xref: /openbmc/linux/block/blk-mq-debugfs.c (revision bbaa836b)
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-tag.h"
15 #include "blk-rq-qos.h"
16 
17 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
18 {
19 	if (stat->nr_samples) {
20 		seq_printf(m, "samples=%d, mean=%llu, min=%llu, max=%llu",
21 			   stat->nr_samples, stat->mean, stat->min, stat->max);
22 	} else {
23 		seq_puts(m, "samples=0");
24 	}
25 }
26 
27 static int queue_poll_stat_show(void *data, struct seq_file *m)
28 {
29 	struct request_queue *q = data;
30 	int bucket;
31 
32 	for (bucket = 0; bucket < (BLK_MQ_POLL_STATS_BKTS / 2); bucket++) {
33 		seq_printf(m, "read  (%d Bytes): ", 1 << (9 + bucket));
34 		print_stat(m, &q->poll_stat[2 * bucket]);
35 		seq_puts(m, "\n");
36 
37 		seq_printf(m, "write (%d Bytes): ",  1 << (9 + bucket));
38 		print_stat(m, &q->poll_stat[2 * bucket + 1]);
39 		seq_puts(m, "\n");
40 	}
41 	return 0;
42 }
43 
44 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
45 	__acquires(&q->requeue_lock)
46 {
47 	struct request_queue *q = m->private;
48 
49 	spin_lock_irq(&q->requeue_lock);
50 	return seq_list_start(&q->requeue_list, *pos);
51 }
52 
53 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
54 {
55 	struct request_queue *q = m->private;
56 
57 	return seq_list_next(v, &q->requeue_list, pos);
58 }
59 
60 static void queue_requeue_list_stop(struct seq_file *m, void *v)
61 	__releases(&q->requeue_lock)
62 {
63 	struct request_queue *q = m->private;
64 
65 	spin_unlock_irq(&q->requeue_lock);
66 }
67 
68 static const struct seq_operations queue_requeue_list_seq_ops = {
69 	.start	= queue_requeue_list_start,
70 	.next	= queue_requeue_list_next,
71 	.stop	= queue_requeue_list_stop,
72 	.show	= blk_mq_debugfs_rq_show,
73 };
74 
75 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
76 			  const char *const *flag_name, int flag_name_count)
77 {
78 	bool sep = false;
79 	int i;
80 
81 	for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
82 		if (!(flags & BIT(i)))
83 			continue;
84 		if (sep)
85 			seq_puts(m, "|");
86 		sep = true;
87 		if (i < flag_name_count && flag_name[i])
88 			seq_puts(m, flag_name[i]);
89 		else
90 			seq_printf(m, "%d", i);
91 	}
92 	return 0;
93 }
94 
95 static int queue_pm_only_show(void *data, struct seq_file *m)
96 {
97 	struct request_queue *q = data;
98 
99 	seq_printf(m, "%d\n", atomic_read(&q->pm_only));
100 	return 0;
101 }
102 
103 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
104 static const char *const blk_queue_flag_name[] = {
105 	QUEUE_FLAG_NAME(STOPPED),
106 	QUEUE_FLAG_NAME(DYING),
107 	QUEUE_FLAG_NAME(NOMERGES),
108 	QUEUE_FLAG_NAME(SAME_COMP),
109 	QUEUE_FLAG_NAME(FAIL_IO),
110 	QUEUE_FLAG_NAME(NONROT),
111 	QUEUE_FLAG_NAME(IO_STAT),
112 	QUEUE_FLAG_NAME(DISCARD),
113 	QUEUE_FLAG_NAME(NOXMERGES),
114 	QUEUE_FLAG_NAME(ADD_RANDOM),
115 	QUEUE_FLAG_NAME(SECERASE),
116 	QUEUE_FLAG_NAME(SAME_FORCE),
117 	QUEUE_FLAG_NAME(DEAD),
118 	QUEUE_FLAG_NAME(INIT_DONE),
119 	QUEUE_FLAG_NAME(STABLE_WRITES),
120 	QUEUE_FLAG_NAME(POLL),
121 	QUEUE_FLAG_NAME(WC),
122 	QUEUE_FLAG_NAME(FUA),
123 	QUEUE_FLAG_NAME(DAX),
124 	QUEUE_FLAG_NAME(STATS),
125 	QUEUE_FLAG_NAME(POLL_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 after blk_cleanup_queue() has called
154 	 * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
155 	 * triggering a use-after-free.
