]>
Commit | Line | Data |
---|---|---|
bd166ef1 JA |
1 | /* |
2 | * blk-mq scheduling framework | |
3 | * | |
4 | * Copyright (C) 2016 Jens Axboe | |
5 | */ | |
6 | #include <linux/kernel.h> | |
7 | #include <linux/module.h> | |
8 | #include <linux/blk-mq.h> | |
9 | ||
10 | #include <trace/events/block.h> | |
11 | ||
12 | #include "blk.h" | |
13 | #include "blk-mq.h" | |
d332ce09 | 14 | #include "blk-mq-debugfs.h" |
bd166ef1 JA |
15 | #include "blk-mq-sched.h" |
16 | #include "blk-mq-tag.h" | |
17 | #include "blk-wbt.h" | |
18 | ||
19 | void blk_mq_sched_free_hctx_data(struct request_queue *q, | |
20 | void (*exit)(struct blk_mq_hw_ctx *)) | |
21 | { | |
22 | struct blk_mq_hw_ctx *hctx; | |
23 | int i; | |
24 | ||
25 | queue_for_each_hw_ctx(q, hctx, i) { | |
26 | if (exit && hctx->sched_data) | |
27 | exit(hctx); | |
28 | kfree(hctx->sched_data); | |
29 | hctx->sched_data = NULL; | |
30 | } | |
31 | } | |
32 | EXPORT_SYMBOL_GPL(blk_mq_sched_free_hctx_data); | |
33 | ||
bd166ef1 | 34 | static void __blk_mq_sched_assign_ioc(struct request_queue *q, |
f1ba8261 PV |
35 | struct request *rq, |
36 | struct bio *bio, | |
37 | struct io_context *ioc) | |
bd166ef1 | 38 | { |
ea511e3c | 39 | struct elevator_queue *e = q->elevator; |
bd166ef1 JA |
40 | struct io_cq *icq; |
41 | ||
42 | spin_lock_irq(q->queue_lock); | |
43 | icq = ioc_lookup_icq(ioc, q); | |
44 | spin_unlock_irq(q->queue_lock); | |
45 | ||
46 | if (!icq) { | |
47 | icq = ioc_create_icq(ioc, q, GFP_ATOMIC); | |
48 | if (!icq) | |
49 | return; | |
50 | } | |
51 | ||
52 | rq->elv.icq = icq; | |
ea511e3c CH |
53 | if (e && e->type->ops.mq.get_rq_priv && |
54 | e->type->ops.mq.get_rq_priv(q, rq, bio)) { | |
55 | rq->elv.icq = NULL; | |
bd166ef1 JA |
56 | return; |
57 | } | |
58 | ||
ea511e3c CH |
59 | rq->rq_flags |= RQF_ELVPRIV; |
60 | get_io_context(icq->ioc); | |
bd166ef1 JA |
61 | } |
62 | ||
d2c0d383 CH |
63 | void blk_mq_sched_assign_ioc(struct request_queue *q, struct request *rq, |
64 | struct bio *bio) | |
bd166ef1 JA |
65 | { |
66 | struct io_context *ioc; | |
67 | ||
68 | ioc = rq_ioc(bio); | |
69 | if (ioc) | |
f1ba8261 | 70 | __blk_mq_sched_assign_ioc(q, rq, bio, ioc); |
bd166ef1 JA |
71 | } |
72 | ||
bd166ef1 JA |
73 | void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) |
74 | { | |
81380ca1 OS |
75 | struct request_queue *q = hctx->queue; |
76 | struct elevator_queue *e = q->elevator; | |
64765a75 JA |
77 | const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request; |
78 | bool did_work = false; | |
bd166ef1 JA |
79 | LIST_HEAD(rq_list); |
80 | ||
81 | if (unlikely(blk_mq_hctx_stopped(hctx))) | |
82 | return; | |
83 | ||
84 | hctx->run++; | |
85 | ||
86 | /* | |
87 | * If we have previous entries on our dispatch list, grab them first for | |
88 | * more fair dispatch. | |
89 | */ | |
90 | if (!list_empty_careful(&hctx->dispatch)) { | |
91 | spin_lock(&hctx->lock); | |
92 | if (!list_empty(&hctx->dispatch)) | |
93 | list_splice_init(&hctx->dispatch, &rq_list); | |
94 | spin_unlock(&hctx->lock); | |
95 | } | |
96 | ||
97 | /* | |
98 | * Only ask the scheduler for requests, if we didn't have residual | |
99 | * requests from the dispatch list. This is to avoid the case where | |
100 | * we only ever dispatch a fraction of the requests available because | |
