]>
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 | ||
67 | inline int elv_try_merge(struct request *__rq, struct bio *bio) | |
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 | } | |
83 | EXPORT_SYMBOL(elv_try_merge); | |
84 | ||
1da177e4 LT |
85 | static struct elevator_type *elevator_find(const char *name) |
86 | { | |
87 | struct elevator_type *e = NULL; | |
88 | struct list_head *entry; | |
89 | ||
1da177e4 LT |
90 | list_for_each(entry, &elv_list) { |
91 | struct elevator_type *__e; | |
92 | ||
93 | __e = list_entry(entry, struct elevator_type, list); | |
94 | ||
95 | if (!strcmp(__e->elevator_name, name)) { | |
96 | e = __e; | |
97 | break; | |
98 | } | |
99 | } | |
1da177e4 LT |
100 | |
101 | return e; | |
102 | } | |
103 | ||
104 | static void elevator_put(struct elevator_type *e) | |
105 | { | |
106 | module_put(e->elevator_owner); | |
107 | } | |
108 | ||
109 | static struct elevator_type *elevator_get(const char *name) | |
110 | { | |
2824bc93 | 111 | struct elevator_type *e; |
1da177e4 | 112 | |
2824bc93 TH |
113 | spin_lock_irq(&elv_list_lock); |
114 | ||
115 | e = elevator_find(name); | |
116 | if (e && !try_module_get(e->elevator_owner)) | |
117 | e = NULL; | |
118 | ||
119 | spin_unlock_irq(&elv_list_lock); | |
1da177e4 LT |
120 | |
121 | return e; | |
122 | } | |
123 | ||
124 | static int elevator_attach(request_queue_t *q, struct elevator_type *e, | |
125 | struct elevator_queue *eq) | |
126 | { | |
127 | int ret = 0; | |
128 | ||
129 | memset(eq, 0, sizeof(*eq)); | |
130 | eq->ops = &e->ops; | |
131 | eq->elevator_type = e; | |
132 | ||
1da177e4 LT |
133 | q->elevator = eq; |
134 | ||
135 | if (eq->ops->elevator_init_fn) | |
136 | ret = eq->ops->elevator_init_fn(q, eq); | |
137 | ||
138 | return ret; | |
139 | } | |
140 | ||
141 | static char chosen_elevator[16]; | |
142 | ||
143 | static void elevator_setup_default(void) | |
144 | { | |
2824bc93 TH |
145 | struct elevator_type *e; |
146 | ||
1da177e4 | 147 | /* |
131dda7f | 148 | * If default has not been set, use the compiled-in selection. |
1da177e4 | 149 | */ |
131dda7f ND |
150 | if (!chosen_elevator[0]) |
151 | strcpy(chosen_elevator, CONFIG_DEFAULT_IOSCHED); | |
152 | ||
153 | /* | |
154 | * If the given scheduler is not available, fall back to no-op. | |
155 | */ | |
cff3ba22 ZA |
156 | if ((e = elevator_find(chosen_elevator))) |
157 | elevator_put(e); | |
158 | else | |
131dda7f | 159 | strcpy(chosen_elevator, "noop"); |
1da177e4 LT |
160 | } |
161 | ||
162 | static int __init elevator_setup(char *str) | |
163 | { | |
164 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); | |
165 | return 0; | |
166 | } | |
167 | ||
168 | __setup("elevator=", elevator_setup); | |
169 | ||
170 | int elevator_init(request_queue_t *q, char *name) | |
171 | { | |
172 | struct elevator_type *e = NULL; | |
173 | struct elevator_queue *eq; | |
174 | int ret = 0; | |
175 | ||
cb98fc8b TH |
176 | INIT_LIST_HEAD(&q->queue_head); |
177 | q->last_merge = NULL; | |
178 | q->end_sector = 0; | |
179 | q->boundary_rq = NULL; | |
cb98fc8b | 180 | |
1da177e4 LT |
181 | elevator_setup_default(); |
182 | ||
183 | if (!name) | |
184 | name = chosen_elevator; | |
185 | ||
186 | e = elevator_get(name); | |
187 | if (!e) | |
188 | return -EINVAL; | |
189 | ||
190 | eq = kmalloc(sizeof(struct elevator_queue), GFP_KERNEL); | |
191 | if (!eq) { | |
b8ea2cb5 | 192 | elevator_put(e); |
1da177e4 LT |
193 | return -ENOMEM; |
194 | } | |
195 | ||
196 | ret = elevator_attach(q, e, eq); | |
197 | if (ret) { | |
198 | kfree(eq); | |
b8ea2cb5 | 199 | elevator_put(e); |
1da177e4 LT |
200 | } |
201 | ||
202 | return ret; | |
203 | } | |
204 | ||
