]>
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 | |
68c43f13 DLM |
86 | static inline bool elv_support_features(unsigned int elv_features, |
87 | unsigned int required_features) | |
8ac0d9a8 | 88 | { |
68c43f13 DLM |
89 | return (required_features & elv_features) == required_features; |
90 | } | |
91 | ||
92 | /** | |
93 | * elevator_match - Test an elevator name and features | |
94 | * @e: Scheduler to test | |
95 | * @name: Elevator name to test | |
96 | * @required_features: Features that the elevator must provide | |
97 | * | |
98 | * Return true is the elevator @e name matches @name and if @e provides all the | |
99 | * the feratures spcified by @required_features. | |
100 | */ | |
101 | static bool elevator_match(const struct elevator_type *e, const char *name, | |
102 | unsigned int required_features) | |
103 | { | |
104 | if (!elv_support_features(e->elevator_features, required_features)) | |
105 | return false; | |
8ac0d9a8 JA |
106 | if (!strcmp(e->elevator_name, name)) |
107 | return true; | |
108 | if (e->elevator_alias && !strcmp(e->elevator_alias, name)) | |
109 | return true; | |
110 | ||
111 | return false; | |
112 | } | |
113 | ||
68c43f13 DLM |
114 | /** |
115 | * elevator_find - Find an elevator | |
116 | * @name: Name of the elevator to find | |
117 | * @required_features: Features that the elevator must provide | |
118 | * | |
119 | * Return the first registered scheduler with name @name and supporting the | |
120 | * features @required_features and NULL otherwise. | |
2527d997 | 121 | */ |
68c43f13 DLM |
122 | static struct elevator_type *elevator_find(const char *name, |
123 | unsigned int required_features) | |
1da177e4 | 124 | { |
a22b169d | 125 | struct elevator_type *e; |
1da177e4 | 126 | |
70cee26e | 127 | list_for_each_entry(e, &elv_list, list) { |
68c43f13 | 128 | if (elevator_match(e, name, required_features)) |
a22b169d | 129 | return e; |
1da177e4 | 130 | } |
1da177e4 | 131 | |
a22b169d | 132 | return NULL; |
1da177e4 LT |
133 | } |
134 | ||
135 | static void elevator_put(struct elevator_type *e) | |
136 | { | |
137 | module_put(e->elevator_owner); | |
138 | } | |
139 | ||
2527d997 JA |
140 | static struct elevator_type *elevator_get(struct request_queue *q, |
141 | const char *name, bool try_loading) | |
1da177e4 | 142 | { |
2824bc93 | 143 | struct elevator_type *e; |
1da177e4 | 144 | |
2a12dcd7 | 145 | spin_lock(&elv_list_lock); |
2824bc93 | 146 | |
68c43f13 | 147 | e = elevator_find(name, q->required_elevator_features); |
21c3c5d2 | 148 | if (!e && try_loading) { |
e1640949 | 149 | spin_unlock(&elv_list_lock); |
490b94be | 150 | request_module("%s-iosched", name); |
e1640949 | 151 | spin_lock(&elv_list_lock); |
68c43f13 | 152 | e = elevator_find(name, q->required_elevator_features); |
e1640949 JA |
153 | } |
154 | ||
2824bc93 TH |
155 | if (e && !try_module_get(e->elevator_owner)) |
156 | e = NULL; | |
157 | ||
2a12dcd7 | 158 | spin_unlock(&elv_list_lock); |
1da177e4 LT |
159 | return e; |
160 | } | |
161 | ||
3d1ab40f AV |
162 | static struct kobj_type elv_ktype; |
163 | ||
d50235b7 | 164 | struct elevator_queue *elevator_alloc(struct request_queue *q, |
165125e1 | 165 | struct elevator_type *e) |
3d1ab40f | 166 | { |
b374d18a | 167 | struct elevator_queue *eq; |
9817064b | 168 | |
c1b511eb | 169 | eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, q->node); |
