]>
Commit | Line | Data |
---|---|---|
320ae51f JA |
1 | #ifndef BLK_MQ_H |
2 | #define BLK_MQ_H | |
3 | ||
4 | #include <linux/blkdev.h> | |
88459642 | 5 | #include <linux/sbitmap.h> |
6a83e74d | 6 | #include <linux/srcu.h> |
320ae51f JA |
7 | |
8 | struct blk_mq_tags; | |
f70ced09 | 9 | struct blk_flush_queue; |
320ae51f | 10 | |
320ae51f JA |
11 | struct blk_mq_hw_ctx { |
12 | struct { | |
13 | spinlock_t lock; | |
14 | struct list_head dispatch; | |
8d354f13 | 15 | unsigned long state; /* BLK_MQ_S_* flags */ |
320ae51f JA |
16 | } ____cacheline_aligned_in_smp; |
17 | ||
27489a3c | 18 | struct work_struct run_work; |
e4043dcf | 19 | cpumask_var_t cpumask; |
506e931f JA |
20 | int next_cpu; |
21 | int next_cpu_batch; | |
320ae51f JA |
22 | |
23 | unsigned long flags; /* BLK_MQ_F_* flags */ | |
24 | ||
bd166ef1 | 25 | void *sched_data; |
320ae51f | 26 | struct request_queue *queue; |
f70ced09 | 27 | struct blk_flush_queue *fq; |
320ae51f JA |
28 | |
29 | void *driver_data; | |
30 | ||
88459642 | 31 | struct sbitmap ctx_map; |
1429d7c9 | 32 | |
4bb659b1 | 33 | struct blk_mq_ctx **ctxs; |
8d354f13 | 34 | unsigned int nr_ctx; |
4bb659b1 | 35 | |
da55f2cc | 36 | wait_queue_t dispatch_wait; |
8537b120 | 37 | atomic_t wait_index; |
320ae51f | 38 | |
320ae51f | 39 | struct blk_mq_tags *tags; |
bd166ef1 | 40 | struct blk_mq_tags *sched_tags; |
320ae51f | 41 | |
6a83e74d BVA |
42 | struct srcu_struct queue_rq_srcu; |
43 | ||
320ae51f JA |
44 | unsigned long queued; |
45 | unsigned long run; | |
8d354f13 | 46 | #define BLK_MQ_MAX_DISPATCH_ORDER 7 |
320ae51f JA |
47 | unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; |
48 | ||
320ae51f | 49 | unsigned int numa_node; |
17ded320 | 50 | unsigned int queue_num; |
320ae51f | 51 | |
0d2602ca JA |
52 | atomic_t nr_active; |
53 | ||
8d354f13 JA |
54 | struct delayed_work delay_work; |
55 | ||
9467f859 | 56 | struct hlist_node cpuhp_dead; |
320ae51f | 57 | struct kobject kobj; |
05229bee | 58 | |
6e219353 | 59 | unsigned long poll_considered; |
05229bee JA |
60 | unsigned long poll_invoked; |
61 | unsigned long poll_success; | |
320ae51f JA |
62 | }; |
63 | ||
24d2f903 | 64 | struct blk_mq_tag_set { |
bdd17e75 | 65 | unsigned int *mq_map; |
f8a5b122 | 66 | const struct blk_mq_ops *ops; |
320ae51f | 67 | unsigned int nr_hw_queues; |
e3a2b3f9 | 68 | unsigned int queue_depth; /* max hw supported */ |
320ae51f JA |
69 | unsigned int reserved_tags; |
70 | unsigned int cmd_size; /* per-request extra data */ | |
71 | int numa_node; | |
72 | unsigned int timeout; | |
73 | unsigned int flags; /* BLK_MQ_F_* */ | |
24d2f903 CH |
74 | void *driver_data; |
75 | ||
76 | struct blk_mq_tags **tags; | |
0d2602ca JA |
77 | |
78 | struct mutex tag_list_lock; | |
79 | struct list_head tag_list; | |
320ae51f JA |
80 | }; |
81 | ||
74c45052 JA |
82 | struct blk_mq_queue_data { |
83 | struct request *rq; | |
84 | struct list_head *list; | |
85 | bool last; | |
86 | }; | |
87 | ||
88 | typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *); | |
0152fb6b | 89 | typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool); |
320ae51f JA |
90 | typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); |
91 | typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); | |
24d2f903 CH |
92 | typedef int (init_request_fn)(void *, struct request *, unsigned int, |
93 | unsigned int, unsigned int); | |
94 | typedef void (exit_request_fn)(void *, struct request *, unsigned int, | |
95 | unsigned int); | |
486cf989 | 96 | typedef int (reinit_request_fn)(void *, struct request *); |
320ae51f | 97 | |
81481eb4 CH |
98 | typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, |
99 | bool); | |
f26cdc85 | 100 | typedef void (busy_tag_iter_fn)(struct request *, void *, bool); |
05229bee | 101 | typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int); |
da695ba2 | 102 | typedef int (map_queues_fn)(struct blk_mq_tag_set *set); |
05229bee | 103 | |
81481eb4 | 104 | |
