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
17static 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
27static 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
44static 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
53static 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
60static 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
68static 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
75static 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
95static 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
104static 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(SCSI_PASSTHROUGH),
128	QUEUE_FLAG_NAME(QUIESCED),
129	QUEUE_FLAG_NAME(PCI_P2PDMA),
130	QUEUE_FLAG_NAME(ZONE_RESETALL),
131	QUEUE_FLAG_NAME(RQ_ALLOC_TIME),
132	QUEUE_FLAG_NAME(HCTX_ACTIVE),
133	QUEUE_FLAG_NAME(NOWAIT),
134};
135#undef QUEUE_FLAG_NAME
136
137static int queue_state_show(void *data, struct seq_file *m)
138{
139	struct request_queue *q = data;
140
141	blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
142		       ARRAY_SIZE(blk_queue_flag_name));
143	seq_puts(m, "\n");
144	return 0;
145}
146
147static ssize_t queue_state_write(void *data, const char __user *buf,
148				 size_t count, loff_t *ppos)
149{
150	struct request_queue *q = data;
151	char opbuf[16] = { }, *op;
152
153	/*
154	 * The "state" attribute is removed after blk_cleanup_queue() has called
155	 * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
156	 * triggering a use-after-free.
157	 */
158	if (blk_queue_dead(q))
159		return -ENOENT;
160
161	if (count >= sizeof(opbuf)) {
162		pr_err("%s: operation too long\n", __func__);
163		goto inval;
164	}
165
166	if (copy_from_user(opbuf, buf, count))
167		return -EFAULT;
168	op = strstrip(opbuf);
169	if (strcmp(op, "run") == 0) {
170		blk_mq_run_hw_queues(q, true);
171	} else if (strcmp(op, "start") == 0) {
172		blk_mq_start_stopped_hw_queues(q, true);
173	} else if (strcmp(op, "kick") == 0) {
174		blk_mq_kick_requeue_list(q);
175	} else {
176		pr_err("%s: unsupported operation '%s'\n", __func__, op);
177inval:
178		pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
179		return -EINVAL;
180	}
181	return count;
182}
183
184static int queue_write_hint_show(void *data, struct seq_file *m)
185{
186	struct request_queue *q = data;
187	int i;
188
189	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
190		seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
191
192	return 0;
193}
194
195static ssize_t queue_write_hint_store(void *data, const char __user *buf,
196				      size_t count, loff_t *ppos)
197{
198	struct request_queue *q = data;
199	int i;
200
201	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
202		q->write_hints[i] = 0;
203
204	return count;
205}
206
207static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
208	{ "poll_stat", 0400, queue_poll_stat_show },
209	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
210	{ "pm_only", 0600, queue_pm_only_show, NULL },
211	{ "state", 0600, queue_state_show, queue_state_write },
212	{ "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
213	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
214	{ },
215};
216
217#define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
218static const char *const hctx_state_name[] = {
219	HCTX_STATE_NAME(STOPPED),
220	HCTX_STATE_NAME(TAG_ACTIVE),
221	HCTX_STATE_NAME(SCHED_RESTART),
222	HCTX_STATE_NAME(INACTIVE),
223};
224#undef HCTX_STATE_NAME
225
226static int hctx_state_show(void *data, struct seq_file *m)
227{
228	struct blk_mq_hw_ctx *hctx = data;
229
230	blk_flags_show(m, hctx->state, hctx_state_name,
231		       ARRAY_SIZE(hctx_state_name));
232	seq_puts(m, "\n");
233	return 0;
234}
235
236#define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
237static const char *const alloc_policy_name[] = {
238	BLK_TAG_ALLOC_NAME(FIFO),
239	BLK_TAG_ALLOC_NAME(RR),
240};
241#undef BLK_TAG_ALLOC_NAME
242
243#define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
244static const char *const hctx_flag_name[] = {
245	HCTX_FLAG_NAME(SHOULD_MERGE),
246	HCTX_FLAG_NAME(TAG_QUEUE_SHARED),
247	HCTX_FLAG_NAME(BLOCKING),
248	HCTX_FLAG_NAME(NO_SCHED),
249	HCTX_FLAG_NAME(STACKING),
250	HCTX_FLAG_NAME(TAG_HCTX_SHARED),
251};
252#undef HCTX_FLAG_NAME
253
254static int hctx_flags_show(void *data, struct seq_file *m)
255{
