]>
Commit | Line | Data |
---|---|---|
3dcf60bc | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
1da177e4 LT |
3 | * Block device elevator/IO-scheduler. |
4 | * | |
5 | * Copyright (C) 2000 Andrea Arcangeli <andrea@suse.de> SuSE | |
6 | * | |
0fe23479 | 7 | * 30042000 Jens Axboe <axboe@kernel.dk> : |
1da177e4 LT |
8 | * |
9 | * Split the elevator a bit so that it is possible to choose a different | |
10 | * one or even write a new "plug in". There are three pieces: | |
11 | * - elevator_fn, inserts a new request in the queue list | |
12 | * - elevator_merge_fn, decides whether a new buffer can be merged with | |
13 | * an existing request | |
14 | * - elevator_dequeue_fn, called when a request is taken off the active list | |
15 | * | |
16 | * 20082000 Dave Jones <davej@suse.de> : | |
17 | * Removed tests for max-bomb-segments, which was breaking elvtune | |
18 | * when run without -bN | |
19 | * | |
20 | * Jens: | |
21 | * - Rework again to work with bio instead of buffer_heads | |
22 | * - loose bi_dev comparisons, partition handling is right now | |
23 | * - completely modularize elevator setup and teardown | |
24 | * | |
25 | */ | |
26 | #include <linux/kernel.h> | |
27 | #include <linux/fs.h> | |
28 | #include <linux/blkdev.h> | |
29 | #include <linux/elevator.h> | |
30 | #include <linux/bio.h> | |
1da177e4 LT |
31 | #include <linux/module.h> |
32 | #include <linux/slab.h> | |
33 | #include <linux/init.h> | |
34 | #include <linux/compiler.h> | |
2056a782 | 35 | #include <linux/blktrace_api.h> |
9817064b | 36 | #include <linux/hash.h> |
0835da67 | 37 | #include <linux/uaccess.h> |
c8158819 | 38 | #include <linux/pm_runtime.h> |
eea8f41c | 39 | #include <linux/blk-cgroup.h> |
1da177e4 | 40 | |
55782138 LZ |
41 | #include <trace/events/block.h> |
42 | ||
242f9dcb | 43 | #include "blk.h" |
bd166ef1 | 44 | #include "blk-mq-sched.h" |
bca6b067 | 45 | #include "blk-pm.h" |
8330cdb0 | 46 | #include "blk-wbt.h" |
242f9dcb | 47 | |
1da177e4 LT |
48 | static DEFINE_SPINLOCK(elv_list_lock); |
49 | static LIST_HEAD(elv_list); | |
50 | ||
9817064b JA |
51 | /* |
52 | * Merge hash stuff. | |
53 | */ | |
83096ebf | 54 | #define rq_hash_key(rq) (blk_rq_pos(rq) + blk_rq_sectors(rq)) |
9817064b | 55 | |
da775265 JA |
56 | /* |
57 | * Query io scheduler to see if the current process issuing bio may be | |
58 | * merged with rq. | |
59 | */ | |
72ef799b | 60 | static int elv_iosched_allow_bio_merge(struct request *rq, struct bio *bio) |
da775265 | 61 | { |
165125e1 | 62 | struct request_queue *q = rq->q; |
b374d18a | 63 | struct elevator_queue *e = q->elevator; |
da775265 | 64 | |
f9cd4bfe JA |
65 | if (e->type->ops.allow_merge) |
66 | return e->type->ops.allow_merge(q, rq, bio); | |
da775265 JA |
67 | |
68 | return 1; | |
69 | } | |
70 | ||
1da177e4 LT |
71 | /* |
72 | * can we safely merge with this request? | |
73 | */ | |
72ef799b | 74 | bool elv_bio_merge_ok(struct request *rq, struct bio *bio) |
1da177e4 | 75 | { |
050c8ea8 | 76 | if (!blk_rq_merge_ok(rq, bio)) |
72ef799b | 77 | return false; |
7ba1ba12 | 78 | |
72ef799b TE |
79 | if (!elv_iosched_allow_bio_merge(rq, bio)) |
80 | return false; | |
1da177e4 | 81 | |
72ef799b | 82 | return true; |
1da177e4 | 83 | } |
72ef799b | 84 | EXPORT_SYMBOL(elv_bio_merge_ok); |
1da177e4 | 85 | |
8ac0d9a8 JA |
86 | static bool elevator_match(const struct elevator_type *e, const char *name) |
