xref: /openbmc/linux/block/blk-mq-debugfs.c (revision bdeeed09)
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_TIMEOUT_EXPIRED),
348 	RQF_NAME(MQ_POLL_SLEPT),
349 };
350 #undef RQF_NAME
351 
352 static const char *const blk_mq_rq_state_name_array[] = {
353 	[MQ_RQ_IDLE]		= "idle",
354 	[MQ_RQ_IN_FLIGHT]	= "in_flight",
355 	[MQ_RQ_COMPLETE]	= "complete",
356 };
357 
358 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
359 {
360 	if (WARN_ON_ONCE((unsigned int)rq_state >
361 			 ARRAY_SIZE(blk_mq_rq_state_name_array)))
362 		return "(?)";
363 	return blk_mq_rq_state_name_array[rq_state];
364 }
365 
366 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
367 {
368 	const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
369 	const unsigned int op = rq->cmd_flags & REQ_OP_MASK;
370 
371 	seq_printf(m, "%p {.op=", rq);
372 	if (op < ARRAY_SIZE(op_name) && op_name[op])
373 		seq_printf(m, "%s", op_name[op]);
374 	else
375 		seq_printf(m, "%d", op);
376 	seq_puts(m, ", .cmd_flags=");
377 	blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
378 		       ARRAY_SIZE(cmd_flag_name));
379 	seq_puts(m, ", .rq_flags=");
380 	blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
381 		       ARRAY_SIZE(rqf_name));
382 	seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
383 	seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
384 		   rq->internal_tag);
385 	if (mq_ops->show_rq)
386 		mq_ops->show_rq(m, rq);
387 	seq_puts(m, "}\n");
388 	return 0;
389 }
390 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
391 
392 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
393 {
394 	return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
395 }
396 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
397 
398 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
399 	__acquires(&hctx->lock)
400 {
401 	struct blk_mq_hw_ctx *hctx = m->private;
402 
403 	spin_lock(&hctx->lock);
404 	return seq_list_start(&hctx->dispatch, *pos);
405 }
406 
407 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
408 {
409 	struct blk_mq_hw_ctx *hctx = m->private;
410 
411 	return seq_list_next(v, &hctx->dispatch, pos);
412 }
413 
414 static void hctx_dispatch_stop(struct seq_file *m, void *v)
415 	__releases(&hctx->lock)
416 {
417 	struct blk_mq_hw_ctx *hctx = m->private;
418 
419 	spin_unlock(&hctx->lock);
420 }
421 
422 static const struct seq_operations hctx_dispatch_seq_ops = {
423 	.start	= hctx_dispatch_start,
424 	.next	= hctx_dispatch_next,
425 	.stop	= hctx_dispatch_stop,
426 	.show	= blk_mq_debugfs_rq_show,
427 };
428 
429 struct show_busy_params {
430 	struct seq_file		*m;
431 	struct blk_mq_hw_ctx	*hctx;
432 };
433 
434 /*
435  * Note: the state of a request may change while this function is in progress,
436  * e.g. due to a concurrent blk_mq_finish_request() call.
