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 | * | |
0fe23479 | 6 | * 30042000 Jens Axboe <axboe@kernel.dk> : |
1da177e4 LT |
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> | |
1da177e4 LT |
30 | #include <linux/module.h> |
31 | #include <linux/slab.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/compiler.h> | |
cb98fc8b | 34 | #include <linux/delay.h> |
2056a782 | 35 | #include <linux/blktrace_api.h> |
9817064b | 36 | #include <linux/hash.h> |
0835da67 | 37 | #include <linux/uaccess.h> |
1da177e4 | 38 | |
242f9dcb JA |
39 | #include "blk.h" |
40 | ||
1da177e4 LT |
41 | static DEFINE_SPINLOCK(elv_list_lock); |
42 | static LIST_HEAD(elv_list); | |
43 | ||
9817064b JA |
44 | /* |
45 | * Merge hash stuff. | |
46 | */ | |
47 | static const int elv_hash_shift = 6; | |
48 | #define ELV_HASH_BLOCK(sec) ((sec) >> 3) | |
4eb166d9 JA |
49 | #define ELV_HASH_FN(sec) \ |
50 | (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift)) | |
9817064b JA |
51 | #define ELV_HASH_ENTRIES (1 << elv_hash_shift) |
52 | #define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors) | |
53 | #define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash)) | |
54 | ||
da775265 JA |
55 | /* |
56 | * Query io scheduler to see if the current process issuing bio may be | |
57 | * merged with rq. | |
58 | */ | |
59 | static int elv_iosched_allow_merge(struct request *rq, struct bio *bio) | |
60 | { | |
165125e1 | 61 | struct request_queue *q = rq->q; |
da775265 JA |
62 | elevator_t *e = q->elevator; |
63 | ||
64 | if (e->ops->elevator_allow_merge_fn) | |
65 | return e->ops->elevator_allow_merge_fn(q, rq, bio); | |
66 | ||
67 | return 1; | |
68 | } | |
69 | ||
1da177e4 LT |
70 | /* |
71 | * can we safely merge with this request? | |
72 | */ | |
72ed0bf6 | 73 | int elv_rq_merge_ok(struct request *rq, struct bio *bio) |
1da177e4 LT |
74 | { |
75 | if (!rq_mergeable(rq)) | |
76 | return 0; | |
77 | ||
e17fc0a1 DW |
78 | /* |
79 | * Don't merge file system requests and discard requests | |
80 | */ | |
81 | if (bio_discard(bio) != bio_discard(rq->bio)) | |
82 | return 0; | |
83 | ||
1da177e4 LT |
84 | /* |
85 | * different data direction or already started, don't merge | |
86 | */ | |
87 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
88 | return 0; | |
89 | ||
90 | /* | |
da775265 | 91 | * must be same device and not a special request |
1da177e4 | 92 | */ |
bb4067e3 | 93 | if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special) |
da775265 JA |
94 | return 0; |
95 | ||
7ba1ba12 MP |
96 | /* |
97 | * only merge integrity protected bio into ditto rq | |
98 | */ | |
99 | if (bio_integrity(bio) != blk_integrity_rq(rq)) | |
100 | return 0; | |
101 | ||
da775265 JA |
102 | if (!elv_iosched_allow_merge(rq, bio)) |
103 | return 0; | |
1da177e4 | 104 | |
da775265 | 105 | return 1; |
1da177e4 LT |
106 | } |
107 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
108 | ||
769db45b | 109 | static inline int elv_try_merge(struct request *__rq, struct bio *bio) |
1da177e4 LT |
110 | { |
111 | int ret = ELEVATOR_NO_MERGE; | |
112 | ||
113 | /* | |
114 | * we can merge and sequence is ok, check if it's possible | |
115 | */ | |
116 | if (elv_rq_merge_ok(__rq, bio)) { | |
117 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
118 | ret = ELEVATOR_BACK_MERGE; | |
119 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
120 | ret = ELEVATOR_FRONT_MERGE; | |
121 | } | |
122 | ||
123 | return ret; | |
124 | } | |
1da177e4 | 125 | |
1da177e4 LT |
126 | static struct elevator_type *elevator_find(const char *name) |
127 | { | |
a22b169d | 128 | struct elevator_type *e; |
1da177e4 | 129 | |
70cee26e | 130 | list_for_each_entry(e, &elv_list, list) { |
a22b169d VT |
131 | if (!strcmp(e->elevator_name, name)) |
132 | return e; | |
1da177e4 | 133 | } |
1da177e4 | 134 | |
a22b169d | 135 | return NULL; |
1da177e4 LT |
136 | } |
137 | ||
138 | static void elevator_put(struct elevator_type *e) | |
139 | { | |
140 | module_put(e->elevator_owner); | |
141 | } | |
142 | ||
143 | static struct elevator_type *elevator_get(const char *name) | |
144 | { | |
2824bc93 | 145 | struct elevator_type *e; |
1da177e4 | 146 | |
2a12dcd7 | 147 | spin_lock(&elv_list_lock); |
2824bc93 TH |
148 | |
149 | e = elevator_find(name); | |
e1640949 JA |
150 | if (!e) { |
151 | char elv[ELV_NAME_MAX + strlen("-iosched")]; | |
152 | ||
153 | spin_unlock(&elv_list_lock); | |
154 | ||
155 | if (!strcmp(name, "anticipatory")) | |
156 | sprintf(elv, "as-iosched"); | |
157 | else | |
158 | sprintf(elv, "%s-iosched", name); | |
159 | ||
e180f594 | 160 | request_module("%s", elv); |
e1640949 JA |
161 | spin_lock(&elv_list_lock); |
162 | e = elevator_find(name); | |
163 | } | |
164 | ||
2824bc93 TH |
165 | if (e && !try_module_get(e->elevator_owner)) |
166 | e = NULL; | |
167 | ||
2a12dcd7 | 168 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
169 | |
170 | return e; | |
171 | } | |
172 | ||
165125e1 JA |
173 | static void *elevator_init_queue(struct request_queue *q, |
174 | struct elevator_queue *eq) | |
1da177e4 | 175 | { |
bb37b94c | 176 | return eq->ops->elevator_init_fn(q); |
bc1c1169 | 177 | } |
1da177e4 | 178 | |