87 | { | |
88 | if (!strcmp(e->elevator_name, name)) | |
89 | return true; | |
90 | if (e->elevator_alias && !strcmp(e->elevator_alias, name)) | |
91 | return true; | |
92 | ||
93 | return false; | |
94 | } | |
95 | ||
2527d997 | 96 | /* |
a1ce35fa | 97 | * Return scheduler with name 'name' |
2527d997 | 98 | */ |
a1ce35fa | 99 | static struct elevator_type *elevator_find(const char *name) |
1da177e4 | 100 | { |
a22b169d | 101 | struct elevator_type *e; |
1da177e4 | 102 | |
70cee26e | 103 | list_for_each_entry(e, &elv_list, list) { |
a1ce35fa | 104 | if (elevator_match(e, name)) |
a22b169d | 105 | return e; |
1da177e4 | 106 | } |
1da177e4 | 107 | |
a22b169d | 108 | return NULL; |
1da177e4 LT |
109 | } |
110 | ||
111 | static void elevator_put(struct elevator_type *e) | |
112 | { | |
113 | module_put(e->elevator_owner); | |
114 | } | |
115 | ||
2527d997 JA |
116 | static struct elevator_type *elevator_get(struct request_queue *q, |
117 | const char *name, bool try_loading) | |
1da177e4 | 118 | { |
2824bc93 | 119 | struct elevator_type *e; |
1da177e4 | 120 | |
2a12dcd7 | 121 | spin_lock(&elv_list_lock); |
2824bc93 | 122 | |
a1ce35fa | 123 | e = elevator_find(name); |
21c3c5d2 | 124 | if (!e && try_loading) { |
e1640949 | 125 | spin_unlock(&elv_list_lock); |
490b94be | 126 | request_module("%s-iosched", name); |
e1640949 | 127 | spin_lock(&elv_list_lock); |
a1ce35fa | 128 | e = elevator_find(name); |
e1640949 JA |
129 | } |
130 | ||
2824bc93 TH |
131 | if (e && !try_module_get(e->elevator_owner)) |
132 | e = NULL; | |
133 | ||
2a12dcd7 | 134 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
135 | return e; |
136 | } | |
137 | ||
484fc254 | 138 | static char chosen_elevator[ELV_NAME_MAX]; |
1da177e4 | 139 | |
5f003976 | 140 | static int __init elevator_setup(char *str) |
1da177e4 | 141 | { |
752a3b79 CE |
142 | /* |
143 | * Be backwards-compatible with previous kernels, so users | |
144 | * won't get the wrong elevator. | |
145 | */ | |
492af635 | 146 | strncpy(chosen_elevator, str, sizeof(chosen_elevator) - 1); |
9b41046c | 147 | return 1; |
1da177e4 LT |
148 | } |
149 | ||
150 | __setup("elevator=", elevator_setup); | |
151 | ||
3d1ab40f AV |
152 | static struct kobj_type elv_ktype; |
153 | ||
d50235b7 | 154 | struct elevator_queue *elevator_alloc(struct request_queue *q, |
165125e1 | 155 | struct elevator_type *e) |
3d1ab40f | 156 | { |
b374d18a | 157 | struct elevator_queue *eq; |
9817064b | 158 | |
c1b511eb | 159 | eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node); |
9817064b | 160 | if (unlikely(!eq)) |
8406a4d5 | 161 | return NULL; |
9817064b | 162 | |
22f746e2 | 163 | eq->type = e; |
f9cb074b | 164 | kobject_init(&eq->kobj, &elv_ktype); |
9817064b | 165 | mutex_init(&eq->sysfs_lock); |
242d98f0 | 166 | hash_init(eq->hash); |
9817064b | 167 | |
3d1ab40f AV |
168 | return eq; |
169 | } | |
d50235b7 | 170 | EXPORT_SYMBOL(elevator_alloc); |
3d1ab40f AV |
171 | |
172 | static void elevator_release(struct kobject *kobj) | |
173 | { | |
b374d18a | 174 | struct elevator_queue *e; |
9817064b | 175 | |
b374d18a | 176 | e = container_of(kobj, struct elevator_queue, kobj); |
22f746e2 | 177 | elevator_put(e->type); |
3d1ab40f AV |
178 | kfree(e); |
179 | } | |
180 | ||
c3e22192 | 181 | void __elevator_exit(struct request_queue *q, struct elevator_queue *e) |
1da177e4 | 182 | { |
