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 #include "blk-rq-qos.h"
  27
  28 static void print_stat(struct seq_file *m, struct blk_rq_stat *stat)
  29 {
  30         if (stat->nr_samples) {
  31                 seq_printf(m, "samples=%d, mean=%lld, min=%llu, max=%llu",
  32                            stat->nr_samples, stat->mean, stat->min, stat->max);
  33         } else {
  34                 seq_puts(m, "samples=0");
  35         }
  36 }
  37
  38 static int queue_poll_stat_show(void *data, struct seq_file *m)
  39 {
  40         struct request_queue *q = data;
  41         int bucket;
  42
  43         for (bucket = 0; bucket < BLK_MQ_POLL_STATS_BKTS/2; bucket++) {
  44                 seq_printf(m, "read  (%d Bytes): ", 1 << (9+bucket));
  45                 print_stat(m, &q->poll_stat[2*bucket]);
  46                 seq_puts(m, "\n");
  47
  48                 seq_printf(m, "write (%d Bytes): ",  1 << (9+bucket));
  49                 print_stat(m, &q->poll_stat[2*bucket+1]);
  50                 seq_puts(m, "\n");
  51         }
  52         return 0;
  53 }
  54
  55 static void *queue_requeue_list_start(struct seq_file *m, loff_t *pos)
  56         __acquires(&q->requeue_lock)
  57 {
  58         struct request_queue *q = m->private;
  59
  60         spin_lock_irq(&q->requeue_lock);
  61         return seq_list_start(&q->requeue_list, *pos);
  62 }
  63
  64 static void *queue_requeue_list_next(struct seq_file *m, void *v, loff_t *pos)
  65 {
  66         struct request_queue *q = m->private;
  67
  68         return seq_list_next(v, &q->requeue_list, pos);
  69 }
  70
  71 static void queue_requeue_list_stop(struct seq_file *m, void *v)
  72         __releases(&q->requeue_lock)
  73 {
  74         struct request_queue *q = m->private;
  75
  76         spin_unlock_irq(&q->requeue_lock);
  77 }
  78
  79 static const struct seq_operations queue_requeue_list_seq_ops = {
  80         .start  = queue_requeue_list_start,
  81         .next   = queue_requeue_list_next,
  82         .stop   = queue_requeue_list_stop,
  83         .show   = blk_mq_debugfs_rq_show,
  84 };
  85
  86 static int blk_flags_show(struct seq_file *m, const unsigned long flags,
  87                           const char *const *flag_name, int flag_name_count)
  88 {
  89         bool sep = false;
  90         int i;
  91
  92         for (i = 0; i < sizeof(flags) * BITS_PER_BYTE; i++) {
  93                 if (!(flags & BIT(i)))
  94                         continue;
  95                 if (sep)
  96                         seq_puts(m, "|");
  97                 sep = true;
  98                 if (i < flag_name_count && flag_name[i])
  99                         seq_puts(m, flag_name[i]);
 100                 else
 101                         seq_printf(m, "%d", i);
 102         }
 103         return 0;
 104 }
 105
 106 static int queue_pm_only_show(void *data, struct seq_file *m)
 107 {
 108         struct request_queue *q = data;
 109
 110         seq_printf(m, "%d\n", atomic_read(&q->pm_only));
 111         return 0;
 112 }
 113
 114 #define QUEUE_FLAG_NAME(name) [QUEUE_FLAG_##name] = #name
 115 static const char *const blk_queue_flag_name[] = {
 116         QUEUE_FLAG_NAME(STOPPED),
 117         QUEUE_FLAG_NAME(DYING),
 118         QUEUE_FLAG_NAME(BIDI),
 119         QUEUE_FLAG_NAME(NOMERGES),
 120         QUEUE_FLAG_NAME(SAME_COMP),
 121         QUEUE_FLAG_NAME(FAIL_IO),
 122         QUEUE_FLAG_NAME(NONROT),
 123         QUEUE_FLAG_NAME(IO_STAT),
 124         QUEUE_FLAG_NAME(DISCARD),
 125         QUEUE_FLAG_NAME(NOXMERGES),
 126         QUEUE_FLAG_NAME(ADD_RANDOM),
 127         QUEUE_FLAG_NAME(SECERASE),
 128         QUEUE_FLAG_NAME(SAME_FORCE),
 129         QUEUE_FLAG_NAME(DEAD),
 130         QUEUE_FLAG_NAME(INIT_DONE),
 131         QUEUE_FLAG_NAME(POLL),
 132         QUEUE_FLAG_NAME(WC),
 133         QUEUE_FLAG_NAME(FUA),
 134         QUEUE_FLAG_NAME(DAX),
 135         QUEUE_FLAG_NAME(STATS),
 136         QUEUE_FLAG_NAME(POLL_STATS),
 137         QUEUE_FLAG_NAME(REGISTERED),
 138         QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
 139         QUEUE_FLAG_NAME(QUIESCED),
 140 };
 141 #undef QUEUE_FLAG_NAME
 142
 143 static int queue_state_show(void *data, struct seq_file *m)
 144 {
 145         struct request_queue *q = data;
 146
 147         blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
 148                        ARRAY_SIZE(blk_queue_flag_name));
 149         seq_puts(m, "\n");
 150         return 0;
 151 }
 152
 153 static ssize_t queue_state_write(void *data, const char __user *buf,
 154                                  size_t count, loff_t *ppos)
 155 {
 156         struct request_queue *q = data;
 157         char opbuf[16] = { }, *op;
 158
 159         /*
 160          * The "state" attribute is removed after blk_cleanup_queue() has called
 161          * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
 162          * triggering a use-after-free.
