]>
Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
1da177e4 LT |
2 | * Block device elevator/IO-scheduler. |
3 | * | |
4 | * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
5 | * | |
6 | * 30042000 Jens Axboe <axboe@suse.de> : | |
7 | * | |
8 | * Split the elevator a bit so that it is possible to choose a different | |
9 | * one or even write a new "plug in". There are three pieces: | |
10 | * - elevator_fn, inserts a new request in the queue list | |
11 | * - elevator_merge_fn, decides whether a new buffer can be merged with | |
12 | * an existing request | |
13 | * - elevator_dequeue_fn, called when a request is taken off the active list | |
14 | * | |
15 | * 20082000 Dave Jones <davej@suse.de> : | |
16 | * Removed tests for max-bomb-segments, which was breaking elvtune | |
17 | * when run without -bN | |
18 | * | |
19 | * Jens: | |
20 | * - Rework again to work with bio instead of buffer_heads | |
21 | * - loose bi_dev comparisons, partition handling is right now | |
22 | * - completely modularize elevator setup and teardown | |
23 | * | |
24 | */ | |
25 | #include <linux/kernel.h> | |
26 | #include <linux/fs.h> | |
27 | #include <linux/blkdev.h> | |
28 | #include <linux/elevator.h> | |
29 | #include <linux/bio.h> | |
30 | #include <linux/config.h> | |
31 | #include <linux/module.h> | |
32 | #include <linux/slab.h> | |
33 | #include <linux/init.h> | |
34 | #include <linux/compiler.h> | |
cb98fc8b | 35 | #include <linux/delay.h> |
1da177e4 LT |
36 | |
37 | #include <asm/uaccess.h> | |
38 | ||
39 | static DEFINE_SPINLOCK(elv_list_lock); | |
40 | static LIST_HEAD(elv_list); | |
41 | ||
42 | /* | |
43 | * can we safely merge with this request? | |
44 | */ | |
45 | inline int elv_rq_merge_ok(struct request *rq, struct bio *bio) | |
46 | { | |
47 | if (!rq_mergeable(rq)) | |
48 | return 0; | |
49 | ||
50 | /* | |
51 | * different data direction or already started, don't merge | |
52 | */ | |
53 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
54 | return 0; | |
55 | ||
56 | /* | |
57 | * same device and no special stuff set, merge is ok | |
58 | */ | |
59 | if (rq->rq_disk == bio->bi_bdev->bd_disk && | |
60 | !rq->waiting && !rq->special) | |
61 | return 1; | |
62 | ||
63 | return 0; | |
64 | } | |
65 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
66 | ||
769db45b | 67 | static inline int elv_try_merge(struct request *__rq, struct bio *bio) |
1da177e4 LT |
68 | { |
69 | int ret = ELEVATOR_NO_MERGE; | |
70 | ||
71 | /* | |
72 | * we can merge and sequence is ok, check if it's possible | |
73 | */ | |
74 | if (elv_rq_merge_ok(__rq, bio)) { | |
75 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
76 | ret = ELEVATOR_BACK_MERGE; | |
77 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
78 | ret = ELEVATOR_FRONT_MERGE; | |
79 | } | |
80 | ||
81 | return ret; | |
82 | } | |
1da177e4 | 83 | |
1da177e4 LT |
84 | static struct elevator_type *elevator_find(const char *name) |
85 | { | |
86 | struct elevator_type *e = NULL; | |
87 | struct list_head *entry; | |
88 | ||
1da177e4 LT |
89 | list_for_each(entry, &elv_list) { |
90 | struct elevator_type *__e; | |
91 | ||
92 | __e = list_entry(entry, struct elevator_type, list); | |
93 | ||
94 | if (!strcmp(__e->elevator_name, name)) { | |
95 | e = __e; | |
96 | break; | |
97 | } | |
98 | } | |
1da177e4 LT |
99 | |
100 | return e; | |
101 | } | |
102 | ||
103 | static void elevator_put(struct elevator_type *e) | |
104 | { | |
105 | module_put(e->elevator_owner); | |
106 | } | |
107 | ||
108 | static struct elevator_type *elevator_get(const char *name) | |
109 | { | |
2824bc93 | 110 | struct elevator_type *e; |
1da177e4 | 111 | |
2824bc93 TH |
112 | spin_lock_irq(&elv_list_lock); |
113 | ||
114 | e = elevator_find(name); | |
115 | if (e && !try_module_get(e->elevator_owner)) | |
116 | e = NULL; | |
117 | ||
118 | spin_unlock_irq(&elv_list_lock); | |