9817064b | 170 | if (unlikely(!eq)) |
8406a4d5 | 171 | return NULL; |
9817064b | 172 | |
22f746e2 | 173 | eq->type = e; |
f9cb074b | 174 | kobject_init(&eq->kobj, &elv_ktype); |
9817064b | 175 | mutex_init(&eq->sysfs_lock); |
242d98f0 | 176 | hash_init(eq->hash); |
9817064b | 177 | |
3d1ab40f AV |
178 | return eq; |
179 | } | |
d50235b7 | 180 | EXPORT_SYMBOL(elevator_alloc); |
3d1ab40f AV |
181 | |
182 | static void elevator_release(struct kobject *kobj) | |
183 | { | |
b374d18a | 184 | struct elevator_queue *e; |
9817064b | 185 | |
b374d18a | 186 | e = container_of(kobj, struct elevator_queue, kobj); |
22f746e2 | 187 | elevator_put(e->type); |
3d1ab40f AV |
188 | kfree(e); |
189 | } | |
190 | ||
c3e22192 | 191 | void __elevator_exit(struct request_queue *q, struct elevator_queue *e) |
1da177e4 | 192 | { |
3d1ab40f | 193 | mutex_lock(&e->sysfs_lock); |
f9cd4bfe | 194 | if (e->type->ops.exit_sched) |
54d5329d | 195 | blk_mq_exit_sched(q, e); |
3d1ab40f | 196 | mutex_unlock(&e->sysfs_lock); |
1da177e4 | 197 | |
3d1ab40f | 198 | kobject_put(&e->kobj); |
1da177e4 | 199 | } |
2e662b65 | 200 | |
9817064b JA |
201 | static inline void __elv_rqhash_del(struct request *rq) |
202 | { | |
242d98f0 | 203 | hash_del(&rq->hash); |
e8064021 | 204 | rq->rq_flags &= ~RQF_HASHED; |
9817064b JA |
205 | } |
206 | ||
70b3ea05 | 207 | void elv_rqhash_del(struct request_queue *q, struct request *rq) |
9817064b JA |
208 | { |
209 | if (ELV_ON_HASH(rq)) | |
210 | __elv_rqhash_del(rq); | |
211 | } | |
bd166ef1 | 212 | EXPORT_SYMBOL_GPL(elv_rqhash_del); |
9817064b | 213 | |
70b3ea05 | 214 | void elv_rqhash_add(struct request_queue *q, struct request *rq) |
9817064b | 215 | { |
b374d18a | 216 | struct elevator_queue *e = q->elevator; |
9817064b JA |
217 | |
218 | BUG_ON(ELV_ON_HASH(rq)); | |
242d98f0 | 219 | hash_add(e->hash, &rq->hash, rq_hash_key(rq)); |
e8064021 | 220 | rq->rq_flags |= RQF_HASHED; |
9817064b | 221 | } |
bd166ef1 | 222 | EXPORT_SYMBOL_GPL(elv_rqhash_add); |
9817064b | 223 | |
70b3ea05 | 224 | void elv_rqhash_reposition(struct request_queue *q, struct request *rq) |
9817064b JA |
225 | { |
226 | __elv_rqhash_del(rq); | |
227 | elv_rqhash_add(q, rq); | |
228 | } | |
229 | ||
70b3ea05 | 230 | struct request *elv_rqhash_find(struct request_queue *q, sector_t offset) |
9817064b | 231 | { |
b374d18a | 232 | struct elevator_queue *e = q->elevator; |
b67bfe0d | 233 | struct hlist_node *next; |
9817064b JA |
234 | struct request *rq; |
235 | ||
ee89f812 | 236 | hash_for_each_possible_safe(e->hash, rq, next, hash, offset) { |
9817064b JA |
237 | BUG_ON(!ELV_ON_HASH(rq)); |
238 | ||
239 | if (unlikely(!rq_mergeable(rq))) { | |
240 | __elv_rqhash_del(rq); | |
241 | continue; | |
242 | } | |
243 | ||
244 | if (rq_hash_key(rq) == offset) | |
245 | return rq; | |
246 | } | |
247 | ||
248 | return NULL; | |
249 | } | |
250 | ||
2e662b65 JA |
251 | /* |
252 | * RB-tree support functions for inserting/lookup/removal of requests | |
253 | * in a sorted RB tree. | |
254 | */ | |
796d5116 | 255 | void elv_rb_add(struct rb_root *root, struct request *rq) |
2e662b65 JA |
256 | { |
257 | struct rb_node **p = &root->rb_node; | |
258 | struct rb_node *parent = NULL; | |
259 | struct request *__rq; | |
260 | ||
261 | while (*p) { | |
262 | parent = *p; | |
263 | __rq = rb_entry(parent, struct request, rb_node); | |
264 | ||
