Merge tag 'sched_ext-for-6.12' of git://git.kernel.org/pub/scm/linux/kernel/git/tj...
[linux-2.6-block.git] / block / blk-mq.h
CommitLineData
b2441318 1/* SPDX-License-Identifier: GPL-2.0 */
320ae51f
JA
2#ifndef INT_BLK_MQ_H
3#define INT_BLK_MQ_H
4
90110e04 5#include <linux/blk-mq.h>
cf43e6be
JA
6#include "blk-stat.h"
7
24d2f903
CH
8struct blk_mq_tag_set;
9
1db4909e
ML
10struct 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
18struct 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
bebe84eb
CH
33enum {
34 BLK_MQ_NO_TAG = -1U,
35 BLK_MQ_TAG_MIN = 1,
36 BLK_MQ_TAG_MAX = BLK_MQ_NO_TAG - 1,
37};
38
3dff6155
JG
39#define BLK_MQ_CPU_WORK_BATCH (8)
40
710fa378
CH
41typedef unsigned int __bitwise blk_insert_t;
42#define BLK_MQ_INSERT_AT_HEAD ((__force blk_insert_t)0x01)
43
3e08773c 44void blk_mq_submit_bio(struct bio *bio);
5a72e899
JA
45int blk_mq_poll(struct request_queue *q, blk_qc_t cookie, struct io_comp_batch *iob,
46 unsigned int flags);
c7e2d94b 47void blk_mq_exit_queue(struct request_queue *q);
e3a2b3f9 48int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr);
aed3ea94 49void blk_mq_wake_waiters(struct request_queue *q);
1fd40b5e
ML
50bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *,
51 unsigned int);
2c3ad667 52void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list);
b347689f
ML
53struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
54 struct blk_mq_ctx *start);
2e315dc0 55void blk_mq_put_rq_ref(struct request *rq);
2c3ad667
JA
56
57/*
58 * Internal helpers for allocating/freeing the request map
59 */
cc71a6f4
JA
60void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
61 unsigned int hctx_idx);
e155b0c2 62void blk_mq_free_rq_map(struct blk_mq_tags *tags);
63064be1
JG
63struct blk_mq_tags *blk_mq_alloc_map_and_rqs(struct blk_mq_tag_set *set,
64 unsigned int hctx_idx, unsigned int depth);
645db34e
JG
65void blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set,
66 struct blk_mq_tags *tags,
67 unsigned int hctx_idx);
396eaf21 68
320ae51f
JA
69/*
70 * CPU -> queue mappings
71 */
ed76e329 72extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
320ae51f 73
b3c661b1
JA
74/*
75 * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
76 * @q: request queue
e20ba6e1 77 * @type: the hctx type index
b3c661b1
JA
78 * @cpu: CPU
79 */
80static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
e20ba6e1 81 enum hctx_type type,
b3c661b1 82 unsigned int cpu)
7d7e0f90 83{
4e5cc99e 84 return xa_load(&q->hctx_table, q->tag_set->map[type].mq_map[cpu]);
7d7e0f90
CH
85}
86
16458cf3 87static inline enum hctx_type blk_mq_get_hctx_type(blk_opf_t opf)
ff2c5660 88{
e20ba6e1
CH
89 enum hctx_type type = HCTX_TYPE_DEFAULT;
90
bb94aea1 91 /*
6ce913fe 92 * The caller ensure that if REQ_POLLED, poll must be enabled.
bb94aea1 93 */
7e923f40 94 if (opf & REQ_POLLED)
e20ba6e1 95 type = HCTX_TYPE_POLL;
7e923f40 96 else if ((opf & REQ_OP_MASK) == REQ_OP_READ)
e20ba6e1 97 type = HCTX_TYPE_READ;
b637108a
ML
98 return type;
99}
100
101/*
102 * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
103 * @q: request queue
7e923f40 104 * @opf: operation type (REQ_OP_*) and flags (e.g. REQ_POLLED).
