Commit | Line | Data |
---|---|---|
3bd94003 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
4cc96131 MS |
2 | /* |
3 | * Copyright (C) 2016 Red Hat, Inc. All rights reserved. | |
4 | * | |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm-core.h" | |
9 | #include "dm-rq.h" | |
10 | ||
4cc96131 MS |
11 | #include <linux/blk-mq.h> |
12 | ||
13 | #define DM_MSG_PREFIX "core-rq" | |
14 | ||
e689fbab MS |
15 | /* |
16 | * One of these is allocated per request. | |
17 | */ | |
18 | struct dm_rq_target_io { | |
19 | struct mapped_device *md; | |
20 | struct dm_target *ti; | |
21 | struct request *orig, *clone; | |
22 | struct kthread_work work; | |
23 | blk_status_t error; | |
24 | union map_info info; | |
25 | struct dm_stats_aux stats_aux; | |
26 | unsigned long duration_jiffies; | |
86a3238c HM |
27 | unsigned int n_sectors; |
28 | unsigned int completed; | |
e689fbab MS |
29 | }; |
30 | ||
4cc96131 MS |
31 | #define DM_MQ_NR_HW_QUEUES 1 |
32 | #define DM_MQ_QUEUE_DEPTH 2048 | |
86a3238c HM |
33 | static unsigned int dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES; |
34 | static unsigned int dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH; | |
4cc96131 MS |
35 | |
36 | /* | |
37 | * Request-based DM's mempools' reserved IOs set by the user. | |
38 | */ | |
39 | #define RESERVED_REQUEST_BASED_IOS 256 | |
86a3238c | 40 | static unsigned int reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; |
4cc96131 | 41 | |
86a3238c | 42 | unsigned int dm_get_reserved_rq_based_ios(void) |
4cc96131 MS |
43 | { |
44 | return __dm_get_module_param(&reserved_rq_based_ios, | |
45 | RESERVED_REQUEST_BASED_IOS, DM_RESERVED_MAX_IOS); | |
46 | } | |
4cc96131 | 47 | |
86a3238c | 48 | static unsigned int dm_get_blk_mq_nr_hw_queues(void) |
4cc96131 MS |
49 | { |
50 | return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32); | |
51 | } | |
52 | ||
86a3238c | 53 | static unsigned int dm_get_blk_mq_queue_depth(void) |
4cc96131 MS |
54 | { |
55 | return __dm_get_module_param(&dm_mq_queue_depth, | |
56 | DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH); | |
57 | } | |
58 | ||
59 | int dm_request_based(struct mapped_device *md) | |
60 | { | |
344e9ffc | 61 | return queue_is_mq(md->queue); |
4cc96131 MS |
62 | } |
63 | ||
6a23e05c | 64 | void dm_start_queue(struct request_queue *q) |
9dbeaeab | 65 | { |
f660174e | 66 | blk_mq_unquiesce_queue(q); |
9dbeaeab MS |
67 | blk_mq_kick_requeue_list(q); |
68 | } | |
69 | ||
6a23e05c | 70 | void dm_stop_queue(struct request_queue *q) |
2397a15a | 71 | { |
7b17c2f7 | 72 | blk_mq_quiesce_queue(q); |
4cc96131 MS |
73 | } |
74 | ||
4cc96131 MS |
75 | /* |
76 | * Partial completion handling for request-based dm | |
77 | */ | |
78 | static void end_clone_bio(struct bio *clone) | |
79 | { | |
80 | struct dm_rq_clone_bio_info *info = | |
81 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
82 | struct dm_rq_target_io *tio = info->tio; | |
4cc96131 | 83 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
4e4cbee9 | 84 | blk_status_t error = clone->bi_status; |
dc6364b5 | 85 | bool is_last = !clone->bi_next; |
4cc96131 MS |
86 | |
87 | bio_put(clone); | |
88 | ||
89 | if (tio->error) | |
90 | /* | |
