Line data Source code
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 : typedef void (rq_end_io_fn)(struct request *, blk_status_t);
18 :
19 : /*
20 : * request flags */
21 : typedef __u32 __bitwise req_flags_t;
22 :
23 : /* drive already may have started this one */
24 : #define RQF_STARTED ((__force req_flags_t)(1 << 1))
25 : /* may not be passed by ioscheduler */
26 : #define RQF_SOFTBARRIER ((__force req_flags_t)(1 << 3))
27 : /* request for flush sequence */
28 : #define RQF_FLUSH_SEQ ((__force req_flags_t)(1 << 4))
29 : /* merge of different types, fail separately */
30 : #define RQF_MIXED_MERGE ((__force req_flags_t)(1 << 5))
31 : /* track inflight for MQ */
32 : #define RQF_MQ_INFLIGHT ((__force req_flags_t)(1 << 6))
33 : /* don't call prep for this one */
34 : #define RQF_DONTPREP ((__force req_flags_t)(1 << 7))
35 : /* vaguely specified driver internal error. Ignored by the block layer */
36 : #define RQF_FAILED ((__force req_flags_t)(1 << 10))
37 : /* don't warn about errors */
38 : #define RQF_QUIET ((__force req_flags_t)(1 << 11))
39 : /* elevator private data attached */
40 : #define RQF_ELVPRIV ((__force req_flags_t)(1 << 12))
41 : /* account into disk and partition IO statistics */
42 : #define RQF_IO_STAT ((__force req_flags_t)(1 << 13))
43 : /* runtime pm request */
44 : #define RQF_PM ((__force req_flags_t)(1 << 15))
45 : /* on IO scheduler merge hash */
46 : #define RQF_HASHED ((__force req_flags_t)(1 << 16))
47 : /* track IO completion time */
48 : #define RQF_STATS ((__force req_flags_t)(1 << 17))
49 : /* Look at ->special_vec for the actual data payload instead of the
50 : bio chain. */
51 : #define RQF_SPECIAL_PAYLOAD ((__force req_flags_t)(1 << 18))
52 : /* The per-zone write lock is held for this request */
53 : #define RQF_ZONE_WRITE_LOCKED ((__force req_flags_t)(1 << 19))
54 : /* already slept for hybrid poll */
55 : #define RQF_MQ_POLL_SLEPT ((__force req_flags_t)(1 << 20))
56 : /* ->timeout has been called, don't expire again */
57 : #define RQF_TIMED_OUT ((__force req_flags_t)(1 << 21))
58 : /* queue has elevator attached */
59 : #define RQF_ELV ((__force req_flags_t)(1 << 22))
60 :
61 : /* flags that prevent us from merging requests: */
62 : #define RQF_NOMERGE_FLAGS \
63 : (RQF_STARTED | RQF_SOFTBARRIER | RQF_FLUSH_SEQ | RQF_SPECIAL_PAYLOAD)
64 :
65 : enum mq_rq_state {
66 : MQ_RQ_IDLE = 0,
67 : MQ_RQ_IN_FLIGHT = 1,
68 : MQ_RQ_COMPLETE = 2,
69 : };
70 :
71 : /*
72 : * Try to put the fields that are referenced together in the same cacheline.
73 : *
74 : * If you modify this structure, make sure to update blk_rq_init() and
75 : * especially blk_mq_rq_ctx_init() to take care of the added fields.
76 : */
77 : struct request {
78 : struct request_queue *q;
79 : struct blk_mq_ctx *mq_ctx;
80 : struct blk_mq_hw_ctx *mq_hctx;
81 :
82 : unsigned int cmd_flags; /* op and common flags */
83 : req_flags_t rq_flags;
84 :
85 : int tag;
86 : int internal_tag;
87 :
88 : unsigned int timeout;
89 :
90 : /* the following two fields are internal, NEVER access directly */
91 : unsigned int __data_len; /* total data len */
92 : sector_t __sector; /* sector cursor */
93 :
94 : struct bio *bio;
95 : struct bio *biotail;
96 :
97 : union {
98 : struct list_head queuelist;
99 : struct request *rq_next;
100 : };
101 :
102 : struct block_device *part;
103 : #ifdef CONFIG_BLK_RQ_ALLOC_TIME
104 : /* Time that the first bio started allocating this request. */
105 : u64 alloc_time_ns;
106 : #endif
107 : /* Time that this request was allocated for this IO. */
108 : u64 start_time_ns;
109 : /* Time that I/O was submitted to the device. */
110 : u64 io_start_time_ns;
111 :
112 : #ifdef CONFIG_BLK_WBT
113 : unsigned short wbt_flags;
114 : #endif
115 : /*
116 : * rq sectors used for blk stats. It has the same value
117 : * with blk_rq_sectors(rq), except that it never be zeroed
118 : * by completion.
119 : */
120 : unsigned short stats_sectors;
121 :
122 : /*
123 : * Number of scatter-gather DMA addr+len pairs after
124 : * physical address coalescing is performed.
125 : */
126 : unsigned short nr_phys_segments;
127 :
128 : #ifdef CONFIG_BLK_DEV_INTEGRITY
129 : unsigned short nr_integrity_segments;
130 : #endif
131 :
132 : #ifdef CONFIG_BLK_INLINE_ENCRYPTION
133 : struct bio_crypt_ctx *crypt_ctx;
134 : struct blk_crypto_keyslot *crypt_keyslot;
135 : #endif
136 :
137 : unsigned short write_hint;
138 : unsigned short ioprio;
139 :
140 : enum mq_rq_state state;
141 : atomic_t ref;
142 :
143 : unsigned long deadline;
144 :
145 : /*
146 : * The hash is used inside the scheduler, and killed once the
147 : * request reaches the dispatch list. The ipi_list is only used
148 : * to queue the request for softirq completion, which is long
149 : * after the request has been unhashed (and even removed from
150 : * the dispatch list).
