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/srcu.h>
8 #include <linux/lockdep.h>
9 
10 struct blk_mq_tags;
11 struct blk_flush_queue;
12 
13 /**
14  * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
15  * block device
16  */
17 struct blk_mq_hw_ctx {
18 	struct {
19 		/** @lock: Protects the dispatch list. */
20 		spinlock_t		lock;
21 		/**
22 		 * @dispatch: Used for requests that are ready to be
23 		 * dispatched to the hardware but for some reason (e.g. lack of
24 		 * resources) could not be sent to the hardware. As soon as the
25 		 * driver can send new requests, requests at this list will
26 		 * be sent first for a fairer dispatch.
27 		 */
28 		struct list_head	dispatch;
29 		 /**
30 		  * @state: BLK_MQ_S_* flags. Defines the state of the hw
31 		  * queue (active, scheduled to restart, stopped).
32 		  */
33 		unsigned long		state;
34 	} ____cacheline_aligned_in_smp;
35 
36 	/**
37 	 * @run_work: Used for scheduling a hardware queue run at a later time.
38 	 */
39 	struct delayed_work	run_work;
40 	/** @cpumask: Map of available CPUs where this hctx can run. */
41 	cpumask_var_t		cpumask;
42 	/**
43 	 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
44 	 * selection from @cpumask.
45 	 */
46 	int			next_cpu;
47 	/**
48 	 * @next_cpu_batch: Counter of how many works left in the batch before
49 	 * changing to the next CPU.
50 	 */
51 	int			next_cpu_batch;
52 
53 	/** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
54 	unsigned long		flags;
55 
56 	/**
57 	 * @sched_data: Pointer owned by the IO scheduler attached to a request
58 	 * queue. It's up to the IO scheduler how to use this pointer.
59 	 */
60 	void			*sched_data;
61 	/**
62 	 * @queue: Pointer to the request queue that owns this hardware context.
63 	 */
64 	struct request_queue	*queue;
65 	/** @fq: Queue of requests that need to perform a flush operation. */
66 	struct blk_flush_queue	*fq;
67 
68 	/**
69 	 * @driver_data: Pointer to data owned by the block driver that created
70 	 * this hctx
71 	 */
72 	void			*driver_data;
73 
74 	/**
75 	 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
76 	 * pending request in that software queue.
77 	 */
78 	struct sbitmap		ctx_map;
79 
80 	/**
81 	 * @dispatch_from: Software queue to be used when no scheduler was
82 	 * selected.
83 	 */
84 	struct blk_mq_ctx	*dispatch_from;
85 	/**
86 	 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
87 	 * decide if the hw_queue is busy using Exponential Weighted Moving
88 	 * Average algorithm.
89 	 */
90 	unsigned int		dispatch_busy;
91 
92 	/** @type: HCTX_TYPE_* flags. Type of hardware queue. */
93 	unsigned short		type;
94 	/** @nr_ctx: Number of software queues. */
95 	unsigned short		nr_ctx;
96 	/** @ctxs: Array of software queues. */
97 	struct blk_mq_ctx	**ctxs;
98 
99 	/** @dispatch_wait_lock: Lock for dispatch_wait queue. */
100 	spinlock_t		dispatch_wait_lock;
101 	/**
102 	 * @dispatch_wait: Waitqueue to put requests when there is no tag
103 	 * available at the moment, to wait for another try in the future.
104 	 */
105 	wait_queue_entry_t	dispatch_wait;
106 
107 	/**
108 	 * @wait_index: Index of next available dispatch_wait queue to insert
109 	 * requests.
110 	 */
111 	atomic_t		wait_index;
112 
113 	/**
114 	 * @tags: Tags owned by the block driver. A tag at this set is only
115 	 * assigned when a request is dispatched from a hardware queue.
116 	 */
117 	struct blk_mq_tags	*tags;
118 	/**
119 	 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
120 	 * scheduler associated with a request queue, a tag is assigned when
121 	 * that request is allocated. Else, this member is not used.
