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/lockdep.h> 8 #include <linux/scatterlist.h> 9 #include <linux/prefetch.h> 10 #include <linux/srcu.h> 11 12 struct blk_mq_tags; 13 struct blk_flush_queue; 14 15 #define BLKDEV_MIN_RQ 4 16 #define BLKDEV_DEFAULT_RQ 128 17 18 enum rq_end_io_ret { 19 RQ_END_IO_NONE, 20 RQ_END_IO_FREE, 21 }; 22 23 typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t); 24 25 /* 26 * request flags */ 27 typedef __u32 __bitwise req_flags_t; 28 29 /* drive already may have started this one */ 30 #define RQF_STARTED ((__force req_flags_t)(1 << 1)) 31 /* may not be passed by ioscheduler */ 32 #define RQF_SOFTBARRIER ((__force req_flags_t)(1 << 3)) 33 /* request for flush sequence */ 34 #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4)) 35 /* merge of different types, fail separately */ 36 #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5)) 37 /* track inflight for MQ */ 38 #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6)) 39 /* don't call prep for this one */ 40 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7)) 41 /* vaguely specified driver internal error. Ignored by the block layer */ 42 #define RQF_FAILED ((__force req_flags_t)(1 << 10)) 43 /* don't warn about errors */ 44 #define RQF_QUIET ((__force req_flags_t)(1 << 11)) 45 /* elevator private data attached */ 46 #define RQF_ELVPRIV ((__force req_flags_t)(1 << 12)) 47 /* account into disk and partition IO statistics */ 48 #define RQF_IO_STAT ((__force req_flags_t)(1 << 13)) 49 /* runtime pm request */ 50 #define RQF_PM ((__force req_flags_t)(1 << 15)) 51 /* on IO scheduler merge hash */ 52 #define RQF_HASHED ((__force req_flags_t)(1 << 16)) 53 /* track IO completion time */ 54 #define RQF_STATS ((__force req_flags_t)(1 << 17)) 55 /* Look at ->special_vec for the actual data payload instead of the 56 bio chain. */ 57 #define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18)) 58 /* The per-zone write lock is held for this request */ 59 #define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19)) 60 /* already slept for hybrid poll */ 61 #define RQF_MQ_POLL_SLEPT ((__force req_flags_t)(1 << 20)) 62 /* ->timeout has been called, don't expire again */ 63 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21)) 64 /* queue has elevator attached */ 65 #define RQF_ELV ((__force req_flags_t)(1 << 22)) 66 #define RQF_RESV ((__force req_flags_t)(1 << 23)) 67 68 /* flags that prevent us from merging requests: */ 69 #define RQF_NOMERGE_FLAGS \ 70 (RQF_STARTED | RQF_SOFTBARRIER | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD) 71 72 enum mq_rq_state { 73 MQ_RQ_IDLE = 0, 74 MQ_RQ_IN_FLIGHT = 1, 75 MQ_RQ_COMPLETE = 2, 76 }; 77 78 /* 79 * Try to put the fields that are referenced together in the same cacheline. 80 * 81 * If you modify this structure, make sure to update blk_rq_init() and 82 * especially blk_mq_rq_ctx_init() to take care of the added fields. 83 */ 84 struct request { 85 struct request_queue *q; 86 struct blk_mq_ctx *mq_ctx; 87 struct blk_mq_hw_ctx *mq_hctx; 88 89 blk_opf_t cmd_flags; /* op and common flags */ 90 req_flags_t rq_flags; 91 92 int tag; 93 int internal_tag; 94 95 unsigned int timeout; 96 97 /* the following two fields are internal, NEVER access directly */ 98 unsigned int __data_len; /* total data len */ 99 sector_t __sector; /* sector cursor */ 100 101 struct bio *bio; 102 struct bio *biotail; 103 104 union { 105 struct list_head queuelist; 106 struct request *rq_next; 107 }; 108 109 struct block_device *part; 110 #ifdef CONFIG_BLK_RQ_ALLOC_TIME 111 /* Time that the first bio started allocating this request. */ 112 u64 alloc_time_ns; 113 #endif 114 /* Time that this request was allocated for this IO. */ 115 u64 start_time_ns; 116 /* Time that I/O was submitted to the device. */ 117 u64 io_start_time_ns; 118 119 #ifdef CONFIG_BLK_WBT 120 unsigned short wbt_flags; 121 #endif 122 /* 123 * rq sectors used for blk stats. It has the same value 124 * with blk_rq_sectors(rq), except that it never be zeroed 125 * by completion. 126 */ 127 unsigned short stats_sectors; 128 129 /* 130 * Number of scatter-gather DMA addr+len pairs after 131 * physical address coalescing is performed. 132 */ 133 unsigned short nr_phys_segments; 134 135 #ifdef CONFIG_BLK_DEV_INTEGRITY 136 unsigned short nr_integrity_segments; 137 #endif 138 139 #ifdef CONFIG_BLK_INLINE_ENCRYPTION 140 struct bio_crypt_ctx *crypt_ctx; 141 struct blk_crypto_keyslot *crypt_keyslot; 142 #endif 143 144 unsigned short ioprio; 145 146 enum mq_rq_state state; 147 atomic_t ref; 148 149 unsigned long deadline; 150 151 /* 152 * The hash is used inside the scheduler, and killed once the 153 * request reaches the dispatch list. The ipi_list is only used 154 * to queue the request for softirq completion, which is long 155 * after the request has been unhashed (and even removed from 156 * the dispatch list). 