block/blk-sysfs.c

   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
  15 #include "blk.h"
  16 #include "blk-mq.h"
  17 #include "blk-mq-debugfs.h"
  18 #include "blk-wbt.h"
  19
  20 struct queue_sysfs_entry {
  21         struct attribute attr;
  22         ssize_t (*show)(struct request_queue *, char *);
  23         ssize_t (*store)(struct request_queue *, const char *, size_t);
  24 };
  25
  26 static ssize_t
  27 queue_var_show(unsigned long var, char *page)
  28 {
  29         return sprintf(page, "%lu\n", var);
  30 }
  31
  32 static ssize_t
  33 queue_var_store(unsigned long *var, const char *page, size_t count)
  34 {
  35         int err;
  36         unsigned long v;
  37
  38         err = kstrtoul(page, 10, &v);
  39         if (err || v > UINT_MAX)
  40                 return -EINVAL;
  41
  42         *var = v;
  43
  44         return count;
  45 }
  46
  47 static ssize_t queue_var_store64(s64 *var, const char *page)
  48 {
  49         int err;
  50         s64 v;
  51
  52         err = kstrtos64(page, 10, &v);
  53         if (err < 0)
  54                 return err;
  55
  56         *var = v;
  57         return 0;
  58 }
  59
  60 static ssize_t queue_requests_show(struct request_queue *q, char *page)
  61 {
  62         return queue_var_show(q->nr_requests, (page));
  63 }
  64
  65 static ssize_t
  66 queue_requests_store(struct request_queue *q, const char *page, size_t count)
  67 {
  68         unsigned long nr;
  69         int ret, err;
  70
  71         if (!q->request_fn && !q->mq_ops)
  72                 return -EINVAL;
  73
  74         ret = queue_var_store(&nr, page, count);
  75         if (ret < 0)
  76                 return ret;
  77
  78         if (nr < BLKDEV_MIN_RQ)
  79                 nr = BLKDEV_MIN_RQ;
  80
  81         if (q->request_fn)
  82                 err = blk_update_nr_requests(q, nr);
  83         else
  84                 err = blk_mq_update_nr_requests(q, nr);
  85
  86         if (err)
  87                 return err;
  88
  89         return ret;
  90 }
  91
  92 static ssize_t queue_ra_show(struct request_queue *q, char *page)
  93 {
  94         unsigned long ra_kb = q->backing_dev_info->ra_pages <<
  95                                         (PAGE_SHIFT - 10);
  96
  97         return queue_var_show(ra_kb, (page));
  98 }
  99
 100 static ssize_t
 101 queue_ra_store(struct request_queue *q, const char *page, size_t count)
 102 {
 103         unsigned long ra_kb;
 104         ssize_t ret = queue_var_store(&ra_kb, page, count);
 105
 106         if (ret < 0)
 107                 return ret;
 108
 109         q->backing_dev_info->ra_pages = ra_kb >> (PAGE_SHIFT - 10);
 110
 111         return ret;
 112 }
 113
 114 static ssize_t queue_max_sectors_show(struct request_queue *q, char *page)
 115 {
 116         int max_sectors_kb = queue_max_sectors(q) >> 1;
 117
 118         return queue_var_show(max_sectors_kb, (page));
 119 }
 120
 121 static ssize_t queue_max_segments_show(struct request_queue *q, char *page)
 122 {
 123         return queue_var_show(queue_max_segments(q), (page));
 124 }
 125
 126 static ssize_t queue_max_discard_segments_show(struct request_queue *q,
 127                 char *page)
 128 {
 129         return queue_var_show(queue_max_discard_segments(q), (page));
 130 }
 131
 132 static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char *page)
 133 {
 134         return queue_var_show(q->limits.max_integrity_segments, (page));
 135 }
 136
 137 static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page)
 138 {
 139         if (blk_queue_cluster(q))
 140                 return queue_var_show(queue_max_segment_size(q), (page));
 141
 142         return queue_var_show(PAGE_SIZE, (page));
 143 }
 144
 145 static ssize_t queue_logical_block_size_show(struct request_queue *q, char *page)
 146 {
 147         return queue_var_show(queue_logical_block_size(q), page);
 148 }
 149
 150 static ssize_t queue_physical_block_size_show(struct request_queue *q, char *page)
 151 {
 152         return queue_var_show(queue_physical_block_size(q), page);
 153 }
 154
 155 static ssize_t queue_chunk_sectors_show(struct request_queue *q, char *page)
 156 {
 157         return queue_var_show(q->limits.chunk_sectors, page);
 158 }
 159
 160 static ssize_t queue_io_min_show(struct request_queue *q, char *page)
 161 {
 162         return queue_var_show(queue_io_min(q), page);
 163 }
 164
 165 static ssize_t queue_io_opt_show(struct request_queue *q, char *page)
