blk-mq: Stop using pointers for blk_mq_tags bitmap tags
[linux-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
cf43e6be 5#include "blk-stat.h"
244c65a3 6#include "blk-mq-tag.h"
cf43e6be 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
JA
27
28 /* incremented at dispatch time */
29 unsigned long rq_dispatched[2];
30 unsigned long rq_merged;
31
32 /* incremented at completion time */
33 unsigned long ____cacheline_aligned_in_smp rq_completed[2];
34
35 struct request_queue *queue;
1db4909e 36 struct blk_mq_ctxs *ctxs;
320ae51f 37 struct kobject kobj;
4bb659b1 38} ____cacheline_aligned_in_smp;
320ae51f 39
c7e2d94b 40void blk_mq_exit_queue(struct request_queue *q);
e3a2b3f9 41int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr);
aed3ea94 42void blk_mq_wake_waiters(struct request_queue *q);
1fd40b5e
ML
43bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *,
44 unsigned int);
e6c98712
BVA
45void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
46 bool kick_requeue_list);
2c3ad667 47void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list);
b347689f
ML
48struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
49 struct blk_mq_ctx *start);
2e315dc0 50void blk_mq_put_rq_ref(struct request *rq);
2c3ad667
JA
51
52/*
53 * Internal helpers for allocating/freeing the request map
54 */
cc71a6f4
JA
55void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
56 unsigned int hctx_idx);
e155b0c2 57void blk_mq_free_rq_map(struct blk_mq_tags *tags);
63064be1
JG
58struct blk_mq_tags *blk_mq_alloc_map_and_rqs(struct blk_mq_tag_set *set,
59 unsigned int hctx_idx, unsigned int depth);
645db34e
JG
60void blk_mq_free_map_and_rqs(struct blk_mq_tag_set *set,
61 struct blk_mq_tags *tags,
62 unsigned int hctx_idx);
2c3ad667
JA
63/*
64 * Internal helpers for request insertion into sw queues
65 */
66void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
67 bool at_head);
01e99aec
ML
68void blk_mq_request_bypass_insert(struct request *rq, bool at_head,
69 bool run_queue);
bd166ef1
JA
70void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx,
71 struct list_head *list);
320ae51f 72
fd9c40f6
BVA
73/* Used by blk_insert_cloned_request() to issue request directly */
74blk_status_t blk_mq_request_issue_directly(struct request *rq, bool last);
6ce3dd6e
ML
75void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
76 struct list_head *list);
396eaf21 77
320ae51f
JA
78/*
79 * CPU -> queue mappings
80 */
ed76e329 81extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
320ae51f 82
b3c661b1
JA
83/*
84 * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
85 * @q: request queue
e20ba6e1 86 * @type: the hctx type index
b3c661b1
JA
87 * @cpu: CPU
88 */
89static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
e20ba6e1 90 enum hctx_type type,
b3c661b1 91 unsigned int cpu)
7d7e0f90 92{
e20ba6e1 93 return q->queue_hw_ctx[q->tag_set->map[type].mq_map[cpu]];
7d7e0f90
CH
94}
95
b3c661b1
JA
96/*
97 * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
98 * @q: request queue
99 * @flags: request command flags
d220a214 100 * @ctx: software queue cpu ctx
b3c661b1
JA
101 */
102static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
103 unsigned int flags,
8ccdf4a3 104 struct blk_mq_ctx *ctx)
ff2c5660 105{
e20ba6e1
CH
106 enum hctx_type type = HCTX_TYPE_DEFAULT;
107
bb94aea1
JW
108 /*
109 * The caller ensure that if REQ_HIPRI, poll must be enabled.
110 */
111 if (flags & REQ_HIPRI)
e20ba6e1 112 type = HCTX_TYPE_POLL;
bb94aea1 113 else if ((flags & REQ_OP_MASK) == REQ_OP_READ)
e20ba6e1 114 type = HCTX_TYPE_READ;
5aceaeb2 115
8ccdf4a3 116 return ctx->hctxs[type];
ff2c5660
JA
117}
118
67aec14c
JA
119/*
120 * sysfs helpers
121 */
737f98cf 122extern void blk_mq_sysfs_init(struct request_queue *q);
7ea5fe31 123extern void blk_mq_sysfs_deinit(struct request_queue *q);
2d0364c8 124extern int __blk_mq_register_dev(struct device *dev, struct request_queue *q);
67aec14c
JA
125extern int blk_mq_sysfs_register(struct request_queue *q);
126extern void blk_mq_sysfs_unregister(struct request_queue *q);
868f2f0b 127extern void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx);
67aec14c 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;
f9afca4d 153 unsigned int cmd_flags;
cb96a42c
ML
154
155 /* input & output parameter */
156 struct blk_mq_ctx *ctx;
157 struct blk_mq_hw_ctx *hctx;
158};
159
32bc15af
JG
160static inline bool blk_mq_is_sbitmap_shared(unsigned int flags)
161{
162 return flags & BLK_MQ_F_TAG_HCTX_SHARED;
163}
164
4941115b
JA
165static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data)
166{
42fdc5e4 167 if (data->q->elevator)
bd166ef1
JA
168 return data->hctx->sched_tags;
169
4941115b
JA
170 return data->hctx->tags;
171}
172
5d1b25c1
BVA
173static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx)
174{
175 return test_bit(BLK_MQ_S_STOPPED, &hctx->state);
176}
177
19c66e59
ML
178static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx)
179{
180 return hctx->nr_ctx && hctx->tags;
181}
182
8446fe92
CH
183unsigned int blk_mq_in_flight(struct request_queue *q,
184 struct block_device *part);
185void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
186 unsigned int inflight[2]);
f299b7c7 187
2a5a24aa
ML
188static inline void blk_mq_put_dispatch_budget(struct request_queue *q,
