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
11 struct blk_mq_tags;
12 struct blk_flush_queue;
13
14 #define BLKDEV_MIN_RQ 4
15 #define BLKDEV_DEFAULT_RQ 128
16
17 enum rq_end_io_ret {
18 RQ_END_IO_NONE,
19 RQ_END_IO_FREE,
20 };
21
22 typedef enum rq_end_io_ret (rq_end_io_fn)(struct request *, blk_status_t);
23
24 /*
25 * request flags */
26 typedef __u32 __bitwise req_flags_t;
27
28 /* drive already may have started this one */
29 #define RQF_STARTED ((__force req_flags_t)(1 << 1))
30 /* may not be passed by ioscheduler */
31 #define RQF_SOFTBARRIER ((__force req_flags_t)(1 << 3))
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 /* track inflight for MQ */
37 #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6))
38 /* don't call prep for this one */
39 #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
40 /* vaguely specified driver internal error. Ignored by the block layer */
41 #define RQF_FAILED ((__force req_flags_t)(1 << 10))
42 /* don't warn about errors */
43 #define RQF_QUIET ((__force req_flags_t)(1 << 11))
44 /* elevator private data attached */
45 #define RQF_ELVPRIV ((__force req_flags_t)(1 << 12))
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 per-zone write lock is held for this request */
58 #define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19))
59 /* already slept for hybrid poll */
60 #define RQF_MQ_POLL_SLEPT ((__force req_flags_t)(1 << 20))
61 /* ->timeout has been called, don't expire again */
62 #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
63 /* queue has elevator attached */
64 #define RQF_ELV ((__force req_flags_t)(1 << 22))
65 #define RQF_RESV ((__force req_flags_t)(1 << 23))
66
67 /* flags that prevent us from merging requests: */
68 #define RQF_NOMERGE_FLAGS \
69 (RQF_STARTED | RQF_SOFTBARRIER | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
70
71 enum mq_rq_state {
72 MQ_RQ_IDLE = 0,
73 MQ_RQ_IN_FLIGHT = 1,
74 MQ_RQ_COMPLETE = 2,
75 };
76
77 /*
78 * Try to put the fields that are referenced together in the same cacheline.
79 *
80 * If you modify this structure, make sure to update blk_rq_init() and
81 * especially blk_mq_rq_ctx_init() to take care of the added fields.
82 */
83 struct request {
84 struct request_queue *q;
85 struct blk_mq_ctx *mq_ctx;
86 struct blk_mq_hw_ctx *mq_hctx;
87
88 blk_opf_t cmd_flags; /* op and common flags */
89 req_flags_t rq_flags;
90
91 int tag;
92 int internal_tag;
93
94 unsigned int timeout;
95
96 /* the following two fields are internal, NEVER access directly */
97 unsigned int __data_len; /* total data len */
98 sector_t __sector; /* sector cursor */
99
100 struct bio *bio;
101 struct bio *biotail;
102
103 union {
104 struct list_head queuelist;
105 struct request *rq_next;
106 };
107
108 struct block_device *part;
109 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
110 /* Time that the first bio started allocating this request. */
111 u64 alloc_time_ns;
112 #endif
113 /* Time that this request was allocated for this IO. */
114 u64 start_time_ns;
115 /* Time that I/O was submitted to the device. */
116 u64 io_start_time_ns;
117
118 #ifdef CONFIG_BLK_WBT
119 unsigned short wbt_flags;
120 #endif
121 /*
122 * rq sectors used for blk stats. It has the same value
123 * with blk_rq_sectors(rq), except that it never be zeroed
124 * by completion.
125 */
126 unsigned short stats_sectors;
127
128 /*
129 * Number of scatter-gather DMA addr+len pairs after
130 * physical address coalescing is performed.
131 */
132 unsigned short nr_phys_segments;
133
134 #ifdef CONFIG_BLK_DEV_INTEGRITY
135 unsigned short nr_integrity_segments;
136 #endif
137
138 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
139 struct bio_crypt_ctx *crypt_ctx;
140 struct blk_crypto_keyslot *crypt_keyslot;
141 #endif
142
143 unsigned short write_hint;
144 unsigned short ioprio;
145
146 enum mq_rq_state state;
147 atomic_t ref;
148
149 unsigned long deadline;
150
151 /*
152 * The hash is used inside the scheduler, and killed once the
153 * request reaches the dispatch list. The ipi_list is only used
154 * to queue the request for softirq completion, which is long
155 * after the request has been unhashed (and even removed from
156 * the dispatch list).
157 */
158 union {
159 struct hlist_node hash; /* merge hash */
160 struct llist_node ipi_list;
161 };
162
163 /*
164 * The rb_node is only used inside the io scheduler, requests
165 * are pruned when moved to the dispatch queue. So let the
166 * completion_data share space with the rb_node.