165125e1 | 179 | static void elevator_attach(struct request_queue *q, struct elevator_queue *eq, |
bc1c1169 JA |
180 | void *data) |
181 | { | |
1da177e4 | 182 | q->elevator = eq; |
bc1c1169 | 183 | eq->elevator_data = data; |
1da177e4 LT |
184 | } |
185 | ||
186 | static char chosen_elevator[16]; | |
187 | ||
5f003976 | 188 | static int __init elevator_setup(char *str) |
1da177e4 | 189 | { |
752a3b79 CE |
190 | /* |
191 | * Be backwards-compatible with previous kernels, so users | |
192 | * won't get the wrong elevator. | |
193 | */ | |
5f003976 | 194 | if (!strcmp(str, "as")) |
752a3b79 | 195 | strcpy(chosen_elevator, "anticipatory"); |
cff3ba22 | 196 | else |
5f003976 | 197 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); |
9b41046c | 198 | return 1; |
1da177e4 LT |
199 | } |
200 | ||
201 | __setup("elevator=", elevator_setup); | |
202 | ||
3d1ab40f AV |
203 | static struct kobj_type elv_ktype; |
204 | ||
165125e1 JA |
205 | static elevator_t *elevator_alloc(struct request_queue *q, |
206 | struct elevator_type *e) | |
3d1ab40f | 207 | { |
9817064b JA |
208 | elevator_t *eq; |
209 | int i; | |
210 | ||
94f6030c | 211 | eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL | __GFP_ZERO, q->node); |
9817064b JA |
212 | if (unlikely(!eq)) |
213 | goto err; | |
214 | ||
9817064b JA |
215 | eq->ops = &e->ops; |
216 | eq->elevator_type = e; | |
f9cb074b | 217 | kobject_init(&eq->kobj, &elv_ktype); |
9817064b JA |
218 | mutex_init(&eq->sysfs_lock); |
219 | ||
b5deef90 JA |
220 | eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES, |
221 | GFP_KERNEL, q->node); | |
9817064b JA |
222 | if (!eq->hash) |
223 | goto err; | |
224 | ||
225 | for (i = 0; i < ELV_HASH_ENTRIES; i++) | |
226 | INIT_HLIST_HEAD(&eq->hash[i]); | |
227 | ||
3d1ab40f | 228 | return eq; |
9817064b JA |
229 | err: |
230 | kfree(eq); | |
231 | elevator_put(e); | |
232 | return NULL; | |
3d1ab40f AV |
233 | } |
234 | ||
235 | static void elevator_release(struct kobject *kobj) | |
236 | { | |
237 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
9817064b | 238 | |
3d1ab40f | 239 | elevator_put(e->elevator_type); |
9817064b | 240 | kfree(e->hash); |
3d1ab40f AV |
241 | kfree(e); |
242 | } | |
243 | ||
165125e1 | 244 | int elevator_init(struct request_queue *q, char *name) |
1da177e4 LT |
245 | { |
246 | struct elevator_type *e = NULL; | |
247 | struct elevator_queue *eq; | |
248 | int ret = 0; | |
bc1c1169 | 249 | void *data; |
1da177e4 | 250 | |
cb98fc8b TH |
251 | INIT_LIST_HEAD(&q->queue_head); |
252 | q->last_merge = NULL; | |
253 | q->end_sector = 0; | |
254 | q->boundary_rq = NULL; | |
cb98fc8b | 255 | |
4eb166d9 JA |
256 | if (name) { |
257 | e = elevator_get(name); | |
258 | if (!e) | |
259 | return -EINVAL; | |
260 | } | |
1da177e4 | 261 | |
4eb166d9 JA |
262 | if (!e && *chosen_elevator) { |
263 | e = elevator_get(chosen_elevator); | |
264 | if (!e) | |
265 | printk(KERN_ERR "I/O scheduler %s not found\n", | |
266 | chosen_elevator); | |
267 | } | |
248d5ca5 | 268 | |
4eb166d9 JA |
269 | if (!e) { |
270 | e = elevator_get(CONFIG_DEFAULT_IOSCHED); | |
271 | if (!e) { | |
272 | printk(KERN_ERR | |
273 | "Default I/O scheduler not found. " \ | |
274 | "Using noop.\n"); | |
275 | e = elevator_get("noop"); | |
276 | } | |
5f003976 ND |
277 | } |
278 | ||
b5deef90 | 279 | eq = elevator_alloc(q, e); |
3d1ab40f | 280 | if (!eq) |
1da177e4 | 281 | return -ENOMEM; |
1da177e4 | 282 | |
bc1c1169 JA |
283 | data = elevator_init_queue(q, eq); |
284 | if (!data) { | |
3d1ab40f | 285 | kobject_put(&eq->kobj); |
bc1c1169 JA |
286 | return -ENOMEM; |
287 | } | |
1da177e4 | 288 | |
bc1c1169 | 289 | elevator_attach(q, eq, data); |
1da177e4 LT |
290 | return ret; |
291 | } | |
2e662b65 JA |
292 | EXPORT_SYMBOL(elevator_init); |
293 | ||
1da177e4 LT |
294 | void elevator_exit(elevator_t *e) |
295 | { | |
3d1ab40f | 296 | mutex_lock(&e->sysfs_lock); |
1da177e4 LT |
297 | if (e->ops->elevator_exit_fn) |
298 | e->ops->elevator_exit_fn(e); | |
3d1ab40f AV |
299 | e->ops = NULL; |
300 | mutex_unlock(&e->sysfs_lock); | |
1da177e4 | 301 | |
3d1ab40f | 302 | kobject_put(&e->kobj); |
1da177e4 | 303 | } |
2e662b65 JA |
304 | EXPORT_SYMBOL(elevator_exit); |
305 | ||
165125e1 | 306 | static void elv_activate_rq(struct request_queue *q, struct request *rq) |
cad97516 JA |
307 | { |
308 | elevator_t *e = q->elevator; | |
309 | ||
310 | if (e->ops->elevator_activate_req_fn) | |
311 | e->ops->elevator_activate_req_fn(q, rq); | |
312 | } | |
313 | ||
165125e1 | 314 | static void elv_deactivate_rq(struct request_queue *q, struct request *rq) |
cad97516 JA |
315 | { |
316 | elevator_t *e = q->elevator; | |
317 | ||
318 | if (e->ops->elevator_deactivate_req_fn) | |
319 | e->ops->elevator_deactivate_req_fn(q, rq); | |
320 | } | |
321 | ||
9817064b JA |
322 | static inline void __elv_rqhash_del(struct request *rq) |
323 | { | |
324 | hlist_del_init(&rq->hash); | |
325 | } | |
326 | ||
165125e1 | 327 | static void elv_rqhash_del(struct request_queue *q, struct request *rq) |
9817064b JA |
328 | { |
329 | if (ELV_ON_HASH(rq)) | |
330 | __elv_rqhash_del(rq); | |
331 | } | |
332 | ||
165125e1 | 333 | static void elv_rqhash_add(struct request_queue *q, struct request *rq) |
9817064b JA |
