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