xref: /openbmc/linux/block/blk-sysfs.c (revision 72d6e4871f0457dc9e498f8437ea344ee6b78075)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Functions related to sysfs handling
4  */
5 #include <linux/kernel.h>
6 #include <linux/slab.h>
7 #include <linux/module.h>
8 #include <linux/bio.h>
9 #include <linux/blkdev.h>
10 #include <linux/backing-dev.h>
11 #include <linux/blktrace_api.h>
12 #include <linux/blk-mq.h>
13 #include <linux/blk-cgroup.h>
14 #include <linux/debugfs.h>
15 
16 #include "blk.h"
17 #include "blk-mq.h"
18 #include "blk-mq-debugfs.h"
19 #include "blk-wbt.h"
20 
21 struct queue_sysfs_entry {
22 	struct attribute attr;
23 	ssize_t (*show)(struct request_queue *, char *);
24 	ssize_t (*store)(struct request_queue *, const char *, size_t);
25 };
26 
27 static ssize_t
28 queue_var_show(unsigned long var, char *page)
29 {
30 	return sprintf(page, "%lu\n", var);
31 }
32 
33 static ssize_t
34 queue_var_store(unsigned long *var, const char *page, size_t count)
35 {
36 	int err;
37 	unsigned long v;
38 
39 	err = kstrtoul(page, 10, &v);
40 	if (err || v > UINT_MAX)
41 		return -EINVAL;
42 
43 	*var = v;
44 
45 	return count;
46 }
47 
48 static ssize_t queue_var_store64(s64 *var, const char *page)
49 {
50 	int err;
51 	s64 v;
52 
53 	err = kstrtos64(page, 10, &v);
54 	if (err < 0)
55 		return err;
56 
57 	*var = v;
58 	return 0;
59 }
60 
61 static ssize_t queue_requests_show(struct request_queue *q, char *page)
62 {
63 	return queue_var_show(q->nr_requests, (page));
64 }
65 
66 static ssize_t
67 queue_requests_store(struct request_queue *q, const char *page, size_t count)
68 {
69 	unsigned long nr;
70 	int ret, err;
71 
72 	if (!queue_is_mq(q))
73 		return -EINVAL;
74 
75 	ret = queue_var_store(&nr, page, count);
76 	if (ret < 0)
77 		return ret;
78 
79 	if (nr < BLKDEV_MIN_RQ)
80 		nr = BLKDEV_MIN_RQ;
81 
82 	err = blk_mq_update_nr_requests(q, nr);
83 	if (err)
84 		return err;
85 
86 	return ret;
87 }
88 
89 static ssize_t queue_ra_show(struct request_queue *q, char *page)
90 {
91 	unsigned long ra_kb = q->backing_dev_info->ra_pages <<
92 					(PAGE_SHIFT - 10);
93 
94 	return queue_var_show(ra_kb, (page));
95 }
96 
97 static ssize_t
98 queue_ra_store(struct request_queue *q, const char *page, size_t count)
99 {
100 	unsigned long ra_kb;
101 	ssize_t ret = queue_var_store(&ra_kb, page, count);
102 
103 	if (ret < 0)
104 		return ret;
105 
106 	q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
107 
108 	return ret;
109 }
110 
111 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
112 {
113 	int max_sectors_kb = queue_max_sectors(q) >> 1;
114 
115 	return queue_var_show(max_sectors_kb, (page));
116 }
117 
118 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
119 {
120 	return queue_var_show(queue_max_segments(q), (page));
121 }
122 
123 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
124 		char *page)
125 {
126 	return queue_var_show(queue_max_discard_segments(q), (page));
127 }
128 
129 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
130 {
131 	return queue_var_show(q->limits.max_integrity_segments, (page));
132 }
133 
134 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
135 {
136 	return queue_var_show(queue_max_segment_size(q), (page));
137 }
138 
139 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
140 {
141 	return queue_var_show(queue_logical_block_size(q), page);
142 }
143 
144 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
145 {
146 	return queue_var_show(queue_physical_block_size(q), page);
147 }
148 
149 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
150 {
151 	return queue_var_show(q->limits.chunk_sectors, page);
152 }
153 
154 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
155 {
156 	return queue_var_show(queue_io_min(q), page);
157 }
158 
159 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
160 {
161 	return queue_var_show(queue_io_opt(q), page);
162 }
163 
164 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
165 {
166 	return queue_var_show(q->limits.discard_granularity, page);
167 }
168 
169 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
170 {
171 
172 	return sprintf(page, "%llu\n",
173 		(unsigned long long)q->limits.max_hw_discard_sectors << 9);
174 }
175 
176 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
177 {
178 	return sprintf(page, "%llu\n",
179 		       (unsigned long long)q->limits.max_discard_sectors << 9);