256	struct blk_mq_hw_ctx *hctx = data;
257	const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
258
259	seq_puts(m, "alloc_policy=");
260	if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
261	    alloc_policy_name[alloc_policy])
262		seq_puts(m, alloc_policy_name[alloc_policy]);
263	else
264		seq_printf(m, "%d", alloc_policy);
265	seq_puts(m, " ");
266	blk_flags_show(m,
267		       hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
268		       hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
269	seq_puts(m, "\n");
270	return 0;
271}
272
273#define CMD_FLAG_NAME(name) [__REQ_##name] = #name
274static const char *const cmd_flag_name[] = {
275	CMD_FLAG_NAME(FAILFAST_DEV),
276	CMD_FLAG_NAME(FAILFAST_TRANSPORT),
277	CMD_FLAG_NAME(FAILFAST_DRIVER),
278	CMD_FLAG_NAME(SYNC),
279	CMD_FLAG_NAME(META),
280	CMD_FLAG_NAME(PRIO),
281	CMD_FLAG_NAME(NOMERGE),
282	CMD_FLAG_NAME(IDLE),
283	CMD_FLAG_NAME(INTEGRITY),
284	CMD_FLAG_NAME(FUA),
285	CMD_FLAG_NAME(PREFLUSH),
286	CMD_FLAG_NAME(RAHEAD),
287	CMD_FLAG_NAME(BACKGROUND),
288	CMD_FLAG_NAME(NOWAIT),
289	CMD_FLAG_NAME(NOUNMAP),
290	CMD_FLAG_NAME(HIPRI),
291};
292#undef CMD_FLAG_NAME
293
294#define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
295static const char *const rqf_name[] = {
296	RQF_NAME(SORTED),
297	RQF_NAME(STARTED),
298	RQF_NAME(SOFTBARRIER),
299	RQF_NAME(FLUSH_SEQ),
300	RQF_NAME(MIXED_MERGE),
301	RQF_NAME(MQ_INFLIGHT),
302	RQF_NAME(DONTPREP),
303	RQF_NAME(FAILED),
304	RQF_NAME(QUIET),
305	RQF_NAME(ELVPRIV),
306	RQF_NAME(IO_STAT),
307	RQF_NAME(ALLOCED),
308	RQF_NAME(PM),
309	RQF_NAME(HASHED),
310	RQF_NAME(STATS),
311	RQF_NAME(SPECIAL_PAYLOAD),
312	RQF_NAME(ZONE_WRITE_LOCKED),
313	RQF_NAME(MQ_POLL_SLEPT),
314};
315#undef RQF_NAME
316
317static const char *const blk_mq_rq_state_name_array[] = {
318	[MQ_RQ_IDLE]		= "idle",
319	[MQ_RQ_IN_FLIGHT]	= "in_flight",
320	[MQ_RQ_COMPLETE]	= "complete",
321};
322
323static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
324{
325	if (WARN_ON_ONCE((unsigned int)rq_state >=
326			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
327		return "(?)";
328	return blk_mq_rq_state_name_array[rq_state];
329}
330
331int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
332{
333	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
334	const unsigned int op = req_op(rq);
335	const char *op_str = blk_op_str(op);
336
337	seq_printf(m, "%p {.op=", rq);
338	if (strcmp(op_str, "UNKNOWN") == 0)
339		seq_printf(m, "%u", op);
340	else
341		seq_printf(m, "%s", op_str);
342	seq_puts(m, ", .cmd_flags=");
343	blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
344		       ARRAY_SIZE(cmd_flag_name));
345	seq_puts(m, ", .rq_flags=");
346	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
347		       ARRAY_SIZE(rqf_name));
348	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
349	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
350		   rq->internal_tag);
351	if (mq_ops->show_rq)
352		mq_ops->show_rq(m, rq);
353	seq_puts(m, "}\n");
354	return 0;
355}
356EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
357
358int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
359{
360	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
361}
362EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
363
364static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
365	__acquires(&hctx->lock)
366{
367	struct blk_mq_hw_ctx *hctx = m->private;
368
369	spin_lock(&hctx->lock);
370	return seq_list_start(&hctx->dispatch, *pos);
371}
372
373static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
374{
375	struct blk_mq_hw_ctx *hctx = m->private;
376
377	return seq_list_next(v, &hctx->dispatch, pos);
378}
379
380static void hctx_dispatch_stop(struct seq_file *m, void *v)
381	__releases(&hctx->lock)
382{
383	struct blk_mq_hw_ctx *hctx = m->private;
384
385	spin_unlock(&hctx->lock);
386}
387
388static const struct seq_operations hctx_dispatch_seq_ops = {
389	.start	= hctx_dispatch_start,
390	.next	= hctx_dispatch_next,
391	.stop	= hctx_dispatch_stop,
392	.show	= blk_mq_debugfs_rq_show,
393};
394
395struct show_busy_params {
396	struct seq_file		*m;
397	struct blk_mq_hw_ctx	*hctx;
398};
399
400/*
401 * Note: the state of a request may change while this function is in progress,
402 * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
403 * keep iterating requests.
