]>
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 JA |
38 | { |
39 | struct io_cq *icq; | |
40 | ||
41 | spin_lock_irq(q->queue_lock); | |
42 | icq = ioc_lookup_icq(ioc, q); | |
43 | spin_unlock_irq(q->queue_lock); | |
44 | ||
45 | if (!icq) { | |
46 | icq = ioc_create_icq(ioc, q, GFP_ATOMIC); | |
47 | if (!icq) | |
48 | return; | |
49 | } | |
50 | ||
51 | rq->elv.icq = icq; | |
f1ba8261 | 52 | if (!blk_mq_sched_get_rq_priv(q, rq, bio)) { |
bd166ef1 JA |
53 | rq->rq_flags |= RQF_ELVPRIV; |
54 | get_io_context(icq->ioc); | |
55 | return; | |
56 | } | |
57 | ||
58 | rq->elv.icq = NULL; | |
59 | } | |
60 | ||
61 | static void blk_mq_sched_assign_ioc(struct request_queue *q, | |
62 | struct request *rq, struct bio *bio) | |
63 | { | |
64 | struct io_context *ioc; | |
65 | ||
66 | ioc = rq_ioc(bio); | |
67 | if (ioc) | |
f1ba8261 | 68 | __blk_mq_sched_assign_ioc(q, rq, bio, ioc); |
bd166ef1 JA |
69 | } |
70 | ||
71 | struct request *blk_mq_sched_get_request(struct request_queue *q, | |
72 | struct bio *bio, | |
73 | unsigned int op, | |
74 | struct blk_mq_alloc_data *data) | |
75 | { | |
76 | struct elevator_queue *e = q->elevator; | |
bd166ef1 | 77 | struct request *rq; |
bd166ef1 JA |
78 | |
79 | blk_queue_enter_live(q); | |
6d2809d5 OS |
80 | data->q = q; |
81 | if (likely(!data->ctx)) | |
82 | data->ctx = blk_mq_get_ctx(q); | |
83 | if (likely(!data->hctx)) | |
84 | data->hctx = blk_mq_map_queue(q, data->ctx->cpu); | |
bd166ef1 | 85 | |
9f2779bf | 86 | if (e) { |
bd166ef1 JA |
87 | data->flags |= BLK_MQ_REQ_INTERNAL; |
88 | ||
89 | /* | |
90 | * Flush requests are special and go directly to the | |
91 | * dispatch list. | |
92 | */ | |
f73f44eb | 93 | if (!op_is_flush(op) && e->type->ops.mq.get_request) { |
bd166ef1 JA |
94 | rq = e->type->ops.mq.get_request(q, op, data); |
95 | if (rq) | |
96 | rq->rq_flags |= RQF_QUEUED; | |
97 | } else | |
98 | rq = __blk_mq_alloc_request(data, op); | |
99 | } else { | |
100 | rq = __blk_mq_alloc_request(data, op); | |
bd166ef1 JA |
101 | } |
102 | ||
103 | if (rq) { | |
f73f44eb | 104 | if (!op_is_flush(op)) { |
bd166ef1 JA |
105 | rq->elv.icq = NULL; |
106 | if (e && e->type->icq_cache) | |
107 | blk_mq_sched_assign_ioc(q, rq, bio); | |
108 | } | |
109 | data->hctx->queued++; | |
110 | return rq; | |
111 | } | |
112 | ||
113 | blk_queue_exit(q); | |
114 | return NULL; | |
115 | } | |
116 | ||
117 | void blk_mq_sched_put_request(struct request *rq) | |
118 | { | |
119 | struct request_queue *q = rq->q; | |
120 | struct elevator_queue *e = q->elevator; | |
121 | ||
122 | if (rq->rq_flags & RQF_ELVPRIV) { | |
123 | blk_mq_sched_put_rq_priv(rq->q, rq); | |
124 | if (rq->elv.icq) { | |
125 | put_io_context(rq->elv.icq->ioc); | |
126 | rq->elv.icq = NULL; | |
127 | } | |
128 | } | |
129 | ||
130 | if ((rq->rq_flags & RQF_QUEUED) && e && e->type->ops.mq.put_request) | |
131 | e->type->ops.mq.put_request(rq); | |
132 | else | |
133 | blk_mq_finish_request(rq); | |
134 | } | |
135 | ||
136 | void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) | |
137 | { | |
81380ca1 OS |
138 | struct request_queue *q = hctx->queue; |
139 | struct elevator_queue *e = q->elevator; | |
64765a75 JA |
140 | const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request; |
141 | bool did_work = false; | |
bd166ef1 JA |
142 | LIST_HEAD(rq_list); |
143 | ||
144 | if (unlikely(blk_mq_hctx_stopped(hctx))) | |
145 | return; | |
146 | ||
147 | hctx->run++; | |
148 | ||
