block: Hold invalidate_lock in BLKZEROOUT ioctl
[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 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 33void blk_mq_submit_bio(struct bio *bio);
5a72e899
JA
34int blk_mq_poll(struct request_queue *q, blk_qc_t cookie, struct io_comp_batch *iob,
35 unsigned int flags);
c7e2d94b 36void blk_mq_exit_queue(struct request_queue *q);
e3a2b3f9 37int blk_mq_update_nr_requests(struct request_queue *q, unsigned int nr);
aed3ea94 38void blk_mq_wake_waiters(struct request_queue *q);
1fd40b5e
ML
39bool blk_mq_dispatch_rq_list(struct blk_mq_hw_ctx *hctx, struct list_head *,
40 unsigned int);
e6c98712
BVA
41void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
42 bool kick_requeue_list);
2c3ad667 43void blk_mq_flush_busy_ctxs(struct blk_mq_hw_ctx *hctx, struct list_head *list);
b347689f
ML
44struct request *blk_mq_dequeue_from_ctx(struct blk_mq_hw_ctx *hctx,
45 struct blk_mq_ctx *start);
2e315dc0 46void blk_mq_put_rq_ref(struct request *rq);
2c3ad667
JA
47
48/*
49 * Internal helpers for allocating/freeing the request map
50 */
cc71a6f4
JA
51void blk_mq_free_rqs(struct blk_mq_tag_set *set, struct blk_mq_tags *tags,
52 unsigned int hctx_idx);
e155b0c2 53void blk_mq_free_rq_map(struct blk_mq_tags *tags);
63064be1
JG
54struct 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
56void 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 */
62void __blk_mq_insert_request(struct blk_mq_hw_ctx *hctx, struct request *rq,
63 bool at_head);
01e99aec
ML
64void blk_mq_request_bypass_insert(struct request *rq, bool at_head,
65 bool run_queue);
bd166ef1
JA
66void blk_mq_insert_requests(struct blk_mq_hw_ctx *hctx, struct blk_mq_ctx *ctx,
67 struct list_head *list);
320ae51f 68
fd9c40f6
BVA
69/* Used by blk_insert_cloned_request() to issue request directly */
70blk_status_t blk_mq_request_issue_directly(struct request *rq, bool last);
6ce3dd6e
ML
71void blk_mq_try_issue_list_directly(struct blk_mq_hw_ctx *hctx,
72 struct list_head *list);
396eaf21 73
320ae51f
JA
74/*
75 * CPU -> queue mappings
76 */
ed76e329 77extern int blk_mq_hw_queue_to_node(struct blk_mq_queue_map *qmap, unsigned int);
320ae51f 78
b3c661b1
JA
79/*
80 * blk_mq_map_queue_type() - map (hctx_type,cpu) to hardware queue
81 * @q: request queue
e20ba6e1 82 * @type: the hctx type index
b3c661b1
JA
83 * @cpu: CPU
84 */
85static inline struct blk_mq_hw_ctx *blk_mq_map_queue_type(struct request_queue *q,
e20ba6e1 86 enum hctx_type type,
b3c661b1 87 unsigned int cpu)
7d7e0f90 88{
e20ba6e1 89 return q->queue_hw_ctx[q->tag_set->map[type].mq_map[cpu]];
7d7e0f90
CH
90}
91
b3c661b1
JA
92/*
93 * blk_mq_map_queue() - map (cmd_flags,type) to hardware queue
94 * @q: request queue
95 * @flags: request command flags
d220a214 96 * @ctx: software queue cpu ctx
b3c661b1
JA
97 */
98static inline struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *q,
99 unsigned int flags,
8ccdf4a3 100 struct blk_mq_ctx *ctx)
ff2c5660 101{
e20ba6e1
CH
102 enum hctx_type type = HCTX_TYPE_DEFAULT;
103
bb94aea1 104 /*
6ce913fe 105 * The caller ensure that if REQ_POLLED, poll must be enabled.
bb94aea1 106 */
6ce913fe 107 if (flags & REQ_POLLED)
e20ba6e1 108 type = HCTX_TYPE_POLL;
bb94aea1 109 else if ((flags & REQ_OP_MASK) == REQ_OP_READ)
e20ba6e1 110 type = HCTX_TYPE_READ;
5aceaeb2 111
8ccdf4a3 112 return ctx->hctxs[type];
ff2c5660
JA
113}
114
67aec14c
JA
115/*
116 * sysfs helpers
117 */
737f98cf 118extern void blk_mq_sysfs_init(struct request_queue *q);
7ea5fe31 119extern void blk_mq_sysfs_deinit(struct request_queue *q);
2d0364c8 120extern int __blk_mq_register_dev(struct device *dev, struct request_queue *q);
67aec14c
JA
121extern int blk_mq_sysfs_register(struct request_queue *q);
122extern void blk_mq_sysfs_unregister(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
e09aae7e
ML
127void blk_mq_release(struct request_queue *q);
128
1aecfe48
ML
129static inline struct blk_mq_ctx *__blk_mq_get_ctx(struct request_queue *q,
130 unsigned int cpu)
131{
132 return per_cpu_ptr(q->queue_ctx, cpu);
133}
134
135/*
136 * This assumes per-cpu software queueing queues. They could be per-node
137 * as well, for instance. For now this is hardcoded as-is. Note that we don't
138 * care about preemption, since we know the ctx's are persistent. This does
139 * mean that we can't rely on ctx always matching the currently running CPU.
