blkparse.c

   1 /*
   2  * block queue tracing parse application
   3  *
   4  * Copyright (C) 2005 Jens Axboe <axboe@suse.de>
   5  * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
   6  *
   7  *  This program is free software; you can redistribute it and/or modify
   8  *  it under the terms of the GNU General Public License as published by
   9  *  the Free Software Foundation; either version 2 of the License, or
  10  *  (at your option) any later version.
  11  *
  12  *  This program is distributed in the hope that it will be useful,
  13  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *  GNU General Public License for more details.
  16  *
  17  *  You should have received a copy of the GNU General Public License
  18  *  along with this program; if not, write to the Free Software
  19  *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
  20  *
  21  */
  22 #include <sys/types.h>
  23 #include <sys/stat.h>
  24 #include <unistd.h>
  25 #include <stdio.h>
  26 #include <fcntl.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <getopt.h>
  30 #include <errno.h>
  31 #include <signal.h>
  32 #include <locale.h>
  33 #include <libgen.h>
  34
  35 #include "blktrace.h"
  36 #include "rbtree.h"
  37 #include "jhash.h"
  38
  39 static char blkparse_version[] = "1.0.1";
  40
  41 struct skip_info {
  42         unsigned long start, end;
  43         struct skip_info *prev, *next;
  44 };
  45
  46 struct per_dev_info {
  47         dev_t dev;
  48         char *name;
  49
  50         int backwards;
  51         unsigned long long events;
  52         unsigned long long first_reported_time;
  53         unsigned long long last_reported_time;
  54         unsigned long long last_read_time;
  55         struct io_stats io_stats;
  56         unsigned long skips;
  57         unsigned long long seq_skips;
  58         unsigned int max_depth[2];
  59         unsigned int cur_depth[2];
  60
  61         struct rb_root rb_track;
  62
  63         int nfiles;
  64         int ncpus;
  65
  66         unsigned long *cpu_map;
  67         unsigned int cpu_map_max;
  68
  69         struct per_cpu_info *cpus;
  70 };
  71
  72 /*
  73  * some duplicated effort here, we can unify this hash and the ppi hash later
  74  */
  75 struct process_pid_map {
  76         pid_t pid;
  77         char comm[16];
  78         struct process_pid_map *hash_next, *list_next;
  79 };
  80
  81 #define PPM_HASH_SHIFT  (8)
  82 #define PPM_HASH_SIZE   (1 << PPM_HASH_SHIFT)
  83 #define PPM_HASH_MASK   (PPM_HASH_SIZE - 1)
  84 static struct process_pid_map *ppm_hash_table[PPM_HASH_SIZE];
  85
  86 struct per_process_info {
  87         struct process_pid_map *ppm;
  88         struct io_stats io_stats;
  89         struct per_process_info *hash_next, *list_next;
  90         int more_than_one;
  91
  92         /*
  93          * individual io stats
  94          */
  95         unsigned long long longest_allocation_wait[2];
  96         unsigned long long longest_dispatch_wait[2];
  97         unsigned long long longest_completion_wait[2];
  98 };
  99
 100 #define PPI_HASH_SHIFT  (8)
 101 #define PPI_HASH_SIZE   (1 << PPI_HASH_SHIFT)
 102 #define PPI_HASH_MASK   (PPI_HASH_SIZE - 1)
 103 static struct per_process_info *ppi_hash_table[PPI_HASH_SIZE];
 104 static struct per_process_info *ppi_list;
 105 static int ppi_list_entries;
 106
 107 static struct option l_opts[] = {
 108         {
 109                 .name = "act-mask",
 110                 .has_arg = required_argument,
 111                 .flag = NULL,
 112                 .val = 'a'
 113         },
 114         {
 115                 .name = "set-mask",
 116                 .has_arg = required_argument,
 117                 .flag = NULL,
 118                 .val = 'A'
 119         },
 120         {
 121                 .name = "batch",
 122                 .has_arg = required_argument,
 123                 .flag = NULL,
 124                 .val = 'b'
 125         },
 126         {
 127                 .name = "input-directory",
 128                 .has_arg = required_argument,
 129                 .flag = NULL,
 130                 .val = 'D'
 131         },
 132         {
 133                 .name = "dump-binary",
 134                 .has_arg = required_argument,
 135                 .flag = NULL,
 136                 .val = 'd'
 137         },
 138         {
 139                 .name = "format",
 140                 .has_arg = required_argument,
 141                 .flag = NULL,
 142                 .val = 'f'
 143         },
 144         {
 145                 .name = "format-spec",
 146                 .has_arg = required_argument,
 147                 .flag = NULL,
 148                 .val = 'F'
 149         },
 150         {