1da177e4 LT |
119 | |
120 | return e; | |
121 | } | |
122 | ||
123 | static int elevator_attach(request_queue_t *q, struct elevator_type *e, | |
124 | struct elevator_queue *eq) | |
125 | { | |
126 | int ret = 0; | |
127 | ||
128 | memset(eq, 0, sizeof(*eq)); | |
129 | eq->ops = &e->ops; | |
130 | eq->elevator_type = e; | |
131 | ||
1da177e4 LT |
132 | q->elevator = eq; |
133 | ||
134 | if (eq->ops->elevator_init_fn) | |
135 | ret = eq->ops->elevator_init_fn(q, eq); | |
136 | ||
137 | return ret; | |
138 | } | |
139 | ||
140 | static char chosen_elevator[16]; | |
141 | ||
142 | static void elevator_setup_default(void) | |
143 | { | |
2824bc93 TH |
144 | struct elevator_type *e; |
145 | ||
1da177e4 | 146 | /* |
131dda7f | 147 | * If default has not been set, use the compiled-in selection. |
1da177e4 | 148 | */ |
131dda7f ND |
149 | if (!chosen_elevator[0]) |
150 | strcpy(chosen_elevator, CONFIG_DEFAULT_IOSCHED); | |
151 | ||
752a3b79 CE |
152 | /* |
153 | * Be backwards-compatible with previous kernels, so users | |
154 | * won't get the wrong elevator. | |
155 | */ | |
156 | if (!strcmp(chosen_elevator, "as")) | |
157 | strcpy(chosen_elevator, "anticipatory"); | |
158 | ||
131dda7f | 159 | /* |
b7bfcf7c | 160 | * If the given scheduler is not available, fall back to the default |
131dda7f | 161 | */ |
cff3ba22 ZA |
162 | if ((e = elevator_find(chosen_elevator))) |
163 | elevator_put(e); | |
164 | else | |
b7bfcf7c | 165 | strcpy(chosen_elevator, CONFIG_DEFAULT_IOSCHED); |
1da177e4 LT |
166 | } |
167 | ||
168 | static int __init elevator_setup(char *str) | |
169 | { | |
170 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
171 | return 0; | |
172 | } | |
173 | ||
174 | __setup("elevator=", elevator_setup); | |
175 | ||
176 | int elevator_init(request_queue_t *q, char *name) | |
177 | { | |
178 | struct elevator_type *e = NULL; | |
179 | struct elevator_queue *eq; | |
180 | int ret = 0; | |
181 | ||
cb98fc8b TH |
182 | INIT_LIST_HEAD(&q->queue_head); |
183 | q->last_merge = NULL; | |
184 | q->end_sector = 0; | |
185 | q->boundary_rq = NULL; | |
cb98fc8b | 186 | |
1da177e4 LT |
187 | elevator_setup_default(); |
188 | ||
189 | if (!name) | |
190 | name = chosen_elevator; | |
191 | ||
192 | e = elevator_get(name); | |
193 | if (!e) | |
194 | return -EINVAL; | |
195 | ||
196 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
197 | if (!eq) { | |
b8ea2cb5 | 198 | elevator_put(e); |
1da177e4 LT |
199 | return -ENOMEM; |
200 | } | |
201 | ||
202 | ret = elevator_attach(q, e, eq); | |
203 | if (ret) { | |
204 | kfree(eq); | |
b8ea2cb5 | 205 | elevator_put(e); |
1da177e4 LT |
206 | } |
207 | ||
208 | return ret; | |
209 | } | |
210 | ||
211 | void elevator_exit(elevator_t *e) | |
212 | { | |
213 | if (e->ops->elevator_exit_fn) | |
214 | e->ops->elevator_exit_fn(e); | |
215 | ||
216 | elevator_put(e->elevator_type); | |
217 | e->elevator_type = NULL; | |
218 | kfree(e); | |
219 | } | |
220 | ||
8922e16c TH |
221 | /* |
222 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
223 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
224 | * appended to the dispatch queue. To be used by specific elevators. | |
225 | */ | |
1b47f531 | 226 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
227 | { |
228 | sector_t boundary; | |
8922e16c TH |
229 | struct list_head *entry; |
230 | ||
06b86245 TH |
231 | if (q->last_merge == rq) |
232 | q->last_merge = NULL; | |
15853af9 | 233 | q->nr_sorted--; |
06b86245 | 234 | |
1b47f531 | 235 | boundary = q->end_sector; |
cb19833d | 236 | |
8922e16c TH |
237 | list_for_each_prev(entry, &q->queue_head) { |
238 | struct request *pos = list_entry_rq(entry); | |
239 | ||
240 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
241 | break; | |
242 | if (rq->sector >= boundary) { | |
243 | if (pos->sector < boundary) | |
244 | continue; | |
245 | } else { | |
246 | if (pos->sector >= boundary) | |
