xref: /linux/block/blk-mq-debugfs.c (revision 460ea8980511c01c1551012b9a6ec6a06d02da59)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2017 Facebook
4  */
5 
6 #include <linux/kernel.h>
7 #include <linux/blkdev.h>
8 #include <linux/debugfs.h>
9 
10 #include "blk.h"
11 #include "blk-mq.h"
12 #include "blk-mq-debugfs.h"
13 #include "blk-mq-sched.h"
14 #include "blk-rq-qos.h"
15 
16 static int queue_poll_stat_show(void *data, struct seq_file *m)
17 {
18 	return 0;
19 }
20 
21 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
22 	__acquires(&q->requeue_lock)
23 {
24 	struct request_queue *q = m->private;
25 
26 	spin_lock_irq(&q->requeue_lock);
27 	return seq_list_start(&q->requeue_list, *pos);
28 }
29 
30 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
31 {
32 	struct request_queue *q = m->private;
33 
34 	return seq_list_next(v, &q->requeue_list, pos);
35 }
36 
37 static void queue_requeue_list_stop(struct seq_file *m, void *v)
38 	__releases(&q->requeue_lock)
39 {
40 	struct request_queue *q = m->private;
41 
42 	spin_unlock_irq(&q->requeue_lock);
43 }
44 
45 static const struct seq_operations queue_requeue_list_seq_ops = {
46 	.start	= queue_requeue_list_start,
47 	.next	= queue_requeue_list_next,
48 	.stop	= queue_requeue_list_stop,
49 	.show	= blk_mq_debugfs_rq_show,
50 };
51 
52 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
53 			  const char *const *flag_name, int flag_name_count)
54 {
55 	bool sep = false;
56 	int i;
57 
58 	for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
59 		if (!(flags & BIT(i)))
60 			continue;
61 		if (sep)
62 			seq_puts(m, "|");
63 		sep = true;
64 		if (i < flag_name_count && flag_name[i])
65 			seq_puts(m, flag_name[i]);
66 		else
67 			seq_printf(m, "%d", i);
68 	}
69 	return 0;
70 }
71 
72 static int queue_pm_only_show(void *data, struct seq_file *m)
73 {
74 	struct request_queue *q = data;
75 
76 	seq_printf(m, "%d\n", atomic_read(&q->pm_only));
77 	return 0;
78 }
79 
80 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
81 static const char *const blk_queue_flag_name[] = {
82 	QUEUE_FLAG_NAME(STOPPED),
83 	QUEUE_FLAG_NAME(DYING),
84 	QUEUE_FLAG_NAME(NOMERGES),
85 	QUEUE_FLAG_NAME(SAME_COMP),
86 	QUEUE_FLAG_NAME(FAIL_IO),
87 	QUEUE_FLAG_NAME(NONROT),
88 	QUEUE_FLAG_NAME(IO_STAT),
89 	QUEUE_FLAG_NAME(NOXMERGES),
90 	QUEUE_FLAG_NAME(ADD_RANDOM),
91 	QUEUE_FLAG_NAME(SYNCHRONOUS),
92 	QUEUE_FLAG_NAME(SAME_FORCE),
93 	QUEUE_FLAG_NAME(INIT_DONE),
94 	QUEUE_FLAG_NAME(STABLE_WRITES),
95 	QUEUE_FLAG_NAME(POLL),
96 	QUEUE_FLAG_NAME(WC),
97 	QUEUE_FLAG_NAME(FUA),
98 	QUEUE_FLAG_NAME(DAX),
99 	QUEUE_FLAG_NAME(STATS),
100 	QUEUE_FLAG_NAME(REGISTERED),
101 	QUEUE_FLAG_NAME(QUIESCED),
102 	QUEUE_FLAG_NAME(PCI_P2PDMA),
103 	QUEUE_FLAG_NAME(ZONE_RESETALL),
104 	QUEUE_FLAG_NAME(RQ_ALLOC_TIME),
105 	QUEUE_FLAG_NAME(HCTX_ACTIVE),
106 	QUEUE_FLAG_NAME(NOWAIT),
107 	QUEUE_FLAG_NAME(SQ_SCHED),
108 	QUEUE_FLAG_NAME(SKIP_TAGSET_QUIESCE),
109 };
110 #undef QUEUE_FLAG_NAME
111 
112 static int queue_state_show(void *data, struct seq_file *m)
113 {
114 	struct request_queue *q = data;
115 
116 	blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
117 		       ARRAY_SIZE(blk_queue_flag_name));
118 	seq_puts(m, "\n");
119 	return 0;
120 }
121 
122 static ssize_t queue_state_write(void *data, const char __user *buf,
123 				 size_t count, loff_t *ppos)
124 {
125 	struct request_queue *q = data;
126 	char opbuf[16] = { }, *op;
127 
128 	/*
129 	 * The "state" attribute is removed when the queue is removed.  Don't
130 	 * allow setting the state on a dying queue to avoid a use-after-free.