157 */ 158 union { 159 struct hlist_node hash; /* merge hash */ 160 struct llist_node ipi_list; 161 }; 162 163 /* 164 * The rb_node is only used inside the io scheduler, requests 165 * are pruned when moved to the dispatch queue. So let the 166 * completion_data share space with the rb_node. 167 */ 168 union { 169 struct rb_node rb_node; /* sort/lookup */ 170 struct bio_vec special_vec; 171 void *completion_data; 172 }; 173 174 175 /* 176 * Three pointers are available for the IO schedulers, if they need 177 * more they have to dynamically allocate it. Flush requests are 178 * never put on the IO scheduler. So let the flush fields share 179 * space with the elevator data. 180 */ 181 union { 182 struct { 183 struct io_cq *icq; 184 void *priv[2]; 185 } elv; 186 187 struct { 188 unsigned int seq; 189 struct list_head list; 190 rq_end_io_fn *saved_end_io; 191 } flush; 192 }; 193 194 union { 195 struct __call_single_data csd; 196 u64 fifo_time; 197 }; 198 199 /* 200 * completion callback. 201 */ 202 rq_end_io_fn *end_io; 203 void *end_io_data; 204 }; 205 206 static inline enum req_op req_op(const struct request *req) 207 { 208 return req->cmd_flags & REQ_OP_MASK; 209 } 210 211 static inline bool blk_rq_is_passthrough(struct request *rq) 212 { 213 return blk_op_is_passthrough(req_op(rq)); 214 } 215 216 static inline unsigned short req_get_ioprio(struct request *req) 217 { 218 return req->ioprio; 219 } 220 221 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ) 222 223 #define rq_dma_dir(rq) \ 224 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE) 225 226 #define rq_list_add(listptr, rq) do { \ 227 (rq)->rq_next = *(listptr); \ 228 *(listptr) = rq; \ 229 } while (0) 230 231 #define rq_list_pop(listptr) \ 232 ({ \ 233 struct request *__req = NULL; \ 234 if ((listptr) && *(listptr)) { \ 235 __req = *(listptr); \ 236 *(listptr) = __req->rq_next; \ 237 } \ 238 __req; \ 239 }) 240 241 #define rq_list_peek(listptr) \ 242 ({ \ 243 struct request *__req = NULL; \ 244 if ((listptr) && *(listptr)) \ 245 __req = *(listptr); \ 246 __req; \ 247 }) 248 249 #define rq_list_for_each(listptr, pos) \ 250 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos)) 251 252 #define rq_list_for_each_safe(listptr, pos, nxt) \ 253 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \ 254 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL) 255 256 #define rq_list_next(rq) (rq)->rq_next 257 #define rq_list_empty(list) ((list) == (struct request *) NULL) 258 259 /** 260 * rq_list_move() - move a struct request from one list to another 261 * @src: The source list @rq is currently in 262 * @dst: The destination list that @rq will be appended to 263 * @rq: The request to move 264 * @prev: The request preceding @rq in @src (NULL if @rq is the head) 265 */ 266 static inline void rq_list_move(struct request **src, struct request **dst, 267 struct request *rq, struct request *prev) 268 { 269 if (prev) 270 prev->rq_next = rq->rq_next; 271 else 272 *src = rq->rq_next; 273 rq_list_add(dst, rq); 274 } 275 276 /** 277 * enum blk_eh_timer_return - How the timeout handler should proceed 278 * @BLK_EH_DONE: The block driver completed the command or will complete it at 279 * a later time. 280 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the 281 * request to complete. 282 */ 283 enum blk_eh_timer_return { 284 BLK_EH_DONE, 285 BLK_EH_RESET_TIMER, 286 }; 287 288 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */ 289 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */ 290 291 /** 292 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware 293 * block device 294 */ 295 struct blk_mq_hw_ctx { 296 struct { 297 /** @lock: Protects the dispatch list. */ 298 spinlock_t lock; 299 /** 300 * @dispatch: Used for requests that are ready to be 301 * dispatched to the hardware but for some reason (e.g. lack of 302 * resources) could not be sent to the hardware. As soon as the 303 * driver can send new requests, requests at this list will 304 * be sent first for a fairer dispatch. 305 */ 306 struct list_head dispatch; 307 /** 308 * @state: BLK_MQ_S_* flags. Defines the state of the hw 309 * queue (active, scheduled to restart, stopped). 310 */ 311 unsigned long state; 312 } ____cacheline_aligned_in_smp; 313 314 /** 315 * @run_work: Used for scheduling a hardware queue run at a later time. 316 */ 317 struct delayed_work run_work; 318 /** @cpumask: Map of available CPUs where this hctx can run. */ 319 cpumask_var_t cpumask; 320 /** 321 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU 322 * selection from @cpumask. 323 */ 324 int next_cpu; 325 /** 326 * @next_cpu_batch: Counter of how many works left in the batch before 327 * changing to the next CPU. 328 */ 329 int next_cpu_batch; 330 331 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */ 332 unsigned long flags; 333 334 /** 335 * @sched_data: Pointer owned by the IO scheduler attached to a request 336 * queue. It's up to the IO scheduler how to use this pointer. 