xref: /linux/block/blk-sysfs.c (revision bbcd53c960713507ae764bf81970651b5577b95a)
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 static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
268 {
269 	return queue_var_show(q->limits.virt_boundary_mask, (page));
270 }
271 
272 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)				\
273 static ssize_t								\
274 queue_##name##_show(struct request_queue *q, char *page)		\
275 {									\
276 	int bit;							\
277 	bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);		\
278 	return queue_var_show(neg ? !bit : bit, page);			\
279 }									\
280 static ssize_t								\
281 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
282 {									\
283 	unsigned long val;						\
284 	ssize_t ret;							\
285 	ret = queue_var_store(&val, page, count);			\
286 	if (ret < 0)							\
287 		 return ret;						\
288 	if (neg)							\
289 		val = !val;						\
290 									\
291 	if (val)							\
292 		blk_queue_flag_set(QUEUE_FLAG_##flag, q);		\
293 	else								\
294 		blk_queue_flag_clear(QUEUE_FLAG_##flag, q);		\
295 	return ret;							\
296 }
297 
298 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
299 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
300 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
301 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
302 #undef QUEUE_SYSFS_BIT_FNS
303 
304 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
305 {
306 	switch (blk_queue_zoned_model(q)) {
307 	case BLK_ZONED_HA:
308 		return sprintf(page, "host-aware\n");
309 	case BLK_ZONED_HM:
310 		return sprintf(page, "host-managed\n");
311 	default:
312 		return sprintf(page, "none\n");
313 	}
314 }
315 
316 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
317 {
318 	return queue_var_show(blk_queue_nr_zones(q), page);
319 }
320 
321 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
322 {
323 	return queue_var_show(queue_max_open_zones(q), page);
324 }
325 
326 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
327 {
328 	return queue_var_show(queue_max_active_zones(q), page);
329 }
330 
331 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
332 {
333 	return queue_var_show((blk_queue_nomerges(q) << 1) |
334 			       blk_queue_noxmerges(q), page);
335 }
336 
337 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
338 				    size_t count)
339 {
340 	unsigned long nm;
341 	ssize_t ret = queue_var_store(&nm, page, count);
342 
343 	if (ret < 0)
344 		return ret;
345 
346 	blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
347 	blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
348 	if (nm == 2)
349 		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
350 	else if (nm)
351 		blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
352 
353 	return ret;
354 }
355 
356 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
357 {
358 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
359 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
360 
361 	return queue_var_show(set << force, page);
362 }
363 
364 static ssize_t
365 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
366 {
367 	ssize_t ret = -EINVAL;
368 #ifdef CONFIG_SMP
369 	unsigned long val;
370 
371 	ret = queue_var_store(&val, page, count);
372 	if (ret < 0)
373 		return ret;
374 
375 	if (val == 2) {
376 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
377 		blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
378 	} else if (val == 1) {
379 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
380 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
381 	} else if (val == 0) {
382 		blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
383 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
384 	}
385 #endif
386 	return ret;
387 }
388 
389 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
390 {
391 	int val;
392 
393 	if (q->poll_nsec == BLK_MQ_POLL_CLASSIC)
394 		val = BLK_MQ_POLL_CLASSIC;
395 	else
396 		val = q->poll_nsec / 1000;
397 
398 	return sprintf(page, "%d\n", val);
399 }
400 
401 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
402 				size_t count)
403 {
404 	int err, val;
405 
406 	if (!q->mq_ops || !q->mq_ops->poll)
407 		return -EINVAL;
408 
409 	err = kstrtoint(page, 10, &val);
410 	if (err < 0)
411 		return err;
412 
413 	if (val == BLK_MQ_POLL_CLASSIC)
414 		q->poll_nsec = BLK_MQ_POLL_CLASSIC;
415 	else if (val >= 0)
416 		q->poll_nsec = val * 1000;
417 	else
418 		return -EINVAL;
419 
420 	return count;
421 }
422 
423 static ssize_t queue_poll_show(struct request_queue *q, char *page)
424 {
425 	return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
426 }
427 
428 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
429 				size_t count)
430 {
431 	unsigned long poll_on;
432 	ssize_t ret;
433 
434 	if (!q->tag_set || q->tag_set->nr_maps <= HCTX_TYPE_POLL ||
435 	    !q->tag_set->map[HCTX_TYPE_POLL].nr_queues)