205 | void elevator_exit(elevator_t *e) | |
206 | { | |
207 | if (e->ops->elevator_exit_fn) | |
208 | e->ops->elevator_exit_fn(e); | |
209 | ||
210 | elevator_put(e->elevator_type); | |
211 | e->elevator_type = NULL; | |
212 | kfree(e); | |
213 | } | |
214 | ||
8922e16c TH |
215 | /* |
216 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
217 | * entry. If sort != 0, rq is sort-inserted; otherwise, rq will be | |
218 | * appended to the dispatch queue. To be used by specific elevators. | |
219 | */ | |
1b47f531 | 220 | void elv_dispatch_sort(request_queue_t *q, struct request *rq) |
8922e16c TH |
221 | { |
222 | sector_t boundary; | |
8922e16c TH |
223 | struct list_head *entry; |
224 | ||
06b86245 TH |
225 | if (q->last_merge == rq) |
226 | q->last_merge = NULL; | |
15853af9 | 227 | q->nr_sorted--; |
06b86245 | 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); | |
15853af9 | 286 | q->nr_sorted--; |
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 | ||
797e7dbb | 307 | __elv_add_request(q, rq, ELEVATOR_INSERT_REQUEUE, 0); |
1da177e4 LT |
308 | } |
309 | ||
15853af9 TH |
310 | static void elv_drain_elevator(request_queue_t *q) |
311 | { | |
312 | static int printed; | |
313 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
314 | ; | |
315 | if (q->nr_sorted == 0) | |
316 | return; | |
317 | if (printed++ < 10) { | |
318 | printk(KERN_ERR "%s: forced dispatching is broken " | |
319 | "(nr_sorted=%u), please report this\n", | |
320 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
321 | } | |
322 | } | |
323 | ||
1da177e4 LT |
324 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, |
325 | int plug) | |
326 | { | |
797e7dbb TH |
327 | struct list_head *pos; |
328 | unsigned ordseq; | |
329 | ||
330 | if (q->ordcolor) | |
331 | rq->flags |= REQ_ORDERED_COLOR; | |
332 | ||
8922e16c | 333 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
797e7dbb TH |
334 | /* |
335 | * toggle ordered color | |
336 | */ | |
337 | q->ordcolor ^= 1; | |
338 | ||
8922e16c TH |
339 | /* |
340 | * barriers implicitly indicate back insertion | |
341 | */ | |
342 | if (where == ELEVATOR_INSERT_SORT) | |
343 | where = ELEVATOR_INSERT_BACK; | |
344 | ||
345 | /* | |
1b47f531 | 346 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
347 | */ |
348 | if (blk_fs_request(rq)) { | |
1b47f531 | 349 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
350 | q->boundary_rq = rq; |
351 | } | |
cb98fc8b TH |
352 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
353 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
354 | |
355 | if (plug) | |
356 | blk_plug_device(q); | |
357 | ||
358 | rq->q = q; | |
359 | ||
8922e16c TH |
360 | switch (where) { |
361 | case ELEVATOR_INSERT_FRONT: | |
362 | rq->flags |= REQ_SOFTBARRIER; | |
363 | ||
364 | list_add(&rq->queuelist, &q->queue_head); | |
365 | break; | |
366 | ||
367 | case ELEVATOR_INSERT_BACK: | |
368 | rq->flags |= REQ_SOFTBARRIER; | |
15853af9 | 369 | elv_drain_elevator(q); |
8922e16c TH |
370 | list_add_tail(&rq->queuelist, &q->queue_head); |
371 | /* | |
372 | * We kick the queue here for the following reasons. | |
373 | * - The elevator might have returned NULL previously | |
374 | * to delay requests and returned them now. As the | |
375 | * queue wasn't empty before this request, ll_rw_blk | |
376 | * won't run the queue on return, resulting in hang. | |
377 | * - Usually, back inserted requests won't be merged | |
378 | * with anything. There's no point in delaying queue | |
379 | * processing. | |
380 | */ | |
381 | blk_remove_plug(q); | |
382 | q->request_fn(q); | |
383 | break; | |
384 | ||
385 | case ELEVATOR_INSERT_SORT: | |
386 | BUG_ON(!blk_fs_request(rq)); | |
