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 | ||
5f003976 | 142 | static int __init elevator_setup(char *str) |
1da177e4 | 143 | { |
752a3b79 CE |
144 | /* |
145 | * Be backwards-compatible with previous kernels, so users | |
146 | * won't get the wrong elevator. | |
147 | */ | |
5f003976 | 148 | if (!strcmp(str, "as")) |
752a3b79 | 149 | strcpy(chosen_elevator, "anticipatory"); |
cff3ba22 | 150 | else |
5f003976 | 151 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); |
1da177e4 LT |
152 | return 0; |
153 | } | |
154 | ||
155 | __setup("elevator=", elevator_setup); | |
156 | ||
157 | int elevator_init(request_queue_t *q, char *name) | |
158 | { | |
159 | struct elevator_type *e = NULL; | |
160 | struct elevator_queue *eq; | |
161 | int ret = 0; | |
162 | ||
cb98fc8b TH |
163 | INIT_LIST_HEAD(&q->queue_head); |
164 | q->last_merge = NULL; | |
165 | q->end_sector = 0; | |
166 | q->boundary_rq = NULL; | |
cb98fc8b | 167 | |
5f003976 | 168 | if (name && !(e = elevator_get(name))) |
1da177e4 LT |
169 | return -EINVAL; |
170 | ||
248d5ca5 ND |
171 | if (!e && *chosen_elevator && !(e = elevator_get(chosen_elevator))) |
172 | printk("I/O scheduler %s not found\n", chosen_elevator); | |
173 | ||
174 | if (!e && !(e = elevator_get(CONFIG_DEFAULT_IOSCHED))) { | |
175 | printk("Default I/O scheduler not found, using no-op\n"); | |
176 | e = elevator_get("noop"); | |
5f003976 ND |
177 | } |
178 | ||
1da177e4 LT |
179 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); |
180 | if (!eq) { | |
b8ea2cb5 | 181 | elevator_put(e); |
1da177e4 LT |
182 | return -ENOMEM; |
183 | } | |
184 | ||
185 | ret = elevator_attach(q, e, eq); | |
186 | if (ret) { | |
187 | kfree(eq); | |
b8ea2cb5 | 188 | elevator_put(e); |
1da177e4 LT |
189 | } |
190 | ||
191 | return ret; | |
192 | } | |
193 | ||
194 | void elevator_exit(elevator_t *e) | |
195 | { | |
196 | if (e->ops->elevator_exit_fn) | |
197 | e->ops->elevator_exit_fn(e); | |
198 | ||
199 | elevator_put(e->elevator_type); | |
200 | e->elevator_type = NULL; | |
201 | kfree(e); | |
202 | } | |
203 | ||
8922e16c TH |
204 | /* |
205 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
206 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
207 | * appended to the dispatch queue. To be used by specific elevators. | |
208 | */ | |
1b47f531 | 209 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
210 | { |
211 | sector_t boundary; | |
8922e16c TH |
212 | struct list_head *entry; |
213 | ||
06b86245 TH |
214 | if (q->last_merge == rq) |
215 | q->last_merge = NULL; | |
15853af9 | 216 | q->nr_sorted--; |
06b86245 | 217 | |
1b47f531 | 218 | boundary = q->end_sector; |
cb19833d | 219 | |
8922e16c TH |
220 | list_for_each_prev(entry, &q->queue_head) { |
221 | struct request *pos = list_entry_rq(entry); | |
222 | ||
223 | if (pos->flags & (REQ_SOFTBARRIER|REQ_HARDBARRIER|REQ_STARTED)) | |
224 | break; | |
225 | if (rq->sector >= boundary) { | |
226 | if (pos->sector < boundary) | |
227 | continue; | |
228 | } else { | |
229 | if (pos->sector >= boundary) | |
230 | break; | |
231 | } | |
232 | if (rq->sector >= pos->sector) | |
233 | break; | |
234 | } | |
235 | ||
236 | list_add(&rq->queuelist, entry); | |
237 | } | |
238 | ||
1da177e4 LT |
239 | int elv_merge(request_queue_t *q, struct request **req, struct bio *bio) |
240 | { | |
241 | elevator_t *e = q->elevator; | |
06b86245 TH |
242 | int ret; |
243 | ||
244 | if (q->last_merge) { | |
245 | ret = elv_try_merge(q->last_merge, bio); | |
246 | if (ret != ELEVATOR_NO_MERGE) { | |
247 | *req = q->last_merge; | |
248 | return ret; | |
249 | } | |
250 | } | |
1da177e4 LT |
251 | |
252 | if (e->ops->elevator_merge_fn) | |
253 | return e->ops->elevator_merge_fn(q, req, bio); | |
254 | ||
255 | return ELEVATOR_NO_MERGE; | |