 166 {
 167         return queue_var_show(queue_io_opt(q), page);
 168 }
 169
 170 static ssize_t queue_discard_granularity_show(struct request_queue *q, char *page)
 171 {
 172         return queue_var_show(q->limits.discard_granularity, page);
 173 }
 174
 175 static ssize_t queue_discard_max_hw_show(struct request_queue *q, char *page)
 176 {
 177
 178         return sprintf(page, "%llu\n",
 179                 (unsigned long long)q->limits.max_hw_discard_sectors << 9);
 180 }
 181
 182 static ssize_t queue_discard_max_show(struct request_queue *q, char *page)
 183 {
 184         return sprintf(page, "%llu\n",
 185                        (unsigned long long)q->limits.max_discard_sectors << 9);
 186 }
 187
 188 static ssize_t queue_discard_max_store(struct request_queue *q,
 189                                        const char *page, size_t count)
 190 {
 191         unsigned long max_discard;
 192         ssize_t ret = queue_var_store(&max_discard, page, count);
 193
 194         if (ret < 0)
 195                 return ret;
 196
 197         if (max_discard & (q->limits.discard_granularity - 1))
 198                 return -EINVAL;
 199
 200         max_discard >>= 9;
 201         if (max_discard > UINT_MAX)
 202                 return -EINVAL;
 203
 204         if (max_discard > q->limits.max_hw_discard_sectors)
 205                 max_discard = q->limits.max_hw_discard_sectors;
 206
 207         q->limits.max_discard_sectors = max_discard;
 208         return ret;
 209 }
 210
 211 static ssize_t queue_discard_zeroes_data_show(struct request_queue *q, char *page)
 212 {
 213         return queue_var_show(0, page);
 214 }
 215
 216 static ssize_t queue_write_same_max_show(struct request_queue *q, char *page)
 217 {
 218         return sprintf(page, "%llu\n",
 219                 (unsigned long long)q->limits.max_write_same_sectors << 9);
 220 }
 221
 222 static ssize_t queue_write_zeroes_max_show(struct request_queue *q, char *page)
 223 {
 224         return sprintf(page, "%llu\n",
 225                 (unsigned long long)q->limits.max_write_zeroes_sectors << 9);
 226 }
 227
 228 static ssize_t
 229 queue_max_sectors_store(struct request_queue *q, const char *page, size_t count)
 230 {
 231         unsigned long max_sectors_kb,
 232                 max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1,
 233                         page_kb = 1 << (PAGE_SHIFT - 10);
 234         ssize_t ret = queue_var_store(&max_sectors_kb, page, count);
 235
 236         if (ret < 0)
 237                 return ret;
 238
 239         max_hw_sectors_kb = min_not_zero(max_hw_sectors_kb, (unsigned long)
 240                                          q->limits.max_dev_sectors >> 1);
 241
 242         if (max_sectors_kb > max_hw_sectors_kb || max_sectors_kb < page_kb)
 243                 return -EINVAL;
 244
 245         spin_lock_irq(q->queue_lock);
 246         q->limits.max_sectors = max_sectors_kb << 1;
 247         q->backing_dev_info->io_pages = max_sectors_kb >> (PAGE_SHIFT - 10);
 248         spin_unlock_irq(q->queue_lock);
 249
 250         return ret;
 251 }
 252
 253 static ssize_t queue_max_hw_sectors_show(struct request_queue *q, char *page)
 254 {
 255         int max_hw_sectors_kb = queue_max_hw_sectors(q) >> 1;
 256
 257         return queue_var_show(max_hw_sectors_kb, (page));
 258 }
 259
 260 #define QUEUE_SYSFS_BIT_FNS(name, flag, neg)                            \
 261 static ssize_t                                                          \
 262 queue_show_##name(struct request_queue *q, char *page)                  \
 263 {                                                                       \
 264         int bit;                                                        \
 265         bit = test_bit(QUEUE_FLAG_##flag, &q->queue_flags);             \
 266         return queue_var_show(neg ? !bit : bit, page);                  \
 267 }                                                                       \
 268 static ssize_t                                                          \
 269 queue_store_##name(struct request_queue *q, const char *page, size_t count) \
 270 {                                                                       \
 271         unsigned long val;                                              \
 272         ssize_t ret;                                                    \
 273         ret = queue_var_store(&val, page, count);                       \