387 | rq->flags |= REQ_SORTED; | |
15853af9 | 388 | q->nr_sorted++; |
06b86245 TH |
389 | if (q->last_merge == NULL && rq_mergeable(rq)) |
390 | q->last_merge = rq; | |
ca23509f TH |
391 | /* |
392 | * Some ioscheds (cfq) run q->request_fn directly, so | |
393 | * rq cannot be accessed after calling | |
394 | * elevator_add_req_fn. | |
395 | */ | |
396 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
397 | break; |
398 | ||
797e7dbb TH |
399 | case ELEVATOR_INSERT_REQUEUE: |
400 | /* | |
401 | * If ordered flush isn't in progress, we do front | |
402 | * insertion; otherwise, requests should be requeued | |
403 | * in ordseq order. | |
404 | */ | |
405 | rq->flags |= REQ_SOFTBARRIER; | |
406 | ||
407 | if (q->ordseq == 0) { | |
408 | list_add(&rq->queuelist, &q->queue_head); | |
409 | break; | |
410 | } | |
411 | ||
412 | ordseq = blk_ordered_req_seq(rq); | |
413 | ||
414 | list_for_each(pos, &q->queue_head) { | |
415 | struct request *pos_rq = list_entry_rq(pos); | |
416 | if (ordseq <= blk_ordered_req_seq(pos_rq)) | |
417 | break; | |
418 | } | |
419 | ||
420 | list_add_tail(&rq->queuelist, pos); | |
421 | break; | |
422 | ||
8922e16c TH |
423 | default: |
424 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
425 | __FUNCTION__, where); | |
426 | BUG(); | |
427 | } | |
428 | ||
429 | if (blk_queue_plugged(q)) { | |
430 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
431 | - q->in_flight; | |
432 | ||
433 | if (nrq >= q->unplug_thresh) | |
434 | __generic_unplug_device(q); | |
435 | } | |
1da177e4 LT |
436 | } |
437 | ||
438 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
439 | int plug) | |
440 | { | |
441 | unsigned long flags; | |
442 | ||
443 | spin_lock_irqsave(q->queue_lock, flags); | |
444 | __elv_add_request(q, rq, where, plug); | |
445 | spin_unlock_irqrestore(q->queue_lock, flags); | |
446 | } | |
447 | ||
448 | static inline struct request *__elv_next_request(request_queue_t *q) | |
449 | { | |
8922e16c TH |
450 | struct request *rq; |
451 | ||
797e7dbb TH |
452 | while (1) { |
453 | while (!list_empty(&q->queue_head)) { | |
454 | rq = list_entry_rq(q->queue_head.next); | |
455 | if (blk_do_ordered(q, &rq)) | |
456 | return rq; | |
457 | } | |
1da177e4 | 458 | |
797e7dbb TH |
459 | if (!q->elevator->ops->elevator_dispatch_fn(q, 0)) |
460 | return NULL; | |
1da177e4 | 461 | } |
1da177e4 LT |
462 | } |
463 | ||
464 | struct request *elv_next_request(request_queue_t *q) | |
465 | { | |
466 | struct request *rq; | |
467 | int ret; | |
468 | ||
469 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
470 | if (!(rq->flags & REQ_STARTED)) { |
471 | elevator_t *e = q->elevator; | |
472 | ||
473 | /* | |
474 | * This is the first time the device driver | |
475 | * sees this request (possibly after | |
476 | * requeueing). Notify IO scheduler. | |
477 | */ | |
478 | if (blk_sorted_rq(rq) && | |
479 | e->ops->elevator_activate_req_fn) | |
480 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 481 | |
8922e16c TH |
482 | /* |
483 | * just mark as started even if we don't start | |
484 | * it, a request that has been delayed should | |
485 | * not be passed by new incoming requests | |
486 | */ | |
487 | rq->flags |= REQ_STARTED; | |
488 | } | |
1da177e4 | 489 | |
8922e16c | 490 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 491 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
492 | q->boundary_rq = NULL; |
493 | } | |
1da177e4 LT |
494 | |
495 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
496 | break; | |
497 | ||
498 | ret = q->prep_rq_fn(q, rq); | |
499 | if (ret == BLKPREP_OK) { | |
500 | break; | |
501 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