149 | /* | |
150 | * If we have previous entries on our dispatch list, grab them first for | |
151 | * more fair dispatch. | |
152 | */ | |
153 | if (!list_empty_careful(&hctx->dispatch)) { | |
154 | spin_lock(&hctx->lock); | |
155 | if (!list_empty(&hctx->dispatch)) | |
156 | list_splice_init(&hctx->dispatch, &rq_list); | |
157 | spin_unlock(&hctx->lock); | |
158 | } | |
159 | ||
160 | /* | |
161 | * Only ask the scheduler for requests, if we didn't have residual | |
162 | * requests from the dispatch list. This is to avoid the case where | |
163 | * we only ever dispatch a fraction of the requests available because | |
164 | * of low device queue depth. Once we pull requests out of the IO | |
165 | * scheduler, we can no longer merge or sort them. So it's best to | |
166 | * leave them there for as long as we can. Mark the hw queue as | |
167 | * needing a restart in that case. | |
168 | */ | |
c13660a0 | 169 | if (!list_empty(&rq_list)) { |
d38d3515 | 170 | blk_mq_sched_mark_restart_hctx(hctx); |
81380ca1 | 171 | did_work = blk_mq_dispatch_rq_list(q, &rq_list); |
64765a75 | 172 | } else if (!has_sched_dispatch) { |
c13660a0 | 173 | blk_mq_flush_busy_ctxs(hctx, &rq_list); |
81380ca1 | 174 | blk_mq_dispatch_rq_list(q, &rq_list); |
64765a75 JA |
175 | } |
176 | ||
177 | /* | |
178 | * We want to dispatch from the scheduler if we had no work left | |
179 | * on the dispatch list, OR if we did have work but weren't able | |
180 | * to make progress. | |
181 | */ | |
182 | if (!did_work && has_sched_dispatch) { | |
c13660a0 JA |
183 | do { |
184 | struct request *rq; | |
185 | ||
186 | rq = e->type->ops.mq.dispatch_request(hctx); | |
187 | if (!rq) | |
188 | break; | |
189 | list_add(&rq->queuelist, &rq_list); | |
81380ca1 | 190 | } while (blk_mq_dispatch_rq_list(q, &rq_list)); |
c13660a0 | 191 | } |
bd166ef1 JA |
192 | } |
193 | ||
e4d750c9 JA |
194 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, |
195 | struct request **merged_request) | |
bd166ef1 JA |
196 | { |
197 | struct request *rq; | |
bd166ef1 | 198 | |
34fe7c05 CH |
199 | switch (elv_merge(q, &rq, bio)) { |
200 | case ELEVATOR_BACK_MERGE: | |
bd166ef1 JA |
201 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
202 | return false; | |
34fe7c05 CH |
203 | if (!bio_attempt_back_merge(q, rq, bio)) |
204 | return false; | |
205 | *merged_request = attempt_back_merge(q, rq); | |
206 | if (!*merged_request) | |
207 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
208 | return true; | |
209 | case ELEVATOR_FRONT_MERGE: | |
bd166ef1 JA |
210 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
211 | return false; | |
34fe7c05 CH |
212 | if (!bio_attempt_front_merge(q, rq, bio)) |
213 | return false; | |
214 | *merged_request = attempt_front_merge(q, rq); | |
215 | if (!*merged_request) | |
216 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
217 | return true; | |
218 | default: | |
219 | return false; | |
bd166ef1 | 220 | } |
bd166ef1 JA |
221 | } |
222 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); | |
223 | ||
9bddeb2a ML |
224 | /* |
225 | * Reverse check our software queue for entries that we could potentially | |
226 | * merge with. Currently includes a hand-wavy stop count of 8, to not spend | |
227 | * too much time checking for merges. | |
228 | */ | |
229 | static bool blk_mq_attempt_merge(struct request_queue *q, | |
230 | struct blk_mq_ctx *ctx, struct bio *bio) | |
231 | { | |
232 | struct request *rq; | |