 274         if (ret < 0)                                                    \
 275                  return ret;                                            \
 276         if (neg)                                                        \
 277                 val = !val;                                             \
 278                                                                         \
 279         if (val)                                                        \
 280                 blk_queue_flag_set(QUEUE_FLAG_##flag, q);               \
 281         else                                                            \
 282                 blk_queue_flag_clear(QUEUE_FLAG_##flag, q);             \
 283         return ret;                                                     \
 284 }
 285
 286 QUEUE_SYSFS_BIT_FNS(nonrot, NONROT, 1);
 287 QUEUE_SYSFS_BIT_FNS(random, ADD_RANDOM, 0);
 288 QUEUE_SYSFS_BIT_FNS(iostats, IO_STAT, 0);
 289 #undef QUEUE_SYSFS_BIT_FNS
 290
 291 static ssize_t queue_zoned_show(struct request_queue *q, char *page)
 292 {
 293         switch (blk_queue_zoned_model(q)) {
 294         case BLK_ZONED_HA:
 295                 return sprintf(page, "host-aware\n");
 296         case BLK_ZONED_HM:
 297                 return sprintf(page, "host-managed\n");
 298         default:
 299                 return sprintf(page, "none\n");
 300         }
 301 }
 302
 303 static ssize_t queue_nr_zones_show(struct request_queue *q, char *page)
 304 {
 305         return queue_var_show(blk_queue_nr_zones(q), page);
 306 }
 307
 308 static ssize_t queue_nomerges_show(struct request_queue *q, char *page)
 309 {
 310         return queue_var_show((blk_queue_nomerges(q) << 1) |
 311                                blk_queue_noxmerges(q), page);
 312 }
 313
 314 static ssize_t queue_nomerges_store(struct request_queue *q, const char *page,
 315                                     size_t count)
 316 {
 317         unsigned long nm;
 318         ssize_t ret = queue_var_store(&nm, page, count);
 319
 320         if (ret < 0)
 321                 return ret;
 322
 323         spin_lock_irq(q->queue_lock);
 324         queue_flag_clear(QUEUE_FLAG_NOMERGES, q);
 325         queue_flag_clear(QUEUE_FLAG_NOXMERGES, q);
 326         if (nm == 2)
 327                 queue_flag_set(QUEUE_FLAG_NOMERGES, q);
 328         else if (nm)
 329                 queue_flag_set(QUEUE_FLAG_NOXMERGES, q);
 330         spin_unlock_irq(q->queue_lock);
 331
 332         return ret;
 333 }
 334
 335 static ssize_t queue_rq_affinity_show(struct request_queue *q, char *page)
 336 {
 337         bool set = test_bit(QUEUE_FLAG_SAME_COMP, &q->queue_flags);
 338         bool force = test_bit(QUEUE_FLAG_SAME_FORCE, &q->queue_flags);
 339
 340         return queue_var_show(set << force, page);
 341 }
 342
 343 static ssize_t
 344 queue_rq_affinity_store(struct request_queue *q, const char *page, size_t count)
 345 {
 346         ssize_t ret = -EINVAL;
 347 #ifdef CONFIG_SMP
 348         unsigned long val;
 349
 350         ret = queue_var_store(&val, page, count);
 351         if (ret < 0)
 352                 return ret;
 353
 354         spin_lock_irq(q->queue_lock);
 355         if (val == 2) {
 356                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 357                 queue_flag_set(QUEUE_FLAG_SAME_FORCE, q);
 358         } else if (val == 1) {
 359                 queue_flag_set(QUEUE_FLAG_SAME_COMP, q);
 360                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 361         } else if (val == 0) {
 362                 queue_flag_clear(QUEUE_FLAG_SAME_COMP, q);
 363                 queue_flag_clear(QUEUE_FLAG_SAME_FORCE, q);
 364         }
 365         spin_unlock_irq(q->queue_lock);
 366 #endif
 367         return ret;
 368 }
 369
 370 static ssize_t queue_poll_delay_show(struct request_queue *q, char *page)
 371 {
 372         int val;
 373
 374         if (q->poll_nsec == -1)
 375                 val = -1;
 376         else
 377                 val = q->poll_nsec / 1000;
 378
 379         return sprintf(page, "%d\n", val);
 380 }
 381
 382 static ssize_t queue_poll_delay_store(struct request_queue *q, const char *page,
 383                                 size_t count)
 384 {
 385         int err, val;
 386
 387         if (!q->mq_ops || !q->mq_ops->poll)
 388                 return -EINVAL;
 389
 390         err = kstrtoint(page, 10, &val);
 391         if (err < 0)
 392                 return err;
 393
 394         if (val == -1)
