]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blame - block/blk-mq.h
bpf: Add oversize check before call kvcalloc()
[mirror_ubuntu-hirsute-kernel.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);
4e4f3102 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);
1c0706a7 57void blk_mq_free_rq_map(struct blk_mq_tags *tags, unsigned int flags);
cc71a6f4
JA
58struct blk_mq_tags *blk_mq_alloc_rq_map(struct blk_mq_tag_set *set,
59 unsigned int hctx_idx,
60 unsigned int nr_tags,
1c0706a7
JG
61 unsigned int reserved_tags,
62 unsigned int flags);
cc71a6f4
JA
63int blk_mq_alloc_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
64 unsigned int hctx_idx, unsigned int depth);
2c3ad667
JA
65
66/*
67 * Internal helpers for request insertion into sw queues
68 */
69void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
70 bool at_head);
01e99aec
ML
71void blk_mq_request_bypass_insert(struct request *rq, bool at_head,
72 bool run_queue);
bd166ef1
JA
73void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx,
74 struct list_head *list);
320ae51f 75
fd9c40f6
BVA
76/* Used by blk_insert_cloned_request() to issue request directly */
77blk_status_t blk_mq_request_issue_directly(struct request *rq, bool last);
6ce3dd6e
ML
78void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
79 struct list_head *list);
396eaf21 80
320ae51f
JA
81/*
82 * CPU -> queue mappings
83 */
ed76e329 84extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
320ae51f 85
b3c661b1
JA
86/*
87 * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
88 * @q: request queue
e20ba6e1 89 * @type: the hctx type index
b3c661b1
JA
90 * @cpu: CPU
91 */
92static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
e20ba6e1 93 enum hctx_type type,
b3c661b1 94 unsigned int cpu)
7d7e0f90 95{
e20ba6e1 96 return q->queue_hw_ctx[q->tag_set->map[type].mq_map[cpu]];
7d7e0f90
CH
97}
98
b3c661b1
JA
99/*
100 * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
101 * @q: request queue
102 * @flags: request command flags
d220a214 103 * @ctx: software queue cpu ctx
b3c661b1
JA
104 */
105static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
106 unsigned int flags,
8ccdf4a3 107 struct blk_mq_ctx *ctx)
ff2c5660 108{
e20ba6e1
CH
109 enum hctx_type type = HCTX_TYPE_DEFAULT;
110
bb94aea1
JW
111 /*
112 * The caller ensure that if REQ_HIPRI, poll must be enabled.
113 */
114 if (flags & REQ_HIPRI)
e20ba6e1 115 type = HCTX_TYPE_POLL;
bb94aea1 116 else if ((flags & REQ_OP_MASK) == REQ_OP_READ)
e20ba6e1 117 type = HCTX_TYPE_READ;
5aceaeb2 118
8ccdf4a3 119 return ctx->hctxs[type];
ff2c5660
JA
120}
121
67aec14c
JA
122/*
123 * sysfs helpers
124 */
737f98cf 125extern void blk_mq_sysfs_init(struct request_queue *q);
7ea5fe31 126extern void blk_mq_sysfs_deinit(struct request_queue *q);
2d0364c8 127extern int __blk_mq_register_dev(struct device *dev, struct request_queue *q);
67aec14c
JA
128extern int blk_mq_sysfs_register(struct request_queue *q);
129extern void blk_mq_sysfs_unregister(struct request_queue *q);
868f2f0b 130extern void blk_mq_hctx_kobj_init(struct blk_mq_hw_ctx *hctx);
67aec14c 131
e09aae7e
ML
132void blk_mq_release(struct request_queue *q);
133
1aecfe48
ML
134static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
135 unsigned int cpu)
136{
137 return per_cpu_ptr(q->queue_ctx, cpu);
138}
139
140/*
141 * This assumes per-cpu software queueing queues. They could be per-node
142 * as well, for instance. For now this is hardcoded as-is. Note that we don't
143 * care about preemption, since we know the ctx's are persistent. This does
144 * mean that we can't rely on ctx always matching the currently running CPU.
