]>
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 | ||
e2b3fa5a | 34 | void blk_mq_sched_assign_ioc(struct request *rq) |
bd166ef1 | 35 | { |
44e8c2bf | 36 | struct request_queue *q = rq->q; |
0c62bff1 | 37 | struct io_context *ioc; |
bd166ef1 JA |
38 | struct io_cq *icq; |
39 | ||
0c62bff1 JA |
40 | /* |
41 | * May not have an IO context if it's a passthrough request | |
42 | */ | |
43 | ioc = current->io_context; | |
44 | if (!ioc) | |
45 | return; | |
46 | ||
0d945c1f | 47 | spin_lock_irq(&q->queue_lock); |
bd166ef1 | 48 | icq = ioc_lookup_icq(ioc, q); |
0d945c1f | 49 | spin_unlock_irq(&q->queue_lock); |
bd166ef1 JA |
50 | |
51 | if (!icq) { | |
52 | icq = ioc_create_icq(ioc, q, GFP_ATOMIC); | |
53 | if (!icq) | |
54 | return; | |
55 | } | |
ea511e3c | 56 | get_io_context(icq->ioc); |
44e8c2bf | 57 | rq->elv.icq = icq; |
bd166ef1 JA |
58 | } |
59 | ||
8e8320c9 JA |
60 | /* |
61 | * Mark a hardware queue as needing a restart. For shared queues, maintain | |
62 | * a count of how many hardware queues are marked for restart. | |
63 | */ | |
7211aef8 | 64 | void blk_mq_sched_mark_restart_hctx(struct blk_mq_hw_ctx *hctx) |
8e8320c9 JA |
65 | { |
66 | if (test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
67 | return; | |
68 | ||
97889f9a | 69 | set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); |
8e8320c9 | 70 | } |
7211aef8 | 71 | EXPORT_SYMBOL_GPL(blk_mq_sched_mark_restart_hctx); |
8e8320c9 | 72 | |
97889f9a | 73 | void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx) |
8e8320c9 JA |
74 | { |
75 | if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
97889f9a ML |
76 | return; |
77 | clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); | |
8e8320c9 | 78 | |
97889f9a | 79 | blk_mq_run_hw_queue(hctx, true); |
8e8320c9 JA |
80 | } |
81 | ||
1f460b63 ML |
82 | /* |
83 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
84 | * its queue by itself in its completion handler, so we don't need to | |
85 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
86 | */ | |
87 | static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx) | |
caf8eb0d ML |
88 | { |
89 | struct request_queue *q = hctx->queue; | |
90 | struct elevator_queue *e = q->elevator; | |
91 | LIST_HEAD(rq_list); | |
92 | ||
93 | do { | |
de148297 | 94 | struct request *rq; |
caf8eb0d | 95 | |
f9cd4bfe | 96 | if (e->type->ops.has_work && !e->type->ops.has_work(hctx)) |
caf8eb0d | 97 | break; |
de148297 | 98 | |
88022d72 | 99 | if (!blk_mq_get_dispatch_budget(hctx)) |
1f460b63 | 100 | break; |
de148297 | 101 | |
f9cd4bfe | 102 | rq = e->type->ops.dispatch_request(hctx); |
de148297 ML |
103 | if (!rq) { |
104 | blk_mq_put_dispatch_budget(hctx); | |
105 | break; | |
de148297 ML |
106 | } |
107 | ||
108 | /* | |
109 | * Now this rq owns the budget which has to be released | |
110 | * if this rq won't be queued to driver via .queue_rq() | |
111 | * in blk_mq_dispatch_rq_list(). | |
112 | */ | |
caf8eb0d | 113 | list_add(&rq->queuelist, &rq_list); |
de148297 | 114 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); |
caf8eb0d ML |
115 | } |
116 | ||
b347689f ML |
117 | static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx, |
118 | struct blk_mq_ctx *ctx) | |