b637108a
ML
105 * @ctx: software queue cpu ctx
106 */
107static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
16458cf3 108 blk_opf_t opf,
b637108a
ML
109 struct blk_mq_ctx *ctx)
110{
7e923f40 111 return ctx->hctxs[blk_mq_get_hctx_type(opf)];
ff2c5660
JA
112}
113
67aec14c
JA
114/*
115 * sysfs helpers
116 */
737f98cf 117extern void blk_mq_sysfs_init(struct request_queue *q);
7ea5fe31 118extern void blk_mq_sysfs_deinit(struct request_queue *q);
8682b92e
CH
119int blk_mq_sysfs_register(struct gendisk *disk);
120void blk_mq_sysfs_unregister(struct gendisk *disk);
eaa870f9
CH
121int blk_mq_sysfs_register_hctxs(struct request_queue *q);
122void blk_mq_sysfs_unregister_hctxs(struct request_queue *q);
868f2f0b 123extern void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx);
47c122e3 124void blk_mq_free_plug_rqs(struct blk_plug *plug);
dbb6f764 125void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
67aec14c 126
2a19b28f
ML
127void blk_mq_cancel_work_sync(struct request_queue *q);
128
e09aae7e
ML
129void blk_mq_release(struct request_queue *q);
130
1aecfe48
ML
131static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
132 unsigned int cpu)
133{
134 return per_cpu_ptr(q->queue_ctx, cpu);
135}
136
137/*
138 * This assumes per-cpu software queueing queues. They could be per-node
139 * as well, for instance. For now this is hardcoded as-is. Note that we don't
140 * care about preemption, since we know the ctx's are persistent. This does
141 * mean that we can't rely on ctx always matching the currently running CPU.
142 */
143static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
144{
c05f4220 145 return __blk_mq_get_ctx(q, raw_smp_processor_id());
1aecfe48
ML
146}
147
cb96a42c
ML
148struct blk_mq_alloc_data {
149 /* input parameter */
150 struct request_queue *q;
9a95e4ef 151 blk_mq_req_flags_t flags;
229a9287 152 unsigned int shallow_depth;
16458cf3 153 blk_opf_t cmd_flags;
ecaf97f4 154 req_flags_t rq_flags;
cb96a42c 155
47c122e3
JA
156 /* allocate multiple requests/tags in one go */
157 unsigned int nr_tags;
158 struct request **cached_rq;
159
cb96a42c
ML
160 /* input & output parameter */
161 struct blk_mq_ctx *ctx;
162 struct blk_mq_hw_ctx *hctx;
163};
164
bebe84eb
CH
165struct blk_mq_tags *blk_mq_init_tags(unsigned int nr_tags,
166 unsigned int reserved_tags, int node, int alloc_policy);
167void blk_mq_free_tags(struct blk_mq_tags *tags);
168int blk_mq_init_bitmaps(struct sbitmap_queue *bitmap_tags,
169 struct sbitmap_queue *breserved_tags, unsigned int queue_depth,
170 unsigned int reserved, int node, int alloc_policy);
171
172unsigned int blk_mq_get_tag(struct blk_mq_alloc_data *data);
173unsigned long blk_mq_get_tags(struct blk_mq_alloc_data *data, int nr_tags,
174 unsigned int *offset);
175void blk_mq_put_tag(struct blk_mq_tags *tags, struct blk_mq_ctx *ctx,
176 unsigned int tag);
177void blk_mq_put_tags(struct blk_mq_tags *tags, int *tag_array, int nr_tags);
178int blk_mq_tag_update_depth(struct blk_mq_hw_ctx *hctx,
179 struct blk_mq_tags **tags, unsigned int depth, bool can_grow);
180void blk_mq_tag_resize_shared_tags(struct blk_mq_tag_set *set,
181 unsigned int size);
182void blk_mq_tag_update_sched_shared_tags(struct request_queue *q);
183
184void blk_mq_tag_wakeup_all(struct blk_mq_tags *tags, bool);
185void blk_mq_queue_tag_busy_iter(struct request_queue *q, busy_tag_iter_fn *fn,
186 void *priv);
187void blk_mq_all_tag_iter(struct blk_mq_tags *tags, busy_tag_iter_fn *fn,
188 void *priv);
189
190static inline struct sbq_wait_state *bt_wait_ptr(struct sbitmap_queue *bt,
191 struct blk_mq_hw_ctx *hctx)
192{
193 if (!hctx)
194 return &bt->ws[0];
195 return sbq_wait_ptr(bt, &hctx->wait_index);
196}
197
198void __blk_mq_tag_busy(struct blk_mq_hw_ctx *);
