1 /* SPDX-License-Identifier: GPL-2.0 */
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>
13 struct blk_flush_queue;
15 #define BLKDEV_MIN_RQ 4
16 #define BLKDEV_DEFAULT_RQ 128
23 typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
27 typedef __u32 __bitwise req_flags_t;
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 /* track inflight for MQ */
36 #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6))
37 /* don't call prep for this one */
38 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
39 /* use hctx->sched_tags */
40 #define RQF_SCHED_TAGS ((__force req_flags_t)(1 << 8))
41 /* use an I/O scheduler for this request */
42 #define RQF_USE_SCHED ((__force req_flags_t)(1 << 9))
43 /* vaguely specified driver internal error. Ignored by the block layer */
44 #define RQF_FAILED ((__force req_flags_t)(1 << 10))
45 /* don't warn about errors */
46 #define RQF_QUIET ((__force req_flags_t)(1 << 11))
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
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 /* ->timeout has been called, don't expire again */
61 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
62 #define RQF_RESV ((__force req_flags_t)(1 << 23))
64 /* flags that prevent us from merging requests: */
65 #define RQF_NOMERGE_FLAGS \
66 (RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
75 * Try to put the fields that are referenced together in the same cacheline.
77 * If you modify this structure, make sure to update blk_rq_init() and
78 * especially blk_mq_rq_ctx_init() to take care of the added fields.
81 struct request_queue *q;
82 struct blk_mq_ctx *mq_ctx;
83 struct blk_mq_hw_ctx *mq_hctx;
85 blk_opf_t cmd_flags; /* op and common flags */
93 /* the following two fields are internal, NEVER access directly */
94 unsigned int __data_len; /* total data len */
95 sector_t __sector; /* sector cursor */
101 struct list_head queuelist;
102 struct request *rq_next;
105 struct block_device *part;
106 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
107 /* Time that the first bio started allocating this request. */
110 /* Time that this request was allocated for this IO. */
112 /* Time that I/O was submitted to the device. */
113 u64 io_start_time_ns;
115 #ifdef CONFIG_BLK_WBT
116 unsigned short wbt_flags;
119 * rq sectors used for blk stats. It has the same value
120 * with blk_rq_sectors(rq), except that it never be zeroed
123 unsigned short stats_sectors;
126 * Number of scatter-gather DMA addr+len pairs after
127 * physical address coalescing is performed.
129 unsigned short nr_phys_segments;
131 #ifdef CONFIG_BLK_DEV_INTEGRITY
132 unsigned short nr_integrity_segments;
135 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
136 struct bio_crypt_ctx *crypt_ctx;
137 struct blk_crypto_keyslot *crypt_keyslot;
140 unsigned short ioprio;
142 enum mq_rq_state state;
145 unsigned long deadline;
148 * The hash is used inside the scheduler, and killed once the
149 * request reaches the dispatch list. The ipi_list is only used
150 * to queue the request for softirq completion, which is long
151 * after the request has been unhashed (and even removed from
152 * the dispatch list).
155 struct hlist_node hash; /* merge hash */
156 struct llist_node ipi_list;
160 * The rb_node is only used inside the io scheduler, requests
161 * are pruned when moved to the dispatch queue. So let the
162 * completion_data share space with the rb_node.
165 struct rb_node rb_node; /* sort/lookup */
166 struct bio_vec special_vec;
167 void *completion_data;
171 * Three pointers are available for the IO schedulers, if they need
172 * more they have to dynamically allocate it.
181 struct list_head list;
182 rq_end_io_fn *saved_end_io;
186 struct __call_single_data csd;
191 * completion callback.
