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
12struct blk_mq_tags;
13struct blk_flush_queue;
14
15#define BLKDEV_MIN_RQ 4
16#define BLKDEV_DEFAULT_RQ 128
17
18enum rq_end_io_ret {
19 RQ_END_IO_NONE,
20 RQ_END_IO_FREE,
21};
22
23typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
24
25/*
26 * request flags */
27typedef __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/* request for flush sequence */
32#define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4))
33/* merge of different types, fail separately */
34#define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5))
35/* don't call prep for this one */
36#define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
37/* use hctx->sched_tags */
38#define RQF_SCHED_TAGS ((__force req_flags_t)(1 << 8))
39/* use an I/O scheduler for this request */
40#define RQF_USE_SCHED ((__force req_flags_t)(1 << 9))
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/* account into disk and partition IO statistics */
46#define RQF_IO_STAT ((__force req_flags_t)(1 << 13))
47/* runtime pm request */
48#define RQF_PM ((__force req_flags_t)(1 << 15))
49/* on IO scheduler merge hash */
50#define RQF_HASHED ((__force req_flags_t)(1 << 16))
51/* track IO completion time */
52#define RQF_STATS ((__force req_flags_t)(1 << 17))
53/* Look at ->special_vec for the actual data payload instead of the
54 bio chain. */
55#define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18))
56/* The per-zone write lock is held for this request */
57#define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19))
58/* ->timeout has been called, don't expire again */
59#define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
60#define RQF_RESV ((__force req_flags_t)(1 << 23))
61
62/* flags that prevent us from merging requests: */
63#define RQF_NOMERGE_FLAGS \
64 (RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
65
66enum mq_rq_state {
67 MQ_RQ_IDLE = 0,
68 MQ_RQ_IN_FLIGHT = 1,
69 MQ_RQ_COMPLETE = 2,
70};
71
72/*
73 * Try to put the fields that are referenced together in the same cacheline.
74 *
75 * If you modify this structure, make sure to update blk_rq_init() and
76 * especially blk_mq_rq_ctx_init() to take care of the added fields.
77 */
78struct request {
79 struct request_queue *q;
80 struct blk_mq_ctx *mq_ctx;
81 struct blk_mq_hw_ctx *mq_hctx;
82
83 blk_opf_t cmd_flags; /* op and common flags */
84 req_flags_t rq_flags;
85
86 int tag;
87 int internal_tag;
88
89 unsigned int timeout;
90
91 /* the following two fields are internal, NEVER access directly */
92 unsigned int __data_len; /* total data len */
93 sector_t __sector; /* sector cursor */
94
95 struct bio *bio;
96 struct bio *biotail;
97
98 union {
99 struct list_head queuelist;
100 struct request *rq_next;
101 };
102
103 struct block_device *part;
104#ifdef CONFIG_BLK_RQ_ALLOC_TIME
105 /* Time that the first bio started allocating this request. */
106 u64 alloc_time_ns;
107#endif
108 /* Time that this request was allocated for this IO. */
109 u64 start_time_ns;
110 /* Time that I/O was submitted to the device. */
111 u64 io_start_time_ns;
112
113#ifdef CONFIG_BLK_WBT
114 unsigned short wbt_flags;
115#endif
116 /*
117 * rq sectors used for blk stats. It has the same value
118 * with blk_rq_sectors(rq), except that it never be zeroed
119 * by completion.
120 */
121 unsigned short stats_sectors;
122
123 /*
124 * Number of scatter-gather DMA addr+len pairs after
125 * physical address coalescing is performed.
126 */
127 unsigned short nr_phys_segments;
128
129#ifdef CONFIG_BLK_DEV_INTEGRITY
130 unsigned short nr_integrity_segments;
131#endif
132
133#ifdef CONFIG_BLK_INLINE_ENCRYPTION
134 struct bio_crypt_ctx *crypt_ctx;
135 struct blk_crypto_keyslot *crypt_keyslot;
136#endif
137
138 unsigned short ioprio;
139
140 enum mq_rq_state state;
141 atomic_t ref;
142
143 unsigned long deadline;
144
145 /*
146 * The hash is used inside the scheduler, and killed once the
147 * request reaches the dispatch list. The ipi_list is only used
148 * to queue the request for softirq completion, which is long
149 * after the request has been unhashed (and even removed from
150 * the dispatch list).
151 */
152 union {
153 struct hlist_node hash; /* merge hash */
154 struct llist_node ipi_list;
155 };
156
157 /*
158 * The rb_node is only used inside the io scheduler, requests
159 * are pruned when moved to the dispatch queue. special_vec must
160 * only be used if RQF_SPECIAL_PAYLOAD is set, and those cannot be
161 * insert into an IO scheduler.
