xref: /openbmc/linux/block/blk-sysfs.c (revision 0545810f)
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_var_store64(s64 *var, const char *page)
51 {
52 	int err;
53 	s64 v;
54 
55 	err = kstrtos64(page, 10, &v);
56 	if (err < 0)
57 		return err;
58 
59 	*var = v;
60 	return 0;
61 }
62 
63 static ssize_t queue_requests_show(struct request_queue *q, char *page)
64 {
65 	return queue_var_show(q->nr_requests, page);
66 }
67 
68 static ssize_t
69 queue_requests_store(struct request_queue *q, const char *page, size_t count)
70 {
71 	unsigned long nr;
72 	int ret, err;
73 
74 	if (!queue_is_mq(q))
75 		return -EINVAL;
76 
77 	ret = queue_var_store(&nr, page, count);
78 	if (ret < 0)
79 		return ret;
80 
81 	if (nr < BLKDEV_MIN_RQ)
82 		nr = BLKDEV_MIN_RQ;
83 
84 	err = blk_mq_update_nr_requests(q, nr);
85 	if (err)
86 		return err;
87 
88 	return ret;
89 }
90 
91 static ssize_t queue_ra_show(struct request_queue *q, char *page)
92 {
93 	unsigned long ra_kb;
94 
95 	if (!q->disk)
96 		return -EINVAL;
97 	ra_kb = q->disk->bdi->ra_pages << (PAGE_SHIFT - 10);
98 	return queue_var_show(ra_kb, page);
99 }
100 
101 static ssize_t
102 queue_ra_store(struct request_queue *q, const char *page, size_t count)
103 {
104 	unsigned long ra_kb;
105 	ssize_t ret;
106 
107 	if (!q->disk)
108 		return -EINVAL;
109 	ret = queue_var_store(&ra_kb, page, count);
110 	if (ret < 0)
111 		return ret;
112 	q->disk->bdi->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
113 	return ret;
114 }
115 
116 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
117 {
118 	int max_sectors_kb = queue_max_sectors(q) >> 1;
119 
120 	return queue_var_show(max_sectors_kb, page);
121 }
122 
123 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
124 {
125 	return queue_var_show(queue_max_segments(q), page);
126 }
127 
128 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
129 		char *page)
130 {
131 	return queue_var_show(queue_max_discard_segments(q), page);
132 }
133 
134 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
135 {
136 	return queue_var_show(q->limits.max_integrity_segments, page);
137 }
138 
139 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
140 {
141 	return queue_var_show(queue_max_segment_size(q), page);
142 }
143 
144 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
145 {
146 	return queue_var_show(queue_logical_block_size(q), page);
147 }
148 
149 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
150 {
151 	return queue_var_show(queue_physical_block_size(q), page);
152 }
153 
154 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
155 {
156 	return queue_var_show(q->limits.chunk_sectors, page);
157 }
158 
159 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
160 {
161 	return queue_var_show(queue_io_min(q), page);
162 }
163 
164 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
165 {
166 	return queue_var_show(queue_io_opt(q), page);
167 }
168 
169 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
170 {
171 	return queue_var_show(q->limits.discard_granularity, page);
172 }
173 
174 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
175 {
176 
177 	return sprintf(page, "%llu\n",
178 		(unsigned long long)q->limits.max_hw_discard_sectors << 9);
179 }
180 
181 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
182 {
183 	return sprintf(page, "%llu\n",
184 		       (unsigned long long)q->limits.max_discard_sectors << 9);
185 }
186 
187 static ssize_t queue_discard_max_store(struct request_queue *q,
188 				       const char *page, size_t count)
189 {
190 	unsigned long max_discard;
191 	ssize_t ret = queue_var_store(&max_discard, page, count);
192 
193 	if (ret < 0)
194 		return ret;
195 
196 	if (max_discard & (q->limits.discard_granularity - 1))
197 		return -EINVAL;
198 
199 	max_discard >>= 9;
200 	if (max_discard > UINT_MAX)
201 		return -EINVAL;
202 
203 	if (max_discard > q->limits.max_hw_discard_sectors)
204 		max_discard = q->limits.max_hw_discard_sectors;
205 
206 	q->limits.max_discard_sectors = max_discard;
207 	return ret;
208 }
209 
210 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
211 {