 151                 .name = "hash-by-name",
 152                 .has_arg = no_argument,
 153                 .flag = NULL,
 154                 .val = 'h'
 155         },
 156         {
 157                 .name = "input",
 158                 .has_arg = required_argument,
 159                 .flag = NULL,
 160                 .val = 'i'
 161         },
 162         {
 163                 .name = "no-msgs",
 164                 .has_arg = no_argument,
 165                 .flag = NULL,
 166                 .val = 'M'
 167         },
 168         {
 169                 .name = "output",
 170                 .has_arg = required_argument,
 171                 .flag = NULL,
 172                 .val = 'o'
 173         },
 174         {
 175                 .name = "no-text-output",
 176                 .has_arg = no_argument,
 177                 .flag = NULL,
 178                 .val = 'O'
 179         },
 180         {
 181                 .name = "quiet",
 182                 .has_arg = no_argument,
 183                 .flag = NULL,
 184                 .val = 'q'
 185         },
 186         {
 187                 .name = "per-program-stats",
 188                 .has_arg = no_argument,
 189                 .flag = NULL,
 190                 .val = 's'
 191         },
 192         {
 193                 .name = "track-ios",
 194                 .has_arg = no_argument,
 195                 .flag = NULL,
 196                 .val = 't'
 197         },
 198         {
 199                 .name = "stopwatch",
 200                 .has_arg = required_argument,
 201                 .flag = NULL,
 202                 .val = 'w'
 203         },
 204         {
 205                 .name = "verbose",
 206                 .has_arg = no_argument,
 207                 .flag = NULL,
 208                 .val = 'v'
 209         },
 210         {
 211                 .name = "version",
 212                 .has_arg = no_argument,
 213                 .flag = NULL,
 214                 .val = 'V'
 215         },
 216         {
 217                 .name = NULL,
 218         }
 219 };
 220
 221 /*
 222  * for sorting the displayed output
 223  */
 224 struct trace {
 225         struct blk_io_trace *bit;
 226         struct rb_node rb_node;
 227         struct trace *next;
 228         unsigned long read_sequence;
 229 };
 230
 231 static struct rb_root rb_sort_root;
 232 static unsigned long rb_sort_entries;
 233
 234 static struct trace *trace_list;
 235
 236 /*
 237  * allocation cache
 238  */
 239 static struct blk_io_trace *bit_alloc_list;
 240 static struct trace *t_alloc_list;
 241
 242 /*
 243  * for tracking individual ios
 244  */
 245 struct io_track {
 246         struct rb_node rb_node;
 247
 248         struct process_pid_map *ppm;
 249         __u64 sector;
 250         unsigned long long allocation_time;
 251         unsigned long long queue_time;
 252         unsigned long long dispatch_time;
 253         unsigned long long completion_time;
 254 };
 255
 256 static int ndevices;
 257 static struct per_dev_info *devices;
 258 static char *get_dev_name(struct per_dev_info *, char *, int);
 259 static int trace_rb_insert_last(struct per_dev_info *, struct trace *);
 260
 261 FILE *ofp = NULL;
 262 static char *output_name;
 263 static char *input_dir;
 264
 265 static unsigned long long genesis_time;
 266 static unsigned long long last_allowed_time;
 267 static unsigned long long stopwatch_start;      /* start from zero by default */
 268 static unsigned long long stopwatch_end = -1ULL;        /* "infinity" */
 269 static unsigned long read_sequence;
 270
 271 static int per_process_stats;
 272 static int per_device_and_cpu_stats = 1;
 273 static int track_ios;
 274 static int ppi_hash_by_pid = 1;
 275 static int verbose;
 276 static unsigned int act_mask = -1U;
 277 static int stats_printed;
 278 static int bin_output_msgs = 1;
 279 int data_is_native = -1;
 280
 281 static FILE *dump_fp;
 282 static char *dump_binary;
 283
 284 static unsigned int t_alloc_cache;
 285 static unsigned int bit_alloc_cache;
 286
 287 #define RB_BATCH_DEFAULT        (512)
 288 static unsigned int rb_batch = RB_BATCH_DEFAULT;
 289
 290 static int pipeline;
 291 static char *pipename;
 292
 293 static int text_output = 1;
 294
 295 #define is_done()       (*(volatile int *)(&done))
 296 static volatile int done;
 297
 298 struct timespec         abs_start_time;
 299 static unsigned long long start_timestamp;
 300
 301 static int have_drv_data = 0;
 302
 303 #define JHASH_RANDOM    (0x3af5f2ee)
 304
 305 #define CPUS_PER_LONG   (8 * sizeof(unsigned long))
 306 #define CPU_IDX(cpu)    ((cpu) / CPUS_PER_LONG)
 307 #define CPU_BIT(cpu)    ((cpu) & (CPUS_PER_LONG - 1))
 308
 309 static void output_binary(void *buf, int len)
