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 14 * block device 15 */ 16 struct blk_mq_hw_ctx { 17 struct { 18 /** @lock: Protects the dispatch list. */ 19 spinlock_t lock; 20 /** 21 * @dispatch: Used for requests that are ready to be 22 * dispatched to the hardware but for some reason (e.g. lack of 23 * resources) could not be sent to the hardware. As soon as the 24 * driver can send new requests, requests at this list will 25 * be sent first for a fairer dispatch. 26 */ 27 struct list_head dispatch; 28 /** 29 * @state: BLK_MQ_S_* flags. Defines the state of the hw 30 * queue (active, scheduled to restart, stopped). 31 */ 32 unsigned long state; 33 } ____cacheline_aligned_in_smp; 34 35 /** 36 * @run_work: Used for scheduling a hardware queue run at a later time. 37 */ 38 struct delayed_work run_work; 39 /** @cpumask: Map of available CPUs where this hctx can run. */ 40 cpumask_var_t cpumask; 41 /** 42 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU 43 * selection from @cpumask. 44 */ 45 int next_cpu; 46 /** 47 * @next_cpu_batch: Counter of how many works left in the batch before 48 * changing to the next CPU. 49 */ 50 int next_cpu_batch; 51 52 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */ 53 unsigned long flags; 54 55 /** 56 * @sched_data: Pointer owned by the IO scheduler attached to a request 57 * queue. It's up to the IO scheduler how to use this pointer. 58 */ 59 void *sched_data; 60 /** 61 * @queue: Pointer to the request queue that owns this hardware context. 62 */ 63 struct request_queue *queue; 64 /** @fq: Queue of requests that need to perform a flush operation. */ 65 struct blk_flush_queue *fq; 66 67 /** 68 * @driver_data: Pointer to data owned by the block driver that created 69 * this hctx 70 */ 71 void *driver_data; 72 73 /** 74 * @ctx_map: Bitmap for each software queue. If bit is on, there is a 75 * pending request in that software queue. 76 */ 77 struct sbitmap ctx_map; 78 79 /** 80 * @dispatch_from: Software queue to be used when no scheduler was 81 * selected. 82 */ 83 struct blk_mq_ctx *dispatch_from; 84 /** 85 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to 86 * decide if the hw_queue is busy using Exponential Weighted Moving 87 * Average algorithm. 88 */ 89 unsigned int dispatch_busy; 90 91 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */ 92 unsigned short type; 93 /** @nr_ctx: Number of software queues. */ 94 unsigned short nr_ctx; 95 /** @ctxs: Array of software queues. */ 96 struct blk_mq_ctx **ctxs; 97 98 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */ 99 spinlock_t dispatch_wait_lock; 100 /** 101 * @dispatch_wait: Waitqueue to put requests when there is no tag 102 * available at the moment, to wait for another try in the future. 103 */ 104 wait_queue_entry_t dispatch_wait; 105 106 /** 107 * @wait_index: Index of next available dispatch_wait queue to insert 108 * requests. 109 */ 110 atomic_t wait_index; 111 112 /** 113 * @tags: Tags owned by the block driver. A tag at this set is only 114 * assigned when a request is dispatched from a hardware queue. 115 */ 116 struct blk_mq_tags *tags; 117 /** 118 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O 119 * scheduler associated with a request queue, a tag is assigned when 120 * that request is allocated. Else, this member is not used. 121 */ 122 struct blk_mq_tags *sched_tags; 123 124 /** @queued: Number of queued requests. */ 125 unsigned long queued; 126 /** @run: Number of dispatched requests. */ 127 unsigned long run; 128 #define BLK_MQ_MAX_DISPATCH_ORDER 7 129 /** @dispatched: Number of dispatch requests by queue. */ 130 unsigned long dispatched[BLK_MQ_MAX_DISPATCH_ORDER]; 131 132 /** @numa_node: NUMA node the storage adapter has been connected to. */ 133 unsigned int numa_node; 134 /** @queue_num: Index of this hardware queue. */ 135 unsigned int queue_num; 136 137 /** 138 * @nr_active: Number of active requests. Only used when a tag set is 139 * shared across request queues. 