193 rq_end_io_fn *end_io;
197 static inline enum req_op req_op(const struct request *req)
199 return req->cmd_flags & REQ_OP_MASK;
202 static inline bool blk_rq_is_passthrough(struct request *rq)
204 return blk_op_is_passthrough(rq->cmd_flags);
207 static inline unsigned short req_get_ioprio(struct request *req)
212 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
214 #define rq_dma_dir(rq) \
215 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
217 #define rq_list_add(listptr, rq) do { \
218 (rq)->rq_next = *(listptr); \
222 #define rq_list_add_tail(lastpptr, rq) do { \
223 (rq)->rq_next = NULL; \
225 *(lastpptr) = &rq->rq_next; \
228 #define rq_list_pop(listptr) \
230 struct request *__req = NULL; \
231 if ((listptr) && *(listptr)) { \
232 __req = *(listptr); \
233 *(listptr) = __req->rq_next; \
238 #define rq_list_peek(listptr) \
240 struct request *__req = NULL; \
241 if ((listptr) && *(listptr)) \
242 __req = *(listptr); \
246 #define rq_list_for_each(listptr, pos) \
247 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
249 #define rq_list_for_each_safe(listptr, pos, nxt) \
250 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
251 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
253 #define rq_list_next(rq) (rq)->rq_next
254 #define rq_list_empty(list) ((list) == (struct request *) NULL)
257 * rq_list_move() - move a struct request from one list to another
258 * @src: The source list @rq is currently in
259 * @dst: The destination list that @rq will be appended to
260 * @rq: The request to move
261 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
263 static inline void rq_list_move(struct request **src, struct request **dst,
264 struct request *rq, struct request *prev)
267 prev->rq_next = rq->rq_next;
270 rq_list_add(dst, rq);
274 * enum blk_eh_timer_return - How the timeout handler should proceed
275 * @BLK_EH_DONE: The block driver completed the command or will complete it at
277 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
278 * request to complete.
280 enum blk_eh_timer_return {
285 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
286 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
289 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
292 struct blk_mq_hw_ctx {
294 /** @lock: Protects the dispatch list. */
297 * @dispatch: Used for requests that are ready to be
298 * dispatched to the hardware but for some reason (e.g. lack of
299 * resources) could not be sent to the hardware. As soon as the
300 * driver can send new requests, requests at this list will
301 * be sent first for a fairer dispatch.
303 struct list_head dispatch;
305 * @state: BLK_MQ_S_* flags. Defines the state of the hw
306 * queue (active, scheduled to restart, stopped).
309 } ____cacheline_aligned_in_smp;
312 * @run_work: Used for scheduling a hardware queue run at a later time.
314 struct delayed_work run_work;
315 /** @cpumask: Map of available CPUs where this hctx can run. */
316 cpumask_var_t cpumask;
318 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
319 * selection from @cpumask.
323 * @next_cpu_batch: Counter of how many works left in the batch before
324 * changing to the next CPU.
328 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
332 * @sched_data: Pointer owned by the IO scheduler attached to a request
333 * queue. It's up to the IO scheduler how to use this pointer.
337 * @queue: Pointer to the request queue that owns this hardware context.
339 struct request_queue *queue;
340 /** @fq: Queue of requests that need to perform a flush operation. */
341 struct blk_flush_queue *fq;
344 * @driver_data: Pointer to data owned by the block driver that created
350 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
351 * pending request in that software queue.
353 struct sbitmap ctx_map;
356 * @dispatch_from: Software queue to be used when no scheduler was
359 struct blk_mq_ctx *dispatch_from;
361 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
362 * decide if the hw_queue is busy using Exponential Weighted Moving
365 unsigned int dispatch_busy;
367 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
369 /** @nr_ctx: Number of software queues. */
370 unsigned short nr_ctx;
371 /** @ctxs: Array of software queues. */
372 struct blk_mq_ctx **ctxs;
374 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
375 spinlock_t dispatch_wait_lock;
377 * @dispatch_wait: Waitqueue to put requests when there is no tag
378 * available at the moment, to wait for another try in the future.
380 wait_queue_entry_t dispatch_wait;
383 * @wait_index: Index of next available dispatch_wait queue to insert
389 * @tags: Tags owned by the block driver. A tag at this set is only
390 * assigned when a request is dispatched from a hardware queue.
392 struct blk_mq_tags *tags;
394 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
395 * scheduler associated with a request queue, a tag is assigned when
396 * that request is allocated. Else, this member is not used.