256 | } | |
257 | ||
258 | void elv_merged_request(request_queue_t *q, struct request *rq) | |
259 | { | |
260 | elevator_t *e = q->elevator; | |
261 | ||
262 | if (e->ops->elevator_merged_fn) | |
263 | e->ops->elevator_merged_fn(q, rq); | |
06b86245 TH |
264 | |
265 | q->last_merge = rq; | |
1da177e4 LT |
266 | } |
267 | ||
268 | void elv_merge_requests(request_queue_t *q, struct request *rq, | |
269 | struct request *next) | |
270 | { | |
271 | elevator_t *e = q->elevator; | |
272 | ||
1da177e4 LT |
273 | if (e->ops->elevator_merge_req_fn) |
274 | e->ops->elevator_merge_req_fn(q, rq, next); | |
15853af9 | 275 | q->nr_sorted--; |
06b86245 TH |
276 | |
277 | q->last_merge = rq; | |
1da177e4 LT |
278 | } |
279 | ||
8922e16c | 280 | void elv_requeue_request(request_queue_t *q, struct request *rq) |
1da177e4 LT |
281 | { |
282 | elevator_t *e = q->elevator; | |
283 | ||
284 | /* | |
285 | * it already went through dequeue, we need to decrement the | |
286 | * in_flight count again | |
287 | */ | |
8922e16c | 288 | if (blk_account_rq(rq)) { |
1da177e4 | 289 | q->in_flight--; |
8922e16c TH |
290 | if (blk_sorted_rq(rq) && e->ops->elevator_deactivate_req_fn) |
291 | e->ops->elevator_deactivate_req_fn(q, rq); | |
292 | } | |
1da177e4 LT |
293 | |
294 | rq->flags &= ~REQ_STARTED; | |
295 | ||
797e7dbb | 296 | __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE, 0); |
1da177e4 LT |
297 | } |
298 | ||
15853af9 TH |
299 | static void elv_drain_elevator(request_queue_t *q) |
300 | { | |
301 | static int printed; | |
302 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
303 | ; | |
304 | if (q->nr_sorted == 0) | |
305 | return; | |
306 | if (printed++ < 10) { | |
307 | printk(KERN_ERR "%s: forced dispatching is broken " | |
308 | "(nr_sorted=%u), please report this\n", | |
309 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
310 | } | |
311 | } | |
312 | ||
1da177e4 LT |
313 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, |
314 | int plug) | |
315 | { | |
797e7dbb TH |
316 | struct list_head *pos; |
317 | unsigned ordseq; | |
318 | ||
319 | if (q->ordcolor) | |
320 | rq->flags |= REQ_ORDERED_COLOR; | |
321 | ||
8922e16c | 322 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
797e7dbb TH |
323 | /* |
324 | * toggle ordered color | |
325 | */ | |
238e7db9 TH |
326 | if (blk_barrier_rq(rq)) |
327 | q->ordcolor ^= 1; | |
797e7dbb | 328 | |
8922e16c TH |
329 | /* |
330 | * barriers implicitly indicate back insertion | |
331 | */ | |
332 | if (where == ELEVATOR_INSERT_SORT) | |
333 | where = ELEVATOR_INSERT_BACK; | |
334 | ||
335 | /* | |
1b47f531 | 336 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
337 | */ |
338 | if (blk_fs_request(rq)) { | |
1b47f531 | 339 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
340 | q->boundary_rq = rq; |
341 | } | |
cb98fc8b TH |
342 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
343 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
344 | |
345 | if (plug) | |
346 | blk_plug_device(q); | |
347 | ||
348 | rq->q = q; | |
349 | ||
8922e16c TH |
350 | switch (where) { |
351 | case ELEVATOR_INSERT_FRONT: | |
352 | rq->flags |= REQ_SOFTBARRIER; | |
353 | ||
354 | list_add(&rq->queuelist, &q->queue_head); | |
355 | break; | |
356 | ||
357 | case ELEVATOR_INSERT_BACK: | |
358 | rq->flags |= REQ_SOFTBARRIER; | |
15853af9 | 359 | elv_drain_elevator(q); |
8922e16c TH |
360 | list_add_tail(&rq->queuelist, &q->queue_head); |
361 | /* | |
362 | * We kick the queue here for the following reasons. | |
363 | * - The elevator might have returned NULL previously | |
364 | * to delay requests and returned them now. As the | |
365 | * queue wasn't empty before this request, ll_rw_blk | |
366 | * won't run the queue on return, resulting in hang. | |
