xref: /linux/block/blk-mq-debugfs.c (revision 856e7c4b619af622d56b3b454f7bec32a170ac99)
1 /*
2  * Copyright (C) 2017 Facebook
3  *
4  * This program is free software; you can redistribute it and/or
5  * modify it under the terms of the GNU General Public
6  * License v2 as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful,
9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  *
13  * You should have received a copy of the GNU General Public License
14  * along with this program.  If not, see <https://www.gnu.org/licenses/>.
15  */
16 
17 #include <linux/kernel.h>
18 #include <linux/blkdev.h>
19 #include <linux/debugfs.h>
20 
21 #include <linux/blk-mq.h>
22 #include "blk.h"
23 #include "blk-mq.h"
24 #include "blk-mq-debugfs.h"
25 #include "blk-mq-tag.h"
26 
27 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
28 {
29 	if (stat->nr_samples) {
30 		seq_printf(m, "samples=%d, mean=%lld, min=%llu, max=%llu",
31 			   stat->nr_samples, stat->mean, stat->min, stat->max);
32 	} else {
33 		seq_puts(m, "samples=0");
34 	}
35 }
36 
37 static int queue_poll_stat_show(void *data, struct seq_file *m)
38 {
39 	struct request_queue *q = data;
40 	int bucket;
41 
42 	for (bucket = 0; bucket < BLK_MQ_POLL_STATS_BKTS/2; bucket++) {
43 		seq_printf(m, "read  (%d Bytes): ", 1 << (9+bucket));
44 		print_stat(m, &q->poll_stat[2*bucket]);
45 		seq_puts(m, "\n");
46 
47 		seq_printf(m, "write (%d Bytes): ",  1 << (9+bucket));
48 		print_stat(m, &q->poll_stat[2*bucket+1]);
49 		seq_puts(m, "\n");
50 	}
51 	return 0;
52 }
53 
54 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
55 	__acquires(&q->requeue_lock)
56 {
57 	struct request_queue *q = m->private;
58 
59 	spin_lock_irq(&q->requeue_lock);
60 	return seq_list_start(&q->requeue_list, *pos);
61 }
62 
63 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
64 {
65 	struct request_queue *q = m->private;
66 
67 	return seq_list_next(v, &q->requeue_list, pos);
68 }
69 
70 static void queue_requeue_list_stop(struct seq_file *m, void *v)
71 	__releases(&q->requeue_lock)
72 {
73 	struct request_queue *q = m->private;
74 
75 	spin_unlock_irq(&q->requeue_lock);
76 }
77 
78 static const struct seq_operations queue_requeue_list_seq_ops = {
79 	.start	= queue_requeue_list_start,
80 	.next	= queue_requeue_list_next,
81 	.stop	= queue_requeue_list_stop,
82 	.show	= blk_mq_debugfs_rq_show,
83 };
84 
85 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
86 			  const char *const *flag_name, int flag_name_count)
87 {
88 	bool sep = false;
89 	int i;
90 
91 	for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
92 		if (!(flags & BIT(i)))
93 			continue;
94 		if (sep)
95 			seq_puts(m, "|");
96 		sep = true;
97 		if (i < flag_name_count && flag_name[i])
98 			seq_puts(m, flag_name[i]);
99 		else
100 			seq_printf(m, "%d", i);
101 	}
102 	return 0;
103 }
104 
105 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
106 static const char *const blk_queue_flag_name[] = {
107 	QUEUE_FLAG_NAME(QUEUED),
108 	QUEUE_FLAG_NAME(STOPPED),
109 	QUEUE_FLAG_NAME(DYING),
110 	QUEUE_FLAG_NAME(BYPASS),
111 	QUEUE_FLAG_NAME(BIDI),
112 	QUEUE_FLAG_NAME(NOMERGES),
113 	QUEUE_FLAG_NAME(SAME_COMP),
