]>
Commit | Line | Data |
---|---|---|
4cc96131 MS |
1 | /* |
2 | * Copyright (C) 2016 Red Hat, Inc. All rights reserved. | |
3 | * | |
4 | * This file is released under the GPL. | |
5 | */ | |
6 | ||
7 | #include "dm-core.h" | |
8 | #include "dm-rq.h" | |
9 | ||
10 | #include <linux/elevator.h> /* for rq_end_sector() */ | |
11 | #include <linux/blk-mq.h> | |
12 | ||
13 | #define DM_MSG_PREFIX "core-rq" | |
14 | ||
15 | #define DM_MQ_NR_HW_QUEUES 1 | |
16 | #define DM_MQ_QUEUE_DEPTH 2048 | |
17 | static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES; | |
18 | static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH; | |
19 | ||
20 | /* | |
21 | * Request-based DM's mempools' reserved IOs set by the user. | |
22 | */ | |
23 | #define RESERVED_REQUEST_BASED_IOS 256 | |
24 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
25 | ||
b23df0d0 | 26 | static bool use_blk_mq = IS_ENABLED(CONFIG_DM_MQ_DEFAULT); |
4cc96131 MS |
27 | |
28 | bool dm_use_blk_mq_default(void) | |
29 | { | |
30 | return use_blk_mq; | |
31 | } | |
32 | ||
33 | bool dm_use_blk_mq(struct mapped_device *md) | |
34 | { | |
35 | return md->use_blk_mq; | |
36 | } | |
37 | EXPORT_SYMBOL_GPL(dm_use_blk_mq); | |
38 | ||
39 | unsigned dm_get_reserved_rq_based_ios(void) | |
40 | { | |
41 | return __dm_get_module_param(&reserved_rq_based_ios, | |
42 | RESERVED_REQUEST_BASED_IOS, DM_RESERVED_MAX_IOS); | |
43 | } | |
44 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
45 | ||
46 | static unsigned dm_get_blk_mq_nr_hw_queues(void) | |
47 | { | |
48 | return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32); | |
49 | } | |
50 | ||
51 | static unsigned dm_get_blk_mq_queue_depth(void) | |
52 | { | |
53 | return __dm_get_module_param(&dm_mq_queue_depth, | |
54 | DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH); | |
55 | } | |
56 | ||
57 | int dm_request_based(struct mapped_device *md) | |
58 | { | |
59 | return blk_queue_stackable(md->queue); | |
60 | } | |
61 | ||
62 | static void dm_old_start_queue(struct request_queue *q) | |
63 | { | |
64 | unsigned long flags; | |
65 | ||
66 | spin_lock_irqsave(q->queue_lock, flags); | |
67 | if (blk_queue_stopped(q)) | |
68 | blk_start_queue(q); | |
69 | spin_unlock_irqrestore(q->queue_lock, flags); | |
70 | } | |
71 | ||
9dbeaeab MS |
72 | static void dm_mq_start_queue(struct request_queue *q) |
73 | { | |
9dbeaeab MS |
74 | blk_mq_start_stopped_hw_queues(q, true); |
75 | blk_mq_kick_requeue_list(q); | |
76 | } | |
77 | ||
4cc96131 MS |
78 | void dm_start_queue(struct request_queue *q) |
79 | { | |
80 | if (!q->mq_ops) | |
81 | dm_old_start_queue(q); | |
9dbeaeab MS |
82 | else |
83 | dm_mq_start_queue(q); | |
4cc96131 MS |
84 | } |
85 | ||
86 | static void dm_old_stop_queue(struct request_queue *q) | |
87 | { | |
88 | unsigned long flags; | |
89 | ||
90 | spin_lock_irqsave(q->queue_lock, flags); | |
c533f249 BVA |
91 | if (!blk_queue_stopped(q)) |
92 | blk_stop_queue(q); | |
4cc96131 MS |
93 | spin_unlock_irqrestore(q->queue_lock, flags); |
94 | } | |
95 | ||
2397a15a BVA |
96 | static void dm_mq_stop_queue(struct request_queue *q) |
97 | { | |
f0d33ab7 | 98 | if (blk_mq_queue_stopped(q)) |
4cc96131 | 99 | return; |
2397a15a | 100 | |
7b17c2f7 | 101 | blk_mq_quiesce_queue(q); |
4cc96131 MS |
102 | } |
103 | ||
104 | void dm_stop_queue(struct request_queue *q) | |
105 | { | |
106 | if (!q->mq_ops) | |
107 | dm_old_stop_queue(q); | |
2397a15a BVA |
108 | else |
109 | dm_mq_stop_queue(q); | |
4cc96131 MS |
110 | } |
111 | ||
4cc96131 MS |
112 | /* |
113 | * Partial completion handling for request-based dm | |
114 | */ | |
115 | static void end_clone_bio(struct bio *clone) | |
116 | { | |
117 | struct dm_rq_clone_bio_info *info = | |
118 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