91 | * An error has already been detected on the request. | |
92 | * Once error occurred, just let clone->end_io() handle | |
93 | * the remainder. | |
94 | */ | |
95 | return; | |
96 | else if (error) { | |
97 | /* | |
98 | * Don't notice the error to the upper layer yet. | |
99 | * The error handling decision is made by the target driver, | |
100 | * when the request is completed. | |
101 | */ | |
102 | tio->error = error; | |
dc6364b5 | 103 | goto exit; |
4cc96131 MS |
104 | } |
105 | ||
106 | /* | |
107 | * I/O for the bio successfully completed. | |
108 | * Notice the data completion to the upper layer. | |
109 | */ | |
dc6364b5 | 110 | tio->completed += nr_bytes; |
4cc96131 MS |
111 | |
112 | /* | |
113 | * Update the original request. | |
d370ad23 | 114 | * Do not use blk_mq_end_request() here, because it may complete |
4cc96131 MS |
115 | * the original request before the clone, and break the ordering. |
116 | */ | |
dc6364b5 ML |
117 | if (is_last) |
118 | exit: | |
119 | blk_update_request(tio->orig, BLK_STS_OK, tio->completed); | |
4cc96131 MS |
120 | } |
121 | ||
122 | static struct dm_rq_target_io *tio_from_request(struct request *rq) | |
123 | { | |
eb8db831 | 124 | return blk_mq_rq_to_pdu(rq); |
4cc96131 MS |
125 | } |
126 | ||
127 | static void rq_end_stats(struct mapped_device *md, struct request *orig) | |
128 | { | |
129 | if (unlikely(dm_stats_used(&md->stats))) { | |
130 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
131 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
132 | dm_stats_account_io(&md->stats, rq_data_dir(orig), | |
133 | blk_rq_pos(orig), tio->n_sectors, true, | |
134 | tio->duration_jiffies, &tio->stats_aux); | |
135 | } | |
136 | } | |
137 | ||
138 | /* | |
139 | * Don't touch any member of the md after calling this function because | |
140 | * the md may be freed in dm_put() at the end of this function. | |
141 | * Or do dm_get() before calling this function and dm_put() later. | |
142 | */ | |
2adc5c55 | 143 | static void rq_completed(struct mapped_device *md) |
4cc96131 | 144 | { |
4cc96131 MS |
145 | /* |
146 | * dm_put() must be at the end of this function. See the comment above | |
147 | */ | |
148 | dm_put(md); | |
149 | } | |
150 | ||
4cc96131 MS |
151 | /* |
152 | * Complete the clone and the original request. | |
153 | * Must be called without clone's queue lock held, | |
154 | * see end_clone_request() for more details. | |
155 | */ | |
2a842aca | 156 | static void dm_end_request(struct request *clone, blk_status_t error) |
4cc96131 | 157 | { |
4cc96131 MS |
158 | struct dm_rq_target_io *tio = clone->end_io_data; |
159 | struct mapped_device *md = tio->md; | |
160 | struct request *rq = tio->orig; | |
161 | ||
eb8db831 | 162 | blk_rq_unprep_clone(clone); |
5de719e3 | 163 | tio->ti->type->release_clone_rq(clone, NULL); |
4cc96131 | 164 | |
4cc96131 | 165 | rq_end_stats(md, rq); |
6a23e05c | 166 | blk_mq_end_request(rq, error); |
2adc5c55 | 167 | rq_completed(md); |
4cc96131 MS |
168 | } |
169 | ||
e0c10752 | 170 | static void __dm_mq_kick_requeue_list(struct request_queue *q, unsigned long msecs) |
4cc96131 | 171 | { |
52d7f1b5 | 172 | blk_mq_delay_kick_requeue_list(q, msecs); |
4cc96131 MS |
