Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
320ae51f JA |
2 | #ifndef INT_BLK_MQ_H |
3 | #define INT_BLK_MQ_H | |
4 | ||
cf43e6be | 5 | #include "blk-stat.h" |
244c65a3 | 6 | #include "blk-mq-tag.h" |
cf43e6be | 7 | |
24d2f903 CH |
8 | struct blk_mq_tag_set; |
9 | ||
1db4909e ML |
10 | struct blk_mq_ctxs { |
11 | struct kobject kobj; | |
12 | struct blk_mq_ctx __percpu *queue_ctx; | |
13 | }; | |
14 | ||
fe644072 LW |
15 | /** |
16 | * struct blk_mq_ctx - State for a software queue facing the submitting CPUs | |
17 | */ | |
320ae51f JA |
18 | struct blk_mq_ctx { |
19 | struct { | |
20 | spinlock_t lock; | |
c16d6b5a ML |
21 | struct list_head rq_lists[HCTX_MAX_TYPES]; |
22 | } ____cacheline_aligned_in_smp; | |
320ae51f JA |
23 | |
24 | unsigned int cpu; | |
f31967f0 | 25 | unsigned short index_hw[HCTX_MAX_TYPES]; |
8ccdf4a3 | 26 | struct blk_mq_hw_ctx *hctxs[HCTX_MAX_TYPES]; |
320ae51f | 27 | |
320ae51f | 28 | struct request_queue *queue; |
1db4909e | 29 | struct blk_mq_ctxs *ctxs; |
320ae51f | 30 | struct kobject kobj; |
4bb659b1 | 31 | } ____cacheline_aligned_in_smp; |
320ae51f | 32 | |
3e08773c | 33 | void blk_mq_submit_bio(struct bio *bio); |
5a72e899 JA |
34 | int blk_mq_poll(struct request_queue *q, blk_qc_t cookie, struct io_comp_batch *iob, |
35 | unsigned int flags); | |
c7e2d94b | 36 | void blk_mq_exit_queue(struct request_queue *q); |
e3a2b3f9 | 37 | int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr); |
aed3ea94 | 38 | void blk_mq_wake_waiters(struct request_queue *q); |
1fd40b5e ML |
39 | bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *, |
40 | unsigned int); | |
e6c98712 BVA |
41 | void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, |
42 | bool kick_requeue_list); | |
2c3ad667 | 43 | void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list); |
b347689f ML |
44 | struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx, |
45 | struct blk_mq_ctx *start); | |
2e315dc0 | 46 | void blk_mq_put_rq_ref(struct request *rq); |
2c3ad667 JA |
47 | |
48 | /* | |
49 | * Internal helpers for allocating/freeing the request map | |
50 | */ | |
cc71a6f4 JA |
51 | void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, |
52 | unsigned int hctx_idx); | |
e155b0c2 | 53 | void blk_mq_free_rq_map(struct blk_mq_tags *tags); |
63064be1 JG |
54 | struct blk_mq_tags *blk_mq_alloc_map_and_rqs(struct blk_mq_tag_set *set, |
55 | unsigned int hctx_idx, unsigned int depth); | |
645db34e JG |
56 | void blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set, |
57 | struct blk_mq_tags *tags, | |
58 | unsigned int hctx_idx); | |
2c3ad667 JA |
59 | /* |
60 | * Internal helpers for request insertion into sw queues | |
61 | */ | |
62 | void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq, | |
63 | bool at_head); | |
01e99aec ML |
64 | void blk_mq_request_bypass_insert(struct request *rq, bool at_head, |
65 | bool run_queue); | |
bd166ef1 JA |
66 | void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx, |
67 | struct list_head *list); | |
6ce3dd6e ML |
68 | void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx, |
69 | struct list_head *list); | |
396eaf21 | 70 | |
320ae51f JA |
71 | /* |
72 | * CPU -> queue mappings | |
73 | */ | |
