xref: /linux/include/linux/blk-mq.h (revision 3ef18b236690af5f6427c5b6d8636881116aa73a)
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 
13 struct blk_mq_tags;
14 struct blk_flush_queue;
15 
16 #define BLKDEV_MIN_RQ	4
17 #define BLKDEV_DEFAULT_RQ	128
18 
19 enum rq_end_io_ret {
20 	RQ_END_IO_NONE,
21 	RQ_END_IO_FREE,
22 };
23 
24 typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
25 
26 /*
27  * request flags */
28 typedef __u32 __bitwise req_flags_t;
29 
30 /* Keep rqf_name[] in sync with the definitions below */
31 enum rqf_flags {
32 	/* drive already may have started this one */
33 	__RQF_STARTED,
34 	/* request for flush sequence */
35 	__RQF_FLUSH_SEQ,
36 	/* merge of different types, fail separately */
37 	__RQF_MIXED_MERGE,
38 	/* don't call prep for this one */
39 	__RQF_DONTPREP,
40 	/* use hctx->sched_tags */
41 	__RQF_SCHED_TAGS,
42 	/* use an I/O scheduler for this request */
43 	__RQF_USE_SCHED,
44 	/* vaguely specified driver internal error.  Ignored by block layer */
45 	__RQF_FAILED,
46 	/* don't warn about errors */
47 	__RQF_QUIET,
48 	/* account into disk and partition IO statistics */
49 	__RQF_IO_STAT,
50 	/* runtime pm request */
51 	__RQF_PM,
52 	/* on IO scheduler merge hash */
53 	__RQF_HASHED,
54 	/* track IO completion time */
55 	__RQF_STATS,
56 	/* Look at ->special_vec for the actual data payload instead of the
57 	   bio chain. */
58 	__RQF_SPECIAL_PAYLOAD,
59 	/* request completion needs to be signaled to zone write plugging. */
60 	__RQF_ZONE_WRITE_PLUGGING,
61 	/* ->timeout has been called, don't expire again */
62 	__RQF_TIMED_OUT,
63 	__RQF_RESV,
64 	__RQF_BITS
65 };
66 
67 #define RQF_STARTED		((__force req_flags_t)(1 << __RQF_STARTED))
68 #define RQF_FLUSH_SEQ		((__force req_flags_t)(1 << __RQF_FLUSH_SEQ))
69 #define RQF_MIXED_MERGE		((__force req_flags_t)(1 << __RQF_MIXED_MERGE))
70 #define RQF_DONTPREP		((__force req_flags_t)(1 << __RQF_DONTPREP))
71 #define RQF_SCHED_TAGS		((__force req_flags_t)(1 << __RQF_SCHED_TAGS))
72 #define RQF_USE_SCHED		((__force req_flags_t)(1 << __RQF_USE_SCHED))
73 #define RQF_FAILED		((__force req_flags_t)(1 << __RQF_FAILED))
74 #define RQF_QUIET		((__force req_flags_t)(1 << __RQF_QUIET))
75 #define RQF_IO_STAT		((__force req_flags_t)(1 << __RQF_IO_STAT))
76 #define RQF_PM			((__force req_flags_t)(1 << __RQF_PM))
77 #define RQF_HASHED		((__force req_flags_t)(1 << __RQF_HASHED))
78 #define RQF_STATS		((__force req_flags_t)(1 << __RQF_STATS))
79 #define RQF_SPECIAL_PAYLOAD	\
80 			((__force req_flags_t)(1 << __RQF_SPECIAL_PAYLOAD))
81 #define RQF_ZONE_WRITE_PLUGGING	\
82 			((__force req_flags_t)(1 << __RQF_ZONE_WRITE_PLUGGING))
83 #define RQF_TIMED_OUT		((__force req_flags_t)(1 << __RQF_TIMED_OUT))
84 #define RQF_RESV		((__force req_flags_t)(1 << __RQF_RESV))
85 
86 /* flags that prevent us from merging requests: */
87 #define RQF_NOMERGE_FLAGS \
88 	(RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
89 
90 enum mq_rq_state {
91 	MQ_RQ_IDLE		= 0,
92 	MQ_RQ_IN_FLIGHT		= 1,
93 	MQ_RQ_COMPLETE		= 2,
94 };
95 
96 /*
97  * Try to put the fields that are referenced together in the same cacheline.
98  *
99  * If you modify this structure, make sure to update blk_rq_init() and
100  * especially blk_mq_rq_ctx_init() to take care of the added fields.
101  */
102 struct request {
103 	struct request_queue *q;
104 	struct blk_mq_ctx *mq_ctx;
105 	struct blk_mq_hw_ctx *mq_hctx;
106 
107 	blk_opf_t cmd_flags;		/* op and common flags */
108 	req_flags_t rq_flags;
109 
110 	int tag;
111 	int internal_tag;
112 
113 	unsigned int timeout;
114 
115 	/* the following two fields are internal, NEVER access directly */
116 	unsigned int __data_len;	/* total data len */
117 	sector_t __sector;		/* sector cursor */
118 
119 	struct bio *bio;
120 	struct bio *biotail;
121 
122 	union {
123 		struct list_head queuelist;
124 		struct request *rq_next;
125 	};
126 
127 	struct block_device *part;
128 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
129 	/* Time that the first bio started allocating this request. */
130 	u64 alloc_time_ns;
131 #endif
132 	/* Time that this request was allocated for this IO. */
133 	u64 start_time_ns;
134 	/* Time that I/O was submitted to the device. */
135 	u64 io_start_time_ns;
136 
137 #ifdef CONFIG_BLK_WBT
138 	unsigned short wbt_flags;
139 #endif
140 	/*
141 	 * rq sectors used for blk stats. It has the same value
142 	 * with blk_rq_sectors(rq), except that it never be zeroed
143 	 * by completion.
