| 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 | unsigned int dispatch_busy; |
| 39 | |
| 40 | unsigned short type; |
| 41 | unsigned short nr_ctx; |
| 42 | struct blk_mq_ctx **ctxs; |
| 43 | |
| 44 | spinlock_t dispatch_wait_lock; |
| 45 | wait_queue_entry_t dispatch_wait; |
| 46 | atomic_t wait_index; |
| 47 | |
| 48 | struct blk_mq_tags *tags; |
| 49 | struct blk_mq_tags *sched_tags; |
| 50 | |
| 51 | unsigned long queued; |
| 52 | unsigned long run; |
| 53 | #define BLK_MQ_MAX_DISPATCH_ORDER 7 |
| 54 | unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; |
| 55 | |
| 56 | unsigned int numa_node; |
| 57 | unsigned int queue_num; |
| 58 | |
| 59 | atomic_t nr_active; |
| 60 | |
| 61 | struct hlist_node cpuhp_dead; |
| 62 | struct kobject kobj; |
| 63 | |
| 64 | unsigned long poll_considered; |
| 65 | unsigned long poll_invoked; |
| 66 | unsigned long poll_success; |
| 67 | |
| 68 | #ifdef CONFIG_BLK_DEBUG_FS |
| 69 | struct dentry *debugfs_dir; |
| 70 | struct dentry *sched_debugfs_dir; |
| 71 | #endif |
| 72 | |
| 73 | struct list_head hctx_list; |
| 74 | |
| 75 | /* Must be the last member - see also blk_mq_hw_ctx_size(). */ |
| 76 | struct srcu_struct srcu[0]; |
| 77 | }; |
| 78 | |
| 79 | /** |
| 80 | * struct blk_mq_queue_map - ctx -> hctx mapping |
| 81 | * @mq_map: CPU ID to hardware queue index map. This is an array |
| 82 | * with nr_cpu_ids elements. Each element has a value in the range |
| 83 | * [@queue_offset, @queue_offset + @nr_queues). |
| 84 | * @nr_queues: Number of hardware queues to map CPU IDs onto. |
| 85 | * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe |
| 86 | * driver to map each hardware queue type (enum hctx_type) onto a distinct |
| 87 | * set of hardware queues. |
| 88 | */ |
| 89 | struct blk_mq_queue_map { |
| 90 | unsigned int *mq_map; |
| 91 | unsigned int nr_queues; |
| 92 | unsigned int queue_offset; |
| 93 | }; |
| 94 | |
| 95 | enum hctx_type { |
| 96 | HCTX_TYPE_DEFAULT, /* all I/O not otherwise accounted for */ |
| 97 | HCTX_TYPE_READ, /* just for READ I/O */ |
| 98 | HCTX_TYPE_POLL, /* polled I/O of any kind */ |
| 99 | |
| 100 | HCTX_MAX_TYPES, |
| 101 | }; |
| 102 | |
| 103 | /** |
| 104 | * struct blk_mq_tag_set - tag set that can be shared between request queues |
| 105 | * @map: One or more ctx -> hctx mappings. One map exists for each |
| 106 | * hardware queue type (enum hctx_type) that the driver wishes |
| 107 | * to support. There are no restrictions on maps being of the |
| 108 | * same size, and it's perfectly legal to share maps between |
| 109 | * types. |
| 110 | * @nr_maps: Number of elements in the @map array. A number in the range |
| 111 | * [1, HCTX_MAX_TYPES]. |
| 112 | * @ops: Pointers to functions that implement block driver behavior. |
| 113 | * @nr_hw_queues: Number of hardware queues supported by the block driver that |
| 114 | * owns this data structure. |
| 115 | * @queue_depth: Number of tags per hardware queue, reserved tags included. |
| 116 | * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag |
| 117 | * allocations. |
| 118 | * @cmd_size: Number of additional bytes to allocate per request. The block |
| 119 | * driver owns these additional bytes. |
| 120 | * @numa_node: NUMA node the storage adapter has been connected to. |
| 121 | * @timeout: Request processing timeout in jiffies. |
| 122 | * @flags: Zero or more BLK_MQ_F_* flags. |
| 123 | * @driver_data: Pointer to data owned by the block driver that created this |
| 124 | * tag set. |
| 125 | * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues |
| 126 | * elements. |
| 127 | * @tag_list_lock: Serializes tag_list accesses. |
| 128 | * @tag_list: List of the request queues that use this tag set. See also |
| 129 | * request_queue.tag_set_list. |
| 130 | */ |
| 131 | struct blk_mq_tag_set { |
