]>
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 | ||
62 | if (hctx->flags & BLK_MQ_F_TAG_SHARED) { | |
63 | struct request_queue *q = hctx->queue; | |
64 | ||
65 | if (!test_and_set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
66 | atomic_inc(&q->shared_hctx_restart); | |
67 | } else | |
68 | set_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); | |
69 | } | |
70 | ||
358a3a6b | 71 | void blk_mq_sched_restart(struct blk_mq_hw_ctx *hctx) |
8e8320c9 JA |
72 | { |
73 | if (!test_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state)) | |
358a3a6b | 74 | return; |
8e8320c9 | 75 | |
358a3a6b | 76 | clear_bit(BLK_MQ_S_SCHED_RESTART, &hctx->state); |
8e8320c9 JA |
77 | |
78 | if (blk_mq_hctx_has_pending(hctx)) { | |
79 | blk_mq_run_hw_queue(hctx, true); | |
358a3a6b | 80 | return; |
8e8320c9 | 81 | } |
8e8320c9 JA |
82 | } |
83 | ||
1f460b63 ML |
84 | /* |
85 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
86 | * its queue by itself in its completion handler, so we don't need to | |
87 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
88 | */ | |
89 | static void blk_mq_do_dispatch_sched(struct blk_mq_hw_ctx *hctx) | |
caf8eb0d ML |
90 | { |
91 | struct request_queue *q = hctx->queue; | |
92 | struct elevator_queue *e = q->elevator; | |
93 | LIST_HEAD(rq_list); | |
94 | ||
95 | do { | |
de148297 ML |
96 | struct request *rq; |
97 | blk_status_t ret; | |
caf8eb0d | 98 | |
de148297 ML |
99 | if (e->type->ops.mq.has_work && |
100 | !e->type->ops.mq.has_work(hctx)) | |
caf8eb0d | 101 | break; |
de148297 ML |
102 | |
103 | ret = blk_mq_get_dispatch_budget(hctx); | |
104 | if (ret == BLK_STS_RESOURCE) | |
1f460b63 | 105 | break; |
de148297 ML |
106 | |
107 | rq = e->type->ops.mq.dispatch_request(hctx); | |
108 | if (!rq) { | |
109 | blk_mq_put_dispatch_budget(hctx); | |
110 | break; | |
111 | } else if (ret != BLK_STS_OK) { | |
112 | blk_mq_end_request(rq, ret); | |
113 | continue; | |
114 | } | |
115 | ||
116 | /* | |
117 | * Now this rq owns the budget which has to be released | |
118 | * if this rq won't be queued to driver via .queue_rq() | |
119 | * in blk_mq_dispatch_rq_list(). | |
120 | */ | |
caf8eb0d | 121 | list_add(&rq->queuelist, &rq_list); |
de148297 | 122 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); |
caf8eb0d ML |
123 | } |
124 | ||
b347689f ML |
125 | static struct blk_mq_ctx *blk_mq_next_ctx(struct blk_mq_hw_ctx *hctx, |
126 | struct blk_mq_ctx *ctx) | |
127 | { | |
128 | unsigned idx = ctx->index_hw; | |
129 | ||
130 | if (++idx == hctx->nr_ctx) | |
131 | idx = 0; | |
132 | ||
133 | return hctx->ctxs[idx]; | |
134 | } | |
135 | ||
1f460b63 ML |
136 | /* |
137 | * Only SCSI implements .get_budget and .put_budget, and SCSI restarts | |
138 | * its queue by itself in its completion handler, so we don't need to | |
139 | * restart queue if .get_budget() returns BLK_STS_NO_RESOURCE. | |
140 | */ | |
141 | static void blk_mq_do_dispatch_ctx(struct blk_mq_hw_ctx *hctx) | |
b347689f ML |
142 | { |
143 | struct request_queue *q = hctx->queue; | |
144 | LIST_HEAD(rq_list); | |
145 | struct blk_mq_ctx *ctx = READ_ONCE(hctx->dispatch_from); | |
146 | ||
147 | do { | |
148 | struct request *rq; | |
149 | blk_status_t ret; | |
150 | ||
151 | if (!sbitmap_any_bit_set(&hctx->ctx_map)) | |
152 | break; | |
153 | ||
154 | ret = blk_mq_get_dispatch_budget(hctx); | |
