block/blk-mq-debugfs.c

   1 /*
   2  * Copyright (C) 2017 Facebook
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License
  14  * along with this program.  If not, see <https://www.gnu.org/licenses/>.
  15  */
  16
  17 #include <linux/kernel.h>
  18 #include <linux/blkdev.h>
  19 #include <linux/debugfs.h>
  20
  21 #include <linux/blk-mq.h>
  22 #include "blk.h"
  23 #include "blk-mq.h"
  24 #include "blk-mq-debugfs.h"
  25 #include "blk-mq-tag.h"
  26
  27 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
  28 {
  29         if (stat->nr_samples) {
  30                 seq_printf(m, "samples=%d, mean=%lld, min=%llu, max=%llu",
  31                            stat->nr_samples, stat->mean, stat->min, stat->max);
  32         } else {
  33                 seq_puts(m, "samples=0");
  34         }
  35 }
  36
  37 static int queue_poll_stat_show(void *data, struct seq_file *m)
  38 {
  39         struct request_queue *q = data;
  40         int bucket;
  41
  42         for (bucket = 0; bucket < BLK_MQ_POLL_STATS_BKTS/2; bucket++) {
  43                 seq_printf(m, "read  (%d Bytes): ", 1 << (9+bucket));
  44                 print_stat(m, &q->poll_stat[2*bucket]);
  45                 seq_puts(m, "\n");
  46
  47                 seq_printf(m, "write (%d Bytes): ",  1 << (9+bucket));
  48                 print_stat(m, &q->poll_stat[2*bucket+1]);
  49                 seq_puts(m, "\n");
  50         }
  51         return 0;
  52 }
  53
  54 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
  55         __acquires(&q->requeue_lock)
  56 {
  57         struct request_queue *q = m->private;
  58
  59         spin_lock_irq(&q->requeue_lock);
  60         return seq_list_start(&q->requeue_list, *pos);
  61 }
  62
  63 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
  64 {
  65         struct request_queue *q = m->private;
  66
  67         return seq_list_next(v, &q->requeue_list, pos);
  68 }
  69
  70 static void queue_requeue_list_stop(struct seq_file *m, void *v)
  71         __releases(&q->requeue_lock)
  72 {
  73         struct request_queue *q = m->private;
  74
  75         spin_unlock_irq(&q->requeue_lock);
  76 }
  77
  78 static const struct seq_operations queue_requeue_list_seq_ops = {
  79         .start  = queue_requeue_list_start,
  80         .next   = queue_requeue_list_next,
  81         .stop   = queue_requeue_list_stop,
  82         .show   = blk_mq_debugfs_rq_show,
  83 };
  84
  85 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
  86                           const char *const *flag_name, int flag_name_count)
  87 {
  88         bool sep = false;
  89         int i;
  90
  91         for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
  92                 if (!(flags & BIT(i)))
  93                         continue;
  94                 if (sep)
  95                         seq_puts(m, "|");
  96                 sep = true;
  97                 if (i < flag_name_count && flag_name[i])
  98                         seq_puts(m, flag_name[i]);
  99                 else
 100                         seq_printf(m, "%d", i);
 101         }
 102         return 0;
 103 }
 104
 105 static int queue_pm_only_show(void *data, struct seq_file *m)
 106 {
 107         struct request_queue *q = data;
 108
 109         seq_printf(m, "%d\n", atomic_read(&q->pm_only));
 110         return 0;
 111 }
 112
 113 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
 114 static const char *const blk_queue_flag_name[] = {
 115         QUEUE_FLAG_NAME(QUEUED),
 116         QUEUE_FLAG_NAME(STOPPED),
 117         QUEUE_FLAG_NAME(DYING),
 118         QUEUE_FLAG_NAME(BYPASS),
 119         QUEUE_FLAG_NAME(BIDI),
