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 {
59 struct blk_mq_ops *ops;
60 unsigned int nr_hw_queues;
61 unsigned int queue_depth; /* max hw supported */
62 unsigned int reserved_tags;
63 unsigned int cmd_size; /* per-request extra data */
66 unsigned int flags; /* BLK_MQ_F_* */
69 struct blk_mq_tags **tags;
71 struct mutex tag_list_lock;
72 struct list_head tag_list;
75 struct blk_mq_queue_data {
77 struct list_head *list;
81 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *);
82 typedef struct blk_mq_hw_ctx *(map_queue_fn)(struct request_queue *, const int);
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);
102 queue_rq_fn *queue_rq;
105 * Map to specific hardware queue
107 map_queue_fn *map_queue;
110 * Called on request timeout
115 * Called to poll for completion of a specific tag.
119 softirq_done_fn *complete;
122 * Called when the block layer side of a hardware queue has been
123 * set up, allowing the driver to allocate/init matching structures.
124 * Ditto for exit/teardown.
126 init_hctx_fn *init_hctx;
127 exit_hctx_fn *exit_hctx;
130 * Called for every command allocated by the block layer to allow
131 * the driver to set up driver specific data.
133 * Tag greater than or equal to queue_depth is for setting up
136 * Ditto for exit/teardown.
138 init_request_fn *init_request;
139 exit_request_fn *exit_request;
140 reinit_request_fn *reinit_request;
144 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */
145 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */
146 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */
148 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
149 BLK_MQ_F_TAG_SHARED = 1 << 1,
150 BLK_MQ_F_SG_MERGE = 1 << 2,
151 BLK_MQ_F_DEFER_ISSUE = 1 << 4,
152 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
153 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
155 BLK_MQ_S_STOPPED = 0,
156 BLK_MQ_S_TAG_ACTIVE = 1,
158 BLK_MQ_MAX_DEPTH = 10240,
160 BLK_MQ_CPU_WORK_BATCH = 8,
162 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
163 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
164 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
165 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
166 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
167 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
169 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
170 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
171 struct request_queue *q);
172 int blk_mq_register_dev(struct device *, struct request_queue *);
173 void blk_mq_unregister_dev(struct device *, struct request_queue *);
175 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
176 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
178 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
180 void blk_mq_insert_request(struct request *, bool, bool, bool);
181 void blk_mq_free_request(struct request *rq);
182 void blk_mq_free_hctx_request(struct blk_mq_hw_ctx *, struct request *rq);
183 bool blk_mq_can_queue(struct blk_mq_hw_ctx *);
186 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */
187 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */
190 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
192 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
193 unsigned int flags, unsigned int hctx_idx);
194 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
195 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
198 BLK_MQ_UNIQUE_TAG_BITS = 16,
199 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
202 u32 blk_mq_unique_tag(struct request *rq);
204 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
206 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
209 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
211 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
214 struct blk_mq_hw_ctx *blk_mq_map_queue(struct request_queue *, const int ctx_index);
216 int blk_mq_request_started(struct request *rq);
217 void blk_mq_start_request(struct request *rq);
218 void blk_mq_end_request(struct request *rq, int error);
219 void __blk_mq_end_request(struct request *rq, int error);
221 void blk_mq_requeue_request(struct request *rq);
222 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head);
223 void blk_mq_cancel_requeue_work(struct request_queue *q);
224 void blk_mq_kick_requeue_list(struct request_queue *q);
225 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
226 void blk_mq_abort_requeue_list(struct request_queue *q);
227 void blk_mq_complete_request(struct request *rq, int error);
229 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
230 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
231 void blk_mq_stop_hw_queues(struct request_queue *q);
232 void blk_mq_start_hw_queues(struct request_queue *q);
233 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
234 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
235 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
236 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
237 busy_tag_iter_fn *fn, void *priv);
238 void blk_mq_freeze_queue(struct request_queue *q);
239 void blk_mq_unfreeze_queue(struct request_queue *q);
240 void blk_mq_freeze_queue_start(struct request_queue *q);
241 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set);
243 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
246 * Driver command data is immediately after the request. So subtract request
247 * size to get back to the original request, add request size to get the PDU.
249 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
251 return pdu - sizeof(struct request);
253 static inline void *blk_mq_rq_to_pdu(struct request *rq)
258 #define queue_for_each_hw_ctx(q, hctx, i) \
259 for ((i) = 0; (i) < (q)->nr_hw_queues && \
260 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
262 #define hctx_for_each_ctx(hctx, ctx, i) \
263 for ((i) = 0; (i) < (hctx)->nr_ctx && \
264 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)