155 | if (ret == BLK_STS_RESOURCE) | |
1f460b63 | 156 | break; |
b347689f ML |
157 | |
158 | rq = blk_mq_dequeue_from_ctx(hctx, ctx); | |
159 | if (!rq) { | |
160 | blk_mq_put_dispatch_budget(hctx); | |
161 | break; | |
162 | } else if (ret != BLK_STS_OK) { | |
163 | blk_mq_end_request(rq, ret); | |
164 | continue; | |
165 | } | |
166 | ||
167 | /* | |
168 | * Now this rq owns the budget which has to be released | |
169 | * if this rq won't be queued to driver via .queue_rq() | |
170 | * in blk_mq_dispatch_rq_list(). | |
171 | */ | |
172 | list_add(&rq->queuelist, &rq_list); | |
173 | ||
174 | /* round robin for fair dispatch */ | |
175 | ctx = blk_mq_next_ctx(hctx, rq->mq_ctx); | |
176 | ||
177 | } while (blk_mq_dispatch_rq_list(q, &rq_list, true)); | |
178 | ||
179 | WRITE_ONCE(hctx->dispatch_from, ctx); | |
b347689f ML |
180 | } |
181 | ||
de148297 | 182 | /* return true if hw queue need to be run again */ |
1f460b63 | 183 | void blk_mq_sched_dispatch_requests(struct blk_mq_hw_ctx *hctx) |
bd166ef1 | 184 | { |
81380ca1 OS |
185 | struct request_queue *q = hctx->queue; |
186 | struct elevator_queue *e = q->elevator; | |
64765a75 | 187 | const bool has_sched_dispatch = e && e->type->ops.mq.dispatch_request; |
bd166ef1 JA |
188 | LIST_HEAD(rq_list); |
189 | ||
f4560ffe ML |
190 | /* RCU or SRCU read lock is needed before checking quiesced flag */ |
191 | if (unlikely(blk_mq_hctx_stopped(hctx) || blk_queue_quiesced(q))) | |
1f460b63 | 192 | return; |
bd166ef1 JA |
193 | |
194 | hctx->run++; | |
195 | ||
196 | /* | |
197 | * If we have previous entries on our dispatch list, grab them first for | |
198 | * more fair dispatch. | |
199 | */ | |
200 | if (!list_empty_careful(&hctx->dispatch)) { | |
201 | spin_lock(&hctx->lock); | |
202 | if (!list_empty(&hctx->dispatch)) | |
203 | list_splice_init(&hctx->dispatch, &rq_list); | |
204 | spin_unlock(&hctx->lock); | |
205 | } | |
206 | ||
207 | /* | |
208 | * Only ask the scheduler for requests, if we didn't have residual | |
209 | * requests from the dispatch list. This is to avoid the case where | |
210 | * we only ever dispatch a fraction of the requests available because | |
211 | * of low device queue depth. Once we pull requests out of the IO | |
212 | * scheduler, we can no longer merge or sort them. So it's best to | |
213 | * leave them there for as long as we can. Mark the hw queue as | |
214 | * needing a restart in that case. | |
caf8eb0d ML |
215 | * |
216 | * We want to dispatch from the scheduler if there was nothing | |
217 | * on the dispatch list or we were able to dispatch from the | |
218 | * dispatch list. | |
bd166ef1 | 219 | */ |
c13660a0 | 220 | if (!list_empty(&rq_list)) { |
d38d3515 | 221 | blk_mq_sched_mark_restart_hctx(hctx); |
b347689f ML |
222 | if (blk_mq_dispatch_rq_list(q, &rq_list, false)) { |
223 | if (has_sched_dispatch) | |
1f460b63 | 224 | blk_mq_do_dispatch_sched(hctx); |
b347689f | 225 | else |
1f460b63 | 226 | blk_mq_do_dispatch_ctx(hctx); |
b347689f | 227 | } |
caf8eb0d | 228 | } else if (has_sched_dispatch) { |
1f460b63 | 229 | blk_mq_do_dispatch_sched(hctx); |
b347689f ML |
230 | } else if (q->mq_ops->get_budget) { |
231 | /* | |
232 | * If we need to get budget before queuing request, we | |
233 | * dequeue request one by one from sw queue for avoiding | |