114 	QUEUE_FLAG_NAME(FAIL_IO),
115 	QUEUE_FLAG_NAME(NONROT),
116 	QUEUE_FLAG_NAME(IO_STAT),
117 	QUEUE_FLAG_NAME(DISCARD),
118 	QUEUE_FLAG_NAME(NOXMERGES),
119 	QUEUE_FLAG_NAME(ADD_RANDOM),
120 	QUEUE_FLAG_NAME(SECERASE),
121 	QUEUE_FLAG_NAME(SAME_FORCE),
122 	QUEUE_FLAG_NAME(DEAD),
123 	QUEUE_FLAG_NAME(INIT_DONE),
124 	QUEUE_FLAG_NAME(NO_SG_MERGE),
125 	QUEUE_FLAG_NAME(POLL),
126 	QUEUE_FLAG_NAME(WC),
127 	QUEUE_FLAG_NAME(FUA),
128 	QUEUE_FLAG_NAME(FLUSH_NQ),
129 	QUEUE_FLAG_NAME(DAX),
130 	QUEUE_FLAG_NAME(STATS),
131 	QUEUE_FLAG_NAME(POLL_STATS),
132 	QUEUE_FLAG_NAME(REGISTERED),
133 	QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
134 	QUEUE_FLAG_NAME(QUIESCED),
135 	QUEUE_FLAG_NAME(PREEMPT_ONLY),
136 };
137 #undef QUEUE_FLAG_NAME
138 
139 static int queue_state_show(void *data, struct seq_file *m)
140 {
141 	struct request_queue *q = data;
142 
143 	blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
144 		       ARRAY_SIZE(blk_queue_flag_name));
145 	seq_puts(m, "\n");
146 	return 0;
147 }
148 
149 static ssize_t queue_state_write(void *data, const char __user *buf,
150 				 size_t count, loff_t *ppos)
151 {
152 	struct request_queue *q = data;
153 	char opbuf[16] = { }, *op;
154 
155 	/*
156 	 * The "state" attribute is removed after blk_cleanup_queue() has called
157 	 * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
158 	 * triggering a use-after-free.
159 	 */
160 	if (blk_queue_dead(q))
161 		return -ENOENT;
162 
163 	if (count >= sizeof(opbuf)) {
164 		pr_err("%s: operation too long\n", __func__);
165 		goto inval;
166 	}
167 
168 	if (copy_from_user(opbuf, buf, count))
169 		return -EFAULT;
170 	op = strstrip(opbuf);
171 	if (strcmp(op, "run") == 0) {
172 		blk_mq_run_hw_queues(q, true);
173 	} else if (strcmp(op, "start") == 0) {
174 		blk_mq_start_stopped_hw_queues(q, true);
175 	} else if (strcmp(op, "kick") == 0) {
176 		blk_mq_kick_requeue_list(q);
177 	} else {
178 		pr_err("%s: unsupported operation '%s'\n", __func__, op);
179 inval:
180 		pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
181 		return -EINVAL;
182 	}
183 	return count;
184 }
185 
186 static int queue_write_hint_show(void *data, struct seq_file *m)
187 {
188 	struct request_queue *q = data;
189 	int i;
190 
191 	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
192 		seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
193 
194 	return 0;
195 }
196 
197 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
198 				      size_t count, loff_t *ppos)
199 {
200 	struct request_queue *q = data;
201 	int i;
202 
203 	for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
204 		q->write_hints[i] = 0;
205 
206 	return count;
207 }
208 
209 static int queue_zone_wlock_show(void *data, struct seq_file *m)
210 {
211 	struct request_queue *q = data;
212 	unsigned int i;
213 
214 	if (!q->seq_zones_wlock)
215 		return 0;
216 
217 	for (i = 0; i < blk_queue_nr_zones(q); i++)
218 		if (test_bit(i, q->seq_zones_wlock))
219 			seq_printf(m, "%u\n", i);
220 
221 	return 0;
222 }
223 
224 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
225 	{ "poll_stat", 0400, queue_poll_stat_show },
