Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
320ae51f JA |
2 | #ifndef BLK_MQ_H |
3 | #define BLK_MQ_H | |
4 | ||
5 | #include <linux/blkdev.h> | |
88459642 | 6 | #include <linux/sbitmap.h> |
6a83e74d | 7 | #include <linux/srcu.h> |
320ae51f JA |
8 | |
9 | struct blk_mq_tags; | |
f70ced09 | 10 | struct blk_flush_queue; |
320ae51f | 11 | |
fe644072 LW |
12 | /** |
13 | * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware block device | |
14 | */ | |
320ae51f JA |
15 | struct blk_mq_hw_ctx { |
16 | struct { | |
17 | spinlock_t lock; | |
18 | struct list_head dispatch; | |
8d354f13 | 19 | unsigned long state; /* BLK_MQ_S_* flags */ |
320ae51f JA |
20 | } ____cacheline_aligned_in_smp; |
21 | ||
9f993737 | 22 | struct delayed_work run_work; |
e4043dcf | 23 | cpumask_var_t cpumask; |
506e931f JA |
24 | int next_cpu; |
25 | int next_cpu_batch; | |
320ae51f JA |
26 | |
27 | unsigned long flags; /* BLK_MQ_F_* flags */ | |
28 | ||
bd166ef1 | 29 | void *sched_data; |
320ae51f | 30 | struct request_queue *queue; |
f70ced09 | 31 | struct blk_flush_queue *fq; |
320ae51f JA |
32 | |
33 | void *driver_data; | |
34 | ||
88459642 | 35 | struct sbitmap ctx_map; |
1429d7c9 | 36 | |
b347689f | 37 | struct blk_mq_ctx *dispatch_from; |
6e768717 | 38 | unsigned int dispatch_busy; |
b347689f | 39 | |
8d354f13 | 40 | unsigned int nr_ctx; |
6e768717 | 41 | struct blk_mq_ctx **ctxs; |
4bb659b1 | 42 | |
5815839b | 43 | spinlock_t dispatch_wait_lock; |
eb619fdb | 44 | wait_queue_entry_t dispatch_wait; |
8537b120 | 45 | atomic_t wait_index; |
320ae51f | 46 | |
320ae51f | 47 | struct blk_mq_tags *tags; |
bd166ef1 | 48 | struct blk_mq_tags *sched_tags; |
320ae51f JA |
49 | |
50 | unsigned long queued; | |
51 | unsigned long run; | |
8d354f13 | 52 | #define BLK_MQ_MAX_DISPATCH_ORDER 7 |
320ae51f JA |
53 | unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; |
54 | ||
320ae51f | 55 | unsigned int numa_node; |
17ded320 | 56 | unsigned int queue_num; |
320ae51f | 57 | |
0d2602ca | 58 | atomic_t nr_active; |
1d9bd516 | 59 | unsigned int nr_expired; |
0d2602ca | 60 | |
9467f859 | 61 | struct hlist_node cpuhp_dead; |
320ae51f | 62 | struct kobject kobj; |
05229bee | 63 | |
6e219353 | 64 | unsigned long poll_considered; |
05229bee JA |
65 | unsigned long poll_invoked; |
66 | unsigned long poll_success; | |
9c1051aa OS |
67 | |
68 | #ifdef CONFIG_BLK_DEBUG_FS | |
69 | struct dentry *debugfs_dir; | |
d332ce09 | 70 | struct dentry *sched_debugfs_dir; |
9c1051aa | 71 | #endif |
07319678 BVA |
72 | |
73 | /* Must be the last member - see also blk_mq_hw_ctx_size(). */ | |
05707b64 | 74 | struct srcu_struct srcu[0]; |
320ae51f JA |
75 | }; |
76 | ||
24d2f903 | 77 | struct blk_mq_tag_set { |
bdd17e75 | 78 | unsigned int *mq_map; |
f8a5b122 | 79 | const struct blk_mq_ops *ops; |
320ae51f | 80 | unsigned int nr_hw_queues; |
e3a2b3f9 | 81 | unsigned int queue_depth; /* max hw supported */ |
320ae51f JA |
82 | unsigned int reserved_tags; |
83 | unsigned int cmd_size; /* per-request extra data */ | |
84 | int numa_node; | |
85 | unsigned int timeout; | |
86 | unsigned int flags; /* BLK_MQ_F_* */ | |
24d2f903 CH |
87 | void *driver_data; |
88 | ||
89 | struct blk_mq_tags **tags; | |
0d2602ca JA |
90 | |
91 | struct mutex tag_list_lock; | |
92 | struct list_head tag_list; | |
320ae51f JA |
93 | }; |
94 | ||
74c45052 JA |
95 | struct blk_mq_queue_data { |
96 | struct request *rq; | |
74c45052 JA |
97 | bool last; |
98 | }; | |
99 | ||
fc17b653 CH |
100 | typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *, |