337 */ 338 void *sched_data; 339 /** 340 * @queue: Pointer to the request queue that owns this hardware context. 341 */ 342 struct request_queue *queue; 343 /** @fq: Queue of requests that need to perform a flush operation. */ 344 struct blk_flush_queue *fq; 345 346 /** 347 * @driver_data: Pointer to data owned by the block driver that created 348 * this hctx 349 */ 350 void *driver_data; 351 352 /** 353 * @ctx_map: Bitmap for each software queue. If bit is on, there is a 354 * pending request in that software queue. 355 */ 356 struct sbitmap ctx_map; 357 358 /** 359 * @dispatch_from: Software queue to be used when no scheduler was 360 * selected. 361 */ 362 struct blk_mq_ctx *dispatch_from; 363 /** 364 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to 365 * decide if the hw_queue is busy using Exponential Weighted Moving 366 * Average algorithm. 367 */ 368 unsigned int dispatch_busy; 369 370 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */ 371 unsigned short type; 372 /** @nr_ctx: Number of software queues. */ 373 unsigned short nr_ctx; 374 /** @ctxs: Array of software queues. */ 375 struct blk_mq_ctx **ctxs; 376 377 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */ 378 spinlock_t dispatch_wait_lock; 379 /** 380 * @dispatch_wait: Waitqueue to put requests when there is no tag 381 * available at the moment, to wait for another try in the future. 382 */ 383 wait_queue_entry_t dispatch_wait; 384 385 /** 386 * @wait_index: Index of next available dispatch_wait queue to insert 387 * requests. 388 */ 389 atomic_t wait_index; 390 391 /** 392 * @tags: Tags owned by the block driver. A tag at this set is only 393 * assigned when a request is dispatched from a hardware queue. 394 */ 395 struct blk_mq_tags *tags; 396 /** 397 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O 398 * scheduler associated with a request queue, a tag is assigned when 399 * that request is allocated. Else, this member is not used. 400 */ 401 struct blk_mq_tags *sched_tags; 402 403 /** @queued: Number of queued requests. */ 404 unsigned long queued; 405 /** @run: Number of dispatched requests. */ 406 unsigned long run; 407 408 /** @numa_node: NUMA node the storage adapter has been connected to. */ 409 unsigned int numa_node; 410 /** @queue_num: Index of this hardware queue. */ 411 unsigned int queue_num; 412 413 /** 414 * @nr_active: Number of active requests. Only used when a tag set is 415 * shared across request queues. 416 */ 417 atomic_t nr_active; 418 419 /** @cpuhp_online: List to store request if CPU is going to die */ 420 struct hlist_node cpuhp_online; 421 /** @cpuhp_dead: List to store request if some CPU die. */ 422 struct hlist_node cpuhp_dead; 423 /** @kobj: Kernel object for sysfs. */ 424 struct kobject kobj; 425 426 #ifdef CONFIG_BLK_DEBUG_FS 427 /** 428 * @debugfs_dir: debugfs directory for this hardware queue. Named 429 * as cpu<cpu_number>. 430 */ 431 struct dentry *debugfs_dir; 432 /** @sched_debugfs_dir: debugfs directory for the scheduler. */ 433 struct dentry *sched_debugfs_dir; 434 #endif 435 436 /** 437 * @hctx_list: if this hctx is not in use, this is an entry in 438 * q->unused_hctx_list. 439 */ 440 struct list_head hctx_list; 441 }; 442 443 /** 444 * struct blk_mq_queue_map - Map software queues to hardware queues 445 * @mq_map: CPU ID to hardware queue index map. This is an array 446 * with nr_cpu_ids elements. Each element has a value in the range 447 * [@queue_offset, @queue_offset + @nr_queues). 448 * @nr_queues: Number of hardware queues to map CPU IDs onto. 449 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe 450 * driver to map each hardware queue type (enum hctx_type) onto a distinct 451 * set of hardware queues. 452 */ 453 struct blk_mq_queue_map { 454 unsigned int *mq_map; 455 unsigned int nr_queues; 456 unsigned int queue_offset; 457 }; 458 459 /** 460 * enum hctx_type - Type of hardware queue 461 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for. 462 * @HCTX_TYPE_READ: Just for READ I/O. 463 * @HCTX_TYPE_POLL: Polled I/O of any kind. 464 * @HCTX_MAX_TYPES: Number of types of hctx. 465 */ 466 enum hctx_type { 467 HCTX_TYPE_DEFAULT, 468 HCTX_TYPE_READ, 469 HCTX_TYPE_POLL, 470 471 HCTX_MAX_TYPES, 472 }; 473 474 /** 475 * struct blk_mq_tag_set - tag set that can be shared between request queues 476 * @ops: Pointers to functions that implement block driver behavior. 477 * @map: One or more ctx -> hctx mappings. One map exists for each 478 * hardware queue type (enum hctx_type) that the driver wishes 479 * to support. There are no restrictions on maps being of the 480 * same size, and it's perfectly legal to share maps between 481 * types. 