162 */
163 union {
164 struct rb_node rb_node; /* sort/lookup */
165 struct bio_vec special_vec;
166 };
167
168 /*
169 * Three pointers are available for the IO schedulers, if they need
170 * more they have to dynamically allocate it.
171 */
172 struct {
173 struct io_cq *icq;
174 void *priv[2];
175 } elv;
176
177 struct {
178 unsigned int seq;
179 rq_end_io_fn *saved_end_io;
180 } flush;
181
182 u64 fifo_time;
183
184 /*
185 * completion callback.
186 */
187 rq_end_io_fn *end_io;
188 void *end_io_data;
189};
190
191static inline enum req_op req_op(const struct request *req)
192{
193 return req->cmd_flags & REQ_OP_MASK;
194}
195
196static inline bool blk_rq_is_passthrough(struct request *rq)
197{
198 return blk_op_is_passthrough(op: rq->cmd_flags);
199}
200
201static inline unsigned short req_get_ioprio(struct request *req)
202{
203 return req->ioprio;
204}
205
206#define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
207
208#define rq_dma_dir(rq) \
209 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
210
211#define rq_list_add(listptr, rq) do { \
212 (rq)->rq_next = *(listptr); \
213 *(listptr) = rq; \
214} while (0)
215
216#define rq_list_add_tail(lastpptr, rq) do { \
217 (rq)->rq_next = NULL; \
218 **(lastpptr) = rq; \
219 *(lastpptr) = &rq->rq_next; \
220} while (0)
221
222#define rq_list_pop(listptr) \
223({ \
224 struct request *__req = NULL; \
225 if ((listptr) && *(listptr)) { \
226 __req = *(listptr); \
227 *(listptr) = __req->rq_next; \
228 } \
229 __req; \
230})
231
232#define rq_list_peek(listptr) \
233({ \
234 struct request *__req = NULL; \
235 if ((listptr) && *(listptr)) \
236 __req = *(listptr); \
237 __req; \
238})
239
240#define rq_list_for_each(listptr, pos) \
241 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
242
243#define rq_list_for_each_safe(listptr, pos, nxt) \
244 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
245 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
246
247#define rq_list_next(rq) (rq)->rq_next
248#define rq_list_empty(list) ((list) == (struct request *) NULL)
249
250/**
251 * rq_list_move() - move a struct request from one list to another
252 * @src: The source list @rq is currently in
253 * @dst: The destination list that @rq will be appended to
254 * @rq: The request to move
255 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
256 */
257static inline void rq_list_move(struct request **src, struct request **dst,
258 struct request *rq, struct request *prev)
259{
260 if (prev)
261 prev->rq_next = rq->rq_next;
262 else
263 *src = rq->rq_next;
264 rq_list_add(dst, rq);
265}
266
267/**
268 * enum blk_eh_timer_return - How the timeout handler should proceed
269 * @BLK_EH_DONE: The block driver completed the command or will complete it at
270 * a later time.
271 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
272 * request to complete.
273 */
274enum blk_eh_timer_return {
275 BLK_EH_DONE,
276 BLK_EH_RESET_TIMER,
277};
278
279#define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
280#define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
281
282/**
283 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
284 * block device
285 */
286struct blk_mq_hw_ctx {
287 struct {
288 /** @lock: Protects the dispatch list. */
289 spinlock_t lock;
290 /**
291 * @dispatch: Used for requests that are ready to be
292 * dispatched to the hardware but for some reason (e.g. lack of
293 * resources) could not be sent to the hardware. As soon as the
294 * driver can send new requests, requests at this list will
295 * be sent first for a fairer dispatch.
296 */
297 struct list_head dispatch;
298 /**
299 * @state: BLK_MQ_S_* flags. Defines the state of the hw
300 * queue (active, scheduled to restart, stopped).
301 */
302 unsigned long state;
303 } ____cacheline_aligned_in_smp;
304
305 /**
306 * @run_work: Used for scheduling a hardware queue run at a later time.
307 */
308 struct delayed_work run_work;
309 /** @cpumask: Map of available CPUs where this hctx can run. */
310 cpumask_var_t cpumask;
311 /**
312 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
313 * selection from @cpumask.
314 */
315 int next_cpu;
316 /**
317 * @next_cpu_batch: Counter of how many works left in the batch before
318 * changing to the next CPU.
319 */
320 int next_cpu_batch;
321
322 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
323 unsigned long flags;
324
325 /**
326 * @sched_data: Pointer owned by the IO scheduler attached to a request
327 * queue. It's up to the IO scheduler how to use this pointer.