101 | * of low device queue depth. Once we pull requests out of the IO | |
102 | * scheduler, we can no longer merge or sort them. So it's best to | |
103 | * leave them there for as long as we can. Mark the hw queue as | |
104 | * needing a restart in that case. | |
105 | */ | |
c13660a0 | 106 | if (!list_empty(&rq_list)) { |
d38d3515 | 107 | blk_mq_sched_mark_restart_hctx(hctx); |
81380ca1 | 108 | did_work = blk_mq_dispatch_rq_list(q, &rq_list); |
64765a75 | 109 | } else if (!has_sched_dispatch) { |
c13660a0 | 110 | blk_mq_flush_busy_ctxs(hctx, &rq_list); |
81380ca1 | 111 | blk_mq_dispatch_rq_list(q, &rq_list); |
64765a75 JA |
112 | } |
113 | ||
114 | /* | |
115 | * We want to dispatch from the scheduler if we had no work left | |
116 | * on the dispatch list, OR if we did have work but weren't able | |
117 | * to make progress. | |
118 | */ | |
119 | if (!did_work && has_sched_dispatch) { | |
c13660a0 JA |
120 | do { |
121 | struct request *rq; | |
122 | ||
123 | rq = e->type->ops.mq.dispatch_request(hctx); | |
124 | if (!rq) | |
125 | break; | |
126 | list_add(&rq->queuelist, &rq_list); | |
81380ca1 | 127 | } while (blk_mq_dispatch_rq_list(q, &rq_list)); |
c13660a0 | 128 | } |
bd166ef1 JA |
129 | } |
130 | ||
e4d750c9 JA |
131 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, |
132 | struct request **merged_request) | |
bd166ef1 JA |
133 | { |
134 | struct request *rq; | |
bd166ef1 | 135 | |
34fe7c05 CH |
136 | switch (elv_merge(q, &rq, bio)) { |
137 | case ELEVATOR_BACK_MERGE: | |
bd166ef1 JA |
138 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
139 | return false; | |
34fe7c05 CH |
140 | if (!bio_attempt_back_merge(q, rq, bio)) |
141 | return false; | |
142 | *merged_request = attempt_back_merge(q, rq); | |
143 | if (!*merged_request) | |
144 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
145 | return true; | |
146 | case ELEVATOR_FRONT_MERGE: | |
bd166ef1 JA |
147 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
148 | return false; | |
34fe7c05 CH |
149 | if (!bio_attempt_front_merge(q, rq, bio)) |
150 | return false; | |
151 | *merged_request = attempt_front_merge(q, rq); | |
152 | if (!*merged_request) | |
153 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
154 | return true; | |
155 | default: | |
156 | return false; | |
bd166ef1 | 157 | } |
bd166ef1 JA |
158 | } |
159 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); | |
160 | ||
9bddeb2a ML |
161 | /* |
162 | * Reverse check our software queue for entries that we could potentially | |
163 | * merge with. Currently includes a hand-wavy stop count of 8, to not spend | |
164 | * too much time checking for merges. | |
165 | */ | |
166 | static bool blk_mq_attempt_merge(struct request_queue *q, | |
167 | struct blk_mq_ctx *ctx, struct bio *bio) | |
168 | { | |
169 | struct request *rq; | |
170 | int checked = 8; | |
171 | ||
172 | list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) { | |
173 | bool merged = false; | |
174 | ||
175 | if (!checked--) | |
176 | break; | |
177 | ||
178 | if (!blk_rq_merge_ok(rq, bio)) | |
179 | continue; | |
180 | ||
181 | switch (blk_try_merge(rq, bio)) { | |
182 | case ELEVATOR_BACK_MERGE: | |
183 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
184 | merged = bio_attempt_back_merge(q, rq, bio); | |
185 | break; | |
186 | case ELEVATOR_FRONT_MERGE: | |