247 | break; | |
248 | } | |
249 | if (rq->sector >= pos->sector) | |
250 | break; | |
251 | } | |
252 | ||
253 | list_add(&rq->queuelist, entry); | |
254 | } | |
255 | ||
1da177e4 LT |
256 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
257 | { | |
258 | elevator_t *e = q->elevator; | |
06b86245 TH |
259 | int ret; |
260 | ||
261 | if (q->last_merge) { | |
262 | ret = elv_try_merge(q->last_merge, bio); | |
263 | if (ret != ELEVATOR_NO_MERGE) { | |
264 | *req = q->last_merge; | |
265 | return ret; | |
266 | } | |
267 | } | |
1da177e4 LT |
268 | |
269 | if (e->ops->elevator_merge_fn) | |
270 | return e->ops->elevator_merge_fn(q, req, bio); | |
271 | ||
272 | return ELEVATOR_NO_MERGE; | |
273 | } | |
274 | ||
275 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
276 | { | |
277 | elevator_t *e = q->elevator; | |
278 | ||
279 | if (e->ops->elevator_merged_fn) | |
280 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
281 | |
282 | q->last_merge = rq; | |
1da177e4 LT |
283 | } |
284 | ||
285 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
286 | struct request *next) | |
287 | { | |
288 | elevator_t *e = q->elevator; | |
289 | ||
1da177e4 LT |
290 | if (e->ops->elevator_merge_req_fn) |
291 | e->ops->elevator_merge_req_fn(q, rq, next); | |
15853af9 | 292 | q->nr_sorted--; |
06b86245 TH |
293 | |
294 | q->last_merge = rq; | |
1da177e4 LT |
295 | } |
296 | ||
8922e16c | 297 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
298 | { |
299 | elevator_t *e = q->elevator; | |
300 | ||
301 | /* | |
302 | * it already went through dequeue, we need to decrement the | |
303 | * in_flight count again | |
304 | */ | |
8922e16c | 305 | if (blk_account_rq(rq)) { |
1da177e4 | 306 | q->in_flight--; |
8922e16c TH |
307 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
308 | e->ops->elevator_deactivate_req_fn(q, rq); | |
309 | } | |
1da177e4 LT |
310 | |
311 | rq->flags &= ~REQ_STARTED; | |
312 | ||
797e7dbb | 313 | __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE, 0); |
1da177e4 LT |
314 | } |
315 | ||
15853af9 TH |
316 | static void elv_drain_elevator(request_queue_t *q) |
317 | { | |
318 | static int printed; | |
319 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
320 | ; | |
321 | if (q->nr_sorted == 0) | |
322 | return; | |
323 | if (printed++ < 10) { | |
324 | printk(KERN_ERR "%s: forced dispatching is broken " | |
325 | "(nr_sorted=%u), please report this\n", | |
326 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
327 | } | |
328 | } | |
329 | ||
1da177e4 LT |
330 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, |
331 | int plug) | |
332 | { | |
797e7dbb TH |
333 | struct list_head *pos; |
334 | unsigned ordseq; | |
335 | ||
336 | if (q->ordcolor) | |
337 | rq->flags |= REQ_ORDERED_COLOR; | |
338 | ||
8922e16c | 339 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
797e7dbb TH |
340 | /* |
341 | * toggle ordered color | |
342 | */ | |
343 | q->ordcolor ^= 1; | |
344 | ||
8922e16c TH |
345 | /* |
346 | * barriers implicitly indicate back insertion | |
347 | */ | |
348 | if (where == ELEVATOR_INSERT_SORT) | |
349 | where = ELEVATOR_INSERT_BACK; | |
350 | ||
351 | /* | |
1b47f531 | 352 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
353 | */ |
354 | if (blk_fs_request(rq)) { | |
1b47f531 | 355 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
356 | q->boundary_rq = rq; |
357 | } | |
cb98fc8b TH |
358 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
359 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
360 | |
361 | if (plug) | |
362 | blk_plug_device(q); | |
363 | ||
364 | rq->q = q; | |
365 | ||
8922e16c TH |
366 | switch (where) { |
367 | case ELEVATOR_INSERT_FRONT: | |
368 | rq->flags |= REQ_SOFTBARRIER; | |
369 | ||
370 | list_add(&rq->queuelist, &q->queue_head); | |
371 | break; | |
372 | ||
373 | case ELEVATOR_INSERT_BACK: | |