226 	{ "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
227 	{ "state", 0600, queue_state_show, queue_state_write },
228 	{ "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
229 	{ "zone_wlock", 0400, queue_zone_wlock_show, NULL },
230 	{ },
231 };
232 
233 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
234 static const char *const hctx_state_name[] = {
235 	HCTX_STATE_NAME(STOPPED),
236 	HCTX_STATE_NAME(TAG_ACTIVE),
237 	HCTX_STATE_NAME(SCHED_RESTART),
238 };
239 #undef HCTX_STATE_NAME
240 
241 static int hctx_state_show(void *data, struct seq_file *m)
242 {
243 	struct blk_mq_hw_ctx *hctx = data;
244 
245 	blk_flags_show(m, hctx->state, hctx_state_name,
246 		       ARRAY_SIZE(hctx_state_name));
247 	seq_puts(m, "\n");
248 	return 0;
249 }
250 
251 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
252 static const char *const alloc_policy_name[] = {
253 	BLK_TAG_ALLOC_NAME(FIFO),
254 	BLK_TAG_ALLOC_NAME(RR),
255 };
256 #undef BLK_TAG_ALLOC_NAME
257 
258 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
259 static const char *const hctx_flag_name[] = {
260 	HCTX_FLAG_NAME(SHOULD_MERGE),
261 	HCTX_FLAG_NAME(TAG_SHARED),
262 	HCTX_FLAG_NAME(SG_MERGE),
263 	HCTX_FLAG_NAME(BLOCKING),
264 	HCTX_FLAG_NAME(NO_SCHED),
265 };
266 #undef HCTX_FLAG_NAME
267 
268 static int hctx_flags_show(void *data, struct seq_file *m)
269 {
270 	struct blk_mq_hw_ctx *hctx = data;
271 	const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
272 
273 	seq_puts(m, "alloc_policy=");
274 	if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
275 	    alloc_policy_name[alloc_policy])
276 		seq_puts(m, alloc_policy_name[alloc_policy]);
277 	else
278 		seq_printf(m, "%d", alloc_policy);
279 	seq_puts(m, " ");
280 	blk_flags_show(m,
281 		       hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
282 		       hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
283 	seq_puts(m, "\n");
284 	return 0;
285 }
286 
287 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
288 static const char *const op_name[] = {
289 	REQ_OP_NAME(READ),
290 	REQ_OP_NAME(WRITE),
291 	REQ_OP_NAME(FLUSH),
292 	REQ_OP_NAME(DISCARD),
293 	REQ_OP_NAME(ZONE_REPORT),
294 	REQ_OP_NAME(SECURE_ERASE),
295 	REQ_OP_NAME(ZONE_RESET),
296 	REQ_OP_NAME(WRITE_SAME),
297 	REQ_OP_NAME(WRITE_ZEROES),
298 	REQ_OP_NAME(SCSI_IN),
299 	REQ_OP_NAME(SCSI_OUT),
300 	REQ_OP_NAME(DRV_IN),
301 	REQ_OP_NAME(DRV_OUT),
302 };
303 #undef REQ_OP_NAME
304 
305 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
306 static const char *const cmd_flag_name[] = {
307 	CMD_FLAG_NAME(FAILFAST_DEV),
308 	CMD_FLAG_NAME(FAILFAST_TRANSPORT),
309 	CMD_FLAG_NAME(FAILFAST_DRIVER),
310 	CMD_FLAG_NAME(SYNC),
311 	CMD_FLAG_NAME(META),
312 	CMD_FLAG_NAME(PRIO),
313 	CMD_FLAG_NAME(NOMERGE),
314 	CMD_FLAG_NAME(IDLE),
315 	CMD_FLAG_NAME(INTEGRITY),
316 	CMD_FLAG_NAME(FUA),
317 	CMD_FLAG_NAME(PREFLUSH),
318 	CMD_FLAG_NAME(RAHEAD),
319 	CMD_FLAG_NAME(BACKGROUND),
320 	CMD_FLAG_NAME(NOUNMAP),
321 	CMD_FLAG_NAME(NOWAIT),
322 };
323 #undef CMD_FLAG_NAME
324 
325 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
326 static const char *const rqf_name[] = {
327 	RQF_NAME(SORTED),