 163          */
 164         if (blk_queue_dead(q))
 165                 return -ENOENT;
 166
 167         if (count >= sizeof(opbuf)) {
 168                 pr_err("%s: operation too long\n", __func__);
 169                 goto inval;
 170         }
 171
 172         if (copy_from_user(opbuf, buf, count))
 173                 return -EFAULT;
 174         op = strstrip(opbuf);
 175         if (strcmp(op, "run") == 0) {
 176                 blk_mq_run_hw_queues(q, true);
 177         } else if (strcmp(op, "start") == 0) {
 178                 blk_mq_start_stopped_hw_queues(q, true);
 179         } else if (strcmp(op, "kick") == 0) {
 180                 blk_mq_kick_requeue_list(q);
 181         } else {
 182                 pr_err("%s: unsupported operation '%s'\n", __func__, op);
 183 inval:
 184                 pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
 185                 return -EINVAL;
 186         }
 187         return count;
 188 }
 189
 190 static int queue_write_hint_show(void *data, struct seq_file *m)
 191 {
 192         struct request_queue *q = data;
 193         int i;
 194
 195         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 196                 seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
 197
 198         return 0;
 199 }
 200
 201 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
 202                                       size_t count, loff_t *ppos)
 203 {
 204         struct request_queue *q = data;
 205         int i;
 206
 207         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 208                 q->write_hints[i] = 0;
 209
 210         return count;
 211 }
 212
 213 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
 214         { "poll_stat", 0400, queue_poll_stat_show },
 215         { "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
 216         { "pm_only", 0600, queue_pm_only_show, NULL },
 217         { "state", 0600, queue_state_show, queue_state_write },
 218         { "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
 219         { "zone_wlock", 0400, queue_zone_wlock_show, NULL },
 220         { },
 221 };
 222
 223 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
 224 static const char *const hctx_state_name[] = {
 225         HCTX_STATE_NAME(STOPPED),
 226         HCTX_STATE_NAME(TAG_ACTIVE),
 227         HCTX_STATE_NAME(SCHED_RESTART),
 228 };
 229 #undef HCTX_STATE_NAME
 230
 231 static int hctx_state_show(void *data, struct seq_file *m)
 232 {
 233         struct blk_mq_hw_ctx *hctx = data;
 234
 235         blk_flags_show(m, hctx->state, hctx_state_name,
 236                        ARRAY_SIZE(hctx_state_name));
 237         seq_puts(m, "\n");
 238         return 0;
 239 }
 240
 241 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
 242 static const char *const alloc_policy_name[] = {
 243         BLK_TAG_ALLOC_NAME(FIFO),
 244         BLK_TAG_ALLOC_NAME(RR),
 245 };
 246 #undef BLK_TAG_ALLOC_NAME
 247
 248 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
 249 static const char *const hctx_flag_name[] = {
 250         HCTX_FLAG_NAME(SHOULD_MERGE),
 251         HCTX_FLAG_NAME(TAG_SHARED),
 252         HCTX_FLAG_NAME(BLOCKING),
 253         HCTX_FLAG_NAME(NO_SCHED),
 254 };
 255 #undef HCTX_FLAG_NAME
 256
 257 static int hctx_flags_show(void *data, struct seq_file *m)
 258 {
 259         struct blk_mq_hw_ctx *hctx = data;
 260         const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
 261
 262         seq_puts(m, "alloc_policy=");
 263         if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
 264             alloc_policy_name[alloc_policy])
 265                 seq_puts(m, alloc_policy_name[alloc_policy]);
 266         else
 267                 seq_printf(m, "%d", alloc_policy);
 268         seq_puts(m, " ");
 269         blk_flags_show(m,
 270                        hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
 271                        hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
 272         seq_puts(m, "\n");
 273         return 0;
 274 }
 275
 276 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
 277 static const char *const op_name[] = {
 278         REQ_OP_NAME(READ),
 279         REQ_OP_NAME(WRITE),
 280         REQ_OP_NAME(FLUSH),
 281         REQ_OP_NAME(DISCARD),
 282         REQ_OP_NAME(SECURE_ERASE),
 283         REQ_OP_NAME(ZONE_RESET),
 284         REQ_OP_NAME(WRITE_SAME),
 285         REQ_OP_NAME(WRITE_ZEROES),
 286         REQ_OP_NAME(SCSI_IN),
 287         REQ_OP_NAME(SCSI_OUT),
 288         REQ_OP_NAME(DRV_IN),
 289         REQ_OP_NAME(DRV_OUT),
 290 };
 291 #undef REQ_OP_NAME
 292
 293 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