404 */
405static bool hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
406{
407	const struct show_busy_params *params = data;
408
409	if (rq->mq_hctx == params->hctx)
410		__blk_mq_debugfs_rq_show(params->m, rq);
411
412	return true;
413}
414
415static int hctx_busy_show(void *data, struct seq_file *m)
416{
417	struct blk_mq_hw_ctx *hctx = data;
418	struct show_busy_params params = { .m = m, .hctx = hctx };
419
420	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
421				&params);
422
423	return 0;
424}
425
426static const char *const hctx_types[] = {
427	[HCTX_TYPE_DEFAULT]	= "default",
428	[HCTX_TYPE_READ]	= "read",
429	[HCTX_TYPE_POLL]	= "poll",
430};
431
432static int hctx_type_show(void *data, struct seq_file *m)
433{
434	struct blk_mq_hw_ctx *hctx = data;
435
436	BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
437	seq_printf(m, "%s\n", hctx_types[hctx->type]);
438	return 0;
439}
440
441static int hctx_ctx_map_show(void *data, struct seq_file *m)
442{
443	struct blk_mq_hw_ctx *hctx = data;
444
445	sbitmap_bitmap_show(&hctx->ctx_map, m);
446	return 0;
447}
448
449static void blk_mq_debugfs_tags_show(struct seq_file *m,
450				     struct blk_mq_tags *tags)
451{
452	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
453	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
454	seq_printf(m, "active_queues=%d\n",
455		   atomic_read(&tags->active_queues));
456
457	seq_puts(m, "\nbitmap_tags:\n");
458	sbitmap_queue_show(tags->bitmap_tags, m);
459
460	if (tags->nr_reserved_tags) {
461		seq_puts(m, "\nbreserved_tags:\n");
462		sbitmap_queue_show(tags->breserved_tags, m);
463	}
464}
465
466static int hctx_tags_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->tags)
476		blk_mq_debugfs_tags_show(m, hctx->tags);
477	mutex_unlock(&q->sysfs_lock);
478
479out:
480	return res;
481}
482
483static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
484{
485	struct blk_mq_hw_ctx *hctx = data;
486	struct request_queue *q = hctx->queue;
487	int res;
488
489	res = mutex_lock_interruptible(&q->sysfs_lock);
490	if (res)
491		goto out;
492	if (hctx->tags)
493		sbitmap_bitmap_show(&hctx->tags->bitmap_tags->sb, m);
494	mutex_unlock(&q->sysfs_lock);
495
496out:
497	return res;
498}
499
500static int hctx_sched_tags_show(void *data, struct seq_file *m)
501{
502	struct blk_mq_hw_ctx *hctx = data;
503	struct request_queue *q = hctx->queue;
504	int res;
505
506	res = mutex_lock_interruptible(&q->sysfs_lock);
507	if (res)
508		goto out;
509	if (hctx->sched_tags)
510		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
511	mutex_unlock(&q->sysfs_lock);
512
513out:
514	return res;
515}
516
517static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
518{
519	struct blk_mq_hw_ctx *hctx = data;
520	struct request_queue *q = hctx->queue;
521	int res;
522
523	res = mutex_lock_interruptible(&q->sysfs_lock);
524	if (res)
525		goto out;
526	if (hctx->sched_tags)
527		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags->sb, m);
528	mutex_unlock(&q->sysfs_lock);
529
530out:
531	return res;
532}
533
534static int hctx_io_poll_show(void *data, struct seq_file *m)
535{
536	struct blk_mq_hw_ctx *hctx = data;
537
538	seq_printf(m, "considered=%lu\n", hctx->poll_considered);
539	seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
540	seq_printf(m, "success=%lu\n", hctx->poll_success);
541	return 0;
542}
543
544static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
545				  size_t count, loff_t *ppos)
546{
547	struct blk_mq_hw_ctx *hctx = data;
548
549	hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
550	return count;
551}
552
553static int hctx_dispatched_show(void *data, struct seq_file *m)
554{
555	struct blk_mq_hw_ctx *hctx = data;
556	int i;
557
558	seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
559
560	for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
561		unsigned int d = 1U << (i - 1);
562
563		seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
564	}
565
566	seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
567	return 0;
568}
569
570static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
571				     size_t count, loff_t *ppos)
572{
573	struct blk_mq_hw_ctx *hctx = data;
574	int i;
575
576	for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