 310 {
 311         if (dump_binary) {
 312                 size_t n = fwrite(buf, len, 1, dump_fp);
 313                 if (n != 1) {
 314                         perror(dump_binary);
 315                         fclose(dump_fp);
 316                         dump_binary = NULL;
 317                 }
 318         }
 319 }
 320
 321 static void resize_cpu_info(struct per_dev_info *pdi, int cpu)
 322 {
 323         struct per_cpu_info *cpus = pdi->cpus;
 324         int ncpus = pdi->ncpus;
 325         int new_count = cpu + 1;
 326         int new_space, size;
 327         char *new_start;
 328
 329         size = new_count * sizeof(struct per_cpu_info);
 330         cpus = realloc(cpus, size);
 331         if (!cpus) {
 332                 char name[20];
 333                 fprintf(stderr, "Out of memory, CPU info for device %s (%d)\n",
 334                         get_dev_name(pdi, name, sizeof(name)), size);
 335                 exit(1);
 336         }
 337
 338         new_start = (char *)cpus + (ncpus * sizeof(struct per_cpu_info));
 339         new_space = (new_count - ncpus) * sizeof(struct per_cpu_info);
 340         memset(new_start, 0, new_space);
 341
 342         pdi->ncpus = new_count;
 343         pdi->cpus = cpus;
 344
 345         for (new_count = 0; new_count < pdi->ncpus; new_count++) {
 346                 struct per_cpu_info *pci = &pdi->cpus[new_count];
 347
 348                 if (!pci->fd) {
 349                         pci->fd = -1;
 350                         memset(&pci->rb_last, 0, sizeof(pci->rb_last));
 351                         pci->rb_last_entries = 0;
 352                         pci->last_sequence = -1;
 353                 }
 354         }
 355 }
 356
 357 static struct per_cpu_info *get_cpu_info(struct per_dev_info *pdi, int cpu)
 358 {
 359         struct per_cpu_info *pci;
 360
 361         if (cpu >= pdi->ncpus)
 362                 resize_cpu_info(pdi, cpu);
 363
 364         pci = &pdi->cpus[cpu];
 365         pci->cpu = cpu;
 366         return pci;
 367 }
 368
 369
 370 static int resize_devices(char *name)
 371 {
 372         int size = (ndevices + 1) * sizeof(struct per_dev_info);
 373
 374         devices = realloc(devices, size);
 375         if (!devices) {
 376                 fprintf(stderr, "Out of memory, device %s (%d)\n", name, size);
 377                 return 1;
 378         }
 379         memset(&devices[ndevices], 0, sizeof(struct per_dev_info));
 380         devices[ndevices].name = name;
 381         ndevices++;
 382         return 0;
 383 }
 384
 385 static struct per_dev_info *get_dev_info(dev_t dev)
 386 {
 387         struct per_dev_info *pdi;
 388         int i;
 389
 390         for (i = 0; i < ndevices; i++) {
 391                 if (!devices[i].dev)
 392                         devices[i].dev = dev;
 393                 if (devices[i].dev == dev)
 394                         return &devices[i];
 395         }
 396
 397         if (resize_devices(NULL))
 398                 return NULL;
 399
 400         pdi = &devices[ndevices - 1];
 401         pdi->dev = dev;
 402         pdi->first_reported_time = 0;
 403         pdi->last_read_time = 0;
 404
 405         return pdi;
 406 }
 407
 408 static void insert_skip(struct per_cpu_info *pci, unsigned long start,
 409                         unsigned long end)
 410 {
 411         struct skip_info *sip;
 412
 413         for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
 414                 if (end == (sip->start - 1)) {
 415                         sip->start = start;
 416                         return;
 417                 } else if (start == (sip->end + 1)) {
 418                         sip->end = end;
 419                         return;
 420                 }
 421         }
 422
 423         sip = malloc(sizeof(struct skip_info));
 424         sip->start = start;
 425         sip->end = end;
 426         sip->prev = sip->next = NULL;
 427         if (pci->skips_tail == NULL)
 428                 pci->skips_head = pci->skips_tail = sip;
 429         else {
 430                 sip->prev = pci->skips_tail;
 431                 pci->skips_tail->next = sip;
 432                 pci->skips_tail = sip;
 433         }
 434 }
 435
 436 static void remove_sip(struct per_cpu_info *pci, struct skip_info *sip)
 437 {
 438         if (sip->prev == NULL) {
 439                 if (sip->next == NULL)
 440                         pci->skips_head = pci->skips_tail = NULL;
 441                 else {
 442                         pci->skips_head = sip->next;
 443                         sip->next->prev = NULL;
 444                 }
 445         } else if (sip->next == NULL) {
 446                 pci->skips_tail = sip->prev;
 447                 sip->prev->next = NULL;