212 	return queue_var_show(0, page);
213 }
214 
215 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
216 {
217 	return queue_var_show(0, page);
218 }
219 
220 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
221 {
222 	return sprintf(page, "%llu\n",
223 		(unsigned long long)q->limits.max_write_zeroes_sectors << 9);
224 }
225 
226 static ssize_t queue_zone_write_granularity_show(struct request_queue *q,
227 						 char *page)
228 {
229 	return queue_var_show(queue_zone_write_granularity(q), page);
230 }
231 
232 static ssize_t queue_zone_append_max_show(struct request_queue *q, char *page)
233 {
234 	unsigned long long max_sectors = q->limits.max_zone_append_sectors;
235 
236 	return sprintf(page, "%llu\n", max_sectors << SECTOR_SHIFT);
237 }
238 
239 static ssize_t
240 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
241 {
242 	unsigned long var;
243 	unsigned int max_sectors_kb,
244 		max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
245 			page_kb = 1 << (PAGE_SHIFT - 10);
246 	ssize_t ret = queue_var_store(&var, page, count);
247 
248 	if (ret < 0)
249 		return ret;
250 
251 	max_sectors_kb = (unsigned int)var;
252 	max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb,
253 					 q->limits.max_dev_sectors >> 1);
254 	if (max_sectors_kb == 0) {
255 		q->limits.max_user_sectors = 0;
256 		max_sectors_kb = min(max_hw_sectors_kb,
257 				     BLK_DEF_MAX_SECTORS >> 1);
258 	} else {
259 		if (max_sectors_kb > max_hw_sectors_kb ||
260 		    max_sectors_kb < page_kb)
261 			return -EINVAL;
262 		q->limits.max_user_sectors = max_sectors_kb << 1;
263 	}
264 
265 	spin_lock_irq(&q->queue_lock);
266 	q->limits.max_sectors = max_sectors_kb << 1;
267 	if (q->disk)
268 		q->disk->bdi->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
269 	spin_unlock_irq(&q->queue_lock);
270 
271 	return ret;
272 }
273 
274 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
275 {
276 	int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
277 
278 	return queue_var_show(max_hw_sectors_kb, page);
279 }
280 
281 static ssize_t queue_virt_boundary_mask_show(struct request_queue *q, char *page)
282 {
283 	return queue_var_show(q->limits.virt_boundary_mask, page);
284 }
285 
286 static ssize_t queue_dma_alignment_show(struct request_queue *q, char *page)
287 {
288 	return queue_var_show(queue_dma_alignment(q), page);
289 }
290 
291 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)				\
292 static ssize_t								\
293 queue_##name##_show(struct request_queue *q, char *page)		\
294 {									\
295 	int bit;							\
296 	bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);		\
297 	return queue_var_show(neg ? !bit : bit, page);			\
298 }									\
299 static ssize_t								\
300 queue_##name##_store(struct request_queue *q, const char *page, size_t count) \
301 {									\
302 	unsigned long val;						\
303 	ssize_t ret;							\
304 	ret = queue_var_store(&val, page, count);			\
305 	if (ret < 0)							\
306 		 return ret;						\
307 	if (neg)							\
308 		val = !val;						\
309 									\
310 	if (val)							\
311 		blk_queue_flag_set(QUEUE_FLAG_##flag, q);		\
312 	else								\
313 		blk_queue_flag_clear(QUEUE_FLAG_##flag, q);		\
314 	return ret;							\
315 }
316 
317 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
318 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
319 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
320 QUEUE_SYSFS_BIT_FNS(stable_writes, STABLE_WRITES, 0);
321 #undef QUEUE_SYSFS_BIT_FNS
322 
323 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
324 {
325 	switch (blk_queue_zoned_model(q)) {
326 	case BLK_ZONED_HA:
327 		return sprintf(page, "host-aware\n");
328 	case BLK_ZONED_HM:
329 		return sprintf(page, "host-managed\n");
330 	default:
331 		return sprintf(page, "none\n");
332 	}
333 }
334 
335 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
336 {
337 	return queue_var_show(disk_nr_zones(q->disk), page);
338 }
339 
340 static ssize_t queue_max_open_zones_show(struct request_queue *q, char *page)
341 {
342 	return queue_var_show(bdev_max_open_zones(q->disk->part0), page);
343 }
344 