101 | const struct blk_mq_queue_data *); | |
88022d72 | 102 | typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *); |
de148297 | 103 | typedef void (put_budget_fn)(struct blk_mq_hw_ctx *); |
0152fb6b | 104 | typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool); |
320ae51f JA |
105 | typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); |
106 | typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); | |
d6296d39 | 107 | typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *, |
24d2f903 | 108 | unsigned int, unsigned int); |
d6296d39 | 109 | typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *, |
24d2f903 | 110 | unsigned int); |
320ae51f | 111 | |
81481eb4 CH |
112 | typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, |
113 | bool); | |
f26cdc85 | 114 | typedef void (busy_tag_iter_fn)(struct request *, void *, bool); |
05229bee | 115 | typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int); |
da695ba2 | 116 | typedef int (map_queues_fn)(struct blk_mq_tag_set *set); |
05229bee | 117 | |
81481eb4 | 118 | |
320ae51f JA |
119 | struct blk_mq_ops { |
120 | /* | |
121 | * Queue request | |
122 | */ | |
123 | queue_rq_fn *queue_rq; | |
124 | ||
de148297 ML |
125 | /* |
126 | * Reserve budget before queue request, once .queue_rq is | |
127 | * run, it is driver's responsibility to release the | |
128 | * reserved budget. Also we have to handle failure case | |
129 | * of .get_budget for avoiding I/O deadlock. | |
130 | */ | |
131 | get_budget_fn *get_budget; | |
132 | put_budget_fn *put_budget; | |
133 | ||
320ae51f JA |
134 | /* |
135 | * Called on request timeout | |
136 | */ | |
0152fb6b | 137 | timeout_fn *timeout; |
320ae51f | 138 | |
05229bee JA |
139 | /* |
140 | * Called to poll for completion of a specific tag. | |
141 | */ | |
142 | poll_fn *poll; | |
143 | ||
30a91cb4 CH |
144 | softirq_done_fn *complete; |
145 | ||
320ae51f JA |
146 | /* |
147 | * Called when the block layer side of a hardware queue has been | |
148 | * set up, allowing the driver to allocate/init matching structures. | |
149 | * Ditto for exit/teardown. | |
150 | */ | |
151 | init_hctx_fn *init_hctx; | |
152 | exit_hctx_fn *exit_hctx; | |
e9b267d9 CH |
153 | |
154 | /* | |
155 | * Called for every command allocated by the block layer to allow | |
156 | * the driver to set up driver specific data. | |
f70ced09 ML |
157 | * |
158 | * Tag greater than or equal to queue_depth is for setting up | |
159 | * flush request. | |
160 | * | |
e9b267d9 CH |
161 | * Ditto for exit/teardown. |
162 | */ | |
163 | init_request_fn *init_request; | |
164 | exit_request_fn *exit_request; | |
d280bab3 BVA |
165 | /* Called from inside blk_get_request() */ |
166 | void (*initialize_rq_fn)(struct request *rq); | |
da695ba2 CH |
167 | |
168 | map_queues_fn *map_queues; | |
2836ee4b BVA |
169 | |
170 | #ifdef CONFIG_BLK_DEBUG_FS | |
171 | /* | |
172 | * Used by the debugfs implementation to show driver-specific | |
173 | * information about a request. | |
174 | */ | |
175 | void (*show_rq)(struct seq_file *m, struct request *rq); | |
176 | #endif | |
320ae51f JA |
177 | }; |
178 | ||
179 | enum { | |
320ae51f | 180 | BLK_MQ_F_SHOULD_MERGE = 1 << 0, |
8a58d1f1 JA |
181 | BLK_MQ_F_TAG_SHARED = 1 << 1, |
182 | BLK_MQ_F_SG_MERGE = 1 << 2, | |
1b792f2f | 183 | BLK_MQ_F_BLOCKING = 1 << 5, |
d3484991 | 184 | BLK_MQ_F_NO_SCHED = 1 << 6, |
24391c0d SL |
185 | BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, |
186 | BLK_MQ_F_ALLOC_POLICY_BITS = 1, | |
320ae51f | 187 | |