151 : */
152 : union {
153 : struct hlist_node hash; /* merge hash */
154 : struct llist_node ipi_list;
155 : };
156 :
157 : /*
158 : * The rb_node is only used inside the io scheduler, requests
159 : * are pruned when moved to the dispatch queue. So let the
160 : * completion_data share space with the rb_node.
161 : */
162 : union {
163 : struct rb_node rb_node; /* sort/lookup */
164 : struct bio_vec special_vec;
165 : void *completion_data;
166 : };
167 :
168 :
169 : /*
170 : * Three pointers are available for the IO schedulers, if they need
171 : * more they have to dynamically allocate it. Flush requests are
172 : * never put on the IO scheduler. So let the flush fields share
173 : * space with the elevator data.
174 : */
175 : union {
176 : struct {
177 : struct io_cq *icq;
178 : void *priv[2];
179 : } elv;
180 :
181 : struct {
182 : unsigned int seq;
183 : struct list_head list;
184 : rq_end_io_fn *saved_end_io;
185 : } flush;
186 : };
187 :
188 : union {
189 : struct __call_single_data csd;
190 : u64 fifo_time;
191 : };
192 :
193 : /*
194 : * completion callback.
195 : */
196 : rq_end_io_fn *end_io;
197 : void *end_io_data;
198 : };
199 :
200 : #define req_op(req) \
201 : ((req)->cmd_flags & REQ_OP_MASK)
202 :
203 : static inline bool blk_rq_is_passthrough(struct request *rq)
204 : {
205 0 : return blk_op_is_passthrough(req_op(rq));
206 : }
207 :
208 : static inline unsigned short req_get_ioprio(struct request *req)
209 : {
210 : return req->ioprio;
211 : }
212 :
213 : #define rq_data_dir(rq) (op_is_write(req_op(rq)) ? WRITE : READ)
214 :
215 : #define rq_dma_dir(rq) \
216 : (op_is_write(req_op(rq)) ? DMA_TO_DEVICE : DMA_FROM_DEVICE)
217 :
218 : #define rq_list_add(listptr, rq) do { \
219 : (rq)->rq_next = *(listptr); \
220 : *(listptr) = rq; \
221 : } while (0)
222 :
223 : #define rq_list_pop(listptr) \
224 : ({ \
225 : struct request *__req = NULL; \
226 : if ((listptr) && *(listptr)) { \
227 : __req = *(listptr); \
228 : *(listptr) = __req->rq_next; \
229 : } \
230 : __req; \
231 : })
232 :
233 : #define rq_list_peek(listptr) \
234 : ({ \
235 : struct request *__req = NULL; \
236 : if ((listptr) && *(listptr)) \
237 : __req = *(listptr); \
238 : __req; \
239 : })
240 :
241 : #define rq_list_for_each(listptr, pos) \
242 : for (pos = rq_list_peek((listptr)); pos; pos = rq_list_next(pos))
243 :
244 : #define rq_list_for_each_safe(listptr, pos, nxt) \
245 : for (pos = rq_list_peek((listptr)), nxt = rq_list_next(pos); \
246 : pos; pos = nxt, nxt = pos ? rq_list_next(pos) : NULL)
247 :
248 : #define rq_list_next(rq) (rq)->rq_next
249 : #define rq_list_empty(list) ((list) == (struct request *) NULL)
250 :
251 : /**
252 : * rq_list_move() - move a struct request from one list to another
253 : * @src: The source list @rq is currently in
254 : * @dst: The destination list that @rq will be appended to
255 : * @rq: The request to move
256 : * @prev: The request preceding @rq in @src (NULL if @rq is the head)
257 : */
258 : static inline void rq_list_move(struct request **src, struct request **dst,
259 : struct request *rq, struct request *prev)
260 : {
261 : if (prev)
262 : prev->rq_next = rq->rq_next;
263 : else
264 : *src = rq->rq_next;
265 : rq_list_add(dst, rq);
266 : }
267 :
268 : enum blk_eh_timer_return {
269 : BLK_EH_DONE, /* drivers has completed the command */
270 : BLK_EH_RESET_TIMER, /* reset timer and try again */
271 : };
272 :
273 : #define BLK_TAG_ALLOC_FIFO 0 /* allocate starting from 0 */
274 : #define BLK_TAG_ALLOC_RR 1 /* allocate starting from last allocated tag */
275 :
276 : /**
277 : * struct blk_mq_hw_ctx - State for a hardware queue facing the hardware
278 : * block device
279 : */
280 : struct blk_mq_hw_ctx {
281 : struct {
282 : /** @lock: Protects the dispatch list. */
283 : spinlock_t lock;
284 : /**
285 : * @dispatch: Used for requests that are ready to be
286 : * dispatched to the hardware but for some reason (e.g. lack of
287 : * resources) could not be sent to the hardware. As soon as the
288 : * driver can send new requests, requests at this list will
289 : * be sent first for a fairer dispatch.
290 : */
291 : struct list_head dispatch;
292 : /**
293 : * @state: BLK_MQ_S_* flags. Defines the state of the hw
294 : * queue (active, scheduled to restart, stopped).
295 : */
296 : unsigned long state;
297 : } ____cacheline_aligned_in_smp;
298 :
299 : /**
300 : * @run_work: Used for scheduling a hardware queue run at a later time.
301 : */
302 : struct delayed_work run_work;
303 : /** @cpumask: Map of available CPUs where this hctx can run. */
304 : cpumask_var_t cpumask;
305 : /**
306 : * @next_cpu: Used by blk_mq_hctx_next_cpu() for round-robin CPU
307 : * selection from @cpumask.
308 : */
309 : int next_cpu;
310 : /**
311 : * @next_cpu_batch: Counter of how many works left in the batch before
312 : * changing to the next CPU.
313 : */
314 : int next_cpu_batch;
315 :
316 : /** @flags: BLK_MQ_F_* flags. Defines the behaviour of the queue. */
317 : unsigned long flags;
318 :
319 : /**
320 : * @sched_data: Pointer owned by the IO scheduler attached to a request
321 : * queue. It's up to the IO scheduler how to use this pointer.