119 | struct dm_rq_target_io *tio = info->tio; | |
120 | struct bio *bio = info->orig; | |
121 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; | |
122 | int error = clone->bi_error; | |
123 | ||
124 | bio_put(clone); | |
125 | ||
126 | if (tio->error) | |
127 | /* | |
128 | * An error has already been detected on the request. | |
129 | * Once error occurred, just let clone->end_io() handle | |
130 | * the remainder. | |
131 | */ | |
132 | return; | |
133 | else if (error) { | |
134 | /* | |
135 | * Don't notice the error to the upper layer yet. | |
136 | * The error handling decision is made by the target driver, | |
137 | * when the request is completed. | |
138 | */ | |
139 | tio->error = error; | |
140 | return; | |
141 | } | |
142 | ||
143 | /* | |
144 | * I/O for the bio successfully completed. | |
145 | * Notice the data completion to the upper layer. | |
146 | */ | |
147 | ||
148 | /* | |
149 | * bios are processed from the head of the list. | |
150 | * So the completing bio should always be rq->bio. | |
151 | * If it's not, something wrong is happening. | |
152 | */ | |
153 | if (tio->orig->bio != bio) | |
154 | DMERR("bio completion is going in the middle of the request"); | |
155 | ||
156 | /* | |
157 | * Update the original request. | |
158 | * Do not use blk_end_request() here, because it may complete | |
159 | * the original request before the clone, and break the ordering. | |
160 | */ | |
161 | blk_update_request(tio->orig, 0, nr_bytes); | |
162 | } | |
163 | ||
164 | static struct dm_rq_target_io *tio_from_request(struct request *rq) | |
165 | { | |
eb8db831 | 166 | return blk_mq_rq_to_pdu(rq); |
4cc96131 MS |
167 | } |
168 | ||
169 | static void rq_end_stats(struct mapped_device *md, struct request *orig) | |
170 | { | |
171 | if (unlikely(dm_stats_used(&md->stats))) { | |
172 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
173 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
174 | dm_stats_account_io(&md->stats, rq_data_dir(orig), | |
175 | blk_rq_pos(orig), tio->n_sectors, true, | |
176 | tio->duration_jiffies, &tio->stats_aux); | |
177 | } | |
178 | } | |
179 | ||
180 | /* | |
181 | * Don't touch any member of the md after calling this function because | |
182 | * the md may be freed in dm_put() at the end of this function. | |
183 | * Or do dm_get() before calling this function and dm_put() later. | |
184 | */ | |
185 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) | |
186 | { | |
d15bb3a6 BVA |
187 | struct request_queue *q = md->queue; |
188 | unsigned long flags; | |
189 | ||
4cc96131 MS |
190 | atomic_dec(&md->pending[rw]); |
191 | ||
192 | /* nudge anyone waiting on suspend queue */ | |
193 | if (!md_in_flight(md)) | |
194 | wake_up(&md->wait); | |
195 | ||
196 | /* | |
197 | * Run this off this callpath, as drivers could invoke end_io while | |
198 | * inside their request_fn (and holding the queue lock). Calling | |
199 | * back into ->request_fn() could deadlock attempting to grab the | |
200 | * queue lock again. | |
201 | */ | |
d15bb3a6 BVA |
202 | if (!q->mq_ops && run_queue) { |
203 | spin_lock_irqsave(q->queue_lock, flags); | |
204 | blk_run_queue_async(q); | |
205 | spin_unlock_irqrestore(q->queue_lock, flags); | |
206 | } | |
4cc96131 MS |
207 | |
208 | /* | |
209 | * dm_put() must be at the end of this function. See the comment above | |
210 | */ | |
211 | dm_put(md); | |
212 | } | |
213 | ||
4cc96131 MS |
214 | /* |
215 | * Complete the clone and the original request. | |
216 | * Must be called without clone's queue lock held, | |
217 | * see end_clone_request() for more details. | |
218 | */ | |
219 | static void dm_end_request(struct request *clone, int error) | |
220 | { | |
221 | int rw = rq_data_dir(clone); | |
222 | struct dm_rq_target_io *tio = clone->end_io_data; | |
