Merge branch 'utilities' of git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux
[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
90typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
0152fb6b 91typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
320ae51f
JA
92typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
93typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
d6296d39 94typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *,
24d2f903 95 unsigned int, unsigned int);
d6296d39 96typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *,
24d2f903 97 unsigned int);
486cf989 98typedef int (reinit_request_fn)(void *, struct request *);
320ae51f 99
81481eb4
CH
100typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
101 bool);
f26cdc85 102typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
05229bee 103typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
da695ba2 104typedef int (map_queues_fn)(struct blk_mq_tag_set *set);
05229bee 105
81481eb4 106
320ae51f
JA
107struct blk_mq_ops {
108 /*
109 * Queue request
110 */
111 queue_rq_fn *queue_rq;
112
320ae51f
JA
113 /*
114 * Called on request timeout
115 */
0152fb6b 116 timeout_fn *timeout;
320ae51f 117
05229bee
JA
118 /*
119 * Called to poll for completion of a specific tag.
120 */
121 poll_fn *poll;
122
30a91cb4
CH
123 softirq_done_fn *complete;
124
320ae51f
JA
125 /*
126 * Called when the block layer side of a hardware queue has been
127 * set up, allowing the driver to allocate/init matching structures.
128 * Ditto for exit/teardown.
129 */
130 init_hctx_fn *init_hctx;
131 exit_hctx_fn *exit_hctx;
e9b267d9
CH
132
133 /*
134 * Called for every command allocated by the block layer to allow
135 * the driver to set up driver specific data.
f70ced09
ML
136 *
137 * Tag greater than or equal to queue_depth is for setting up
138 * flush request.
139 *
e9b267d9
CH
140 * Ditto for exit/teardown.
141 */
142 init_request_fn *init_request;
143 exit_request_fn *exit_request;
486cf989 144 reinit_request_fn *reinit_request;
da695ba2
CH
145
146 map_queues_fn *map_queues;
2836ee4b
BVA
147
148#ifdef CONFIG_BLK_DEBUG_FS
149 /*
150 * Used by the debugfs implementation to show driver-specific
151 * information about a request.
152 */
153 void (*show_rq)(struct seq_file *m, struct request *rq);
154#endif
320ae51f
JA
155};
156
157enum {
158 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
159 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
160 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
161
162 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
8a58d1f1
JA
163 BLK_MQ_F_TAG_SHARED = 1 << 1,
164 BLK_MQ_F_SG_MERGE = 1 << 2,
1b792f2f 165 BLK_MQ_F_BLOCKING = 1 << 5,
d3484991 166 BLK_MQ_F_NO_SCHED = 1 << 6,
24391c0d
SL
167 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
168 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
320ae51f 169
5d12f905 170 BLK_MQ_S_STOPPED = 0,
0d2602ca 171 BLK_MQ_S_TAG_ACTIVE = 1,
bd166ef1 172 BLK_MQ_S_SCHED_RESTART = 2,
da55f2cc 173 BLK_MQ_S_TAG_WAITING = 3,
21c6e939 174 BLK_MQ_S_START_ON_RUN = 4,
320ae51f 175
a4391c64 176 BLK_MQ_MAX_DEPTH = 10240,
506e931f
JA
177
178 BLK_MQ_CPU_WORK_BATCH = 8,
320ae51f 179};
24391c0d
SL
180#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
181 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
182 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
183#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
184 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
185 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
320ae51f 186
24d2f903 187struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
b62c21b7
MS
188struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
189 struct request_queue *q);
b21d5b30
MB
190int blk_mq_register_dev(struct device *, struct request_queue *);
191void blk_mq_unregister_dev(struct device *, struct request_queue *);
320ae51f 192
24d2f903
CH
193int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
194void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
195
320ae51f
JA
196void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
197
320ae51f
JA
198void blk_mq_free_request(struct request *rq);
199bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
6f3b0e8b
CH
200
201enum {
202 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
203 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
bd166ef1 204 BLK_MQ_REQ_INTERNAL = (1 << 2), /* allocate internal/sched tag */
6f3b0e8b
CH
205};
206
4ce01dd1 207struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
6f3b0e8b 208 unsigned int flags);
1f5bd336
ML
209struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
210 unsigned int flags, unsigned int hctx_idx);
0e62f51f 211struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
320ae51f 212
205fb5f5
BVA
213enum {
214 BLK_MQ_UNIQUE_TAG_BITS = 16,
215 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
216};
217
218u32 blk_mq_unique_tag(struct request *rq);
219
220static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
221{
222 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
223}
224
225static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
226{
227 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
228}
229
320ae51f 230
973c0191 231int blk_mq_request_started(struct request *rq);
e2490073 232void blk_mq_start_request(struct request *rq);
c8a446ad
CH
233void blk_mq_end_request(struct request *rq, int error);
234void __blk_mq_end_request(struct request *rq, int error);
320ae51f 235
2b053aca
BVA
236void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
237void blk_mq_add_to_requeue_list(struct request *rq, bool at_head,
238 bool kick_requeue_list);
6fca6a61 239void blk_mq_kick_requeue_list(struct request_queue *q);
2849450a 240void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
1885b24d 241void blk_mq_abort_requeue_list(struct request_queue *q);
08e0029a 242void blk_mq_complete_request(struct request *rq);
30a91cb4 243
fd001443 244bool blk_mq_queue_stopped(struct request_queue *q);
320ae51f
JA
245void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
246void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
280d45f6 247void blk_mq_stop_hw_queues(struct request_queue *q);
2f268556 248void blk_mq_start_hw_queues(struct request_queue *q);
ae911c5e 249void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
1b4a3258 250void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
7587a5ae 251void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
5b727272 252void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
b94ec296 253void blk_mq_run_hw_queues(struct request_queue *q, bool async);
70f4db63 254void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
e0489487
SG
255void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
256 busy_tag_iter_fn *fn, void *priv);
c761d96b 257void blk_mq_freeze_queue(struct request_queue *q);
b4c6a028 258void blk_mq_unfreeze_queue(struct request_queue *q);
1671d522 259void blk_freeze_queue_start(struct request_queue *q);
6bae363e 260void blk_mq_freeze_queue_wait(struct request_queue *q);
f91328c4
KB
261int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
262 unsigned long timeout);
486cf989 263int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
320ae51f 264
9e5a7e22 265int blk_mq_map_queues(struct blk_mq_tag_set *set);
868f2f0b
KB
266void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
267
320ae51f
JA
268/*
269 * Driver command data is immediately after the request. So subtract request
2963e3f7 270 * size to get back to the original request, add request size to get the PDU.
320ae51f
JA
271 */
272static inline struct request *blk_mq_rq_from_pdu(void *pdu)
273{
274 return pdu - sizeof(struct request);
275}
276static inline void *blk_mq_rq_to_pdu(struct request *rq)
277{
2963e3f7 278 return rq + 1;
320ae51f
JA
279}
280
320ae51f 281#define queue_for_each_hw_ctx(q, hctx, i) \
0d0b7d42
JA
282 for ((i) = 0; (i) < (q)->nr_hw_queues && \
283 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
320ae51f 284
320ae51f 285#define hctx_for_each_ctx(hctx, ctx, i) \
0d0b7d42
JA
286 for ((i) = 0; (i) < (hctx)->nr_ctx && \
287 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
320ae51f 288
320ae51f 289#endif