482 * @nr_maps: Number of elements in the @map array. A number in the range 483 * [1, HCTX_MAX_TYPES]. 484 * @nr_hw_queues: Number of hardware queues supported by the block driver that 485 * owns this data structure. 486 * @queue_depth: Number of tags per hardware queue, reserved tags included. 487 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag 488 * allocations. 489 * @cmd_size: Number of additional bytes to allocate per request. The block 490 * driver owns these additional bytes. 491 * @numa_node: NUMA node the storage adapter has been connected to. 492 * @timeout: Request processing timeout in jiffies. 493 * @flags: Zero or more BLK_MQ_F_* flags. 494 * @driver_data: Pointer to data owned by the block driver that created this 495 * tag set. 496 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues 497 * elements. 498 * @shared_tags: 499 * Shared set of tags. Has @nr_hw_queues elements. If set, 500 * shared by all @tags. 501 * @tag_list_lock: Serializes tag_list accesses. 502 * @tag_list: List of the request queues that use this tag set. See also 503 * request_queue.tag_set_list. 504 * @srcu: Use as lock when type of the request queue is blocking 505 * (BLK_MQ_F_BLOCKING). 506 */ 507 struct blk_mq_tag_set { 508 const struct blk_mq_ops *ops; 509 struct blk_mq_queue_map map[HCTX_MAX_TYPES]; 510 unsigned int nr_maps; 511 unsigned int nr_hw_queues; 512 unsigned int queue_depth; 513 unsigned int reserved_tags; 514 unsigned int cmd_size; 515 int numa_node; 516 unsigned int timeout; 517 unsigned int flags; 518 void *driver_data; 519 520 struct blk_mq_tags **tags; 521 522 struct blk_mq_tags *shared_tags; 523 524 struct mutex tag_list_lock; 525 struct list_head tag_list; 526 struct srcu_struct *srcu; 527 }; 528 529 /** 530 * struct blk_mq_queue_data - Data about a request inserted in a queue 531 * 532 * @rq: Request pointer. 533 * @last: If it is the last request in the queue. 534 */ 535 struct blk_mq_queue_data { 536 struct request *rq; 537 bool last; 538 }; 539 540 typedef bool (busy_tag_iter_fn)(struct request *, void *); 541 542 /** 543 * struct blk_mq_ops - Callback functions that implements block driver 544 * behaviour. 545 */ 546 struct blk_mq_ops { 547 /** 548 * @queue_rq: Queue a new request from block IO. 549 */ 550 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *, 551 const struct blk_mq_queue_data *); 552 553 /** 554 * @commit_rqs: If a driver uses bd->last to judge when to submit 555 * requests to hardware, it must define this function. In case of errors 556 * that make us stop issuing further requests, this hook serves the 557 * purpose of kicking the hardware (which the last request otherwise 558 * would have done). 559 */ 560 void (*commit_rqs)(struct blk_mq_hw_ctx *); 561 562 /** 563 * @queue_rqs: Queue a list of new requests. Driver is guaranteed 564 * that each request belongs to the same queue. If the driver doesn't 565 * empty the @rqlist completely, then the rest will be queued 566 * individually by the block layer upon return. 567 */ 568 void (*queue_rqs)(struct request **rqlist); 569 570 /** 571 * @get_budget: Reserve budget before queue request, once .queue_rq is 572 * run, it is driver's responsibility to release the 573 * reserved budget. Also we have to handle failure case 574 * of .get_budget for avoiding I/O deadlock. 575 */ 576 int (*get_budget)(struct request_queue *); 577 578 /** 579 * @put_budget: Release the reserved budget. 580 */ 581 void (*put_budget)(struct request_queue *, int); 582 583 /** 584 * @set_rq_budget_token: store rq's budget token 585 */ 586 void (*set_rq_budget_token)(struct request *, int); 587 /** 588 * @get_rq_budget_token: retrieve rq's budget token 589 */ 590 int (*get_rq_budget_token)(struct request *); 591 592 /** 593 * @timeout: Called on request timeout. 594 */ 595 enum blk_eh_timer_return (*timeout)(struct request *); 596 597 /** 598 * @poll: Called to poll for completion of a specific tag. 599 */ 600 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *); 601 602 /** 603 * @complete: Mark the request as complete. 604 */ 605 void (*complete)(struct request *); 606 607 /** 608 * @init_hctx: Called when the block layer side of a hardware queue has 609 * been set up, allowing the driver to allocate/init matching 610 * structures. 611 */ 612 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int); 613 /** 614 * @exit_hctx: Ditto for exit/teardown. 615 */ 616 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int); 617 618 /** 619 * @init_request: Called for every command allocated by the block layer 620 * to allow the driver to set up driver specific data. 621 * 622 * Tag greater than or equal to queue_depth is for setting up 623 * flush request. 624 */ 625 int (*init_request)(struct blk_mq_tag_set *set, struct request *, 626 unsigned int, unsigned int); 627 /** 628 * @exit_request: Ditto for exit/teardown. 