437  */
438 static void hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
439 {
440 	const struct show_busy_params *params = data;
441 
442 	if (blk_mq_map_queue(rq->q, rq->mq_ctx->cpu) == params->hctx &&
443 	    blk_mq_rq_state(rq) != MQ_RQ_IDLE)
444 		__blk_mq_debugfs_rq_show(params->m,
445 					 list_entry_rq(&rq->queuelist));
446 }
447 
448 static int hctx_busy_show(void *data, struct seq_file *m)
449 {
450 	struct blk_mq_hw_ctx *hctx = data;
451 	struct show_busy_params params = { .m = m, .hctx = hctx };
452 
453 	blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
454 				&params);
455 
456 	return 0;
457 }
458 
459 static int hctx_ctx_map_show(void *data, struct seq_file *m)
460 {
461 	struct blk_mq_hw_ctx *hctx = data;
462 
463 	sbitmap_bitmap_show(&hctx->ctx_map, m);
464 	return 0;
465 }
466 
467 static void blk_mq_debugfs_tags_show(struct seq_file *m,
468 				     struct blk_mq_tags *tags)
469 {
470 	seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
471 	seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
472 	seq_printf(m, "active_queues=%d\n",
473 		   atomic_read(&tags->active_queues));
474 
475 	seq_puts(m, "\nbitmap_tags:\n");
476 	sbitmap_queue_show(&tags->bitmap_tags, m);
477 
478 	if (tags->nr_reserved_tags) {
479 		seq_puts(m, "\nbreserved_tags:\n");
480 		sbitmap_queue_show(&tags->breserved_tags, m);
481 	}
482 }
483 
484 static int hctx_tags_show(void *data, struct seq_file *m)
485 {
486 	struct blk_mq_hw_ctx *hctx = data;
487 	struct request_queue *q = hctx->queue;
488 	int res;
489 
490 	res = mutex_lock_interruptible(&q->sysfs_lock);
491 	if (res)
492 		goto out;
493 	if (hctx->tags)
494 		blk_mq_debugfs_tags_show(m, hctx->tags);
495 	mutex_unlock(&q->sysfs_lock);
496 
497 out:
498 	return res;
499 }
500 
501 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
502 {
503 	struct blk_mq_hw_ctx *hctx = data;
504 	struct request_queue *q = hctx->queue;
505 	int res;
506 
507 	res = mutex_lock_interruptible(&q->sysfs_lock);
508 	if (res)
509 		goto out;
510 	if (hctx->tags)
511 		sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
512 	mutex_unlock(&q->sysfs_lock);
513 
514 out:
515 	return res;
516 }
517 
518 static int hctx_sched_tags_show(void *data, struct seq_file *m)
519 {
520 	struct blk_mq_hw_ctx *hctx = data;
521 	struct request_queue *q = hctx->queue;
522 	int res;
523 
524 	res = mutex_lock_interruptible(&q->sysfs_lock);
525 	if (res)
526 		goto out;
527 	if (hctx->sched_tags)
528 		blk_mq_debugfs_tags_show(m, hctx->sched_tags);
529 	mutex_unlock(&q->sysfs_lock);
530 
531 out:
532 	return res;
533 }
534 
535 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
536 {
537 	struct blk_mq_hw_ctx *hctx = data;
538 	struct request_queue *q = hctx->queue;
539 	int res;
540 
541 	res = mutex_lock_interruptible(&q->sysfs_lock);
542 	if (res)
543 		goto out;
544 	if (hctx->sched_tags)
545 		sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
546 	mutex_unlock(&q->sysfs_lock);
547 
548 out:
549 	return res;
550 }
551 
552 static int hctx_io_poll_show(void *data, struct seq_file *m)
553 {
554 	struct blk_mq_hw_ctx *hctx = data;
555 
556 	seq_printf(m, "considered=%lu\n", hctx->poll_considered);
557 	seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
558 	seq_printf(m, "success=%lu\n", hctx->poll_success);
559 	return 0;
560 }
561 
562 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
563 				  size_t count, loff_t *ppos)
564 {
565 	struct blk_mq_hw_ctx *hctx = data;
566 
567 	hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
568 	return count;
569 }
570 
571 static int hctx_dispatched_show(void *data, struct seq_file *m)
572 {
573 	struct blk_mq_hw_ctx *hctx = data;
574 	int i;
575 
576 	seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
577 
578 	for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
579 		unsigned int d = 1U << (i - 1);
580 
581 		seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
582 	}
583 
584 	seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