398 struct blk_mq_tags *sched_tags;
400 /** @queued: Number of queued requests. */
401 unsigned long queued;
402 /** @run: Number of dispatched requests. */
405 /** @numa_node: NUMA node the storage adapter has been connected to. */
406 unsigned int numa_node;
407 /** @queue_num: Index of this hardware queue. */
408 unsigned int queue_num;
411 * @nr_active: Number of active requests. Only used when a tag set is
412 * shared across request queues.
416 /** @cpuhp_online: List to store request if CPU is going to die */
417 struct hlist_node cpuhp_online;
418 /** @cpuhp_dead: List to store request if some CPU die. */
419 struct hlist_node cpuhp_dead;
420 /** @kobj: Kernel object for sysfs. */
423 #ifdef CONFIG_BLK_DEBUG_FS
425 * @debugfs_dir: debugfs directory for this hardware queue. Named
426 * as cpu<cpu_number>.
428 struct dentry *debugfs_dir;
429 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
430 struct dentry *sched_debugfs_dir;
434 * @hctx_list: if this hctx is not in use, this is an entry in
435 * q->unused_hctx_list.
437 struct list_head hctx_list;
441 * struct blk_mq_queue_map - Map software queues to hardware queues
442 * @mq_map: CPU ID to hardware queue index map. This is an array
443 * with nr_cpu_ids elements. Each element has a value in the range
444 * [@queue_offset, @queue_offset + @nr_queues).
445 * @nr_queues: Number of hardware queues to map CPU IDs onto.
446 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
447 * driver to map each hardware queue type (enum hctx_type) onto a distinct
448 * set of hardware queues.
450 struct blk_mq_queue_map {
451 unsigned int *mq_map;
452 unsigned int nr_queues;
453 unsigned int queue_offset;
457 * enum hctx_type - Type of hardware queue
458 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
459 * @HCTX_TYPE_READ: Just for READ I/O.
460 * @HCTX_TYPE_POLL: Polled I/O of any kind.
461 * @HCTX_MAX_TYPES: Number of types of hctx.
472 * struct blk_mq_tag_set - tag set that can be shared between request queues
473 * @ops: Pointers to functions that implement block driver behavior.
474 * @map: One or more ctx -> hctx mappings. One map exists for each
475 * hardware queue type (enum hctx_type) that the driver wishes
476 * to support. There are no restrictions on maps being of the
477 * same size, and it's perfectly legal to share maps between
479 * @nr_maps: Number of elements in the @map array. A number in the range
480 * [1, HCTX_MAX_TYPES].
481 * @nr_hw_queues: Number of hardware queues supported by the block driver that
482 * owns this data structure.
483 * @queue_depth: Number of tags per hardware queue, reserved tags included.
484 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
486 * @cmd_size: Number of additional bytes to allocate per request. The block
487 * driver owns these additional bytes.
488 * @numa_node: NUMA node the storage adapter has been connected to.
489 * @timeout: Request processing timeout in jiffies.
490 * @flags: Zero or more BLK_MQ_F_* flags.
491 * @driver_data: Pointer to data owned by the block driver that created this
493 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
496 * Shared set of tags. Has @nr_hw_queues elements. If set,
497 * shared by all @tags.
498 * @tag_list_lock: Serializes tag_list accesses.
499 * @tag_list: List of the request queues that use this tag set. See also
500 * request_queue.tag_set_list.
501 * @srcu: Use as lock when type of the request queue is blocking
502 * (BLK_MQ_F_BLOCKING).
504 struct blk_mq_tag_set {
505 const struct blk_mq_ops *ops;
506 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
507 unsigned int nr_maps;
508 unsigned int nr_hw_queues;
509 unsigned int queue_depth;
510 unsigned int reserved_tags;
511 unsigned int cmd_size;
513 unsigned int timeout;
517 struct blk_mq_tags **tags;
519 struct blk_mq_tags *shared_tags;
521 struct mutex tag_list_lock;
522 struct list_head tag_list;
523 struct srcu_struct *srcu;
527 * struct blk_mq_queue_data - Data about a request inserted in a queue
529 * @rq: Request pointer.
530 * @last: If it is the last request in the queue.
532 struct blk_mq_queue_data {
537 typedef bool (busy_tag_iter_fn)(struct request *, void *);
540 * struct blk_mq_ops - Callback functions that implements block driver
545 * @queue_rq: Queue a new request from block IO.