 294 static const char *const cmd_flag_name[] = {
 295         CMD_FLAG_NAME(FAILFAST_DEV),
 296         CMD_FLAG_NAME(FAILFAST_TRANSPORT),
 297         CMD_FLAG_NAME(FAILFAST_DRIVER),
 298         CMD_FLAG_NAME(SYNC),
 299         CMD_FLAG_NAME(META),
 300         CMD_FLAG_NAME(PRIO),
 301         CMD_FLAG_NAME(NOMERGE),
 302         CMD_FLAG_NAME(IDLE),
 303         CMD_FLAG_NAME(INTEGRITY),
 304         CMD_FLAG_NAME(FUA),
 305         CMD_FLAG_NAME(PREFLUSH),
 306         CMD_FLAG_NAME(RAHEAD),
 307         CMD_FLAG_NAME(BACKGROUND),
 308         CMD_FLAG_NAME(NOWAIT),
 309         CMD_FLAG_NAME(NOUNMAP),
 310         CMD_FLAG_NAME(HIPRI),
 311 };
 312 #undef CMD_FLAG_NAME
 313
 314 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
 315 static const char *const rqf_name[] = {
 316         RQF_NAME(SORTED),
 317         RQF_NAME(STARTED),
 318         RQF_NAME(SOFTBARRIER),
 319         RQF_NAME(FLUSH_SEQ),
 320         RQF_NAME(MIXED_MERGE),
 321         RQF_NAME(MQ_INFLIGHT),
 322         RQF_NAME(DONTPREP),
 323         RQF_NAME(PREEMPT),
 324         RQF_NAME(COPY_USER),
 325         RQF_NAME(FAILED),
 326         RQF_NAME(QUIET),
 327         RQF_NAME(ELVPRIV),
 328         RQF_NAME(IO_STAT),
 329         RQF_NAME(ALLOCED),
 330         RQF_NAME(PM),
 331         RQF_NAME(HASHED),
 332         RQF_NAME(STATS),
 333         RQF_NAME(SPECIAL_PAYLOAD),
 334         RQF_NAME(ZONE_WRITE_LOCKED),
 335         RQF_NAME(MQ_POLL_SLEPT),
 336 };
 337 #undef RQF_NAME
 338
 339 static const char *const blk_mq_rq_state_name_array[] = {
 340         [MQ_RQ_IDLE]            = "idle",
 341         [MQ_RQ_IN_FLIGHT]       = "in_flight",
 342         [MQ_RQ_COMPLETE]        = "complete",
 343 };
 344
 345 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
 346 {
 347         if (WARN_ON_ONCE((unsigned int)rq_state >=
 348                          ARRAY_SIZE(blk_mq_rq_state_name_array)))
 349                 return "(?)";
 350         return blk_mq_rq_state_name_array[rq_state];
 351 }
 352
 353 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
 354 {
 355         const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
 356         const unsigned int op = rq->cmd_flags & REQ_OP_MASK;
 357
 358         seq_printf(m, "%p {.op=", rq);
 359         if (op < ARRAY_SIZE(op_name) && op_name[op])
 360                 seq_printf(m, "%s", op_name[op]);
 361         else
 362                 seq_printf(m, "%d", op);
 363         seq_puts(m, ", .cmd_flags=");
 364         blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
 365                        ARRAY_SIZE(cmd_flag_name));
 366         seq_puts(m, ", .rq_flags=");
 367         blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
 368                        ARRAY_SIZE(rqf_name));
 369         seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
 370         seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
 371                    rq->internal_tag);
 372         if (mq_ops->show_rq)
 373                 mq_ops->show_rq(m, rq);
 374         seq_puts(m, "}\n");
 375         return 0;
 376 }
 377 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
 378
 379 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
 380 {
 381         return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
 382 }
 383 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
 384
 385 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
 386         __acquires(&hctx->lock)
 387 {
 388         struct blk_mq_hw_ctx *hctx = m->private;
 389
 390         spin_lock(&hctx->lock);
 391         return seq_list_start(&hctx->dispatch, *pos);
 392 }
 393
 394 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
 395 {
 396         struct blk_mq_hw_ctx *hctx = m->private;
 397
 398         return seq_list_next(v, &hctx->dispatch, pos);
 399 }
 400
 401 static void hctx_dispatch_stop(struct seq_file *m, void *v)
 402         __releases(&hctx->lock)
 403 {
 404         struct blk_mq_hw_ctx *hctx = m->private;
 405
 406         spin_unlock(&hctx->lock);
 407 }
 408
 409 static const struct seq_operations hctx_dispatch_seq_ops = {
 410         .start  = hctx_dispatch_start,
 411         .next   = hctx_dispatch_next,
 412         .stop   = hctx_dispatch_stop,
 413         .show   = blk_mq_debugfs_rq_show,
 414 };
 415
 416 struct show_busy_params {
 417         struct seq_file         *m;
 418         struct blk_mq_hw_ctx    *hctx;
 419 };
 420
 421 /*
 422  * Note: the state of a request may change while this function is in progress,
 423  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
 424  * keep iterating requests.
 425  */