 395                 q->poll_nsec = -1;
 396         else
 397                 q->poll_nsec = val * 1000;
 398
 399         return count;
 400 }
 401
 402 static ssize_t queue_poll_show(struct request_queue *q, char *page)
 403 {
 404         return queue_var_show(test_bit(QUEUE_FLAG_POLL, &q->queue_flags), page);
 405 }
 406
 407 static ssize_t queue_poll_store(struct request_queue *q, const char *page,
 408                                 size_t count)
 409 {
 410         unsigned long poll_on;
 411         ssize_t ret;
 412
 413         if (!q->mq_ops || !q->mq_ops->poll)
 414                 return -EINVAL;
 415
 416         ret = queue_var_store(&poll_on, page, count);
 417         if (ret < 0)
 418                 return ret;
 419
 420         if (poll_on)
 421                 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
 422         else
 423                 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
 424
 425         return ret;
 426 }
 427
 428 static ssize_t queue_wb_lat_show(struct request_queue *q, char *page)
 429 {
 430         if (!wbt_rq_qos(q))
 431                 return -EINVAL;
 432
 433         return sprintf(page, "%llu\n", div_u64(wbt_get_min_lat(q), 1000));
 434 }
 435
 436 static ssize_t queue_wb_lat_store(struct request_queue *q, const char *page,
 437                                   size_t count)
 438 {
 439         struct rq_qos *rqos;
 440         ssize_t ret;
 441         s64 val;
 442
 443         ret = queue_var_store64(&val, page);
 444         if (ret < 0)
 445                 return ret;
 446         if (val < -1)
 447                 return -EINVAL;
 448
 449         rqos = wbt_rq_qos(q);
 450         if (!rqos) {
 451                 ret = wbt_init(q);
 452                 if (ret)
 453                         return ret;
 454         }
 455
 456         if (val == -1)
 457                 val = wbt_default_latency_nsec(q);
 458         else if (val >= 0)
 459                 val *= 1000ULL;
 460
 461         /*
 462          * Ensure that the queue is idled, in case the latency update
 463          * ends up either enabling or disabling wbt completely. We can't
 464          * have IO inflight if that happens.
 465          */
 466         if (q->mq_ops) {
 467                 blk_mq_freeze_queue(q);
 468                 blk_mq_quiesce_queue(q);
 469         } else
 470                 blk_queue_bypass_start(q);
 471
 472         wbt_set_min_lat(q, val);
 473         wbt_update_limits(q);
 474
 475         if (q->mq_ops) {
 476                 blk_mq_unquiesce_queue(q);
 477                 blk_mq_unfreeze_queue(q);
 478         } else
 479                 blk_queue_bypass_end(q);
 480
 481         return count;
 482 }
 483
 484 static ssize_t queue_wc_show(struct request_queue *q, char *page)
 485 {
 486         if (test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 487                 return sprintf(page, "write back\n");
 488
 489         return sprintf(page, "write through\n");
 490 }
 491
 492 static ssize_t queue_wc_store(struct request_queue *q, const char *page,
 493                               size_t count)
 494 {
 495         int set = -1;
 496
 497         if (!strncmp(page, "write back", 10))
 498                 set = 1;
 499         else if (!strncmp(page, "write through", 13) ||
 500                  !strncmp(page, "none", 4))
 501                 set = 0;
 502
 503         if (set == -1)
 504                 return -EINVAL;
 505
 506         if (set)
 507                 blk_queue_flag_set(QUEUE_FLAG_WC, q);
 508         else
 509                 blk_queue_flag_clear(QUEUE_FLAG_WC, q);
 510
 511         return count;
 512 }
 513
 514 static ssize_t queue_fua_show(struct request_queue *q, char *page)
 515 {
 516         return sprintf(page, "%u\n", test_bit(QUEUE_FLAG_FUA, &q->queue_flags));
 517 }
 518
 519 static ssize_t queue_dax_show(struct request_queue *q, char *page)
 520 {
 521         return queue_var_show(blk_queue_dax(q), page);
 522 }
 523
 524 static struct queue_sysfs_entry queue_requests_entry = {
 525         .attr = {.name = "nr_requests", .mode = 0644 },
 526         .show = queue_requests_show,
 527         .store = queue_requests_store,
 528 };
 529
 530 static struct queue_sysfs_entry queue_ra_entry = {
 531         .attr = {.name = "read_ahead_kb", .mode = 0644 },
 532         .show = queue_ra_show,
 533         .store = queue_ra_store,
 534 };
 535
 536 static struct queue_sysfs_entry queue_max_sectors_entry = {