436 		return -EINVAL;
437 
438 	ret = queue_var_store(&poll_on, page, count);
439 	if (ret < 0)
440 		return ret;
441 
442 	if (poll_on) {
443 		blk_queue_flag_set(QUEUE_FLAG_POLL, q);
444 	} else {
445 		blk_mq_freeze_queue(q);
446 		blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
447 		blk_mq_unfreeze_queue(q);
448 	}
449 
450 	return ret;
451 }
452 
453 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
454 {
455 	return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
456 }
457 
458 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
459 				  size_t count)
460 {
461 	unsigned int val;
462 	int err;
463 
464 	err = kstrtou32(page, 10, &val);
465 	if (err || val == 0)
466 		return -EINVAL;
467 
468 	blk_queue_rq_timeout(q, msecs_to_jiffies(val));
469 
470 	return count;
471 }
472 
473 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
474 {
475 	if (!wbt_rq_qos(q))
476 		return -EINVAL;
477 
478 	return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
479 }
480 
481 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
482 				  size_t count)
483 {
484 	struct rq_qos *rqos;
485 	ssize_t ret;
486 	s64 val;
487 
488 	ret = queue_var_store64(&val, page);
489 	if (ret < 0)
490 		return ret;
491 	if (val < -1)
492 		return -EINVAL;
493 
494 	rqos = wbt_rq_qos(q);
495 	if (!rqos) {
496 		ret = wbt_init(q);
497 		if (ret)
498 			return ret;
499 	}
500 
501 	if (val == -1)
502 		val = wbt_default_latency_nsec(q);
503 	else if (val >= 0)
504 		val *= 1000ULL;
505 
506 	if (wbt_get_min_lat(q) == val)
507 		return count;
508 
509 	/*
510 	 * Ensure that the queue is idled, in case the latency update
511 	 * ends up either enabling or disabling wbt completely. We can't
512 	 * have IO inflight if that happens.
513 	 */
514 	blk_mq_freeze_queue(q);
515 	blk_mq_quiesce_queue(q);
516 
517 	wbt_set_min_lat(q, val);
518 
519 	blk_mq_unquiesce_queue(q);
520 	blk_mq_unfreeze_queue(q);
521 
522 	return count;
523 }
524 
525 static ssize_t queue_wc_show(struct request_queue *q, char *page)
526 {
527 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
528 		return sprintf(page, "write back\n");
529 
530 	return sprintf(page, "write through\n");
531 }
532 
533 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
534 			      size_t count)
535 {
536 	int set = -1;
537 
538 	if (!strncmp(page, "write back", 10))
539 		set = 1;
540 	else if (!strncmp(page, "write through", 13) ||
541 		 !strncmp(page, "none", 4))
542 		set = 0;
543 
544 	if (set == -1)
545 		return -EINVAL;
546 
547 	if (set)
548 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
549 	else
550 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
551 
552 	return count;
553 }
554 
555 static ssize_t queue_fua_show(struct request_queue *q, char *page)
556 {
557 	return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
558 }
559 
560 static ssize_t queue_dax_show(struct request_queue *q, char *page)
561 {
562 	return queue_var_show(blk_queue_dax(q), page);
563 }
564 
565 #define QUEUE_RO_ENTRY(_prefix, _name)			\
566 static struct queue_sysfs_entry _prefix##_entry = {	\
567 	.attr	= { .name = _name, .mode = 0444 },	\
568 	.show	= _prefix##_show,			\
569 };
570 
571 #define QUEUE_RW_ENTRY(_prefix, _name)			\
572 static struct queue_sysfs_entry _prefix##_entry = {	\
573 	.attr	= { .name = _name, .mode = 0644 },	\
574 	.show	= _prefix##_show,			\
575 	.store	= _prefix##_store,			\
576 };
577 
578 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
579 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
580 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
581 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
582 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
583 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
584 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
585 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
586 
587 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
588 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
589 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
590 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
591 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
592 
593 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
594 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
595 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
596 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
597 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
598 
599 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
600 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
601 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
602 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
603 
604 QUEUE_RO_ENTRY(queue_zoned, "zoned");
605 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
606 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
607 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
