1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef BLK_MQ_H 3 #define BLK_MQ_H 4 5 #include <linux/blkdev.h> 6 #include <linux/sbitmap.h> 7 #include <linux/srcu.h> 8 9 struct blk_mq_tags; 10 struct blk_flush_queue; 11 12 /** 13 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware block device 14 */ 15 struct blk_mq_hw_ctx { 16 struct { 17 spinlock_t lock; 18 struct list_head dispatch; 19 unsigned long state; /* BLK_MQ_S_* flags */ 20 } ____cacheline_aligned_in_smp; 21 22 struct delayed_work run_work; 23 cpumask_var_t cpumask; 24 int next_cpu; 25 int next_cpu_batch; 26 27 unsigned long flags; /* BLK_MQ_F_* flags */ 28 29 void *sched_data; 30 struct request_queue *queue; 31 struct blk_flush_queue *fq; 32 33 void *driver_data; 34 35 struct sbitmap ctx_map; 36 37 struct blk_mq_ctx *dispatch_from; 38 39 struct blk_mq_ctx **ctxs; 40 unsigned int nr_ctx; 41 42 wait_queue_entry_t dispatch_wait; 43 atomic_t wait_index; 44 45 struct blk_mq_tags *tags; 46 struct blk_mq_tags *sched_tags; 47 48 unsigned long queued; 49 unsigned long run; 50 #define BLK_MQ_MAX_DISPATCH_ORDER 7 51 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; 52 53 unsigned int numa_node; 54 unsigned int queue_num; 55 56 atomic_t nr_active; 57 unsigned int nr_expired; 58 59 struct hlist_node cpuhp_dead; 60 struct kobject kobj; 61 62 unsigned long poll_considered; 63 unsigned long poll_invoked; 64 unsigned long poll_success; 65 66 #ifdef CONFIG_BLK_DEBUG_FS 67 struct dentry *debugfs_dir; 68 struct dentry *sched_debugfs_dir; 69 #endif 70 71 /* Must be the last member - see also blk_mq_hw_ctx_size(). */ 72 struct srcu_struct srcu[0]; 73 }; 74 75 struct blk_mq_tag_set { 76 unsigned int *mq_map; 77 const struct blk_mq_ops *ops; 78 unsigned int nr_hw_queues; 79 unsigned int queue_depth; /* max hw supported */ 80 unsigned int reserved_tags; 81 unsigned int cmd_size; /* per-request extra data */ 82 int numa_node; 83 unsigned int timeout; 84 unsigned int flags; /* BLK_MQ_F_* */ 85 void *driver_data; 86 87 struct blk_mq_tags **tags; 88 89 struct mutex tag_list_lock; 90 struct list_head tag_list; 91 }; 92 93 struct blk_mq_queue_data { 94 struct request *rq; 95 bool last; 96 }; 97 98 typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *, 99 const struct blk_mq_queue_data *); 100 typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *); 101 typedef void (put_budget_fn)(struct blk_mq_hw_ctx *); 102 typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool); 103 typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); 104 typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); 105 typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *, 106 unsigned int, unsigned int); 107 typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *, 108 unsigned int); 109 110 typedef void (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, 111 bool); 112 typedef void (busy_tag_iter_fn)(struct request *, void *, bool); 113 typedef int (poll_fn)(struct blk_mq_hw_ctx *, unsigned int); 114 typedef int (map_queues_fn)(struct blk_mq_tag_set *set); 115 116 117 struct blk_mq_ops { 118 /* 119 * Queue request 120 */ 121 queue_rq_fn *queue_rq; 122 123 /* 124 * Reserve budget before queue request, once .queue_rq is 125 * run, it is driver's responsibility to release the 126 * reserved budget. Also we have to handle failure case 127 * of .get_budget for avoiding I/O deadlock. 128 */ 129 get_budget_fn *get_budget; 130 put_budget_fn *put_budget; 131 132 /* 133 * Called on request timeout 134 */ 135 timeout_fn *timeout; 136 137 /* 138 * Called to poll for completion of a specific tag. 139 */ 140 poll_fn *poll; 141 142 softirq_done_fn *complete; 143 144 /* 145 * Called when the block layer side of a hardware queue has been 146 * set up, allowing the driver to allocate/init matching structures. 147 * Ditto for exit/teardown. 148 */ 149 init_hctx_fn *init_hctx; 150 exit_hctx_fn *exit_hctx; 151 152 /* 153 * Called for every command allocated by the block layer to allow 154 * the driver to set up driver specific data. 155 * 156 * Tag greater than or equal to queue_depth is for setting up 157 * flush request. 158 * 159 * Ditto for exit/teardown. 160 */ 161 init_request_fn *init_request; 162 exit_request_fn *exit_request; 163 /* Called from inside blk_get_request() */ 164 void (*initialize_rq_fn)(struct request *rq); 165 166 map_queues_fn *map_queues; 167 168 #ifdef CONFIG_BLK_DEBUG_FS 169 /* 170 * Used by the debugfs implementation to show driver-specific 171 * information about a request. 