180 }
181 
182 static ssize_t queue_discard_max_store(struct request_queue *q,
183 				       const char *page, size_t count)
184 {
185 	unsigned long max_discard;
186 	ssize_t ret = queue_var_store(&max_discard, page, count);
187 
188 	if (ret < 0)
189 		return ret;
190 
191 	if (max_discard & (q->limits.discard_granularity - 1))
192 		return -EINVAL;
193 
194 	max_discard >>= 9;
195 	if (max_discard > UINT_MAX)
196 		return -EINVAL;
197 
198 	if (max_discard > q->limits.max_hw_discard_sectors)
199 		max_discard = q->limits.max_hw_discard_sectors;
200 
201 	q->limits.max_discard_sectors = max_discard;
202 	return ret;
203 }
204 
205 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
206 {
207 	return queue_var_show(0, page);
208 }
209 
210 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
211 {
212 	return sprintf(page, "%llu\n",
213 		(unsigned long long)q->limits.max_write_same_sectors << 9);
214 }
215 
216 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
217 {
218 	return sprintf(page, "%llu\n",
219 		(unsigned long long)q->limits.max_write_zeroes_sectors << 9);
220 }
221 
222 static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
223 						 char *page)
224 {
225 	return queue_var_show(queue_zone_write_granularity(q), page);
226 }
227 
228 static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
229 {
230 	unsigned long long max_sectors = q->limits.max_zone_append_sectors;
231 
232 	return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
233 }
234 
235 static ssize_t
236 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
237 {
238 	unsigned long max_sectors_kb,
239 		max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
240 			page_kb = 1 << (PAGE_SHIFT - 10);
241 	ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
242 
243 	if (ret < 0)
244 		return ret;
245 
246 	max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
247 					 q->limits.max_dev_sectors >> 1);
248 
249 	if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
250 		return -EINVAL;
251 
252 	spin_lock_irq(&q->queue_lock);
253 	q->limits.max_sectors = max_sectors_kb << 1;
254 	q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
255 	spin_unlock_irq(&q->queue_lock);
256 
257 	return ret;
258 }
259 
260 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
261 {
262 	int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
263 
264 	return queue_var_show(max_hw_sectors_kb, (page));
265 }
266 
267 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)				\
268 static ssize_t								\
269 queue_##name##_show(struct request_queue *q, char *page)		\
270 {									\
271 	int bit;							\
272 	bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);		\
273 	return queue_var_show(neg ? !bit : bit, page);			\
274 }									\
275 static ssize_t								\
276 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
277 {									\
278 	unsigned long val;						\
279 	ssize_t ret;							\
280 	ret = queue_var_store(&val, page, count);			\
281 	if (ret < 0)							\
282 		 return ret;						\
283 	if (neg)							\
284 		val = !val;						\
285 									\
286 	if (val)							\
287 		blk_queue_flag_set(QUEUE_FLAG_##flag, q);		\
288 	else								\
289 		blk_queue_flag_clear(QUEUE_FLAG_##flag, q);		\
290 	return ret;							\
291 }
292 
293 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
294 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
295 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
296 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
297 #undef QUEUE_SYSFS_BIT_FNS
298 
299 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
300 {
301 	switch (blk_queue_zoned_model(q)) {
302 	case BLK_ZONED_HA:
303 		return sprintf(page, "host-aware\n");
304 	case BLK_ZONED_HM:
305 		return sprintf(page, "host-managed\n");
306 	default:
307 		return sprintf(page, "none\n");
308 	}
309 }
310 
311 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
312 {
313 	return queue_var_show(blk_queue_nr_zones(q), page);
314 }
315 
316 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
317 {
318 	return queue_var_show(queue_max_open_zones(q), page);
319 }
320 
321 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
322 {
323 	return queue_var_show(queue_max_active_zones(q), page);
324 }
325 
326 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
327 {
328 	return queue_var_show((blk_queue_nomerges(q) << 1) |