131 	 */
132 	if (blk_queue_dying(q))
133 		return -ENOENT;
134 
135 	if (count >= sizeof(opbuf)) {
136 		pr_err("%s: operation too long\n", __func__);
137 		goto inval;
138 	}
139 
140 	if (copy_from_user(opbuf, buf, count))
141 		return -EFAULT;
142 	op = strstrip(opbuf);
143 	if (strcmp(op, "run") == 0) {
144 		blk_mq_run_hw_queues(q, true);
145 	} else if (strcmp(op, "start") == 0) {
146 		blk_mq_start_stopped_hw_queues(q, true);
147 	} else if (strcmp(op, "kick") == 0) {
148 		blk_mq_kick_requeue_list(q);
149 	} else {
150 		pr_err("%s: unsupported operation '%s'\n", __func__, op);
151 inval:
152 		pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
153 		return -EINVAL;
154 	}
155 	return count;
156 }
157 
158 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
159 	{ "poll_stat", 0400, queue_poll_stat_show },
160 	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
161 	{ "pm_only", 0600, queue_pm_only_show, NULL },
162 	{ "state", 0600, queue_state_show, queue_state_write },
163 	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
164 	{ },
165 };
166 
167 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
168 static const char *const hctx_state_name[] = {
169 	HCTX_STATE_NAME(STOPPED),
170 	HCTX_STATE_NAME(TAG_ACTIVE),
171 	HCTX_STATE_NAME(SCHED_RESTART),
172 	HCTX_STATE_NAME(INACTIVE),
173 };
174 #undef HCTX_STATE_NAME
175 
176 static int hctx_state_show(void *data, struct seq_file *m)
177 {
178 	struct blk_mq_hw_ctx *hctx = data;
179 
180 	blk_flags_show(m, hctx->state, hctx_state_name,
181 		       ARRAY_SIZE(hctx_state_name));
182 	seq_puts(m, "\n");
183 	return 0;
184 }
185 
186 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
187 static const char *const alloc_policy_name[] = {
188 	BLK_TAG_ALLOC_NAME(FIFO),
189 	BLK_TAG_ALLOC_NAME(RR),
190 };
191 #undef BLK_TAG_ALLOC_NAME
192 
193 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
194 static const char *const hctx_flag_name[] = {
195 	HCTX_FLAG_NAME(SHOULD_MERGE),
196 	HCTX_FLAG_NAME(TAG_QUEUE_SHARED),
197 	HCTX_FLAG_NAME(BLOCKING),
198 	HCTX_FLAG_NAME(NO_SCHED),
199 	HCTX_FLAG_NAME(STACKING),
200 	HCTX_FLAG_NAME(TAG_HCTX_SHARED),
201 };
202 #undef HCTX_FLAG_NAME
203 
204 static int hctx_flags_show(void *data, struct seq_file *m)
205 {
206 	struct blk_mq_hw_ctx *hctx = data;
207 	const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
208 
209 	seq_puts(m, "alloc_policy=");
210 	if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
211 	    alloc_policy_name[alloc_policy])
212 		seq_puts(m, alloc_policy_name[alloc_policy]);
213 	else
214 		seq_printf(m, "%d", alloc_policy);
215 	seq_puts(m, " ");
216 	blk_flags_show(m,
217 		       hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
218 		       hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
219 	seq_puts(m, "\n");
220 	return 0;
221 }
222 
223 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
224 static const char *const cmd_flag_name[] = {
225 	CMD_FLAG_NAME(FAILFAST_DEV),
226 	CMD_FLAG_NAME(FAILFAST_TRANSPORT),
227 	CMD_FLAG_NAME(FAILFAST_DRIVER),
228 	CMD_FLAG_NAME(SYNC),
229 	CMD_FLAG_NAME(META),
230 	CMD_FLAG_NAME(PRIO),
231 	CMD_FLAG_NAME(NOMERGE),
232 	CMD_FLAG_NAME(IDLE),
233 	CMD_FLAG_NAME(INTEGRITY),
234 	CMD_FLAG_NAME(FUA),
235 	CMD_FLAG_NAME(PREFLUSH),
236 	CMD_FLAG_NAME(RAHEAD),
237 	CMD_FLAG_NAME(BACKGROUND),
238 	CMD_FLAG_NAME(NOWAIT),