322 : */
323 : void *sched_data;
324 : /**
325 : * @queue: Pointer to the request queue that owns this hardware context.
326 : */
327 : struct request_queue *queue;
328 : /** @fq: Queue of requests that need to perform a flush operation. */
329 : struct blk_flush_queue *fq;
330 :
331 : /**
332 : * @driver_data: Pointer to data owned by the block driver that created
333 : * this hctx
334 : */
335 : void *driver_data;
336 :
337 : /**
338 : * @ctx_map: Bitmap for each software queue. If bit is on, there is a
339 : * pending request in that software queue.
340 : */
341 : struct sbitmap ctx_map;
342 :
343 : /**
344 : * @dispatch_from: Software queue to be used when no scheduler was
345 : * selected.
346 : */
347 : struct blk_mq_ctx *dispatch_from;
348 : /**
349 : * @dispatch_busy: Number used by blk_mq_update_dispatch_busy() to
350 : * decide if the hw_queue is busy using Exponential Weighted Moving
351 : * Average algorithm.
352 : */
353 : unsigned int dispatch_busy;
354 :
355 : /** @type: HCTX_TYPE_* flags. Type of hardware queue. */
356 : unsigned short type;
357 : /** @nr_ctx: Number of software queues. */
358 : unsigned short nr_ctx;
359 : /** @ctxs: Array of software queues. */
360 : struct blk_mq_ctx **ctxs;
361 :
362 : /** @dispatch_wait_lock: Lock for dispatch_wait queue. */
363 : spinlock_t dispatch_wait_lock;
364 : /**
365 : * @dispatch_wait: Waitqueue to put requests when there is no tag
366 : * available at the moment, to wait for another try in the future.
367 : */
368 : wait_queue_entry_t dispatch_wait;
369 :
370 : /**
371 : * @wait_index: Index of next available dispatch_wait queue to insert
372 : * requests.
373 : */
374 : atomic_t wait_index;
375 :
376 : /**
377 : * @tags: Tags owned by the block driver. A tag at this set is only
378 : * assigned when a request is dispatched from a hardware queue.
379 : */
380 : struct blk_mq_tags *tags;
381 : /**
382 : * @sched_tags: Tags owned by I/O scheduler. If there is an I/O
383 : * scheduler associated with a request queue, a tag is assigned when
384 : * that request is allocated. Else, this member is not used.
385 : */
386 : struct blk_mq_tags *sched_tags;
387 :
388 : /** @queued: Number of queued requests. */
389 : unsigned long queued;
390 : /** @run: Number of dispatched requests. */
391 : unsigned long run;
392 :
393 : /** @numa_node: NUMA node the storage adapter has been connected to. */
394 : unsigned int numa_node;
395 : /** @queue_num: Index of this hardware queue. */
396 : unsigned int queue_num;
397 :
398 : /**
399 : * @nr_active: Number of active requests. Only used when a tag set is
400 : * shared across request queues.
401 : */
402 : atomic_t nr_active;
403 :
404 : /** @cpuhp_online: List to store request if CPU is going to die */
405 : struct hlist_node cpuhp_online;
406 : /** @cpuhp_dead: List to store request if some CPU die. */
407 : struct hlist_node cpuhp_dead;
408 : /** @kobj: Kernel object for sysfs. */
409 : struct kobject kobj;
410 :
411 : #ifdef CONFIG_BLK_DEBUG_FS
412 : /**
413 : * @debugfs_dir: debugfs directory for this hardware queue. Named
414 : * as cpu<cpu_number>.
415 : */
416 : struct dentry *debugfs_dir;
417 : /** @sched_debugfs_dir: debugfs directory for the scheduler. */
418 : struct dentry *sched_debugfs_dir;
419 : #endif
420 :
421 : /**
422 : * @hctx_list: if this hctx is not in use, this is an entry in
423 : * q->unused_hctx_list.
424 : */
425 : struct list_head hctx_list;
426 : };
427 :
428 : /**
429 : * struct blk_mq_queue_map - Map software queues to hardware queues
430 : * @mq_map: CPU ID to hardware queue index map. This is an array
431 : * with nr_cpu_ids elements. Each element has a value in the range
432 : * [@queue_offset, @queue_offset + @nr_queues).
433 : * @nr_queues: Number of hardware queues to map CPU IDs onto.
434 : * @queue_offset: First hardware queue to map onto. Used by the PCIe NVMe
435 : * driver to map each hardware queue type (enum hctx_type) onto a distinct
436 : * set of hardware queues.
437 : */
438 : struct blk_mq_queue_map {
439 : unsigned int *mq_map;
440 : unsigned int nr_queues;
441 : unsigned int queue_offset;
442 : };
443 :
444 : /**
445 : * enum hctx_type - Type of hardware queue
446 : * @HCTX_TYPE_DEFAULT: All I/O not otherwise accounted for.
447 : * @HCTX_TYPE_READ: Just for READ I/O.
448 : * @HCTX_TYPE_POLL: Polled I/O of any kind.
449 : * @HCTX_MAX_TYPES: Number of types of hctx.
450 : */
451 : enum hctx_type {
452 : HCTX_TYPE_DEFAULT,
453 : HCTX_TYPE_READ,
454 : HCTX_TYPE_POLL,
455 :
456 : HCTX_MAX_TYPES,
457 : };
458 :
459 : /**
460 : * struct blk_mq_tag_set - tag set that can be shared between request queues
461 : * @map: One or more ctx -> hctx mappings. One map exists for each
462 : * hardware queue type (enum hctx_type) that the driver wishes
463 : * to support. There are no restrictions on maps being of the
464 : * same size, and it's perfectly legal to share maps between
465 : * types.
466 : * @nr_maps: Number of elements in the @map array. A number in the range
467 : * [1, HCTX_MAX_TYPES].
468 : * @ops: Pointers to functions that implement block driver behavior.
