block/blk-mq-debugfs.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2017 Facebook
   4  */
   5
   6 #include <linux/kernel.h>
   7 #include <linux/blkdev.h>
   8 #include <linux/debugfs.h>
   9
  10 #include <linux/blk-mq.h>
  11 #include "blk.h"
  12 #include "blk-mq.h"
  13 #include "blk-mq-debugfs.h"
  14 #include "blk-mq-tag.h"
  15 #include "blk-rq-qos.h"
  16
  17 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
  18 {
  19         if (stat->nr_samples) {
  20                 seq_printf(m, "samples=%d, mean=%llu, min=%llu, max=%llu",
  21                            stat->nr_samples, stat->mean, stat->min, stat->max);
  22         } else {
  23                 seq_puts(m, "samples=0");
  24         }
  25 }
  26
  27 static int queue_poll_stat_show(void *data, struct seq_file *m)
  28 {
  29         struct request_queue *q = data;
  30         int bucket;
  31
  32         for (bucket = 0; bucket < (BLK_MQ_POLL_STATS_BKTS / 2); bucket++) {
  33                 seq_printf(m, "read  (%d Bytes): ", 1 << (9 + bucket));
  34                 print_stat(m, &q->poll_stat[2 * bucket]);
  35                 seq_puts(m, "\n");
  36
  37                 seq_printf(m, "write (%d Bytes): ",  1 << (9 + bucket));
  38                 print_stat(m, &q->poll_stat[2 * bucket + 1]);
  39                 seq_puts(m, "\n");
  40         }
  41         return 0;
  42 }
  43
  44 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
  45         __acquires(&q->requeue_lock)
  46 {
  47         struct request_queue *q = m->private;
  48
  49         spin_lock_irq(&q->requeue_lock);
  50         return seq_list_start(&q->requeue_list, *pos);
  51 }
  52
  53 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
  54 {
  55         struct request_queue *q = m->private;
  56
  57         return seq_list_next(v, &q->requeue_list, pos);
  58 }
  59
  60 static void queue_requeue_list_stop(struct seq_file *m, void *v)
  61         __releases(&q->requeue_lock)
  62 {
  63         struct request_queue *q = m->private;
  64
  65         spin_unlock_irq(&q->requeue_lock);
  66 }
  67
  68 static const struct seq_operations queue_requeue_list_seq_ops = {
  69         .start  = queue_requeue_list_start,
  70         .next   = queue_requeue_list_next,
  71         .stop   = queue_requeue_list_stop,
  72         .show   = blk_mq_debugfs_rq_show,
  73 };
  74
  75 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
  76                           const char *const *flag_name, int flag_name_count)
  77 {
  78         bool sep = false;
  79         int i;
  80
  81         for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
  82                 if (!(flags & BIT(i)))
  83                         continue;
  84                 if (sep)
  85                         seq_puts(m, "|");
  86                 sep = true;
  87                 if (i < flag_name_count && flag_name[i])
  88                         seq_puts(m, flag_name[i]);
  89                 else
  90                         seq_printf(m, "%d", i);
  91         }
  92         return 0;
  93 }
  94
  95 static int queue_pm_only_show(void *data, struct seq_file *m)
  96 {
  97         struct request_queue *q = data;
  98
  99         seq_printf(m, "%d\n", atomic_read(&q->pm_only));
 100         return 0;
 101 }
 102
 103 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
 104 static const char *const blk_queue_flag_name[] = {
 105         QUEUE_FLAG_NAME(STOPPED),
 106         QUEUE_FLAG_NAME(DYING),
 107         QUEUE_FLAG_NAME(NOMERGES),
 108         QUEUE_FLAG_NAME(SAME_COMP),
 109         QUEUE_FLAG_NAME(FAIL_IO),
 110         QUEUE_FLAG_NAME(NONROT),
 111         QUEUE_FLAG_NAME(IO_STAT),
 112         QUEUE_FLAG_NAME(DISCARD),
 113         QUEUE_FLAG_NAME(NOXMERGES),
 114         QUEUE_FLAG_NAME(ADD_RANDOM),
 115         QUEUE_FLAG_NAME(SECERASE),
 116         QUEUE_FLAG_NAME(SAME_FORCE),
 117         QUEUE_FLAG_NAME(DEAD),
 118         QUEUE_FLAG_NAME(INIT_DONE),
 119         QUEUE_FLAG_NAME(STABLE_WRITES),
 120         QUEUE_FLAG_NAME(POLL),
 121         QUEUE_FLAG_NAME(WC),
 122         QUEUE_FLAG_NAME(FUA),
 123         QUEUE_FLAG_NAME(DAX),
 124         QUEUE_FLAG_NAME(STATS),
 125         QUEUE_FLAG_NAME(POLL_STATS),
 126         QUEUE_FLAG_NAME(REGISTERED),
 127         QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
