used by specific elevators. | |
221 | */ | |
1b47f531 | 222 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
223 | { |
224 | sector_t boundary; | |
8922e16c TH |
225 | struct list_head *entry; |
226 | ||
06b86245 TH |
227 | if (q->last_merge == rq) |
228 | q->last_merge = NULL; | |
229 | ||
1b47f531 | 230 | boundary = q->end_sector; |
cb19833d | 231 | |
8922e16c TH |
232 | list_for_each_prev(entry, &q->queue_head) { |
233 | struct request *pos = list_entry_rq(entry); | |
234 | ||
235 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
236 | break; | |
237 | if (rq->sector >= boundary) { | |
238 | if (pos->sector < boundary) | |
239 | continue; | |
240 | } else { | |
241 | if (pos->sector >= boundary) | |
242 | break; | |
243 | } | |
244 | if (rq->sector >= pos->sector) | |
245 | break; | |
246 | } | |
247 | ||
248 | list_add(&rq->queuelist, entry); | |
249 | } | |
250 | ||
1da177e4 LT |
251 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
252 | { | |
253 | elevator_t *e = q->elevator; | |
06b86245 TH |
254 | int ret; |
255 | ||
256 | if (q->last_merge) { | |
257 | ret = elv_try_merge(q->last_merge, bio); | |
258 | if (ret != ELEVATOR_NO_MERGE) { | |
259 | *req = q->last_merge; | |
260 | return ret; | |
261 | } | |
262 | } | |
1da177e4 LT |
263 | |
264 | if (e->ops->elevator_merge_fn) | |
265 | return e->ops->elevator_merge_fn(q, req, bio); | |
266 | ||
267 | return ELEVATOR_NO_MERGE; | |
268 | } | |
269 | ||
270 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
271 | { | |
272 | elevator_t *e = q->elevator; | |
273 | ||
274 | if (e->ops->elevator_merged_fn) | |
275 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
276 | |
277 | q->last_merge = rq; | |
1da177e4 LT |
278 | } |
279 | ||
280 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
281 | struct request *next) | |
282 | { | |
283 | elevator_t *e = q->elevator; | |
284 | ||
1da177e4 LT |
285 | if (e->ops->elevator_merge_req_fn) |
286 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 TH |
287 | |
288 | q->last_merge = rq; | |
1da177e4 LT |
289 | } |
290 | ||
8922e16c | 291 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
292 | { |
293 | elevator_t *e = q->elevator; | |
294 | ||
295 | /* | |
296 | * it already went through dequeue, we need to decrement the | |
297 | * in_flight count again | |
298 | */ | |
8922e16c | 299 | if (blk_account_rq(rq)) { |
1da177e4 | 300 | q->in_flight--; |
8922e16c TH |
301 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
302 | e->ops->elevator_deactivate_req_fn(q, rq); | |
303 | } | |
1da177e4 LT |
304 | |
305 | rq->flags &= ~REQ_STARTED; | |
306 | ||
1da177e4 LT |
307 | /* |
308 | * if this is the flush, requeue the original instead and drop the flush | |
309 | */ | |
310 | if (rq->flags & REQ_BAR_FLUSH) { | |
311 | clear_bit(QUEUE_FLAG_FLUSH, &q->queue_flags); | |
312 | rq = rq->end_io_data; | |
313 | } | |
314 | ||
8922e16c | 315 | __elv_add_request(q, rq, ELEVATOR_INSERT_FRONT, 0); |
1da177e4 LT |
316 | } |
317 | ||
318 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, | |
319 | int plug) | |
320 | { | |
8922e16c TH |
321 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
322 | /* | |
323 | * barriers implicitly indicate back insertion | |
324 | */ | |
325 | if (where == ELEVATOR_INSERT_SORT) | |
326 | where = ELEVATOR_INSERT_BACK; | |
327 | ||
328 | /* | |
1b47f531 | 329 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
330 | */ |
331 | if (blk_fs_request(rq)) { | |
1b47f531 | 332 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
333 | q->boundary_rq = rq; |
334 | } | |
cb98fc8b TH |
335 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
336 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
337 | |
338 | if (plug) | |
339 | blk_plug_device(q); | |