233 | int checked = 8; | |
234 | ||
235 | list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) { | |
236 | bool merged = false; | |
237 | ||
238 | if (!checked--) | |
239 | break; | |
240 | ||
241 | if (!blk_rq_merge_ok(rq, bio)) | |
242 | continue; | |
243 | ||
244 | switch (blk_try_merge(rq, bio)) { | |
245 | case ELEVATOR_BACK_MERGE: | |
246 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
247 | merged = bio_attempt_back_merge(q, rq, bio); | |
248 | break; | |
249 | case ELEVATOR_FRONT_MERGE: | |
250 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
251 | merged = bio_attempt_front_merge(q, rq, bio); | |
252 | break; | |
253 | case ELEVATOR_DISCARD_MERGE: | |
254 | merged = bio_attempt_discard_merge(q, rq, bio); | |
255 | break; | |
256 | default: | |
257 | continue; | |
258 | } | |
259 | ||
260 | if (merged) | |
261 | ctx->rq_merged++; | |
262 | return merged; | |
263 | } | |
264 | ||
265 | return false; | |
266 | } | |
267 | ||
bd166ef1 JA |
268 | bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) |
269 | { | |
270 | struct elevator_queue *e = q->elevator; | |
9bddeb2a ML |
271 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
272 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
273 | bool ret = false; | |
bd166ef1 | 274 | |
9bddeb2a | 275 | if (e && e->type->ops.mq.bio_merge) { |
bd166ef1 JA |
276 | blk_mq_put_ctx(ctx); |
277 | return e->type->ops.mq.bio_merge(hctx, bio); | |
278 | } | |
279 | ||
9bddeb2a ML |
280 | if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) { |
281 | /* default per sw-queue merge */ | |
282 | spin_lock(&ctx->lock); | |
283 | ret = blk_mq_attempt_merge(q, ctx, bio); | |
284 | spin_unlock(&ctx->lock); | |
285 | } | |
286 | ||
287 | blk_mq_put_ctx(ctx); | |
288 | return ret; | |
bd166ef1 JA |
289 | } |
290 | ||
291 | bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) | |
292 | { | |
293 | return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); | |
294 | } | |
295 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); | |
296 | ||
297 | void blk_mq_sched_request_inserted(struct request *rq) | |
298 | { | |
299 | trace_block_rq_insert(rq->q, rq); | |
300 | } | |
301 | EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); | |
302 | ||
0cacba6c OS |
303 | static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, |
304 | struct request *rq) | |
bd166ef1 JA |
305 | { |
306 | if (rq->tag == -1) { | |
307 | rq->rq_flags |= RQF_SORTED; | |
308 | return false; | |
309 | } | |
310 | ||
311 | /* | |
312 | * If we already have a real request tag, send directly to | |
313 | * the dispatch list. | |
314 | */ | |
315 | spin_lock(&hctx->lock); | |
316 | list_add(&rq->queuelist, &hctx->dispatch); | |
317 | spin_unlock(&hctx->lock); | |
318 | return true; | |
319 | } | |
bd166ef1 | 320 | |
6d8c6c0f | 321 | static bool blk_mq_sched_restart_hctx(struct blk_mq_hw_ctx *hctx) |
50e1dab8 JA |
322 | { |
323 | if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) { | |
324 | clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); | |
6d8c6c0f | 325 | if (blk_mq_hctx_has_pending(hctx)) { |
50e1dab8 | 326 | blk_mq_run_hw_queue(hctx, true); |
6d8c6c0f BVA |
327 | return true; |
328 | } | |
50e1dab8 | 329 | } |
6d8c6c0f | 330 | return false; |
50e1dab8 JA |
331 | } |
332 | ||
6d8c6c0f BVA |
333 | /** |
334 | * list_for_each_entry_rcu_rr - iterate in a round-robin fashion over rcu list | |
335 | * @pos: loop cursor. | |