167 */
168 union {
169 struct rb_node rb_node; /* sort/lookup */
170 struct bio_vec special_vec;
171 void *completion_data;
172 };
173
174
175 /*
176 * Three pointers are available for the IO schedulers, if they need
177 * more they have to dynamically allocate it. Flush requests are
178 * never put on the IO scheduler. So let the flush fields share
179 * space with the elevator data.
180 */
181 union {
182 struct {
183 struct io_cq *icq;
184 void *priv[2];
185 } elv;
186
187 struct {
188 unsigned int seq;
189 struct list_head list;
190 rq_end_io_fn *saved_end_io;
191 } flush;
192 };
193
194 union {
195 struct __call_single_data csd;
196 u64 fifo_time;
197 };
198
199 /*
200 * completion callback.
201 */
202 rq_end_io_fn *end_io;
203 void *end_io_data;
204 };
205
req_op(const struct request * req)206 static inline enum req_op req_op(const struct request *req)
207 {
208 return req->cmd_flags & REQ_OP_MASK;
209 }
210
blk_rq_is_passthrough(struct request * rq)211 static inline bool blk_rq_is_passthrough(struct request *rq)
212 {
213 return blk_op_is_passthrough(req_op(rq));
214 }
215
req_get_ioprio(struct request * req)216 static inline unsigned short req_get_ioprio(struct request *req)
217 {
218 return req->ioprio;
219 }
220
221 #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
222
223 #define rq_dma_dir(rq) \
224 (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
225
226 #define rq_list_add(listptr, rq) do { \
227 (rq)->rq_next = *(listptr); \
228 *(listptr) = rq; \
229 } while (0)
230
231 #define rq_list_pop(listptr) \
232 ({ \
233 struct request *__req = NULL; \
234 if ((listptr) && *(listptr)) { \
235 __req = *(listptr); \
236 *(listptr) = __req->rq_next; \
237 } \
238 __req; \
239 })
240
241 #define rq_list_peek(listptr) \
242 ({ \
243 struct request *__req = NULL; \
244 if ((listptr) && *(listptr)) \
245 __req = *(listptr); \
246 __req; \
247 })
248
249 #define rq_list_for_each(listptr, pos) \
250 for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
251
252 #define rq_list_for_each_safe(listptr, pos, nxt) \
253 for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
254 pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
255
256 #define rq_list_next(rq) (rq)->rq_next
257 #define rq_list_empty(list) ((list) == (struct request *) NULL)
258
259 /**
260 * rq_list_move() - move a struct request from one list to another
261 * @src: The source list @rq is currently in
262 * @dst: The destination list that @rq will be appended to
263 * @rq: The request to move
264 * @prev: The request preceding @rq in @src (NULL if @rq is the head)
265 */
rq_list_move(struct request ** src,struct request ** dst,struct request * rq,struct request * prev)266 static inline void rq_list_move(struct request **src, struct request **dst,
267 struct request *rq, struct request *prev)
268 {
269 if (prev)
270 prev->rq_next = rq->rq_next;
271 else
272 *src = rq->rq_next;
273 rq_list_add(dst, rq);
274 }
275
276 /**
277 * enum blk_eh_timer_return - How the timeout handler should proceed
278 * @BLK_EH_DONE: The block driver completed the command or will complete it at
279 * a later time.
280 * @BLK_EH_RESET_TIMER: Reset the request timer and continue waiting for the
281 * request to complete.
282 */
283 enum blk_eh_timer_return {
284 BLK_EH_DONE,
285 BLK_EH_RESET_TIMER,
286 };
287
288 #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
289 #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
290
291 /**
292 * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
293 * block device
294 */
295 struct blk_mq_hw_ctx {
296 struct {
297 /** @lock: Protects the dispatch list. */
298 spinlock_t lock;
299 /**
300 * @dispatch: Used for requests that are ready to be
301 * dispatched to the hardware but for some reason (e.g. lack of
302 * resources) could not be sent to the hardware. As soon as the
303 * driver can send new requests, requests at this list will
304 * be sent first for a fairer dispatch.
305 */
306 struct list_head dispatch;
307 /**
308 * @state: BLK_MQ_S_* flags. Defines the state of the hw
309 * queue (active, scheduled to restart, stopped).
310 */
311 unsigned long state;
312 } ____cacheline_aligned_in_smp;
313
314 /**
315 * @run_work: Used for scheduling a hardware queue run at a later time.
316 */
317 struct delayed_work run_work;
318 /** @cpumask: Map of available CPUs where this hctx can run. */
319 cpumask_var_t cpumask;
320 /**
321 * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
322 * selection from @cpumask.
323 */
324 int next_cpu;
325 /**
326 * @next_cpu_batch: Counter of how many works left in the batch before
327 * changing to the next CPU.
328 */
329 int next_cpu_batch;
330
331 /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
332 unsigned long flags;
333
334 /**
335 * @sched_data: Pointer owned by the IO scheduler attached to a request
336 * queue. It's up to the IO scheduler how to use this pointer.
337 */
338 void *sched_data;
339 /**
340 * @queue: Pointer to the request queue that owns this hardware context.