374 | rq->flags |= REQ_SOFTBARRIER; | |
15853af9 | 375 | elv_drain_elevator(q); |
8922e16c TH |
376 | list_add_tail(&rq->queuelist, &q->queue_head); |
377 | /* | |
378 | * We kick the queue here for the following reasons. | |
379 | * - The elevator might have returned NULL previously | |
380 | * to delay requests and returned them now. As the | |
381 | * queue wasn't empty before this request, ll_rw_blk | |
382 | * won't run the queue on return, resulting in hang. | |
383 | * - Usually, back inserted requests won't be merged | |
384 | * with anything. There's no point in delaying queue | |
385 | * processing. | |
386 | */ | |
387 | blk_remove_plug(q); | |
388 | q->request_fn(q); | |
389 | break; | |
390 | ||
391 | case ELEVATOR_INSERT_SORT: | |
392 | BUG_ON(!blk_fs_request(rq)); | |
393 | rq->flags |= REQ_SORTED; | |
15853af9 | 394 | q->nr_sorted++; |
06b86245 TH |
395 | if (q->last_merge == NULL && rq_mergeable(rq)) |
396 | q->last_merge = rq; | |
ca23509f TH |
397 | /* |
398 | * Some ioscheds (cfq) run q->request_fn directly, so | |
399 | * rq cannot be accessed after calling | |
400 | * elevator_add_req_fn. | |
401 | */ | |
402 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
403 | break; |
404 | ||
797e7dbb TH |
405 | case ELEVATOR_INSERT_REQUEUE: |
406 | /* | |
407 | * If ordered flush isn't in progress, we do front | |
408 | * insertion; otherwise, requests should be requeued | |
409 | * in ordseq order. | |
410 | */ | |
411 | rq->flags |= REQ_SOFTBARRIER; | |
412 | ||
413 | if (q->ordseq == 0) { | |
414 | list_add(&rq->queuelist, &q->queue_head); | |
415 | break; | |
416 | } | |
417 | ||
418 | ordseq = blk_ordered_req_seq(rq); | |
419 | ||
420 | list_for_each(pos, &q->queue_head) { | |
421 | struct request *pos_rq = list_entry_rq(pos); | |
422 | if (ordseq <= blk_ordered_req_seq(pos_rq)) | |
423 | break; | |
424 | } | |
425 | ||
426 | list_add_tail(&rq->queuelist, pos); | |
427 | break; | |
428 | ||
8922e16c TH |
429 | default: |
430 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
431 | __FUNCTION__, where); | |
432 | BUG(); | |
433 | } | |
434 | ||
435 | if (blk_queue_plugged(q)) { | |
436 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
437 | - q->in_flight; | |
438 | ||
439 | if (nrq >= q->unplug_thresh) | |
440 | __generic_unplug_device(q); | |
441 | } | |
1da177e4 LT |
442 | } |
443 | ||
444 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
445 | int plug) | |
446 | { | |
447 | unsigned long flags; | |
448 | ||
449 | spin_lock_irqsave(q->queue_lock, flags); | |
450 | __elv_add_request(q, rq, where, plug); | |
451 | spin_unlock_irqrestore(q->queue_lock, flags); | |
452 | } | |
453 | ||
454 | static inline struct request *__elv_next_request(request_queue_t *q) | |
455 | { | |
8922e16c TH |
456 | struct request *rq; |
457 | ||
797e7dbb TH |
458 | while (1) { |
459 | while (!list_empty(&q->queue_head)) { | |
460 | rq = list_entry_rq(q->queue_head.next); | |
461 | if (blk_do_ordered(q, &rq)) | |
462 | return rq; | |
463 | } | |
1da177e4 | 464 | |
797e7dbb TH |
465 | if (!q->elevator->ops->elevator_dispatch_fn(q, 0)) |
466 | return NULL; | |
1da177e4 | 467 | } |
1da177e4 LT |
468 | } |
469 | ||
470 | struct request *elv_next_request(request_queue_t *q) | |
471 | { | |
472 | struct request *rq; | |
473 | int ret; | |
474 | ||
475 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
476 | if (!(rq->flags & REQ_STARTED)) { |
477 | elevator_t *e = q->elevator; | |
478 | ||
479 | /* | |
480 | * This is the first time the device driver | |
481 | * sees this request (possibly after | |
482 | * requeueing). Notify IO scheduler. | |
483 | */ | |
484 | if (blk_sorted_rq(rq) && | |
485 | e->ops->elevator_activate_req_fn) | |
486 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 487 | |
8922e16c TH |
488 | /* |
489 | * just mark as started even if we don't start | |