367 | * - Usually, back inserted requests won't be merged | |
368 | * with anything. There's no point in delaying queue | |
369 | * processing. | |
370 | */ | |
371 | blk_remove_plug(q); | |
372 | q->request_fn(q); | |
373 | break; | |
374 | ||
375 | case ELEVATOR_INSERT_SORT: | |
376 | BUG_ON(!blk_fs_request(rq)); | |
377 | rq->flags |= REQ_SORTED; | |
15853af9 | 378 | q->nr_sorted++; |
06b86245 TH |
379 | if (q->last_merge == NULL && rq_mergeable(rq)) |
380 | q->last_merge = rq; | |
ca23509f TH |
381 | /* |
382 | * Some ioscheds (cfq) run q->request_fn directly, so | |
383 | * rq cannot be accessed after calling | |
384 | * elevator_add_req_fn. | |
385 | */ | |
386 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
387 | break; |
388 | ||
797e7dbb TH |
389 | case ELEVATOR_INSERT_REQUEUE: |
390 | /* | |
391 | * If ordered flush isn't in progress, we do front | |
392 | * insertion; otherwise, requests should be requeued | |
393 | * in ordseq order. | |
394 | */ | |
395 | rq->flags |= REQ_SOFTBARRIER; | |
396 | ||
397 | if (q->ordseq == 0) { | |
398 | list_add(&rq->queuelist, &q->queue_head); | |
399 | break; | |
400 | } | |
401 | ||
402 | ordseq = blk_ordered_req_seq(rq); | |
403 | ||
404 | list_for_each(pos, &q->queue_head) { | |
405 | struct request *pos_rq = list_entry_rq(pos); | |
406 | if (ordseq <= blk_ordered_req_seq(pos_rq)) | |
407 | break; | |
408 | } | |
409 | ||
410 | list_add_tail(&rq->queuelist, pos); | |
411 | break; | |
412 | ||
8922e16c TH |
413 | default: |
414 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
415 | __FUNCTION__, where); | |
416 | BUG(); | |
417 | } | |
418 | ||
419 | if (blk_queue_plugged(q)) { | |
420 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
421 | - q->in_flight; | |
422 | ||
423 | if (nrq >= q->unplug_thresh) | |
424 | __generic_unplug_device(q); | |
425 | } | |
1da177e4 LT |
426 | } |
427 | ||
428 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
429 | int plug) | |
430 | { | |
431 | unsigned long flags; | |
432 | ||
433 | spin_lock_irqsave(q->queue_lock, flags); | |
434 | __elv_add_request(q, rq, where, plug); | |
435 | spin_unlock_irqrestore(q->queue_lock, flags); | |
436 | } | |
437 | ||
438 | static inline struct request *__elv_next_request(request_queue_t *q) | |
439 | { | |
8922e16c TH |
440 | struct request *rq; |
441 | ||
797e7dbb TH |
442 | while (1) { |
443 | while (!list_empty(&q->queue_head)) { | |
444 | rq = list_entry_rq(q->queue_head.next); | |
445 | if (blk_do_ordered(q, &rq)) | |
446 | return rq; | |
447 | } | |
1da177e4 | 448 | |
797e7dbb TH |
449 | if (!q->elevator->ops->elevator_dispatch_fn(q, 0)) |
450 | return NULL; | |
1da177e4 | 451 | } |
1da177e4 LT |
452 | } |
453 | ||
454 | struct request *elv_next_request(request_queue_t *q) | |
455 | { | |
456 | struct request *rq; | |
457 | int ret; | |
458 | ||
459 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
460 | if (!(rq->flags & REQ_STARTED)) { |
461 | elevator_t *e = q->elevator; | |
462 | ||
463 | /* | |
464 | * This is the first time the device driver | |
465 | * sees this request (possibly after | |
466 | * requeueing). Notify IO scheduler. | |
467 | */ | |
468 | if (blk_sorted_rq(rq) && | |
469 | e->ops->elevator_activate_req_fn) | |
470 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 471 | |
8922e16c TH |
472 | /* |
473 | * just mark as started even if we don't start | |
474 | * it, a request that has been delayed should | |
475 | * not be passed by new incoming requests | |
476 | */ | |
477 | rq->flags |= REQ_STARTED; | |
478 | } | |
1da177e4 | 479 | |
8922e16c | 480 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 481 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
482 | q->boundary_rq = NULL; |
483 | } | |
1da177e4 LT |
484 | |