345 static ssize_t queue_max_active_zones_show(struct request_queue *q, char *page)
346 {
347 	return queue_var_show(bdev_max_active_zones(q->disk->part0), page);
348 }
349 
350 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
351 {
352 	return queue_var_show((blk_queue_nomerges(q) << 1) |
353 			       blk_queue_noxmerges(q), page);
354 }
355 
356 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
357 				    size_t count)
358 {
359 	unsigned long nm;
360 	ssize_t ret = queue_var_store(&nm, page, count);
361 
362 	if (ret < 0)
363 		return ret;
364 
365 	blk_queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
366 	blk_queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
367 	if (nm == 2)
368 		blk_queue_flag_set(QUEUE_FLAG_NOMERGES, q);
369 	else if (nm)
370 		blk_queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
371 
372 	return ret;
373 }
374 
375 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
376 {
377 	bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
378 	bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
379 
380 	return queue_var_show(set << force, page);
381 }
382 
383 static ssize_t
384 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
385 {
386 	ssize_t ret = -EINVAL;
387 #ifdef CONFIG_SMP
388 	unsigned long val;
389 
390 	ret = queue_var_store(&val, page, count);
391 	if (ret < 0)
392 		return ret;
393 
394 	if (val == 2) {
395 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
396 		blk_queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
397 	} else if (val == 1) {
398 		blk_queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
399 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
400 	} else if (val == 0) {
401 		blk_queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
402 		blk_queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
403 	}
404 #endif
405 	return ret;
406 }
407 
408 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
409 {
410 	return sprintf(page, "%d\n", -1);
411 }
412 
413 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
414 				size_t count)
415 {
416 	return count;
417 }
418 
419 static ssize_t queue_poll_show(struct request_queue *q, char *page)
420 {
421 	return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
422 }
423 
424 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
425 				size_t count)
426 {
427 	if (!test_bit(QUEUE_FLAG_POLL, &q->queue_flags))
428 		return -EINVAL;
429 	pr_info_ratelimited("writes to the poll attribute are ignored.\n");
430 	pr_info_ratelimited("please use driver specific parameters instead.\n");
431 	return count;
432 }
433 
434 static ssize_t queue_io_timeout_show(struct request_queue *q, char *page)
435 {
436 	return sprintf(page, "%u\n", jiffies_to_msecs(q->rq_timeout));
437 }
438 
439 static ssize_t queue_io_timeout_store(struct request_queue *q, const char *page,
440 				  size_t count)
441 {
442 	unsigned int val;
443 	int err;
444 
445 	err = kstrtou32(page, 10, &val);
446 	if (err || val == 0)
447 		return -EINVAL;
448 
449 	blk_queue_rq_timeout(q, msecs_to_jiffies(val));
450 
451 	return count;
452 }
453 
454 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
455 {
456 	if (!wbt_rq_qos(q))
457 		return -EINVAL;
458 
459 	if (wbt_disabled(q))
460 		return sprintf(page, "0\n");
461 
462 	return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
463 }
464 
465 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
466 				  size_t count)
467 {
468 	struct rq_qos *rqos;
469 	ssize_t ret;
470 	s64 val;
471 
472 	ret = queue_var_store64(&val, page);
473 	if (ret < 0)
474 		return ret;
475 	if (val < -1)
476 		return -EINVAL;
477 
478 	rqos = wbt_rq_qos(q);
479 	if (!rqos) {
480 		ret = wbt_init(q->disk);
481 		if (ret)
482 			return ret;
483 	}
484 
485 	if (val == -1)
486 		val = wbt_default_latency_nsec(q);
487 	else if (val >= 0)
488 		val *= 1000ULL;
489 
490 	if (wbt_get_min_lat(q) == val)
491 		return count;
492 
493 	/*
494 	 * Ensure that the queue is idled, in case the latency update
495 	 * ends up either enabling or disabling wbt completely. We can't
496 	 * have IO inflight if that happens.
