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> |
1da177e4 LT |
37 | |
38 | #include <asm/uaccess.h> | |
39 | ||
40 | static DEFINE_SPINLOCK(elv_list_lock); | |
41 | static LIST_HEAD(elv_list); | |
42 | ||
9817064b JA |
43 | /* |
44 | * Merge hash stuff. | |
45 | */ | |
46 | static const int elv_hash_shift = 6; | |
47 | #define ELV_HASH_BLOCK(sec) ((sec) >> 3) | |
4eb166d9 JA |
48 | #define ELV_HASH_FN(sec) \ |
49 | (hash_long(ELV_HASH_BLOCK((sec)), elv_hash_shift)) | |
9817064b JA |
50 | #define ELV_HASH_ENTRIES (1 << elv_hash_shift) |
51 | #define rq_hash_key(rq) ((rq)->sector + (rq)->nr_sectors) | |
52 | #define ELV_ON_HASH(rq) (!hlist_unhashed(&(rq)->hash)) | |
53 | ||
da775265 JA |
54 | /* |
55 | * Query io scheduler to see if the current process issuing bio may be | |
56 | * merged with rq. | |
57 | */ | |
58 | static int elv_iosched_allow_merge(struct request *rq, struct bio *bio) | |
59 | { | |
165125e1 | 60 | struct request_queue *q = rq->q; |
da775265 JA |
61 | elevator_t *e = q->elevator; |
62 | ||
63 | if (e->ops->elevator_allow_merge_fn) | |
64 | return e->ops->elevator_allow_merge_fn(q, rq, bio); | |
65 | ||
66 | return 1; | |
67 | } | |
68 | ||
1da177e4 LT |
69 | /* |
70 | * can we safely merge with this request? | |
71 | */ | |
72ed0bf6 | 72 | int elv_rq_merge_ok(struct request *rq, struct bio *bio) |
1da177e4 LT |
73 | { |
74 | if (!rq_mergeable(rq)) | |
75 | return 0; | |
76 | ||
e17fc0a1 DW |
77 | /* |
78 | * Don't merge file system requests and discard requests | |
79 | */ | |
80 | if (bio_discard(bio) != bio_discard(rq->bio)) | |
81 | return 0; | |
82 | ||
1da177e4 LT |
83 | /* |
84 | * different data direction or already started, don't merge | |
85 | */ | |
86 | if (bio_data_dir(bio) != rq_data_dir(rq)) | |
87 | return 0; | |
88 | ||
89 | /* | |
da775265 | 90 | * must be same device and not a special request |
1da177e4 | 91 | */ |
bb4067e3 | 92 | if (rq->rq_disk != bio->bi_bdev->bd_disk || rq->special) |
da775265 JA |
93 | return 0; |
94 | ||
7ba1ba12 MP |
95 | /* |
96 | * only merge integrity protected bio into ditto rq | |
97 | */ | |
98 | if (bio_integrity(bio) != blk_integrity_rq(rq)) | |
99 | return 0; | |
100 | ||
da775265 JA |
101 | if (!elv_iosched_allow_merge(rq, bio)) |
102 | return 0; | |
1da177e4 | 103 | |
da775265 | 104 | return 1; |
1da177e4 LT |
105 | } |
106 | EXPORT_SYMBOL(elv_rq_merge_ok); | |
107 | ||
769db45b | 108 | static inline int elv_try_merge(struct request *__rq, struct bio *bio) |
1da177e4 LT |
109 | { |
110 | int ret = ELEVATOR_NO_MERGE; | |
111 | ||
112 | /* | |
113 | * we can merge and sequence is ok, check if it's possible | |
114 | */ | |
115 | if (elv_rq_merge_ok(__rq, bio)) { | |
116 | if (__rq->sector + __rq->nr_sectors == bio->bi_sector) | |
117 | ret = ELEVATOR_BACK_MERGE; | |
118 | else if (__rq->sector - bio_sectors(bio) == bio->bi_sector) | |
119 | ret = ELEVATOR_FRONT_MERGE; | |
120 | } | |
121 | ||
122 | return ret; | |
123 | } | |
1da177e4 | 124 | |
1da177e4 LT |
125 | static struct elevator_type *elevator_find(const char *name) |
126 | { | |
a22b169d | 127 | struct elevator_type *e; |
1da177e4 | 128 | |
70cee26e | 129 | list_for_each_entry(e, &elv_list, list) { |
a22b169d VT |
130 | if (!strcmp(e->elevator_name, name)) |
131 | return e; | |
1da177e4 | 132 | } |
1da177e4 | 133 | |
a22b169d | 134 | return NULL; |
1da177e4 LT |
135 | } |
136 | ||
137 | static void elevator_put(struct elevator_type *e) | |
138 | { | |
139 | module_put(e->elevator_owner); | |
140 | } | |
141 | ||
142 | static struct elevator_type *elevator_get(const char *name) | |
143 | { | |
2824bc93 | 144 | struct elevator_type *e; |
1da177e4 | 145 | |
2a12dcd7 | 146 | spin_lock(&elv_list_lock); |
2824bc93 TH |
147 | |
148 | e = elevator_find(name); | |
e1640949 JA |
149 | if (!e) { |
150 | char elv[ELV_NAME_MAX + strlen("-iosched")]; | |
151 | ||
152 | spin_unlock(&elv_list_lock); | |
153 | ||
154 | if (!strcmp(name, "anticipatory")) | |
155 | sprintf(elv, "as-iosched"); | |
156 | else | |
157 | sprintf(elv, "%s-iosched", name); | |
158 | ||
e180f594 | 159 | request_module("%s", elv); |
e1640949 JA |
160 | spin_lock(&elv_list_lock); |
161 | e = elevator_find(name); | |
162 | } | |
163 | ||
2824bc93 TH |
164 | if (e && !try_module_get(e->elevator_owner)) |
165 | e = NULL; | |
166 | ||
2a12dcd7 | 167 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
168 | |
169 | return e; | |
170 | } | |
171 | ||
165125e1 JA |
172 | static void *elevator_init_queue(struct request_queue *q, |
173 | struct elevator_queue *eq) | |
1da177e4 | 174 | { |
bb37b94c | 175 | return eq->ops->elevator_init_fn(q); |
bc1c1169 | 176 | } |
1da177e4 | 177 | |
165125e1 | 178 | static void elevator_attach(struct request_queue *q, struct elevator_queue *eq, |
bc1c1169 JA |
179 | void *data) |
180 | { | |
1da177e4 | 181 | q->elevator = eq; |
bc1c1169 | 182 | eq->elevator_data = data; |
1da177e4 LT |
183 | } |
184 | ||
185 | static char chosen_elevator[16]; | |
186 | ||
5f003976 | 187 | static int __init elevator_setup(char *str) |
1da177e4 | 188 | { |
752a3b79 CE |
189 | /* |