 120         QUEUE_FLAG_NAME(NOMERGES),
 121         QUEUE_FLAG_NAME(SAME_COMP),
 122         QUEUE_FLAG_NAME(FAIL_IO),
 123         QUEUE_FLAG_NAME(NONROT),
 124         QUEUE_FLAG_NAME(IO_STAT),
 125         QUEUE_FLAG_NAME(DISCARD),
 126         QUEUE_FLAG_NAME(NOXMERGES),
 127         QUEUE_FLAG_NAME(ADD_RANDOM),
 128         QUEUE_FLAG_NAME(SECERASE),
 129         QUEUE_FLAG_NAME(SAME_FORCE),
 130         QUEUE_FLAG_NAME(DEAD),
 131         QUEUE_FLAG_NAME(INIT_DONE),
 132         QUEUE_FLAG_NAME(NO_SG_MERGE),
 133         QUEUE_FLAG_NAME(POLL),
 134         QUEUE_FLAG_NAME(WC),
 135         QUEUE_FLAG_NAME(FUA),
 136         QUEUE_FLAG_NAME(FLUSH_NQ),
 137         QUEUE_FLAG_NAME(DAX),
 138         QUEUE_FLAG_NAME(STATS),
 139         QUEUE_FLAG_NAME(POLL_STATS),
 140         QUEUE_FLAG_NAME(REGISTERED),
 141         QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
 142         QUEUE_FLAG_NAME(QUIESCED),
 143 };
 144 #undef QUEUE_FLAG_NAME
 145
 146 static int queue_state_show(void *data, struct seq_file *m)
 147 {
 148         struct request_queue *q = data;
 149
 150         blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
 151                        ARRAY_SIZE(blk_queue_flag_name));
 152         seq_puts(m, "\n");
 153         return 0;
 154 }
 155
 156 static ssize_t queue_state_write(void *data, const char __user *buf,
 157                                  size_t count, loff_t *ppos)
 158 {
 159         struct request_queue *q = data;
 160         char opbuf[16] = { }, *op;
 161
 162         /*
 163          * The "state" attribute is removed after blk_cleanup_queue() has called
 164          * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
 165          * triggering a use-after-free.
 166          */
 167         if (blk_queue_dead(q))
 168                 return -ENOENT;
 169
 170         if (count >= sizeof(opbuf)) {
 171                 pr_err("%s: operation too long\n", __func__);
 172                 goto inval;
 173         }
 174
 175         if (copy_from_user(opbuf, buf, count))
 176                 return -EFAULT;
 177         op = strstrip(opbuf);
 178         if (strcmp(op, "run") == 0) {
 179                 blk_mq_run_hw_queues(q, true);
 180         } else if (strcmp(op, "start") == 0) {
 181                 blk_mq_start_stopped_hw_queues(q, true);
 182         } else if (strcmp(op, "kick") == 0) {
 183                 blk_mq_kick_requeue_list(q);
 184         } else {
 185                 pr_err("%s: unsupported operation '%s'\n", __func__, op);
 186 inval:
 187                 pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
 188                 return -EINVAL;
 189         }
 190         return count;
 191 }
 192
 193 static int queue_write_hint_show(void *data, struct seq_file *m)
 194 {
 195         struct request_queue *q = data;
 196         int i;
 197
 198         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 199                 seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
 200
 201         return 0;
 202 }
 203
 204 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
 205                                       size_t count, loff_t *ppos)
 206 {
 207         struct request_queue *q = data;
 208         int i;
 209
 210         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 211                 q->write_hints[i] = 0;
 212
 213         return count;
 214 }
 215
 216 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
 217         { "poll_stat", 0400, queue_poll_stat_show },
 218         { "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
 219         { "pm_only", 0600, queue_pm_only_show, NULL },
 220         { "state", 0600, queue_state_show, queue_state_write },