497 	 */
498 	blk_mq_freeze_queue(q);
499 	blk_mq_quiesce_queue(q);
500 
501 	wbt_set_min_lat(q, val);
502 
503 	blk_mq_unquiesce_queue(q);
504 	blk_mq_unfreeze_queue(q);
505 
506 	return count;
507 }
508 
509 static ssize_t queue_wc_show(struct request_queue *q, char *page)
510 {
511 	if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
512 		return sprintf(page, "write back\n");
513 
514 	return sprintf(page, "write through\n");
515 }
516 
517 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
518 			      size_t count)
519 {
520 	int set = -1;
521 
522 	if (!strncmp(page, "write back", 10))
523 		set = 1;
524 	else if (!strncmp(page, "write through", 13) ||
525 		 !strncmp(page, "none", 4))
526 		set = 0;
527 
528 	if (set == -1)
529 		return -EINVAL;
530 
531 	if (set)
532 		blk_queue_flag_set(QUEUE_FLAG_WC, q);
533 	else
534 		blk_queue_flag_clear(QUEUE_FLAG_WC, q);
535 
536 	return count;
537 }
538 
539 static ssize_t queue_fua_show(struct request_queue *q, char *page)
540 {
541 	return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
542 }
543 
544 static ssize_t queue_dax_show(struct request_queue *q, char *page)
545 {
546 	return queue_var_show(blk_queue_dax(q), page);
547 }
548 
549 #define QUEUE_RO_ENTRY(_prefix, _name)			\
550 static struct queue_sysfs_entry _prefix##_entry = {	\
551 	.attr	= { .name = _name, .mode = 0444 },	\
552 	.show	= _prefix##_show,			\
553 };
554 
555 #define QUEUE_RW_ENTRY(_prefix, _name)			\
556 static struct queue_sysfs_entry _prefix##_entry = {	\
557 	.attr	= { .name = _name, .mode = 0644 },	\
558 	.show	= _prefix##_show,			\
559 	.store	= _prefix##_store,			\
560 };
561 
562 QUEUE_RW_ENTRY(queue_requests, "nr_requests");
563 QUEUE_RW_ENTRY(queue_ra, "read_ahead_kb");
564 QUEUE_RW_ENTRY(queue_max_sectors, "max_sectors_kb");
565 QUEUE_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
566 QUEUE_RO_ENTRY(queue_max_segments, "max_segments");
567 QUEUE_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
568 QUEUE_RO_ENTRY(queue_max_segment_size, "max_segment_size");
569 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
570 
571 QUEUE_RO_ENTRY(queue_logical_block_size, "logical_block_size");
572 QUEUE_RO_ENTRY(queue_physical_block_size, "physical_block_size");
573 QUEUE_RO_ENTRY(queue_chunk_sectors, "chunk_sectors");
574 QUEUE_RO_ENTRY(queue_io_min, "minimum_io_size");
575 QUEUE_RO_ENTRY(queue_io_opt, "optimal_io_size");
576 
577 QUEUE_RO_ENTRY(queue_max_discard_segments, "max_discard_segments");
578 QUEUE_RO_ENTRY(queue_discard_granularity, "discard_granularity");
579 QUEUE_RO_ENTRY(queue_discard_max_hw, "discard_max_hw_bytes");
580 QUEUE_RW_ENTRY(queue_discard_max, "discard_max_bytes");
581 QUEUE_RO_ENTRY(queue_discard_zeroes_data, "discard_zeroes_data");
582 
583 QUEUE_RO_ENTRY(queue_write_same_max, "write_same_max_bytes");
584 QUEUE_RO_ENTRY(queue_write_zeroes_max, "write_zeroes_max_bytes");