140 */ 141 atomic_t nr_active; 142 /** 143 * @elevator_queued: Number of queued requests on hctx. 144 */ 145 atomic_t elevator_queued; 146 147 /** @cpuhp_online: List to store request if CPU is going to die */ 148 struct hlist_node cpuhp_online; 149 /** @cpuhp_dead: List to store request if some CPU die. */ 150 struct hlist_node cpuhp_dead; 151 /** @kobj: Kernel object for sysfs. */ 152 struct kobject kobj; 153 154 /** @poll_considered: Count times blk_poll() was called. */ 155 unsigned long poll_considered; 156 /** @poll_invoked: Count how many requests blk_poll() polled. */ 157 unsigned long poll_invoked; 158 /** @poll_success: Count how many polled requests were completed. */ 159 unsigned long poll_success; 160 161 #ifdef CONFIG_BLK_DEBUG_FS 162 /** 163 * @debugfs_dir: debugfs directory for this hardware queue. Named 164 * as cpu<cpu_number>. 165 */ 166 struct dentry *debugfs_dir; 167 /** @sched_debugfs_dir: debugfs directory for the scheduler. */ 168 struct dentry *sched_debugfs_dir; 169 #endif 170 171 /** 172 * @hctx_list: if this hctx is not in use, this is an entry in 173 * q->unused_hctx_list. 174 */ 175 struct list_head hctx_list; 176 177 /** 178 * @srcu: Sleepable RCU. Use as lock when type of the hardware queue is 179 * blocking (BLK_MQ_F_BLOCKING). Must be the last member - see also 180 * blk_mq_hw_ctx_size(). 181 */ 182 struct srcu_struct srcu[]; 183 }; 184 185 /** 186 * struct blk_mq_queue_map - Map software queues to hardware queues 187 * @mq_map: CPU ID to hardware queue index map. This is an array 188 * with nr_cpu_ids elements. Each element has a value in the range 189 * [@queue_offset, @queue_offset + @nr_queues). 190 * @nr_queues: Number of hardware queues to map CPU IDs onto. 191 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe 192 * driver to map each hardware queue type (enum hctx_type) onto a distinct 193 * set of hardware queues. 194 */ 195 struct blk_mq_queue_map { 196 unsigned int *mq_map; 197 unsigned int nr_queues; 198 unsigned int queue_offset; 199 }; 200 201 /** 202 * enum hctx_type - Type of hardware queue 203 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for. 204 * @HCTX_TYPE_READ: Just for READ I/O. 205 * @HCTX_TYPE_POLL: Polled I/O of any kind. 206 * @HCTX_MAX_TYPES: Number of types of hctx. 207 */ 208 enum hctx_type { 209 HCTX_TYPE_DEFAULT, 210 HCTX_TYPE_READ, 211 HCTX_TYPE_POLL, 212 213 HCTX_MAX_TYPES, 214 }; 215 216 /** 217 * struct blk_mq_tag_set - tag set that can be shared between request queues 218 * @map: One or more ctx -> hctx mappings. One map exists for each 219 * hardware queue type (enum hctx_type) that the driver wishes 220 * to support. There are no restrictions on maps being of the 221 * same size, and it's perfectly legal to share maps between 222 * types. 223 * @nr_maps: Number of elements in the @map array. A number in the range 224 * [1, HCTX_MAX_TYPES]. 225 * @ops: Pointers to functions that implement block driver behavior. 226 * @nr_hw_queues: Number of hardware queues supported by the block driver that 227 * owns this data structure. 228 * @queue_depth: Number of tags per hardware queue, reserved tags included. 229 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag 230 * allocations. 231 * @cmd_size: Number of additional bytes to allocate per request. The block 232 * driver owns these additional bytes. 233 * @numa_node: NUMA node the storage adapter has been connected to. 234 * @timeout: Request processing timeout in jiffies. 235 * @flags: Zero or more BLK_MQ_F_* flags. 236 * @driver_data: Pointer to data owned by the block driver that created this 237 * tag set. 238 * @__bitmap_tags: A shared tags sbitmap, used over all hctx's 239 * @__breserved_tags: 240 * A shared reserved tags sbitmap, used over all hctx's 241 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues 242 * elements. 