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 /* drive already may have started this one */
31 #define RQF_STARTED ((__force req_flags_t)(1 << 1))
32 /* request for flush sequence */
33 #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4))
34 /* merge of different types, fail separately */
35 #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5))
36 /* don't call prep for this one */
37 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
38 /* use hctx->sched_tags */
39 #define RQF_SCHED_TAGS ((__force req_flags_t)(1 << 8))
40 /* use an I/O scheduler for this request */
41 #define RQF_USE_SCHED ((__force req_flags_t)(1 << 9))
42 /* vaguely specified driver internal error. Ignored by the block layer */
43 #define RQF_FAILED ((__force req_flags_t)(1 << 10))
44 /* don't warn about errors */
45 #define RQF_QUIET ((__force req_flags_t)(1 << 11))
46 /* account into disk and partition IO statistics */
47 #define RQF_IO_STAT ((__force req_flags_t)(1 << 13))
48 /* runtime pm request */
49 #define RQF_PM ((__force req_flags_t)(1 << 15))
50 /* on IO scheduler merge hash */
51 #define RQF_HASHED ((__force req_flags_t)(1 << 16))
52 /* track IO completion time */
53 #define RQF_STATS ((__force req_flags_t)(1 << 17))
54 /* Look at ->special_vec for the actual data payload instead of the
55 bio chain. */
56 #define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18))
57 /* The request completion needs to be signaled to zone write pluging. */
58 #define RQF_ZONE_WRITE_PLUGGING ((__force req_flags_t)(1 << 20))
59 /* ->timeout has been called, don't expire again */
60 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
61 #define RQF_RESV ((__force req_flags_t)(1 << 23))
62
63 /* flags that prevent us from merging requests: */
64 #define RQF_NOMERGE_FLAGS \
65 (RQF_STARTED | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
66
67 enum mq_rq_state {
68 MQ_RQ_IDLE = 0,
69 MQ_RQ_IN_FLIGHT = 1,
70 MQ_RQ_COMPLETE = 2,
71 };
72
73 /*
74 * Try to put the fields that are referenced together in the same cacheline.
75 *
76 * If you modify this structure, make sure to update blk_rq_init() and
77 * especially blk_mq_rq_ctx_init() to take care of the added fields.
78 */
79 struct request {
80 struct request_queue *q;
81 struct blk_mq_ctx *mq_ctx;
82 struct blk_mq_hw_ctx *mq_hctx;
83
84 blk_opf_t cmd_flags; /* op and common flags */
85 req_flags_t rq_flags;
86
87 int tag;
88 int internal_tag;
89
90 unsigned int timeout;
91
92 /* the following two fields are internal, NEVER access directly */
93 unsigned int __data_len; /* total data len */
94 sector_t __sector; /* sector cursor */
95
96 struct bio *bio;
97 struct bio *biotail;
98
99 union {
100 struct list_head queuelist;
101 struct request *rq_next;
102 };
103
104 struct block_device *part;
105 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
106 /* Time that the first bio started allocating this request. */
107 u64 alloc_time_ns;
108 #endif
109 /* Time that this request was allocated for this IO. */
110 u64 start_time_ns;
111 /* Time that I/O was submitted to the device. */
112 u64 io_start_time_ns;
113
114 #ifdef CONFIG_BLK_WBT
115 unsigned short wbt_flags;
116 #endif
117 /*
118 * rq sectors used for blk stats. It has the same value
119 * with blk_rq_sectors(rq), except that it never be zeroed
120 * by completion.
121 */
122 unsigned short stats_sectors;
123
124 /*
125 * Number of scatter-gather DMA addr+len pairs after
126 * physical address coalescing is performed.
127 */
128 unsigned short nr_phys_segments;
129
130 #ifdef CONFIG_BLK_DEV_INTEGRITY
131 unsigned short nr_integrity_segments;
132 #endif
133
134 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
135 struct bio_crypt_ctx *crypt_ctx;
136 struct blk_crypto_keyslot *crypt_keyslot;
137 #endif
138
139 enum rw_hint write_hint;
140 unsigned short ioprio;
141
142 enum mq_rq_state state;
143 atomic_t ref;
144
145 unsigned long deadline;
146
147 /*
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).
153 */
154 union {
155 struct hlist_node hash; /* merge hash */
156 struct llist_node ipi_list;
157 };
158
159 /*
160 * The rb_node is only used inside the io scheduler, requests
161 * are pruned when moved to the dispatch queue. special_vec must
162 * only be used if RQF_SPECIAL_PAYLOAD is set, and those cannot be
163 * insert into an IO scheduler.
