4 #include <linux/blkdev.h>
5 #include <linux/sbitmap.h>
8 struct blk_flush_queue
;
10 struct blk_mq_hw_ctx
{
13 struct list_head dispatch
;
14 unsigned long state
; /* BLK_MQ_S_* flags */
15 } ____cacheline_aligned_in_smp
;
17 struct work_struct run_work
;
18 cpumask_var_t cpumask
;
22 unsigned long flags
; /* BLK_MQ_F_* flags */
24 struct request_queue
*queue
;
25 struct blk_flush_queue
*fq
;
29 struct sbitmap ctx_map
;
31 struct blk_mq_ctx
**ctxs
;
36 struct blk_mq_tags
*tags
;
40 #define BLK_MQ_MAX_DISPATCH_ORDER 7
41 unsigned long dispatched
[BLK_MQ_MAX_DISPATCH_ORDER
];
43 unsigned int numa_node
;
44 unsigned int queue_num
;
48 struct delayed_work delay_work
;
50 struct hlist_node cpuhp_dead
;
53 unsigned long poll_considered
;
54 unsigned long poll_invoked
;
55 unsigned long poll_success
;
58 struct blk_mq_tag_set
{
60 struct blk_mq_ops
*ops
;
61 unsigned int nr_hw_queues
;
62 unsigned int queue_depth
; /* max hw supported */
63 unsigned int reserved_tags
;
64 unsigned int cmd_size
; /* per-request extra data */
67 unsigned int flags
; /* BLK_MQ_F_* */
70 struct blk_mq_tags
**tags
;
72 struct mutex tag_list_lock
;
73 struct list_head tag_list
;
76 struct blk_mq_queue_data
{
78 struct list_head
*list
;
82 typedef int (queue_rq_fn
)(struct blk_mq_hw_ctx
*, const struct blk_mq_queue_data
*);
83 typedef enum blk_eh_timer_return (timeout_fn
)(struct request
*, bool);
84 typedef int (init_hctx_fn
)(struct blk_mq_hw_ctx
*, void *, unsigned int);
85 typedef void (exit_hctx_fn
)(struct blk_mq_hw_ctx
*, unsigned int);
86 typedef int (init_request_fn
)(void *, struct request
*, unsigned int,
87 unsigned int, unsigned int);
88 typedef void (exit_request_fn
)(void *, struct request
*, unsigned int,
90 typedef int (reinit_request_fn
)(void *, struct request
*);
92 typedef void (busy_iter_fn
)(struct blk_mq_hw_ctx
*, struct request
*, void *,
94 typedef void (busy_tag_iter_fn
)(struct request
*, void *, bool);
95 typedef int (poll_fn
)(struct blk_mq_hw_ctx
*, unsigned int);
96 typedef int (map_queues_fn
)(struct blk_mq_tag_set
*set
);
103 queue_rq_fn
*queue_rq
;
106 * Called on request timeout
111 * Called to poll for completion of a specific tag.
115 softirq_done_fn
*complete
;
118 * Called when the block layer side of a hardware queue has been
119 * set up, allowing the driver to allocate/init matching structures.
120 * Ditto for exit/teardown.
122 init_hctx_fn
*init_hctx
;
123 exit_hctx_fn
*exit_hctx
;
126 * Called for every command allocated by the block layer to allow
127 * the driver to set up driver specific data.
129 * Tag greater than or equal to queue_depth is for setting up
132 * Ditto for exit/teardown.
