xref: /openbmc/linux/block/blk-mq-debugfs.c (revision 0d4bb5e4)
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 <linux/blk-mq.h>
11 #include "blk.h"
12 #include "blk-mq.h"
13 #include "blk-mq-debugfs.h"
14 #include "blk-mq-sched.h"
15 #include "blk-mq-tag.h"
16 #include "blk-rq-qos.h"
17 
18 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
19 {
20 	if (stat->nr_samples) {
21 		seq_printf(m, "samples=%d, mean=%llu, min=%llu, max=%llu",
22 			   stat->nr_samples, stat->mean, stat->min, stat->max);
23 	} else {
24 		seq_puts(m, "samples=0");
25 	}
26 }
27 
28 static int queue_poll_stat_show(void *data, struct seq_file *m)
29 {
30 	struct request_queue *q = data;
31 	int bucket;
32 
33 	if (!q->poll_stat)
34 		return 0;
35 
36 	for (bucket = 0; bucket < (BLK_MQ_POLL_STATS_BKTS / 2); bucket++) {
37 		seq_printf(m, "read  (%d Bytes): ", 1 << (9 + bucket));
38 		print_stat(m, &q->poll_stat[2 * bucket]);
39 		seq_puts(m, "\n");
40 
41 		seq_printf(m, "write (%d Bytes): ",  1 << (9 + bucket));
42 		print_stat(m, &q->poll_stat[2 * bucket + 1]);
43 		seq_puts(m, "\n");
44 	}
45 	return 0;
46 }
47 
48 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
49 	__acquires(&q->requeue_lock)
50 {
51 	struct request_queue *q = m->private;
52 
53 	spin_lock_irq(&q->requeue_lock);
54 	return seq_list_start(&q->requeue_list, *pos);
55 }
56 
57 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
58 {
59 	struct request_queue *q = m->private;
60 
61 	return seq_list_next(v, &q->requeue_list, pos);
62 }
63 
64 static void queue_requeue_list_stop(struct seq_file *m, void *v)
65 	__releases(&q->requeue_lock)
66 {
67 	struct request_queue *q = m->private;
68 
69 	spin_unlock_irq(&q->requeue_lock);
70 }
71 
72 static const struct seq_operations queue_requeue_list_seq_ops = {
73 	.start	= queue_requeue_list_start,
74 	.next	= queue_requeue_list_next,
75 	.stop	= queue_requeue_list_stop,
76 	.show	= blk_mq_debugfs_rq_show,
77 };
78 
79 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
80 			  const char *const *flag_name, int flag_name_count)
81 {
82 	bool sep = false;
83 	int i;
84 
85 	for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
86 		if (!(flags & BIT(i)))
87 			continue;
88 		if (sep)
89 			seq_puts(m, "|");
90 		sep = true;
91 		if (i < flag_name_count && flag_name[i])
92 			seq_puts(m, flag_name[i]);
93 		else
94 			seq_printf(m, "%d", i);
95 	}
96 	return 0;
97 }
98 
99 static int queue_pm_only_show(void *data, struct seq_file *m)
100 {
101 	struct request_queue *q = data;
102 
103 	seq_printf(m, "%d\n", atomic_read(&q->pm_only));
104 	return 0;
105 }
106 
107 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
108 static const char *const blk_queue_flag_name[] = {
109 	QUEUE_FLAG_NAME(STOPPED),
110 	QUEUE_FLAG_NAME(DYING),
111 	QUEUE_FLAG_NAME(NOMERGES),
112 	QUEUE_FLAG_NAME(SAME_COMP),
113 	QUEUE_FLAG_NAME(FAIL_IO),
114 	QUEUE_FLAG_NAME(NONROT),
115 	QUEUE_FLAG_NAME(IO_STAT),
116 	QUEUE_FLAG_NAME(NOXMERGES),
117 	QUEUE_FLAG_NAME(ADD_RANDOM),
118 	QUEUE_FLAG_NAME(SAME_FORCE),
119 	QUEUE_FLAG_NAME(INIT_DONE),
120 	QUEUE_FLAG_NAME(STABLE_WRITES),
121 	QUEUE_FLAG_NAME(POLL),
122 	QUEUE_FLAG_NAME(WC),
123 	QUEUE_FLAG_NAME(FUA),
124 	QUEUE_FLAG_NAME(DAX),
125 	QUEUE_FLAG_NAME(STATS),
126 	QUEUE_FLAG_NAME(REGISTERED),
127 	QUEUE_FLAG_NAME(QUIESCED),
128 	QUEUE_FLAG_NAME(PCI_P2PDMA),
129 	QUEUE_FLAG_NAME(ZONE_RESETALL),