122 	 */
123 	struct blk_mq_tags	*sched_tags;
124 
125 	/** @queued: Number of queued requests. */
126 	unsigned long		queued;
127 	/** @run: Number of dispatched requests. */
128 	unsigned long		run;
129 #define BLK_MQ_MAX_DISPATCH_ORDER	7
130 	/** @dispatched: Number of dispatch requests by queue. */
131 	unsigned long		dispatched[BLK_MQ_MAX_DISPATCH_ORDER];
132 
133 	/** @numa_node: NUMA node the storage adapter has been connected to. */
134 	unsigned int		numa_node;
135 	/** @queue_num: Index of this hardware queue. */
136 	unsigned int		queue_num;
137 
138 	/**
139 	 * @nr_active: Number of active requests. Only used when a tag set is
140 	 * shared across request queues.
141 	 */
142 	atomic_t		nr_active;
143 
144 	/** @cpuhp_online: List to store request if CPU is going to die */
145 	struct hlist_node	cpuhp_online;
146 	/** @cpuhp_dead: List to store request if some CPU die. */
147 	struct hlist_node	cpuhp_dead;
148 	/** @kobj: Kernel object for sysfs. */
149 	struct kobject		kobj;
150 
151 	/** @poll_considered: Count times blk_poll() was called. */
152 	unsigned long		poll_considered;
153 	/** @poll_invoked: Count how many requests blk_poll() polled. */
154 	unsigned long		poll_invoked;
155 	/** @poll_success: Count how many polled requests were completed. */
156 	unsigned long		poll_success;
157 
158 #ifdef CONFIG_BLK_DEBUG_FS
159 	/**
160 	 * @debugfs_dir: debugfs directory for this hardware queue. Named
161 	 * as cpu<cpu_number>.
162 	 */
163 	struct dentry		*debugfs_dir;
164 	/** @sched_debugfs_dir:	debugfs directory for the scheduler. */
165 	struct dentry		*sched_debugfs_dir;
166 #endif
167 
168 	/**
169 	 * @hctx_list: if this hctx is not in use, this is an entry in
170 	 * q->unused_hctx_list.
171 	 */
172 	struct list_head	hctx_list;
173 
174 	/**
175 	 * @srcu: Sleepable RCU. Use as lock when type of the hardware queue is
176 	 * blocking (BLK_MQ_F_BLOCKING). Must be the last member - see also
177 	 * blk_mq_hw_ctx_size().
178 	 */
179 	struct srcu_struct	srcu[];
180 };
181 
182 /**
183  * struct blk_mq_queue_map - Map software queues to hardware queues
184  * @mq_map:       CPU ID to hardware queue index map. This is an array
185  *	with nr_cpu_ids elements. Each element has a value in the range
186  *	[@queue_offset, @queue_offset + @nr_queues).
187  * @nr_queues:    Number of hardware queues to map CPU IDs onto.
188  * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
189  *	driver to map each hardware queue type (enum hctx_type) onto a distinct
190  *	set of hardware queues.
191  */
192 struct blk_mq_queue_map {
193 	unsigned int *mq_map;
194 	unsigned int nr_queues;
195 	unsigned int queue_offset;
196 };
197 
198 /**
199  * enum hctx_type - Type of hardware queue
200  * @HCTX_TYPE_DEFAULT:	All I/O not otherwise accounted for.
201  * @HCTX_TYPE_READ:	Just for READ I/O.
202  * @HCTX_TYPE_POLL:	Polled I/O of any kind.
203  * @HCTX_MAX_TYPES:	Number of types of hctx.
204  */
205 enum hctx_type {
206 	HCTX_TYPE_DEFAULT,
207 	HCTX_TYPE_READ,
208 	HCTX_TYPE_POLL,
209 
210 	HCTX_MAX_TYPES,
211 };
212 
213 /**
214  * struct blk_mq_tag_set - tag set that can be shared between request queues
215  * @map:	   One or more ctx -> hctx mappings. One map exists for each
216  *		   hardware queue type (enum hctx_type) that the driver wishes
217  *		   to support. There are no restrictions on maps being of the
218  *		   same size, and it's perfectly legal to share maps between
219  *		   types.