 221         { "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
 222         { "zone_wlock", 0400, queue_zone_wlock_show, NULL },
 223         { },
 224 };
 225
 226 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
 227 static const char *const hctx_state_name[] = {
 228         HCTX_STATE_NAME(STOPPED),
 229         HCTX_STATE_NAME(TAG_ACTIVE),
 230         HCTX_STATE_NAME(SCHED_RESTART),
 231 };
 232 #undef HCTX_STATE_NAME
 233
 234 static int hctx_state_show(void *data, struct seq_file *m)
 235 {
 236         struct blk_mq_hw_ctx *hctx = data;
 237
 238         blk_flags_show(m, hctx->state, hctx_state_name,
 239                        ARRAY_SIZE(hctx_state_name));
 240         seq_puts(m, "\n");
 241         return 0;
 242 }
 243
 244 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
 245 static const char *const alloc_policy_name[] = {
 246         BLK_TAG_ALLOC_NAME(FIFO),
 247         BLK_TAG_ALLOC_NAME(RR),
 248 };
 249 #undef BLK_TAG_ALLOC_NAME
 250
 251 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
 252 static const char *const hctx_flag_name[] = {
 253         HCTX_FLAG_NAME(SHOULD_MERGE),
 254         HCTX_FLAG_NAME(TAG_SHARED),
 255         HCTX_FLAG_NAME(SG_MERGE),
 256         HCTX_FLAG_NAME(BLOCKING),
 257         HCTX_FLAG_NAME(NO_SCHED),
 258 };
 259 #undef HCTX_FLAG_NAME
 260
 261 static int hctx_flags_show(void *data, struct seq_file *m)
 262 {
 263         struct blk_mq_hw_ctx *hctx = data;
 264         const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
 265
 266         seq_puts(m, "alloc_policy=");
 267         if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
 268             alloc_policy_name[alloc_policy])
 269                 seq_puts(m, alloc_policy_name[alloc_policy]);
 270         else
 271                 seq_printf(m, "%d", alloc_policy);
 272         seq_puts(m, " ");
 273         blk_flags_show(m,
 274                        hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
 275                        hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
 276         seq_puts(m, "\n");
 277         return 0;
 278 }
 279
 280 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
 281 static const char *const op_name[] = {
 282         REQ_OP_NAME(READ),
 283         REQ_OP_NAME(WRITE),
 284         REQ_OP_NAME(FLUSH),
 285         REQ_OP_NAME(DISCARD),
 286         REQ_OP_NAME(ZONE_REPORT),
 287         REQ_OP_NAME(SECURE_ERASE),
 288         REQ_OP_NAME(ZONE_RESET),
 289         REQ_OP_NAME(WRITE_SAME),
 290         REQ_OP_NAME(WRITE_ZEROES),
 291         REQ_OP_NAME(SCSI_IN),
 292         REQ_OP_NAME(SCSI_OUT),
 293         REQ_OP_NAME(DRV_IN),
 294         REQ_OP_NAME(DRV_OUT),
 295 };
 296 #undef REQ_OP_NAME
 297
 298 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
 299 static const char *const cmd_flag_name[] = {
 300         CMD_FLAG_NAME(FAILFAST_DEV),
 301         CMD_FLAG_NAME(FAILFAST_TRANSPORT),
 302         CMD_FLAG_NAME(FAILFAST_DRIVER),
 303         CMD_FLAG_NAME(SYNC),
 304         CMD_FLAG_NAME(META),
 305         CMD_FLAG_NAME(PRIO),
 306         CMD_FLAG_NAME(NOMERGE),
 307         CMD_FLAG_NAME(IDLE),
 308         CMD_FLAG_NAME(INTEGRITY),
 309         CMD_FLAG_NAME(FUA),
 310         CMD_FLAG_NAME(PREFLUSH),
 311         CMD_FLAG_NAME(RAHEAD),
 312         CMD_FLAG_NAME(BACKGROUND),
 313         CMD_FLAG_NAME(NOUNMAP),
 314         CMD_FLAG_NAME(NOWAIT),
 315 };
 316 #undef CMD_FLAG_NAME
 317
 318 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
 319 static const char *const rqf_name[] = {
 320         RQF_NAME(SORTED),
 321         RQF_NAME(STARTED),
 322         RQF_NAME(QUEUED),
 323         RQF_NAME(SOFTBARRIER),