340 | ||
341 | rq->q = q; | |
342 | ||
8922e16c TH |
343 | switch (where) { |
344 | case ELEVATOR_INSERT_FRONT: | |
345 | rq->flags |= REQ_SOFTBARRIER; | |
346 | ||
347 | list_add(&rq->queuelist, &q->queue_head); | |
348 | break; | |
349 | ||
350 | case ELEVATOR_INSERT_BACK: | |
351 | rq->flags |= REQ_SOFTBARRIER; | |
352 | ||
353 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
354 | ; | |
355 | list_add_tail(&rq->queuelist, &q->queue_head); | |
356 | /* | |
357 | * We kick the queue here for the following reasons. | |
358 | * - The elevator might have returned NULL previously | |
359 | * to delay requests and returned them now. As the | |
360 | * queue wasn't empty before this request, ll_rw_blk | |
361 | * won't run the queue on return, resulting in hang. | |
362 | * - Usually, back inserted requests won't be merged | |
363 | * with anything. There's no point in delaying queue | |
364 | * processing. | |
365 | */ | |
366 | blk_remove_plug(q); | |
367 | q->request_fn(q); | |
368 | break; | |
369 | ||
370 | case ELEVATOR_INSERT_SORT: | |
371 | BUG_ON(!blk_fs_request(rq)); | |
372 | rq->flags |= REQ_SORTED; | |
06b86245 TH |
373 | if (q->last_merge == NULL && rq_mergeable(rq)) |
374 | q->last_merge = rq; | |
ca23509f TH |
375 | /* |
376 | * Some ioscheds (cfq) run q->request_fn directly, so | |
377 | * rq cannot be accessed after calling | |
378 | * elevator_add_req_fn. | |
379 | */ | |
380 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
381 | break; |
382 | ||
383 | default: | |
384 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
385 | __FUNCTION__, where); | |
386 | BUG(); | |
387 | } | |
388 | ||
389 | if (blk_queue_plugged(q)) { | |
390 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
391 | - q->in_flight; | |
392 | ||
393 | if (nrq >= q->unplug_thresh) | |
394 | __generic_unplug_device(q); | |
395 | } | |
1da177e4 LT |
396 | } |
397 | ||
398 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
399 | int plug) | |
400 | { | |
401 | unsigned long flags; | |
402 | ||
403 | spin_lock_irqsave(q->queue_lock, flags); | |
404 | __elv_add_request(q, rq, where, plug); | |
405 | spin_unlock_irqrestore(q->queue_lock, flags); | |
406 | } | |
407 | ||
408 | static inline struct request *__elv_next_request(request_queue_t *q) | |
409 | { | |
8922e16c TH |
410 | struct request *rq; |
411 | ||
412 | if (unlikely(list_empty(&q->queue_head) && | |
413 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
414 | return NULL; | |
415 | ||
416 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
417 | |
418 | /* | |
419 | * if this is a barrier write and the device has to issue a | |
420 | * flush sequence to support it, check how far we are | |
421 | */ | |
8922e16c | 422 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
423 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
424 | ||
425 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
426 | !blk_barrier_preflush(rq)) | |
427 | rq = blk_start_pre_flush(q, rq); | |
428 | } | |
429 | ||
430 | return rq; | |
431 | } | |
432 | ||
433 | struct request *elv_next_request(request_queue_t *q) | |
434 | { | |
435 | struct request *rq; | |
436 | int ret; | |
437 | ||
438 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
439 | if (!(rq->flags & REQ_STARTED)) { |
440 | elevator_t *e = q->elevator; | |
441 | ||
442 | /* | |
443 | * This is the first time the device driver | |
444 | * sees this request (possibly after | |
445 | * requeueing). Notify IO scheduler. | |
446 | */ | |
447 | if (blk_sorted_rq(rq) && | |
448 | e->ops->elevator_activate_req_fn) | |
449 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 450 | |
8922e16c TH |
451 | /* |
452 | * just mark as started even if we don't start | |