220  * @nr_maps:	   Number of elements in the @map array. A number in the range
221  *		   [1, HCTX_MAX_TYPES].
222  * @ops:	   Pointers to functions that implement block driver behavior.
223  * @nr_hw_queues:  Number of hardware queues supported by the block driver that
224  *		   owns this data structure.
225  * @queue_depth:   Number of tags per hardware queue, reserved tags included.
226  * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
227  *		   allocations.
228  * @cmd_size:	   Number of additional bytes to allocate per request. The block
229  *		   driver owns these additional bytes.
230  * @numa_node:	   NUMA node the storage adapter has been connected to.
231  * @timeout:	   Request processing timeout in jiffies.
232  * @flags:	   Zero or more BLK_MQ_F_* flags.
233  * @driver_data:   Pointer to data owned by the block driver that created this
234  *		   tag set.
235  * @active_queues_shared_sbitmap:
236  * 		   number of active request queues per tag set.
237  * @__bitmap_tags: A shared tags sbitmap, used over all hctx's
238  * @__breserved_tags:
239  *		   A shared reserved tags sbitmap, used over all hctx's
240  * @tags:	   Tag sets. One tag set per hardware queue. Has @nr_hw_queues
241  *		   elements.
242  * @tag_list_lock: Serializes tag_list accesses.
243  * @tag_list:	   List of the request queues that use this tag set. See also
244  *		   request_queue.tag_set_list.
245  */
246 struct blk_mq_tag_set {
247 	struct blk_mq_queue_map	map[HCTX_MAX_TYPES];
248 	unsigned int		nr_maps;
249 	const struct blk_mq_ops	*ops;
250 	unsigned int		nr_hw_queues;
251 	unsigned int		queue_depth;
252 	unsigned int		reserved_tags;
253 	unsigned int		cmd_size;
254 	int			numa_node;
255 	unsigned int		timeout;
256 	unsigned int		flags;
257 	void			*driver_data;
258 	atomic_t		active_queues_shared_sbitmap;
259 
260 	struct sbitmap_queue	__bitmap_tags;
261 	struct sbitmap_queue	__breserved_tags;
262 	struct blk_mq_tags	**tags;
263 
264 	struct mutex		tag_list_lock;
265 	struct list_head	tag_list;
266 };
267 
268 /**
269  * struct blk_mq_queue_data - Data about a request inserted in a queue
270  *
271  * @rq:   Request pointer.
272  * @last: If it is the last request in the queue.
273  */
274 struct blk_mq_queue_data {
275 	struct request *rq;
276 	bool last;
277 };
278 
279 typedef bool (busy_iter_fn)(struct blk_mq_hw_ctx *, struct request *, void *,
280 		bool);
281 typedef bool (busy_tag_iter_fn)(struct request *, void *, bool);
282 
283 /**
284  * struct blk_mq_ops - Callback functions that implements block driver
285  * behaviour.
286  */
287 struct blk_mq_ops {
288 	/**
289 	 * @queue_rq: Queue a new request from block IO.
290 	 */
291 	blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
292 				 const struct blk_mq_queue_data *);
293 
294 	/**
295 	 * @commit_rqs: If a driver uses bd->last to judge when to submit
296 	 * requests to hardware, it must define this function. In case of errors
297 	 * that make us stop issuing further requests, this hook serves the
298 	 * purpose of kicking the hardware (which the last request otherwise
299 	 * would have done).
300 	 */
301 	void (*commit_rqs)(struct blk_mq_hw_ctx *);
302 
303 	/**
304 	 * @get_budget: Reserve budget before queue request, once .queue_rq is
305 	 * run, it is driver's responsibility to release the
306 	 * reserved budget. Also we have to handle failure case
307 	 * of .get_budget for avoiding I/O deadlock.