144 	 */
145 	unsigned short stats_sectors;
146 
147 	/*
148 	 * Number of scatter-gather DMA addr+len pairs after
149 	 * physical address coalescing is performed.
150 	 */
151 	unsigned short nr_phys_segments;
152 	unsigned short nr_integrity_segments;
153 
154 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
155 	struct bio_crypt_ctx *crypt_ctx;
156 	struct blk_crypto_keyslot *crypt_keyslot;
157 #endif
158 
159 	enum mq_rq_state state;
160 	atomic_t ref;
161 
162 	unsigned long deadline;
163 
164 	/*
165 	 * The hash is used inside the scheduler, and killed once the
166 	 * request reaches the dispatch list. The ipi_list is only used
167 	 * to queue the request for softirq completion, which is long
168 	 * after the request has been unhashed (and even removed from
169 	 * the dispatch list).
170 	 */
171 	union {
172 		struct hlist_node hash;	/* merge hash */
173 		struct llist_node ipi_list;
174 	};
175 
176 	/*
177 	 * The rb_node is only used inside the io scheduler, requests
178 	 * are pruned when moved to the dispatch queue. special_vec must
179 	 * only be used if RQF_SPECIAL_PAYLOAD is set, and those cannot be
180 	 * insert into an IO scheduler.
181 	 */
182 	union {
183 		struct rb_node rb_node;	/* sort/lookup */
184 		struct bio_vec special_vec;
185 	};
186 
187 	/*
188 	 * Three pointers are available for the IO schedulers, if they need
189 	 * more they have to dynamically allocate it.
190 	 */
191 	struct {
192 		struct io_cq		*icq;
193 		void			*priv[2];
194 	} elv;
195 
196 	struct {
197 		unsigned int		seq;
198 		rq_end_io_fn		*saved_end_io;
199 	} flush;
200 
201 	u64 fifo_time;
202 
203 	/*
204 	 * completion callback.
205 	 */
206 	rq_end_io_fn *end_io;
207 	void *end_io_data;
208 };
209 
req_op(const struct request * req)210 static inline enum req_op req_op(const struct request *req)
211 {
212 	return req->cmd_flags & REQ_OP_MASK;
213 }
214 
blk_rq_is_passthrough(struct request * rq)215 static inline bool blk_rq_is_passthrough(struct request *rq)
216 {
217 	return blk_op_is_passthrough(rq->cmd_flags);
218 }
219 
req_get_ioprio(struct request * req)220 static inline unsigned short req_get_ioprio(struct request *req)
221 {
222 	if (req->bio)
223 		return req->bio->bi_ioprio;
224 	return 0;
225 }
226 
227 #define rq_data_dir(rq)		(op_is_write(req_op(rq)) ? WRITE : READ)
228 
229 #define rq_dma_dir(rq) \
230 	(op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
231 
rq_list_empty(const struct rq_list * rl)232 static inline int rq_list_empty(const struct rq_list *rl)
233 {
234 	return rl->head == NULL;
235 }
236 
rq_list_init(struct rq_list * rl)237 static inline void rq_list_init(struct rq_list *rl)
238 {
239 	rl->head = NULL;
240 	rl->tail = NULL;
241 }
242 
rq_list_add_tail(struct rq_list * rl,struct request * rq)243 static inline void rq_list_add_tail(struct rq_list *rl, struct request *rq)
244 {
245 	rq->rq_next = NULL;
246 	if (rl->tail)
247 		rl->tail->rq_next = rq;
248 	else
249 		rl->head = rq;
250 	rl->tail = rq;
251 }
252 
rq_list_add_head(struct rq_list * rl,struct request * rq)253 static inline void rq_list_add_head(struct rq_list *rl, struct request *rq)
254 {
255 	rq->rq_next = rl->head;
256 	rl->head = rq;
257 	if (!rl->tail)
258 		rl->tail = rq;
259 }
260 
rq_list_pop(struct rq_list * rl)261 static inline struct request *rq_list_pop(struct rq_list *rl)
262 {
263 	struct request *rq = rl->head;
264 
265 	if (rq) {
266 		rl->head = rl->head->rq_next;
267 		if (!rl->head)
268 			rl->tail = NULL;
269 		rq->rq_next = NULL;
270 	}
271 
272 	return rq;
273 }
274 
rq_list_peek(struct rq_list * rl)275 static inline struct request *rq_list_peek(struct rq_list *rl)
276 {
277 	return rl->head;
278 }
279 
280 #define rq_list_for_each(rl, pos)					\
281 	for (pos = rq_list_peek((rl)); (pos); pos = pos->rq_next)
282 
283 #define rq_list_for_each_safe(rl, pos, nxt)				\
284 	for (pos = rq_list_peek((rl)), nxt = pos->rq_next;		\
285 		pos; pos = nxt, nxt = pos ? pos->rq_next : NULL)
286 
287 /**
288  * enum blk_eh_timer_return - How the timeout handler should proceed
289  * @BLK_EH_DONE: The block driver completed the command or will complete it at
290  *	a later time.