485 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
486 | break; | |
487 | ||
488 | ret = q->prep_rq_fn(q, rq); | |
489 | if (ret == BLKPREP_OK) { | |
490 | break; | |
491 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
492 | /* |
493 | * the request may have been (partially) prepped. | |
494 | * we need to keep this request in the front to | |
8922e16c TH |
495 | * avoid resource deadlock. REQ_STARTED will |
496 | * prevent other fs requests from passing this one. | |
2e759cd4 | 497 | */ |
1da177e4 LT |
498 | rq = NULL; |
499 | break; | |
500 | } else if (ret == BLKPREP_KILL) { | |
501 | int nr_bytes = rq->hard_nr_sectors << 9; | |
502 | ||
503 | if (!nr_bytes) | |
504 | nr_bytes = rq->data_len; | |
505 | ||
506 | blkdev_dequeue_request(rq); | |
507 | rq->flags |= REQ_QUIET; | |
508 | end_that_request_chunk(rq, 0, nr_bytes); | |
8ffdc655 | 509 | end_that_request_last(rq, 0); |
1da177e4 LT |
510 | } else { |
511 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
512 | ret); | |
513 | break; | |
514 | } | |
515 | } | |
516 | ||
517 | return rq; | |
518 | } | |
519 | ||
8922e16c | 520 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 521 | { |
8922e16c TH |
522 | BUG_ON(list_empty(&rq->queuelist)); |
523 | ||
524 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
525 | |
526 | /* | |
527 | * the time frame between a request being removed from the lists | |
528 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 529 | * the driver side. |
1da177e4 LT |
530 | */ |
531 | if (blk_account_rq(rq)) | |
532 | q->in_flight++; | |
1da177e4 LT |
533 | } |
534 | ||
535 | int elv_queue_empty(request_queue_t *q) | |
536 | { | |
537 | elevator_t *e = q->elevator; | |
538 | ||
8922e16c TH |
539 | if (!list_empty(&q->queue_head)) |
540 | return 0; | |
541 | ||
1da177e4 LT |
542 | if (e->ops->elevator_queue_empty_fn) |
543 | return e->ops->elevator_queue_empty_fn(q); | |
544 | ||
8922e16c | 545 | return 1; |
1da177e4 LT |
546 | } |
547 | ||
548 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
549 | { | |
1da177e4 LT |
550 | elevator_t *e = q->elevator; |
551 | ||
552 | if (e->ops->elevator_latter_req_fn) | |
553 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
554 | return NULL; |
555 | } | |
556 | ||
557 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
558 | { | |
1da177e4 LT |
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); | |
1da177e4 LT |
563 | return NULL; |
564 | } | |
565 | ||
22e2c507 | 566 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 567 | gfp_t gfp_mask) |
1da177e4 LT |
568 | { |
569 | elevator_t *e = q->elevator; | |
570 | ||
571 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 572 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
573 | |
574 | rq->elevator_private = NULL; | |
575 | return 0; | |
576 | } | |
577 | ||
578 | void elv_put_request(request_queue_t *q, struct request *rq) | |
579 | { | |
580 | elevator_t *e = q->elevator; | |
581 | ||
582 | if (e->ops->elevator_put_req_fn) | |
583 | e->ops->elevator_put_req_fn(q, rq); | |
584 | } | |
585 | ||
22e2c507 | 586 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
587 | { |
588 | elevator_t *e = q->elevator; | |
589 | ||
590 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 591 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
592 | |
593 | return ELV_MQUEUE_MAY; | |
594 | } | |
595 | ||
596 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
597 | { | |
598 | elevator_t *e = q->elevator; | |
599 | ||
600 | /* | |
601 | * request is released from the driver, io must be done | |
602 | */ | |
8922e16c | 603 | if (blk_account_rq(rq)) { |
1da177e4 | 604 | q->in_flight--; |
1bc691d3 TH |
605 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
606 | e->ops->elevator_completed_req_fn(q, rq); | |
