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#include <linux/rw_hint.h>
12#include <linux/rwsem.h>
13
14struct blk_mq_tags;
15struct blk_flush_queue;
16
17#define BLKDEV_MIN_RQ 4
18#define BLKDEV_DEFAULT_RQ 128
19
20enum rq_end_io_ret {
21 RQ_END_IO_NONE,
22 RQ_END_IO_FREE,
23};
24
25typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
26
27/*
28 * request flags */
29typedef __u32 __bitwise req_flags_t;
30
31/* Keep rqf_name[] in sync with the definitions below */
32enum rqf_flags {
33 /* drive already may have started this one */
34 __RQF_STARTED,
35 /* request for flush sequence */
36 __RQF_FLUSH_SEQ,
37 /* merge of different types, fail separately */
38 __RQF_MIXED_MERGE,
39 /* don't call prep for this one */
40 __RQF_DONTPREP,
41 /* use hctx->sched_tags */
42 __RQF_SCHED_TAGS,
43 /* use an I/O scheduler for this request */
44 __RQF_USE_SCHED,
45 /* vaguely specified driver internal error. Ignored by block layer */
46 __RQF_FAILED,
47 /* don't warn about errors */
48 __RQF_QUIET,
49 /* account into disk and partition IO statistics */
50 __RQF_IO_STAT,
51 /* runtime pm request */
52 __RQF_PM,
53 /* on IO scheduler merge hash */
54 __RQF_HASHED,
55 /* track IO completion time */
56 __RQF_STATS,
57 /* Look at ->special_vec for the actual data payload instead of the
58 bio chain. */
59 __RQF_SPECIAL_PAYLOAD,
60 /* request completion needs to be signaled to zone write plugging. */
61 __RQF_ZONE_WRITE_PLUGGING,
62 /* ->timeout has been called, don't expire again */
63 __RQF_TIMED_OUT,
64 __RQF_RESV,
65 __RQF_BITS
66};
67
68#define RQF_STARTED ((__force req_flags_t)(1 << __RQF_STARTED))
69#define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << __RQF_FLUSH_SEQ))
70#define RQF_MIXED_MERGE ((__force req_flags_t)(1 << __RQF_MIXED_MERGE))
71#define RQF_DONTPREP ((__force req_flags_t)(1 << __RQF_DONTPREP))
72#define RQF_SCHED_TAGS ((__force req_flags_t)(1 << __RQF_SCHED_TAGS))
73#define RQF_USE_SCHED ((__force req_flags_t)(1 << __RQF_USE_SCHED))
74#define RQF_FAILED ((__force req_flags_t)(1 << __RQF_FAILED))
75#define RQF_QUIET ((__force req_flags_t)(1 << __RQF_QUIET))
76#define RQF_IO_STAT ((__force req_flags_t)(1 << __RQF_IO_STAT))
77#define RQF_PM ((__force req_flags_t)(1 << __RQF_PM))
78#define RQF_HASHED ((__force req_flags_t)(1 << __RQF_HASHED))
79#define RQF_STATS ((__force req_flags_t)(1 << __RQF_STATS))
80#define RQF_SPECIAL_PAYLOAD \
81 ((__force req_flags_t)(1 << __RQF_SPECIAL_PAYLOAD))
82#define RQF_ZONE_WRITE_PLUGGING \
83 ((__force req_flags_t)(1 << __RQF_ZONE_WRITE_PLUGGING))
84#define RQF_TIMED_OUT ((__force req_flags_t)(1 << __RQF_TIMED_OUT))
85#define RQF_RESV ((__force req_flags_t)(1 << __RQF_RESV))
86
87/* flags that prevent us from merging requests: */
88#define RQF_NOMERGE_FLAGS \
89 (RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
90
91enum mq_rq_state {
92 MQ_RQ_IDLE = 0,
93 MQ_RQ_IN_FLIGHT = 1,
94 MQ_RQ_COMPLETE = 2,
95};
96
97/*
98 * Try to put the fields that are referenced together in the same cacheline.
99 *
100 * If you modify this structure, make sure to update blk_rq_init() and
101 * especially blk_mq_rq_ctx_init() to take care of the added fields.
102 */
103struct request {
104 struct request_queue *q;
105 struct blk_mq_ctx *mq_ctx;
106 struct blk_mq_hw_ctx *mq_hctx;
107
108 blk_opf_t cmd_flags; /* op and common flags */
109 req_flags_t rq_flags;
110
111 int tag;
112 int internal_tag;
113
114 unsigned int timeout;
115
116 /* the following two fields are internal, NEVER access directly */
117 unsigned int __data_len; /* total data len */
118 sector_t __sector; /* sector cursor */
119
120 struct bio *bio;
121 struct bio *biotail;
122
123 union {
124 struct list_head queuelist;
125 struct request *rq_next;
126 };
127
128 struct block_device *part;
129#ifdef CONFIG_BLK_RQ_ALLOC_TIME
130 /* Time that the first bio started allocating this request. */
131 u64 alloc_time_ns;
132#endif
133 /* Time that this request was allocated for this IO. */
134 u64 start_time_ns;
135 /* Time that I/O was submitted to the device. */
136 u64 io_start_time_ns;
137
138#ifdef CONFIG_BLK_WBT
139 unsigned short wbt_flags;
140#endif
141 /*
142 * rq sectors used for blk stats. It has the same value
143 * with blk_rq_sectors(rq), except that it never be zeroed
144 * by completion.
