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(NO_SG_MERGE),
 132         QUEUE_FLAG_NAME(POLL),
 133         QUEUE_FLAG_NAME(WC),
 134         QUEUE_FLAG_NAME(FUA),
 135         QUEUE_FLAG_NAME(FLUSH_NQ),
 136         QUEUE_FLAG_NAME(DAX),
 137         QUEUE_FLAG_NAME(STATS),
 138         QUEUE_FLAG_NAME(POLL_STATS),
 139         QUEUE_FLAG_NAME(REGISTERED),
 140         QUEUE_FLAG_NAME(SCSI_PASSTHROUGH),
 141         QUEUE_FLAG_NAME(QUIESCED),
 142 };
 143 #undef QUEUE_FLAG_NAME
 144
 145 static int queue_state_show(void *data, struct seq_file *m)
 146 {
 147         struct request_queue *q = data;
 148
 149         blk_flags_show(m, q->queue_flags, blk_queue_flag_name,
 150                        ARRAY_SIZE(blk_queue_flag_name));
 151         seq_puts(m, "\n");
 152         return 0;
 153 }
 154
 155 static ssize_t queue_state_write(void *data, const char __user *buf,
 156                                  size_t count, loff_t *ppos)
 157 {
 158         struct request_queue *q = data;
 159         char opbuf[16] = { }, *op;
 160
 161         /*
 162          * The "state" attribute is removed after blk_cleanup_queue() has called
 163          * blk_mq_free_queue(). Return if QUEUE_FLAG_DEAD has been set to avoid
 164          * triggering a use-after-free.
 165          */
 166         if (blk_queue_dead(q))
 167                 return -ENOENT;
 168
 169         if (count >= sizeof(opbuf)) {
 170                 pr_err("%s: operation too long\n", __func__);
 171                 goto inval;
 172         }
 173
 174         if (copy_from_user(opbuf, buf, count))
 175                 return -EFAULT;
 176         op = strstrip(opbuf);
 177         if (strcmp(op, "run") == 0) {
 178                 blk_mq_run_hw_queues(q, true);
 179         } else if (strcmp(op, "start") == 0) {
 180                 blk_mq_start_stopped_hw_queues(q, true);
 181         } else if (strcmp(op, "kick") == 0) {
 182                 blk_mq_kick_requeue_list(q);
 183         } else {
 184                 pr_err("%s: unsupported operation '%s'\n", __func__, op);
 185 inval:
 186                 pr_err("%s: use 'run', 'start' or 'kick'\n", __func__);
 187                 return -EINVAL;
 188         }
 189         return count;
 190 }
 191
 192 static int queue_write_hint_show(void *data, struct seq_file *m)
 193 {
 194         struct request_queue *q = data;
 195         int i;
 196
 197         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 198                 seq_printf(m, "hint%d: %llu\n", i, q->write_hints[i]);
 199
 200         return 0;
 201 }
 202
 203 static ssize_t queue_write_hint_store(void *data, const char __user *buf,
 204                                       size_t count, loff_t *ppos)
 205 {
 206         struct request_queue *q = data;
 207         int i;
 208
 209         for (i = 0; i < BLK_MAX_WRITE_HINTS; i++)
 210                 q->write_hints[i] = 0;
 211
 212         return count;
 213 }
 214
 215 static const struct blk_mq_debugfs_attr blk_mq_debugfs_queue_attrs[] = {
 216         { "poll_stat", 0400, queue_poll_stat_show },
 217         { "requeue_list", 0400, .seq_ops = &queue_requeue_list_seq_ops },
 218         { "pm_only", 0600, queue_pm_only_show, NULL },
 219         { "state", 0600, queue_state_show, queue_state_write },
 220         { "write_hints", 0600, queue_write_hint_show, queue_write_hint_store },
 221         { "zone_wlock", 0400, queue_zone_wlock_show, NULL },
 222         { },
 223 };
 224
 225 #define HCTX_STATE_NAME(name) [BLK_MQ_S_##name] = #name
 226 static const char *const hctx_state_name[] = {