| 132 | struct blk_mq_queue_map map[HCTX_MAX_TYPES]; |
| 133 | unsigned int nr_maps; |
| 134 | const struct blk_mq_ops *ops; |
| 135 | unsigned int nr_hw_queues; |
| 136 | unsigned int queue_depth; |
| 137 | unsigned int reserved_tags; |
| 138 | unsigned int cmd_size; |
| 139 | int numa_node; |
| 140 | unsigned int timeout; |
| 141 | unsigned int flags; |
| 142 | void *driver_data; |
| 143 | |
| 144 | struct blk_mq_tags **tags; |
| 145 | |
| 146 | struct mutex tag_list_lock; |
| 147 | struct list_head tag_list; |
| 148 | }; |
| 149 | |
| 150 | struct blk_mq_queue_data { |
| 151 | struct request *rq; |
| 152 | bool last; |
| 153 | }; |
| 154 | |
| 155 | typedef blk_status_t (queue_rq_fn)(struct blk_mq_hw_ctx *, |
| 156 | const struct blk_mq_queue_data *); |
| 157 | typedef void (commit_rqs_fn)(struct blk_mq_hw_ctx *); |
| 158 | typedef bool (get_budget_fn)(struct blk_mq_hw_ctx *); |
| 159 | typedef void (put_budget_fn)(struct blk_mq_hw_ctx *); |
| 160 | typedef enum blk_eh_timer_return (timeout_fn)(struct request *, bool); |
| 161 | typedef int (init_hctx_fn)(struct blk_mq_hw_ctx *, void *, unsigned int); |
| 162 | typedef void (exit_hctx_fn)(struct blk_mq_hw_ctx *, unsigned int); |
| 163 | typedef int (init_request_fn)(struct blk_mq_tag_set *set, struct request *, |
| 164 | unsigned int, unsigned int); |
| 165 | typedef void (exit_request_fn)(struct blk_mq_tag_set *set, struct request *, |
| 166 | unsigned int); |
| 167 | |
| 168 | typedef bool (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, |
| 169 | bool); |
| 170 | typedef bool (busy_tag_iter_fn)(struct request *, void *, bool); |
| 171 | typedef int (poll_fn)(struct blk_mq_hw_ctx *); |
| 172 | typedef int (map_queues_fn)(struct blk_mq_tag_set *set); |
| 173 | typedef bool (busy_fn)(struct request_queue *); |
| 174 | typedef void (complete_fn)(struct request *); |
| 175 | typedef void (cleanup_rq_fn)(struct request *); |
| 176 | |
| 177 | |
| 178 | struct blk_mq_ops { |
| 179 | /* |
| 180 | * Queue request |
| 181 | */ |
| 182 | queue_rq_fn *queue_rq; |
| 183 | |
| 184 | /* |
| 185 | * If a driver uses bd->last to judge when to submit requests to |
| 186 | * hardware, it must define this function. In case of errors that |
| 187 | * make us stop issuing further requests, this hook serves the |
| 188 | * purpose of kicking the hardware (which the last request otherwise |
| 189 | * would have done). |
| 190 | */ |
| 191 | commit_rqs_fn *commit_rqs; |
| 192 | |
| 193 | /* |
| 194 | * Reserve budget before queue request, once .queue_rq is |
| 195 | * run, it is driver's responsibility to release the |
| 196 | * reserved budget. Also we have to handle failure case |
| 197 | * of .get_budget for avoiding I/O deadlock. |
| 198 | */ |
| 199 | get_budget_fn *get_budget; |
| 200 | put_budget_fn *put_budget; |
| 201 | |
| 202 | /* |
| 203 | * Called on request timeout |
| 204 | */ |
| 205 | timeout_fn *timeout; |
| 206 | |
| 207 | /* |
| 208 | * Called to poll for completion of a specific tag. |
| 209 | */ |
| 210 | poll_fn *poll; |
| 211 | |
| 212 | complete_fn *complete; |
| 213 | |
| 214 | /* |
| 215 | * Called when the block layer side of a hardware queue has been |
| 216 | * set up, allowing the driver to allocate/init matching structures. |
| 217 | * Ditto for exit/teardown. |
| 218 | */ |
| 219 | init_hctx_fn *init_hctx; |
| 220 | exit_hctx_fn *exit_hctx; |
| 221 | |
| 222 | /* |
| 223 | * Called for every command allocated by the block layer to allow |
| 224 | * the driver to set up driver specific data. |
| 225 | * |
| 226 | * Tag greater than or equal to queue_depth is for setting up |
| 227 | * flush request. |
| 228 | * |
| 229 | * Ditto for exit/teardown. |
| 230 | */ |
| 231 | init_request_fn *init_request; |
| 232 | exit_request_fn *exit_request; |
| 233 | /* Called from inside blk_get_request() */ |