334 | { |
335 | elevator_t *e = q->elevator; | |
336 | ||
337 | BUG_ON(ELV_ON_HASH(rq)); | |
338 | hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]); | |
339 | } | |
340 | ||
165125e1 | 341 | static void elv_rqhash_reposition(struct request_queue *q, struct request *rq) |
9817064b JA |
342 | { |
343 | __elv_rqhash_del(rq); | |
344 | elv_rqhash_add(q, rq); | |
345 | } | |
346 | ||
165125e1 | 347 | static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) |
9817064b JA |
348 | { |
349 | elevator_t *e = q->elevator; | |
350 | struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)]; | |
351 | struct hlist_node *entry, *next; | |
352 | struct request *rq; | |
353 | ||
354 | hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) { | |
355 | BUG_ON(!ELV_ON_HASH(rq)); | |
356 | ||
357 | if (unlikely(!rq_mergeable(rq))) { | |
358 | __elv_rqhash_del(rq); | |
359 | continue; | |
360 | } | |
361 | ||
362 | if (rq_hash_key(rq) == offset) | |
363 | return rq; | |
364 | } | |
365 | ||
366 | return NULL; | |
367 | } | |
368 | ||
2e662b65 JA |
369 | /* |
370 | * RB-tree support functions for inserting/lookup/removal of requests | |
371 | * in a sorted RB tree. | |
372 | */ | |
373 | struct request *elv_rb_add(struct rb_root *root, struct request *rq) | |
374 | { | |
375 | struct rb_node **p = &root->rb_node; | |
376 | struct rb_node *parent = NULL; | |
377 | struct request *__rq; | |
378 | ||
379 | while (*p) { | |
380 | parent = *p; | |
381 | __rq = rb_entry(parent, struct request, rb_node); | |
382 | ||
383 | if (rq->sector < __rq->sector) | |
384 | p = &(*p)->rb_left; | |
385 | else if (rq->sector > __rq->sector) | |
386 | p = &(*p)->rb_right; | |
387 | else | |
388 | return __rq; | |
389 | } | |
390 | ||
391 | rb_link_node(&rq->rb_node, parent, p); | |
392 | rb_insert_color(&rq->rb_node, root); | |
393 | return NULL; | |
394 | } | |
2e662b65 JA |
395 | EXPORT_SYMBOL(elv_rb_add); |
396 | ||
397 | void elv_rb_del(struct rb_root *root, struct request *rq) | |
398 | { | |
399 | BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); | |
400 | rb_erase(&rq->rb_node, root); | |
401 | RB_CLEAR_NODE(&rq->rb_node); | |
402 | } | |
2e662b65 JA |
403 | EXPORT_SYMBOL(elv_rb_del); |
404 | ||
405 | struct request *elv_rb_find(struct rb_root *root, sector_t sector) | |
406 | { | |
407 | struct rb_node *n = root->rb_node; | |
408 | struct request *rq; | |
409 | ||
410 | while (n) { | |
411 | rq = rb_entry(n, struct request, rb_node); | |
412 | ||
413 | if (sector < rq->sector) | |
414 | n = n->rb_left; | |
415 | else if (sector > rq->sector) | |
416 | n = n->rb_right; | |
417 | else | |
418 | return rq; | |
419 | } | |
420 | ||
421 | return NULL; | |
422 | } | |
2e662b65 JA |
423 | EXPORT_SYMBOL(elv_rb_find); |
424 | ||
8922e16c TH |
425 | /* |
426 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
dbe7f76d | 427 | * entry. rq is sort instead into the dispatch queue. To be used by |
2e662b65 | 428 | * specific elevators. |
8922e16c | 429 | */ |
165125e1 | 430 | void elv_dispatch_sort(struct request_queue *q, struct request *rq) |
8922e16c TH |
431 | { |
432 | sector_t boundary; | |
8922e16c | 433 | struct list_head *entry; |
4eb166d9 | 434 | int stop_flags; |
8922e16c | 435 | |
06b86245 TH |
436 | if (q->last_merge == rq) |
437 | q->last_merge = NULL; | |
9817064b JA |
438 | |
439 | elv_rqhash_del(q, rq); | |
440 | ||
15853af9 | 441 | q->nr_sorted--; |
06b86245 | 442 | |
1b47f531 | 443 | boundary = q->end_sector; |
4eb166d9 | 444 | stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED; |
8922e16c TH |
445 | list_for_each_prev(entry, &q->queue_head) { |
446 | struct request *pos = list_entry_rq(entry); | |
447 | ||
e17fc0a1 DW |
448 | if (blk_discard_rq(rq) != blk_discard_rq(pos)) |
449 | break; | |
783660b2 JA |
450 | if (rq_data_dir(rq) != rq_data_dir(pos)) |
451 | break; | |
4eb166d9 | 452 | if (pos->cmd_flags & stop_flags) |
8922e16c TH |
453 | break; |
454 | if (rq->sector >= boundary) { | |
455 | if (pos->sector < boundary) | |
456 | continue; | |
457 | } else { | |
458 | if (pos->sector >= boundary) | |
459 | break; | |
460 | } | |
461 | if (rq->sector >= pos->sector) | |
462 | break; | |
463 | } | |
464 | ||
465 | list_add(&rq->queuelist, entry); | |
466 | } | |
2e662b65 JA |
467 | EXPORT_SYMBOL(elv_dispatch_sort); |
468 | ||
9817064b | 469 | /* |
2e662b65 JA |
470 | * Insert rq into dispatch queue of q. Queue lock must be held on |
471 | * entry. rq is added to the back of the dispatch queue. To be used by | |
472 | * specific elevators. | |
9817064b JA |
473 | */ |
474 | void elv_dispatch_add_tail(struct request_queue *q, struct request *rq) | |
475 | { | |
476 | if (q->last_merge == rq) | |
477 | q->last_merge = NULL; | |
478 | ||
479 | elv_rqhash_del(q, rq); | |
480 | ||
481 | q->nr_sorted--; | |
482 | ||
483 | q->end_sector = rq_end_sector(rq); | |
484 | q->boundary_rq = rq; | |
485 | list_add_tail(&rq->queuelist, &q->queue_head); | |
486 | } | |
2e662b65 JA |
487 | EXPORT_SYMBOL(elv_dispatch_add_tail); |
488 | ||
165125e1 | 489 | int elv_merge(struct request_queue *q, struct request **req, struct bio *bio) |
1da177e4 LT |
490 | { |
491 | elevator_t *e = q->elevator; | |
9817064b | 492 | struct request *__rq; |
