xref: /openbmc/linux/block/blk-sysfs.c (revision 61f4d204)
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/debugfs.h>
13 
14 #include "blk.h"
15 #include "blk-mq.h"
16 #include "blk-mq-debugfs.h"
17 #include "blk-mq-sched.h"
18 #include "blk-rq-qos.h"
19 #include "blk-wbt.h"
20 #include "blk-cgroup.h"
21 #include "blk-throttle.h"
22 
23 struct queue_sysfs_entry {
24 	struct attribute attr;
25 	ssize_t (*show)(struct request_queue *, char *);
26 	ssize_t (*store)(struct request_queue *, const char *, size_t);
27 };
28 
29 static ssize_t
30 queue_var_show(unsigned long var, char *page)
31 {
32 	return sprintf(page, "%lu\n", var);
33 }
34 
35 static ssize_t
36 queue_var_store(unsigned long *var, const char *page, size_t count)
37 {
38 	int err;
39 	unsigned long v;
40 
41 	err = kstrtoul(page, 10, &v);
42 	if (err || v > UINT_MAX)
43 		return -EINVAL;
44 
45 	*var = v;
46 
47 	return count;
48 }
49 
50 static ssize_t queue_requests_show(struct request_queue *q, char *page)
51 {
52 	return queue_var_show(q->nr_requests, page);
53 }
54 
55 static ssize_t
56 queue_requests_store(struct request_queue *q, const char *page, size_t count)
57 {
58 	unsigned long nr;
59 	int ret, err;
60 
61 	if (!queue_is_mq(q))
62 		return -EINVAL;
63 
64 	ret = queue_var_store(&nr, page, count);
65 	if (ret < 0)
66 		return ret;
67 
68 	if (nr < BLKDEV_MIN_RQ)
69 		nr = BLKDEV_MIN_RQ;
70 
71 	err = blk_mq_update_nr_requests(q, nr);
72 	if (err)
73 		return err;
74 
75 	return ret;
76 }
77 
78 static ssize_t queue_ra_show(struct request_queue *q, char *page)
79 {
80 	unsigned long ra_kb;
81 
82 	if (!q->disk)
83 		return -EINVAL;
84 	ra_kb = q->disk->bdi->ra_pages << (PAGE_SHIFT - 10);
85 	return queue_var_show(ra_kb, page);
86 }
87 
88 static ssize_t
89 queue_ra_store(struct request_queue *q, const char *page, size_t count)
90 {
91 	unsigned long ra_kb;
92 	ssize_t ret;
93 
94 	if (!q->disk)
95 		return -EINVAL;
96 	ret = queue_var_store(&ra_kb, page, count);
97 	if (ret < 0)
98 		return ret;
99 	q->disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
100 	return ret;
101 }
102 
103 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
104 {
105 	int max_sectors_kb = queue_max_sectors(q) >> 1;
106 
107 	return queue_var_show(max_sectors_kb, page);
108 }
109 
110 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
111 {
112 	return queue_var_show(queue_max_segments(q), page);
113 }
114 
115 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
116 		char *page)
117 {
118 	return queue_var_show(queue_max_discard_segments(q), page);
119 }
120 
121 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
122 {
123 	return queue_var_show(q->limits.max_integrity_segments, page);
124 }
125 
126 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
127 {
128 	return queue_var_show(queue_max_segment_size(q), page);
129 }
130 
131 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
132 {
133 	return queue_var_show(queue_logical_block_size(q), page);
134 }
135 
136 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
137 {
138 	return queue_var_show(queue_physical_block_size(q), page);
139 }
140 
141 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
142 {
143 	return queue_var_show(q->limits.chunk_sectors, page);
144 }
145 
146 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
147 {
148 	return queue_var_show(queue_io_min(q), page);
149 }
150 
151 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
152 {
153 	return queue_var_show(queue_io_opt(q), page);
154 }
155 
156 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
157 {
158 	return queue_var_show(q->limits.discard_granularity, page);
159 }
160 
161 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
162 {
163 
164 	return sprintf(page, "%llu\n",
165 		(unsigned long long)q->limits.max_hw_discard_sectors << 9);
166 }
167 
168 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
169 {
170 	return sprintf(page, "%llu\n",
171 		       (unsigned long long)q->limits.max_discard_sectors << 9);
172 }
173 
174 static ssize_t queue_discard_max_store(struct request_queue *q,
175 				       const char *page, size_t count)
176 {
177 	unsigned long max_discard;
178 	ssize_t ret = queue_var_store(&max_discard, page, count);
179 
180 	if (ret < 0)
181 		return ret;
182 
183 	if (max_discard & (q->limits.discard_granularity - 1))
184 		return -EINVAL;
185 
186 	max_discard >>= 9;
187 	if (max_discard > UINT_MAX)
188 		return -EINVAL;
189 