328 	RQF_NAME(STARTED),
329 	RQF_NAME(QUEUED),
330 	RQF_NAME(SOFTBARRIER),
331 	RQF_NAME(FLUSH_SEQ),
332 	RQF_NAME(MIXED_MERGE),
333 	RQF_NAME(MQ_INFLIGHT),
334 	RQF_NAME(DONTPREP),
335 	RQF_NAME(PREEMPT),
336 	RQF_NAME(COPY_USER),
337 	RQF_NAME(FAILED),
338 	RQF_NAME(QUIET),
339 	RQF_NAME(ELVPRIV),
340 	RQF_NAME(IO_STAT),
341 	RQF_NAME(ALLOCED),
342 	RQF_NAME(PM),
343 	RQF_NAME(HASHED),
344 	RQF_NAME(STATS),
345 	RQF_NAME(SPECIAL_PAYLOAD),
346 	RQF_NAME(ZONE_WRITE_LOCKED),
347 	RQF_NAME(MQ_POLL_SLEPT),
348 };
349 #undef RQF_NAME
350 
351 static const char *const blk_mq_rq_state_name_array[] = {
352 	[MQ_RQ_IDLE]		= "idle",
353 	[MQ_RQ_IN_FLIGHT]	= "in_flight",
354 	[MQ_RQ_COMPLETE]	= "complete",
355 };
356 
357 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
358 {
359 	if (WARN_ON_ONCE((unsigned int)rq_state >
360 			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
361 		return "(?)";
362 	return blk_mq_rq_state_name_array[rq_state];
363 }
364 
365 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
366 {
367 	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
368 	const unsigned int op = rq->cmd_flags & REQ_OP_MASK;
369 
370 	seq_printf(m, "%p {.op=", rq);
371 	if (op < ARRAY_SIZE(op_name) && op_name[op])
372 		seq_printf(m, "%s", op_name[op]);
373 	else
374 		seq_printf(m, "%d", op);
375 	seq_puts(m, ", .cmd_flags=");
376 	blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
377 		       ARRAY_SIZE(cmd_flag_name));
378 	seq_puts(m, ", .rq_flags=");
379 	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
380 		       ARRAY_SIZE(rqf_name));
381 	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
382 	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
383 		   rq->internal_tag);
384 	if (mq_ops->show_rq)
385 		mq_ops->show_rq(m, rq);
386 	seq_puts(m, "}\n");
387 	return 0;
388 }
389 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
390 
391 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
392 {
393 	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
394 }
395 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
396 
397 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
398 	__acquires(&hctx->lock)
399 {
400 	struct blk_mq_hw_ctx *hctx = m->private;
401 
402 	spin_lock(&hctx->lock);
403 	return seq_list_start(&hctx->dispatch, *pos);
404 }
405 
406 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
407 {
408 	struct blk_mq_hw_ctx *hctx = m->private;
409 
410 	return seq_list_next(v, &hctx->dispatch, pos);
411 }
412 
413 static void hctx_dispatch_stop(struct seq_file *m, void *v)
414 	__releases(&hctx->lock)
415 {
416 	struct blk_mq_hw_ctx *hctx = m->private;
417 
418 	spin_unlock(&hctx->lock);
419 }
420 
421 static const struct seq_operations hctx_dispatch_seq_ops = {
422 	.start	= hctx_dispatch_start,
423 	.next	= hctx_dispatch_next,
424 	.stop	= hctx_dispatch_stop,
425 	.show	= blk_mq_debugfs_rq_show,
426 };
427 
428 struct show_busy_params {
429 	struct seq_file		*m;
430 	struct blk_mq_hw_ctx	*hctx;
431 };
432 
433 /*
434  * Note: the state of a request may change while this function is in progress,
435  * e.g. due to a concurrent blk_mq_finish_request() call.