145 */
146 unsigned short stats_sectors;
147
148 /*
149 * Number of scatter-gather DMA addr+len pairs after
150 * physical address coalescing is performed.
151 */
152 unsigned short nr_phys_segments;
153 unsigned short nr_integrity_segments;
154
155 /*
156 * The lowest set bit for address gaps between physical segments. This
157 * provides information necessary for dma optimization opprotunities,
158 * like for testing if the segments can be coalesced against the
159 * device's iommu granule.
160 */
161 unsigned char phys_gap_bit;
162
163#ifdef CONFIG_BLK_INLINE_ENCRYPTION
164 struct bio_crypt_ctx *crypt_ctx;
165 struct blk_crypto_keyslot *crypt_keyslot;
166#endif
167
168 enum mq_rq_state state;
169 atomic_t ref;
170
171 unsigned long deadline;
172
173 /*
174 * The hash is used inside the scheduler, and killed once the
175 * request reaches the dispatch list. The ipi_list is only used
176 * to queue the request for softirq completion, which is long
177 * after the request has been unhashed (and even removed from
178 * the dispatch list).
179 */
180 union {
181 struct hlist_node hash; /* merge hash */
182 struct llist_node ipi_list;
183 };
184
185 /*
186 * The rb_node is only used inside the io scheduler, requests
187 * are pruned when moved to the dispatch queue. special_vec must
188 * only be used if RQF_SPECIAL_PAYLOAD is set, and those cannot be
189 * insert into an IO scheduler.
190 */
191 union {
192 struct rb_node rb_node; /* sort/lookup */
193 struct bio_vec special_vec;
194 };
195
196 /*
197 * Three pointers are available for the IO schedulers, if they need
198 * more they have to dynamically allocate it.
199 */
200 struct {
201 struct io_cq *icq;
202 void *priv[2];
203 } elv;
204
205 struct {
206 unsigned int seq;
207 rq_end_io_fn *saved_end_io;
208 } flush;
209
210 u64 fifo_time;
211
212 /*
213 * completion callback.
214 */
215 rq_end_io_fn *end_io;
216 void *end_io_data;
217};
218
219/*
220 * Returns a mask with all bits starting at req->phys_gap_bit set to 1.
221 */
222static inline unsigned long req_phys_gap_mask(const struct request *req)
223{
224 return ~(((1 << req->phys_gap_bit) >> 1) - 1);
225}
226
227static inline enum req_op req_op(const struct request *req)
228{
229 return req->cmd_flags & REQ_OP_MASK;
230}
231
232static inline bool blk_rq_is_passthrough(struct request *rq)
233{
234 return blk_op_is_passthrough(op: rq->cmd_flags);
235}
236
237static inline unsigned short req_get_ioprio(struct request *req)
238{
239 if (req->bio)
240 return req->bio->bi_ioprio;
241 return 0;
242}
243
244#define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
245
246#define rq_dma_dir(rq) \
247 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
248
249static inline int rq_list_empty(const struct rq_list *rl)
250{
251 return rl->head == NULL;
252}
253
254static inline void rq_list_init(struct rq_list *rl)
255{
256 rl->head = NULL;
257 rl->tail = NULL;
258}
259
260static inline void rq_list_add_tail(struct rq_list *rl, struct request *rq)
261{
262 rq->rq_next = NULL;
263 if (rl->tail)
264 rl->tail->rq_next = rq;
265 else
266 rl->head = rq;
267 rl->tail = rq;
268}
269
270static inline void rq_list_add_head(struct rq_list *rl, struct request *rq)
271{
272 rq->rq_next = rl->head;
273 rl->head = rq;
274 if (!rl->tail)
275 rl->tail = rq;
276}
277
278static inline struct request *rq_list_pop(struct rq_list *rl)
279{
280 struct request *rq = rl->head;
281
282 if (rq) {
283 rl->head = rl->head->rq_next;
284 if (!rl->head)
285 rl->tail = NULL;
286 rq->rq_next = NULL;
287 }
288
289 return rq;
290}
291
292static inline struct request *rq_list_peek(struct rq_list *rl)
293{
294 return rl->head;
295}
296
297#define rq_list_for_each(rl, pos) \
298 for (pos = rq_list_peek((rl)); (pos); pos = pos->rq_next)
299
300#define rq_list_for_each_safe(rl, pos, nxt) \
301 for (pos = rq_list_peek((rl)), nxt = pos->rq_next; \
302 pos; pos = nxt, nxt = pos ? pos->rq_next : NULL)
303
304/**
305 * enum blk_eh_timer_return - How the timeout handler should proceed
306 * @BLK_EH_DONE: The block driver completed the command or will complete it at
307 * a later time.