164 */
165 union {
166 struct rb_node rb_node; /* sort/lookup */
167 struct bio_vec special_vec;
168 };
169
170 /*
171 * Three pointers are available for the IO schedulers, if they need
172 * more they have to dynamically allocate it.
173 */
174 struct {
175 struct io_cq *icq;
176 void *priv[2];
177 } elv;
178
179 struct {
180 unsigned int seq;
181 rq_end_io_fn *saved_end_io;
182 } flush;
183
184 u64 fifo_time;
185
186 /*
187 * completion callback.
188 */
189 rq_end_io_fn *end_io;
190 void *end_io_data;
191 };
192
req_op(const struct request * req)193 static inline enum req_op req_op(const struct request *req)
194 {
195 return req->cmd_flags & REQ_OP_MASK;
196 }
197
blk_rq_is_passthrough(struct request * rq)198 static inline bool blk_rq_is_passthrough(struct request *rq)
199 {
200 return blk_op_is_passthrough(rq->cmd_flags);
201 }
202
req_get_ioprio(struct request * req)203 static inline unsigned short req_get_ioprio(struct request *req)
204 {
205 return req->ioprio;
206 }
207
208 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
209
210 #define rq_dma_dir(rq) \
211 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
212
213 #define rq_list_add(listptr, rq) do { \
214 (rq)->rq_next = *(listptr); \
215 *(listptr) = rq; \
216 } while (0)
217
218 #define rq_list_add_tail(lastpptr, rq) do { \
219 (rq)->rq_next = NULL; \
220 **(lastpptr) = rq; \
221 *(lastpptr) = &rq->rq_next; \
222 } while (0)
223
224 #define rq_list_pop(listptr) \
225 ({ \
226 struct request *__req = NULL; \
227 if ((listptr) && *(listptr)) { \
228 __req = *(listptr); \
229 *(listptr) = __req->rq_next; \
230 } \
231 __req; \
232 })
233
234 #define rq_list_peek(listptr) \
235 ({ \
236 struct request *__req = NULL; \
237 if ((listptr) && *(listptr)) \
238 __req = *(listptr); \
239 __req; \
240 })
241
242 #define rq_list_for_each(listptr, pos) \
243 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
244
245 #define rq_list_for_each_safe(listptr, pos, nxt) \
246 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
247 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
248
249 #define rq_list_next(rq) (rq)->rq_next
250 #define rq_list_empty(list) ((list) == (struct request *) NULL)
251
252 /**
253 * rq_list_move() - move a struct request from one list to another
254 * @src: The source list @rq is currently in
255 * @dst: The destination list that @rq will be appended to
256 * @rq: The request to move
257 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
258 */
rq_list_move(struct request ** src,struct request ** dst,struct request * rq,struct request * prev)259 static inline void rq_list_move(struct request **src, struct request **dst,
260 struct request *rq, struct request *prev)
261 {
262 if (prev)
263 prev->rq_next = rq->rq_next;
264 else
265 *src = rq->rq_next;
266 rq_list_add(dst, rq);
267 }
268
269 /**
270 * enum blk_eh_timer_return - How the timeout handler should proceed
271 * @BLK_EH_DONE: The block driver completed the command or will complete it at
272 * a later time.
273 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
274 * request to complete.
275 */
276 enum blk_eh_timer_return {
277 BLK_EH_DONE,
278 BLK_EH_RESET_TIMER,
279 };
280
281 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
282 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
283
284 /**
285 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
286 * block device
287 */
288 struct blk_mq_hw_ctx {
289 struct {
290 /** @lock: Protects the dispatch list. */
291 spinlock_t lock;
292 /**
293 * @dispatch: Used for requests that are ready to be
294 * dispatched to the hardware but for some reason (e.g. lack of
295 * resources) could not be sent to the hardware. As soon as the
296 * driver can send new requests, requests at this list will
297 * be sent first for a fairer dispatch.
298 */
299 struct list_head dispatch;
300 /**
301 * @state: BLK_MQ_S_* flags. Defines the state of the hw
302 * queue (active, scheduled to restart, stopped).
303 */
304 unsigned long state;
305 } ____cacheline_aligned_in_smp;
306
307 /**
308 * @run_work: Used for scheduling a hardware queue run at a later time.
309 */
310 struct delayed_work run_work;
311 /** @cpumask: Map of available CPUs where this hctx can run. */
312 cpumask_var_t cpumask;
313 /**
314 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
315 * selection from @cpumask.
316 */
317 int next_cpu;
318 /**
319 * @next_cpu_batch: Counter of how many works left in the batch before
320 * changing to the next CPU.