320ae51f JA |
105 | struct blk_mq_ops { |
106 | /* | |
107 | * Queue request | |
108 | */ | |
109 | queue_rq_fn *queue_rq; | |
110 | ||
320ae51f JA |
111 | /* |
112 | * Called on request timeout | |
113 | */ | |
0152fb6b | 114 | timeout_fn *timeout; |
320ae51f | 115 | |
05229bee JA |
116 | /* |
117 | * Called to poll for completion of a specific tag. | |
118 | */ | |
119 | poll_fn *poll; | |
120 | ||
30a91cb4 CH |
121 | softirq_done_fn *complete; |
122 | ||
320ae51f JA |
123 | /* |
124 | * Called when the block layer side of a hardware queue has been | |
125 | * set up, allowing the driver to allocate/init matching structures. | |
126 | * Ditto for exit/teardown. | |
127 | */ | |
128 | init_hctx_fn *init_hctx; | |
129 | exit_hctx_fn *exit_hctx; | |
e9b267d9 CH |
130 | |
131 | /* | |
132 | * Called for every command allocated by the block layer to allow | |
133 | * the driver to set up driver specific data. | |
f70ced09 ML |
134 | * |
135 | * Tag greater than or equal to queue_depth is for setting up | |
136 | * flush request. | |
137 | * | |
e9b267d9 CH |
138 | * Ditto for exit/teardown. |
139 | */ | |
140 | init_request_fn *init_request; | |
141 | exit_request_fn *exit_request; | |
486cf989 | 142 | reinit_request_fn *reinit_request; |
da695ba2 CH |
143 | |
144 | map_queues_fn *map_queues; | |
320ae51f JA |
145 | }; |
146 | ||
147 | enum { | |
148 | BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */ | |
149 | BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */ | |
150 | BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */ | |
151 | ||
152 | BLK_MQ_F_SHOULD_MERGE = 1 << 0, | |
8a58d1f1 JA |
153 | BLK_MQ_F_TAG_SHARED = 1 << 1, |
154 | BLK_MQ_F_SG_MERGE = 1 << 2, | |
1b792f2f | 155 | BLK_MQ_F_BLOCKING = 1 << 5, |
d3484991 | 156 | BLK_MQ_F_NO_SCHED = 1 << 6, |
24391c0d SL |
157 | BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, |
158 | BLK_MQ_F_ALLOC_POLICY_BITS = 1, | |
320ae51f | 159 | |
5d12f905 | 160 | BLK_MQ_S_STOPPED = 0, |
0d2602ca | 161 | BLK_MQ_S_TAG_ACTIVE = 1, |
bd166ef1 | 162 | BLK_MQ_S_SCHED_RESTART = 2, |
da55f2cc | 163 | BLK_MQ_S_TAG_WAITING = 3, |
320ae51f | 164 | |
a4391c64 | 165 | BLK_MQ_MAX_DEPTH = 10240, |
506e931f JA |
166 | |
167 | BLK_MQ_CPU_WORK_BATCH = 8, | |
320ae51f | 168 | }; |
24391c0d SL |
169 | #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ |
170 | ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ | |
171 | ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) | |
172 | #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ | |
173 | ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ | |
174 | << BLK_MQ_F_ALLOC_POLICY_START_BIT) | |
320ae51f | 175 | |
24d2f903 | 176 | struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); |
b62c21b7 MS |
177 | struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, |
178 | struct request_queue *q); | |
b21d5b30 MB |
179 | int blk_mq_register_dev(struct device *, struct request_queue *); |
180 | void blk_mq_unregister_dev(struct device *, struct request_queue *); | |
320ae51f | 181 | |
24d2f903 CH |
182 | int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); |
183 | void blk_mq_free_tag_set(struct blk_mq_tag_set *set); | |
184 | ||
320ae51f JA |
185 | void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); |
186 | ||
320ae51f JA |
187 | void blk_mq_free_request(struct request *rq); |
188 | bool blk_mq_can_queue(struct blk_mq_hw_ctx *); | |
6f3b0e8b CH |
189 | |
190 | enum { | |
191 | BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */ | |
192 | BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */ | |
bd166ef1 | 193 | BLK_MQ_REQ_INTERNAL = (1 << 2), /* allocate internal/sched tag */ |
6f3b0e8b CH |
194 | }; |
195 | ||
4ce01dd1 | 196 | struct request *blk_mq_alloc_request(struct request_queue *q, int rw, |
6f3b0e8b | 197 | unsigned int flags); |
1f5bd336 ML |
198 | struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op, |
199 | unsigned int flags, unsigned int hctx_idx); | |