308 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
309 * request to complete.
310 */
311enum blk_eh_timer_return {
312 BLK_EH_DONE,
313 BLK_EH_RESET_TIMER,
314};
315
316/**
317 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
318 * block device
319 */
320struct blk_mq_hw_ctx {
321 struct {
322 /** @lock: Protects the dispatch list. */
323 spinlock_t lock;
324 /**
325 * @dispatch: Used for requests that are ready to be
326 * dispatched to the hardware but for some reason (e.g. lack of
327 * resources) could not be sent to the hardware. As soon as the
328 * driver can send new requests, requests at this list will
329 * be sent first for a fairer dispatch.
330 */
331 struct list_head dispatch;
332 /**
333 * @state: BLK_MQ_S_* flags. Defines the state of the hw
334 * queue (active, scheduled to restart, stopped).
335 */
336 unsigned long state;
337 } ____cacheline_aligned_in_smp;
338
339 /**
340 * @run_work: Used for scheduling a hardware queue run at a later time.
341 */
342 struct delayed_work run_work;
343 /** @cpumask: Map of available CPUs where this hctx can run. */
344 cpumask_var_t cpumask;
345 /**
346 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
347 * selection from @cpumask.
348 */
349 int next_cpu;
350 /**
351 * @next_cpu_batch: Counter of how many works left in the batch before
352 * changing to the next CPU.
353 */
354 int next_cpu_batch;
355
356 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
357 unsigned long flags;
358
359 /**
360 * @sched_data: Pointer owned by the IO scheduler attached to a request
361 * queue. It's up to the IO scheduler how to use this pointer.
362 */
363 void *sched_data;
364 /**
365 * @queue: Pointer to the request queue that owns this hardware context.
366 */
367 struct request_queue *queue;
368 /** @fq: Queue of requests that need to perform a flush operation. */
369 struct blk_flush_queue *fq;
370
371 /**
372 * @driver_data: Pointer to data owned by the block driver that created
373 * this hctx
374 */
375 void *driver_data;
376
377 /**
378 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
379 * pending request in that software queue.
380 */
381 struct sbitmap ctx_map;
382
383 /**
384 * @dispatch_from: Software queue to be used when no scheduler was
385 * selected.
386 */
387 struct blk_mq_ctx *dispatch_from;
388 /**
389 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
390 * decide if the hw_queue is busy using Exponential Weighted Moving
391 * Average algorithm.
392 */
393 unsigned int dispatch_busy;
394
395 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
396 unsigned short type;
397 /** @nr_ctx: Number of software queues. */
398 unsigned short nr_ctx;
399 /** @ctxs: Array of software queues. */
400 struct blk_mq_ctx **ctxs;
401
402 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
403 spinlock_t dispatch_wait_lock;
404 /**
405 * @dispatch_wait: Waitqueue to put requests when there is no tag
406 * available at the moment, to wait for another try in the future.
407 */
408 wait_queue_entry_t dispatch_wait;
409
410 /**
411 * @wait_index: Index of next available dispatch_wait queue to insert
412 * requests.
413 */
414 atomic_t wait_index;
415
416 /**
417 * @tags: Tags owned by the block driver. A tag at this set is only
418 * assigned when a request is dispatched from a hardware queue.
419 */
420 struct blk_mq_tags *tags;
421 /**
422 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
423 * scheduler associated with a request queue, a tag is assigned when
424 * that request is allocated. Else, this member is not used.
425 */
426 struct blk_mq_tags *sched_tags;
427
428 /** @numa_node: NUMA node the storage adapter has been connected to. */
429 unsigned int numa_node;
430 /** @queue_num: Index of this hardware queue. */
431 unsigned int queue_num;
432
433 /**
434 * @nr_active: Number of active requests. Only used when a tag set is
435 * shared across request queues.
436 */
437 atomic_t nr_active;
438
439 /** @cpuhp_online: List to store request if CPU is going to die */
440 struct hlist_node cpuhp_online;
441 /** @cpuhp_dead: List to store request if some CPU die. */
442 struct hlist_node cpuhp_dead;
443 /** @kobj: Kernel object for sysfs. */
444 struct kobject kobj;
445
446#ifdef CONFIG_BLK_DEBUG_FS
447 /**
448 * @debugfs_dir: debugfs directory for this hardware queue. Named
449 * as cpu<cpu_number>.
450 */
451 struct dentry *debugfs_dir;
452 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
453 struct dentry *sched_debugfs_dir;
454#endif
455
456 /**
457 * @hctx_list: if this hctx is not in use, this is an entry in
458 * q->unused_hctx_list.
459 */
460 struct list_head hctx_list;
461};
462
463/**
464 * struct blk_mq_queue_map - Map software queues to hardware queues
465 * @mq_map: CPU ID to hardware queue index map. This is an array
466 * with nr_cpu_ids elements. Each element has a value in the range
467 * [@queue_offset, @queue_offset + @nr_queues).
