block: stop using bio_data() in blk_write_same_mergeable
[linux-2.6-block.git] / include / linux / blk-mq.h
CommitLineData
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
8struct blk_mq_tags;
f70ced09 9struct blk_flush_queue;
320ae51f 10
320ae51f
JA
11struct 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
9f993737 18 struct delayed_work 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
9467f859 54 struct hlist_node cpuhp_dead;
320ae51f 55 struct kobject kobj;
05229bee 56
6e219353 57 unsigned long poll_considered;
05229bee
JA
58 unsigned long poll_invoked;
59 unsigned long poll_success;
9c1051aa
OS
60
61#ifdef CONFIG_BLK_DEBUG_FS
62 struct dentry *debugfs_dir;
d332ce09 63 struct dentry *sched_debugfs_dir;
9c1051aa 64#endif
320ae51f
JA
65};
66
24d2f903 67struct blk_mq_tag_set {
bdd17e75 68 unsigned int *mq_map;
f8a5b122 69 const struct blk_mq_ops *ops;
320ae51f 70 unsigned int nr_hw_queues;
e3a2b3f9 71 unsigned int queue_depth; /* max hw supported */
320ae51f
JA
72 unsigned int reserved_tags;
73 unsigned int cmd_size; /* per-request extra data */
74 int numa_node;
75 unsigned int timeout;
76 unsigned int flags; /* BLK_MQ_F_* */
24d2f903
CH
77 void *driver_data;
78
79 struct blk_mq_tags **tags;
0d2602ca
JA
80
81 struct mutex tag_list_lock;
82 struct list_head tag_list;
320ae51f
JA
83};
84
74c45052
JA
85struct blk_mq_queue_data {
86 struct request *rq;
74c45052
JA
87 bool last;
88};
89
fc17b653
CH
90typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *,
91 const struct blk_mq_queue_data *);
0152fb6b 92typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
320ae51f
JA
93typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
94typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
d6296d39 95typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
24d2f903 96 unsigned int, unsigned int);
d6296d39 97typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
24d2f903 98 unsigned int);
486cf989 99typedef int (reinit_request_fn)(void *, struct request *);
320ae51f 100
81481eb4
CH
101typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
102 bool);
f26cdc85 103typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
05229bee 104typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
da695ba2 105typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
05229bee 106
81481eb4 107
320ae51f
JA
108struct blk_mq_ops {
109 /*
110 * Queue request
111 */
112 queue_rq_fn *queue_rq;
113
320ae51f
JA
114 /*
115 * Called on request timeout
116 */
0152fb6b 117 timeout_fn *timeout;
320ae51f 118
05229bee
JA
119 /*
120 * Called to poll for completion of a specific tag.
121 */
122 poll_fn *poll;
123
30a91cb4
CH
124 softirq_done_fn *complete;
125
320ae51f
JA
126 /*
127 * Called when the block layer side of a hardware queue has been
128 * set up, allowing the driver to allocate/init matching structures.
129 * Ditto for exit/teardown.
130 */
131 init_hctx_fn *init_hctx;
132 exit_hctx_fn *exit_hctx;
e9b267d9
CH
133
134 /*
135 * Called for every command allocated by the block layer to allow
136 * the driver to set up driver specific data.
f70ced09
ML
137 *
138 * Tag greater than or equal to queue_depth is for setting up
139 * flush request.
140 *
e9b267d9
CH
141 * Ditto for exit/teardown.
142 */
143 init_request_fn *init_request;
144 exit_request_fn *exit_request;
486cf989 145 reinit_request_fn *reinit_request;
da695ba2
CH
146
147 map_queues_fn *map_queues;
2836ee4b
BVA
148
149#ifdef CONFIG_BLK_DEBUG_FS
150 /*
151 * Used by the debugfs implementation to show driver-specific
152 * information about a request.