06b86245 TH |
493 | int ret; |
494 | ||
9817064b JA |
495 | /* |
496 | * First try one-hit cache. | |
497 | */ | |
06b86245 TH |
498 | if (q->last_merge) { |
499 | ret = elv_try_merge(q->last_merge, bio); | |
500 | if (ret != ELEVATOR_NO_MERGE) { | |
501 | *req = q->last_merge; | |
502 | return ret; | |
503 | } | |
504 | } | |
1da177e4 | 505 | |
ac9fafa1 AB |
506 | if (blk_queue_nomerges(q)) |
507 | return ELEVATOR_NO_MERGE; | |
508 | ||
9817064b JA |
509 | /* |
510 | * See if our hash lookup can find a potential backmerge. | |
511 | */ | |
512 | __rq = elv_rqhash_find(q, bio->bi_sector); | |
513 | if (__rq && elv_rq_merge_ok(__rq, bio)) { | |
514 | *req = __rq; | |
515 | return ELEVATOR_BACK_MERGE; | |
516 | } | |
517 | ||
1da177e4 LT |
518 | if (e->ops->elevator_merge_fn) |
519 | return e->ops->elevator_merge_fn(q, req, bio); | |
520 | ||
521 | return ELEVATOR_NO_MERGE; | |
522 | } | |
523 | ||
165125e1 | 524 | void elv_merged_request(struct request_queue *q, struct request *rq, int type) |
1da177e4 LT |
525 | { |
526 | elevator_t *e = q->elevator; | |
527 | ||
528 | if (e->ops->elevator_merged_fn) | |
2e662b65 | 529 | e->ops->elevator_merged_fn(q, rq, type); |
06b86245 | 530 | |
2e662b65 JA |
531 | if (type == ELEVATOR_BACK_MERGE) |
532 | elv_rqhash_reposition(q, rq); | |
9817064b | 533 | |
06b86245 | 534 | q->last_merge = rq; |
1da177e4 LT |
535 | } |
536 | ||
165125e1 | 537 | void elv_merge_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
538 | struct request *next) |
539 | { | |
540 | elevator_t *e = q->elevator; | |
541 | ||
1da177e4 LT |
542 | if (e->ops->elevator_merge_req_fn) |
543 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 | 544 | |
9817064b JA |
545 | elv_rqhash_reposition(q, rq); |
546 | elv_rqhash_del(q, next); | |
547 | ||
548 | q->nr_sorted--; | |
06b86245 | 549 | q->last_merge = rq; |
1da177e4 LT |
550 | } |
551 | ||
165125e1 | 552 | void elv_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 553 | { |
1da177e4 LT |
554 | /* |
555 | * it already went through dequeue, we need to decrement the | |
556 | * in_flight count again | |
557 | */ | |
8922e16c | 558 | if (blk_account_rq(rq)) { |
1da177e4 | 559 | q->in_flight--; |
cad97516 JA |
560 | if (blk_sorted_rq(rq)) |
561 | elv_deactivate_rq(q, rq); | |
8922e16c | 562 | } |
1da177e4 | 563 | |
4aff5e23 | 564 | rq->cmd_flags &= ~REQ_STARTED; |
1da177e4 | 565 | |
30e9656c | 566 | elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE); |
1da177e4 LT |
567 | } |
568 | ||
165125e1 | 569 | static void elv_drain_elevator(struct request_queue *q) |
15853af9 TH |
570 | { |
571 | static int printed; | |
572 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
573 | ; | |
574 | if (q->nr_sorted == 0) | |
575 | return; | |
576 | if (printed++ < 10) { | |
577 | printk(KERN_ERR "%s: forced dispatching is broken " | |
578 | "(nr_sorted=%u), please report this\n", | |
579 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
580 | } | |
581 | } | |
582 | ||
165125e1 | 583 | void elv_insert(struct request_queue *q, struct request *rq, int where) |
1da177e4 | 584 | { |
797e7dbb TH |
585 | struct list_head *pos; |
586 | unsigned ordseq; | |
dac07ec1 | 587 | int unplug_it = 1; |
797e7dbb | 588 | |
2056a782 JA |
589 | blk_add_trace_rq(q, rq, BLK_TA_INSERT); |
590 | ||
1da177e4 LT |
591 | rq->q = q; |
592 | ||
8922e16c TH |
593 | switch (where) { |
594 | case ELEVATOR_INSERT_FRONT: | |
4aff5e23 | 595 | rq->cmd_flags |= REQ_SOFTBARRIER; |
8922e16c TH |
596 | |
597 | list_add(&rq->queuelist, &q->queue_head); | |
598 | break; | |
599 | ||
600 | case ELEVATOR_INSERT_BACK: | |
4aff5e23 | 601 | rq->cmd_flags |= REQ_SOFTBARRIER; |
15853af9 | 602 | elv_drain_elevator(q); |
8922e16c TH |
603 | list_add_tail(&rq->queuelist, &q->queue_head); |
604 | /* | |
605 | * We kick the queue here for the following reasons. | |
606 | * - The elevator might have returned NULL previously | |
607 | * to delay requests and returned them now. As the | |
608 | * queue wasn't empty before this request, ll_rw_blk | |
609 | * won't run the queue on return, resulting in hang. | |
610 | * - Usually, back inserted requests won't be merged | |
611 | * with anything. There's no point in delaying queue | |
612 | * processing. | |
613 | */ | |
614 | blk_remove_plug(q); | |
615 | q->request_fn(q); | |
616 | break; | |
617 | ||
618 | case ELEVATOR_INSERT_SORT: | |
e17fc0a1 | 619 | BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq)); |
4aff5e23 | 620 | rq->cmd_flags |= REQ_SORTED; |
15853af9 | 621 | q->nr_sorted++; |
9817064b JA |
622 | if (rq_mergeable(rq)) { |
623 | elv_rqhash_add(q, rq); | |
624 | if (!q->last_merge) | |
625 | q->last_merge = rq; | |
626 | } | |
627 | ||
ca23509f TH |
628 | /* |
629 | * Some ioscheds (cfq) run q->request_fn directly, so | |
630 | * rq cannot be accessed after calling | |
631 | * elevator_add_req_fn. | |
632 | */ | |
633 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
634 | break; |
635 | ||
797e7dbb TH |
636 | case ELEVATOR_INSERT_REQUEUE: |
637 | /* | |
638 | * If ordered flush isn't in progress, we do front | |
639 | * insertion; otherwise, requests should be requeued | |
640 | * in ordseq order. | |
641 | */ | |