 128         QUEUE_FLAG_NAME(QUIESCED),
 129         QUEUE_FLAG_NAME(PCI_P2PDMA),
 130         QUEUE_FLAG_NAME(ZONE_RESETALL),
 131         QUEUE_FLAG_NAME(RQ_ALLOC_TIME),
 132         QUEUE_FLAG_NAME(NOWAIT),
 133 };
 134 #undef QUEUE_FLAG_NAME
 135
 136 static int queue_state_show(void *data, struct seq_file *m)
 137 {
 138         struct request_queue *q = data;
 139
 140         blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
 141                        ARRAY_SIZE(blk_queue_flag_name));
 142         seq_puts(m, "\n");
 143         return 0;
 144 }
 145
 146 static ssize_t queue_state_write(void *data, const char __user *buf,
 147                                  size_t count, loff_t *ppos)
 148 {
 149         struct request_queue *q = data;
 150         char opbuf[16] = { }, *op;
 151
 152         /*
 153          * The "state" attribute is removed after blk_cleanup_queue() has called
 154          * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
 155          * triggering a use-after-free.
 156          */
 157         if (blk_queue_dead(q))
 158                 return -ENOENT;
 159
 160         if (count >= sizeof(opbuf)) {
 161                 pr_err("%s: operation too long\n", __func__);
 162                 goto inval;
 163         }
 164
 165         if (copy_from_user(opbuf, buf, count))
 166                 return -EFAULT;
 167         op = strstrip(opbuf);
 168         if (strcmp(op, "run") == 0) {
 169                 blk_mq_run_hw_queues(q, true);
 170         } else if (strcmp(op, "start") == 0) {
 171                 blk_mq_start_stopped_hw_queues(q, true);
 172         } else if (strcmp(op, "kick") == 0) {
 173                 blk_mq_kick_requeue_list(q);
 174         } else {
 175                 pr_err("%s: unsupported operation '%s'\n", __func__, op);
 176 inval:
 177                 pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
 178                 return -EINVAL;
 179         }
 180         return count;
 181 }
 182
 183 static int queue_write_hint_show(void *data, struct seq_file *m)
 184 {
 185         struct request_queue *q = data;
 186         int i;
 187
 188         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 189                 seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
 190
 191         return 0;
 192 }
 193
 194 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
 195                                       size_t count, loff_t *ppos)
 196 {
 197         struct request_queue *q = data;
 198         int i;
 199
 200         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 201                 q->write_hints[i] = 0;
 202
 203         return count;
 204 }
 205
 206 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
 207         { "poll_stat", 0400, queue_poll_stat_show },
 208         { "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
 209         { "pm_only", 0600, queue_pm_only_show, NULL },
 210         { "state", 0600, queue_state_show, queue_state_write },
 211         { "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
 212         { "zone_wlock", 0400, queue_zone_wlock_show, NULL },
 213         { },
 214 };
 215
 216 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
 217 static const char *const hctx_state_name[] = {
 218         HCTX_STATE_NAME(STOPPED),
 219         HCTX_STATE_NAME(TAG_ACTIVE),
 220         HCTX_STATE_NAME(SCHED_RESTART),
 221         HCTX_STATE_NAME(INACTIVE),
 222 };
 223 #undef HCTX_STATE_NAME
 224
 225 static int hctx_state_show(void *data, struct seq_file *m)
 226 {
 227         struct blk_mq_hw_ctx *hctx = data;
 228
 229         blk_flags_show(m, hctx->state, hctx_state_name,
 230                        ARRAY_SIZE(hctx_state_name));
 231         seq_puts(m, "\n");
 232         return 0;
 233 }
 234
 235 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
 236 static const char *const alloc_policy_name[] = {
 237         BLK_TAG_ALLOC_NAME(FIFO),
 238         BLK_TAG_ALLOC_NAME(RR),
 239 };
 240 #undef BLK_TAG_ALLOC_NAME
 241
 242 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
 243 static const char *const hctx_flag_name[] = {
 244         HCTX_FLAG_NAME(SHOULD_MERGE),
 245         HCTX_FLAG_NAME(TAG_QUEUE_SHARED),
 246         HCTX_FLAG_NAME(BLOCKING),
 247         HCTX_FLAG_NAME(NO_SCHED),