3d1ab40f | 183 | mutex_lock(&e->sysfs_lock); |
f9cd4bfe | 184 | if (e->type->ops.exit_sched) |
54d5329d | 185 | blk_mq_exit_sched(q, e); |
3d1ab40f | 186 | mutex_unlock(&e->sysfs_lock); |
1da177e4 | 187 | |
3d1ab40f | 188 | kobject_put(&e->kobj); |
1da177e4 | 189 | } |
2e662b65 | 190 | |
9817064b JA |
191 | static inline void __elv_rqhash_del(struct request *rq) |
192 | { | |
242d98f0 | 193 | hash_del(&rq->hash); |
e8064021 | 194 | rq->rq_flags &= ~RQF_HASHED; |
9817064b JA |
195 | } |
196 | ||
70b3ea05 | 197 | void elv_rqhash_del(struct request_queue *q, struct request *rq) |
9817064b JA |
198 | { |
199 | if (ELV_ON_HASH(rq)) | |
200 | __elv_rqhash_del(rq); | |
201 | } | |
bd166ef1 | 202 | EXPORT_SYMBOL_GPL(elv_rqhash_del); |
9817064b | 203 | |
70b3ea05 | 204 | void elv_rqhash_add(struct request_queue *q, struct request *rq) |
9817064b | 205 | { |
b374d18a | 206 | struct elevator_queue *e = q->elevator; |
9817064b JA |
207 | |
208 | BUG_ON(ELV_ON_HASH(rq)); | |
242d98f0 | 209 | hash_add(e->hash, &rq->hash, rq_hash_key(rq)); |
e8064021 | 210 | rq->rq_flags |= RQF_HASHED; |
9817064b | 211 | } |
bd166ef1 | 212 | EXPORT_SYMBOL_GPL(elv_rqhash_add); |
9817064b | 213 | |
70b3ea05 | 214 | void elv_rqhash_reposition(struct request_queue *q, struct request *rq) |
9817064b JA |
215 | { |
216 | __elv_rqhash_del(rq); | |
217 | elv_rqhash_add(q, rq); | |
218 | } | |
219 | ||
70b3ea05 | 220 | struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) |
9817064b | 221 | { |
b374d18a | 222 | struct elevator_queue *e = q->elevator; |
b67bfe0d | 223 | struct hlist_node *next; |
9817064b JA |
224 | struct request *rq; |
225 | ||
ee89f812 | 226 | hash_for_each_possible_safe(e->hash, rq, next, hash, offset) { |
9817064b JA |
227 | BUG_ON(!ELV_ON_HASH(rq)); |
228 | ||
229 | if (unlikely(!rq_mergeable(rq))) { | |
230 | __elv_rqhash_del(rq); | |
231 | continue; | |
232 | } | |
233 | ||
234 | if (rq_hash_key(rq) == offset) | |
235 | return rq; | |
236 | } | |
237 | ||
238 | return NULL; | |
239 | } | |
240 | ||
2e662b65 JA |
241 | /* |
242 | * RB-tree support functions for inserting/lookup/removal of requests | |
243 | * in a sorted RB tree. | |
244 | */ | |
796d5116 | 245 | void elv_rb_add(struct rb_root *root, struct request *rq) |
2e662b65 JA |
246 | { |
247 | struct rb_node **p = &root->rb_node; | |
248 | struct rb_node *parent = NULL; | |
249 | struct request *__rq; | |
250 | ||
251 | while (*p) { | |
252 | parent = *p; | |
253 | __rq = rb_entry(parent, struct request, rb_node); | |
254 | ||
83096ebf | 255 | if (blk_rq_pos(rq) < blk_rq_pos(__rq)) |
2e662b65 | 256 | p = &(*p)->rb_left; |
796d5116 | 257 | else if (blk_rq_pos(rq) >= blk_rq_pos(__rq)) |
2e662b65 | 258 | p = &(*p)->rb_right; |
2e662b65 JA |
259 | } |
260 | ||
261 | rb_link_node(&rq->rb_node, parent, p); | |
262 | rb_insert_color(&rq->rb_node, root); | |
2e662b65 | 263 | } |
2e662b65 JA |
264 | EXPORT_SYMBOL(elv_rb_add); |
265 | ||
266 | void elv_rb_del(struct rb_root *root, struct request *rq) | |
267 | { | |
268 | BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); | |
269 | rb_erase(&rq->rb_node, root); | |
270 | RB_CLEAR_NODE(&rq->rb_node); | |
271 | } | |
2e662b65 JA |
272 | EXPORT_SYMBOL(elv_rb_del); |
273 | ||
274 | struct request *elv_rb_find(struct rb_root *root, sector_t sector) | |
275 | { | |
276 | struct rb_node *n = root->rb_node; | |
277 | struct request *rq; | |