436  */
437 static void hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
438 {
439 	const struct show_busy_params *params = data;
440 
441 	if (blk_mq_map_queue(rq->q, rq->mq_ctx->cpu) == params->hctx &&
442 	    blk_mq_rq_state(rq) != MQ_RQ_IDLE)
443 		__blk_mq_debugfs_rq_show(params->m,
444 					 list_entry_rq(&rq->queuelist));
445 }
446 
447 static int hctx_busy_show(void *data, struct seq_file *m)
448 {
449 	struct blk_mq_hw_ctx *hctx = data;
450 	struct show_busy_params params = { .m = m, .hctx = hctx };
451 
452 	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
453 				&params);
454 
455 	return 0;
456 }
457 
458 static int hctx_ctx_map_show(void *data, struct seq_file *m)
459 {
460 	struct blk_mq_hw_ctx *hctx = data;
461 
462 	sbitmap_bitmap_show(&hctx->ctx_map, m);
463 	return 0;
464 }
465 
466 static void blk_mq_debugfs_tags_show(struct seq_file *m,
467 				     struct blk_mq_tags *tags)
468 {
469 	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
470 	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
471 	seq_printf(m, "active_queues=%d\n",
472 		   atomic_read(&tags->active_queues));
473 
474 	seq_puts(m, "\nbitmap_tags:\n");
475 	sbitmap_queue_show(&tags->bitmap_tags, m);
476 
477 	if (tags->nr_reserved_tags) {
478 		seq_puts(m, "\nbreserved_tags:\n");
479 		sbitmap_queue_show(&tags->breserved_tags, m);
480 	}
481 }
482 
483 static int hctx_tags_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 		blk_mq_debugfs_tags_show(m, hctx->tags);
494 	mutex_unlock(&q->sysfs_lock);
495 
496 out:
497 	return res;
498 }
499 
500 static int hctx_tags_bitmap_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->tags)
510 		sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
511 	mutex_unlock(&q->sysfs_lock);
512 
513 out:
514 	return res;
515 }
516 
517 static int hctx_sched_tags_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 		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
528 	mutex_unlock(&q->sysfs_lock);
529 
530 out:
531 	return res;
532 }
533 
534 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
535 {
536 	struct blk_mq_hw_ctx *hctx = data;
537 	struct request_queue *q = hctx->queue;
538 	int res;
539 
540 	res = mutex_lock_interruptible(&q->sysfs_lock);
541 	if (res)
542 		goto out;
543 	if (hctx->sched_tags)
544 		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
545 	mutex_unlock(&q->sysfs_lock);
546 
547 out:
548 	return res;
549 }
550 
551 static int hctx_io_poll_show(void *data, struct seq_file *m)
552 {
553 	struct blk_mq_hw_ctx *hctx = data;
554 
555 	seq_printf(m, "considered=%lu\n", hctx->poll_considered);
556 	seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
557 	seq_printf(m, "success=%lu\n", hctx->poll_success);
558 	return 0;
559 }
560 
561 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
562 				  size_t count, loff_t *ppos)
563 {
564 	struct blk_mq_hw_ctx *hctx = data;
565 
566 	hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
567 	return count;
568 }
569 
570 static int hctx_dispatched_show(void *data, struct seq_file *m)
571 {
572 	struct blk_mq_hw_ctx *hctx = data;
573 	int i;
574 
575 	seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
576 
577 	for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
578 		unsigned int d = 1U << (i - 1);
579 
580 		seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
581 	}
582 
583 	seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
584 	return 0;
585 }
586 
587 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
588 				     size_t count, loff_t *ppos)
589 {
590 	struct blk_mq_hw_ctx *hctx = data;
591 	int i;
592 
593 	for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
594 		hctx->dispatched[i] = 0;