83096ebf | 265 | if (blk_rq_pos(rq) < blk_rq_pos(__rq)) |
2e662b65 | 266 | p = &(*p)->rb_left; |
796d5116 | 267 | else if (blk_rq_pos(rq) >= blk_rq_pos(__rq)) |
2e662b65 | 268 | p = &(*p)->rb_right; |
2e662b65 JA |
269 | } |
270 | ||
271 | rb_link_node(&rq->rb_node, parent, p); | |
272 | rb_insert_color(&rq->rb_node, root); | |
2e662b65 | 273 | } |
2e662b65 JA |
274 | EXPORT_SYMBOL(elv_rb_add); |
275 | ||
276 | void elv_rb_del(struct rb_root *root, struct request *rq) | |
277 | { | |
278 | BUG_ON(RB_EMPTY_NODE(&rq->rb_node)); | |
279 | rb_erase(&rq->rb_node, root); | |
280 | RB_CLEAR_NODE(&rq->rb_node); | |
281 | } | |
2e662b65 JA |
282 | EXPORT_SYMBOL(elv_rb_del); |
283 | ||
284 | struct request *elv_rb_find(struct rb_root *root, sector_t sector) | |
285 | { | |
286 | struct rb_node *n = root->rb_node; | |
287 | struct request *rq; | |
288 | ||
289 | while (n) { | |
290 | rq = rb_entry(n, struct request, rb_node); | |
291 | ||
83096ebf | 292 | if (sector < blk_rq_pos(rq)) |
2e662b65 | 293 | n = n->rb_left; |
83096ebf | 294 | else if (sector > blk_rq_pos(rq)) |
2e662b65 JA |
295 | n = n->rb_right; |
296 | else | |
297 | return rq; | |
298 | } | |
299 | ||
300 | return NULL; | |
301 | } | |
2e662b65 JA |
302 | EXPORT_SYMBOL(elv_rb_find); |
303 | ||
34fe7c05 CH |
304 | enum elv_merge elv_merge(struct request_queue *q, struct request **req, |
305 | struct bio *bio) | |
1da177e4 | 306 | { |
b374d18a | 307 | struct elevator_queue *e = q->elevator; |
9817064b | 308 | struct request *__rq; |
06b86245 | 309 | |
488991e2 AB |
310 | /* |
311 | * Levels of merges: | |
312 | * nomerges: No merges at all attempted | |
313 | * noxmerges: Only simple one-hit cache try | |
314 | * merges: All merge tries attempted | |
315 | */ | |
7460d389 | 316 | if (blk_queue_nomerges(q) || !bio_mergeable(bio)) |
488991e2 AB |
317 | return ELEVATOR_NO_MERGE; |
318 | ||
9817064b JA |
319 | /* |
320 | * First try one-hit cache. | |
321 | */ | |
72ef799b | 322 | if (q->last_merge && elv_bio_merge_ok(q->last_merge, bio)) { |
34fe7c05 CH |
323 | enum elv_merge ret = blk_try_merge(q->last_merge, bio); |
324 | ||
06b86245 TH |
325 | if (ret != ELEVATOR_NO_MERGE) { |
326 | *req = q->last_merge; | |
327 | return ret; | |
328 | } | |
329 | } | |
1da177e4 | 330 | |
488991e2 | 331 | if (blk_queue_noxmerges(q)) |
ac9fafa1 AB |
332 | return ELEVATOR_NO_MERGE; |
333 | ||
9817064b JA |
334 | /* |
335 | * See if our hash lookup can find a potential backmerge. | |
336 | */ | |
4f024f37 | 337 | __rq = elv_rqhash_find(q, bio->bi_iter.bi_sector); |
72ef799b | 338 | if (__rq && elv_bio_merge_ok(__rq, bio)) { |
9817064b JA |
339 | *req = __rq; |
340 | return ELEVATOR_BACK_MERGE; | |
341 | } | |
342 | ||
f9cd4bfe JA |
343 | if (e->type->ops.request_merge) |
344 | return e->type->ops.request_merge(q, req, bio); | |
1da177e4 LT |
345 | |
346 | return ELEVATOR_NO_MERGE; | |
347 | } | |
348 | ||
5e84ea3a JA |
349 | /* |
350 | * Attempt to do an insertion back merge. Only check for the case where | |
351 | * we can append 'rq' to an existing request, so we can throw 'rq' away | |
352 | * afterwards. | |
353 | * | |
354 | * Returns true if we merged, false otherwise | |
355 | */ | |
bd166ef1 | 356 | bool elv_attempt_insert_merge(struct request_queue *q, struct request *rq) |
5e84ea3a JA |
357 | { |
358 | struct request *__rq; | |
bee0393c | 359 | bool ret; |
5e84ea3a JA |
360 | |