223 | struct mapped_device *md = tio->md; | |
224 | struct request *rq = tio->orig; | |
225 | ||
eb8db831 CH |
226 | blk_rq_unprep_clone(clone); |
227 | tio->ti->type->release_clone_rq(clone); | |
4cc96131 | 228 | |
4cc96131 MS |
229 | rq_end_stats(md, rq); |
230 | if (!rq->q->mq_ops) | |
231 | blk_end_request_all(rq, error); | |
232 | else | |
233 | blk_mq_end_request(rq, error); | |
234 | rq_completed(md, rw, true); | |
235 | } | |
236 | ||
4cc96131 MS |
237 | /* |
238 | * Requeue the original request of a clone. | |
239 | */ | |
240 | static void dm_old_requeue_request(struct request *rq) | |
241 | { | |
242 | struct request_queue *q = rq->q; | |
243 | unsigned long flags; | |
244 | ||
245 | spin_lock_irqsave(q->queue_lock, flags); | |
246 | blk_requeue_request(q, rq); | |
247 | blk_run_queue_async(q); | |
248 | spin_unlock_irqrestore(q->queue_lock, flags); | |
249 | } | |
250 | ||
e0c10752 | 251 | static void __dm_mq_kick_requeue_list(struct request_queue *q, unsigned long msecs) |
4cc96131 | 252 | { |
52d7f1b5 | 253 | blk_mq_delay_kick_requeue_list(q, msecs); |
4cc96131 MS |
254 | } |
255 | ||
e0c10752 MS |
256 | void dm_mq_kick_requeue_list(struct mapped_device *md) |
257 | { | |
258 | __dm_mq_kick_requeue_list(dm_get_md_queue(md), 0); | |
259 | } | |
260 | EXPORT_SYMBOL(dm_mq_kick_requeue_list); | |
261 | ||
262 | static void dm_mq_delay_requeue_request(struct request *rq, unsigned long msecs) | |
263 | { | |
2b053aca | 264 | blk_mq_requeue_request(rq, false); |
e0c10752 MS |
265 | __dm_mq_kick_requeue_list(rq->q, msecs); |
266 | } | |
267 | ||
fbc39b4c | 268 | static void dm_requeue_original_request(struct dm_rq_target_io *tio, bool delay_requeue) |
4cc96131 | 269 | { |
fbc39b4c MS |
270 | struct mapped_device *md = tio->md; |
271 | struct request *rq = tio->orig; | |
4cc96131 MS |
272 | int rw = rq_data_dir(rq); |
273 | ||
274 | rq_end_stats(md, rq); | |
eb8db831 CH |
275 | if (tio->clone) { |
276 | blk_rq_unprep_clone(tio->clone); | |
277 | tio->ti->type->release_clone_rq(tio->clone); | |
278 | } | |
4cc96131 MS |
279 | |
280 | if (!rq->q->mq_ops) | |
281 | dm_old_requeue_request(rq); | |
282 | else | |
06eb061f | 283 | dm_mq_delay_requeue_request(rq, delay_requeue ? 100/*ms*/ : 0); |
4cc96131 MS |
284 | |
285 | rq_completed(md, rw, false); | |
286 | } | |
287 | ||
288 | static void dm_done(struct request *clone, int error, bool mapped) | |
289 | { | |
7ed8578a | 290 | int r = DM_ENDIO_DONE; |
4cc96131 MS |
291 | struct dm_rq_target_io *tio = clone->end_io_data; |
292 | dm_request_endio_fn rq_end_io = NULL; | |
293 | ||
294 | if (tio->ti) { | |
295 | rq_end_io = tio->ti->type->rq_end_io; | |
296 | ||
297 | if (mapped && rq_end_io) | |
298 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
299 | } | |
300 | ||
7ed8578a | 301 | if (unlikely(error == -EREMOTEIO)) { |
ac62d620 CH |
302 | if (req_op(clone) == REQ_OP_WRITE_SAME && |
303 | !clone->q->limits.max_write_same_sectors) | |
304 | disable_write_same(tio->md); | |
305 | if (req_op(clone) == REQ_OP_WRITE_ZEROES && | |
306 | !clone->q->limits.max_write_zeroes_sectors) | |
307 | disable_write_zeroes(tio->md); | |
308 | } | |
4cc96131 | 309 | |
7ed8578a CH |
310 | switch (r) { |
311 | case DM_ENDIO_DONE: | |
4cc96131 | 312 | /* The target wants to complete the I/O */ |
7ed8578a CH |
313 | dm_end_request(clone, error); |
314 | break; | |
315 | case DM_ENDIO_INCOMPLETE: | |
4cc96131 MS |
316 | /* The target will handle the I/O */ |
317 | return; | |
7ed8578a | 318 | case DM_ENDIO_REQUEUE: |
4cc96131 | 319 | /* The target wants to requeue the I/O */ |
fbc39b4c | 320 | dm_requeue_original_request(tio, false); |
7ed8578a CH |
321 | break; |
322 | default: | |
4cc96131 MS |
323 | DMWARN("unimplemented target endio return value: %d", r); |
324 | BUG(); | |