 537         .attr = {.name = "max_sectors_kb", .mode = 0644 },
 538         .show = queue_max_sectors_show,
 539         .store = queue_max_sectors_store,
 540 };
 541
 542 static struct queue_sysfs_entry queue_max_hw_sectors_entry = {
 543         .attr = {.name = "max_hw_sectors_kb", .mode = 0444 },
 544         .show = queue_max_hw_sectors_show,
 545 };
 546
 547 static struct queue_sysfs_entry queue_max_segments_entry = {
 548         .attr = {.name = "max_segments", .mode = 0444 },
 549         .show = queue_max_segments_show,
 550 };
 551
 552 static struct queue_sysfs_entry queue_max_discard_segments_entry = {
 553         .attr = {.name = "max_discard_segments", .mode = 0444 },
 554         .show = queue_max_discard_segments_show,
 555 };
 556
 557 static struct queue_sysfs_entry queue_max_integrity_segments_entry = {
 558         .attr = {.name = "max_integrity_segments", .mode = 0444 },
 559         .show = queue_max_integrity_segments_show,
 560 };
 561
 562 static struct queue_sysfs_entry queue_max_segment_size_entry = {
 563         .attr = {.name = "max_segment_size", .mode = 0444 },
 564         .show = queue_max_segment_size_show,
 565 };
 566
 567 static struct queue_sysfs_entry queue_iosched_entry = {
 568         .attr = {.name = "scheduler", .mode = 0644 },
 569         .show = elv_iosched_show,
 570         .store = elv_iosched_store,
 571 };
 572
 573 static struct queue_sysfs_entry queue_hw_sector_size_entry = {
 574         .attr = {.name = "hw_sector_size", .mode = 0444 },
 575         .show = queue_logical_block_size_show,
 576 };
 577
 578 static struct queue_sysfs_entry queue_logical_block_size_entry = {
 579         .attr = {.name = "logical_block_size", .mode = 0444 },
 580         .show = queue_logical_block_size_show,
 581 };
 582
 583 static struct queue_sysfs_entry queue_physical_block_size_entry = {
 584         .attr = {.name = "physical_block_size", .mode = 0444 },
 585         .show = queue_physical_block_size_show,
 586 };
 587
 588 static struct queue_sysfs_entry queue_chunk_sectors_entry = {
 589         .attr = {.name = "chunk_sectors", .mode = 0444 },
 590         .show = queue_chunk_sectors_show,
 591 };
 592
 593 static struct queue_sysfs_entry queue_io_min_entry = {
 594         .attr = {.name = "minimum_io_size", .mode = 0444 },
 595         .show = queue_io_min_show,
 596 };
 597
 598 static struct queue_sysfs_entry queue_io_opt_entry = {
 599         .attr = {.name = "optimal_io_size", .mode = 0444 },
 600         .show = queue_io_opt_show,
 601 };
 602
 603 static struct queue_sysfs_entry queue_discard_granularity_entry = {
 604         .attr = {.name = "discard_granularity", .mode = 0444 },
 605         .show = queue_discard_granularity_show,
 606 };
 607
 608 static struct queue_sysfs_entry queue_discard_max_hw_entry = {
 609         .attr = {.name = "discard_max_hw_bytes", .mode = 0444 },
 610         .show = queue_discard_max_hw_show,
 611 };
 612
 613 static struct queue_sysfs_entry queue_discard_max_entry = {
 614         .attr = {.name = "discard_max_bytes", .mode = 0644 },
 615         .show = queue_discard_max_show,
 616         .store = queue_discard_max_store,
 617 };
 618
 619 static struct queue_sysfs_entry queue_discard_zeroes_data_entry = {
 620         .attr = {.name = "discard_zeroes_data", .mode = 0444 },
 621         .show = queue_discard_zeroes_data_show,
 622 };
 623
 624 static struct queue_sysfs_entry queue_write_same_max_entry = {
 625         .attr = {.name = "write_same_max_bytes", .mode = 0444 },
 626         .show = queue_write_same_max_show,
 627 };
 628
 629 static struct queue_sysfs_entry queue_write_zeroes_max_entry = {
 630         .attr = {.name = "write_zeroes_max_bytes", .mode = 0444 },
 631         .show = queue_write_zeroes_max_show,
 632 };
 633
 634 static struct queue_sysfs_entry queue_nonrot_entry = {
 635         .attr = {.name = "rotational", .mode = 0644 },
 636         .show = queue_show_nonrot,
 637         .store = queue_store_nonrot,
 638 };
 639
 640 static struct queue_sysfs_entry queue_zoned_entry = {
 641         .attr = {.name = "zoned", .mode = 0444 },
 642         .show = queue_zoned_show,
 643 };
 644
 645 static struct queue_sysfs_entry queue_nr_zones_entry = {
 646         .attr = {.name = "nr_zones", .mode = 0444 },
 647         .show = queue_nr_zones_show,
 648 };
 649
 650 static struct queue_sysfs_entry queue_nomerges_entry = {
 651         .attr = {.name = "nomerges", .mode = 0644 },
 652         .show = queue_nomerges_show,