321 */
322 int next_cpu_batch;
323
324 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
325 unsigned long flags;
326
327 /**
328 * @sched_data: Pointer owned by the IO scheduler attached to a request
329 * queue. It's up to the IO scheduler how to use this pointer.
330 */
331 void *sched_data;
332 /**
333 * @queue: Pointer to the request queue that owns this hardware context.
334 */
335 struct request_queue *queue;
336 /** @fq: Queue of requests that need to perform a flush operation. */
337 struct blk_flush_queue *fq;
338
339 /**
340 * @driver_data: Pointer to data owned by the block driver that created
341 * this hctx
342 */
343 void *driver_data;
344
345 /**
346 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
347 * pending request in that software queue.
348 */
349 struct sbitmap ctx_map;
350
351 /**
352 * @dispatch_from: Software queue to be used when no scheduler was
353 * selected.
354 */
355 struct blk_mq_ctx *dispatch_from;
356 /**
357 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
358 * decide if the hw_queue is busy using Exponential Weighted Moving
359 * Average algorithm.
360 */
361 unsigned int dispatch_busy;
362
363 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
364 unsigned short type;
365 /** @nr_ctx: Number of software queues. */
366 unsigned short nr_ctx;
367 /** @ctxs: Array of software queues. */
368 struct blk_mq_ctx **ctxs;
369
370 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
371 spinlock_t dispatch_wait_lock;
372 /**
373 * @dispatch_wait: Waitqueue to put requests when there is no tag
374 * available at the moment, to wait for another try in the future.
375 */
376 wait_queue_entry_t dispatch_wait;
377
378 /**
379 * @wait_index: Index of next available dispatch_wait queue to insert
380 * requests.
381 */
382 atomic_t wait_index;
383
384 /**
385 * @tags: Tags owned by the block driver. A tag at this set is only
386 * assigned when a request is dispatched from a hardware queue.
387 */
388 struct blk_mq_tags *tags;
389 /**
390 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
391 * scheduler associated with a request queue, a tag is assigned when
392 * that request is allocated. Else, this member is not used.
393 */
394 struct blk_mq_tags *sched_tags;
395
396 /** @numa_node: NUMA node the storage adapter has been connected to. */
397 unsigned int numa_node;
398 /** @queue_num: Index of this hardware queue. */
399 unsigned int queue_num;
400
401 /**
402 * @nr_active: Number of active requests. Only used when a tag set is
403 * shared across request queues.
404 */
405 atomic_t nr_active;
406
407 /** @cpuhp_online: List to store request if CPU is going to die */
408 struct hlist_node cpuhp_online;
409 /** @cpuhp_dead: List to store request if some CPU die. */
410 struct hlist_node cpuhp_dead;
411 /** @kobj: Kernel object for sysfs. */
412 struct kobject kobj;
413
414 #ifdef CONFIG_BLK_DEBUG_FS
415 /**
416 * @debugfs_dir: debugfs directory for this hardware queue. Named
417 * as cpu<cpu_number>.
418 */
419 struct dentry *debugfs_dir;
420 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
421 struct dentry *sched_debugfs_dir;
422 #endif
423
424 /**
425 * @hctx_list: if this hctx is not in use, this is an entry in
426 * q->unused_hctx_list.
427 */
428 struct list_head hctx_list;
429 };
430
431 /**
432 * struct blk_mq_queue_map - Map software queues to hardware queues
433 * @mq_map: CPU ID to hardware queue index map. This is an array
434 * with nr_cpu_ids elements. Each element has a value in the range
435 * [@queue_offset, @queue_offset + @nr_queues).
436 * @nr_queues: Number of hardware queues to map CPU IDs onto.
437 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
438 * driver to map each hardware queue type (enum hctx_type) onto a distinct
439 * set of hardware queues.
440 */
441 struct blk_mq_queue_map {
442 unsigned int *mq_map;
443 unsigned int nr_queues;
444 unsigned int queue_offset;
445 };
446
447 /**
448 * enum hctx_type - Type of hardware queue
449 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
450 * @HCTX_TYPE_READ: Just for READ I/O.
451 * @HCTX_TYPE_POLL: Polled I/O of any kind.
452 * @HCTX_MAX_TYPES: Number of types of hctx.
453 */
454 enum hctx_type {
455 HCTX_TYPE_DEFAULT,
456 HCTX_TYPE_READ,
457 HCTX_TYPE_POLL,
458
459 HCTX_MAX_TYPES,
460 };
461
462 /**
463 * struct blk_mq_tag_set - tag set that can be shared between request queues
464 * @ops: Pointers to functions that implement block driver behavior.
