blk-mq: improve layout of blk_mq_hw_ctx
[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>
5
6struct blk_mq_tags;
f70ced09 7struct blk_flush_queue;
320ae51f
JA
8
9struct blk_mq_cpu_notifier {
10 struct list_head list;
11 void *data;
e814e71b 12 int (*notify)(void *data, unsigned long action, unsigned int cpu);
320ae51f
JA
13};
14
1429d7c9 15struct blk_mq_ctxmap {
569fd0ce 16 unsigned int size;
1429d7c9
JA
17 unsigned int bits_per_word;
18 struct blk_align_bitmap *map;
19};
20
320ae51f
JA
21struct blk_mq_hw_ctx {
22 struct {
23 spinlock_t lock;
24 struct list_head dispatch;
8d354f13 25 unsigned long state; /* BLK_MQ_S_* flags */
320ae51f
JA
26 } ____cacheline_aligned_in_smp;
27
27489a3c 28 struct work_struct run_work;
e4043dcf 29 cpumask_var_t cpumask;
506e931f
JA
30 int next_cpu;
31 int next_cpu_batch;
320ae51f
JA
32
33 unsigned long flags; /* BLK_MQ_F_* flags */
34
35 struct request_queue *queue;
f70ced09 36 struct blk_flush_queue *fq;
320ae51f
JA
37
38 void *driver_data;
39
1429d7c9
JA
40 struct blk_mq_ctxmap ctx_map;
41
4bb659b1 42 struct blk_mq_ctx **ctxs;
8d354f13 43 unsigned int nr_ctx;
4bb659b1 44
8537b120 45 atomic_t wait_index;
320ae51f 46
320ae51f
JA
47 struct blk_mq_tags *tags;
48
49 unsigned long queued;
50 unsigned long run;
8d354f13 51#define BLK_MQ_MAX_DISPATCH_ORDER 7
320ae51f
JA
52 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
53
320ae51f 54 unsigned int numa_node;
17ded320 55 unsigned int queue_num;
320ae51f 56
0d2602ca
JA
57 atomic_t nr_active;
58
8d354f13
JA
59 struct delayed_work delay_work;
60
320ae51f
JA
61 struct blk_mq_cpu_notifier cpu_notifier;
62 struct kobject kobj;
05229bee
JA
63
64 unsigned long poll_invoked;
65 unsigned long poll_success;
320ae51f
JA
66};
67
24d2f903 68struct blk_mq_tag_set {
320ae51f
JA
69 struct blk_mq_ops *ops;
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;
87 struct list_head *list;
88 bool last;
89};
90
91typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
320ae51f 92typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
0152fb6b 93typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
320ae51f
JA
94typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
95typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
24d2f903
CH
96typedef int (init_request_fn)(void *, struct request *, unsigned int,
97 unsigned int, unsigned int);
98typedef void (exit_request_fn)(void *, struct request *, unsigned int,
99 unsigned int);
486cf989 100typedef int (reinit_request_fn)(void *, struct request *);
320ae51f 101
81481eb4
CH
102typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
103 bool);
f26cdc85 104typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
05229bee
JA
105typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
106
81481eb4 107
320ae51f
JA
108struct blk_mq_ops {
109 /*
110 * Queue request
111 */
112 queue_rq_fn *queue_rq;
113
114 /*
115 * Map to specific hardware queue
116 */
117 map_queue_fn *map_queue;
118
119 /*
120 * Called on request timeout
121 */
0152fb6b 122 timeout_fn *timeout;
320ae51f 123
05229bee
JA
124 /*
125 * Called to poll for completion of a specific tag.
126 */
127 poll_fn *poll;
128
30a91cb4
CH
129 softirq_done_fn *complete;
130
320ae51f
JA
131 /*
132 * Called when the block layer side of a hardware queue has been
133 * set up, allowing the driver to allocate/init matching structures.
134 * Ditto for exit/teardown.
135 */
136 init_hctx_fn *init_hctx;
137 exit_hctx_fn *exit_hctx;
e9b267d9
CH
138
139 /*
140 * Called for every command allocated by the block layer to allow
141 * the driver to set up driver specific data.
f70ced09
ML
142 *
143 * Tag greater than or equal to queue_depth is for setting up
144 * flush request.