119 | { | |
f31967f0 | 120 | unsigned short idx = ctx->index_hw[hctx->type]; |
b347689f ML |
121 | |
122 | if (++idx == hctx->nr_ctx) | |
123 | idx = 0; | |
124 | ||
125 | return hctx->ctxs[idx]; | |
126 | } | |
127 | ||
1f460b63 ML |
128 | /* |
129 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
130 | * its queue by itself in its completion handler, so we don't need to | |
131 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
132 | */ | |
133 | static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx) | |
b347689f ML |
134 | { |
135 | struct request_queue *q = hctx->queue; | |
136 | LIST_HEAD(rq_list); | |
137 | struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from); | |
138 | ||
139 | do { | |
140 | struct request *rq; | |
b347689f ML |
141 | |
142 | if (!sbitmap_any_bit_set(&hctx->ctx_map)) | |
143 | break; | |
144 | ||
88022d72 | 145 | if (!blk_mq_get_dispatch_budget(hctx)) |
1f460b63 | 146 | break; |
b347689f ML |
147 | |
148 | rq = blk_mq_dequeue_from_ctx(hctx, ctx); | |
149 | if (!rq) { | |
150 | blk_mq_put_dispatch_budget(hctx); | |
151 | break; | |
b347689f ML |
152 | } |
153 | ||
154 | /* | |
155 | * Now this rq owns the budget which has to be released | |
156 | * if this rq won't be queued to driver via .queue_rq() | |
157 | * in blk_mq_dispatch_rq_list(). | |
158 | */ | |
159 | list_add(&rq->queuelist, &rq_list); | |
160 | ||
161 | /* round robin for fair dispatch */ | |
162 | ctx = blk_mq_next_ctx(hctx, rq->mq_ctx); | |
163 | ||
164 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); | |
165 | ||
166 | WRITE_ONCE(hctx->dispatch_from, ctx); | |
b347689f ML |
167 | } |
168 | ||
1f460b63 | 169 | void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) |
bd166ef1 | 170 | { |
81380ca1 OS |
171 | struct request_queue *q = hctx->queue; |
172 | struct elevator_queue *e = q->elevator; | |
f9cd4bfe | 173 | const bool has_sched_dispatch = e && e->type->ops.dispatch_request; |
bd166ef1 JA |
174 | LIST_HEAD(rq_list); |
175 | ||
f4560ffe ML |
176 | /* RCU or SRCU read lock is needed before checking quiesced flag */ |
177 | if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q))) | |
1f460b63 | 178 | return; |
bd166ef1 JA |
179 | |
180 | hctx->run++; | |
181 | ||
182 | /* | |
183 | * If we have previous entries on our dispatch list, grab them first for | |
184 | * more fair dispatch. | |
185 | */ | |
186 | if (!list_empty_careful(&hctx->dispatch)) { | |
187 | spin_lock(&hctx->lock); | |
188 | if (!list_empty(&hctx->dispatch)) | |
189 | list_splice_init(&hctx->dispatch, &rq_list); | |
190 | spin_unlock(&hctx->lock); | |
191 | } | |
192 | ||
193 | /* | |
194 | * Only ask the scheduler for requests, if we didn't have residual | |
195 | * requests from the dispatch list. This is to avoid the case where | |
196 | * we only ever dispatch a fraction of the requests available because | |
197 | * of low device queue depth. Once we pull requests out of the IO | |
198 | * scheduler, we can no longer merge or sort them. So it's best to | |
199 | * leave them there for as long as we can. Mark the hw queue as | |
200 | * needing a restart in that case. | |
caf8eb0d ML |
201 | * |
202 | * We want to dispatch from the scheduler if there was nothing | |