361 | if (blk_queue_nomerges(q)) | |
362 | return false; | |
363 | ||
364 | /* | |
365 | * First try one-hit cache. | |
366 | */ | |
367 | if (q->last_merge && blk_attempt_req_merge(q, q->last_merge, rq)) | |
368 | return true; | |
369 | ||
370 | if (blk_queue_noxmerges(q)) | |
371 | return false; | |
372 | ||
bee0393c | 373 | ret = false; |
5e84ea3a JA |
374 | /* |
375 | * See if our hash lookup can find a potential backmerge. | |
376 | */ | |
bee0393c SL |
377 | while (1) { |
378 | __rq = elv_rqhash_find(q, blk_rq_pos(rq)); | |
379 | if (!__rq || !blk_attempt_req_merge(q, __rq, rq)) | |
380 | break; | |
381 | ||
382 | /* The merged request could be merged with others, try again */ | |
383 | ret = true; | |
384 | rq = __rq; | |
385 | } | |
27419322 | 386 | |
bee0393c | 387 | return ret; |
5e84ea3a JA |
388 | } |
389 | ||
34fe7c05 CH |
390 | void elv_merged_request(struct request_queue *q, struct request *rq, |
391 | enum elv_merge type) | |
1da177e4 | 392 | { |
b374d18a | 393 | struct elevator_queue *e = q->elevator; |
1da177e4 | 394 | |
f9cd4bfe JA |
395 | if (e->type->ops.request_merged) |
396 | e->type->ops.request_merged(q, rq, type); | |
06b86245 | 397 | |
2e662b65 JA |
398 | if (type == ELEVATOR_BACK_MERGE) |
399 | elv_rqhash_reposition(q, rq); | |
9817064b | 400 | |
06b86245 | 401 | q->last_merge = rq; |
1da177e4 LT |
402 | } |
403 | ||
165125e1 | 404 | void elv_merge_requests(struct request_queue *q, struct request *rq, |
1da177e4 LT |
405 | struct request *next) |
406 | { | |
b374d18a | 407 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 408 | |
f9cd4bfe JA |
409 | if (e->type->ops.requests_merged) |
410 | e->type->ops.requests_merged(q, rq, next); | |
06b86245 | 411 | |
9817064b | 412 | elv_rqhash_reposition(q, rq); |
06b86245 | 413 | q->last_merge = rq; |
1da177e4 LT |
414 | } |
415 | ||
165125e1 | 416 | struct request *elv_latter_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.next_request) |
421 | return e->type->ops.next_request(q, rq); | |
bd166ef1 | 422 | |
1da177e4 LT |
423 | return NULL; |
424 | } | |
425 | ||
165125e1 | 426 | struct request *elv_former_request(struct request_queue *q, struct request *rq) |
1da177e4 | 427 | { |
b374d18a | 428 | struct elevator_queue *e = q->elevator; |
1da177e4 | 429 | |
f9cd4bfe JA |
430 | if (e->type->ops.former_request) |
431 | return e->type->ops.former_request(q, rq); | |
bd166ef1 | 432 | |
a1ce35fa | 433 | return NULL; |
1da177e4 LT |
434 | } |
435 | ||
3d1ab40f AV |
436 | #define to_elv(atr) container_of((atr), struct elv_fs_entry, attr) |
437 | ||
438 | static ssize_t | |
439 | elv_attr_show(struct kobject *kobj, struct attribute *attr, char *page) | |
1da177e4 | 440 | { |
3d1ab40f | 441 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 442 | struct elevator_queue *e; |
3d1ab40f AV |
443 | ssize_t error; |
444 | ||
445 | if (!entry->show) | |
446 | return -EIO; | |
447 | ||
b374d18a | 448 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 449 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 450 | error = e->type ? entry->show(e, page) : -ENOENT; |
3d1ab40f AV |
451 | mutex_unlock(&e->sysfs_lock); |
452 | return error; | |
453 | } | |
1da177e4 | 454 | |
3d1ab40f AV |
455 | static ssize_t |
456 | elv_attr_store(struct kobject *kobj, struct attribute *attr, | |
457 | const char *page, size_t length) | |
458 | { | |
3d1ab40f | 459 | struct elv_fs_entry *entry = to_elv(attr); |