608 
609 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
610 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
611 QUEUE_RW_ENTRY(queue_poll, "io_poll");
612 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
613 QUEUE_RW_ENTRY(queue_wc, "write_cache");
614 QUEUE_RO_ENTRY(queue_fua, "fua");
615 QUEUE_RO_ENTRY(queue_dax, "dax");
616 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
617 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
618 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
619 
620 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
621 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
622 #endif
623 
624 /* legacy alias for logical_block_size: */
625 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
626 	.attr = {.name = "hw_sector_size", .mode = 0444 },
627 	.show = queue_logical_block_size_show,
628 };
629 
630 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
631 QUEUE_RW_ENTRY(queue_iostats, "iostats");
632 QUEUE_RW_ENTRY(queue_random, "add_random");
633 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
634 
635 static struct attribute *queue_attrs[] = {
636 	&queue_requests_entry.attr,
637 	&queue_ra_entry.attr,
638 	&queue_max_hw_sectors_entry.attr,
639 	&queue_max_sectors_entry.attr,
640 	&queue_max_segments_entry.attr,
641 	&queue_max_discard_segments_entry.attr,
642 	&queue_max_integrity_segments_entry.attr,
643 	&queue_max_segment_size_entry.attr,
644 	&elv_iosched_entry.attr,
645 	&queue_hw_sector_size_entry.attr,
646 	&queue_logical_block_size_entry.attr,
647 	&queue_physical_block_size_entry.attr,
648 	&queue_chunk_sectors_entry.attr,
649 	&queue_io_min_entry.attr,
650 	&queue_io_opt_entry.attr,
651 	&queue_discard_granularity_entry.attr,
652 	&queue_discard_max_entry.attr,
653 	&queue_discard_max_hw_entry.attr,
654 	&queue_discard_zeroes_data_entry.attr,
655 	&queue_write_same_max_entry.attr,
656 	&queue_write_zeroes_max_entry.attr,
657 	&queue_zone_append_max_entry.attr,
658 	&queue_zone_write_granularity_entry.attr,
659 	&queue_nonrot_entry.attr,
660 	&queue_zoned_entry.attr,
661 	&queue_nr_zones_entry.attr,
662 	&queue_max_open_zones_entry.attr,
663 	&queue_max_active_zones_entry.attr,
664 	&queue_nomerges_entry.attr,
665 	&queue_rq_affinity_entry.attr,
666 	&queue_iostats_entry.attr,
667 	&queue_stable_writes_entry.attr,
668 	&queue_random_entry.attr,
669 	&queue_poll_entry.attr,
670 	&queue_wc_entry.attr,
671 	&queue_fua_entry.attr,
672 	&queue_dax_entry.attr,
673 	&queue_wb_lat_entry.attr,
674 	&queue_poll_delay_entry.attr,
675 	&queue_io_timeout_entry.attr,
676 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
677 	&blk_throtl_sample_time_entry.attr,
678 #endif
679 	&queue_virt_boundary_mask_entry.attr,
680 	NULL,
681 };
682 
683 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
684 				int n)
685 {
686 	struct request_queue *q =
687 		container_of(kobj, struct request_queue, kobj);
688 
689 	if (attr == &queue_io_timeout_entry.attr &&
690 		(!q->mq_ops || !q->mq_ops->timeout))
691 			return 0;
692 
693 	if ((attr == &queue_max_open_zones_entry.attr ||
694 	     attr == &queue_max_active_zones_entry.attr) &&
695 	    !blk_queue_is_zoned(q))
696 		return 0;
697 
698 	return attr->mode;
699 }
700 
701 static struct attribute_group queue_attr_group = {
702 	.attrs = queue_attrs,
703 	.is_visible = queue_attr_visible,
704 };
705 
706 
707 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
708 
709 static ssize_t
710 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
711 {
712 	struct queue_sysfs_entry *entry = to_queue(attr);
713 	struct request_queue *q =
714 		container_of(kobj, struct request_queue, kobj);
715 	ssize_t res;
716 
717 	if (!entry->show)
718 		return -EIO;
719 	mutex_lock(&q->sysfs_lock);
720 	res = entry->show(q, page);
721 	mutex_unlock(&q->sysfs_lock);
722 	return res;
723 }
724 
725 static ssize_t
726 queue_attr_store(struct kobject *kobj, struct attribute *attr,
727 		    const char *page, size_t length)
728 {
729 	struct queue_sysfs_entry *entry = to_queue(attr);
730 	struct request_queue *q;
731 	ssize_t res;
732 
733 	if (!entry->store)
734 		return -EIO;
735 
736 	q = container_of(kobj, struct request_queue, kobj);
737 	mutex_lock(&q->sysfs_lock);
738 	res = entry->store(q, page, length);
739 	mutex_unlock(&q->sysfs_lock);
740 	return res;
741 }
742 
743 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
744 {
745 	struct request_queue *q = container_of(rcu_head, struct request_queue,
746 					       rcu_head);
747 	kmem_cache_free(blk_requestq_cachep, q);
748 }
749 
750 /* Unconfigure the I/O scheduler and dissociate from the cgroup controller. */
751 static void blk_exit_queue(struct request_queue *q)
752 {
753 	/*
754 	 * Since the I/O scheduler exit code may access cgroup information,
755 	 * perform I/O scheduler exit before disassociating from the block
756 	 * cgroup controller.