629 */ 630 void (*exit_request)(struct blk_mq_tag_set *set, struct request *, 631 unsigned int); 632 633 /** 634 * @cleanup_rq: Called before freeing one request which isn't completed 635 * yet, and usually for freeing the driver private data. 636 */ 637 void (*cleanup_rq)(struct request *); 638 639 /** 640 * @busy: If set, returns whether or not this queue currently is busy. 641 */ 642 bool (*busy)(struct request_queue *); 643 644 /** 645 * @map_queues: This allows drivers specify their own queue mapping by 646 * overriding the setup-time function that builds the mq_map. 647 */ 648 void (*map_queues)(struct blk_mq_tag_set *set); 649 650 #ifdef CONFIG_BLK_DEBUG_FS 651 /** 652 * @show_rq: Used by the debugfs implementation to show driver-specific 653 * information about a request. 654 */ 655 void (*show_rq)(struct seq_file *m, struct request *rq); 656 #endif 657 }; 658 659 enum { 660 BLK_MQ_F_SHOULD_MERGE = 1 << 0, 661 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1, 662 /* 663 * Set when this device requires underlying blk-mq device for 664 * completing IO: 665 */ 666 BLK_MQ_F_STACKING = 1 << 2, 667 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3, 668 BLK_MQ_F_BLOCKING = 1 << 5, 669 /* Do not allow an I/O scheduler to be configured. */ 670 BLK_MQ_F_NO_SCHED = 1 << 6, 671 /* 672 * Select 'none' during queue registration in case of a single hwq 673 * or shared hwqs instead of 'mq-deadline'. 674 */ 675 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7, 676 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8, 677 BLK_MQ_F_ALLOC_POLICY_BITS = 1, 678 679 BLK_MQ_S_STOPPED = 0, 680 BLK_MQ_S_TAG_ACTIVE = 1, 681 BLK_MQ_S_SCHED_RESTART = 2, 682 683 /* hw queue is inactive after all its CPUs become offline */ 684 BLK_MQ_S_INACTIVE = 3, 685 686 BLK_MQ_MAX_DEPTH = 10240, 687 688 BLK_MQ_CPU_WORK_BATCH = 8, 689 }; 690 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \ 691 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \ 692 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) 693 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \ 694 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \ 695 << BLK_MQ_F_ALLOC_POLICY_START_BIT) 696 697 #define BLK_MQ_NO_HCTX_IDX (-1U) 698 699 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata, 700 struct lock_class_key *lkclass); 701 #define blk_mq_alloc_disk(set, queuedata) \ 702 ({ \ 703 static struct lock_class_key __key; \ 704 \ 705 __blk_mq_alloc_disk(set, queuedata, &__key); \ 706 }) 707 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q, 708 struct lock_class_key *lkclass); 709 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *); 710 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set, 711 struct request_queue *q); 712 void blk_mq_destroy_queue(struct request_queue *); 713 714 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set); 715 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set, 716 const struct blk_mq_ops *ops, unsigned int queue_depth, 717 unsigned int set_flags); 718 void blk_mq_free_tag_set(struct blk_mq_tag_set *set); 719 720 void blk_mq_free_request(struct request *rq); 721 722 bool blk_mq_queue_inflight(struct request_queue *q); 723 724 enum { 725 /* return when out of requests */ 726 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0), 727 /* allocate from reserved pool */ 728 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1), 729 /* set RQF_PM */ 730 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2), 731 }; 732 733 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf, 734 blk_mq_req_flags_t flags); 735 struct request *blk_mq_alloc_request_hctx(struct request_queue *q, 736 blk_opf_t opf, blk_mq_req_flags_t flags, 737 unsigned int hctx_idx); 738 739 /* 740 * Tag address space map. 741 */ 742 struct blk_mq_tags { 743 unsigned int nr_tags; 744 unsigned int nr_reserved_tags; 745 746 atomic_t active_queues; 747 748 struct sbitmap_queue bitmap_tags; 749 struct sbitmap_queue breserved_tags; 750 751 struct request **rqs; 752 struct request **static_rqs; 753 struct list_head page_list; 754 755 /* 756 * used to clear request reference in rqs[] before freeing one 757 * request pool 758 */ 759 spinlock_t lock; 760 }; 761 762 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, 763 unsigned int tag) 764 { 765 if (tag < tags->nr_tags) { 766 prefetch(tags->rqs[tag]); 767 return tags->rqs[tag]; 768 } 769 770 return NULL; 771 } 772 773 enum { 774 BLK_MQ_UNIQUE_TAG_BITS = 16, 775 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1, 776 }; 777 778 u32 blk_mq_unique_tag(struct request *rq); 779 780 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag) 781 { 782 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS; 783 } 784 785 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag) 786 { 787 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK; 788 } 789 790 /** 791 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request 792 * @rq: target request. 