341 */
342 struct request_queue *queue;
343 /** @fq: Queue of requests that need to perform a flush operation. */
344 struct blk_flush_queue *fq;
345
346 /**
347 * @driver_data: Pointer to data owned by the block driver that created
348 * this hctx
349 */
350 void *driver_data;
351
352 /**
353 * @ctx_map: Bitmap for each software queue. If bit is on, there is a
354 * pending request in that software queue.
355 */
356 struct sbitmap ctx_map;
357
358 /**
359 * @dispatch_from: Software queue to be used when no scheduler was
360 * selected.
361 */
362 struct blk_mq_ctx *dispatch_from;
363 /**
364 * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
365 * decide if the hw_queue is busy using Exponential Weighted Moving
366 * Average algorithm.
367 */
368 unsigned int dispatch_busy;
369
370 /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
371 unsigned short type;
372 /** @nr_ctx: Number of software queues. */
373 unsigned short nr_ctx;
374 /** @ctxs: Array of software queues. */
375 struct blk_mq_ctx **ctxs;
376
377 /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
378 spinlock_t dispatch_wait_lock;
379 /**
380 * @dispatch_wait: Waitqueue to put requests when there is no tag
381 * available at the moment, to wait for another try in the future.
382 */
383 wait_queue_entry_t dispatch_wait;
384
385 /**
386 * @wait_index: Index of next available dispatch_wait queue to insert
387 * requests.
388 */
389 atomic_t wait_index;
390
391 /**
392 * @tags: Tags owned by the block driver. A tag at this set is only
393 * assigned when a request is dispatched from a hardware queue.
394 */
395 struct blk_mq_tags *tags;
396 /**
397 * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
398 * scheduler associated with a request queue, a tag is assigned when
399 * that request is allocated. Else, this member is not used.
400 */
401 struct blk_mq_tags *sched_tags;
402
403 /** @queued: Number of queued requests. */
404 unsigned long queued;
405 /** @run: Number of dispatched requests. */
406 unsigned long run;
407
408 /** @numa_node: NUMA node the storage adapter has been connected to. */
409 unsigned int numa_node;
410 /** @queue_num: Index of this hardware queue. */
411 unsigned int queue_num;
412
413 /**
414 * @nr_active: Number of active requests. Only used when a tag set is
415 * shared across request queues.
416 */
417 atomic_t nr_active;
418
419 /** @cpuhp_online: List to store request if CPU is going to die */
420 struct hlist_node cpuhp_online;
421 /** @cpuhp_dead: List to store request if some CPU die. */
422 struct hlist_node cpuhp_dead;
423 /** @kobj: Kernel object for sysfs. */
424 struct kobject kobj;
425
426 #ifdef CONFIG_BLK_DEBUG_FS
427 /**
428 * @debugfs_dir: debugfs directory for this hardware queue. Named
429 * as cpu<cpu_number>.
430 */
431 struct dentry *debugfs_dir;
432 /** @sched_debugfs_dir: debugfs directory for the scheduler. */
433 struct dentry *sched_debugfs_dir;
434 #endif
435
436 /**
437 * @hctx_list: if this hctx is not in use, this is an entry in
438 * q->unused_hctx_list.
439 */
440 struct list_head hctx_list;
441 };
442
443 /**
444 * struct blk_mq_queue_map - Map software queues to hardware queues
445 * @mq_map: CPU ID to hardware queue index map. This is an array
446 * with nr_cpu_ids elements. Each element has a value in the range
447 * [@queue_offset, @queue_offset + @nr_queues).
448 * @nr_queues: Number of hardware queues to map CPU IDs onto.
449 * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
450 * driver to map each hardware queue type (enum hctx_type) onto a distinct
451 * set of hardware queues.
452 */
453 struct blk_mq_queue_map {
454 unsigned int *mq_map;
455 unsigned int nr_queues;
456 unsigned int queue_offset;
457 };
458
459 /**
460 * enum hctx_type - Type of hardware queue
461 * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
462 * @HCTX_TYPE_READ: Just for READ I/O.
463 * @HCTX_TYPE_POLL: Polled I/O of any kind.
464 * @HCTX_MAX_TYPES: Number of types of hctx.
465 */
466 enum hctx_type {
467 HCTX_TYPE_DEFAULT,
468 HCTX_TYPE_READ,
469 HCTX_TYPE_POLL,
470
471 HCTX_MAX_TYPES,
472 };
473
474 /**
475 * struct blk_mq_tag_set - tag set that can be shared between request queues
476 * @map: One or more ctx -> hctx mappings. One map exists for each
477 * hardware queue type (enum hctx_type) that the driver wishes
478 * to support. There are no restrictions on maps being of the
479 * same size, and it's perfectly legal to share maps between
480 * types.
481 * @nr_maps: Number of elements in the @map array. A number in the range
482 * [1, HCTX_MAX_TYPES].
483 * @ops: Pointers to functions that implement block driver behavior.