190 	if (max_discard > q->limits.max_hw_discard_sectors)
191 		max_discard = q->limits.max_hw_discard_sectors;
192 
193 	q->limits.max_discard_sectors = max_discard;
194 	return ret;
195 }
196 
197 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
198 {
199 	return queue_var_show(0, page);
200 }
201 
202 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
203 {
204 	return queue_var_show(0, page);
205 }
206 
207 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
208 {
209 	return sprintf(page, "%llu\n",
210 		(unsigned long long)q->limits.max_write_zeroes_sectors << 9);
211 }
212 
213 static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
214 						 char *page)
215 {
216 	return queue_var_show(queue_zone_write_granularity(q), page);
217 }
218 
219 static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
220 {
221 	unsigned long long max_sectors = q->limits.max_zone_append_sectors;
222 
223 	return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
224 }
225 
226 static ssize_t
227 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
228 {
229 	unsigned long var;
230 	unsigned int max_sectors_kb,
231 		max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
232 			page_kb = 1 << (PAGE_SHIFT - 10);
233 	ssize_t ret = queue_var_store(&var, page, count);
234 
235 	if (ret < 0)
236 		return ret;
237 
238 	max_sectors_kb = (unsigned int)var;
239 	max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb,
240 					 q->limits.max_dev_sectors >> 1);
241 	if (max_sectors_kb == 0) {
242 		q->limits.max_user_sectors = 0;
243 		max_sectors_kb = min(max_hw_sectors_kb,
244 				     BLK_DEF_MAX_SECTORS >> 1);
245 	} else {
246 		if (max_sectors_kb > max_hw_sectors_kb ||
247 		    max_sectors_kb < page_kb)
248 			return -EINVAL;
249 		q->limits.max_user_sectors = max_sectors_kb << 1;
250 	}
251 
252 	spin_lock_irq(&q->queue_lock);
253 	q->limits.max_sectors = max_sectors_kb << 1;
254 	if (q->disk)
255 		q->disk->bdi->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
256 	spin_unlock_irq(&q->queue_lock);
257 
258 	return ret;
259 }
260 
261 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
262 {
263 	int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
264 
265 	return queue_var_show(max_hw_sectors_kb, page);
266 }
267 
268 static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
269 {
270 	return queue_var_show(q->limits.virt_boundary_mask, page);
271 }
272 
273 static ssize_t queue_dma_alignment_show(struct request_queue *q, char *page)
274 {
275 	return queue_var_show(queue_dma_alignment(q), page);
276 }
277 
278 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)				\
279 static ssize_t								\
280 queue_##name##_show(struct request_queue *q, char *page)		\
281 {									\
282 	int bit;							\
283 	bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);		\
284 	return queue_var_show(neg ? !bit : bit, page);			\
285 }									\
286 static ssize_t								\
287 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
288 {									\
289 	unsigned long val;						\
290 	ssize_t ret;							\
291 	ret = queue_var_store(&val, page, count);			\
292 	if (ret < 0)							\
293 		 return ret;						\
294 	if (neg)							\
295 		val = !val;						\
296 									\
297 	if (val)							\
298 		blk_queue_flag_set(QUEUE_FLAG_##flag, q);		\
299 	else								\
300 		blk_queue_flag_clear(QUEUE_FLAG_##flag, q);		\
301 	return ret;							\
302 }
303 
304 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
305 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
306 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
307 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
308 #undef QUEUE_SYSFS_BIT_FNS
309 
310 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
311 {
312 	switch (blk_queue_zoned_model(q)) {
313 	case BLK_ZONED_HA:
314 		return sprintf(page, "host-aware\n");
315 	case BLK_ZONED_HM:
316 		return sprintf(page, "host-managed\n");
317 	default:
318 		return sprintf(page, "none\n");
319 	}
320 }
321 
322 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
323 {
324 	return queue_var_show(disk_nr_zones(q->disk), page);
325 }
326 
327 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
328 {
329 	return queue_var_show(bdev_max_open_zones(q->disk->part0), page);
330 }
331 
332 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
333 {
334 	return queue_var_show(bdev_max_active_zones(q->disk->part0), page);
335 }
336 
337 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
338 {
