4 #include <linux/blkdev.h>
5 #include <linux/sbitmap.h>
6 #include <linux/srcu.h>
9 struct blk_flush_queue
;
11 struct blk_mq_hw_ctx
{
14 struct list_head dispatch
;
15 unsigned long state
; /* BLK_MQ_S_* flags */
16 } ____cacheline_aligned_in_smp
;
18 struct work_struct run_work
;
19 cpumask_var_t cpumask
;
23 unsigned long flags
; /* BLK_MQ_F_* flags */
26 struct request_queue
*queue
;
27 struct blk_flush_queue
*fq
;
31 struct sbitmap ctx_map
;
33 struct blk_mq_ctx
**ctxs
;
36 wait_queue_t dispatch_wait
;
39 struct blk_mq_tags
*tags
;
40 struct blk_mq_tags
*sched_tags
;
42 struct srcu_struct queue_rq_srcu
;
46 #define BLK_MQ_MAX_DISPATCH_ORDER 7
47 unsigned long dispatched
[BLK_MQ_MAX_DISPATCH_ORDER
];
49 unsigned int numa_node
;
50 unsigned int queue_num
;
54 struct delayed_work delayed_run_work
;
55 struct delayed_work delay_work
;
57 struct hlist_node cpuhp_dead
;
60 unsigned long poll_considered
;
61 unsigned long poll_invoked
;
62 unsigned long poll_success
;
65 struct blk_mq_tag_set
{
67 const struct blk_mq_ops
*ops
;
68 unsigned int nr_hw_queues
;
69 unsigned int queue_depth
; /* max hw supported */
70 unsigned int reserved_tags
;
71 unsigned int cmd_size
; /* per-request extra data */
74 unsigned int flags
; /* BLK_MQ_F_* */
77 struct blk_mq_tags
**tags
;
79 struct mutex tag_list_lock
;
80 struct list_head tag_list
;
83 struct blk_mq_queue_data
{
85 struct list_head
*list
;
89 typedef int (queue_rq_fn
)(struct blk_mq_hw_ctx
*, const struct blk_mq_queue_data
*);
90 typedef enum blk_eh_timer_return (timeout_fn
)(struct request
*, bool);
91 typedef int (init_hctx_fn
)(struct blk_mq_hw_ctx
*, void *, unsigned int);
92 typedef void (exit_hctx_fn
)(struct blk_mq_hw_ctx
*, unsigned int);
93 typedef int (init_request_fn
)(void *, struct request
*, unsigned int,
94 unsigned int, unsigned int);
95 typedef void (exit_request_fn
)(void *, struct request
*, unsigned int,
97 typedef int (reinit_request_fn
)(void *, struct request
*);
99 typedef void (busy_iter_fn
)(struct blk_mq_hw_ctx
*, struct request
*, void *,
101 typedef void (busy_tag_iter_fn
)(struct request
*, void *, bool);
102 typedef int (poll_fn
)(struct blk_mq_hw_ctx
*, unsigned int);
103 typedef int (map_queues_fn
)(struct blk_mq_tag_set
*set
);
110 queue_rq_fn
*queue_rq
;
113 * Called on request timeout
118 * Called to poll for completion of a specific tag.
122 softirq_done_fn
*complete
;
125 * Called when the block layer side of a hardware queue has been
126 * set up, allowing the driver to allocate/init matching structures.
127 * Ditto for exit/teardown.
129 init_hctx_fn
*init_hctx
;
130 exit_hctx_fn
*exit_hctx
;
133 * Called for every command allocated by the block layer to allow
134 * the driver to set up driver specific data.
136 * Tag greater than or equal to queue_depth is for setting up
139 * Ditto for exit/teardown.