469 : * @nr_hw_queues: Number of hardware queues supported by the block driver that
470 : * owns this data structure.
471 : * @queue_depth: Number of tags per hardware queue, reserved tags included.
472 : * @reserved_tags: Number of tags to set aside for BLK_MQ_REQ_RESERVED tag
473 : * allocations.
474 : * @cmd_size: Number of additional bytes to allocate per request. The block
475 : * driver owns these additional bytes.
476 : * @numa_node: NUMA node the storage adapter has been connected to.
477 : * @timeout: Request processing timeout in jiffies.
478 : * @flags: Zero or more BLK_MQ_F_* flags.
479 : * @driver_data: Pointer to data owned by the block driver that created this
480 : * tag set.
481 : * @tags: Tag sets. One tag set per hardware queue. Has @nr_hw_queues
482 : * elements.
483 : * @shared_tags:
484 : * Shared set of tags. Has @nr_hw_queues elements. If set,
485 : * shared by all @tags.
486 : * @tag_list_lock: Serializes tag_list accesses.
487 : * @tag_list: List of the request queues that use this tag set. See also
488 : * request_queue.tag_set_list.
489 : */
490 : struct blk_mq_tag_set {
491 : struct blk_mq_queue_map map[HCTX_MAX_TYPES];
492 : unsigned int nr_maps;
493 : const struct blk_mq_ops *ops;
494 : unsigned int nr_hw_queues;
495 : unsigned int queue_depth;
496 : unsigned int reserved_tags;
497 : unsigned int cmd_size;
498 : int numa_node;
499 : unsigned int timeout;
500 : unsigned int flags;
501 : void *driver_data;
502 :
503 : struct blk_mq_tags **tags;
504 :
505 : struct blk_mq_tags *shared_tags;
506 :
507 : struct mutex tag_list_lock;
508 : struct list_head tag_list;
509 : };
510 :
511 : /**
512 : * struct blk_mq_queue_data - Data about a request inserted in a queue
513 : *
514 : * @rq: Request pointer.
515 : * @last: If it is the last request in the queue.
516 : */
517 : struct blk_mq_queue_data {
518 : struct request *rq;
519 : bool last;
520 : };
521 :
522 : typedef bool (busy_tag_iter_fn)(struct request *, void *, bool);
523 :
524 : /**
525 : * struct blk_mq_ops - Callback functions that implements block driver
526 : * behaviour.
527 : */
528 : struct blk_mq_ops {
529 : /**
530 : * @queue_rq: Queue a new request from block IO.
531 : */
532 : blk_status_t (*queue_rq)(struct blk_mq_hw_ctx *,
533 : const struct blk_mq_queue_data *);
534 :
535 : /**
536 : * @commit_rqs: If a driver uses bd->last to judge when to submit
537 : * requests to hardware, it must define this function. In case of errors
538 : * that make us stop issuing further requests, this hook serves the
539 : * purpose of kicking the hardware (which the last request otherwise
540 : * would have done).
541 : */
542 : void (*commit_rqs)(struct blk_mq_hw_ctx *);
543 :
544 : /**
545 : * @queue_rqs: Queue a list of new requests. Driver is guaranteed
546 : * that each request belongs to the same queue. If the driver doesn't
547 : * empty the @rqlist completely, then the rest will be queued
548 : * individually by the block layer upon return.
549 : */
550 : void (*queue_rqs)(struct request **rqlist);
551 :
552 : /**
553 : * @get_budget: Reserve budget before queue request, once .queue_rq is
554 : * run, it is driver's responsibility to release the
555 : * reserved budget. Also we have to handle failure case
556 : * of .get_budget for avoiding I/O deadlock.
557 : */
558 : int (*get_budget)(struct request_queue *);
559 :
560 : /**
561 : * @put_budget: Release the reserved budget.
562 : */
563 : void (*put_budget)(struct request_queue *, int);
564 :
565 : /**
566 : * @set_rq_budget_token: store rq's budget token
567 : */
568 : void (*set_rq_budget_token)(struct request *, int);
569 : /**
570 : * @get_rq_budget_token: retrieve rq's budget token
571 : */
572 : int (*get_rq_budget_token)(struct request *);
573 :
574 : /**
575 : * @timeout: Called on request timeout.
576 : */
577 : enum blk_eh_timer_return (*timeout)(struct request *, bool);
578 :
579 : /**
580 : * @poll: Called to poll for completion of a specific tag.
581 : */
582 : int (*poll)(struct blk_mq_hw_ctx *, struct io_comp_batch *);
583 :
584 : /**
585 : * @complete: Mark the request as complete.
586 : */
587 : void (*complete)(struct request *);
588 :
589 : /**
590 : * @init_hctx: Called when the block layer side of a hardware queue has
591 : * been set up, allowing the driver to allocate/init matching
592 : * structures.
593 : */
594 : int (*init_hctx)(struct blk_mq_hw_ctx *, void *, unsigned int);
595 : /**
596 : * @exit_hctx: Ditto for exit/teardown.
597 : */
598 : void (*exit_hctx)(struct blk_mq_hw_ctx *, unsigned int);
599 :
600 : /**
601 : * @init_request: Called for every command allocated by the block layer
602 : * to allow the driver to set up driver specific data.
603 : *
604 : * Tag greater than or equal to queue_depth is for setting up
605 : * flush request.
606 : */
607 : int (*init_request)(struct blk_mq_tag_set *set, struct request *,
608 : unsigned int, unsigned int);
609 : /**
610 : * @exit_request: Ditto for exit/teardown.
611 : */
612 : void (*exit_request)(struct blk_mq_tag_set *set, struct request *,
613 : unsigned int);
614 :
615 : /**
616 : * @cleanup_rq: Called before freeing one request which isn't completed
617 : * yet, and usually for freeing the driver private data.
618 : */
619 : void (*cleanup_rq)(struct request *);
620 :
621 : /**
622 : * @busy: If set, returns whether or not this queue currently is busy.