328 */
329 void *sched_data;
330 /**
331 * @queue: Pointer to the request queue that owns this hardware context.
332 */
333 struct request_queue *queue;
334 /** @fq: Queue of requests that need to perform a flush operation. */
335 struct blk_flush_queue *fq;
336
337 /**
338 * @driver_data: Pointer to data owned by the block driver that created
339 * this hctx
340 */
341 void *driver_data;
342
343 /**
344 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
345 * pending request in that software queue.
346 */
347 struct sbitmap ctx_map;
348
349 /**
350 * @dispatch_from: Software queue to be used when no scheduler was
351 * selected.
352 */
353 struct blk_mq_ctx *dispatch_from;
354 /**
355 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
356 * decide if the hw_queue is busy using Exponential Weighted Moving
357 * Average algorithm.
358 */
359 unsigned int dispatch_busy;
360
361 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
362 unsigned short type;
363 /** @nr_ctx: Number of software queues. */
364 unsigned short nr_ctx;
365 /** @ctxs: Array of software queues. */
366 struct blk_mq_ctx **ctxs;
367
368 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
369 spinlock_t dispatch_wait_lock;
370 /**
371 * @dispatch_wait: Waitqueue to put requests when there is no tag
372 * available at the moment, to wait for another try in the future.
373 */
374 wait_queue_entry_t dispatch_wait;
375
376 /**
377 * @wait_index: Index of next available dispatch_wait queue to insert
378 * requests.
379 */
380 atomic_t wait_index;
381
382 /**
383 * @tags: Tags owned by the block driver. A tag at this set is only
384 * assigned when a request is dispatched from a hardware queue.
385 */
386 struct blk_mq_tags *tags;
387 /**
388 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
389 * scheduler associated with a request queue, a tag is assigned when
390 * that request is allocated. Else, this member is not used.
391 */
392 struct blk_mq_tags *sched_tags;
393
394 /** @run: Number of dispatched requests. */
395 unsigned long run;
396
397 /** @numa_node: NUMA node the storage adapter has been connected to. */
398 unsigned int numa_node;
399 /** @queue_num: Index of this hardware queue. */
400 unsigned int queue_num;
401
402 /**
403 * @nr_active: Number of active requests. Only used when a tag set is
404 * shared across request queues.
405 */
406 atomic_t nr_active;
407
408 /** @cpuhp_online: List to store request if CPU is going to die */
409 struct hlist_node cpuhp_online;
410 /** @cpuhp_dead: List to store request if some CPU die. */
411 struct hlist_node cpuhp_dead;
412 /** @kobj: Kernel object for sysfs. */
413 struct kobject kobj;
414
415#ifdef CONFIG_BLK_DEBUG_FS
416 /**
417 * @debugfs_dir: debugfs directory for this hardware queue. Named
418 * as cpu<cpu_number>.
419 */
420 struct dentry *debugfs_dir;
421 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
422 struct dentry *sched_debugfs_dir;
423#endif
424
425 /**
426 * @hctx_list: if this hctx is not in use, this is an entry in
427 * q->unused_hctx_list.
428 */
429 struct list_head hctx_list;
430};
431
432/**
433 * struct blk_mq_queue_map - Map software queues to hardware queues
434 * @mq_map: CPU ID to hardware queue index map. This is an array
435 * with nr_cpu_ids elements. Each element has a value in the range
436 * [@queue_offset, @queue_offset + @nr_queues).
437 * @nr_queues: Number of hardware queues to map CPU IDs onto.
438 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
439 * driver to map each hardware queue type (enum hctx_type) onto a distinct
440 * set of hardware queues.
441 */
442struct blk_mq_queue_map {
443 unsigned int *mq_map;
444 unsigned int nr_queues;
445 unsigned int queue_offset;
446};
447
448/**
449 * enum hctx_type - Type of hardware queue
450 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
451 * @HCTX_TYPE_READ: Just for READ I/O.
452 * @HCTX_TYPE_POLL: Polled I/O of any kind.
453 * @HCTX_MAX_TYPES: Number of types of hctx.
454 */
455enum hctx_type {
456 HCTX_TYPE_DEFAULT,
457 HCTX_TYPE_READ,
458 HCTX_TYPE_POLL,
459
460 HCTX_MAX_TYPES,
461};
462
463/**
464 * struct blk_mq_tag_set - tag set that can be shared between request queues
465 * @ops: Pointers to functions that implement block driver behavior.