130 	QUEUE_FLAG_NAME(RQ_ALLOC_TIME),
131 	QUEUE_FLAG_NAME(HCTX_ACTIVE),
132 	QUEUE_FLAG_NAME(NOWAIT),
133 };
134 #undef QUEUE_FLAG_NAME
135 
136 static int queue_state_show(void *data, struct seq_file *m)
137 {
138 	struct request_queue *q = data;
139 
140 	blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
141 		       ARRAY_SIZE(blk_queue_flag_name));
142 	seq_puts(m, "\n");
143 	return 0;
144 }
145 
146 static ssize_t queue_state_write(void *data, const char __user *buf,
147 				 size_t count, loff_t *ppos)
148 {
149 	struct request_queue *q = data;
150 	char opbuf[16] = { }, *op;
151 
152 	/*
153 	 * The "state" attribute is removed when the queue is removed.  Don't
154 	 * allow setting the state on a dying queue to avoid a use-after-free.
155 	 */
156 	if (blk_queue_dying(q))
157 		return -ENOENT;
158 
159 	if (count >= sizeof(opbuf)) {
160 		pr_err("%s: operation too long\n", __func__);
161 		goto inval;
162 	}
163 
164 	if (copy_from_user(opbuf, buf, count))
165 		return -EFAULT;
166 	op = strstrip(opbuf);
167 	if (strcmp(op, "run") == 0) {
168 		blk_mq_run_hw_queues(q, true);
169 	} else if (strcmp(op, "start") == 0) {
170 		blk_mq_start_stopped_hw_queues(q, true);
171 	} else if (strcmp(op, "kick") == 0) {
172 		blk_mq_kick_requeue_list(q);
173 	} else {
174 		pr_err("%s: unsupported operation '%s'\n", __func__, op);
175 inval:
176 		pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
177 		return -EINVAL;
178 	}
179 	return count;
180 }
181 
182 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
183 	{ "poll_stat", 0400, queue_poll_stat_show },
184 	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
185 	{ "pm_only", 0600, queue_pm_only_show, NULL },
186 	{ "state", 0600, queue_state_show, queue_state_write },
187 	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
188 	{ },
189 };
190 
191 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
192 static const char *const hctx_state_name[] = {
193 	HCTX_STATE_NAME(STOPPED),
194 	HCTX_STATE_NAME(TAG_ACTIVE),
195 	HCTX_STATE_NAME(SCHED_RESTART),
196 	HCTX_STATE_NAME(INACTIVE),
197 };
198 #undef HCTX_STATE_NAME
199 
200 static int hctx_state_show(void *data, struct seq_file *m)
201 {
202 	struct blk_mq_hw_ctx *hctx = data;
203 
204 	blk_flags_show(m, hctx->state, hctx_state_name,
205 		       ARRAY_SIZE(hctx_state_name));
206 	seq_puts(m, "\n");
207 	return 0;
208 }
209 
210 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
211 static const char *const alloc_policy_name[] = {
212 	BLK_TAG_ALLOC_NAME(FIFO),
213 	BLK_TAG_ALLOC_NAME(RR),
214 };
215 #undef BLK_TAG_ALLOC_NAME
216 
217 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
218 static const char *const hctx_flag_name[] = {
219 	HCTX_FLAG_NAME(SHOULD_MERGE),
220 	HCTX_FLAG_NAME(TAG_QUEUE_SHARED),
221 	HCTX_FLAG_NAME(BLOCKING),
222 	HCTX_FLAG_NAME(NO_SCHED),
223 	HCTX_FLAG_NAME(STACKING),
224 	HCTX_FLAG_NAME(TAG_HCTX_SHARED),
225 };
226 #undef HCTX_FLAG_NAME
227 
228 static int hctx_flags_show(void *data, struct seq_file *m)
229 {
230 	struct blk_mq_hw_ctx *hctx = data;
231 	const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
232 
233 	seq_puts(m, "alloc_policy=");
234 	if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
235 	    alloc_policy_name[alloc_policy])
236 		seq_puts(m, alloc_policy_name[alloc_policy]);
237 	else
238 		seq_printf(m, "%d", alloc_policy);