291  * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
292  *	request to complete.
293  */
294 enum blk_eh_timer_return {
295 	BLK_EH_DONE,
296 	BLK_EH_RESET_TIMER,
297 };
298 
299 /**
300  * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
301  * block device
302  */
303 struct blk_mq_hw_ctx {
304 	struct {
305 		/** @lock: Protects the dispatch list. */
306 		spinlock_t		lock;
307 		/**
308 		 * @dispatch: Used for requests that are ready to be
309 		 * dispatched to the hardware but for some reason (e.g. lack of
310 		 * resources) could not be sent to the hardware. As soon as the
311 		 * driver can send new requests, requests at this list will
312 		 * be sent first for a fairer dispatch.
313 		 */
314 		struct list_head	dispatch;
315 		 /**
316 		  * @state: BLK_MQ_S_* flags. Defines the state of the hw
317 		  * queue (active, scheduled to restart, stopped).
318 		  */
319 		unsigned long		state;
320 	} ____cacheline_aligned_in_smp;
321 
322 	/**
323 	 * @run_work: Used for scheduling a hardware queue run at a later time.
324 	 */
325 	struct delayed_work	run_work;
326 	/** @cpumask: Map of available CPUs where this hctx can run. */
327 	cpumask_var_t		cpumask;
328 	/**
329 	 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
330 	 * selection from @cpumask.
331 	 */
332 	int			next_cpu;
333 	/**
334 	 * @next_cpu_batch: Counter of how many works left in the batch before
335 	 * changing to the next CPU.
336 	 */
337 	int			next_cpu_batch;
338 
339 	/** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
340 	unsigned long		flags;
341 
342 	/**
343 	 * @sched_data: Pointer owned by the IO scheduler attached to a request
344 	 * queue. It's up to the IO scheduler how to use this pointer.
345 	 */
346 	void			*sched_data;
347 	/**
348 	 * @queue: Pointer to the request queue that owns this hardware context.
349 	 */
350 	struct request_queue	*queue;
351 	/** @fq: Queue of requests that need to perform a flush operation. */
352 	struct blk_flush_queue	*fq;
353 
354 	/**
355 	 * @driver_data: Pointer to data owned by the block driver that created
356 	 * this hctx
357 	 */
358 	void			*driver_data;
359 
360 	/**
361 	 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
362 	 * pending request in that software queue.
363 	 */
364 	struct sbitmap		ctx_map;
365 
366 	/**
367 	 * @dispatch_from: Software queue to be used when no scheduler was
368 	 * selected.
369 	 */
370 	struct blk_mq_ctx	*dispatch_from;
371 	/**
372 	 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
373 	 * decide if the hw_queue is busy using Exponential Weighted Moving
374 	 * Average algorithm.
375 	 */
376 	unsigned int		dispatch_busy;
377 
378 	/** @type: HCTX_TYPE_* flags. Type of hardware queue. */
379 	unsigned short		type;
380 	/** @nr_ctx: Number of software queues. */
381 	unsigned short		nr_ctx;
382 	/** @ctxs: Array of software queues. */
383 	struct blk_mq_ctx	**ctxs;
384 
385 	/** @dispatch_wait_lock: Lock for dispatch_wait queue. */
386 	spinlock_t		dispatch_wait_lock;
387 	/**
388 	 * @dispatch_wait: Waitqueue to put requests when there is no tag
389 	 * available at the moment, to wait for another try in the future.
390 	 */
391 	wait_queue_entry_t	dispatch_wait;
392 
393 	/**
394 	 * @wait_index: Index of next available dispatch_wait queue to insert
395 	 * requests.
396 	 */
397 	atomic_t		wait_index;
398 
399 	/**
400 	 * @tags: Tags owned by the block driver. A tag at this set is only
401 	 * assigned when a request is dispatched from a hardware queue.
402 	 */
403 	struct blk_mq_tags	*tags;
404 	/**
405 	 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
406 	 * scheduler associated with a request queue, a tag is assigned when
407 	 * that request is allocated. Else, this member is not used.
408 	 */
409 	struct blk_mq_tags	*sched_tags;
410 
411 	/** @numa_node: NUMA node the storage adapter has been connected to. */
412 	unsigned int		numa_node;
413 	/** @queue_num: Index of this hardware queue. */
414 	unsigned int		queue_num;
415 
416 	/**
417 	 * @nr_active: Number of active requests. Only used when a tag set is
418 	 * shared across request queues.
419 	 */
420 	atomic_t		nr_active;
421 
422 	/** @cpuhp_online: List to store request if CPU is going to die */
423 	struct hlist_node	cpuhp_online;
424 	/** @cpuhp_dead: List to store request if some CPU die. */
425 	struct hlist_node	cpuhp_dead;
426 	/** @kobj: Kernel object for sysfs. */
427 	struct kobject		kobj;
428 
429 #ifdef CONFIG_BLK_DEBUG_FS
430 	/**
431 	 * @debugfs_dir: debugfs directory for this hardware queue. Named
432 	 * as cpu<cpu_number>.
433 	 */
434 	struct dentry		*debugfs_dir;
435 	/** @sched_debugfs_dir:	debugfs directory for the scheduler. */
436 	struct dentry		*sched_debugfs_dir;
437 #endif
438 
439 	/**
440 	 * @hctx_list: if this hctx is not in use, this is an entry in
441 	 * q->unused_hctx_list.
