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 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 | 40 | void blk_mq_exit_queue(struct request_queue *q); |
e3a2b3f9 | 41 | int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr); |
aed3ea94 | 42 | void blk_mq_wake_waiters(struct request_queue *q); |
1fd40b5e ML |
43 | bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *, |
44 | unsigned int); | |
e6c98712 BVA |
45 | void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, |
46 | bool kick_requeue_list); | |
2c3ad667 | 47 | void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list); |
b347689f ML |
48 | struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx, |
49 | struct blk_mq_ctx *start); | |
2c3ad667 JA |
50 | |
51 | /* | |
52 | * Internal helpers for allocating/freeing the request map | |
53 | */ | |
cc71a6f4 JA |
54 | void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, |
55 | unsigned int hctx_idx); | |
56 | void blk_mq_free_rq_map(struct blk_mq_tags *tags); | |
57 | struct blk_mq_tags *blk_mq_alloc_rq_map(struct blk_mq_tag_set *set, | |
58 | unsigned int hctx_idx, | |
59 | unsigned int nr_tags, | |
60 | unsigned int reserved_tags); | |
61 | int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags, | |
62 | unsigned int hctx_idx, unsigned int depth); | |
2c3ad667 JA |
63 | |
64 | /* | |
65 | * Internal helpers for request insertion into sw queues | |
66 | */ | |
67 | void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq, | |
68 | bool at_head); | |
01e99aec ML |
69 | void blk_mq_request_bypass_insert(struct request *rq, bool at_head, |
70 | bool run_queue); | |
bd166ef1 JA |
71 | void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx, |
72 | struct list_head *list); | |
320ae51f | 73 | |
fd9c40f6 BVA |
74 | /* Used by blk_insert_cloned_request() to issue request directly */ |
75 | blk_status_t blk_mq_request_issue_directly(struct request *rq, bool last); | |
6ce3dd6e ML |
76 | void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx, |
77 | struct list_head *list); | |
396eaf21 | 78 | |
320ae51f JA |
79 | /* |
80 | * CPU -> queue mappings | |
81 | */ | |
ed76e329 | 82 | extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int); |
320ae51f | 83 | |
b3c661b1 JA |
84 | /* |
85 | * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue | |
86 | * @q: request queue | |
e20ba6e1 | 87 | * @type: the hctx type index |
b3c661b1 JA |
88 | * @cpu: CPU |
89 | */ | |
90 | static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q, | |
e20ba6e1 | 91 | enum hctx_type type, |
b3c661b1 | 92 | unsigned int cpu) |
7d7e0f90 | 93 | { |
e20ba6e1 | 94 | return q->queue_hw_ctx[q->tag_set->map[type].mq_map[cpu]]; |
7d7e0f90 CH |
95 | } |
96 | ||
b3c661b1 JA |
97 | /* |
98 | * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue | |
99 | * @q: request queue | |
100 | * @flags: request command flags | |
8ccdf4a3 | 101 | * @cpu: cpu ctx |
b3c661b1 JA |
102 | */ |
103 | static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q, | |
104 | unsigned int flags, | |
8ccdf4a3 | 105 | struct blk_mq_ctx *ctx) |
ff2c5660 | 106 | { |
e20ba6e1 CH |
107 | enum hctx_type type = HCTX_TYPE_DEFAULT; |
108 | ||
bb94aea1 JW |
109 | /* |
110 | * The caller ensure that if REQ_HIPRI, poll must be enabled. | |
111 | */ | |
112 | if (flags & REQ_HIPRI) | |
e20ba6e1 | 113 | type = HCTX_TYPE_POLL; |
bb94aea1 | 114 | else if ((flags & REQ_OP_MASK) == REQ_OP_READ) |
e20ba6e1 | 115 | type = HCTX_TYPE_READ; |
5aceaeb2 | 116 | |
8ccdf4a3 | 117 | return ctx->hctxs[type]; |
ff2c5660 JA |
118 | } |
119 | ||
67aec14c JA |
120 | /* |
121 | * sysfs helpers | |
122 | */ | |
737f98cf | 123 | extern void blk_mq_sysfs_init(struct request_queue *q); |
7ea5fe31 | 124 | extern void blk_mq_sysfs_deinit(struct request_queue *q); |
2d0364c8 | 125 | extern int __blk_mq_register_dev(struct device *dev, struct request_queue *q); |
67aec14c JA |
126 | extern int blk_mq_sysfs_register(struct request_queue *q); |
127 | extern void blk_mq_sysfs_unregister(struct request_queue *q); | |
868f2f0b | 128 | extern void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx); |
67aec14c | 129 | |
e09aae7e ML |
130 | void blk_mq_release(struct request_queue *q); |
131 | ||
1aecfe48 ML |
132 | static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q, |
133 | unsigned int cpu) | |
134 | { | |
135 | return per_cpu_ptr(q->queue_ctx, cpu); | |
136 | } | |
137 | ||
138 | /* | |
139 | * This assumes per-cpu software queueing queues. They could be per-node | |
140 | * as well, for instance. For now this is hardcoded as-is. Note that we don't | |
141 | * care about preemption, since we know the ctx's are persistent. This does | |