595 	return count;
596 }
597 
598 static int hctx_queued_show(void *data, struct seq_file *m)
599 {
600 	struct blk_mq_hw_ctx *hctx = data;
601 
602 	seq_printf(m, "%lu\n", hctx->queued);
603 	return 0;
604 }
605 
606 static ssize_t hctx_queued_write(void *data, const char __user *buf,
607 				 size_t count, loff_t *ppos)
608 {
609 	struct blk_mq_hw_ctx *hctx = data;
610 
611 	hctx->queued = 0;
612 	return count;
613 }
614 
615 static int hctx_run_show(void *data, struct seq_file *m)
616 {
617 	struct blk_mq_hw_ctx *hctx = data;
618 
619 	seq_printf(m, "%lu\n", hctx->run);
620 	return 0;
621 }
622 
623 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
624 			      loff_t *ppos)
625 {
626 	struct blk_mq_hw_ctx *hctx = data;
627 
628 	hctx->run = 0;
629 	return count;
630 }
631 
632 static int hctx_active_show(void *data, struct seq_file *m)
633 {
634 	struct blk_mq_hw_ctx *hctx = data;
635 
636 	seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
637 	return 0;
638 }
639 
640 static void *ctx_rq_list_start(struct seq_file *m, loff_t *pos)
641 	__acquires(&ctx->lock)
642 {
643 	struct blk_mq_ctx *ctx = m->private;
644 
645 	spin_lock(&ctx->lock);
646 	return seq_list_start(&ctx->rq_list, *pos);
647 }
648 
649 static void *ctx_rq_list_next(struct seq_file *m, void *v, loff_t *pos)
650 {
651 	struct blk_mq_ctx *ctx = m->private;
652 
653 	return seq_list_next(v, &ctx->rq_list, pos);
654 }
655 
656 static void ctx_rq_list_stop(struct seq_file *m, void *v)
657 	__releases(&ctx->lock)
658 {
659 	struct blk_mq_ctx *ctx = m->private;
660 
661 	spin_unlock(&ctx->lock);
662 }
663 
664 static const struct seq_operations ctx_rq_list_seq_ops = {
665 	.start	= ctx_rq_list_start,
666 	.next	= ctx_rq_list_next,
667 	.stop	= ctx_rq_list_stop,
668 	.show	= blk_mq_debugfs_rq_show,
669 };
670 static int ctx_dispatched_show(void *data, struct seq_file *m)
671 {
672 	struct blk_mq_ctx *ctx = data;
673 
674 	seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
675 	return 0;
676 }
677 
678 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
679 				    size_t count, loff_t *ppos)
680 {
681 	struct blk_mq_ctx *ctx = data;
682 
683 	ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
684 	return count;
685 }
686 
687 static int ctx_merged_show(void *data, struct seq_file *m)
688 {
689 	struct blk_mq_ctx *ctx = data;
690 
691 	seq_printf(m, "%lu\n", ctx->rq_merged);
692 	return 0;
693 }
694 
695 static ssize_t ctx_merged_write(void *data, const char __user *buf,
696 				size_t count, loff_t *ppos)
697 {
698 	struct blk_mq_ctx *ctx = data;
699 
700 	ctx->rq_merged = 0;
701 	return count;
702 }
703 
704 static int ctx_completed_show(void *data, struct seq_file *m)
705 {
706 	struct blk_mq_ctx *ctx = data;
707 
708 	seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
709 	return 0;
710 }
711 
712 static ssize_t ctx_completed_write(void *data, const char __user *buf,
713 				   size_t count, loff_t *ppos)
714 {
715 	struct blk_mq_ctx *ctx = data;
716 
717 	ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
718 	return count;
719 }
720 
721 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
722 {
723 	const struct blk_mq_debugfs_attr *attr = m->private;
724 	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
725 
726 	return attr->show(data, m);
727 }
728 
729 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
730 				    size_t count, loff_t *ppos)
731 {
732 	struct seq_file *m = file->private_data;
733 	const struct blk_mq_debugfs_attr *attr = m->private;
734 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
735 
736 	/*
737 	 * Attributes that only implement .seq_ops are read-only and 'attr' is
738 	 * the same with 'data' in this case.