190 | * Be backwards-compatible with previous kernels, so users | |
191 | * won't get the wrong elevator. | |
192 | */ | |
5f003976 | 193 | if (!strcmp(str, "as")) |
752a3b79 | 194 | strcpy(chosen_elevator, "anticipatory"); |
cff3ba22 | 195 | else |
5f003976 | 196 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); |
9b41046c | 197 | return 1; |
1da177e4 LT |
198 | } |
199 | ||
200 | __setup("elevator=", elevator_setup); | |
201 | ||
3d1ab40f AV |
202 | static struct kobj_type elv_ktype; |
203 | ||
165125e1 JA |
204 | static elevator_t *elevator_alloc(struct request_queue *q, |
205 | struct elevator_type *e) | |
3d1ab40f | 206 | { |
9817064b JA |
207 | elevator_t *eq; |
208 | int i; | |
209 | ||
94f6030c | 210 | eq = kmalloc_node(sizeof(elevator_t), GFP_KERNEL | __GFP_ZERO, q->node); |
9817064b JA |
211 | if (unlikely(!eq)) |
212 | goto err; | |
213 | ||
9817064b JA |
214 | eq->ops = &e->ops; |
215 | eq->elevator_type = e; | |
f9cb074b | 216 | kobject_init(&eq->kobj, &elv_ktype); |
9817064b JA |
217 | mutex_init(&eq->sysfs_lock); |
218 | ||
b5deef90 JA |
219 | eq->hash = kmalloc_node(sizeof(struct hlist_head) * ELV_HASH_ENTRIES, |
220 | GFP_KERNEL, q->node); | |
9817064b JA |
221 | if (!eq->hash) |
222 | goto err; | |
223 | ||
224 | for (i = 0; i < ELV_HASH_ENTRIES; i++) | |
225 | INIT_HLIST_HEAD(&eq->hash[i]); | |
226 | ||
3d1ab40f | 227 | return eq; |
9817064b JA |
228 | err: |
229 | kfree(eq); | |
230 | elevator_put(e); | |
231 | return NULL; | |
3d1ab40f AV |
232 | } |
233 | ||
234 | static void elevator_release(struct kobject *kobj) | |
235 | { | |
236 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
9817064b | 237 | |
3d1ab40f | 238 | elevator_put(e->elevator_type); |
9817064b | 239 | kfree(e->hash); |
3d1ab40f AV |
240 | kfree(e); |
241 | } | |
242 | ||
165125e1 | 243 | int elevator_init(struct request_queue *q, char *name) |
1da177e4 LT |
244 | { |
245 | struct elevator_type *e = NULL; | |
246 | struct elevator_queue *eq; | |
247 | int ret = 0; | |
bc1c1169 | 248 | void *data; |
1da177e4 | 249 | |
cb98fc8b TH |
250 | INIT_LIST_HEAD(&q->queue_head); |
251 | q->last_merge = NULL; | |
252 | q->end_sector = 0; | |
253 | q->boundary_rq = NULL; | |
cb98fc8b | 254 | |
4eb166d9 JA |
255 | if (name) { |
256 | e = elevator_get(name); | |
257 | if (!e) | |
258 | return -EINVAL; | |
259 | } | |
1da177e4 | 260 | |
4eb166d9 JA |
261 | if (!e && *chosen_elevator) { |
262 | e = elevator_get(chosen_elevator); | |
263 | if (!e) | |
264 | printk(KERN_ERR "I/O scheduler %s not found\n", | |
265 | chosen_elevator); | |
266 | } | |
248d5ca5 | 267 | |
4eb166d9 JA |
268 | if (!e) { |
269 | e = elevator_get(CONFIG_DEFAULT_IOSCHED); | |
270 | if (!e) { | |
271 | printk(KERN_ERR | |
272 | "Default I/O scheduler not found. " \ | |
273 | "Using noop.\n"); | |
274 | e = elevator_get("noop"); | |
275 | } | |
5f003976 ND |
276 | } |
277 | ||
b5deef90 | 278 | eq = elevator_alloc(q, e); |
3d1ab40f | 279 | if (!eq) |
1da177e4 | 280 | return -ENOMEM; |
1da177e4 | 281 | |
bc1c1169 JA |
282 | data = elevator_init_queue(q, eq); |
283 | if (!data) { | |
3d1ab40f | 284 | kobject_put(&eq->kobj); |
bc1c1169 JA |
285 | return -ENOMEM; |
286 | } | |
1da177e4 | 287 | |
bc1c1169 | 288 | elevator_attach(q, eq, data); |
1da177e4 LT |
289 | return ret; |
290 | } | |
2e662b65 JA |
291 | EXPORT_SYMBOL(elevator_init); |
292 | ||
1da177e4 LT |
293 | void elevator_exit(elevator_t *e) |
294 | { | |
3d1ab40f | 295 | mutex_lock(&e->sysfs_lock); |
1da177e4 LT |
296 | if (e->ops->elevator_exit_fn) |
297 | e->ops->elevator_exit_fn(e); | |
3d1ab40f AV |
298 | e->ops = NULL; |
299 | mutex_unlock(&e->sysfs_lock); | |
1da177e4 | 300 | |
3d1ab40f | 301 | kobject_put(&e->kobj); |
1da177e4 | 302 | } |
2e662b65 JA |
303 | EXPORT_SYMBOL(elevator_exit); |
304 | ||
165125e1 | 305 | static void elv_activate_rq(struct request_queue *q, struct request *rq) |
cad97516 JA |
306 | { |
307 | elevator_t *e = q->elevator; | |
308 | ||
309 | if (e->ops->elevator_activate_req_fn) | |
310 | e->ops->elevator_activate_req_fn(q, rq); | |
311 | } | |
312 | ||
165125e1 | 313 | static void elv_deactivate_rq(struct request_queue *q, struct request *rq) |
cad97516 JA |
314 | { |
315 | elevator_t *e = q->elevator; | |
316 | ||
317 | if (e->ops->elevator_deactivate_req_fn) | |
318 | e->ops->elevator_deactivate_req_fn(q, rq); | |
319 | } | |
320 | ||
9817064b JA |
321 | static inline void __elv_rqhash_del(struct request *rq) |
322 | { | |
323 | hlist_del_init(&rq->hash); | |
324 | } | |
325 | ||
165125e1 | 326 | static void elv_rqhash_del(struct request_queue *q, struct request *rq) |
9817064b JA |
327 | { |
328 | if (ELV_ON_HASH(rq)) | |
329 | __elv_rqhash_del(rq); | |
330 | } | |
331 | ||
165125e1 | 332 | static void elv_rqhash_add(struct request_queue *q, struct request *rq) |
9817064b JA |
333 | { |
334 | elevator_t *e = q->elevator; | |
335 | ||
336 | BUG_ON(ELV_ON_HASH(rq)); | |
337 | hlist_add_head(&rq->hash, &e->hash[ELV_HASH_FN(rq_hash_key(rq))]); | |
338 | } | |
339 | ||
165125e1 | 340 | static void elv_rqhash_reposition(struct request_queue *q, struct request *rq) |