607 | } | |
797e7dbb | 608 | |
1bc691d3 TH |
609 | /* |
610 | * Check if the queue is waiting for fs requests to be | |
611 | * drained for flush sequence. | |
612 | */ | |
613 | if (unlikely(q->ordseq)) { | |
614 | struct request *first_rq = list_entry_rq(q->queue_head.next); | |
615 | if (q->in_flight == 0 && | |
797e7dbb TH |
616 | blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN && |
617 | blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) { | |
618 | blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0); | |
619 | q->request_fn(q); | |
620 | } | |
8922e16c | 621 | } |
1da177e4 LT |
622 | } |
623 | ||
624 | int elv_register_queue(struct request_queue *q) | |
625 | { | |
626 | elevator_t *e = q->elevator; | |
627 | ||
628 | e->kobj.parent = kobject_get(&q->kobj); | |
629 | if (!e->kobj.parent) | |
630 | return -EBUSY; | |
631 | ||
632 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
633 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
634 | ||
635 | return kobject_register(&e->kobj); | |
636 | } | |
637 | ||
638 | void elv_unregister_queue(struct request_queue *q) | |
639 | { | |
640 | if (q) { | |
641 | elevator_t *e = q->elevator; | |
642 | kobject_unregister(&e->kobj); | |
643 | kobject_put(&q->kobj); | |
644 | } | |
645 | } | |
646 | ||
647 | int elv_register(struct elevator_type *e) | |
648 | { | |
2824bc93 | 649 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
650 | if (elevator_find(e->elevator_name)) |
651 | BUG(); | |
1da177e4 LT |
652 | list_add_tail(&e->list, &elv_list); |
653 | spin_unlock_irq(&elv_list_lock); | |
654 | ||
655 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
5f003976 ND |
656 | if (!strcmp(e->elevator_name, chosen_elevator) || |
657 | (!*chosen_elevator && | |
658 | !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED))) | |
659 | printk(" (default)"); | |
1da177e4 LT |
660 | printk("\n"); |
661 | return 0; | |
662 | } | |
663 | EXPORT_SYMBOL_GPL(elv_register); | |
664 | ||
665 | void elv_unregister(struct elevator_type *e) | |
666 | { | |
83521d3e CH |
667 | struct task_struct *g, *p; |
668 | ||
669 | /* | |
670 | * Iterate every thread in the process to remove the io contexts. | |
671 | */ | |
672 | read_lock(&tasklist_lock); | |
673 | do_each_thread(g, p) { | |
674 | struct io_context *ioc = p->io_context; | |
675 | if (ioc && ioc->cic) { | |
676 | ioc->cic->exit(ioc->cic); | |
677 | ioc->cic->dtor(ioc->cic); | |
678 | ioc->cic = NULL; | |
679 | } | |
680 | if (ioc && ioc->aic) { | |
681 | ioc->aic->exit(ioc->aic); | |
682 | ioc->aic->dtor(ioc->aic); | |
683 | ioc->aic = NULL; | |
684 | } | |
685 | } while_each_thread(g, p); | |
686 | read_unlock(&tasklist_lock); | |
687 | ||
1da177e4 LT |
688 | spin_lock_irq(&elv_list_lock); |
689 | list_del_init(&e->list); | |
690 | spin_unlock_irq(&elv_list_lock); | |
691 | } | |
692 | EXPORT_SYMBOL_GPL(elv_unregister); | |
693 | ||
694 | /* | |
695 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
696 | * we don't free the old io scheduler, before we have allocated what we | |
697 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 698 | * one, if the new one fails init for some reason. |
1da177e4 LT |
699 | */ |
700 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
701 | { | |
cb98fc8b | 702 | elevator_t *old_elevator, *e; |
1da177e4 | 703 | |
cb98fc8b TH |
704 | /* |
705 | * Allocate new elevator | |
706 | */ | |
707 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
708 | if (!e) |
709 | goto error; | |
710 | ||
711 | /* | |
cb98fc8b | 712 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 713 | */ |
cb98fc8b TH |
714 | spin_lock_irq(q->queue_lock); |
715 | ||
64521d1a | 716 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 717 | |
15853af9 | 718 | elv_drain_elevator(q); |