585 QUEUE_RO_ENTRY(queue_zone_append_max, "zone_append_max_bytes");
586 QUEUE_RO_ENTRY(queue_zone_write_granularity, "zone_write_granularity");
587 
588 QUEUE_RO_ENTRY(queue_zoned, "zoned");
589 QUEUE_RO_ENTRY(queue_nr_zones, "nr_zones");
590 QUEUE_RO_ENTRY(queue_max_open_zones, "max_open_zones");
591 QUEUE_RO_ENTRY(queue_max_active_zones, "max_active_zones");
592 
593 QUEUE_RW_ENTRY(queue_nomerges, "nomerges");
594 QUEUE_RW_ENTRY(queue_rq_affinity, "rq_affinity");
595 QUEUE_RW_ENTRY(queue_poll, "io_poll");
596 QUEUE_RW_ENTRY(queue_poll_delay, "io_poll_delay");
597 QUEUE_RW_ENTRY(queue_wc, "write_cache");
598 QUEUE_RO_ENTRY(queue_fua, "fua");
599 QUEUE_RO_ENTRY(queue_dax, "dax");
600 QUEUE_RW_ENTRY(queue_io_timeout, "io_timeout");
601 QUEUE_RW_ENTRY(queue_wb_lat, "wbt_lat_usec");
602 QUEUE_RO_ENTRY(queue_virt_boundary_mask, "virt_boundary_mask");
603 QUEUE_RO_ENTRY(queue_dma_alignment, "dma_alignment");
604 
605 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
606 QUEUE_RW_ENTRY(blk_throtl_sample_time, "throttle_sample_time");
607 #endif
608 
609 /* legacy alias for logical_block_size: */
610 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
611 	.attr = {.name = "hw_sector_size", .mode = 0444 },
612 	.show = queue_logical_block_size_show,
613 };
614 
615 QUEUE_RW_ENTRY(queue_nonrot, "rotational");
616 QUEUE_RW_ENTRY(queue_iostats, "iostats");
617 QUEUE_RW_ENTRY(queue_random, "add_random");
618 QUEUE_RW_ENTRY(queue_stable_writes, "stable_writes");
619 
620 static struct attribute *queue_attrs[] = {
621 	&queue_requests_entry.attr,
622 	&queue_ra_entry.attr,
623 	&queue_max_hw_sectors_entry.attr,
624 	&queue_max_sectors_entry.attr,
625 	&queue_max_segments_entry.attr,
626 	&queue_max_discard_segments_entry.attr,
627 	&queue_max_integrity_segments_entry.attr,
628 	&queue_max_segment_size_entry.attr,
629 	&elv_iosched_entry.attr,
630 	&queue_hw_sector_size_entry.attr,
631 	&queue_logical_block_size_entry.attr,
632 	&queue_physical_block_size_entry.attr,
633 	&queue_chunk_sectors_entry.attr,
634 	&queue_io_min_entry.attr,
635 	&queue_io_opt_entry.attr,
636 	&queue_discard_granularity_entry.attr,
637 	&queue_discard_max_entry.attr,
638 	&queue_discard_max_hw_entry.attr,
639 	&queue_discard_zeroes_data_entry.attr,
640 	&queue_write_same_max_entry.attr,
641 	&queue_write_zeroes_max_entry.attr,
642 	&queue_zone_append_max_entry.attr,
643 	&queue_zone_write_granularity_entry.attr,
644 	&queue_nonrot_entry.attr,
645 	&queue_zoned_entry.attr,
646 	&queue_nr_zones_entry.attr,
647 	&queue_max_open_zones_entry.attr,
648 	&queue_max_active_zones_entry.attr,
649 	&queue_nomerges_entry.attr,
650 	&queue_rq_affinity_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_wb_lat_entry.attr,
659 	&queue_poll_delay_entry.attr,
660 	&queue_io_timeout_entry.attr,
661 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
662 	&blk_throtl_sample_time_entry.attr,