156 	 */
157 	if (blk_queue_dead(q))
158 		return -ENOENT;
159 
160 	if (count >= sizeof(opbuf)) {
161 		pr_err("%s: operation too long\n", __func__);
162 		goto inval;
163 	}
164 
165 	if (copy_from_user(opbuf, buf, count))
166 		return -EFAULT;
167 	op = strstrip(opbuf);
168 	if (strcmp(op, "run") == 0) {
169 		blk_mq_run_hw_queues(q, true);
170 	} else if (strcmp(op, "start") == 0) {
171 		blk_mq_start_stopped_hw_queues(q, true);
172 	} else if (strcmp(op, "kick") == 0) {
173 		blk_mq_kick_requeue_list(q);
174 	} else {
175 		pr_err("%s: unsupported operation '%s'\n", __func__, op);
176 inval:
177 		pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
178 		return -EINVAL;
179 	}
180 	return count;
181 }
182 
183 static int queue_write_hint_show(void *data, struct seq_file *m)
184 {
185 	struct request_queue *q = data;
186 	int i;
187 
188 	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
189 		seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
190 
191 	return 0;
192 }
193 
194 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
195 				      size_t count, loff_t *ppos)
196 {
197 	struct request_queue *q = data;
198 	int i;
199 
200 	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
201 		q->write_hints[i] = 0;
202 
203 	return count;
204 }
205 
206 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
207 	{ "poll_stat", 0400, queue_poll_stat_show },
208 	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
209 	{ "pm_only", 0600, queue_pm_only_show, NULL },
210 	{ "state", 0600, queue_state_show, queue_state_write },
211 	{ "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
212 	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
213 	{ },
214 };
215 
216 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
217 static const char *const hctx_state_name[] = {
218 	HCTX_STATE_NAME(STOPPED),
219 	HCTX_STATE_NAME(TAG_ACTIVE),
220 	HCTX_STATE_NAME(SCHED_RESTART),
221 	HCTX_STATE_NAME(INACTIVE),
222 };
223 #undef HCTX_STATE_NAME
224 
225 static int hctx_state_show(void *data, struct seq_file *m)
226 {
227 	struct blk_mq_hw_ctx *hctx = data;
228 
229 	blk_flags_show(m, hctx->state, hctx_state_name,
230 		       ARRAY_SIZE(hctx_state_name));
231 	seq_puts(m, "\n");
232 	return 0;
233 }
234 
235 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
236 static const char *const alloc_policy_name[] = {
237 	BLK_TAG_ALLOC_NAME(FIFO),
238 	BLK_TAG_ALLOC_NAME(RR),
239 };
240 #undef BLK_TAG_ALLOC_NAME
241 
242 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
243 static const char *const hctx_flag_name[] = {
244 	HCTX_FLAG_NAME(SHOULD_MERGE),
245 	HCTX_FLAG_NAME(TAG_QUEUE_SHARED),
246 	HCTX_FLAG_NAME(BLOCKING),
247 	HCTX_FLAG_NAME(NO_SCHED),
248 	HCTX_FLAG_NAME(STACKING),
249 	HCTX_FLAG_NAME(TAG_HCTX_SHARED),
250 };
251 #undef HCTX_FLAG_NAME
252 
253 static int hctx_flags_show(void *data, struct seq_file *m)
254 {
255 	struct blk_mq_hw_ctx *hctx = data;
256 	const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
257 
258 	seq_puts(m, "alloc_policy=");
259 	if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
260 	    alloc_policy_name[alloc_policy])
261 		seq_puts(m, alloc_policy_name[alloc_policy]);
262 	else
263 		seq_printf(m, "%d", alloc_policy);
264 	seq_puts(m, " ");
265 	blk_flags_show(m,
266 		       hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
267 		       hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
268 	seq_puts(m, "\n");
269 	return 0;
270 }
271 
272 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
273 static const char *const cmd_flag_name[] = {
274 	CMD_FLAG_NAME(FAILFAST_DEV),
275 	CMD_FLAG_NAME(FAILFAST_TRANSPORT),