442 	 */
443 	struct list_head	hctx_list;
444 };
445 
446 /**
447  * struct blk_mq_queue_map - Map software queues to hardware queues
448  * @mq_map:       CPU ID to hardware queue index map. This is an array
449  *	with nr_cpu_ids elements. Each element has a value in the range
450  *	[@queue_offset, @queue_offset + @nr_queues).
451  * @nr_queues:    Number of hardware queues to map CPU IDs onto.
452  * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
453  *	driver to map each hardware queue type (enum hctx_type) onto a distinct
454  *	set of hardware queues.
455  */
456 struct blk_mq_queue_map {
457 	unsigned int *mq_map;
458 	unsigned int nr_queues;
459 	unsigned int queue_offset;
460 };
461 
462 /**
463  * enum hctx_type - Type of hardware queue
464  * @HCTX_TYPE_DEFAULT:	All I/O not otherwise accounted for.
465  * @HCTX_TYPE_READ:	Just for READ I/O.
466  * @HCTX_TYPE_POLL:	Polled I/O of any kind.
467  * @HCTX_MAX_TYPES:	Number of types of hctx.
468  */
469 enum hctx_type {
470 	HCTX_TYPE_DEFAULT,
471 	HCTX_TYPE_READ,
472 	HCTX_TYPE_POLL,
473 
474 	HCTX_MAX_TYPES,
475 };
476 
477 /**
478  * struct blk_mq_tag_set - tag set that can be shared between request queues
479  * @ops:	   Pointers to functions that implement block driver behavior.
480  * @map:	   One or more ctx -> hctx mappings. One map exists for each
481  *		   hardware queue type (enum hctx_type) that the driver wishes
482  *		   to support. There are no restrictions on maps being of the
483  *		   same size, and it's perfectly legal to share maps between
484  *		   types.
485  * @nr_maps:	   Number of elements in the @map array. A number in the range
486  *		   [1, HCTX_MAX_TYPES].
487  * @nr_hw_queues:  Number of hardware queues supported by the block driver that
488  *		   owns this data structure.
489  * @queue_depth:   Number of tags per hardware queue, reserved tags included.
490  * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
491  *		   allocations.
492  * @cmd_size:	   Number of additional bytes to allocate per request. The block
493  *		   driver owns these additional bytes.
494  * @numa_node:	   NUMA node the storage adapter has been connected to.
495  * @timeout:	   Request processing timeout in jiffies.
496  * @flags:	   Zero or more BLK_MQ_F_* flags.
497  * @driver_data:   Pointer to data owned by the block driver that created this
498  *		   tag set.
499  * @tags:	   Tag sets. One tag set per hardware queue. Has @nr_hw_queues
500  *		   elements.
501  * @shared_tags:
502  *		   Shared set of tags. Has @nr_hw_queues elements. If set,
503  *		   shared by all @tags.
504  * @tag_list_lock: Serializes tag_list accesses.
505  * @tag_list:	   List of the request queues that use this tag set. See also
506  *		   request_queue.tag_set_list.
507  * @srcu:	   Use as lock when type of the request queue is blocking
508  *		   (BLK_MQ_F_BLOCKING).
509  */
510 struct blk_mq_tag_set {
511 	const struct blk_mq_ops	*ops;
512 	struct blk_mq_queue_map	map[HCTX_MAX_TYPES];
513 	unsigned int		nr_maps;
514 	unsigned int		nr_hw_queues;
515 	unsigned int		queue_depth;
516 	unsigned int		reserved_tags;
517 	unsigned int		cmd_size;
518 	int			numa_node;
519 	unsigned int		timeout;
520 	unsigned int		flags;
521 	void			*driver_data;
522 
523 	struct blk_mq_tags	**tags;
524 
525 	struct blk_mq_tags	*shared_tags;
526 
527 	struct mutex		tag_list_lock;
528 	struct list_head	tag_list;
529 	struct srcu_struct	*srcu;
530 };
531 
532 /**
533  * struct blk_mq_queue_data - Data about a request inserted in a queue
534  *
535  * @rq:   Request pointer.
536  * @last: If it is the last request in the queue.
537  */
538 struct blk_mq_queue_data {
539 	struct request *rq;
540 	bool last;
541 };
542 
543 typedef bool (busy_tag_iter_fn)(struct request *, void *);
544 
545 /**
546  * struct blk_mq_ops - Callback functions that implements block driver
547  * behaviour.
548  */
549 struct blk_mq_ops {
550 	/**
551 	 * @queue_rq: Queue a new request from block IO.
552 	 */
553 	blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
554 				 const struct blk_mq_queue_data *);
555 
556 	/**
557 	 * @commit_rqs: If a driver uses bd->last to judge when to submit
558 	 * requests to hardware, it must define this function. In case of errors
559 	 * that make us stop issuing further requests, this hook serves the
560 	 * purpose of kicking the hardware (which the last request otherwise
561 	 * would have done).
562 	 */
563 	void (*commit_rqs)(struct blk_mq_hw_ctx *);
564 
565 	/**
566 	 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
567 	 * that each request belongs to the same queue. If the driver doesn't
568 	 * empty the @rqlist completely, then the rest will be queued
569 	 * individually by the block layer upon return.
570 	 */
571 	void (*queue_rqs)(struct rq_list *rqlist);
572 
573 	/**
574 	 * @get_budget: Reserve budget before queue request, once .queue_rq is
575 	 * run, it is driver's responsibility to release the
576 	 * reserved budget. Also we have to handle failure case
577 	 * of .get_budget for avoiding I/O deadlock.