663 #endif
664 	&queue_virt_boundary_mask_entry.attr,
665 	&queue_dma_alignment_entry.attr,
666 	NULL,
667 };
668 
669 static umode_t queue_attr_visible(struct kobject *kobj, struct attribute *attr,
670 				int n)
671 {
672 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
673 	struct request_queue *q = disk->queue;
674 
675 	if (attr == &queue_io_timeout_entry.attr &&
676 		(!q->mq_ops || !q->mq_ops->timeout))
677 			return 0;
678 
679 	if ((attr == &queue_max_open_zones_entry.attr ||
680 	     attr == &queue_max_active_zones_entry.attr) &&
681 	    !blk_queue_is_zoned(q))
682 		return 0;
683 
684 	return attr->mode;
685 }
686 
687 static struct attribute_group queue_attr_group = {
688 	.attrs = queue_attrs,
689 	.is_visible = queue_attr_visible,
690 };
691 
692 
693 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
694 
695 static ssize_t
696 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
697 {
698 	struct queue_sysfs_entry *entry = to_queue(attr);
699 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
700 	struct request_queue *q = disk->queue;
701 	ssize_t res;
702 
703 	if (!entry->show)
704 		return -EIO;
705 	mutex_lock(&q->sysfs_lock);
706 	res = entry->show(q, page);
707 	mutex_unlock(&q->sysfs_lock);
708 	return res;
709 }
710 
711 static ssize_t
712 queue_attr_store(struct kobject *kobj, struct attribute *attr,
713 		    const char *page, size_t length)
714 {
715 	struct queue_sysfs_entry *entry = to_queue(attr);
716 	struct gendisk *disk = container_of(kobj, struct gendisk, queue_kobj);
717 	struct request_queue *q = disk->queue;
718 	ssize_t res;
719 
720 	if (!entry->store)
721 		return -EIO;
722 
723 	mutex_lock(&q->sysfs_lock);
724 	res = entry->store(q, page, length);
725 	mutex_unlock(&q->sysfs_lock);
726 	return res;
727 }
728 
729 static const struct sysfs_ops queue_sysfs_ops = {
730 	.show	= queue_attr_show,
731 	.store	= queue_attr_store,
732 };
733 
734 static const struct attribute_group *blk_queue_attr_groups[] = {
735 	&queue_attr_group,
736 	NULL
737 };
738 
739 static void blk_queue_release(struct kobject *kobj)
740 {
741 	/* nothing to do here, all data is associated with the parent gendisk */
742 }
743 
744 static const struct kobj_type blk_queue_ktype = {
745 	.default_groups = blk_queue_attr_groups,
746 	.sysfs_ops	= &queue_sysfs_ops,
747 	.release	= blk_queue_release,
748 };
749 
750 static void blk_debugfs_remove(struct gendisk *disk)
751 {
752 	struct request_queue *q = disk->queue;
753 
754 	mutex_lock(&q->debugfs_mutex);
755 	blk_trace_shutdown(q);
756 	debugfs_remove_recursive(q->debugfs_dir);
757 	q->debugfs_dir = NULL;
758 	q->sched_debugfs_dir = NULL;
759 	q->rqos_debugfs_dir = NULL;
760 	mutex_unlock(&q->debugfs_mutex);
761 }
762 
763 /**
764  * blk_register_queue - register a block layer queue with sysfs
765  * @disk: Disk of which the request queue should be registered with sysfs.