339 	return queue_var_show((blk_queue_nomerges(q) << 1) |
340 			       blk_queue_noxmerges(q), page);
341 }
342 
343 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
344 				    size_t count)
345 {
346 	unsigned long nm;
347 	ssize_t ret = queue_var_store(&nm, page, count);
348 
349 	if (ret < 0)
350 		return ret;
351 
352 	blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
353 	blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
354 	if (nm == 2)
355 		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
356 	else if (nm)
357 		blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
358 
359 	return ret;
360 }
361 
362 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
363 {
364 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
365 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
366 
367 	return queue_var_show(set << force, page);
368 }
369 
370 static ssize_t
371 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
372 {
373 	ssize_t ret = -EINVAL;
374 #ifdef CONFIG_SMP
375 	unsigned long val;
376 
377 	ret = queue_var_store(&val, page, count);
378 	if (ret < 0)
379 		return ret;
380 
381 	if (val == 2) {
382 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
383 		blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
384 	} else if (val == 1) {
385 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
386 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
387 	} else if (val == 0) {
388 		blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
389 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
390 	}
391 #endif
392 	return ret;
393 }
394 
395 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
396 {
397 	return sprintf(page, "%d\n", -1);
398 }
399 
400 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
401 				size_t count)
402 {
403 	return count;
404 }
405 
406 static ssize_t queue_poll_show(struct request_queue *q, char *page)
407 {
408 	return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
409 }
410 
411 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
412 				size_t count)
413 {
414 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
415 		return -EINVAL;
416 	pr_info_ratelimited("writes to the poll attribute are ignored.\n");
417 	pr_info_ratelimited("please use driver specific parameters instead.\n");
418 	return count;
419 }
420 
421 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
422 {
423 	return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
424 }
425 
426 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
427 				  size_t count)
428 {
429 	unsigned int val;
430 	int err;
431 
432 	err = kstrtou32(page, 10, &val);
433 	if (err || val == 0)
434 		return -EINVAL;
435 
436 	blk_queue_rq_timeout(q, msecs_to_jiffies(val));
437 
438 	return count;
439 }
440 
441 static ssize_t queue_wc_show(struct request_queue *q, char *page)
442 {
443 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
444 		return sprintf(page, "write back\n");
445 
446 	return sprintf(page, "write through\n");
447 }
448 
449 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
450 			      size_t count)
451 {
452 	int set = -1;
453 
454 	if (!strncmp(page, "write back", 10))
455 		set = 1;
456 	else if (!strncmp(page, "write through", 13) ||
457 		 !strncmp(page, "none", 4))
458 		set = 0;
459 
460 	if (set == -1)
461 		return -EINVAL;
462 
463 	if (set)
464 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
465 	else
466 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
467 
468 	return count;
469 }
470 
471 static ssize_t queue_fua_show(struct request_queue *q, char *page)
472 {
473 	return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
474 }
475 
476 static ssize_t queue_dax_show(struct request_queue *q, char *page)
477 {
478 	return queue_var_show(blk_queue_dax(q), page);
479 }
480 
481 #define QUEUE_RO_ENTRY(_prefix, _name)			\
482 static struct queue_sysfs_entry _prefix##_entry = {	\
483 	.attr	= { .name = _name, .mode = 0444 },	\
484 	.show	= _prefix##_show,			\
485 };
486 
487 #define QUEUE_RW_ENTRY(_prefix, _name)			\
488 static struct queue_sysfs_entry _prefix##_entry = {	\
489 	.attr	= { .name = _name, .mode = 0644 },	\
490 	.show	= _prefix##_show,			\
491 	.store	= _prefix##_store,			\
492 };
493 
494 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
495 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
496 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
497 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