793 */ 794 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq) 795 { 796 return READ_ONCE(rq->state); 797 } 798 799 static inline int blk_mq_request_started(struct request *rq) 800 { 801 return blk_mq_rq_state(rq) != MQ_RQ_IDLE; 802 } 803 804 static inline int blk_mq_request_completed(struct request *rq) 805 { 806 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE; 807 } 808 809 /* 810 * 811 * Set the state to complete when completing a request from inside ->queue_rq. 812 * This is used by drivers that want to ensure special complete actions that 813 * need access to the request are called on failure, e.g. by nvme for 814 * multipathing. 815 */ 816 static inline void blk_mq_set_request_complete(struct request *rq) 817 { 818 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE); 819 } 820 821 /* 822 * Complete the request directly instead of deferring it to softirq or 823 * completing it another CPU. Useful in preemptible instead of an interrupt. 824 */ 825 static inline void blk_mq_complete_request_direct(struct request *rq, 826 void (*complete)(struct request *rq)) 827 { 828 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE); 829 complete(rq); 830 } 831 832 void blk_mq_start_request(struct request *rq); 833 void blk_mq_end_request(struct request *rq, blk_status_t error); 834 void __blk_mq_end_request(struct request *rq, blk_status_t error); 835 void blk_mq_end_request_batch(struct io_comp_batch *ib); 836 837 /* 838 * Only need start/end time stamping if we have iostat or 839 * blk stats enabled, or using an IO scheduler. 840 */ 841 static inline bool blk_mq_need_time_stamp(struct request *rq) 842 { 843 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_ELV)); 844 } 845 846 static inline bool blk_mq_is_reserved_rq(struct request *rq) 847 { 848 return rq->rq_flags & RQF_RESV; 849 } 850 851 /* 852 * Batched completions only work when there is no I/O error and no special 853 * ->end_io handler. 854 */ 855 static inline bool blk_mq_add_to_batch(struct request *req, 856 struct io_comp_batch *iob, int ioerror, 857 void (*complete)(struct io_comp_batch *)) 858 { 859 if (!iob || (req->rq_flags & RQF_ELV) || ioerror || 860 (req->end_io && !blk_rq_is_passthrough(req))) 861 return false; 862 863 if (!iob->complete) 864 iob->complete = complete; 865 else if (iob->complete != complete) 866 return false; 867 iob->need_ts |= blk_mq_need_time_stamp(req); 868 rq_list_add(&iob->req_list, req); 869 return true; 870 } 871 872 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list); 873 void blk_mq_kick_requeue_list(struct request_queue *q); 874 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs); 875 void blk_mq_complete_request(struct request *rq); 876 bool blk_mq_complete_request_remote(struct request *rq); 877 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx); 878 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx); 879 void blk_mq_stop_hw_queues(struct request_queue *q); 880 void blk_mq_start_hw_queues(struct request_queue *q); 881 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 882 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async); 883 void blk_mq_quiesce_queue(struct request_queue *q); 884 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set); 885 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set); 886 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set); 887 void blk_mq_unquiesce_queue(struct request_queue *q); 888 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs); 889 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async); 890 void blk_mq_run_hw_queues(struct request_queue *q, bool async); 891 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs); 892 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset, 893 busy_tag_iter_fn *fn, void *priv); 894 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset); 895 void blk_mq_freeze_queue(struct request_queue *q); 896 void blk_mq_unfreeze_queue(struct request_queue *q); 897 void blk_freeze_queue_start(struct request_queue *q); 898 void blk_mq_freeze_queue_wait(struct request_queue *q); 899 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q, 900 unsigned long timeout); 901 902 void blk_mq_map_queues(struct blk_mq_queue_map *qmap); 903 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues); 904 905 void blk_mq_quiesce_queue_nowait(struct request_queue *q); 906 907 unsigned int blk_mq_rq_cpu(struct request *rq); 908 909 bool __blk_should_fake_timeout(struct request_queue *q); 910 static inline bool blk_should_fake_timeout(struct request_queue *q) 911 { 912 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) && 913 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags)) 914 return __blk_should_fake_timeout(q); 915 return false; 916 } 917 918 /** 919 * blk_mq_rq_from_pdu - cast a PDU to a request 920 * @pdu: the PDU (Protocol Data Unit) to be casted 921 * 922 * Return: request 923 * 924 * Driver command data is immediately after the request. So subtract request 925 * size to get back to the original request. 