ed76e329 | 74 | extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int); |
320ae51f | 75 | |
b3c661b1 JA |
76 | /* |
77 | * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue | |
78 | * @q: request queue | |
e20ba6e1 | 79 | * @type: the hctx type index |
b3c661b1 JA |
80 | * @cpu: CPU |
81 | */ | |
82 | static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q, | |
e20ba6e1 | 83 | enum hctx_type type, |
b3c661b1 | 84 | unsigned int cpu) |
7d7e0f90 | 85 | { |
4e5cc99e | 86 | return xa_load(&q->hctx_table, q->tag_set->map[type].mq_map[cpu]); |
7d7e0f90 CH |
87 | } |
88 | ||
16458cf3 | 89 | static inline enum hctx_type blk_mq_get_hctx_type(blk_opf_t opf) |
ff2c5660 | 90 | { |
e20ba6e1 CH |
91 | enum hctx_type type = HCTX_TYPE_DEFAULT; |
92 | ||
bb94aea1 | 93 | /* |
6ce913fe | 94 | * The caller ensure that if REQ_POLLED, poll must be enabled. |
bb94aea1 | 95 | */ |
7e923f40 | 96 | if (opf & REQ_POLLED) |
e20ba6e1 | 97 | type = HCTX_TYPE_POLL; |
7e923f40 | 98 | else if ((opf & REQ_OP_MASK) == REQ_OP_READ) |
e20ba6e1 | 99 | type = HCTX_TYPE_READ; |
b637108a ML |
100 | return type; |
101 | } | |
102 | ||
103 | /* | |
104 | * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue | |
105 | * @q: request queue | |
7e923f40 | 106 | * @opf: operation type (REQ_OP_*) and flags (e.g. REQ_POLLED). |
b637108a ML |
107 | * @ctx: software queue cpu ctx |
108 | */ | |
109 | static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q, | |
16458cf3 | 110 | blk_opf_t opf, |
b637108a ML |
111 | struct blk_mq_ctx *ctx) |
112 | { | |
7e923f40 | 113 | return ctx->hctxs[blk_mq_get_hctx_type(opf)]; |
ff2c5660 JA |
114 | } |
115 | ||
67aec14c JA |
116 | /* |
117 | * sysfs helpers | |
118 | */ | |
737f98cf | 119 | extern void blk_mq_sysfs_init(struct request_queue *q); |
7ea5fe31 | 120 | extern void blk_mq_sysfs_deinit(struct request_queue *q); |
8682b92e CH |
121 | int blk_mq_sysfs_register(struct gendisk *disk); |
122 | void blk_mq_sysfs_unregister(struct gendisk *disk); | |
eaa870f9 CH |
123 | int blk_mq_sysfs_register_hctxs(struct request_queue *q); |
124 | void blk_mq_sysfs_unregister_hctxs(struct request_queue *q); | |
868f2f0b | 125 | extern void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx); |
47c122e3 | 126 | void blk_mq_free_plug_rqs(struct blk_plug *plug); |
dbb6f764 | 127 | void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); |
67aec14c | 128 | |
2a19b28f ML |
129 | void blk_mq_cancel_work_sync(struct request_queue *q); |
130 | ||
e09aae7e ML |
131 | void blk_mq_release(struct request_queue *q); |
132 | ||
1aecfe48 ML |
133 | static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q, |
134 | unsigned int cpu) | |
135 | { | |
136 | return per_cpu_ptr(q->queue_ctx, cpu); | |
137 | } | |
138 | ||
139 | /* | |
140 | * This assumes per-cpu software queueing queues. They could be per-node | |
141 | * as well, for instance. For now this is hardcoded as-is. Note that we don't | |
142 | * care about preemption, since we know the ctx's are persistent. This does | |
143 | * mean that we can't rely on ctx always matching the currently running CPU. | |
144 | */ | |
145 | static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q) | |
146 | { | |
c05f4220 | 147 | return __blk_mq_get_ctx(q, raw_smp_processor_id()); |