239 	CMD_FLAG_NAME(NOUNMAP),
240 	CMD_FLAG_NAME(POLLED),
241 };
242 #undef CMD_FLAG_NAME
243 
244 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
245 static const char *const rqf_name[] = {
246 	RQF_NAME(STARTED),
247 	RQF_NAME(FLUSH_SEQ),
248 	RQF_NAME(MIXED_MERGE),
249 	RQF_NAME(MQ_INFLIGHT),
250 	RQF_NAME(DONTPREP),
251 	RQF_NAME(SCHED_TAGS),
252 	RQF_NAME(USE_SCHED),
253 	RQF_NAME(FAILED),
254 	RQF_NAME(QUIET),
255 	RQF_NAME(IO_STAT),
256 	RQF_NAME(PM),
257 	RQF_NAME(HASHED),
258 	RQF_NAME(STATS),
259 	RQF_NAME(SPECIAL_PAYLOAD),
260 	RQF_NAME(ZONE_WRITE_LOCKED),
261 	RQF_NAME(TIMED_OUT),
262 	RQF_NAME(RESV),
263 };
264 #undef RQF_NAME
265 
266 static const char *const blk_mq_rq_state_name_array[] = {
267 	[MQ_RQ_IDLE]		= "idle",
268 	[MQ_RQ_IN_FLIGHT]	= "in_flight",
269 	[MQ_RQ_COMPLETE]	= "complete",
270 };
271 
272 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
273 {
274 	if (WARN_ON_ONCE((unsigned int)rq_state >=
275 			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
276 		return "(?)";
277 	return blk_mq_rq_state_name_array[rq_state];
278 }
279 
280 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
281 {
282 	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
283 	const enum req_op op = req_op(rq);
284 	const char *op_str = blk_op_str(op);
285 
286 	seq_printf(m, "%p {.op=", rq);
287 	if (strcmp(op_str, "UNKNOWN") == 0)
288 		seq_printf(m, "%u", op);
289 	else
290 		seq_printf(m, "%s", op_str);
291 	seq_puts(m, ", .cmd_flags=");
292 	blk_flags_show(m, (__force unsigned int)(rq->cmd_flags & ~REQ_OP_MASK),
293 		       cmd_flag_name, ARRAY_SIZE(cmd_flag_name));
294 	seq_puts(m, ", .rq_flags=");
295 	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
296 		       ARRAY_SIZE(rqf_name));
297 	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
298 	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
299 		   rq->internal_tag);
300 	if (mq_ops->show_rq)
301 		mq_ops->show_rq(m, rq);
302 	seq_puts(m, "}\n");
303 	return 0;
304 }
305 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
306 
307 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
308 {
309 	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
310 }
311 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
312 
313 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
314 	__acquires(&hctx->lock)
315 {
316 	struct blk_mq_hw_ctx *hctx = m->private;
317 
318 	spin_lock(&hctx->lock);
319 	return seq_list_start(&hctx->dispatch, *pos);
320 }
321 
322 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
323 {
324 	struct blk_mq_hw_ctx *hctx = m->private;
325 
326 	return seq_list_next(v, &hctx->dispatch, pos);
327 }
328 
329 static void hctx_dispatch_stop(struct seq_file *m, void *v)
330 	__releases(&hctx->lock)
331 {
332 	struct blk_mq_hw_ctx *hctx = m->private;
333 
334 	spin_unlock(&hctx->lock);
335 }
336 
337 static const struct seq_operations hctx_dispatch_seq_ops = {
338 	.start	= hctx_dispatch_start,
339 	.next	= hctx_dispatch_next,
340 	.stop	= hctx_dispatch_stop,
341 	.show	= blk_mq_debugfs_rq_show,
342 };
343 
344 struct show_busy_params {
345 	struct seq_file		*m;
346 	struct blk_mq_hw_ctx	*hctx;
347 };
348 
349 /*
350  * Note: the state of a request may change while this function is in progress,
351  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
352  * keep iterating requests.