498 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
499 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
500 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
501 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
502 
503 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
504 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
505 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
506 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
507 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
508 
509 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
510 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
511 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
512 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
513 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
514 
515 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
516 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
517 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
518 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
519 
520 QUEUE_RO_ENTRY(queue_zoned, "zoned");
521 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
522 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
523 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
524 
525 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
526 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
527 QUEUE_RW_ENTRY(queue_poll, "io_poll");
528 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
529 QUEUE_RW_ENTRY(queue_wc, "write_cache");
530 QUEUE_RO_ENTRY(queue_fua, "fua");
531 QUEUE_RO_ENTRY(queue_dax, "dax");
532 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
533 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
534 QUEUE_RO_ENTRY(queue_dma_alignment, "dma_alignment");
535 
536 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
537 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
538 #endif
539 
540 /* legacy alias for logical_block_size: */
541 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
542 	.attr = {.name = "hw_sector_size", .mode = 0444 },
543 	.show = queue_logical_block_size_show,
544 };
545 
546 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
547 QUEUE_RW_ENTRY(queue_iostats, "iostats");
548 QUEUE_RW_ENTRY(queue_random, "add_random");
549 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
550 
551 #ifdef CONFIG_BLK_WBT
552 static ssize_t queue_var_store64(s64 *var, const char *page)
553 {
554 	int err;
555 	s64 v;
556 
557 	err = kstrtos64(page, 10, &v);
558 	if (err < 0)
559 		return err;
560 
561 	*var = v;
562 	return 0;
563 }
564 
565 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
566 {
567 	if (!wbt_rq_qos(q))
568 		return -EINVAL;
569 
570 	if (wbt_disabled(q))
571 		return sprintf(page, "0\n");
572 
573 	return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
574 }
575 
576 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
577 				  size_t count)
578 {
579 	struct rq_qos *rqos;
580 	ssize_t ret;
581 	s64 val;
582 
583 	ret = queue_var_store64(&val, page);
584 	if (ret < 0)
585 		return ret;
586 	if (val < -1)
587 		return -EINVAL;
588 
589 	rqos = wbt_rq_qos(q);
590 	if (!rqos) {
591 		ret = wbt_init(q->disk);
592 		if (ret)
593 			return ret;
594 	}
595 
596 	if (val == -1)
597 		val = wbt_default_latency_nsec(q);
598 	else if (val >= 0)
599 		val *= 1000ULL;
600 
601 	if (wbt_get_min_lat(q) == val)
602 		return count;
603 
604 	/*
605 	 * Ensure that the queue is idled, in case the latency update
606 	 * ends up either enabling or disabling wbt completely. We can't
607 	 * have IO inflight if that happens.
608 	 */
609 	blk_mq_freeze_queue(q);
610 	blk_mq_quiesce_queue(q);
611 
612 	wbt_set_min_lat(q, val);
613 
614 	blk_mq_unquiesce_queue(q);
615 	blk_mq_unfreeze_queue(q);
616 
617 	return count;
618 }
619 
620 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
621 #endif
622 
623 static struct attribute *queue_attrs[] = {
624 	&queue_ra_entry.attr,
625 	&queue_max_hw_sectors_entry.attr,
626 	&queue_max_sectors_entry.attr,
627 	&queue_max_segments_entry.attr,
628 	&queue_max_discard_segments_entry.attr,
629 	&queue_max_integrity_segments_entry.attr,
630 	&queue_max_segment_size_entry.attr,
631 	&queue_hw_sector_size_entry.attr,
632 	&queue_logical_block_size_entry.attr,
633 	&queue_physical_block_size_entry.attr,
634 	&queue_chunk_sectors_entry.attr,
635 	&queue_io_min_entry.attr,
636 	&queue_io_opt_entry.attr,