468 * @nr_queues: Number of hardware queues to map CPU IDs onto.
469 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
470 * driver to map each hardware queue type (enum hctx_type) onto a distinct
471 * set of hardware queues.
472 */
473struct blk_mq_queue_map {
474 unsigned int *mq_map;
475 unsigned int nr_queues;
476 unsigned int queue_offset;
477};
478
479/**
480 * enum hctx_type - Type of hardware queue
481 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
482 * @HCTX_TYPE_READ: Just for READ I/O.
483 * @HCTX_TYPE_POLL: Polled I/O of any kind.
484 * @HCTX_MAX_TYPES: Number of types of hctx.
485 */
486enum hctx_type {
487 HCTX_TYPE_DEFAULT,
488 HCTX_TYPE_READ,
489 HCTX_TYPE_POLL,
490
491 HCTX_MAX_TYPES,
492};
493
494/**
495 * struct blk_mq_tag_set - tag set that can be shared between request queues
496 * @ops: Pointers to functions that implement block driver behavior.
497 * @map: One or more ctx -> hctx mappings. One map exists for each
498 * hardware queue type (enum hctx_type) that the driver wishes
499 * to support. There are no restrictions on maps being of the
500 * same size, and it's perfectly legal to share maps between
501 * types.
502 * @nr_maps: Number of elements in the @map array. A number in the range
503 * [1, HCTX_MAX_TYPES].
504 * @nr_hw_queues: Number of hardware queues supported by the block driver that
505 * owns this data structure.
506 * @queue_depth: Number of tags per hardware queue, reserved tags included.
507 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
508 * allocations.
509 * @cmd_size: Number of additional bytes to allocate per request. The block
510 * driver owns these additional bytes.
511 * @numa_node: NUMA node the storage adapter has been connected to.
512 * @timeout: Request processing timeout in jiffies.
513 * @flags: Zero or more BLK_MQ_F_* flags.
514 * @driver_data: Pointer to data owned by the block driver that created this
515 * tag set.
516 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
517 * elements.
518 * @shared_tags:
519 * Shared set of tags. Has @nr_hw_queues elements. If set,
520 * shared by all @tags.
521 * @tag_list_lock: Serializes tag_list accesses.
522 * @tag_list: List of the request queues that use this tag set. See also
523 * request_queue.tag_set_list.
524 * @srcu: Use as lock when type of the request queue is blocking
525 * (BLK_MQ_F_BLOCKING).
526 * @tags_srcu: SRCU used to defer freeing of tags page_list to prevent
527 * use-after-free when iterating tags.
528 * @update_nr_hwq_lock:
529 * Synchronize updating nr_hw_queues with add/del disk &
530 * switching elevator.
531 */
532struct blk_mq_tag_set {
533 const struct blk_mq_ops *ops;
534 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
535 unsigned int nr_maps;
536 unsigned int nr_hw_queues;
537 unsigned int queue_depth;
538 unsigned int reserved_tags;
539 unsigned int cmd_size;
540 int numa_node;
541 unsigned int timeout;
542 unsigned int flags;
543 void *driver_data;
544
545 struct blk_mq_tags **tags;
546
547 struct blk_mq_tags *shared_tags;
548
549 struct mutex tag_list_lock;
550 struct list_head tag_list;
551 struct srcu_struct *srcu;
552 struct srcu_struct tags_srcu;
553
554 struct rw_semaphore update_nr_hwq_lock;
555};
556
557/**
558 * struct blk_mq_queue_data - Data about a request inserted in a queue
559 *
560 * @rq: Request pointer.
561 * @last: If it is the last request in the queue.
562 */
563struct blk_mq_queue_data {
564 struct request *rq;
565 bool last;
566};
567
568typedef bool (busy_tag_iter_fn)(struct request *, void *);
569
570/**
571 * struct blk_mq_ops - Callback functions that implements block driver
572 * behaviour.
573 */
574struct blk_mq_ops {
575 /**
576 * @queue_rq: Queue a new request from block IO.
577 */
578 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
579 const struct blk_mq_queue_data *);
580
581 /**
582 * @commit_rqs: If a driver uses bd->last to judge when to submit
583 * requests to hardware, it must define this function. In case of errors
584 * that make us stop issuing further requests, this hook serves the
585 * purpose of kicking the hardware (which the last request otherwise
586 * would have done).
587 */
588 void (*commit_rqs)(struct blk_mq_hw_ctx *);
589
590 /**
591 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
592 * that each request belongs to the same queue. If the driver doesn't
593 * empty the @rqlist completely, then the rest will be queued
594 * individually by the block layer upon return.
595 */
596 void (*queue_rqs)(struct rq_list *rqlist);
597
598 /**
599 * @get_budget: Reserve budget before queue request, once .queue_rq is
600 * run, it is driver's responsibility to release the
601 * reserved budget. Also we have to handle failure case
602 * of .get_budget for avoiding I/O deadlock.