 248         HCTX_FLAG_NAME(STACKING),
 249         HCTX_FLAG_NAME(TAG_HCTX_SHARED),
 250 };
 251 #undef HCTX_FLAG_NAME
 252
 253 static int hctx_flags_show(void *data, struct seq_file *m)
 254 {
 255         struct blk_mq_hw_ctx *hctx = data;
 256         const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
 257
 258         seq_puts(m, "alloc_policy=");
 259         if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
 260             alloc_policy_name[alloc_policy])
 261                 seq_puts(m, alloc_policy_name[alloc_policy]);
 262         else
 263                 seq_printf(m, "%d", alloc_policy);
 264         seq_puts(m, " ");
 265         blk_flags_show(m,
 266                        hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
 267                        hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
 268         seq_puts(m, "\n");
 269         return 0;
 270 }
 271
 272 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
 273 static const char *const cmd_flag_name[] = {
 274         CMD_FLAG_NAME(FAILFAST_DEV),
 275         CMD_FLAG_NAME(FAILFAST_TRANSPORT),
 276         CMD_FLAG_NAME(FAILFAST_DRIVER),
 277         CMD_FLAG_NAME(SYNC),
 278         CMD_FLAG_NAME(META),
 279         CMD_FLAG_NAME(PRIO),
 280         CMD_FLAG_NAME(NOMERGE),
 281         CMD_FLAG_NAME(IDLE),
 282         CMD_FLAG_NAME(INTEGRITY),
 283         CMD_FLAG_NAME(FUA),
 284         CMD_FLAG_NAME(PREFLUSH),
 285         CMD_FLAG_NAME(RAHEAD),
 286         CMD_FLAG_NAME(BACKGROUND),
 287         CMD_FLAG_NAME(NOWAIT),
 288         CMD_FLAG_NAME(NOUNMAP),
 289         CMD_FLAG_NAME(HIPRI),
 290 };
 291 #undef CMD_FLAG_NAME
 292
 293 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
 294 static const char *const rqf_name[] = {
 295         RQF_NAME(STARTED),
 296         RQF_NAME(SOFTBARRIER),
 297         RQF_NAME(FLUSH_SEQ),
 298         RQF_NAME(MIXED_MERGE),
 299         RQF_NAME(MQ_INFLIGHT),
 300         RQF_NAME(DONTPREP),
 301         RQF_NAME(FAILED),
 302         RQF_NAME(QUIET),
 303         RQF_NAME(ELVPRIV),
 304         RQF_NAME(IO_STAT),
 305         RQF_NAME(PM),
 306         RQF_NAME(HASHED),
 307         RQF_NAME(STATS),
 308         RQF_NAME(SPECIAL_PAYLOAD),
 309         RQF_NAME(ZONE_WRITE_LOCKED),
 310         RQF_NAME(MQ_POLL_SLEPT),
 311 };
 312 #undef RQF_NAME
 313
 314 static const char *const blk_mq_rq_state_name_array[] = {
 315         [MQ_RQ_IDLE]            = "idle",
 316         [MQ_RQ_IN_FLIGHT]       = "in_flight",
 317         [MQ_RQ_COMPLETE]        = "complete",
 318 };
 319
 320 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
 321 {
 322         if (WARN_ON_ONCE((unsigned int)rq_state >=
 323                          ARRAY_SIZE(blk_mq_rq_state_name_array)))
 324                 return "(?)";
 325         return blk_mq_rq_state_name_array[rq_state];
 326 }
 327
 328 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
 329 {
 330         const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
 331         const unsigned int op = req_op(rq);
 332         const char *op_str = blk_op_str(op);
 333
 334         seq_printf(m, "%p {.op=", rq);
 335         if (strcmp(op_str, "UNKNOWN") == 0)
 336                 seq_printf(m, "%u", op);
 337         else
 338                 seq_printf(m, "%s", op_str);
 339         seq_puts(m, ", .cmd_flags=");
 340         blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
 341                        ARRAY_SIZE(cmd_flag_name));
 342         seq_puts(m, ", .rq_flags=");
 343         blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
 344                        ARRAY_SIZE(rqf_name));
 345         seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
 346         seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
 347                    rq->internal_tag);
 348         if (mq_ops->show_rq)
 349                 mq_ops->show_rq(m, rq);
 350         seq_puts(m, "}\n");
 351         return 0;
 352 }
 353 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
 354
 355 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
 356 {
 357         return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
 358 }
 359 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
 360
 361 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
 362         __acquires(&hctx->lock)
 363 {