466 * @map: One or more ctx -> hctx mappings. One map exists for each
467 * hardware queue type (enum hctx_type) that the driver wishes
468 * to support. There are no restrictions on maps being of the
469 * same size, and it's perfectly legal to share maps between
470 * types.
471 * @nr_maps: Number of elements in the @map array. A number in the range
472 * [1, HCTX_MAX_TYPES].
473 * @nr_hw_queues: Number of hardware queues supported by the block driver that
474 * owns this data structure.
475 * @queue_depth: Number of tags per hardware queue, reserved tags included.
476 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
477 * allocations.
478 * @cmd_size: Number of additional bytes to allocate per request. The block
479 * driver owns these additional bytes.
480 * @numa_node: NUMA node the storage adapter has been connected to.
481 * @timeout: Request processing timeout in jiffies.
482 * @flags: Zero or more BLK_MQ_F_* flags.
483 * @driver_data: Pointer to data owned by the block driver that created this
484 * tag set.
485 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
486 * elements.
487 * @shared_tags:
488 * Shared set of tags. Has @nr_hw_queues elements. If set,
489 * shared by all @tags.
490 * @tag_list_lock: Serializes tag_list accesses.
491 * @tag_list: List of the request queues that use this tag set. See also
492 * request_queue.tag_set_list.
493 * @srcu: Use as lock when type of the request queue is blocking
494 * (BLK_MQ_F_BLOCKING).
495 */
496struct blk_mq_tag_set {
497 const struct blk_mq_ops *ops;
498 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
499 unsigned int nr_maps;
500 unsigned int nr_hw_queues;
501 unsigned int queue_depth;
502 unsigned int reserved_tags;
503 unsigned int cmd_size;
504 int numa_node;
505 unsigned int timeout;
506 unsigned int flags;
507 void *driver_data;
508
509 struct blk_mq_tags **tags;
510
511 struct blk_mq_tags *shared_tags;
512
513 struct mutex tag_list_lock;
514 struct list_head tag_list;
515 struct srcu_struct *srcu;
516};
517
518/**
519 * struct blk_mq_queue_data - Data about a request inserted in a queue
520 *
521 * @rq: Request pointer.
522 * @last: If it is the last request in the queue.
523 */
524struct blk_mq_queue_data {
525 struct request *rq;
526 bool last;
527};
528
529typedef bool (busy_tag_iter_fn)(struct request *, void *);
530
531/**
532 * struct blk_mq_ops - Callback functions that implements block driver
533 * behaviour.
534 */
535struct blk_mq_ops {
536 /**
537 * @queue_rq: Queue a new request from block IO.
538 */
539 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
540 const struct blk_mq_queue_data *);
541
542 /**
543 * @commit_rqs: If a driver uses bd->last to judge when to submit
544 * requests to hardware, it must define this function. In case of errors
545 * that make us stop issuing further requests, this hook serves the
546 * purpose of kicking the hardware (which the last request otherwise
547 * would have done).
548 */
549 void (*commit_rqs)(struct blk_mq_hw_ctx *);
550
551 /**
552 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
553 * that each request belongs to the same queue. If the driver doesn't
554 * empty the @rqlist completely, then the rest will be queued
555 * individually by the block layer upon return.
556 */
557 void (*queue_rqs)(struct request **rqlist);
558
559 /**
560 * @get_budget: Reserve budget before queue request, once .queue_rq is
561 * run, it is driver's responsibility to release the
562 * reserved budget. Also we have to handle failure case
563 * of .get_budget for avoiding I/O deadlock.
564 */
565 int (*get_budget)(struct request_queue *);
566
567 /**
568 * @put_budget: Release the reserved budget.
569 */
570 void (*put_budget)(struct request_queue *, int);
571
572 /**
573 * @set_rq_budget_token: store rq's budget token
574 */
575 void (*set_rq_budget_token)(struct request *, int);
576 /**
577 * @get_rq_budget_token: retrieve rq's budget token
578 */
579 int (*get_rq_budget_token)(struct request *);
580
581 /**
582 * @timeout: Called on request timeout.
583 */
584 enum blk_eh_timer_return (*timeout)(struct request *);
585
586 /**
587 * @poll: Called to poll for completion of a specific tag.
588 */
589 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
590
591 /**
592 * @complete: Mark the request as complete.
593 */
594 void (*complete)(struct request *);
595
596 /**
597 * @init_hctx: Called when the block layer side of a hardware queue has
598 * been set up, allowing the driver to allocate/init matching
599 * structures.
600 */
601 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
602 /**
603 * @exit_hctx: Ditto for exit/teardown.
604 */
605 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
606
607 /**
608 * @init_request: Called for every command allocated by the block layer
609 * to allow the driver to set up driver specific data.