 227         HCTX_STATE_NAME(STOPPED),
 228         HCTX_STATE_NAME(TAG_ACTIVE),
 229         HCTX_STATE_NAME(SCHED_RESTART),
 230 };
 231 #undef HCTX_STATE_NAME
 232
 233 static int hctx_state_show(void *data, struct seq_file *m)
 234 {
 235         struct blk_mq_hw_ctx *hctx = data;
 236
 237         blk_flags_show(m, hctx->state, hctx_state_name,
 238                        ARRAY_SIZE(hctx_state_name));
 239         seq_puts(m, "\n");
 240         return 0;
 241 }
 242
 243 #define BLK_TAG_ALLOC_NAME(name) [BLK_TAG_ALLOC_##name] = #name
 244 static const char *const alloc_policy_name[] = {
 245         BLK_TAG_ALLOC_NAME(FIFO),
 246         BLK_TAG_ALLOC_NAME(RR),
 247 };
 248 #undef BLK_TAG_ALLOC_NAME
 249
 250 #define HCTX_FLAG_NAME(name) [ilog2(BLK_MQ_F_##name)] = #name
 251 static const char *const hctx_flag_name[] = {
 252         HCTX_FLAG_NAME(SHOULD_MERGE),
 253         HCTX_FLAG_NAME(TAG_SHARED),
 254         HCTX_FLAG_NAME(SG_MERGE),
 255         HCTX_FLAG_NAME(BLOCKING),
 256         HCTX_FLAG_NAME(NO_SCHED),
 257 };
 258 #undef HCTX_FLAG_NAME
 259
 260 static int hctx_flags_show(void *data, struct seq_file *m)
 261 {
 262         struct blk_mq_hw_ctx *hctx = data;
 263         const int alloc_policy = BLK_MQ_FLAG_TO_ALLOC_POLICY(hctx->flags);
 264
 265         seq_puts(m, "alloc_policy=");
 266         if (alloc_policy < ARRAY_SIZE(alloc_policy_name) &&
 267             alloc_policy_name[alloc_policy])
 268                 seq_puts(m, alloc_policy_name[alloc_policy]);
 269         else
 270                 seq_printf(m, "%d", alloc_policy);
 271         seq_puts(m, " ");
 272         blk_flags_show(m,
 273                        hctx->flags ^ BLK_ALLOC_POLICY_TO_MQ_FLAG(alloc_policy),
 274                        hctx_flag_name, ARRAY_SIZE(hctx_flag_name));
 275         seq_puts(m, "\n");
 276         return 0;
 277 }
 278
 279 #define REQ_OP_NAME(name) [REQ_OP_##name] = #name
 280 static const char *const op_name[] = {
 281         REQ_OP_NAME(READ),
 282         REQ_OP_NAME(WRITE),
 283         REQ_OP_NAME(FLUSH),
 284         REQ_OP_NAME(DISCARD),
 285         REQ_OP_NAME(SECURE_ERASE),
 286         REQ_OP_NAME(ZONE_RESET),
 287         REQ_OP_NAME(WRITE_SAME),
 288         REQ_OP_NAME(WRITE_ZEROES),
 289         REQ_OP_NAME(SCSI_IN),
 290         REQ_OP_NAME(SCSI_OUT),
 291         REQ_OP_NAME(DRV_IN),
 292         REQ_OP_NAME(DRV_OUT),
 293 };
 294 #undef REQ_OP_NAME
 295
 296 #define CMD_FLAG_NAME(name) [__REQ_##name] = #name
 297 static const char *const cmd_flag_name[] = {
 298         CMD_FLAG_NAME(FAILFAST_DEV),
 299         CMD_FLAG_NAME(FAILFAST_TRANSPORT),
 300         CMD_FLAG_NAME(FAILFAST_DRIVER),
 301         CMD_FLAG_NAME(SYNC),
 302         CMD_FLAG_NAME(META),
 303         CMD_FLAG_NAME(PRIO),
 304         CMD_FLAG_NAME(NOMERGE),
 305         CMD_FLAG_NAME(IDLE),
 306         CMD_FLAG_NAME(INTEGRITY),
 307         CMD_FLAG_NAME(FUA),
 308         CMD_FLAG_NAME(PREFLUSH),
 309         CMD_FLAG_NAME(RAHEAD),
 310         CMD_FLAG_NAME(BACKGROUND),
 311         CMD_FLAG_NAME(NOWAIT),
 312         CMD_FLAG_NAME(NOUNMAP),
 313         CMD_FLAG_NAME(HIPRI),
 314 };
 315 #undef CMD_FLAG_NAME
 316
 317 #define RQF_NAME(name) [ilog2((__force u32)RQF_##name)] = #name
 318 static const char *const rqf_name[] = {
 319         RQF_NAME(SORTED),
 320         RQF_NAME(STARTED),
 321         RQF_NAME(SOFTBARRIER),
 322         RQF_NAME(FLUSH_SEQ),
 323         RQF_NAME(MIXED_MERGE),
 324         RQF_NAME(MQ_INFLIGHT),
 325         RQF_NAME(DONTPREP),
 326         RQF_NAME(PREEMPT),
 327         RQF_NAME(COPY_USER),