465 * @map: One or more ctx -> hctx mappings. One map exists for each
466 * hardware queue type (enum hctx_type) that the driver wishes
467 * to support. There are no restrictions on maps being of the
468 * same size, and it's perfectly legal to share maps between
469 * types.
470 * @nr_maps: Number of elements in the @map array. A number in the range
471 * [1, HCTX_MAX_TYPES].
472 * @nr_hw_queues: Number of hardware queues supported by the block driver that
473 * owns this data structure.
474 * @queue_depth: Number of tags per hardware queue, reserved tags included.
475 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
476 * allocations.
477 * @cmd_size: Number of additional bytes to allocate per request. The block
478 * driver owns these additional bytes.
479 * @numa_node: NUMA node the storage adapter has been connected to.
480 * @timeout: Request processing timeout in jiffies.
481 * @flags: Zero or more BLK_MQ_F_* flags.
482 * @driver_data: Pointer to data owned by the block driver that created this
483 * tag set.
484 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
485 * elements.
486 * @shared_tags:
487 * Shared set of tags. Has @nr_hw_queues elements. If set,
488 * shared by all @tags.
489 * @tag_list_lock: Serializes tag_list accesses.
490 * @tag_list: List of the request queues that use this tag set. See also
491 * request_queue.tag_set_list.
492 * @srcu: Use as lock when type of the request queue is blocking
493 * (BLK_MQ_F_BLOCKING).
494 */
495 struct blk_mq_tag_set {
496 const struct blk_mq_ops *ops;
497 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
498 unsigned int nr_maps;
499 unsigned int nr_hw_queues;
500 unsigned int queue_depth;
501 unsigned int reserved_tags;
502 unsigned int cmd_size;
503 int numa_node;
504 unsigned int timeout;
505 unsigned int flags;
506 void *driver_data;
507
508 struct blk_mq_tags **tags;
509
510 struct blk_mq_tags *shared_tags;
511
512 struct mutex tag_list_lock;
513 struct list_head tag_list;
514 struct srcu_struct *srcu;
515 };
516
517 /**
518 * struct blk_mq_queue_data - Data about a request inserted in a queue
519 *
520 * @rq: Request pointer.
521 * @last: If it is the last request in the queue.
522 */
523 struct blk_mq_queue_data {
524 struct request *rq;
525 bool last;
526 };
527
528 typedef bool (busy_tag_iter_fn)(struct request *, void *);
529
530 /**
531 * struct blk_mq_ops - Callback functions that implements block driver
532 * behaviour.
533 */
534 struct blk_mq_ops {
535 /**
536 * @queue_rq: Queue a new request from block IO.
537 */
538 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
539 const struct blk_mq_queue_data *);
540
541 /**
542 * @commit_rqs: If a driver uses bd->last to judge when to submit
543 * requests to hardware, it must define this function. In case of errors
544 * that make us stop issuing further requests, this hook serves the
545 * purpose of kicking the hardware (which the last request otherwise
546 * would have done).
547 */
548 void (*commit_rqs)(struct blk_mq_hw_ctx *);
549
550 /**
551 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
552 * that each request belongs to the same queue. If the driver doesn't
553 * empty the @rqlist completely, then the rest will be queued
554 * individually by the block layer upon return.
555 */
556 void (*queue_rqs)(struct request **rqlist);
557
558 /**
559 * @get_budget: Reserve budget before queue request, once .queue_rq is
560 * run, it is driver's responsibility to release the
561 * reserved budget. Also we have to handle failure case
562 * of .get_budget for avoiding I/O deadlock.
563 */
564 int (*get_budget)(struct request_queue *);
565
566 /**
567 * @put_budget: Release the reserved budget.
568 */
569 void (*put_budget)(struct request_queue *, int);
570
571 /**
572 * @set_rq_budget_token: store rq's budget token
573 */
574 void (*set_rq_budget_token)(struct request *, int);
575 /**
576 * @get_rq_budget_token: retrieve rq's budget token
577 */
578 int (*get_rq_budget_token)(struct request *);
579
580 /**
581 * @timeout: Called on request timeout.
582 */
583 enum blk_eh_timer_return (*timeout)(struct request *);
584
585 /**
586 * @poll: Called to poll for completion of a specific tag.
587 */
588 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
589
590 /**
591 * @complete: Mark the request as complete.
592 */
593 void (*complete)(struct request *);
594
595 /**
596 * @init_hctx: Called when the block layer side of a hardware queue has
597 * been set up, allowing the driver to allocate/init matching
598 * structures.