173 | } |
174 | ||
e0c10752 MS |
175 | void dm_mq_kick_requeue_list(struct mapped_device *md) |
176 | { | |
33bd6f06 | 177 | __dm_mq_kick_requeue_list(md->queue, 0); |
e0c10752 MS |
178 | } |
179 | EXPORT_SYMBOL(dm_mq_kick_requeue_list); | |
180 | ||
181 | static void dm_mq_delay_requeue_request(struct request *rq, unsigned long msecs) | |
182 | { | |
2b053aca | 183 | blk_mq_requeue_request(rq, false); |
e0c10752 MS |
184 | __dm_mq_kick_requeue_list(rq->q, msecs); |
185 | } | |
186 | ||
fbc39b4c | 187 | static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_requeue) |
4cc96131 | 188 | { |
fbc39b4c MS |
189 | struct mapped_device *md = tio->md; |
190 | struct request *rq = tio->orig; | |
d5c27f3f | 191 | unsigned long delay_ms = delay_requeue ? 100 : 0; |
4cc96131 MS |
192 | |
193 | rq_end_stats(md, rq); | |
eb8db831 CH |
194 | if (tio->clone) { |
195 | blk_rq_unprep_clone(tio->clone); | |
5de719e3 | 196 | tio->ti->type->release_clone_rq(tio->clone, NULL); |
eb8db831 | 197 | } |
4cc96131 | 198 | |
6a23e05c | 199 | dm_mq_delay_requeue_request(rq, delay_ms); |
2adc5c55 | 200 | rq_completed(md); |
4cc96131 MS |
201 | } |
202 | ||
2a842aca | 203 | static void dm_done(struct request *clone, blk_status_t error, bool mapped) |
4cc96131 | 204 | { |
7ed8578a | 205 | int r = DM_ENDIO_DONE; |
4cc96131 MS |
206 | struct dm_rq_target_io *tio = clone->end_io_data; |
207 | dm_request_endio_fn rq_end_io = NULL; | |
208 | ||
209 | if (tio->ti) { | |
210 | rq_end_io = tio->ti->type->rq_end_io; | |
211 | ||
212 | if (mapped && rq_end_io) | |
213 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
214 | } | |
215 | ||
2a842aca | 216 | if (unlikely(error == BLK_STS_TARGET)) { |
bcb44433 MS |
217 | if (req_op(clone) == REQ_OP_DISCARD && |
218 | !clone->q->limits.max_discard_sectors) | |
219 | disable_discard(tio->md); | |
bcb44433 MS |
220 | else if (req_op(clone) == REQ_OP_WRITE_ZEROES && |
221 | !clone->q->limits.max_write_zeroes_sectors) | |
ac62d620 CH |
222 | disable_write_zeroes(tio->md); |
223 | } | |
4cc96131 | 224 | |
7ed8578a CH |
225 | switch (r) { |
226 | case DM_ENDIO_DONE: | |
4cc96131 | 227 | /* The target wants to complete the I/O */ |
7ed8578a CH |
228 | dm_end_request(clone, error); |
229 | break; | |
230 | case DM_ENDIO_INCOMPLETE: | |
4cc96131 MS |
231 | /* The target will handle the I/O */ |
232 | return; | |
7ed8578a | 233 | case DM_ENDIO_REQUEUE: |
4cc96131 | 234 | /* The target wants to requeue the I/O */ |
fbc39b4c | 235 | dm_requeue_original_request(tio, false); |
7ed8578a | 236 | break; |
ac514ffc MS |
237 | case DM_ENDIO_DELAY_REQUEUE: |
238 | /* The target wants to requeue the I/O after a delay */ | |
239 | dm_requeue_original_request(tio, true); | |
240 | break; | |
7ed8578a | 241 | default: |
43e6c111 | 242 | DMCRIT("unimplemented target endio return value: %d", r); |
4cc96131 MS |
243 | BUG(); |
244 | } | |
245 | } | |
246 | ||
247 | /* | |
248 | * Request completion handler for request-based dm | |
249 | */ | |
250 | static void dm_softirq_done(struct request *rq) | |
251 | { | |
252 | bool mapped = true; | |
253 | struct dm_rq_target_io *tio = tio_from_request(rq); | |
254 | struct request *clone = tio->clone; | |
4cc96131 MS |
255 | |