9817064b JA |
341 | { |
342 | __elv_rqhash_del(rq); | |
343 | elv_rqhash_add(q, rq); | |
344 | } | |
345 | ||
165125e1 | 346 | static struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) |
9817064b JA |
347 | { |
348 | elevator_t *e = q->elevator; | |
349 | struct hlist_head *hash_list = &e->hash[ELV_HASH_FN(offset)]; | |
350 | struct hlist_node *entry, *next; | |
351 | struct request *rq; | |
352 | ||
353 | hlist_for_each_entry_safe(rq, entry, next, hash_list, hash) { | |
354 | BUG_ON(!ELV_ON_HASH(rq)); | |
355 | ||
356 | if (unlikely(!rq_mergeable(rq))) { | |
357 | __elv_rqhash_del(rq); | |
358 | continue; | |
359 | } | |
360 | ||
361 | if (rq_hash_key(rq) == offset) | |
362 | return rq; | |
363 | } | |
364 | ||
365 | return NULL; | |
366 | } | |
367 | ||
2e662b65 JA |
368 | /* |
369 | * RB-tree support functions for inserting/lookup/removal of requests | |
370 | * in a sorted RB tree. | |
371 | */ | |
372 | struct request *elv_rb_add(struct rb_root *root, struct request *rq) | |
373 | { | |
374 | struct rb_node **p = &root->rb_node; | |
375 | struct rb_node *parent = NULL; | |
376 | struct request *__rq; | |
377 | ||
378 | while (*p) { | |
379 | parent = *p; | |
380 | __rq = rb_entry(parent, struct request, rb_node); | |
381 | ||
382 | if (rq->sector < __rq->sector) | |
383 | p = &(*p)->rb_left; | |
384 | else if (rq->sector > __rq->sector) | |
385 | p = &(*p)->rb_right; | |
386 | else | |
387 | return __rq; | |
388 | } | |
389 | ||
390 | rb_link_node(&rq->rb_node, parent, p); | |
391 | rb_insert_color(&rq->rb_node, root); | |
392 | return NULL; | |
393 | } | |
2e662b65 JA |
394 | EXPORT_SYMBOL(elv_rb_add); |
395 | ||
396 | void elv_rb_del(struct rb_root *root, struct request *rq) | |
397 | { | |
398 | BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); | |
399 | rb_erase(&rq->rb_node, root); | |
400 | RB_CLEAR_NODE(&rq->rb_node); | |
401 | } | |
2e662b65 JA |
402 | EXPORT_SYMBOL(elv_rb_del); |
403 | ||
404 | struct request *elv_rb_find(struct rb_root *root, sector_t sector) | |
405 | { | |
406 | struct rb_node *n = root->rb_node; | |
407 | struct request *rq; | |
408 | ||
409 | while (n) { | |
410 | rq = rb_entry(n, struct request, rb_node); | |
411 | ||
412 | if (sector < rq->sector) | |
413 | n = n->rb_left; | |
414 | else if (sector > rq->sector) | |
415 | n = n->rb_right; | |
416 | else | |
417 | return rq; | |
418 | } | |
419 | ||
420 | return NULL; | |
421 | } | |
2e662b65 JA |
422 | EXPORT_SYMBOL(elv_rb_find); |
423 | ||
8922e16c TH |
424 | /* |
425 | * Insert rq into dispatch queue of q. Queue lock must be held on | |
dbe7f76d | 426 | * entry. rq is sort instead into the dispatch queue. To be used by |
2e662b65 | 427 | * specific elevators. |
8922e16c | 428 | */ |
165125e1 | 429 | void elv_dispatch_sort(struct request_queue *q, struct request *rq) |
8922e16c TH |
430 | { |
431 | sector_t boundary; | |
8922e16c | 432 | struct list_head *entry; |
4eb166d9 | 433 | int stop_flags; |
8922e16c | 434 | |
06b86245 TH |
435 | if (q->last_merge == rq) |
436 | q->last_merge = NULL; | |
9817064b JA |
437 | |
438 | elv_rqhash_del(q, rq); | |
439 | ||
15853af9 | 440 | q->nr_sorted--; |
06b86245 | 441 | |
1b47f531 | 442 | boundary = q->end_sector; |
4eb166d9 | 443 | stop_flags = REQ_SOFTBARRIER | REQ_HARDBARRIER | REQ_STARTED; |
8922e16c TH |
444 | list_for_each_prev(entry, &q->queue_head) { |
445 | struct request *pos = list_entry_rq(entry); | |
446 | ||
e17fc0a1 DW |
447 | if (blk_discard_rq(rq) != blk_discard_rq(pos)) |
448 | break; | |
783660b2 JA |
449 | if (rq_data_dir(rq) != rq_data_dir(pos)) |
450 | break; | |
4eb166d9 | 451 | if (pos->cmd_flags & stop_flags) |
8922e16c TH |
452 | break; |
453 | if (rq->sector >= boundary) { | |
454 | if (pos->sector < boundary) | |
455 | continue; | |
456 | } else { | |
457 | if (pos->sector >= boundary) | |
458 | break; | |
459 | } | |
460 | if (rq->sector >= pos->sector) | |
461 | break; | |
462 | } | |
463 | ||
464 | list_add(&rq->queuelist, entry); | |
465 | } | |
2e662b65 JA |
466 | EXPORT_SYMBOL(elv_dispatch_sort); |
467 | ||
9817064b | 468 | /* |
2e662b65 JA |
469 | * Insert rq into dispatch queue of q. Queue lock must be held on |
470 | * entry. rq is added to the back of the dispatch queue. To be used by | |
471 | * specific elevators. | |
9817064b JA |
472 | */ |
473 | void elv_dispatch_add_tail(struct request_queue *q, struct request *rq) | |
474 | { | |
475 | if (q->last_merge == rq) | |
476 | q->last_merge = NULL; | |
477 | ||
478 | elv_rqhash_del(q, rq); | |
479 | ||
480 | q->nr_sorted--; | |
481 | ||
482 | q->end_sector = rq_end_sector(rq); | |
483 | q->boundary_rq = rq; | |
484 | list_add_tail(&rq->queuelist, &q->queue_head); | |
485 | } | |
2e662b65 JA |
486 | EXPORT_SYMBOL(elv_dispatch_add_tail); |
487 | ||
165125e1 | 488 | int elv_merge(struct request_queue *q, struct request **req, struct bio *bio) |
1da177e4 LT |
489 | { |
490 | elevator_t *e = q->elevator; | |
9817064b | 491 | struct request *__rq; |
06b86245 TH |
492 | int ret; |
493 | ||
9817064b JA |
494 | /* |
495 | * First try one-hit cache. | |