308 	 */
309 	int (*get_budget)(struct request_queue *);
310 
311 	/**
312 	 * @put_budget: Release the reserved budget.
313 	 */
314 	void (*put_budget)(struct request_queue *, int);
315 
316 	/**
317 	 * @set_rq_budget_token: store rq's budget token
318 	 */
319 	void (*set_rq_budget_token)(struct request *, int);
320 	/**
321 	 * @get_rq_budget_token: retrieve rq's budget token
322 	 */
323 	int (*get_rq_budget_token)(struct request *);
324 
325 	/**
326 	 * @timeout: Called on request timeout.
327 	 */
328 	enum blk_eh_timer_return (*timeout)(struct request *, bool);
329 
330 	/**
331 	 * @poll: Called to poll for completion of a specific tag.
332 	 */
333 	int (*poll)(struct blk_mq_hw_ctx *);
334 
335 	/**
336 	 * @complete: Mark the request as complete.
337 	 */
338 	void (*complete)(struct request *);
339 
340 	/**
341 	 * @init_hctx: Called when the block layer side of a hardware queue has
342 	 * been set up, allowing the driver to allocate/init matching
343 	 * structures.
344 	 */
345 	int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
346 	/**
347 	 * @exit_hctx: Ditto for exit/teardown.
348 	 */
349 	void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
350 
351 	/**
352 	 * @init_request: Called for every command allocated by the block layer
353 	 * to allow the driver to set up driver specific data.
354 	 *
355 	 * Tag greater than or equal to queue_depth is for setting up
356 	 * flush request.
357 	 */
358 	int (*init_request)(struct blk_mq_tag_set *set, struct request *,
359 			    unsigned int, unsigned int);
360 	/**
361 	 * @exit_request: Ditto for exit/teardown.
362 	 */
363 	void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
364 			     unsigned int);
365 
366 	/**
367 	 * @initialize_rq_fn: Called from inside blk_get_request().
368 	 */
369 	void (*initialize_rq_fn)(struct request *rq);
370 
371 	/**
372 	 * @cleanup_rq: Called before freeing one request which isn't completed
373 	 * yet, and usually for freeing the driver private data.
374 	 */
375 	void (*cleanup_rq)(struct request *);
376 
377 	/**
378 	 * @busy: If set, returns whether or not this queue currently is busy.
379 	 */
380 	bool (*busy)(struct request_queue *);
381 
382 	/**
383 	 * @map_queues: This allows drivers specify their own queue mapping by
384 	 * overriding the setup-time function that builds the mq_map.
385 	 */
386 	int (*map_queues)(struct blk_mq_tag_set *set);
387 
388 #ifdef CONFIG_BLK_DEBUG_FS
389 	/**
390 	 * @show_rq: Used by the debugfs implementation to show driver-specific
391 	 * information about a request.
392 	 */
393 	void (*show_rq)(struct seq_file *m, struct request *rq);
394 #endif
395 };
396 
397 enum {
398 	BLK_MQ_F_SHOULD_MERGE	= 1 << 0,
399 	BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
400 	/*
401 	 * Set when this device requires underlying blk-mq device for
402 	 * completing IO:
403 	 */
404 	BLK_MQ_F_STACKING	= 1 << 2,
405 	BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
406 	BLK_MQ_F_BLOCKING	= 1 << 5,
407 	/* Do not allow an I/O scheduler to be configured. */
408 	BLK_MQ_F_NO_SCHED	= 1 << 6,
409 	/*
410 	 * Select 'none' during queue registration in case of a single hwq
411 	 * or shared hwqs instead of 'mq-deadline'.