623 : */
624 : bool (*busy)(struct request_queue *);
625 :
626 : /**
627 : * @map_queues: This allows drivers specify their own queue mapping by
628 : * overriding the setup-time function that builds the mq_map.
629 : */
630 : int (*map_queues)(struct blk_mq_tag_set *set);
631 :
632 : #ifdef CONFIG_BLK_DEBUG_FS
633 : /**
634 : * @show_rq: Used by the debugfs implementation to show driver-specific
635 : * information about a request.
636 : */
637 : void (*show_rq)(struct seq_file *m, struct request *rq);
638 : #endif
639 : };
640 :
641 : enum {
642 : BLK_MQ_F_SHOULD_MERGE = 1 << 0,
643 : BLK_MQ_F_TAG_QUEUE_SHARED = 1 << 1,
644 : /*
645 : * Set when this device requires underlying blk-mq device for
646 : * completing IO:
647 : */
648 : BLK_MQ_F_STACKING = 1 << 2,
649 : BLK_MQ_F_TAG_HCTX_SHARED = 1 << 3,
650 : BLK_MQ_F_BLOCKING = 1 << 5,
651 : /* Do not allow an I/O scheduler to be configured. */
652 : BLK_MQ_F_NO_SCHED = 1 << 6,
653 : /*
654 : * Select 'none' during queue registration in case of a single hwq
655 : * or shared hwqs instead of 'mq-deadline'.
656 : */
657 : BLK_MQ_F_NO_SCHED_BY_DEFAULT = 1 << 7,
658 : BLK_MQ_F_ALLOC_POLICY_START_BIT = 8,
659 : BLK_MQ_F_ALLOC_POLICY_BITS = 1,
660 :
661 : BLK_MQ_S_STOPPED = 0,
662 : BLK_MQ_S_TAG_ACTIVE = 1,
663 : BLK_MQ_S_SCHED_RESTART = 2,
664 :
665 : /* hw queue is inactive after all its CPUs become offline */
666 : BLK_MQ_S_INACTIVE = 3,
667 :
668 : BLK_MQ_MAX_DEPTH = 10240,
669 :
670 : BLK_MQ_CPU_WORK_BATCH = 8,
671 : };
672 : #define BLK_MQ_FLAG_TO_ALLOC_POLICY(flags) \
673 : ((flags >> BLK_MQ_F_ALLOC_POLICY_START_BIT) & \
674 : ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1))
675 : #define BLK_ALLOC_POLICY_TO_MQ_FLAG(policy) \
676 : ((policy & ((1 << BLK_MQ_F_ALLOC_POLICY_BITS) - 1)) \
677 : << BLK_MQ_F_ALLOC_POLICY_START_BIT)
678 :
679 : #define BLK_MQ_NO_HCTX_IDX (-1U)
680 :
681 : struct gendisk *__blk_mq_alloc_disk(struct blk_mq_tag_set *set, void *queuedata,
682 : struct lock_class_key *lkclass);
683 : #define blk_mq_alloc_disk(set, queuedata) \
684 : ({ \
685 : static struct lock_class_key __key; \
686 : \
687 : __blk_mq_alloc_disk(set, queuedata, &__key); \
688 : })
689 : struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
690 : int blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
691 : struct request_queue *q);
692 : void blk_mq_unregister_dev(struct device *, struct request_queue *);
693 :
694 : int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set);
695 : int blk_mq_alloc_sq_tag_set(struct blk_mq_tag_set *set,
696 : const struct blk_mq_ops *ops, unsigned int queue_depth,
697 : unsigned int set_flags);
698 : void blk_mq_free_tag_set(struct blk_mq_tag_set *set);
699 :
700 : void blk_mq_free_request(struct request *rq);
701 :
702 : bool blk_mq_queue_inflight(struct request_queue *q);
703 :
704 : enum {
705 : /* return when out of requests */
706 : BLK_MQ_REQ_NOWAIT = (__force blk_mq_req_flags_t)(1 << 0),
707 : /* allocate from reserved pool */
708 : BLK_MQ_REQ_RESERVED = (__force blk_mq_req_flags_t)(1 << 1),
709 : /* set RQF_PM */
710 : BLK_MQ_REQ_PM = (__force blk_mq_req_flags_t)(1 << 2),
711 : };
712 :
713 : struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
714 : blk_mq_req_flags_t flags);
715 : struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
716 : unsigned int op, blk_mq_req_flags_t flags,
717 : unsigned int hctx_idx);
718 :
719 : /*
720 : * Tag address space map.
721 : */
722 : struct blk_mq_tags {
723 : unsigned int nr_tags;
724 : unsigned int nr_reserved_tags;
725 :
726 : atomic_t active_queues;
727 :
728 : struct sbitmap_queue bitmap_tags;
729 : struct sbitmap_queue breserved_tags;
730 :
731 : struct request **rqs;
732 : struct request **static_rqs;
733 : struct list_head page_list;
734 :
735 : /*
736 : * used to clear request reference in rqs[] before freeing one
737 : * request pool
738 : */
739 : spinlock_t lock;
740 : };
741 :
742 : static inline struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags,
743 : unsigned int tag)
744 : {
745 0 : if (tag < tags->nr_tags) {
746 0 : prefetch(tags->rqs[tag]);
747 : return tags->rqs[tag];
748 : }
749 :
750 : return NULL;
751 : }
752 :
753 : enum {
754 : BLK_MQ_UNIQUE_TAG_BITS = 16,
755 : BLK_MQ_UNIQUE_TAG_MASK = (1 << BLK_MQ_UNIQUE_TAG_BITS) - 1,
756 : };
757 :
758 : u32 blk_mq_unique_tag(struct request *rq);
759 :
760 : static inline u16 blk_mq_unique_tag_to_hwq(u32 unique_tag)
761 : {
762 : return unique_tag >> BLK_MQ_UNIQUE_TAG_BITS;
763 : }
764 :
765 : static inline u16 blk_mq_unique_tag_to_tag(u32 unique_tag)
766 : {
767 : return unique_tag & BLK_MQ_UNIQUE_TAG_MASK;
768 : }
769 :
770 : /**
771 : * blk_mq_rq_state() - read the current MQ_RQ_* state of a request
772 : * @rq: target request.