603 */
604 int (*get_budget)(struct request_queue *);
605
606 /**
607 * @put_budget: Release the reserved budget.
608 */
609 void (*put_budget)(struct request_queue *, int);
610
611 /**
612 * @set_rq_budget_token: store rq's budget token
613 */
614 void (*set_rq_budget_token)(struct request *, int);
615 /**
616 * @get_rq_budget_token: retrieve rq's budget token
617 */
618 int (*get_rq_budget_token)(struct request *);
619
620 /**
621 * @timeout: Called on request timeout.
622 */
623 enum blk_eh_timer_return (*timeout)(struct request *);
624
625 /**
626 * @poll: Called to poll for completion of a specific tag.
627 */
628 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
629
630 /**
631 * @complete: Mark the request as complete.
632 */
633 void (*complete)(struct request *);
634
635 /**
636 * @init_hctx: Called when the block layer side of a hardware queue has
637 * been set up, allowing the driver to allocate/init matching
638 * structures.
639 */
640 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
641 /**
642 * @exit_hctx: Ditto for exit/teardown.
643 */
644 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
645
646 /**
647 * @init_request: Called for every command allocated by the block layer
648 * to allow the driver to set up driver specific data.
649 *
650 * Tag greater than or equal to queue_depth is for setting up
651 * flush request.
652 */
653 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
654 unsigned int, unsigned int);
655 /**
656 * @exit_request: Ditto for exit/teardown.
657 */
658 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
659 unsigned int);
660
661 /**
662 * @cleanup_rq: Called before freeing one request which isn't completed
663 * yet, and usually for freeing the driver private data.
664 */
665 void (*cleanup_rq)(struct request *);
666
667 /**
668 * @busy: If set, returns whether or not this queue currently is busy.
669 */
670 bool (*busy)(struct request_queue *);
671
672 /**
673 * @map_queues: This allows drivers specify their own queue mapping by
674 * overriding the setup-time function that builds the mq_map.
675 */
676 void (*map_queues)(struct blk_mq_tag_set *set);
677
678#ifdef CONFIG_BLK_DEBUG_FS
679 /**
680 * @show_rq: Used by the debugfs implementation to show driver-specific
681 * information about a request.
682 */
683 void (*show_rq)(struct seq_file *m, struct request *rq);
684#endif
685};
686
687/* Keep hctx_flag_name[] in sync with the definitions below */
688enum {
689 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
690 /*
691 * Set when this device requires underlying blk-mq device for
692 * completing IO:
693 */
694 BLK_MQ_F_STACKING = 1 << 2,
695 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
696 BLK_MQ_F_BLOCKING = 1 << 4,
697
698 /*
699 * Alloc tags on a round-robin base instead of the first available one.
700 */
701 BLK_MQ_F_TAG_RR = 1 << 5,
702
703 /*
704 * Select 'none' during queue registration in case of a single hwq
705 * or shared hwqs instead of 'mq-deadline'.
706 */
707 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 6,
708
709 BLK_MQ_F_MAX = 1 << 7,
710};
711
712#define BLK_MQ_MAX_DEPTH (10240)
713#define BLK_MQ_NO_HCTX_IDX (-1U)
714
715enum {
716 /* Keep hctx_state_name[] in sync with the definitions below */
717 BLK_MQ_S_STOPPED,
718 BLK_MQ_S_TAG_ACTIVE,
719 BLK_MQ_S_SCHED_RESTART,
720 /* hw queue is inactive after all its CPUs become offline */
721 BLK_MQ_S_INACTIVE,
722 BLK_MQ_S_MAX
723};
724
725struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set,
726 struct queue_limits *lim, void *queuedata,
727 struct lock_class_key *lkclass);
728#define blk_mq_alloc_disk(set, lim, queuedata) \
729({ \
730 static struct lock_class_key __key; \
731 \
732 __blk_mq_alloc_disk(set, lim, queuedata, &__key); \
733})
734struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
735 struct lock_class_key *lkclass);
736struct request_queue *blk_mq_alloc_queue(struct blk_mq_tag_set *set,
737 struct queue_limits *lim, void *queuedata);
738int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
739 struct request_queue *q);
740void blk_mq_destroy_queue(struct request_queue *);
741
742int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
743int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
744 const struct blk_mq_ops *ops, unsigned int queue_depth,
745 unsigned int set_flags);
746void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
747
748void blk_mq_free_request(struct request *rq);
749int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
750 unsigned int poll_flags);
751
752bool blk_mq_queue_inflight(struct request_queue *q);
753
754enum {
755 /* return when out of requests */
756 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
757 /* allocate from reserved pool */
758 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
759 /* set RQF_PM */
760 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
761};
762
763struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
764 blk_mq_req_flags_t flags);
765struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
766 blk_opf_t opf, blk_mq_req_flags_t flags,
767 unsigned int hctx_idx);
768
769/*
770 * Tag address space map.