 448         } else {
 449                 sip->prev->next = sip->next;
 450                 sip->next->prev = sip->prev;
 451         }
 452
 453         sip->prev = sip->next = NULL;
 454         free(sip);
 455 }
 456
 457 #define IN_SKIP(sip,seq) (((sip)->start <= (seq)) && ((seq) <= sip->end))
 458 static int check_current_skips(struct per_cpu_info *pci, unsigned long seq)
 459 {
 460         struct skip_info *sip;
 461
 462         for (sip = pci->skips_tail; sip != NULL; sip = sip->prev) {
 463                 if (IN_SKIP(sip, seq)) {
 464                         if (sip->start == seq) {
 465                                 if (sip->end == seq)
 466                                         remove_sip(pci, sip);
 467                                 else
 468                                         sip->start += 1;
 469                         } else if (sip->end == seq)
 470                                 sip->end -= 1;
 471                         else {
 472                                 sip->end = seq - 1;
 473                                 insert_skip(pci, seq + 1, sip->end);
 474                         }
 475                         return 1;
 476                 }
 477         }
 478
 479         return 0;
 480 }
 481
 482 static void collect_pdi_skips(struct per_dev_info *pdi)
 483 {
 484         struct skip_info *sip;
 485         int cpu;
 486
 487         pdi->skips = 0;
 488         pdi->seq_skips = 0;
 489
 490         for (cpu = 0; cpu < pdi->ncpus; cpu++) {
 491                 struct per_cpu_info *pci = &pdi->cpus[cpu];
 492
 493                 for (sip = pci->skips_head; sip != NULL; sip = sip->next) {
 494                         pdi->skips++;
 495                         pdi->seq_skips += (sip->end - sip->start + 1);
 496                         if (verbose)
 497                                 fprintf(stderr,"(%d,%d): skipping %lu -> %lu\n",
 498                                         MAJOR(pdi->dev), MINOR(pdi->dev),
 499                                         sip->start, sip->end);
 500                 }
 501         }
 502 }
 503
 504 static void cpu_mark_online(struct per_dev_info *pdi, unsigned int cpu)
 505 {
 506         if (cpu >= pdi->cpu_map_max || !pdi->cpu_map) {
 507                 int new_max = (cpu + CPUS_PER_LONG) & ~(CPUS_PER_LONG - 1);
 508                 unsigned long *map = malloc(new_max / sizeof(long));
 509
 510                 memset(map, 0, new_max / sizeof(long));
 511
 512                 if (pdi->cpu_map) {
 513                         memcpy(map, pdi->cpu_map, pdi->cpu_map_max / sizeof(long));
 514                         free(pdi->cpu_map);
 515                 }
 516
 517                 pdi->cpu_map = map;
 518                 pdi->cpu_map_max = new_max;
 519         }
 520
 521         pdi->cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
 522 }
 523
 524 static inline void cpu_mark_offline(struct per_dev_info *pdi, int cpu)
 525 {
 526         pdi->cpu_map[CPU_IDX(cpu)] &= ~(1UL << CPU_BIT(cpu));
 527 }
 528
 529 static inline int cpu_is_online(struct per_dev_info *pdi, int cpu)
 530 {
 531         return (pdi->cpu_map[CPU_IDX(cpu)] & (1UL << CPU_BIT(cpu))) != 0;
 532 }
 533
 534 static inline int ppm_hash_pid(pid_t pid)
 535 {
 536         return jhash_1word(pid, JHASH_RANDOM) & PPM_HASH_MASK;
 537 }
 538
 539 static struct process_pid_map *find_ppm(pid_t pid)
 540 {
 541         const int hash_idx = ppm_hash_pid(pid);
 542         struct process_pid_map *ppm;
 543
 544         ppm = ppm_hash_table[hash_idx];
 545         while (ppm) {
 546                 if (ppm->pid == pid)
 547                         return ppm;
 548
 549                 ppm = ppm->hash_next;
 550         }
 551
 552         return NULL;
 553 }
 554
 555 static struct process_pid_map *add_ppm_hash(pid_t pid, const char *name)
 556 {
 557         const int hash_idx = ppm_hash_pid(pid);
 558         struct process_pid_map *ppm;
 559
 560         ppm = find_ppm(pid);
 561         if (!ppm) {
 562                 ppm = malloc(sizeof(*ppm));
 563                 memset(ppm, 0, sizeof(*ppm));
 564                 ppm->pid = pid;
 565                 strcpy(ppm->comm, name);
 566                 ppm->hash_next = ppm_hash_table[hash_idx];
 567                 ppm_hash_table[hash_idx] = ppm;
 568         }
 569
 570         return ppm;
 571 }
 572
 573 static void handle_notify(struct blk_io_trace *bit)
 574 {
 575         void    *payload = (caddr_t) bit + sizeof(*bit);
 576         __u32   two32[2];
 577
 578         switch (bit->action) {
 579         case BLK_TN_PROCESS:
 580                 add_ppm_hash(bit->pid, payload);
 581                 break;
 582
 583         case BLK_TN_TIMESTAMP:
 584                 if (bit->pdu_len != sizeof(two32))
