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 | ||
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 | ||
15853af9 TH |
318 | static void elv_drain_elevator(request_queue_t *q) |
319 | { | |
320 | static int printed; | |
321 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
322 | ; | |
323 | if (q->nr_sorted == 0) | |
324 | return; | |
325 | if (printed++ < 10) { | |
326 | printk(KERN_ERR "%s: forced dispatching is broken " | |
327 | "(nr_sorted=%u), please report this\n", | |
328 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
329 | } | |
330 | } | |
331 | ||
1da177e4 LT |
332 | void __elv_add_request(request_queue_t *q, struct request *rq, int where, |
333 | int plug) | |
334 | { | |
8922e16c TH |
335 | if (rq->flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
336 | /* | |
337 | * barriers implicitly indicate back insertion | |
338 | */ | |
339 | if (where == ELEVATOR_INSERT_SORT) | |
340 | where = ELEVATOR_INSERT_BACK; | |
341 | ||
342 | /* | |
1b47f531 | 343 | * this request is scheduling boundary, update end_sector |
8922e16c TH |
344 | */ |
345 | if (blk_fs_request(rq)) { | |
1b47f531 | 346 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
347 | q->boundary_rq = rq; |
348 | } | |
cb98fc8b TH |
349 | } else if (!(rq->flags & REQ_ELVPRIV) && where == ELEVATOR_INSERT_SORT) |
350 | where = ELEVATOR_INSERT_BACK; | |
1da177e4 LT |
351 | |
352 | if (plug) | |
353 | blk_plug_device(q); | |
354 | ||
355 | rq->q = q; | |
356 | ||
8922e16c TH |
357 | switch (where) { |
358 | case ELEVATOR_INSERT_FRONT: | |
359 | rq->flags |= REQ_SOFTBARRIER; | |
360 | ||
361 | list_add(&rq->queuelist, &q->queue_head); | |
362 | break; | |
363 | ||
364 | case ELEVATOR_INSERT_BACK: | |
365 | rq->flags |= REQ_SOFTBARRIER; | |
15853af9 | 366 | elv_drain_elevator(q); |
8922e16c TH |
367 | list_add_tail(&rq->queuelist, &q->queue_head); |
368 | /* | |
369 | * We kick the queue here for the following reasons. | |
370 | * - The elevator might have returned NULL previously | |
371 | * to delay requests and returned them now. As the | |
372 | * queue wasn't empty before this request, ll_rw_blk | |
373 | * won't run the queue on return, resulting in hang. | |
374 | * - Usually, back inserted requests won't be merged | |
375 | * with anything. There's no point in delaying queue | |
376 | * processing. | |
377 | */ | |
378 | blk_remove_plug(q); | |
379 | q->request_fn(q); | |
380 | break; | |
381 | ||
382 | case ELEVATOR_INSERT_SORT: | |
383 | BUG_ON(!blk_fs_request(rq)); | |
384 | rq->flags |= REQ_SORTED; | |
15853af9 | 385 | q->nr_sorted++; |
06b86245 TH |
386 | if (q->last_merge == NULL && rq_mergeable(rq)) |
387 | q->last_merge = rq; | |
ca23509f TH |
388 | /* |
389 | * Some ioscheds (cfq) run q->request_fn directly, so | |
390 | * rq cannot be accessed after calling | |
391 | * elevator_add_req_fn. | |
392 | */ | |
393 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
394 | break; |
395 | ||
396 | default: | |
397 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
398 | __FUNCTION__, where); | |
399 | BUG(); | |
400 | } | |
401 | ||
402 | if (blk_queue_plugged(q)) { | |
403 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] | |
404 | - q->in_flight; | |
405 | ||
406 | if (nrq >= q->unplug_thresh) | |
407 | __generic_unplug_device(q); | |
408 | } | |
1da177e4 LT |