203 | * on the dispatch list or we were able to dispatch from the | |
204 | * dispatch list. | |
bd166ef1 | 205 | */ |
c13660a0 | 206 | if (!list_empty(&rq_list)) { |
d38d3515 | 207 | blk_mq_sched_mark_restart_hctx(hctx); |
b347689f ML |
208 | if (blk_mq_dispatch_rq_list(q, &rq_list, false)) { |
209 | if (has_sched_dispatch) | |
1f460b63 | 210 | blk_mq_do_dispatch_sched(hctx); |
b347689f | 211 | else |
1f460b63 | 212 | blk_mq_do_dispatch_ctx(hctx); |
b347689f | 213 | } |
caf8eb0d | 214 | } else if (has_sched_dispatch) { |
1f460b63 | 215 | blk_mq_do_dispatch_sched(hctx); |
6e768717 ML |
216 | } else if (hctx->dispatch_busy) { |
217 | /* dequeue request one by one from sw queue if queue is busy */ | |
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, | |
c16d6b5a | 305 | struct blk_mq_hw_ctx *hctx, |
9c558734 JA |
306 | struct blk_mq_ctx *ctx, struct bio *bio) |
307 | { | |
c16d6b5a ML |
308 | enum hctx_type type = hctx->type; |
309 | ||
9c558734 JA |
310 | lockdep_assert_held(&ctx->lock); |
311 | ||
c16d6b5a | 312 | if (blk_mq_bio_list_merge(q, &ctx->rq_lists[type], bio)) { |
9c558734 JA |
313 | ctx->rq_merged++; |
314 | return true; | |
315 | } | |
316 | ||
317 | return false; | |
318 | } | |
9bddeb2a | 319 | |
bd166ef1 JA |
320 | bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) |
321 | { | |
322 | struct elevator_queue *e = q->elevator; | |
9bddeb2a | 323 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
8ccdf4a3 | 324 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, bio->bi_opf, ctx); |
9bddeb2a | 325 | bool ret = false; |
c16d6b5a | 326 | enum hctx_type type; |
bd166ef1 | 327 | |
f9cd4bfe | 328 | if (e && e->type->ops.bio_merge) { |
bd166ef1 | 329 | blk_mq_put_ctx(ctx); |
f9cd4bfe | 330 | return e->type->ops.bio_merge(hctx, bio); |
bd166ef1 JA |
331 | } |
332 | ||
c16d6b5a | 333 | type = hctx->type; |
b04f50ab | 334 | if ((hctx->flags & BLK_MQ_F_SHOULD_MERGE) && |
c16d6b5a | 335 | !list_empty_careful(&ctx->rq_lists[type])) { |
9bddeb2a ML |
336 | /* default per sw-queue merge */ |
337 | spin_lock(&ctx->lock); | |
c16d6b5a | 338 | ret = blk_mq_attempt_merge(q, hctx, ctx, bio); |
9bddeb2a ML |
339 | spin_unlock(&ctx->lock); |
340 | } | |
341 | ||
342 | blk_mq_put_ctx(ctx); | |
343 | return ret; | |
bd166ef1 JA |
344 | } |
345 | ||
346 | bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) | |
347 | { | |
348 | return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); | |
349 | } | |
350 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); | |
351 | ||
352 | void blk_mq_sched_request_inserted(struct request *rq) | |
353 | { | |
354 | trace_block_rq_insert(rq->q, rq); | |
355 | } | |
356 | EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); | |
357 | ||
0cacba6c | 358 | static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, |
a6a252e6 | 359 | bool has_sched, |
0cacba6c | 360 | struct request *rq) |
bd166ef1 | 361 | { |
a6a252e6 ML |
362 | /* dispatch flush rq directly */ |
363 | if (rq->rq_flags & RQF_FLUSH_SEQ) { | |
364 | spin_lock(&hctx->lock); | |
365 | list_add(&rq->queuelist, &hctx->dispatch); | |
366 | spin_unlock(&hctx->lock); | |
367 | return true; | |
368 | } | |
369 | ||