329 			       blk_queue_noxmerges(q), page);
330 }
331 
332 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
333 				    size_t count)
334 {
335 	unsigned long nm;
336 	ssize_t ret = queue_var_store(&nm, page, count);
337 
338 	if (ret < 0)
339 		return ret;
340 
341 	blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
342 	blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
343 	if (nm == 2)
344 		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
345 	else if (nm)
346 		blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
347 
348 	return ret;
349 }
350 
351 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
352 {
353 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
354 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
355 
356 	return queue_var_show(set << force, page);
357 }
358 
359 static ssize_t
360 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
361 {
362 	ssize_t ret = -EINVAL;
363 #ifdef CONFIG_SMP
364 	unsigned long val;
365 
366 	ret = queue_var_store(&val, page, count);
367 	if (ret < 0)
368 		return ret;
369 
370 	if (val == 2) {
371 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
372 		blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
373 	} else if (val == 1) {
374 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
375 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
376 	} else if (val == 0) {
377 		blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
378 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
379 	}
380 #endif
381 	return ret;
382 }
383 
384 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
385 {
386 	int val;
387 
388 	if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
389 		val = BLK_MQ_POLL_CLASSIC;
390 	else
391 		val = q->poll_nsec / 1000;
392 
393 	return sprintf(page, "%d\n", val);
394 }
395 
396 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
397 				size_t count)
398 {
399 	int err, val;
400 
401 	if (!q->mq_ops || !q->mq_ops->poll)
402 		return -EINVAL;
403 
404 	err = kstrtoint(page, 10, &val);
405 	if (err < 0)
406 		return err;
407 
408 	if (val == BLK_MQ_POLL_CLASSIC)
409 		q->poll_nsec = BLK_MQ_POLL_CLASSIC;
410 	else if (val >= 0)
411 		q->poll_nsec = val * 1000;
412 	else
413 		return -EINVAL;
414 
415 	return count;
416 }
417 
418 static ssize_t queue_poll_show(struct request_queue *q, char *page)
419 {
420 	return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
421 }
422 
423 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
424 				size_t count)
425 {
426 	unsigned long poll_on;
427 	ssize_t ret;
428 
429 	if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
430 	    !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
431 		return -EINVAL;
432 
433 	ret = queue_var_store(&poll_on, page, count);
434 	if (ret < 0)
435 		return ret;
436 
437 	if (poll_on)
438 		blk_queue_flag_set(QUEUE_FLAG_POLL, q);
439 	else
440 		blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
441 
442 	return ret;
443 }
444 
445 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
446 {
447 	return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
448 }
449 
450 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
451 				  size_t count)
452 {
453 	unsigned int val;
454 	int err;
455 
456 	err = kstrtou32(page, 10, &val);
457 	if (err || val == 0)
458 		return -EINVAL;
459 
460 	blk_queue_rq_timeout(q, msecs_to_jiffies(val));
461 
462 	return count;
463 }
464 
465 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
466 {
467 	if (!wbt_rq_qos(q))
468 		return -EINVAL;
469 
470 	return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
471 }
472 
473 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
474 				  size_t count)
475 {
476 	struct rq_qos *rqos;
477 	ssize_t ret;
478 	s64 val;
479 
480 	ret = queue_var_store64(&val, page);
481 	if (ret < 0)
482 		return ret;
483 	if (val < -1)
484 		return -EINVAL;
485 
486 	rqos = wbt_rq_qos(q);
487 	if (!rqos) {
488 		ret = wbt_init(q);
489 		if (ret)
490 			return ret;
491 	}
492 
493 	if (val == -1)
494 		val = wbt_default_latency_nsec(q);
495 	else if (val >= 0)
496 		val *= 1000ULL;
497 
498 	if (wbt_get_min_lat(q) == val)
499 		return count;
500 
501 	/*
502 	 * Ensure that the queue is idled, in case the latency update
503 	 * ends up either enabling or disabling wbt completely. We can't
504 	 * have IO inflight if that happens.