484 * @nr_hw_queues: Number of hardware queues supported by the block driver that
485 * owns this data structure.
486 * @queue_depth: Number of tags per hardware queue, reserved tags included.
487 * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
488 * allocations.
489 * @cmd_size: Number of additional bytes to allocate per request. The block
490 * driver owns these additional bytes.
491 * @numa_node: NUMA node the storage adapter has been connected to.
492 * @timeout: Request processing timeout in jiffies.
493 * @flags: Zero or more BLK_MQ_F_* flags.
494 * @driver_data: Pointer to data owned by the block driver that created this
495 * tag set.
496 * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
497 * elements.
498 * @shared_tags:
499 * Shared set of tags. Has @nr_hw_queues elements. If set,
500 * shared by all @tags.
501 * @tag_list_lock: Serializes tag_list accesses.
502 * @tag_list: List of the request queues that use this tag set. See also
503 * request_queue.tag_set_list.
504 */
505 struct blk_mq_tag_set {
506 struct blk_mq_queue_map map[HCTX_MAX_TYPES];
507 unsigned int nr_maps;
508 const struct blk_mq_ops *ops;
509 unsigned int nr_hw_queues;
510 unsigned int queue_depth;
511 unsigned int reserved_tags;
512 unsigned int cmd_size;
513 int numa_node;
514 unsigned int timeout;
515 unsigned int flags;
516 void *driver_data;
517
518 struct blk_mq_tags **tags;
519
520 struct blk_mq_tags *shared_tags;
521
522 struct mutex tag_list_lock;
523 struct list_head tag_list;
524 };
525
526 /**
527 * struct blk_mq_queue_data - Data about a request inserted in a queue
528 *
529 * @rq: Request pointer.
530 * @last: If it is the last request in the queue.
531 */
532 struct blk_mq_queue_data {
533 struct request *rq;
534 bool last;
535 };
536
537 typedef bool (busy_tag_iter_fn)(struct request *, void *);
538
539 /**
540 * struct blk_mq_ops - Callback functions that implements block driver
541 * behaviour.
542 */
543 struct blk_mq_ops {
544 /**
545 * @queue_rq: Queue a new request from block IO.
546 */
547 blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
548 const struct blk_mq_queue_data *);
549
550 /**
551 * @commit_rqs: If a driver uses bd->last to judge when to submit
552 * requests to hardware, it must define this function. In case of errors
553 * that make us stop issuing further requests, this hook serves the
554 * purpose of kicking the hardware (which the last request otherwise
555 * would have done).
556 */
557 void (*commit_rqs)(struct blk_mq_hw_ctx *);
558
559 /**
560 * @queue_rqs: Queue a list of new requests. Driver is guaranteed
561 * that each request belongs to the same queue. If the driver doesn't
562 * empty the @rqlist completely, then the rest will be queued
563 * individually by the block layer upon return.
564 */
565 void (*queue_rqs)(struct request **rqlist);
566
567 /**
568 * @get_budget: Reserve budget before queue request, once .queue_rq is
569 * run, it is driver's responsibility to release the
570 * reserved budget. Also we have to handle failure case
571 * of .get_budget for avoiding I/O deadlock.
572 */
573 int (*get_budget)(struct request_queue *);
574
575 /**
576 * @put_budget: Release the reserved budget.
577 */
578 void (*put_budget)(struct request_queue *, int);
579
580 /**
581 * @set_rq_budget_token: store rq's budget token
582 */
583 void (*set_rq_budget_token)(struct request *, int);
584 /**
585 * @get_rq_budget_token: retrieve rq's budget token
586 */
587 int (*get_rq_budget_token)(struct request *);
588
589 /**
590 * @timeout: Called on request timeout.
591 */
592 enum blk_eh_timer_return (*timeout)(struct request *);
593
594 /**
595 * @poll: Called to poll for completion of a specific tag.
596 */
597 int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
598
599 /**
600 * @complete: Mark the request as complete.
601 */
602 void (*complete)(struct request *);
603
604 /**
605 * @init_hctx: Called when the block layer side of a hardware queue has
606 * been set up, allowing the driver to allocate/init matching
607 * structures.
608 */
609 int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
610 /**
611 * @exit_hctx: Ditto for exit/teardown.
612 */
613 void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
614
615 /**
616 * @init_request: Called for every command allocated by the block layer
617 * to allow the driver to set up driver specific data.
618 *
619 * Tag greater than or equal to queue_depth is for setting up
620 * flush request.
621 */
622 int (*init_request)(struct blk_mq_tag_set *set, struct request *,
623 unsigned int, unsigned int);
624 /**
625 * @exit_request: Ditto for exit/teardown.
626 */
627 void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
628 unsigned int);
629
630 /**
631 * @cleanup_rq: Called before freeing one request which isn't completed
632 * yet, and usually for freeing the driver private data.
633 */
634 void (*cleanup_rq)(struct request *);
635
636 /**
637 * @busy: If set, returns whether or not this queue currently is busy.