| 234 | void (*initialize_rq_fn)(struct request *rq); |
| 235 | |
| 236 | /* |
| 237 | * Called before freeing one request which isn't completed yet, |
| 238 | * and usually for freeing the driver private data |
| 239 | */ |
| 240 | cleanup_rq_fn *cleanup_rq; |
| 241 | |
| 242 | /* |
| 243 | * If set, returns whether or not this queue currently is busy |
| 244 | */ |
| 245 | busy_fn *busy; |
| 246 | |
| 247 | map_queues_fn *map_queues; |
| 248 | |
| 249 | #ifdef CONFIG_BLK_DEBUG_FS |
| 250 | /* |
| 251 | * Used by the debugfs implementation to show driver-specific |
| 252 | * information about a request. |
| 253 | */ |
| 254 | void (*show_rq)(struct seq_file *m, struct request *rq); |
| 255 | #endif |
| 256 | }; |
| 257 | |
| 258 | enum { |
| 259 | BLK_MQ_F_SHOULD_MERGE = 1 << 0, |
| 260 | BLK_MQ_F_TAG_SHARED = 1 << 1, |
| 261 | BLK_MQ_F_BLOCKING = 1 << 5, |
| 262 | BLK_MQ_F_NO_SCHED = 1 << 6, |
| 263 | BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, |
| 264 | BLK_MQ_F_ALLOC_POLICY_BITS = 1, |
| 265 | |
| 266 | BLK_MQ_S_STOPPED = 0, |
| 267 | BLK_MQ_S_TAG_ACTIVE = 1, |
| 268 | BLK_MQ_S_SCHED_RESTART = 2, |
| 269 | |
| 270 | BLK_MQ_MAX_DEPTH = 10240, |
| 271 | |
| 272 | BLK_MQ_CPU_WORK_BATCH = 8, |
| 273 | }; |
| 274 | #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ |
| 275 | ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ |
| 276 | ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) |
| 277 | #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ |
| 278 | ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ |
| 279 | << BLK_MQ_F_ALLOC_POLICY_START_BIT) |
| 280 | |
| 281 | struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); |
| 282 | struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, |
| 283 | struct request_queue *q, |
| 284 | bool elevator_init); |
| 285 | struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set, |
| 286 | const struct blk_mq_ops *ops, |
| 287 | unsigned int queue_depth, |
| 288 | unsigned int set_flags); |
| 289 | void blk_mq_unregister_dev(struct device *, struct request_queue *); |
| 290 | |
| 291 | int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); |
| 292 | void blk_mq_free_tag_set(struct blk_mq_tag_set *set); |
| 293 | |
| 294 | void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); |
| 295 | |
| 296 | void blk_mq_free_request(struct request *rq); |
| 297 | bool blk_mq_can_queue(struct blk_mq_hw_ctx *); |
| 298 | |
| 299 | bool blk_mq_queue_inflight(struct request_queue *q); |
| 300 | |
| 301 | enum { |
| 302 | /* return when out of requests */ |
| 303 | BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0), |
| 304 | /* allocate from reserved pool */ |
| 305 | BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1), |
| 306 | /* allocate internal/sched tag */ |
| 307 | BLK_MQ_REQ_INTERNAL = (__force blk_mq_req_flags_t)(1 << 2), |
| 308 | /* set RQF_PREEMPT */ |
| 309 | BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3), |
| 310 | }; |
| 311 | |
| 312 | struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op, |
| 313 | blk_mq_req_flags_t flags); |
| 314 | struct request *blk_mq_alloc_request_hctx(struct request_queue *q, |
| 315 | unsigned int op, blk_mq_req_flags_t flags, |
| 316 | unsigned int hctx_idx); |
| 317 | struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); |
| 318 | |
| 319 | enum { |
| 320 | BLK_MQ_UNIQUE_TAG_BITS = 16, |
| 321 | BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, |
| 322 | }; |
| 323 | |
| 324 | u32 blk_mq_unique_tag(struct request *rq); |
| 325 | |
| 326 | static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) |
| 327 | { |
| 328 | return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; |
| 329 | } |
| 330 | |
| 331 | static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) |
| 332 | { |