502 | /* |
503 | * the request may have been (partially) prepped. | |
504 | * we need to keep this request in the front to | |
8922e16c TH |
505 | * avoid resource deadlock. REQ_STARTED will |
506 | * prevent other fs requests from passing this one. | |
2e759cd4 | 507 | */ |
1da177e4 LT |
508 | rq = NULL; |
509 | break; | |
510 | } else if (ret == BLKPREP_KILL) { | |
511 | int nr_bytes = rq->hard_nr_sectors << 9; | |
512 | ||
513 | if (!nr_bytes) | |
514 | nr_bytes = rq->data_len; | |
515 | ||
516 | blkdev_dequeue_request(rq); | |
517 | rq->flags |= REQ_QUIET; | |
518 | end_that_request_chunk(rq, 0, nr_bytes); | |
8ffdc655 | 519 | end_that_request_last(rq, 0); |
1da177e4 LT |
520 | } else { |
521 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
522 | ret); | |
523 | break; | |
524 | } | |
525 | } | |
526 | ||
527 | return rq; | |
528 | } | |
529 | ||
8922e16c | 530 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 531 | { |
8922e16c TH |
532 | BUG_ON(list_empty(&rq->queuelist)); |
533 | ||
534 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
535 | |
536 | /* | |
537 | * the time frame between a request being removed from the lists | |
538 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 539 | * the driver side. |
1da177e4 LT |
540 | */ |
541 | if (blk_account_rq(rq)) | |
542 | q->in_flight++; | |
1da177e4 LT |
543 | } |
544 | ||
545 | int elv_queue_empty(request_queue_t *q) | |
546 | { | |
547 | elevator_t *e = q->elevator; | |
548 | ||
8922e16c TH |
549 | if (!list_empty(&q->queue_head)) |
550 | return 0; | |
551 | ||
1da177e4 LT |
552 | if (e->ops->elevator_queue_empty_fn) |
553 | return e->ops->elevator_queue_empty_fn(q); | |
554 | ||
8922e16c | 555 | return 1; |
1da177e4 LT |
556 | } |
557 | ||
558 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
559 | { | |
1da177e4 LT |
560 | elevator_t *e = q->elevator; |
561 | ||
562 | if (e->ops->elevator_latter_req_fn) | |
563 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
564 | return NULL; |
565 | } | |
566 | ||
567 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
568 | { | |
1da177e4 LT |
569 | elevator_t *e = q->elevator; |
570 | ||
571 | if (e->ops->elevator_former_req_fn) | |
572 | return e->ops->elevator_former_req_fn(q, rq); | |
1da177e4 LT |
573 | return NULL; |
574 | } | |
575 | ||
22e2c507 | 576 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 577 | gfp_t gfp_mask) |
1da177e4 LT |
578 | { |
579 | elevator_t *e = q->elevator; | |
580 | ||
581 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 582 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
583 | |
584 | rq->elevator_private = NULL; | |
585 | return 0; | |
586 | } | |
587 | ||
588 | void elv_put_request(request_queue_t *q, struct request *rq) | |
589 | { | |
590 | elevator_t *e = q->elevator; | |
591 | ||
592 | if (e->ops->elevator_put_req_fn) | |
593 | e->ops->elevator_put_req_fn(q, rq); | |
594 | } | |
595 | ||
22e2c507 | 596 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
597 | { |
598 | elevator_t *e = q->elevator; | |
599 | ||
600 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 601 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
602 | |
603 | return ELV_MQUEUE_MAY; | |
604 | } | |
605 | ||
606 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
607 | { | |
608 | elevator_t *e = q->elevator; | |
609 | ||
610 | /* | |
611 | * request is released from the driver, io must be done | |
612 | */ | |
8922e16c | 613 | if (blk_account_rq(rq)) { |
797e7dbb TH |
614 | struct request *first_rq = list_entry_rq(q->queue_head.next); |
615 | ||
1da177e4 | 616 | q->in_flight--; |
797e7dbb TH |
617 | |
618 | /* | |