239 	seq_puts(m, " ");
240 	blk_flags_show(m,
241 		       hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
242 		       hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
243 	seq_puts(m, "\n");
244 	return 0;
245 }
246 
247 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
248 static const char *const cmd_flag_name[] = {
249 	CMD_FLAG_NAME(FAILFAST_DEV),
250 	CMD_FLAG_NAME(FAILFAST_TRANSPORT),
251 	CMD_FLAG_NAME(FAILFAST_DRIVER),
252 	CMD_FLAG_NAME(SYNC),
253 	CMD_FLAG_NAME(META),
254 	CMD_FLAG_NAME(PRIO),
255 	CMD_FLAG_NAME(NOMERGE),
256 	CMD_FLAG_NAME(IDLE),
257 	CMD_FLAG_NAME(INTEGRITY),
258 	CMD_FLAG_NAME(FUA),
259 	CMD_FLAG_NAME(PREFLUSH),
260 	CMD_FLAG_NAME(RAHEAD),
261 	CMD_FLAG_NAME(BACKGROUND),
262 	CMD_FLAG_NAME(NOWAIT),
263 	CMD_FLAG_NAME(NOUNMAP),
264 	CMD_FLAG_NAME(POLLED),
265 };
266 #undef CMD_FLAG_NAME
267 
268 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
269 static const char *const rqf_name[] = {
270 	RQF_NAME(STARTED),
271 	RQF_NAME(SOFTBARRIER),
272 	RQF_NAME(FLUSH_SEQ),
273 	RQF_NAME(MIXED_MERGE),
274 	RQF_NAME(MQ_INFLIGHT),
275 	RQF_NAME(DONTPREP),
276 	RQF_NAME(FAILED),
277 	RQF_NAME(QUIET),
278 	RQF_NAME(ELVPRIV),
279 	RQF_NAME(IO_STAT),
280 	RQF_NAME(PM),
281 	RQF_NAME(HASHED),
282 	RQF_NAME(STATS),
283 	RQF_NAME(SPECIAL_PAYLOAD),
284 	RQF_NAME(ZONE_WRITE_LOCKED),
285 	RQF_NAME(MQ_POLL_SLEPT),
286 	RQF_NAME(ELV),
287 };
288 #undef RQF_NAME
289 
290 static const char *const blk_mq_rq_state_name_array[] = {
291 	[MQ_RQ_IDLE]		= "idle",
292 	[MQ_RQ_IN_FLIGHT]	= "in_flight",
293 	[MQ_RQ_COMPLETE]	= "complete",
294 };
295 
296 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
297 {
298 	if (WARN_ON_ONCE((unsigned int)rq_state >=
299 			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
300 		return "(?)";
301 	return blk_mq_rq_state_name_array[rq_state];
302 }
303 
304 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
305 {
306 	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
307 	const enum req_op op = req_op(rq);
308 	const char *op_str = blk_op_str(op);
309 
310 	seq_printf(m, "%p {.op=", rq);
311 	if (strcmp(op_str, "UNKNOWN") == 0)
312 		seq_printf(m, "%u", op);
313 	else
314 		seq_printf(m, "%s", op_str);
315 	seq_puts(m, ", .cmd_flags=");
316 	blk_flags_show(m, (__force unsigned int)(rq->cmd_flags & ~REQ_OP_MASK),
317 		       cmd_flag_name, ARRAY_SIZE(cmd_flag_name));
318 	seq_puts(m, ", .rq_flags=");
319 	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
320 		       ARRAY_SIZE(rqf_name));
321 	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
322 	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
323 		   rq->internal_tag);
324 	if (mq_ops->show_rq)
325 		mq_ops->show_rq(m, rq);
326 	seq_puts(m, "}\n");
327 	return 0;
328 }
329 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
330 
331 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
332 {
333 	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
334 }
335 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
336 
337 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
338 	__acquires(&hctx->lock)
339 {
340 	struct blk_mq_hw_ctx *hctx = m->private;
341 
342 	spin_lock(&hctx->lock);
343 	return seq_list_start(&hctx->dispatch, *pos);
344 }
345 
346 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
347 {
348 	struct blk_mq_hw_ctx *hctx = m->private;