256 | if (!clone) { | |
61febef4 JA |
257 | struct mapped_device *md = tio->md; |
258 | ||
259 | rq_end_stats(md, rq); | |
6a23e05c | 260 | blk_mq_end_request(rq, tio->error); |
2adc5c55 | 261 | rq_completed(md); |
4cc96131 MS |
262 | return; |
263 | } | |
264 | ||
e8064021 | 265 | if (rq->rq_flags & RQF_FAILED) |
4cc96131 MS |
266 | mapped = false; |
267 | ||
268 | dm_done(clone, tio->error, mapped); | |
269 | } | |
270 | ||
271 | /* | |
272 | * Complete the clone and the original request with the error status | |
273 | * through softirq context. | |
274 | */ | |
2a842aca | 275 | static void dm_complete_request(struct request *rq, blk_status_t error) |
4cc96131 MS |
276 | { |
277 | struct dm_rq_target_io *tio = tio_from_request(rq); | |
278 | ||
279 | tio->error = error; | |
15f73f5b CH |
280 | if (likely(!blk_should_fake_timeout(rq->q))) |
281 | blk_mq_complete_request(rq); | |
4cc96131 MS |
282 | } |
283 | ||
284 | /* | |
285 | * Complete the not-mapped clone and the original request with the error status | |
286 | * through softirq context. | |
287 | * Target's rq_end_io() function isn't called. | |
6a23e05c | 288 | * This may be used when the target's clone_and_map_rq() function fails. |
4cc96131 | 289 | */ |
2a842aca | 290 | static void dm_kill_unmapped_request(struct request *rq, blk_status_t error) |
4cc96131 | 291 | { |
e8064021 | 292 | rq->rq_flags |= RQF_FAILED; |
4cc96131 MS |
293 | dm_complete_request(rq, error); |
294 | } | |
295 | ||
de671d61 JA |
296 | static enum rq_end_io_ret end_clone_request(struct request *clone, |
297 | blk_status_t error) | |
4cc96131 MS |
298 | { |
299 | struct dm_rq_target_io *tio = clone->end_io_data; | |
300 | ||
4cc96131 | 301 | dm_complete_request(tio->orig, error); |
de671d61 | 302 | return RQ_END_IO_NONE; |
4cc96131 MS |
303 | } |
304 | ||
4cc96131 MS |
305 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
306 | void *data) | |
307 | { | |
308 | struct dm_rq_target_io *tio = data; | |
309 | struct dm_rq_clone_bio_info *info = | |
310 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
311 | ||
312 | info->orig = bio_orig; | |
313 | info->tio = tio; | |
314 | bio->bi_end_io = end_clone_bio; | |
315 | ||
316 | return 0; | |
317 | } | |
318 | ||
319 | static int setup_clone(struct request *clone, struct request *rq, | |
320 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
321 | { | |
322 | int r; | |
323 | ||
29dec90a | 324 | r = blk_rq_prep_clone(clone, rq, &tio->md->mempools->bs, gfp_mask, |
4cc96131 MS |
325 | dm_rq_bio_constructor, tio); |
326 | if (r) | |
327 | return r; | |
328 | ||
4cc96131 MS |
329 | clone->end_io = end_clone_request; |
330 | clone->end_io_data = tio; | |
331 | ||
332 | tio->clone = clone; | |
333 | ||
334 | return 0; | |
335 | } | |
336 | ||
4cc96131 MS |
337 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
338 | struct mapped_device *md) | |
339 | { | |
340 | tio->md = md; | |
341 | tio->ti = NULL; | |
342 | tio->clone = NULL; | |
343 | tio->orig = rq; | |
344 | tio->error = 0; | |
dc6364b5 | 345 | tio->completed = 0; |
4cc96131 MS |
346 | /* |
347 | * Avoid initializing info for blk-mq; it passes | |
348 | * target-specific data through info.ptr | |