599 */
600 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
601 /**
602 * @exit_hctx: Ditto for exit/teardown.
603 */
604 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
605
606 /**
607 * @init_request: Called for every command allocated by the block layer
608 * to allow the driver to set up driver specific data.
609 *
610 * Tag greater than or equal to queue_depth is for setting up
611 * flush request.
612 */
613 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
614 unsigned int, unsigned int);
615 /**
616 * @exit_request: Ditto for exit/teardown.
617 */
618 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
619 unsigned int);
620
621 /**
622 * @cleanup_rq: Called before freeing one request which isn't completed
623 * yet, and usually for freeing the driver private data.
624 */
625 void (*cleanup_rq)(struct request *);
626
627 /**
628 * @busy: If set, returns whether or not this queue currently is busy.
629 */
630 bool (*busy)(struct request_queue *);
631
632 /**
633 * @map_queues: This allows drivers specify their own queue mapping by
634 * overriding the setup-time function that builds the mq_map.
635 */
636 void (*map_queues)(struct blk_mq_tag_set *set);
637
638 #ifdef CONFIG_BLK_DEBUG_FS
639 /**
640 * @show_rq: Used by the debugfs implementation to show driver-specific
641 * information about a request.
642 */
643 void (*show_rq)(struct seq_file *m, struct request *rq);
644 #endif
645 };
646
647 enum {
648 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
649 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
650 /*
651 * Set when this device requires underlying blk-mq device for
652 * completing IO:
653 */
654 BLK_MQ_F_STACKING = 1 << 2,
655 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
656 BLK_MQ_F_BLOCKING = 1 << 5,
657 /* Do not allow an I/O scheduler to be configured. */
658 BLK_MQ_F_NO_SCHED = 1 << 6,
659 /*
660 * Select 'none' during queue registration in case of a single hwq
661 * or shared hwqs instead of 'mq-deadline'.
662 */
663 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
664 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
665 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
666
667 BLK_MQ_S_STOPPED = 0,
668 BLK_MQ_S_TAG_ACTIVE = 1,
669 BLK_MQ_S_SCHED_RESTART = 2,
670
671 /* hw queue is inactive after all its CPUs become offline */
672 BLK_MQ_S_INACTIVE = 3,
673
674 BLK_MQ_MAX_DEPTH = 10240,
675
676 BLK_MQ_CPU_WORK_BATCH = 8,
677 };
678 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
679 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
680 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
681 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
682 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
683 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
684
685 #define BLK_MQ_NO_HCTX_IDX (-1U)
686
687 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set,
688 struct queue_limits *lim, void *queuedata,
689 struct lock_class_key *lkclass);
690 #define blk_mq_alloc_disk(set, lim, queuedata) \
691 ({ \
692 static struct lock_class_key __key; \
693 \
694 __blk_mq_alloc_disk(set, lim, queuedata, &__key); \
695 })
696 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
697 struct lock_class_key *lkclass);
698 struct request_queue *blk_mq_alloc_queue(struct blk_mq_tag_set *set,
699 struct queue_limits *lim, void *queuedata);
700 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
701 struct request_queue *q);
702 void blk_mq_destroy_queue(struct request_queue *);
703
704 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
705 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
706 const struct blk_mq_ops *ops, unsigned int queue_depth,
707 unsigned int set_flags);
708 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
709
710 void blk_mq_free_request(struct request *rq);
711 int blk_rq_poll(struct request *rq, struct io_comp_batch *iob,
712 unsigned int poll_flags);
713
714 bool blk_mq_queue_inflight(struct request_queue *q);
715
716 enum {
717 /* return when out of requests */
718 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
719 /* allocate from reserved pool */
720 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
721 /* set RQF_PM */
722 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
723 };
724
725 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
726 blk_mq_req_flags_t flags);
727 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
728 blk_opf_t opf, blk_mq_req_flags_t flags,
729 unsigned int hctx_idx);
730
731 /*
732 * Tag address space map.