 364         struct blk_mq_hw_ctx *hctx = m->private;
 365
 366         spin_lock(&hctx->lock);
 367         return seq_list_start(&hctx->dispatch, *pos);
 368 }
 369
 370 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
 371 {
 372         struct blk_mq_hw_ctx *hctx = m->private;
 373
 374         return seq_list_next(v, &hctx->dispatch, pos);
 375 }
 376
 377 static void hctx_dispatch_stop(struct seq_file *m, void *v)
 378         __releases(&hctx->lock)
 379 {
 380         struct blk_mq_hw_ctx *hctx = m->private;
 381
 382         spin_unlock(&hctx->lock);
 383 }
 384
 385 static const struct seq_operations hctx_dispatch_seq_ops = {
 386         .start  = hctx_dispatch_start,
 387         .next   = hctx_dispatch_next,
 388         .stop   = hctx_dispatch_stop,
 389         .show   = blk_mq_debugfs_rq_show,
 390 };
 391
 392 struct show_busy_params {
 393         struct seq_file         *m;
 394         struct blk_mq_hw_ctx    *hctx;
 395 };
 396
 397 /*
 398  * Note: the state of a request may change while this function is in progress,
 399  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
 400  * keep iterating requests.
 401  */
 402 static bool hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 403 {
 404         const struct show_busy_params *params = data;
 405
 406         if (rq->mq_hctx == params->hctx)
 407                 __blk_mq_debugfs_rq_show(params->m, rq);
 408
 409         return true;
 410 }
 411
 412 static int hctx_busy_show(void *data, struct seq_file *m)
 413 {
 414         struct blk_mq_hw_ctx *hctx = data;
 415         struct show_busy_params params = { .m = m, .hctx = hctx };
 416
 417         blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
 418                                 &params);
 419
 420         return 0;
 421 }
 422
 423 static const char *const hctx_types[] = {
 424         [HCTX_TYPE_DEFAULT]     = "default",
 425         [HCTX_TYPE_READ]        = "read",
 426         [HCTX_TYPE_POLL]        = "poll",
 427 };
 428
 429 static int hctx_type_show(void *data, struct seq_file *m)
 430 {
 431         struct blk_mq_hw_ctx *hctx = data;
 432
 433         BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
 434         seq_printf(m, "%s\n", hctx_types[hctx->type]);
 435         return 0;
 436 }
 437
 438 static int hctx_ctx_map_show(void *data, struct seq_file *m)
 439 {
 440         struct blk_mq_hw_ctx *hctx = data;
 441
 442         sbitmap_bitmap_show(&hctx->ctx_map, m);
 443         return 0;
 444 }
 445
 446 static void blk_mq_debugfs_tags_show(struct seq_file *m,
 447                                      struct blk_mq_tags *tags)
 448 {
 449         seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
 450         seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
 451         seq_printf(m, "active_queues=%d\n",
 452                    atomic_read(&tags->active_queues));
 453
 454         seq_puts(m, "\nbitmap_tags:\n");
 455         sbitmap_queue_show(tags->bitmap_tags, m);
 456
 457         if (tags->nr_reserved_tags) {
 458                 seq_puts(m, "\nbreserved_tags:\n");
 459                 sbitmap_queue_show(tags->breserved_tags, m);
 460         }
 461 }
 462
 463 static int hctx_tags_show(void *data, struct seq_file *m)
 464 {
 465         struct blk_mq_hw_ctx *hctx = data;
 466         struct request_queue *q = hctx->queue;
 467         int res;
 468
 469         res = mutex_lock_interruptible(&q->sysfs_lock);
 470         if (res)
 471                 goto out;
 472         if (hctx->tags)
 473                 blk_mq_debugfs_tags_show(m, hctx->tags);
 474         mutex_unlock(&q->sysfs_lock);
 475
 476 out:
 477         return res;
 478 }
 479
 480 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
 481 {
 482         struct blk_mq_hw_ctx *hctx = data;
 483         struct request_queue *q = hctx->queue;
 484         int res;
 485
 486         res = mutex_lock_interruptible(&q->sysfs_lock);
 487         if (res)
 488                 goto out;
 489         if (hctx->tags)
 490                 sbitmap_bitmap_show(&hctx->tags->bitmap_tags->sb, m);
 491         mutex_unlock(&q->sysfs_lock);
 492
 493 out:
 494         return res;
 495 }
 496
 497 static int hctx_sched_tags_show(void *data, struct seq_file *m)
 498 {
 499         struct blk_mq_hw_ctx *hctx = data;