409 | } |
410 | ||
411 | void elv_add_request(request_queue_t *q, struct request *rq, int where, | |
412 | int plug) | |
413 | { | |
414 | unsigned long flags; | |
415 | ||
416 | spin_lock_irqsave(q->queue_lock, flags); | |
417 | __elv_add_request(q, rq, where, plug); | |
418 | spin_unlock_irqrestore(q->queue_lock, flags); | |
419 | } | |
420 | ||
421 | static inline struct request *__elv_next_request(request_queue_t *q) | |
422 | { | |
8922e16c TH |
423 | struct request *rq; |
424 | ||
425 | if (unlikely(list_empty(&q->queue_head) && | |
426 | !q->elevator->ops->elevator_dispatch_fn(q, 0))) | |
427 | return NULL; | |
428 | ||
429 | rq = list_entry_rq(q->queue_head.next); | |
1da177e4 LT |
430 | |
431 | /* | |
432 | * if this is a barrier write and the device has to issue a | |
433 | * flush sequence to support it, check how far we are | |
434 | */ | |
8922e16c | 435 | if (blk_fs_request(rq) && blk_barrier_rq(rq)) { |
1da177e4 LT |
436 | BUG_ON(q->ordered == QUEUE_ORDERED_NONE); |
437 | ||
438 | if (q->ordered == QUEUE_ORDERED_FLUSH && | |
439 | !blk_barrier_preflush(rq)) | |
440 | rq = blk_start_pre_flush(q, rq); | |
441 | } | |
442 | ||
443 | return rq; | |
444 | } | |
445 | ||
446 | struct request *elv_next_request(request_queue_t *q) | |
447 | { | |
448 | struct request *rq; | |
449 | int ret; | |
450 | ||
451 | while ((rq = __elv_next_request(q)) != NULL) { | |
8922e16c TH |
452 | if (!(rq->flags & REQ_STARTED)) { |
453 | elevator_t *e = q->elevator; | |
454 | ||
455 | /* | |
456 | * This is the first time the device driver | |
457 | * sees this request (possibly after | |
458 | * requeueing). Notify IO scheduler. | |
459 | */ | |
460 | if (blk_sorted_rq(rq) && | |
461 | e->ops->elevator_activate_req_fn) | |
462 | e->ops->elevator_activate_req_fn(q, rq); | |
1da177e4 | 463 | |
8922e16c TH |
464 | /* |
465 | * just mark as started even if we don't start | |
466 | * it, a request that has been delayed should | |
467 | * not be passed by new incoming requests | |
468 | */ | |
469 | rq->flags |= REQ_STARTED; | |
470 | } | |
1da177e4 | 471 | |
8922e16c | 472 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 473 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
474 | q->boundary_rq = NULL; |
475 | } | |
1da177e4 LT |
476 | |
477 | if ((rq->flags & REQ_DONTPREP) || !q->prep_rq_fn) | |
478 | break; | |
479 | ||
480 | ret = q->prep_rq_fn(q, rq); | |
481 | if (ret == BLKPREP_OK) { | |
482 | break; | |
483 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
484 | /* |
485 | * the request may have been (partially) prepped. | |
486 | * we need to keep this request in the front to | |
8922e16c TH |
487 | * avoid resource deadlock. REQ_STARTED will |
488 | * prevent other fs requests from passing this one. | |
2e759cd4 | 489 | */ |
1da177e4 LT |
490 | rq = NULL; |
491 | break; | |
492 | } else if (ret == BLKPREP_KILL) { | |
493 | int nr_bytes = rq->hard_nr_sectors << 9; | |
494 | ||
495 | if (!nr_bytes) | |
496 | nr_bytes = rq->data_len; | |
497 | ||
498 | blkdev_dequeue_request(rq); | |
499 | rq->flags |= REQ_QUIET; | |
500 | end_that_request_chunk(rq, 0, nr_bytes); | |
501 | end_that_request_last(rq); | |
502 | } else { | |
503 | printk(KERN_ERR "%s: bad return=%d\n", __FUNCTION__, | |
504 | ret); | |
505 | break; | |
506 | } | |
507 | } | |
508 | ||
509 | return rq; | |