243 * @tag_list_lock: Serializes tag_list accesses. 244 * @tag_list: List of the request queues that use this tag set. See also 245 * request_queue.tag_set_list. 246 */ 247 struct blk_mq_tag_set { 248 struct blk_mq_queue_map map[HCTX_MAX_TYPES]; 249 unsigned int nr_maps; 250 const struct blk_mq_ops *ops; 251 unsigned int nr_hw_queues; 252 unsigned int queue_depth; 253 unsigned int reserved_tags; 254 unsigned int cmd_size; 255 int numa_node; 256 unsigned int timeout; 257 unsigned int flags; 258 void *driver_data; 259 atomic_t active_queues_shared_sbitmap; 260 261 struct sbitmap_queue __bitmap_tags; 262 struct sbitmap_queue __breserved_tags; 263 struct blk_mq_tags **tags; 264 265 struct mutex tag_list_lock; 266 struct list_head tag_list; 267 }; 268 269 /** 270 * struct blk_mq_queue_data - Data about a request inserted in a queue 271 * 272 * @rq: Request pointer. 273 * @last: If it is the last request in the queue. 274 */ 275 struct blk_mq_queue_data { 276 struct request *rq; 277 bool last; 278 }; 279 280 typedef bool (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *, 281 bool); 282 typedef bool (busy_tag_iter_fn)(struct request *, void *, bool); 283 284 /** 285 * struct blk_mq_ops - Callback functions that implements block driver 286 * behaviour. 287 */ 288 struct blk_mq_ops { 289 /** 290 * @queue_rq: Queue a new request from block IO. 291 */ 292 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *, 293 const struct blk_mq_queue_data *); 294 295 /** 296 * @commit_rqs: If a driver uses bd->last to judge when to submit 297 * requests to hardware, it must define this function. In case of errors 298 * that make us stop issuing further requests, this hook serves the 299 * purpose of kicking the hardware (which the last request otherwise 300 * would have done). 301 */ 302 void (*commit_rqs)(struct blk_mq_hw_ctx *); 303 304 /** 305 * @get_budget: Reserve budget before queue request, once .queue_rq is 306 * run, it is driver's responsibility to release the 307 * reserved budget. Also we have to handle failure case 308 * of .get_budget for avoiding I/O deadlock. 309 */ 310 bool (*get_budget)(struct request_queue *); 311 312 /** 313 * @put_budget: Release the reserved budget. 314 */ 315 void (*put_budget)(struct request_queue *); 316 317 /** 318 * @timeout: Called on request timeout. 319 */ 320 enum blk_eh_timer_return (*timeout)(struct request *, bool); 321 322 /** 323 * @poll: Called to poll for completion of a specific tag. 324 */ 325 int (*poll)(struct blk_mq_hw_ctx *); 326 327 /** 328 * @complete: Mark the request as complete. 329 */ 330 void (*complete)(struct request *); 331 332 /** 333 * @init_hctx: Called when the block layer side of a hardware queue has 334 * been set up, allowing the driver to allocate/init matching 335 * structures. 336 */ 337 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int); 338 /** 339 * @exit_hctx: Ditto for exit/teardown. 340 */ 341 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int); 342 343 /** 344 * @init_request: Called for every command allocated by the block layer 345 * to allow the driver to set up driver specific data. 346 * 347 * Tag greater than or equal to queue_depth is for setting up 348 * flush request. 349 */ 350 int (*init_request)(struct blk_mq_tag_set *set, struct request *, 351 unsigned int, unsigned int); 352 /** 353 * @exit_request: Ditto for exit/teardown. 354 */ 355 void (*exit_request)(struct blk_mq_tag_set *set, struct request *, 356 unsigned int); 357 358 /** 359 * @initialize_rq_fn: Called from inside blk_get_request(). 