349 
350 	return seq_list_next(v, &hctx->dispatch, pos);
351 }
352 
353 static void hctx_dispatch_stop(struct seq_file *m, void *v)
354 	__releases(&hctx->lock)
355 {
356 	struct blk_mq_hw_ctx *hctx = m->private;
357 
358 	spin_unlock(&hctx->lock);
359 }
360 
361 static const struct seq_operations hctx_dispatch_seq_ops = {
362 	.start	= hctx_dispatch_start,
363 	.next	= hctx_dispatch_next,
364 	.stop	= hctx_dispatch_stop,
365 	.show	= blk_mq_debugfs_rq_show,
366 };
367 
368 struct show_busy_params {
369 	struct seq_file		*m;
370 	struct blk_mq_hw_ctx	*hctx;
371 };
372 
373 /*
374  * Note: the state of a request may change while this function is in progress,
375  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
376  * keep iterating requests.
377  */
378 static bool hctx_show_busy_rq(struct request *rq, void *data)
379 {
380 	const struct show_busy_params *params = data;
381 
382 	if (rq->mq_hctx == params->hctx)
383 		__blk_mq_debugfs_rq_show(params->m, rq);
384 
385 	return true;
386 }
387 
388 static int hctx_busy_show(void *data, struct seq_file *m)
389 {
390 	struct blk_mq_hw_ctx *hctx = data;
391 	struct show_busy_params params = { .m = m, .hctx = hctx };
392 
393 	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
394 				&params);
395 
396 	return 0;
397 }
398 
399 static const char *const hctx_types[] = {
400 	[HCTX_TYPE_DEFAULT]	= "default",
401 	[HCTX_TYPE_READ]	= "read",
402 	[HCTX_TYPE_POLL]	= "poll",
403 };
404 
405 static int hctx_type_show(void *data, struct seq_file *m)
406 {
407 	struct blk_mq_hw_ctx *hctx = data;
408 
409 	BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
410 	seq_printf(m, "%s\n", hctx_types[hctx->type]);
411 	return 0;
412 }
413 
414 static int hctx_ctx_map_show(void *data, struct seq_file *m)
415 {
416 	struct blk_mq_hw_ctx *hctx = data;
417 
418 	sbitmap_bitmap_show(&hctx->ctx_map, m);
419 	return 0;
420 }
421 
422 static void blk_mq_debugfs_tags_show(struct seq_file *m,
423 				     struct blk_mq_tags *tags)
424 {
425 	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
426 	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
427 	seq_printf(m, "active_queues=%d\n",
428 		   atomic_read(&tags->active_queues));
429 
430 	seq_puts(m, "\nbitmap_tags:\n");
431 	sbitmap_queue_show(&tags->bitmap_tags, m);
432 
433 	if (tags->nr_reserved_tags) {
434 		seq_puts(m, "\nbreserved_tags:\n");
435 		sbitmap_queue_show(&tags->breserved_tags, m);
436 	}
437 }
438 
439 static int hctx_tags_show(void *data, struct seq_file *m)
440 {
441 	struct blk_mq_hw_ctx *hctx = data;
442 	struct request_queue *q = hctx->queue;
443 	int res;
444 
445 	res = mutex_lock_interruptible(&q->sysfs_lock);
446 	if (res)
447 		goto out;
448 	if (hctx->tags)
449 		blk_mq_debugfs_tags_show(m, hctx->tags);
450 	mutex_unlock(&q->sysfs_lock);
451 
452 out:
453 	return res;
454 }
455 
456 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
457 {
458 	struct blk_mq_hw_ctx *hctx = data;
459 	struct request_queue *q = hctx->queue;
460 	int res;
461 
462 	res = mutex_lock_interruptible(&q->sysfs_lock);
463 	if (res)
464 		goto out;
465 	if (hctx->tags)
466 		sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
467 	mutex_unlock(&q->sysfs_lock);
468 
469 out:
470 	return res;
471 }
472 
473 static int hctx_sched_tags_show(void *data, struct seq_file *m)
474 {
475 	struct blk_mq_hw_ctx *hctx = data;