 324         RQF_NAME(FLUSH_SEQ),
 325         RQF_NAME(MIXED_MERGE),
 326         RQF_NAME(MQ_INFLIGHT),
 327         RQF_NAME(DONTPREP),
 328         RQF_NAME(PREEMPT),
 329         RQF_NAME(COPY_USER),
 330         RQF_NAME(FAILED),
 331         RQF_NAME(QUIET),
 332         RQF_NAME(ELVPRIV),
 333         RQF_NAME(IO_STAT),
 334         RQF_NAME(ALLOCED),
 335         RQF_NAME(PM),
 336         RQF_NAME(HASHED),
 337         RQF_NAME(STATS),
 338         RQF_NAME(SPECIAL_PAYLOAD),
 339         RQF_NAME(ZONE_WRITE_LOCKED),
 340         RQF_NAME(MQ_POLL_SLEPT),
 341 };
 342 #undef RQF_NAME
 343
 344 static const char *const blk_mq_rq_state_name_array[] = {
 345         [MQ_RQ_IDLE]            = "idle",
 346         [MQ_RQ_IN_FLIGHT]       = "in_flight",
 347         [MQ_RQ_COMPLETE]        = "complete",
 348 };
 349
 350 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
 351 {
 352         if (WARN_ON_ONCE((unsigned int)rq_state >=
 353                          ARRAY_SIZE(blk_mq_rq_state_name_array)))
 354                 return "(?)";
 355         return blk_mq_rq_state_name_array[rq_state];
 356 }
 357
 358 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
 359 {
 360         const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
 361         const unsigned int op = rq->cmd_flags & REQ_OP_MASK;
 362
 363         seq_printf(m, "%p {.op=", rq);
 364         if (op < ARRAY_SIZE(op_name) && op_name[op])
 365                 seq_printf(m, "%s", op_name[op]);
 366         else
 367                 seq_printf(m, "%d", op);
 368         seq_puts(m, ", .cmd_flags=");
 369         blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
 370                        ARRAY_SIZE(cmd_flag_name));
 371         seq_puts(m, ", .rq_flags=");
 372         blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
 373                        ARRAY_SIZE(rqf_name));
 374         seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
 375         seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
 376                    rq->internal_tag);
 377         if (mq_ops->show_rq)
 378                 mq_ops->show_rq(m, rq);
 379         seq_puts(m, "}\n");
 380         return 0;
 381 }
 382 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
 383
 384 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
 385 {
 386         return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
 387 }
 388 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
 389
 390 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
 391         __acquires(&hctx->lock)
 392 {
 393         struct blk_mq_hw_ctx *hctx = m->private;
 394
 395         spin_lock(&hctx->lock);
 396         return seq_list_start(&hctx->dispatch, *pos);
 397 }
 398
 399 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
 400 {
 401         struct blk_mq_hw_ctx *hctx = m->private;
 402
 403         return seq_list_next(v, &hctx->dispatch, pos);
 404 }
 405
 406 static void hctx_dispatch_stop(struct seq_file *m, void *v)
 407         __releases(&hctx->lock)
 408 {
 409         struct blk_mq_hw_ctx *hctx = m->private;
 410
 411         spin_unlock(&hctx->lock);
 412 }
 413
 414 static const struct seq_operations hctx_dispatch_seq_ops = {
 415         .start  = hctx_dispatch_start,
 416         .next   = hctx_dispatch_next,
 417         .stop   = hctx_dispatch_stop,
 418         .show   = blk_mq_debugfs_rq_show,
 419 };
 420
 421 struct show_busy_params {
 422         struct seq_file         *m;
 423         struct blk_mq_hw_ctx    *hctx;
 424 };
 425
 426 /*
 427  * Note: the state of a request may change while this function is in progress,
 428  * e.g. due to a concurrent blk_mq_finish_request() call.