619 | * Check if the queue is waiting for fs requests to be | |
620 | * drained for flush sequence. | |
621 | */ | |
622 | if (q->ordseq && q->in_flight == 0 && | |
623 | blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN && | |
624 | blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) { | |
625 | blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0); | |
626 | q->request_fn(q); | |
627 | } | |
628 | ||
8922e16c TH |
629 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
630 | e->ops->elevator_completed_req_fn(q, rq); | |
631 | } | |
1da177e4 LT |
632 | } |
633 | ||
634 | int elv_register_queue(struct request_queue *q) | |
635 | { | |
636 | elevator_t *e = q->elevator; | |
637 | ||
638 | e->kobj.parent = kobject_get(&q->kobj); | |
639 | if (!e->kobj.parent) | |
640 | return -EBUSY; | |
641 | ||
642 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
643 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
644 | ||
645 | return kobject_register(&e->kobj); | |
646 | } | |
647 | ||
648 | void elv_unregister_queue(struct request_queue *q) | |
649 | { | |
650 | if (q) { | |
651 | elevator_t *e = q->elevator; | |
652 | kobject_unregister(&e->kobj); | |
653 | kobject_put(&q->kobj); | |
654 | } | |
655 | } | |
656 | ||
657 | int elv_register(struct elevator_type *e) | |
658 | { | |
2824bc93 | 659 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
660 | if (elevator_find(e->elevator_name)) |
661 | BUG(); | |
1da177e4 LT |
662 | list_add_tail(&e->list, &elv_list); |
663 | spin_unlock_irq(&elv_list_lock); | |
664 | ||
665 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
666 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
667 | printk(" (default)"); | |
668 | printk("\n"); | |
669 | return 0; | |
670 | } | |
671 | EXPORT_SYMBOL_GPL(elv_register); | |
672 | ||
673 | void elv_unregister(struct elevator_type *e) | |
674 | { | |
83521d3e CH |
675 | struct task_struct *g, *p; |
676 | ||
677 | /* | |
678 | * Iterate every thread in the process to remove the io contexts. | |
679 | */ | |
680 | read_lock(&tasklist_lock); | |
681 | do_each_thread(g, p) { | |
682 | struct io_context *ioc = p->io_context; | |
683 | if (ioc && ioc->cic) { | |
684 | ioc->cic->exit(ioc->cic); | |
685 | ioc->cic->dtor(ioc->cic); | |
686 | ioc->cic = NULL; | |
687 | } | |
688 | if (ioc && ioc->aic) { | |
689 | ioc->aic->exit(ioc->aic); | |
690 | ioc->aic->dtor(ioc->aic); | |
691 | ioc->aic = NULL; | |
692 | } | |
693 | } while_each_thread(g, p); | |
694 | read_unlock(&tasklist_lock); | |
695 | ||
1da177e4 LT |
696 | spin_lock_irq(&elv_list_lock); |
697 | list_del_init(&e->list); | |
698 | spin_unlock_irq(&elv_list_lock); | |
699 | } | |
700 | EXPORT_SYMBOL_GPL(elv_unregister); | |
701 | ||
702 | /* | |
703 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
704 | * we don't free the old io scheduler, before we have allocated what we | |
705 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 706 | * one, if the new one fails init for some reason. |
1da177e4 LT |
707 | */ |
708 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
709 | { | |
cb98fc8b | 710 | elevator_t *old_elevator, *e; |
1da177e4 | 711 | |
cb98fc8b TH |
712 | /* |
713 | * Allocate new elevator | |
714 | */ | |
715 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
716 | if (!e) |
717 | goto error; | |
718 | ||
719 | /* | |
cb98fc8b | 720 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 721 | */ |
cb98fc8b TH |
722 | spin_lock_irq(q->queue_lock); |
723 | ||
64521d1a | 724 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 725 | |
15853af9 | 726 | elv_drain_elevator(q); |
cb98fc8b TH |
727 | |
728 | while (q->rq.elvpriv) { | |
407df2aa TH |
729 | blk_remove_plug(q); |