 585                         return;
 586                 memcpy(two32, payload, sizeof(two32));
 587                 if (!data_is_native) {
 588                         two32[0] = be32_to_cpu(two32[0]);
 589                         two32[1] = be32_to_cpu(two32[1]);
 590                 }
 591                 start_timestamp = bit->time;
 592                 abs_start_time.tv_sec  = two32[0];
 593                 abs_start_time.tv_nsec = two32[1];
 594                 if (abs_start_time.tv_nsec < 0) {
 595                         abs_start_time.tv_sec--;
 596                         abs_start_time.tv_nsec += 1000000000;
 597                 }
 598
 599                 break;
 600
 601         case BLK_TN_MESSAGE:
 602                 if (bit->pdu_len > 0) {
 603                         char msg[bit->pdu_len+1];
 604
 605                         memcpy(msg, (char *)payload, bit->pdu_len);
 606                         msg[bit->pdu_len] = '\0';
 607
 608                         fprintf(ofp,
 609                                 "%3d,%-3d %2d %8s %5d.%09lu %5u %2s %3s %s\n",
 610                                 MAJOR(bit->device), MINOR(bit->device),
 611                                 bit->cpu, "0", (int) SECONDS(bit->time),
 612                                 (unsigned long) NANO_SECONDS(bit->time),
 613                                 0, "m", "N", msg);
 614                 }
 615                 break;
 616
 617         default:
 618                 /* Ignore unknown notify events */
 619                 ;
 620         }
 621 }
 622
 623 char *find_process_name(pid_t pid)
 624 {
 625         struct process_pid_map *ppm = find_ppm(pid);
 626
 627         if (ppm)
 628                 return ppm->comm;
 629
 630         return NULL;
 631 }
 632
 633 static inline int ppi_hash_pid(pid_t pid)
 634 {
 635         return jhash_1word(pid, JHASH_RANDOM) & PPI_HASH_MASK;
 636 }
 637
 638 static inline int ppi_hash_name(const char *name)
 639 {
 640         return jhash(name, 16, JHASH_RANDOM) & PPI_HASH_MASK;
 641 }
 642
 643 static inline int ppi_hash(struct per_process_info *ppi)
 644 {
 645         struct process_pid_map *ppm = ppi->ppm;
 646
 647         if (ppi_hash_by_pid)
 648                 return ppi_hash_pid(ppm->pid);
 649
 650         return ppi_hash_name(ppm->comm);
 651 }
 652
 653 static inline void add_ppi_to_hash(struct per_process_info *ppi)
 654 {
 655         const int hash_idx = ppi_hash(ppi);
 656
 657         ppi->hash_next = ppi_hash_table[hash_idx];
 658         ppi_hash_table[hash_idx] = ppi;
 659 }
 660
 661 static inline void add_ppi_to_list(struct per_process_info *ppi)
 662 {
 663         ppi->list_next = ppi_list;
 664         ppi_list = ppi;
 665         ppi_list_entries++;
 666 }
 667
 668 static struct per_process_info *find_ppi_by_name(char *name)
 669 {
 670         const int hash_idx = ppi_hash_name(name);
 671         struct per_process_info *ppi;
 672
 673         ppi = ppi_hash_table[hash_idx];
 674         while (ppi) {
 675                 struct process_pid_map *ppm = ppi->ppm;
 676
 677                 if (!strcmp(ppm->comm, name))
 678                         return ppi;
 679
 680                 ppi = ppi->hash_next;
 681         }
 682
 683         return NULL;
 684 }
 685
 686 static struct per_process_info *find_ppi_by_pid(pid_t pid)
 687 {
 688         const int hash_idx = ppi_hash_pid(pid);
 689         struct per_process_info *ppi;
 690
 691         ppi = ppi_hash_table[hash_idx];
 692         while (ppi) {
 693                 struct process_pid_map *ppm = ppi->ppm;
 694
 695                 if (ppm->pid == pid)
 696                         return ppi;
 697
 698                 ppi = ppi->hash_next;
 699         }
 700
 701         return NULL;
 702 }
 703
 704 static struct per_process_info *find_ppi(pid_t pid)
 705 {
 706         struct per_process_info *ppi;
 707         char *name;
 708
 709         if (ppi_hash_by_pid)
 710                 return find_ppi_by_pid(pid);
 711
 712         name = find_process_name(pid);
 713         if (!name)
 714                 return NULL;
 715
 716         ppi = find_ppi_by_name(name);
 717         if (ppi && ppi->ppm->pid != pid)
 718                 ppi->more_than_one = 1;
 719
 720         return ppi;
 721 }
 722
 723 /*
 724  * struct trace and blktrace allocation cache, we do potentially
 725  * millions of mallocs for these structures while only using at most
 726  * a few thousand at the time
 727  */
 728 static inline void t_free(struct trace *t)
 729 {
 730         if (t_alloc_cache < 1024) {
 731                 t->next = t_alloc_list;
 732                 t_alloc_list = t;
 733                 t_alloc_cache++;
 734         } else
 735                 free(t);
 736 }
 737