510 | } | |
511 | ||
8922e16c | 512 | void elv_dequeue_request(request_queue_t *q, struct request *rq) |
1da177e4 | 513 | { |
8922e16c TH |
514 | BUG_ON(list_empty(&rq->queuelist)); |
515 | ||
516 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
517 | |
518 | /* | |
519 | * the time frame between a request being removed from the lists | |
520 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 521 | * the driver side. |
1da177e4 LT |
522 | */ |
523 | if (blk_account_rq(rq)) | |
524 | q->in_flight++; | |
1da177e4 LT |
525 | } |
526 | ||
527 | int elv_queue_empty(request_queue_t *q) | |
528 | { | |
529 | elevator_t *e = q->elevator; | |
530 | ||
8922e16c TH |
531 | if (!list_empty(&q->queue_head)) |
532 | return 0; | |
533 | ||
1da177e4 LT |
534 | if (e->ops->elevator_queue_empty_fn) |
535 | return e->ops->elevator_queue_empty_fn(q); | |
536 | ||
8922e16c | 537 | return 1; |
1da177e4 LT |
538 | } |
539 | ||
540 | struct request *elv_latter_request(request_queue_t *q, struct request *rq) | |
541 | { | |
1da177e4 LT |
542 | elevator_t *e = q->elevator; |
543 | ||
544 | if (e->ops->elevator_latter_req_fn) | |
545 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
546 | return NULL; |
547 | } | |
548 | ||
549 | struct request *elv_former_request(request_queue_t *q, struct request *rq) | |
550 | { | |
1da177e4 LT |
551 | elevator_t *e = q->elevator; |
552 | ||
553 | if (e->ops->elevator_former_req_fn) | |
554 | return e->ops->elevator_former_req_fn(q, rq); | |
1da177e4 LT |
555 | return NULL; |
556 | } | |
557 | ||
22e2c507 | 558 | int elv_set_request(request_queue_t *q, struct request *rq, struct bio *bio, |
8267e268 | 559 | gfp_t gfp_mask) |
1da177e4 LT |
560 | { |
561 | elevator_t *e = q->elevator; | |
562 | ||
563 | if (e->ops->elevator_set_req_fn) | |
22e2c507 | 564 | return e->ops->elevator_set_req_fn(q, rq, bio, gfp_mask); |
1da177e4 LT |
565 | |
566 | rq->elevator_private = NULL; | |
567 | return 0; | |
568 | } | |
569 | ||
570 | void elv_put_request(request_queue_t *q, struct request *rq) | |
571 | { | |
572 | elevator_t *e = q->elevator; | |
573 | ||
574 | if (e->ops->elevator_put_req_fn) | |
575 | e->ops->elevator_put_req_fn(q, rq); | |
576 | } | |
577 | ||
22e2c507 | 578 | int elv_may_queue(request_queue_t *q, int rw, struct bio *bio) |
1da177e4 LT |
579 | { |
580 | elevator_t *e = q->elevator; | |
581 | ||
582 | if (e->ops->elevator_may_queue_fn) | |
22e2c507 | 583 | return e->ops->elevator_may_queue_fn(q, rw, bio); |
1da177e4 LT |
584 | |
585 | return ELV_MQUEUE_MAY; | |
586 | } | |
587 | ||
588 | void elv_completed_request(request_queue_t *q, struct request *rq) | |
589 | { | |
590 | elevator_t *e = q->elevator; | |
591 | ||
592 | /* | |
593 | * request is released from the driver, io must be done | |
594 | */ | |
8922e16c | 595 | if (blk_account_rq(rq)) { |
1da177e4 | 596 | q->in_flight--; |
8922e16c TH |
597 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
598 | e->ops->elevator_completed_req_fn(q, rq); | |
599 | } | |
1da177e4 LT |
600 | } |
601 | ||
602 | int elv_register_queue(struct request_queue *q) | |
603 | { | |
604 | elevator_t *e = q->elevator; | |
605 | ||
606 | e->kobj.parent = kobject_get(&q->kobj); | |
607 | if (!e->kobj.parent) | |
608 | return -EBUSY; | |
609 | ||
610 | snprintf(e->kobj.name, KOBJ_NAME_LEN, "%s", "iosched"); | |