278 | ||
279 | while (n) { | |
280 | rq = rb_entry(n, struct request, rb_node); | |
281 | ||
83096ebf | 282 | if (sector < blk_rq_pos(rq)) |
2e662b65 | 283 | n = n->rb_left; |
83096ebf | 284 | else if (sector > blk_rq_pos(rq)) |
2e662b65 JA |
285 | n = n->rb_right; |
286 | else | |
287 | return rq; | |
288 | } | |
289 | ||
290 | return NULL; | |
291 | } | |
2e662b65 JA |
292 | EXPORT_SYMBOL(elv_rb_find); |
293 | ||
34fe7c05 CH |
294 | enum elv_merge elv_merge(struct request_queue *q, struct request **req, |
295 | struct bio *bio) | |
1da177e4 | 296 | { |
b374d18a | 297 | struct elevator_queue *e = q->elevator; |
9817064b | 298 | struct request *__rq; |
06b86245 | 299 | |
488991e2 AB |
300 | /* |
301 | * Levels of merges: | |
302 | * nomerges: No merges at all attempted | |
303 | * noxmerges: Only simple one-hit cache try | |
304 | * merges: All merge tries attempted | |
305 | */ | |
7460d389 | 306 | if (blk_queue_nomerges(q) || !bio_mergeable(bio)) |
488991e2 AB |
307 | return ELEVATOR_NO_MERGE; |
308 | ||
9817064b JA |
309 | /* |
310 | * First try one-hit cache. | |
311 | */ | |
72ef799b | 312 | if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) { |
34fe7c05 CH |
313 | enum elv_merge ret = blk_try_merge(q->last_merge, bio); |
314 | ||
06b86245 TH |
315 | if (ret != ELEVATOR_NO_MERGE) { |
316 | *req = q->last_merge; | |
317 | return ret; | |
318 | } | |
319 | } | |
1da177e4 | 320 | |
488991e2 | 321 | if (blk_queue_noxmerges(q)) |
ac9fafa1 AB |
322 | return ELEVATOR_NO_MERGE; |
323 | ||
9817064b JA |
324 | /* |
325 | * See if our hash lookup can find a potential backmerge. | |
326 | */ | |
4f024f37 | 327 | __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector); |
72ef799b | 328 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b JA |
329 | *req = __rq; |
330 | return ELEVATOR_BACK_MERGE; | |
331 | } | |
332 | ||
f9cd4bfe JA |
333 | if (e->type->ops.request_merge) |
334 | return e->type->ops.request_merge(q, req, bio); | |
1da177e4 LT |
335 | |
336 | return ELEVATOR_NO_MERGE; | |
337 | } | |
338 | ||
5e84ea3a JA |
339 | /* |
340 | * Attempt to do an insertion back merge. Only check for the case where | |
341 | * we can append 'rq' to an existing request, so we can throw 'rq' away | |
342 | * afterwards. | |
343 | * | |
344 | * Returns true if we merged, false otherwise | |
345 | */ | |
bd166ef1 | 346 | bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq) |
5e84ea3a JA |
347 | { |
348 | struct request *__rq; | |
bee0393c | 349 | bool ret; |
5e84ea3a JA |
350 | |
351 | if (blk_queue_nomerges(q)) | |
352 | return false; | |
353 | ||
354 | /* | |
355 | * First try one-hit cache. | |
356 | */ | |
357 | if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) | |
358 | return true; | |
359 | ||
360 | if (blk_queue_noxmerges(q)) | |
361 | return false; | |
362 | ||
bee0393c | 363 | ret = false; |
5e84ea3a JA |
364 | /* |
365 | * See if our hash lookup can find a potential backmerge. | |
366 | */ | |
bee0393c SL |
367 | while (1) { |
368 | __rq = elv_rqhash_find(q, blk_rq_pos(rq)); | |
369 | if (!__rq || !blk_attempt_req_merge(q, __rq, rq)) | |
370 | break; | |
371 | ||
372 | /* The merged request could be merged with others, try again */ | |
373 | ret = true; | |
374 | rq = __rq; | |
375 | } | |
27419322 | 376 | |
bee0393c | 377 | return ret; |
5e84ea3a JA |
378 | } |
379 | ||
34fe7c05 CH |