638 */
639 bool (*busy)(struct request_queue *);
640
641 /**
642 * @map_queues: This allows drivers specify their own queue mapping by
643 * overriding the setup-time function that builds the mq_map.
644 */
645 void (*map_queues)(struct blk_mq_tag_set *set);
646
647 #ifdef CONFIG_BLK_DEBUG_FS
648 /**
649 * @show_rq: Used by the debugfs implementation to show driver-specific
650 * information about a request.
651 */
652 void (*show_rq)(struct seq_file *m, struct request *rq);
653 #endif
654 };
655
656 enum {
657 BLK_MQ_F_SHOULD_MERGE = 1 << 0,
658 BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
659 /*
660 * Set when this device requires underlying blk-mq device for
661 * completing IO:
662 */
663 BLK_MQ_F_STACKING = 1 << 2,
664 BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
665 BLK_MQ_F_BLOCKING = 1 << 5,
666 /* Do not allow an I/O scheduler to be configured. */
667 BLK_MQ_F_NO_SCHED = 1 << 6,
668 /*
669 * Select 'none' during queue registration in case of a single hwq
670 * or shared hwqs instead of 'mq-deadline'.
671 */
672 BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
673 BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
674 BLK_MQ_F_ALLOC_POLICY_BITS = 1,
675
676 BLK_MQ_S_STOPPED = 0,
677 BLK_MQ_S_TAG_ACTIVE = 1,
678 BLK_MQ_S_SCHED_RESTART = 2,
679
680 /* hw queue is inactive after all its CPUs become offline */
681 BLK_MQ_S_INACTIVE = 3,
682
683 BLK_MQ_MAX_DEPTH = 10240,
684
685 BLK_MQ_CPU_WORK_BATCH = 8,
686 };
687 #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
688 ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
689 ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
690 #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
691 ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
692 << BLK_MQ_F_ALLOC_POLICY_START_BIT)
693
694 #define BLK_MQ_NO_HCTX_IDX (-1U)
695
696 struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
697 struct lock_class_key *lkclass);
698 #define blk_mq_alloc_disk(set, queuedata) \
699 ({ \
700 static struct lock_class_key __key; \
701 \
702 __blk_mq_alloc_disk(set, queuedata, &__key); \
703 })
704 struct gendisk *blk_mq_alloc_disk_for_queue(struct request_queue *q,
705 struct lock_class_key *lkclass);
706 struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
707 int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
708 struct request_queue *q);
709 void blk_mq_destroy_queue(struct request_queue *);
710
711 int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
712 int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
713 const struct blk_mq_ops *ops, unsigned int queue_depth,
714 unsigned int set_flags);
715 void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
716
717 void blk_mq_free_request(struct request *rq);
718
719 bool blk_mq_queue_inflight(struct request_queue *q);
720
721 enum {
722 /* return when out of requests */
723 BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
724 /* allocate from reserved pool */
725 BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
726 /* set RQF_PM */
727 BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
728 };
729
730 struct request *blk_mq_alloc_request(struct request_queue *q, blk_opf_t opf,
731 blk_mq_req_flags_t flags);
732 struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
733 blk_opf_t opf, blk_mq_req_flags_t flags,
734 unsigned int hctx_idx);
735
736 /*
737 * Tag address space map.
738 */
739 struct blk_mq_tags {
740 unsigned int nr_tags;
741 unsigned int nr_reserved_tags;
742
743 atomic_t active_queues;
744
745 struct sbitmap_queue bitmap_tags;
746 struct sbitmap_queue breserved_tags;
747
748 struct request **rqs;
749 struct request **static_rqs;
750 struct list_head page_list;
751
752 /*
753 * used to clear request reference in rqs[] before freeing one
754 * request pool
755 */
756 spinlock_t lock;
757 };
758
blk_mq_tag_to_rq(struct blk_mq_tags * tags,unsigned int tag)759 static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
760 unsigned int tag)
761 {
762 if (tag < tags->nr_tags) {
763 prefetch(tags->rqs[tag]);
764 return tags->rqs[tag];
765 }
766
767 return NULL;
768 }
769
770 enum {
771 BLK_MQ_UNIQUE_TAG_BITS = 16,
772 BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
773 };
774
775 u32 blk_mq_unique_tag(struct request *rq);
776
blk_mq_unique_tag_to_hwq(u32 unique_tag)777 static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
778 {
779 return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
780 }
781
blk_mq_unique_tag_to_tag(u32 unique_tag)782 static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
783 {
784 return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
785 }
786
787 /**
788 * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
789 * @rq: target request.
790 */
blk_mq_rq_state(struct request * rq)791 static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
792 {
793 return READ_ONCE(rq->state);
794 }
795
blk_mq_request_started(struct request * rq)796 static inline int blk_mq_request_started(struct request *rq)
797 {
798 return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
799 }
800
blk_mq_request_completed(struct request * rq)801 static inline int blk_mq_request_completed(struct request *rq)
802 {
803 return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
804 }
805
806 /*
807 *
808 * Set the state to complete when completing a request from inside ->queue_rq.