325 | } | |
326 | } | |
327 | ||
328 | /* | |
329 | * Request completion handler for request-based dm | |
330 | */ | |
331 | static void dm_softirq_done(struct request *rq) | |
332 | { | |
333 | bool mapped = true; | |
334 | struct dm_rq_target_io *tio = tio_from_request(rq); | |
335 | struct request *clone = tio->clone; | |
336 | int rw; | |
337 | ||
338 | if (!clone) { | |
61febef4 JA |
339 | struct mapped_device *md = tio->md; |
340 | ||
341 | rq_end_stats(md, rq); | |
4cc96131 | 342 | rw = rq_data_dir(rq); |
eb8db831 | 343 | if (!rq->q->mq_ops) |
4cc96131 | 344 | blk_end_request_all(rq, tio->error); |
eb8db831 | 345 | else |
4cc96131 | 346 | blk_mq_end_request(rq, tio->error); |
61febef4 | 347 | rq_completed(md, rw, false); |
4cc96131 MS |
348 | return; |
349 | } | |
350 | ||
e8064021 | 351 | if (rq->rq_flags & RQF_FAILED) |
4cc96131 MS |
352 | mapped = false; |
353 | ||
354 | dm_done(clone, tio->error, mapped); | |
355 | } | |
356 | ||
357 | /* | |
358 | * Complete the clone and the original request with the error status | |
359 | * through softirq context. | |
360 | */ | |
361 | static void dm_complete_request(struct request *rq, int error) | |
362 | { | |
363 | struct dm_rq_target_io *tio = tio_from_request(rq); | |
364 | ||
365 | tio->error = error; | |
366 | if (!rq->q->mq_ops) | |
367 | blk_complete_request(rq); | |
368 | else | |
08e0029a | 369 | blk_mq_complete_request(rq); |
4cc96131 MS |
370 | } |
371 | ||
372 | /* | |
373 | * Complete the not-mapped clone and the original request with the error status | |
374 | * through softirq context. | |
375 | * Target's rq_end_io() function isn't called. | |
376 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. | |
377 | */ | |
378 | static void dm_kill_unmapped_request(struct request *rq, int error) | |
379 | { | |
e8064021 | 380 | rq->rq_flags |= RQF_FAILED; |
4cc96131 MS |
381 | dm_complete_request(rq, error); |
382 | } | |
383 | ||
384 | /* | |
385 | * Called with the clone's queue lock held (in the case of .request_fn) | |
386 | */ | |
387 | static void end_clone_request(struct request *clone, int error) | |
388 | { | |
389 | struct dm_rq_target_io *tio = clone->end_io_data; | |
390 | ||
4cc96131 MS |
391 | /* |
392 | * Actual request completion is done in a softirq context which doesn't | |
393 | * hold the clone's queue lock. Otherwise, deadlock could occur because: | |
394 | * - another request may be submitted by the upper level driver | |
395 | * of the stacking during the completion | |
396 | * - the submission which requires queue lock may be done | |
397 | * against this clone's queue | |
398 | */ | |
399 | dm_complete_request(tio->orig, error); | |
400 | } | |
401 | ||
402 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) | |
403 | { | |
404 | int r; | |
405 | ||
406 | if (blk_queue_io_stat(clone->q)) | |
e8064021 | 407 | clone->rq_flags |= RQF_IO_STAT; |
4cc96131 MS |
408 | |
409 | clone->start_time = jiffies; | |
410 | r = blk_insert_cloned_request(clone->q, clone); | |
411 | if (r) | |
412 | /* must complete clone in terms of original request */ | |
413 | dm_complete_request(rq, r); | |
414 | } | |
415 | ||
416 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
417 | void *data) | |
418 | { | |
419 | struct dm_rq_target_io *tio = data; | |
420 | struct dm_rq_clone_bio_info *info = | |
421 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
422 | ||
423 | info->orig = bio_orig; | |
424 | info->tio = tio; | |
425 | bio->bi_end_io = end_clone_bio; | |
426 | ||
427 | return 0; | |
428 | } | |
429 | ||
430 | static int setup_clone(struct request *clone, struct request *rq, | |
431 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
432 | { | |
433 | int r; | |
434 | ||
435 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