 426 static bool hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 427 {
 428         const struct show_busy_params *params = data;
 429
 430         if (rq->mq_hctx == params->hctx)
 431                 __blk_mq_debugfs_rq_show(params->m,
 432                                          list_entry_rq(&rq->queuelist));
 433
 434         return true;
 435 }
 436
 437 static int hctx_busy_show(void *data, struct seq_file *m)
 438 {
 439         struct blk_mq_hw_ctx *hctx = data;
 440         struct show_busy_params params = { .m = m, .hctx = hctx };
 441
 442         blk_mq_tagset_busy_iter(hctx->queue->tag_set, hctx_show_busy_rq,
 443                                 &params);
 444
 445         return 0;
 446 }
 447
 448 static const char *const hctx_types[] = {
 449         [HCTX_TYPE_DEFAULT]     = "default",
 450         [HCTX_TYPE_READ]        = "read",
 451         [HCTX_TYPE_POLL]        = "poll",
 452 };
 453
 454 static int hctx_type_show(void *data, struct seq_file *m)
 455 {
 456         struct blk_mq_hw_ctx *hctx = data;
 457
 458         BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
 459         seq_printf(m, "%s\n", hctx_types[hctx->type]);
 460         return 0;
 461 }
 462
 463 static int hctx_ctx_map_show(void *data, struct seq_file *m)
 464 {
 465         struct blk_mq_hw_ctx *hctx = data;
 466
 467         sbitmap_bitmap_show(&hctx->ctx_map, m);
 468         return 0;
 469 }
 470
 471 static void blk_mq_debugfs_tags_show(struct seq_file *m,
 472                                      struct blk_mq_tags *tags)
 473 {
 474         seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
 475         seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
 476         seq_printf(m, "active_queues=%d\n",
 477                    atomic_read(&tags->active_queues));
 478
 479         seq_puts(m, "\nbitmap_tags:\n");
 480         sbitmap_queue_show(&tags->bitmap_tags, m);
 481
 482         if (tags->nr_reserved_tags) {
 483                 seq_puts(m, "\nbreserved_tags:\n");
 484                 sbitmap_queue_show(&tags->breserved_tags, m);
 485         }
 486 }
 487
 488 static int hctx_tags_show(void *data, struct seq_file *m)
 489 {
 490         struct blk_mq_hw_ctx *hctx = data;
 491         struct request_queue *q = hctx->queue;
 492         int res;
 493
 494         res = mutex_lock_interruptible(&q->sysfs_lock);
 495         if (res)
 496                 goto out;
 497         if (hctx->tags)
 498                 blk_mq_debugfs_tags_show(m, hctx->tags);
 499         mutex_unlock(&q->sysfs_lock);
 500
 501 out:
 502         return res;
 503 }
 504
 505 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
 506 {
 507         struct blk_mq_hw_ctx *hctx = data;
 508         struct request_queue *q = hctx->queue;
 509         int res;
 510
 511         res = mutex_lock_interruptible(&q->sysfs_lock);
 512         if (res)
 513                 goto out;
 514         if (hctx->tags)
 515                 sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
 516         mutex_unlock(&q->sysfs_lock);
 517
 518 out:
 519         return res;
 520 }
 521
 522 static int hctx_sched_tags_show(void *data, struct seq_file *m)
 523 {
 524         struct blk_mq_hw_ctx *hctx = data;
 525         struct request_queue *q = hctx->queue;
 526         int res;
 527
 528         res = mutex_lock_interruptible(&q->sysfs_lock);
 529         if (res)
 530                 goto out;
 531         if (hctx->sched_tags)
 532                 blk_mq_debugfs_tags_show(m, hctx->sched_tags);
 533         mutex_unlock(&q->sysfs_lock);
 534
 535 out:
 536         return res;
 537 }
 538
 539 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
 540 {
 541         struct blk_mq_hw_ctx *hctx = data;
 542         struct request_queue *q = hctx->queue;
 543         int res;
 544
 545         res = mutex_lock_interruptible(&q->sysfs_lock);
 546         if (res)
 547                 goto out;
 548         if (hctx->sched_tags)
 549                 sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
 550         mutex_unlock(&q->sysfs_lock);
 551
 552 out:
 553         return res;
 554 }
 555
 556 static int hctx_io_poll_show(void *data, struct seq_file *m)
 557 {
 558         struct blk_mq_hw_ctx *hctx = data;
 559
 560         seq_printf(m, "considered=%lu\n", hctx->poll_considered);
 561         seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
 562         seq_printf(m, "success=%lu\n", hctx->poll_success);
 563         return 0;
 564 }
 565
 566 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
 567                                   size_t count, loff_t *ppos)
 568 {
 569         struct blk_mq_hw_ctx *hctx = data;