 429  */
 430 static void hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 431 {
 432         const struct show_busy_params *params = data;
 433
 434         if (blk_mq_map_queue(rq->q, rq->mq_ctx->cpu) == params->hctx &&
 435             blk_mq_rq_state(rq) != MQ_RQ_IDLE)
 436                 __blk_mq_debugfs_rq_show(params->m,
 437                                          list_entry_rq(&rq->queuelist));
 438 }
 439
 440 static int hctx_busy_show(void *data, struct seq_file *m)
 441 {
 442         struct blk_mq_hw_ctx *hctx = data;
 443         struct show_busy_params params = { .m = m, .hctx = hctx };
 444
 445         blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
 446                                 &params);
 447
 448         return 0;
 449 }
 450
 451 static int hctx_ctx_map_show(void *data, struct seq_file *m)
 452 {
 453         struct blk_mq_hw_ctx *hctx = data;
 454
 455         sbitmap_bitmap_show(&hctx->ctx_map, m);
 456         return 0;
 457 }
 458
 459 static void blk_mq_debugfs_tags_show(struct seq_file *m,
 460                                      struct blk_mq_tags *tags)
 461 {
 462         seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
 463         seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
 464         seq_printf(m, "active_queues=%d\n",
 465                    atomic_read(&tags->active_queues));
 466
 467         seq_puts(m, "\nbitmap_tags:\n");
 468         sbitmap_queue_show(&tags->bitmap_tags, m);
 469
 470         if (tags->nr_reserved_tags) {
 471                 seq_puts(m, "\nbreserved_tags:\n");
 472                 sbitmap_queue_show(&tags->breserved_tags, m);
 473         }
 474 }
 475
 476 static int hctx_tags_show(void *data, struct seq_file *m)
 477 {
 478         struct blk_mq_hw_ctx *hctx = data;
 479         struct request_queue *q = hctx->queue;
 480         int res;
 481
 482         res = mutex_lock_interruptible(&q->sysfs_lock);
 483         if (res)
 484                 goto out;
 485         if (hctx->tags)
 486                 blk_mq_debugfs_tags_show(m, hctx->tags);
 487         mutex_unlock(&q->sysfs_lock);
 488
 489 out:
 490         return res;
 491 }
 492
 493 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
 494 {
 495         struct blk_mq_hw_ctx *hctx = data;
 496         struct request_queue *q = hctx->queue;
 497         int res;
 498
 499         res = mutex_lock_interruptible(&q->sysfs_lock);
 500         if (res)
 501                 goto out;
 502         if (hctx->tags)
 503                 sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
 504         mutex_unlock(&q->sysfs_lock);
 505
 506 out:
 507         return res;
 508 }
 509
 510 static int hctx_sched_tags_show(void *data, struct seq_file *m)
 511 {
 512         struct blk_mq_hw_ctx *hctx = data;
 513         struct request_queue *q = hctx->queue;
 514         int res;
 515
 516         res = mutex_lock_interruptible(&q->sysfs_lock);
 517         if (res)
 518                 goto out;
 519         if (hctx->sched_tags)
 520                 blk_mq_debugfs_tags_show(m, hctx->sched_tags);
 521         mutex_unlock(&q->sysfs_lock);
 522
 523 out:
 524         return res;
 525 }
 526
 527 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
 528 {
 529         struct blk_mq_hw_ctx *hctx = data;
 530         struct request_queue *q = hctx->queue;
 531         int res;
 532
 533         res = mutex_lock_interruptible(&q->sysfs_lock);
 534         if (res)
 535                 goto out;
 536         if (hctx->sched_tags)
 537                 sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
 538         mutex_unlock(&q->sysfs_lock);
 539
 540 out:
 541         return res;
 542 }
 543
 544 static int hctx_io_poll_show(void *data, struct seq_file *m)
 545 {
 546         struct blk_mq_hw_ctx *hctx = data;
 547
 548         seq_printf(m, "considered=%lu\n", hctx->poll_considered);
 549         seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
 550         seq_printf(m, "success=%lu\n", hctx->poll_success);
 551         return 0;
 552 }
 553
 554 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
 555                                   size_t count, loff_t *ppos)
 556 {
 557         struct blk_mq_hw_ctx *hctx = data;
 558
 559         hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
 560         return count;
 561 }
 562
 563 static int hctx_dispatched_show(void *data, struct seq_file *m)
 564 {
 565         struct blk_mq_hw_ctx *hctx = data;
 566         int i;
 567
 568         seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
 569
 570         for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
 571                 unsigned int d = 1U << (i - 1);
 572
 573                 seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
 574         }
 575
 576         seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
 577         return 0;
 578 }
 579