b374d18a | 460 | struct elevator_queue *e; |
3d1ab40f | 461 | ssize_t error; |
1da177e4 | 462 | |
3d1ab40f AV |
463 | if (!entry->store) |
464 | return -EIO; | |
1da177e4 | 465 | |
b374d18a | 466 | e = container_of(kobj, struct elevator_queue, kobj); |
3d1ab40f | 467 | mutex_lock(&e->sysfs_lock); |
22f746e2 | 468 | error = e->type ? entry->store(e, page, length) : -ENOENT; |
3d1ab40f AV |
469 | mutex_unlock(&e->sysfs_lock); |
470 | return error; | |
471 | } | |
472 | ||
52cf25d0 | 473 | static const struct sysfs_ops elv_sysfs_ops = { |
3d1ab40f AV |
474 | .show = elv_attr_show, |
475 | .store = elv_attr_store, | |
476 | }; | |
477 | ||
478 | static struct kobj_type elv_ktype = { | |
479 | .sysfs_ops = &elv_sysfs_ops, | |
480 | .release = elevator_release, | |
481 | }; | |
482 | ||
cecf5d87 ML |
483 | /* |
484 | * elv_register_queue is called from either blk_register_queue or | |
485 | * elevator_switch, elevator switch is prevented from being happen | |
486 | * in the two paths, so it is safe to not hold q->sysfs_lock. | |
487 | */ | |
488 | int elv_register_queue(struct request_queue *q, bool uevent) | |
3d1ab40f | 489 | { |
5a5bafdc | 490 | struct elevator_queue *e = q->elevator; |
3d1ab40f AV |
491 | int error; |
492 | ||
b2d6db58 | 493 | error = kobject_add(&e->kobj, &q->kobj, "%s", "iosched"); |
3d1ab40f | 494 | if (!error) { |
22f746e2 | 495 | struct elv_fs_entry *attr = e->type->elevator_attrs; |
3d1ab40f | 496 | if (attr) { |
e572ec7e AV |
497 | while (attr->attr.name) { |
498 | if (sysfs_create_file(&e->kobj, &attr->attr)) | |
3d1ab40f | 499 | break; |
e572ec7e | 500 | attr++; |
3d1ab40f AV |
501 | } |
502 | } | |
cecf5d87 ML |
503 | if (uevent) |
504 | kobject_uevent(&e->kobj, KOBJ_ADD); | |
505 | ||
430c62fb | 506 | e->registered = 1; |
3d1ab40f AV |
507 | } |
508 | return error; | |
1da177e4 | 509 | } |
bc1c1169 | 510 | |
cecf5d87 ML |
511 | /* |
512 | * elv_unregister_queue is called from either blk_unregister_queue or | |
513 | * elevator_switch, elevator switch is prevented from being happen | |
514 | * in the two paths, so it is safe to not hold q->sysfs_lock. | |
515 | */ | |
1da177e4 LT |
516 | void elv_unregister_queue(struct request_queue *q) |
517 | { | |
f8fc877d TH |
518 | if (q) { |
519 | struct elevator_queue *e = q->elevator; | |
520 | ||
521 | kobject_uevent(&e->kobj, KOBJ_REMOVE); | |
522 | kobject_del(&e->kobj); | |
cecf5d87 | 523 | |
f8fc877d | 524 | e->registered = 0; |
8330cdb0 JK |
525 | /* Re-enable throttling in case elevator disabled it */ |
526 | wbt_enable_default(q); | |
f8fc877d | 527 | } |
1da177e4 LT |
528 | } |
529 | ||
e567bf71 | 530 | int elv_register(struct elevator_type *e) |
1da177e4 | 531 | { |
3d3c2379 TH |
532 | /* create icq_cache if requested */ |
533 | if (e->icq_size) { | |
534 | if (WARN_ON(e->icq_size < sizeof(struct io_cq)) || | |
535 | WARN_ON(e->icq_align < __alignof__(struct io_cq))) | |
536 | return -EINVAL; | |
537 | ||
538 | snprintf(e->icq_cache_name, sizeof(e->icq_cache_name), | |
539 | "%s_io_cq", e->elevator_name); | |
540 | e->icq_cache = kmem_cache_create(e->icq_cache_name, e->icq_size, | |
541 | e->icq_align, 0, NULL); | |
542 | if (!e->icq_cache) | |
543 | return -ENOMEM; | |
544 | } | |
545 | ||
546 | /* register, don't allow duplicate names */ | |
2a12dcd7 | 547 | spin_lock(&elv_list_lock); |