134 init_request_fn
*init_request
;
135 exit_request_fn
*exit_request
;
136 reinit_request_fn
*reinit_request
;
138 map_queues_fn
*map_queues
;
142 BLK_MQ_RQ_QUEUE_OK
= 0, /* queued fine */
143 BLK_MQ_RQ_QUEUE_BUSY
= 1, /* requeue IO for later */
144 BLK_MQ_RQ_QUEUE_ERROR
= 2, /* end IO with error */
146 BLK_MQ_F_SHOULD_MERGE
= 1 << 0,
147 BLK_MQ_F_TAG_SHARED
= 1 << 1,
148 BLK_MQ_F_SG_MERGE
= 1 << 2,
149 BLK_MQ_F_DEFER_ISSUE
= 1 << 4,
150 BLK_MQ_F_BLOCKING
= 1 << 5,
151 BLK_MQ_F_ALLOC_POLICY_START_BIT
= 8,
152 BLK_MQ_F_ALLOC_POLICY_BITS
= 1,
154 BLK_MQ_S_STOPPED
= 0,
155 BLK_MQ_S_TAG_ACTIVE
= 1,
157 BLK_MQ_MAX_DEPTH
= 10240,
159 BLK_MQ_CPU_WORK_BATCH
= 8,
161 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
162 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
163 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
164 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
165 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
166 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
168 struct request_queue
*blk_mq_init_queue(struct blk_mq_tag_set
*);
169 struct request_queue
*blk_mq_init_allocated_queue(struct blk_mq_tag_set
*set
,
170 struct request_queue
*q
);
171 int blk_mq_register_dev(struct device
*, struct request_queue
*);
172 void blk_mq_unregister_dev(struct device
*, struct request_queue
*);
174 int blk_mq_alloc_tag_set(struct blk_mq_tag_set
*set
);
175 void blk_mq_free_tag_set(struct blk_mq_tag_set
*set
);
177 void blk_mq_flush_plug_list(struct blk_plug
*plug
, bool from_schedule
);
179 void blk_mq_insert_request(struct request
*, bool, bool, bool);
180 void blk_mq_free_request(struct request
*rq
);
181 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx
*, struct request
*rq
);
182 bool blk_mq_can_queue(struct blk_mq_hw_ctx
*);
185 BLK_MQ_REQ_NOWAIT
= (1 << 0), /* return when out of requests */
186 BLK_MQ_REQ_RESERVED
= (1 << 1), /* allocate from reserved pool */
189 struct request
*blk_mq_alloc_request(struct request_queue
*q
, int rw
,
191 struct request
*blk_mq_alloc_request_hctx(struct request_queue
*q
, int op
,
192 unsigned int flags
, unsigned int hctx_idx
);
193 struct request
*blk_mq_tag_to_rq(struct blk_mq_tags
*tags
, unsigned int tag
);
196 BLK_MQ_UNIQUE_TAG_BITS
= 16,
197 BLK_MQ_UNIQUE_TAG_MASK
= (1 << BLK_MQ_UNIQUE_TAG_BITS
) - 1,
200 u32
blk_mq_unique_tag(struct request
*rq
);
202 static inline u16
blk_mq_unique_tag_to_hwq(u32 unique_tag
)
204 return unique_tag
>> BLK_MQ_UNIQUE_TAG_BITS
;
207 static inline u16
blk_mq_unique_tag_to_tag(u32 unique_tag
)
209 return unique_tag
& BLK_MQ_UNIQUE_TAG_MASK
;
213 int blk_mq_request_started(struct request
*rq
);
214 void blk_mq_start_request(struct request
*rq
);
215 void blk_mq_end_request(struct request
*rq
, int error
);
216 void __blk_mq_end_request(struct request
*rq
, int error
);
218 void blk_mq_requeue_request(struct request
*rq
);
219 void blk_mq_add_to_requeue_list(struct request
*rq
, bool at_head
);
220 void blk_mq_cancel_requeue_work(struct request_queue
*q
);
221 void blk_mq_kick_requeue_list(struct request_queue
*q
);
222 void blk_mq_delay_kick_requeue_list(struct request_queue
*q
, unsigned long msecs
);
223 void blk_mq_abort_requeue_list(struct request_queue
*q
);
224 void blk_mq_complete_request(struct request
*rq
, int error
);
226 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx
*hctx
);
227 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx
*hctx
);
228 void blk_mq_stop_hw_queues(struct request_queue
*q
);
229 void blk_mq_start_hw_queues(struct request_queue
*q
);
230 void blk_mq_start_stopped_hw_queues(struct request_queue
*q
, bool async
);
231 void blk_mq_run_hw_queues(struct request_queue
*q
, bool async
);
232 void blk_mq_delay_queue(struct blk_mq_hw_ctx
*hctx
, unsigned long msecs
);
233 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set
*tagset
,
234 busy_tag_iter_fn
*fn
, void *priv
);
235 void blk_mq_freeze_queue(struct request_queue
*q
);
236 void blk_mq_unfreeze_queue(struct request_queue
*q
);
237 void blk_mq_freeze_queue_start(struct request_queue
*q
);
238 int blk_mq_reinit_tagset(struct blk_mq_tag_set
*set
);
240 int blk_mq_map_queues(struct blk_mq_tag_set
*set
);
241 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set
*set
, int nr_hw_queues
);
244 * Driver command data is immediately after the request. So subtract request
245 * size to get back to the original request, add request size to get the PDU.
247 static inline struct request
*blk_mq_rq_from_pdu(void *pdu
)
249 return pdu
- sizeof(struct request
);
251 static inline void *blk_mq_rq_to_pdu(struct request
*rq
)
256 #define queue_for_each_hw_ctx(q, hctx, i) \
257 for ((i) = 0; (i) < (q)->nr_hw_queues && \
258 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
260 #define hctx_for_each_ctx(hctx, ctx, i) \
261 for ((i) = 0; (i) < (hctx)->nr_ctx && \
262 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)