 328         RQF_NAME(FAILED),
 329         RQF_NAME(QUIET),
 330         RQF_NAME(ELVPRIV),
 331         RQF_NAME(IO_STAT),
 332         RQF_NAME(ALLOCED),
 333         RQF_NAME(PM),
 334         RQF_NAME(HASHED),
 335         RQF_NAME(STATS),
 336         RQF_NAME(SPECIAL_PAYLOAD),
 337         RQF_NAME(ZONE_WRITE_LOCKED),
 338         RQF_NAME(MQ_POLL_SLEPT),
 339 };
 340 #undef RQF_NAME
 341
 342 static const char *const blk_mq_rq_state_name_array[] = {
 343         [MQ_RQ_IDLE]            = "idle",
 344         [MQ_RQ_IN_FLIGHT]       = "in_flight",
 345         [MQ_RQ_COMPLETE]        = "complete",
 346 };
 347
 348 static const char *blk_mq_rq_state_name(enum mq_rq_state rq_state)
 349 {
 350         if (WARN_ON_ONCE((unsigned int)rq_state >=
 351                          ARRAY_SIZE(blk_mq_rq_state_name_array)))
 352                 return "(?)";
 353         return blk_mq_rq_state_name_array[rq_state];
 354 }
 355
 356 int __blk_mq_debugfs_rq_show(struct seq_file *m, struct request *rq)
 357 {
 358         const struct blk_mq_ops *const mq_ops = rq->q->mq_ops;
 359         const unsigned int op = rq->cmd_flags & REQ_OP_MASK;
 360
 361         seq_printf(m, "%p {.op=", rq);
 362         if (op < ARRAY_SIZE(op_name) && op_name[op])
 363                 seq_printf(m, "%s", op_name[op]);
 364         else
 365                 seq_printf(m, "%d", op);
 366         seq_puts(m, ", .cmd_flags=");
 367         blk_flags_show(m, rq->cmd_flags & ~REQ_OP_MASK, cmd_flag_name,
 368                        ARRAY_SIZE(cmd_flag_name));
 369         seq_puts(m, ", .rq_flags=");
 370         blk_flags_show(m, (__force unsigned int)rq->rq_flags, rqf_name,
 371                        ARRAY_SIZE(rqf_name));
 372         seq_printf(m, ", .state=%s", blk_mq_rq_state_name(blk_mq_rq_state(rq)));
 373         seq_printf(m, ", .tag=%d, .internal_tag=%d", rq->tag,
 374                    rq->internal_tag);
 375         if (mq_ops->show_rq)
 376                 mq_ops->show_rq(m, rq);
 377         seq_puts(m, "}\n");
 378         return 0;
 379 }
 380 EXPORT_SYMBOL_GPL(__blk_mq_debugfs_rq_show);
 381
 382 int blk_mq_debugfs_rq_show(struct seq_file *m, void *v)
 383 {
 384         return __blk_mq_debugfs_rq_show(m, list_entry_rq(v));
 385 }
 386 EXPORT_SYMBOL_GPL(blk_mq_debugfs_rq_show);
 387
 388 static void *hctx_dispatch_start(struct seq_file *m, loff_t *pos)
 389         __acquires(&hctx->lock)
 390 {
 391         struct blk_mq_hw_ctx *hctx = m->private;
 392
 393         spin_lock(&hctx->lock);
 394         return seq_list_start(&hctx->dispatch, *pos);
 395 }
 396
 397 static void *hctx_dispatch_next(struct seq_file *m, void *v, loff_t *pos)
 398 {
 399         struct blk_mq_hw_ctx *hctx = m->private;
 400
 401         return seq_list_next(v, &hctx->dispatch, pos);
 402 }
 403
 404 static void hctx_dispatch_stop(struct seq_file *m, void *v)
 405         __releases(&hctx->lock)
 406 {
 407         struct blk_mq_hw_ctx *hctx = m->private;
 408
 409         spin_unlock(&hctx->lock);
 410 }
 411
 412 static const struct seq_operations hctx_dispatch_seq_ops = {
 413         .start  = hctx_dispatch_start,
 414         .next   = hctx_dispatch_next,
 415         .stop   = hctx_dispatch_stop,
 416         .show   = blk_mq_debugfs_rq_show,
 417 };
 418
 419 struct show_busy_params {
 420         struct seq_file         *m;
 421         struct blk_mq_hw_ctx    *hctx;
 422 };
 423
 424 /*
 425  * Note: the state of a request may change while this function is in progress,
 426  * e.g. due to a concurrent blk_mq_finish_request() call. Returns true to
 427  * keep iterating requests.