 570
 571         hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
 572         return count;
 573 }
 574
 575 static int hctx_dispatched_show(void *data, struct seq_file *m)
 576 {
 577         struct blk_mq_hw_ctx *hctx = data;
 578         int i;
 579
 580         seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
 581
 582         for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
 583                 unsigned int d = 1U << (i - 1);
 584
 585                 seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
 586         }
 587
 588         seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
 589         return 0;
 590 }
 591
 592 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
 593                                      size_t count, loff_t *ppos)
 594 {
 595         struct blk_mq_hw_ctx *hctx = data;
 596         int i;
 597
 598         for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
 599                 hctx->dispatched[i] = 0;
 600         return count;
 601 }
 602
 603 static int hctx_queued_show(void *data, struct seq_file *m)
 604 {
 605         struct blk_mq_hw_ctx *hctx = data;
 606
 607         seq_printf(m, "%lu\n", hctx->queued);
 608         return 0;
 609 }
 610
 611 static ssize_t hctx_queued_write(void *data, const char __user *buf,
 612                                  size_t count, loff_t *ppos)
 613 {
 614         struct blk_mq_hw_ctx *hctx = data;
 615
 616         hctx->queued = 0;
 617         return count;
 618 }
 619
 620 static int hctx_run_show(void *data, struct seq_file *m)
 621 {
 622         struct blk_mq_hw_ctx *hctx = data;
 623
 624         seq_printf(m, "%lu\n", hctx->run);
 625         return 0;
 626 }
 627
 628 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
 629                               loff_t *ppos)
 630 {
 631         struct blk_mq_hw_ctx *hctx = data;
 632
 633         hctx->run = 0;
 634         return count;
 635 }
 636
 637 static int hctx_active_show(void *data, struct seq_file *m)
 638 {
 639         struct blk_mq_hw_ctx *hctx = data;
 640
 641         seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
 642         return 0;
 643 }
 644
 645 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
 646 {
 647         struct blk_mq_hw_ctx *hctx = data;
 648
 649         seq_printf(m, "%u\n", hctx->dispatch_busy);
 650         return 0;
 651 }
 652
 653 #define CTX_RQ_SEQ_OPS(name, type)                                      \
 654 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
 655         __acquires(&ctx->lock)                                          \
 656 {                                                                       \
 657         struct blk_mq_ctx *ctx = m->private;                            \
 658                                                                         \
 659         spin_lock(&ctx->lock);                                          \
 660         return seq_list_start(&ctx->rq_lists[type], *pos);              \
 661 }                                                                       \
 662                                                                         \
 663 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,     \
 664                                      loff_t *pos)                       \
 665 {                                                                       \
 666         struct blk_mq_ctx *ctx = m->private;                            \
 667                                                                         \
 668         return seq_list_next(v, &ctx->rq_lists[type], pos);             \
 669 }                                                                       \
 670                                                                         \
 671 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)      \
 672         __releases(&ctx->lock)                                          \
 673 {                                                                       \
 674         struct blk_mq_ctx *ctx = m->private;                            \
 675                                                                         \
 676         spin_unlock(&ctx->lock);                                        \
 677 }                                                                       \
 678                                                                         \
 679 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {     \
 680         .start  = ctx_##name##_rq_list_start,                           \
 681         .next   = ctx_##name##_rq_list_next,                            \
 682         .stop   = ctx_##name##_rq_list_stop,                            \