199void __blk_mq_tag_idle(struct blk_mq_hw_ctx *);
200
201static inline void blk_mq_tag_busy(struct blk_mq_hw_ctx *hctx)
202{
203 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
204 __blk_mq_tag_busy(hctx);
205}
206
207static inline void blk_mq_tag_idle(struct blk_mq_hw_ctx *hctx)
208{
209 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
210 __blk_mq_tag_idle(hctx);
211}
212
213static inline bool blk_mq_tag_is_reserved(struct blk_mq_tags *tags,
214 unsigned int tag)
215{
216 return tag < tags->nr_reserved_tags;
217}
218
079a2e3e 219static inline bool blk_mq_is_shared_tags(unsigned int flags)
32bc15af
JG
220{
221 return flags & BLK_MQ_F_TAG_HCTX_SHARED;
222}
223
4941115b
JA
224static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data)
225{
dd6216bb
CH
226 if (data->rq_flags & RQF_SCHED_TAGS)
227 return data->hctx->sched_tags;
228 return data->hctx->tags;
4941115b
JA
229}
230
5d1b25c1
BVA
231static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx)
232{
233 return test_bit(BLK_MQ_S_STOPPED, &hctx->state);
234}
235
19c66e59
ML
236static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx)
237{
238 return hctx->nr_ctx && hctx->tags;
239}
240
8446fe92
CH
241unsigned int blk_mq_in_flight(struct request_queue *q,
242 struct block_device *part);
243void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
244 unsigned int inflight[2]);
f299b7c7 245
2a5a24aa
ML
246static inline void blk_mq_put_dispatch_budget(struct request_queue *q,
247 int budget_token)
de148297 248{
de148297 249 if (q->mq_ops->put_budget)
2a5a24aa 250 q->mq_ops->put_budget(q, budget_token);
de148297
ML
251}
252
2a5a24aa 253static inline int blk_mq_get_dispatch_budget(struct request_queue *q)
de148297 254{
de148297 255 if (q->mq_ops->get_budget)
65c76369 256 return q->mq_ops->get_budget(q);
2a5a24aa
ML
257 return 0;
258}
259
260static inline void blk_mq_set_rq_budget_token(struct request *rq, int token)
261{
262 if (token < 0)
263 return;
264
265 if (rq->q->mq_ops->set_rq_budget_token)
266 rq->q->mq_ops->set_rq_budget_token(rq, token);
267}
268
269static inline int blk_mq_get_rq_budget_token(struct request *rq)
270{
271 if (rq->q->mq_ops->get_rq_budget_token)
272 return rq->q->mq_ops->get_rq_budget_token(rq);
273 return -1;
de148297
ML
274}
275
b8643d68
CZ
276static inline void __blk_mq_add_active_requests(struct blk_mq_hw_ctx *hctx,
277 int val)
bccf5e26 278{
079a2e3e 279 if (blk_mq_is_shared_tags(hctx->flags))
b8643d68 280 atomic_add(val, &hctx->queue->nr_active_requests_shared_tags);
bccf5e26 281 else
b8643d68
CZ
282 atomic_add(val, &hctx->nr_active);
283}
284
285static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
286{
287 __blk_mq_add_active_requests(hctx, 1);
bccf5e26
JG
288}
289
3b87c6ea
ML
290static inline void __blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx,
291 int val)
bccf5e26 292{
079a2e3e 293 if (blk_mq_is_shared_tags(hctx->flags))
3b87c6ea 294 atomic_sub(val, &hctx->queue->nr_active_requests_shared_tags);
bccf5e26 295 else
3b87c6ea
ML
296 atomic_sub(val, &hctx->nr_active);
297}
298
299static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
300{
301 __blk_mq_sub_active_requests(hctx, 1);
bccf5e26
JG
302}
303
b8643d68
CZ
304static inline void blk_mq_add_active_requests(struct blk_mq_hw_ctx *hctx,
305 int val)
306{
307 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
308 __blk_mq_add_active_requests(hctx, val);
309}
310
311static inline void blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
312{
313 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
314 __blk_mq_inc_active_requests(hctx);
315}
316
317static inline void blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx,
318 int val)
319{
320 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