0e62f51f | 200 | struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); |
320ae51f | 201 | |
205fb5f5 BVA |
202 | enum { |
203 | BLK_MQ_UNIQUE_TAG_BITS = 16, | |
204 | BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, | |
205 | }; | |
206 | ||
207 | u32 blk_mq_unique_tag(struct request *rq); | |
208 | ||
209 | static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) | |
210 | { | |
211 | return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; | |
212 | } | |
213 | ||
214 | static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) | |
215 | { | |
216 | return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; | |
217 | } | |
218 | ||
320ae51f | 219 | |
973c0191 | 220 | int blk_mq_request_started(struct request *rq); |
e2490073 | 221 | void blk_mq_start_request(struct request *rq); |
c8a446ad CH |
222 | void blk_mq_end_request(struct request *rq, int error); |
223 | void __blk_mq_end_request(struct request *rq, int error); | |
320ae51f | 224 | |
2b053aca BVA |
225 | void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); |
226 | void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, | |
227 | bool kick_requeue_list); | |
6fca6a61 | 228 | void blk_mq_kick_requeue_list(struct request_queue *q); |
2849450a | 229 | void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); |
1885b24d | 230 | void blk_mq_abort_requeue_list(struct request_queue *q); |
f4829a9b | 231 | void blk_mq_complete_request(struct request *rq, int error); |
30a91cb4 | 232 | |
fd001443 | 233 | bool blk_mq_queue_stopped(struct request_queue *q); |
320ae51f JA |
234 | void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); |
235 | void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); | |
280d45f6 | 236 | void blk_mq_stop_hw_queues(struct request_queue *q); |
2f268556 | 237 | void blk_mq_start_hw_queues(struct request_queue *q); |
ae911c5e | 238 | void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); |
1b4a3258 | 239 | void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); |
b94ec296 | 240 | void blk_mq_run_hw_queues(struct request_queue *q, bool async); |
70f4db63 | 241 | void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); |
e0489487 SG |
242 | void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, |
243 | busy_tag_iter_fn *fn, void *priv); | |
c761d96b | 244 | void blk_mq_freeze_queue(struct request_queue *q); |
b4c6a028 KB |
245 | void blk_mq_unfreeze_queue(struct request_queue *q); |
246 | void blk_mq_freeze_queue_start(struct request_queue *q); | |
6bae363e | 247 | void blk_mq_freeze_queue_wait(struct request_queue *q); |
f91328c4 KB |
248 | int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, |
249 | unsigned long timeout); | |
486cf989 | 250 | int blk_mq_reinit_tagset(struct blk_mq_tag_set *set); |
320ae51f | 251 | |
9e5a7e22 | 252 | int blk_mq_map_queues(struct blk_mq_tag_set *set); |
868f2f0b KB |
253 | void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); |
254 | ||
320ae51f JA |
255 | /* |
256 | * Driver command data is immediately after the request. So subtract request | |
2963e3f7 | 257 | * size to get back to the original request, add request size to get the PDU. |
320ae51f JA |
258 | */ |
259 | static inline struct request *blk_mq_rq_from_pdu(void *pdu) | |
260 | { | |
261 | return pdu - sizeof(struct request); | |
262 | } | |
263 | static inline void *blk_mq_rq_to_pdu(struct request *rq) | |
264 | { | |
2963e3f7 | 265 | return rq + 1; |
320ae51f JA |
266 | } |
267 | ||
320ae51f | 268 | #define queue_for_each_hw_ctx(q, hctx, i) \ |
0d0b7d42 JA |
269 | for ((i) = 0; (i) < (q)->nr_hw_queues && \ |
270 | ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) | |
320ae51f | 271 | |
320ae51f | 272 | #define hctx_for_each_ctx(hctx, ctx, i) \ |
0d0b7d42 JA |
273 | for ((i) = 0; (i) < (hctx)->nr_ctx && \ |
274 | ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) | |
320ae51f | 275 | |
320ae51f | 276 | #endif |