349 | * (see: dm_mq_init_request) | |
350 | */ | |
351 | if (!md->init_tio_pdu) | |
352 | memset(&tio->info, 0, sizeof(tio->info)); | |
4cc96131 MS |
353 | } |
354 | ||
4cc96131 MS |
355 | /* |
356 | * Returns: | |
a8ac51e4 MS |
357 | * DM_MAPIO_* : the request has been processed as indicated |
358 | * DM_MAPIO_REQUEUE : the original request needs to be immediately requeued | |
4cc96131 MS |
359 | * < 0 : the request was completed due to failure |
360 | */ | |
fbc39b4c | 361 | static int map_request(struct dm_rq_target_io *tio) |
4cc96131 MS |
362 | { |
363 | int r; | |
364 | struct dm_target *ti = tio->ti; | |
fbc39b4c MS |
365 | struct mapped_device *md = tio->md; |
366 | struct request *rq = tio->orig; | |
4cc96131 | 367 | struct request *clone = NULL; |
396eaf21 | 368 | blk_status_t ret; |
4cc96131 | 369 | |
eb8db831 | 370 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); |
4cc96131 MS |
371 | switch (r) { |
372 | case DM_MAPIO_SUBMITTED: | |
373 | /* The target has taken the I/O to submit by itself later */ | |
374 | break; | |
375 | case DM_MAPIO_REMAPPED: | |
eb8db831 CH |
376 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
377 | /* -ENOMEM */ | |
5de719e3 | 378 | ti->type->release_clone_rq(clone, &tio->info); |
eb8db831 CH |
379 | return DM_MAPIO_REQUEUE; |
380 | } | |
381 | ||
4cc96131 | 382 | /* The target has remapped the I/O so dispatch it */ |
a54895fa | 383 | trace_block_rq_remap(clone, disk_devt(dm_disk(md)), |
4cc96131 | 384 | blk_rq_pos(rq)); |
9f9adea7 CH |
385 | ret = blk_insert_cloned_request(clone); |
386 | switch (ret) { | |
387 | case BLK_STS_OK: | |
388 | break; | |
389 | case BLK_STS_RESOURCE: | |
390 | case BLK_STS_DEV_RESOURCE: | |
396eaf21 | 391 | blk_rq_unprep_clone(clone); |
226b4fc7 | 392 | blk_mq_cleanup_rq(clone); |
5de719e3 | 393 | tio->ti->type->release_clone_rq(clone, &tio->info); |
396eaf21 | 394 | tio->clone = NULL; |
34743bfd | 395 | return DM_MAPIO_REQUEUE; |
9f9adea7 CH |
396 | default: |
397 | /* must complete clone in terms of original request */ | |
398 | dm_complete_request(rq, ret); | |
396eaf21 | 399 | } |
4cc96131 MS |
400 | break; |
401 | case DM_MAPIO_REQUEUE: | |
402 | /* The target wants to requeue the I/O */ | |
a8ac51e4 MS |
403 | break; |
404 | case DM_MAPIO_DELAY_REQUEUE: | |
405 | /* The target wants to requeue the I/O after a delay */ | |
fbc39b4c | 406 | dm_requeue_original_request(tio, true); |
4cc96131 | 407 | break; |
412445ac | 408 | case DM_MAPIO_KILL: |
4cc96131 | 409 | /* The target wants to complete the I/O */ |
2a842aca | 410 | dm_kill_unmapped_request(rq, BLK_STS_IOERR); |
ece07280 | 411 | break; |
412445ac | 412 | default: |
43e6c111 | 413 | DMCRIT("unimplemented target map return value: %d", r); |
412445ac | 414 | BUG(); |
4cc96131 MS |
415 | } |
416 | ||
a8ac51e4 | 417 | return r; |
4cc96131 MS |
418 | } |
419 | ||
6a23e05c JA |
420 | /* DEPRECATED: previously used for request-based merge heuristic in dm_request_fn() */ |
421 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
422 | { | |
423 | return sprintf(buf, "%u\n", 0); | |
424 | } | |
425 | ||
426 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
427 | const char *buf, size_t count) | |
428 | { | |
429 | return count; | |