578 	 */
579 	int (*get_budget)(struct request_queue *);
580 
581 	/**
582 	 * @put_budget: Release the reserved budget.
583 	 */
584 	void (*put_budget)(struct request_queue *, int);
585 
586 	/**
587 	 * @set_rq_budget_token: store rq's budget token
588 	 */
589 	void (*set_rq_budget_token)(struct request *, int);
590 	/**
591 	 * @get_rq_budget_token: retrieve rq's budget token
592 	 */
593 	int (*get_rq_budget_token)(struct request *);
594 
595 	/**
596 	 * @timeout: Called on request timeout.
597 	 */
598 	enum blk_eh_timer_return (*timeout)(struct request *);
599 
600 	/**
601 	 * @poll: Called to poll for completion of a specific tag.
602 	 */
603 	int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
604 
605 	/**
606 	 * @complete: Mark the request as complete.
607 	 */
608 	void (*complete)(struct request *);
609 
610 	/**
611 	 * @init_hctx: Called when the block layer side of a hardware queue has
612 	 * been set up, allowing the driver to allocate/init matching
613 	 * structures.
614 	 */
615 	int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
616 	/**
617 	 * @exit_hctx: Ditto for exit/teardown.
618 	 */
619 	void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
620 
621 	/**
622 	 * @init_request: Called for every command allocated by the block layer
623 	 * to allow the driver to set up driver specific data.
624 	 *
625 	 * Tag greater than or equal to queue_depth is for setting up
626 	 * flush request.
627 	 */
628 	int (*init_request)(struct blk_mq_tag_set *set, struct request *,
629 			    unsigned int, unsigned int);
630 	/**
631 	 * @exit_request: Ditto for exit/teardown.
632 	 */
633 	void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
634 			     unsigned int);
635 
636 	/**
637 	 * @cleanup_rq: Called before freeing one request which isn't completed
638 	 * yet, and usually for freeing the driver private data.
639 	 */
640 	void (*cleanup_rq)(struct request *);
641 
642 	/**
643 	 * @busy: If set, returns whether or not this queue currently is busy.
644 	 */
645 	bool (*busy)(struct request_queue *);
646 
647 	/**
648 	 * @map_queues: This allows drivers specify their own queue mapping by
649 	 * overriding the setup-time function that builds the mq_map.
650 	 */
651 	void (*map_queues)(struct blk_mq_tag_set *set);
652 
653 #ifdef CONFIG_BLK_DEBUG_FS
654 	/**
655 	 * @show_rq: Used by the debugfs implementation to show driver-specific
656 	 * information about a request.
657 	 */
658 	void (*show_rq)(struct seq_file *m, struct request *rq);
659 #endif
660 };
661 
662 /* Keep hctx_flag_name[] in sync with the definitions below */
663 enum {
664 	BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
665 	/*
666 	 * Set when this device requires underlying blk-mq device for
667 	 * completing IO:
668 	 */
669 	BLK_MQ_F_STACKING	= 1 << 2,
670 	BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
671 	BLK_MQ_F_BLOCKING	= 1 << 4,
672 
673 	/*
674 	 * Alloc tags on a round-robin base instead of the first available one.
675 	 */
676 	BLK_MQ_F_TAG_RR		= 1 << 5,
677 
678 	/*
679 	 * Select 'none' during queue registration in case of a single hwq
680 	 * or shared hwqs instead of 'mq-deadline'.
681 	 */
682 	BLK_MQ_F_NO_SCHED_BY_DEFAULT	= 1 << 6,
683 
684 	BLK_MQ_F_MAX = 1 << 7,
685 };
686 
687 #define BLK_MQ_MAX_DEPTH	(10240)
688 #define BLK_MQ_NO_HCTX_IDX	(-1U)
689 
690 enum {
691 	/* Keep hctx_state_name[] in sync with the definitions below */
692 	BLK_MQ_S_STOPPED,
693 	BLK_MQ_S_TAG_ACTIVE,
694 	BLK_MQ_S_SCHED_RESTART,
695 	/* hw queue is inactive after all its CPUs become offline */
696 	BLK_MQ_S_INACTIVE,
697 	BLK_MQ_S_MAX
698 };
699 
700 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set,
701 		struct queue_limits *lim, void *queuedata,
702 		struct lock_class_key *lkclass);
703 #define blk_mq_alloc_disk(set, lim, queuedata)				\
704 ({									\
705 	static struct lock_class_key __key;				\
706 									\
707 	__blk_mq_alloc_disk(set, lim, queuedata, &__key);		\
708 })
709 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
710 		struct lock_class_key *lkclass);
711 struct request_queue *blk_mq_alloc_queue(struct blk_mq_tag_set *set,
712 		struct queue_limits *lim, void *queuedata);
713 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
714 		struct request_queue *q);
715 void blk_mq_destroy_queue(struct request_queue *);
716 
717 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
718 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
719 		const struct blk_mq_ops *ops, unsigned int queue_depth,
720 		unsigned int set_flags);
721 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
722 
723 void blk_mq_free_request(struct request *rq);
724 int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
725 		unsigned int poll_flags);
726 
727 bool blk_mq_queue_inflight(struct request_queue *q);
728 
729 enum {
730 	/* return when out of requests */
731 	BLK_MQ_REQ_NOWAIT	= (__force blk_mq_req_flags_t)(1 << 0),
732 	/* allocate from reserved pool */
733 	BLK_MQ_REQ_RESERVED	= (__force blk_mq_req_flags_t)(1 << 1),
734 	/* set RQF_PM */
735 	BLK_MQ_REQ_PM		= (__force blk_mq_req_flags_t)(1 << 2),
736 };
737 
738 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
739 		blk_mq_req_flags_t flags);
740 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
741 		blk_opf_t opf, blk_mq_req_flags_t flags,
742 		unsigned int hctx_idx);
743 
744 /*
745  * Tag address space map.