336 | * @skip: the list element that will not be examined. Iteration starts at | |
337 | * @skip->next. | |
338 | * @head: head of the list to examine. This list must have at least one | |
339 | * element, namely @skip. | |
340 | * @member: name of the list_head structure within typeof(*pos). | |
341 | */ | |
342 | #define list_for_each_entry_rcu_rr(pos, skip, head, member) \ | |
343 | for ((pos) = (skip); \ | |
344 | (pos = (pos)->member.next != (head) ? list_entry_rcu( \ | |
345 | (pos)->member.next, typeof(*pos), member) : \ | |
346 | list_entry_rcu((pos)->member.next->next, typeof(*pos), member)), \ | |
347 | (pos) != (skip); ) | |
50e1dab8 | 348 | |
6d8c6c0f BVA |
349 | /* |
350 | * Called after a driver tag has been freed to check whether a hctx needs to | |
351 | * be restarted. Restarts @hctx if its tag set is not shared. Restarts hardware | |
352 | * queues in a round-robin fashion if the tag set of @hctx is shared with other | |
353 | * hardware queues. | |
354 | */ | |
355 | void blk_mq_sched_restart(struct blk_mq_hw_ctx *const hctx) | |
356 | { | |
357 | struct blk_mq_tags *const tags = hctx->tags; | |
358 | struct blk_mq_tag_set *const set = hctx->queue->tag_set; | |
359 | struct request_queue *const queue = hctx->queue, *q; | |
360 | struct blk_mq_hw_ctx *hctx2; | |
361 | unsigned int i, j; | |
362 | ||
363 | if (set->flags & BLK_MQ_F_TAG_SHARED) { | |
364 | rcu_read_lock(); | |
365 | list_for_each_entry_rcu_rr(q, queue, &set->tag_list, | |
366 | tag_set_list) { | |
367 | queue_for_each_hw_ctx(q, hctx2, i) | |
368 | if (hctx2->tags == tags && | |
369 | blk_mq_sched_restart_hctx(hctx2)) | |
370 | goto done; | |
371 | } | |
372 | j = hctx->queue_num + 1; | |
373 | for (i = 0; i < queue->nr_hw_queues; i++, j++) { | |
374 | if (j == queue->nr_hw_queues) | |
375 | j = 0; | |
376 | hctx2 = queue->queue_hw_ctx[j]; | |
377 | if (hctx2->tags == tags && | |
378 | blk_mq_sched_restart_hctx(hctx2)) | |
379 | break; | |
d38d3515 | 380 | } |
6d8c6c0f BVA |
381 | done: |
382 | rcu_read_unlock(); | |
d38d3515 | 383 | } else { |
50e1dab8 | 384 | blk_mq_sched_restart_hctx(hctx); |
50e1dab8 JA |
385 | } |
386 | } | |
387 | ||
bd6737f1 JA |
388 | /* |
389 | * Add flush/fua to the queue. If we fail getting a driver tag, then | |
390 | * punt to the requeue list. Requeue will re-invoke us from a context | |
391 | * that's safe to block from. | |
392 | */ | |
393 | static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx, | |
394 | struct request *rq, bool can_block) | |
395 | { | |
396 | if (blk_mq_get_driver_tag(rq, &hctx, can_block)) { | |
397 | blk_insert_flush(rq); | |
398 | blk_mq_run_hw_queue(hctx, true); | |
399 | } else | |
c7a571b4 | 400 | blk_mq_add_to_requeue_list(rq, false, true); |
bd6737f1 JA |
401 | } |
402 | ||
403 | void blk_mq_sched_insert_request(struct request *rq, bool at_head, | |
404 | bool run_queue, bool async, bool can_block) | |
405 | { | |
406 | struct request_queue *q = rq->q; | |
407 | struct elevator_queue *e = q->elevator; | |
408 | struct blk_mq_ctx *ctx = rq->mq_ctx; | |
409 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
410 | ||
f3a8ab7d | 411 | if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) { |
bd6737f1 JA |
412 | blk_mq_sched_insert_flush(hctx, rq, can_block); |
413 | return; | |
414 | } | |
415 | ||
0cacba6c OS |
416 | if (e && blk_mq_sched_bypass_insert(hctx, rq)) |
417 | goto run; | |
418 | ||
bd6737f1 JA |
419 | if (e && e->type->ops.mq.insert_requests) { |