 683         .show   = blk_mq_debugfs_rq_show,                               \
 684 }
 685
 686 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
 687 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
 688 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
 689
 690 static int ctx_dispatched_show(void *data, struct seq_file *m)
 691 {
 692         struct blk_mq_ctx *ctx = data;
 693
 694         seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
 695         return 0;
 696 }
 697
 698 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
 699                                     size_t count, loff_t *ppos)
 700 {
 701         struct blk_mq_ctx *ctx = data;
 702
 703         ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
 704         return count;
 705 }
 706
 707 static int ctx_merged_show(void *data, struct seq_file *m)
 708 {
 709         struct blk_mq_ctx *ctx = data;
 710
 711         seq_printf(m, "%lu\n", ctx->rq_merged);
 712         return 0;
 713 }
 714
 715 static ssize_t ctx_merged_write(void *data, const char __user *buf,
 716                                 size_t count, loff_t *ppos)
 717 {
 718         struct blk_mq_ctx *ctx = data;
 719
 720         ctx->rq_merged = 0;
 721         return count;
 722 }
 723
 724 static int ctx_completed_show(void *data, struct seq_file *m)
 725 {
 726         struct blk_mq_ctx *ctx = data;
 727
 728         seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
 729         return 0;
 730 }
 731
 732 static ssize_t ctx_completed_write(void *data, const char __user *buf,
 733                                    size_t count, loff_t *ppos)
 734 {
 735         struct blk_mq_ctx *ctx = data;
 736
 737         ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
 738         return count;
 739 }
 740
 741 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
 742 {
 743         const struct blk_mq_debugfs_attr *attr = m->private;
 744         void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
 745
 746         return attr->show(data, m);
 747 }
 748
 749 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
 750                                     size_t count, loff_t *ppos)
 751 {
 752         struct seq_file *m = file->private_data;
 753         const struct blk_mq_debugfs_attr *attr = m->private;
 754         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 755
 756         /*
 757          * Attributes that only implement .seq_ops are read-only and 'attr' is
 758          * the same with 'data' in this case.
 759          */
 760         if (attr == data || !attr->write)
 761                 return -EPERM;
 762
 763         return attr->write(data, buf, count, ppos);
 764 }
 765
 766 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
 767 {
 768         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 769         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 770         struct seq_file *m;
 771         int ret;
 772
 773         if (attr->seq_ops) {
 774                 ret = seq_open(file, attr->seq_ops);
 775                 if (!ret) {
 776                         m = file->private_data;
 777                         m->private = data;
 778                 }
 779                 return ret;
 780         }
 781
 782         if (WARN_ON_ONCE(!attr->show))
 783                 return -EPERM;
 784
 785         return single_open(file, blk_mq_debugfs_show, inode->i_private);
 786 }
 787
 788 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
 789 {
 790         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 791
 792         if (attr->show)
 793                 return single_release(inode, file);
 794         else
 795                 return seq_release(inode, file);
 796 }
 797
 798 static const struct file_operations blk_mq_debugfs_fops = {
 799         .open           = blk_mq_debugfs_open,
 800         .read           = seq_read,
 801         .write          = blk_mq_debugfs_write,
 802         .llseek         = seq_lseek,
 803         .release        = blk_mq_debugfs_release,
 804 };
 805
 806 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
 807         {"state", 0400, hctx_state_show},
 808         {"flags", 0400, hctx_flags_show},
 809         {"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
 810         {"busy", 0400, hctx_busy_show},
 811         {"ctx_map", 0400, hctx_ctx_map_show},
 812         {"tags", 0400, hctx_tags_show},
 813         {"tags_bitmap", 0400, hctx_tags_bitmap_show},
 814         {"sched_tags", 0400, hctx_sched_tags_show},
 815         {"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