 428  */
 429 static bool hctx_show_busy_rq(struct request *rq, void *data, bool reserved)
 430 {
 431         const struct show_busy_params *params = data;
 432
 433         if (rq->mq_hctx == params->hctx)
 434                 __blk_mq_debugfs_rq_show(params->m,
 435                                          list_entry_rq(&rq->queuelist));
 436
 437         return true;
 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 const char *const hctx_types[] = {
 452         [HCTX_TYPE_DEFAULT]     = "default",
 453         [HCTX_TYPE_READ]        = "read",
 454         [HCTX_TYPE_POLL]        = "poll",
 455 };
 456
 457 static int hctx_type_show(void *data, struct seq_file *m)
 458 {
 459         struct blk_mq_hw_ctx *hctx = data;
 460
 461         BUILD_BUG_ON(ARRAY_SIZE(hctx_types) != HCTX_MAX_TYPES);
 462         seq_printf(m, "%s\n", hctx_types[hctx->type]);
 463         return 0;
 464 }
 465
 466 static int hctx_ctx_map_show(void *data, struct seq_file *m)
 467 {
 468         struct blk_mq_hw_ctx *hctx = data;
 469
 470         sbitmap_bitmap_show(&hctx->ctx_map, m);
 471         return 0;
 472 }
 473
 474 static void blk_mq_debugfs_tags_show(struct seq_file *m,
 475                                      struct blk_mq_tags *tags)
 476 {
 477         seq_printf(m, "nr_tags=%u\n", tags->nr_tags);
 478         seq_printf(m, "nr_reserved_tags=%u\n", tags->nr_reserved_tags);
 479         seq_printf(m, "active_queues=%d\n",
 480                    atomic_read(&tags->active_queues));
 481
 482         seq_puts(m, "\nbitmap_tags:\n");
 483         sbitmap_queue_show(&tags->bitmap_tags, m);
 484
 485         if (tags->nr_reserved_tags) {
 486                 seq_puts(m, "\nbreserved_tags:\n");
 487                 sbitmap_queue_show(&tags->breserved_tags, m);
 488         }
 489 }
 490
 491 static int hctx_tags_show(void *data, struct seq_file *m)
 492 {
 493         struct blk_mq_hw_ctx *hctx = data;
 494         struct request_queue *q = hctx->queue;
 495         int res;
 496
 497         res = mutex_lock_interruptible(&q->sysfs_lock);
 498         if (res)
 499                 goto out;
 500         if (hctx->tags)
 501                 blk_mq_debugfs_tags_show(m, hctx->tags);
 502         mutex_unlock(&q->sysfs_lock);
 503
 504 out:
 505         return res;
 506 }
 507
 508 static int hctx_tags_bitmap_show(void *data, struct seq_file *m)
 509 {
 510         struct blk_mq_hw_ctx *hctx = data;
 511         struct request_queue *q = hctx->queue;
 512         int res;
 513
 514         res = mutex_lock_interruptible(&q->sysfs_lock);
 515         if (res)
 516                 goto out;
 517         if (hctx->tags)
 518                 sbitmap_bitmap_show(&hctx->tags->bitmap_tags.sb, m);
 519         mutex_unlock(&q->sysfs_lock);
 520
 521 out:
 522         return res;
 523 }
 524
 525 static int hctx_sched_tags_show(void *data, struct seq_file *m)
 526 {
 527         struct blk_mq_hw_ctx *hctx = data;
 528         struct request_queue *q = hctx->queue;
 529         int res;
 530
 531         res = mutex_lock_interruptible(&q->sysfs_lock);
 532         if (res)
 533                 goto out;
 534         if (hctx->sched_tags)
 535                 blk_mq_debugfs_tags_show(m, hctx->sched_tags);
 536         mutex_unlock(&q->sysfs_lock);
 537
 538 out:
 539         return res;
 540 }
 541
 542 static int hctx_sched_tags_bitmap_show(void *data, struct seq_file *m)
 543 {
 544         struct blk_mq_hw_ctx *hctx = data;
 545         struct request_queue *q = hctx->queue;
 546         int res;
 547
 548         res = mutex_lock_interruptible(&q->sysfs_lock);
 549         if (res)
 550                 goto out;
 551         if (hctx->sched_tags)
 552                 sbitmap_bitmap_show(&hctx->sched_tags->bitmap_tags.sb, m);
 553         mutex_unlock(&q->sysfs_lock);
 554
 555 out:
 556         return res;
 557 }
 558
 559 static int hctx_io_poll_show(void *data, struct seq_file *m)
 560 {
 561         struct blk_mq_hw_ctx *hctx = data;
 562
 563         seq_printf(m, "considered=%lu\n", hctx->poll_considered);
 564         seq_printf(m, "invoked=%lu\n", hctx->poll_invoked);
 565         seq_printf(m, "success=%lu\n", hctx->poll_success);
 566         return 0;
 567 }
 568
 569 static ssize_t hctx_io_poll_write(void *data, const char __user *buf,
 570                                   size_t count, loff_t *ppos)
 571 {
 572         struct blk_mq_hw_ctx *hctx = data;
 573
 574         hctx->poll_considered = hctx->poll_invoked = hctx->poll_success = 0;
 575         return count;
 576 }
 577
 578 static int hctx_dispatched_show(void *data, struct seq_file *m)
 579 {
 580         struct blk_mq_hw_ctx *hctx = data;
 581         int i;
 582