420 | LIST_HEAD(list); | |
421 | ||
422 | list_add(&rq->queuelist, &list); | |
423 | e->type->ops.mq.insert_requests(hctx, &list, at_head); | |
424 | } else { | |
425 | spin_lock(&ctx->lock); | |
426 | __blk_mq_insert_request(hctx, rq, at_head); | |
427 | spin_unlock(&ctx->lock); | |
428 | } | |
429 | ||
0cacba6c | 430 | run: |
bd6737f1 JA |
431 | if (run_queue) |
432 | blk_mq_run_hw_queue(hctx, async); | |
433 | } | |
434 | ||
435 | void blk_mq_sched_insert_requests(struct request_queue *q, | |
436 | struct blk_mq_ctx *ctx, | |
437 | struct list_head *list, bool run_queue_async) | |
438 | { | |
439 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
440 | struct elevator_queue *e = hctx->queue->elevator; | |
441 | ||
0cacba6c OS |
442 | if (e) { |
443 | struct request *rq, *next; | |
444 | ||
445 | /* | |
446 | * We bypass requests that already have a driver tag assigned, | |
447 | * which should only be flushes. Flushes are only ever inserted | |
448 | * as single requests, so we shouldn't ever hit the | |
449 | * WARN_ON_ONCE() below (but let's handle it just in case). | |
450 | */ | |
451 | list_for_each_entry_safe(rq, next, list, queuelist) { | |
452 | if (WARN_ON_ONCE(rq->tag != -1)) { | |
453 | list_del_init(&rq->queuelist); | |
454 | blk_mq_sched_bypass_insert(hctx, rq); | |
455 | } | |
456 | } | |
457 | } | |
458 | ||
bd6737f1 JA |
459 | if (e && e->type->ops.mq.insert_requests) |
460 | e->type->ops.mq.insert_requests(hctx, list, false); | |
461 | else | |
462 | blk_mq_insert_requests(hctx, ctx, list); | |
463 | ||
464 | blk_mq_run_hw_queue(hctx, run_queue_async); | |
465 | } | |
466 | ||
bd166ef1 JA |
467 | static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, |
468 | struct blk_mq_hw_ctx *hctx, | |
469 | unsigned int hctx_idx) | |
470 | { | |
471 | if (hctx->sched_tags) { | |
472 | blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); | |
473 | blk_mq_free_rq_map(hctx->sched_tags); | |
474 | hctx->sched_tags = NULL; | |
475 | } | |
476 | } | |
477 | ||
6917ff0b OS |
478 | static int blk_mq_sched_alloc_tags(struct request_queue *q, |
479 | struct blk_mq_hw_ctx *hctx, | |
480 | unsigned int hctx_idx) | |
481 | { | |
482 | struct blk_mq_tag_set *set = q->tag_set; | |
483 | int ret; | |
484 | ||
485 | hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, | |
486 | set->reserved_tags); | |
487 | if (!hctx->sched_tags) | |
488 | return -ENOMEM; | |
489 | ||
490 | ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); | |
491 | if (ret) | |
492 | blk_mq_sched_free_tags(set, hctx, hctx_idx); | |
493 | ||
494 | return ret; | |
495 | } | |
496 | ||
54d5329d | 497 | static void blk_mq_sched_tags_teardown(struct request_queue *q) |
bd166ef1 JA |
498 | { |
499 | struct blk_mq_tag_set *set = q->tag_set; | |
500 | struct blk_mq_hw_ctx *hctx; | |
6917ff0b OS |
501 | int i; |
502 | ||
503 | queue_for_each_hw_ctx(q, hctx, i) | |
504 | blk_mq_sched_free_tags(set, hctx, i); | |
505 | } | |
506 | ||
93252632 OS |
507 | int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, |
508 | unsigned int hctx_idx) | |
509 | { | |
510 | struct elevator_queue *e = q->elevator; | |
ee056f98 | 511 | int ret; |
93252632 OS |
512 | |
513 | if (!e) | |
514 | return 0; | |
515 | ||
ee056f98 OS |
516 | ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx); |
517 | if (ret) | |
518 | return ret; | |
519 | ||
520 | if (e->type->ops.mq.init_hctx) { | |
521 | ret = e->type->ops.mq.init_hctx(hctx, hctx_idx); | |