353  */
354 static bool hctx_show_busy_rq(struct request *rq, void *data)
355 {
356 	const struct show_busy_params *params = data;
357 
358 	if (rq->mq_hctx == params->hctx)
359 		__blk_mq_debugfs_rq_show(params->m, rq);
360 
361 	return true;
362 }
363 
364 static int hctx_busy_show(void *data, struct seq_file *m)
365 {
366 	struct blk_mq_hw_ctx *hctx = data;
367 	struct show_busy_params params = { .m = m, .hctx = hctx };
368 
369 	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
370 				&params);
371 
372 	return 0;
373 }
374 
375 static const char *const hctx_types[] = {
376 	[HCTX_TYPE_DEFAULT]	= "default",
377 	[HCTX_TYPE_READ]	= "read",
378 	[HCTX_TYPE_POLL]	= "poll",
379 };
380 
381 static int hctx_type_show(void *data, struct seq_file *m)
382 {
383 	struct blk_mq_hw_ctx *hctx = data;
384 
385 	BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
386 	seq_printf(m, "%s\n", hctx_types[hctx->type]);
387 	return 0;
388 }
389 
390 static int hctx_ctx_map_show(void *data, struct seq_file *m)
391 {
392 	struct blk_mq_hw_ctx *hctx = data;
393 
394 	sbitmap_bitmap_show(&hctx->ctx_map, m);
395 	return 0;
396 }
397 
398 static void blk_mq_debugfs_tags_show(struct seq_file *m,
399 				     struct blk_mq_tags *tags)
400 {
401 	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
402 	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
403 	seq_printf(m, "active_queues=%d\n",
404 		   READ_ONCE(tags->active_queues));
405 
406 	seq_puts(m, "\nbitmap_tags:\n");
407 	sbitmap_queue_show(&tags->bitmap_tags, m);
408 
409 	if (tags->nr_reserved_tags) {
410 		seq_puts(m, "\nbreserved_tags:\n");
411 		sbitmap_queue_show(&tags->breserved_tags, m);
412 	}
413 }
414 
415 static int hctx_tags_show(void *data, struct seq_file *m)
416 {
417 	struct blk_mq_hw_ctx *hctx = data;
418 	struct request_queue *q = hctx->queue;
419 	int res;
420 
421 	res = mutex_lock_interruptible(&q->sysfs_lock);
422 	if (res)
423 		goto out;
424 	if (hctx->tags)
425 		blk_mq_debugfs_tags_show(m, hctx->tags);
426 	mutex_unlock(&q->sysfs_lock);
427 
428 out:
429 	return res;
430 }
431 
432 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
433 {
434 	struct blk_mq_hw_ctx *hctx = data;
435 	struct request_queue *q = hctx->queue;
436 	int res;
437 
438 	res = mutex_lock_interruptible(&q->sysfs_lock);
439 	if (res)
440 		goto out;
441 	if (hctx->tags)
442 		sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
443 	mutex_unlock(&q->sysfs_lock);
444 
445 out:
446 	return res;
447 }
448 
449 static int hctx_sched_tags_show(void *data, struct seq_file *m)
450 {
451 	struct blk_mq_hw_ctx *hctx = data;
452 	struct request_queue *q = hctx->queue;
453 	int res;
454 
455 	res = mutex_lock_interruptible(&q->sysfs_lock);
456 	if (res)
457 		goto out;
458 	if (hctx->sched_tags)
459 		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
460 	mutex_unlock(&q->sysfs_lock);
461 
462 out:
463 	return res;
464 }
465 
466 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
467 {