 738 static inline struct trace *t_alloc(void)
 739 {
 740         struct trace *t = t_alloc_list;
 741
 742         if (t) {
 743                 t_alloc_list = t->next;
 744                 t_alloc_cache--;
 745                 return t;
 746         }
 747
 748         return malloc(sizeof(*t));
 749 }
 750
 751 static inline void bit_free(struct blk_io_trace *bit)
 752 {
 753         if (bit_alloc_cache < 1024 && !bit->pdu_len) {
 754                 /*
 755                  * abuse a 64-bit field for a next pointer for the free item
 756                  */
 757                 bit->time = (__u64) (unsigned long) bit_alloc_list;
 758                 bit_alloc_list = (struct blk_io_trace *) bit;
 759                 bit_alloc_cache++;
 760         } else
 761                 free(bit);
 762 }
 763
 764 static inline struct blk_io_trace *bit_alloc(void)
 765 {
 766         struct blk_io_trace *bit = bit_alloc_list;
 767
 768         if (bit) {
 769                 bit_alloc_list = (struct blk_io_trace *) (unsigned long) \
 770                                  bit->time;
 771                 bit_alloc_cache--;
 772                 return bit;
 773         }
 774
 775         return malloc(sizeof(*bit));
 776 }
 777
 778 static inline void __put_trace_last(struct per_dev_info *pdi, struct trace *t)
 779 {
 780         struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
 781
 782         rb_erase(&t->rb_node, &pci->rb_last);
 783         pci->rb_last_entries--;
 784
 785         bit_free(t->bit);
 786         t_free(t);
 787 }
 788
 789 static void put_trace(struct per_dev_info *pdi, struct trace *t)
 790 {
 791         rb_erase(&t->rb_node, &rb_sort_root);
 792         rb_sort_entries--;
 793
 794         trace_rb_insert_last(pdi, t);
 795 }
 796
 797 static inline int trace_rb_insert(struct trace *t, struct rb_root *root)
 798 {
 799         struct rb_node **p = &root->rb_node;
 800         struct rb_node *parent = NULL;
 801         struct trace *__t;
 802
 803         while (*p) {
 804                 parent = *p;
 805
 806                 __t = rb_entry(parent, struct trace, rb_node);
 807
 808                 if (t->bit->time < __t->bit->time)
 809                         p = &(*p)->rb_left;
 810                 else if (t->bit->time > __t->bit->time)
 811                         p = &(*p)->rb_right;
 812                 else if (t->bit->device < __t->bit->device)
 813                         p = &(*p)->rb_left;
 814                 else if (t->bit->device > __t->bit->device)
 815                         p = &(*p)->rb_right;
 816                 else if (t->bit->sequence < __t->bit->sequence)
 817                         p = &(*p)->rb_left;
 818                 else    /* >= sequence */
 819                         p = &(*p)->rb_right;
 820         }
 821
 822         rb_link_node(&t->rb_node, parent, p);
 823         rb_insert_color(&t->rb_node, root);
 824         return 0;
 825 }
 826
 827 static inline int trace_rb_insert_sort(struct trace *t)
 828 {
 829         if (!trace_rb_insert(t, &rb_sort_root)) {
 830                 rb_sort_entries++;
 831                 return 0;
 832         }
 833
 834         return 1;
 835 }
 836
 837 static int trace_rb_insert_last(struct per_dev_info *pdi, struct trace *t)
 838 {
 839         struct per_cpu_info *pci = get_cpu_info(pdi, t->bit->cpu);
 840
 841         if (trace_rb_insert(t, &pci->rb_last))
 842                 return 1;
 843
 844         pci->rb_last_entries++;
 845
 846         if (pci->rb_last_entries > rb_batch * pdi->nfiles) {
 847                 struct rb_node *n = rb_first(&pci->rb_last);
 848
 849                 t = rb_entry(n, struct trace, rb_node);
 850                 __put_trace_last(pdi, t);
 851         }
 852
 853         return 0;
 854 }
 855
 856 static struct trace *trace_rb_find(dev_t device, unsigned long sequence,
 857                                    struct rb_root *root, int order)
 858 {
 859         struct rb_node *n = root->rb_node;
 860         struct rb_node *prev = NULL;
 861         struct trace *__t;
 862
 863         while (n) {
 864                 __t = rb_entry(n, struct trace, rb_node);
 865                 prev = n;
 866
 867                 if (device < __t->bit->device)
 868                         n = n->rb_left;
 869                 else if (device > __t->bit->device)
 870                         n = n->rb_right;
 871                 else if (sequence < __t->bit->sequence)
 872                         n = n->rb_left;
 873                 else if (sequence > __t->bit->sequence)
 874                         n = n->rb_right;
 875                 else
 876                         return __t;
 877         }
 878
 879         /*
 880          * hack - the list may not be sequence ordered because some