412 	 */
413 	BLK_MQ_F_NO_SCHED_BY_DEFAULT	= 1 << 7,
414 	BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
415 	BLK_MQ_F_ALLOC_POLICY_BITS = 1,
416 
417 	BLK_MQ_S_STOPPED	= 0,
418 	BLK_MQ_S_TAG_ACTIVE	= 1,
419 	BLK_MQ_S_SCHED_RESTART	= 2,
420 
421 	/* hw queue is inactive after all its CPUs become offline */
422 	BLK_MQ_S_INACTIVE	= 3,
423 
424 	BLK_MQ_MAX_DEPTH	= 10240,
425 
426 	BLK_MQ_CPU_WORK_BATCH	= 8,
427 };
428 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
429 	((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
430 		((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
431 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
432 	((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
433 		<< BLK_MQ_F_ALLOC_POLICY_START_BIT)
434 
435 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
436 		struct lock_class_key *lkclass);
437 #define blk_mq_alloc_disk(set, queuedata)				\
438 ({									\
439 	static struct lock_class_key __key;				\
440 									\
441 	__blk_mq_alloc_disk(set, queuedata, &__key);			\
442 })
443 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
444 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
445 		struct request_queue *q);
446 void blk_mq_unregister_dev(struct device *, struct request_queue *);
447 
448 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
449 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
450 		const struct blk_mq_ops *ops, unsigned int queue_depth,
451 		unsigned int set_flags);
452 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
453 
454 void blk_mq_flush_plug_list(struct blk_plug *plug, bool from_schedule);
455 
456 void blk_mq_free_request(struct request *rq);
457 
458 bool blk_mq_queue_inflight(struct request_queue *q);
459 
460 enum {
461 	/* return when out of requests */
462 	BLK_MQ_REQ_NOWAIT	= (__force blk_mq_req_flags_t)(1 << 0),
463 	/* allocate from reserved pool */
464 	BLK_MQ_REQ_RESERVED	= (__force blk_mq_req_flags_t)(1 << 1),
465 	/* set RQF_PM */
466 	BLK_MQ_REQ_PM		= (__force blk_mq_req_flags_t)(1 << 2),
467 };
468 
469 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
470 		blk_mq_req_flags_t flags);
471 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
472 		unsigned int op, blk_mq_req_flags_t flags,
473 		unsigned int hctx_idx);
474 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
475 
476 enum {
477 	BLK_MQ_UNIQUE_TAG_BITS = 16,
478 	BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
479 };
480 
481 u32 blk_mq_unique_tag(struct request *rq);
482 
blk_mq_unique_tag_to_hwq(u32 unique_tag)483 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
484 {
485 	return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
486 }
487 
blk_mq_unique_tag_to_tag(u32 unique_tag)488 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
489 {
490 	return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
491 }
492 
493 /**
494  * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
495  * @rq: target request.
496  */
blk_mq_rq_state(struct request * rq)497 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
498 {
499 	return READ_ONCE(rq->state);
500 }
501 
blk_mq_request_started(struct request * rq)502 static inline int blk_mq_request_started(struct request *rq)
503 {
504 	return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
505 }
506 
blk_mq_request_completed(struct request * rq)507 static inline int blk_mq_request_completed(struct request *rq)
508 {
509 	return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
510 }
511 
512 /*
513  *
514  * Set the state to complete when completing a request from inside ->queue_rq.
515  * This is used by drivers that want to ensure special complete actions that
516  * need access to the request are called on failure, e.g. by nvme for
517  * multipathing.