733 */
734 struct blk_mq_tags {
735 unsigned int nr_tags;
736 unsigned int nr_reserved_tags;
737 unsigned int active_queues;
738
739 struct sbitmap_queue bitmap_tags;
740 struct sbitmap_queue breserved_tags;
741
742 struct request **rqs;
743 struct request **static_rqs;
744 struct list_head page_list;
745
746 /*
747 * used to clear request reference in rqs[] before freeing one
748 * request pool
749 */
750 spinlock_t lock;
751 };
752
blk_mq_tag_to_rq(struct blk_mq_tags * tags,unsigned int tag)753 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
754 unsigned int tag)
755 {
756 if (tag < tags->nr_tags) {
757 prefetch(tags->rqs[tag]);
758 return tags->rqs[tag];
759 }
760
761 return NULL;
762 }
763
764 enum {
765 BLK_MQ_UNIQUE_TAG_BITS = 16,
766 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
767 };
768
769 u32 blk_mq_unique_tag(struct request *rq);
770
blk_mq_unique_tag_to_hwq(u32 unique_tag)771 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
772 {
773 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
774 }
775
blk_mq_unique_tag_to_tag(u32 unique_tag)776 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
777 {
778 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
779 }
780
781 /**
782 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
783 * @rq: target request.
784 */
blk_mq_rq_state(struct request * rq)785 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
786 {
787 return READ_ONCE(rq->state);
788 }
789
blk_mq_request_started(struct request * rq)790 static inline int blk_mq_request_started(struct request *rq)
791 {
792 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
793 }
794
blk_mq_request_completed(struct request * rq)795 static inline int blk_mq_request_completed(struct request *rq)
796 {
797 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
798 }
799
800 /*
801 *
802 * Set the state to complete when completing a request from inside ->queue_rq.
803 * This is used by drivers that want to ensure special complete actions that
804 * need access to the request are called on failure, e.g. by nvme for
805 * multipathing.
806 */
blk_mq_set_request_complete(struct request * rq)807 static inline void blk_mq_set_request_complete(struct request *rq)
808 {
809 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
810 }
811
812 /*
813 * Complete the request directly instead of deferring it to softirq or
814 * completing it another CPU. Useful in preemptible instead of an interrupt.
815 */
blk_mq_complete_request_direct(struct request * rq,void (* complete)(struct request * rq))816 static inline void blk_mq_complete_request_direct(struct request *rq,
817 void (*complete)(struct request *rq))
818 {
819 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
820 complete(rq);
821 }
822
823 void blk_mq_start_request(struct request *rq);
824 void blk_mq_end_request(struct request *rq, blk_status_t error);
825 void __blk_mq_end_request(struct request *rq, blk_status_t error);
826 void blk_mq_end_request_batch(struct io_comp_batch *ib);
827
828 /*
829 * Only need start/end time stamping if we have iostat or
830 * blk stats enabled, or using an IO scheduler.
831 */
blk_mq_need_time_stamp(struct request * rq)832 static inline bool blk_mq_need_time_stamp(struct request *rq)
833 {
834 /*
835 * passthrough io doesn't use iostat accounting, cgroup stats
836 * and io scheduler functionalities.
837 */
838 if (blk_rq_is_passthrough(rq))
839 return false;
840 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_USE_SCHED));
841 }
842
blk_mq_is_reserved_rq(struct request * rq)843 static inline bool blk_mq_is_reserved_rq(struct request *rq)
844 {
845 return rq->rq_flags & RQF_RESV;
846 }
847
848 /*
849 * Batched completions only work when there is no I/O error and no special
850 * ->end_io handler.
851 */
blk_mq_add_to_batch(struct request * req,struct io_comp_batch * iob,int ioerror,void (* complete)(struct io_comp_batch *))852 static inline bool blk_mq_add_to_batch(struct request *req,
853 struct io_comp_batch *iob, int ioerror,
854 void (*complete)(struct io_comp_batch *))
855 {
856 /*
857 * blk_mq_end_request_batch() can't end request allocated from
858 * sched tags
859 */
860 if (!iob || (req->rq_flags & RQF_SCHED_TAGS) || ioerror ||
861 (req->end_io && !blk_rq_is_passthrough(req)))
862 return false;
863
864 if (!iob->complete)
865 iob->complete = complete;
866 else if (iob->complete != complete)
867 return false;
868 iob->need_ts |= blk_mq_need_time_stamp(req);
869 rq_list_add(&iob->req_list, req);
870 return true;
871 }
872
873 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
874 void blk_mq_kick_requeue_list(struct request_queue *q);
875 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
876 void blk_mq_complete_request(struct request *rq);
877 bool blk_mq_complete_request_remote(struct request *rq);
878 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
879 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
880 void blk_mq_stop_hw_queues(struct request_queue *q);
881 void blk_mq_start_hw_queues(struct request_queue *q);
882 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
883 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
884 void blk_mq_quiesce_queue(struct request_queue *q);
885 void blk_mq_wait_quiesce_done(struct blk_mq_tag_set *set);
886 void blk_mq_quiesce_tagset(struct blk_mq_tag_set *set);
887 void blk_mq_unquiesce_tagset(struct blk_mq_tag_set *set);
888 void blk_mq_unquiesce_queue(struct request_queue *q);
889 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
890 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
891 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
892 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
893 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
894 busy_tag_iter_fn *fn, void *priv);
895 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
896 void blk_mq_freeze_queue(struct request_queue *q);
897 void blk_mq_unfreeze_queue(struct request_queue *q);
898 void blk_freeze_queue_start(struct request_queue *q);
899 void blk_mq_freeze_queue_wait(struct request_queue *q);
900 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
901 unsigned long timeout);
902
903 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
904 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
905
906 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
907
908 unsigned int blk_mq_rq_cpu(struct request *rq);
909
910 bool __blk_should_fake_timeout(struct request_queue *q);
blk_should_fake_timeout(struct request_queue * q)911 static inline bool blk_should_fake_timeout(struct request_queue *q)
912 {
913 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
914 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
915 return __blk_should_fake_timeout(q);
916 return false;
917 }
918
919 /**
920 * blk_mq_rq_from_pdu - cast a PDU to a request
921 * @pdu: the PDU (Protocol Data Unit) to be casted
922 *
923 * Return: request
924 *
925 * Driver command data is immediately after the request. So subtract request
926 * size to get back to the original request.