 816         {"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
 817         {"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
 818         {"queued", 0600, hctx_queued_show, hctx_queued_write},
 819         {"run", 0600, hctx_run_show, hctx_run_write},
 820         {"active", 0400, hctx_active_show},
 821         {"dispatch_busy", 0400, hctx_dispatch_busy_show},
 822         {"type", 0400, hctx_type_show},
 823         {},
 824 };
 825
 826 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
 827         {"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
 828         {"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
 829         {"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
 830         {"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
 831         {"merged", 0600, ctx_merged_show, ctx_merged_write},
 832         {"completed", 0600, ctx_completed_show, ctx_completed_write},
 833         {},
 834 };
 835
 836 static bool debugfs_create_files(struct dentry *parent, void *data,
 837                                  const struct blk_mq_debugfs_attr *attr)
 838 {
 839         if (IS_ERR_OR_NULL(parent))
 840                 return false;
 841
 842         d_inode(parent)->i_private = data;
 843
 844         for (; attr->name; attr++) {
 845                 if (!debugfs_create_file(attr->name, attr->mode, parent,
 846                                          (void *)attr, &blk_mq_debugfs_fops))
 847                         return false;
 848         }
 849         return true;
 850 }
 851
 852 int blk_mq_debugfs_register(struct request_queue *q)
 853 {
 854         struct blk_mq_hw_ctx *hctx;
 855         int i;
 856
 857         if (!blk_debugfs_root)
 858                 return -ENOENT;
 859
 860         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 861                                             blk_debugfs_root);
 862         if (!q->debugfs_dir)
 863                 return -ENOMEM;
 864
 865         if (!debugfs_create_files(q->debugfs_dir, q,
 866                                   blk_mq_debugfs_queue_attrs))
 867                 goto err;
 868
 869         /*
 870          * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
 871          * didn't exist yet (because we don't know what to name the directory
 872          * until the queue is registered to a gendisk).
 873          */
 874         if (q->elevator && !q->sched_debugfs_dir)
 875                 blk_mq_debugfs_register_sched(q);
 876
 877         /* Similarly, blk_mq_init_hctx() couldn't do this previously. */
 878         queue_for_each_hw_ctx(q, hctx, i) {
 879                 if (!hctx->debugfs_dir && blk_mq_debugfs_register_hctx(q, hctx))
 880                         goto err;
 881                 if (q->elevator && !hctx->sched_debugfs_dir &&
 882                     blk_mq_debugfs_register_sched_hctx(q, hctx))
 883                         goto err;
 884         }
 885
 886         if (q->rq_qos) {
 887                 struct rq_qos *rqos = q->rq_qos;
 888
 889                 while (rqos) {
 890                         blk_mq_debugfs_register_rqos(rqos);
 891                         rqos = rqos->next;
 892                 }
 893         }
 894
 895         return 0;
 896
 897 err:
 898         blk_mq_debugfs_unregister(q);
 899         return -ENOMEM;
 900 }
 901
 902 void blk_mq_debugfs_unregister(struct request_queue *q)
 903 {
 904         debugfs_remove_recursive(q->debugfs_dir);
 905         q->sched_debugfs_dir = NULL;
 906         q->debugfs_dir = NULL;
 907 }
 908
 909 static int blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
 910                                        struct blk_mq_ctx *ctx)
 911 {
 912         struct dentry *ctx_dir;
 913         char name[20];
 914
 915         snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
 916         ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
 917         if (!ctx_dir)
 918                 return -ENOMEM;
 919
 920         if (!debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs))
 921                 return -ENOMEM;
 922
 923         return 0;
 924 }
 925
 926 int blk_mq_debugfs_register_hctx(struct request_queue *q,
 927                                  struct blk_mq_hw_ctx *hctx)
 928 {
 929         struct blk_mq_ctx *ctx;
 930         char name[20];
 931         int i;
 932
 933         if (!q->debugfs_dir)
 934                 return -ENOENT;
 935
 936         snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
 937         hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
 938         if (!hctx->debugfs_dir)
 939                 return -ENOMEM;
 940
 941         if (!debugfs_create_files(hctx->debugfs_dir, hctx,