518  */
blk_mq_set_request_complete(struct request * rq)519 static inline void blk_mq_set_request_complete(struct request *rq)
520 {
521 	WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
522 }
523 
524 void blk_mq_start_request(struct request *rq);
525 void blk_mq_end_request(struct request *rq, blk_status_t error);
526 void __blk_mq_end_request(struct request *rq, blk_status_t error);
527 
528 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
529 void blk_mq_kick_requeue_list(struct request_queue *q);
530 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
531 void blk_mq_complete_request(struct request *rq);
532 bool blk_mq_complete_request_remote(struct request *rq);
533 bool blk_mq_queue_stopped(struct request_queue *q);
534 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
535 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
536 void blk_mq_stop_hw_queues(struct request_queue *q);
537 void blk_mq_start_hw_queues(struct request_queue *q);
538 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
539 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
540 void blk_mq_quiesce_queue(struct request_queue *q);
541 void blk_mq_unquiesce_queue(struct request_queue *q);
542 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
543 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
544 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
545 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
546 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
547 		busy_tag_iter_fn *fn, void *priv);
548 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
549 void blk_mq_freeze_queue(struct request_queue *q);
550 void blk_mq_unfreeze_queue(struct request_queue *q);
551 void blk_freeze_queue_start(struct request_queue *q);
552 void blk_mq_freeze_queue_wait(struct request_queue *q);
553 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
554 				     unsigned long timeout);
555 
556 int blk_mq_map_queues(struct blk_mq_queue_map *qmap);
557 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
558 
559 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
560 
561 unsigned int blk_mq_rq_cpu(struct request *rq);
562 
563 bool __blk_should_fake_timeout(struct request_queue *q);
blk_should_fake_timeout(struct request_queue * q)564 static inline bool blk_should_fake_timeout(struct request_queue *q)
565 {
566 	if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
567 	    test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
568 		return __blk_should_fake_timeout(q);
569 	return false;
570 }
571 
572 /**
573  * blk_mq_rq_from_pdu - cast a PDU to a request
574  * @pdu: the PDU (Protocol Data Unit) to be casted
575  *
576  * Return: request
577  *
578  * Driver command data is immediately after the request. So subtract request
579  * size to get back to the original request.
580  */
blk_mq_rq_from_pdu(void * pdu)581 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
582 {
583 	return pdu - sizeof(struct request);
584 }
585 
586 /**
587  * blk_mq_rq_to_pdu - cast a request to a PDU
588  * @rq: the request to be casted
589  *
590  * Return: pointer to the PDU
591  *
592  * Driver command data is immediately after the request. So add request to get
593  * the PDU.
594  */
blk_mq_rq_to_pdu(struct request * rq)595 static inline void *blk_mq_rq_to_pdu(struct request *rq)
596 {
597 	return rq + 1;
598 }
599 
600 #define queue_for_each_hw_ctx(q, hctx, i)				\
601 	for ((i) = 0; (i) < (q)->nr_hw_queues &&			\
602 	     ({ hctx = (q)->queue_hw_ctx[i]; 1; }); (i)++)
603 
604 #define hctx_for_each_ctx(hctx, ctx, i)					\
605 	for ((i) = 0; (i) < (hctx)->nr_ctx &&				\
606 	     ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
607 
request_to_qc_t(struct blk_mq_hw_ctx * hctx,struct request * rq)608 static inline blk_qc_t request_to_qc_t(struct blk_mq_hw_ctx *hctx,
609 		struct request *rq)
610 {
611 	if (rq->tag != -1)
612 		return rq->tag | (hctx->queue_num << BLK_QC_T_SHIFT);
613 
614 	return rq->internal_tag | (hctx->queue_num << BLK_QC_T_SHIFT) |
615 			BLK_QC_T_INTERNAL;
616 }
617 
blk_mq_cleanup_rq(struct request * rq)618 static inline void blk_mq_cleanup_rq(struct request *rq)
619 {
620 	if (rq->q->mq_ops->cleanup_rq)
621 		rq->q->mq_ops->cleanup_rq(rq);
622 }
623 
blk_rq_bio_prep(struct request * rq,struct bio * bio,unsigned int nr_segs)624 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
625 		unsigned int nr_segs)
626 {
627 	rq->nr_phys_segments = nr_segs;
628 	rq->__data_len = bio->bi_iter.bi_size;
629 	rq->bio = rq->biotail = bio;
630 	rq->ioprio = bio_prio(bio);
631 
632 	if (bio->bi_bdev)
633 		rq->rq_disk = bio->bi_bdev->bd_disk;
634 }
635 
636 blk_qc_t blk_mq_submit_bio(struct bio *bio);
637 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
638 		struct lock_class_key *key);
639 
640 #endif
641