611 | e->kobj.ktype = e->elevator_type->elevator_ktype; | |
612 | ||
613 | return kobject_register(&e->kobj); | |
614 | } | |
615 | ||
616 | void elv_unregister_queue(struct request_queue *q) | |
617 | { | |
618 | if (q) { | |
619 | elevator_t *e = q->elevator; | |
620 | kobject_unregister(&e->kobj); | |
621 | kobject_put(&q->kobj); | |
622 | } | |
623 | } | |
624 | ||
625 | int elv_register(struct elevator_type *e) | |
626 | { | |
2824bc93 | 627 | spin_lock_irq(&elv_list_lock); |
1da177e4 LT |
628 | if (elevator_find(e->elevator_name)) |
629 | BUG(); | |
1da177e4 LT |
630 | list_add_tail(&e->list, &elv_list); |
631 | spin_unlock_irq(&elv_list_lock); | |
632 | ||
633 | printk(KERN_INFO "io scheduler %s registered", e->elevator_name); | |
634 | if (!strcmp(e->elevator_name, chosen_elevator)) | |
635 | printk(" (default)"); | |
636 | printk("\n"); | |
637 | return 0; | |
638 | } | |
639 | EXPORT_SYMBOL_GPL(elv_register); | |
640 | ||
641 | void elv_unregister(struct elevator_type *e) | |
642 | { | |
83521d3e CH |
643 | struct task_struct *g, *p; |
644 | ||
645 | /* | |
646 | * Iterate every thread in the process to remove the io contexts. | |
647 | */ | |
648 | read_lock(&tasklist_lock); | |
649 | do_each_thread(g, p) { | |
650 | struct io_context *ioc = p->io_context; | |
651 | if (ioc && ioc->cic) { | |
652 | ioc->cic->exit(ioc->cic); | |
653 | ioc->cic->dtor(ioc->cic); | |
654 | ioc->cic = NULL; | |
655 | } | |
656 | if (ioc && ioc->aic) { | |
657 | ioc->aic->exit(ioc->aic); | |
658 | ioc->aic->dtor(ioc->aic); | |
659 | ioc->aic = NULL; | |
660 | } | |
661 | } while_each_thread(g, p); | |
662 | read_unlock(&tasklist_lock); | |
663 | ||
1da177e4 LT |
664 | spin_lock_irq(&elv_list_lock); |
665 | list_del_init(&e->list); | |
666 | spin_unlock_irq(&elv_list_lock); | |
667 | } | |
668 | EXPORT_SYMBOL_GPL(elv_unregister); | |
669 | ||
670 | /* | |
671 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
672 | * we don't free the old io scheduler, before we have allocated what we | |
673 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 674 | * one, if the new one fails init for some reason. |
1da177e4 LT |
675 | */ |
676 | static void elevator_switch(request_queue_t *q, struct elevator_type *new_e) | |
677 | { | |
cb98fc8b | 678 | elevator_t *old_elevator, *e; |
1da177e4 | 679 | |
cb98fc8b TH |
680 | /* |
681 | * Allocate new elevator | |
682 | */ | |
683 | e = kmalloc(sizeof(elevator_t), GFP_KERNEL); | |
1da177e4 LT |
684 | if (!e) |
685 | goto error; | |
686 | ||
687 | /* | |
cb98fc8b | 688 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 689 | */ |
cb98fc8b TH |
690 | spin_lock_irq(q->queue_lock); |
691 | ||
64521d1a | 692 | set_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 693 | |
15853af9 | 694 | elv_drain_elevator(q); |
cb98fc8b TH |
695 | |
696 | while (q->rq.elvpriv) { | |
407df2aa TH |
697 | blk_remove_plug(q); |
698 | q->request_fn(q); | |
cb98fc8b | 699 | spin_unlock_irq(q->queue_lock); |
64521d1a | 700 | msleep(10); |
cb98fc8b | 701 | spin_lock_irq(q->queue_lock); |
15853af9 | 702 | elv_drain_elevator(q); |
cb98fc8b TH |
703 | } |
704 | ||
705 | spin_unlock_irq(q->queue_lock); | |
1da177e4 LT |
706 | |
707 | /* | |
708 | * unregister old elevator data | |
709 | */ | |
710 | elv_unregister_queue(q); | |