505 	 */
506 	blk_mq_freeze_queue(q);
507 	blk_mq_quiesce_queue(q);
508 
509 	wbt_set_min_lat(q, val);
510 
511 	blk_mq_unquiesce_queue(q);
512 	blk_mq_unfreeze_queue(q);
513 
514 	return count;
515 }
516 
517 static ssize_t queue_wc_show(struct request_queue *q, char *page)
518 {
519 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
520 		return sprintf(page, "write back\n");
521 
522 	return sprintf(page, "write through\n");
523 }
524 
525 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
526 			      size_t count)
527 {
528 	int set = -1;
529 
530 	if (!strncmp(page, "write back", 10))
531 		set = 1;
532 	else if (!strncmp(page, "write through", 13) ||
533 		 !strncmp(page, "none", 4))
534 		set = 0;
535 
536 	if (set == -1)
537 		return -EINVAL;
538 
539 	if (set)
540 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
541 	else
542 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
543 
544 	return count;
545 }
546 
547 static ssize_t queue_fua_show(struct request_queue *q, char *page)
548 {
549 	return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
550 }
551 
552 static ssize_t queue_dax_show(struct request_queue *q, char *page)
553 {
554 	return queue_var_show(blk_queue_dax(q), page);
555 }
556 
557 #define QUEUE_RO_ENTRY(_prefix, _name)			\
558 static struct queue_sysfs_entry _prefix##_entry = {	\
559 	.attr	= { .name = _name, .mode = 0444 },	\
560 	.show	= _prefix##_show,			\
561 };
562 
563 #define QUEUE_RW_ENTRY(_prefix, _name)			\
564 static struct queue_sysfs_entry _prefix##_entry = {	\
565 	.attr	= { .name = _name, .mode = 0644 },	\
566 	.show	= _prefix##_show,			\
567 	.store	= _prefix##_store,			\
568 };
569 
570 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
571 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
572 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
573 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
574 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
575 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
576 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
577 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
578 
579 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
580 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
581 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
582 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
583 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
584 
585 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
586 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
587 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
588 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
589 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
590 
591 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
592 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
593 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
594 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
595 
596 QUEUE_RO_ENTRY(queue_zoned, "zoned");
597 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
598 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
599 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
600 
601 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
602 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
603 QUEUE_RW_ENTRY(queue_poll, "io_poll");
604 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
605 QUEUE_RW_ENTRY(queue_wc, "write_cache");
606 QUEUE_RO_ENTRY(queue_fua, "fua");
607 QUEUE_RO_ENTRY(queue_dax, "dax");
608 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
609 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
610 
611 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
612 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
613 #endif
614 
615 /* legacy alias for logical_block_size: */
616 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
617 	.attr = {.name = "hw_sector_size", .mode = 0444 },
618 	.show = queue_logical_block_size_show,