496 | */ | |
06b86245 TH |
497 | if (q->last_merge) { |
498 | ret = elv_try_merge(q->last_merge, bio); | |
499 | if (ret != ELEVATOR_NO_MERGE) { | |
500 | *req = q->last_merge; | |
501 | return ret; | |
502 | } | |
503 | } | |
1da177e4 | 504 | |
ac9fafa1 AB |
505 | if (blk_queue_nomerges(q)) |
506 | return ELEVATOR_NO_MERGE; | |
507 | ||
9817064b JA |
508 | /* |
509 | * See if our hash lookup can find a potential backmerge. | |
510 | */ | |
511 | __rq = elv_rqhash_find(q, bio->bi_sector); | |
512 | if (__rq && elv_rq_merge_ok(__rq, bio)) { | |
513 | *req = __rq; | |
514 | return ELEVATOR_BACK_MERGE; | |
515 | } | |
516 | ||
1da177e4 LT |
517 | if (e->ops->elevator_merge_fn) |
518 | return e->ops->elevator_merge_fn(q, req, bio); | |
519 | ||
520 | return ELEVATOR_NO_MERGE; | |
521 | } | |
522 | ||
165125e1 | 523 | void elv_merged_request(struct request_queue *q, struct request *rq, int type) |
1da177e4 LT |
524 | { |
525 | elevator_t *e = q->elevator; | |
526 | ||
527 | if (e->ops->elevator_merged_fn) | |
2e662b65 | 528 | e->ops->elevator_merged_fn(q, rq, type); |
06b86245 | 529 | |
2e662b65 JA |
530 | if (type == ELEVATOR_BACK_MERGE) |
531 | elv_rqhash_reposition(q, rq); | |
9817064b | 532 | |
06b86245 | 533 | q->last_merge = rq; |
1da177e4 LT |
534 | } |
535 | ||
165125e1 | 536 | void elv_merge_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
537 | struct request *next) |
538 | { | |
539 | elevator_t *e = q->elevator; | |
540 | ||
1da177e4 LT |
541 | if (e->ops->elevator_merge_req_fn) |
542 | e->ops->elevator_merge_req_fn(q, rq, next); | |
06b86245 | 543 | |
9817064b JA |
544 | elv_rqhash_reposition(q, rq); |
545 | elv_rqhash_del(q, next); | |
546 | ||
547 | q->nr_sorted--; | |
06b86245 | 548 | q->last_merge = rq; |
1da177e4 LT |
549 | } |
550 | ||
165125e1 | 551 | void elv_requeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 552 | { |
1da177e4 LT |
553 | /* |
554 | * it already went through dequeue, we need to decrement the | |
555 | * in_flight count again | |
556 | */ | |
8922e16c | 557 | if (blk_account_rq(rq)) { |
1da177e4 | 558 | q->in_flight--; |
cad97516 JA |
559 | if (blk_sorted_rq(rq)) |
560 | elv_deactivate_rq(q, rq); | |
8922e16c | 561 | } |
1da177e4 | 562 | |
4aff5e23 | 563 | rq->cmd_flags &= ~REQ_STARTED; |
1da177e4 | 564 | |
30e9656c | 565 | elv_insert(q, rq, ELEVATOR_INSERT_REQUEUE); |
1da177e4 LT |
566 | } |
567 | ||
165125e1 | 568 | static void elv_drain_elevator(struct request_queue *q) |
15853af9 TH |
569 | { |
570 | static int printed; | |
571 | while (q->elevator->ops->elevator_dispatch_fn(q, 1)) | |
572 | ; | |
573 | if (q->nr_sorted == 0) | |
574 | return; | |
575 | if (printed++ < 10) { | |
576 | printk(KERN_ERR "%s: forced dispatching is broken " | |
577 | "(nr_sorted=%u), please report this\n", | |
578 | q->elevator->elevator_type->elevator_name, q->nr_sorted); | |
579 | } | |
580 | } | |
581 | ||
165125e1 | 582 | void elv_insert(struct request_queue *q, struct request *rq, int where) |
1da177e4 | 583 | { |
797e7dbb TH |
584 | struct list_head *pos; |
585 | unsigned ordseq; | |
dac07ec1 | 586 | int unplug_it = 1; |
797e7dbb | 587 | |
2056a782 JA |
588 | blk_add_trace_rq(q, rq, BLK_TA_INSERT); |
589 | ||
1da177e4 LT |
590 | rq->q = q; |
591 | ||
8922e16c TH |
592 | switch (where) { |
593 | case ELEVATOR_INSERT_FRONT: | |
4aff5e23 | 594 | rq->cmd_flags |= REQ_SOFTBARRIER; |
8922e16c TH |
595 | |
596 | list_add(&rq->queuelist, &q->queue_head); | |
597 | break; | |
598 | ||
599 | case ELEVATOR_INSERT_BACK: | |
4aff5e23 | 600 | rq->cmd_flags |= REQ_SOFTBARRIER; |
15853af9 | 601 | elv_drain_elevator(q); |
8922e16c TH |
602 | list_add_tail(&rq->queuelist, &q->queue_head); |
603 | /* | |
604 | * We kick the queue here for the following reasons. | |
605 | * - The elevator might have returned NULL previously | |
606 | * to delay requests and returned them now. As the | |
607 | * queue wasn't empty before this request, ll_rw_blk | |
608 | * won't run the queue on return, resulting in hang. | |
609 | * - Usually, back inserted requests won't be merged | |
610 | * with anything. There's no point in delaying queue | |
611 | * processing. | |
612 | */ | |
613 | blk_remove_plug(q); | |
614 | q->request_fn(q); | |
615 | break; | |
616 | ||
617 | case ELEVATOR_INSERT_SORT: | |
e17fc0a1 | 618 | BUG_ON(!blk_fs_request(rq) && !blk_discard_rq(rq)); |
4aff5e23 | 619 | rq->cmd_flags |= REQ_SORTED; |
15853af9 | 620 | q->nr_sorted++; |
9817064b JA |
621 | if (rq_mergeable(rq)) { |
622 | elv_rqhash_add(q, rq); | |
623 | if (!q->last_merge) | |
624 | q->last_merge = rq; | |
625 | } | |
626 | ||
ca23509f TH |
627 | /* |
628 | * Some ioscheds (cfq) run q->request_fn directly, so | |
629 | * rq cannot be accessed after calling | |
630 | * elevator_add_req_fn. | |
631 | */ | |
632 | q->elevator->ops->elevator_add_req_fn(q, rq); | |
8922e16c TH |
633 | break; |
634 | ||
797e7dbb TH |
635 | case ELEVATOR_INSERT_REQUEUE: |
636 | /* | |
637 | * If ordered flush isn't in progress, we do front | |
638 | * insertion; otherwise, requests should be requeued | |
