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 | ||
44e8c2bf | 34 | void blk_mq_sched_assign_ioc(struct request *rq, struct bio *bio) |
bd166ef1 | 35 | { |
44e8c2bf CH |
36 | struct request_queue *q = rq->q; |
37 | struct io_context *ioc = rq_ioc(bio); | |
bd166ef1 JA |
38 | struct io_cq *icq; |
39 | ||
40 | spin_lock_irq(q->queue_lock); | |
41 | icq = ioc_lookup_icq(ioc, q); | |
42 | spin_unlock_irq(q->queue_lock); | |
43 | ||
44 | if (!icq) { | |
45 | icq = ioc_create_icq(ioc, q, GFP_ATOMIC); | |
46 | if (!icq) | |
47 | return; | |
48 | } | |
ea511e3c | 49 | get_io_context(icq->ioc); |
44e8c2bf | 50 | rq->elv.icq = icq; |
bd166ef1 JA |
51 | } |
52 | ||
8e8320c9 JA |
53 | /* |
54 | * Mark a hardware queue as needing a restart. For shared queues, maintain | |
55 | * a count of how many hardware queues are marked for restart. | |
56 | */ | |
57 | static void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx) | |
58 | { | |
59 | if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
60 | return; | |
61 | ||
97889f9a | 62 | set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); |
8e8320c9 JA |
63 | } |
64 | ||
97889f9a | 65 | void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx) |
8e8320c9 JA |
66 | { |
67 | if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
97889f9a ML |
68 | return; |
69 | clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); | |
8e8320c9 | 70 | |
97889f9a | 71 | blk_mq_run_hw_queue(hctx, true); |
8e8320c9 JA |
72 | } |
73 | ||
1f460b63 ML |
74 | /* |
75 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
76 | * its queue by itself in its completion handler, so we don't need to | |
77 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
78 | */ | |
79 | static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx) | |
caf8eb0d ML |
80 | { |
81 | struct request_queue *q = hctx->queue; | |
82 | struct elevator_queue *e = q->elevator; | |
83 | LIST_HEAD(rq_list); | |
84 | ||
85 | do { | |
de148297 | 86 | struct request *rq; |
caf8eb0d | 87 | |
de148297 ML |
88 | if (e->type->ops.mq.has_work && |
89 | !e->type->ops.mq.has_work(hctx)) | |
caf8eb0d | 90 | break; |
de148297 | 91 | |
88022d72 | 92 | if (!blk_mq_get_dispatch_budget(hctx)) |
1f460b63 | 93 | break; |
de148297 ML |
94 | |
95 | rq = e->type->ops.mq.dispatch_request(hctx); | |
96 | if (!rq) { | |
97 | blk_mq_put_dispatch_budget(hctx); | |
98 | break; | |
de148297 ML |
99 | } |
100 | ||
101 | /* | |
102 | * Now this rq owns the budget which has to be released | |
103 | * if this rq won't be queued to driver via .queue_rq() | |
104 | * in blk_mq_dispatch_rq_list(). | |
105 | */ | |
caf8eb0d | 106 | list_add(&rq->queuelist, &rq_list); |
de148297 | 107 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); |
caf8eb0d ML |
108 | } |
109 | ||
b347689f ML |
110 | static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx, |
111 | struct blk_mq_ctx *ctx) | |
112 | { | |
113 | unsigned idx = ctx->index_hw; | |
114 | ||
115 | if (++idx == hctx->nr_ctx) | |
116 | idx = 0; | |
117 | ||
118 | return hctx->ctxs[idx]; | |
119 | } | |
120 | ||
1f460b63 ML |
121 | /* |
122 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
123 | * its queue by itself in its completion handler, so we don't need to | |