142 | * mean that we can't rely on ctx always matching the currently running CPU. | |
143 | */ | |
144 | static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q) | |
145 | { | |
c05f4220 | 146 | return __blk_mq_get_ctx(q, raw_smp_processor_id()); |
1aecfe48 ML |
147 | } |
148 | ||
cb96a42c ML |
149 | struct blk_mq_alloc_data { |
150 | /* input parameter */ | |
151 | struct request_queue *q; | |
9a95e4ef | 152 | blk_mq_req_flags_t flags; |
229a9287 | 153 | unsigned int shallow_depth; |
f9afca4d | 154 | unsigned int cmd_flags; |
cb96a42c ML |
155 | |
156 | /* input & output parameter */ | |
157 | struct blk_mq_ctx *ctx; | |
158 | struct blk_mq_hw_ctx *hctx; | |
159 | }; | |
160 | ||
4941115b JA |
161 | static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data) |
162 | { | |
42fdc5e4 | 163 | if (data->q->elevator) |
bd166ef1 JA |
164 | return data->hctx->sched_tags; |
165 | ||
4941115b JA |
166 | return data->hctx->tags; |
167 | } | |
168 | ||
5d1b25c1 BVA |
169 | static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx) |
170 | { | |
171 | return test_bit(BLK_MQ_S_STOPPED, &hctx->state); | |
172 | } | |
173 | ||
19c66e59 ML |
174 | static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx) |
175 | { | |
176 | return hctx->nr_ctx && hctx->tags; | |
177 | } | |
178 | ||
e016b782 | 179 | unsigned int blk_mq_in_flight(struct request_queue *q, struct hd_struct *part); |
bf0ddaba OS |
180 | void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part, |
181 | unsigned int inflight[2]); | |
f299b7c7 | 182 | |
65c76369 | 183 | static inline void blk_mq_put_dispatch_budget(struct request_queue *q) |
de148297 | 184 | { |
de148297 | 185 | if (q->mq_ops->put_budget) |
65c76369 | 186 | q->mq_ops->put_budget(q); |
de148297 ML |
187 | } |
188 | ||
65c76369 | 189 | static inline bool blk_mq_get_dispatch_budget(struct request_queue *q) |
de148297 | 190 | { |
de148297 | 191 | if (q->mq_ops->get_budget) |
65c76369 | 192 | return q->mq_ops->get_budget(q); |
88022d72 | 193 | return true; |
de148297 ML |
194 | } |
195 | ||
4e2f62e5 JA |
196 | static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx, |
197 | struct request *rq) | |
198 | { | |
199 | blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag); | |
200 | rq->tag = BLK_MQ_NO_TAG; | |
201 | ||
202 | if (rq->rq_flags & RQF_MQ_INFLIGHT) { | |
203 | rq->rq_flags &= ~RQF_MQ_INFLIGHT; | |
204 | atomic_dec(&hctx->nr_active); | |
205 | } | |
206 | } | |
207 | ||
208 | static inline void blk_mq_put_driver_tag(struct request *rq) | |
209 | { | |
210 | if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG) | |
211 | return; | |
212 | ||
213 | __blk_mq_put_driver_tag(rq->mq_hctx, rq); | |
214 | } | |
215 | ||
ed76e329 | 216 | static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap) |
0da73d00 MI |
217 | { |
218 | int cpu; | |
219 | ||
220 | for_each_possible_cpu(cpu) | |
ed76e329 | 221 | qmap->mq_map[cpu] = 0; |
0da73d00 MI |
222 | } |
223 | ||
b49773e7 DLM |
224 | /* |
225 | * blk_mq_plug() - Get caller context plug | |
226 | * @q: request queue | |
227 | * @bio : the bio being submitted by the caller context | |
228 | * | |
229 | * Plugging, by design, may delay the insertion of BIOs into the elevator in | |
230 | * order to increase BIO merging opportunities. This however can cause BIO | |
231 | * insertion order to change from the order in which submit_bio() is being | |
232 | * executed in the case of multiple contexts concurrently issuing BIOs to a | |
233 | * device, even if these context are synchronized to tightly control BIO issuing | |
234 | * order. While this is not a problem with regular block devices, this ordering | |
235 | * change can cause write BIO failures with zoned block devices as these | |
236 | * require sequential write patterns to zones. Prevent this from happening by | |
237 | * ignoring the plug state of a BIO issuing context if the target request queue | |
238 | * is for a zoned block device and the BIO to plug is a write operation. | |
239 | * | |
240 | * Return current->plug if the bio can be plugged and NULL otherwise | |
241 | */ | |
242 | static inline struct blk_plug *blk_mq_plug(struct request_queue *q, | |
243 | struct bio *bio) | |
244 | { | |
245 | /* | |
246 | * For regular block devices or read operations, use the context plug | |
247 | * which may be NULL if blk_start_plug() was not executed. | |
248 | */ | |
249 | if (!blk_queue_is_zoned(q) || !op_is_write(bio_op(bio))) | |
250 | return current->plug; | |
251 | ||
252 | /* Zoned block device write operation case: do not plug the BIO */ | |
253 | return NULL; | |
254 | } | |
255 | ||
320ae51f | 256 | #endif |