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