 881          * events don't have sequence and time matched. so we end up
 882          * being a little off in the rb lookup here, because we don't
 883          * know the time we are looking for. compensate by browsing
 884          * a little ahead from the last entry to find the match
 885          */
 886         if (order && prev) {
 887                 int max = 5;
 888
 889                 while (((n = rb_next(prev)) != NULL) && max--) {
 890                         __t = rb_entry(n, struct trace, rb_node);
 891
 892                         if (__t->bit->device == device &&
 893                             __t->bit->sequence == sequence)
 894                                 return __t;
 895
 896                         prev = n;
 897                 }
 898         }
 899
 900         return NULL;
 901 }
 902
 903 static inline struct trace *trace_rb_find_last(struct per_dev_info *pdi,
 904                                                struct per_cpu_info *pci,
 905                                                unsigned long seq)
 906 {
 907         return trace_rb_find(pdi->dev, seq, &pci->rb_last, 0);
 908 }
 909
 910 static inline int track_rb_insert(struct per_dev_info *pdi,struct io_track *iot)
 911 {
 912         struct rb_node **p = &pdi->rb_track.rb_node;
 913         struct rb_node *parent = NULL;
 914         struct io_track *__iot;
 915
 916         while (*p) {
 917                 parent = *p;
 918                 __iot = rb_entry(parent, struct io_track, rb_node);
 919
 920                 if (iot->sector < __iot->sector)
 921                         p = &(*p)->rb_left;
 922                 else if (iot->sector > __iot->sector)
 923                         p = &(*p)->rb_right;
 924                 else {
 925                         fprintf(stderr,
 926                                 "sector alias (%Lu) on device %d,%d!\n",
 927                                 (unsigned long long) iot->sector,
 928                                 MAJOR(pdi->dev), MINOR(pdi->dev));
 929                         return 1;
 930                 }
 931         }
 932
 933         rb_link_node(&iot->rb_node, parent, p);
 934         rb_insert_color(&iot->rb_node, &pdi->rb_track);
 935         return 0;
 936 }
 937
 938 static struct io_track *__find_track(struct per_dev_info *pdi, __u64 sector)
 939 {
 940         struct rb_node *n = pdi->rb_track.rb_node;
 941         struct io_track *__iot;
 942
 943         while (n) {
 944                 __iot = rb_entry(n, struct io_track, rb_node);
 945
 946                 if (sector < __iot->sector)
 947                         n = n->rb_left;
 948                 else if (sector > __iot->sector)
 949                         n = n->rb_right;
 950                 else
 951                         return __iot;
 952         }
 953
 954         return NULL;
 955 }
 956
 957 static struct io_track *find_track(struct per_dev_info *pdi, pid_t pid,
 958                                    __u64 sector)
 959 {
 960         struct io_track *iot;
 961
 962         iot = __find_track(pdi, sector);
 963         if (!iot) {
 964                 iot = malloc(sizeof(*iot));
 965                 iot->ppm = find_ppm(pid);
 966                 if (!iot->ppm)
 967                         iot->ppm = add_ppm_hash(pid, "unknown");
 968                 iot->sector = sector;
 969                 track_rb_insert(pdi, iot);
 970         }
 971
 972         return iot;
 973 }
 974
 975 static void log_track_frontmerge(struct per_dev_info *pdi,
 976                                  struct blk_io_trace *t)
 977 {
 978         struct io_track *iot;
 979
 980         if (!track_ios)
 981                 return;
 982
 983         iot = __find_track(pdi, t->sector + t_sec(t));
 984         if (!iot) {
 985                 if (verbose)
 986                         fprintf(stderr, "merge not found for (%d,%d): %llu\n",
 987                                 MAJOR(pdi->dev), MINOR(pdi->dev),
 988                                 (unsigned long long) t->sector + t_sec(t));
 989                 return;
 990         }
 991
 992         rb_erase(&iot->rb_node, &pdi->rb_track);
 993         iot->sector -= t_sec(t);
 994         track_rb_insert(pdi, iot);
 995 }
 996
 997 static void log_track_getrq(struct per_dev_info *pdi, struct blk_io_trace *t)
 998 {
 999         struct io_track *iot;
1000
1001         if (!track_ios)
1002                 return;
1003
1004         iot = find_track(pdi, t->pid, t->sector);
1005         iot->allocation_time = t->time;
1006 }
1007
1008 static inline int is_remapper(struct per_dev_info *pdi)
1009 {
1010         int major = MAJOR(pdi->dev);
1011
1012         return (major == 253 || major == 9);
1013 }
1014
1015 /*
1016  * for md/dm setups, the interesting cycle is Q -> C. So track queueing