 500         struct request_queue *q = hctx->queue;
 501         int res;
 502
 503         res = mutex_lock_interruptible(&q->sysfs_lock);
 504         if (res)
 505                 goto out;
 506         if (hctx->sched_tags)
 507                 blk_mq_debugfs_tags_show(m, hctx->sched_tags);
 508         mutex_unlock(&q->sysfs_lock);
 509
 510 out:
 511         return res;
 512 }
 513
 514 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
 515 {
 516         struct blk_mq_hw_ctx *hctx = data;
 517         struct request_queue *q = hctx->queue;
 518         int res;
 519
 520         res = mutex_lock_interruptible(&q->sysfs_lock);
 521         if (res)
 522                 goto out;
 523         if (hctx->sched_tags)
 524                 sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags->sb, m);
 525         mutex_unlock(&q->sysfs_lock);
 526
 527 out:
 528         return res;
 529 }
 530
 531 static int hctx_io_poll_show(void *data, struct seq_file *m)
 532 {
 533         struct blk_mq_hw_ctx *hctx = data;
 534
 535         seq_printf(m, "considered=%lu\n", hctx->poll_considered);
 536         seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
 537         seq_printf(m, "success=%lu\n", hctx->poll_success);
 538         return 0;
 539 }
 540
 541 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
 542                                   size_t count, loff_t *ppos)
 543 {
 544         struct blk_mq_hw_ctx *hctx = data;
 545
 546         hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
 547         return count;
 548 }
 549
 550 static int hctx_dispatched_show(void *data, struct seq_file *m)
 551 {
 552         struct blk_mq_hw_ctx *hctx = data;
 553         int i;
 554
 555         seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
 556
 557         for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
 558                 unsigned int d = 1U << (i - 1);
 559
 560                 seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
 561         }
 562
 563         seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
 564         return 0;
 565 }
 566
 567 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
 568                                      size_t count, loff_t *ppos)
 569 {
 570         struct blk_mq_hw_ctx *hctx = data;
 571         int i;
 572
 573         for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
 574                 hctx->dispatched[i] = 0;
 575         return count;
 576 }
 577
 578 static int hctx_queued_show(void *data, struct seq_file *m)
 579 {
 580         struct blk_mq_hw_ctx *hctx = data;
 581
 582         seq_printf(m, "%lu\n", hctx->queued);
 583         return 0;
 584 }
 585
 586 static ssize_t hctx_queued_write(void *data, const char __user *buf,
 587                                  size_t count, loff_t *ppos)
 588 {
 589         struct blk_mq_hw_ctx *hctx = data;
 590
 591         hctx->queued = 0;
 592         return count;
 593 }
 594
 595 static int hctx_run_show(void *data, struct seq_file *m)
 596 {
 597         struct blk_mq_hw_ctx *hctx = data;
 598
 599         seq_printf(m, "%lu\n", hctx->run);
 600         return 0;
 601 }
 602
 603 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
 604                               loff_t *ppos)
 605 {
 606         struct blk_mq_hw_ctx *hctx = data;
 607
 608         hctx->run = 0;
 609         return count;
 610 }
 611
 612 static int hctx_active_show(void *data, struct seq_file *m)
 613 {
 614         struct blk_mq_hw_ctx *hctx = data;
 615
 616         seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
 617         return 0;
 618 }
 619
 620 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
 621 {
 622         struct blk_mq_hw_ctx *hctx = data;
 623
 624         seq_printf(m, "%u\n", hctx->dispatch_busy);
 625         return 0;
 626 }
 627
 628 #define CTX_RQ_SEQ_OPS(name, type)                                      \
 629 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
 630         __acquires(&ctx->lock)                                          \
 631 {                                                                       \
 632         struct blk_mq_ctx *ctx = m->private;                            \
 633                                                                         \
 634         spin_lock(&ctx->lock);                                          \