380 | void elv_merged_request(struct request_queue *q, struct request *rq, |
381 | enum elv_merge type) | |
1da177e4 | 382 | { |
b374d18a | 383 | struct elevator_queue *e = q->elevator; |
1da177e4 | 384 | |
f9cd4bfe JA |
385 | if (e->type->ops.request_merged) |
386 | e->type->ops.request_merged(q, rq, type); | |
06b86245 | 387 | |
2e662b65 JA |
388 | if (type == ELEVATOR_BACK_MERGE) |
389 | elv_rqhash_reposition(q, rq); | |
9817064b | 390 | |
06b86245 | 391 | q->last_merge = rq; |
1da177e4 LT |
392 | } |
393 | ||
165125e1 | 394 | void elv_merge_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
395 | struct request *next) |
396 | { | |
b374d18a | 397 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 398 | |
f9cd4bfe JA |
399 | if (e->type->ops.requests_merged) |
400 | e->type->ops.requests_merged(q, rq, next); | |
06b86245 | 401 | |
9817064b | 402 | elv_rqhash_reposition(q, rq); |
06b86245 | 403 | q->last_merge = rq; |
1da177e4 LT |
404 | } |
405 | ||
165125e1 | 406 | struct request *elv_latter_request(struct request_queue *q, struct request *rq) |
1da177e4 | 407 | { |
b374d18a | 408 | struct elevator_queue *e = q->elevator; |
1da177e4 | 409 | |
f9cd4bfe JA |
410 | if (e->type->ops.next_request) |
411 | return e->type->ops.next_request(q, rq); | |
bd166ef1 | 412 | |
1da177e4 LT |
413 | return NULL; |
414 | } | |
415 | ||
165125e1 | 416 | struct request *elv_former_request(struct request_queue *q, struct request *rq) |
1da177e4 | 417 | { |
b374d18a | 418 | struct elevator_queue *e = q->elevator; |
1da177e4 | 419 | |
f9cd4bfe JA |
420 | if (e->type->ops.former_request) |
421 | return e->type->ops.former_request(q, rq); | |
bd166ef1 | 422 | |
a1ce35fa | 423 | return NULL; |
1da177e4 LT |
424 | } |
425 | ||
3d1ab40f AV |
426 | #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) |
427 | ||
428 | static ssize_t | |
429 | elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1da177e4 | 430 | { |
3d1ab40f | 431 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 432 | struct elevator_queue *e; |
3d1ab40f AV |
433 | ssize_t error; |
434 | ||
435 | if (!entry->show) | |
436 | return -EIO; | |
437 | ||
b374d18a | 438 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 439 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 440 | error = e->type ? entry->show(e, page) : -ENOENT; |
3d1ab40f AV |
441 | mutex_unlock(&e->sysfs_lock); |
442 | return error; | |
443 | } | |
1da177e4 | 444 | |
3d1ab40f AV |
445 | static ssize_t |
446 | elv_attr_store(struct kobject *kobj, struct attribute *attr, | |
447 | const char *page, size_t length) | |
448 | { | |
3d1ab40f | 449 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 450 | struct elevator_queue *e; |
3d1ab40f | 451 | ssize_t error; |
1da177e4 | 452 | |
3d1ab40f AV |
453 | if (!entry->store) |
454 | return -EIO; | |
1da177e4 | 455 | |
b374d18a | 456 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 457 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 458 | error = e->type ? entry->store(e, page, length) : -ENOENT; |
3d1ab40f AV |
459 | mutex_unlock(&e->sysfs_lock); |
460 | return error; | |
461 | } | |
462 | ||
52cf25d0 | 463 | static const struct sysfs_ops elv_sysfs_ops = { |
3d1ab40f AV |
464 | .show = elv_attr_show, |
465 | .store = elv_attr_store, | |
466 | }; | |
467 | ||
468 | static struct kobj_type elv_ktype = { | |
469 | .sysfs_ops = &elv_sysfs_ops, | |