145 *
e9b267d9
CH
146 * Ditto for exit/teardown.
147 */
148 init_request_fn *init_request;
149 exit_request_fn *exit_request;
486cf989 150 reinit_request_fn *reinit_request;
320ae51f
JA
151};
152
153enum {
154 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
155 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
156 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
157
158 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
8a58d1f1
JA
159 BLK_MQ_F_TAG_SHARED = 1 << 1,
160 BLK_MQ_F_SG_MERGE = 1 << 2,
e167dfb5 161 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
24391c0d
SL
162 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
163 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
320ae51f 164
5d12f905 165 BLK_MQ_S_STOPPED = 0,
0d2602ca 166 BLK_MQ_S_TAG_ACTIVE = 1,
320ae51f 167
a4391c64 168 BLK_MQ_MAX_DEPTH = 10240,
506e931f
JA
169
170 BLK_MQ_CPU_WORK_BATCH = 8,
320ae51f 171};
24391c0d
SL
172#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
173 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
174 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
175#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
176 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
177 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
320ae51f 178
24d2f903 179struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
b62c21b7
MS
180struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
181 struct request_queue *q);
320ae51f
JA
182int blk_mq_register_disk(struct gendisk *);
183void blk_mq_unregister_disk(struct gendisk *);
320ae51f 184
24d2f903
CH
185int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
186void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
187
320ae51f
JA
188void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
189
feb71dae 190void blk_mq_insert_request(struct request *, bool, bool, bool);
320ae51f 191void blk_mq_free_request(struct request *rq);
7c7f2f2b 192void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
320ae51f 193bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
6f3b0e8b
CH
194
195enum {
196 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
197 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
198};
199
4ce01dd1 200struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
6f3b0e8b 201 unsigned int flags);
1f5bd336
ML
202struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
203 unsigned int flags, unsigned int hctx_idx);
0e62f51f 204struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
f26cdc85 205struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
320ae51f 206
205fb5f5
BVA
207enum {
208 BLK_MQ_UNIQUE_TAG_BITS = 16,
209 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
210};
211
212u32 blk_mq_unique_tag(struct request *rq);
213
214static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
215{
216 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
217}
218
219static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
220{
221 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
222}
223
320ae51f 224struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
f14bbe77 225struct blk_mq_hw_ctx *blk_mq_alloc_single_hw_queue(struct blk_mq_tag_set *, unsigned int, int);
320ae51f 226
973c0191 227int blk_mq_request_started(struct request *rq);
e2490073 228void blk_mq_start_request(struct request *rq);
c8a446ad
CH
229void blk_mq_end_request(struct request *rq, int error);
230void __blk_mq_end_request(struct request *rq, int error);
320ae51f 231
ed0791b2 232void blk_mq_requeue_request(struct request *rq);
6fca6a61 233void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
c68ed59f 234void blk_mq_cancel_requeue_work(struct request_queue *q);
6fca6a61 235void blk_mq_kick_requeue_list(struct request_queue *q);
1885b24d 236void blk_mq_abort_requeue_list(struct request_queue *q);
f4829a9b 237void blk_mq_complete_request(struct request *rq, int error);
30a91cb4 238
320ae51f
JA
239void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
240void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
280d45f6 241void blk_mq_stop_hw_queues(struct request_queue *q);
2f268556 242void blk_mq_start_hw_queues(struct request_queue *q);
1b4a3258 243void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
b94ec296 244void blk_mq_run_hw_queues(struct request_queue *q, bool async);
70f4db63 245void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
e0489487
SG
246void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
247 busy_tag_iter_fn *fn, void *priv);
c761d96b 248void blk_mq_freeze_queue(struct request_queue *q);
b4c6a028
KB
249void blk_mq_unfreeze_queue(struct request_queue *q);
250void blk_mq_freeze_queue_start(struct request_queue *q);
486cf989 251int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
320ae51f 252
868f2f0b
KB
253void 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 */
259static inline struct request *blk_mq_rq_from_pdu(void *pdu)
260{
261 return pdu - sizeof(struct request);
262}
263static 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