490 | * it, a request that has been delayed should | |
491 | * not be passed by new incoming requests | |
492 | */ | |
493 | rq->flags |= REQ_STARTED; | |
494 | } | |
1da177e4 | 495 | |
8922e16c | 496 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 497 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
498 | q->boundary_rq = NULL; |
499 | } | |
1da177e4 LT |
500 | |
501 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
502 | break; | |
503 | ||
504 | ret = q->prep_rq_fn(q, rq); | |
505 | if (ret == BLKPREP_OK) { | |
506 | break; | |
507 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
508 | /* |
509 | * the request may have been (partially) prepped. | |
510 | * we need to keep this request in the front to | |
8922e16c TH |
511 | * avoid resource deadlock. REQ_STARTED will |
512 | * prevent other fs requests from passing this one. | |
2e759cd4 | 513 | */ |
1da177e4 LT |
514 | rq = NULL; |
515 | break; | |
516 | } else if (ret == BLKPREP_KILL) { | |
517 | int nr_bytes = rq->hard_nr_sectors << 9; | |
518 | ||
519 | if (!nr_bytes) | |
520 | nr_bytes = rq->data_len; | |
521 | ||
522 | blkdev_dequeue_request(rq); | |
523 | rq->flags |= REQ_QUIET; | |
524 | end_that_request_chunk(rq, 0, nr_bytes); | |
8ffdc655 | 525 | end_that_request_last(rq, 0); |
1da177e4 LT |
526 | } else { |
527 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
528 | ret); | |
529 | break; | |
530 | } | |
531 | } | |
532 | ||
533 | return rq; | |
534 | } | |
535 | ||
8922e16c | 536 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 537 | { |
8922e16c TH |
538 | BUG_ON(list_empty(&rq->queuelist)); |
539 | ||
540 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
541 | |
542 | /* | |
543 | * the time frame between a request being removed from the lists | |
544 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 545 | * the driver side. |
1da177e4 LT |
546 | */ |
547 | if (blk_account_rq(rq)) | |
548 | q->in_flight++; | |
1da177e4 LT |
549 | } |
550 | ||
551 | int elv_queue_empty(request_queue_t *q) | |
552 | { | |
553 | elevator_t *e = q->elevator; | |
554 | ||
8922e16c TH |
555 | if (!list_empty(&q->queue_head)) |
556 | return 0; | |
557 | ||
1da177e4 LT |
558 | if (e->ops->elevator_queue_empty_fn) |
559 | return e->ops->elevator_queue_empty_fn(q); | |
560 | ||
8922e16c | 561 | return 1; |
1da177e4 LT |
562 | } |
563 | ||
564 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
565 | { | |
1da177e4 LT |
566 | elevator_t *e = q->elevator; |
567 | ||
568 | if (e->ops->elevator_latter_req_fn) | |
569 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
570 | return NULL; |
571 | } | |
572 | ||
573 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
574 | { | |
1da177e4 LT |
575 | elevator_t *e = q->elevator; |
576 | ||
577 | if (e->ops->elevator_former_req_fn) | |
578 | return e->ops->elevator_former_req_fn(q, rq); | |
1da177e4 LT |
579 | return NULL; |
580 | } | |
581 | ||
22e2c507 | 582 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 583 | gfp_t gfp_mask) |
1da177e4 LT |
584 | { |
585 | elevator_t *e = q->elevator; | |
586 | ||
587 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 588 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
589 | |
590 | rq->elevator_private = NULL; | |
591 | return 0; | |
592 | } | |
593 | ||
594 | void elv_put_request(request_queue_t *q, struct request *rq) | |
595 | { | |
596 | elevator_t *e = q->elevator; | |
597 | ||
598 | if (e->ops->elevator_put_req_fn) | |
599 | e->ops->elevator_put_req_fn(q, rq); | |
600 | } | |
601 | ||
22e2c507 | 602 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
603 | { |
604 | elevator_t *e = q->elevator; | |
605 | ||
606 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 607 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
608 | |
609 | return ELV_MQUEUE_MAY; | |
610 | } | |
611 | ||
612 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
613 | { | |