187 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
188 | merged = bio_attempt_front_merge(q, rq, bio); | |
189 | break; | |
190 | case ELEVATOR_DISCARD_MERGE: | |
191 | merged = bio_attempt_discard_merge(q, rq, bio); | |
192 | break; | |
193 | default: | |
194 | continue; | |
195 | } | |
196 | ||
197 | if (merged) | |
198 | ctx->rq_merged++; | |
199 | return merged; | |
200 | } | |
201 | ||
202 | return false; | |
203 | } | |
204 | ||
bd166ef1 JA |
205 | bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) |
206 | { | |
207 | struct elevator_queue *e = q->elevator; | |
9bddeb2a ML |
208 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
209 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
210 | bool ret = false; | |
bd166ef1 | 211 | |
9bddeb2a | 212 | if (e && e->type->ops.mq.bio_merge) { |
bd166ef1 JA |
213 | blk_mq_put_ctx(ctx); |
214 | return e->type->ops.mq.bio_merge(hctx, bio); | |
215 | } | |
216 | ||
9bddeb2a ML |
217 | if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) { |
218 | /* default per sw-queue merge */ | |
219 | spin_lock(&ctx->lock); | |
220 | ret = blk_mq_attempt_merge(q, ctx, bio); | |
221 | spin_unlock(&ctx->lock); | |
222 | } | |
223 | ||
224 | blk_mq_put_ctx(ctx); | |
225 | return ret; | |
bd166ef1 JA |
226 | } |
227 | ||
228 | bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) | |
229 | { | |
230 | return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); | |
231 | } | |
232 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); | |
233 | ||
234 | void blk_mq_sched_request_inserted(struct request *rq) | |
235 | { | |
236 | trace_block_rq_insert(rq->q, rq); | |
237 | } | |
238 | EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); | |
239 | ||
0cacba6c OS |
240 | static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, |
241 | struct request *rq) | |
bd166ef1 JA |
242 | { |
243 | if (rq->tag == -1) { | |
244 | rq->rq_flags |= RQF_SORTED; | |
245 | return false; | |
246 | } | |
247 | ||
248 | /* | |
249 | * If we already have a real request tag, send directly to | |
250 | * the dispatch list. | |
251 | */ | |
252 | spin_lock(&hctx->lock); | |
253 | list_add(&rq->queuelist, &hctx->dispatch); | |
254 | spin_unlock(&hctx->lock); | |
255 | return true; | |
256 | } | |
bd166ef1 | 257 | |
6d8c6c0f | 258 | static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx) |
50e1dab8 JA |
259 | { |
260 | if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) { | |
261 | clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); | |
6d8c6c0f | 262 | if (blk_mq_hctx_has_pending(hctx)) { |
50e1dab8 | 263 | blk_mq_run_hw_queue(hctx, true); |
6d8c6c0f BVA |
264 | return true; |
265 | } | |
50e1dab8 | 266 | } |
6d8c6c0f | 267 | return false; |
50e1dab8 JA |
268 | } |
269 | ||
6d8c6c0f BVA |
270 | /** |
271 | * list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list | |
272 | * @pos: loop cursor. | |
273 | * @skip: the list element that will not be examined. Iteration starts at | |
274 | * @skip->next. | |
275 | * @head: head of the list to examine. This list must have at least one | |
276 | * element, namely @skip. | |
277 | * @member: name of the list_head structure within typeof(*pos). | |
278 | */ | |
279 | #define list_for_each_entry_rcu_rr(pos, skip, head, member) \ | |
280 | for ((pos) = (skip); \ | |