739 	 */
740 	if (attr == data || !attr->write)
741 		return -EPERM;
742 
743 	return attr->write(data, buf, count, ppos);
744 }
745 
746 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
747 {
748 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
749 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
750 	struct seq_file *m;
751 	int ret;
752 
753 	if (attr->seq_ops) {
754 		ret = seq_open(file, attr->seq_ops);
755 		if (!ret) {
756 			m = file->private_data;
757 			m->private = data;
758 		}
759 		return ret;
760 	}
761 
762 	if (WARN_ON_ONCE(!attr->show))
763 		return -EPERM;
764 
765 	return single_open(file, blk_mq_debugfs_show, inode->i_private);
766 }
767 
768 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
769 {
770 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
771 
772 	if (attr->show)
773 		return single_release(inode, file);
774 	else
775 		return seq_release(inode, file);
776 }
777 
778 static const struct file_operations blk_mq_debugfs_fops = {
779 	.open		= blk_mq_debugfs_open,
780 	.read		= seq_read,
781 	.write		= blk_mq_debugfs_write,
782 	.llseek		= seq_lseek,
783 	.release	= blk_mq_debugfs_release,
784 };
785 
786 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
787 	{"state", 0400, hctx_state_show},
788 	{"flags", 0400, hctx_flags_show},
789 	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
790 	{"busy", 0400, hctx_busy_show},
791 	{"ctx_map", 0400, hctx_ctx_map_show},
792 	{"tags", 0400, hctx_tags_show},
793 	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
794 	{"sched_tags", 0400, hctx_sched_tags_show},
795 	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
796 	{"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
797 	{"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
798 	{"queued", 0600, hctx_queued_show, hctx_queued_write},
799 	{"run", 0600, hctx_run_show, hctx_run_write},
800 	{"active", 0400, hctx_active_show},
801 	{},
802 };
803 
804 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
805 	{"rq_list", 0400, .seq_ops = &ctx_rq_list_seq_ops},
806 	{"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
807 	{"merged", 0600, ctx_merged_show, ctx_merged_write},
808 	{"completed", 0600, ctx_completed_show, ctx_completed_write},
809 	{},
810 };
811 
812 static bool debugfs_create_files(struct dentry *parent, void *data,
813 				 const struct blk_mq_debugfs_attr *attr)
814 {
815 	d_inode(parent)->i_private = data;
816 
817 	for (; attr->name; attr++) {
818 		if (!debugfs_create_file(attr->name, attr->mode, parent,
819 					 (void *)attr, &blk_mq_debugfs_fops))
820 			return false;
821 	}
822 	return true;
823 }
824 
825 int blk_mq_debugfs_register(struct request_queue *q)
826 {
827 	struct blk_mq_hw_ctx *hctx;
828 	int i;
829 
830 	if (!blk_debugfs_root)
831 		return -ENOENT;
832 
833 	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
834 					    blk_debugfs_root);
835 	if (!q->debugfs_dir)
836 		return -ENOMEM;
837 
838 	if (!debugfs_create_files(q->debugfs_dir, q,
839 				  blk_mq_debugfs_queue_attrs))
840 		goto err;
841 
842 	/*
843 	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
844 	 * didn't exist yet (because we don't know what to name the directory
845 	 * until the queue is registered to a gendisk).