140 */
141static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
142{
c05f4220 143 return __blk_mq_get_ctx(q, raw_smp_processor_id());
1aecfe48
ML
144}
145
cb96a42c
ML
146struct blk_mq_alloc_data {
147 /* input parameter */
148 struct request_queue *q;
9a95e4ef 149 blk_mq_req_flags_t flags;
229a9287 150 unsigned int shallow_depth;
f9afca4d 151 unsigned int cmd_flags;
56f8da64 152 unsigned int rq_flags;
cb96a42c 153
47c122e3
JA
154 /* allocate multiple requests/tags in one go */
155 unsigned int nr_tags;
156 struct request **cached_rq;
157
cb96a42c
ML
158 /* input & output parameter */
159 struct blk_mq_ctx *ctx;
160 struct blk_mq_hw_ctx *hctx;
161};
162
079a2e3e 163static inline bool blk_mq_is_shared_tags(unsigned int flags)
32bc15af
JG
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{
56f8da64
JA
170 if (!(data->rq_flags & RQF_ELV))
171 return data->hctx->tags;
172 return data->hctx->sched_tags;
4941115b
JA
173}
174
5d1b25c1
BVA
175static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx)
176{
177 return test_bit(BLK_MQ_S_STOPPED, &hctx->state);
178}
179
19c66e59
ML
180static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx)
181{
182 return hctx->nr_ctx && hctx->tags;
183}
184
8446fe92
CH
185unsigned int blk_mq_in_flight(struct request_queue *q,
186 struct block_device *part);
187void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
188 unsigned int inflight[2]);
f299b7c7 189
2a5a24aa
ML
190static inline void blk_mq_put_dispatch_budget(struct request_queue *q,
191 int budget_token)
de148297 192{
de148297 193 if (q->mq_ops->put_budget)
2a5a24aa 194 q->mq_ops->put_budget(q, budget_token);
de148297
ML
195}
196
2a5a24aa 197static inline int 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);
2a5a24aa
ML
201 return 0;
202}
203
204static inline void blk_mq_set_rq_budget_token(struct request *rq, int token)
205{
206 if (token < 0)
207 return;
208
209 if (rq->q->mq_ops->set_rq_budget_token)
210 rq->q->mq_ops->set_rq_budget_token(rq, token);
211}
212
213static inline int blk_mq_get_rq_budget_token(struct request *rq)
214{
215 if (rq->q->mq_ops->get_rq_budget_token)
216 return rq->q->mq_ops->get_rq_budget_token(rq);
217 return -1;
de148297
ML
218}
219
bccf5e26
JG
220static inline void __blk_mq_inc_active_requests(struct blk_mq_hw_ctx *hctx)
221{
079a2e3e
JG
222 if (blk_mq_is_shared_tags(hctx->flags))
223 atomic_inc(&hctx->queue->nr_active_requests_shared_tags);
bccf5e26
JG
224 else
225 atomic_inc(&hctx->nr_active);
226}
227
3b87c6ea
ML
228static inline void __blk_mq_sub_active_requests(struct blk_mq_hw_ctx *hctx,
229 int val)
bccf5e26 230{
079a2e3e 231 if (blk_mq_is_shared_tags(hctx->flags))
3b87c6ea 232 atomic_sub(val, &hctx->queue->nr_active_requests_shared_tags);
bccf5e26 233 else
3b87c6ea
ML
234 atomic_sub(val, &hctx->nr_active);
235}
236
237static inline void __blk_mq_dec_active_requests(struct blk_mq_hw_ctx *hctx)
238{
239 __blk_mq_sub_active_requests(hctx, 1);
bccf5e26
JG
240}
241
242static inline int __blk_mq_active_requests(struct blk_mq_hw_ctx *hctx)
243{
079a2e3e
JG
244 if (blk_mq_is_shared_tags(hctx->flags))
245 return atomic_read(&hctx->queue->nr_active_requests_shared_tags);
bccf5e26
JG
246 return atomic_read(&hctx->nr_active);
247}
4e2f62e5
JA
248static inline void __blk_mq_put_driver_tag(struct blk_mq_hw_ctx *hctx,
249 struct request *rq)
250{
251 blk_mq_put_tag(hctx->tags, rq->mq_ctx, rq->tag);
252 rq->tag = BLK_MQ_NO_TAG;
253
254 if (rq->rq_flags & RQF_MQ_INFLIGHT) {
255 rq->rq_flags &= ~RQF_MQ_INFLIGHT;
bccf5e26 256 __blk_mq_dec_active_requests(hctx);
4e2f62e5
JA
257 }
258}
259
260static inline void blk_mq_put_driver_tag(struct request *rq)
261{
262 if (rq->tag == BLK_MQ_NO_TAG || rq->internal_tag == BLK_MQ_NO_TAG)