610 *
611 * Tag greater than or equal to queue_depth is for setting up
612 * flush request.
613 */
614 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
615 unsigned int, unsigned int);
616 /**
617 * @exit_request: Ditto for exit/teardown.
618 */
619 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
620 unsigned int);
621
622 /**
623 * @cleanup_rq: Called before freeing one request which isn't completed
624 * yet, and usually for freeing the driver private data.
625 */
626 void (*cleanup_rq)(struct request *);
627
628 /**
629 * @busy: If set, returns whether or not this queue currently is busy.
630 */
631 bool (*busy)(struct request_queue *);
632
633 /**
634 * @map_queues: This allows drivers specify their own queue mapping by
635 * overriding the setup-time function that builds the mq_map.
636 */
637 void (*map_queues)(struct blk_mq_tag_set *set);
638
639#ifdef CONFIG_BLK_DEBUG_FS
640 /**
641 * @show_rq: Used by the debugfs implementation to show driver-specific
642 * information about a request.
643 */
644 void (*show_rq)(struct seq_file *m, struct request *rq);
645#endif
646};
647
648enum {
649 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
650 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
651 /*
652 * Set when this device requires underlying blk-mq device for
653 * completing IO:
654 */
655 BLK_MQ_F_STACKING = 1 << 2,
656 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
657 BLK_MQ_F_BLOCKING = 1 << 5,
658 /* Do not allow an I/O scheduler to be configured. */
659 BLK_MQ_F_NO_SCHED = 1 << 6,
660 /*
661 * Select 'none' during queue registration in case of a single hwq
662 * or shared hwqs instead of 'mq-deadline'.
663 */
664 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
665 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
666 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
667
668 BLK_MQ_S_STOPPED = 0,
669 BLK_MQ_S_TAG_ACTIVE = 1,
670 BLK_MQ_S_SCHED_RESTART = 2,
671
672 /* hw queue is inactive after all its CPUs become offline */
673 BLK_MQ_S_INACTIVE = 3,
674
675 BLK_MQ_MAX_DEPTH = 10240,
676
677 BLK_MQ_CPU_WORK_BATCH = 8,
678};
679#define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
680 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
681 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
682#define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
683 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
684 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
685
686#define BLK_MQ_NO_HCTX_IDX (-1U)
687
688struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
689 struct lock_class_key *lkclass);
690#define blk_mq_alloc_disk(set, queuedata) \
691({ \
692 static struct lock_class_key __key; \
693 \
694 __blk_mq_alloc_disk(set, queuedata, &__key); \
695})
696struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
697 struct lock_class_key *lkclass);
698struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
699int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
700 struct request_queue *q);
701void blk_mq_destroy_queue(struct request_queue *);
702
703int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
704int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
705 const struct blk_mq_ops *ops, unsigned int queue_depth,
706 unsigned int set_flags);
707void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
708
709void blk_mq_free_request(struct request *rq);
710int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
711 unsigned int poll_flags);
712
713bool blk_mq_queue_inflight(struct request_queue *q);
714
715enum {
716 /* return when out of requests */
717 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
718 /* allocate from reserved pool */
719 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
720 /* set RQF_PM */
721 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
722};
723
724struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
725 blk_mq_req_flags_t flags);
726struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
727 blk_opf_t opf, blk_mq_req_flags_t flags,
728 unsigned int hctx_idx);
729
730/*
731 * Tag address space map.
732 */
733struct blk_mq_tags {
734 unsigned int nr_tags;
735 unsigned int nr_reserved_tags;
736 unsigned int active_queues;
737
738 struct sbitmap_queue bitmap_tags;
739 struct sbitmap_queue breserved_tags;
740
741 struct request **rqs;
742 struct request **static_rqs;
743 struct list_head page_list;
744
745 /*
746 * used to clear request reference in rqs[] before freeing one
747 * request pool
748 */
749 spinlock_t lock;
750};
751
752static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
753 unsigned int tag)
754{
755 if (tag < tags->nr_tags) {
756 prefetch(tags->rqs[tag]);
757 return tags->rqs[tag];
758 }
759
760 return NULL;
761}
762
763enum {
764 BLK_MQ_UNIQUE_TAG_BITS = 16,
765 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
766};
767
768u32 blk_mq_unique_tag(struct request *rq);
769
770static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
771{
772 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
773}
774
775static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
776{
777 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
778}
779
780/**
781 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
782 * @rq: target request.