577		hctx->dispatched[i] = 0;
578	return count;
579}
580
581static int hctx_queued_show(void *data, struct seq_file *m)
582{
583	struct blk_mq_hw_ctx *hctx = data;
584
585	seq_printf(m, "%lu\n", hctx->queued);
586	return 0;
587}
588
589static ssize_t hctx_queued_write(void *data, const char __user *buf,
590				 size_t count, loff_t *ppos)
591{
592	struct blk_mq_hw_ctx *hctx = data;
593
594	hctx->queued = 0;
595	return count;
596}
597
598static int hctx_run_show(void *data, struct seq_file *m)
599{
600	struct blk_mq_hw_ctx *hctx = data;
601
602	seq_printf(m, "%lu\n", hctx->run);
603	return 0;
604}
605
606static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
607			      loff_t *ppos)
608{
609	struct blk_mq_hw_ctx *hctx = data;
610
611	hctx->run = 0;
612	return count;
613}
614
615static int hctx_active_show(void *data, struct seq_file *m)
616{
617	struct blk_mq_hw_ctx *hctx = data;
618
619	seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
620	return 0;
621}
622
623static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
624{
625	struct blk_mq_hw_ctx *hctx = data;
626
627	seq_printf(m, "%u\n", hctx->dispatch_busy);
628	return 0;
629}
630
631#define CTX_RQ_SEQ_OPS(name, type)					\
632static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
633	__acquires(&ctx->lock)						\
634{									\
635	struct blk_mq_ctx *ctx = m->private;				\
636									\
637	spin_lock(&ctx->lock);						\
638	return seq_list_start(&ctx->rq_lists[type], *pos);		\
639}									\
640									\
641static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,	\
642				     loff_t *pos)			\
643{									\
644	struct blk_mq_ctx *ctx = m->private;				\
645									\
646	return seq_list_next(v, &ctx->rq_lists[type], pos);		\
647}									\
648									\
649static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)	\
650	__releases(&ctx->lock)						\
651{									\
652	struct blk_mq_ctx *ctx = m->private;				\
653									\
654	spin_unlock(&ctx->lock);					\
655}									\
656									\
657static const struct seq_operations ctx_##name##_rq_list_seq_ops = {	\
658	.start	= ctx_##name##_rq_list_start,				\
659	.next	= ctx_##name##_rq_list_next,				\
660	.stop	= ctx_##name##_rq_list_stop,				\
661	.show	= blk_mq_debugfs_rq_show,				\
662}
663
664CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
665CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
666CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
667
668static int ctx_dispatched_show(void *data, struct seq_file *m)
669{
670	struct blk_mq_ctx *ctx = data;
671
672	seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
673	return 0;
674}
675
676static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
677				    size_t count, loff_t *ppos)
678{
679	struct blk_mq_ctx *ctx = data;
680
681	ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
682	return count;
683}
684
685static int ctx_merged_show(void *data, struct seq_file *m)
686{
687	struct blk_mq_ctx *ctx = data;
688
689	seq_printf(m, "%lu\n", ctx->rq_merged);
690	return 0;
691}
692
693static ssize_t ctx_merged_write(void *data, const char __user *buf,
694				size_t count, loff_t *ppos)
695{
696	struct blk_mq_ctx *ctx = data;
697
698	ctx->rq_merged = 0;
699	return count;
700}
701
702static int ctx_completed_show(void *data, struct seq_file *m)
703{
704	struct blk_mq_ctx *ctx = data;
705
706	seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
707	return 0;
708}
709
710static ssize_t ctx_completed_write(void *data, const char __user *buf,
711				   size_t count, loff_t *ppos)
712{
713	struct blk_mq_ctx *ctx = data;
714
715	ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
716	return count;
717}
718
719static int blk_mq_debugfs_show(struct seq_file *m, void *v)
720{
721	const struct blk_mq_debugfs_attr *attr = m->private;
722	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
723
724	return attr->show(data, m);
725}
726
727static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
728				    size_t count, loff_t *ppos)
729{
730	struct seq_file *m = file->private_data;
731	const struct blk_mq_debugfs_attr *attr = m->private;
732	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
733
734	/*
735	 * Attributes that only implement .seq_ops are read-only and 'attr' is
736	 * the same with 'data' in this case.