746  */
747 struct blk_mq_tags {
748 	unsigned int nr_tags;
749 	unsigned int nr_reserved_tags;
750 	unsigned int active_queues;
751 
752 	struct sbitmap_queue bitmap_tags;
753 	struct sbitmap_queue breserved_tags;
754 
755 	struct request **rqs;
756 	struct request **static_rqs;
757 	struct list_head page_list;
758 
759 	/*
760 	 * used to clear request reference in rqs[] before freeing one
761 	 * request pool
762 	 */
763 	spinlock_t lock;
764 };
765 
blk_mq_tag_to_rq(struct blk_mq_tags * tags,unsigned int tag)766 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
767 					       unsigned int tag)
768 {
769 	if (tag < tags->nr_tags) {
770 		prefetch(tags->rqs[tag]);
771 		return tags->rqs[tag];
772 	}
773 
774 	return NULL;
775 }
776 
777 enum {
778 	BLK_MQ_UNIQUE_TAG_BITS = 16,
779 	BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
780 };
781 
782 u32 blk_mq_unique_tag(struct request *rq);
783 
blk_mq_unique_tag_to_hwq(u32 unique_tag)784 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
785 {
786 	return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
787 }
788 
blk_mq_unique_tag_to_tag(u32 unique_tag)789 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
790 {
791 	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
792 }
793 
794 /**
795  * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
796  * @rq: target request.
797  */
blk_mq_rq_state(struct request * rq)798 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
799 {
800 	return READ_ONCE(rq->state);
801 }
802 
blk_mq_request_started(struct request * rq)803 static inline int blk_mq_request_started(struct request *rq)
804 {
805 	return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
806 }
807 
blk_mq_request_completed(struct request * rq)808 static inline int blk_mq_request_completed(struct request *rq)
809 {
810 	return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
811 }
812 
813 /*
814  *
815  * Set the state to complete when completing a request from inside ->queue_rq.
816  * This is used by drivers that want to ensure special complete actions that
817  * need access to the request are called on failure, e.g. by nvme for
818  * multipathing.
819  */
blk_mq_set_request_complete(struct request * rq)820 static inline void blk_mq_set_request_complete(struct request *rq)
821 {
822 	WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
823 }
824 
825 /*
826  * Complete the request directly instead of deferring it to softirq or
827  * completing it another CPU. Useful in preemptible instead of an interrupt.
828  */
blk_mq_complete_request_direct(struct request * rq,void (* complete)(struct request * rq))829 static inline void blk_mq_complete_request_direct(struct request *rq,
830 		   void (*complete)(struct request *rq))
831 {
832 	WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
833 	complete(rq);
834 }
835 
836 void blk_mq_start_request(struct request *rq);
837 void blk_mq_end_request(struct request *rq, blk_status_t error);
838 void __blk_mq_end_request(struct request *rq, blk_status_t error);
839 void blk_mq_end_request_batch(struct io_comp_batch *ib);
840 
841 /*
842  * Only need start/end time stamping if we have iostat or
843  * blk stats enabled, or using an IO scheduler.
844  */
blk_mq_need_time_stamp(struct request * rq)845 static inline bool blk_mq_need_time_stamp(struct request *rq)
846 {
847 	return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
848 }
849 
blk_mq_is_reserved_rq(struct request * rq)850 static inline bool blk_mq_is_reserved_rq(struct request *rq)
851 {
852 	return rq->rq_flags & RQF_RESV;
853 }
854 
855 /**
856  * blk_mq_add_to_batch() - add a request to the completion batch
857  * @req: The request to add to batch
858  * @iob: The batch to add the request
859  * @is_error: Specify true if the request failed with an error
860  * @complete: The completaion handler for the request
861  *
862  * Batched completions only work when there is no I/O error and no special
863  * ->end_io handler.