430 | } | |
431 | ||
4cc96131 MS |
432 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
433 | { | |
6a23e05c | 434 | blk_mq_start_request(orig); |
4cc96131 | 435 | |
4cc96131 MS |
436 | if (unlikely(dm_stats_used(&md->stats))) { |
437 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
438 | tio->duration_jiffies = jiffies; | |
439 | tio->n_sectors = blk_rq_sectors(orig); | |
440 | dm_stats_account_io(&md->stats, rq_data_dir(orig), | |
441 | blk_rq_pos(orig), tio->n_sectors, false, 0, | |
442 | &tio->stats_aux); | |
443 | } | |
444 | ||
445 | /* | |
446 | * Hold the md reference here for the in-flight I/O. | |
447 | * We can't rely on the reference count by device opener, | |
448 | * because the device may be closed during the request completion | |
449 | * when all bios are completed. | |
450 | * See the comment in rq_completed() too. | |
451 | */ | |
452 | dm_get(md); | |
453 | } | |
454 | ||
6a23e05c JA |
455 | static int dm_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, |
456 | unsigned int hctx_idx, unsigned int numa_node) | |
eb8db831 | 457 | { |
6a23e05c | 458 | struct mapped_device *md = set->driver_data; |
eb8db831 CH |
459 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); |
460 | ||
461 | /* | |
462 | * Must initialize md member of tio, otherwise it won't | |
463 | * be available in dm_mq_queue_rq. | |
464 | */ | |
465 | tio->md = md; | |
466 | ||
467 | if (md->init_tio_pdu) { | |
468 | /* target-specific per-io data is immediately after the tio */ | |
469 | tio->info.ptr = tio + 1; | |
470 | } | |
471 | ||
472 | return 0; | |
473 | } | |
474 | ||
fc17b653 | 475 | static blk_status_t dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, |
4cc96131 MS |
476 | const struct blk_mq_queue_data *bd) |
477 | { | |
478 | struct request *rq = bd->rq; | |
479 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
480 | struct mapped_device *md = tio->md; | |
481 | struct dm_target *ti = md->immutable_target; | |
482 | ||
b4459b11 ML |
483 | /* |
484 | * blk-mq's unquiesce may come from outside events, such as | |
485 | * elevator switch, updating nr_requests or others, and request may | |
486 | * come during suspend, so simply ask for blk-mq to requeue it. | |
487 | */ | |
488 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) | |
489 | return BLK_STS_RESOURCE; | |
490 | ||
4cc96131 MS |
491 | if (unlikely(!ti)) { |
492 | int srcu_idx; | |
fa247089 | 493 | struct dm_table *map; |
4cc96131 | 494 | |
fa247089 MS |
495 | map = dm_get_live_table(md, &srcu_idx); |
496 | if (unlikely(!map)) { | |
497 | dm_put_live_table(md, srcu_idx); | |
498 | return BLK_STS_RESOURCE; | |
499 | } | |
4cc96131 MS |
500 | ti = dm_table_find_target(map, 0); |
501 | dm_put_live_table(md, srcu_idx); | |
502 | } | |
503 | ||
504 | if (ti->type->busy && ti->type->busy(ti)) | |
fc17b653 | 505 | return BLK_STS_RESOURCE; |
4cc96131 MS |
506 | |
507 | dm_start_request(md, rq); | |
508 | ||
509 | /* Init tio using md established in .init_request */ | |
510 | init_tio(tio, rq, md); | |
511 | ||
512 | /* | |
513 | * Establish tio->ti before calling map_request(). | |
514 | */ | |
515 | tio->ti = ti; | |
516 | ||
517 | /* Direct call is fine since .queue_rq allows allocations */ | |