476 	struct request_queue *q = hctx->queue;
477 	int res;
478 
479 	res = mutex_lock_interruptible(&q->sysfs_lock);
480 	if (res)
481 		goto out;
482 	if (hctx->sched_tags)
483 		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
484 	mutex_unlock(&q->sysfs_lock);
485 
486 out:
487 	return res;
488 }
489 
490 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
491 {
492 	struct blk_mq_hw_ctx *hctx = data;
493 	struct request_queue *q = hctx->queue;
494 	int res;
495 
496 	res = mutex_lock_interruptible(&q->sysfs_lock);
497 	if (res)
498 		goto out;
499 	if (hctx->sched_tags)
500 		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
501 	mutex_unlock(&q->sysfs_lock);
502 
503 out:
504 	return res;
505 }
506 
507 static int hctx_run_show(void *data, struct seq_file *m)
508 {
509 	struct blk_mq_hw_ctx *hctx = data;
510 
511 	seq_printf(m, "%lu\n", hctx->run);
512 	return 0;
513 }
514 
515 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
516 			      loff_t *ppos)
517 {
518 	struct blk_mq_hw_ctx *hctx = data;
519 
520 	hctx->run = 0;
521 	return count;
522 }
523 
524 static int hctx_active_show(void *data, struct seq_file *m)
525 {
526 	struct blk_mq_hw_ctx *hctx = data;
527 
528 	seq_printf(m, "%d\n", __blk_mq_active_requests(hctx));
529 	return 0;
530 }
531 
532 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
533 {
534 	struct blk_mq_hw_ctx *hctx = data;
535 
536 	seq_printf(m, "%u\n", hctx->dispatch_busy);
537 	return 0;
538 }
539 
540 #define CTX_RQ_SEQ_OPS(name, type)					\
541 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
542 	__acquires(&ctx->lock)						\
543 {									\
544 	struct blk_mq_ctx *ctx = m->private;				\
545 									\
546 	spin_lock(&ctx->lock);						\
547 	return seq_list_start(&ctx->rq_lists[type], *pos);		\
548 }									\
549 									\
550 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,	\
551 				     loff_t *pos)			\
552 {									\
553 	struct blk_mq_ctx *ctx = m->private;				\
554 									\
555 	return seq_list_next(v, &ctx->rq_lists[type], pos);		\
556 }									\
557 									\
558 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)	\
559 	__releases(&ctx->lock)						\
560 {									\
561 	struct blk_mq_ctx *ctx = m->private;				\
562 									\
563 	spin_unlock(&ctx->lock);					\
564 }									\
565 									\
566 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {	\
567 	.start	= ctx_##name##_rq_list_start,				\
568 	.next	= ctx_##name##_rq_list_next,				\
569 	.stop	= ctx_##name##_rq_list_stop,				\
570 	.show	= blk_mq_debugfs_rq_show,				\
571 }
572 
573 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
574 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
575 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
576 
577 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
578 {
579 	const struct blk_mq_debugfs_attr *attr = m->private;
580 	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
581 
582 	return attr->show(data, m);
583 }
584 
585 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
586 				    size_t count, loff_t *ppos)
587 {
588 	struct seq_file *m = file->private_data;
589 	const struct blk_mq_debugfs_attr *attr = m->private;
590 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
591 
592 	/*
593 	 * Attributes that only implement .seq_ops are read-only and 'attr' is
594 	 * the same with 'data' in this case.