468 	struct blk_mq_hw_ctx *hctx = data;
469 	struct request_queue *q = hctx->queue;
470 	int res;
471 
472 	res = mutex_lock_interruptible(&q->sysfs_lock);
473 	if (res)
474 		goto out;
475 	if (hctx->sched_tags)
476 		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
477 	mutex_unlock(&q->sysfs_lock);
478 
479 out:
480 	return res;
481 }
482 
483 static int hctx_run_show(void *data, struct seq_file *m)
484 {
485 	struct blk_mq_hw_ctx *hctx = data;
486 
487 	seq_printf(m, "%lu\n", hctx->run);
488 	return 0;
489 }
490 
491 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
492 			      loff_t *ppos)
493 {
494 	struct blk_mq_hw_ctx *hctx = data;
495 
496 	hctx->run = 0;
497 	return count;
498 }
499 
500 static int hctx_active_show(void *data, struct seq_file *m)
501 {
502 	struct blk_mq_hw_ctx *hctx = data;
503 
504 	seq_printf(m, "%d\n", __blk_mq_active_requests(hctx));
505 	return 0;
506 }
507 
508 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
509 {
510 	struct blk_mq_hw_ctx *hctx = data;
511 
512 	seq_printf(m, "%u\n", hctx->dispatch_busy);
513 	return 0;
514 }
515 
516 #define CTX_RQ_SEQ_OPS(name, type)					\
517 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
518 	__acquires(&ctx->lock)						\
519 {									\
520 	struct blk_mq_ctx *ctx = m->private;				\
521 									\
522 	spin_lock(&ctx->lock);						\
523 	return seq_list_start(&ctx->rq_lists[type], *pos);		\
524 }									\
525 									\
526 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,	\
527 				     loff_t *pos)			\
528 {									\
529 	struct blk_mq_ctx *ctx = m->private;				\
530 									\
531 	return seq_list_next(v, &ctx->rq_lists[type], pos);		\
532 }									\
533 									\
534 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)	\
535 	__releases(&ctx->lock)						\
536 {									\
537 	struct blk_mq_ctx *ctx = m->private;				\
538 									\
539 	spin_unlock(&ctx->lock);					\
540 }									\
541 									\
542 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {	\
543 	.start	= ctx_##name##_rq_list_start,				\
544 	.next	= ctx_##name##_rq_list_next,				\
545 	.stop	= ctx_##name##_rq_list_stop,				\
546 	.show	= blk_mq_debugfs_rq_show,				\
547 }
548 
549 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
550 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
551 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
552 
553 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
554 {
555 	const struct blk_mq_debugfs_attr *attr = m->private;
556 	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
557 
558 	return attr->show(data, m);
559 }
560 
561 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
562 				    size_t count, loff_t *ppos)
563 {
564 	struct seq_file *m = file->private_data;
565 	const struct blk_mq_debugfs_attr *attr = m->private;
566 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
567 
568 	/*
569 	 * Attributes that only implement .seq_ops are read-only and 'attr' is
570 	 * the same with 'data' in this case.