5d12f905 | 188 | BLK_MQ_S_STOPPED = 0, |
0d2602ca | 189 | BLK_MQ_S_TAG_ACTIVE = 1, |
bd166ef1 | 190 | BLK_MQ_S_SCHED_RESTART = 2, |
320ae51f | 191 | |
a4391c64 | 192 | BLK_MQ_MAX_DEPTH = 10240, |
506e931f JA |
193 | |
194 | BLK_MQ_CPU_WORK_BATCH = 8, | |
320ae51f | 195 | }; |
24391c0d SL |
196 | #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ |
197 | ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ | |
198 | ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) | |
199 | #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ | |
200 | ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ | |
201 | << BLK_MQ_F_ALLOC_POLICY_START_BIT) | |
320ae51f | 202 | |
24d2f903 | 203 | struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); |
b62c21b7 MS |
204 | struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, |
205 | struct request_queue *q); | |
b21d5b30 MB |
206 | int blk_mq_register_dev(struct device *, struct request_queue *); |
207 | void blk_mq_unregister_dev(struct device *, struct request_queue *); | |
320ae51f | 208 | |
24d2f903 CH |
209 | int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); |
210 | void blk_mq_free_tag_set(struct blk_mq_tag_set *set); | |
211 | ||
320ae51f JA |
212 | void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); |
213 | ||
320ae51f JA |
214 | void blk_mq_free_request(struct request *rq); |
215 | bool blk_mq_can_queue(struct blk_mq_hw_ctx *); | |
6f3b0e8b CH |
216 | |
217 | enum { | |
9a95e4ef BVA |
218 | /* return when out of requests */ |
219 | BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0), | |
220 | /* allocate from reserved pool */ | |
221 | BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1), | |
222 | /* allocate internal/sched tag */ | |
223 | BLK_MQ_REQ_INTERNAL = (__force blk_mq_req_flags_t)(1 << 2), | |
224 | /* set RQF_PREEMPT */ | |
225 | BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3), | |
6f3b0e8b CH |
226 | }; |
227 | ||
cd6ce148 | 228 | struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op, |
9a95e4ef | 229 | blk_mq_req_flags_t flags); |
cd6ce148 | 230 | struct request *blk_mq_alloc_request_hctx(struct request_queue *q, |
9a95e4ef BVA |
231 | unsigned int op, blk_mq_req_flags_t flags, |
232 | unsigned int hctx_idx); | |
0e62f51f | 233 | struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); |
320ae51f | 234 | |
205fb5f5 BVA |
235 | enum { |
236 | BLK_MQ_UNIQUE_TAG_BITS = 16, | |
237 | BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, | |
238 | }; | |
239 | ||
240 | u32 blk_mq_unique_tag(struct request *rq); | |
241 | ||
242 | static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) | |
243 | { | |
244 | return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; | |
245 | } | |
246 | ||
247 | static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) | |
248 | { | |
249 | return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; | |
250 | } | |
251 | ||
320ae51f | 252 | |
973c0191 | 253 | int blk_mq_request_started(struct request *rq); |
e2490073 | 254 | void blk_mq_start_request(struct request *rq); |
2a842aca CH |
255 | void blk_mq_end_request(struct request *rq, blk_status_t error); |
256 | void __blk_mq_end_request(struct request *rq, blk_status_t error); | |
320ae51f | 257 | |
2b053aca BVA |
258 | void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); |
259 | void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, | |
260 | bool kick_requeue_list); | |
6fca6a61 | 261 | void blk_mq_kick_requeue_list(struct request_queue *q); |