 653         .store = queue_nomerges_store,
 654 };
 655
 656 static struct queue_sysfs_entry queue_rq_affinity_entry = {
 657         .attr = {.name = "rq_affinity", .mode = 0644 },
 658         .show = queue_rq_affinity_show,
 659         .store = queue_rq_affinity_store,
 660 };
 661
 662 static struct queue_sysfs_entry queue_iostats_entry = {
 663         .attr = {.name = "iostats", .mode = 0644 },
 664         .show = queue_show_iostats,
 665         .store = queue_store_iostats,
 666 };
 667
 668 static struct queue_sysfs_entry queue_random_entry = {
 669         .attr = {.name = "add_random", .mode = 0644 },
 670         .show = queue_show_random,
 671         .store = queue_store_random,
 672 };
 673
 674 static struct queue_sysfs_entry queue_poll_entry = {
 675         .attr = {.name = "io_poll", .mode = 0644 },
 676         .show = queue_poll_show,
 677         .store = queue_poll_store,
 678 };
 679
 680 static struct queue_sysfs_entry queue_poll_delay_entry = {
 681         .attr = {.name = "io_poll_delay", .mode = 0644 },
 682         .show = queue_poll_delay_show,
 683         .store = queue_poll_delay_store,
 684 };
 685
 686 static struct queue_sysfs_entry queue_wc_entry = {
 687         .attr = {.name = "write_cache", .mode = 0644 },
 688         .show = queue_wc_show,
 689         .store = queue_wc_store,
 690 };
 691
 692 static struct queue_sysfs_entry queue_fua_entry = {
 693         .attr = {.name = "fua", .mode = 0444 },
 694         .show = queue_fua_show,
 695 };
 696
 697 static struct queue_sysfs_entry queue_dax_entry = {
 698         .attr = {.name = "dax", .mode = 0444 },
 699         .show = queue_dax_show,
 700 };
 701
 702 static struct queue_sysfs_entry queue_wb_lat_entry = {
 703         .attr = {.name = "wbt_lat_usec", .mode = 0644 },
 704         .show = queue_wb_lat_show,
 705         .store = queue_wb_lat_store,
 706 };
 707
 708 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 709 static struct queue_sysfs_entry throtl_sample_time_entry = {
 710         .attr = {.name = "throttle_sample_time", .mode = 0644 },
 711         .show = blk_throtl_sample_time_show,
 712         .store = blk_throtl_sample_time_store,
 713 };
 714 #endif
 715
 716 static struct attribute *default_attrs[] = {
 717         &queue_requests_entry.attr,
 718         &queue_ra_entry.attr,
 719         &queue_max_hw_sectors_entry.attr,
 720         &queue_max_sectors_entry.attr,
 721         &queue_max_segments_entry.attr,
 722         &queue_max_discard_segments_entry.attr,
 723         &queue_max_integrity_segments_entry.attr,
 724         &queue_max_segment_size_entry.attr,
 725         &queue_iosched_entry.attr,
 726         &queue_hw_sector_size_entry.attr,
 727         &queue_logical_block_size_entry.attr,
 728         &queue_physical_block_size_entry.attr,
 729         &queue_chunk_sectors_entry.attr,
 730         &queue_io_min_entry.attr,
 731         &queue_io_opt_entry.attr,
 732         &queue_discard_granularity_entry.attr,
 733         &queue_discard_max_entry.attr,
 734         &queue_discard_max_hw_entry.attr,
 735         &queue_discard_zeroes_data_entry.attr,
 736         &queue_write_same_max_entry.attr,
 737         &queue_write_zeroes_max_entry.attr,
 738         &queue_nonrot_entry.attr,
 739         &queue_zoned_entry.attr,
 740         &queue_nr_zones_entry.attr,
 741         &queue_nomerges_entry.attr,
 742         &queue_rq_affinity_entry.attr,
 743         &queue_iostats_entry.attr,
 744         &queue_random_entry.attr,
 745         &queue_poll_entry.attr,
 746         &queue_wc_entry.attr,
 747         &queue_fua_entry.attr,
 748         &queue_dax_entry.attr,
 749         &queue_wb_lat_entry.attr,
 750         &queue_poll_delay_entry.attr,
 751 #ifdef CONFIG_BLK_DEV_THROTTLING_LOW
 752         &throtl_sample_time_entry.attr,
 753 #endif
 754         NULL,
 755 };
 756
 757 #define to_queue(atr) container_of((atr), struct queue_sysfs_entry, attr)
 758
 759 static ssize_t
 760 queue_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
 761 {
 762         struct queue_sysfs_entry *entry = to_queue(attr);
 763         struct request_queue *q =
 764                 container_of(kobj, struct request_queue, kobj);
 765         ssize_t res;
 766
 767         if (!entry->show)
 768                 return -EIO;
 769         mutex_lock(&q->sysfs_lock);
 770         if (blk_queue_dying(q)) {
 771                 mutex_unlock(&q->sysfs_lock);
 772                 return -ENOENT;
 773         }