619 };
620 
621 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
622 QUEUE_RW_ENTRY(queue_iostats, "iostats");
623 QUEUE_RW_ENTRY(queue_random, "add_random");
624 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
625 
626 static struct attribute *queue_attrs[] = {
627 	&queue_requests_entry.attr,
628 	&queue_ra_entry.attr,
629 	&queue_max_hw_sectors_entry.attr,
630 	&queue_max_sectors_entry.attr,
631 	&queue_max_segments_entry.attr,
632 	&queue_max_discard_segments_entry.attr,
633 	&queue_max_integrity_segments_entry.attr,
634 	&queue_max_segment_size_entry.attr,
635 	&elv_iosched_entry.attr,
636 	&queue_hw_sector_size_entry.attr,
637 	&queue_logical_block_size_entry.attr,
638 	&queue_physical_block_size_entry.attr,
639 	&queue_chunk_sectors_entry.attr,
640 	&queue_io_min_entry.attr,
641 	&queue_io_opt_entry.attr,
642 	&queue_discard_granularity_entry.attr,
643 	&queue_discard_max_entry.attr,
644 	&queue_discard_max_hw_entry.attr,
645 	&queue_discard_zeroes_data_entry.attr,
646 	&queue_write_same_max_entry.attr,
647 	&queue_write_zeroes_max_entry.attr,
648 	&queue_zone_append_max_entry.attr,
649 	&queue_zone_write_granularity_entry.attr,
650 	&queue_nonrot_entry.attr,
651 	&queue_zoned_entry.attr,
652 	&queue_nr_zones_entry.attr,
653 	&queue_max_open_zones_entry.attr,
654 	&queue_max_active_zones_entry.attr,
655 	&queue_nomerges_entry.attr,
656 	&queue_rq_affinity_entry.attr,
657 	&queue_iostats_entry.attr,
658 	&queue_stable_writes_entry.attr,
659 	&queue_random_entry.attr,
660 	&queue_poll_entry.attr,
661 	&queue_wc_entry.attr,
662 	&queue_fua_entry.attr,
663 	&queue_dax_entry.attr,
664 	&queue_wb_lat_entry.attr,
665 	&queue_poll_delay_entry.attr,
666 	&queue_io_timeout_entry.attr,
667 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
668 	&blk_throtl_sample_time_entry.attr,
669 #endif
670 	NULL,
671 };
672 
673 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
674 				int n)
675 {
676 	struct request_queue *q =
677 		container_of(kobj, struct request_queue, kobj);
678 
679 	if (attr == &queue_io_timeout_entry.attr &&
680 		(!q->mq_ops || !q->mq_ops->timeout))
681 			return 0;
682 
683 	if ((attr == &queue_max_open_zones_entry.attr ||
684 	     attr == &queue_max_active_zones_entry.attr) &&
685 	    !blk_queue_is_zoned(q))
686 		return 0;
687 
688 	return attr->mode;
689 }
690 
691 static struct attribute_group queue_attr_group = {
692 	.attrs = queue_attrs,
693 	.is_visible = queue_attr_visible,
694 };
695 
696 
697 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
698 
699 static ssize_t
700 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
701 {
702 	struct queue_sysfs_entry *entry = to_queue(attr);
703 	struct request_queue *q =
704 		container_of(kobj, struct request_queue, kobj);
705 	ssize_t res;
706 
707 	if (!entry->show)
708 		return -EIO;
709 	mutex_lock(&q->sysfs_lock);
710 	res = entry->show(q, page);
711 	mutex_unlock(&q->sysfs_lock);
712 	return res;
713 }
714 
715 static ssize_t
716 queue_attr_store(struct kobject *kobj, struct attribute *attr,
717 		    const char *page, size_t length)
718 {
719 	struct queue_sysfs_entry *entry = to_queue(attr);
720 	struct request_queue *q;
721 	ssize_t res;
722 
723 	if (!entry->store)
724 		return -EIO;
725 
726 	q = container_of(kobj, struct request_queue, kobj);
727 	mutex_lock(&q->sysfs_lock);
728 	res = entry->store(q, page, length);
729 	mutex_unlock(&q->sysfs_lock);
730 	return res;
731 }
732 
733 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
734 {
735 	struct request_queue *q = container_of(rcu_head, struct request_queue,
736 					       rcu_head);
737 	kmem_cache_free(blk_requestq_cachep, q);
738 }
739 
740 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
741 static void blk_exit_queue(struct request_queue *q)
742 {
743 	/*
744 	 * Since the I/O scheduler exit code may access cgroup information,
745 	 * perform I/O scheduler exit before disassociating from the block
746 	 * cgroup controller.