864  *
865  * Return: true when the request was added to the batch, otherwise false
866  */
blk_mq_add_to_batch(struct request * req,struct io_comp_batch * iob,bool is_error,void (* complete)(struct io_comp_batch *))867 static inline bool blk_mq_add_to_batch(struct request *req,
868 				       struct io_comp_batch *iob, bool is_error,
869 				       void (*complete)(struct io_comp_batch *))
870 {
871 	/*
872 	 * Check various conditions that exclude batch processing:
873 	 * 1) No batch container
874 	 * 2) Has scheduler data attached
875 	 * 3) Not a passthrough request and end_io set
876 	 * 4) Not a passthrough request and failed with an error
877 	 */
878 	if (!iob)
879 		return false;
880 	if (req->rq_flags & RQF_SCHED_TAGS)
881 		return false;
882 	if (!blk_rq_is_passthrough(req)) {
883 		if (req->end_io)
884 			return false;
885 		if (is_error)
886 			return false;
887 	}
888 
889 	if (!iob->complete)
890 		iob->complete = complete;
891 	else if (iob->complete != complete)
892 		return false;
893 	iob->need_ts |= blk_mq_need_time_stamp(req);
894 	rq_list_add_tail(&iob->req_list, req);
895 	return true;
896 }
897 
898 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
899 void blk_mq_kick_requeue_list(struct request_queue *q);
900 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
901 void blk_mq_complete_request(struct request *rq);
902 bool blk_mq_complete_request_remote(struct request *rq);
903 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
904 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
905 void blk_mq_stop_hw_queues(struct request_queue *q);
906 void blk_mq_start_hw_queues(struct request_queue *q);
907 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
908 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
909 void blk_mq_quiesce_queue(struct request_queue *q);
910 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
911 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
912 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
913 void blk_mq_unquiesce_queue(struct request_queue *q);
914 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
915 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
916 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
917 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
918 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
919 		busy_tag_iter_fn *fn, void *priv);
920 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
921 void blk_mq_freeze_queue_nomemsave(struct request_queue *q);
922 void blk_mq_unfreeze_queue_nomemrestore(struct request_queue *q);
923 static inline unsigned int __must_check
blk_mq_freeze_queue(struct request_queue * q)924 blk_mq_freeze_queue(struct request_queue *q)
925 {
926 	unsigned int memflags = memalloc_noio_save();
927 
928 	blk_mq_freeze_queue_nomemsave(q);
929 	return memflags;
930 }
931 static inline void
blk_mq_unfreeze_queue(struct request_queue * q,unsigned int memflags)932 blk_mq_unfreeze_queue(struct request_queue *q, unsigned int memflags)
933 {
934 	blk_mq_unfreeze_queue_nomemrestore(q);
935 	memalloc_noio_restore(memflags);
936 }
937 void blk_freeze_queue_start(struct request_queue *q);
938 void blk_mq_freeze_queue_wait(struct request_queue *q);
939 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
940 				     unsigned long timeout);
941 void blk_mq_unfreeze_queue_non_owner(struct request_queue *q);
942 void blk_freeze_queue_start_non_owner(struct request_queue *q);
943 
944 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
945 void blk_mq_map_hw_queues(struct blk_mq_queue_map *qmap,
946 			  struct device *dev, unsigned int offset);
947 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
948 
949 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
950 
951 unsigned int blk_mq_rq_cpu(struct request *rq);
952 
953 bool __blk_should_fake_timeout(struct request_queue *q);
blk_should_fake_timeout(struct request_queue * q)954 static inline bool blk_should_fake_timeout(struct request_queue *q)
955 {
956 	if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
957 	    test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
958 		return __blk_should_fake_timeout(q);
959 	return false;
960 }
961 
962 /**
963  * blk_mq_rq_from_pdu - cast a PDU to a request
964  * @pdu: the PDU (Protocol Data Unit) to be casted
965  *
966  * Return: request
967  *
968  * Driver command data is immediately after the request. So subtract request
969  * size to get back to the original request.
970  */
blk_mq_rq_from_pdu(void * pdu)971 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
972 {
973 	return pdu - sizeof(struct request);
974 }
975 
976 /**
977  * blk_mq_rq_to_pdu - cast a request to a PDU
978  * @rq: the request to be casted
979  *
980  * Return: pointer to the PDU
981  *
982  * Driver command data is immediately after the request. So add request to get
983  * the PDU.
984  */
blk_mq_rq_to_pdu(struct request * rq)985 static inline void *blk_mq_rq_to_pdu(struct request *rq)
986 {
987 	return rq + 1;
988 }
989 
990 #define queue_for_each_hw_ctx(q, hctx, i)				\
991 	xa_for_each(&(q)->hctx_table, (i), (hctx))
992 
993 #define hctx_for_each_ctx(hctx, ctx, i)					\
994 	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
995 	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
996 
blk_mq_cleanup_rq(struct request * rq)997 static inline void blk_mq_cleanup_rq(struct request *rq)
998 {
999 	if (rq->q->mq_ops->cleanup_rq)
1000 		rq->q->mq_ops->cleanup_rq(rq);
1001 }
1002 
1003 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
1004 		struct lock_class_key *key);
1005 
rq_is_sync(struct request * rq)1006 static inline bool rq_is_sync(struct request *rq)
1007 {
1008 	return op_is_sync(rq->cmd_flags);
1009 }
1010 
1011 void blk_rq_init(struct request_queue *q, struct request *rq);
1012 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
1013 		struct bio_set *bs, gfp_t gfp_mask,
1014 		int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
1015 void blk_rq_unprep_clone(struct request *rq);
1016 blk_status_t blk_insert_cloned_request(struct request *rq);
1017 
1018 struct rq_map_data {
1019 	struct page **pages;
1020 	unsigned long offset;
1021 	unsigned short page_order;
1022 	unsigned short nr_entries;