436 | dm_rq_bio_constructor, tio); | |
437 | if (r) | |
438 | return r; | |
439 | ||
4cc96131 MS |
440 | clone->end_io = end_clone_request; |
441 | clone->end_io_data = tio; | |
442 | ||
443 | tio->clone = clone; | |
444 | ||
445 | return 0; | |
446 | } | |
447 | ||
4cc96131 MS |
448 | static void map_tio_request(struct kthread_work *work); |
449 | ||
450 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, | |
451 | struct mapped_device *md) | |
452 | { | |
453 | tio->md = md; | |
454 | tio->ti = NULL; | |
455 | tio->clone = NULL; | |
456 | tio->orig = rq; | |
457 | tio->error = 0; | |
458 | /* | |
459 | * Avoid initializing info for blk-mq; it passes | |
460 | * target-specific data through info.ptr | |
461 | * (see: dm_mq_init_request) | |
462 | */ | |
463 | if (!md->init_tio_pdu) | |
464 | memset(&tio->info, 0, sizeof(tio->info)); | |
465 | if (md->kworker_task) | |
3989144f | 466 | kthread_init_work(&tio->work, map_tio_request); |
4cc96131 MS |
467 | } |
468 | ||
4cc96131 MS |
469 | /* |
470 | * Returns: | |
a8ac51e4 MS |
471 | * DM_MAPIO_* : the request has been processed as indicated |
472 | * DM_MAPIO_REQUEUE : the original request needs to be immediately requeued | |
4cc96131 MS |
473 | * < 0 : the request was completed due to failure |
474 | */ | |
fbc39b4c | 475 | static int map_request(struct dm_rq_target_io *tio) |
4cc96131 MS |
476 | { |
477 | int r; | |
478 | struct dm_target *ti = tio->ti; | |
fbc39b4c MS |
479 | struct mapped_device *md = tio->md; |
480 | struct request *rq = tio->orig; | |
4cc96131 MS |
481 | struct request *clone = NULL; |
482 | ||
eb8db831 | 483 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); |
4cc96131 MS |
484 | switch (r) { |
485 | case DM_MAPIO_SUBMITTED: | |
486 | /* The target has taken the I/O to submit by itself later */ | |
487 | break; | |
488 | case DM_MAPIO_REMAPPED: | |
eb8db831 CH |
489 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
490 | /* -ENOMEM */ | |
491 | ti->type->release_clone_rq(clone); | |
492 | return DM_MAPIO_REQUEUE; | |
493 | } | |
494 | ||
4cc96131 MS |
495 | /* The target has remapped the I/O so dispatch it */ |
496 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), | |
497 | blk_rq_pos(rq)); | |
498 | dm_dispatch_clone_request(clone, rq); | |
499 | break; | |
500 | case DM_MAPIO_REQUEUE: | |
501 | /* The target wants to requeue the I/O */ | |
a8ac51e4 MS |
502 | break; |
503 | case DM_MAPIO_DELAY_REQUEUE: | |
504 | /* The target wants to requeue the I/O after a delay */ | |
fbc39b4c | 505 | dm_requeue_original_request(tio, true); |
4cc96131 | 506 | break; |
412445ac | 507 | case DM_MAPIO_KILL: |
4cc96131 | 508 | /* The target wants to complete the I/O */ |
412445ac CH |
509 | dm_kill_unmapped_request(rq, -EIO); |
510 | default: | |
511 | DMWARN("unimplemented target map return value: %d", r); | |
512 | BUG(); | |
4cc96131 MS |
513 | } |
514 | ||
a8ac51e4 | 515 | return r; |
4cc96131 MS |
516 | } |
517 | ||
518 | static void dm_start_request(struct mapped_device *md, struct request *orig) | |
519 | { | |
520 | if (!orig->q->mq_ops) | |
521 | blk_start_request(orig); | |
522 | else | |
523 | blk_mq_start_request(orig); | |
524 | atomic_inc(&md->pending[rq_data_dir(orig)]); | |
525 | ||
526 | if (md->seq_rq_merge_deadline_usecs) { | |
527 | md->last_rq_pos = rq_end_sector(orig); | |
528 | md->last_rq_rw = rq_data_dir(orig); | |
529 | md->last_rq_start_time = ktime_get(); | |
530 | } | |
531 | ||
532 | if (unlikely(dm_stats_used(&md->stats))) { | |
533 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
534 | tio->duration_jiffies = jiffies; | |
535 | tio->n_sectors = blk_rq_sectors(orig); | |
536 | dm_stats_account_io(&md->stats, rq_data_dir(orig), | |
537 | blk_rq_pos(orig), tio->n_sectors, false, 0, | |
538 | &tio->stats_aux); | |