470 | .release = elevator_release, | |
471 | }; | |
472 | ||
5a5bafdc | 473 | int elv_register_queue(struct request_queue *q) |
3d1ab40f | 474 | { |
5a5bafdc | 475 | struct elevator_queue *e = q->elevator; |
3d1ab40f AV |
476 | int error; |
477 | ||
14a23498 BVA |
478 | lockdep_assert_held(&q->sysfs_lock); |
479 | ||
b2d6db58 | 480 | error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); |
3d1ab40f | 481 | if (!error) { |
22f746e2 | 482 | struct elv_fs_entry *attr = e->type->elevator_attrs; |
3d1ab40f | 483 | if (attr) { |
e572ec7e AV |
484 | while (attr->attr.name) { |
485 | if (sysfs_create_file(&e->kobj, &attr->attr)) | |
3d1ab40f | 486 | break; |
e572ec7e | 487 | attr++; |
3d1ab40f AV |
488 | } |
489 | } | |
490 | kobject_uevent(&e->kobj, KOBJ_ADD); | |
430c62fb | 491 | e->registered = 1; |
3d1ab40f AV |
492 | } |
493 | return error; | |
1da177e4 | 494 | } |
bc1c1169 | 495 | |
1da177e4 LT |
496 | void elv_unregister_queue(struct request_queue *q) |
497 | { | |
14a23498 BVA |
498 | lockdep_assert_held(&q->sysfs_lock); |
499 | ||
f8fc877d TH |
500 | if (q) { |
501 | struct elevator_queue *e = q->elevator; | |
502 | ||
503 | kobject_uevent(&e->kobj, KOBJ_REMOVE); | |
504 | kobject_del(&e->kobj); | |
505 | e->registered = 0; | |
8330cdb0 JK |
506 | /* Re-enable throttling in case elevator disabled it */ |
507 | wbt_enable_default(q); | |
f8fc877d | 508 | } |
1da177e4 LT |
509 | } |
510 | ||
e567bf71 | 511 | int elv_register(struct elevator_type *e) |
1da177e4 | 512 | { |
3d3c2379 TH |
513 | /* create icq_cache if requested */ |
514 | if (e->icq_size) { | |
515 | if (WARN_ON(e->icq_size < sizeof(struct io_cq)) || | |
516 | WARN_ON(e->icq_align < __alignof__(struct io_cq))) | |
517 | return -EINVAL; | |
518 | ||
519 | snprintf(e->icq_cache_name, sizeof(e->icq_cache_name), | |
520 | "%s_io_cq", e->elevator_name); | |
521 | e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size, | |
522 | e->icq_align, 0, NULL); | |
523 | if (!e->icq_cache) | |
524 | return -ENOMEM; | |
525 | } | |
526 | ||
527 | /* register, don't allow duplicate names */ | |
2a12dcd7 | 528 | spin_lock(&elv_list_lock); |
a1ce35fa | 529 | if (elevator_find(e->elevator_name)) { |
3d3c2379 | 530 | spin_unlock(&elv_list_lock); |
62d2a194 | 531 | kmem_cache_destroy(e->icq_cache); |
3d3c2379 TH |
532 | return -EBUSY; |
533 | } | |
1da177e4 | 534 | list_add_tail(&e->list, &elv_list); |
2a12dcd7 | 535 | spin_unlock(&elv_list_lock); |
1da177e4 | 536 | |
d0b0a81a HT |
537 | printk(KERN_INFO "io scheduler %s registered\n", e->elevator_name); |
538 | ||
3d3c2379 | 539 | return 0; |
1da177e4 LT |
540 | } |
541 | EXPORT_SYMBOL_GPL(elv_register); | |
542 | ||
543 | void elv_unregister(struct elevator_type *e) | |
544 | { | |
3d3c2379 | 545 | /* unregister */ |
2a12dcd7 | 546 | spin_lock(&elv_list_lock); |
1da177e4 | 547 | list_del_init(&e->list); |
2a12dcd7 | 548 | spin_unlock(&elv_list_lock); |
3d3c2379 TH |
549 | |
550 | /* | |
551 | * Destroy icq_cache if it exists. icq's are RCU managed. Make | |
552 | * sure all RCU operations are complete before proceeding. | |
553 | */ | |
554 | if (e->icq_cache) { | |
555 | rcu_barrier(); | |
556 | kmem_cache_destroy(e->icq_cache); | |
557 | e->icq_cache = NULL; | |
558 | } | |
1da177e4 LT |
559 | } |
560 | EXPORT_SYMBOL_GPL(elv_unregister); | |
561 | ||
d48ece20 | 562 | int elevator_switch_mq(struct request_queue *q, |