926 */ 927 static inline struct request *blk_mq_rq_from_pdu(void *pdu) 928 { 929 return pdu - sizeof(struct request); 930 } 931 932 /** 933 * blk_mq_rq_to_pdu - cast a request to a PDU 934 * @rq: the request to be casted 935 * 936 * Return: pointer to the PDU 937 * 938 * Driver command data is immediately after the request. So add request to get 939 * the PDU. 940 */ 941 static inline void *blk_mq_rq_to_pdu(struct request *rq) 942 { 943 return rq + 1; 944 } 945 946 #define queue_for_each_hw_ctx(q, hctx, i) \ 947 xa_for_each(&(q)->hctx_table, (i), (hctx)) 948 949 #define hctx_for_each_ctx(hctx, ctx, i) \ 950 for ((i) = 0; (i) < (hctx)->nr_ctx && \ 951 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++) 952 953 static inline void blk_mq_cleanup_rq(struct request *rq) 954 { 955 if (rq->q->mq_ops->cleanup_rq) 956 rq->q->mq_ops->cleanup_rq(rq); 957 } 958 959 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio, 960 unsigned int nr_segs) 961 { 962 rq->nr_phys_segments = nr_segs; 963 rq->__data_len = bio->bi_iter.bi_size; 964 rq->bio = rq->biotail = bio; 965 rq->ioprio = bio_prio(bio); 966 } 967 968 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx, 969 struct lock_class_key *key); 970 971 static inline bool rq_is_sync(struct request *rq) 972 { 973 return op_is_sync(rq->cmd_flags); 974 } 975 976 void blk_rq_init(struct request_queue *q, struct request *rq); 977 int blk_rq_prep_clone(struct request *rq, struct request *rq_src, 978 struct bio_set *bs, gfp_t gfp_mask, 979 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data); 980 void blk_rq_unprep_clone(struct request *rq); 981 blk_status_t blk_insert_cloned_request(struct request *rq); 982 983 struct rq_map_data { 984 struct page **pages; 985 unsigned long offset; 986 unsigned short page_order; 987 unsigned short nr_entries; 988 bool null_mapped; 989 bool from_user; 990 }; 991 992 int blk_rq_map_user(struct request_queue *, struct request *, 993 struct rq_map_data *, void __user *, unsigned long, gfp_t); 994 int blk_rq_map_user_io(struct request *, struct rq_map_data *, 995 void __user *, unsigned long, gfp_t, bool, int, bool, int); 996 int blk_rq_map_user_iov(struct request_queue *, struct request *, 997 struct rq_map_data *, const struct iov_iter *, gfp_t); 998 int blk_rq_unmap_user(struct bio *); 999 int blk_rq_map_kern(struct request_queue *, struct request *, void *, 1000 unsigned int, gfp_t); 1001 int blk_rq_append_bio(struct request *rq, struct bio *bio); 1002 void blk_execute_rq_nowait(struct request *rq, bool at_head); 1003 blk_status_t blk_execute_rq(struct request *rq, bool at_head); 1004 bool blk_rq_is_poll(struct request *rq); 1005 1006 struct req_iterator { 1007 struct bvec_iter iter; 1008 struct bio *bio; 1009 }; 1010 1011 #define __rq_for_each_bio(_bio, rq) \ 1012 if ((rq->bio)) \ 1013 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next) 1014 1015 #define rq_for_each_segment(bvl, _rq, _iter) \ 1016 __rq_for_each_bio(_iter.bio, _rq) \ 1017 bio_for_each_segment(bvl, _iter.bio, _iter.iter) 1018 1019 #define rq_for_each_bvec(bvl, _rq, _iter) \ 1020 __rq_for_each_bio(_iter.bio, _rq) \ 1021 bio_for_each_bvec(bvl, _iter.bio, _iter.iter) 1022 1023 #define rq_iter_last(bvec, _iter) \ 1024 (_iter.bio->bi_next == NULL && \ 1025 bio_iter_last(bvec, _iter.iter)) 1026 1027 /* 1028 * blk_rq_pos() : the current sector 1029 * blk_rq_bytes() : bytes left in the entire request 1030 * blk_rq_cur_bytes() : bytes left in the current segment 1031 * blk_rq_sectors() : sectors left in the entire request 1032 * blk_rq_cur_sectors() : sectors left in the current segment 1033 * blk_rq_stats_sectors() : sectors of the entire request used for stats 1034 */ 1035 static inline sector_t blk_rq_pos(const struct request *rq) 1036 { 1037 return rq->__sector; 1038 } 1039 1040 static inline unsigned int blk_rq_bytes(const struct request *rq) 1041 { 1042 return rq->__data_len; 1043 } 1044 1045 static inline int blk_rq_cur_bytes(const struct request *rq) 1046 { 1047 if (!rq->bio) 1048 return 0; 1049 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */ 1050 return rq->bio->bi_iter.bi_size; 1051 return bio_iovec(rq->bio).bv_len; 1052 } 1053 1054 static inline unsigned int blk_rq_sectors(const struct request *rq) 1055 { 1056 return blk_rq_bytes(rq) >> SECTOR_SHIFT; 1057 } 1058 1059 static inline unsigned int blk_rq_cur_sectors(const struct request *rq) 1060 { 1061 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT; 1062 } 1063 1064 static inline unsigned int blk_rq_stats_sectors(const struct request *rq) 1065 { 1066 return rq->stats_sectors; 1067 } 1068 1069 /* 1070 * Some commands like WRITE SAME have a payload or data transfer size which 1071 * is different from the size of the request. Any driver that supports such 1072 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to 1073 * calculate the data transfer size. 1074 */ 1075 static inline unsigned int blk_rq_payload_bytes(struct request *rq) 1076 { 1077 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1078 return rq->special_vec.bv_len; 1079 return blk_rq_bytes(rq); 1080 } 1081 1082 /* 1083 * Return the first full biovec in the request. The caller needs to check that 1084 * there are any bvecs before calling this helper. 