809 * This is used by drivers that want to ensure special complete actions that
810 * need access to the request are called on failure, e.g. by nvme for
811 * multipathing.
812 */
blk_mq_set_request_complete(struct request * rq)813 static inline void blk_mq_set_request_complete(struct request *rq)
814 {
815 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
816 }
817
818 /*
819 * Complete the request directly instead of deferring it to softirq or
820 * completing it another CPU. Useful in preemptible instead of an interrupt.
821 */
blk_mq_complete_request_direct(struct request * rq,void (* complete)(struct request * rq))822 static inline void blk_mq_complete_request_direct(struct request *rq,
823 void (*complete)(struct request *rq))
824 {
825 WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
826 complete(rq);
827 }
828
829 void blk_mq_start_request(struct request *rq);
830 void blk_mq_end_request(struct request *rq, blk_status_t error);
831 void __blk_mq_end_request(struct request *rq, blk_status_t error);
832 void blk_mq_end_request_batch(struct io_comp_batch *ib);
833
834 /*
835 * Only need start/end time stamping if we have iostat or
836 * blk stats enabled, or using an IO scheduler.
837 */
blk_mq_need_time_stamp(struct request * rq)838 static inline bool blk_mq_need_time_stamp(struct request *rq)
839 {
840 return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_ELV));
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 if (!iob || (req->rq_flags & RQF_ELV) || ioerror ||
857 (req->end_io && !blk_rq_is_passthrough(req)))
858 return false;
859
860 if (!iob->complete)
861 iob->complete = complete;
862 else if (iob->complete != complete)
863 return false;
864 iob->need_ts |= blk_mq_need_time_stamp(req);
865 rq_list_add(&iob->req_list, req);
866 return true;
867 }
868
869 void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
870 void blk_mq_kick_requeue_list(struct request_queue *q);
871 void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
872 void blk_mq_complete_request(struct request *rq);
873 bool blk_mq_complete_request_remote(struct request *rq);
874 void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
875 void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
876 void blk_mq_stop_hw_queues(struct request_queue *q);
877 void blk_mq_start_hw_queues(struct request_queue *q);
878 void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
879 void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
880 void blk_mq_quiesce_queue(struct request_queue *q);
881 void blk_mq_wait_quiesce_done(struct request_queue *q);
882 void blk_mq_unquiesce_queue(struct request_queue *q);
883 void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
884 void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
885 void blk_mq_run_hw_queues(struct request_queue *q, bool async);
886 void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
887 void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
888 busy_tag_iter_fn *fn, void *priv);
889 void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
890 void blk_mq_freeze_queue(struct request_queue *q);
891 void blk_mq_unfreeze_queue(struct request_queue *q);
892 void blk_freeze_queue_start(struct request_queue *q);
893 void blk_mq_freeze_queue_wait(struct request_queue *q);
894 int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
895 unsigned long timeout);
896
897 void blk_mq_map_queues(struct blk_mq_queue_map *qmap);
898 void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
899
900 void blk_mq_quiesce_queue_nowait(struct request_queue *q);
901
902 unsigned int blk_mq_rq_cpu(struct request *rq);
903
904 bool __blk_should_fake_timeout(struct request_queue *q);
blk_should_fake_timeout(struct request_queue * q)905 static inline bool blk_should_fake_timeout(struct request_queue *q)
906 {
907 if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
908 test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
909 return __blk_should_fake_timeout(q);
910 return false;
911 }
912
913 /**
914 * blk_mq_rq_from_pdu - cast a PDU to a request
915 * @pdu: the PDU (Protocol Data Unit) to be casted
916 *
917 * Return: request
918 *
919 * Driver command data is immediately after the request. So subtract request
920 * size to get back to the original request.
921 */
blk_mq_rq_from_pdu(void * pdu)922 static inline struct request *blk_mq_rq_from_pdu(void *pdu)
923 {
924 return pdu - sizeof(struct request);
925 }
926
927 /**
928 * blk_mq_rq_to_pdu - cast a request to a PDU
929 * @rq: the request to be casted
930 *
931 * Return: pointer to the PDU
932 *
933 * Driver command data is immediately after the request. So add request to get
934 * the PDU.