927 */
blk_mq_rq_from_pdu(void * pdu)928 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
929 {
930 return pdu - sizeof(struct request);
931 }
932
933 /**
934 * blk_mq_rq_to_pdu - cast a request to a PDU
935 * @rq: the request to be casted
936 *
937 * Return: pointer to the PDU
938 *
939 * Driver command data is immediately after the request. So add request to get
940 * the PDU.
941 */
blk_mq_rq_to_pdu(struct request * rq)942 static inline void *blk_mq_rq_to_pdu(struct request *rq)
943 {
944 return rq + 1;
945 }
946
947 #define queue_for_each_hw_ctx(q, hctx, i) \
948 xa_for_each(&(q)->hctx_table, (i), (hctx))
949
950 #define hctx_for_each_ctx(hctx, ctx, i) \
951 for ((i) = 0; (i) < (hctx)->nr_ctx && \
952 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
953
blk_mq_cleanup_rq(struct request * rq)954 static inline void blk_mq_cleanup_rq(struct request *rq)
955 {
956 if (rq->q->mq_ops->cleanup_rq)
957 rq->q->mq_ops->cleanup_rq(rq);
958 }
959
blk_rq_bio_prep(struct request * rq,struct bio * bio,unsigned int nr_segs)960 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
961 unsigned int nr_segs)
962 {
963 rq->nr_phys_segments = nr_segs;
964 rq->__data_len = bio->bi_iter.bi_size;
965 rq->bio = rq->biotail = bio;
966 rq->ioprio = bio_prio(bio);
967 }
968
969 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
970 struct lock_class_key *key);
971
rq_is_sync(struct request * rq)972 static inline bool rq_is_sync(struct request *rq)
973 {
974 return op_is_sync(rq->cmd_flags);
975 }
976
977 void blk_rq_init(struct request_queue *q, struct request *rq);
978 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
979 struct bio_set *bs, gfp_t gfp_mask,
980 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
981 void blk_rq_unprep_clone(struct request *rq);
982 blk_status_t blk_insert_cloned_request(struct request *rq);
983
984 struct rq_map_data {
985 struct page **pages;
986 unsigned long offset;
987 unsigned short page_order;
988 unsigned short nr_entries;
989 bool null_mapped;
990 bool from_user;
991 };
992
993 int blk_rq_map_user(struct request_queue *, struct request *,
994 struct rq_map_data *, void __user *, unsigned long, gfp_t);
995 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
996 void __user *, unsigned long, gfp_t, bool, int, bool, int);
997 int blk_rq_map_user_iov(struct request_queue *, struct request *,
998 struct rq_map_data *, const struct iov_iter *, gfp_t);
999 int blk_rq_unmap_user(struct bio *);
1000 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
1001 unsigned int, gfp_t);
1002 int blk_rq_append_bio(struct request *rq, struct bio *bio);
1003 void blk_execute_rq_nowait(struct request *rq, bool at_head);
1004 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
1005 bool blk_rq_is_poll(struct request *rq);
1006
1007 struct req_iterator {
1008 struct bvec_iter iter;
1009 struct bio *bio;
1010 };
1011
1012 #define __rq_for_each_bio(_bio, rq) \
1013 if ((rq->bio)) \
1014 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1015
1016 #define rq_for_each_segment(bvl, _rq, _iter) \
1017 __rq_for_each_bio(_iter.bio, _rq) \
1018 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1019
1020 #define rq_for_each_bvec(bvl, _rq, _iter) \
1021 __rq_for_each_bio(_iter.bio, _rq) \
1022 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1023
1024 #define rq_iter_last(bvec, _iter) \
1025 (_iter.bio->bi_next == NULL && \
1026 bio_iter_last(bvec, _iter.iter))
1027
1028 /*
1029 * blk_rq_pos() : the current sector
1030 * blk_rq_bytes() : bytes left in the entire request
1031 * blk_rq_cur_bytes() : bytes left in the current segment
1032 * blk_rq_sectors() : sectors left in the entire request
1033 * blk_rq_cur_sectors() : sectors left in the current segment
1034 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1035 */
blk_rq_pos(const struct request * rq)1036 static inline sector_t blk_rq_pos(const struct request *rq)
1037 {
1038 return rq->__sector;
1039 }
1040
blk_rq_bytes(const struct request * rq)1041 static inline unsigned int blk_rq_bytes(const struct request *rq)