757 	 */
758 	if (q->elevator) {
759 		ioc_clear_queue(q);
760 		__elevator_exit(q, q->elevator);
761 	}
762 
763 	/*
764 	 * Remove all references to @q from the block cgroup controller before
765 	 * restoring @q->queue_lock to avoid that restoring this pointer causes
766 	 * e.g. blkcg_print_blkgs() to crash.
767 	 */
768 	blkcg_exit_queue(q);
769 
770 	/*
771 	 * Since the cgroup code may dereference the @q->backing_dev_info
772 	 * pointer, only decrease its reference count after having removed the
773 	 * association with the block cgroup controller.
774 	 */
775 	bdi_put(q->backing_dev_info);
776 }
777 
778 /**
779  * blk_release_queue - releases all allocated resources of the request_queue
780  * @kobj: pointer to a kobject, whose container is a request_queue
781  *
782  * This function releases all allocated resources of the request queue.
783  *
784  * The struct request_queue refcount is incremented with blk_get_queue() and
785  * decremented with blk_put_queue(). Once the refcount reaches 0 this function
786  * is called.
787  *
788  * For drivers that have a request_queue on a gendisk and added with
789  * __device_add_disk() the refcount to request_queue will reach 0 with
790  * the last put_disk() called by the driver. For drivers which don't use
791  * __device_add_disk() this happens with blk_cleanup_queue().
792  *
793  * Drivers exist which depend on the release of the request_queue to be
794  * synchronous, it should not be deferred.
795  *
796  * Context: can sleep
797  */
798 static void blk_release_queue(struct kobject *kobj)
799 {
800 	struct request_queue *q =
801 		container_of(kobj, struct request_queue, kobj);
802 
803 	might_sleep();
804 
805 	if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
806 		blk_stat_remove_callback(q, q->poll_cb);
807 	blk_stat_free_callback(q->poll_cb);
808 
809 	blk_free_queue_stats(q->stats);
810 
811 	if (queue_is_mq(q)) {
812 		struct blk_mq_hw_ctx *hctx;
813 		int i;
814 
815 		cancel_delayed_work_sync(&q->requeue_work);
816 
817 		queue_for_each_hw_ctx(q, hctx, i)
818 			cancel_delayed_work_sync(&hctx->run_work);
819 	}
820 
821 	blk_exit_queue(q);
822 
823 	blk_queue_free_zone_bitmaps(q);
824 
825 	if (queue_is_mq(q))
826 		blk_mq_release(q);
827 
828 	blk_trace_shutdown(q);
829 	mutex_lock(&q->debugfs_mutex);
830 	debugfs_remove_recursive(q->debugfs_dir);
831 	mutex_unlock(&q->debugfs_mutex);
832 
833 	if (queue_is_mq(q))
834 		blk_mq_debugfs_unregister(q);
835 
836 	bioset_exit(&q->bio_split);
837 
838 	ida_simple_remove(&blk_queue_ida, q->id);
839 	call_rcu(&q->rcu_head, blk_free_queue_rcu);
840 }
841 
842 static const struct sysfs_ops queue_sysfs_ops = {
843 	.show	= queue_attr_show,
844 	.store	= queue_attr_store,
845 };
846 
847 struct kobj_type blk_queue_ktype = {
848 	.sysfs_ops	= &queue_sysfs_ops,
849 	.release	= blk_release_queue,
850 };
851 
852 /**
853  * blk_register_queue - register a block layer queue with sysfs
854  * @disk: Disk of which the request queue should be registered with sysfs.