153 */
154 void (*show_rq)(struct seq_file *m, struct request *rq);
155#endif
320ae51f
JA
156};
157
158enum {
320ae51f 159 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
8a58d1f1
JA
160 BLK_MQ_F_TAG_SHARED = 1 << 1,
161 BLK_MQ_F_SG_MERGE = 1 << 2,
1b792f2f 162 BLK_MQ_F_BLOCKING = 1 << 5,
d3484991 163 BLK_MQ_F_NO_SCHED = 1 << 6,
24391c0d
SL
164 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
165 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
320ae51f 166
5d12f905 167 BLK_MQ_S_STOPPED = 0,
0d2602ca 168 BLK_MQ_S_TAG_ACTIVE = 1,
bd166ef1 169 BLK_MQ_S_SCHED_RESTART = 2,
da55f2cc 170 BLK_MQ_S_TAG_WAITING = 3,
21c6e939 171 BLK_MQ_S_START_ON_RUN = 4,
320ae51f 172
a4391c64 173 BLK_MQ_MAX_DEPTH = 10240,
506e931f
JA
174
175 BLK_MQ_CPU_WORK_BATCH = 8,
320ae51f 176};
24391c0d
SL
177#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
178 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
179 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
180#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
181 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
182 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
320ae51f 183
24d2f903 184struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
b62c21b7
MS
185struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
186 struct request_queue *q);
b21d5b30
MB
187int blk_mq_register_dev(struct device *, struct request_queue *);
188void blk_mq_unregister_dev(struct device *, struct request_queue *);
320ae51f 189
24d2f903
CH
190int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
191void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
192
320ae51f
JA
193void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
194
320ae51f
JA
195void blk_mq_free_request(struct request *rq);
196bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
6f3b0e8b
CH
197
198enum {
199 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
200 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
bd166ef1 201 BLK_MQ_REQ_INTERNAL = (1 << 2), /* allocate internal/sched tag */
6f3b0e8b
CH
202};
203
4ce01dd1 204struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
6f3b0e8b 205 unsigned int flags);
1f5bd336
ML
206struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
207 unsigned int flags, unsigned int hctx_idx);
0e62f51f 208struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
320ae51f 209
205fb5f5
BVA
210enum {
211 BLK_MQ_UNIQUE_TAG_BITS = 16,
212 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
213};
214
215u32 blk_mq_unique_tag(struct request *rq);
216
217static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
218{
219 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
220}
221
222static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
223{
224 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
225}
226
320ae51f 227
973c0191 228int blk_mq_request_started(struct request *rq);
e2490073 229void blk_mq_start_request(struct request *rq);
2a842aca
CH
230void blk_mq_end_request(struct request *rq, blk_status_t error);
231void __blk_mq_end_request(struct request *rq, blk_status_t error);
320ae51f 232
2b053aca
BVA
233void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
234void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
235 bool kick_requeue_list);
6fca6a61 236void blk_mq_kick_requeue_list(struct request_queue *q);
2849450a 237void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
08e0029a 238void blk_mq_complete_request(struct request *rq);
30a91cb4 239
fd001443 240bool blk_mq_queue_stopped(struct request_queue *q);
320ae51f
JA
241void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
242void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
280d45f6 243void blk_mq_stop_hw_queues(struct request_queue *q);
2f268556 244void blk_mq_start_hw_queues(struct request_queue *q);
ae911c5e 245void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
1b4a3258 246void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
97e01209 247void blk_mq_quiesce_queue(struct request_queue *q);
e4e73913 248void blk_mq_unquiesce_queue(struct request_queue *q);
7587a5ae 249void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
5b727272 250void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
b94ec296 251void blk_mq_run_hw_queues(struct request_queue *q, bool async);
70f4db63 252void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
e0489487
SG
253void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
254 busy_tag_iter_fn *fn, void *priv);
c761d96b 255void blk_mq_freeze_queue(struct request_queue *q);
b4c6a028 256void blk_mq_unfreeze_queue(struct request_queue *q);
1671d522 257void blk_freeze_queue_start(struct request_queue *q);
6bae363e 258void blk_mq_freeze_queue_wait(struct request_queue *q);
f91328c4
KB
259int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
260 unsigned long timeout);
486cf989 261int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
320ae51f 262
9e5a7e22 263int blk_mq_map_queues(struct blk_mq_tag_set *set);
868f2f0b
KB
264void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
265
4f084b41
ML
266/*
267 * FIXME: this helper is just for working around mpt3sas.
268 */
269static inline void blk_mq_quiesce_queue_nowait(struct request_queue *q)
270{
f4560ffe
ML
271 spin_lock_irq(q->queue_lock);
272 queue_flag_set(QUEUE_FLAG_QUIESCED, q);
273 spin_unlock_irq(q->queue_lock);
4f084b41
ML
274}
275
320ae51f
JA
276/*
277 * Driver command data is immediately after the request. So subtract request
2963e3f7 278 * size to get back to the original request, add request size to get the PDU.
320ae51f
JA
279 */
280static inline struct request *blk_mq_rq_from_pdu(void *pdu)
281{
282 return pdu - sizeof(struct request);
283}
284static inline void *blk_mq_rq_to_pdu(struct request *rq)
285{
2963e3f7 286 return rq + 1;
320ae51f
JA
287}
288
320ae51f 289#define queue_for_each_hw_ctx(q, hctx, i) \
0d0b7d42
JA
290 for ((i) = 0; (i) < (q)->nr_hw_queues && \
291 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
320ae51f 292
320ae51f 293#define hctx_for_each_ctx(hctx, ctx, i) \
0d0b7d42
JA
294 for ((i) = 0; (i) < (hctx)->nr_ctx && \
295 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
320ae51f 296
320ae51f 297#endif