 635         return seq_list_start(&ctx->rq_lists[type], *pos);              \
 636 }                                                                       \
 637                                                                         \
 638 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,     \
 639                                      loff_t *pos)                       \
 640 {                                                                       \
 641         struct blk_mq_ctx *ctx = m->private;                            \
 642                                                                         \
 643         return seq_list_next(v, &ctx->rq_lists[type], pos);             \
 644 }                                                                       \
 645                                                                         \
 646 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)      \
 647         __releases(&ctx->lock)                                          \
 648 {                                                                       \
 649         struct blk_mq_ctx *ctx = m->private;                            \
 650                                                                         \
 651         spin_unlock(&ctx->lock);                                        \
 652 }                                                                       \
 653                                                                         \
 654 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {     \
 655         .start  = ctx_##name##_rq_list_start,                           \
 656         .next   = ctx_##name##_rq_list_next,                            \
 657         .stop   = ctx_##name##_rq_list_stop,                            \
 658         .show   = blk_mq_debugfs_rq_show,                               \
 659 }
 660
 661 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
 662 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
 663 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
 664
 665 static int ctx_dispatched_show(void *data, struct seq_file *m)
 666 {
 667         struct blk_mq_ctx *ctx = data;
 668
 669         seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
 670         return 0;
 671 }
 672
 673 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
 674                                     size_t count, loff_t *ppos)
 675 {
 676         struct blk_mq_ctx *ctx = data;
 677
 678         ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
 679         return count;
 680 }
 681
 682 static int ctx_merged_show(void *data, struct seq_file *m)
 683 {
 684         struct blk_mq_ctx *ctx = data;
 685
 686         seq_printf(m, "%lu\n", ctx->rq_merged);
 687         return 0;
 688 }
 689
 690 static ssize_t ctx_merged_write(void *data, const char __user *buf,
 691                                 size_t count, loff_t *ppos)
 692 {
 693         struct blk_mq_ctx *ctx = data;
 694
 695         ctx->rq_merged = 0;
 696         return count;
 697 }
 698
 699 static int ctx_completed_show(void *data, struct seq_file *m)
 700 {
 701         struct blk_mq_ctx *ctx = data;
 702
 703         seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
 704         return 0;
 705 }
 706
 707 static ssize_t ctx_completed_write(void *data, const char __user *buf,
 708                                    size_t count, loff_t *ppos)
 709 {
 710         struct blk_mq_ctx *ctx = data;
 711
 712         ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
 713         return count;
 714 }
 715
 716 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
 717 {
 718         const struct blk_mq_debugfs_attr *attr = m->private;
 719         void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
 720
 721         return attr->show(data, m);
 722 }
 723
 724 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
 725                                     size_t count, loff_t *ppos)
 726 {
 727         struct seq_file *m = file->private_data;
 728         const struct blk_mq_debugfs_attr *attr = m->private;
 729         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 730
 731         /*
 732          * Attributes that only implement .seq_ops are read-only and 'attr' is
 733          * the same with 'data' in this case.