637 	&queue_discard_granularity_entry.attr,
638 	&queue_discard_max_entry.attr,
639 	&queue_discard_max_hw_entry.attr,
640 	&queue_discard_zeroes_data_entry.attr,
641 	&queue_write_same_max_entry.attr,
642 	&queue_write_zeroes_max_entry.attr,
643 	&queue_zone_append_max_entry.attr,
644 	&queue_zone_write_granularity_entry.attr,
645 	&queue_nonrot_entry.attr,
646 	&queue_zoned_entry.attr,
647 	&queue_nr_zones_entry.attr,
648 	&queue_max_open_zones_entry.attr,
649 	&queue_max_active_zones_entry.attr,
650 	&queue_nomerges_entry.attr,
651 	&queue_iostats_entry.attr,
652 	&queue_stable_writes_entry.attr,
653 	&queue_random_entry.attr,
654 	&queue_poll_entry.attr,
655 	&queue_wc_entry.attr,
656 	&queue_fua_entry.attr,
657 	&queue_dax_entry.attr,
658 	&queue_poll_delay_entry.attr,
659 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
660 	&blk_throtl_sample_time_entry.attr,
661 #endif
662 	&queue_virt_boundary_mask_entry.attr,
663 	&queue_dma_alignment_entry.attr,
664 	NULL,
665 };
666 
667 static struct attribute *blk_mq_queue_attrs[] = {
668 	&queue_requests_entry.attr,
669 	&elv_iosched_entry.attr,
670 	&queue_rq_affinity_entry.attr,
671 	&queue_io_timeout_entry.attr,
672 #ifdef CONFIG_BLK_WBT
673 	&queue_wb_lat_entry.attr,
674 #endif
675 	NULL,
676 };
677 
678 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
679 				int n)
680 {
681 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
682 	struct request_queue *q = disk->queue;
683 
684 	if ((attr == &queue_max_open_zones_entry.attr ||
685 	     attr == &queue_max_active_zones_entry.attr) &&
686 	    !blk_queue_is_zoned(q))
687 		return 0;
688 
689 	return attr->mode;
690 }
691 
692 static umode_t blk_mq_queue_attr_visible(struct kobject *kobj,
693 					 struct attribute *attr, int n)
694 {
695 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
696 	struct request_queue *q = disk->queue;
697 
698 	if (!queue_is_mq(q))
699 		return 0;
700 
701 	if (attr == &queue_io_timeout_entry.attr && !q->mq_ops->timeout)
702 		return 0;
703 
704 	return attr->mode;
705 }
706 
707 static struct attribute_group queue_attr_group = {
708 	.attrs = queue_attrs,
709 	.is_visible = queue_attr_visible,
710 };
711 
712 static struct attribute_group blk_mq_queue_attr_group = {
713 	.attrs = blk_mq_queue_attrs,
714 	.is_visible = blk_mq_queue_attr_visible,
715 };
716 
717 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
718 
719 static ssize_t
720 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
721 {
722 	struct queue_sysfs_entry *entry = to_queue(attr);
723 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
724 	struct request_queue *q = disk->queue;
725 	ssize_t res;
726 
727 	if (!entry->show)
728 		return -EIO;
729 	mutex_lock(&q->sysfs_lock);
730 	res = entry->show(q, page);
731 	mutex_unlock(&q->sysfs_lock);
732 	return res;
733 }
734 
735 static ssize_t
736 queue_attr_store(struct kobject *kobj, struct attribute *attr,
737 		    const char *page, size_t length)
738 {
739 	struct queue_sysfs_entry *entry = to_queue(attr);
740 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
741 	struct request_queue *q = disk->queue;
742 	ssize_t res;
743 
744 	if (!entry->store)
745 		return -EIO;
746 
747 	mutex_lock(&q->sysfs_lock);
748 	res = entry->store(q, page, length);
749 	mutex_unlock(&q->sysfs_lock);
750 	return res;
751 }
752 
753 static const struct sysfs_ops queue_sysfs_ops = {
754 	.show	= queue_attr_show,
755 	.store	= queue_attr_store,
756 };
757 
758 static const struct attribute_group *blk_queue_attr_groups[] = {
759 	&queue_attr_group,
760 	&blk_mq_queue_attr_group,
761 	NULL
762 };
763 
764 static void blk_queue_release(struct kobject *kobj)
765 {
766 	/* nothing to do here, all data is associated with the parent gendisk */
767 }
768 
769 static const struct kobj_type blk_queue_ktype = {
770 	.default_groups = blk_queue_attr_groups,
771 	.sysfs_ops	= &queue_sysfs_ops,
772 	.release	= blk_queue_release,
773 };
774 
775 static void blk_debugfs_remove(struct gendisk *disk)
776 {
777 	struct request_queue *q = disk->queue;
778 
779 	mutex_lock(&q->debugfs_mutex);
780 	blk_trace_shutdown(q);
781 	debugfs_remove_recursive(q->debugfs_dir);
782 	q->debugfs_dir = NULL;
783 	q->sched_debugfs_dir = NULL;
784 	q->rqos_debugfs_dir = NULL;
785 	mutex_unlock(&q->debugfs_mutex);
786 }
787 
788 /**
789  * blk_register_queue - register a block layer queue with sysfs
790  * @disk: Disk of which the request queue should be registered with sysfs.