 942                                   blk_mq_debugfs_hctx_attrs))
 943                 goto err;
 944
 945         hctx_for_each_ctx(hctx, ctx, i) {
 946                 if (blk_mq_debugfs_register_ctx(hctx, ctx))
 947                         goto err;
 948         }
 949
 950         return 0;
 951
 952 err:
 953         blk_mq_debugfs_unregister_hctx(hctx);
 954         return -ENOMEM;
 955 }
 956
 957 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
 958 {
 959         debugfs_remove_recursive(hctx->debugfs_dir);
 960         hctx->sched_debugfs_dir = NULL;
 961         hctx->debugfs_dir = NULL;
 962 }
 963
 964 int blk_mq_debugfs_register_hctxs(struct request_queue *q)
 965 {
 966         struct blk_mq_hw_ctx *hctx;
 967         int i;
 968
 969         queue_for_each_hw_ctx(q, hctx, i) {
 970                 if (blk_mq_debugfs_register_hctx(q, hctx))
 971                         return -ENOMEM;
 972         }
 973
 974         return 0;
 975 }
 976
 977 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
 978 {
 979         struct blk_mq_hw_ctx *hctx;
 980         int i;
 981
 982         queue_for_each_hw_ctx(q, hctx, i)
 983                 blk_mq_debugfs_unregister_hctx(hctx);
 984 }
 985
 986 int blk_mq_debugfs_register_sched(struct request_queue *q)
 987 {
 988         struct elevator_type *e = q->elevator->type;
 989
 990         if (!q->debugfs_dir)
 991                 return -ENOENT;
 992
 993         if (!e->queue_debugfs_attrs)
 994                 return 0;
 995
 996         q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
 997         if (!q->sched_debugfs_dir)
 998                 return -ENOMEM;
 999
1000         if (!debugfs_create_files(q->sched_debugfs_dir, q,
1001                                   e->queue_debugfs_attrs))
1002                 goto err;
1003
1004         return 0;
1005
1006 err:
1007         blk_mq_debugfs_unregister_sched(q);
1008         return -ENOMEM;
1009 }
1010
1011 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
1012 {
1013         debugfs_remove_recursive(q->sched_debugfs_dir);
1014         q->sched_debugfs_dir = NULL;
1015 }
1016
1017 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
1018 {
1019         debugfs_remove_recursive(rqos->debugfs_dir);
1020         rqos->debugfs_dir = NULL;
1021 }
1022
1023 int blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
1024 {
1025         struct request_queue *q = rqos->q;
1026         const char *dir_name = rq_qos_id_to_name(rqos->id);
1027
1028         if (!q->debugfs_dir)
1029                 return -ENOENT;
1030
1031         if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
1032                 return 0;
1033
1034         if (!q->rqos_debugfs_dir) {
1035                 q->rqos_debugfs_dir = debugfs_create_dir("rqos",
1036                                                          q->debugfs_dir);
1037                 if (!q->rqos_debugfs_dir)
1038                         return -ENOMEM;
1039         }
1040
1041         rqos->debugfs_dir = debugfs_create_dir(dir_name,
1042                                                rqos->q->rqos_debugfs_dir);
1043         if (!rqos->debugfs_dir)
1044                 return -ENOMEM;
1045
1046         if (!debugfs_create_files(rqos->debugfs_dir, rqos,
1047                                   rqos->ops->debugfs_attrs))
1048                 goto err;
1049         return 0;
1050  err:
1051         blk_mq_debugfs_unregister_rqos(rqos);
1052         return -ENOMEM;
1053 }
1054
1055 void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q)
1056 {
1057         debugfs_remove_recursive(q->rqos_debugfs_dir);
1058         q->rqos_debugfs_dir = NULL;
1059 }
1060
1061 int blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
1062                                        struct blk_mq_hw_ctx *hctx)
1063 {
1064         struct elevator_type *e = q->elevator->type;
1065
1066         if (!hctx->debugfs_dir)
1067                 return -ENOENT;
1068
1069         if (!e->hctx_debugfs_attrs)
1070                 return 0;
1071
1072         hctx->sched_debugfs_dir = debugfs_create_dir("sched",
1073                                                      hctx->debugfs_dir);
1074         if (!hctx->sched_debugfs_dir)
1075                 return -ENOMEM;
1076
1077         if (!debugfs_create_files(hctx->sched_debugfs_dir, hctx,
1078                                   e->hctx_debugfs_attrs))
1079                 return -ENOMEM;
1080
1081         return 0;
1082 }
1083
1084 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1085 {
1086         debugfs_remove_recursive(hctx->sched_debugfs_dir);
1087         hctx->sched_debugfs_dir = NULL;
1088 }