595 	 */
596 	if (attr == data || !attr->write)
597 		return -EPERM;
598 
599 	return attr->write(data, buf, count, ppos);
600 }
601 
602 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
603 {
604 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
605 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
606 	struct seq_file *m;
607 	int ret;
608 
609 	if (attr->seq_ops) {
610 		ret = seq_open(file, attr->seq_ops);
611 		if (!ret) {
612 			m = file->private_data;
613 			m->private = data;
614 		}
615 		return ret;
616 	}
617 
618 	if (WARN_ON_ONCE(!attr->show))
619 		return -EPERM;
620 
621 	return single_open(file, blk_mq_debugfs_show, inode->i_private);
622 }
623 
624 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
625 {
626 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
627 
628 	if (attr->show)
629 		return single_release(inode, file);
630 
631 	return seq_release(inode, file);
632 }
633 
634 static const struct file_operations blk_mq_debugfs_fops = {
635 	.open		= blk_mq_debugfs_open,
636 	.read		= seq_read,
637 	.write		= blk_mq_debugfs_write,
638 	.llseek		= seq_lseek,
639 	.release	= blk_mq_debugfs_release,
640 };
641 
642 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
643 	{"state", 0400, hctx_state_show},
644 	{"flags", 0400, hctx_flags_show},
645 	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
646 	{"busy", 0400, hctx_busy_show},
647 	{"ctx_map", 0400, hctx_ctx_map_show},
648 	{"tags", 0400, hctx_tags_show},
649 	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
650 	{"sched_tags", 0400, hctx_sched_tags_show},
651 	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
652 	{"run", 0600, hctx_run_show, hctx_run_write},
653 	{"active", 0400, hctx_active_show},
654 	{"dispatch_busy", 0400, hctx_dispatch_busy_show},
655 	{"type", 0400, hctx_type_show},
656 	{},
657 };
658 
659 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
660 	{"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
661 	{"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
662 	{"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
663 	{},
664 };
665 
666 static void debugfs_create_files(struct dentry *parent, void *data,
667 				 const struct blk_mq_debugfs_attr *attr)
668 {
669 	if (IS_ERR_OR_NULL(parent))
670 		return;
671 
672 	d_inode(parent)->i_private = data;
673 
674 	for (; attr->name; attr++)
675 		debugfs_create_file(attr->name, attr->mode, parent,
676 				    (void *)attr, &blk_mq_debugfs_fops);
677 }
678 
679 void blk_mq_debugfs_register(struct request_queue *q)
680 {
681 	struct blk_mq_hw_ctx *hctx;
682 	unsigned long i;
683 
684 	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
685 
686 	/*
687 	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
688 	 * didn't exist yet (because we don't know what to name the directory
689 	 * until the queue is registered to a gendisk).
690 	 */
691 	if (q->elevator && !q->sched_debugfs_dir)
692 		blk_mq_debugfs_register_sched(q);
693 
694 	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
695 	queue_for_each_hw_ctx(q, hctx, i) {
696 		if (!hctx->debugfs_dir)
697 			blk_mq_debugfs_register_hctx(q, hctx);
698 		if (q->elevator && !hctx->sched_debugfs_dir)
699 			blk_mq_debugfs_register_sched_hctx(q, hctx);
700 	}
701 
702 	if (q->rq_qos) {
703 		struct rq_qos *rqos = q->rq_qos;
704 
705 		while (rqos) {
706 			blk_mq_debugfs_register_rqos(rqos);
707 			rqos = rqos->next;
708 		}
709 	}
710 }
711 
712 static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
713 					struct blk_mq_ctx *ctx)
714 {
715 	struct dentry *ctx_dir;
716 	char name[20];
717 
718 	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
719 	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
720 
721 	debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs);
722 }
723 
724 void blk_mq_debugfs_register_hctx(struct request_queue *q,
725 				  struct blk_mq_hw_ctx *hctx)
726 {
727 	struct blk_mq_ctx *ctx;
728 	char name[20];
729 	int i;
730 
731 	if (!q->debugfs_dir)
732 		return;
733 
734 	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
735 	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
736 
737 	debugfs_create_files(hctx->debugfs_dir, hctx, blk_mq_debugfs_hctx_attrs);
738 
739 	hctx_for_each_ctx(hctx, ctx, i)
740 		blk_mq_debugfs_register_ctx(hctx, ctx);
741 }
742 
743 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
744 {
745 	if (!hctx->queue->debugfs_dir)
746 		return;
747 	debugfs_remove_recursive(hctx->debugfs_dir);
748 	hctx->sched_debugfs_dir = NULL;
749 	hctx->debugfs_dir = NULL;
750 }
751 
752 void blk_mq_debugfs_register_hctxs(struct request_queue *q)
753 {
754 	struct blk_mq_hw_ctx *hctx;
755 	unsigned long i;
756 
757 	queue_for_each_hw_ctx(q, hctx, i)
758 		blk_mq_debugfs_register_hctx(q, hctx);
759 }
760 
761 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
762 {
763 	struct blk_mq_hw_ctx *hctx;
764 	unsigned long i;
765 
766 	queue_for_each_hw_ctx(q, hctx, i)
767 		blk_mq_debugfs_unregister_hctx(hctx);
768 }
769 
770 void blk_mq_debugfs_register_sched(struct request_queue *q)
771 {
772 	struct elevator_type *e = q->elevator->type;
773 
774 	lockdep_assert_held(&q->debugfs_mutex);
775 
776 	/*
777 	 * If the parent directory has not been created yet, return, we will be
778 	 * called again later on and the directory/files will be created then.