281 | (pos = (pos)->member.next != (head) ? list_entry_rcu( \ | |
282 | (pos)->member.next, typeof(*pos), member) : \ | |
283 | list_entry_rcu((pos)->member.next->next, typeof(*pos), member)), \ | |
284 | (pos) != (skip); ) | |
50e1dab8 | 285 | |
6d8c6c0f BVA |
286 | /* |
287 | * Called after a driver tag has been freed to check whether a hctx needs to | |
288 | * be restarted. Restarts @hctx if its tag set is not shared. Restarts hardware | |
289 | * queues in a round-robin fashion if the tag set of @hctx is shared with other | |
290 | * hardware queues. | |
291 | */ | |
292 | void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx) | |
293 | { | |
294 | struct blk_mq_tags *const tags = hctx->tags; | |
295 | struct blk_mq_tag_set *const set = hctx->queue->tag_set; | |
296 | struct request_queue *const queue = hctx->queue, *q; | |
297 | struct blk_mq_hw_ctx *hctx2; | |
298 | unsigned int i, j; | |
299 | ||
300 | if (set->flags & BLK_MQ_F_TAG_SHARED) { | |
301 | rcu_read_lock(); | |
302 | list_for_each_entry_rcu_rr(q, queue, &set->tag_list, | |
303 | tag_set_list) { | |
304 | queue_for_each_hw_ctx(q, hctx2, i) | |
305 | if (hctx2->tags == tags && | |
306 | blk_mq_sched_restart_hctx(hctx2)) | |
307 | goto done; | |
308 | } | |
309 | j = hctx->queue_num + 1; | |
310 | for (i = 0; i < queue->nr_hw_queues; i++, j++) { | |
311 | if (j == queue->nr_hw_queues) | |
312 | j = 0; | |
313 | hctx2 = queue->queue_hw_ctx[j]; | |
314 | if (hctx2->tags == tags && | |
315 | blk_mq_sched_restart_hctx(hctx2)) | |
316 | break; | |
d38d3515 | 317 | } |
6d8c6c0f BVA |
318 | done: |
319 | rcu_read_unlock(); | |
d38d3515 | 320 | } else { |
50e1dab8 | 321 | blk_mq_sched_restart_hctx(hctx); |
50e1dab8 JA |
322 | } |
323 | } | |
324 | ||
bd6737f1 JA |
325 | /* |
326 | * Add flush/fua to the queue. If we fail getting a driver tag, then | |
327 | * punt to the requeue list. Requeue will re-invoke us from a context | |
328 | * that's safe to block from. | |
329 | */ | |
330 | static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx, | |
331 | struct request *rq, bool can_block) | |
332 | { | |
333 | if (blk_mq_get_driver_tag(rq, &hctx, can_block)) { | |
334 | blk_insert_flush(rq); | |
335 | blk_mq_run_hw_queue(hctx, true); | |
336 | } else | |
c7a571b4 | 337 | blk_mq_add_to_requeue_list(rq, false, true); |
bd6737f1 JA |
338 | } |
339 | ||
340 | void blk_mq_sched_insert_request(struct request *rq, bool at_head, | |
341 | bool run_queue, bool async, bool can_block) | |
342 | { | |
343 | struct request_queue *q = rq->q; | |
344 | struct elevator_queue *e = q->elevator; | |
345 | struct blk_mq_ctx *ctx = rq->mq_ctx; | |
346 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
347 | ||
f3a8ab7d | 348 | if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) { |
bd6737f1 JA |
349 | blk_mq_sched_insert_flush(hctx, rq, can_block); |
350 | return; | |
351 | } | |
352 | ||
0cacba6c OS |
353 | if (e && blk_mq_sched_bypass_insert(hctx, rq)) |
354 | goto run; | |
355 | ||
bd6737f1 JA |
356 | if (e && e->type->ops.mq.insert_requests) { |
357 | LIST_HEAD(list); | |
358 | ||
359 | list_add(&rq->queuelist, &list); | |
360 | e->type->ops.mq.insert_requests(hctx, &list, at_head); | |
361 | } else { | |
362 | spin_lock(&ctx->lock); | |
363 | __blk_mq_insert_request(hctx, rq, at_head); | |