 583         seq_printf(m, "%8u\t%lu\n", 0U, hctx->dispatched[0]);
 584
 585         for (i = 1; i < BLK_MQ_MAX_DISPATCH_ORDER - 1; i++) {
 586                 unsigned int d = 1U << (i - 1);
 587
 588                 seq_printf(m, "%8u\t%lu\n", d, hctx->dispatched[i]);
 589         }
 590
 591         seq_printf(m, "%8u+\t%lu\n", 1U << (i - 1), hctx->dispatched[i]);
 592         return 0;
 593 }
 594
 595 static ssize_t hctx_dispatched_write(void *data, const char __user *buf,
 596                                      size_t count, loff_t *ppos)
 597 {
 598         struct blk_mq_hw_ctx *hctx = data;
 599         int i;
 600
 601         for (i = 0; i < BLK_MQ_MAX_DISPATCH_ORDER; i++)
 602                 hctx->dispatched[i] = 0;
 603         return count;
 604 }
 605
 606 static int hctx_queued_show(void *data, struct seq_file *m)
 607 {
 608         struct blk_mq_hw_ctx *hctx = data;
 609
 610         seq_printf(m, "%lu\n", hctx->queued);
 611         return 0;
 612 }
 613
 614 static ssize_t hctx_queued_write(void *data, const char __user *buf,
 615                                  size_t count, loff_t *ppos)
 616 {
 617         struct blk_mq_hw_ctx *hctx = data;
 618
 619         hctx->queued = 0;
 620         return count;
 621 }
 622
 623 static int hctx_run_show(void *data, struct seq_file *m)
 624 {
 625         struct blk_mq_hw_ctx *hctx = data;
 626
 627         seq_printf(m, "%lu\n", hctx->run);
 628         return 0;
 629 }
 630
 631 static ssize_t hctx_run_write(void *data, const char __user *buf, size_t count,
 632                               loff_t *ppos)
 633 {
 634         struct blk_mq_hw_ctx *hctx = data;
 635
 636         hctx->run = 0;
 637         return count;
 638 }
 639
 640 static int hctx_active_show(void *data, struct seq_file *m)
 641 {
 642         struct blk_mq_hw_ctx *hctx = data;
 643
 644         seq_printf(m, "%d\n", atomic_read(&hctx->nr_active));
 645         return 0;
 646 }
 647
 648 static int hctx_dispatch_busy_show(void *data, struct seq_file *m)
 649 {
 650         struct blk_mq_hw_ctx *hctx = data;
 651
 652         seq_printf(m, "%u\n", hctx->dispatch_busy);
 653         return 0;
 654 }
 655
 656 #define CTX_RQ_SEQ_OPS(name, type)                                      \
 657 static void *ctx_##name##_rq_list_start(struct seq_file *m, loff_t *pos) \
 658         __acquires(&ctx->lock)                                          \
 659 {                                                                       \
 660         struct blk_mq_ctx *ctx = m->private;                            \
 661                                                                         \
 662         spin_lock(&ctx->lock);                                          \
 663         return seq_list_start(&ctx->rq_lists[type], *pos);              \
 664 }                                                                       \
 665                                                                         \
 666 static void *ctx_##name##_rq_list_next(struct seq_file *m, void *v,     \
 667                                      loff_t *pos)                       \
 668 {                                                                       \
 669         struct blk_mq_ctx *ctx = m->private;                            \
 670                                                                         \
 671         return seq_list_next(v, &ctx->rq_lists[type], pos);             \
 672 }                                                                       \
 673                                                                         \
 674 static void ctx_##name##_rq_list_stop(struct seq_file *m, void *v)      \
 675         __releases(&ctx->lock)                                          \
 676 {                                                                       \
 677         struct blk_mq_ctx *ctx = m->private;                            \
 678                                                                         \
 679         spin_unlock(&ctx->lock);                                        \
 680 }                                                                       \
 681                                                                         \
 682 static const struct seq_operations ctx_##name##_rq_list_seq_ops = {     \
 683         .start  = ctx_##name##_rq_list_start,                           \
 684         .next   = ctx_##name##_rq_list_next,                            \
 685         .stop   = ctx_##name##_rq_list_stop,                            \
 686         .show   = blk_mq_debugfs_rq_show,                               \
 687 }
 688
 689 CTX_RQ_SEQ_OPS(default, HCTX_TYPE_DEFAULT);