124 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
125 | */ | |
126 | static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx) | |
b347689f ML |
127 | { |
128 | struct request_queue *q = hctx->queue; | |
129 | LIST_HEAD(rq_list); | |
130 | struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from); | |
131 | ||
132 | do { | |
133 | struct request *rq; | |
b347689f ML |
134 | |
135 | if (!sbitmap_any_bit_set(&hctx->ctx_map)) | |
136 | break; | |
137 | ||
88022d72 | 138 | if (!blk_mq_get_dispatch_budget(hctx)) |
1f460b63 | 139 | break; |
b347689f ML |
140 | |
141 | rq = blk_mq_dequeue_from_ctx(hctx, ctx); | |
142 | if (!rq) { | |
143 | blk_mq_put_dispatch_budget(hctx); | |
144 | break; | |
b347689f ML |
145 | } |
146 | ||
147 | /* | |
148 | * Now this rq owns the budget which has to be released | |
149 | * if this rq won't be queued to driver via .queue_rq() | |
150 | * in blk_mq_dispatch_rq_list(). | |
151 | */ | |
152 | list_add(&rq->queuelist, &rq_list); | |
153 | ||
154 | /* round robin for fair dispatch */ | |
155 | ctx = blk_mq_next_ctx(hctx, rq->mq_ctx); | |
156 | ||
157 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); | |
158 | ||
159 | WRITE_ONCE(hctx->dispatch_from, ctx); | |
b347689f ML |
160 | } |
161 | ||
1f460b63 | 162 | void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) |
bd166ef1 | 163 | { |
81380ca1 OS |
164 | struct request_queue *q = hctx->queue; |
165 | struct elevator_queue *e = q->elevator; | |
64765a75 | 166 | const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request; |
bd166ef1 JA |
167 | LIST_HEAD(rq_list); |
168 | ||
f4560ffe ML |
169 | /* RCU or SRCU read lock is needed before checking quiesced flag */ |
170 | if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q))) | |
1f460b63 | 171 | return; |
bd166ef1 JA |
172 | |
173 | hctx->run++; | |
174 | ||
175 | /* | |
176 | * If we have previous entries on our dispatch list, grab them first for | |
177 | * more fair dispatch. | |
178 | */ | |
179 | if (!list_empty_careful(&hctx->dispatch)) { | |
180 | spin_lock(&hctx->lock); | |
181 | if (!list_empty(&hctx->dispatch)) | |
182 | list_splice_init(&hctx->dispatch, &rq_list); | |
183 | spin_unlock(&hctx->lock); | |
184 | } | |
185 | ||
186 | /* | |
187 | * Only ask the scheduler for requests, if we didn't have residual | |
188 | * requests from the dispatch list. This is to avoid the case where | |
189 | * we only ever dispatch a fraction of the requests available because | |
190 | * of low device queue depth. Once we pull requests out of the IO | |
191 | * scheduler, we can no longer merge or sort them. So it's best to | |
192 | * leave them there for as long as we can. Mark the hw queue as | |
193 | * needing a restart in that case. | |
caf8eb0d ML |
194 | * |
195 | * We want to dispatch from the scheduler if there was nothing | |
196 | * on the dispatch list or we were able to dispatch from the | |
197 | * dispatch list. | |
bd166ef1 | 198 | */ |
c13660a0 | 199 | if (!list_empty(&rq_list)) { |
d38d3515 | 200 | blk_mq_sched_mark_restart_hctx(hctx); |
b347689f ML |
201 | if (blk_mq_dispatch_rq_list(q, &rq_list, false)) { |
202 | if (has_sched_dispatch) | |
1f460b63 | 203 | blk_mq_do_dispatch_sched(hctx); |
b347689f | 204 | else |