747 	 */
748 	if (q->elevator) {
749 		ioc_clear_queue(q);
750 		__elevator_exit(q, q->elevator);
751 	}
752 
753 	/*
754 	 * Remove all references to @q from the block cgroup controller before
755 	 * restoring @q->queue_lock to avoid that restoring this pointer causes
756 	 * e.g. blkcg_print_blkgs() to crash.
757 	 */
758 	blkcg_exit_queue(q);
759 
760 	/*
761 	 * Since the cgroup code may dereference the @q->backing_dev_info
762 	 * pointer, only decrease its reference count after having removed the
763 	 * association with the block cgroup controller.
764 	 */
765 	bdi_put(q->backing_dev_info);
766 }
767 
768 /**
769  * blk_release_queue - releases all allocated resources of the request_queue
770  * @kobj: pointer to a kobject, whose container is a request_queue
771  *
772  * This function releases all allocated resources of the request queue.
773  *
774  * The struct request_queue refcount is incremented with blk_get_queue() and
775  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
776  * is called.
777  *
778  * For drivers that have a request_queue on a gendisk and added with
779  * __device_add_disk() the refcount to request_queue will reach 0 with
780  * the last put_disk() called by the driver. For drivers which don't use
781  * __device_add_disk() this happens with blk_cleanup_queue().
782  *
783  * Drivers exist which depend on the release of the request_queue to be
784  * synchronous, it should not be deferred.
785  *
786  * Context: can sleep
787  */
788 static void blk_release_queue(struct kobject *kobj)
789 {
790 	struct request_queue *q =
791 		container_of(kobj, struct request_queue, kobj);
792 
793 	might_sleep();
794 
795 	if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
796 		blk_stat_remove_callback(q, q->poll_cb);
797 	blk_stat_free_callback(q->poll_cb);
798 
799 	blk_free_queue_stats(q->stats);
800 
801 	if (queue_is_mq(q)) {
802 		struct blk_mq_hw_ctx *hctx;
803 		int i;
804 
805 		cancel_delayed_work_sync(&q->requeue_work);
806 
807 		queue_for_each_hw_ctx(q, hctx, i)
808 			cancel_delayed_work_sync(&hctx->run_work);
809 	}
810 
811 	blk_exit_queue(q);
812 
813 	blk_queue_free_zone_bitmaps(q);
814 
815 	if (queue_is_mq(q))
816 		blk_mq_release(q);
817 
818 	blk_trace_shutdown(q);
819 	mutex_lock(&q->debugfs_mutex);
820 	debugfs_remove_recursive(q->debugfs_dir);
821 	mutex_unlock(&q->debugfs_mutex);
822 
823 	if (queue_is_mq(q))
824 		blk_mq_debugfs_unregister(q);
825 
826 	bioset_exit(&q->bio_split);
827 
828 	ida_simple_remove(&blk_queue_ida, q->id);
829 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
830 }
831 
832 static const struct sysfs_ops queue_sysfs_ops = {
833 	.show	= queue_attr_show,
834 	.store	= queue_attr_store,
835 };
836 
837 struct kobj_type blk_queue_ktype = {
838 	.sysfs_ops	= &queue_sysfs_ops,
839 	.release	= blk_release_queue,
840 };
841 
842 /**
843  * blk_register_queue - register a block layer queue with sysfs
844  * @disk: Disk of which the request queue should be registered with sysfs.
845  */
846 int blk_register_queue(struct gendisk *disk)
847 {
848 	int ret;
849 	struct device *dev = disk_to_dev(disk);
850 	struct request_queue *q = disk->queue;
851 
852 	if (WARN_ON(!q))
853 		return -ENXIO;
854 
855 	WARN_ONCE(blk_queue_registered(q),
856 		  "%s is registering an already registered queue\n",
857 		  kobject_name(&dev->kobj));
858 
859 	/*
860 	 * SCSI probing may synchronously create and destroy a lot of
861 	 * request_queues for non-existent devices.  Shutting down a fully
862 	 * functional queue takes measureable wallclock time as RCU grace
863 	 * periods are involved.  To avoid excessive latency in these
864 	 * cases, a request_queue starts out in a degraded mode which is
865 	 * faster to shut down and is made fully functional here as
866 	 * request_queues for non-existent devices never get registered.