360 */ 361 void (*initialize_rq_fn)(struct request *rq); 362 363 /** 364 * @cleanup_rq: Called before freeing one request which isn't completed 365 * yet, and usually for freeing the driver private data. 366 */ 367 void (*cleanup_rq)(struct request *); 368 369 /** 370 * @busy: If set, returns whether or not this queue currently is busy. 371 */ 372 bool (*busy)(struct request_queue *); 373 374 /** 375 * @map_queues: This allows drivers specify their own queue mapping by 376 * overriding the setup-time function that builds the mq_map. 377 */ 378 int (*map_queues)(struct blk_mq_tag_set *set); 379 380 #ifdef CONFIG_BLK_DEBUG_FS 381 /** 382 * @show_rq: Used by the debugfs implementation to show driver-specific 383 * information about a request. 384 */ 385 void (*show_rq)(struct seq_file *m, struct request *rq); 386 #endif 387 }; 388 389 enum { 390 BLK_MQ_F_SHOULD_MERGE = 1 << 0, 391 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1, 392 /* 393 * Set when this device requires underlying blk-mq device for 394 * completing IO: 395 */ 396 BLK_MQ_F_STACKING = 1 << 2, 397 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3, 398 BLK_MQ_F_BLOCKING = 1 << 5, 399 BLK_MQ_F_NO_SCHED = 1 << 6, 400 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, 401 BLK_MQ_F_ALLOC_POLICY_BITS = 1, 402 403 BLK_MQ_S_STOPPED = 0, 404 BLK_MQ_S_TAG_ACTIVE = 1, 405 BLK_MQ_S_SCHED_RESTART = 2, 406 407 /* hw queue is inactive after all its CPUs become offline */ 408 BLK_MQ_S_INACTIVE = 3, 409 410 BLK_MQ_MAX_DEPTH = 10240, 411 412 BLK_MQ_CPU_WORK_BATCH = 8, 413 }; 414 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ 415 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ 416 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) 417 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ 418 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ 419 << BLK_MQ_F_ALLOC_POLICY_START_BIT) 420 421 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); 422 struct request_queue *blk_mq_init_queue_data(struct blk_mq_tag_set *set, 423 void *queuedata); 424 struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, 425 struct request_queue *q, 426 bool elevator_init); 427 struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set, 428 const struct blk_mq_ops *ops, 429 unsigned int queue_depth, 430 unsigned int set_flags); 431 void blk_mq_unregister_dev(struct device *, struct request_queue *); 432 433 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); 434 void blk_mq_free_tag_set(struct blk_mq_tag_set *set); 435 436 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule); 437 438 void blk_mq_free_request(struct request *rq); 439 440 bool blk_mq_queue_inflight(struct request_queue *q); 441 442 enum { 443 /* return when out of requests */ 444 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0), 445 /* allocate from reserved pool */ 446 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1), 447 /* set RQF_PREEMPT */ 448 BLK_MQ_REQ_PREEMPT = (__force blk_mq_req_flags_t)(1 << 3), 449 }; 450 451 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op, 452 blk_mq_req_flags_t flags); 453 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, 454 unsigned int op, blk_mq_req_flags_t flags, 455 unsigned int hctx_idx); 456 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag); 457 458 enum { 459 BLK_MQ_UNIQUE_TAG_BITS = 16, 460 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, 461 }; 462 463 u32 blk_mq_unique_tag(struct request *rq); 464 465 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) 466 { 467 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; 468 } 469 470 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) 471 { 472 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; 473 } 474 475 /** 476 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request 477 * @rq: target request. 