276 	CMD_FLAG_NAME(FAILFAST_DRIVER),
277 	CMD_FLAG_NAME(SYNC),
278 	CMD_FLAG_NAME(META),
279 	CMD_FLAG_NAME(PRIO),
280 	CMD_FLAG_NAME(NOMERGE),
281 	CMD_FLAG_NAME(IDLE),
282 	CMD_FLAG_NAME(INTEGRITY),
283 	CMD_FLAG_NAME(FUA),
284 	CMD_FLAG_NAME(PREFLUSH),
285 	CMD_FLAG_NAME(RAHEAD),
286 	CMD_FLAG_NAME(BACKGROUND),
287 	CMD_FLAG_NAME(NOWAIT),
288 	CMD_FLAG_NAME(NOUNMAP),
289 	CMD_FLAG_NAME(POLLED),
290 };
291 #undef CMD_FLAG_NAME
292 
293 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
294 static const char *const rqf_name[] = {
295 	RQF_NAME(STARTED),
296 	RQF_NAME(SOFTBARRIER),
297 	RQF_NAME(FLUSH_SEQ),
298 	RQF_NAME(MIXED_MERGE),
299 	RQF_NAME(MQ_INFLIGHT),
300 	RQF_NAME(DONTPREP),
301 	RQF_NAME(FAILED),
302 	RQF_NAME(QUIET),
303 	RQF_NAME(ELVPRIV),
304 	RQF_NAME(IO_STAT),
305 	RQF_NAME(PM),
306 	RQF_NAME(HASHED),
307 	RQF_NAME(STATS),
308 	RQF_NAME(SPECIAL_PAYLOAD),
309 	RQF_NAME(ZONE_WRITE_LOCKED),
310 	RQF_NAME(MQ_POLL_SLEPT),
311 	RQF_NAME(ELV),
312 };
313 #undef RQF_NAME
314 
315 static const char *const blk_mq_rq_state_name_array[] = {
316 	[MQ_RQ_IDLE]		= "idle",
317 	[MQ_RQ_IN_FLIGHT]	= "in_flight",
318 	[MQ_RQ_COMPLETE]	= "complete",
319 };
320 
321 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
322 {
323 	if (WARN_ON_ONCE((unsigned int)rq_state >=
324 			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
325 		return "(?)";
326 	return blk_mq_rq_state_name_array[rq_state];
327 }
328 
329 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
330 {
331 	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
332 	const unsigned int op = req_op(rq);
333 	const char *op_str = blk_op_str(op);
334 
335 	seq_printf(m, "%p {.op=", rq);
336 	if (strcmp(op_str, "UNKNOWN") == 0)
337 		seq_printf(m, "%u", op);
338 	else
339 		seq_printf(m, "%s", op_str);
340 	seq_puts(m, ", .cmd_flags=");
341 	blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
342 		       ARRAY_SIZE(cmd_flag_name));
343 	seq_puts(m, ", .rq_flags=");
344 	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
345 		       ARRAY_SIZE(rqf_name));
346 	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
347 	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
348 		   rq->internal_tag);
349 	if (mq_ops->show_rq)
350 		mq_ops->show_rq(m, rq);
351 	seq_puts(m, "}\n");
352 	return 0;
353 }
354 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
355 
356 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
357 {
358 	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
359 }
360 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
361 
362 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
363 	__acquires(&hctx->lock)
364 {
365 	struct blk_mq_hw_ctx *hctx = m->private;
366 
367 	spin_lock(&hctx->lock);
368 	return seq_list_start(&hctx->dispatch, *pos);
369 }
370 
371 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
372 {
373 	struct blk_mq_hw_ctx *hctx = m->private;
374 
375 	return seq_list_next(v, &hctx->dispatch, pos);
376 }
377 
378 static void hctx_dispatch_stop(struct seq_file *m, void *v)
379 	__releases(&hctx->lock)
380 {
381 	struct blk_mq_hw_ctx *hctx = m->private;
382 
383 	spin_unlock(&hctx->lock);
384 }
385 
386 static const struct seq_operations hctx_dispatch_seq_ops = {
387 	.start	= hctx_dispatch_start,
388 	.next	= hctx_dispatch_next,
389 	.stop	= hctx_dispatch_stop,
390 	.show	= blk_mq_debugfs_rq_show,
391 };
392 
393 struct show_busy_params {
394 	struct seq_file		*m;
395 	struct blk_mq_hw_ctx	*hctx;
396 };
397 
398 /*
399  * Note: the state of a request may change while this function is in progress,
400  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
401  * keep iterating requests.