547 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
548 const struct blk_mq_queue_data *);
551 * @commit_rqs: If a driver uses bd->last to judge when to submit
552 * requests to hardware, it must define this function. In case of errors
553 * that make us stop issuing further requests, this hook serves the
554 * purpose of kicking the hardware (which the last request otherwise
557 void (*commit_rqs)(struct blk_mq_hw_ctx *);
560 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
561 * that each request belongs to the same queue. If the driver doesn't
562 * empty the @rqlist completely, then the rest will be queued
563 * individually by the block layer upon return.
565 void (*queue_rqs)(struct request **rqlist);
568 * @get_budget: Reserve budget before queue request, once .queue_rq is
569 * run, it is driver's responsibility to release the
570 * reserved budget. Also we have to handle failure case
571 * of .get_budget for avoiding I/O deadlock.
573 int (*get_budget)(struct request_queue *);
576 * @put_budget: Release the reserved budget.
578 void (*put_budget)(struct request_queue *, int);
581 * @set_rq_budget_token: store rq's budget token
583 void (*set_rq_budget_token)(struct request *, int);
585 * @get_rq_budget_token: retrieve rq's budget token
587 int (*get_rq_budget_token)(struct request *);
590 * @timeout: Called on request timeout.
592 enum blk_eh_timer_return (*timeout)(struct request *);
595 * @poll: Called to poll for completion of a specific tag.
597 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
600 * @complete: Mark the request as complete.
602 void (*complete)(struct request *);
605 * @init_hctx: Called when the block layer side of a hardware queue has
606 * been set up, allowing the driver to allocate/init matching
609 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
611 * @exit_hctx: Ditto for exit/teardown.
613 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
616 * @init_request: Called for every command allocated by the block layer
617 * to allow the driver to set up driver specific data.
619 * Tag greater than or equal to queue_depth is for setting up
622 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
623 unsigned int, unsigned int);
625 * @exit_request: Ditto for exit/teardown.
627 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
631 * @cleanup_rq: Called before freeing one request which isn't completed
632 * yet, and usually for freeing the driver private data.
634 void (*cleanup_rq)(struct request *);
637 * @busy: If set, returns whether or not this queue currently is busy.
639 bool (*busy)(struct request_queue *);
642 * @map_queues: This allows drivers specify their own queue mapping by
643 * overriding the setup-time function that builds the mq_map.
645 void (*map_queues)(struct blk_mq_tag_set *set);
647 #ifdef CONFIG_BLK_DEBUG_FS
649 * @show_rq: Used by the debugfs implementation to show driver-specific
650 * information about a request.
652 void (*show_rq)(struct seq_file *m, struct request *rq);
657 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
658 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
660 * Set when this device requires underlying blk-mq device for
663 BLK_MQ_F_STACKING = 1 << 2,
664 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
665 BLK_MQ_F_BLOCKING = 1 << 5,
666 /* Do not allow an I/O scheduler to be configured. */
667 BLK_MQ_F_NO_SCHED = 1 << 6,
669 * Select 'none' during queue registration in case of a single hwq
670 * or shared hwqs instead of 'mq-deadline'.
672 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
673 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
674 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
676 BLK_MQ_S_STOPPED = 0,
677 BLK_MQ_S_TAG_ACTIVE = 1,
678 BLK_MQ_S_SCHED_RESTART = 2,
680 /* hw queue is inactive after all its CPUs become offline */
681 BLK_MQ_S_INACTIVE = 3,
683 BLK_MQ_MAX_DEPTH = 10240,
685 BLK_MQ_CPU_WORK_BATCH = 8,
687 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
688 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
689 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
690 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
691 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
692 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
694 #define BLK_MQ_NO_HCTX_IDX (-1U)
696 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
697 struct lock_class_key *lkclass);
698 #define blk_mq_alloc_disk(set, queuedata) \
700 static struct lock_class_key __key; \
702 __blk_mq_alloc_disk(set, queuedata, &__key); \
704 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
705 struct lock_class_key *lkclass);
706 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
707 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
708 struct request_queue *q);
709 void blk_mq_destroy_queue(struct request_queue *);
711 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
712 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
713 const struct blk_mq_ops *ops, unsigned int queue_depth,
714 unsigned int set_flags);
715 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
717 void blk_mq_free_request(struct request *rq);
718 int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
719 unsigned int poll_flags);
721 bool blk_mq_queue_inflight(struct request_queue *q);
724 /* return when out of requests */
725 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
726 /* allocate from reserved pool */
727 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
729 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
732 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
733 blk_mq_req_flags_t flags);
734 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
735 blk_opf_t opf, blk_mq_req_flags_t flags,
736 unsigned int hctx_idx);
739 * Tag address space map.