1f460b63 | 205 | blk_mq_do_dispatch_ctx(hctx); |
b347689f | 206 | } |
caf8eb0d | 207 | } else if (has_sched_dispatch) { |
1f460b63 | 208 | blk_mq_do_dispatch_sched(hctx); |
b347689f ML |
209 | } else if (q->mq_ops->get_budget) { |
210 | /* | |
211 | * If we need to get budget before queuing request, we | |
212 | * dequeue request one by one from sw queue for avoiding | |
213 | * to mess up I/O merge when dispatch runs out of resource. | |
214 | * | |
215 | * TODO: get more budgets, and dequeue more requests in | |
216 | * one time. | |
217 | */ | |
1f460b63 | 218 | blk_mq_do_dispatch_ctx(hctx); |
caf8eb0d | 219 | } else { |
c13660a0 | 220 | blk_mq_flush_busy_ctxs(hctx, &rq_list); |
de148297 | 221 | blk_mq_dispatch_rq_list(q, &rq_list, false); |
64765a75 | 222 | } |
bd166ef1 JA |
223 | } |
224 | ||
e4d750c9 JA |
225 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, |
226 | struct request **merged_request) | |
bd166ef1 JA |
227 | { |
228 | struct request *rq; | |
bd166ef1 | 229 | |
34fe7c05 CH |
230 | switch (elv_merge(q, &rq, bio)) { |
231 | case ELEVATOR_BACK_MERGE: | |
bd166ef1 JA |
232 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
233 | return false; | |
34fe7c05 CH |
234 | if (!bio_attempt_back_merge(q, rq, bio)) |
235 | return false; | |
236 | *merged_request = attempt_back_merge(q, rq); | |
237 | if (!*merged_request) | |
238 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
239 | return true; | |
240 | case ELEVATOR_FRONT_MERGE: | |
bd166ef1 JA |
241 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
242 | return false; | |
34fe7c05 CH |
243 | if (!bio_attempt_front_merge(q, rq, bio)) |
244 | return false; | |
245 | *merged_request = attempt_front_merge(q, rq); | |
246 | if (!*merged_request) | |
247 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
248 | return true; | |
bea99a50 KB |
249 | case ELEVATOR_DISCARD_MERGE: |
250 | return bio_attempt_discard_merge(q, rq, bio); | |
34fe7c05 CH |
251 | default: |
252 | return false; | |
bd166ef1 | 253 | } |
bd166ef1 JA |
254 | } |
255 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); | |
256 | ||
9bddeb2a | 257 | /* |
9c558734 JA |
258 | * Iterate list of requests and see if we can merge this bio with any |
259 | * of them. | |
9bddeb2a | 260 | */ |
9c558734 JA |
261 | bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list, |
262 | struct bio *bio) | |
9bddeb2a ML |
263 | { |
264 | struct request *rq; | |
265 | int checked = 8; | |
266 | ||
9c558734 | 267 | list_for_each_entry_reverse(rq, list, queuelist) { |
9bddeb2a ML |
268 | bool merged = false; |
269 | ||
270 | if (!checked--) | |
271 | break; | |
272 | ||
273 | if (!blk_rq_merge_ok(rq, bio)) | |
274 | continue; | |
275 | ||
276 | switch (blk_try_merge(rq, bio)) { | |
277 | case ELEVATOR_BACK_MERGE: | |
278 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
279 | merged = bio_attempt_back_merge(q, rq, bio); | |
280 | break; | |
281 | case ELEVATOR_FRONT_MERGE: | |
282 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
283 | merged = bio_attempt_front_merge(q, rq, bio); | |
284 | break; | |
285 | case ELEVATOR_DISCARD_MERGE: | |
286 | merged = bio_attempt_discard_merge(q, rq, bio); | |
287 | break; | |
288 | default: | |
289 | continue; | |
290 | } | |
291 | ||
9bddeb2a ML |
