1 #ifndef BLK_MQ_H 2 #define BLK_MQ_H 3 4 #include <linux/blkdev.h> 5 #include <linux/sbitmap.h> 6 #include <linux/srcu.h> 7 8 struct blk_mq_tags; 9 struct blk_flush_queue; 10 11 struct blk_mq_hw_ctx { 12 struct { 13 spinlock_t lock; 14 struct list_head dispatch; 15 unsigned long state; /* BLK_MQ_S_* flags */ 16 } ____cacheline_aligned_in_smp; 17 18 struct delayed_work run_work; 19 cpumask_var_t cpumask; 20 int next_cpu; 21 int next_cpu_batch; 22 23 unsigned long flags; /* BLK_MQ_F_* flags */ 24 25 void *sched_data; 26 struct request_queue *queue; 27 struct blk_flush_queue *fq; 28 29 void *driver_data; 30 31 struct sbitmap ctx_map; 32 33 struct blk_mq_ctx **ctxs; 34 unsigned int nr_ctx; 35 36 wait_queue_t dispatch_wait; 37 atomic_t wait_index; 38 39 struct blk_mq_tags *tags; 40 struct blk_mq_tags *sched_tags; 41 42 struct srcu_struct queue_rq_srcu; 43 44 unsigned long queued; 45 unsigned long run; 46 #define BLK_MQ_MAX_DISPATCH_ORDER 7 47 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; 48 49 unsigned int numa_node; 50 unsigned int queue_num; 51 52 atomic_t nr_active; 53 54 struct hlist_node cpuhp_dead; 55 struct kobject kobj; 56 57 unsigned long poll_considered; 58 unsigned long poll_invoked; 59 unsigned long poll_success; 60 }; 61 62 struct blk_mq_tag_set { 63 unsigned int *mq_map; 64 const struct blk_mq_ops *ops; 65 unsigned int nr_hw_queues; 66 unsigned int queue_depth; /* max hw supported */ 67 unsigned int reserved_tags; 68 unsigned int cmd_size; /* per-request extra data */ 69 int numa_node; 70 unsigned int timeout; 71 unsigned int flags; /* BLK_MQ_F_* */ 72 void *driver_data; 73 74 struct blk_mq_tags **tags; 75 76 struct mutex tag_list_lock; 77 struct list_head tag_list; 78 }; 79 80 struct blk_mq_queue_data { 81 struct request *rq; 82 bool last; 83 }; 84 85 typedef int (queue_rq_fn)(struct blk_mq_hw_ctx *, const struct blk_mq_queue_data *); 86 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool); 87 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); 88 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); 89 typedef int (init_request_fn)(void *, struct request *, unsigned int, 90 unsigned int, unsigned int); 91 typedef void (exit_request_fn)(void *, struct request *, unsigned int, 92 unsigned int); 93 typedef int (reinit_request_fn)(void *, struct request *); 94 95 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, 96 bool); 97 typedef void (busy_tag_iter_fn)(struct request *, void *, bool); 98 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int); 99 typedef int (map_queues_fn)(struct blk_mq_tag_set *set); 100 101 102 struct blk_mq_ops { 103 /* 104 * Queue request 105 */ 106 queue_rq_fn *queue_rq; 107 108 /* 109 * Called on request timeout 110 */ 111 timeout_fn *timeout; 112 113 /* 114 * Called to poll for completion of a specific tag. 115 */ 116 poll_fn *poll; 117 118 softirq_done_fn *complete; 119 120 /* 121 * Called when the block layer side of a hardware queue has been 122 * set up, allowing the driver to allocate/init matching structures. 123 * Ditto for exit/teardown. 124 */ 125 init_hctx_fn *init_hctx; 126 exit_hctx_fn *exit_hctx; 127 128 /* 129 * Called for every command allocated by the block layer to allow 130 * the driver to set up driver specific data. 131 * 132 * Tag greater than or equal to queue_depth is for setting up 133 * flush request. 134 * 135 * Ditto for exit/teardown. 136 */ 137 init_request_fn *init_request; 138 exit_request_fn *exit_request; 139 reinit_request_fn *reinit_request; 140 141 map_queues_fn *map_queues; 142 143 #ifdef CONFIG_BLK_DEBUG_FS 144 /* 145 * Used by the debugfs implementation to show driver-specific 146 * information about a request. 147 */ 148 void (*show_rq)(struct seq_file *m, struct request *rq); 149 #endif 150 }; 151 152 enum { 153 BLK_MQ_RQ_QUEUE_OK = 0, /* queued fine */ 154 BLK_MQ_RQ_QUEUE_BUSY = 1, /* requeue IO for later */ 155 BLK_MQ_RQ_QUEUE_ERROR = 2, /* end IO with error */ 156 157 BLK_MQ_F_SHOULD_MERGE = 1 << 0, 158 BLK_MQ_F_TAG_SHARED = 1 << 1, 159 BLK_MQ_F_SG_MERGE = 1 << 2, 160 BLK_MQ_F_BLOCKING = 1 << 5, 161 BLK_MQ_F_NO_SCHED = 1 << 6, 162 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, 163 BLK_MQ_F_ALLOC_POLICY_BITS = 1, 164 165 BLK_MQ_S_STOPPED = 0, 166 BLK_MQ_S_TAG_ACTIVE = 1, 167 BLK_MQ_S_SCHED_RESTART = 2, 168 BLK_MQ_S_TAG_WAITING = 3, 169 BLK_MQ_S_START_ON_RUN = 4, 170 171 BLK_MQ_MAX_DEPTH = 10240, 172 173 BLK_MQ_CPU_WORK_BATCH = 8, 174 }; 175 