263 return;
264
265 __blk_mq_put_driver_tag(rq->mq_hctx, rq);
266}
267
a808a9d5
JA
268bool __blk_mq_get_driver_tag(struct blk_mq_hw_ctx *hctx, struct request *rq);
269
270static inline bool blk_mq_get_driver_tag(struct request *rq)
271{
272 struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
273
274 if (rq->tag != BLK_MQ_NO_TAG &&
275 !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED)) {
276 hctx->tags->rqs[rq->tag] = rq;
277 return true;
278 }
279
280 return __blk_mq_get_driver_tag(hctx, rq);
281}
61347154 282
ed76e329 283static inline void blk_mq_clear_mq_map(struct blk_mq_queue_map *qmap)
0da73d00
MI
284{
285 int cpu;
286
287 for_each_possible_cpu(cpu)
ed76e329 288 qmap->mq_map[cpu] = 0;
0da73d00
MI
289}
290
b49773e7
DLM
291/*
292 * blk_mq_plug() - Get caller context plug
293 * @q: request queue
294 * @bio : the bio being submitted by the caller context
295 *
296 * Plugging, by design, may delay the insertion of BIOs into the elevator in
297 * order to increase BIO merging opportunities. This however can cause BIO
298 * insertion order to change from the order in which submit_bio() is being
299 * executed in the case of multiple contexts concurrently issuing BIOs to a
300 * device, even if these context are synchronized to tightly control BIO issuing
301 * order. While this is not a problem with regular block devices, this ordering
302 * change can cause write BIO failures with zoned block devices as these
303 * require sequential write patterns to zones. Prevent this from happening by
304 * ignoring the plug state of a BIO issuing context if the target request queue
305 * is for a zoned block device and the BIO to plug is a write operation.
306 *
307 * Return current->plug if the bio can be plugged and NULL otherwise
308 */
309static inline struct blk_plug *blk_mq_plug(struct request_queue *q,
310 struct bio *bio)
311{
312 /*
313 * For regular block devices or read operations, use the context plug
314 * which may be NULL if blk_start_plug() was not executed.
315 */
316 if (!blk_queue_is_zoned(q) || !op_is_write(bio_op(bio)))
317 return current->plug;
318
319 /* Zoned block device write operation case: do not plug the BIO */
320 return NULL;
321}
322
fd2ef39c
JK
323/* Free all requests on the list */
324static inline void blk_mq_free_requests(struct list_head *list)
325{
326 while (!list_empty(list)) {
327 struct request *rq = list_entry_rq(list->next);
328
329 list_del_init(&rq->queuelist);
330 blk_mq_free_request(rq);
331 }
332}
333
a0235d23
JG
334/*
335 * For shared tag users, we track the number of currently active users
336 * and attempt to provide a fair share of the tag depth for each of them.
337 */
338static inline bool hctx_may_queue(struct blk_mq_hw_ctx *hctx,
339 struct sbitmap_queue *bt)
340{
341 unsigned int depth, users;
342
343 if (!hctx || !(hctx->flags & BLK_MQ_F_TAG_QUEUE_SHARED))
344 return true;
a0235d23
JG
345
346 /*
347 * Don't try dividing an ant
348 */
349 if (bt->sb.depth == 1)
350 return true;
351
079a2e3e 352 if (blk_mq_is_shared_tags(hctx->flags)) {
f1b49fdc 353 struct request_queue *q = hctx->queue;
f1b49fdc 354
2569063c 355 if (!test_bit(QUEUE_FLAG_HCTX_ACTIVE, &q->queue_flags))
f1b49fdc 356 return true;
f1b49fdc
JG
357 } else {
358 if (!test_bit(BLK_MQ_S_TAG_ACTIVE, &hctx->state))
359 return true;
f1b49fdc
JG
360 }
361
e155b0c2
JG
362 users = atomic_read(&hctx->tags->active_queues);
363
a0235d23
JG
364 if (!users)
365 return true;
366
367 /*
368 * Allow at least some tags
369 */
370 depth = max((bt->sb.depth + users - 1) / users, 4U);
bccf5e26 371 return __blk_mq_active_requests(hctx) < depth;
a0235d23
JG
372}
373
374
320ae51f 375#endif