773 : */
774 : static inline enum mq_rq_state blk_mq_rq_state(struct request *rq)
775 : {
776 0 : return READ_ONCE(rq->state);
777 : }
778 :
779 : static inline int blk_mq_request_started(struct request *rq)
780 : {
781 0 : return blk_mq_rq_state(rq) != MQ_RQ_IDLE;
782 : }
783 :
784 : static inline int blk_mq_request_completed(struct request *rq)
785 : {
786 0 : return blk_mq_rq_state(rq) == MQ_RQ_COMPLETE;
787 : }
788 :
789 : /*
790 : *
791 : * Set the state to complete when completing a request from inside ->queue_rq.
792 : * This is used by drivers that want to ensure special complete actions that
793 : * need access to the request are called on failure, e.g. by nvme for
794 : * multipathing.
795 : */
796 : static inline void blk_mq_set_request_complete(struct request *rq)
797 : {
798 : WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
799 : }
800 :
801 : /*
802 : * Complete the request directly instead of deferring it to softirq or
803 : * completing it another CPU. Useful in preemptible instead of an interrupt.
804 : */
805 : static inline void blk_mq_complete_request_direct(struct request *rq,
806 : void (*complete)(struct request *rq))
807 : {
808 : WRITE_ONCE(rq->state, MQ_RQ_COMPLETE);
809 : complete(rq);
810 : }
811 :
812 : void blk_mq_start_request(struct request *rq);
813 : void blk_mq_end_request(struct request *rq, blk_status_t error);
814 : void __blk_mq_end_request(struct request *rq, blk_status_t error);
815 : void blk_mq_end_request_batch(struct io_comp_batch *ib);
816 :
817 : /*
818 : * Only need start/end time stamping if we have iostat or
819 : * blk stats enabled, or using an IO scheduler.
820 : */
821 : static inline bool blk_mq_need_time_stamp(struct request *rq)
822 : {
823 0 : return (rq->rq_flags & (RQF_IO_STAT | RQF_STATS | RQF_ELV));
824 : }
825 :
826 : /*
827 : * Batched completions only work when there is no I/O error and no special
828 : * ->end_io handler.
829 : */
830 : static inline bool blk_mq_add_to_batch(struct request *req,
831 : struct io_comp_batch *iob, int ioerror,
832 : void (*complete)(struct io_comp_batch *))
833 : {
834 : if (!iob || (req->rq_flags & RQF_ELV) || req->end_io || ioerror)
835 : return false;
836 : if (!iob->complete)
837 : iob->complete = complete;
838 : else if (iob->complete != complete)
839 : return false;
840 : iob->need_ts |= blk_mq_need_time_stamp(req);
841 : rq_list_add(&iob->req_list, req);
842 : return true;
843 : }
844 :
845 : void blk_mq_requeue_request(struct request *rq, bool kick_requeue_list);
846 : void blk_mq_kick_requeue_list(struct request_queue *q);
847 : void blk_mq_delay_kick_requeue_list(struct request_queue *q, unsigned long msecs);
848 : void blk_mq_complete_request(struct request *rq);
849 : bool blk_mq_complete_request_remote(struct request *rq);
850 : bool blk_mq_queue_stopped(struct request_queue *q);
851 : void blk_mq_stop_hw_queue(struct blk_mq_hw_ctx *hctx);
852 : void blk_mq_start_hw_queue(struct blk_mq_hw_ctx *hctx);
853 : void blk_mq_stop_hw_queues(struct request_queue *q);
854 : void blk_mq_start_hw_queues(struct request_queue *q);
855 : void blk_mq_start_stopped_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
856 : void blk_mq_start_stopped_hw_queues(struct request_queue *q, bool async);
857 : void blk_mq_quiesce_queue(struct request_queue *q);
858 : void blk_mq_wait_quiesce_done(struct request_queue *q);
859 : void blk_mq_unquiesce_queue(struct request_queue *q);
860 : void blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs);
861 : void blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async);
862 : void blk_mq_run_hw_queues(struct request_queue *q, bool async);
863 : void blk_mq_delay_run_hw_queues(struct request_queue *q, unsigned long msecs);
864 : void blk_mq_tagset_busy_iter(struct blk_mq_tag_set *tagset,
865 : busy_tag_iter_fn *fn, void *priv);
866 : void blk_mq_tagset_wait_completed_request(struct blk_mq_tag_set *tagset);
867 : void blk_mq_freeze_queue(struct request_queue *q);
868 : void blk_mq_unfreeze_queue(struct request_queue *q);
869 : void blk_freeze_queue_start(struct request_queue *q);
870 : void blk_mq_freeze_queue_wait(struct request_queue *q);
871 : int blk_mq_freeze_queue_wait_timeout(struct request_queue *q,
872 : unsigned long timeout);
873 :
874 : int blk_mq_map_queues(struct blk_mq_queue_map *qmap);
875 : void blk_mq_update_nr_hw_queues(struct blk_mq_tag_set *set, int nr_hw_queues);
876 :
877 : void blk_mq_quiesce_queue_nowait(struct request_queue *q);
878 :
879 : unsigned int blk_mq_rq_cpu(struct request *rq);
880 :
881 : bool __blk_should_fake_timeout(struct request_queue *q);
882 : static inline bool blk_should_fake_timeout(struct request_queue *q)
883 : {
884 : if (IS_ENABLED(CONFIG_FAIL_IO_TIMEOUT) &&
885 : test_bit(QUEUE_FLAG_FAIL_IO, &q->queue_flags))
886 : return __blk_should_fake_timeout(q);
887 : return false;
888 : }
889 :
890 : /**
891 : * blk_mq_rq_from_pdu - cast a PDU to a request
892 : * @pdu: the PDU (Protocol Data Unit) to be casted
893 : *
894 : * Return: request
895 : *
896 : * Driver command data is immediately after the request. So subtract request
897 : * size to get back to the original request.