402  */
403 static bool hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
404 {
405 	const struct show_busy_params *params = data;
406 
407 	if (rq->mq_hctx == params->hctx)
408 		__blk_mq_debugfs_rq_show(params->m, rq);
409 
410 	return true;
411 }
412 
413 static int hctx_busy_show(void *data, struct seq_file *m)
414 {
415 	struct blk_mq_hw_ctx *hctx = data;
416 	struct show_busy_params params = { .m = m, .hctx = hctx };
417 
418 	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
419 				&params);
420 
421 	return 0;
422 }
423 
424 static const char *const hctx_types[] = {
425 	[HCTX_TYPE_DEFAULT]	= "default",
426 	[HCTX_TYPE_READ]	= "read",
427 	[HCTX_TYPE_POLL]	= "poll",
428 };
429 
430 static int hctx_type_show(void *data, struct seq_file *m)
431 {
432 	struct blk_mq_hw_ctx *hctx = data;
433 
434 	BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
435 	seq_printf(m, "%s\n", hctx_types[hctx->type]);
436 	return 0;
437 }
438 
439 static int hctx_ctx_map_show(void *data, struct seq_file *m)
440 {
441 	struct blk_mq_hw_ctx *hctx = data;
442 
443 	sbitmap_bitmap_show(&hctx->ctx_map, m);
444 	return 0;
445 }
446 
447 static void blk_mq_debugfs_tags_show(struct seq_file *m,
448 				     struct blk_mq_tags *tags)
449 {
450 	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
451 	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
452 	seq_printf(m, "active_queues=%d\n",
453 		   atomic_read(&tags->active_queues));
454 
455 	seq_puts(m, "\nbitmap_tags:\n");
456 	sbitmap_queue_show(&tags->bitmap_tags, m);
457 
458 	if (tags->nr_reserved_tags) {
459 		seq_puts(m, "\nbreserved_tags:\n");
460 		sbitmap_queue_show(&tags->breserved_tags, m);
461 	}
462 }
463 
464 static int hctx_tags_show(void *data, struct seq_file *m)
465 {
466 	struct blk_mq_hw_ctx *hctx = data;
467 	struct request_queue *q = hctx->queue;
468 	int res;
469 
470 	res = mutex_lock_interruptible(&q->sysfs_lock);
471 	if (res)
472 		goto out;
473 	if (hctx->tags)
474 		blk_mq_debugfs_tags_show(m, hctx->tags);
475 	mutex_unlock(&q->sysfs_lock);
476 
477 out:
478 	return res;
479 }
480 
481 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
482 {
483 	struct blk_mq_hw_ctx *hctx = data;
484 	struct request_queue *q = hctx->queue;
485 	int res;
486 
487 	res = mutex_lock_interruptible(&q->sysfs_lock);
488 	if (res)
489 		goto out;
490 	if (hctx->tags)
491 		sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
492 	mutex_unlock(&q->sysfs_lock);
493 
494 out:
495 	return res;
496 }
497 
498 static int hctx_sched_tags_show(void *data, struct seq_file *m)
499 {
500 	struct blk_mq_hw_ctx *hctx = data;
501 	struct request_queue *q = hctx->queue;
502 	int res;
503 
504 	res = mutex_lock_interruptible(&q->sysfs_lock);
505 	if (res)
506 		goto out;
507 	if (hctx->sched_tags)
508 		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
509 	mutex_unlock(&q->sysfs_lock);
510 
511 out:
512 	return res;
513 }
514 
515 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
516 {