771 */
772struct blk_mq_tags {
773 unsigned int nr_tags;
774 unsigned int nr_reserved_tags;
775 unsigned int active_queues;
776
777 struct sbitmap_queue bitmap_tags;
778 struct sbitmap_queue breserved_tags;
779
780 struct request **rqs;
781 struct request **static_rqs;
782 struct list_head page_list;
783
784 /*
785 * used to clear request reference in rqs[] before freeing one
786 * request pool
787 */
788 spinlock_t lock;
789 struct rcu_head rcu_head;
790};
791
792static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
793 unsigned int tag)
794{
795 if (tag < tags->nr_tags) {
796 prefetch(tags->rqs[tag]);
797 return tags->rqs[tag];
798 }
799
800 return NULL;
801}
802
803enum {
804 BLK_MQ_UNIQUE_TAG_BITS = 16,
805 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
806};
807
808u32 blk_mq_unique_tag(struct request *rq);
809
810static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
811{
812 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
813}
814
815static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
816{
817 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
818}
819
820/**
821 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
822 * @rq: target request.
823 */
824static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
825{
826 return READ_ONCE(rq->state);
827}
828
829static inline int blk_mq_request_started(struct request *rq)
830{
831 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
832}
833
834static inline int blk_mq_request_completed(struct request *rq)
835{
836 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
837}
838
839/*
840 *
841 * Set the state to complete when completing a request from inside ->queue_rq.
842 * This is used by drivers that want to ensure special complete actions that
843 * need access to the request are called on failure, e.g. by nvme for
844 * multipathing.
845 */
846static inline void blk_mq_set_request_complete(struct request *rq)
847{
848 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
849}
850
851/*
852 * Complete the request directly instead of deferring it to softirq or
853 * completing it another CPU. Useful in preemptible instead of an interrupt.
854 */
855static inline void blk_mq_complete_request_direct(struct request *rq,
856 void (*complete)(struct request *rq))
857{
858 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
859 complete(rq);
860}
861
862void blk_mq_start_request(struct request *rq);
863void blk_mq_end_request(struct request *rq, blk_status_t error);
864void __blk_mq_end_request(struct request *rq, blk_status_t error);
865void blk_mq_end_request_batch(struct io_comp_batch *ib);
866
867/*
868 * Only need start/end time stamping if we have iostat or
869 * blk stats enabled, or using an IO scheduler.
870 */
871static inline bool blk_mq_need_time_stamp(struct request *rq)
872{
873 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
874}
875
876static inline bool blk_mq_is_reserved_rq(struct request *rq)
877{
878 return rq->rq_flags & RQF_RESV;
879}
880
881/**
882 * blk_mq_add_to_batch() - add a request to the completion batch
883 * @req: The request to add to batch
884 * @iob: The batch to add the request
885 * @is_error: Specify true if the request failed with an error
886 * @complete: The completaion handler for the request
887 *
888 * Batched completions only work when there is no I/O error and no special
889 * ->end_io handler.
890 *
891 * Return: true when the request was added to the batch, otherwise false
892 */
893static inline bool blk_mq_add_to_batch(struct request *req,
894 struct io_comp_batch *iob, bool is_error,
895 void (*complete)(struct io_comp_batch *))
896{
897 /*
898 * Check various conditions that exclude batch processing:
899 * 1) No batch container
900 * 2) Has scheduler data attached
901 * 3) Not a passthrough request and end_io set
902 * 4) Not a passthrough request and failed with an error
903 */
904 if (!iob)
905 return false;
906 if (req->rq_flags & RQF_SCHED_TAGS)
907 return false;
908 if (!blk_rq_is_passthrough(rq: req)) {
909 if (req->end_io)
910 return false;
911 if (is_error)
912 return false;
913 }
914
915 if (!iob->complete)
916 iob->complete = complete;
917 else if (iob->complete != complete)
918 return false;
919 iob->need_ts |= blk_mq_need_time_stamp(rq: req);
920 rq_list_add_tail(rl: &iob->req_list, rq: req);
921 return true;
922}
923
924void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
925void blk_mq_kick_requeue_list(struct request_queue *q);
926void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
927void blk_mq_complete_request(struct request *rq);
928bool blk_mq_complete_request_remote(struct request *rq);
929void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
930void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
931void blk_mq_stop_hw_queues(struct request_queue *q);
932void blk_mq_start_hw_queues(struct request_queue *q);
933void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
934void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
935void blk_mq_quiesce_queue(struct request_queue *q);
936void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
937void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