145 */
146static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
147{
c05f4220 148 return __blk_mq_get_ctx(q, raw_smp_processor_id());
1aecfe48
ML
149}
150
cb96a42c
ML
151struct blk_mq_alloc_data {
152 /* input parameter */
153 struct request_queue *q;
9a95e4ef 154 blk_mq_req_flags_t flags;
229a9287 155 unsigned int shallow_depth;
f9afca4d 156 unsigned int cmd_flags;
cb96a42c
ML
157
158 /* input & output parameter */
159 struct blk_mq_ctx *ctx;
160 struct blk_mq_hw_ctx *hctx;
161};
162
32bc15af
JG
163static inline bool blk_mq_is_sbitmap_shared(unsigned int flags)
164{
165 return flags & BLK_MQ_F_TAG_HCTX_SHARED;
166}
167
4941115b
JA
168static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data)
169{
42fdc5e4 170 if (data->q->elevator)
bd166ef1
JA
171 return data->hctx->sched_tags;
172
4941115b
JA
173 return data->hctx->tags;
174}
175
5d1b25c1
BVA
176static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx)
177{
178 return test_bit(BLK_MQ_S_STOPPED, &hctx->state);
179}
180
19c66e59
ML
181static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx)
182{
183 return hctx->nr_ctx && hctx->tags;
184}
185
8446fe92
CH
186unsigned int blk_mq_in_flight(struct request_queue *q,
187 struct block_device *part);
188void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
189 unsigned int inflight[2]);
f299b7c7 190
65c76369 191static inline void blk_mq_put_dispatch_budget(struct request_queue *q)
de148297 192{
de148297 193 if (q->mq_ops->put_budget)
65c76369 194 q->mq_ops->put_budget(q);
de148297
ML
195}
196
65c76369 197static inline bool blk_mq_get_dispatch_budget(struct request_queue *q)
de148297 198{
de148297 199 if (q->mq_ops->get_budget)
65c76369 200 return q->mq_ops->get_budget(q);
88022d72 201 return true;
de148297
ML
202}
203
bccf5e26
JG
204static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
205{
206 if (blk_mq_is_sbitmap_shared(hctx->flags))
207 atomic_inc(&hctx->queue->nr_active_requests_shared_sbitmap);
208 else
209 atomic_inc(&hctx->nr_active);
210}
211
212static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
213{
214 if (blk_mq_is_sbitmap_shared(hctx->flags))
215 atomic_dec(&hctx->queue->nr_active_requests_shared_sbitmap);
216 else
217 atomic_dec(&hctx->nr_active);
218}
219
220static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)
221{
222 if (blk_mq_is_sbitmap_shared(hctx->flags))
223 return atomic_read(&hctx->queue->nr_active_requests_shared_sbitmap);
224 return atomic_read(&hctx->nr_active);
225}
4e2f62e5
JA
226static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx,
227 struct request *rq)
228{
229 blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag);
230 rq->tag = BLK_MQ_NO_TAG;
231
232 if (rq->rq_flags & RQF_MQ_INFLIGHT) {
233 rq->rq_flags &= ~RQF_MQ_INFLIGHT;
bccf5e26 234 __blk_mq_dec_active_requests(hctx);
4e2f62e5
JA
235 }
236}
237
238static inline void blk_mq_put_driver_tag(struct request *rq)
239{
240 if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG)
241 return;
242
243 __blk_mq_put_driver_tag(rq->mq_hctx, rq);
244}
245
ed76e329 246static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
0da73d00
MI
247{
248 int cpu;
249
250 for_each_possible_cpu(cpu)
ed76e329 251 qmap->mq_map[cpu] = 0;
0da73d00
MI
252}
253
b49773e7
DLM
254/*
255 * blk_mq_plug() - Get caller context plug
256 * @q: request queue
257 * @bio : the bio being submitted by the caller context
258 *
259 * Plugging, by design, may delay the insertion of BIOs into the elevator in
260 * order to increase BIO merging opportunities. This however can cause BIO
261 * insertion order to change from the order in which submit_bio() is being
262 * executed in the case of multiple contexts concurrently issuing BIOs to a
263 * device, even if these context are synchronized to tightly control BIO issuing
264 * order. While this is not a problem with regular block devices, this ordering
265 * change can cause write BIO failures with zoned block devices as these
266 * require sequential write patterns to zones. Prevent this from happening by
267 * ignoring the plug state of a BIO issuing context if the target request queue
268 * is for a zoned block device and the BIO to plug is a write operation.
269 *
270 * Return current->plug if the bio can be plugged and NULL otherwise
271 */
272static inline struct blk_plug *blk_mq_plug(struct request_queue *q,
273 struct bio *bio)
274{
275 /*
276 * For regular block devices or read operations, use the context plug
277 * which may be NULL if blk_start_plug() was not executed.
278 */
279 if (!blk_queue_is_zoned(q) || !op_is_write(bio_op(bio)))
280 return current->plug;
281
282 /* Zoned block device write operation case: do not plug the BIO */
283 return NULL;
284}
285
a0235d23
JG
286/*
287 * For shared tag users, we track the number of currently active users
288 * and attempt to provide a fair share of the tag depth for each of them.
289 */
290static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
291 struct sbitmap_queue *bt)
292{
293 unsigned int depth, users;
294
295 if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED))
296 return true;
a0235d23
JG
297
298 /*
299 * Don't try dividing an ant
300 */
301 if (bt->sb.depth == 1)
302 return true;
303
f1b49fdc
JG
304 if (blk_mq_is_sbitmap_shared(hctx->flags)) {
305 struct request_queue *q = hctx->queue;
306 struct blk_mq_tag_set *set = q->tag_set;
307
2569063c 308 if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags))
f1b49fdc
JG
309 return true;
310 users = atomic_read(&set->active_queues_shared_sbitmap);
311 } else {
312 if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
313 return true;
314 users = atomic_read(&hctx->tags->active_queues);
315 }
316
a0235d23
JG
317 if (!users)
318 return true;
319
320 /*
321 * Allow at least some tags
322 */
323 depth = max((bt->sb.depth + users - 1) / users, 4U);
bccf5e26 324 return __blk_mq_active_requests(hctx) < depth;
a0235d23
JG
325}
326
327
320ae51f 328#endif