54d5329d OS |
563 | struct elevator_type *new_e) |
564 | { | |
565 | int ret; | |
566 | ||
14a23498 BVA |
567 | lockdep_assert_held(&q->sysfs_lock); |
568 | ||
54d5329d OS |
569 | if (q->elevator) { |
570 | if (q->elevator->registered) | |
571 | elv_unregister_queue(q); | |
572 | ioc_clear_queue(q); | |
573 | elevator_exit(q, q->elevator); | |
574 | } | |
575 | ||
576 | ret = blk_mq_init_sched(q, new_e); | |
577 | if (ret) | |
578 | goto out; | |
579 | ||
580 | if (new_e) { | |
581 | ret = elv_register_queue(q); | |
582 | if (ret) { | |
583 | elevator_exit(q, q->elevator); | |
584 | goto out; | |
585 | } | |
586 | } | |
587 | ||
588 | if (new_e) | |
589 | blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name); | |
590 | else | |
591 | blk_add_trace_msg(q, "elv switch: none"); | |
592 | ||
593 | out: | |
54d5329d | 594 | return ret; |
54d5329d OS |
595 | } |
596 | ||
131d08e1 CH |
597 | /* |
598 | * For blk-mq devices, we default to using mq-deadline, if available, for single | |
599 | * queue devices. If deadline isn't available OR we have multiple queues, | |
600 | * default to "none". | |
601 | */ | |
602 | int elevator_init_mq(struct request_queue *q) | |
603 | { | |
604 | struct elevator_type *e; | |
605 | int err = 0; | |
606 | ||
607 | if (q->nr_hw_queues != 1) | |
608 | return 0; | |
609 | ||
610 | /* | |
611 | * q->sysfs_lock must be held to provide mutual exclusion between | |
612 | * elevator_switch() and here. | |
613 | */ | |
614 | mutex_lock(&q->sysfs_lock); | |
615 | if (unlikely(q->elevator)) | |
616 | goto out_unlock; | |
617 | ||
618 | e = elevator_get(q, "mq-deadline", false); | |
619 | if (!e) | |
620 | goto out_unlock; | |
621 | ||
622 | err = blk_mq_init_sched(q, e); | |
623 | if (err) | |
624 | elevator_put(e); | |
625 | out_unlock: | |
626 | mutex_unlock(&q->sysfs_lock); | |
627 | return err; | |
628 | } | |
629 | ||
630 | ||
1da177e4 LT |
631 | /* |
632 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
633 | * we don't free the old io scheduler, before we have allocated what we | |
634 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 635 | * one, if the new one fails init for some reason. |
1da177e4 | 636 | */ |
165125e1 | 637 | static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) |
1da177e4 | 638 | { |
e8989fae | 639 | int err; |
1da177e4 | 640 | |
14a23498 BVA |
641 | lockdep_assert_held(&q->sysfs_lock); |
642 | ||
a1ce35fa JA |
643 | blk_mq_freeze_queue(q); |
644 | blk_mq_quiesce_queue(q); | |
4722dc52 | 645 | |
a1ce35fa | 646 | err = elevator_switch_mq(q, new_e); |
1da177e4 | 647 | |
a1ce35fa JA |
648 | blk_mq_unquiesce_queue(q); |
649 | blk_mq_unfreeze_queue(q); | |
75ad23bc | 650 | |
5dd531a0 | 651 | return err; |
1da177e4 LT |
652 | } |
653 | ||
5dd531a0 JA |
654 | /* |
655 | * Switch this queue to the given IO scheduler. | |
656 | */ | |
7c8a3679 | 657 | static int __elevator_change(struct request_queue *q, const char *name) |
1da177e4 LT |
658 | { |
659 | char elevator_name[ELV_NAME_MAX]; | |
660 | struct elevator_type *e; | |
661 | ||
e9a823fb DJ |
662 | /* Make sure queue is not in the middle of being removed */ |
663 | if (!test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags)) | |
664 | return -ENOENT; | |
665 | ||
bd166ef1 JA |
666 | /* |
667 | * Special case for mq, turn off scheduling | |
668 | */ | |
fbd72127 AZ |