453 | * it, a request that has been delayed should | |
454 | * not be passed by new incoming requests | |
455 | */ | |
456 | rq->flags |= REQ_STARTED; | |
457 | } | |
1da177e4 | 458 | |
8922e16c | 459 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 460 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
461 | q->boundary_rq = NULL; |
462 | } | |
1da177e4 LT |
463 | |
464 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
465 | break; | |
466 | ||
467 | ret = q->prep_rq_fn(q, rq); | |
468 | if (ret == BLKPREP_OK) { | |
469 | break; | |
470 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
471 | /* |
472 | * the request may have been (partially) prepped. | |
473 | * we need to keep this request in the front to | |
8922e16c TH |
474 | * avoid resource deadlock. REQ_STARTED will |
475 | * prevent other fs requests from passing this one. | |
2e759cd4 | 476 | */ |
1da177e4 LT |
477 | rq = NULL; |
478 | break; | |
479 | } else if (ret == BLKPREP_KILL) { | |
480 | int nr_bytes = rq->hard_nr_sectors << 9; | |
481 | ||
482 | if (!nr_bytes) | |
483 | nr_bytes = rq->data_len; | |
484 | ||
485 | blkdev_dequeue_request(rq); | |
486 | rq->flags |= REQ_QUIET; | |
487 | end_that_request_chunk(rq, 0, nr_bytes); | |
488 | end_that_request_last(rq); | |
489 | } else { | |
490 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
491 | ret); | |
492 | break; | |
493 | } | |
494 | } | |
495 | ||
496 | return rq; | |
497 | } | |
498 | ||
8922e16c | 499 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 500 | { |
8922e16c TH |
501 | BUG_ON(list_empty(&rq->queuelist)); |
502 | ||
503 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
504 | |
505 | /* | |
506 | * the time frame between a request being removed from the lists | |
507 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 508 | * the driver side. |
1da177e4 LT |
509 | */ |
510 | if (blk_account_rq(rq)) | |
511 | q->in_flight++; | |
1da177e4 LT |
512 | } |
513 | ||
514 | int elv_queue_empty(request_queue_t *q) | |
515 | { | |
516 | elevator_t *e = q->elevator; | |
517 | ||
8922e16c TH |
518 | if (!list_empty(&q->queue_head)) |
519 | return 0; | |
520 | ||
1da177e4 LT |
521 | if (e->ops->elevator_queue_empty_fn) |
522 | return e->ops->elevator_queue_empty_fn(q); | |
523 | ||
8922e16c | 524 | return 1; |
1da177e4 LT |
525 | } |
526 | ||
527 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
528 | { | |
529 | struct list_head *next; | |
530 | ||
531 | elevator_t *e = q->elevator; | |
532 | ||
533 | if (e->ops->elevator_latter_req_fn) | |
534 | return e->ops->elevator_latter_req_fn(q, rq); | |
535 | ||
536 | next = rq->queuelist.next; | |
537 | if (next != &q->queue_head && next != &rq->queuelist) | |
538 | return list_entry_rq(next); | |
539 | ||
540 | return NULL; | |
541 | } | |
542 | ||
543 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
544 | { | |
545 | struct list_head *prev; | |
546 | ||
547 | elevator_t *e = q->elevator; | |
548 | ||
549 | if (e->ops->elevator_former_req_fn) | |
550 | return e->ops->elevator_former_req_fn(q, rq); | |
551 | ||
552 | prev = rq->queuelist.prev; | |
553 | if (prev != &q->queue_head && prev != &rq->queuelist) | |
554 | return list_entry_rq(prev); | |
555 | ||
556 | return NULL; | |
557 | } | |
558 | ||
22e2c507 | 559 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 560 | gfp_t gfp_mask) |
1da177e4 LT |
561 | { |
562 | elevator_t *e = q->elevator; | |
563 | ||
564 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 565 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
566 | |
567 | rq->elevator_private = NULL; | |
568 | return 0; | |
569 | } | |
570 | ||
571 | void elv_put_request(request_queue_t *q, struct request *rq) | |
572 | { | |
573 | elevator_t *e = q->elevator; | |
574 | ||