639 | * in ordseq order. | |
640 | */ | |
4aff5e23 | 641 | rq->cmd_flags |= REQ_SOFTBARRIER; |
797e7dbb | 642 | |
95543179 LV |
643 | /* |
644 | * Most requeues happen because of a busy condition, | |
645 | * don't force unplug of the queue for that case. | |
646 | */ | |
647 | unplug_it = 0; | |
648 | ||
797e7dbb TH |
649 | if (q->ordseq == 0) { |
650 | list_add(&rq->queuelist, &q->queue_head); | |
651 | break; | |
652 | } | |
653 | ||
654 | ordseq = blk_ordered_req_seq(rq); | |
655 | ||
656 | list_for_each(pos, &q->queue_head) { | |
657 | struct request *pos_rq = list_entry_rq(pos); | |
658 | if (ordseq <= blk_ordered_req_seq(pos_rq)) | |
659 | break; | |
660 | } | |
661 | ||
662 | list_add_tail(&rq->queuelist, pos); | |
663 | break; | |
664 | ||
8922e16c TH |
665 | default: |
666 | printk(KERN_ERR "%s: bad insertion point %d\n", | |
24c03d47 | 667 | __func__, where); |
8922e16c TH |
668 | BUG(); |
669 | } | |
670 | ||
dac07ec1 | 671 | if (unplug_it && blk_queue_plugged(q)) { |
8922e16c TH |
672 | int nrq = q->rq.count[READ] + q->rq.count[WRITE] |
673 | - q->in_flight; | |
674 | ||
675 | if (nrq >= q->unplug_thresh) | |
676 | __generic_unplug_device(q); | |
677 | } | |
1da177e4 LT |
678 | } |
679 | ||
165125e1 | 680 | void __elv_add_request(struct request_queue *q, struct request *rq, int where, |
30e9656c TH |
681 | int plug) |
682 | { | |
683 | if (q->ordcolor) | |
4aff5e23 | 684 | rq->cmd_flags |= REQ_ORDERED_COLOR; |
30e9656c | 685 | |
4aff5e23 | 686 | if (rq->cmd_flags & (REQ_SOFTBARRIER | REQ_HARDBARRIER)) { |
30e9656c TH |
687 | /* |
688 | * toggle ordered color | |
689 | */ | |
690 | if (blk_barrier_rq(rq)) | |
691 | q->ordcolor ^= 1; | |
692 | ||
693 | /* | |
694 | * barriers implicitly indicate back insertion | |
695 | */ | |
696 | if (where == ELEVATOR_INSERT_SORT) | |
697 | where = ELEVATOR_INSERT_BACK; | |
698 | ||
699 | /* | |
700 | * this request is scheduling boundary, update | |
701 | * end_sector | |
702 | */ | |
e17fc0a1 | 703 | if (blk_fs_request(rq) || blk_discard_rq(rq)) { |
30e9656c TH |
704 | q->end_sector = rq_end_sector(rq); |
705 | q->boundary_rq = rq; | |
706 | } | |
4eb166d9 JA |
707 | } else if (!(rq->cmd_flags & REQ_ELVPRIV) && |
708 | where == ELEVATOR_INSERT_SORT) | |
30e9656c TH |
709 | where = ELEVATOR_INSERT_BACK; |
710 | ||
711 | if (plug) | |
712 | blk_plug_device(q); | |
713 | ||
714 | elv_insert(q, rq, where); | |
715 | } | |
2e662b65 JA |
716 | EXPORT_SYMBOL(__elv_add_request); |
717 | ||
165125e1 | 718 | void elv_add_request(struct request_queue *q, struct request *rq, int where, |
1da177e4 LT |
719 | int plug) |
720 | { | |
721 | unsigned long flags; | |
722 | ||
723 | spin_lock_irqsave(q->queue_lock, flags); | |
724 | __elv_add_request(q, rq, where, plug); | |
725 | spin_unlock_irqrestore(q->queue_lock, flags); | |
726 | } | |
2e662b65 JA |
727 | EXPORT_SYMBOL(elv_add_request); |
728 | ||
165125e1 | 729 | static inline struct request *__elv_next_request(struct request_queue *q) |
1da177e4 | 730 | { |
8922e16c TH |
731 | struct request *rq; |
732 | ||
797e7dbb TH |
733 | while (1) { |
734 | while (!list_empty(&q->queue_head)) { | |
735 | rq = list_entry_rq(q->queue_head.next); | |
736 | if (blk_do_ordered(q, &rq)) | |
737 | return rq; | |
738 | } | |
1da177e4 | 739 | |
797e7dbb TH |
740 | if (!q->elevator->ops->elevator_dispatch_fn(q, 0)) |
741 | return NULL; | |
1da177e4 | 742 | } |
1da177e4 LT |
743 | } |
744 | ||
165125e1 | 745 | struct request *elv_next_request(struct request_queue *q) |
1da177e4 LT |
746 | { |
747 | struct request *rq; | |
748 | int ret; | |
749 | ||
750 | while ((rq = __elv_next_request(q)) != NULL) { | |
bf2de6f5 JA |
751 | /* |
752 | * Kill the empty barrier place holder, the driver must | |
753 | * not ever see it. | |
754 | */ | |
755 | if (blk_empty_barrier(rq)) { | |
756 | end_queued_request(rq, 1); | |
757 | continue; | |
758 | } | |
4aff5e23 | 759 | if (!(rq->cmd_flags & REQ_STARTED)) { |
8922e16c TH |
760 | /* |
761 | * This is the first time the device driver | |
762 | * sees this request (possibly after | |
763 | * requeueing). Notify IO scheduler. | |
764 | */ | |
cad97516 JA |
765 | if (blk_sorted_rq(rq)) |
766 | elv_activate_rq(q, rq); | |
1da177e4 | 767 | |
8922e16c TH |
768 | /* |
769 | * just mark as started even if we don't start | |
770 | * it, a request that has been delayed should | |
771 | * not be passed by new incoming requests | |
772 | */ | |
4aff5e23 | 773 | rq->cmd_flags |= REQ_STARTED; |
2056a782 | 774 | blk_add_trace_rq(q, rq, BLK_TA_ISSUE); |
8922e16c | 775 | } |
1da177e4 | 776 | |
8922e16c | 777 | if (!q->boundary_rq || q->boundary_rq == rq) { |
1b47f531 | 778 | q->end_sector = rq_end_sector(rq); |
8922e16c TH |
779 | q->boundary_rq = NULL; |
780 | } | |
1da177e4 | 781 | |
fa0ccd83 JB |
782 | if (rq->cmd_flags & REQ_DONTPREP) |
783 | break; | |
784 | ||
785 | if (q->dma_drain_size && rq->data_len) { | |
786 | /* | |
787 | * make sure space for the drain appears we | |
788 | * know we can do this because max_hw_segments | |
789 | * has been adjusted to be one fewer than the | |