846 	 */
847 	if (q->elevator && !q->sched_debugfs_dir)
848 		blk_mq_debugfs_register_sched(q);
849 
850 	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
851 	queue_for_each_hw_ctx(q, hctx, i) {
852 		if (!hctx->debugfs_dir && blk_mq_debugfs_register_hctx(q, hctx))
853 			goto err;
854 		if (q->elevator && !hctx->sched_debugfs_dir &&
855 		    blk_mq_debugfs_register_sched_hctx(q, hctx))
856 			goto err;
857 	}
858 
859 	return 0;
860 
861 err:
862 	blk_mq_debugfs_unregister(q);
863 	return -ENOMEM;
864 }
865 
866 void blk_mq_debugfs_unregister(struct request_queue *q)
867 {
868 	debugfs_remove_recursive(q->debugfs_dir);
869 	q->sched_debugfs_dir = NULL;
870 	q->debugfs_dir = NULL;
871 }
872 
873 static int blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
874 				       struct blk_mq_ctx *ctx)
875 {
876 	struct dentry *ctx_dir;
877 	char name[20];
878 
879 	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
880 	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
881 	if (!ctx_dir)
882 		return -ENOMEM;
883 
884 	if (!debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs))
885 		return -ENOMEM;
886 
887 	return 0;
888 }
889 
890 int blk_mq_debugfs_register_hctx(struct request_queue *q,
891 				 struct blk_mq_hw_ctx *hctx)
892 {
893 	struct blk_mq_ctx *ctx;
894 	char name[20];
895 	int i;
896 
897 	if (!q->debugfs_dir)
898 		return -ENOENT;
899 
900 	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
901 	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
902 	if (!hctx->debugfs_dir)
903 		return -ENOMEM;
904 
905 	if (!debugfs_create_files(hctx->debugfs_dir, hctx,
906 				  blk_mq_debugfs_hctx_attrs))
907 		goto err;
908 
909 	hctx_for_each_ctx(hctx, ctx, i) {
910 		if (blk_mq_debugfs_register_ctx(hctx, ctx))
911 			goto err;
912 	}
913 
914 	return 0;
915 
916 err:
917 	blk_mq_debugfs_unregister_hctx(hctx);
918 	return -ENOMEM;
919 }
920 
921 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
922 {
923 	debugfs_remove_recursive(hctx->debugfs_dir);
924 	hctx->sched_debugfs_dir = NULL;
925 	hctx->debugfs_dir = NULL;
926 }
927 
928 int blk_mq_debugfs_register_hctxs(struct request_queue *q)
929 {
930 	struct blk_mq_hw_ctx *hctx;
931 	int i;
932 
933 	queue_for_each_hw_ctx(q, hctx, i) {
934 		if (blk_mq_debugfs_register_hctx(q, hctx))
935 			return -ENOMEM;
936 	}
937 
938 	return 0;
939 }
940 
941 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
942 {
943 	struct blk_mq_hw_ctx *hctx;
944 	int i;
945 
946 	queue_for_each_hw_ctx(q, hctx, i)
947 		blk_mq_debugfs_unregister_hctx(hctx);
948 }
949 
950 int blk_mq_debugfs_register_sched(struct request_queue *q)
951 {
952 	struct elevator_type *e = q->elevator->type;
953 
954 	if (!q->debugfs_dir)
955 		return -ENOENT;
956 
957 	if (!e->queue_debugfs_attrs)
958 		return 0;
959 
960 	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
961 	if (!q->sched_debugfs_dir)
962 		return -ENOMEM;
963 
964 	if (!debugfs_create_files(q->sched_debugfs_dir, q,
965 				  e->queue_debugfs_attrs))
966 		goto err;
967 
968 	return 0;
969 
970 err:
971 	blk_mq_debugfs_unregister_sched(q);
972 	return -ENOMEM;
973 }
974 
975 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
976 {
977 	debugfs_remove_recursive(q->sched_debugfs_dir);
978 	q->sched_debugfs_dir = NULL;
979 }
980 
981 int blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
982 				       struct blk_mq_hw_ctx *hctx)
983 {
984 	struct elevator_type *e = q->elevator->type;
985 
986 	if (!hctx->debugfs_dir)
987 		return -ENOENT;
988 
989 	if (!e->hctx_debugfs_attrs)
990 		return 0;
991 
992 	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
993 						     hctx->debugfs_dir);
994 	if (!hctx->sched_debugfs_dir)
995 		return -ENOMEM;
996 
997 	if (!debugfs_create_files(hctx->sched_debugfs_dir, hctx,
998 				  e->hctx_debugfs_attrs))
999 		return -ENOMEM;
1000 
1001 	return 0;
1002 }
1003 
1004 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1005 {
1006 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
1007 	hctx->sched_debugfs_dir = NULL;
1008 }
1009