68c43f13 | 548 | if (elevator_find(e->elevator_name, 0)) { |
3d3c2379 | 549 | spin_unlock(&elv_list_lock); |
62d2a194 | 550 | kmem_cache_destroy(e->icq_cache); |
3d3c2379 TH |
551 | return -EBUSY; |
552 | } | |
1da177e4 | 553 | list_add_tail(&e->list, &elv_list); |
2a12dcd7 | 554 | spin_unlock(&elv_list_lock); |
1da177e4 | 555 | |
d0b0a81a HT |
556 | printk(KERN_INFO "io scheduler %s registered\n", e->elevator_name); |
557 | ||
3d3c2379 | 558 | return 0; |
1da177e4 LT |
559 | } |
560 | EXPORT_SYMBOL_GPL(elv_register); | |
561 | ||
562 | void elv_unregister(struct elevator_type *e) | |
563 | { | |
3d3c2379 | 564 | /* unregister */ |
2a12dcd7 | 565 | spin_lock(&elv_list_lock); |
1da177e4 | 566 | list_del_init(&e->list); |
2a12dcd7 | 567 | spin_unlock(&elv_list_lock); |
3d3c2379 TH |
568 | |
569 | /* | |
570 | * Destroy icq_cache if it exists. icq's are RCU managed. Make | |
571 | * sure all RCU operations are complete before proceeding. | |
572 | */ | |
573 | if (e->icq_cache) { | |
574 | rcu_barrier(); | |
575 | kmem_cache_destroy(e->icq_cache); | |
576 | e->icq_cache = NULL; | |
577 | } | |
1da177e4 LT |
578 | } |
579 | EXPORT_SYMBOL_GPL(elv_unregister); | |
580 | ||
d48ece20 | 581 | int elevator_switch_mq(struct request_queue *q, |
54d5329d OS |
582 | struct elevator_type *new_e) |
583 | { | |
584 | int ret; | |
585 | ||
14a23498 BVA |
586 | lockdep_assert_held(&q->sysfs_lock); |
587 | ||
54d5329d | 588 | if (q->elevator) { |
b89f625e | 589 | if (q->elevator->registered) |
54d5329d | 590 | elv_unregister_queue(q); |
cecf5d87 | 591 | |
54d5329d OS |
592 | ioc_clear_queue(q); |
593 | elevator_exit(q, q->elevator); | |
594 | } | |
595 | ||
596 | ret = blk_mq_init_sched(q, new_e); | |
597 | if (ret) | |
598 | goto out; | |
599 | ||
600 | if (new_e) { | |
cecf5d87 | 601 | ret = elv_register_queue(q, true); |
54d5329d OS |
602 | if (ret) { |
603 | elevator_exit(q, q->elevator); | |
604 | goto out; | |
605 | } | |
606 | } | |
607 | ||
608 | if (new_e) | |
609 | blk_add_trace_msg(q, "elv switch: %s", new_e->elevator_name); | |
610 | else | |
611 | blk_add_trace_msg(q, "elv switch: none"); | |
612 | ||
613 | out: | |
54d5329d | 614 | return ret; |
54d5329d OS |
615 | } |
616 | ||
61db437d DLM |
617 | static inline bool elv_support_iosched(struct request_queue *q) |
618 | { | |
619 | if (q->tag_set && (q->tag_set->flags & BLK_MQ_F_NO_SCHED)) | |
620 | return false; | |
621 | return true; | |
622 | } | |
623 | ||
131d08e1 | 624 | /* |
a0958ba7 DLM |
625 | * For single queue devices, default to using mq-deadline. If we have multiple |
626 | * queues or mq-deadline is not available, default to "none". | |
627 | */ | |
628 | static struct elevator_type *elevator_get_default(struct request_queue *q) | |
629 | { | |
630 | if (q->nr_hw_queues != 1) | |
631 | return NULL; | |
632 | ||
633 | return elevator_get(q, "mq-deadline", false); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Get the first elevator providing the features required by the request queue. | |
638 | * Default to "none" if no matching elevator is found. | |
639 | */ | |
640 | static struct elevator_type *elevator_get_by_features(struct request_queue *q) | |
641 | { | |
a2614255 | 642 | struct elevator_type *e, *found = NULL; |
a0958ba7 DLM |
643 | |
644 | spin_lock(&elv_list_lock); | |
645 | ||
646 | list_for_each_entry(e, &elv_list, list) { | |
647 | if (elv_support_features(e->elevator_features, | |