935 */
blk_mq_rq_to_pdu(struct request * rq)936 static inline void *blk_mq_rq_to_pdu(struct request *rq)
937 {
938 return rq + 1;
939 }
940
941 #define queue_for_each_hw_ctx(q, hctx, i) \
942 xa_for_each(&(q)->hctx_table, (i), (hctx))
943
944 #define hctx_for_each_ctx(hctx, ctx, i) \
945 for ((i) = 0; (i) < (hctx)->nr_ctx && \
946 ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
947
blk_mq_cleanup_rq(struct request * rq)948 static inline void blk_mq_cleanup_rq(struct request *rq)
949 {
950 if (rq->q->mq_ops->cleanup_rq)
951 rq->q->mq_ops->cleanup_rq(rq);
952 }
953
blk_rq_bio_prep(struct request * rq,struct bio * bio,unsigned int nr_segs)954 static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
955 unsigned int nr_segs)
956 {
957 rq->nr_phys_segments = nr_segs;
958 rq->__data_len = bio->bi_iter.bi_size;
959 rq->bio = rq->biotail = bio;
960 rq->ioprio = bio_prio(bio);
961 }
962
963 void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
964 struct lock_class_key *key);
965
rq_is_sync(struct request * rq)966 static inline bool rq_is_sync(struct request *rq)
967 {
968 return op_is_sync(rq->cmd_flags);
969 }
970
971 void blk_rq_init(struct request_queue *q, struct request *rq);
972 int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
973 struct bio_set *bs, gfp_t gfp_mask,
974 int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
975 void blk_rq_unprep_clone(struct request *rq);
976 blk_status_t blk_insert_cloned_request(struct request *rq);
977
978 struct rq_map_data {
979 struct page **pages;
980 unsigned long offset;
981 unsigned short page_order;
982 unsigned short nr_entries;
983 bool null_mapped;
984 bool from_user;
985 };
986
987 int blk_rq_map_user(struct request_queue *, struct request *,
988 struct rq_map_data *, void __user *, unsigned long, gfp_t);
989 int blk_rq_map_user_io(struct request *, struct rq_map_data *,
990 void __user *, unsigned long, gfp_t, bool, int, bool, int);
991 int blk_rq_map_user_iov(struct request_queue *, struct request *,
992 struct rq_map_data *, const struct iov_iter *, gfp_t);
993 int blk_rq_unmap_user(struct bio *);
994 int blk_rq_map_kern(struct request_queue *, struct request *, void *,
995 unsigned int, gfp_t);
996 int blk_rq_append_bio(struct request *rq, struct bio *bio);
997 void blk_execute_rq_nowait(struct request *rq, bool at_head);
998 blk_status_t blk_execute_rq(struct request *rq, bool at_head);
999 bool blk_rq_is_poll(struct request *rq);
1000
1001 struct req_iterator {
1002 struct bvec_iter iter;
1003 struct bio *bio;
1004 };
1005
1006 #define __rq_for_each_bio(_bio, rq) \
1007 if ((rq->bio)) \
1008 for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
1009
1010 #define rq_for_each_segment(bvl, _rq, _iter) \
1011 __rq_for_each_bio(_iter.bio, _rq) \
1012 bio_for_each_segment(bvl, _iter.bio, _iter.iter)
1013
1014 #define rq_for_each_bvec(bvl, _rq, _iter) \
1015 __rq_for_each_bio(_iter.bio, _rq) \
1016 bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
1017
1018 #define rq_iter_last(bvec, _iter) \
1019 (_iter.bio->bi_next == NULL && \
1020 bio_iter_last(bvec, _iter.iter))
1021
1022 /*
1023 * blk_rq_pos() : the current sector
1024 * blk_rq_bytes() : bytes left in the entire request
1025 * blk_rq_cur_bytes() : bytes left in the current segment
1026 * blk_rq_sectors() : sectors left in the entire request
1027 * blk_rq_cur_sectors() : sectors left in the current segment
1028 * blk_rq_stats_sectors() : sectors of the entire request used for stats
1029 */
blk_rq_pos(const struct request * rq)1030 static inline sector_t blk_rq_pos(const struct request *rq)
1031 {
1032 return rq->__sector;
1033 }
1034
blk_rq_bytes(const struct request * rq)1035 static inline unsigned int blk_rq_bytes(const struct request *rq)
1036 {
1037 return rq->__data_len;
1038 }
1039
blk_rq_cur_bytes(const struct request * rq)1040 static inline int blk_rq_cur_bytes(const struct request *rq)
1041 {
1042 if (!rq->bio)
1043 return 0;
1044 if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1045 return rq->bio->bi_iter.bi_size;
1046 return bio_iovec(rq->bio).bv_len;
1047 }
1048
blk_rq_sectors(const struct request * rq)1049 static inline unsigned int blk_rq_sectors(const struct request *rq)
1050 {
1051 return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1052 }
1053
blk_rq_cur_sectors(const struct request * rq)1054 static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1055 {
1056 return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1057 }
1058
blk_rq_stats_sectors(const struct request * rq)1059 static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1060 {
1061 return rq->stats_sectors;
1062 }
1063
1064 /*
1065 * Some commands like WRITE SAME have a payload or data transfer size which
1066 * is different from the size of the request. Any driver that supports such
1067 * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1068 * calculate the data transfer size.
1069 */
blk_rq_payload_bytes(struct request * rq)1070 static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1071 {
1072 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1073 return rq->special_vec.bv_len;
1074 return blk_rq_bytes(rq);
1075 }
1076
1077 /*
1078 * Return the first full biovec in the request. The caller needs to check that
1079 * there are any bvecs before calling this helper.