938void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
939void blk_mq_unquiesce_queue(struct request_queue *q);
940void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
941void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
942void blk_mq_run_hw_queues(struct request_queue *q, bool async);
943void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
944void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
945 busy_tag_iter_fn *fn, void *priv);
946void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
947void blk_mq_freeze_queue_nomemsave(struct request_queue *q);
948void blk_mq_unfreeze_queue_nomemrestore(struct request_queue *q);
949static inline unsigned int __must_check
950blk_mq_freeze_queue(struct request_queue *q)
951{
952 unsigned int memflags = memalloc_noio_save();
953
954 blk_mq_freeze_queue_nomemsave(q);
955 return memflags;
956}
957static inline void
958blk_mq_unfreeze_queue(struct request_queue *q, unsigned int memflags)
959{
960 blk_mq_unfreeze_queue_nomemrestore(q);
961 memalloc_noio_restore(flags: memflags);
962}
963void blk_freeze_queue_start(struct request_queue *q);
964void blk_mq_freeze_queue_wait(struct request_queue *q);
965int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
966 unsigned long timeout);
967void blk_mq_unfreeze_queue_non_owner(struct request_queue *q);
968void blk_freeze_queue_start_non_owner(struct request_queue *q);
969
970unsigned int blk_mq_num_possible_queues(unsigned int max_queues);
971unsigned int blk_mq_num_online_queues(unsigned int max_queues);
972void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
973void blk_mq_map_hw_queues(struct blk_mq_queue_map *qmap,
974 struct device *dev, unsigned int offset);
975void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
976
977void blk_mq_quiesce_queue_nowait(struct request_queue *q);
978
979unsigned int blk_mq_rq_cpu(struct request *rq);
980
981bool __blk_should_fake_timeout(struct request_queue *q);
982static inline bool blk_should_fake_timeout(struct request_queue *q)
983{
984 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
985 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
986 return __blk_should_fake_timeout(q);
987 return false;
988}
989
990/**
991 * blk_mq_rq_from_pdu - cast a PDU to a request
992 * @pdu: the PDU (Protocol Data Unit) to be casted
993 *
994 * Return: request
995 *
996 * Driver command data is immediately after the request. So subtract request
997 * size to get back to the original request.
998 */
999static inline struct request *blk_mq_rq_from_pdu(void *pdu)
1000{
1001 return pdu - sizeof(struct request);
1002}
1003
1004/**
1005 * blk_mq_rq_to_pdu - cast a request to a PDU
1006 * @rq: the request to be casted
1007 *
1008 * Return: pointer to the PDU
1009 *
1010 * Driver command data is immediately after the request. So add request to get
1011 * the PDU.
1012 */
1013static inline void *blk_mq_rq_to_pdu(struct request *rq)
1014{
1015 return rq + 1;
1016}
1017
1018static inline struct blk_mq_hw_ctx *queue_hctx(struct request_queue *q, int id)
1019{
1020 struct blk_mq_hw_ctx *hctx;
1021
1022 rcu_read_lock();
1023 hctx = rcu_dereference(q->queue_hw_ctx)[id];
1024 rcu_read_unlock();
1025
1026 return hctx;
1027}
1028
1029#define queue_for_each_hw_ctx(q, hctx, i) \
1030 for ((i) = 0; (i) < (q)->nr_hw_queues && \
1031 ({ hctx = queue_hctx((q), i); 1; }); (i)++)
1032
1033#define hctx_for_each_ctx(hctx, ctx, i) \
1034 for ((i) = 0; (i) < (hctx)->nr_ctx && \
1035 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
1036
1037static inline void blk_mq_cleanup_rq(struct request *rq)
1038{
1039 if (rq->q->mq_ops->cleanup_rq)
1040 rq->q->mq_ops->cleanup_rq(rq);
1041}
1042
1043void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
1044 struct lock_class_key *key);
1045
1046static inline bool rq_is_sync(struct request *rq)
1047{
1048 return op_is_sync(op: rq->cmd_flags);
1049}
1050
1051void blk_rq_init(struct request_queue *q, struct request *rq);
1052int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1053 struct bio_set *bs, gfp_t gfp_mask,
1054 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
1055void blk_rq_unprep_clone(struct request *rq);
1056blk_status_t blk_insert_cloned_request(struct request *rq);
1057
1058struct rq_map_data {
1059 struct page **pages;
1060 unsigned long offset;
1061 unsigned short page_order;
1062 unsigned short nr_entries;
1063 bool null_mapped;
1064 bool from_user;
1065};
1066
1067int blk_rq_map_user(struct request_queue *, struct request *,
1068 struct rq_map_data *, void __user *, unsigned long, gfp_t);
1069int blk_rq_map_user_io(struct request *, struct rq_map_data *,
1070 void __user *, unsigned long, gfp_t, bool, int, bool, int);
1071int blk_rq_map_user_iov(struct request_queue *, struct request *,
1072 struct rq_map_data *, const struct iov_iter *, gfp_t);
1073int blk_rq_unmap_user(struct bio *);
1074int blk_rq_map_kern(struct request *rq, void *kbuf, unsigned int len,
1075 gfp_t gfp);
1076int blk_rq_append_bio(struct request *rq, struct bio *bio);
1077void blk_execute_rq_nowait(struct request *rq, bool at_head);
1078blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1079bool blk_rq_is_poll(struct request *rq);