539 | } | |
540 | ||
541 | /* | |
542 | * Hold the md reference here for the in-flight I/O. | |
543 | * We can't rely on the reference count by device opener, | |
544 | * because the device may be closed during the request completion | |
545 | * when all bios are completed. | |
546 | * See the comment in rq_completed() too. | |
547 | */ | |
548 | dm_get(md); | |
549 | } | |
550 | ||
eb8db831 CH |
551 | static int __dm_rq_init_rq(struct mapped_device *md, struct request *rq) |
552 | { | |
553 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
554 | ||
555 | /* | |
556 | * Must initialize md member of tio, otherwise it won't | |
557 | * be available in dm_mq_queue_rq. | |
558 | */ | |
559 | tio->md = md; | |
560 | ||
561 | if (md->init_tio_pdu) { | |
562 | /* target-specific per-io data is immediately after the tio */ | |
563 | tio->info.ptr = tio + 1; | |
564 | } | |
565 | ||
566 | return 0; | |
567 | } | |
568 | ||
569 | static int dm_rq_init_rq(struct request_queue *q, struct request *rq, gfp_t gfp) | |
570 | { | |
571 | return __dm_rq_init_rq(q->rq_alloc_data, rq); | |
572 | } | |
573 | ||
4cc96131 MS |
574 | static void map_tio_request(struct kthread_work *work) |
575 | { | |
576 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); | |
4cc96131 | 577 | |
fbc39b4c MS |
578 | if (map_request(tio) == DM_MAPIO_REQUEUE) |
579 | dm_requeue_original_request(tio, false); | |
4cc96131 MS |
580 | } |
581 | ||
582 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
583 | { | |
584 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
585 | } | |
586 | ||
587 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 | |
588 | ||
589 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
590 | const char *buf, size_t count) | |
591 | { | |
592 | unsigned deadline; | |
593 | ||
e83068a5 | 594 | if (dm_get_md_type(md) != DM_TYPE_REQUEST_BASED) |
4cc96131 MS |
595 | return count; |
596 | ||
597 | if (kstrtouint(buf, 10, &deadline)) | |
598 | return -EINVAL; | |
599 | ||
600 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
601 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
602 | ||
603 | md->seq_rq_merge_deadline_usecs = deadline; | |
604 | ||
605 | return count; | |
606 | } | |
607 | ||
608 | static bool dm_old_request_peeked_before_merge_deadline(struct mapped_device *md) | |
609 | { | |
610 | ktime_t kt_deadline; | |
611 | ||
612 | if (!md->seq_rq_merge_deadline_usecs) | |
613 | return false; | |
614 | ||
615 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
616 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
617 | ||
618 | return !ktime_after(ktime_get(), kt_deadline); | |
619 | } | |
620 | ||
621 | /* | |
622 | * q->request_fn for old request-based dm. | |
623 | * Called with the queue lock held. | |
624 | */ | |
625 | static void dm_old_request_fn(struct request_queue *q) | |
626 | { | |
627 | struct mapped_device *md = q->queuedata; | |
628 | struct dm_target *ti = md->immutable_target; | |
629 | struct request *rq; | |
630 | struct dm_rq_target_io *tio; | |
631 | sector_t pos = 0; | |
632 | ||
633 | if (unlikely(!ti)) { | |
634 | int srcu_idx; | |
635 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
636 | ||
4087a1ff MS |
637 | if (unlikely(!map)) { |
638 | dm_put_live_table(md, srcu_idx); | |
639 | return; | |
640 | } | |
4cc96131 MS |
641 | ti = dm_table_find_target(map, pos); |
642 | dm_put_live_table(md, srcu_idx); | |
643 | } | |
644 | ||
645 | /* | |
646 | * For suspend, check blk_queue_stopped() and increment | |
647 | * ->pending within a single queue_lock not to increment the | |
648 | * number of in-flight I/Os after the queue is stopped in | |
649 | * dm_suspend(). | |
650 | */ | |
651 | while (!blk_queue_stopped(q)) { | |
652 | rq = blk_peek_request(q); | |
653 | if (!rq) | |