791  */
792 int blk_register_queue(struct gendisk *disk)
793 {
794 	struct request_queue *q = disk->queue;
795 	int ret;
796 
797 	mutex_lock(&q->sysfs_dir_lock);
798 	kobject_init(&disk->queue_kobj, &blk_queue_ktype);
799 	ret = kobject_add(&disk->queue_kobj, &disk_to_dev(disk)->kobj, "queue");
800 	if (ret < 0)
801 		goto out_put_queue_kobj;
802 
803 	if (queue_is_mq(q)) {
804 		ret = blk_mq_sysfs_register(disk);
805 		if (ret)
806 			goto out_put_queue_kobj;
807 	}
808 	mutex_lock(&q->sysfs_lock);
809 
810 	mutex_lock(&q->debugfs_mutex);
811 	q->debugfs_dir = debugfs_create_dir(disk->disk_name, blk_debugfs_root);
812 	if (queue_is_mq(q))
813 		blk_mq_debugfs_register(q);
814 	mutex_unlock(&q->debugfs_mutex);
815 
816 	ret = disk_register_independent_access_ranges(disk);
817 	if (ret)
818 		goto out_debugfs_remove;
819 
820 	if (q->elevator) {
821 		ret = elv_register_queue(q, false);
822 		if (ret)
823 			goto out_unregister_ia_ranges;
824 	}
825 
826 	ret = blk_crypto_sysfs_register(disk);
827 	if (ret)
828 		goto out_elv_unregister;
829 
830 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
831 	wbt_enable_default(disk);
832 	blk_throtl_register(disk);
833 
834 	/* Now everything is ready and send out KOBJ_ADD uevent */
835 	kobject_uevent(&disk->queue_kobj, KOBJ_ADD);
836 	if (q->elevator)
837 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
838 	mutex_unlock(&q->sysfs_lock);
839 	mutex_unlock(&q->sysfs_dir_lock);
840 
841 	/*
842 	 * SCSI probing may synchronously create and destroy a lot of
843 	 * request_queues for non-existent devices.  Shutting down a fully
844 	 * functional queue takes measureable wallclock time as RCU grace
845 	 * periods are involved.  To avoid excessive latency in these
846 	 * cases, a request_queue starts out in a degraded mode which is
847 	 * faster to shut down and is made fully functional here as
848 	 * request_queues for non-existent devices never get registered.
849 	 */
850 	if (!blk_queue_init_done(q)) {
851 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
852 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
853 	}
854 
855 	return ret;
856 
857 out_elv_unregister:
858 	elv_unregister_queue(q);
859 out_unregister_ia_ranges:
860 	disk_unregister_independent_access_ranges(disk);
861 out_debugfs_remove:
862 	blk_debugfs_remove(disk);
863 	mutex_unlock(&q->sysfs_lock);
864 out_put_queue_kobj:
865 	kobject_put(&disk->queue_kobj);
866 	mutex_unlock(&q->sysfs_dir_lock);
867 	return ret;
868 }
869 
870 /**
871  * blk_unregister_queue - counterpart of blk_register_queue()
872  * @disk: Disk of which the request queue should be unregistered from sysfs.
873  *
874  * Note: the caller is responsible for guaranteeing that this function is called
875  * after blk_register_queue() has finished.
876  */
877 void blk_unregister_queue(struct gendisk *disk)
878 {
879 	struct request_queue *q = disk->queue;
880 
881 	if (WARN_ON(!q))
882 		return;
883 
884 	/* Return early if disk->queue was never registered. */
885 	if (!blk_queue_registered(q))
886 		return;
887 
888 	/*
889 	 * Since sysfs_remove_dir() prevents adding new directory entries
890 	 * before removal of existing entries starts, protect against
891 	 * concurrent elv_iosched_store() calls.
892 	 */
893 	mutex_lock(&q->sysfs_lock);
894 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
895 	mutex_unlock(&q->sysfs_lock);
896 
897 	mutex_lock(&q->sysfs_dir_lock);
898 	/*
899 	 * Remove the sysfs attributes before unregistering the queue data
900 	 * structures that can be modified through sysfs.
901 	 */
902 	if (queue_is_mq(q))
903 		blk_mq_sysfs_unregister(disk);
904 	blk_crypto_sysfs_unregister(disk);
905 
906 	mutex_lock(&q->sysfs_lock);
907 	elv_unregister_queue(q);
908 	disk_unregister_independent_access_ranges(disk);
909 	mutex_unlock(&q->sysfs_lock);
910 
911 	/* Now that we've deleted all child objects, we can delete the queue. */
912 	kobject_uevent(&disk->queue_kobj, KOBJ_REMOVE);
913 	kobject_del(&disk->queue_kobj);
914 	mutex_unlock(&q->sysfs_dir_lock);
915 
916 	blk_debugfs_remove(disk);
917 }
918