a2614255 JA |
648 | q->required_elevator_features)) { |
649 | found = e; | |
a0958ba7 | 650 | break; |
a2614255 | 651 | } |
a0958ba7 DLM |
652 | } |
653 | ||
a2614255 JA |
654 | if (found && !try_module_get(found->elevator_owner)) |
655 | found = NULL; | |
a0958ba7 DLM |
656 | |
657 | spin_unlock(&elv_list_lock); | |
a2614255 | 658 | return found; |
a0958ba7 DLM |
659 | } |
660 | ||
661 | /* | |
662 | * For a device queue that has no required features, use the default elevator | |
663 | * settings. Otherwise, use the first elevator available matching the required | |
664 | * features. If no suitable elevator is find or if the chosen elevator | |
665 | * initialization fails, fall back to the "none" elevator (no elevator). | |
131d08e1 | 666 | */ |
954b4a5c | 667 | void elevator_init_mq(struct request_queue *q) |
131d08e1 CH |
668 | { |
669 | struct elevator_type *e; | |
954b4a5c | 670 | int err; |
131d08e1 | 671 | |
61db437d | 672 | if (!elv_support_iosched(q)) |
954b4a5c | 673 | return; |
61db437d | 674 | |
c48dac13 ML |
675 | WARN_ON_ONCE(test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags)); |
676 | ||
131d08e1 | 677 | if (unlikely(q->elevator)) |
954b4a5c | 678 | return; |
131d08e1 | 679 | |
a0958ba7 DLM |
680 | if (!q->required_elevator_features) |
681 | e = elevator_get_default(q); | |
682 | else | |
683 | e = elevator_get_by_features(q); | |
131d08e1 | 684 | if (!e) |
954b4a5c | 685 | return; |
131d08e1 | 686 | |
737eb78e DLM |
687 | blk_mq_freeze_queue(q); |
688 | blk_mq_quiesce_queue(q); | |
689 | ||
131d08e1 | 690 | err = blk_mq_init_sched(q, e); |
737eb78e DLM |
691 | |
692 | blk_mq_unquiesce_queue(q); | |
693 | blk_mq_unfreeze_queue(q); | |
694 | ||
954b4a5c DLM |
695 | if (err) { |
696 | pr_warn("\"%s\" elevator initialization failed, " | |
697 | "falling back to \"none\"\n", e->elevator_name); | |
131d08e1 | 698 | elevator_put(e); |
954b4a5c | 699 | } |
131d08e1 CH |
700 | } |
701 | ||
702 | ||
1da177e4 LT |
703 | /* |
704 | * switch to new_e io scheduler. be careful not to introduce deadlocks - | |
705 | * we don't free the old io scheduler, before we have allocated what we | |
706 | * need for the new one. this way we have a chance of going back to the old | |
cb98fc8b | 707 | * one, if the new one fails init for some reason. |
1da177e4 | 708 | */ |
165125e1 | 709 | static int elevator_switch(struct request_queue *q, struct elevator_type *new_e) |
1da177e4 | 710 | { |
e8989fae | 711 | int err; |
1da177e4 | 712 | |
14a23498 BVA |
713 | lockdep_assert_held(&q->sysfs_lock); |
714 | ||
a1ce35fa JA |
715 | blk_mq_freeze_queue(q); |
716 | blk_mq_quiesce_queue(q); | |
4722dc52 | 717 | |
a1ce35fa | 718 | err = elevator_switch_mq(q, new_e); |
1da177e4 | 719 | |
a1ce35fa JA |
720 | blk_mq_unquiesce_queue(q); |
721 | blk_mq_unfreeze_queue(q); | |
75ad23bc | 722 | |
5dd531a0 | 723 | return err; |
1da177e4 LT |
724 | } |
725 | ||
5dd531a0 JA |
726 | /* |
727 | * Switch this queue to the given IO scheduler. | |
728 | */ | |
7c8a3679 | 729 | static int __elevator_change(struct request_queue *q, const char *name) |
1da177e4 LT |
730 | { |
731 | char elevator_name[ELV_NAME_MAX]; | |
732 | struct elevator_type *e; | |
733 | ||
e9a823fb | 734 | /* Make sure queue is not in the middle of being removed */ |
58c898ba | 735 | if (!blk_queue_registered(q)) |
e9a823fb DJ |