766  */
767 int blk_register_queue(struct gendisk *disk)
768 {
769 	struct request_queue *q = disk->queue;
770 	int ret;
771 
772 	mutex_lock(&q->sysfs_dir_lock);
773 	kobject_init(&disk->queue_kobj, &blk_queue_ktype);
774 	ret = kobject_add(&disk->queue_kobj, &disk_to_dev(disk)->kobj, "queue");
775 	if (ret < 0)
776 		goto out_put_queue_kobj;
777 
778 	if (queue_is_mq(q)) {
779 		ret = blk_mq_sysfs_register(disk);
780 		if (ret)
781 			goto out_put_queue_kobj;
782 	}
783 	mutex_lock(&q->sysfs_lock);
784 
785 	mutex_lock(&q->debugfs_mutex);
786 	q->debugfs_dir = debugfs_create_dir(disk->disk_name, blk_debugfs_root);
787 	if (queue_is_mq(q))
788 		blk_mq_debugfs_register(q);
789 	mutex_unlock(&q->debugfs_mutex);
790 
791 	ret = disk_register_independent_access_ranges(disk);
792 	if (ret)
793 		goto out_debugfs_remove;
794 
795 	if (q->elevator) {
796 		ret = elv_register_queue(q, false);
797 		if (ret)
798 			goto out_unregister_ia_ranges;
799 	}
800 
801 	ret = blk_crypto_sysfs_register(disk);
802 	if (ret)
803 		goto out_elv_unregister;
804 
805 	blk_queue_flag_set(QUEUE_FLAG_REGISTERED, q);
806 	wbt_enable_default(disk);
807 	blk_throtl_register(disk);
808 
809 	/* Now everything is ready and send out KOBJ_ADD uevent */
810 	kobject_uevent(&disk->queue_kobj, KOBJ_ADD);
811 	if (q->elevator)
812 		kobject_uevent(&q->elevator->kobj, KOBJ_ADD);
813 	mutex_unlock(&q->sysfs_lock);
814 	mutex_unlock(&q->sysfs_dir_lock);
815 
816 	/*
817 	 * SCSI probing may synchronously create and destroy a lot of
818 	 * request_queues for non-existent devices.  Shutting down a fully
819 	 * functional queue takes measureable wallclock time as RCU grace
820 	 * periods are involved.  To avoid excessive latency in these
821 	 * cases, a request_queue starts out in a degraded mode which is
822 	 * faster to shut down and is made fully functional here as
823 	 * request_queues for non-existent devices never get registered.
824 	 */
825 	if (!blk_queue_init_done(q)) {
826 		blk_queue_flag_set(QUEUE_FLAG_INIT_DONE, q);
827 		percpu_ref_switch_to_percpu(&q->q_usage_counter);
828 	}
829 
830 	return ret;
831 
832 out_elv_unregister:
833 	elv_unregister_queue(q);
834 out_unregister_ia_ranges:
835 	disk_unregister_independent_access_ranges(disk);
836 out_debugfs_remove:
837 	blk_debugfs_remove(disk);
838 	mutex_unlock(&q->sysfs_lock);
839 out_put_queue_kobj:
840 	kobject_put(&disk->queue_kobj);
841 	mutex_unlock(&q->sysfs_dir_lock);
842 	return ret;
843 }
844 
845 /**
846  * blk_unregister_queue - counterpart of blk_register_queue()
847  * @disk: Disk of which the request queue should be unregistered from sysfs.
848  *
849  * Note: the caller is responsible for guaranteeing that this function is called
850  * after blk_register_queue() has finished.
851  */
852 void blk_unregister_queue(struct gendisk *disk)
853 {
854 	struct request_queue *q = disk->queue;
855 
856 	if (WARN_ON(!q))
857 		return;
858 
859 	/* Return early if disk->queue was never registered. */
860 	if (!blk_queue_registered(q))
861 		return;
862 
863 	/*
864 	 * Since sysfs_remove_dir() prevents adding new directory entries
865 	 * before removal of existing entries starts, protect against
866 	 * concurrent elv_iosched_store() calls.
867 	 */
868 	mutex_lock(&q->sysfs_lock);
869 	blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
870 	mutex_unlock(&q->sysfs_lock);
871 
872 	mutex_lock(&q->sysfs_dir_lock);
873 	/*
874 	 * Remove the sysfs attributes before unregistering the queue data
875 	 * structures that can be modified through sysfs.
876 	 */
877 	if (queue_is_mq(q))
878 		blk_mq_sysfs_unregister(disk);
879 	blk_crypto_sysfs_unregister(disk);
880 
881 	mutex_lock(&q->sysfs_lock);
882 	elv_unregister_queue(q);
883 	disk_unregister_independent_access_ranges(disk);
884 	mutex_unlock(&q->sysfs_lock);
885 
886 	/* Now that we've deleted all child objects, we can delete the queue. */
887 	kobject_uevent(&disk->queue_kobj, KOBJ_REMOVE);
888 	kobject_del(&disk->queue_kobj);
889 	mutex_unlock(&q->sysfs_dir_lock);
890 
891 	blk_debugfs_remove(disk);
892 }
893