 734          */
 735         if (attr == data || !attr->write)
 736                 return -EPERM;
 737
 738         return attr->write(data, buf, count, ppos);
 739 }
 740
 741 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
 742 {
 743         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 744         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 745         struct seq_file *m;
 746         int ret;
 747
 748         if (attr->seq_ops) {
 749                 ret = seq_open(file, attr->seq_ops);
 750                 if (!ret) {
 751                         m = file->private_data;
 752                         m->private = data;
 753                 }
 754                 return ret;
 755         }
 756
 757         if (WARN_ON_ONCE(!attr->show))
 758                 return -EPERM;
 759
 760         return single_open(file, blk_mq_debugfs_show, inode->i_private);
 761 }
 762
 763 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
 764 {
 765         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 766
 767         if (attr->show)
 768                 return single_release(inode, file);
 769
 770         return seq_release(inode, file);
 771 }
 772
 773 static const struct file_operations blk_mq_debugfs_fops = {
 774         .open           = blk_mq_debugfs_open,
 775         .read           = seq_read,
 776         .write          = blk_mq_debugfs_write,
 777         .llseek         = seq_lseek,
 778         .release        = blk_mq_debugfs_release,
 779 };
 780
 781 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
 782         {"state", 0400, hctx_state_show},
 783         {"flags", 0400, hctx_flags_show},
 784         {"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
 785         {"busy", 0400, hctx_busy_show},
 786         {"ctx_map", 0400, hctx_ctx_map_show},
 787         {"tags", 0400, hctx_tags_show},
 788         {"tags_bitmap", 0400, hctx_tags_bitmap_show},
 789         {"sched_tags", 0400, hctx_sched_tags_show},
 790         {"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
 791         {"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
 792         {"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
 793         {"queued", 0600, hctx_queued_show, hctx_queued_write},
 794         {"run", 0600, hctx_run_show, hctx_run_write},
 795         {"active", 0400, hctx_active_show},
 796         {"dispatch_busy", 0400, hctx_dispatch_busy_show},
 797         {"type", 0400, hctx_type_show},
 798         {},
 799 };
 800
 801 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
 802         {"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
 803         {"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
 804         {"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
 805         {"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
 806         {"merged", 0600, ctx_merged_show, ctx_merged_write},
 807         {"completed", 0600, ctx_completed_show, ctx_completed_write},
 808         {},
 809 };
 810
 811 static void debugfs_create_files(struct dentry *parent, void *data,
 812                                  const struct blk_mq_debugfs_attr *attr)
 813 {
 814         if (IS_ERR_OR_NULL(parent))
 815                 return;
 816
 817         d_inode(parent)->i_private = data;
 818
 819         for (; attr->name; attr++)
 820                 debugfs_create_file(attr->name, attr->mode, parent,
 821                                     (void *)attr, &blk_mq_debugfs_fops);
 822 }
 823
 824 void blk_mq_debugfs_register(struct request_queue *q)
 825 {
 826         struct blk_mq_hw_ctx *hctx;
 827         int i;
 828
 829         debugfs_create_files(q->debugfs_dir, q, blk_mq_debugfs_queue_attrs);
 830
 831         /*
 832          * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
 833          * didn't exist yet (because we don't know what to name the directory
 834          * until the queue is registered to a gendisk).
 835          */
 836         if (q->elevator && !q->sched_debugfs_dir)
 837                 blk_mq_debugfs_register_sched(q);
 838
 839         /* Similarly, blk_mq_init_hctx() couldn't do this previously. */
 840         queue_for_each_hw_ctx(q, hctx, i) {
 841                 if (!hctx->debugfs_dir)
 842                         blk_mq_debugfs_register_hctx(q, hctx);
 843                 if (q->elevator && !hctx->sched_debugfs_dir)
 844                         blk_mq_debugfs_register_sched_hctx(q, hctx);
 845         }
 846
 847         if (q->rq_qos) {
 848                 struct rq_qos *rqos = q->rq_qos;
 849
 850                 while (rqos) {
 851                         blk_mq_debugfs_register_rqos(rqos);
 852                         rqos = rqos->next;
 853                 }
 854         }
 855 }
 856
 857 void blk_mq_debugfs_unregister(struct request_queue *q)
 858 {
 859         q->sched_debugfs_dir = NULL;
 860 }
 861
 862 static void blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
 863                                         struct blk_mq_ctx *ctx)