737	 */
738	if (attr == data || !attr->write)
739		return -EPERM;
740
741	return attr->write(data, buf, count, ppos);
742}
743
744static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
745{
746	const struct blk_mq_debugfs_attr *attr = inode->i_private;
747	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
748	struct seq_file *m;
749	int ret;
750
751	if (attr->seq_ops) {
752		ret = seq_open(file, attr->seq_ops);
753		if (!ret) {
754			m = file->private_data;
755			m->private = data;
756		}
757		return ret;
758	}
759
760	if (WARN_ON_ONCE(!attr->show))
761		return -EPERM;
762
763	return single_open(file, blk_mq_debugfs_show, inode->i_private);
764}
765
766static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
767{
768	const struct blk_mq_debugfs_attr *attr = inode->i_private;
769
770	if (attr->show)
771		return single_release(inode, file);
772
773	return seq_release(inode, file);
774}
775
776static const struct file_operations blk_mq_debugfs_fops = {
777	.open		= blk_mq_debugfs_open,
778	.read		= seq_read,
779	.write		= blk_mq_debugfs_write,
780	.llseek		= seq_lseek,
781	.release	= blk_mq_debugfs_release,
782};
783
784static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
785	{"state", 0400, hctx_state_show},
786	{"flags", 0400, hctx_flags_show},
787	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
788	{"busy", 0400, hctx_busy_show},
789	{"ctx_map", 0400, hctx_ctx_map_show},
790	{"tags", 0400, hctx_tags_show},
791	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
792	{"sched_tags", 0400, hctx_sched_tags_show},
793	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
794	{"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
795	{"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
796	{"queued", 0600, hctx_queued_show, hctx_queued_write},
797	{"run", 0600, hctx_run_show, hctx_run_write},
798	{"active", 0400, hctx_active_show},
799	{"dispatch_busy", 0400, hctx_dispatch_busy_show},
800	{"type", 0400, hctx_type_show},
801	{},
802};
803
804static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
805	{"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
806	{"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
807	{"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
808	{"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
809	{"merged", 0600, ctx_merged_show, ctx_merged_write},
810	{"completed", 0600, ctx_completed_show, ctx_completed_write},
811	{},
812};
813
814static void debugfs_create_files(struct dentry *parent, void *data,
815				 const struct blk_mq_debugfs_attr *attr)
816{
817	if (IS_ERR_OR_NULL(parent))
818		return;
819
820	d_inode(parent)->i_private = data;
821
822	for (; attr->name; attr++)
823		debugfs_create_file(attr->name, attr->mode, parent,
824				    (void *)attr, &blk_mq_debugfs_fops);
825}
826
827void blk_mq_debugfs_register(struct request_queue *q)
828{
829	struct blk_mq_hw_ctx *hctx;
830	int i;
831
832	debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
833
834	/*
835	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
836	 * didn't exist yet (because we don't know what to name the directory
837	 * until the queue is registered to a gendisk).