234 | * to mess up I/O merge when dispatch runs out of resource. | |
235 | * | |
236 | * TODO: get more budgets, and dequeue more requests in | |
237 | * one time. | |
238 | */ | |
1f460b63 | 239 | blk_mq_do_dispatch_ctx(hctx); |
caf8eb0d | 240 | } else { |
c13660a0 | 241 | blk_mq_flush_busy_ctxs(hctx, &rq_list); |
de148297 | 242 | blk_mq_dispatch_rq_list(q, &rq_list, false); |
64765a75 | 243 | } |
bd166ef1 JA |
244 | } |
245 | ||
e4d750c9 JA |
246 | bool blk_mq_sched_try_merge(struct request_queue *q, struct bio *bio, |
247 | struct request **merged_request) | |
bd166ef1 JA |
248 | { |
249 | struct request *rq; | |
bd166ef1 | 250 | |
34fe7c05 CH |
251 | switch (elv_merge(q, &rq, bio)) { |
252 | case ELEVATOR_BACK_MERGE: | |
bd166ef1 JA |
253 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
254 | return false; | |
34fe7c05 CH |
255 | if (!bio_attempt_back_merge(q, rq, bio)) |
256 | return false; | |
257 | *merged_request = attempt_back_merge(q, rq); | |
258 | if (!*merged_request) | |
259 | elv_merged_request(q, rq, ELEVATOR_BACK_MERGE); | |
260 | return true; | |
261 | case ELEVATOR_FRONT_MERGE: | |
bd166ef1 JA |
262 | if (!blk_mq_sched_allow_merge(q, rq, bio)) |
263 | return false; | |
34fe7c05 CH |
264 | if (!bio_attempt_front_merge(q, rq, bio)) |
265 | return false; | |
266 | *merged_request = attempt_front_merge(q, rq); | |
267 | if (!*merged_request) | |
268 | elv_merged_request(q, rq, ELEVATOR_FRONT_MERGE); | |
269 | return true; | |
270 | default: | |
271 | return false; | |
bd166ef1 | 272 | } |
bd166ef1 JA |
273 | } |
274 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_merge); | |
275 | ||
9bddeb2a ML |
276 | /* |
277 | * Reverse check our software queue for entries that we could potentially | |
278 | * merge with. Currently includes a hand-wavy stop count of 8, to not spend | |
279 | * too much time checking for merges. | |
280 | */ | |
281 | static bool blk_mq_attempt_merge(struct request_queue *q, | |
282 | struct blk_mq_ctx *ctx, struct bio *bio) | |
283 | { | |
284 | struct request *rq; | |
285 | int checked = 8; | |
286 | ||
7b607814 BVA |
287 | lockdep_assert_held(&ctx->lock); |
288 | ||
9bddeb2a ML |
289 | list_for_each_entry_reverse(rq, &ctx->rq_list, queuelist) { |
290 | bool merged = false; | |
291 | ||
292 | if (!checked--) | |
293 | break; | |
294 | ||
295 | if (!blk_rq_merge_ok(rq, bio)) | |
296 | continue; | |
297 | ||
298 | switch (blk_try_merge(rq, bio)) { | |
299 | case ELEVATOR_BACK_MERGE: | |
300 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
301 | merged = bio_attempt_back_merge(q, rq, bio); | |
302 | break; | |
303 | case ELEVATOR_FRONT_MERGE: | |
304 | if (blk_mq_sched_allow_merge(q, rq, bio)) | |
305 | merged = bio_attempt_front_merge(q, rq, bio); | |
306 | break; | |
307 | case ELEVATOR_DISCARD_MERGE: | |
308 | merged = bio_attempt_discard_merge(q, rq, bio); | |
309 | break; | |
310 | default: | |
311 | continue; | |
312 | } | |
313 | ||
314 | if (merged) | |
315 | ctx->rq_merged++; | |
316 | return merged; | |
317 | } | |
318 | ||
319 | return false; | |
320 | } | |
321 | ||
bd166ef1 JA |
322 | bool __blk_mq_sched_bio_merge(struct request_queue *q, struct bio *bio) |
323 | { | |
324 | struct elevator_queue *e = q->elevator; | |
9bddeb2a ML |
325 | struct blk_mq_ctx *ctx = blk_mq_get_ctx(q); |
326 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