779 	 */
780 	if (!q->debugfs_dir)
781 		return;
782 
783 	if (!e->queue_debugfs_attrs)
784 		return;
785 
786 	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
787 
788 	debugfs_create_files(q->sched_debugfs_dir, q, e->queue_debugfs_attrs);
789 }
790 
791 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
792 {
793 	lockdep_assert_held(&q->debugfs_mutex);
794 
795 	debugfs_remove_recursive(q->sched_debugfs_dir);
796 	q->sched_debugfs_dir = NULL;
797 }
798 
799 static const char *rq_qos_id_to_name(enum rq_qos_id id)
800 {
801 	switch (id) {
802 	case RQ_QOS_WBT:
803 		return "wbt";
804 	case RQ_QOS_LATENCY:
805 		return "latency";
806 	case RQ_QOS_COST:
807 		return "cost";
808 	case RQ_QOS_IOPRIO:
809 		return "ioprio";
810 	}
811 	return "unknown";
812 }
813 
814 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
815 {
816 	lockdep_assert_held(&rqos->q->debugfs_mutex);
817 
818 	if (!rqos->q->debugfs_dir)
819 		return;
820 	debugfs_remove_recursive(rqos->debugfs_dir);
821 	rqos->debugfs_dir = NULL;
822 }
823 
824 void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
825 {
826 	struct request_queue *q = rqos->q;
827 	const char *dir_name = rq_qos_id_to_name(rqos->id);
828 
829 	lockdep_assert_held(&q->debugfs_mutex);
830 
831 	if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
832 		return;
833 
834 	if (!q->rqos_debugfs_dir)
835 		q->rqos_debugfs_dir = debugfs_create_dir("rqos",
836 							 q->debugfs_dir);
837 
838 	rqos->debugfs_dir = debugfs_create_dir(dir_name,
839 					       rqos->q->rqos_debugfs_dir);
840 
841 	debugfs_create_files(rqos->debugfs_dir, rqos, rqos->ops->debugfs_attrs);
842 }
843 
844 void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
845 					struct blk_mq_hw_ctx *hctx)
846 {
847 	struct elevator_type *e = q->elevator->type;
848 
849 	lockdep_assert_held(&q->debugfs_mutex);
850 
851 	/*
852 	 * If the parent debugfs directory has not been created yet, return;
853 	 * We will be called again later on with appropriate parent debugfs
854 	 * directory from blk_register_queue()
855 	 */
856 	if (!hctx->debugfs_dir)
857 		return;
858 
859 	if (!e->hctx_debugfs_attrs)
860 		return;
861 
862 	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
863 						     hctx->debugfs_dir);
864 	debugfs_create_files(hctx->sched_debugfs_dir, hctx,
865 			     e->hctx_debugfs_attrs);
866 }
867 
868 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
869 {
870 	lockdep_assert_held(&hctx->queue->debugfs_mutex);
871 
872 	if (!hctx->queue->debugfs_dir)
873 		return;
874 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
875 	hctx->sched_debugfs_dir = NULL;
876 }
877