923218f6 | 370 | if (has_sched) |
bd166ef1 | 371 | rq->rq_flags |= RQF_SORTED; |
bd166ef1 | 372 | |
a6a252e6 | 373 | return false; |
bd166ef1 | 374 | } |
bd166ef1 | 375 | |
bd6737f1 | 376 | void blk_mq_sched_insert_request(struct request *rq, bool at_head, |
9e97d295 | 377 | bool run_queue, bool async) |
bd6737f1 JA |
378 | { |
379 | struct request_queue *q = rq->q; | |
380 | struct elevator_queue *e = q->elevator; | |
381 | struct blk_mq_ctx *ctx = rq->mq_ctx; | |
ea4f995e | 382 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; |
bd6737f1 | 383 | |
a6a252e6 ML |
384 | /* flush rq in flush machinery need to be dispatched directly */ |
385 | if (!(rq->rq_flags & RQF_FLUSH_SEQ) && op_is_flush(rq->cmd_flags)) { | |
923218f6 ML |
386 | blk_insert_flush(rq); |
387 | goto run; | |
bd6737f1 JA |
388 | } |
389 | ||
923218f6 ML |
390 | WARN_ON(e && (rq->tag != -1)); |
391 | ||
a6a252e6 | 392 | if (blk_mq_sched_bypass_insert(hctx, !!e, rq)) |
0cacba6c OS |
393 | goto run; |
394 | ||
f9cd4bfe | 395 | if (e && e->type->ops.insert_requests) { |
bd6737f1 JA |
396 | LIST_HEAD(list); |
397 | ||
398 | list_add(&rq->queuelist, &list); | |
f9cd4bfe | 399 | e->type->ops.insert_requests(hctx, &list, at_head); |
bd6737f1 JA |
400 | } else { |
401 | spin_lock(&ctx->lock); | |
402 | __blk_mq_insert_request(hctx, rq, at_head); | |
403 | spin_unlock(&ctx->lock); | |
404 | } | |
405 | ||
0cacba6c | 406 | run: |
bd6737f1 JA |
407 | if (run_queue) |
408 | blk_mq_run_hw_queue(hctx, async); | |
409 | } | |
410 | ||
67cae4c9 | 411 | void blk_mq_sched_insert_requests(struct blk_mq_hw_ctx *hctx, |
bd6737f1 JA |
412 | struct blk_mq_ctx *ctx, |
413 | struct list_head *list, bool run_queue_async) | |
414 | { | |
f9afca4d | 415 | struct elevator_queue *e; |
bd6737f1 | 416 | |
f9afca4d | 417 | e = hctx->queue->elevator; |
f9cd4bfe JA |
418 | if (e && e->type->ops.insert_requests) |
419 | e->type->ops.insert_requests(hctx, list, false); | |
6ce3dd6e ML |
420 | else { |
421 | /* | |
422 | * try to issue requests directly if the hw queue isn't | |
423 | * busy in case of 'none' scheduler, and this way may save | |
424 | * us one extra enqueue & dequeue to sw queue. | |
425 | */ | |
fd9c40f6 | 426 | if (!hctx->dispatch_busy && !e && !run_queue_async) { |
6ce3dd6e | 427 | blk_mq_try_issue_list_directly(hctx, list); |
fd9c40f6 BVA |
428 | if (list_empty(list)) |
429 | return; | |
430 | } | |
431 | blk_mq_insert_requests(hctx, ctx, list); | |
6ce3dd6e | 432 | } |
bd6737f1 JA |
433 | |
434 | blk_mq_run_hw_queue(hctx, run_queue_async); | |
435 | } | |
436 | ||
bd166ef1 JA |
437 | static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, |
438 | struct blk_mq_hw_ctx *hctx, | |
439 | unsigned int hctx_idx) | |
440 | { | |
441 | if (hctx->sched_tags) { | |
442 | blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); | |
443 | blk_mq_free_rq_map(hctx->sched_tags); | |
444 | hctx->sched_tags = NULL; | |
445 | } | |
446 | } | |
447 | ||
6917ff0b OS |
448 | static int blk_mq_sched_alloc_tags(struct request_queue *q, |
449 | struct blk_mq_hw_ctx *hctx, | |
450 | unsigned int hctx_idx) | |
451 | { | |
452 | struct blk_mq_tag_set *set = q->tag_set; | |
453 | int ret; | |
454 | ||