898 : */
899 : static inline struct request *blk_mq_rq_from_pdu(void *pdu)
900 : {
901 : return pdu - sizeof(struct request);
902 : }
903 :
904 : /**
905 : * blk_mq_rq_to_pdu - cast a request to a PDU
906 : * @rq: the request to be casted
907 : *
908 : * Return: pointer to the PDU
909 : *
910 : * Driver command data is immediately after the request. So add request to get
911 : * the PDU.
912 : */
913 : static inline void *blk_mq_rq_to_pdu(struct request *rq)
914 : {
915 : return rq + 1;
916 : }
917 :
918 : #define queue_for_each_hw_ctx(q, hctx, i) \
919 : xa_for_each(&(q)->hctx_table, (i), (hctx))
920 :
921 : #define hctx_for_each_ctx(hctx, ctx, i) \
922 : for ((i) = 0; (i) < (hctx)->nr_ctx && \
923 : ({ ctx = (hctx)->ctxs[(i)]; 1; }); (i)++)
924 :
925 : static inline void blk_mq_cleanup_rq(struct request *rq)
926 : {
927 : if (rq->q->mq_ops->cleanup_rq)
928 : rq->q->mq_ops->cleanup_rq(rq);
929 : }
930 :
931 : static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio,
932 : unsigned int nr_segs)
933 : {
934 0 : rq->nr_phys_segments = nr_segs;
935 0 : rq->__data_len = bio->bi_iter.bi_size;
936 0 : rq->bio = rq->biotail = bio;
937 0 : rq->ioprio = bio_prio(bio);
938 : }
939 :
940 : void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
941 : struct lock_class_key *key);
942 :
943 : static inline bool rq_is_sync(struct request *rq)
944 : {
945 : return op_is_sync(rq->cmd_flags);
946 : }
947 :
948 : void blk_rq_init(struct request_queue *q, struct request *rq);
949 : int blk_rq_prep_clone(struct request *rq, struct request *rq_src,
950 : struct bio_set *bs, gfp_t gfp_mask,
951 : int (*bio_ctr)(struct bio *, struct bio *, void *), void *data);
952 : void blk_rq_unprep_clone(struct request *rq);
953 : blk_status_t blk_insert_cloned_request(struct request *rq);
954 :
955 : struct rq_map_data {
956 : struct page **pages;
957 : int page_order;
958 : int nr_entries;
959 : unsigned long offset;
960 : int null_mapped;
961 : int from_user;
962 : };
963 :
964 : int blk_rq_map_user(struct request_queue *, struct request *,
965 : struct rq_map_data *, void __user *, unsigned long, gfp_t);
966 : int blk_rq_map_user_iov(struct request_queue *, struct request *,
967 : struct rq_map_data *, const struct iov_iter *, gfp_t);
968 : int blk_rq_unmap_user(struct bio *);
969 : int blk_rq_map_kern(struct request_queue *, struct request *, void *,
970 : unsigned int, gfp_t);
971 : int blk_rq_append_bio(struct request *rq, struct bio *bio);
972 : void blk_execute_rq_nowait(struct request *rq, bool at_head,
973 : rq_end_io_fn *end_io);
974 : blk_status_t blk_execute_rq(struct request *rq, bool at_head);
975 :
976 : struct req_iterator {
977 : struct bvec_iter iter;
978 : struct bio *bio;
979 : };
980 :
981 : #define __rq_for_each_bio(_bio, rq) \
982 : if ((rq->bio)) \
983 : for (_bio = (rq)->bio; _bio; _bio = _bio->bi_next)
984 :
985 : #define rq_for_each_segment(bvl, _rq, _iter) \
986 : __rq_for_each_bio(_iter.bio, _rq) \
987 : bio_for_each_segment(bvl, _iter.bio, _iter.iter)
988 :
989 : #define rq_for_each_bvec(bvl, _rq, _iter) \
990 : __rq_for_each_bio(_iter.bio, _rq) \
991 : bio_for_each_bvec(bvl, _iter.bio, _iter.iter)
992 :
993 : #define rq_iter_last(bvec, _iter) \
994 : (_iter.bio->bi_next == NULL && \
995 : bio_iter_last(bvec, _iter.iter))
996 :
997 : /*
998 : * blk_rq_pos() : the current sector
999 : * blk_rq_bytes() : bytes left in the entire request
1000 : * blk_rq_cur_bytes() : bytes left in the current segment
1001 : * blk_rq_sectors() : sectors left in the entire request
1002 : * blk_rq_cur_sectors() : sectors left in the current segment
1003 : * blk_rq_stats_sectors() : sectors of the entire request used for stats
1004 : */
1005 : static inline sector_t blk_rq_pos(const struct request *rq)
1006 : {
1007 : return rq->__sector;
1008 : }
1009 :
1010 : static inline unsigned int blk_rq_bytes(const struct request *rq)
1011 : {
1012 : return rq->__data_len;
1013 : }
1014 :
1015 0 : static inline int blk_rq_cur_bytes(const struct request *rq)
1016 : {
1017 0 : if (!rq->bio)
1018 : return 0;
1019 0 : if (!bio_has_data(rq->bio)) /* dataless requests such as discard */
1020 0 : return rq->bio->bi_iter.bi_size;
1021 0 : return bio_iovec(rq->bio).bv_len;
1022 : }
1023 :
1024 : static inline unsigned int blk_rq_sectors(const struct request *rq)
1025 : {
1026 0 : return blk_rq_bytes(rq) >> SECTOR_SHIFT;
1027 : }
1028 :
1029 : static inline unsigned int blk_rq_cur_sectors(const struct request *rq)
1030 : {
1031 0 : return blk_rq_cur_bytes(rq) >> SECTOR_SHIFT;
1032 : }
1033 :
1034 : static inline unsigned int blk_rq_stats_sectors(const struct request *rq)
1035 : {
1036 0 : return rq->stats_sectors;
1037 : }
1038 :
1039 : /*
1040 : * Some commands like WRITE SAME have a payload or data transfer size which
1041 : * is different from the size of the request. Any driver that supports such
1042 : * commands using the RQF_SPECIAL_PAYLOAD flag needs to use this helper to
1043 : * calculate the data transfer size.