cb98fc8b TH |
719 | |
720 | while (q->rq.elvpriv) { | |
407df2aa TH |
721 | blk_remove_plug(q); |
722 | q->request_fn(q); | |
cb98fc8b | 723 | spin_unlock_irq(q->queue_lock); |
64521d1a | 724 | msleep(10); |
cb98fc8b | 725 | spin_lock_irq(q->queue_lock); |
15853af9 | 726 | elv_drain_elevator(q); |
cb98fc8b TH |
727 | } |
728 | ||
729 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
730 | |
731 | /* | |
732 | * unregister old elevator data | |
733 | */ | |
734 | elv_unregister_queue(q); | |
735 | old_elevator = q->elevator; | |
736 | ||
1da177e4 LT |
737 | /* |
738 | * attach and start new elevator | |
739 | */ | |
740 | if (elevator_attach(q, new_e, e)) | |
741 | goto fail; | |
742 | ||
743 | if (elv_register_queue(q)) | |
744 | goto fail_register; | |
745 | ||
746 | /* | |
cb98fc8b | 747 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
748 | */ |
749 | elevator_exit(old_elevator); | |
64521d1a | 750 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
751 | return; |
752 | ||
753 | fail_register: | |
754 | /* | |
755 | * switch failed, exit the new io scheduler and reattach the old | |
756 | * one again (along with re-adding the sysfs dir) | |
757 | */ | |
758 | elevator_exit(e); | |
cb98fc8b | 759 | e = NULL; |
1da177e4 LT |
760 | fail: |
761 | q->elevator = old_elevator; | |
762 | elv_register_queue(q); | |
64521d1a | 763 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 764 | kfree(e); |
1da177e4 | 765 | error: |
1da177e4 LT |
766 | elevator_put(new_e); |
767 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
768 | } | |
769 | ||
770 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
771 | { | |
772 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 773 | size_t len; |
1da177e4 LT |
774 | struct elevator_type *e; |
775 | ||
be561235 TH |
776 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
777 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
778 | len = strlen(elevator_name); | |
1da177e4 | 779 | |
be561235 TH |
780 | if (len && elevator_name[len - 1] == '\n') |
781 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
782 | |
783 | e = elevator_get(elevator_name); | |
784 | if (!e) { | |
785 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
786 | return -EINVAL; | |
787 | } | |
788 | ||
2ca7d93b ND |
789 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
790 | elevator_put(e); | |
1da177e4 | 791 | return count; |
2ca7d93b | 792 | } |
1da177e4 LT |
793 | |
794 | elevator_switch(q, e); | |
795 | return count; | |
796 | } | |
797 | ||
798 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
799 | { | |
800 | elevator_t *e = q->elevator; | |
801 | struct elevator_type *elv = e->elevator_type; | |
802 | struct list_head *entry; | |
803 | int len = 0; | |
804 | ||
805 | spin_lock_irq(q->queue_lock); | |
806 | list_for_each(entry, &elv_list) { | |
807 | struct elevator_type *__e; | |
808 | ||
809 | __e = list_entry(entry, struct elevator_type, list); | |
810 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
811 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
812 | else | |
813 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
814 | } | |
815 | spin_unlock_irq(q->queue_lock); | |
816 | ||
817 | len += sprintf(len+name, "\n"); | |
818 | return len; | |
819 | } | |
820 | ||
1b47f531 | 821 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
822 | EXPORT_SYMBOL(elv_add_request); |
823 | EXPORT_SYMBOL(__elv_add_request); | |
824 | EXPORT_SYMBOL(elv_requeue_request); | |
825 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 826 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
827 | EXPORT_SYMBOL(elv_queue_empty); |
828 | EXPORT_SYMBOL(elv_completed_request); | |
829 | EXPORT_SYMBOL(elevator_exit); | |
830 | EXPORT_SYMBOL(elevator_init); |