292 | return merged; |
293 | } | |
294 | ||
295 | return false; | |
296 | } | |
9c558734 JA |
297 | EXPORT_SYMBOL_GPL(blk_mq_bio_list_merge); |
298 | ||
299 | /* | |
300 | * Reverse check our software queue for entries that we could potentially | |
301 | * merge with. Currently includes a hand-wavy stop count of 8, to not spend | |
302 | * too much time checking for merges. | |
303 | */ | |
304 | static bool blk_mq_attempt_merge(struct request_queue *q, | |
305 | struct blk_mq_ctx *ctx, struct bio *bio) | |
306 | { | |
307 | lockdep_assert_held(&ctx->lock); | |
308 | ||
309 | if (blk_mq_bio_list_merge(q, &ctx->rq_list, bio)) { | |
310 | ctx->rq_merged++; | |
311 | return true; | |
312 | } | |
313 | ||
314 | return false; | |
315 | } | |
9bddeb2a | 316 | |
bd166ef1 JA |
317 | bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) |
318 | { | |
319 | struct elevator_queue *e = q->elevator; | |
9bddeb2a ML |
320 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
321 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
322 | bool ret = false; | |
bd166ef1 | 323 | |
9bddeb2a | 324 | if (e && e->type->ops.mq.bio_merge) { |
bd166ef1 JA |
325 | blk_mq_put_ctx(ctx); |
326 | return e->type->ops.mq.bio_merge(hctx, bio); | |
327 | } | |
328 | ||
b04f50ab ML |
329 | if ((hctx->flags & BLK_MQ_F_SHOULD_MERGE) && |
330 | !list_empty_careful(&ctx->rq_list)) { | |
9bddeb2a ML |
331 | /* default per sw-queue merge */ |
332 | spin_lock(&ctx->lock); | |
333 | ret = blk_mq_attempt_merge(q, ctx, bio); | |
334 | spin_unlock(&ctx->lock); | |
335 | } | |
336 | ||
337 | blk_mq_put_ctx(ctx); | |
338 | return ret; | |
bd166ef1 JA |
339 | } |
340 | ||
341 | bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) | |
342 | { | |
343 | return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); | |
344 | } | |
345 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); | |
346 | ||
347 | void blk_mq_sched_request_inserted(struct request *rq) | |
348 | { | |
349 | trace_block_rq_insert(rq->q, rq); | |
350 | } | |
351 | EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); | |
352 | ||
0cacba6c | 353 | static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, |
a6a252e6 | 354 | bool has_sched, |
0cacba6c | 355 | struct request *rq) |
bd166ef1 | 356 | { |
a6a252e6 ML |
357 | /* dispatch flush rq directly */ |
358 | if (rq->rq_flags & RQF_FLUSH_SEQ) { | |
359 | spin_lock(&hctx->lock); | |
360 | list_add(&rq->queuelist, &hctx->dispatch); | |
361 | spin_unlock(&hctx->lock); | |
362 | return true; | |
363 | } | |
364 | ||
923218f6 | 365 | if (has_sched) |
bd166ef1 | 366 | rq->rq_flags |= RQF_SORTED; |
bd166ef1 | 367 | |
a6a252e6 | 368 | return false; |
bd166ef1 | 369 | } |
bd166ef1 | 370 | |
bd6737f1 | 371 | void blk_mq_sched_insert_request(struct request *rq, bool at_head, |
9e97d295 | 372 | bool run_queue, bool async) |
bd6737f1 JA |
373 | { |
374 | struct request_queue *q = rq->q; | |
375 | struct elevator_queue *e = q->elevator; | |
376 | struct blk_mq_ctx *ctx = rq->mq_ctx; | |
377 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
378 | ||
a6a252e6 ML |
379 | /* flush rq in flush machinery need to be dispatched directly */ |
380 | if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) { | |
923218f6 ML |