4aff5e23 | 642 | rq->cmd_flags |= REQ_SOFTBARRIER; |
797e7dbb | 643 | |
95543179 LV |
644 | /* |
645 | * Most requeues happen because of a busy condition, | |
646 | * don't force unplug of the queue for that case. | |
647 | */ | |
648 | unplug_it = 0; | |
649 | ||
797e7dbb TH |
650 | if (q->ordseq == 0) { |
651 | list_add(&rq->queuelist, &q->queue_head); | |
652 | break; | |
653 | } | |
654 | ||
655 | ordseq = blk_ordered_req_seq(rq); | |
656 | ||
657 | list_for_each(pos, &q->queue_head) { | |
658 | struct request *pos_rq = list_entry_rq(pos); | |
659 | if (ordseq <= blk_ordered_req_seq(pos_rq)) | |
660 | break; | |
661 | } | |
662 | ||
663 | list_add_tail(&rq->queuelist, pos); | |
664 | break; | |
665 | ||
8922e16c TH |
666 | default: |
667 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
24c03d47 | 668 | __func__, where); |
8922e16c TH |
669 | BUG(); |
670 | } | |
671 | ||
dac07ec1 | 672 | if (unplug_it && blk_queue_plugged(q)) { |
8922e16c TH |
673 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] |
674 | - q->in_flight; | |
675 | ||
676 | if (nrq >= q->unplug_thresh) | |
677 | __generic_unplug_device(q); | |
678 | } | |
1da177e4 LT |
679 | } |
680 | ||
165125e1 | 681 | void __elv_add_request(struct request_queue *q, struct request *rq, int where, |
30e9656c TH |
682 | int plug) |
683 | { | |
684 | if (q->ordcolor) | |
4aff5e23 | 685 | rq->cmd_flags |= REQ_ORDERED_COLOR; |
30e9656c | 686 | |
4aff5e23 | 687 | if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
30e9656c TH |
688 | /* |
689 | * toggle ordered color | |
690 | */ | |
691 | if (blk_barrier_rq(rq)) | |
692 | q->ordcolor ^= 1; | |
693 | ||
694 | /* | |
695 | * barriers implicitly indicate back insertion | |
696 | */ | |
697 | if (where == ELEVATOR_INSERT_SORT) | |
698 | where = ELEVATOR_INSERT_BACK; | |
699 | ||
700 | /* | |
701 | * this request is scheduling boundary, update | |
702 | * end_sector | |
703 | */ | |
e17fc0a1 | 704 | if (blk_fs_request(rq) || blk_discard_rq(rq)) { |
30e9656c TH |
705 | q->end_sector = rq_end_sector(rq); |
706 | q->boundary_rq = rq; | |
707 | } | |
4eb166d9 JA |
708 | } else if (!(rq->cmd_flags & REQ_ELVPRIV) && |
709 | where == ELEVATOR_INSERT_SORT) | |
30e9656c TH |
710 | where = ELEVATOR_INSERT_BACK; |
711 | ||
712 | if (plug) | |
713 | blk_plug_device(q); | |
714 | ||
715 | elv_insert(q, rq, where); | |
716 | } | |
2e662b65 JA |
717 | EXPORT_SYMBOL(__elv_add_request); |
718 | ||
165125e1 | 719 | void elv_add_request(struct request_queue *q, struct request *rq, int where, |
1da177e4 LT |
720 | int plug) |
721 | { | |
722 | unsigned long flags; | |
723 | ||
724 | spin_lock_irqsave(q->queue_lock, flags); | |
725 | __elv_add_request(q, rq, where, plug); | |
726 | spin_unlock_irqrestore(q->queue_lock, flags); | |
727 | } | |
2e662b65 JA |
728 | EXPORT_SYMBOL(elv_add_request); |
729 | ||
165125e1 | 730 | static inline struct request *__elv_next_request(struct request_queue *q) |
1da177e4 | 731 | { |
8922e16c TH |
732 | struct request *rq; |
733 | ||
797e7dbb TH |
734 | while (1) { |
735 | while (!list_empty(&q->queue_head)) { | |
736 | rq = list_entry_rq(q->queue_head.next); | |
737 | if (blk_do_ordered(q, &rq)) | |
738 | return rq; | |
739 | } | |
1da177e4 | 740 | |
797e7dbb TH |
741 | if (!q->elevator->ops->elevator_dispatch_fn(q, 0)) |
742 | return NULL; | |
1da177e4 | 743 | } |
1da177e4 LT |
744 | } |
745 | ||
165125e1 | 746 | struct request *elv_next_request(struct request_queue *q) |
1da177e4 LT |
747 | { |
748 | struct request *rq; | |
749 | int ret; | |
750 | ||
751 | while ((rq = __elv_next_request(q)) != NULL) { | |
bf2de6f5 JA |
752 | /* |
753 | * Kill the empty barrier place holder, the driver must | |
754 | * not ever see it. | |
755 | */ | |
756 | if (blk_empty_barrier(rq)) { | |
99cd3386 | 757 | __blk_end_request(rq, 0, blk_rq_bytes(rq)); |
bf2de6f5 JA |
758 | continue; |
759 | } | |
4aff5e23 | 760 | if (!(rq->cmd_flags & REQ_STARTED)) { |
8922e16c TH |
761 | /* |
762 | * This is the first time the device driver | |
763 | * sees this request (possibly after | |
764 | * requeueing). Notify IO scheduler. | |
765 | */ | |
cad97516 JA |
766 | if (blk_sorted_rq(rq)) |
767 | elv_activate_rq(q, rq); | |
1da177e4 | 768 | |
8922e16c TH |
769 | /* |
770 | * just mark as started even if we don't start | |
771 | * it, a request that has been delayed should | |
772 | * not be passed by new incoming requests | |
773 | */ | |
4aff5e23 | 774 | rq->cmd_flags |= REQ_STARTED; |
2056a782 | 775 | blk_add_trace_rq(q, rq, BLK_TA_ISSUE); |
242f9dcb JA |
776 | |
777 | /* | |
778 | * We are now handing the request to the hardware, | |
779 | * add the timeout handler | |
780 | */ | |
781 | blk_add_timer(rq); | |
8922e16c | 782 | } |
1da177e4 | 783 | |
8922e16c | 784 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 785 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
786 | q->boundary_rq = NULL; |
787 | } | |
1da177e4 | 788 | |
fa0ccd83 JB |
789 | if (rq->cmd_flags & REQ_DONTPREP) |
790 | break; | |
791 | ||
792 | if (q->dma_drain_size && rq->data_len) { | |
793 | /* | |
794 | * make sure space for the drain appears we | |
795 | * know we can do this because max_hw_segments | |