1aecfe48 ML |
148 | } |
149 | ||
cb96a42c ML |
150 | struct blk_mq_alloc_data { |
151 | /* input parameter */ | |
152 | struct request_queue *q; | |
9a95e4ef | 153 | blk_mq_req_flags_t flags; |
229a9287 | 154 | unsigned int shallow_depth; |
16458cf3 | 155 | blk_opf_t cmd_flags; |
ecaf97f4 | 156 | req_flags_t rq_flags; |
cb96a42c | 157 | |
47c122e3 JA |
158 | /* allocate multiple requests/tags in one go */ |
159 | unsigned int nr_tags; | |
160 | struct request **cached_rq; | |
161 | ||
cb96a42c ML |
162 | /* input & output parameter */ |
163 | struct blk_mq_ctx *ctx; | |
164 | struct blk_mq_hw_ctx *hctx; | |
165 | }; | |
166 | ||
079a2e3e | 167 | static inline bool blk_mq_is_shared_tags(unsigned int flags) |
32bc15af JG |
168 | { |
169 | return flags & BLK_MQ_F_TAG_HCTX_SHARED; | |
170 | } | |
171 | ||
4941115b JA |
172 | static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data) |
173 | { | |
56f8da64 JA |
174 | if (!(data->rq_flags & RQF_ELV)) |
175 | return data->hctx->tags; | |
176 | return data->hctx->sched_tags; | |
4941115b JA |
177 | } |
178 | ||
5d1b25c1 BVA |
179 | static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx) |
180 | { | |
181 | return test_bit(BLK_MQ_S_STOPPED, &hctx->state); | |
182 | } | |
183 | ||
19c66e59 ML |
184 | static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx) |
185 | { | |
186 | return hctx->nr_ctx && hctx->tags; | |
187 | } | |
188 | ||
8446fe92 CH |
189 | unsigned int blk_mq_in_flight(struct request_queue *q, |
190 | struct block_device *part); | |
191 | void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part, | |
192 | unsigned int inflight[2]); | |
f299b7c7 | 193 | |
2a5a24aa ML |
194 | static inline void blk_mq_put_dispatch_budget(struct request_queue *q, |
195 | int budget_token) | |
de148297 | 196 | { |
de148297 | 197 | if (q->mq_ops->put_budget) |
2a5a24aa | 198 | q->mq_ops->put_budget(q, budget_token); |
de148297 ML |
199 | } |
200 | ||
2a5a24aa | 201 | static inline int blk_mq_get_dispatch_budget(struct request_queue *q) |
de148297 | 202 | { |
de148297 | 203 | if (q->mq_ops->get_budget) |
65c76369 | 204 | return q->mq_ops->get_budget(q); |
2a5a24aa ML |
205 | return 0; |
206 | } | |
207 | ||
208 | static inline void blk_mq_set_rq_budget_token(struct request *rq, int token) | |
209 | { | |
210 | if (token < 0) | |
211 | return; | |
212 | ||
213 | if (rq->q->mq_ops->set_rq_budget_token) | |
214 | rq->q->mq_ops->set_rq_budget_token(rq, token); | |
215 | } | |
216 | ||
217 | static inline int blk_mq_get_rq_budget_token(struct request *rq) | |
218 | { | |
219 | if (rq->q->mq_ops->get_rq_budget_token) | |
220 | return rq->q->mq_ops->get_rq_budget_token(rq); | |
221 | return -1; | |
de148297 ML |
222 | } |
223 | ||
bccf5e26 JG |
224 | static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx) |
225 | { | |
079a2e3e JG |
226 | if (blk_mq_is_shared_tags(hctx->flags)) |
227 | atomic_inc(&hctx->queue->nr_active_requests_shared_tags); | |
bccf5e26 JG |
228 | else |
229 | atomic_inc(&hctx->nr_active); | |
230 | } | |
231 | ||
3b87c6ea ML |
232 | static inline void __blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx, |
233 | int val) | |
bccf5e26 | 234 | { |
079a2e3e | 235 | if (blk_mq_is_shared_tags(hctx->flags)) |