517 	struct blk_mq_hw_ctx *hctx = data;
518 	struct request_queue *q = hctx->queue;
519 	int res;
520 
521 	res = mutex_lock_interruptible(&q->sysfs_lock);
522 	if (res)
523 		goto out;
524 	if (hctx->sched_tags)
525 		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
526 	mutex_unlock(&q->sysfs_lock);
527 
528 out:
529 	return res;
530 }
531 
532 static int hctx_run_show(void *data, struct seq_file *m)
533 {
534 	struct blk_mq_hw_ctx *hctx = data;
535 
536 	seq_printf(m, "%lu\n", hctx->run);
537 	return 0;
538 }
539 
540 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
541 			      loff_t *ppos)
542 {
543 	struct blk_mq_hw_ctx *hctx = data;
544 
545 	hctx->run = 0;
546 	return count;
547 }
548 
549 static int hctx_active_show(void *data, struct seq_file *m)
550 {
551 	struct blk_mq_hw_ctx *hctx = data;
552 
553 	seq_printf(m, "%d\n", __blk_mq_active_requests(hctx));
554 	return 0;
555 }
556 
557 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
558 {
559 	struct blk_mq_hw_ctx *hctx = data;
560 
561 	seq_printf(m, "%u\n", hctx->dispatch_busy);
562 	return 0;
563 }
564 
565 #define CTX_RQ_SEQ_OPS(name, type)					\
566 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
567 	__acquires(&ctx->lock)						\
568 {									\
569 	struct blk_mq_ctx *ctx = m->private;				\
570 									\
571 	spin_lock(&ctx->lock);						\
572 	return seq_list_start(&ctx->rq_lists[type], *pos);		\
573 }									\
574 									\
575 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,	\
576 				     loff_t *pos)			\
577 {									\
578 	struct blk_mq_ctx *ctx = m->private;				\
579 									\
580 	return seq_list_next(v, &ctx->rq_lists[type], pos);		\
581 }									\
582 									\
583 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)	\
584 	__releases(&ctx->lock)						\
585 {									\
586 	struct blk_mq_ctx *ctx = m->private;				\
587 									\
588 	spin_unlock(&ctx->lock);					\
589 }									\
590 									\
591 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {	\
592 	.start	= ctx_##name##_rq_list_start,				\
593 	.next	= ctx_##name##_rq_list_next,				\
594 	.stop	= ctx_##name##_rq_list_stop,				\
595 	.show	= blk_mq_debugfs_rq_show,				\
596 }
597 
598 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
599 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
600 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
601 
602 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
603 {
604 	const struct blk_mq_debugfs_attr *attr = m->private;
605 	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
606 
607 	return attr->show(data, m);
608 }
609 
610 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
611 				    size_t count, loff_t *ppos)
612 {
613 	struct seq_file *m = file->private_data;
614 	const struct blk_mq_debugfs_attr *attr = m->private;
615 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
616 
617 	/*
618 	 * Attributes that only implement .seq_ops are read-only and 'attr' is
619 	 * the same with 'data' in this case.