364 | spin_unlock(&ctx->lock); | |
365 | } | |
366 | ||
0cacba6c | 367 | run: |
bd6737f1 JA |
368 | if (run_queue) |
369 | blk_mq_run_hw_queue(hctx, async); | |
370 | } | |
371 | ||
372 | void blk_mq_sched_insert_requests(struct request_queue *q, | |
373 | struct blk_mq_ctx *ctx, | |
374 | struct list_head *list, bool run_queue_async) | |
375 | { | |
376 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
377 | struct elevator_queue *e = hctx->queue->elevator; | |
378 | ||
0cacba6c OS |
379 | if (e) { |
380 | struct request *rq, *next; | |
381 | ||
382 | /* | |
383 | * We bypass requests that already have a driver tag assigned, | |
384 | * which should only be flushes. Flushes are only ever inserted | |
385 | * as single requests, so we shouldn't ever hit the | |
386 | * WARN_ON_ONCE() below (but let's handle it just in case). | |
387 | */ | |
388 | list_for_each_entry_safe(rq, next, list, queuelist) { | |
389 | if (WARN_ON_ONCE(rq->tag != -1)) { | |
390 | list_del_init(&rq->queuelist); | |
391 | blk_mq_sched_bypass_insert(hctx, rq); | |
392 | } | |
393 | } | |
394 | } | |
395 | ||
bd6737f1 JA |
396 | if (e && e->type->ops.mq.insert_requests) |
397 | e->type->ops.mq.insert_requests(hctx, list, false); | |
398 | else | |
399 | blk_mq_insert_requests(hctx, ctx, list); | |
400 | ||
401 | blk_mq_run_hw_queue(hctx, run_queue_async); | |
402 | } | |
403 | ||
bd166ef1 JA |
404 | static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, |
405 | struct blk_mq_hw_ctx *hctx, | |
406 | unsigned int hctx_idx) | |
407 | { | |
408 | if (hctx->sched_tags) { | |
409 | blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); | |
410 | blk_mq_free_rq_map(hctx->sched_tags); | |
411 | hctx->sched_tags = NULL; | |
412 | } | |
413 | } | |
414 | ||
6917ff0b OS |
415 | static int blk_mq_sched_alloc_tags(struct request_queue *q, |
416 | struct blk_mq_hw_ctx *hctx, | |
417 | unsigned int hctx_idx) | |
418 | { | |
419 | struct blk_mq_tag_set *set = q->tag_set; | |
420 | int ret; | |
421 | ||
422 | hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, | |
423 | set->reserved_tags); | |
424 | if (!hctx->sched_tags) | |
425 | return -ENOMEM; | |
426 | ||
427 | ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); | |
428 | if (ret) | |
429 | blk_mq_sched_free_tags(set, hctx, hctx_idx); | |
430 | ||
431 | return ret; | |
432 | } | |
433 | ||
54d5329d | 434 | static void blk_mq_sched_tags_teardown(struct request_queue *q) |
bd166ef1 JA |
435 | { |
436 | struct blk_mq_tag_set *set = q->tag_set; | |
437 | struct blk_mq_hw_ctx *hctx; | |
6917ff0b OS |
438 | int i; |
439 | ||
440 | queue_for_each_hw_ctx(q, hctx, i) | |
441 | blk_mq_sched_free_tags(set, hctx, i); | |
442 | } | |
443 | ||
93252632 OS |
444 | int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, |
445 | unsigned int hctx_idx) | |
446 | { | |
447 | struct elevator_queue *e = q->elevator; | |
ee056f98 | 448 | int ret; |
93252632 OS |
449 | |
450 | if (!e) | |
451 | return 0; | |
452 | ||
ee056f98 OS |
453 | ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx); |
454 | if (ret) | |
455 | return ret; | |
456 | ||
457 | if (e->type->ops.mq.init_hctx) { | |
458 | ret = e->type->ops.mq.init_hctx(hctx, hctx_idx); | |
459 | if (ret) { | |
460 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); | |
461 | return ret; | |
462 | } | |
