4 #include <linux/blkdev.h>
7 struct blk_flush_queue;
9 struct blk_mq_cpu_notifier {
10 struct list_head list;
12 int (*notify)(void *data, unsigned long action, unsigned int cpu);
15 struct blk_mq_ctxmap {
17 unsigned int bits_per_word;
18 struct blk_align_bitmap *map;
21 struct blk_mq_hw_ctx {
24 struct list_head dispatch;
25 unsigned long state; /* BLK_MQ_S_* flags */
26 } ____cacheline_aligned_in_smp;
28 struct work_struct run_work;
29 cpumask_var_t cpumask;
33 unsigned long flags; /* BLK_MQ_F_* flags */
35 struct request_queue *queue;
36 struct blk_flush_queue *fq;
40 struct blk_mq_ctxmap ctx_map;
42 struct blk_mq_ctx **ctxs;
47 struct blk_mq_tags *tags;
51 #define BLK_MQ_MAX_DISPATCH_ORDER 7
52 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
54 unsigned int numa_node;
55 unsigned int queue_num;
59 struct delayed_work delay_work;
61 struct blk_mq_cpu_notifier cpu_notifier;
64 unsigned long poll_considered;
65 unsigned long poll_invoked;
66 unsigned long poll_success;
69 struct blk_mq_tag_set {
71 struct blk_mq_ops *ops;
72 unsigned int nr_hw_queues;
73 unsigned int queue_depth; /* max hw supported */
74 unsigned int reserved_tags;
75 unsigned int cmd_size; /* per-request extra data */
78 unsigned int flags; /* BLK_MQ_F_* */
81 struct blk_mq_tags **tags;
83 struct mutex tag_list_lock;
84 struct list_head tag_list;
87 struct blk_mq_queue_data {
89 struct list_head *list;
93 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
94 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool);
95 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int);
96 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int);
97 typedef int (init_request_fn)(void *, struct request *, unsigned int,
98 unsigned int, unsigned int);
99 typedef void (exit_request_fn)(void *, struct request *, unsigned int,
101 typedef int (reinit_request_fn)(void *, struct request *);
103 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
105 typedef void (busy_tag_iter_fn)(struct request *, void *, bool);
106 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int);
113 queue_rq_fn *queue_rq;
116 * Called on request timeout
121 * Called to poll for completion of a specific tag.
125 softirq_done_fn *complete;
128 * Called when the block layer side of a hardware queue has been
129 * set up, allowing the driver to allocate/init matching structures.
130 * Ditto for exit/teardown.
132 init_hctx_fn *init_hctx;
133 exit_hctx_fn *exit_hctx;
136 * Called for every command allocated by the block layer to allow
137 * the driver to set up driver specific data.
139 * Tag greater than or equal to queue_depth is for setting up
142 * Ditto for exit/teardown.
144 init_request_fn *init_request;
145 exit_request_fn *exit_request;
146 reinit_request_fn *reinit_request;
150 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
151 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
152 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
154 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
155 BLK_MQ_F_TAG_SHARED = 1 << 1,
156 BLK_MQ_F_SG_MERGE = 1 << 2,
157 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
158 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
159 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
161 BLK_MQ_S_STOPPED = 0,
162 BLK_MQ_S_TAG_ACTIVE = 1,
164 BLK_MQ_MAX_DEPTH = 10240,
166 BLK_MQ_CPU_WORK_BATCH = 8,
168 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
169 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
170 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
171 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
172 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
173 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
175 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
176 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
177 struct request_queue *q);
178 int blk_mq_register_disk(struct gendisk *);
179 void blk_mq_unregister_disk(struct gendisk *);
181 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
182 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
184 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
186 void blk_mq_insert_request(struct request *, bool, bool, bool);
187 void blk_mq_free_request(struct request *rq);
188 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
189 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
192 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
193 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
196 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
198 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
199 unsigned int flags, unsigned int hctx_idx);
200 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
201 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
204 BLK_MQ_UNIQUE_TAG_BITS = 16,
205 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
208 u32 blk_mq_unique_tag(struct request *rq);
210 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
212 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
215 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
217 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
221 int blk_mq_request_started(struct request *rq);
222 void blk_mq_start_request(struct request *rq);
223 void blk_mq_end_request(struct request *rq, int error);
224 void __blk_mq_end_request(struct request *rq, int error);
226 void blk_mq_requeue_request(struct request *rq);
227 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
228 void blk_mq_cancel_requeue_work(struct request_queue *q);
229 void blk_mq_kick_requeue_list(struct request_queue *q);
230 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
231 void blk_mq_abort_requeue_list(struct request_queue *q);
232 void blk_mq_complete_request(struct request *rq, int error);
234 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
235 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
236 void blk_mq_stop_hw_queues(struct request_queue *q);
237 void blk_mq_start_hw_queues(struct request_queue *q);
238 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
239 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
240 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
241 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
242 busy_tag_iter_fn *fn, void *priv);
243 void blk_mq_freeze_queue(struct request_queue *q);
244 void blk_mq_unfreeze_queue(struct request_queue *q);
245 void blk_mq_freeze_queue_start(struct request_queue *q);
246 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
248 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
251 * Driver command data is immediately after the request. So subtract request
252 * size to get back to the original request, add request size to get the PDU.
254 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
256 return pdu - sizeof(struct request);
258 static inline void *blk_mq_rq_to_pdu(struct request *rq)
263 #define queue_for_each_hw_ctx(q, hctx, i) \
264 for ((i) = 0; (i) < (q)->nr_hw_queues && \
265 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
267 #define hctx_for_each_ctx(hctx, ctx, i) \
268 for ((i) = 0; (i) < (hctx)->nr_ctx && \
269 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)