620 	 */
621 	if (attr == data || !attr->write)
622 		return -EPERM;
623 
624 	return attr->write(data, buf, count, ppos);
625 }
626 
627 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
628 {
629 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
630 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
631 	struct seq_file *m;
632 	int ret;
633 
634 	if (attr->seq_ops) {
635 		ret = seq_open(file, attr->seq_ops);
636 		if (!ret) {
637 			m = file->private_data;
638 			m->private = data;
639 		}
640 		return ret;
641 	}
642 
643 	if (WARN_ON_ONCE(!attr->show))
644 		return -EPERM;
645 
646 	return single_open(file, blk_mq_debugfs_show, inode->i_private);
647 }
648 
649 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
650 {
651 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
652 
653 	if (attr->show)
654 		return single_release(inode, file);
655 
656 	return seq_release(inode, file);
657 }
658 
659 static const struct file_operations blk_mq_debugfs_fops = {
660 	.open		= blk_mq_debugfs_open,
661 	.read		= seq_read,
662 	.write		= blk_mq_debugfs_write,
663 	.llseek		= seq_lseek,
664 	.release	= blk_mq_debugfs_release,
665 };
666 
667 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
668 	{"state", 0400, hctx_state_show},
669 	{"flags", 0400, hctx_flags_show},
670 	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
671 	{"busy", 0400, hctx_busy_show},
672 	{"ctx_map", 0400, hctx_ctx_map_show},
673 	{"tags", 0400, hctx_tags_show},
674 	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
675 	{"sched_tags", 0400, hctx_sched_tags_show},
676 	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
677 	{"run", 0600, hctx_run_show, hctx_run_write},
678 	{"active", 0400, hctx_active_show},
679 	{"dispatch_busy", 0400, hctx_dispatch_busy_show},
680 	{"type", 0400, hctx_type_show},
681 	{},
682 };
683 
684 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
685 	{"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
686 	{"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
687 	{"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
688 	{},
689 };
690 
691 static void debugfs_create_files(struct dentry *parent, void *data,
692 				 const struct blk_mq_debugfs_attr *attr)
693 {
694 	if (IS_ERR_OR_NULL(parent))
695 		return;
696 
697 	d_inode(parent)->i_private = data;
698 
699 	for (; attr->name; attr++)
700 		debugfs_create_file(attr->name, attr->mode, parent,
701 				    (void *)attr, &blk_mq_debugfs_fops);
702 }
703 
704 void blk_mq_debugfs_register(struct request_queue *q)
705 {
706 	struct blk_mq_hw_ctx *hctx;
707 	int i;
708 
709 	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
710 
711 	/*
712 	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
713 	 * didn't exist yet (because we don't know what to name the directory
714 	 * until the queue is registered to a gendisk).
715 	 */
716 	if (q->elevator && !q->sched_debugfs_dir)
717 		blk_mq_debugfs_register_sched(q);
718 
719 	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
720 	queue_for_each_hw_ctx(q, hctx, i) {
721 		if (!hctx->debugfs_dir)
722 			blk_mq_debugfs_register_hctx(q, hctx);
723 		if (q->elevator && !hctx->sched_debugfs_dir)
724 			blk_mq_debugfs_register_sched_hctx(q, hctx);
725 	}
726 
727 	if (q->rq_qos) {
728 		struct rq_qos *rqos = q->rq_qos;
729 
730 		while (rqos) {
731 			blk_mq_debugfs_register_rqos(rqos);
732 			rqos = rqos->next;
733 		}
734 	}
735 }
736 
737 void blk_mq_debugfs_unregister(struct request_queue *q)
738 {
739 	q->sched_debugfs_dir = NULL;
740 }
741 
742 static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
743 					struct blk_mq_ctx *ctx)
744 {
745 	struct dentry *ctx_dir;
746 	char name[20];
747 
748 	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
749 	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
750 
751 	debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs);