783 */
784static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
785{
786 return READ_ONCE(rq->state);
787}
788
789static inline int blk_mq_request_started(struct request *rq)
790{
791 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
792}
793
794static inline int blk_mq_request_completed(struct request *rq)
795{
796 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
797}
798
799/*
800 *
801 * Set the state to complete when completing a request from inside ->queue_rq.
802 * This is used by drivers that want to ensure special complete actions that
803 * need access to the request are called on failure, e.g. by nvme for
804 * multipathing.
805 */
806static inline void blk_mq_set_request_complete(struct request *rq)
807{
808 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
809}
810
811/*
812 * Complete the request directly instead of deferring it to softirq or
813 * completing it another CPU. Useful in preemptible instead of an interrupt.
814 */
815static inline void blk_mq_complete_request_direct(struct request *rq,
816 void (*complete)(struct request *rq))
817{
818 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
819 complete(rq);
820}
821
822void blk_mq_start_request(struct request *rq);
823void blk_mq_end_request(struct request *rq, blk_status_t error);
824void __blk_mq_end_request(struct request *rq, blk_status_t error);
825void blk_mq_end_request_batch(struct io_comp_batch *ib);
826
827/*
828 * Only need start/end time stamping if we have iostat or
829 * blk stats enabled, or using an IO scheduler.
830 */
831static inline bool blk_mq_need_time_stamp(struct request *rq)
832{
833 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
834}
835
836static inline bool blk_mq_is_reserved_rq(struct request *rq)
837{
838 return rq->rq_flags & RQF_RESV;
839}
840
841/*
842 * Batched completions only work when there is no I/O error and no special
843 * ->end_io handler.
844 */
845static inline bool blk_mq_add_to_batch(struct request *req,
846 struct io_comp_batch *iob, int ioerror,
847 void (*complete)(struct io_comp_batch *))
848{
849 /*
850 * blk_mq_end_request_batch() can't end request allocated from
851 * sched tags
852 */
853 if (!iob || (req->rq_flags & RQF_SCHED_TAGS) || ioerror ||
854 (req->end_io && !blk_rq_is_passthrough(rq: req)))
855 return false;
856
857 if (!iob->complete)
858 iob->complete = complete;
859 else if (iob->complete != complete)
860 return false;
861 iob->need_ts |= blk_mq_need_time_stamp(rq: req);
862 rq_list_add(&iob->req_list, req);
863 return true;
864}
865
866void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
867void blk_mq_kick_requeue_list(struct request_queue *q);
868void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
869void blk_mq_complete_request(struct request *rq);
870bool blk_mq_complete_request_remote(struct request *rq);
871void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
872void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
873void blk_mq_stop_hw_queues(struct request_queue *q);
874void blk_mq_start_hw_queues(struct request_queue *q);
875void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
876void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
877void blk_mq_quiesce_queue(struct request_queue *q);
878void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
879void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
880void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
881void blk_mq_unquiesce_queue(struct request_queue *q);
882void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
883void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
884void blk_mq_run_hw_queues(struct request_queue *q, bool async);
885void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
886void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
887 busy_tag_iter_fn *fn, void *priv);
888void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
889void blk_mq_freeze_queue(struct request_queue *q);
890void blk_mq_unfreeze_queue(struct request_queue *q);
891void blk_freeze_queue_start(struct request_queue *q);
892void blk_mq_freeze_queue_wait(struct request_queue *q);
893int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
894 unsigned long timeout);
895
896void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
897void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
898
899void blk_mq_quiesce_queue_nowait(struct request_queue *q);
900
901unsigned int blk_mq_rq_cpu(struct request *rq);
902
903bool __blk_should_fake_timeout(struct request_queue *q);
904static inline bool blk_should_fake_timeout(struct request_queue *q)
905{
906 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
907 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
908 return __blk_should_fake_timeout(q);
909 return false;
910}
911
912/**
913 * blk_mq_rq_from_pdu - cast a PDU to a request
914 * @pdu: the PDU (Protocol Data Unit) to be casted
915 *
916 * Return: request
917 *
918 * Driver command data is immediately after the request. So subtract request
919 * size to get back to the original request.
920 */
921static inline struct request *blk_mq_rq_from_pdu(void *pdu)
922{
923 return pdu - sizeof(struct request);
924}
925
926/**
927 * blk_mq_rq_to_pdu - cast a request to a PDU
928 * @rq: the request to be casted
929 *
930 * Return: pointer to the PDU
931 *
932 * Driver command data is immediately after the request. So add request to get
933 * the PDU.