614 | elevator_t *e = q->elevator; | |
615 | ||
616 | /* | |
617 | * request is released from the driver, io must be done | |
618 | */ | |
8922e16c | 619 | if (blk_account_rq(rq)) { |
1da177e4 | 620 | q->in_flight--; |
1bc691d3 TH |
621 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
622 | e->ops->elevator_completed_req_fn(q, rq); | |
623 | } | |
797e7dbb | 624 | |
1bc691d3 TH |
625 | /* |
626 | * Check if the queue is waiting for fs requests to be | |
627 | * drained for flush sequence. | |
628 | */ | |
629 | if (unlikely(q->ordseq)) { | |
630 | struct request *first_rq = list_entry_rq(q->queue_head.next); | |
631 | if (q->in_flight == 0 && | |
797e7dbb TH |
632 | blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN && |
633 | blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) { | |
634 | blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0); | |
635 | q->request_fn(q); | |
636 | } | |
8922e16c | 637 | } |
1da177e4 LT |
638 | } |
639 | ||
640 | int elv_register_queue(struct request_queue *q) | |
641 | { | |
642 | elevator_t *e = q->elevator; | |
643 | ||
644 | e->kobj.parent = kobject_get(&q->kobj); | |
645 | if (!e->kobj.parent) | |
646 | return -EBUSY; | |
647 | ||
648 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
649 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
650 | ||
651 | return kobject_register(&e->kobj); | |
652 | } | |
653 | ||
654 | void elv_unregister_queue(struct request_queue *q) | |
655 | { | |
656 | if (q) { | |
657 | elevator_t *e = q->elevator; | |
658 | kobject_unregister(&e->kobj); | |
659 | kobject_put(&q->kobj); | |
660 | } | |
661 | } | |
662 | ||
663 | int elv_register(struct elevator_type *e) | |
664 | { | |
2824bc93 | 665 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
666 | if (elevator_find(e->elevator_name)) |
667 | BUG(); | |
1da177e4 LT |
668 | list_add_tail(&e->list, &elv_list); |
669 | spin_unlock_irq(&elv_list_lock); | |
670 | ||
671 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
672 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
673 | printk(" (default)"); | |
674 | printk("\n"); | |
675 | return 0; | |
676 | } | |
677 | EXPORT_SYMBOL_GPL(elv_register); | |
678 | ||
679 | void elv_unregister(struct elevator_type *e) | |
680 | { | |
83521d3e CH |
681 | struct task_struct *g, *p; |
682 | ||
683 | /* | |
684 | * Iterate every thread in the process to remove the io contexts. | |
685 | */ | |
686 | read_lock(&tasklist_lock); | |
687 | do_each_thread(g, p) { | |
688 | struct io_context *ioc = p->io_context; | |
689 | if (ioc && ioc->cic) { | |
690 | ioc->cic->exit(ioc->cic); | |
691 | ioc->cic->dtor(ioc->cic); | |
692 | ioc->cic = NULL; | |
693 | } | |
694 | if (ioc && ioc->aic) { | |
695 | ioc->aic->exit(ioc->aic); | |
696 | ioc->aic->dtor(ioc->aic); | |
697 | ioc->aic = NULL; | |
698 | } | |
699 | } while_each_thread(g, p); | |
700 | read_unlock(&tasklist_lock); | |
701 | ||
1da177e4 LT |
702 | spin_lock_irq(&elv_list_lock); |
703 | list_del_init(&e->list); | |
704 | spin_unlock_irq(&elv_list_lock); | |
705 | } | |
706 | EXPORT_SYMBOL_GPL(elv_unregister); | |
707 | ||
708 | /* | |
709 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
710 | * we don't free the old io scheduler, before we have allocated what we | |
711 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 712 | * one, if the new one fails init for some reason. |
1da177e4 LT |
713 | */ |
714 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
715 | { | |
cb98fc8b | 716 | elevator_t *old_elevator, *e; |
1da177e4 | 717 | |
cb98fc8b TH |
718 | /* |
719 | * Allocate new elevator | |
720 | */ | |
721 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
722 | if (!e) |
723 | goto error; | |
724 | ||
725 | /* | |
cb98fc8b | 726 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 727 | */ |
cb98fc8b TH |
728 | spin_lock_irq(q->queue_lock); |
729 | ||