742 unsigned int nr_tags;
743 unsigned int nr_reserved_tags;
744 unsigned int active_queues;
746 struct sbitmap_queue bitmap_tags;
747 struct sbitmap_queue breserved_tags;
749 struct request **rqs;
750 struct request **static_rqs;
751 struct list_head page_list;
754 * used to clear request reference in rqs[] before freeing one
760 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
763 if (tag < tags->nr_tags) {
764 prefetch(tags->rqs[tag]);
765 return tags->rqs[tag];
772 BLK_MQ_UNIQUE_TAG_BITS = 16,
773 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
776 u32 blk_mq_unique_tag(struct request *rq);
778 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
780 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
783 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
785 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
789 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
790 * @rq: target request.
792 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
794 return READ_ONCE(rq->state);
797 static inline int blk_mq_request_started(struct request *rq)
799 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
802 static inline int blk_mq_request_completed(struct request *rq)
804 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
809 * Set the state to complete when completing a request from inside ->queue_rq.
810 * This is used by drivers that want to ensure special complete actions that
811 * need access to the request are called on failure, e.g. by nvme for
814 static inline void blk_mq_set_request_complete(struct request *rq)
816 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
820 * Complete the request directly instead of deferring it to softirq or
821 * completing it another CPU. Useful in preemptible instead of an interrupt.
823 static inline void blk_mq_complete_request_direct(struct request *rq,
824 void (*complete)(struct request *rq))
826 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
830 void blk_mq_start_request(struct request *rq);
831 void blk_mq_end_request(struct request *rq, blk_status_t error);
832 void __blk_mq_end_request(struct request *rq, blk_status_t error);
833 void blk_mq_end_request_batch(struct io_comp_batch *ib);
836 * Only need start/end time stamping if we have iostat or
837 * blk stats enabled, or using an IO scheduler.
839 static inline bool blk_mq_need_time_stamp(struct request *rq)
841 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
844 static inline bool blk_mq_is_reserved_rq(struct request *rq)
846 return rq->rq_flags & RQF_RESV;
850 * Batched completions only work when there is no I/O error and no special
853 static inline bool blk_mq_add_to_batch(struct request *req,
854 struct io_comp_batch *iob, int ioerror,
855 void (*complete)(struct io_comp_batch *))
858 * blk_mq_end_request_batch() can't end request allocated from
861 if (!iob || (req->rq_flags & RQF_SCHED_TAGS) || ioerror ||
862 (req->end_io && !blk_rq_is_passthrough(req)))
866 iob->complete = complete;
867 else if (iob->complete != complete)
869 iob->need_ts |= blk_mq_need_time_stamp(req);
870 rq_list_add(&iob->req_list, req);
874 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
875 void blk_mq_kick_requeue_list(struct request_queue *q);
876 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
877 void blk_mq_complete_request(struct request *rq);
878 bool blk_mq_complete_request_remote(struct request *rq);
879 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
880 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
881 void blk_mq_stop_hw_queues(struct request_queue *q);
882 void blk_mq_start_hw_queues(struct request_queue *q);
883 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
884 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
885 void blk_mq_quiesce_queue(struct request_queue *q);
886 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
887 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
888 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
889 void blk_mq_unquiesce_queue(struct request_queue *q);
890 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
891 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
892 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
893 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
894 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
895 busy_tag_iter_fn *fn, void *priv);
896 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
897 void blk_mq_freeze_queue(struct request_queue *q);
898 void blk_mq_unfreeze_queue(struct request_queue *q);
899 void blk_freeze_queue_start(struct request_queue *q);
900 void blk_mq_freeze_queue_wait(struct request_queue *q);
901 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
902 unsigned long timeout);
904 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
905 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
907 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
909 unsigned int blk_mq_rq_cpu(struct request *rq);
911 bool __blk_should_fake_timeout(struct request_queue *q);
912 static inline bool blk_should_fake_timeout(struct request_queue *q)
914 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
915 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
916 return __blk_should_fake_timeout(q);
921 * blk_mq_rq_from_pdu - cast a PDU to a request
922 * @pdu: the PDU (Protocol Data Unit) to be casted
926 * Driver command data is immediately after the request. So subtract request
927 * size to get back to the original request.