654 | return; | |
655 | ||
656 | /* always use block 0 to find the target for flushes for now */ | |
657 | pos = 0; | |
658 | if (req_op(rq) != REQ_OP_FLUSH) | |
659 | pos = blk_rq_pos(rq); | |
660 | ||
661 | if ((dm_old_request_peeked_before_merge_deadline(md) && | |
4f9c74c6 | 662 | md_in_flight(md) && rq->bio && !bio_multiple_segments(rq->bio) && |
4cc96131 MS |
663 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) || |
664 | (ti->type->busy && ti->type->busy(ti))) { | |
bd9f55ea | 665 | blk_delay_queue(q, 10); |
4cc96131 MS |
666 | return; |
667 | } | |
668 | ||
669 | dm_start_request(md, rq); | |
670 | ||
671 | tio = tio_from_request(rq); | |
eb8db831 | 672 | init_tio(tio, rq, md); |
4cc96131 MS |
673 | /* Establish tio->ti before queuing work (map_tio_request) */ |
674 | tio->ti = ti; | |
3989144f | 675 | kthread_queue_work(&md->kworker, &tio->work); |
4cc96131 MS |
676 | BUG_ON(!irqs_disabled()); |
677 | } | |
678 | } | |
679 | ||
680 | /* | |
681 | * Fully initialize a .request_fn request-based queue. | |
682 | */ | |
eb8db831 | 683 | int dm_old_init_request_queue(struct mapped_device *md, struct dm_table *t) |
4cc96131 | 684 | { |
eb8db831 CH |
685 | struct dm_target *immutable_tgt; |
686 | ||
4cc96131 | 687 | /* Fully initialize the queue */ |
eb8db831 CH |
688 | md->queue->cmd_size = sizeof(struct dm_rq_target_io); |
689 | md->queue->rq_alloc_data = md; | |
5ea708d1 | 690 | md->queue->request_fn = dm_old_request_fn; |
eb8db831 CH |
691 | md->queue->init_rq_fn = dm_rq_init_rq; |
692 | ||
693 | immutable_tgt = dm_table_get_immutable_target(t); | |
694 | if (immutable_tgt && immutable_tgt->per_io_data_size) { | |
695 | /* any target-specific per-io data is immediately after the tio */ | |
696 | md->queue->cmd_size += immutable_tgt->per_io_data_size; | |
697 | md->init_tio_pdu = true; | |
698 | } | |
5ea708d1 | 699 | if (blk_init_allocated_queue(md->queue) < 0) |
4cc96131 MS |
700 | return -EINVAL; |
701 | ||
702 | /* disable dm_old_request_fn's merge heuristic by default */ | |
703 | md->seq_rq_merge_deadline_usecs = 0; | |
704 | ||
705 | dm_init_normal_md_queue(md); | |
706 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
4cc96131 MS |
707 | |
708 | /* Initialize the request-based DM worker thread */ | |
3989144f | 709 | kthread_init_worker(&md->kworker); |
4cc96131 MS |
710 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, |
711 | "kdmwork-%s", dm_device_name(md)); | |
937fa62e MS |
712 | if (IS_ERR(md->kworker_task)) { |
713 | int error = PTR_ERR(md->kworker_task); | |
714 | md->kworker_task = NULL; | |
715 | return error; | |
716 | } | |
4cc96131 MS |
717 | |
718 | elv_register_queue(md->queue); | |
719 | ||
720 | return 0; | |
721 | } | |
722 | ||
d6296d39 CH |
723 | static int dm_mq_init_request(struct blk_mq_tag_set *set, struct request *rq, |
724 | unsigned int hctx_idx, unsigned int numa_node) | |
4cc96131 | 725 | { |
d6296d39 | 726 | return __dm_rq_init_rq(set->driver_data, rq); |
4cc96131 MS |
727 | } |
728 | ||
729 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
730 | const struct blk_mq_queue_data *bd) | |
731 | { | |
732 | struct request *rq = bd->rq; | |
733 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
734 | struct mapped_device *md = tio->md; | |
735 | struct dm_target *ti = md->immutable_target; | |
736 | ||
737 | if (unlikely(!ti)) { | |
738 | int srcu_idx; | |
739 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
740 | ||
741 | ti = dm_table_find_target(map, 0); | |
742 | dm_put_live_table(md, srcu_idx); | |
743 | } | |
744 | ||
745 | if (ti->type->busy && ti->type->busy(ti)) | |
746 | return BLK_MQ_RQ_QUEUE_BUSY; | |
747 | ||
748 | dm_start_request(md, rq); | |
749 | ||
750 | /* Init tio using md established in .init_request */ | |
751 | init_tio(tio, rq, md); | |