934 */
935static inline void *blk_mq_rq_to_pdu(struct request *rq)
936{
937 return rq + 1;
938}
939
940#define queue_for_each_hw_ctx(q, hctx, i) \
941 xa_for_each(&(q)->hctx_table, (i), (hctx))
942
943#define hctx_for_each_ctx(hctx, ctx, i) \
944 for ((i) = 0; (i) < (hctx)->nr_ctx && \
945 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
946
947static inline void blk_mq_cleanup_rq(struct request *rq)
948{
949 if (rq->q->mq_ops->cleanup_rq)
950 rq->q->mq_ops->cleanup_rq(rq);
951}
952
953static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
954 unsigned int nr_segs)
955{
956 rq->nr_phys_segments = nr_segs;
957 rq->__data_len = bio->bi_iter.bi_size;
958 rq->bio = rq->biotail = bio;
959 rq->ioprio = bio_prio(bio);
960}
961
962void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
963 struct lock_class_key *key);
964
965static inline bool rq_is_sync(struct request *rq)
966{
967 return op_is_sync(op: rq->cmd_flags);
968}
969
970void blk_rq_init(struct request_queue *q, struct request *rq);
971int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
972 struct bio_set *bs, gfp_t gfp_mask,
973 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
974void blk_rq_unprep_clone(struct request *rq);
975blk_status_t blk_insert_cloned_request(struct request *rq);
976
977struct rq_map_data {
978 struct page **pages;
979 unsigned long offset;
980 unsigned short page_order;
981 unsigned short nr_entries;
982 bool null_mapped;
983 bool from_user;
984};
985
986int blk_rq_map_user(struct request_queue *, struct request *,
987 struct rq_map_data *, void __user *, unsigned long, gfp_t);
988int blk_rq_map_user_io(struct request *, struct rq_map_data *,
989 void __user *, unsigned long, gfp_t, bool, int, bool, int);
990int blk_rq_map_user_iov(struct request_queue *, struct request *,
991 struct rq_map_data *, const struct iov_iter *, gfp_t);
992int blk_rq_unmap_user(struct bio *);
993int blk_rq_map_kern(struct request_queue *, struct request *, void *,
994 unsigned int, gfp_t);
995int blk_rq_append_bio(struct request *rq, struct bio *bio);
996void blk_execute_rq_nowait(struct request *rq, bool at_head);
997blk_status_t blk_execute_rq(struct request *rq, bool at_head);
998bool blk_rq_is_poll(struct request *rq);
999
1000struct req_iterator {
1001 struct bvec_iter iter;
1002 struct bio *bio;
1003};
1004
1005#define __rq_for_each_bio(_bio, rq) \
1006 if ((rq->bio)) \
1007 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1008
1009#define rq_for_each_segment(bvl, _rq, _iter) \
1010 __rq_for_each_bio(_iter.bio, _rq) \
1011 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1012
1013#define rq_for_each_bvec(bvl, _rq, _iter) \
1014 __rq_for_each_bio(_iter.bio, _rq) \
1015 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1016
1017#define rq_iter_last(bvec, _iter) \
1018 (_iter.bio->bi_next == NULL && \
1019 bio_iter_last(bvec, _iter.iter))
1020
1021/*
1022 * blk_rq_pos() : the current sector
1023 * blk_rq_bytes() : bytes left in the entire request
1024 * blk_rq_cur_bytes() : bytes left in the current segment
1025 * blk_rq_sectors() : sectors left in the entire request
1026 * blk_rq_cur_sectors() : sectors left in the current segment
1027 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1028 */
1029static inline sector_t blk_rq_pos(const struct request *rq)
1030{
1031 return rq->__sector;
1032}
1033
1034static inline unsigned int blk_rq_bytes(const struct request *rq)
1035{
1036 return rq->__data_len;
1037}
1038
1039static inline int blk_rq_cur_bytes(const struct request *rq)
1040{
1041 if (!rq->bio)
1042 return 0;
1043 if (!bio_has_data(bio: rq->bio)) /* dataless requests such as discard */
1044 return rq->bio->bi_iter.bi_size;
1045 return bio_iovec(rq->bio).bv_len;
1046}
1047
1048static inline unsigned int blk_rq_sectors(const struct request *rq)
1049{
1050 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1051}
1052
1053static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1054{
1055 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1056}
1057
1058static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1059{
1060 return rq->stats_sectors;
1061}
1062
1063/*
1064 * Some commands like WRITE SAME have a payload or data transfer size which
1065 * is different from the size of the request. Any driver that supports such
1066 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1067 * calculate the data transfer size.
1068 */
1069static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1070{
1071 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1072 return rq->special_vec.bv_len;
1073 return blk_rq_bytes(rq);
1074}
1075
1076/*
1077 * Return the first full biovec in the request. The caller needs to check that
1078 * there are any bvecs before calling this helper.