796 | * has been adjusted to be one fewer than the | |
797 | * device can handle | |
798 | */ | |
799 | rq->nr_phys_segments++; | |
fa0ccd83 JB |
800 | } |
801 | ||
802 | if (!q->prep_rq_fn) | |
1da177e4 LT |
803 | break; |
804 | ||
805 | ret = q->prep_rq_fn(q, rq); | |
806 | if (ret == BLKPREP_OK) { | |
807 | break; | |
808 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
809 | /* |
810 | * the request may have been (partially) prepped. | |
811 | * we need to keep this request in the front to | |
8922e16c TH |
812 | * avoid resource deadlock. REQ_STARTED will |
813 | * prevent other fs requests from passing this one. | |
2e759cd4 | 814 | */ |
fa0ccd83 JB |
815 | if (q->dma_drain_size && rq->data_len && |
816 | !(rq->cmd_flags & REQ_DONTPREP)) { | |
817 | /* | |
818 | * remove the space for the drain we added | |
819 | * so that we don't add it again | |
820 | */ | |
821 | --rq->nr_phys_segments; | |
fa0ccd83 JB |
822 | } |
823 | ||
1da177e4 LT |
824 | rq = NULL; |
825 | break; | |
826 | } else if (ret == BLKPREP_KILL) { | |
4aff5e23 | 827 | rq->cmd_flags |= REQ_QUIET; |
99cd3386 | 828 | __blk_end_request(rq, -EIO, blk_rq_bytes(rq)); |
1da177e4 | 829 | } else { |
24c03d47 | 830 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); |
1da177e4 LT |
831 | break; |
832 | } | |
833 | } | |
834 | ||
835 | return rq; | |
836 | } | |
2e662b65 JA |
837 | EXPORT_SYMBOL(elv_next_request); |
838 | ||
165125e1 | 839 | void elv_dequeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 840 | { |
8922e16c | 841 | BUG_ON(list_empty(&rq->queuelist)); |
9817064b | 842 | BUG_ON(ELV_ON_HASH(rq)); |
8922e16c TH |
843 | |
844 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
845 | |
846 | /* | |
847 | * the time frame between a request being removed from the lists | |
848 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 849 | * the driver side. |
1da177e4 LT |
850 | */ |
851 | if (blk_account_rq(rq)) | |
852 | q->in_flight++; | |
1da177e4 | 853 | } |
2e662b65 JA |
854 | EXPORT_SYMBOL(elv_dequeue_request); |
855 | ||
165125e1 | 856 | int elv_queue_empty(struct request_queue *q) |
1da177e4 LT |
857 | { |
858 | elevator_t *e = q->elevator; | |
859 | ||
8922e16c TH |
860 | if (!list_empty(&q->queue_head)) |
861 | return 0; | |
862 | ||
1da177e4 LT |
863 | if (e->ops->elevator_queue_empty_fn) |
864 | return e->ops->elevator_queue_empty_fn(q); | |
865 | ||
8922e16c | 866 | return 1; |
1da177e4 | 867 | } |
2e662b65 JA |
868 | EXPORT_SYMBOL(elv_queue_empty); |
869 | ||
165125e1 | 870 | struct request *elv_latter_request(struct request_queue *q, struct request *rq) |
1da177e4 | 871 | { |
1da177e4 LT |
872 | elevator_t *e = q->elevator; |
873 | ||
874 | if (e->ops->elevator_latter_req_fn) | |
875 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
876 | return NULL; |
877 | } | |
878 | ||
165125e1 | 879 | struct request *elv_former_request(struct request_queue *q, struct request *rq) |
1da177e4 | 880 | { |
1da177e4 LT |
881 | elevator_t *e = q->elevator; |
882 | ||
883 | if (e->ops->elevator_former_req_fn) | |
884 | return e->ops->elevator_former_req_fn(q, rq); | |
1da177e4 LT |
885 | return NULL; |
886 | } | |
887 | ||
165125e1 | 888 | int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
889 | { |
890 | elevator_t *e = q->elevator; | |
891 | ||
892 | if (e->ops->elevator_set_req_fn) | |
cb78b285 | 893 | return e->ops->elevator_set_req_fn(q, rq, gfp_mask); |
1da177e4 LT |
894 | |
895 | rq->elevator_private = NULL; | |
896 | return 0; | |
897 | } | |
898 | ||
165125e1 | 899 | void elv_put_request(struct request_queue *q, struct request *rq) |
1da177e4 LT |
900 | { |
901 | elevator_t *e = q->elevator; | |
902 | ||
903 | if (e->ops->elevator_put_req_fn) | |
bb37b94c | 904 | e->ops->elevator_put_req_fn(rq); |
1da177e4 LT |
905 | } |
906 | ||
165125e1 | 907 | int elv_may_queue(struct request_queue *q, int rw) |
1da177e4 LT |
908 | { |
909 | elevator_t *e = q->elevator; | |
910 | ||
911 | if (e->ops->elevator_may_queue_fn) | |
cb78b285 | 912 | return e->ops->elevator_may_queue_fn(q, rw); |
1da177e4 LT |
913 | |
914 | return ELV_MQUEUE_MAY; | |
915 | } | |
916 | ||
11914a53 MA |
917 | void elv_abort_queue(struct request_queue *q) |
918 | { | |
919 | struct request *rq; | |
920 | ||
921 | while (!list_empty(&q->queue_head)) { | |
922 | rq = list_entry_rq(q->queue_head.next); | |
923 | rq->cmd_flags |= REQ_QUIET; | |
924 | blk_add_trace_rq(q, rq, BLK_TA_ABORT); | |
99cd3386 | 925 | __blk_end_request(rq, -EIO, blk_rq_bytes(rq)); |
11914a53 MA |
926 | } |
927 | } | |
928 | EXPORT_SYMBOL(elv_abort_queue); | |
929 | ||
165125e1 | 930 | void elv_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 LT |
931 | { |
932 | elevator_t *e = q->elevator; | |
933 | ||
934 | /* | |
935 | * request is released from the driver, io must be done | |
936 | */ | |
8922e16c | 937 | if (blk_account_rq(rq)) { |
1da177e4 | 938 | q->in_flight--; |
1bc691d3 TH |
939 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
940 | e->ops->elevator_completed_req_fn(q, rq); | |
941 | } | |
797e7dbb | 942 | |
1bc691d3 TH |
943 | /* |
944 | * Check if the queue is waiting for fs requests to be | |
945 | * drained for flush sequence. | |
946 | */ | |
947 | if (unlikely(q->ordseq)) { | |