1080 */
req_bvec(struct request * rq)1081 static inline struct bio_vec req_bvec(struct request *rq)
1082 {
1083 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1084 return rq->special_vec;
1085 return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1086 }
1087
blk_rq_count_bios(struct request * rq)1088 static inline unsigned int blk_rq_count_bios(struct request *rq)
1089 {
1090 unsigned int nr_bios = 0;
1091 struct bio *bio;
1092
1093 __rq_for_each_bio(bio, rq)
1094 nr_bios++;
1095
1096 return nr_bios;
1097 }
1098
1099 void blk_steal_bios(struct bio_list *list, struct request *rq);
1100
1101 /*
1102 * Request completion related functions.
1103 *
1104 * blk_update_request() completes given number of bytes and updates
1105 * the request without completing it.
1106 */
1107 bool blk_update_request(struct request *rq, blk_status_t error,
1108 unsigned int nr_bytes);
1109 void blk_abort_request(struct request *);
1110
1111 /*
1112 * Number of physical segments as sent to the device.
1113 *
1114 * Normally this is the number of discontiguous data segments sent by the
1115 * submitter. But for data-less command like discard we might have no
1116 * actual data segments submitted, but the driver might have to add it's
1117 * own special payload. In that case we still return 1 here so that this
1118 * special payload will be mapped.
1119 */
blk_rq_nr_phys_segments(struct request * rq)1120 static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1121 {
1122 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1123 return 1;
1124 return rq->nr_phys_segments;
1125 }
1126
1127 /*
1128 * Number of discard segments (or ranges) the driver needs to fill in.
1129 * Each discard bio merged into a request is counted as one segment.
1130 */
blk_rq_nr_discard_segments(struct request * rq)1131 static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1132 {
1133 return max_t(unsigned short, rq->nr_phys_segments, 1);
1134 }
1135
1136 int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1137 struct scatterlist *sglist, struct scatterlist **last_sg);
blk_rq_map_sg(struct request_queue * q,struct request * rq,struct scatterlist * sglist)1138 static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1139 struct scatterlist *sglist)
1140 {
1141 struct scatterlist *last_sg = NULL;
1142
1143 return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1144 }
1145 void blk_dump_rq_flags(struct request *, char *);
1146
1147 #ifdef CONFIG_BLK_DEV_ZONED
blk_rq_zone_no(struct request * rq)1148 static inline unsigned int blk_rq_zone_no(struct request *rq)
1149 {
1150 return disk_zone_no(rq->q->disk, blk_rq_pos(rq));
1151 }
1152
blk_rq_zone_is_seq(struct request * rq)1153 static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1154 {
1155 return disk_zone_is_seq(rq->q->disk, blk_rq_pos(rq));
1156 }
1157
1158 bool blk_req_needs_zone_write_lock(struct request *rq);
1159 bool blk_req_zone_write_trylock(struct request *rq);
1160 void __blk_req_zone_write_lock(struct request *rq);
1161 void __blk_req_zone_write_unlock(struct request *rq);
1162
blk_req_zone_write_lock(struct request * rq)1163 static inline void blk_req_zone_write_lock(struct request *rq)
1164 {
1165 if (blk_req_needs_zone_write_lock(rq))
1166 __blk_req_zone_write_lock(rq);
1167 }
1168
blk_req_zone_write_unlock(struct request * rq)1169 static inline void blk_req_zone_write_unlock(struct request *rq)
1170 {
1171 if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1172 __blk_req_zone_write_unlock(rq);
1173 }
1174
blk_req_zone_is_write_locked(struct request * rq)1175 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1176 {
1177 return rq->q->disk->seq_zones_wlock &&
1178 test_bit(blk_rq_zone_no(rq), rq->q->disk->seq_zones_wlock);
1179 }
1180
blk_req_can_dispatch_to_zone(struct request * rq)1181 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1182 {
1183 if (!blk_req_needs_zone_write_lock(rq))
1184 return true;
1185 return !blk_req_zone_is_write_locked(rq);
1186 }
1187 #else /* CONFIG_BLK_DEV_ZONED */
blk_req_needs_zone_write_lock(struct request * rq)1188 static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1189 {
1190 return false;
1191 }
1192
blk_req_zone_write_lock(struct request * rq)1193 static inline void blk_req_zone_write_lock(struct request *rq)
1194 {
1195 }
1196
blk_req_zone_write_unlock(struct request * rq)1197 static inline void blk_req_zone_write_unlock(struct request *rq)
1198 {
1199 }
blk_req_zone_is_write_locked(struct request * rq)1200 static inline bool blk_req_zone_is_write_locked(struct request *rq)
1201 {
1202 return false;
1203 }
1204
blk_req_can_dispatch_to_zone(struct request * rq)1205 static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1206 {
1207 return true;
1208 }
1209 #endif /* CONFIG_BLK_DEV_ZONED */
1210
1211 #endif /* BLK_MQ_H */
1212