64521d1a | 730 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 731 | |
15853af9 | 732 | elv_drain_elevator(q); |
cb98fc8b TH |
733 | |
734 | while (q->rq.elvpriv) { | |
407df2aa TH |
735 | blk_remove_plug(q); |
736 | q->request_fn(q); | |
cb98fc8b | 737 | spin_unlock_irq(q->queue_lock); |
64521d1a | 738 | msleep(10); |
cb98fc8b | 739 | spin_lock_irq(q->queue_lock); |
15853af9 | 740 | elv_drain_elevator(q); |
cb98fc8b TH |
741 | } |
742 | ||
743 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
744 | |
745 | /* | |
746 | * unregister old elevator data | |
747 | */ | |
748 | elv_unregister_queue(q); | |
749 | old_elevator = q->elevator; | |
750 | ||
1da177e4 LT |
751 | /* |
752 | * attach and start new elevator | |
753 | */ | |
754 | if (elevator_attach(q, new_e, e)) | |
755 | goto fail; | |
756 | ||
757 | if (elv_register_queue(q)) | |
758 | goto fail_register; | |
759 | ||
760 | /* | |
cb98fc8b | 761 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
762 | */ |
763 | elevator_exit(old_elevator); | |
64521d1a | 764 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
765 | return; |
766 | ||
767 | fail_register: | |
768 | /* | |
769 | * switch failed, exit the new io scheduler and reattach the old | |
770 | * one again (along with re-adding the sysfs dir) | |
771 | */ | |
772 | elevator_exit(e); | |
cb98fc8b | 773 | e = NULL; |
1da177e4 LT |
774 | fail: |
775 | q->elevator = old_elevator; | |
776 | elv_register_queue(q); | |
64521d1a | 777 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 778 | kfree(e); |
1da177e4 | 779 | error: |
1da177e4 LT |
780 | elevator_put(new_e); |
781 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
782 | } | |
783 | ||
784 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
785 | { | |
786 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 787 | size_t len; |
1da177e4 LT |
788 | struct elevator_type *e; |
789 | ||
be561235 TH |
790 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
791 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
792 | len = strlen(elevator_name); | |
1da177e4 | 793 | |
be561235 TH |
794 | if (len && elevator_name[len - 1] == '\n') |
795 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
796 | |
797 | e = elevator_get(elevator_name); | |
798 | if (!e) { | |
799 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
800 | return -EINVAL; | |
801 | } | |
802 | ||
2ca7d93b ND |
803 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
804 | elevator_put(e); | |
1da177e4 | 805 | return count; |
2ca7d93b | 806 | } |
1da177e4 LT |
807 | |
808 | elevator_switch(q, e); | |
809 | return count; | |
810 | } | |
811 | ||
812 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
813 | { | |
814 | elevator_t *e = q->elevator; | |
815 | struct elevator_type *elv = e->elevator_type; | |
816 | struct list_head *entry; | |
817 | int len = 0; | |
818 | ||
819 | spin_lock_irq(q->queue_lock); | |
820 | list_for_each(entry, &elv_list) { | |
821 | struct elevator_type *__e; | |
822 | ||
823 | __e = list_entry(entry, struct elevator_type, list); | |
824 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
825 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
826 | else | |
827 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
828 | } | |
829 | spin_unlock_irq(q->queue_lock); | |
830 | ||
831 | len += sprintf(len+name, "\n"); | |
832 | return len; | |
833 | } | |
834 | ||
1b47f531 | 835 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
836 | EXPORT_SYMBOL(elv_add_request); |
837 | EXPORT_SYMBOL(__elv_add_request); | |
838 | EXPORT_SYMBOL(elv_requeue_request); | |
839 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 840 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
841 | EXPORT_SYMBOL(elv_queue_empty); |
842 | EXPORT_SYMBOL(elv_completed_request); | |
843 | EXPORT_SYMBOL(elevator_exit); | |
844 | EXPORT_SYMBOL(elevator_init); |