 580 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
 581                                      size_t count, loff_t *ppos)
 582 {
 583         struct blk_mq_hw_ctx *hctx = data;
 584         int i;
 585
 586         for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
 587                 hctx->dispatched[i] = 0;
 588         return count;
 589 }
 590
 591 static int hctx_queued_show(void *data, struct seq_file *m)
 592 {
 593         struct blk_mq_hw_ctx *hctx = data;
 594
 595         seq_printf(m, "%lu\n", hctx->queued);
 596         return 0;
 597 }
 598
 599 static ssize_t hctx_queued_write(void *data, const char __user *buf,
 600                                  size_t count, loff_t *ppos)
 601 {
 602         struct blk_mq_hw_ctx *hctx = data;
 603
 604         hctx->queued = 0;
 605         return count;
 606 }
 607
 608 static int hctx_run_show(void *data, struct seq_file *m)
 609 {
 610         struct blk_mq_hw_ctx *hctx = data;
 611
 612         seq_printf(m, "%lu\n", hctx->run);
 613         return 0;
 614 }
 615
 616 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
 617                               loff_t *ppos)
 618 {
 619         struct blk_mq_hw_ctx *hctx = data;
 620
 621         hctx->run = 0;
 622         return count;
 623 }
 624
 625 static int hctx_active_show(void *data, struct seq_file *m)
 626 {
 627         struct blk_mq_hw_ctx *hctx = data;
 628
 629         seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
 630         return 0;
 631 }
 632
 633 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
 634 {
 635         struct blk_mq_hw_ctx *hctx = data;
 636
 637         seq_printf(m, "%u\n", hctx->dispatch_busy);
 638         return 0;
 639 }
 640
 641 static void *ctx_rq_list_start(struct seq_file *m, loff_t *pos)
 642         __acquires(&ctx->lock)
 643 {
 644         struct blk_mq_ctx *ctx = m->private;
 645
 646         spin_lock(&ctx->lock);
 647         return seq_list_start(&ctx->rq_list, *pos);
 648 }
 649
 650 static void *ctx_rq_list_next(struct seq_file *m, void *v, loff_t *pos)
 651 {
 652         struct blk_mq_ctx *ctx = m->private;
 653
 654         return seq_list_next(v, &ctx->rq_list, pos);
 655 }
 656
 657 static void ctx_rq_list_stop(struct seq_file *m, void *v)
 658         __releases(&ctx->lock)
 659 {
 660         struct blk_mq_ctx *ctx = m->private;
 661
 662         spin_unlock(&ctx->lock);
 663 }
 664
 665 static const struct seq_operations ctx_rq_list_seq_ops = {
 666         .start  = ctx_rq_list_start,
 667         .next   = ctx_rq_list_next,
 668         .stop   = ctx_rq_list_stop,
 669         .show   = blk_mq_debugfs_rq_show,
 670 };
 671 static int ctx_dispatched_show(void *data, struct seq_file *m)
 672 {
 673         struct blk_mq_ctx *ctx = data;
 674
 675         seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
 676         return 0;
 677 }
 678
 679 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
 680                                     size_t count, loff_t *ppos)
 681 {
 682         struct blk_mq_ctx *ctx = data;
 683
 684         ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
 685         return count;
 686 }
 687
 688 static int ctx_merged_show(void *data, struct seq_file *m)
 689 {
 690         struct blk_mq_ctx *ctx = data;
 691
 692         seq_printf(m, "%lu\n", ctx->rq_merged);
 693         return 0;
 694 }
 695
 696 static ssize_t ctx_merged_write(void *data, const char __user *buf,
 697                                 size_t count, loff_t *ppos)
 698 {
 699         struct blk_mq_ctx *ctx = data;
 700
 701         ctx->rq_merged = 0;
 702         return count;
 703 }
 704
 705 static int ctx_completed_show(void *data, struct seq_file *m)
 706 {
 707         struct blk_mq_ctx *ctx = data;
 708
 709         seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
 710         return 0;
 711 }
 712
 713 static ssize_t ctx_completed_write(void *data, const char __user *buf,
 714                                    size_t count, loff_t *ppos)
 715 {
 716         struct blk_mq_ctx *ctx = data;
 717
 718         ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
 719         return count;
 720 }
 721
 722 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
 723 {
 724         const struct blk_mq_debugfs_attr *attr = m->private;
 725         void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
 726
 727         return attr->show(data, m);
 728 }
 729
 730 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
 731                                     size_t count, loff_t *ppos)
 732 {
 733         struct seq_file *m = file->private_data;
 734         const struct blk_mq_debugfs_attr *attr = m->private;
 735         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 736
 737         /*
 738          * Attributes that only implement .seq_ops are read-only and 'attr' is
 739          * the same with 'data' in this case.