2849450a | 262 | void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); |
08e0029a | 263 | void blk_mq_complete_request(struct request *rq); |
9c558734 JA |
264 | bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list, |
265 | struct bio *bio); | |
fd001443 | 266 | bool blk_mq_queue_stopped(struct request_queue *q); |
320ae51f JA |
267 | void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); |
268 | void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); | |
280d45f6 | 269 | void blk_mq_stop_hw_queues(struct request_queue *q); |
2f268556 | 270 | void blk_mq_start_hw_queues(struct request_queue *q); |
ae911c5e | 271 | void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); |
1b4a3258 | 272 | void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); |
97e01209 | 273 | void blk_mq_quiesce_queue(struct request_queue *q); |
e4e73913 | 274 | void blk_mq_unquiesce_queue(struct request_queue *q); |
7587a5ae | 275 | void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); |
79f720a7 | 276 | bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); |
b94ec296 | 277 | void blk_mq_run_hw_queues(struct request_queue *q, bool async); |
e0489487 SG |
278 | void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, |
279 | busy_tag_iter_fn *fn, void *priv); | |
c761d96b | 280 | void blk_mq_freeze_queue(struct request_queue *q); |
b4c6a028 | 281 | void blk_mq_unfreeze_queue(struct request_queue *q); |
1671d522 | 282 | void blk_freeze_queue_start(struct request_queue *q); |
6bae363e | 283 | void blk_mq_freeze_queue_wait(struct request_queue *q); |
f91328c4 KB |
284 | int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, |
285 | unsigned long timeout); | |
320ae51f | 286 | |
9e5a7e22 | 287 | int blk_mq_map_queues(struct blk_mq_tag_set *set); |
868f2f0b KB |
288 | void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); |
289 | ||
852ec809 | 290 | void blk_mq_quiesce_queue_nowait(struct request_queue *q); |
4f084b41 | 291 | |
0fc09f92 KB |
292 | /** |
293 | * blk_mq_mark_complete() - Set request state to complete | |
294 | * @rq: request to set to complete state | |
295 | * | |
296 | * Returns true if request state was successfully set to complete. If | |
297 | * successful, the caller is responsibile for seeing this request is ended, as | |
298 | * blk_mq_complete_request will not work again. | |
299 | */ | |
300 | static inline bool blk_mq_mark_complete(struct request *rq) | |
301 | { | |
302 | return cmpxchg(&rq->state, MQ_RQ_IN_FLIGHT, MQ_RQ_COMPLETE) == | |
303 | MQ_RQ_IN_FLIGHT; | |
304 | } | |
305 | ||
320ae51f JA |
306 | /* |
307 | * Driver command data is immediately after the request. So subtract request | |
2963e3f7 | 308 | * size to get back to the original request, add request size to get the PDU. |
320ae51f JA |
309 | */ |
310 | static inline struct request *blk_mq_rq_from_pdu(void *pdu) | |
311 | { | |
312 | return pdu - sizeof(struct request); | |
313 | } | |
314 | static inline void *blk_mq_rq_to_pdu(struct request *rq) | |
315 | { | |
2963e3f7 | 316 | return rq + 1; |
320ae51f JA |
317 | } |
318 | ||
320ae51f | 319 | #define queue_for_each_hw_ctx(q, hctx, i) \ |
0d0b7d42 JA |
320 | for ((i) = 0; (i) < (q)->nr_hw_queues && \ |
321 | ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) | |
320ae51f | 322 | |
320ae51f | 323 | #define hctx_for_each_ctx(hctx, ctx, i) \ |
0d0b7d42 JA |
324 | for ((i) = 0; (i) < (hctx)->nr_ctx && \ |
325 | ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) | |
320ae51f | 326 | |
320ae51f | 327 | #endif |