711 | old_elevator = q->elevator; | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * attach and start new elevator | |
715 | */ | |
716 | if (elevator_attach(q, new_e, e)) | |
717 | goto fail; | |
718 | ||
719 | if (elv_register_queue(q)) | |
720 | goto fail_register; | |
721 | ||
722 | /* | |
cb98fc8b | 723 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
724 | */ |
725 | elevator_exit(old_elevator); | |
64521d1a | 726 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
1da177e4 LT |
727 | return; |
728 | ||
729 | fail_register: | |
730 | /* | |
731 | * switch failed, exit the new io scheduler and reattach the old | |
732 | * one again (along with re-adding the sysfs dir) | |
733 | */ | |
734 | elevator_exit(e); | |
cb98fc8b | 735 | e = NULL; |
1da177e4 LT |
736 | fail: |
737 | q->elevator = old_elevator; | |
738 | elv_register_queue(q); | |
64521d1a | 739 | clear_bit(QUEUE_FLAG_ELVSWITCH, &q->queue_flags); |
cb98fc8b | 740 | kfree(e); |
1da177e4 | 741 | error: |
1da177e4 LT |
742 | elevator_put(new_e); |
743 | printk(KERN_ERR "elevator: switch to %s failed\n",new_e->elevator_name); | |
744 | } | |
745 | ||
746 | ssize_t elv_iosched_store(request_queue_t *q, const char *name, size_t count) | |
747 | { | |
748 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 749 | size_t len; |
1da177e4 LT |
750 | struct elevator_type *e; |
751 | ||
be561235 TH |
752 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
753 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
754 | len = strlen(elevator_name); | |
1da177e4 | 755 | |
be561235 TH |
756 | if (len && elevator_name[len - 1] == '\n') |
757 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
758 | |
759 | e = elevator_get(elevator_name); | |
760 | if (!e) { | |
761 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
762 | return -EINVAL; | |
763 | } | |
764 | ||
2ca7d93b ND |
765 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
766 | elevator_put(e); | |
1da177e4 | 767 | return count; |
2ca7d93b | 768 | } |
1da177e4 LT |
769 | |
770 | elevator_switch(q, e); | |
771 | return count; | |
772 | } | |
773 | ||
774 | ssize_t elv_iosched_show(request_queue_t *q, char *name) | |
775 | { | |
776 | elevator_t *e = q->elevator; | |
777 | struct elevator_type *elv = e->elevator_type; | |
778 | struct list_head *entry; | |
779 | int len = 0; | |
780 | ||
781 | spin_lock_irq(q->queue_lock); | |
782 | list_for_each(entry, &elv_list) { | |
783 | struct elevator_type *__e; | |
784 | ||
785 | __e = list_entry(entry, struct elevator_type, list); | |
786 | if (!strcmp(elv->elevator_name, __e->elevator_name)) | |
787 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
788 | else | |
789 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
790 | } | |
791 | spin_unlock_irq(q->queue_lock); | |
792 | ||
793 | len += sprintf(len+name, "\n"); | |
794 | return len; | |
795 | } | |
796 | ||
1b47f531 | 797 | EXPORT_SYMBOL(elv_dispatch_sort); |
1da177e4 LT |
798 | EXPORT_SYMBOL(elv_add_request); |
799 | EXPORT_SYMBOL(__elv_add_request); | |
800 | EXPORT_SYMBOL(elv_requeue_request); | |
801 | EXPORT_SYMBOL(elv_next_request); | |
8922e16c | 802 | EXPORT_SYMBOL(elv_dequeue_request); |
1da177e4 LT |
803 | EXPORT_SYMBOL(elv_queue_empty); |
804 | EXPORT_SYMBOL(elv_completed_request); | |
805 | EXPORT_SYMBOL(elevator_exit); | |
806 | EXPORT_SYMBOL(elevator_init); |