867 	 */
868 	if (!blk_queue_init_done(q)) {
869 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
870 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
871 	}
872 
873 	blk_queue_update_readahead(q);
874 
875 	ret = blk_trace_init_sysfs(dev);
876 	if (ret)
877 		return ret;
878 
879 	mutex_lock(&q->sysfs_dir_lock);
880 
881 	ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
882 	if (ret < 0) {
883 		blk_trace_remove_sysfs(dev);
884 		goto unlock;
885 	}
886 
887 	ret = sysfs_create_group(&q->kobj, &queue_attr_group);
888 	if (ret) {
889 		blk_trace_remove_sysfs(dev);
890 		kobject_del(&q->kobj);
891 		kobject_put(&dev->kobj);
892 		goto unlock;
893 	}
894 
895 	mutex_lock(&q->debugfs_mutex);
896 	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
897 					    blk_debugfs_root);
898 	mutex_unlock(&q->debugfs_mutex);
899 
900 	if (queue_is_mq(q)) {
901 		__blk_mq_register_dev(dev, q);
902 		blk_mq_debugfs_register(q);
903 	}
904 
905 	mutex_lock(&q->sysfs_lock);
906 	if (q->elevator) {
907 		ret = elv_register_queue(q, false);
908 		if (ret) {
909 			mutex_unlock(&q->sysfs_lock);
910 			mutex_unlock(&q->sysfs_dir_lock);
911 			kobject_del(&q->kobj);
912 			blk_trace_remove_sysfs(dev);
913 			kobject_put(&dev->kobj);
914 			return ret;
915 		}
916 	}
917 
918 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
919 	wbt_enable_default(q);
920 	blk_throtl_register_queue(q);
921 
922 	/* Now everything is ready and send out KOBJ_ADD uevent */
923 	kobject_uevent(&q->kobj, KOBJ_ADD);
924 	if (q->elevator)
925 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
926 	mutex_unlock(&q->sysfs_lock);
927 
928 	ret = 0;
929 unlock:
930 	mutex_unlock(&q->sysfs_dir_lock);
931 	return ret;
932 }
933 EXPORT_SYMBOL_GPL(blk_register_queue);
934 
935 /**
936  * blk_unregister_queue - counterpart of blk_register_queue()
937  * @disk: Disk of which the request queue should be unregistered from sysfs.
938  *
939  * Note: the caller is responsible for guaranteeing that this function is called
940  * after blk_register_queue() has finished.
941  */
942 void blk_unregister_queue(struct gendisk *disk)
943 {
944 	struct request_queue *q = disk->queue;
945 
946 	if (WARN_ON(!q))
947 		return;
948 
949 	/* Return early if disk->queue was never registered. */
950 	if (!blk_queue_registered(q))
951 		return;
952 
953 	/*
954 	 * Since sysfs_remove_dir() prevents adding new directory entries
955 	 * before removal of existing entries starts, protect against
956 	 * concurrent elv_iosched_store() calls.
957 	 */
958 	mutex_lock(&q->sysfs_lock);
959 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
960 	mutex_unlock(&q->sysfs_lock);
961 
962 	mutex_lock(&q->sysfs_dir_lock);
963 	/*
964 	 * Remove the sysfs attributes before unregistering the queue data
965 	 * structures that can be modified through sysfs.
966 	 */
967 	if (queue_is_mq(q))
968 		blk_mq_unregister_dev(disk_to_dev(disk), q);
969 
970 	kobject_uevent(&q->kobj, KOBJ_REMOVE);
971 	kobject_del(&q->kobj);
972 	blk_trace_remove_sysfs(disk_to_dev(disk));
973 
974 	mutex_lock(&q->sysfs_lock);
975 	if (q->elevator)
976 		elv_unregister_queue(q);
977 	mutex_unlock(&q->sysfs_lock);
978 	mutex_unlock(&q->sysfs_dir_lock);
979 
980 	kobject_put(&disk_to_dev(disk)->kobj);
981 }
982