141 init_request_fn
*init_request
;
142 exit_request_fn
*exit_request
;
143 reinit_request_fn
*reinit_request
;
145 map_queues_fn
*map_queues
;
149 BLK_MQ_RQ_QUEUE_OK
= 0, /* queued fine */
150 BLK_MQ_RQ_QUEUE_BUSY
= 1, /* requeue IO for later */
151 BLK_MQ_RQ_QUEUE_ERROR
= 2, /* end IO with error */
153 BLK_MQ_F_SHOULD_MERGE
= 1 << 0,
154 BLK_MQ_F_TAG_SHARED
= 1 << 1,
155 BLK_MQ_F_SG_MERGE
= 1 << 2,
156 BLK_MQ_F_DEFER_ISSUE
= 1 << 4,
157 BLK_MQ_F_BLOCKING
= 1 << 5,
158 BLK_MQ_F_NO_SCHED
= 1 << 6,
159 BLK_MQ_F_ALLOC_POLICY_START_BIT
= 8,
160 BLK_MQ_F_ALLOC_POLICY_BITS
= 1,
162 BLK_MQ_S_STOPPED
= 0,
163 BLK_MQ_S_TAG_ACTIVE
= 1,
164 BLK_MQ_S_SCHED_RESTART
= 2,
165 BLK_MQ_S_TAG_WAITING
= 3,
167 BLK_MQ_MAX_DEPTH
= 10240,
169 BLK_MQ_CPU_WORK_BATCH
= 8,
171 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
172 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
173 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
174 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
175 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
176 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
178 struct request_queue
*blk_mq_init_queue(struct blk_mq_tag_set
*);
179 struct request_queue
*blk_mq_init_allocated_queue(struct blk_mq_tag_set
*set
,
180 struct request_queue
*q
);
181 int blk_mq_register_dev(struct device
*, struct request_queue
*);
182 void blk_mq_unregister_dev(struct device
*, struct request_queue
*);
184 int blk_mq_alloc_tag_set(struct blk_mq_tag_set
*set
);
185 void blk_mq_free_tag_set(struct blk_mq_tag_set
*set
);
187 void blk_mq_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
);
189 void blk_mq_free_request(struct request
*rq
);
190 bool blk_mq_can_queue(struct blk_mq_hw_ctx
*);
193 BLK_MQ_REQ_NOWAIT
= (1 << 0), /* return when out of requests */
194 BLK_MQ_REQ_RESERVED
= (1 << 1), /* allocate from reserved pool */
195 BLK_MQ_REQ_INTERNAL
= (1 << 2), /* allocate internal/sched tag */
198 struct request
*blk_mq_alloc_request(struct request_queue
*q
, int rw
,
200 struct request
*blk_mq_alloc_request_hctx(struct request_queue
*q
, int op
,
201 unsigned int flags
, unsigned int hctx_idx
);
202 struct request
*blk_mq_tag_to_rq(struct blk_mq_tags
*tags
, unsigned int tag
);
205 BLK_MQ_UNIQUE_TAG_BITS
= 16,
206 BLK_MQ_UNIQUE_TAG_MASK
= (1 << BLK_MQ_UNIQUE_TAG_BITS
) - 1,
209 u32
blk_mq_unique_tag(struct request
*rq
);
211 static inline u16
blk_mq_unique_tag_to_hwq(u32 unique_tag
)
213 return unique_tag
>> BLK_MQ_UNIQUE_TAG_BITS
;
216 static inline u16
blk_mq_unique_tag_to_tag(u32 unique_tag
)
218 return unique_tag
& BLK_MQ_UNIQUE_TAG_MASK
;
222 int blk_mq_request_started(struct request
*rq
);
223 void blk_mq_start_request(struct request
*rq
);
224 void blk_mq_end_request(struct request
*rq
, int error
);
225 void __blk_mq_end_request(struct request
*rq
, int error
);
227 void blk_mq_requeue_request(struct request
*rq
, bool kick_requeue_list
);
228 void blk_mq_add_to_requeue_list(struct request
*rq
, bool at_head
,
229 bool kick_requeue_list
);
230 void blk_mq_kick_requeue_list(struct request_queue
*q
);
231 void blk_mq_delay_kick_requeue_list(struct request_queue
*q
, unsigned long msecs
);
232 void blk_mq_abort_requeue_list(struct request_queue
*q
);
233 void blk_mq_complete_request(struct request
*rq
, int error
);
235 bool blk_mq_queue_stopped(struct request_queue
*q
);
236 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx
*hctx
);
237 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx
*hctx
);
238 void blk_mq_stop_hw_queues(struct request_queue
*q
);
239 void blk_mq_start_hw_queues(struct request_queue
*q
);
240 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx
*hctx
, bool async
);
241 void blk_mq_start_stopped_hw_queues(struct request_queue
*q
, bool async
);
242 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx
*hctx
, unsigned long msecs
);
243 void blk_mq_run_hw_queues(struct request_queue
*q
, bool async
);
244 void blk_mq_delay_queue(struct blk_mq_hw_ctx
*hctx
, unsigned long msecs
);
245 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set
*tagset
,
246 busy_tag_iter_fn
*fn
, void *priv
);
247 void blk_mq_freeze_queue(struct request_queue
*q
);
248 void blk_mq_unfreeze_queue(struct request_queue
*q
);
249 void blk_mq_freeze_queue_start(struct request_queue
*q
);
250 void blk_mq_freeze_queue_wait(struct request_queue
*q
);
251 int blk_mq_freeze_queue_wait_timeout(struct request_queue
*q
,
252 unsigned long timeout
);
253 int blk_mq_reinit_tagset(struct blk_mq_tag_set
*set
);
255 int blk_mq_map_queues(struct blk_mq_tag_set
*set
);
256 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set
*set
, int nr_hw_queues
);
259 * Driver command data is immediately after the request. So subtract request
260 * size to get back to the original request, add request size to get the PDU.
262 static inline struct request
*blk_mq_rq_from_pdu(void *pdu
)
264 return pdu
- sizeof(struct request
);
266 static inline void *blk_mq_rq_to_pdu(struct request
*rq
)
271 #define queue_for_each_hw_ctx(q, hctx, i) \
272 for ((i) = 0; (i) < (q)->nr_hw_queues && \
273 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
275 #define hctx_for_each_ctx(hctx, ctx, i) \
276 for ((i) = 0; (i) < (hctx)->nr_ctx && \
277 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)