 740          */
 741         if (attr == data || !attr->write)
 742                 return -EPERM;
 743
 744         return attr->write(data, buf, count, ppos);
 745 }
 746
 747 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
 748 {
 749         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 750         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 751         struct seq_file *m;
 752         int ret;
 753
 754         if (attr->seq_ops) {
 755                 ret = seq_open(file, attr->seq_ops);
 756                 if (!ret) {
 757                         m = file->private_data;
 758                         m->private = data;
 759                 }
 760                 return ret;
 761         }
 762
 763         if (WARN_ON_ONCE(!attr->show))
 764                 return -EPERM;
 765
 766         return single_open(file, blk_mq_debugfs_show, inode->i_private);
 767 }
 768
 769 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
 770 {
 771         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 772
 773         if (attr->show)
 774                 return single_release(inode, file);
 775         else
 776                 return seq_release(inode, file);
 777 }
 778
 779 static const struct file_operations blk_mq_debugfs_fops = {
 780         .open           = blk_mq_debugfs_open,
 781         .read           = seq_read,
 782         .write          = blk_mq_debugfs_write,
 783         .llseek         = seq_lseek,
 784         .release        = blk_mq_debugfs_release,
 785 };
 786
 787 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
 788         {"state", 0400, hctx_state_show},
 789         {"flags", 0400, hctx_flags_show},
 790         {"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
 791         {"busy", 0400, hctx_busy_show},
 792         {"ctx_map", 0400, hctx_ctx_map_show},
 793         {"tags", 0400, hctx_tags_show},
 794         {"tags_bitmap", 0400, hctx_tags_bitmap_show},
 795         {"sched_tags", 0400, hctx_sched_tags_show},
 796         {"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
 797         {"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
 798         {"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
 799         {"queued", 0600, hctx_queued_show, hctx_queued_write},
 800         {"run", 0600, hctx_run_show, hctx_run_write},
 801         {"active", 0400, hctx_active_show},
 802         {"dispatch_busy", 0400, hctx_dispatch_busy_show},
 803         {},
 804 };
 805
 806 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
 807         {"rq_list", 0400, .seq_ops = &ctx_rq_list_seq_ops},
 808         {"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
 809         {"merged", 0600, ctx_merged_show, ctx_merged_write},
 810         {"completed", 0600, ctx_completed_show, ctx_completed_write},
 811         {},
 812 };
 813
 814 static bool debugfs_create_files(struct dentry *parent, void *data,
 815                                  const struct blk_mq_debugfs_attr *attr)
 816 {
 817         d_inode(parent)->i_private = data;
 818
 819         for (; attr->name; attr++) {
 820                 if (!debugfs_create_file(attr->name, attr->mode, parent,
 821                                          (void *)attr, &blk_mq_debugfs_fops))
 822                         return false;
 823         }
 824         return true;
 825 }
 826
 827 int blk_mq_debugfs_register(struct request_queue *q)
 828 {
 829         struct blk_mq_hw_ctx *hctx;
 830         int i;
 831
 832         if (!blk_debugfs_root)
 833                 return -ENOENT;
 834
 835         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 836                                             blk_debugfs_root);
 837         if (!q->debugfs_dir)
 838                 return -ENOMEM;
 839
 840         if (!debugfs_create_files(q->debugfs_dir, q,
 841                                   blk_mq_debugfs_queue_attrs))
 842                 goto err;
 843
 844         /*
 845          * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
 846          * didn't exist yet (because we don't know what to name the directory
 847          * until the queue is registered to a gendisk).