327 | bool ret = false; | |
bd166ef1 | 328 | |
9bddeb2a | 329 | if (e && e->type->ops.mq.bio_merge) { |
bd166ef1 JA |
330 | blk_mq_put_ctx(ctx); |
331 | return e->type->ops.mq.bio_merge(hctx, bio); | |
332 | } | |
333 | ||
9bddeb2a ML |
334 | if (hctx->flags & BLK_MQ_F_SHOULD_MERGE) { |
335 | /* default per sw-queue merge */ | |
336 | spin_lock(&ctx->lock); | |
337 | ret = blk_mq_attempt_merge(q, ctx, bio); | |
338 | spin_unlock(&ctx->lock); | |
339 | } | |
340 | ||
341 | blk_mq_put_ctx(ctx); | |
342 | return ret; | |
bd166ef1 JA |
343 | } |
344 | ||
345 | bool blk_mq_sched_try_insert_merge(struct request_queue *q, struct request *rq) | |
346 | { | |
347 | return rq_mergeable(rq) && elv_attempt_insert_merge(q, rq); | |
348 | } | |
349 | EXPORT_SYMBOL_GPL(blk_mq_sched_try_insert_merge); | |
350 | ||
351 | void blk_mq_sched_request_inserted(struct request *rq) | |
352 | { | |
353 | trace_block_rq_insert(rq->q, rq); | |
354 | } | |
355 | EXPORT_SYMBOL_GPL(blk_mq_sched_request_inserted); | |
356 | ||
0cacba6c OS |
357 | static bool blk_mq_sched_bypass_insert(struct blk_mq_hw_ctx *hctx, |
358 | struct request *rq) | |
bd166ef1 JA |
359 | { |
360 | if (rq->tag == -1) { | |
361 | rq->rq_flags |= RQF_SORTED; | |
362 | return false; | |
363 | } | |
364 | ||
365 | /* | |
366 | * If we already have a real request tag, send directly to | |
367 | * the dispatch list. | |
368 | */ | |
369 | spin_lock(&hctx->lock); | |
370 | list_add(&rq->queuelist, &hctx->dispatch); | |
371 | spin_unlock(&hctx->lock); | |
372 | return true; | |
373 | } | |
bd166ef1 | 374 | |
bd6737f1 JA |
375 | /* |
376 | * Add flush/fua to the queue. If we fail getting a driver tag, then | |
377 | * punt to the requeue list. Requeue will re-invoke us from a context | |
378 | * that's safe to block from. | |
379 | */ | |
380 | static void blk_mq_sched_insert_flush(struct blk_mq_hw_ctx *hctx, | |
381 | struct request *rq, bool can_block) | |
382 | { | |
383 | if (blk_mq_get_driver_tag(rq, &hctx, can_block)) { | |
384 | blk_insert_flush(rq); | |
385 | blk_mq_run_hw_queue(hctx, true); | |
386 | } else | |
c7a571b4 | 387 | blk_mq_add_to_requeue_list(rq, false, true); |
bd6737f1 JA |
388 | } |
389 | ||
390 | void blk_mq_sched_insert_request(struct request *rq, bool at_head, | |
391 | bool run_queue, bool async, bool can_block) | |
392 | { | |
393 | struct request_queue *q = rq->q; | |
394 | struct elevator_queue *e = q->elevator; | |
395 | struct blk_mq_ctx *ctx = rq->mq_ctx; | |
396 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
397 | ||
f3a8ab7d | 398 | if (rq->tag == -1 && op_is_flush(rq->cmd_flags)) { |
bd6737f1 JA |
399 | blk_mq_sched_insert_flush(hctx, rq, can_block); |
400 | return; | |
401 | } | |
402 | ||
0cacba6c OS |
403 | if (e && blk_mq_sched_bypass_insert(hctx, rq)) |
404 | goto run; | |
405 | ||
bd6737f1 JA |
406 | if (e && e->type->ops.mq.insert_requests) { |
407 | LIST_HEAD(list); | |
408 | ||
409 | list_add(&rq->queuelist, &list); | |
410 | e->type->ops.mq.insert_requests(hctx, &list, at_head); | |
411 | } else { | |
412 | spin_lock(&ctx->lock); | |
413 | __blk_mq_insert_request(hctx, rq, at_head); | |
414 | spin_unlock(&ctx->lock); | |
415 | } | |
416 | ||
0cacba6c | 417 | run: |
bd6737f1 JA |
418 | if (run_queue) |
419 | blk_mq_run_hw_queue(hctx, async); | |
420 | } | |
421 | ||
422 | void blk_mq_sched_insert_requests(struct request_queue *q, | |