929 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
931 return pdu - sizeof(struct request);
935 * blk_mq_rq_to_pdu - cast a request to a PDU
936 * @rq: the request to be casted
938 * Return: pointer to the PDU
940 * Driver command data is immediately after the request. So add request to get
943 static inline void *blk_mq_rq_to_pdu(struct request *rq)
948 #define queue_for_each_hw_ctx(q, hctx, i) \
949 xa_for_each(&(q)->hctx_table, (i), (hctx))
951 #define hctx_for_each_ctx(hctx, ctx, i) \
952 for ((i) = 0; (i) < (hctx)->nr_ctx && \
953 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
955 static inline void blk_mq_cleanup_rq(struct request *rq)
957 if (rq->q->mq_ops->cleanup_rq)
958 rq->q->mq_ops->cleanup_rq(rq);
961 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
962 unsigned int nr_segs)
964 rq->nr_phys_segments = nr_segs;
965 rq->__data_len = bio->bi_iter.bi_size;
966 rq->bio = rq->biotail = bio;
967 rq->ioprio = bio_prio(bio);
970 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
971 struct lock_class_key *key);
973 static inline bool rq_is_sync(struct request *rq)
975 return op_is_sync(rq->cmd_flags);
978 void blk_rq_init(struct request_queue *q, struct request *rq);
979 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
980 struct bio_set *bs, gfp_t gfp_mask,
981 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
982 void blk_rq_unprep_clone(struct request *rq);
983 blk_status_t blk_insert_cloned_request(struct request *rq);
987 unsigned long offset;
988 unsigned short page_order;
989 unsigned short nr_entries;
994 int blk_rq_map_user(struct request_queue *, struct request *,
995 struct rq_map_data *, void __user *, unsigned long, gfp_t);
996 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
997 void __user *, unsigned long, gfp_t, bool, int, bool, int);
998 int blk_rq_map_user_iov(struct request_queue *, struct request *,
999 struct rq_map_data *, const struct iov_iter *, gfp_t);
1000 int blk_rq_unmap_user(struct bio *);
1001 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
1002 unsigned int, gfp_t);
1003 int blk_rq_append_bio(struct request *rq, struct bio *bio);
1004 void blk_execute_rq_nowait(struct request *rq, bool at_head);
1005 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1006 bool blk_rq_is_poll(struct request *rq);
1008 struct req_iterator {
1009 struct bvec_iter iter;
1013 #define __rq_for_each_bio(_bio, rq) \
1015 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1017 #define rq_for_each_segment(bvl, _rq, _iter) \
1018 __rq_for_each_bio(_iter.bio, _rq) \
1019 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1021 #define rq_for_each_bvec(bvl, _rq, _iter) \
1022 __rq_for_each_bio(_iter.bio, _rq) \
1023 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1025 #define rq_iter_last(bvec, _iter) \
1026 (_iter.bio->bi_next == NULL && \
1027 bio_iter_last(bvec, _iter.iter))
1030 * blk_rq_pos() : the current sector
1031 * blk_rq_bytes() : bytes left in the entire request
1032 * blk_rq_cur_bytes() : bytes left in the current segment
1033 * blk_rq_sectors() : sectors left in the entire request
1034 * blk_rq_cur_sectors() : sectors left in the current segment
1035 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1037 static inline sector_t blk_rq_pos(const struct request *rq)
1039 return rq->__sector;
1042 static inline unsigned int blk_rq_bytes(const struct request *rq)
1044 return rq->__data_len;
1047 static inline int blk_rq_cur_bytes(const struct request *rq)
1051 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1052 return rq->bio->bi_iter.bi_size;
1053 return bio_iovec(rq->bio).bv_len;
1056 static inline unsigned int blk_rq_sectors(const struct request *rq)
1058 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1061 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1063 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1066 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1068 return rq->stats_sectors;
1072 * Some commands like WRITE SAME have a payload or data transfer size which
1073 * is different from the size of the request. Any driver that supports such
1074 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1075 * calculate the data transfer size.