838	 */
839	if (q->elevator && !q->sched_debugfs_dir)
840		blk_mq_debugfs_register_sched(q);
841
842	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
843	queue_for_each_hw_ctx(q, hctx, i) {
844		if (!hctx->debugfs_dir)
845			blk_mq_debugfs_register_hctx(q, hctx);
846		if (q->elevator && !hctx->sched_debugfs_dir)
847			blk_mq_debugfs_register_sched_hctx(q, hctx);
848	}
849
850	if (q->rq_qos) {
851		struct rq_qos *rqos = q->rq_qos;
852
853		while (rqos) {
854			blk_mq_debugfs_register_rqos(rqos);
855			rqos = rqos->next;
856		}
857	}
858}
859
860void blk_mq_debugfs_unregister(struct request_queue *q)
861{
862	q->sched_debugfs_dir = NULL;
863}
864
865static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
866					struct blk_mq_ctx *ctx)
867{
868	struct dentry *ctx_dir;
869	char name[20];
870
871	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
872	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
873
874	debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs);
875}
876
877void blk_mq_debugfs_register_hctx(struct request_queue *q,
878				  struct blk_mq_hw_ctx *hctx)
879{
880	struct blk_mq_ctx *ctx;
881	char name[20];
882	int i;
883
884	if (!q->debugfs_dir)
885		return;
886
887	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
888	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
889
890	debugfs_create_files(hctx->debugfs_dir, hctx, blk_mq_debugfs_hctx_attrs);
891
892	hctx_for_each_ctx(hctx, ctx, i)
893		blk_mq_debugfs_register_ctx(hctx, ctx);
894}
895
896void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
897{
898	debugfs_remove_recursive(hctx->debugfs_dir);
899	hctx->sched_debugfs_dir = NULL;
900	hctx->debugfs_dir = NULL;
901}
902
903void blk_mq_debugfs_register_hctxs(struct request_queue *q)
904{
905	struct blk_mq_hw_ctx *hctx;
906	int i;
907
908	queue_for_each_hw_ctx(q, hctx, i)
909		blk_mq_debugfs_register_hctx(q, hctx);
910}
911
912void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
913{
914	struct blk_mq_hw_ctx *hctx;
915	int i;
916
917	queue_for_each_hw_ctx(q, hctx, i)
918		blk_mq_debugfs_unregister_hctx(hctx);
919}
920
921void blk_mq_debugfs_register_sched(struct request_queue *q)
922{
923	struct elevator_type *e = q->elevator->type;
924
925	/*
926	 * If the parent directory has not been created yet, return, we will be
927	 * called again later on and the directory/files will be created then.
928	 */
929	if (!q->debugfs_dir)
930		return;
931
932	if (!e->queue_debugfs_attrs)
933		return;
934
935	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
936
937	debugfs_create_files(q->sched_debugfs_dir, q, e->queue_debugfs_attrs);
938}
939
940void blk_mq_debugfs_unregister_sched(struct request_queue *q)
941{
942	debugfs_remove_recursive(q->sched_debugfs_dir);
943	q->sched_debugfs_dir = NULL;
944}
945
946void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
947{
948	debugfs_remove_recursive(rqos->debugfs_dir);
949	rqos->debugfs_dir = NULL;
950}
951
952void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
953{
954	struct request_queue *q = rqos->q;
955	const char *dir_name = rq_qos_id_to_name(rqos->id);
956
957	if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
958		return;
959
960	if (!q->rqos_debugfs_dir)
961		q->rqos_debugfs_dir = debugfs_create_dir("rqos",
962							 q->debugfs_dir);
963
964	rqos->debugfs_dir = debugfs_create_dir(dir_name,
965					       rqos->q->rqos_debugfs_dir);
966
967	debugfs_create_files(rqos->debugfs_dir, rqos, rqos->ops->debugfs_attrs);
968}
969
970void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q)
971{
972	debugfs_remove_recursive(q->rqos_debugfs_dir);
973	q->rqos_debugfs_dir = NULL;
974}
975
976void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
977					struct blk_mq_hw_ctx *hctx)
978{
979	struct elevator_type *e = q->elevator->type;
980
981	if (!e->hctx_debugfs_attrs)
982		return;
983
984	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
985						     hctx->debugfs_dir);
986	debugfs_create_files(hctx->sched_debugfs_dir, hctx,
987			     e->hctx_debugfs_attrs);
988}
989
990void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
991{
992	debugfs_remove_recursive(hctx->sched_debugfs_dir);
993	hctx->sched_debugfs_dir = NULL;
994}
995