790 | * device can handle | |
791 | */ | |
792 | rq->nr_phys_segments++; | |
fa0ccd83 JB |
793 | } |
794 | ||
795 | if (!q->prep_rq_fn) | |
1da177e4 LT |
796 | break; |
797 | ||
798 | ret = q->prep_rq_fn(q, rq); | |
799 | if (ret == BLKPREP_OK) { | |
800 | break; | |
801 | } else if (ret == BLKPREP_DEFER) { | |
2e759cd4 TH |
802 | /* |
803 | * the request may have been (partially) prepped. | |
804 | * we need to keep this request in the front to | |
8922e16c TH |
805 | * avoid resource deadlock. REQ_STARTED will |
806 | * prevent other fs requests from passing this one. | |
2e759cd4 | 807 | */ |
fa0ccd83 JB |
808 | if (q->dma_drain_size && rq->data_len && |
809 | !(rq->cmd_flags & REQ_DONTPREP)) { | |
810 | /* | |
811 | * remove the space for the drain we added | |
812 | * so that we don't add it again | |
813 | */ | |
814 | --rq->nr_phys_segments; | |
fa0ccd83 JB |
815 | } |
816 | ||
1da177e4 LT |
817 | rq = NULL; |
818 | break; | |
819 | } else if (ret == BLKPREP_KILL) { | |
4aff5e23 | 820 | rq->cmd_flags |= REQ_QUIET; |
a0cd1285 | 821 | end_queued_request(rq, 0); |
1da177e4 | 822 | } else { |
24c03d47 | 823 | printk(KERN_ERR "%s: bad return=%d\n", __func__, ret); |
1da177e4 LT |
824 | break; |
825 | } | |
826 | } | |
827 | ||
828 | return rq; | |
829 | } | |
2e662b65 JA |
830 | EXPORT_SYMBOL(elv_next_request); |
831 | ||
165125e1 | 832 | void elv_dequeue_request(struct request_queue *q, struct request *rq) |
1da177e4 | 833 | { |
8922e16c | 834 | BUG_ON(list_empty(&rq->queuelist)); |
9817064b | 835 | BUG_ON(ELV_ON_HASH(rq)); |
8922e16c TH |
836 | |
837 | list_del_init(&rq->queuelist); | |
1da177e4 LT |
838 | |
839 | /* | |
840 | * the time frame between a request being removed from the lists | |
841 | * and to it is freed is accounted as io that is in progress at | |
8922e16c | 842 | * the driver side. |
1da177e4 LT |
843 | */ |
844 | if (blk_account_rq(rq)) | |
845 | q->in_flight++; | |
1da177e4 | 846 | } |
2e662b65 JA |
847 | EXPORT_SYMBOL(elv_dequeue_request); |
848 | ||
165125e1 | 849 | int elv_queue_empty(struct request_queue *q) |
1da177e4 LT |
850 | { |
851 | elevator_t *e = q->elevator; | |
852 | ||
8922e16c TH |
853 | if (!list_empty(&q->queue_head)) |
854 | return 0; | |
855 | ||
1da177e4 LT |
856 | if (e->ops->elevator_queue_empty_fn) |
857 | return e->ops->elevator_queue_empty_fn(q); | |
858 | ||
8922e16c | 859 | return 1; |
1da177e4 | 860 | } |
2e662b65 JA |
861 | EXPORT_SYMBOL(elv_queue_empty); |
862 | ||
165125e1 | 863 | struct request *elv_latter_request(struct request_queue *q, struct request *rq) |
1da177e4 | 864 | { |
1da177e4 LT |
865 | elevator_t *e = q->elevator; |
866 | ||
867 | if (e->ops->elevator_latter_req_fn) | |
868 | return e->ops->elevator_latter_req_fn(q, rq); | |
1da177e4 LT |
869 | return NULL; |
870 | } | |
871 | ||
165125e1 | 872 | struct request *elv_former_request(struct request_queue *q, struct request *rq) |
1da177e4 | 873 | { |
1da177e4 LT |
874 | elevator_t *e = q->elevator; |
875 | ||
876 | if (e->ops->elevator_former_req_fn) | |
877 | return e->ops->elevator_former_req_fn(q, rq); | |
1da177e4 LT |
878 | return NULL; |
879 | } | |
880 | ||
165125e1 | 881 | int elv_set_request(struct request_queue *q, struct request *rq, gfp_t gfp_mask) |
1da177e4 LT |
882 | { |
883 | elevator_t *e = q->elevator; | |
884 | ||
885 | if (e->ops->elevator_set_req_fn) | |
cb78b285 | 886 | return e->ops->elevator_set_req_fn(q, rq, gfp_mask); |
1da177e4 LT |
887 | |
888 | rq->elevator_private = NULL; | |
889 | return 0; | |
890 | } | |
891 | ||
165125e1 | 892 | void elv_put_request(struct request_queue *q, struct request *rq) |
1da177e4 LT |
893 | { |
894 | elevator_t *e = q->elevator; | |
895 | ||
896 | if (e->ops->elevator_put_req_fn) | |
bb37b94c | 897 | e->ops->elevator_put_req_fn(rq); |
1da177e4 LT |
898 | } |
899 | ||
165125e1 | 900 | int elv_may_queue(struct request_queue *q, int rw) |
1da177e4 LT |
901 | { |
902 | elevator_t *e = q->elevator; | |
903 | ||
904 | if (e->ops->elevator_may_queue_fn) | |
cb78b285 | 905 | return e->ops->elevator_may_queue_fn(q, rw); |
1da177e4 LT |
906 | |
907 | return ELV_MQUEUE_MAY; | |
908 | } | |
909 | ||
165125e1 | 910 | void elv_completed_request(struct request_queue *q, struct request *rq) |
1da177e4 LT |
911 | { |
912 | elevator_t *e = q->elevator; | |
913 | ||
914 | /* | |
915 | * request is released from the driver, io must be done | |
916 | */ | |
8922e16c | 917 | if (blk_account_rq(rq)) { |
1da177e4 | 918 | q->in_flight--; |
1bc691d3 TH |
919 | if (blk_sorted_rq(rq) && e->ops->elevator_completed_req_fn) |
920 | e->ops->elevator_completed_req_fn(q, rq); | |
921 | } | |
797e7dbb | 922 | |
1bc691d3 TH |
923 | /* |
924 | * Check if the queue is waiting for fs requests to be | |
925 | * drained for flush sequence. | |
926 | */ | |
927 | if (unlikely(q->ordseq)) { | |
928 | struct request *first_rq = list_entry_rq(q->queue_head.next); | |
929 | if (q->in_flight == 0 && | |
797e7dbb TH |
930 | blk_ordered_cur_seq(q) == QUEUE_ORDSEQ_DRAIN && |
931 | blk_ordered_req_seq(first_rq) > QUEUE_ORDSEQ_DRAIN) { | |