 864 {
 865         struct dentry *ctx_dir;
 866         char name[20];
 867
 868         snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
 869         ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
 870
 871         debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs);
 872 }
 873
 874 void blk_mq_debugfs_register_hctx(struct request_queue *q,
 875                                   struct blk_mq_hw_ctx *hctx)
 876 {
 877         struct blk_mq_ctx *ctx;
 878         char name[20];
 879         int i;
 880
 881         snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
 882         hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
 883
 884         debugfs_create_files(hctx->debugfs_dir, hctx, blk_mq_debugfs_hctx_attrs);
 885
 886         hctx_for_each_ctx(hctx, ctx, i)
 887                 blk_mq_debugfs_register_ctx(hctx, ctx);
 888 }
 889
 890 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
 891 {
 892         debugfs_remove_recursive(hctx->debugfs_dir);
 893         hctx->sched_debugfs_dir = NULL;
 894         hctx->debugfs_dir = NULL;
 895 }
 896
 897 void blk_mq_debugfs_register_hctxs(struct request_queue *q)
 898 {
 899         struct blk_mq_hw_ctx *hctx;
 900         int i;
 901
 902         queue_for_each_hw_ctx(q, hctx, i)
 903                 blk_mq_debugfs_register_hctx(q, hctx);
 904 }
 905
 906 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
 907 {
 908         struct blk_mq_hw_ctx *hctx;
 909         int i;
 910
 911         queue_for_each_hw_ctx(q, hctx, i)
 912                 blk_mq_debugfs_unregister_hctx(hctx);
 913 }
 914
 915 void blk_mq_debugfs_register_sched(struct request_queue *q)
 916 {
 917         struct elevator_type *e = q->elevator->type;
 918
 919         /*
 920          * If the parent directory has not been created yet, return, we will be
 921          * called again later on and the directory/files will be created then.
 922          */
 923         if (!q->debugfs_dir)
 924                 return;
 925
 926         if (!e->queue_debugfs_attrs)
 927                 return;
 928
 929         q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
 930
 931         debugfs_create_files(q->sched_debugfs_dir, q, e->queue_debugfs_attrs);
 932 }
 933
 934 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
 935 {
 936         debugfs_remove_recursive(q->sched_debugfs_dir);
 937         q->sched_debugfs_dir = NULL;
 938 }
 939
 940 static const char *rq_qos_id_to_name(enum rq_qos_id id)
 941 {
 942         switch (id) {
 943         case RQ_QOS_WBT:
 944                 return "wbt";
 945         case RQ_QOS_LATENCY:
 946                 return "latency";
 947         case RQ_QOS_COST:
 948                 return "cost";
 949         case RQ_QOS_IOPRIO:
 950                 return "ioprio";
 951         }
 952         return "unknown";
 953 }
 954
 955 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
 956 {
 957         debugfs_remove_recursive(rqos->debugfs_dir);
 958         rqos->debugfs_dir = NULL;
 959 }
 960
 961 void blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
 962 {
 963         struct request_queue *q = rqos->q;
 964         const char *dir_name = rq_qos_id_to_name(rqos->id);
 965
 966         if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
 967                 return;
 968
 969         if (!q->rqos_debugfs_dir)
 970                 q->rqos_debugfs_dir = debugfs_create_dir("rqos",
 971                                                          q->debugfs_dir);
 972
 973         rqos->debugfs_dir = debugfs_create_dir(dir_name,
 974                                                rqos->q->rqos_debugfs_dir);
 975
 976         debugfs_create_files(rqos->debugfs_dir, rqos, rqos->ops->debugfs_attrs);
 977 }
 978
 979 void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q)
 980 {
 981         debugfs_remove_recursive(q->rqos_debugfs_dir);
 982         q->rqos_debugfs_dir = NULL;
 983 }
 984
 985 void blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
 986                                         struct blk_mq_hw_ctx *hctx)
 987 {
 988         struct elevator_type *e = q->elevator->type;
 989
 990         /*
 991          * If the parent debugfs directory has not been created yet, return;
 992          * We will be called again later on with appropriate parent debugfs
 993          * directory from blk_register_queue()
 994          */
 995         if (!hctx->debugfs_dir)
 996                 return;
 997
 998         if (!e->hctx_debugfs_attrs)
 999                 return;
1000
1001         hctx->sched_debugfs_dir = debugfs_create_dir("sched",
1002                                                      hctx->debugfs_dir);
1003         debugfs_create_files(hctx->sched_debugfs_dir, hctx,
1004                              e->hctx_debugfs_attrs);
1005 }
1006
1007 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1008 {
1009         debugfs_remove_recursive(hctx->sched_debugfs_dir);
1010         hctx->sched_debugfs_dir = NULL;
1011 }