423 | struct blk_mq_ctx *ctx, | |
424 | struct list_head *list, bool run_queue_async) | |
425 | { | |
426 | struct blk_mq_hw_ctx *hctx = blk_mq_map_queue(q, ctx->cpu); | |
427 | struct elevator_queue *e = hctx->queue->elevator; | |
428 | ||
0cacba6c OS |
429 | if (e) { |
430 | struct request *rq, *next; | |
431 | ||
432 | /* | |
433 | * We bypass requests that already have a driver tag assigned, | |
434 | * which should only be flushes. Flushes are only ever inserted | |
435 | * as single requests, so we shouldn't ever hit the | |
436 | * WARN_ON_ONCE() below (but let's handle it just in case). | |
437 | */ | |
438 | list_for_each_entry_safe(rq, next, list, queuelist) { | |
439 | if (WARN_ON_ONCE(rq->tag != -1)) { | |
440 | list_del_init(&rq->queuelist); | |
441 | blk_mq_sched_bypass_insert(hctx, rq); | |
442 | } | |
443 | } | |
444 | } | |
445 | ||
bd6737f1 JA |
446 | if (e && e->type->ops.mq.insert_requests) |
447 | e->type->ops.mq.insert_requests(hctx, list, false); | |
448 | else | |
449 | blk_mq_insert_requests(hctx, ctx, list); | |
450 | ||
451 | blk_mq_run_hw_queue(hctx, run_queue_async); | |
452 | } | |
453 | ||
bd166ef1 JA |
454 | static void blk_mq_sched_free_tags(struct blk_mq_tag_set *set, |
455 | struct blk_mq_hw_ctx *hctx, | |
456 | unsigned int hctx_idx) | |
457 | { | |
458 | if (hctx->sched_tags) { | |
459 | blk_mq_free_rqs(set, hctx->sched_tags, hctx_idx); | |
460 | blk_mq_free_rq_map(hctx->sched_tags); | |
461 | hctx->sched_tags = NULL; | |
462 | } | |
463 | } | |
464 | ||
6917ff0b OS |
465 | static int blk_mq_sched_alloc_tags(struct request_queue *q, |
466 | struct blk_mq_hw_ctx *hctx, | |
467 | unsigned int hctx_idx) | |
468 | { | |
469 | struct blk_mq_tag_set *set = q->tag_set; | |
470 | int ret; | |
471 | ||
472 | hctx->sched_tags = blk_mq_alloc_rq_map(set, hctx_idx, q->nr_requests, | |
473 | set->reserved_tags); | |
474 | if (!hctx->sched_tags) | |
475 | return -ENOMEM; | |
476 | ||
477 | ret = blk_mq_alloc_rqs(set, hctx->sched_tags, hctx_idx, q->nr_requests); | |
478 | if (ret) | |
479 | blk_mq_sched_free_tags(set, hctx, hctx_idx); | |
480 | ||
481 | return ret; | |
482 | } | |
483 | ||
54d5329d | 484 | static void blk_mq_sched_tags_teardown(struct request_queue *q) |
bd166ef1 JA |
485 | { |
486 | struct blk_mq_tag_set *set = q->tag_set; | |
487 | struct blk_mq_hw_ctx *hctx; | |
6917ff0b OS |
488 | int i; |
489 | ||
490 | queue_for_each_hw_ctx(q, hctx, i) | |
491 | blk_mq_sched_free_tags(set, hctx, i); | |
492 | } | |
493 | ||
93252632 OS |
494 | int blk_mq_sched_init_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, |
495 | unsigned int hctx_idx) | |
496 | { | |
497 | struct elevator_queue *e = q->elevator; | |
ee056f98 | 498 | int ret; |
93252632 OS |
499 | |
500 | if (!e) | |
501 | return 0; | |
502 | ||
ee056f98 OS |
503 | ret = blk_mq_sched_alloc_tags(q, hctx, hctx_idx); |
504 | if (ret) | |
505 | return ret; | |
506 | ||
507 | if (e->type->ops.mq.init_hctx) { | |
508 | ret = e->type->ops.mq.init_hctx(hctx, hctx_idx); | |
509 | if (ret) { | |
510 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); | |
511 | return ret; | |
512 | } | |
513 | } | |
514 | ||
d332ce09 OS |
515 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
516 | ||
ee056f98 | 517 | return 0; |
93252632 OS |
518 | } |
519 | ||