463 | } | |
464 | ||
d332ce09 OS |
465 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
466 | ||
ee056f98 | 467 | return 0; |
93252632 OS |
468 | } |
469 | ||
470 | void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, | |
471 | unsigned int hctx_idx) | |
472 | { | |
473 | struct elevator_queue *e = q->elevator; | |
474 | ||
475 | if (!e) | |
476 | return; | |
477 | ||
d332ce09 OS |
478 | blk_mq_debugfs_unregister_sched_hctx(hctx); |
479 | ||
ee056f98 OS |
480 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { |
481 | e->type->ops.mq.exit_hctx(hctx, hctx_idx); | |
482 | hctx->sched_data = NULL; | |
483 | } | |
484 | ||
93252632 OS |
485 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); |
486 | } | |
487 | ||
6917ff0b OS |
488 | int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) |
489 | { | |
490 | struct blk_mq_hw_ctx *hctx; | |
ee056f98 | 491 | struct elevator_queue *eq; |
6917ff0b OS |
492 | unsigned int i; |
493 | int ret; | |
494 | ||
495 | if (!e) { | |
496 | q->elevator = NULL; | |
497 | return 0; | |
498 | } | |
bd166ef1 JA |
499 | |
500 | /* | |
501 | * Default to 256, since we don't split into sync/async like the | |
502 | * old code did. Additionally, this is a per-hw queue depth. | |
503 | */ | |
504 | q->nr_requests = 2 * BLKDEV_MAX_RQ; | |
505 | ||
bd166ef1 | 506 | queue_for_each_hw_ctx(q, hctx, i) { |
6917ff0b | 507 | ret = blk_mq_sched_alloc_tags(q, hctx, i); |
bd166ef1 | 508 | if (ret) |
6917ff0b | 509 | goto err; |
bd166ef1 JA |
510 | } |
511 | ||
6917ff0b OS |
512 | ret = e->ops.mq.init_sched(q, e); |
513 | if (ret) | |
514 | goto err; | |
bd166ef1 | 515 | |
d332ce09 OS |
516 | blk_mq_debugfs_register_sched(q); |
517 | ||
518 | queue_for_each_hw_ctx(q, hctx, i) { | |
519 | if (e->ops.mq.init_hctx) { | |
ee056f98 OS |
520 | ret = e->ops.mq.init_hctx(hctx, i); |
521 | if (ret) { | |
522 | eq = q->elevator; | |
523 | blk_mq_exit_sched(q, eq); | |
524 | kobject_put(&eq->kobj); | |
525 | return ret; | |
526 | } | |
527 | } | |
d332ce09 | 528 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
ee056f98 OS |
529 | } |
530 | ||
bd166ef1 | 531 | return 0; |
bd166ef1 | 532 | |
6917ff0b | 533 | err: |
54d5329d OS |
534 | blk_mq_sched_tags_teardown(q); |
535 | q->elevator = NULL; | |
6917ff0b | 536 | return ret; |
bd166ef1 | 537 | } |
d3484991 | 538 | |
54d5329d OS |
539 | void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) |
540 | { | |
ee056f98 OS |
541 | struct blk_mq_hw_ctx *hctx; |
542 | unsigned int i; | |
543 | ||
d332ce09 OS |
544 | queue_for_each_hw_ctx(q, hctx, i) { |
545 | blk_mq_debugfs_unregister_sched_hctx(hctx); | |
546 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { | |
547 | e->type->ops.mq.exit_hctx(hctx, i); | |
548 | hctx->sched_data = NULL; | |
ee056f98 OS |
549 | } |
550 | } | |
d332ce09 | 551 | blk_mq_debugfs_unregister_sched(q); |
54d5329d OS |
552 | if (e->type->ops.mq.exit_sched) |
553 | e->type->ops.mq.exit_sched(e); | |
554 | blk_mq_sched_tags_teardown(q); | |
555 | q->elevator = NULL; | |
556 | } | |
557 | ||
d3484991 JA |
558 | int blk_mq_sched_init(struct request_queue *q) |
559 | { | |
560 | int ret; | |
561 | ||
d3484991 JA |
562 | mutex_lock(&q->sysfs_lock); |
563 | ret = elevator_init(q, NULL); | |
564 | mutex_unlock(&q->sysfs_lock); | |
565 | ||
566 | return ret; | |
567 | } |