478 */ 479 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq) 480 { 481 return READ_ONCE(rq->state); 482 } 483 484 static inline int blk_mq_request_started(struct request *rq) 485 { 486 return blk_mq_rq_state(rq) != MQ_RQ_IDLE; 487 } 488 489 static inline int blk_mq_request_completed(struct request *rq) 490 { 491 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE; 492 } 493 494 void blk_mq_start_request(struct request *rq); 495 void blk_mq_end_request(struct request *rq, blk_status_t error); 496 void __blk_mq_end_request(struct request *rq, blk_status_t error); 497 498 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); 499 void blk_mq_kick_requeue_list(struct request_queue *q); 500 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); 501 void blk_mq_complete_request(struct request *rq); 502 bool blk_mq_complete_request_remote(struct request *rq); 503 bool blk_mq_queue_stopped(struct request_queue *q); 504 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); 505 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); 506 void blk_mq_stop_hw_queues(struct request_queue *q); 507 void blk_mq_start_hw_queues(struct request_queue *q); 508 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 509 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); 510 void blk_mq_quiesce_queue(struct request_queue *q); 511 void blk_mq_unquiesce_queue(struct request_queue *q); 512 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); 513 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 514 void blk_mq_run_hw_queues(struct request_queue *q, bool async); 515 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs); 516 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, 517 busy_tag_iter_fn *fn, void *priv); 518 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset); 519 void blk_mq_freeze_queue(struct request_queue *q); 520 void blk_mq_unfreeze_queue(struct request_queue *q); 521 void blk_freeze_queue_start(struct request_queue *q); 522 void blk_mq_freeze_queue_wait(struct request_queue *q); 523 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, 524 unsigned long timeout); 525 526 int blk_mq_map_queues(struct blk_mq_queue_map *qmap); 527 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); 528 529 void blk_mq_quiesce_queue_nowait(struct request_queue *q); 530 531 unsigned int blk_mq_rq_cpu(struct request *rq); 532 533 bool __blk_should_fake_timeout(struct request_queue *q); 534 static inline bool blk_should_fake_timeout(struct request_queue *q) 535 { 536 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) && 537 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags)) 538 return __blk_should_fake_timeout(q); 539 return false; 540 } 541 542 /** 543 * blk_mq_rq_from_pdu - cast a PDU to a request 544 * @pdu: the PDU (Protocol Data Unit) to be casted 545 * 546 * Return: request 547 * 548 * Driver command data is immediately after the request. So subtract request 549 * size to get back to the original request. 550 */ 551 static inline struct request *blk_mq_rq_from_pdu(void *pdu) 552 { 553 return pdu - sizeof(struct request); 554 } 555 556 /** 557 * blk_mq_rq_to_pdu - cast a request to a PDU 558 * @rq: the request to be casted 559 * 560 * Return: pointer to the PDU 561 * 562 * Driver command data is immediately after the request. So add request to get 563 * the PDU. 564 */ 565 static inline void *blk_mq_rq_to_pdu(struct request *rq) 566 { 567 return rq + 1; 568 } 569 570 #define queue_for_each_hw_ctx(q, hctx, i) \ 571 for ((i) = 0; (i) < (q)->nr_hw_queues && \ 572 ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++) 573 574 #define hctx_for_each_ctx(hctx, ctx, i) \ 575 for ((i) = 0; (i) < (hctx)->nr_ctx && \ 576 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) 577 578 static inline blk_qc_t request_to_qc_t(struct blk_mq_hw_ctx *hctx, 579 struct request *rq) 580 { 581 if (rq->tag != -1) 582 return rq->tag | (hctx->queue_num << BLK_QC_T_SHIFT); 583 584 return rq->internal_tag | (hctx->queue_num << BLK_QC_T_SHIFT) | 585 BLK_QC_T_INTERNAL; 586 } 587 588 static inline void blk_mq_cleanup_rq(struct request *rq) 589 { 590 if (rq->q->mq_ops->cleanup_rq) 591 rq->q->mq_ops->cleanup_rq(rq); 592 } 593 594 blk_qc_t blk_mq_submit_bio(struct bio *bio); 595 596 #endif 597