3b87c6ea | 236 | atomic_sub(val, &hctx->queue->nr_active_requests_shared_tags); |
bccf5e26 | 237 | else |
3b87c6ea ML |
238 | atomic_sub(val, &hctx->nr_active); |
239 | } | |
240 | ||
241 | static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx) | |
242 | { | |
243 | __blk_mq_sub_active_requests(hctx, 1); | |
bccf5e26 JG |
244 | } |
245 | ||
246 | static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx) | |
247 | { | |
079a2e3e JG |
248 | if (blk_mq_is_shared_tags(hctx->flags)) |
249 | return atomic_read(&hctx->queue->nr_active_requests_shared_tags); | |
bccf5e26 JG |
250 | return atomic_read(&hctx->nr_active); |
251 | } | |
4e2f62e5 JA |
252 | static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx, |
253 | struct request *rq) | |
254 | { | |
255 | blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag); | |
256 | rq->tag = BLK_MQ_NO_TAG; | |
257 | ||
258 | if (rq->rq_flags & RQF_MQ_INFLIGHT) { | |
259 | rq->rq_flags &= ~RQF_MQ_INFLIGHT; | |
bccf5e26 | 260 | __blk_mq_dec_active_requests(hctx); |
4e2f62e5 JA |
261 | } |
262 | } | |
263 | ||
264 | static inline void blk_mq_put_driver_tag(struct request *rq) | |
265 | { | |
266 | if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG) | |
267 | return; | |
268 | ||
269 | __blk_mq_put_driver_tag(rq->mq_hctx, rq); | |
270 | } | |
271 | ||
a808a9d5 JA |
272 | bool __blk_mq_get_driver_tag(struct blk_mq_hw_ctx *hctx, struct request *rq); |
273 | ||
274 | static inline bool blk_mq_get_driver_tag(struct request *rq) | |
275 | { | |
276 | struct blk_mq_hw_ctx *hctx = rq->mq_hctx; | |
277 | ||
278 | if (rq->tag != BLK_MQ_NO_TAG && | |
279 | !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) { | |
280 | hctx->tags->rqs[rq->tag] = rq; | |
281 | return true; | |
282 | } | |
283 | ||
284 | return __blk_mq_get_driver_tag(hctx, rq); | |
285 | } | |
61347154 | 286 | |
ed76e329 | 287 | static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap) |
0da73d00 MI |
288 | { |
289 | int cpu; | |
290 | ||
291 | for_each_possible_cpu(cpu) | |
ed76e329 | 292 | qmap->mq_map[cpu] = 0; |
0da73d00 MI |
293 | } |
294 | ||
b49773e7 DLM |
295 | /* |
296 | * blk_mq_plug() - Get caller context plug | |
b49773e7 DLM |
297 | * @bio : the bio being submitted by the caller context |
298 | * | |
299 | * Plugging, by design, may delay the insertion of BIOs into the elevator in | |
300 | * order to increase BIO merging opportunities. This however can cause BIO | |
301 | * insertion order to change from the order in which submit_bio() is being | |
302 | * executed in the case of multiple contexts concurrently issuing BIOs to a | |
303 | * device, even if these context are synchronized to tightly control BIO issuing | |
304 | * order. While this is not a problem with regular block devices, this ordering | |
305 | * change can cause write BIO failures with zoned block devices as these | |
306 | * require sequential write patterns to zones. Prevent this from happening by | |
6deacb3b CH |
307 | * ignoring the plug state of a BIO issuing context if it is for a zoned block |
308 | * device and the BIO to plug is a write operation. | |
b49773e7 DLM |
309 | * |
310 | * Return current->plug if the bio can be plugged and NULL otherwise | |
311 | */ | |