575 | if (e->ops->elevator_put_req_fn) | |
576 | e->ops->elevator_put_req_fn(q, rq); | |
577 | } | |
578 | ||
22e2c507 | 579 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
580 | { |
581 | elevator_t *e = q->elevator; | |
582 | ||
583 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 584 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
585 | |
586 | return ELV_MQUEUE_MAY; | |
587 | } | |
588 | ||
589 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
590 | { | |
591 | elevator_t *e = q->elevator; | |
592 | ||
593 | /* | |
594 | * request is released from the driver, io must be done | |
595 | */ | |
8922e16c | 596 | if (blk_account_rq(rq)) { |
1da177e4 | 597 | q->in_flight--; |
8922e16c TH |
598 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
599 | e->ops->elevator_completed_req_fn(q, rq); | |
600 | } | |
1da177e4 LT |
601 | } |
602 | ||
603 | int elv_register_queue(struct request_queue *q) | |
604 | { | |
605 | elevator_t *e = q->elevator; | |
606 | ||
607 | e->kobj.parent = kobject_get(&q->kobj); | |
608 | if (!e->kobj.parent) | |
609 | return -EBUSY; | |
610 | ||
611 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
612 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
613 | ||
614 | return kobject_register(&e->kobj); | |
615 | } | |
616 | ||
617 | void elv_unregister_queue(struct request_queue *q) | |
618 | { | |
619 | if (q) { | |
620 | elevator_t *e = q->elevator; | |
621 | kobject_unregister(&e->kobj); | |
622 | kobject_put(&q->kobj); | |
623 | } | |
624 | } | |
625 | ||
626 | int elv_register(struct elevator_type *e) | |
627 | { | |
2824bc93 | 628 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
629 | if (elevator_find(e->elevator_name)) |
630 | BUG(); | |
1da177e4 LT |
631 | list_add_tail(&e->list, &elv_list); |
632 | spin_unlock_irq(&elv_list_lock); | |
633 | ||
634 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
635 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
636 | printk(" (default)"); | |
637 | printk("\n"); | |
638 | return 0; | |
639 | } | |
640 | EXPORT_SYMBOL_GPL(elv_register); | |
641 | ||
642 | void elv_unregister(struct elevator_type *e) | |
643 | { | |
83521d3e CH |
644 | struct task_struct *g, *p; |
645 | ||
646 | /* | |
647 | * Iterate every thread in the process to remove the io contexts. | |
648 | */ | |
649 | read_lock(&tasklist_lock); | |
650 | do_each_thread(g, p) { | |
651 | struct io_context *ioc = p->io_context; | |
652 | if (ioc && ioc->cic) { | |
653 | ioc->cic->exit(ioc->cic); | |
654 | ioc->cic->dtor(ioc->cic); | |
655 | ioc->cic = NULL; | |
656 | } | |
657 | if (ioc && ioc->aic) { | |
658 | ioc->aic->exit(ioc->aic); | |
659 | ioc->aic->dtor(ioc->aic); | |
660 | ioc->aic = NULL; | |
661 | } | |
662 | } while_each_thread(g, p); | |
663 | read_unlock(&tasklist_lock); | |
664 | ||
1da177e4 LT |
665 | spin_lock_irq(&elv_list_lock); |
666 | list_del_init(&e->list); | |
667 | spin_unlock_irq(&elv_list_lock); | |
668 | } | |
669 | EXPORT_SYMBOL_GPL(elv_unregister); | |
670 | ||
671 | /* | |
672 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
673 | * we don't free the old io scheduler, before we have allocated what we | |
674 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 675 | * one, if the new one fails init for some reason. |
1da177e4 LT |
676 | */ |
677 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
678 | { | |
cb98fc8b | 679 | elevator_t *old_elevator, *e; |
1da177e4 | 680 | |
cb98fc8b TH |
681 | /* |
682 | * Allocate new elevator | |
683 | */ | |
684 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
685 | if (!e) |
686 | goto error; | |
687 | ||
688 | /* | |
cb98fc8b | 689 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 690 | */ |
cb98fc8b TH |
691 | spin_lock_irq(q->queue_lock); |
692 | ||