321 __blk_mq_sub_active_requests(hctx, val);
322}
323
324static inline void blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
325{
326 if (hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)
327 __blk_mq_dec_active_requests(hctx);
328}
329
bccf5e26
JG
330static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)
331{
079a2e3e
JG
332 if (blk_mq_is_shared_tags(hctx->flags))
333 return atomic_read(&hctx->queue->nr_active_requests_shared_tags);
bccf5e26
JG
334 return atomic_read(&hctx->nr_active);
335}
4e2f62e5
JA
336static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx,
337 struct request *rq)
338{
b8643d68 339 blk_mq_dec_active_requests(hctx);
4e2f62e5
JA
340 blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag);
341 rq->tag = BLK_MQ_NO_TAG;
4e2f62e5
JA
342}
343
344static inline void blk_mq_put_driver_tag(struct request *rq)
345{
346 if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG)
347 return;
348
349 __blk_mq_put_driver_tag(rq->mq_hctx, rq);
350}
351
b8643d68 352bool __blk_mq_alloc_driver_tag(struct request *rq);
a808a9d5
JA
353
354static inline bool blk_mq_get_driver_tag(struct request *rq)
355{
b8643d68
CZ
356 if (rq->tag == BLK_MQ_NO_TAG && !__blk_mq_alloc_driver_tag(rq))
357 return false;
a808a9d5 358
b8643d68 359 return true;
a808a9d5 360}
61347154 361
ed76e329 362static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
0da73d00
MI
363{
364 int cpu;
365
366 for_each_possible_cpu(cpu)
ed76e329 367 qmap->mq_map[cpu] = 0;
0da73d00
MI
368}
369
fd2ef39c
JK
370/* Free all requests on the list */
371static inline void blk_mq_free_requests(struct list_head *list)
372{
373 while (!list_empty(list)) {
374 struct request *rq = list_entry_rq(list->next);
375
376 list_del_init(&rq->queuelist);
377 blk_mq_free_request(rq);
378 }
379}
380
a0235d23
JG
381/*
382 * For shared tag users, we track the number of currently active users
383 * and attempt to provide a fair share of the tag depth for each of them.
384 */
385static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
386 struct sbitmap_queue *bt)
387{
388 unsigned int depth, users;
389
390 if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED))
391 return true;
a0235d23
JG
392
393 /*
394 * Don't try dividing an ant
395 */
396 if (bt->sb.depth == 1)
397 return true;
398
079a2e3e 399 if (blk_mq_is_shared_tags(hctx->flags)) {
f1b49fdc 400 struct request_queue *q = hctx->queue;
f1b49fdc 401
2569063c 402 if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags))
f1b49fdc 403 return true;
f1b49fdc
JG
404 } else {
405 if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
406 return true;
f1b49fdc
JG
407 }
408
4f1731df 409 users = READ_ONCE(hctx->tags->active_queues);
a0235d23
JG
410 if (!users)
411 return true;
412
413 /*
414 * Allow at least some tags
415 */
416 depth = max((bt->sb.depth + users - 1) / users, 4U);
bccf5e26 417 return __blk_mq_active_requests(hctx) < depth;
a0235d23
JG
418}
419
2a904d00 420/* run the code block in @dispatch_ops with rcu/srcu read lock held */
41adf531 421#define __blk_mq_run_dispatch_ops(q, check_sleep, dispatch_ops) \
2a904d00 422do { \
80bd4a7a 423 if ((q)->tag_set->flags & BLK_MQ_F_BLOCKING) { \
00e885ef 424 struct blk_mq_tag_set *__tag_set = (q)->tag_set; \
2a904d00
ML
425 int srcu_idx; \
426 \
41adf531 427 might_sleep_if(check_sleep); \
00e885ef 428 srcu_idx = srcu_read_lock(__tag_set->srcu); \
2a904d00 429 (dispatch_ops); \
00e885ef 430 srcu_read_unlock(__tag_set->srcu, srcu_idx); \
80bd4a7a
CH
431 } else { \
432 rcu_read_lock(); \
433 (dispatch_ops); \
434 rcu_read_unlock(); \
2a904d00
ML
435 } \
436} while (0)
a0235d23 437
41adf531
ML
438#define blk_mq_run_dispatch_ops(q, dispatch_ops) \
439 __blk_mq_run_dispatch_ops(q, true, dispatch_ops) \
440
320ae51f 441#endif