752 }
753 
754 void blk_mq_debugfs_register_hctx(struct request_queue *q,
755 				  struct blk_mq_hw_ctx *hctx)
756 {
757 	struct blk_mq_ctx *ctx;
758 	char name[20];
759 	int i;
760 
761 	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
762 	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
763 
764 	debugfs_create_files(hctx->debugfs_dir, hctx, blk_mq_debugfs_hctx_attrs);
765 
766 	hctx_for_each_ctx(hctx, ctx, i)
767 		blk_mq_debugfs_register_ctx(hctx, ctx);
768 }
769 
770 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
771 {
772 	debugfs_remove_recursive(hctx->debugfs_dir);
773 	hctx->sched_debugfs_dir = NULL;
774 	hctx->debugfs_dir = NULL;
775 }
776 
777 void blk_mq_debugfs_register_hctxs(struct request_queue *q)
778 {
779 	struct blk_mq_hw_ctx *hctx;
780 	int i;
781 
782 	queue_for_each_hw_ctx(q, hctx, i)
783 		blk_mq_debugfs_register_hctx(q, hctx);
784 }
785 
786 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
787 {
788 	struct blk_mq_hw_ctx *hctx;
789 	int i;
790 
791 	queue_for_each_hw_ctx(q, hctx, i)
792 		blk_mq_debugfs_unregister_hctx(hctx);
793 }
794 
795 void blk_mq_debugfs_register_sched(struct request_queue *q)
796 {
797 	struct elevator_type *e = q->elevator->type;
798 
799 	/*
800 	 * If the parent directory has not been created yet, return, we will be
801 	 * called again later on and the directory/files will be created then.
802 	 */
803 	if (!q->debugfs_dir)
804 		return;
805 
806 	if (!e->queue_debugfs_attrs)
807 		return;
808 
809 	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
810 
811 	debugfs_create_files(q->sched_debugfs_dir, q, e->queue_debugfs_attrs);
812 }
813 
814 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
815 {
816 	debugfs_remove_recursive(q->sched_debugfs_dir);
817 	q->sched_debugfs_dir = NULL;
818 }
819 
820 static const char *rq_qos_id_to_name(enum rq_qos_id id)
821 {
822 	switch (id) {
823 	case RQ_QOS_WBT:
824 		return "wbt";
825 	case RQ_QOS_LATENCY:
826 		return "latency";
827 	case RQ_QOS_COST:
828 		return "cost";
829 	case RQ_QOS_IOPRIO:
830 		return "ioprio";
831 	}
832 	return "unknown";
833 }
834 
835 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
836 {
837 	debugfs_remove_recursive(rqos->debugfs_dir);
838 	rqos->debugfs_dir = NULL;
839 }
840 
841 void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
842 {
843 	struct request_queue *q = rqos->q;
844 	const char *dir_name = rq_qos_id_to_name(rqos->id);
845 
846 	if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
847 		return;
848 
849 	if (!q->rqos_debugfs_dir)
850 		q->rqos_debugfs_dir = debugfs_create_dir("rqos",
851 							 q->debugfs_dir);
852 
853 	rqos->debugfs_dir = debugfs_create_dir(dir_name,
854 					       rqos->q->rqos_debugfs_dir);
855 
856 	debugfs_create_files(rqos->debugfs_dir, rqos, rqos->ops->debugfs_attrs);
857 }
858 
859 void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q)
860 {
861 	debugfs_remove_recursive(q->rqos_debugfs_dir);
862 	q->rqos_debugfs_dir = NULL;
863 }
864 
865 void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
866 					struct blk_mq_hw_ctx *hctx)
867 {
868 	struct elevator_type *e = q->elevator->type;
869 
870 	/*
871 	 * If the parent debugfs directory has not been created yet, return;
872 	 * We will be called again later on with appropriate parent debugfs
873 	 * directory from blk_register_queue()
874 	 */
875 	if (!hctx->debugfs_dir)
876 		return;
877 
878 	if (!e->hctx_debugfs_attrs)
879 		return;
880 
881 	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
882 						     hctx->debugfs_dir);
883 	debugfs_create_files(hctx->sched_debugfs_dir, hctx,
884 			     e->hctx_debugfs_attrs);
885 }
886 
887 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
888 {
889 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
890 	hctx->sched_debugfs_dir = NULL;
891 }
892