 774         res = entry->show(q, page);
 775         mutex_unlock(&q->sysfs_lock);
 776         return res;
 777 }
 778
 779 static ssize_t
 780 queue_attr_store(struct kobject *kobj, struct attribute *attr,
 781                     const char *page, size_t length)
 782 {
 783         struct queue_sysfs_entry *entry = to_queue(attr);
 784         struct request_queue *q;
 785         ssize_t res;
 786
 787         if (!entry->store)
 788                 return -EIO;
 789
 790         q = container_of(kobj, struct request_queue, kobj);
 791         mutex_lock(&q->sysfs_lock);
 792         if (blk_queue_dying(q)) {
 793                 mutex_unlock(&q->sysfs_lock);
 794                 return -ENOENT;
 795         }
 796         res = entry->store(q, page, length);
 797         mutex_unlock(&q->sysfs_lock);
 798         return res;
 799 }
 800
 801 static void blk_free_queue_rcu(struct rcu_head *rcu_head)
 802 {
 803         struct request_queue *q = container_of(rcu_head, struct request_queue,
 804                                                rcu_head);
 805         kmem_cache_free(blk_requestq_cachep, q);
 806 }
 807
 808 /**
 809  * __blk_release_queue - release a request queue when it is no longer needed
 810  * @work: pointer to the release_work member of the request queue to be released
 811  *
 812  * Description:
 813  *     blk_release_queue is the counterpart of blk_init_queue(). It should be
 814  *     called when a request queue is being released; typically when a block
 815  *     device is being de-registered. Its primary task it to free the queue
 816  *     itself.
 817  *
 818  * Notes:
 819  *     The low level driver must have finished any outstanding requests first
 820  *     via blk_cleanup_queue().
 821  *
 822  *     Although blk_release_queue() may be called with preemption disabled,
 823  *     __blk_release_queue() may sleep.
 824  */
 825 static void __blk_release_queue(struct work_struct *work)
 826 {
 827         struct request_queue *q = container_of(work, typeof(*q), release_work);
 828
 829         if (test_bit(QUEUE_FLAG_POLL_STATS, &q->queue_flags))
 830                 blk_stat_remove_callback(q, q->poll_cb);
 831         blk_stat_free_callback(q->poll_cb);
 832
 833         if (!blk_queue_dead(q)) {
 834                 /*
 835                  * Last reference was dropped without having called
 836                  * blk_cleanup_queue().
 837                  */
 838                 WARN_ONCE(blk_queue_init_done(q),
 839                           "request queue %p has been registered but blk_cleanup_queue() has not been called for that queue\n",
 840                           q);
 841                 blk_exit_queue(q);
 842         }
 843
 844         WARN(blk_queue_root_blkg(q),
 845              "request queue %p is being released but it has not yet been removed from the blkcg controller\n",
 846              q);
 847
 848         blk_free_queue_stats(q->stats);
 849
 850         blk_exit_rl(q, &q->root_rl);
 851
 852         if (q->queue_tags)
 853                 __blk_queue_free_tags(q);
 854
 855         blk_queue_free_zone_bitmaps(q);
 856
 857         if (!q->mq_ops) {
 858                 if (q->exit_rq_fn)
 859                         q->exit_rq_fn(q, q->fq->flush_rq);
 860                 blk_free_flush_queue(q->fq);
 861         } else {
 862                 blk_mq_release(q);
 863         }
 864
 865         blk_trace_shutdown(q);
 866
 867         if (q->mq_ops)
 868                 blk_mq_debugfs_unregister(q);
 869
 870         bioset_exit(&q->bio_split);
 871
 872         ida_simple_remove(&blk_queue_ida, q->id);
 873         call_rcu(&q->rcu_head, blk_free_queue_rcu);
 874 }
 875
 876 static void blk_release_queue(struct kobject *kobj)
 877 {
 878         struct request_queue *q =
 879                 container_of(kobj, struct request_queue, kobj);
 880
 881         INIT_WORK(&q->release_work, __blk_release_queue);
 882         schedule_work(&q->release_work);
 883 }
 884
 885 static const struct sysfs_ops queue_sysfs_ops = {
 886         .show   = queue_attr_show,
 887         .store  = queue_attr_store,
 888 };
 889
 890 struct kobj_type blk_queue_ktype = {
 891         .sysfs_ops      = &queue_sysfs_ops,
 892         .default_attrs  = default_attrs,
 893         .release        = blk_release_queue,
 894 };
 895
 896 /**
 897  * blk_register_queue - register a block layer queue with sysfs
 898  * @disk: Disk of which the request queue should be registered with sysfs.