 690 CTX_RQ_SEQ_OPS(read, HCTX_TYPE_READ);
 691 CTX_RQ_SEQ_OPS(poll, HCTX_TYPE_POLL);
 692
 693 static int ctx_dispatched_show(void *data, struct seq_file *m)
 694 {
 695         struct blk_mq_ctx *ctx = data;
 696
 697         seq_printf(m, "%lu %lu\n", ctx->rq_dispatched[1], ctx->rq_dispatched[0]);
 698         return 0;
 699 }
 700
 701 static ssize_t ctx_dispatched_write(void *data, const char __user *buf,
 702                                     size_t count, loff_t *ppos)
 703 {
 704         struct blk_mq_ctx *ctx = data;
 705
 706         ctx->rq_dispatched[0] = ctx->rq_dispatched[1] = 0;
 707         return count;
 708 }
 709
 710 static int ctx_merged_show(void *data, struct seq_file *m)
 711 {
 712         struct blk_mq_ctx *ctx = data;
 713
 714         seq_printf(m, "%lu\n", ctx->rq_merged);
 715         return 0;
 716 }
 717
 718 static ssize_t ctx_merged_write(void *data, const char __user *buf,
 719                                 size_t count, loff_t *ppos)
 720 {
 721         struct blk_mq_ctx *ctx = data;
 722
 723         ctx->rq_merged = 0;
 724         return count;
 725 }
 726
 727 static int ctx_completed_show(void *data, struct seq_file *m)
 728 {
 729         struct blk_mq_ctx *ctx = data;
 730
 731         seq_printf(m, "%lu %lu\n", ctx->rq_completed[1], ctx->rq_completed[0]);
 732         return 0;
 733 }
 734
 735 static ssize_t ctx_completed_write(void *data, const char __user *buf,
 736                                    size_t count, loff_t *ppos)
 737 {
 738         struct blk_mq_ctx *ctx = data;
 739
 740         ctx->rq_completed[0] = ctx->rq_completed[1] = 0;
 741         return count;
 742 }
 743
 744 static int blk_mq_debugfs_show(struct seq_file *m, void *v)
 745 {
 746         const struct blk_mq_debugfs_attr *attr = m->private;
 747         void *data = d_inode(m->file->f_path.dentry->d_parent)->i_private;
 748
 749         return attr->show(data, m);
 750 }
 751
 752 static ssize_t blk_mq_debugfs_write(struct file *file, const char __user *buf,
 753                                     size_t count, loff_t *ppos)
 754 {
 755         struct seq_file *m = file->private_data;
 756         const struct blk_mq_debugfs_attr *attr = m->private;
 757         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 758
 759         /*
 760          * Attributes that only implement .seq_ops are read-only and 'attr' is
 761          * the same with 'data' in this case.
 762          */
 763         if (attr == data || !attr->write)
 764                 return -EPERM;
 765
 766         return attr->write(data, buf, count, ppos);
 767 }
 768
 769 static int blk_mq_debugfs_open(struct inode *inode, struct file *file)
 770 {
 771         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 772         void *data = d_inode(file->f_path.dentry->d_parent)->i_private;
 773         struct seq_file *m;
 774         int ret;
 775
 776         if (attr->seq_ops) {
 777                 ret = seq_open(file, attr->seq_ops);
 778                 if (!ret) {
 779                         m = file->private_data;
 780                         m->private = data;
 781                 }
 782                 return ret;
 783         }
 784
 785         if (WARN_ON_ONCE(!attr->show))
 786                 return -EPERM;
 787
 788         return single_open(file, blk_mq_debugfs_show, inode->i_private);
 789 }
 790
 791 static int blk_mq_debugfs_release(struct inode *inode, struct file *file)
 792 {
 793         const struct blk_mq_debugfs_attr *attr = inode->i_private;
 794
 795         if (attr->show)
 796                 return single_release(inode, file);
 797         else
 798                 return seq_release(inode, file);
 799 }
 800
 801 static const struct file_operations blk_mq_debugfs_fops = {
 802         .open           = blk_mq_debugfs_open,
 803         .read           = seq_read,
 804         .write          = blk_mq_debugfs_write,
 805         .llseek         = seq_lseek,
 806         .release        = blk_mq_debugfs_release,
 807 };
 808
 809 static const struct blk_mq_debugfs_attr blk_mq_debugfs_hctx_attrs[] = {
 810         {"state", 0400, hctx_state_show},
 811         {"flags", 0400, hctx_flags_show},
 812         {"dispatch", 0400, .seq_ops = &hctx_dispatch_seq_ops},
 813         {"busy", 0400, hctx_busy_show},
 814         {"ctx_map", 0400, hctx_ctx_map_show},
 815         {"tags", 0400, hctx_tags_show},
 816         {"tags_bitmap", 0400, hctx_tags_bitmap_show},
 817         {"sched_tags", 0400, hctx_sched_tags_show},
 818         {"sched_tags_bitmap", 0400, hctx_sched_tags_bitmap_show},
 819         {"io_poll", 0600, hctx_io_poll_show, hctx_io_poll_write},
 820         {"dispatched", 0600, hctx_dispatched_show, hctx_dispatched_write},