6deacb3b | 312 | static inline struct blk_plug *blk_mq_plug( struct bio *bio) |
b49773e7 | 313 | { |
6deacb3b | 314 | /* Zoned block device write operation case: do not plug the BIO */ |
8cafdb5a PR |
315 | if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) && |
316 | bdev_op_is_zoned_write(bio->bi_bdev, bio_op(bio))) | |
6deacb3b CH |
317 | return NULL; |
318 | ||
b49773e7 DLM |
319 | /* |
320 | * For regular block devices or read operations, use the context plug | |
321 | * which may be NULL if blk_start_plug() was not executed. | |
322 | */ | |
6deacb3b | 323 | return current->plug; |
b49773e7 DLM |
324 | } |
325 | ||
fd2ef39c JK |
326 | /* Free all requests on the list */ |
327 | static inline void blk_mq_free_requests(struct list_head *list) | |
328 | { | |
329 | while (!list_empty(list)) { | |
330 | struct request *rq = list_entry_rq(list->next); | |
331 | ||
332 | list_del_init(&rq->queuelist); | |
333 | blk_mq_free_request(rq); | |
334 | } | |
335 | } | |
336 | ||
a0235d23 JG |
337 | /* |
338 | * For shared tag users, we track the number of currently active users | |
339 | * and attempt to provide a fair share of the tag depth for each of them. | |
340 | */ | |
341 | static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx, | |
342 | struct sbitmap_queue *bt) | |
343 | { | |
344 | unsigned int depth, users; | |
345 | ||
346 | if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) | |
347 | return true; | |
a0235d23 JG |
348 | |
349 | /* | |
350 | * Don't try dividing an ant | |
351 | */ | |
352 | if (bt->sb.depth == 1) | |
353 | return true; | |
354 | ||
079a2e3e | 355 | if (blk_mq_is_shared_tags(hctx->flags)) { |
f1b49fdc | 356 | struct request_queue *q = hctx->queue; |
f1b49fdc | 357 | |
2569063c | 358 | if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags)) |
f1b49fdc | 359 | return true; |
f1b49fdc JG |
360 | } else { |
361 | if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state)) | |
362 | return true; | |
f1b49fdc JG |
363 | } |
364 | ||
e155b0c2 JG |
365 | users = atomic_read(&hctx->tags->active_queues); |
366 | ||
a0235d23 JG |
367 | if (!users) |
368 | return true; | |
369 | ||
370 | /* | |
371 | * Allow at least some tags | |
372 | */ | |
373 | depth = max((bt->sb.depth + users - 1) / users, 4U); | |
bccf5e26 | 374 | return __blk_mq_active_requests(hctx) < depth; |
a0235d23 JG |
375 | } |
376 | ||
2a904d00 | 377 | /* run the code block in @dispatch_ops with rcu/srcu read lock held */ |
41adf531 | 378 | #define __blk_mq_run_dispatch_ops(q, check_sleep, dispatch_ops) \ |
2a904d00 | 379 | do { \ |
80bd4a7a | 380 | if ((q)->tag_set->flags & BLK_MQ_F_BLOCKING) { \ |
2a904d00 ML |
381 | int srcu_idx; \ |
382 | \ | |
41adf531 | 383 | might_sleep_if(check_sleep); \ |
80bd4a7a | 384 | srcu_idx = srcu_read_lock((q)->tag_set->srcu); \ |
2a904d00 | 385 | (dispatch_ops); \ |
80bd4a7a CH |
386 | srcu_read_unlock((q)->tag_set->srcu, srcu_idx); \ |
387 | } else { \ | |
388 | rcu_read_lock(); \ | |
389 | (dispatch_ops); \ | |
390 | rcu_read_unlock(); \ | |
2a904d00 ML |
391 | } \ |
392 | } while (0) | |
a0235d23 | 393 | |
41adf531 ML |
394 | #define blk_mq_run_dispatch_ops(q, dispatch_ops) \ |
395 | __blk_mq_run_dispatch_ops(q, true, dispatch_ops) \ | |
396 | ||
320ae51f | 397 | #endif |