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ 176 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ 177 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) 178 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ 179 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ 180 << BLK_MQ_F_ALLOC_POLICY_START_BIT) 181 182 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); 183 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, 184 struct request_queue *q); 185 int blk_mq_register_dev(struct device *, struct request_queue *); 186 void blk_mq_unregister_dev(struct device *, struct request_queue *); 187 188 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); 189 void blk_mq_free_tag_set(struct blk_mq_tag_set *set); 190 191 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); 192 193 void blk_mq_free_request(struct request *rq); 194 bool blk_mq_can_queue(struct blk_mq_hw_ctx *); 195 196 enum { 197 BLK_MQ_REQ_NOWAIT = (1 << 0), /* return when out of requests */ 198 BLK_MQ_REQ_RESERVED = (1 << 1), /* allocate from reserved pool */ 199 BLK_MQ_REQ_INTERNAL = (1 << 2), /* allocate internal/sched tag */ 200 }; 201 202 struct request *blk_mq_alloc_request(struct request_queue *q, int rw, 203 unsigned int flags); 204 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op, 205 unsigned int flags, unsigned int hctx_idx); 206 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); 207 208 enum { 209 BLK_MQ_UNIQUE_TAG_BITS = 16, 210 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, 211 }; 212 213 u32 blk_mq_unique_tag(struct request *rq); 214 215 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) 216 { 217 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; 218 } 219 220 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) 221 { 222 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; 223 } 224 225 226 int blk_mq_request_started(struct request *rq); 227 void blk_mq_start_request(struct request *rq); 228 void blk_mq_end_request(struct request *rq, int error); 229 void __blk_mq_end_request(struct request *rq, int error); 230 231 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); 232 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, 233 bool kick_requeue_list); 234 void blk_mq_kick_requeue_list(struct request_queue *q); 235 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); 236 void blk_mq_abort_requeue_list(struct request_queue *q); 237 void blk_mq_complete_request(struct request *rq); 238 239 bool blk_mq_queue_stopped(struct request_queue *q); 240 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); 241 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); 242 void blk_mq_stop_hw_queues(struct request_queue *q); 243 void blk_mq_start_hw_queues(struct request_queue *q); 244 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 245 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); 246 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); 247 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 248 void blk_mq_run_hw_queues(struct request_queue *q, bool async); 249 void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); 250 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, 251 busy_tag_iter_fn *fn, void *priv); 252 void blk_mq_freeze_queue(struct request_queue *q); 253 void blk_mq_unfreeze_queue(struct request_queue *q); 254 void blk_freeze_queue_start(struct request_queue *q); 255 void blk_mq_freeze_queue_wait(struct request_queue *q); 256 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, 257 unsigned long timeout); 258 int blk_mq_reinit_tagset(struct blk_mq_tag_set *set); 259 260 int blk_mq_map_queues(struct blk_mq_tag_set *set); 261 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); 262 263 /* 264 * Driver command data is immediately after the request. So subtract request 265 * size to get back to the original request, add request size to get the PDU. 266 */ 267 static inline struct request *blk_mq_rq_from_pdu(void *pdu) 268 { 269 return pdu - sizeof(struct request); 270 } 271 static inline void *blk_mq_rq_to_pdu(struct request *rq) 272 { 273 return rq + 1; 274 } 275 276 #define queue_for_each_hw_ctx(q, hctx, i) \ 277 for ((i) = 0; (i) < (q)->nr_hw_queues && \ 278 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) 279 280 #define hctx_for_each_ctx(hctx, ctx, i) \ 281 for ((i) = 0; (i) < (hctx)->nr_ctx && \ 282 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) 283 284 #endif 285