1023 	bool null_mapped;
1024 	bool from_user;
1025 };
1026 
1027 int blk_rq_map_user(struct request_queue *, struct request *,
1028 		struct rq_map_data *, void __user *, unsigned long, gfp_t);
1029 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
1030 		void __user *, unsigned long, gfp_t, bool, int, bool, int);
1031 int blk_rq_map_user_iov(struct request_queue *, struct request *,
1032 		struct rq_map_data *, const struct iov_iter *, gfp_t);
1033 int blk_rq_unmap_user(struct bio *);
1034 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
1035 		unsigned int, gfp_t);
1036 int blk_rq_append_bio(struct request *rq, struct bio *bio);
1037 void blk_execute_rq_nowait(struct request *rq, bool at_head);
1038 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1039 bool blk_rq_is_poll(struct request *rq);
1040 
1041 struct req_iterator {
1042 	struct bvec_iter iter;
1043 	struct bio *bio;
1044 };
1045 
1046 #define __rq_for_each_bio(_bio, rq)	\
1047 	if ((rq->bio))			\
1048 		for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1049 
1050 #define rq_for_each_segment(bvl, _rq, _iter)			\
1051 	__rq_for_each_bio(_iter.bio, _rq)			\
1052 		bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1053 
1054 #define rq_for_each_bvec(bvl, _rq, _iter)			\
1055 	__rq_for_each_bio(_iter.bio, _rq)			\
1056 		bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1057 
1058 #define rq_iter_last(bvec, _iter)				\
1059 		(_iter.bio->bi_next == NULL &&			\
1060 		 bio_iter_last(bvec, _iter.iter))
1061 
1062 /*
1063  * blk_rq_pos()			: the current sector
1064  * blk_rq_bytes()		: bytes left in the entire request
1065  * blk_rq_cur_bytes()		: bytes left in the current segment
1066  * blk_rq_sectors()		: sectors left in the entire request
1067  * blk_rq_cur_sectors()		: sectors left in the current segment
1068  * blk_rq_stats_sectors()	: sectors of the entire request used for stats
1069  */
blk_rq_pos(const struct request * rq)1070 static inline sector_t blk_rq_pos(const struct request *rq)
1071 {
1072 	return rq->__sector;
1073 }
1074 
blk_rq_bytes(const struct request * rq)1075 static inline unsigned int blk_rq_bytes(const struct request *rq)
1076 {
1077 	return rq->__data_len;
1078 }
1079 
blk_rq_cur_bytes(const struct request * rq)1080 static inline int blk_rq_cur_bytes(const struct request *rq)
1081 {
1082 	if (!rq->bio)
1083 		return 0;
1084 	if (!bio_has_data(rq->bio))	/* dataless requests such as discard */
1085 		return rq->bio->bi_iter.bi_size;
1086 	return bio_iovec(rq->bio).bv_len;
1087 }
1088 
blk_rq_sectors(const struct request * rq)1089 static inline unsigned int blk_rq_sectors(const struct request *rq)
1090 {
1091 	return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1092 }
1093 
blk_rq_cur_sectors(const struct request * rq)1094 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1095 {
1096 	return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1097 }
1098 
blk_rq_stats_sectors(const struct request * rq)1099 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1100 {
1101 	return rq->stats_sectors;
1102 }
1103 
1104 /*
1105  * Some commands like WRITE SAME have a payload or data transfer size which
1106  * is different from the size of the request.  Any driver that supports such
1107  * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1108  * calculate the data transfer size.
1109  */
blk_rq_payload_bytes(struct request * rq)1110 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1111 {
1112 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1113 		return rq->special_vec.bv_len;
1114 	return blk_rq_bytes(rq);
1115 }
1116 
1117 /*
1118  * Return the first full biovec in the request.  The caller needs to check that
1119  * there are any bvecs before calling this helper.
1120  */
req_bvec(struct request * rq)1121 static inline struct bio_vec req_bvec(struct request *rq)
1122 {
1123 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1124 		return rq->special_vec;
1125 	return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1126 }
1127 
blk_rq_count_bios(struct request * rq)1128 static inline unsigned int blk_rq_count_bios(struct request *rq)
1129 {
1130 	unsigned int nr_bios = 0;
1131 	struct bio *bio;
1132 
1133 	__rq_for_each_bio(bio, rq)
1134 		nr_bios++;
1135 
1136 	return nr_bios;
1137 }
1138 
1139 void blk_steal_bios(struct bio_list *list, struct request *rq);
1140 
1141 /*
1142  * Request completion related functions.
1143  *
1144  * blk_update_request() completes given number of bytes and updates
1145  * the request without completing it.
1146  */
1147 bool blk_update_request(struct request *rq, blk_status_t error,
1148 			       unsigned int nr_bytes);
1149 void blk_abort_request(struct request *);
1150 
1151 /*
1152  * Number of physical segments as sent to the device.
1153  *
1154  * Normally this is the number of discontiguous data segments sent by the
1155  * submitter.  But for data-less command like discard we might have no
1156  * actual data segments submitted, but the driver might have to add it's
1157  * own special payload.  In that case we still return 1 here so that this
1158  * special payload will be mapped.
1159  */
blk_rq_nr_phys_segments(struct request * rq)1160 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1161 {
1162 	if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1163 		return 1;
1164 	return rq->nr_phys_segments;
1165 }
1166 
1167 /*
1168  * Number of discard segments (or ranges) the driver needs to fill in.
1169  * Each discard bio merged into a request is counted as one segment.
1170  */
blk_rq_nr_discard_segments(struct request * rq)1171 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1172 {
1173 	return max_t(unsigned short, rq->nr_phys_segments, 1);
1174 }
1175 
1176 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1177 		struct scatterlist *sglist, struct scatterlist **last_sg);
blk_rq_map_sg(struct request_queue * q,struct request * rq,struct scatterlist * sglist)1178 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1179 		struct scatterlist *sglist)
1180 {
1181 	struct scatterlist *last_sg = NULL;
1182 
1183 	return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1184 }
1185 void blk_dump_rq_flags(struct request *, char *);
1186 
1187 #endif /* BLK_MQ_H */
1188