752 | ||
753 | /* | |
754 | * Establish tio->ti before calling map_request(). | |
755 | */ | |
756 | tio->ti = ti; | |
757 | ||
758 | /* Direct call is fine since .queue_rq allows allocations */ | |
fbc39b4c | 759 | if (map_request(tio) == DM_MAPIO_REQUEUE) { |
4cc96131 MS |
760 | /* Undo dm_start_request() before requeuing */ |
761 | rq_end_stats(md, rq); | |
762 | rq_completed(md, rq_data_dir(rq), false); | |
6077c2d7 | 763 | blk_mq_delay_run_hw_queue(hctx, 100/*ms*/); |
4cc96131 MS |
764 | return BLK_MQ_RQ_QUEUE_BUSY; |
765 | } | |
766 | ||
767 | return BLK_MQ_RQ_QUEUE_OK; | |
768 | } | |
769 | ||
f363b089 | 770 | static const struct blk_mq_ops dm_mq_ops = { |
4cc96131 | 771 | .queue_rq = dm_mq_queue_rq, |
4cc96131 MS |
772 | .complete = dm_softirq_done, |
773 | .init_request = dm_mq_init_request, | |
774 | }; | |
775 | ||
e83068a5 | 776 | int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t) |
4cc96131 MS |
777 | { |
778 | struct request_queue *q; | |
e83068a5 | 779 | struct dm_target *immutable_tgt; |
4cc96131 MS |
780 | int err; |
781 | ||
e83068a5 | 782 | if (!dm_table_all_blk_mq_devices(t)) { |
4cc96131 MS |
783 | DMERR("request-based dm-mq may only be stacked on blk-mq device(s)"); |
784 | return -EINVAL; | |
785 | } | |
786 | ||
787 | md->tag_set = kzalloc_node(sizeof(struct blk_mq_tag_set), GFP_KERNEL, md->numa_node_id); | |
788 | if (!md->tag_set) | |
789 | return -ENOMEM; | |
790 | ||
791 | md->tag_set->ops = &dm_mq_ops; | |
792 | md->tag_set->queue_depth = dm_get_blk_mq_queue_depth(); | |
793 | md->tag_set->numa_node = md->numa_node_id; | |
794 | md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
795 | md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues(); | |
796 | md->tag_set->driver_data = md; | |
797 | ||
798 | md->tag_set->cmd_size = sizeof(struct dm_rq_target_io); | |
e83068a5 | 799 | immutable_tgt = dm_table_get_immutable_target(t); |
4cc96131 MS |
800 | if (immutable_tgt && immutable_tgt->per_io_data_size) { |
801 | /* any target-specific per-io data is immediately after the tio */ | |
802 | md->tag_set->cmd_size += immutable_tgt->per_io_data_size; | |
803 | md->init_tio_pdu = true; | |
804 | } | |
805 | ||
806 | err = blk_mq_alloc_tag_set(md->tag_set); | |
807 | if (err) | |
808 | goto out_kfree_tag_set; | |
809 | ||
810 | q = blk_mq_init_allocated_queue(md->tag_set, md->queue); | |
811 | if (IS_ERR(q)) { | |
812 | err = PTR_ERR(q); | |
813 | goto out_tag_set; | |
814 | } | |
815 | dm_init_md_queue(md); | |
816 | ||
817 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
23a60124 BVA |
818 | err = blk_mq_register_dev(disk_to_dev(md->disk), q); |
819 | if (err) | |
820 | goto out_cleanup_queue; | |
4cc96131 MS |
821 | |
822 | return 0; | |
823 | ||
23a60124 BVA |
824 | out_cleanup_queue: |
825 | blk_cleanup_queue(q); | |
4cc96131 MS |
826 | out_tag_set: |
827 | blk_mq_free_tag_set(md->tag_set); | |
828 | out_kfree_tag_set: | |
829 | kfree(md->tag_set); | |
830 | ||
831 | return err; | |
832 | } | |
833 | ||
834 | void dm_mq_cleanup_mapped_device(struct mapped_device *md) | |
835 | { | |
836 | if (md->tag_set) { | |
837 | blk_mq_free_tag_set(md->tag_set); | |
838 | kfree(md->tag_set); | |
839 | } | |
840 | } | |
841 | ||
842 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); | |
843 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
844 | ||
845 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); | |
846 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
847 | ||
848 | module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR); | |
849 | MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices"); | |
850 | ||
851 | module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR); | |
852 | MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices"); |