 899  */
 900 int blk_register_queue(struct gendisk *disk)
 901 {
 902         int ret;
 903         struct device *dev = disk_to_dev(disk);
 904         struct request_queue *q = disk->queue;
 905
 906         if (WARN_ON(!q))
 907                 return -ENXIO;
 908
 909         WARN_ONCE(test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags),
 910                   "%s is registering an already registered queue\n",
 911                   kobject_name(&dev->kobj));
 912         queue_flag_set_unlocked(QUEUE_FLAG_REGISTERED, q);
 913
 914         /*
 915          * SCSI probing may synchronously create and destroy a lot of
 916          * request_queues for non-existent devices.  Shutting down a fully
 917          * functional queue takes measureable wallclock time as RCU grace
 918          * periods are involved.  To avoid excessive latency in these
 919          * cases, a request_queue starts out in a degraded mode which is
 920          * faster to shut down and is made fully functional here as
 921          * request_queues for non-existent devices never get registered.
 922          */
 923         if (!blk_queue_init_done(q)) {
 924                 queue_flag_set_unlocked(QUEUE_FLAG_INIT_DONE, q);
 925                 percpu_ref_switch_to_percpu(&q->q_usage_counter);
 926                 blk_queue_bypass_end(q);
 927         }
 928
 929         ret = blk_trace_init_sysfs(dev);
 930         if (ret)
 931                 return ret;
 932
 933         /* Prevent changes through sysfs until registration is completed. */
 934         mutex_lock(&q->sysfs_lock);
 935
 936         ret = kobject_add(&q->kobj, kobject_get(&dev->kobj), "%s", "queue");
 937         if (ret < 0) {
 938                 blk_trace_remove_sysfs(dev);
 939                 goto unlock;
 940         }
 941
 942         if (q->mq_ops) {
 943                 __blk_mq_register_dev(dev, q);
 944                 blk_mq_debugfs_register(q);
 945         }
 946
 947         kobject_uevent(&q->kobj, KOBJ_ADD);
 948
 949         wbt_enable_default(q);
 950
 951         blk_throtl_register_queue(q);
 952
 953         if (q->request_fn || (q->mq_ops && q->elevator)) {
 954                 ret = elv_register_queue(q);
 955                 if (ret) {
 956                         mutex_unlock(&q->sysfs_lock);
 957                         kobject_uevent(&q->kobj, KOBJ_REMOVE);
 958                         kobject_del(&q->kobj);
 959                         blk_trace_remove_sysfs(dev);
 960                         kobject_put(&dev->kobj);
 961                         return ret;
 962                 }
 963         }
 964         ret = 0;
 965 unlock:
 966         mutex_unlock(&q->sysfs_lock);
 967         return ret;
 968 }
 969 EXPORT_SYMBOL_GPL(blk_register_queue);
 970
 971 /**
 972  * blk_unregister_queue - counterpart of blk_register_queue()
 973  * @disk: Disk of which the request queue should be unregistered from sysfs.
 974  *
 975  * Note: the caller is responsible for guaranteeing that this function is called
 976  * after blk_register_queue() has finished.
 977  */
 978 void blk_unregister_queue(struct gendisk *disk)
 979 {
 980         struct request_queue *q = disk->queue;
 981
 982         if (WARN_ON(!q))
 983                 return;
 984
 985         /* Return early if disk->queue was never registered. */
 986         if (!test_bit(QUEUE_FLAG_REGISTERED, &q->queue_flags))
 987                 return;
 988
 989         /*
 990          * Since sysfs_remove_dir() prevents adding new directory entries
 991          * before removal of existing entries starts, protect against
 992          * concurrent elv_iosched_store() calls.
 993          */
 994         mutex_lock(&q->sysfs_lock);
 995
 996         blk_queue_flag_clear(QUEUE_FLAG_REGISTERED, q);
 997
 998         /*
 999          * Remove the sysfs attributes before unregistering the queue data
1000          * structures that can be modified through sysfs.
1001          */
1002         if (q->mq_ops)
1003                 blk_mq_unregister_dev(disk_to_dev(disk), q);
1004         mutex_unlock(&q->sysfs_lock);
1005
1006         kobject_uevent(&q->kobj, KOBJ_REMOVE);
1007         kobject_del(&q->kobj);
1008         blk_trace_remove_sysfs(disk_to_dev(disk));
1009
1010         mutex_lock(&q->sysfs_lock);
1011         if (q->request_fn || (q->mq_ops && q->elevator))
1012                 elv_unregister_queue(q);
1013         mutex_unlock(&q->sysfs_lock);
1014
1015         kobject_put(&disk_to_dev(disk)->kobj);
1016 }