381 | blk_insert_flush(rq); |
382 | goto run; | |
bd6737f1 JA |
383 | } |
384 | ||
923218f6 ML |
385 | WARN_ON(e && (rq->tag != -1)); |
386 | ||
a6a252e6 | 387 | if (blk_mq_sched_bypass_insert(hctx, !!e, rq)) |
0cacba6c OS |
388 | goto run; |
389 | ||
bd6737f1 JA |
390 | if (e && e->type->ops.mq.insert_requests) { |
391 | LIST_HEAD(list); | |
392 | ||
393 | list_add(&rq->queuelist, &list); | |
394 | e->type->ops.mq.insert_requests(hctx, &list, at_head); | |
395 | } else { | |
396 | spin_lock(&ctx->lock); | |
397 | __blk_mq_insert_request(hctx, rq, at_head); | |
398 | spin_unlock(&ctx->lock); | |
399 | } | |
400 | ||
0cacba6c | 401 | run: |
bd6737f1 JA |
402 | if (run_queue) |
403 | blk_mq_run_hw_queue(hctx, async); | |
404 | } | |
405 | ||
406 | void blk_mq_sched_insert_requests(struct request_queue *q, | |
407 | struct blk_mq_ctx *ctx, | |
408 | struct list_head *list, bool run_queue_async) | |
409 | { | |
410 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
411 | struct elevator_queue *e = hctx->queue->elevator; | |
412 | ||
413 | if (e && e->type->ops.mq.insert_requests) | |
414 | e->type->ops.mq.insert_requests(hctx, list, false); | |
415 | else | |
416 | blk_mq_insert_requests(hctx, ctx, list); | |
417 | ||
418 | blk_mq_run_hw_queue(hctx, run_queue_async); | |
419 | } | |
420 | ||
bd166ef1 JA |
421 | static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, |
422 | struct blk_mq_hw_ctx *hctx, | |
423 | unsigned int hctx_idx) | |
424 | { | |
425 | if (hctx->sched_tags) { | |
426 | blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); | |
427 | blk_mq_free_rq_map(hctx->sched_tags); | |
428 | hctx->sched_tags = NULL; | |
429 | } | |
430 | } | |
431 | ||
6917ff0b OS |
432 | static int blk_mq_sched_alloc_tags(struct request_queue *q, |
433 | struct blk_mq_hw_ctx *hctx, | |
434 | unsigned int hctx_idx) | |
435 | { | |
436 | struct blk_mq_tag_set *set = q->tag_set; | |
437 | int ret; | |
438 | ||
439 | hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, | |
440 | set->reserved_tags); | |
441 | if (!hctx->sched_tags) | |
442 | return -ENOMEM; | |
443 | ||
444 | ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); | |
445 | if (ret) | |
446 | blk_mq_sched_free_tags(set, hctx, hctx_idx); | |
447 | ||
448 | return ret; | |
449 | } | |
450 | ||
54d5329d | 451 | static void blk_mq_sched_tags_teardown(struct request_queue *q) |
bd166ef1 JA |
452 | { |
453 | struct blk_mq_tag_set *set = q->tag_set; | |
454 | struct blk_mq_hw_ctx *hctx; | |
6917ff0b OS |
455 | int i; |
456 | ||
457 | queue_for_each_hw_ctx(q, hctx, i) | |
458 | blk_mq_sched_free_tags(set, hctx, i); | |
459 | } | |
460 | ||
93252632 OS |
461 | int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, |
462 | unsigned int hctx_idx) | |
463 | { | |
464 | struct elevator_queue *e = q->elevator; | |
ee056f98 | 465 | int ret; |
93252632 OS |
466 | |
467 | if (!e) | |
468 | return 0; | |
469 | ||
ee056f98 OS |
470 | ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx); |
471 | if (ret) | |
472 | return ret; | |
473 | ||
474 | if (e->type->ops.mq.init_hctx) { | |
475 | ret = e->type->ops.mq.init_hctx(hctx, hctx_idx); | |
476 | if (ret) { | |
477 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); | |
478 | return ret; | |
479 | } | |