855  */
856 int blk_register_queue(struct gendisk *disk)
857 {
858 	int ret;
859 	struct device *dev = disk_to_dev(disk);
860 	struct request_queue *q = disk->queue;
861 
862 	if (WARN_ON(!q))
863 		return -ENXIO;
864 
865 	WARN_ONCE(blk_queue_registered(q),
866 		  "%s is registering an already registered queue\n",
867 		  kobject_name(&dev->kobj));
868 
869 	/*
870 	 * SCSI probing may synchronously create and destroy a lot of
871 	 * request_queues for non-existent devices.  Shutting down a fully
872 	 * functional queue takes measureable wallclock time as RCU grace
873 	 * periods are involved.  To avoid excessive latency in these
874 	 * cases, a request_queue starts out in a degraded mode which is
875 	 * faster to shut down and is made fully functional here as
876 	 * request_queues for non-existent devices never get registered.
877 	 */
878 	if (!blk_queue_init_done(q)) {
879 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
880 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
881 	}
882 
883 	blk_queue_update_readahead(q);
884 
885 	ret = blk_trace_init_sysfs(dev);
886 	if (ret)
887 		return ret;
888 
889 	mutex_lock(&q->sysfs_dir_lock);
890 
891 	ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
892 	if (ret < 0) {
893 		blk_trace_remove_sysfs(dev);
894 		goto unlock;
895 	}
896 
897 	ret = sysfs_create_group(&q->kobj, &queue_attr_group);
898 	if (ret) {
899 		blk_trace_remove_sysfs(dev);
900 		kobject_del(&q->kobj);
901 		kobject_put(&dev->kobj);
902 		goto unlock;
903 	}
904 
905 	mutex_lock(&q->debugfs_mutex);
906 	q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
907 					    blk_debugfs_root);
908 	mutex_unlock(&q->debugfs_mutex);
909 
910 	if (queue_is_mq(q)) {
911 		__blk_mq_register_dev(dev, q);
912 		blk_mq_debugfs_register(q);
913 	}
914 
915 	mutex_lock(&q->sysfs_lock);
916 	if (q->elevator) {
917 		ret = elv_register_queue(q, false);
918 		if (ret) {
919 			mutex_unlock(&q->sysfs_lock);
920 			mutex_unlock(&q->sysfs_dir_lock);
921 			kobject_del(&q->kobj);
922 			blk_trace_remove_sysfs(dev);
923 			kobject_put(&dev->kobj);
924 			return ret;
925 		}
926 	}
927 
928 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
929 	wbt_enable_default(q);
930 	blk_throtl_register_queue(q);
931 
932 	/* Now everything is ready and send out KOBJ_ADD uevent */
933 	kobject_uevent(&q->kobj, KOBJ_ADD);
934 	if (q->elevator)
935 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
936 	mutex_unlock(&q->sysfs_lock);
937 
938 	ret = 0;
939 unlock:
940 	mutex_unlock(&q->sysfs_dir_lock);
941 	return ret;
942 }
943 EXPORT_SYMBOL_GPL(blk_register_queue);
944 
945 /**
946  * blk_unregister_queue - counterpart of blk_register_queue()
947  * @disk: Disk of which the request queue should be unregistered from sysfs.
948  *
949  * Note: the caller is responsible for guaranteeing that this function is called
950  * after blk_register_queue() has finished.
951  */
952 void blk_unregister_queue(struct gendisk *disk)
953 {
954 	struct request_queue *q = disk->queue;
955 
956 	if (WARN_ON(!q))
957 		return;
958 
959 	/* Return early if disk->queue was never registered. */
960 	if (!blk_queue_registered(q))
961 		return;
962 
963 	/*
964 	 * Since sysfs_remove_dir() prevents adding new directory entries
965 	 * before removal of existing entries starts, protect against
966 	 * concurrent elv_iosched_store() calls.
967 	 */
968 	mutex_lock(&q->sysfs_lock);
969 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
970 	mutex_unlock(&q->sysfs_lock);
971 
972 	mutex_lock(&q->sysfs_dir_lock);
973 	/*
974 	 * Remove the sysfs attributes before unregistering the queue data
975 	 * structures that can be modified through sysfs.
976 	 */
977 	if (queue_is_mq(q))
978 		blk_mq_unregister_dev(disk_to_dev(disk), q);
979 
980 	kobject_uevent(&q->kobj, KOBJ_REMOVE);
981 	kobject_del(&q->kobj);
982 	blk_trace_remove_sysfs(disk_to_dev(disk));
983 
984 	mutex_lock(&q->sysfs_lock);
985 	if (q->elevator)
986 		elv_unregister_queue(q);
987 	mutex_unlock(&q->sysfs_lock);
988 	mutex_unlock(&q->sysfs_dir_lock);
989 
990 	kobject_put(&disk_to_dev(disk)->kobj);
991 }
992