585 	return 0;
586 }
587 
588 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
589 				     size_t count, loff_t *ppos)
590 {
591 	struct blk_mq_hw_ctx *hctx = data;
592 	int i;
593 
594 	for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
595 		hctx->dispatched[i] = 0;
596 	return count;
597 }
598 
599 static int hctx_queued_show(void *data, struct seq_file *m)
600 {
601 	struct blk_mq_hw_ctx *hctx = data;
602 
603 	seq_printf(m, "%lu\n", hctx->queued);
604 	return 0;
605 }
606 
607 static ssize_t hctx_queued_write(void *data, const char __user *buf,
608 				 size_t count, loff_t *ppos)
609 {
610 	struct blk_mq_hw_ctx *hctx = data;
611 
612 	hctx->queued = 0;
613 	return count;
614 }
615 
616 static int hctx_run_show(void *data, struct seq_file *m)
617 {
618 	struct blk_mq_hw_ctx *hctx = data;
619 
620 	seq_printf(m, "%lu\n", hctx->run);
621 	return 0;
622 }
623 
624 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
625 			      loff_t *ppos)
626 {
627 	struct blk_mq_hw_ctx *hctx = data;
628 
629 	hctx->run = 0;
630 	return count;
631 }
632 
633 static int hctx_active_show(void *data, struct seq_file *m)
634 {
635 	struct blk_mq_hw_ctx *hctx = data;
636 
637 	seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
638 	return 0;
639 }
640 
641 static void *ctx_rq_list_start(struct seq_file *m, loff_t *pos)
642 	__acquires(&ctx->lock)
643 {
644 	struct blk_mq_ctx *ctx = m->private;
645 
646 	spin_lock(&ctx->lock);
647 	return seq_list_start(&ctx->rq_list, *pos);
648 }
649 
650 static void *ctx_rq_list_next(struct seq_file *m, void *v, loff_t *pos)
651 {
652 	struct blk_mq_ctx *ctx = m->private;
653 
654 	return seq_list_next(v, &ctx->rq_list, pos);
655 }
656 
657 static void ctx_rq_list_stop(struct seq_file *m, void *v)
658 	__releases(&ctx->lock)
659 {
660 	struct blk_mq_ctx *ctx = m->private;
661 
662 	spin_unlock(&ctx->lock);
663 }
664 
665 static const struct seq_operations ctx_rq_list_seq_ops = {
666 	.start	= ctx_rq_list_start,
667 	.next	= ctx_rq_list_next,
668 	.stop	= ctx_rq_list_stop,
669 	.show	= blk_mq_debugfs_rq_show,
670 };
671 static int ctx_dispatched_show(void *data, struct seq_file *m)
672 {
673 	struct blk_mq_ctx *ctx = data;
674 
675 	seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
676 	return 0;
677 }
678 
679 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
680 				    size_t count, loff_t *ppos)
681 {
682 	struct blk_mq_ctx *ctx = data;
683 
684 	ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
685 	return count;
686 }
687 
688 static int ctx_merged_show(void *data, struct seq_file *m)
689 {
690 	struct blk_mq_ctx *ctx = data;
691 
692 	seq_printf(m, "%lu\n", ctx->rq_merged);
693 	return 0;
694 }
695 
696 static ssize_t ctx_merged_write(void *data, const char __user *buf,
697 				size_t count, loff_t *ppos)
698 {
699 	struct blk_mq_ctx *ctx = data;
700 
701 	ctx->rq_merged = 0;
702 	return count;
703 }
704 
705 static int ctx_completed_show(void *data, struct seq_file *m)
706 {
707 	struct blk_mq_ctx *ctx = data;
708 
709 	seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
710 	return 0;
711 }
712 
713 static ssize_t ctx_completed_write(void *data, const char __user *buf,
714 				   size_t count, loff_t *ppos)
715 {
716 	struct blk_mq_ctx *ctx = data;
717 
718 	ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
719 	return count;
720 }
721 
722 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
723 {
724 	const struct blk_mq_debugfs_attr *attr = m->private;
725 	void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
726 
727 	return attr->show(data, m);
728 }
729 
730 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
731 				    size_t count, loff_t *ppos)
732 {
733 	struct seq_file *m = file->private_data;
734 	const struct blk_mq_debugfs_attr *attr = m->private;
735 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
736 
737 	/*
738 	 * Attributes that only implement .seq_ops are read-only and 'attr' is
739 	 * the same with 'data' in this case.