932 | blk_ordered_complete_seq(q, QUEUE_ORDSEQ_DRAIN, 0); | |
933 | q->request_fn(q); | |
934 | } | |
8922e16c | 935 | } |
1da177e4 LT |
936 | } |
937 | ||
3d1ab40f AV |
938 | #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) |
939 | ||
940 | static ssize_t | |
941 | elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1da177e4 | 942 | { |
3d1ab40f AV |
943 | elevator_t *e = container_of(kobj, elevator_t, kobj); |
944 | struct elv_fs_entry *entry = to_elv(attr); | |
945 | ssize_t error; | |
946 | ||
947 | if (!entry->show) | |
948 | return -EIO; | |
949 | ||
950 | mutex_lock(&e->sysfs_lock); | |
951 | error = e->ops ? entry->show(e, page) : -ENOENT; | |
952 | mutex_unlock(&e->sysfs_lock); | |
953 | return error; | |
954 | } | |
1da177e4 | 955 | |
3d1ab40f AV |
956 | static ssize_t |
957 | elv_attr_store(struct kobject *kobj, struct attribute *attr, | |
958 | const char *page, size_t length) | |
959 | { | |
960 | elevator_t *e = container_of(kobj, elevator_t, kobj); | |
961 | struct elv_fs_entry *entry = to_elv(attr); | |
962 | ssize_t error; | |
1da177e4 | 963 | |
3d1ab40f AV |
964 | if (!entry->store) |
965 | return -EIO; | |
1da177e4 | 966 | |
3d1ab40f AV |
967 | mutex_lock(&e->sysfs_lock); |
968 | error = e->ops ? entry->store(e, page, length) : -ENOENT; | |
969 | mutex_unlock(&e->sysfs_lock); | |
970 | return error; | |
971 | } | |
972 | ||
973 | static struct sysfs_ops elv_sysfs_ops = { | |
974 | .show = elv_attr_show, | |
975 | .store = elv_attr_store, | |
976 | }; | |
977 | ||
978 | static struct kobj_type elv_ktype = { | |
979 | .sysfs_ops = &elv_sysfs_ops, | |
980 | .release = elevator_release, | |
981 | }; | |
982 | ||
983 | int elv_register_queue(struct request_queue *q) | |
984 | { | |
985 | elevator_t *e = q->elevator; | |
986 | int error; | |
987 | ||
b2d6db58 | 988 | error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); |
3d1ab40f | 989 | if (!error) { |
e572ec7e | 990 | struct elv_fs_entry *attr = e->elevator_type->elevator_attrs; |
3d1ab40f | 991 | if (attr) { |
e572ec7e AV |
992 | while (attr->attr.name) { |
993 | if (sysfs_create_file(&e->kobj, &attr->attr)) | |
3d1ab40f | 994 | break; |
e572ec7e | 995 | attr++; |
3d1ab40f AV |
996 | } |
997 | } | |
998 | kobject_uevent(&e->kobj, KOBJ_ADD); | |
999 | } | |
1000 | return error; | |
1da177e4 LT |
1001 | } |
1002 | ||
bc1c1169 JA |
1003 | static void __elv_unregister_queue(elevator_t *e) |
1004 | { | |
1005 | kobject_uevent(&e->kobj, KOBJ_REMOVE); | |
1006 | kobject_del(&e->kobj); | |
1007 | } | |
1008 | ||
1da177e4 LT |
1009 | void elv_unregister_queue(struct request_queue *q) |
1010 | { | |
bc1c1169 JA |
1011 | if (q) |
1012 | __elv_unregister_queue(q->elevator); | |
1da177e4 LT |
1013 | } |
1014 | ||
2fdd82bd | 1015 | void elv_register(struct elevator_type *e) |
1da177e4 | 1016 | { |
1ffb96c5 | 1017 | char *def = ""; |
2a12dcd7 JA |
1018 | |
1019 | spin_lock(&elv_list_lock); | |
ce524497 | 1020 | BUG_ON(elevator_find(e->elevator_name)); |
1da177e4 | 1021 | list_add_tail(&e->list, &elv_list); |
2a12dcd7 | 1022 | spin_unlock(&elv_list_lock); |
1da177e4 | 1023 | |
5f003976 ND |
1024 | if (!strcmp(e->elevator_name, chosen_elevator) || |
1025 | (!*chosen_elevator && | |
1026 | !strcmp(e->elevator_name, CONFIG_DEFAULT_IOSCHED))) | |
1ffb96c5 TV |
1027 | def = " (default)"; |
1028 | ||
4eb166d9 JA |
1029 | printk(KERN_INFO "io scheduler %s registered%s\n", e->elevator_name, |
1030 | def); | |
1da177e4 LT |
1031 | } |
1032 | EXPORT_SYMBOL_GPL(elv_register); | |
1033 | ||
1034 | void elv_unregister(struct elevator_type *e) | |
1035 | { | |
83521d3e CH |
1036 | struct task_struct *g, *p; |
1037 | ||
1038 | /* | |
1039 | * Iterate every thread in the process to remove the io contexts. | |
1040 | */ | |
e17a9489 AV |
1041 | if (e->ops.trim) { |
1042 | read_lock(&tasklist_lock); | |
1043 | do_each_thread(g, p) { | |
1044 | task_lock(p); | |
2d8f6131 ON |
1045 | if (p->io_context) |
1046 | e->ops.trim(p->io_context); | |
e17a9489 AV |
1047 | task_unlock(p); |
1048 | } while_each_thread(g, p); | |
1049 | read_unlock(&tasklist_lock); | |
1050 | } | |
83521d3e | 1051 | |
2a12dcd7 | 1052 | spin_lock(&elv_list_lock); |
1da177e4 | 1053 | list_del_init(&e->list); |
2a12dcd7 | 1054 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
1055 | } |
1056 | EXPORT_SYMBOL_GPL(elv_unregister); | |
1057 | ||
1058 | /* | |
1059 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
1060 | * we don't free the old io scheduler, before we have allocated what we | |
1061 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 1062 | * one, if the new one fails init for some reason. |
1da177e4 | 1063 | */ |
165125e1 | 1064 | static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) |
1da177e4 | 1065 | { |
cb98fc8b | 1066 | elevator_t *old_elevator, *e; |
bc1c1169 | 1067 | void *data; |
1da177e4 | 1068 | |
cb98fc8b TH |
1069 | /* |
1070 | * Allocate new elevator | |
1071 | */ | |
b5deef90 | 1072 | e = elevator_alloc(q, new_e); |
1da177e4 | 1073 | if (!e) |
3d1ab40f | 1074 | return 0; |
1da177e4 | 1075 | |
bc1c1169 JA |
1076 | data = elevator_init_queue(q, e); |