571 	 */
572 	if (attr == data || !attr->write)
573 		return -EPERM;
574 
575 	return attr->write(data, buf, count, ppos);
576 }
577 
578 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
579 {
580 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
581 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
582 	struct seq_file *m;
583 	int ret;
584 
585 	if (attr->seq_ops) {
586 		ret = seq_open(file, attr->seq_ops);
587 		if (!ret) {
588 			m = file->private_data;
589 			m->private = data;
590 		}
591 		return ret;
592 	}
593 
594 	if (WARN_ON_ONCE(!attr->show))
595 		return -EPERM;
596 
597 	return single_open(file, blk_mq_debugfs_show, inode->i_private);
598 }
599 
600 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
601 {
602 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
603 
604 	if (attr->show)
605 		return single_release(inode, file);
606 
607 	return seq_release(inode, file);
608 }
609 
610 static const struct file_operations blk_mq_debugfs_fops = {
611 	.open		= blk_mq_debugfs_open,
612 	.read		= seq_read,
613 	.write		= blk_mq_debugfs_write,
614 	.llseek		= seq_lseek,
615 	.release	= blk_mq_debugfs_release,
616 };
617 
618 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
619 	{"state", 0400, hctx_state_show},
620 	{"flags", 0400, hctx_flags_show},
621 	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
622 	{"busy", 0400, hctx_busy_show},
623 	{"ctx_map", 0400, hctx_ctx_map_show},
624 	{"tags", 0400, hctx_tags_show},
625 	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
626 	{"sched_tags", 0400, hctx_sched_tags_show},
627 	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
628 	{"run", 0600, hctx_run_show, hctx_run_write},
629 	{"active", 0400, hctx_active_show},
630 	{"dispatch_busy", 0400, hctx_dispatch_busy_show},
631 	{"type", 0400, hctx_type_show},
632 	{},
633 };
634 
635 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
636 	{"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
637 	{"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
638 	{"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
639 	{},
640 };
641 
642 static void debugfs_create_files(struct dentry *parent, void *data,
643 				 const struct blk_mq_debugfs_attr *attr)
644 {
645 	if (IS_ERR_OR_NULL(parent))
646 		return;
647 
648 	d_inode(parent)->i_private = data;
649 
650 	for (; attr->name; attr++)
651 		debugfs_create_file(attr->name, attr->mode, parent,
652 				    (void *)attr, &blk_mq_debugfs_fops);
653 }
654 
655 void blk_mq_debugfs_register(struct request_queue *q)
656 {
657 	struct blk_mq_hw_ctx *hctx;
658 	unsigned long i;
659 
660 	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
661 
662 	/*
663 	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
664 	 * didn't exist yet (because we don't know what to name the directory
665 	 * until the queue is registered to a gendisk).
666 	 */
667 	if (q->elevator && !q->sched_debugfs_dir)
668 		blk_mq_debugfs_register_sched(q);
669 
670 	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
671 	queue_for_each_hw_ctx(q, hctx, i) {
672 		if (!hctx->debugfs_dir)
673 			blk_mq_debugfs_register_hctx(q, hctx);
674 		if (q->elevator && !hctx->sched_debugfs_dir)
675 			blk_mq_debugfs_register_sched_hctx(q, hctx);
676 	}
677 
678 	if (q->rq_qos) {
679 		struct rq_qos *rqos = q->rq_qos;
680 
681 		while (rqos) {
682 			blk_mq_debugfs_register_rqos(rqos);
683 			rqos = rqos->next;
684 		}
685 	}
686 }
687 
688 static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
689 					struct blk_mq_ctx *ctx)
690 {
691 	struct dentry *ctx_dir;
692 	char name[20];
693 
694 	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
695 	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
696 
697 	debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs);
698 }
699 
700 void blk_mq_debugfs_register_hctx(struct request_queue *q,
701 				  struct blk_mq_hw_ctx *hctx)
702 {
703 	struct blk_mq_ctx *ctx;
704 	char name[20];
705 	int i;
706 
707 	if (!q->debugfs_dir)
708 		return;
709 
710 	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
711 	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
712 
713 	debugfs_create_files(hctx->debugfs_dir, hctx, blk_mq_debugfs_hctx_attrs);
714 
715 	hctx_for_each_ctx(hctx, ctx, i)
716 		blk_mq_debugfs_register_ctx(hctx, ctx);
717 }
718 
719 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
720 {
721 	if (!hctx->queue->debugfs_dir)
722 		return;
723 	debugfs_remove_recursive(hctx->debugfs_dir);
724 	hctx->sched_debugfs_dir = NULL;
725 	hctx->debugfs_dir = NULL;
726 }
727 
728 void blk_mq_debugfs_register_hctxs(struct request_queue *q)
729 {
730 	struct blk_mq_hw_ctx *hctx;
731 	unsigned long i;
732 
733 	queue_for_each_hw_ctx(q, hctx, i)
734 		blk_mq_debugfs_register_hctx(q, hctx);
735 }
736 
737 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
738 {
739 	struct blk_mq_hw_ctx *hctx;
740 	unsigned long i;
741 
742 	queue_for_each_hw_ctx(q, hctx, i)
743 		blk_mq_debugfs_unregister_hctx(hctx);
744 }
745 
746 void blk_mq_debugfs_register_sched(struct request_queue *q)
747 {
748 	struct elevator_type *e = q->elevator->type;
749 
750 	lockdep_assert_held(&q->debugfs_mutex);
751 
752 	/*
753 	 * If the parent directory has not been created yet, return, we will be
754 	 * called again later on and the directory/files will be created then.