 821         {"queued", 0600, hctx_queued_show, hctx_queued_write},
 822         {"run", 0600, hctx_run_show, hctx_run_write},
 823         {"active", 0400, hctx_active_show},
 824         {"dispatch_busy", 0400, hctx_dispatch_busy_show},
 825         {"type", 0400, hctx_type_show},
 826         {},
 827 };
 828
 829 static const struct blk_mq_debugfs_attr blk_mq_debugfs_ctx_attrs[] = {
 830         {"default_rq_list", 0400, .seq_ops = &ctx_default_rq_list_seq_ops},
 831         {"read_rq_list", 0400, .seq_ops = &ctx_read_rq_list_seq_ops},
 832         {"poll_rq_list", 0400, .seq_ops = &ctx_poll_rq_list_seq_ops},
 833         {"dispatched", 0600, ctx_dispatched_show, ctx_dispatched_write},
 834         {"merged", 0600, ctx_merged_show, ctx_merged_write},
 835         {"completed", 0600, ctx_completed_show, ctx_completed_write},
 836         {},
 837 };
 838
 839 static bool debugfs_create_files(struct dentry *parent, void *data,
 840                                  const struct blk_mq_debugfs_attr *attr)
 841 {
 842         if (IS_ERR_OR_NULL(parent))
 843                 return false;
 844
 845         d_inode(parent)->i_private = data;
 846
 847         for (; attr->name; attr++) {
 848                 if (!debugfs_create_file(attr->name, attr->mode, parent,
 849                                          (void *)attr, &blk_mq_debugfs_fops))
 850                         return false;
 851         }
 852         return true;
 853 }
 854
 855 int blk_mq_debugfs_register(struct request_queue *q)
 856 {
 857         struct blk_mq_hw_ctx *hctx;
 858         int i;
 859
 860         if (!blk_debugfs_root)
 861                 return -ENOENT;
 862
 863         q->debugfs_dir = debugfs_create_dir(kobject_name(q->kobj.parent),
 864                                             blk_debugfs_root);
 865         if (!q->debugfs_dir)
 866                 return -ENOMEM;
 867
 868         if (!debugfs_create_files(q->debugfs_dir, q,
 869                                   blk_mq_debugfs_queue_attrs))
 870                 goto err;
 871
 872         /*
 873          * blk_mq_init_sched() attempted to do this already, but q->debugfs_dir
 874          * didn't exist yet (because we don't know what to name the directory
 875          * until the queue is registered to a gendisk).
 876          */
 877         if (q->elevator && !q->sched_debugfs_dir)
 878                 blk_mq_debugfs_register_sched(q);
 879
 880         /* Similarly, blk_mq_init_hctx() couldn't do this previously. */
 881         queue_for_each_hw_ctx(q, hctx, i) {
 882                 if (!hctx->debugfs_dir && blk_mq_debugfs_register_hctx(q, hctx))
 883                         goto err;
 884                 if (q->elevator && !hctx->sched_debugfs_dir &&
 885                     blk_mq_debugfs_register_sched_hctx(q, hctx))
 886                         goto err;
 887         }
 888
 889         if (q->rq_qos) {
 890                 struct rq_qos *rqos = q->rq_qos;
 891
 892                 while (rqos) {
 893                         blk_mq_debugfs_register_rqos(rqos);
 894                         rqos = rqos->next;
 895                 }
 896         }
 897
 898         return 0;
 899
 900 err:
 901         blk_mq_debugfs_unregister(q);
 902         return -ENOMEM;
 903 }
 904
 905 void blk_mq_debugfs_unregister(struct request_queue *q)
 906 {
 907         debugfs_remove_recursive(q->debugfs_dir);
 908         q->sched_debugfs_dir = NULL;
 909         q->debugfs_dir = NULL;
 910 }
 911
 912 static int blk_mq_debugfs_register_ctx(struct blk_mq_hw_ctx *hctx,
 913                                        struct blk_mq_ctx *ctx)
 914 {
 915         struct dentry *ctx_dir;
 916         char name[20];
 917
 918         snprintf(name, sizeof(name), "cpu%u", ctx->cpu);
 919         ctx_dir = debugfs_create_dir(name, hctx->debugfs_dir);
 920         if (!ctx_dir)
 921                 return -ENOMEM;
 922
 923         if (!debugfs_create_files(ctx_dir, ctx, blk_mq_debugfs_ctx_attrs))
 924                 return -ENOMEM;
 925
 926         return 0;
 927 }
 928
 929 int blk_mq_debugfs_register_hctx(struct request_queue *q,
 930                                  struct blk_mq_hw_ctx *hctx)
 931 {
 932         struct blk_mq_ctx *ctx;
 933         char name[20];
 934         int i;
 935
 936         if (!q->debugfs_dir)
 937                 return -ENOENT;
 938
 939         snprintf(name, sizeof(name), "hctx%u", hctx->queue_num);
 940         hctx->debugfs_dir = debugfs_create_dir(name, q->debugfs_dir);
 941         if (!hctx->debugfs_dir)
 942                 return -ENOMEM;
 943
 944         if (!debugfs_create_files(hctx->debugfs_dir, hctx,