669 | if (!strncmp(name, "none", 4)) { |
670 | if (!q->elevator) | |
671 | return 0; | |
bd166ef1 | 672 | return elevator_switch(q, NULL); |
fbd72127 | 673 | } |
cd43e26f | 674 | |
ee2e992c | 675 | strlcpy(elevator_name, name, sizeof(elevator_name)); |
2527d997 | 676 | e = elevator_get(q, strstrip(elevator_name), true); |
340ff321 | 677 | if (!e) |
1da177e4 | 678 | return -EINVAL; |
1da177e4 | 679 | |
8ac0d9a8 | 680 | if (q->elevator && elevator_match(q->elevator->type, elevator_name)) { |
2ca7d93b | 681 | elevator_put(e); |
5dd531a0 | 682 | return 0; |
2ca7d93b | 683 | } |
1da177e4 | 684 | |
5dd531a0 JA |
685 | return elevator_switch(q, e); |
686 | } | |
7c8a3679 | 687 | |
3a5088c8 ML |
688 | static inline bool elv_support_iosched(struct request_queue *q) |
689 | { | |
344e9ffc | 690 | if (q->tag_set && (q->tag_set->flags & BLK_MQ_F_NO_SCHED)) |
3a5088c8 ML |
691 | return false; |
692 | return true; | |
693 | } | |
694 | ||
5dd531a0 JA |
695 | ssize_t elv_iosched_store(struct request_queue *q, const char *name, |
696 | size_t count) | |
697 | { | |
698 | int ret; | |
699 | ||
344e9ffc | 700 | if (!queue_is_mq(q) || !elv_support_iosched(q)) |
5dd531a0 JA |
701 | return count; |
702 | ||
7c8a3679 | 703 | ret = __elevator_change(q, name); |
5dd531a0 JA |
704 | if (!ret) |
705 | return count; | |
706 | ||
5dd531a0 | 707 | return ret; |
1da177e4 LT |
708 | } |
709 | ||
165125e1 | 710 | ssize_t elv_iosched_show(struct request_queue *q, char *name) |
1da177e4 | 711 | { |
b374d18a | 712 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 713 | struct elevator_type *elv = NULL; |
70cee26e | 714 | struct elevator_type *__e; |
1da177e4 LT |
715 | int len = 0; |
716 | ||
344e9ffc | 717 | if (!queue_is_mq(q)) |
cd43e26f MP |
718 | return sprintf(name, "none\n"); |
719 | ||
bd166ef1 JA |
720 | if (!q->elevator) |
721 | len += sprintf(name+len, "[none] "); | |
722 | else | |
723 | elv = e->type; | |
cd43e26f | 724 | |
2a12dcd7 | 725 | spin_lock(&elv_list_lock); |
70cee26e | 726 | list_for_each_entry(__e, &elv_list, list) { |
a1ce35fa | 727 | if (elv && elevator_match(elv, __e->elevator_name)) { |
1da177e4 | 728 | len += sprintf(name+len, "[%s] ", elv->elevator_name); |
bd166ef1 JA |
729 | continue; |
730 | } | |
a1ce35fa | 731 | if (elv_support_iosched(q)) |
1da177e4 LT |
732 | len += sprintf(name+len, "%s ", __e->elevator_name); |
733 | } | |
2a12dcd7 | 734 | spin_unlock(&elv_list_lock); |
1da177e4 | 735 | |
344e9ffc | 736 | if (q->elevator) |
bd166ef1 JA |
737 | len += sprintf(name+len, "none"); |
738 | ||
1da177e4 LT |
739 | len += sprintf(len+name, "\n"); |
740 | return len; | |
741 | } | |
742 | ||
165125e1 JA |
743 | struct request *elv_rb_former_request(struct request_queue *q, |
744 | struct request *rq) | |
2e662b65 JA |
745 | { |
746 | struct rb_node *rbprev = rb_prev(&rq->rb_node); | |
747 | ||
748 | if (rbprev) | |
749 | return rb_entry_rq(rbprev); | |
750 | ||
751 | return NULL; | |
752 | } | |
2e662b65 JA |
753 | EXPORT_SYMBOL(elv_rb_former_request); |
754 | ||
165125e1 JA |
755 | struct request *elv_rb_latter_request(struct request_queue *q, |
756 | struct request *rq) | |
2e662b65 JA |
757 | { |
758 | struct rb_node *rbnext = rb_next(&rq->rb_node); | |
759 | ||
760 | if (rbnext) | |
761 | return rb_entry_rq(rbnext); | |
762 | ||
763 | return NULL; | |
764 | } | |
2e662b65 | 765 | EXPORT_SYMBOL(elv_rb_latter_request); |