1085 */ 1086 static inline struct bio_vec req_bvec(struct request *rq) 1087 { 1088 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1089 return rq->special_vec; 1090 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter); 1091 } 1092 1093 static inline unsigned int blk_rq_count_bios(struct request *rq) 1094 { 1095 unsigned int nr_bios = 0; 1096 struct bio *bio; 1097 1098 __rq_for_each_bio(bio, rq) 1099 nr_bios++; 1100 1101 return nr_bios; 1102 } 1103 1104 void blk_steal_bios(struct bio_list *list, struct request *rq); 1105 1106 /* 1107 * Request completion related functions. 1108 * 1109 * blk_update_request() completes given number of bytes and updates 1110 * the request without completing it. 1111 */ 1112 bool blk_update_request(struct request *rq, blk_status_t error, 1113 unsigned int nr_bytes); 1114 void blk_abort_request(struct request *); 1115 1116 /* 1117 * Number of physical segments as sent to the device. 1118 * 1119 * Normally this is the number of discontiguous data segments sent by the 1120 * submitter. But for data-less command like discard we might have no 1121 * actual data segments submitted, but the driver might have to add it's 1122 * own special payload. In that case we still return 1 here so that this 1123 * special payload will be mapped. 1124 */ 1125 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq) 1126 { 1127 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD) 1128 return 1; 1129 return rq->nr_phys_segments; 1130 } 1131 1132 /* 1133 * Number of discard segments (or ranges) the driver needs to fill in. 1134 * Each discard bio merged into a request is counted as one segment. 1135 */ 1136 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq) 1137 { 1138 return max_t(unsigned short, rq->nr_phys_segments, 1); 1139 } 1140 1141 int __blk_rq_map_sg(struct request_queue *q, struct request *rq, 1142 struct scatterlist *sglist, struct scatterlist **last_sg); 1143 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq, 1144 struct scatterlist *sglist) 1145 { 1146 struct scatterlist *last_sg = NULL; 1147 1148 return __blk_rq_map_sg(q, rq, sglist, &last_sg); 1149 } 1150 void blk_dump_rq_flags(struct request *, char *); 1151 1152 #ifdef CONFIG_BLK_DEV_ZONED 1153 static inline unsigned int blk_rq_zone_no(struct request *rq) 1154 { 1155 return disk_zone_no(rq->q->disk, blk_rq_pos(rq)); 1156 } 1157 1158 static inline unsigned int blk_rq_zone_is_seq(struct request *rq) 1159 { 1160 return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq)); 1161 } 1162 1163 bool blk_req_needs_zone_write_lock(struct request *rq); 1164 bool blk_req_zone_write_trylock(struct request *rq); 1165 void __blk_req_zone_write_lock(struct request *rq); 1166 void __blk_req_zone_write_unlock(struct request *rq); 1167 1168 static inline void blk_req_zone_write_lock(struct request *rq) 1169 { 1170 if (blk_req_needs_zone_write_lock(rq)) 1171 __blk_req_zone_write_lock(rq); 1172 } 1173 1174 static inline void blk_req_zone_write_unlock(struct request *rq) 1175 { 1176 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED) 1177 __blk_req_zone_write_unlock(rq); 1178 } 1179 1180 static inline bool blk_req_zone_is_write_locked(struct request *rq) 1181 { 1182 return rq->q->disk->seq_zones_wlock && 1183 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock); 1184 } 1185 1186 static inline bool blk_req_can_dispatch_to_zone(struct request *rq) 1187 { 1188 if (!blk_req_needs_zone_write_lock(rq)) 1189 return true; 1190 return !blk_req_zone_is_write_locked(rq); 1191 } 1192 #else /* CONFIG_BLK_DEV_ZONED */ 1193 static inline bool blk_req_needs_zone_write_lock(struct request *rq) 1194 { 1195 return false; 1196 } 1197 1198 static inline void blk_req_zone_write_lock(struct request *rq) 1199 { 1200 } 1201 1202 static inline void blk_req_zone_write_unlock(struct request *rq) 1203 { 1204 } 1205 static inline bool blk_req_zone_is_write_locked(struct request *rq) 1206 { 1207 return false; 1208 } 1209 1210 static inline bool blk_req_can_dispatch_to_zone(struct request *rq) 1211 { 1212 return true; 1213 } 1214 #endif /* CONFIG_BLK_DEV_ZONED */ 1215 1216 #endif /* BLK_MQ_H */ 1217