 945                                   blk_mq_debugfs_hctx_attrs))
 946                 goto err;
 947
 948         hctx_for_each_ctx(hctx, ctx, i) {
 949                 if (blk_mq_debugfs_register_ctx(hctx, ctx))
 950                         goto err;
 951         }
 952
 953         return 0;
 954
 955 err:
 956         blk_mq_debugfs_unregister_hctx(hctx);
 957         return -ENOMEM;
 958 }
 959
 960 void blk_mq_debugfs_unregister_hctx(struct blk_mq_hw_ctx *hctx)
 961 {
 962         debugfs_remove_recursive(hctx->debugfs_dir);
 963         hctx->sched_debugfs_dir = NULL;
 964         hctx->debugfs_dir = NULL;
 965 }
 966
 967 int blk_mq_debugfs_register_hctxs(struct request_queue *q)
 968 {
 969         struct blk_mq_hw_ctx *hctx;
 970         int i;
 971
 972         queue_for_each_hw_ctx(q, hctx, i) {
 973                 if (blk_mq_debugfs_register_hctx(q, hctx))
 974                         return -ENOMEM;
 975         }
 976
 977         return 0;
 978 }
 979
 980 void blk_mq_debugfs_unregister_hctxs(struct request_queue *q)
 981 {
 982         struct blk_mq_hw_ctx *hctx;
 983         int i;
 984
 985         queue_for_each_hw_ctx(q, hctx, i)
 986                 blk_mq_debugfs_unregister_hctx(hctx);
 987 }
 988
 989 int blk_mq_debugfs_register_sched(struct request_queue *q)
 990 {
 991         struct elevator_type *e = q->elevator->type;
 992
 993         if (!q->debugfs_dir)
 994                 return -ENOENT;
 995
 996         if (!e->queue_debugfs_attrs)
 997                 return 0;
 998
 999         q->sched_debugfs_dir = debugfs_create_dir("sched", q->debugfs_dir);
1000         if (!q->sched_debugfs_dir)
1001                 return -ENOMEM;
1002
1003         if (!debugfs_create_files(q->sched_debugfs_dir, q,
1004                                   e->queue_debugfs_attrs))
1005                 goto err;
1006
1007         return 0;
1008
1009 err:
1010         blk_mq_debugfs_unregister_sched(q);
1011         return -ENOMEM;
1012 }
1013
1014 void blk_mq_debugfs_unregister_sched(struct request_queue *q)
1015 {
1016         debugfs_remove_recursive(q->sched_debugfs_dir);
1017         q->sched_debugfs_dir = NULL;
1018 }
1019
1020 void blk_mq_debugfs_unregister_rqos(struct rq_qos *rqos)
1021 {
1022         debugfs_remove_recursive(rqos->debugfs_dir);
1023         rqos->debugfs_dir = NULL;
1024 }
1025
1026 int blk_mq_debugfs_register_rqos(struct rq_qos *rqos)
1027 {
1028         struct request_queue *q = rqos->q;
1029         const char *dir_name = rq_qos_id_to_name(rqos->id);
1030
1031         if (!q->debugfs_dir)
1032                 return -ENOENT;
1033
1034         if (rqos->debugfs_dir || !rqos->ops->debugfs_attrs)
1035                 return 0;
1036
1037         if (!q->rqos_debugfs_dir) {
1038                 q->rqos_debugfs_dir = debugfs_create_dir("rqos",
1039                                                          q->debugfs_dir);
1040                 if (!q->rqos_debugfs_dir)
1041                         return -ENOMEM;
1042         }
1043
1044         rqos->debugfs_dir = debugfs_create_dir(dir_name,
1045                                                rqos->q->rqos_debugfs_dir);
1046         if (!rqos->debugfs_dir)
1047                 return -ENOMEM;
1048
1049         if (!debugfs_create_files(rqos->debugfs_dir, rqos,
1050                                   rqos->ops->debugfs_attrs))
1051                 goto err;
1052         return 0;
1053  err:
1054         blk_mq_debugfs_unregister_rqos(rqos);
1055         return -ENOMEM;
1056 }
1057
1058 void blk_mq_debugfs_unregister_queue_rqos(struct request_queue *q)
1059 {
1060         debugfs_remove_recursive(q->rqos_debugfs_dir);
1061         q->rqos_debugfs_dir = NULL;
1062 }
1063
1064 int blk_mq_debugfs_register_sched_hctx(struct request_queue *q,
1065                                        struct blk_mq_hw_ctx *hctx)
1066 {
1067         struct elevator_type *e = q->elevator->type;
1068
1069         if (!hctx->debugfs_dir)
1070                 return -ENOENT;
1071
1072         if (!e->hctx_debugfs_attrs)
1073                 return 0;
1074
1075         hctx->sched_debugfs_dir = debugfs_create_dir("sched",
1076                                                      hctx->debugfs_dir);
1077         if (!hctx->sched_debugfs_dir)
1078                 return -ENOMEM;
1079
1080         if (!debugfs_create_files(hctx->sched_debugfs_dir, hctx,
1081                                   e->hctx_debugfs_attrs))
1082                 return -ENOMEM;
1083
1084         return 0;
1085 }
1086
1087 void blk_mq_debugfs_unregister_sched_hctx(struct blk_mq_hw_ctx *hctx)
1088 {
1089         debugfs_remove_recursive(hctx->sched_debugfs_dir);
1090         hctx->sched_debugfs_dir = NULL;
1091 }