1077 | if (!data) { | |
1078 | kobject_put(&e->kobj); | |
1079 | return 0; | |
1080 | } | |
1081 | ||
1da177e4 | 1082 | /* |
cb98fc8b | 1083 | * Turn on BYPASS and drain all requests w/ elevator private data |
1da177e4 | 1084 | */ |
cb98fc8b TH |
1085 | spin_lock_irq(q->queue_lock); |
1086 | ||
75ad23bc | 1087 | queue_flag_set(QUEUE_FLAG_ELVSWITCH, q); |
cb98fc8b | 1088 | |
15853af9 | 1089 | elv_drain_elevator(q); |
cb98fc8b TH |
1090 | |
1091 | while (q->rq.elvpriv) { | |
407df2aa TH |
1092 | blk_remove_plug(q); |
1093 | q->request_fn(q); | |
cb98fc8b | 1094 | spin_unlock_irq(q->queue_lock); |
64521d1a | 1095 | msleep(10); |
cb98fc8b | 1096 | spin_lock_irq(q->queue_lock); |
15853af9 | 1097 | elv_drain_elevator(q); |
cb98fc8b TH |
1098 | } |
1099 | ||
1da177e4 | 1100 | /* |
bc1c1169 | 1101 | * Remember old elevator. |
1da177e4 | 1102 | */ |
1da177e4 LT |
1103 | old_elevator = q->elevator; |
1104 | ||
1da177e4 LT |
1105 | /* |
1106 | * attach and start new elevator | |
1107 | */ | |
bc1c1169 JA |
1108 | elevator_attach(q, e, data); |
1109 | ||
1110 | spin_unlock_irq(q->queue_lock); | |
1111 | ||
1112 | __elv_unregister_queue(old_elevator); | |
1da177e4 LT |
1113 | |
1114 | if (elv_register_queue(q)) | |
1115 | goto fail_register; | |
1116 | ||
1117 | /* | |
cb98fc8b | 1118 | * finally exit old elevator and turn off BYPASS. |
1da177e4 LT |
1119 | */ |
1120 | elevator_exit(old_elevator); | |
75ad23bc NP |
1121 | spin_lock_irq(q->queue_lock); |
1122 | queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q); | |
1123 | spin_unlock_irq(q->queue_lock); | |
1124 | ||
4722dc52 AB |
1125 | blk_add_trace_msg(q, "elv switch: %s", e->elevator_type->elevator_name); |
1126 | ||
3d1ab40f | 1127 | return 1; |
1da177e4 LT |
1128 | |
1129 | fail_register: | |
1130 | /* | |
1131 | * switch failed, exit the new io scheduler and reattach the old | |
1132 | * one again (along with re-adding the sysfs dir) | |
1133 | */ | |
1134 | elevator_exit(e); | |
1da177e4 LT |
1135 | q->elevator = old_elevator; |
1136 | elv_register_queue(q); | |
75ad23bc NP |
1137 | |
1138 | spin_lock_irq(q->queue_lock); | |
1139 | queue_flag_clear(QUEUE_FLAG_ELVSWITCH, q); | |
1140 | spin_unlock_irq(q->queue_lock); | |
1141 | ||
3d1ab40f | 1142 | return 0; |
1da177e4 LT |
1143 | } |
1144 | ||
165125e1 JA |
1145 | ssize_t elv_iosched_store(struct request_queue *q, const char *name, |
1146 | size_t count) | |
1da177e4 LT |
1147 | { |
1148 | char elevator_name[ELV_NAME_MAX]; | |
be561235 | 1149 | size_t len; |
1da177e4 LT |
1150 | struct elevator_type *e; |
1151 | ||
be561235 TH |
1152 | elevator_name[sizeof(elevator_name) - 1] = '\0'; |
1153 | strncpy(elevator_name, name, sizeof(elevator_name) - 1); | |
1154 | len = strlen(elevator_name); | |
1da177e4 | 1155 | |
be561235 TH |
1156 | if (len && elevator_name[len - 1] == '\n') |
1157 | elevator_name[len - 1] = '\0'; | |
1da177e4 LT |
1158 | |
1159 | e = elevator_get(elevator_name); | |
1160 | if (!e) { | |
1161 | printk(KERN_ERR "elevator: type %s not found\n", elevator_name); | |
1162 | return -EINVAL; | |
1163 | } | |
1164 | ||
2ca7d93b ND |
1165 | if (!strcmp(elevator_name, q->elevator->elevator_type->elevator_name)) { |
1166 | elevator_put(e); | |
1da177e4 | 1167 | return count; |
2ca7d93b | 1168 | } |
1da177e4 | 1169 | |
3d1ab40f | 1170 | if (!elevator_switch(q, e)) |
4eb166d9 JA |
1171 | printk(KERN_ERR "elevator: switch to %s failed\n", |
1172 | elevator_name); | |
1da177e4 LT |
1173 | return count; |
1174 | } | |
1175 | ||
165125e1 | 1176 | ssize_t elv_iosched_show(struct request_queue *q, char *name) |
1da177e4 LT |
1177 | { |
1178 | elevator_t *e = q->elevator; | |
1179 | struct elevator_type *elv = e->elevator_type; | |
70cee26e | 1180 | struct elevator_type *__e; |
1da177e4 LT |
1181 | int len = 0; |
1182 | ||
2a12dcd7 | 1183 | spin_lock(&elv_list_lock); |
70cee26e | 1184 | list_for_each_entry(__e, &elv_list, list) { |
1da177e4 LT |
1185 | if (!strcmp(elv->elevator_name, __e->elevator_name)) |
1186 | len += sprintf(name+len, "[%s] ", elv->elevator_name); | |
1187 | else | |
1188 | len += sprintf(name+len, "%s ", __e->elevator_name); | |
1189 | } | |
2a12dcd7 | 1190 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
1191 | |
1192 | len += sprintf(len+name, "\n"); | |
1193 | return len; | |
1194 | } | |
1195 | ||
165125e1 JA |
1196 | struct request *elv_rb_former_request(struct request_queue *q, |
1197 | struct request *rq) | |
2e662b65 JA |
1198 | { |
1199 | struct rb_node *rbprev = rb_prev(&rq->rb_node); | |
1200 | ||
1201 | if (rbprev) | |
1202 | return rb_entry_rq(rbprev); | |
1203 | ||
1204 | return NULL; | |
1205 | } | |
2e662b65 JA |
1206 | EXPORT_SYMBOL(elv_rb_former_request); |
1207 | ||
165125e1 JA |
1208 | struct request *elv_rb_latter_request(struct request_queue *q, |
1209 | struct request *rq) | |
2e662b65 JA |
1210 | { |
1211 | struct rb_node *rbnext = rb_next(&rq->rb_node); | |
1212 | ||
1213 | if (rbnext) | |
1214 | return rb_entry_rq(rbnext); | |
1215 | ||
1216 | return NULL; | |
1217 | } | |
2e662b65 | 1218 | EXPORT_SYMBOL(elv_rb_latter_request); |