1044 : */
1045 : static inline unsigned int blk_rq_payload_bytes(struct request *rq)
1046 : {
1047 : if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1048 : return rq->special_vec.bv_len;
1049 : return blk_rq_bytes(rq);
1050 : }
1051 :
1052 : /*
1053 : * Return the first full biovec in the request. The caller needs to check that
1054 : * there are any bvecs before calling this helper.
1055 : */
1056 : static inline struct bio_vec req_bvec(struct request *rq)
1057 : {
1058 : if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1059 : return rq->special_vec;
1060 : return mp_bvec_iter_bvec(rq->bio->bi_io_vec, rq->bio->bi_iter);
1061 : }
1062 :
1063 : static inline unsigned int blk_rq_count_bios(struct request *rq)
1064 : {
1065 : unsigned int nr_bios = 0;
1066 : struct bio *bio;
1067 :
1068 : __rq_for_each_bio(bio, rq)
1069 : nr_bios++;
1070 :
1071 : return nr_bios;
1072 : }
1073 :
1074 : void blk_steal_bios(struct bio_list *list, struct request *rq);
1075 :
1076 : /*
1077 : * Request completion related functions.
1078 : *
1079 : * blk_update_request() completes given number of bytes and updates
1080 : * the request without completing it.
1081 : */
1082 : bool blk_update_request(struct request *rq, blk_status_t error,
1083 : unsigned int nr_bytes);
1084 : void blk_abort_request(struct request *);
1085 :
1086 : /*
1087 : * Number of physical segments as sent to the device.
1088 : *
1089 : * Normally this is the number of discontiguous data segments sent by the
1090 : * submitter. But for data-less command like discard we might have no
1091 : * actual data segments submitted, but the driver might have to add it's
1092 : * own special payload. In that case we still return 1 here so that this
1093 : * special payload will be mapped.
1094 : */
1095 : static inline unsigned short blk_rq_nr_phys_segments(struct request *rq)
1096 : {
1097 0 : if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1098 : return 1;
1099 0 : return rq->nr_phys_segments;
1100 : }
1101 :
1102 : /*
1103 : * Number of discard segments (or ranges) the driver needs to fill in.
1104 : * Each discard bio merged into a request is counted as one segment.
1105 : */
1106 : static inline unsigned short blk_rq_nr_discard_segments(struct request *rq)
1107 : {
1108 0 : return max_t(unsigned short, rq->nr_phys_segments, 1);
1109 : }
1110 :
1111 : int __blk_rq_map_sg(struct request_queue *q, struct request *rq,
1112 : struct scatterlist *sglist, struct scatterlist **last_sg);
1113 : static inline int blk_rq_map_sg(struct request_queue *q, struct request *rq,
1114 : struct scatterlist *sglist)
1115 : {
1116 : struct scatterlist *last_sg = NULL;
1117 :
1118 : return __blk_rq_map_sg(q, rq, sglist, &last_sg);
1119 : }
1120 : void blk_dump_rq_flags(struct request *, char *);
1121 :
1122 : #ifdef CONFIG_BLK_DEV_ZONED
1123 : static inline unsigned int blk_rq_zone_no(struct request *rq)
1124 : {
1125 : return blk_queue_zone_no(rq->q, blk_rq_pos(rq));
1126 : }
1127 :
1128 : static inline unsigned int blk_rq_zone_is_seq(struct request *rq)
1129 : {
1130 : return blk_queue_zone_is_seq(rq->q, blk_rq_pos(rq));
1131 : }
1132 :
1133 : bool blk_req_needs_zone_write_lock(struct request *rq);
1134 : bool blk_req_zone_write_trylock(struct request *rq);
1135 : void __blk_req_zone_write_lock(struct request *rq);
1136 : void __blk_req_zone_write_unlock(struct request *rq);
1137 :
1138 : static inline void blk_req_zone_write_lock(struct request *rq)
1139 : {
1140 : if (blk_req_needs_zone_write_lock(rq))
1141 : __blk_req_zone_write_lock(rq);
1142 : }
1143 :
1144 : static inline void blk_req_zone_write_unlock(struct request *rq)
1145 : {
1146 : if (rq->rq_flags & RQF_ZONE_WRITE_LOCKED)
1147 : __blk_req_zone_write_unlock(rq);
1148 : }
1149 :
1150 : static inline bool blk_req_zone_is_write_locked(struct request *rq)
1151 : {
1152 : return rq->q->seq_zones_wlock &&
1153 : test_bit(blk_rq_zone_no(rq), rq->q->seq_zones_wlock);
1154 : }
1155 :
1156 : static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1157 : {
1158 : if (!blk_req_needs_zone_write_lock(rq))
1159 : return true;
1160 : return !blk_req_zone_is_write_locked(rq);
1161 : }
1162 : #else /* CONFIG_BLK_DEV_ZONED */
1163 : static inline bool blk_req_needs_zone_write_lock(struct request *rq)
1164 : {
1165 : return false;
1166 : }
1167 :
1168 : static inline void blk_req_zone_write_lock(struct request *rq)
1169 : {
1170 : }
1171 :
1172 : static inline void blk_req_zone_write_unlock(struct request *rq)
1173 : {
1174 : }
1175 : static inline bool blk_req_zone_is_write_locked(struct request *rq)
1176 : {
1177 : return false;
1178 : }
1179 :
1180 : static inline bool blk_req_can_dispatch_to_zone(struct request *rq)
1181 : {
1182 : return true;
1183 : }
1184 : #endif /* CONFIG_BLK_DEV_ZONED */
1185 :
1186 : #endif /* BLK_MQ_H */
|