1077 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1079 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1080 return rq->special_vec.bv_len;
1081 return blk_rq_bytes(rq);
1085 * Return the first full biovec in the request. The caller needs to check that
1086 * there are any bvecs before calling this helper.
1088 static inline struct bio_vec req_bvec(struct request *rq)
1090 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1091 return rq->special_vec;
1092 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1095 static inline unsigned int blk_rq_count_bios(struct request *rq)
1097 unsigned int nr_bios = 0;
1100 __rq_for_each_bio(bio, rq)
1106 void blk_steal_bios(struct bio_list *list, struct request *rq);
1109 * Request completion related functions.
1111 * blk_update_request() completes given number of bytes and updates
1112 * the request without completing it.
1114 bool blk_update_request(struct request *rq, blk_status_t error,
1115 unsigned int nr_bytes);
1116 void blk_abort_request(struct request *);
1119 * Number of physical segments as sent to the device.
1121 * Normally this is the number of discontiguous data segments sent by the
1122 * submitter. But for data-less command like discard we might have no
1123 * actual data segments submitted, but the driver might have to add it's
1124 * own special payload. In that case we still return 1 here so that this
1125 * special payload will be mapped.
1127 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1129 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1131 return rq->nr_phys_segments;
1135 * Number of discard segments (or ranges) the driver needs to fill in.
1136 * Each discard bio merged into a request is counted as one segment.
1138 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1140 return max_t(unsigned short, rq->nr_phys_segments, 1);
1143 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1144 struct scatterlist *sglist, struct scatterlist **last_sg);
1145 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1146 struct scatterlist *sglist)
1148 struct scatterlist *last_sg = NULL;
1150 return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1152 void blk_dump_rq_flags(struct request *, char *);
1154 #ifdef CONFIG_BLK_DEV_ZONED
1155 static inline unsigned int blk_rq_zone_no(struct request *rq)
1157 return disk_zone_no(rq->q->disk, blk_rq_pos(rq));
1160 static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1162 return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq));
1166 * blk_rq_is_seq_zoned_write() - Check if @rq requires write serialization.
1167 * @rq: Request to examine.
1169 * Note: REQ_OP_ZONE_APPEND requests do not require serialization.
1171 static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1173 return op_needs_zoned_write_locking(req_op(rq)) &&
1174 blk_rq_zone_is_seq(rq);
1177 bool blk_req_needs_zone_write_lock(struct request *rq);
1178 bool blk_req_zone_write_trylock(struct request *rq);
1179 void __blk_req_zone_write_lock(struct request *rq);
1180 void __blk_req_zone_write_unlock(struct request *rq);
1182 static inline void blk_req_zone_write_lock(struct request *rq)
1184 if (blk_req_needs_zone_write_lock(rq))
1185 __blk_req_zone_write_lock(rq);
1188 static inline void blk_req_zone_write_unlock(struct request *rq)
1190 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1191 __blk_req_zone_write_unlock(rq);
1194 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1196 return rq->q->disk->seq_zones_wlock &&
1197 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock);
1200 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1202 if (!blk_req_needs_zone_write_lock(rq))
1204 return !blk_req_zone_is_write_locked(rq);
1206 #else /* CONFIG_BLK_DEV_ZONED */
1207 static inline bool blk_rq_is_seq_zoned_write(struct request *rq)
1212 static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1217 static inline void blk_req_zone_write_lock(struct request *rq)
1221 static inline void blk_req_zone_write_unlock(struct request *rq)
1224 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1229 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1233 #endif /* CONFIG_BLK_DEV_ZONED */
1235 #endif /* BLK_MQ_H */