480 | } | |
481 | ||
d332ce09 OS |
482 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
483 | ||
ee056f98 | 484 | return 0; |
93252632 OS |
485 | } |
486 | ||
487 | void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, | |
488 | unsigned int hctx_idx) | |
489 | { | |
490 | struct elevator_queue *e = q->elevator; | |
491 | ||
492 | if (!e) | |
493 | return; | |
494 | ||
d332ce09 OS |
495 | blk_mq_debugfs_unregister_sched_hctx(hctx); |
496 | ||
ee056f98 OS |
497 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { |
498 | e->type->ops.mq.exit_hctx(hctx, hctx_idx); | |
499 | hctx->sched_data = NULL; | |
500 | } | |
501 | ||
93252632 OS |
502 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); |
503 | } | |
504 | ||
6917ff0b OS |
505 | int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) |
506 | { | |
507 | struct blk_mq_hw_ctx *hctx; | |
ee056f98 | 508 | struct elevator_queue *eq; |
6917ff0b OS |
509 | unsigned int i; |
510 | int ret; | |
511 | ||
512 | if (!e) { | |
513 | q->elevator = NULL; | |
32a50fab | 514 | q->nr_requests = q->tag_set->queue_depth; |
6917ff0b OS |
515 | return 0; |
516 | } | |
bd166ef1 JA |
517 | |
518 | /* | |
32825c45 ML |
519 | * Default to double of smaller one between hw queue_depth and 128, |
520 | * since we don't split into sync/async like the old code did. | |
521 | * Additionally, this is a per-hw queue depth. | |
bd166ef1 | 522 | */ |
32825c45 ML |
523 | q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth, |
524 | BLKDEV_MAX_RQ); | |
bd166ef1 | 525 | |
bd166ef1 | 526 | queue_for_each_hw_ctx(q, hctx, i) { |
6917ff0b | 527 | ret = blk_mq_sched_alloc_tags(q, hctx, i); |
bd166ef1 | 528 | if (ret) |
6917ff0b | 529 | goto err; |
bd166ef1 JA |
530 | } |
531 | ||
6917ff0b OS |
532 | ret = e->ops.mq.init_sched(q, e); |
533 | if (ret) | |
534 | goto err; | |
bd166ef1 | 535 | |
d332ce09 OS |
536 | blk_mq_debugfs_register_sched(q); |
537 | ||
538 | queue_for_each_hw_ctx(q, hctx, i) { | |
539 | if (e->ops.mq.init_hctx) { | |
ee056f98 OS |
540 | ret = e->ops.mq.init_hctx(hctx, i); |
541 | if (ret) { | |
542 | eq = q->elevator; | |
543 | blk_mq_exit_sched(q, eq); | |
544 | kobject_put(&eq->kobj); | |
545 | return ret; | |
546 | } | |
547 | } | |
d332ce09 | 548 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
ee056f98 OS |
549 | } |
550 | ||
bd166ef1 | 551 | return 0; |
bd166ef1 | 552 | |
6917ff0b | 553 | err: |
54d5329d OS |
554 | blk_mq_sched_tags_teardown(q); |
555 | q->elevator = NULL; | |
6917ff0b | 556 | return ret; |
bd166ef1 | 557 | } |
d3484991 | 558 | |
54d5329d OS |
559 | void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) |
560 | { | |
ee056f98 OS |
561 | struct blk_mq_hw_ctx *hctx; |
562 | unsigned int i; | |
563 | ||
d332ce09 OS |
564 | queue_for_each_hw_ctx(q, hctx, i) { |
565 | blk_mq_debugfs_unregister_sched_hctx(hctx); | |
566 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { | |
567 | e->type->ops.mq.exit_hctx(hctx, i); | |
568 | hctx->sched_data = NULL; | |
ee056f98 OS |
569 | } |
570 | } | |
d332ce09 | 571 | blk_mq_debugfs_unregister_sched(q); |
54d5329d OS |
572 | if (e->type->ops.mq.exit_sched) |
573 | e->type->ops.mq.exit_sched(e); | |
574 | blk_mq_sched_tags_teardown(q); | |
575 | q->elevator = NULL; | |
576 | } |