172 */ 173 void (*show_rq)(struct seq_file *m, struct request *rq); 174 #endif 175 }; 176 177 enum { 178 BLK_MQ_F_SHOULD_MERGE = 1 << 0, 179 BLK_MQ_F_TAG_SHARED = 1 << 1, 180 BLK_MQ_F_SG_MERGE = 1 << 2, 181 BLK_MQ_F_BLOCKING = 1 << 5, 182 BLK_MQ_F_NO_SCHED = 1 << 6, 183 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, 184 BLK_MQ_F_ALLOC_POLICY_BITS = 1, 185 186 BLK_MQ_S_STOPPED = 0, 187 BLK_MQ_S_TAG_ACTIVE = 1, 188 BLK_MQ_S_SCHED_RESTART = 2, 189 190 BLK_MQ_MAX_DEPTH = 10240, 191 192 BLK_MQ_CPU_WORK_BATCH = 8, 193 }; 194 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ 195 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ 196 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) 197 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ 198 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ 199 << BLK_MQ_F_ALLOC_POLICY_START_BIT) 200 201 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); 202 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, 203 struct request_queue *q); 204 int blk_mq_register_dev(struct device *, struct request_queue *); 205 void blk_mq_unregister_dev(struct device *, struct request_queue *); 206 207 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); 208 void blk_mq_free_tag_set(struct blk_mq_tag_set *set); 209 210 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); 211 212 void blk_mq_free_request(struct request *rq); 213 bool blk_mq_can_queue(struct blk_mq_hw_ctx *); 214 215 enum { 216 /* return when out of requests */ 217 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0), 218 /* allocate from reserved pool */ 219 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1), 220 /* allocate internal/sched tag */ 221 BLK_MQ_REQ_INTERNAL = (__force blk_mq_req_flags_t)(1 << 2), 222 /* set RQF_PREEMPT */ 223 BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3), 224 }; 225 226 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op, 227 blk_mq_req_flags_t flags); 228 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, 229 unsigned int op, blk_mq_req_flags_t flags, 230 unsigned int hctx_idx); 231 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); 232 233 enum { 234 BLK_MQ_UNIQUE_TAG_BITS = 16, 235 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, 236 }; 237 238 u32 blk_mq_unique_tag(struct request *rq); 239 240 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) 241 { 242 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; 243 } 244 245 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) 246 { 247 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; 248 } 249 250 251 int blk_mq_request_started(struct request *rq); 252 void blk_mq_start_request(struct request *rq); 253 void blk_mq_end_request(struct request *rq, blk_status_t error); 254 void __blk_mq_end_request(struct request *rq, blk_status_t error); 255 256 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); 257 void blk_mq_add_to_requeue_list(struct request *rq, bool at_head, 258 bool kick_requeue_list); 259 void blk_mq_kick_requeue_list(struct request_queue *q); 260 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); 261 void blk_mq_complete_request(struct request *rq); 262 263 bool blk_mq_queue_stopped(struct request_queue *q); 264 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); 265 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); 266 void blk_mq_stop_hw_queues(struct request_queue *q); 267 void blk_mq_start_hw_queues(struct request_queue *q); 268 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 269 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); 270 void blk_mq_quiesce_queue(struct request_queue *q); 271 void blk_mq_unquiesce_queue(struct request_queue *q); 272 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); 273 bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 274 void blk_mq_run_hw_queues(struct request_queue *q, bool async); 275 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, 276 busy_tag_iter_fn *fn, void *priv); 277 void blk_mq_freeze_queue(struct request_queue *q); 278 void blk_mq_unfreeze_queue(struct request_queue *q); 279 void blk_freeze_queue_start(struct request_queue *q); 280 void blk_mq_freeze_queue_wait(struct request_queue *q); 281 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, 282 unsigned long timeout); 283 int blk_mq_tagset_iter(struct blk_mq_tag_set *set, void *data, 284 int (reinit_request)(void *, struct request *)); 285 286 int blk_mq_map_queues(struct blk_mq_tag_set *set); 287 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); 288 289 void blk_mq_quiesce_queue_nowait(struct request_queue *q); 290 291 /* 292 * Driver command data is immediately after the request. So subtract request 293 * size to get back to the original request, add request size to get the PDU. 294 */ 295 static inline struct request *blk_mq_rq_from_pdu(void *pdu) 296 { 297 return pdu - sizeof(struct request); 298 } 299 static inline void *blk_mq_rq_to_pdu(struct request *rq) 300 { 301 return rq + 1; 302 } 303 304 #define queue_for_each_hw_ctx(q, hctx, i) \ 305 for ((i) = 0; (i) < (q)->nr_hw_queues && \ 306 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) 307 308 #define hctx_for_each_ctx(hctx, ctx, i) \ 309 for ((i) = 0; (i) < (hctx)->nr_ctx && \ 310 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) 311 312 #endif 313