189 int budget_token)
de148297 190{
de148297 191 if (q->mq_ops->put_budget)
2a5a24aa 192 q->mq_ops->put_budget(q, budget_token);
de148297
ML
193}
194
2a5a24aa 195static inline int blk_mq_get_dispatch_budget(struct request_queue *q)
de148297 196{
de148297 197 if (q->mq_ops->get_budget)
65c76369 198 return q->mq_ops->get_budget(q);
2a5a24aa
ML
199 return 0;
200}
201
202static inline void blk_mq_set_rq_budget_token(struct request *rq, int token)
203{
204 if (token < 0)
205 return;
206
207 if (rq->q->mq_ops->set_rq_budget_token)
208 rq->q->mq_ops->set_rq_budget_token(rq, token);
209}
210
211static inline int blk_mq_get_rq_budget_token(struct request *rq)
212{
213 if (rq->q->mq_ops->get_rq_budget_token)
214 return rq->q->mq_ops->get_rq_budget_token(rq);
215 return -1;
de148297
ML
216}
217
bccf5e26
JG
218static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
219{
220 if (blk_mq_is_sbitmap_shared(hctx->flags))
221 atomic_inc(&hctx->queue->nr_active_requests_shared_sbitmap);
222 else
223 atomic_inc(&hctx->nr_active);
224}
225
226static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
227{
228 if (blk_mq_is_sbitmap_shared(hctx->flags))
229 atomic_dec(&hctx->queue->nr_active_requests_shared_sbitmap);
230 else
231 atomic_dec(&hctx->nr_active);
232}
233
234static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)
235{
236 if (blk_mq_is_sbitmap_shared(hctx->flags))
237 return atomic_read(&hctx->queue->nr_active_requests_shared_sbitmap);
238 return atomic_read(&hctx->nr_active);
239}
4e2f62e5
JA
240static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx,
241 struct request *rq)
242{
243 blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag);
244 rq->tag = BLK_MQ_NO_TAG;
245
246 if (rq->rq_flags & RQF_MQ_INFLIGHT) {
247 rq->rq_flags &= ~RQF_MQ_INFLIGHT;
bccf5e26 248 __blk_mq_dec_active_requests(hctx);
4e2f62e5
JA
249 }
250}
251
252static inline void blk_mq_put_driver_tag(struct request *rq)
253{
254 if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG)
255 return;
256
257 __blk_mq_put_driver_tag(rq->mq_hctx, rq);
258}
259
61347154
JK
260bool blk_mq_get_driver_tag(struct request *rq);
261
ed76e329 262static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
0da73d00
MI
263{
264 int cpu;
265
266 for_each_possible_cpu(cpu)
ed76e329 267 qmap->mq_map[cpu] = 0;
0da73d00
MI
268}
269
b49773e7
DLM
270/*
271 * blk_mq_plug() - Get caller context plug
272 * @q: request queue
273 * @bio : the bio being submitted by the caller context
274 *
275 * Plugging, by design, may delay the insertion of BIOs into the elevator in
276 * order to increase BIO merging opportunities. This however can cause BIO
277 * insertion order to change from the order in which submit_bio() is being
278 * executed in the case of multiple contexts concurrently issuing BIOs to a
279 * device, even if these context are synchronized to tightly control BIO issuing
280 * order. While this is not a problem with regular block devices, this ordering
281 * change can cause write BIO failures with zoned block devices as these
282 * require sequential write patterns to zones. Prevent this from happening by
283 * ignoring the plug state of a BIO issuing context if the target request queue
284 * is for a zoned block device and the BIO to plug is a write operation.
285 *
286 * Return current->plug if the bio can be plugged and NULL otherwise
287 */
288static inline struct blk_plug *blk_mq_plug(struct request_queue *q,
289 struct bio *bio)
290{
291 /*
292 * For regular block devices or read operations, use the context plug
293 * which may be NULL if blk_start_plug() was not executed.
294 */
295 if (!blk_queue_is_zoned(q) || !op_is_write(bio_op(bio)))
296 return current->plug;
297
298 /* Zoned block device write operation case: do not plug the BIO */
299 return NULL;
300}
301
fd2ef39c
JK
302/* Free all requests on the list */
303static inline void blk_mq_free_requests(struct list_head *list)
304{
305 while (!list_empty(list)) {
306 struct request *rq = list_entry_rq(list->next);
307
308 list_del_init(&rq->queuelist);
309 blk_mq_free_request(rq);
310 }
311}
312
a0235d23
JG
313/*
314 * For shared tag users, we track the number of currently active users
315 * and attempt to provide a fair share of the tag depth for each of them.
316 */
317static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
318 struct sbitmap_queue *bt)
319{
320 unsigned int depth, users;
321
322 if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED))
323 return true;
a0235d23
JG
324
325 /*
326 * Don't try dividing an ant
327 */
328 if (bt->sb.depth == 1)
329 return true;
330
f1b49fdc
JG
331 if (blk_mq_is_sbitmap_shared(hctx->flags)) {
332 struct request_queue *q = hctx->queue;
f1b49fdc 333
2569063c 334 if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags))
f1b49fdc 335 return true;
f1b49fdc
JG
336 } else {
337 if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
338 return true;
f1b49fdc
JG
339 }
340
e155b0c2
JG
341 users = atomic_read(&hctx->tags->active_queues);
342
a0235d23
JG
343 if (!users)
344 return true;
345
346 /*
347 * Allow at least some tags
348 */
349 depth = max((bt->sb.depth + users - 1) / users, 4U);
bccf5e26 350 return __blk_mq_active_requests(hctx) < depth;
a0235d23
JG
351}
352
353
320ae51f 354#endif