| 333 | return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; |
| 334 | } |
| 335 | |
| 336 | |
| 337 | int blk_mq_request_started(struct request *rq); |
| 338 | int blk_mq_request_completed(struct request *rq); |
| 339 | void blk_mq_start_request(struct request *rq); |
| 340 | void blk_mq_end_request(struct request *rq, blk_status_t error); |
| 341 | void __blk_mq_end_request(struct request *rq, blk_status_t error); |
| 342 | |
| 343 | void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); |
| 344 | void blk_mq_kick_requeue_list(struct request_queue *q); |
| 345 | void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); |
| 346 | bool blk_mq_complete_request(struct request *rq); |
| 347 | bool blk_mq_bio_list_merge(struct request_queue *q, struct list_head *list, |
| 348 | struct bio *bio, unsigned int nr_segs); |
| 349 | bool blk_mq_queue_stopped(struct request_queue *q); |
| 350 | void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); |
| 351 | void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); |
| 352 | void blk_mq_stop_hw_queues(struct request_queue *q); |
| 353 | void blk_mq_start_hw_queues(struct request_queue *q); |
| 354 | void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); |
| 355 | void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); |
| 356 | void blk_mq_quiesce_queue(struct request_queue *q); |
| 357 | void blk_mq_unquiesce_queue(struct request_queue *q); |
| 358 | void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); |
| 359 | bool blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); |
| 360 | void blk_mq_run_hw_queues(struct request_queue *q, bool async); |
| 361 | void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, |
| 362 | busy_tag_iter_fn *fn, void *priv); |
| 363 | void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset); |
| 364 | void blk_mq_freeze_queue(struct request_queue *q); |
| 365 | void blk_mq_unfreeze_queue(struct request_queue *q); |
| 366 | void blk_freeze_queue_start(struct request_queue *q); |
| 367 | void blk_mq_freeze_queue_wait(struct request_queue *q); |
| 368 | int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, |
| 369 | unsigned long timeout); |
| 370 | |
| 371 | int blk_mq_map_queues(struct blk_mq_queue_map *qmap); |
| 372 | void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); |
| 373 | |
| 374 | void blk_mq_quiesce_queue_nowait(struct request_queue *q); |
| 375 | |
| 376 | unsigned int blk_mq_rq_cpu(struct request *rq); |
| 377 | |
| 378 | /* |
| 379 | * Driver command data is immediately after the request. So subtract request |
| 380 | * size to get back to the original request, add request size to get the PDU. |
| 381 | */ |
| 382 | static inline struct request *blk_mq_rq_from_pdu(void *pdu) |
| 383 | { |
| 384 | return pdu - sizeof(struct request); |
| 385 | } |
| 386 | static inline void *blk_mq_rq_to_pdu(struct request *rq) |
| 387 | { |
| 388 | return rq + 1; |
| 389 | } |
| 390 | |
| 391 | #define queue_for_each_hw_ctx(q, hctx, i) \ |
| 392 | for ((i) = 0; (i) < (q)->nr_hw_queues && \ |
| 393 | ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) |
| 394 | |
| 395 | #define hctx_for_each_ctx(hctx, ctx, i) \ |
| 396 | for ((i) = 0; (i) < (hctx)->nr_ctx && \ |
| 397 | ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) |
| 398 | |
| 399 | static inline blk_qc_t request_to_qc_t(struct blk_mq_hw_ctx *hctx, |
| 400 | struct request *rq) |
| 401 | { |
| 402 | if (rq->tag != -1) |
| 403 | return rq->tag | (hctx->queue_num << BLK_QC_T_SHIFT); |
| 404 | |
| 405 | return rq->internal_tag | (hctx->queue_num << BLK_QC_T_SHIFT) | |
| 406 | BLK_QC_T_INTERNAL; |
| 407 | } |
| 408 | |
| 409 | static inline void blk_mq_cleanup_rq(struct request *rq) |
| 410 | { |
| 411 | if (rq->q->mq_ops->cleanup_rq) |
| 412 | rq->q->mq_ops->cleanup_rq(rq); |
| 413 | } |
| 414 | |
| 415 | #endif |