740 	 */
741 	if (attr == data || !attr->write)
742 		return -EPERM;
743 
744 	return attr->write(data, buf, count, ppos);
745 }
746 
747 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
748 {
749 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
750 	void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
751 	struct seq_file *m;
752 	int ret;
753 
754 	if (attr->seq_ops) {
755 		ret = seq_open(file, attr->seq_ops);
756 		if (!ret) {
757 			m = file->private_data;
758 			m->private = data;
759 		}
760 		return ret;
761 	}
762 
763 	if (WARN_ON_ONCE(!attr->show))
764 		return -EPERM;
765 
766 	return single_open(file, blk_mq_debugfs_show, inode->i_private);
767 }
768 
769 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
770 {
771 	const struct blk_mq_debugfs_attr *attr = inode->i_private;
772 
773 	if (attr->show)
774 		return single_release(inode, file);
775 	else
776 		return seq_release(inode, file);
777 }
778 
779 static const struct file_operations blk_mq_debugfs_fops = {
780 	.open		= blk_mq_debugfs_open,
781 	.read		= seq_read,
782 	.write		= blk_mq_debugfs_write,
783 	.llseek		= seq_lseek,
784 	.release	= blk_mq_debugfs_release,
785 };
786 
787 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
788 	{"state", 0400, hctx_state_show},
789 	{"flags", 0400, hctx_flags_show},
790 	{"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
791 	{"busy", 0400, hctx_busy_show},
792 	{"ctx_map", 0400, hctx_ctx_map_show},
793 	{"tags", 0400, hctx_tags_show},
794 	{"tags_bitmap", 0400, hctx_tags_bitmap_show},
795 	{"sched_tags", 0400, hctx_sched_tags_show},
796 	{"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
797 	{"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
798 	{"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
799 	{"queued", 0600, hctx_queued_show, hctx_queued_write},
800 	{"run", 0600, hctx_run_show, hctx_run_write},
801 	{"active", 0400, hctx_active_show},
802 	{},
803 };
804 
805 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
806 	{"rq_list", 0400, .seq_ops = &ctx_rq_list_seq_ops},
807 	{"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
808 	{"merged", 0600, ctx_merged_show, ctx_merged_write},
809 	{"completed", 0600, ctx_completed_show, ctx_completed_write},
810 	{},
811 };
812 
813 static bool debugfs_create_files(struct dentry *parent, void *data,
814 				 const struct blk_mq_debugfs_attr *attr)
815 {
816 	d_inode(parent)->i_private = data;
817 
818 	for (; attr->name; attr++) {
819 		if (!debugfs_create_file(attr->name, attr->mode, parent,
820 					 (void *)attr, &blk_mq_debugfs_fops))
821 			return false;
822 	}
823 	return true;
824 }
825 
826 int blk_mq_debugfs_register(struct request_queue *q)
827 {
828 	struct blk_mq_hw_ctx *hctx;
829 	int i;
830 
831 	if (!blk_debugfs_root)
832 		return -ENOENT;
833 
834 	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
835 					    blk_debugfs_root);
836 	if (!q->debugfs_dir)
837 		return -ENOMEM;
838 
839 	if (!debugfs_create_files(q->debugfs_dir, q,
840 				  blk_mq_debugfs_queue_attrs))
841 		goto err;
842 
843 	/*
844 	 * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
845 	 * didn't exist yet (because we don't know what to name the directory
846 	 * until the queue is registered to a gendisk).