736 | return -ENOENT; |
737 | ||
bd166ef1 JA |
738 | /* |
739 | * Special case for mq, turn off scheduling | |
740 | */ | |
fbd72127 AZ |
741 | if (!strncmp(name, "none", 4)) { |
742 | if (!q->elevator) | |
743 | return 0; | |
bd166ef1 | 744 | return elevator_switch(q, NULL); |
fbd72127 | 745 | } |
cd43e26f | 746 | |
ee2e992c | 747 | strlcpy(elevator_name, name, sizeof(elevator_name)); |
2527d997 | 748 | e = elevator_get(q, strstrip(elevator_name), true); |
340ff321 | 749 | if (!e) |
1da177e4 | 750 | return -EINVAL; |
1da177e4 | 751 | |
68c43f13 DLM |
752 | if (q->elevator && |
753 | elevator_match(q->elevator->type, elevator_name, 0)) { | |
2ca7d93b | 754 | elevator_put(e); |
5dd531a0 | 755 | return 0; |
2ca7d93b | 756 | } |
1da177e4 | 757 | |
5dd531a0 JA |
758 | return elevator_switch(q, e); |
759 | } | |
7c8a3679 | 760 | |
5dd531a0 JA |
761 | ssize_t elv_iosched_store(struct request_queue *q, const char *name, |
762 | size_t count) | |
763 | { | |
764 | int ret; | |
765 | ||
344e9ffc | 766 | if (!queue_is_mq(q) || !elv_support_iosched(q)) |
5dd531a0 JA |
767 | return count; |
768 | ||
7c8a3679 | 769 | ret = __elevator_change(q, name); |
5dd531a0 JA |
770 | if (!ret) |
771 | return count; | |
772 | ||
5dd531a0 | 773 | return ret; |
1da177e4 LT |
774 | } |
775 | ||
165125e1 | 776 | ssize_t elv_iosched_show(struct request_queue *q, char *name) |
1da177e4 | 777 | { |
b374d18a | 778 | struct elevator_queue *e = q->elevator; |
bd166ef1 | 779 | struct elevator_type *elv = NULL; |
70cee26e | 780 | struct elevator_type *__e; |
1da177e4 LT |
781 | int len = 0; |
782 | ||
344e9ffc | 783 | if (!queue_is_mq(q)) |
cd43e26f MP |
784 | return sprintf(name, "none\n"); |
785 | ||
bd166ef1 JA |
786 | if (!q->elevator) |
787 | len += sprintf(name+len, "[none] "); | |
788 | else | |
789 | elv = e->type; | |
cd43e26f | 790 | |
2a12dcd7 | 791 | spin_lock(&elv_list_lock); |
70cee26e | 792 | list_for_each_entry(__e, &elv_list, list) { |
68c43f13 | 793 | if (elv && elevator_match(elv, __e->elevator_name, 0)) { |
1da177e4 | 794 | len += sprintf(name+len, "[%s] ", elv->elevator_name); |
bd166ef1 JA |
795 | continue; |
796 | } | |
68c43f13 DLM |
797 | if (elv_support_iosched(q) && |
798 | elevator_match(__e, __e->elevator_name, | |
799 | q->required_elevator_features)) | |
1da177e4 LT |
800 | len += sprintf(name+len, "%s ", __e->elevator_name); |
801 | } | |
2a12dcd7 | 802 | spin_unlock(&elv_list_lock); |
1da177e4 | 803 | |
344e9ffc | 804 | if (q->elevator) |
bd166ef1 JA |
805 | len += sprintf(name+len, "none"); |
806 | ||
1da177e4 LT |
807 | len += sprintf(len+name, "\n"); |
808 | return len; | |
809 | } | |
810 | ||
165125e1 JA |
811 | struct request *elv_rb_former_request(struct request_queue *q, |
812 | struct request *rq) | |
2e662b65 JA |
813 | { |
814 | struct rb_node *rbprev = rb_prev(&rq->rb_node); | |
815 | ||
816 | if (rbprev) | |
817 | return rb_entry_rq(rbprev); | |
818 | ||
819 | return NULL; | |
820 | } | |
2e662b65 JA |
821 | EXPORT_SYMBOL(elv_rb_former_request); |
822 | ||
165125e1 JA |
823 | struct request *elv_rb_latter_request(struct request_queue *q, |
824 | struct request *rq) | |
2e662b65 JA |
825 | { |
826 | struct rb_node *rbnext = rb_next(&rq->rb_node); | |
827 | ||
828 | if (rbnext) | |
829 | return rb_entry_rq(rbnext); | |
830 | ||
831 | return NULL; | |
832 | } | |
2e662b65 | 833 | EXPORT_SYMBOL(elv_rb_latter_request); |