64521d1a | 693 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b TH |
694 | |
695 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
696 | ; | |
697 | ||
698 | while (q->rq.elvpriv) { | |
407df2aa TH |
699 | blk_remove_plug(q); |
700 | q->request_fn(q); | |
cb98fc8b | 701 | spin_unlock_irq(q->queue_lock); |
64521d1a | 702 | msleep(10); |
cb98fc8b TH |
703 | spin_lock_irq(q->queue_lock); |
704 | } | |
705 | ||
706 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
707 | |
708 | /* | |
709 | * unregister old elevator data | |
710 | */ | |
711 | elv_unregister_queue(q); | |
712 | old_elevator = q->elevator; | |
713 | ||
1da177e4 LT |
714 | /* |
715 | * attach and start new elevator | |
716 | */ | |
717 | if (elevator_attach(q, new_e, e)) | |
718 | goto fail; | |
719 | ||
720 | if (elv_register_queue(q)) | |
721 | goto fail_register; | |
722 | ||
723 | /* | |
cb98fc8b | 724 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
725 | */ |
726 | elevator_exit(old_elevator); | |
64521d1a | 727 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
728 | return; |
729 | ||
730 | fail_register: | |
731 | /* | |
732 | * switch failed, exit the new io scheduler and reattach the old | |
733 | * one again (along with re-adding the sysfs dir) | |
734 | */ | |
735 | elevator_exit(e); | |
cb98fc8b | 736 | e = NULL; |
1da177e4 LT |
737 | fail: |
738 | q->elevator = old_elevator; | |
739 | elv_register_queue(q); | |
64521d1a | 740 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 741 | kfree(e); |
1da177e4 | 742 | error: |
1da177e4 LT |
743 | elevator_put(new_e); |
744 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
745 | } | |
746 | ||
747 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
748 | { | |
749 | char elevator_name[ELV_NAME_MAX]; | |
750 | struct elevator_type *e; | |
751 | ||
752 | memset(elevator_name, 0, sizeof(elevator_name)); | |
753 | strncpy(elevator_name, name, sizeof(elevator_name)); | |
754 | ||
755 | if (elevator_name[strlen(elevator_name) - 1] == '\n') | |
756 | elevator_name[strlen(elevator_name) - 1] = '\0'; | |
757 | ||
758 | e = elevator_get(elevator_name); | |
759 | if (!e) { | |
760 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
761 | return -EINVAL; | |
762 | } | |
763 | ||
2ca7d93b ND |
764 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
765 | elevator_put(e); | |
1da177e4 | 766 | return count; |
2ca7d93b | 767 | } |
1da177e4 LT |
768 | |
769 | elevator_switch(q, e); | |
770 | return count; | |
771 | } | |
772 | ||
773 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
774 | { | |
775 | elevator_t *e = q->elevator; | |
776 | struct elevator_type *elv = e->elevator_type; | |
777 | struct list_head *entry; | |
778 | int len = 0; | |
779 | ||
780 | spin_lock_irq(q->queue_lock); | |
781 | list_for_each(entry, &elv_list) { | |
782 | struct elevator_type *__e; | |
783 | ||
784 | __e = list_entry(entry, struct elevator_type, list); | |
785 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
786 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
787 | else | |
788 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
789 | } | |
790 | spin_unlock_irq(q->queue_lock); | |
791 | ||
792 | len += sprintf(len+name, "\n"); | |
793 | return len; | |
794 | } | |
795 | ||
1b47f531 | 796 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
797 | EXPORT_SYMBOL(elv_add_request); |
798 | EXPORT_SYMBOL(__elv_add_request); | |
799 | EXPORT_SYMBOL(elv_requeue_request); | |
800 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 801 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
802 | EXPORT_SYMBOL(elv_queue_empty); |
803 | EXPORT_SYMBOL(elv_completed_request); | |
804 | EXPORT_SYMBOL(elevator_exit); | |
805 | EXPORT_SYMBOL(elevator_init); |