520 | void blk_mq_sched_exit_hctx(struct request_queue *q, struct blk_mq_hw_ctx *hctx, | |
521 | unsigned int hctx_idx) | |
522 | { | |
523 | struct elevator_queue *e = q->elevator; | |
524 | ||
525 | if (!e) | |
526 | return; | |
527 | ||
d332ce09 OS |
528 | blk_mq_debugfs_unregister_sched_hctx(hctx); |
529 | ||
ee056f98 OS |
530 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { |
531 | e->type->ops.mq.exit_hctx(hctx, hctx_idx); | |
532 | hctx->sched_data = NULL; | |
533 | } | |
534 | ||
93252632 OS |
535 | blk_mq_sched_free_tags(q->tag_set, hctx, hctx_idx); |
536 | } | |
537 | ||
6917ff0b OS |
538 | int blk_mq_init_sched(struct request_queue *q, struct elevator_type *e) |
539 | { | |
540 | struct blk_mq_hw_ctx *hctx; | |
ee056f98 | 541 | struct elevator_queue *eq; |
6917ff0b OS |
542 | unsigned int i; |
543 | int ret; | |
544 | ||
545 | if (!e) { | |
546 | q->elevator = NULL; | |
547 | return 0; | |
548 | } | |
bd166ef1 JA |
549 | |
550 | /* | |
32825c45 ML |
551 | * Default to double of smaller one between hw queue_depth and 128, |
552 | * since we don't split into sync/async like the old code did. | |
553 | * Additionally, this is a per-hw queue depth. | |
bd166ef1 | 554 | */ |
32825c45 ML |
555 | q->nr_requests = 2 * min_t(unsigned int, q->tag_set->queue_depth, |
556 | BLKDEV_MAX_RQ); | |
bd166ef1 | 557 | |
bd166ef1 | 558 | queue_for_each_hw_ctx(q, hctx, i) { |
6917ff0b | 559 | ret = blk_mq_sched_alloc_tags(q, hctx, i); |
bd166ef1 | 560 | if (ret) |
6917ff0b | 561 | goto err; |
bd166ef1 JA |
562 | } |
563 | ||
6917ff0b OS |
564 | ret = e->ops.mq.init_sched(q, e); |
565 | if (ret) | |
566 | goto err; | |
bd166ef1 | 567 | |
d332ce09 OS |
568 | blk_mq_debugfs_register_sched(q); |
569 | ||
570 | queue_for_each_hw_ctx(q, hctx, i) { | |
571 | if (e->ops.mq.init_hctx) { | |
ee056f98 OS |
572 | ret = e->ops.mq.init_hctx(hctx, i); |
573 | if (ret) { | |
574 | eq = q->elevator; | |
575 | blk_mq_exit_sched(q, eq); | |
576 | kobject_put(&eq->kobj); | |
577 | return ret; | |
578 | } | |
579 | } | |
d332ce09 | 580 | blk_mq_debugfs_register_sched_hctx(q, hctx); |
ee056f98 OS |
581 | } |
582 | ||
bd166ef1 | 583 | return 0; |
bd166ef1 | 584 | |
6917ff0b | 585 | err: |
54d5329d OS |
586 | blk_mq_sched_tags_teardown(q); |
587 | q->elevator = NULL; | |
6917ff0b | 588 | return ret; |
bd166ef1 | 589 | } |
d3484991 | 590 | |
54d5329d OS |
591 | void blk_mq_exit_sched(struct request_queue *q, struct elevator_queue *e) |
592 | { | |
ee056f98 OS |
593 | struct blk_mq_hw_ctx *hctx; |
594 | unsigned int i; | |
595 | ||
d332ce09 OS |
596 | queue_for_each_hw_ctx(q, hctx, i) { |
597 | blk_mq_debugfs_unregister_sched_hctx(hctx); | |
598 | if (e->type->ops.mq.exit_hctx && hctx->sched_data) { | |
599 | e->type->ops.mq.exit_hctx(hctx, i); | |
600 | hctx->sched_data = NULL; | |
ee056f98 OS |
601 | } |
602 | } | |
d332ce09 | 603 | blk_mq_debugfs_unregister_sched(q); |
54d5329d OS |
604 | if (e->type->ops.mq.exit_sched) |
605 | e->type->ops.mq.exit_sched(e); | |
606 | blk_mq_sched_tags_teardown(q); | |
607 | q->elevator = NULL; | |
608 | } | |
609 | ||
d3484991 JA |
610 | int blk_mq_sched_init(struct request_queue *q) |
611 | { | |
612 | int ret; | |
613 | ||
d3484991 JA |
614 | mutex_lock(&q->sysfs_lock); |
615 | ret = elevator_init(q, NULL); | |
616 | mutex_unlock(&q->sysfs_lock); | |
617 | ||
618 | return ret; | |
619 | } |