948 | struct request *first_rq = list_entry_rq(q->queue_head.next); | |
949 | if (q->in_flight == 0 && | |
797e7dbb TH |
950 | blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN && |
951 | blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) { | |
952 | blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0); | |
953 | q->request_fn(q); | |
954 | } | |
8922e16c | 955 | } |
1da177e4 LT |
956 | } |
957 | ||
3d1ab40f AV |
958 | #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) |
959 | ||
960 | static ssize_t | |
961 | elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1da177e4 | 962 | { |
3d1ab40f AV |
963 | elevator_t *e = container_of(kobj, elevator_t, kobj); |
964 | struct elv_fs_entry *entry = to_elv(attr); | |
965 | ssize_t error; | |
966 | ||
967 | if (!entry->show) | |
968 | return -EIO; | |
969 | ||
970 | mutex_lock(&e->sysfs_lock); | |
971 | error = e->ops ? entry->show(e, page) : -ENOENT; | |
972 | mutex_unlock(&e->sysfs_lock); | |
973 | return error; | |
974 | } | |
1da177e4 | 975 | |
3d1ab40f AV |
976 | static ssize_t |
977 | elv_attr_store(struct kobject *kobj, struct attribute *attr, | |
978 | const char *page, size_t length) | |
979 | { | |
980 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
981 | struct elv_fs_entry *entry = to_elv(attr); | |
982 | ssize_t error; | |
1da177e4 | 983 | |
3d1ab40f AV |
984 | if (!entry->store) |
985 | return -EIO; | |
1da177e4 | 986 | |
3d1ab40f AV |
987 | mutex_lock(&e->sysfs_lock); |
988 | error = e->ops ? entry->store(e, page, length) : -ENOENT; | |
989 | mutex_unlock(&e->sysfs_lock); | |
990 | return error; | |
991 | } | |
992 | ||
993 | static struct sysfs_ops elv_sysfs_ops = { | |
994 | .show = elv_attr_show, | |
995 | .store = elv_attr_store, | |
996 | }; | |
997 | ||
998 | static struct kobj_type elv_ktype = { | |
999 | .sysfs_ops = &elv_sysfs_ops, | |
1000 | .release = elevator_release, | |
1001 | }; | |
1002 | ||
1003 | int elv_register_queue(struct request_queue *q) | |
1004 | { | |
1005 | elevator_t *e = q->elevator; | |
1006 | int error; | |
1007 | ||
b2d6db58 | 1008 | error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); |
3d1ab40f | 1009 | if (!error) { |
e572ec7e | 1010 | struct elv_fs_entry *attr = e->elevator_type->elevator_attrs; |
3d1ab40f | 1011 | if (attr) { |
e572ec7e AV |
1012 | while (attr->attr.name) { |
1013 | if (sysfs_create_file(&e->kobj, &attr->attr)) | |
3d1ab40f | 1014 | break; |
e572ec7e | 1015 | attr++; |
3d1ab40f AV |
1016 | } |
1017 | } | |
1018 | kobject_uevent(&e->kobj, KOBJ_ADD); | |
1019 | } | |
1020 | return error; | |
1da177e4 LT |
1021 | } |
1022 | ||
bc1c1169 JA |
1023 | static void __elv_unregister_queue(elevator_t *e) |
1024 | { | |
1025 | kobject_uevent(&e->kobj, KOBJ_REMOVE); | |
1026 | kobject_del(&e->kobj); | |
1027 | } | |
1028 | ||
1da177e4 LT |
1029 | void elv_unregister_queue(struct request_queue *q) |
1030 | { | |
bc1c1169 JA |
1031 | if (q) |
1032 | __elv_unregister_queue(q->elevator); | |
1da177e4 LT |
1033 | } |
1034 | ||
2fdd82bd | 1035 | void elv_register(struct elevator_type *e) |
1da177e4 | 1036 | { |
1ffb96c5 | 1037 | char *def = ""; |
2a12dcd7 JA |
1038 | |
1039 | spin_lock(&elv_list_lock); | |
ce524497 | 1040 | BUG_ON(elevator_find(e->elevator_name)); |
1da177e4 | 1041 | list_add_tail(&e->list, &elv_list); |
2a12dcd7 | 1042 | spin_unlock(&elv_list_lock); |
1da177e4 | 1043 | |
5f003976 ND |
1044 | if (!strcmp(e->elevator_name, chosen_elevator) || |
1045 | (!*chosen_elevator && | |
1046 | !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED))) | |
1ffb96c5 TV |
1047 | def = " (default)"; |
1048 | ||
4eb166d9 JA |
1049 | printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name, |
1050 | def); | |
1da177e4 LT |
1051 | } |
1052 | EXPORT_SYMBOL_GPL(elv_register); | |
1053 | ||
1054 | void elv_unregister(struct elevator_type *e) | |
1055 | { | |
83521d3e CH |
1056 | struct task_struct *g, *p; |
1057 | ||
1058 | /* | |
1059 | * Iterate every thread in the process to remove the io contexts. | |
1060 | */ | |
e17a9489 AV |
1061 | if (e->ops.trim) { |
1062 | read_lock(&tasklist_lock); | |
1063 | do_each_thread(g, p) { | |
1064 | task_lock(p); | |
2d8f6131 ON |
1065 | if (p->io_context) |
1066 | e->ops.trim(p->io_context); | |
e17a9489 AV |
1067 | task_unlock(p); |
1068 | } while_each_thread(g, p); | |
1069 | read_unlock(&tasklist_lock); | |
1070 | } | |
83521d3e | 1071 | |
2a12dcd7 | 1072 | spin_lock(&elv_list_lock); |
1da177e4 | 1073 | list_del_init(&e->list); |
2a12dcd7 | 1074 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
1075 | } |
1076 | EXPORT_SYMBOL_GPL(elv_unregister); | |
1077 | ||
1078 | /* | |
1079 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
1080 | * we don't free the old io scheduler, before we have allocated what we | |
1081 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 1082 | * one, if the new one fails init for some reason. |
1da177e4 | 1083 | */ |
165125e1 | 1084 | static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) |
1da177e4 | 1085 | { |
cb98fc8b | 1086 | elevator_t *old_elevator, *e; |
bc1c1169 | 1087 | void *data; |
1da177e4 | 1088 | |
cb98fc8b TH |
1089 | /* |
1090 | * Allocate new elevator | |