 848          */
 849         if (q->elevator && !q->sched_debugfs_dir)
 850                 blk_mq_debugfs_register_sched(q);
 851
 852         /* Similarly, blk_mq_init_hctx() couldn't do this previously. */
 853         queue_for_each_hw_ctx(q, hctx, i) {
 854                 if (!hctx->debugfs_dir && blk_mq_debugfs_register_hctx(q, hctx))
 855                         goto err;
 856                 if (q->elevator && !hctx->sched_debugfs_dir &&
 857                     blk_mq_debugfs_register_sched_hctx(q, hctx))
 858                         goto err;
 859         }
 860
 861         return 0;
 862
 863 err:
 864         blk_mq_debugfs_unregister(q);
 865         return -ENOMEM;
 866 }
 867
 868 void blk_mq_debugfs_unregister(struct request_queue *q)
 869 {
 870         debugfs_remove_recursive(q->debugfs_dir);
 871         q->sched_debugfs_dir = NULL;
 872         q->debugfs_dir = NULL;
 873 }
 874
 875 static int blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
 876                                        struct blk_mq_ctx *ctx)
 877 {
 878         struct dentry *ctx_dir;
 879         char name[20];
 880
 881         snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
 882         ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
 883         if (!ctx_dir)
 884                 return -ENOMEM;
 885
 886         if (!debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs))
 887                 return -ENOMEM;
 888
 889         return 0;
 890 }
 891
 892 int blk_mq_debugfs_register_hctx(struct request_queue *q,
 893                                  struct blk_mq_hw_ctx *hctx)
 894 {
 895         struct blk_mq_ctx *ctx;
 896         char name[20];
 897         int i;
 898
 899         if (!q->debugfs_dir)
 900                 return -ENOENT;
 901
 902         snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
 903         hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
 904         if (!hctx->debugfs_dir)
 905                 return -ENOMEM;
 906
 907         if (!debugfs_create_files(hctx->debugfs_dir, hctx,
 908                                   blk_mq_debugfs_hctx_attrs))
 909                 goto err;
 910
 911         hctx_for_each_ctx(hctx, ctx, i) {
 912                 if (blk_mq_debugfs_register_ctx(hctx, ctx))
 913                         goto err;
 914         }
 915
 916         return 0;
 917
 918 err:
 919         blk_mq_debugfs_unregister_hctx(hctx);
 920         return -ENOMEM;
 921 }
 922
 923 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
 924 {
 925         debugfs_remove_recursive(hctx->debugfs_dir);
 926         hctx->sched_debugfs_dir = NULL;
 927         hctx->debugfs_dir = NULL;
 928 }
 929
 930 int blk_mq_debugfs_register_hctxs(struct request_queue *q)
 931 {
 932         struct blk_mq_hw_ctx *hctx;
 933         int i;
 934
 935         queue_for_each_hw_ctx(q, hctx, i) {
 936                 if (blk_mq_debugfs_register_hctx(q, hctx))
 937                         return -ENOMEM;
 938         }
 939
 940         return 0;
 941 }
 942
 943 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
 944 {
 945         struct blk_mq_hw_ctx *hctx;
 946         int i;
 947
 948         queue_for_each_hw_ctx(q, hctx, i)
 949                 blk_mq_debugfs_unregister_hctx(hctx);
 950 }
 951
 952 int blk_mq_debugfs_register_sched(struct request_queue *q)
 953 {
 954         struct elevator_type *e = q->elevator->type;
 955
 956         if (!q->debugfs_dir)
 957                 return -ENOENT;
 958
 959         if (!e->queue_debugfs_attrs)
 960                 return 0;
 961
 962         q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
 963         if (!q->sched_debugfs_dir)
 964                 return -ENOMEM;
 965
 966         if (!debugfs_create_files(q->sched_debugfs_dir, q,
 967                                   e->queue_debugfs_attrs))
 968                 goto err;
 969
 970         return 0;
 971
 972 err:
 973         blk_mq_debugfs_unregister_sched(q);
 974         return -ENOMEM;
 975 }
 976
 977 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
 978 {
 979         debugfs_remove_recursive(q->sched_debugfs_dir);
 980         q->sched_debugfs_dir = NULL;
 981 }
 982
 983 int blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
 984                                        struct blk_mq_hw_ctx *hctx)
 985 {
 986         struct elevator_type *e = q->elevator->type;
 987
 988         if (!hctx->debugfs_dir)
 989                 return -ENOENT;
 990
 991         if (!e->hctx_debugfs_attrs)
 992                 return 0;
 993
 994         hctx->sched_debugfs_dir = debugfs_create_dir("sched",
 995                                                      hctx->debugfs_dir);
 996         if (!hctx->sched_debugfs_dir)
 997                 return -ENOMEM;
 998
 999         if (!debugfs_create_files(hctx->sched_debugfs_dir, hctx,
1000                                   e->hctx_debugfs_attrs))
1001                 return -ENOMEM;
1002
1003         return 0;
1004 }
1005
1006 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1007 {
1008         debugfs_remove_recursive(hctx->sched_debugfs_dir);
1009         hctx->sched_debugfs_dir = NULL;
1010 }