522 | if (ret) { | |
523 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); | |
524 | return ret; | |
525 | } | |
526 | } | |
527 | ||
d332ce09 OS |
528 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
529 | ||
ee056f98 | 530 | return 0; |
93252632 OS |
531 | } |
532 | ||
533 | void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, | |
534 | unsigned int hctx_idx) | |
535 | { | |
536 | struct elevator_queue *e = q->elevator; | |
537 | ||
538 | if (!e) | |
539 | return; | |
540 | ||
d332ce09 OS |
541 | blk_mq_debugfs_unregister_sched_hctx(hctx); |
542 | ||
ee056f98 OS |
543 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { |
544 | e->type->ops.mq.exit_hctx(hctx, hctx_idx); | |
545 | hctx->sched_data = NULL; | |
546 | } | |
547 | ||
93252632 OS |
548 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); |
549 | } | |
550 | ||
6917ff0b OS |
551 | int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) |
552 | { | |
553 | struct blk_mq_hw_ctx *hctx; | |
ee056f98 | 554 | struct elevator_queue *eq; |
6917ff0b OS |
555 | unsigned int i; |
556 | int ret; | |
557 | ||
558 | if (!e) { | |
559 | q->elevator = NULL; | |
560 | return 0; | |
561 | } | |
bd166ef1 JA |
562 | |
563 | /* | |
564 | * Default to 256, since we don't split into sync/async like the | |
565 | * old code did. Additionally, this is a per-hw queue depth. | |
566 | */ | |
567 | q->nr_requests = 2 * BLKDEV_MAX_RQ; | |
568 | ||
bd166ef1 | 569 | queue_for_each_hw_ctx(q, hctx, i) { |
6917ff0b | 570 | ret = blk_mq_sched_alloc_tags(q, hctx, i); |
bd166ef1 | 571 | if (ret) |
6917ff0b | 572 | goto err; |
bd166ef1 JA |
573 | } |
574 | ||
6917ff0b OS |
575 | ret = e->ops.mq.init_sched(q, e); |
576 | if (ret) | |
577 | goto err; | |
bd166ef1 | 578 | |
d332ce09 OS |
579 | blk_mq_debugfs_register_sched(q); |
580 | ||
581 | queue_for_each_hw_ctx(q, hctx, i) { | |
582 | if (e->ops.mq.init_hctx) { | |
ee056f98 OS |
583 | ret = e->ops.mq.init_hctx(hctx, i); |
584 | if (ret) { | |
585 | eq = q->elevator; | |
586 | blk_mq_exit_sched(q, eq); | |
587 | kobject_put(&eq->kobj); | |
588 | return ret; | |
589 | } | |
590 | } | |
d332ce09 | 591 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
ee056f98 OS |
592 | } |
593 | ||
bd166ef1 | 594 | return 0; |
bd166ef1 | 595 | |
6917ff0b | 596 | err: |
54d5329d OS |
597 | blk_mq_sched_tags_teardown(q); |
598 | q->elevator = NULL; | |
6917ff0b | 599 | return ret; |
bd166ef1 | 600 | } |
d3484991 | 601 | |
54d5329d OS |
602 | void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) |
603 | { | |
ee056f98 OS |
604 | struct blk_mq_hw_ctx *hctx; |
605 | unsigned int i; | |
606 | ||
d332ce09 OS |
607 | queue_for_each_hw_ctx(q, hctx, i) { |
608 | blk_mq_debugfs_unregister_sched_hctx(hctx); | |
609 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { | |
610 | e->type->ops.mq.exit_hctx(hctx, i); | |
611 | hctx->sched_data = NULL; | |
ee056f98 OS |
612 | } |
613 | } | |
d332ce09 | 614 | blk_mq_debugfs_unregister_sched(q); |
54d5329d OS |
615 | if (e->type->ops.mq.exit_sched) |
616 | e->type->ops.mq.exit_sched(e); | |
617 | blk_mq_sched_tags_teardown(q); | |
618 | q->elevator = NULL; | |
619 | } | |
620 | ||
d3484991 JA |
621 | int blk_mq_sched_init(struct request_queue *q) |
622 | { | |
623 | int ret; | |
624 | ||
d3484991 JA |
625 | mutex_lock(&q->sysfs_lock); |
626 | ret = elevator_init(q, NULL); | |
627 | mutex_unlock(&q->sysfs_lock); | |
628 | ||
629 | return ret; | |
630 | } |