1080
1081struct req_iterator {
1082 struct bvec_iter iter;
1083 struct bio *bio;
1084};
1085
1086#define __rq_for_each_bio(_bio, rq) \
1087 if ((rq->bio)) \
1088 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1089
1090#define rq_for_each_segment(bvl, _rq, _iter) \
1091 __rq_for_each_bio(_iter.bio, _rq) \
1092 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1093
1094#define rq_for_each_bvec(bvl, _rq, _iter) \
1095 __rq_for_each_bio(_iter.bio, _rq) \
1096 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1097
1098#define rq_iter_last(bvec, _iter) \
1099 (_iter.bio->bi_next == NULL && \
1100 bio_iter_last(bvec, _iter.iter))
1101
1102/*
1103 * blk_rq_pos() : the current sector
1104 * blk_rq_bytes() : bytes left in the entire request
1105 * blk_rq_cur_bytes() : bytes left in the current segment
1106 * blk_rq_sectors() : sectors left in the entire request
1107 * blk_rq_cur_sectors() : sectors left in the current segment
1108 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1109 */
1110static inline sector_t blk_rq_pos(const struct request *rq)
1111{
1112 return rq->__sector;
1113}
1114
1115static inline unsigned int blk_rq_bytes(const struct request *rq)
1116{
1117 return rq->__data_len;
1118}
1119
1120static inline int blk_rq_cur_bytes(const struct request *rq)
1121{
1122 if (!rq->bio)
1123 return 0;
1124 if (!bio_has_data(bio: rq->bio)) /* dataless requests such as discard */
1125 return rq->bio->bi_iter.bi_size;
1126 return bio_iovec(rq->bio).bv_len;
1127}
1128
1129static inline unsigned int blk_rq_sectors(const struct request *rq)
1130{
1131 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1132}
1133
1134static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1135{
1136 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1137}
1138
1139static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1140{
1141 return rq->stats_sectors;
1142}
1143
1144/*
1145 * Some commands like WRITE SAME have a payload or data transfer size which
1146 * is different from the size of the request. Any driver that supports such
1147 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1148 * calculate the data transfer size.
1149 */
1150static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1151{
1152 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1153 return rq->special_vec.bv_len;
1154 return blk_rq_bytes(rq);
1155}
1156
1157/*
1158 * Return the first full biovec in the request. The caller needs to check that
1159 * there are any bvecs before calling this helper.
1160 */
1161static inline struct bio_vec req_bvec(struct request *rq)
1162{
1163 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1164 return rq->special_vec;
1165 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1166}
1167
1168static inline unsigned int blk_rq_count_bios(struct request *rq)
1169{
1170 unsigned int nr_bios = 0;
1171 struct bio *bio;
1172
1173 __rq_for_each_bio(bio, rq)
1174 nr_bios++;
1175
1176 return nr_bios;
1177}
1178
1179void blk_steal_bios(struct bio_list *list, struct request *rq);
1180
1181/*
1182 * Request completion related functions.
1183 *
1184 * blk_update_request() completes given number of bytes and updates
1185 * the request without completing it.
1186 */
1187bool blk_update_request(struct request *rq, blk_status_t error,
1188 unsigned int nr_bytes);
1189void blk_abort_request(struct request *);
1190
1191/*
1192 * Number of physical segments as sent to the device.
1193 *
1194 * Normally this is the number of discontiguous data segments sent by the
1195 * submitter. But for data-less command like discard we might have no
1196 * actual data segments submitted, but the driver might have to add it's
1197 * own special payload. In that case we still return 1 here so that this
1198 * special payload will be mapped.
1199 */
1200static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1201{
1202 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1203 return 1;
1204 return rq->nr_phys_segments;
1205}
1206
1207/*
1208 * Number of discard segments (or ranges) the driver needs to fill in.
1209 * Each discard bio merged into a request is counted as one segment.
1210 */
1211static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1212{
1213 return max_t(unsigned short, rq->nr_phys_segments, 1);
1214}
1215
1216/**
1217 * blk_rq_nr_bvec - return number of bvecs in a request
1218 * @rq: request to calculate bvecs for
1219 *
1220 * Returns the number of bvecs.
1221 */
1222static inline unsigned int blk_rq_nr_bvec(struct request *rq)
1223{
1224 struct req_iterator rq_iter;
1225 struct bio_vec bv;
1226 unsigned int nr_bvec = 0;
1227
1228 rq_for_each_bvec(bv, rq, rq_iter)
1229 nr_bvec++;
1230
1231 return nr_bvec;
1232}
1233
1234int __blk_rq_map_sg(struct request *rq, struct scatterlist *sglist,
1235 struct scatterlist **last_sg);
1236static inline int blk_rq_map_sg(struct request *rq, struct scatterlist *sglist)
1237{
1238 struct scatterlist *last_sg = NULL;
1239
1240 return __blk_rq_map_sg(rq, sglist, last_sg: &last_sg);
1241}
1242void blk_dump_rq_flags(struct request *, char *);
1243
1244#endif /* BLK_MQ_H */
1245

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