1079 */
1080static inline struct bio_vec req_bvec(struct request *rq)
1081{
1082 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1083 return rq->special_vec;
1084 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1085}
1086
1087static inline unsigned int blk_rq_count_bios(struct request *rq)
1088{
1089 unsigned int nr_bios = 0;
1090 struct bio *bio;
1091
1092 __rq_for_each_bio(bio, rq)
1093 nr_bios++;
1094
1095 return nr_bios;
1096}
1097
1098void blk_steal_bios(struct bio_list *list, struct request *rq);
1099
1100/*
1101 * Request completion related functions.
1102 *
1103 * blk_update_request() completes given number of bytes and updates
1104 * the request without completing it.
1105 */
1106bool blk_update_request(struct request *rq, blk_status_t error,
1107 unsigned int nr_bytes);
1108void blk_abort_request(struct request *);
1109
1110/*
1111 * Number of physical segments as sent to the device.
1112 *
1113 * Normally this is the number of discontiguous data segments sent by the
1114 * submitter. But for data-less command like discard we might have no
1115 * actual data segments submitted, but the driver might have to add it's
1116 * own special payload. In that case we still return 1 here so that this
1117 * special payload will be mapped.
1118 */
1119static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1120{
1121 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1122 return 1;
1123 return rq->nr_phys_segments;
1124}
1125
1126/*
1127 * Number of discard segments (or ranges) the driver needs to fill in.
1128 * Each discard bio merged into a request is counted as one segment.
1129 */
1130static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1131{
1132 return max_t(unsigned short, rq->nr_phys_segments, 1);
1133}
1134
1135int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1136 struct scatterlist *sglist, struct scatterlist **last_sg);
1137static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1138 struct scatterlist *sglist)
1139{
1140 struct scatterlist *last_sg = NULL;
1141
1142 return __blk_rq_map_sg(q, rq, sglist, last_sg: &last_sg);
1143}
1144void blk_dump_rq_flags(struct request *, char *);
1145
1146#ifdef CONFIG_BLK_DEV_ZONED
1147static inline unsigned int blk_rq_zone_no(struct request *rq)
1148{
1149 return disk_zone_no(disk: rq->q->disk, sector: blk_rq_pos(rq));
1150}
1151
1152static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1153{
1154 return disk_zone_is_seq(disk: rq->q->disk, sector: blk_rq_pos(rq));
1155}
1156
1157/**
1158 * blk_rq_is_seq_zoned_write() - Check if @rq requires write serialization.
1159 * @rq: Request to examine.
1160 *
1161 * Note: REQ_OP_ZONE_APPEND requests do not require serialization.
1162 */
1163static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1164{
1165 return op_needs_zoned_write_locking(op: req_op(req: rq)) &&
1166 blk_rq_zone_is_seq(rq);
1167}
1168
1169bool blk_req_needs_zone_write_lock(struct request *rq);
1170bool blk_req_zone_write_trylock(struct request *rq);
1171void __blk_req_zone_write_lock(struct request *rq);
1172void __blk_req_zone_write_unlock(struct request *rq);
1173
1174static inline void blk_req_zone_write_lock(struct request *rq)
1175{
1176 if (blk_req_needs_zone_write_lock(rq))
1177 __blk_req_zone_write_lock(rq);
1178}
1179
1180static inline void blk_req_zone_write_unlock(struct request *rq)
1181{
1182 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1183 __blk_req_zone_write_unlock(rq);
1184}
1185
1186static inline bool blk_req_zone_is_write_locked(struct request *rq)
1187{
1188 return rq->q->disk->seq_zones_wlock &&
1189 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock);
1190}
1191
1192static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1193{
1194 if (!blk_req_needs_zone_write_lock(rq))
1195 return true;
1196 return !blk_req_zone_is_write_locked(rq);
1197}
1198#else /* CONFIG_BLK_DEV_ZONED */
1199static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1200{
1201 return false;
1202}
1203
1204static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1205{
1206 return false;
1207}
1208
1209static inline void blk_req_zone_write_lock(struct request *rq)
1210{
1211}
1212
1213static inline void blk_req_zone_write_unlock(struct request *rq)
1214{
1215}
1216static inline bool blk_req_zone_is_write_locked(struct request *rq)
1217{
1218 return false;
1219}
1220
1221static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1222{
1223 return true;
1224}
1225#endif /* CONFIG_BLK_DEV_ZONED */
1226
1227#endif /* BLK_MQ_H */
1228

source code of linux/include/linux/blk-mq.h