1042 {
1043 return rq->__data_len;
1044 }
1045
blk_rq_cur_bytes(const struct request * rq)1046 static inline int blk_rq_cur_bytes(const struct request *rq)
1047 {
1048 if (!rq->bio)
1049 return 0;
1050 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1051 return rq->bio->bi_iter.bi_size;
1052 return bio_iovec(rq->bio).bv_len;
1053 }
1054
blk_rq_sectors(const struct request * rq)1055 static inline unsigned int blk_rq_sectors(const struct request *rq)
1056 {
1057 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1058 }
1059
blk_rq_cur_sectors(const struct request * rq)1060 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1061 {
1062 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1063 }
1064
blk_rq_stats_sectors(const struct request * rq)1065 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1066 {
1067 return rq->stats_sectors;
1068 }
1069
1070 /*
1071 * Some commands like WRITE SAME have a payload or data transfer size which
1072 * is different from the size of the request. Any driver that supports such
1073 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1074 * calculate the data transfer size.
1075 */
blk_rq_payload_bytes(struct request * rq)1076 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1077 {
1078 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1079 return rq->special_vec.bv_len;
1080 return blk_rq_bytes(rq);
1081 }
1082
1083 /*
1084 * Return the first full biovec in the request. The caller needs to check that
1085 * there are any bvecs before calling this helper.
1086 */
req_bvec(struct request * rq)1087 static inline struct bio_vec req_bvec(struct request *rq)
1088 {
1089 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1090 return rq->special_vec;
1091 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1092 }
1093
blk_rq_count_bios(struct request * rq)1094 static inline unsigned int blk_rq_count_bios(struct request *rq)
1095 {
1096 unsigned int nr_bios = 0;
1097 struct bio *bio;
1098
1099 __rq_for_each_bio(bio, rq)
1100 nr_bios++;
1101
1102 return nr_bios;
1103 }
1104
1105 void blk_steal_bios(struct bio_list *list, struct request *rq);
1106
1107 /*
1108 * Request completion related functions.
1109 *
1110 * blk_update_request() completes given number of bytes and updates
1111 * the request without completing it.
1112 */
1113 bool blk_update_request(struct request *rq, blk_status_t error,
1114 unsigned int nr_bytes);
1115 void blk_abort_request(struct request *);
1116
1117 /*
1118 * Number of physical segments as sent to the device.
1119 *
1120 * Normally this is the number of discontiguous data segments sent by the
1121 * submitter. But for data-less command like discard we might have no
1122 * actual data segments submitted, but the driver might have to add it's
1123 * own special payload. In that case we still return 1 here so that this
1124 * special payload will be mapped.
1125 */
blk_rq_nr_phys_segments(struct request * rq)1126 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1127 {
1128 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1129 return 1;
1130 return rq->nr_phys_segments;
1131 }
1132
1133 /*
1134 * Number of discard segments (or ranges) the driver needs to fill in.
1135 * Each discard bio merged into a request is counted as one segment.
1136 */
blk_rq_nr_discard_segments(struct request * rq)1137 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1138 {
1139 return max_t(unsigned short, rq->nr_phys_segments, 1);
1140 }
1141
1142 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1143 struct scatterlist *sglist, struct scatterlist **last_sg);
blk_rq_map_sg(struct request_queue * q,struct request * rq,struct scatterlist * sglist)1144 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1145 struct scatterlist *sglist)
1146 {
1147 struct scatterlist *last_sg = NULL;
1148
1149 return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1150 }
1151 void blk_dump_rq_flags(struct request *, char *);
1152
1153 #endif /* BLK_MQ_H */
1154