847 	 */
848 	if (q->elevator && !q->sched_debugfs_dir)
849 		blk_mq_debugfs_register_sched(q);
850 
851 	/* Similarly, blk_mq_init_hctx() couldn't do this previously. */
852 	queue_for_each_hw_ctx(q, hctx, i) {
853 		if (!hctx->debugfs_dir && blk_mq_debugfs_register_hctx(q, hctx))
854 			goto err;
855 		if (q->elevator && !hctx->sched_debugfs_dir &&
856 		    blk_mq_debugfs_register_sched_hctx(q, hctx))
857 			goto err;
858 	}
859 
860 	return 0;
861 
862 err:
863 	blk_mq_debugfs_unregister(q);
864 	return -ENOMEM;
865 }
866 
867 void blk_mq_debugfs_unregister(struct request_queue *q)
868 {
869 	debugfs_remove_recursive(q->debugfs_dir);
870 	q->sched_debugfs_dir = NULL;
871 	q->debugfs_dir = NULL;
872 }
873 
874 static int blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
875 				       struct blk_mq_ctx *ctx)
876 {
877 	struct dentry *ctx_dir;
878 	char name[20];
879 
880 	snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
881 	ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
882 	if (!ctx_dir)
883 		return -ENOMEM;
884 
885 	if (!debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs))
886 		return -ENOMEM;
887 
888 	return 0;
889 }
890 
891 int blk_mq_debugfs_register_hctx(struct request_queue *q,
892 				 struct blk_mq_hw_ctx *hctx)
893 {
894 	struct blk_mq_ctx *ctx;
895 	char name[20];
896 	int i;
897 
898 	if (!q->debugfs_dir)
899 		return -ENOENT;
900 
901 	snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
902 	hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
903 	if (!hctx->debugfs_dir)
904 		return -ENOMEM;
905 
906 	if (!debugfs_create_files(hctx->debugfs_dir, hctx,
907 				  blk_mq_debugfs_hctx_attrs))
908 		goto err;
909 
910 	hctx_for_each_ctx(hctx, ctx, i) {
911 		if (blk_mq_debugfs_register_ctx(hctx, ctx))
912 			goto err;
913 	}
914 
915 	return 0;
916 
917 err:
918 	blk_mq_debugfs_unregister_hctx(hctx);
919 	return -ENOMEM;
920 }
921 
922 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
923 {
924 	debugfs_remove_recursive(hctx->debugfs_dir);
925 	hctx->sched_debugfs_dir = NULL;
926 	hctx->debugfs_dir = NULL;
927 }
928 
929 int blk_mq_debugfs_register_hctxs(struct request_queue *q)
930 {
931 	struct blk_mq_hw_ctx *hctx;
932 	int i;
933 
934 	queue_for_each_hw_ctx(q, hctx, i) {
935 		if (blk_mq_debugfs_register_hctx(q, hctx))
936 			return -ENOMEM;
937 	}
938 
939 	return 0;
940 }
941 
942 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
943 {
944 	struct blk_mq_hw_ctx *hctx;
945 	int i;
946 
947 	queue_for_each_hw_ctx(q, hctx, i)
948 		blk_mq_debugfs_unregister_hctx(hctx);
949 }
950 
951 int blk_mq_debugfs_register_sched(struct request_queue *q)
952 {
953 	struct elevator_type *e = q->elevator->type;
954 
955 	if (!q->debugfs_dir)
956 		return -ENOENT;
957 
958 	if (!e->queue_debugfs_attrs)
959 		return 0;
960 
961 	q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
962 	if (!q->sched_debugfs_dir)
963 		return -ENOMEM;
964 
965 	if (!debugfs_create_files(q->sched_debugfs_dir, q,
966 				  e->queue_debugfs_attrs))
967 		goto err;
968 
969 	return 0;
970 
971 err:
972 	blk_mq_debugfs_unregister_sched(q);
973 	return -ENOMEM;
974 }
975 
976 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
977 {
978 	debugfs_remove_recursive(q->sched_debugfs_dir);
979 	q->sched_debugfs_dir = NULL;
980 }
981 
982 int blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
983 				       struct blk_mq_hw_ctx *hctx)
984 {
985 	struct elevator_type *e = q->elevator->type;
986 
987 	if (!hctx->debugfs_dir)
988 		return -ENOENT;
989 
990 	if (!e->hctx_debugfs_attrs)
991 		return 0;
992 
993 	hctx->sched_debugfs_dir = debugfs_create_dir("sched",
994 						     hctx->debugfs_dir);
995 	if (!hctx->sched_debugfs_dir)
996 		return -ENOMEM;
997 
998 	if (!debugfs_create_files(hctx->sched_debugfs_dir, hctx,
999 				  e->hctx_debugfs_attrs))
1000 		return -ENOMEM;
1001 
1002 	return 0;
1003 }
1004 
1005 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1006 {
1007 	debugfs_remove_recursive(hctx->sched_debugfs_dir);
1008 	hctx->sched_debugfs_dir = NULL;
1009 }
1010