455 | hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, | |
456 | set->reserved_tags); | |
457 | if (!hctx->sched_tags) | |
458 | return -ENOMEM; | |
459 | ||
460 | ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); | |
461 | if (ret) | |
462 | blk_mq_sched_free_tags(set, hctx, hctx_idx); | |
463 | ||
464 | return ret; | |
465 | } | |
466 | ||
54d5329d | 467 | static void blk_mq_sched_tags_teardown(struct request_queue *q) |
bd166ef1 JA |
468 | { |
469 | struct blk_mq_tag_set *set = q->tag_set; | |
470 | struct blk_mq_hw_ctx *hctx; | |
6917ff0b OS |
471 | int i; |
472 | ||
473 | queue_for_each_hw_ctx(q, hctx, i) | |
474 | blk_mq_sched_free_tags(set, hctx, i); | |
475 | } | |
476 | ||
477 | int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) | |
478 | { | |
479 | struct blk_mq_hw_ctx *hctx; | |
ee056f98 | 480 | struct elevator_queue *eq; |
6917ff0b OS |
481 | unsigned int i; |
482 | int ret; | |
483 | ||
484 | if (!e) { | |
485 | q->elevator = NULL; | |
32a50fab | 486 | q->nr_requests = q->tag_set->queue_depth; |
6917ff0b OS |
487 | return 0; |
488 | } | |
bd166ef1 JA |
489 | |
490 | /* | |
32825c45 ML |
491 | * Default to double of smaller one between hw queue_depth and 128, |
492 | * since we don't split into sync/async like the old code did. | |
493 | * Additionally, this is a per-hw queue depth. | |
bd166ef1 | 494 | */ |
32825c45 ML |
495 | q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth, |
496 | BLKDEV_MAX_RQ); | |
bd166ef1 | 497 | |
bd166ef1 | 498 | queue_for_each_hw_ctx(q, hctx, i) { |
6917ff0b | 499 | ret = blk_mq_sched_alloc_tags(q, hctx, i); |
bd166ef1 | 500 | if (ret) |
6917ff0b | 501 | goto err; |
bd166ef1 JA |
502 | } |
503 | ||
f9cd4bfe | 504 | ret = e->ops.init_sched(q, e); |
6917ff0b OS |
505 | if (ret) |
506 | goto err; | |
bd166ef1 | 507 | |
d332ce09 OS |
508 | blk_mq_debugfs_register_sched(q); |
509 | ||
510 | queue_for_each_hw_ctx(q, hctx, i) { | |
f9cd4bfe JA |
511 | if (e->ops.init_hctx) { |
512 | ret = e->ops.init_hctx(hctx, i); | |
ee056f98 OS |
513 | if (ret) { |
514 | eq = q->elevator; | |
515 | blk_mq_exit_sched(q, eq); | |
516 | kobject_put(&eq->kobj); | |
517 | return ret; | |
518 | } | |
519 | } | |
d332ce09 | 520 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
ee056f98 OS |
521 | } |
522 | ||
bd166ef1 | 523 | return 0; |
bd166ef1 | 524 | |
6917ff0b | 525 | err: |
54d5329d OS |
526 | blk_mq_sched_tags_teardown(q); |
527 | q->elevator = NULL; | |
6917ff0b | 528 | return ret; |
bd166ef1 | 529 | } |
d3484991 | 530 | |
54d5329d OS |
531 | void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) |
532 | { | |
ee056f98 OS |
533 | struct blk_mq_hw_ctx *hctx; |
534 | unsigned int i; | |
535 | ||
d332ce09 OS |
536 | queue_for_each_hw_ctx(q, hctx, i) { |
537 | blk_mq_debugfs_unregister_sched_hctx(hctx); | |
f9cd4bfe JA |
538 | if (e->type->ops.exit_hctx && hctx->sched_data) { |
539 | e->type->ops.exit_hctx(hctx, i); | |
d332ce09 | 540 | hctx->sched_data = NULL; |
ee056f98 OS |
541 | } |
542 | } | |
d332ce09 | 543 | blk_mq_debugfs_unregister_sched(q); |
f9cd4bfe JA |
544 | if (e->type->ops.exit_sched) |
545 | e->type->ops.exit_sched(e); | |
54d5329d OS |
546 | blk_mq_sched_tags_teardown(q); |
547 | q->elevator = NULL; | |
548 | } |