1091 | */ | |
b5deef90 | 1092 | e = elevator_alloc(q, new_e); |
1da177e4 | 1093 | if (!e) |
3d1ab40f | 1094 | return 0; |
1da177e4 | 1095 | |
bc1c1169 JA |
1096 | data = elevator_init_queue(q, e); |
1097 | if (!data) { | |
1098 | kobject_put(&e->kobj); | |
1099 | return 0; | |
1100 | } | |
1101 | ||
1da177e4 | 1102 | /* |
cb98fc8b | 1103 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 1104 | */ |
cb98fc8b TH |
1105 | spin_lock_irq(q->queue_lock); |
1106 | ||
75ad23bc | 1107 | queue_flag_set(QUEUE_FLAG_ELVSWITCH, q); |
cb98fc8b | 1108 | |
15853af9 | 1109 | elv_drain_elevator(q); |
cb98fc8b TH |
1110 | |
1111 | while (q->rq.elvpriv) { | |
407df2aa TH |
1112 | blk_remove_plug(q); |
1113 | q->request_fn(q); | |
cb98fc8b | 1114 | spin_unlock_irq(q->queue_lock); |
64521d1a | 1115 | msleep(10); |
cb98fc8b | 1116 | spin_lock_irq(q->queue_lock); |
15853af9 | 1117 | elv_drain_elevator(q); |
cb98fc8b TH |
1118 | } |
1119 | ||
1da177e4 | 1120 | /* |
bc1c1169 | 1121 | * Remember old elevator. |
1da177e4 | 1122 | */ |
1da177e4 LT |
1123 | old_elevator = q->elevator; |
1124 | ||
1da177e4 LT |
1125 | /* |
1126 | * attach and start new elevator | |
1127 | */ | |
bc1c1169 JA |
1128 | elevator_attach(q, e, data); |
1129 | ||
1130 | spin_unlock_irq(q->queue_lock); | |
1131 | ||
1132 | __elv_unregister_queue(old_elevator); | |
1da177e4 LT |
1133 | |
1134 | if (elv_register_queue(q)) | |
1135 | goto fail_register; | |
1136 | ||
1137 | /* | |
cb98fc8b | 1138 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
1139 | */ |
1140 | elevator_exit(old_elevator); | |
75ad23bc NP |
1141 | spin_lock_irq(q->queue_lock); |
1142 | queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q); | |
1143 | spin_unlock_irq(q->queue_lock); | |
1144 | ||
4722dc52 AB |
1145 | blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name); |
1146 | ||
3d1ab40f | 1147 | return 1; |
1da177e4 LT |
1148 | |
1149 | fail_register: | |
1150 | /* | |
1151 | * switch failed, exit the new io scheduler and reattach the old | |
1152 | * one again (along with re-adding the sysfs dir) | |
1153 | */ | |
1154 | elevator_exit(e); | |
1da177e4 LT |
1155 | q->elevator = old_elevator; |
1156 | elv_register_queue(q); | |
75ad23bc NP |
1157 | |
1158 | spin_lock_irq(q->queue_lock); | |
1159 | queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q); | |
1160 | spin_unlock_irq(q->queue_lock); | |
1161 | ||
3d1ab40f | 1162 | return 0; |
1da177e4 LT |
1163 | } |
1164 | ||
165125e1 JA |
1165 | ssize_t elv_iosched_store(struct request_queue *q, const char *name, |
1166 | size_t count) | |
1da177e4 LT |
1167 | { |
1168 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 1169 | size_t len; |
1da177e4 LT |
1170 | struct elevator_type *e; |
1171 | ||
be561235 TH |
1172 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
1173 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
1174 | len = strlen(elevator_name); | |
1da177e4 | 1175 | |
be561235 TH |
1176 | if (len && elevator_name[len - 1] == '\n') |
1177 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
1178 | |
1179 | e = elevator_get(elevator_name); | |
1180 | if (!e) { | |
1181 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
1182 | return -EINVAL; | |
1183 | } | |
1184 | ||
2ca7d93b ND |
1185 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
1186 | elevator_put(e); | |
1da177e4 | 1187 | return count; |
2ca7d93b | 1188 | } |
1da177e4 | 1189 | |
3d1ab40f | 1190 | if (!elevator_switch(q, e)) |
4eb166d9 JA |
1191 | printk(KERN_ERR "elevator: switch to %s failed\n", |
1192 | elevator_name); | |
1da177e4 LT |
1193 | return count; |
1194 | } | |
1195 | ||
165125e1 | 1196 | ssize_t elv_iosched_show(struct request_queue *q, char *name) |
1da177e4 LT |
1197 | { |
1198 | elevator_t *e = q->elevator; | |
1199 | struct elevator_type *elv = e->elevator_type; | |
70cee26e | 1200 | struct elevator_type *__e; |
1da177e4 LT |
1201 | int len = 0; |
1202 | ||
2a12dcd7 | 1203 | spin_lock(&elv_list_lock); |
70cee26e | 1204 | list_for_each_entry(__e, &elv_list, list) { |
1da177e4 LT |
1205 | if (!strcmp(elv->elevator_name, __e->elevator_name)) |
1206 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
1207 | else | |
1208 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
1209 | } | |
2a12dcd7 | 1210 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
1211 | |
1212 | len += sprintf(len+name, "\n"); | |
1213 | return len; | |
1214 | } | |
1215 | ||
165125e1 JA |
1216 | struct request *elv_rb_former_request(struct request_queue *q, |
1217 | struct request *rq) | |
2e662b65 JA |
1218 | { |
1219 | struct rb_node *rbprev = rb_prev(&rq->rb_node); | |
1220 | ||
1221 | if (rbprev) | |
1222 | return rb_entry_rq(rbprev); | |
1223 | ||
1224 | return NULL; | |
1225 | } | |
2e662b65 JA |
1226 | EXPORT_SYMBOL(elv_rb_former_request); |
1227 | ||
165125e1 JA |
1228 | struct request *elv_rb_latter_request(struct request_queue *q, |
1229 | struct request *rq) | |
2e662b65 JA |
1230 | { |
1231 | struct rb_node *rbnext = rb_next(&rq->rb_node); | |
1232 | ||
1233 | if (rbnext) | |
1234 | return rb_entry_rq(rbnext); | |
1235 | ||
1236 | return NULL; | |
1237 | } | |
2e662b65 | 1238 | EXPORT_SYMBOL(elv_rb_latter_request); |