730 | q->request_fn(q); | |
cb98fc8b | 731 | spin_unlock_irq(q->queue_lock); |
64521d1a | 732 | msleep(10); |
cb98fc8b | 733 | spin_lock_irq(q->queue_lock); |
15853af9 | 734 | elv_drain_elevator(q); |
cb98fc8b TH |
735 | } |
736 | ||
737 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
738 | |
739 | /* | |
740 | * unregister old elevator data | |
741 | */ | |
742 | elv_unregister_queue(q); | |
743 | old_elevator = q->elevator; | |
744 | ||
1da177e4 LT |
745 | /* |
746 | * attach and start new elevator | |
747 | */ | |
748 | if (elevator_attach(q, new_e, e)) | |
749 | goto fail; | |
750 | ||
751 | if (elv_register_queue(q)) | |
752 | goto fail_register; | |
753 | ||
754 | /* | |
cb98fc8b | 755 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
756 | */ |
757 | elevator_exit(old_elevator); | |
64521d1a | 758 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
759 | return; |
760 | ||
761 | fail_register: | |
762 | /* | |
763 | * switch failed, exit the new io scheduler and reattach the old | |
764 | * one again (along with re-adding the sysfs dir) | |
765 | */ | |
766 | elevator_exit(e); | |
cb98fc8b | 767 | e = NULL; |
1da177e4 LT |
768 | fail: |
769 | q->elevator = old_elevator; | |
770 | elv_register_queue(q); | |
64521d1a | 771 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 772 | kfree(e); |
1da177e4 | 773 | error: |
1da177e4 LT |
774 | elevator_put(new_e); |
775 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
776 | } | |
777 | ||
778 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
779 | { | |
780 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 781 | size_t len; |
1da177e4 LT |
782 | struct elevator_type *e; |
783 | ||
be561235 TH |
784 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
785 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
786 | len = strlen(elevator_name); | |
1da177e4 | 787 | |
be561235 TH |
788 | if (len && elevator_name[len - 1] == '\n') |
789 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
790 | |
791 | e = elevator_get(elevator_name); | |
792 | if (!e) { | |
793 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
794 | return -EINVAL; | |
795 | } | |
796 | ||
2ca7d93b ND |
797 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
798 | elevator_put(e); | |
1da177e4 | 799 | return count; |
2ca7d93b | 800 | } |
1da177e4 LT |
801 | |
802 | elevator_switch(q, e); | |
803 | return count; | |
804 | } | |
805 | ||
806 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
807 | { | |
808 | elevator_t *e = q->elevator; | |
809 | struct elevator_type *elv = e->elevator_type; | |
810 | struct list_head *entry; | |
811 | int len = 0; | |
812 | ||
813 | spin_lock_irq(q->queue_lock); | |
814 | list_for_each(entry, &elv_list) { | |
815 | struct elevator_type *__e; | |
816 | ||
817 | __e = list_entry(entry, struct elevator_type, list); | |
818 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
819 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
820 | else | |
821 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
822 | } | |
823 | spin_unlock_irq(q->queue_lock); | |
824 | ||
825 | len += sprintf(len+name, "\n"); | |
826 | return len; | |
827 | } | |
828 | ||
1b47f531 | 829 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
830 | EXPORT_SYMBOL(elv_add_request); |
831 | EXPORT_SYMBOL(__elv_add_request); | |
832 | EXPORT_SYMBOL(elv_requeue_request); | |
833 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 834 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
835 | EXPORT_SYMBOL(elv_queue_empty); |
836 | EXPORT_SYMBOL(elv_completed_request); | |
837 | EXPORT_SYMBOL(elevator_exit); | |
838 | EXPORT_SYMBOL(elevator_init); |