755 	 */
756 	if (!q->debugfs_dir)
757 		return;
758 
759 	if (!e->queue_debugfs_attrs)
760 		return;
761 
762 	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
763 
764 	debugfs_create_files(q->sched_debugfs_dir, q, e->queue_debugfs_attrs);
765 }
766 
767 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
768 {
769 	lockdep_assert_held(&q->debugfs_mutex);
770 
771 	debugfs_remove_recursive(q->sched_debugfs_dir);
772 	q->sched_debugfs_dir = NULL;
773 }
774 
775 static const char *rq_qos_id_to_name(enum rq_qos_id id)
776 {
777 	switch (id) {
778 	case RQ_QOS_WBT:
779 		return "wbt";
780 	case RQ_QOS_LATENCY:
781 		return "latency";
782 	case RQ_QOS_COST:
783 		return "cost";
784 	}
785 	return "unknown";
786 }
787 
788 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
789 {
790 	lockdep_assert_held(&rqos->disk->queue->debugfs_mutex);
791 
792 	if (!rqos->disk->queue->debugfs_dir)
793 		return;
794 	debugfs_remove_recursive(rqos->debugfs_dir);
795 	rqos->debugfs_dir = NULL;
796 }
797 
798 void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
799 {
800 	struct request_queue *q = rqos->disk->queue;
801 	const char *dir_name = rq_qos_id_to_name(rqos->id);
802 
803 	lockdep_assert_held(&q->debugfs_mutex);
804 
805 	if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
806 		return;
807 
808 	if (!q->rqos_debugfs_dir)
809 		q->rqos_debugfs_dir = debugfs_create_dir("rqos",
810 							 q->debugfs_dir);
811 
812 	rqos->debugfs_dir = debugfs_create_dir(dir_name, q->rqos_debugfs_dir);
813 	debugfs_create_files(rqos->debugfs_dir, rqos, rqos->ops->debugfs_attrs);
814 }
815 
816 void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
817 					struct blk_mq_hw_ctx *hctx)
818 {
819 	struct elevator_type *e = q->elevator->type;
820 
821 	lockdep_assert_held(&q->debugfs_mutex);
822 
823 	/*
824 	 * If the parent debugfs directory has not been created yet, return;
825 	 * We will be called again later on with appropriate parent debugfs
826 	 * directory from blk_register_queue()
827 	 */
828 	if (!hctx->debugfs_dir)
829 		return;
830 
831 	if (!e->hctx_debugfs_attrs)
832 		return;
833 
834 	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
835 						     hctx->debugfs_dir);
836 	debugfs_create_files(hctx->sched_debugfs_dir, hctx,
837 			     e->hctx_debugfs_attrs);
838 }
839 
840 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
841 {
842 	lockdep_assert_held(&hctx->queue->debugfs_mutex);
843 
844 	if (!hctx->queue->debugfs_dir)
845 		return;
846 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
847 	hctx->sched_debugfs_dir = NULL;
848 }
849