fbc39b4c | 518 | if (map_request(tio) == DM_MAPIO_REQUEUE) { |
4cc96131 MS |
519 | /* Undo dm_start_request() before requeuing */ |
520 | rq_end_stats(md, rq); | |
2adc5c55 | 521 | rq_completed(md); |
fc17b653 | 522 | return BLK_STS_RESOURCE; |
4cc96131 MS |
523 | } |
524 | ||
fc17b653 | 525 | return BLK_STS_OK; |
4cc96131 MS |
526 | } |
527 | ||
f363b089 | 528 | static const struct blk_mq_ops dm_mq_ops = { |
4cc96131 | 529 | .queue_rq = dm_mq_queue_rq, |
4cc96131 MS |
530 | .complete = dm_softirq_done, |
531 | .init_request = dm_mq_init_request, | |
532 | }; | |
533 | ||
e83068a5 | 534 | int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t) |
4cc96131 | 535 | { |
e83068a5 | 536 | struct dm_target *immutable_tgt; |
4cc96131 MS |
537 | int err; |
538 | ||
4cc96131 MS |
539 | md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id); |
540 | if (!md->tag_set) | |
541 | return -ENOMEM; | |
542 | ||
543 | md->tag_set->ops = &dm_mq_ops; | |
544 | md->tag_set->queue_depth = dm_get_blk_mq_queue_depth(); | |
545 | md->tag_set->numa_node = md->numa_node_id; | |
bf0beec0 | 546 | md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_STACKING; |
4cc96131 MS |
547 | md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues(); |
548 | md->tag_set->driver_data = md; | |
549 | ||
550 | md->tag_set->cmd_size = sizeof(struct dm_rq_target_io); | |
e83068a5 | 551 | immutable_tgt = dm_table_get_immutable_target(t); |
4cc96131 MS |
552 | if (immutable_tgt && immutable_tgt->per_io_data_size) { |
553 | /* any target-specific per-io data is immediately after the tio */ | |
554 | md->tag_set->cmd_size += immutable_tgt->per_io_data_size; | |
555 | md->init_tio_pdu = true; | |
556 | } | |
557 | ||
558 | err = blk_mq_alloc_tag_set(md->tag_set); | |
559 | if (err) | |
560 | goto out_kfree_tag_set; | |
561 | ||
26a9750a CH |
562 | err = blk_mq_init_allocated_queue(md->tag_set, md->queue); |
563 | if (err) | |
4cc96131 | 564 | goto out_tag_set; |
4cc96131 MS |
565 | return 0; |
566 | ||
567 | out_tag_set: | |
568 | blk_mq_free_tag_set(md->tag_set); | |
569 | out_kfree_tag_set: | |
570 | kfree(md->tag_set); | |
8e947c8f | 571 | md->tag_set = NULL; |
4cc96131 MS |
572 | |
573 | return err; | |
574 | } | |
575 | ||
576 | void dm_mq_cleanup_mapped_device(struct mapped_device *md) | |
577 | { | |
578 | if (md->tag_set) { | |
579 | blk_mq_free_tag_set(md->tag_set); | |
580 | kfree(md->tag_set); | |
8e947c8f | 581 | md->tag_set = NULL; |
4cc96131 MS |
582 | } |
583 | } | |
584 | ||
585 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); | |
586 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
587 | ||
6a23e05c JA |
588 | /* Unused, but preserved for userspace compatibility */ |
589 | static bool use_blk_mq = true; | |
4cc96131 MS |
590 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
591 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
592 | ||
593 | module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR); | |
594 | MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices"); | |
595 | ||
596 | module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR); | |
597 | MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices"); |