1017  * time here, as dispatch time
1018  */
1019 static void log_track_queue(struct per_dev_info *pdi, struct blk_io_trace *t)
1020 {
1021         struct io_track *iot;
1022
1023         if (!track_ios)
1024                 return;
1025         if (!is_remapper(pdi))
1026                 return;
1027
1028         iot = find_track(pdi, t->pid, t->sector);
1029         iot->dispatch_time = t->time;
1030 }
1031
1032 /*
1033  * return time between rq allocation and insertion
1034  */
1035 static unsigned long long log_track_insert(struct per_dev_info *pdi,
1036                                            struct blk_io_trace *t)
1037 {
1038         unsigned long long elapsed;
1039         struct io_track *iot;
1040
1041         if (!track_ios)
1042                 return -1;
1043
1044         iot = find_track(pdi, t->pid, t->sector);
1045         iot->queue_time = t->time;
1046
1047         if (!iot->allocation_time)
1048                 return -1;
1049
1050         elapsed = iot->queue_time - iot->allocation_time;
1051
1052         if (per_process_stats) {
1053                 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1054                 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1055
1056                 if (ppi && elapsed > ppi->longest_allocation_wait[w])
1057                         ppi->longest_allocation_wait[w] = elapsed;
1058         }
1059
1060         return elapsed;
1061 }
1062
1063 /*
1064  * return time between queue and issue
1065  */
1066 static unsigned long long log_track_issue(struct per_dev_info *pdi,
1067                                           struct blk_io_trace *t)
1068 {
1069         unsigned long long elapsed;
1070         struct io_track *iot;
1071
1072         if (!track_ios)
1073                 return -1;
1074         if ((t->action & BLK_TC_ACT(BLK_TC_FS)) == 0)
1075                 return -1;
1076
1077         iot = __find_track(pdi, t->sector);
1078         if (!iot) {
1079                 if (verbose)
1080                         fprintf(stderr, "issue not found for (%d,%d): %llu\n",
1081                                 MAJOR(pdi->dev), MINOR(pdi->dev),
1082                                 (unsigned long long) t->sector);
1083                 return -1;
1084         }
1085
1086         iot->dispatch_time = t->time;
1087         elapsed = iot->dispatch_time - iot->queue_time;
1088
1089         if (per_process_stats) {
1090                 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1091                 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1092
1093                 if (ppi && elapsed > ppi->longest_dispatch_wait[w])
1094                         ppi->longest_dispatch_wait[w] = elapsed;
1095         }
1096
1097         return elapsed;
1098 }
1099
1100 /*
1101  * return time between dispatch and complete
1102  */
1103 static unsigned long long log_track_complete(struct per_dev_info *pdi,
1104                                              struct blk_io_trace *t)
1105 {
1106         unsigned long long elapsed;
1107         struct io_track *iot;
1108
1109         if (!track_ios)
1110                 return -1;
1111
1112         iot = __find_track(pdi, t->sector);
1113         if (!iot) {
1114                 if (verbose)
1115                         fprintf(stderr,"complete not found for (%d,%d): %llu\n",
1116                                 MAJOR(pdi->dev), MINOR(pdi->dev),
1117                                 (unsigned long long) t->sector);
1118                 return -1;
1119         }
1120
1121         iot->completion_time = t->time;
1122         elapsed = iot->completion_time - iot->dispatch_time;
1123
1124         if (per_process_stats) {
1125                 struct per_process_info *ppi = find_ppi(iot->ppm->pid);
1126                 int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1127
1128                 if (ppi && elapsed > ppi->longest_completion_wait[w])
1129                         ppi->longest_completion_wait[w] = elapsed;
1130         }
1131
1132         /*
1133          * kill the trace, we don't need it after completion
1134          */
1135         rb_erase(&iot->rb_node, &pdi->rb_track);
1136         free(iot);
1137
1138         return elapsed;
1139 }
1140
1141
1142 static struct io_stats *find_process_io_stats(pid_t pid)
1143 {
1144         struct per_process_info *ppi = find_ppi(pid);
1145
1146         if (!ppi) {
1147                 ppi = malloc(sizeof(*ppi));
1148                 memset(ppi, 0, sizeof(*ppi));
1149                 ppi->ppm = find_ppm(pid);
1150                 if (!ppi->ppm)
1151                         ppi->ppm = add_ppm_hash(pid, "unknown");
1152                 add_ppi_to_hash(ppi);
1153                 add_ppi_to_list(ppi);
1154         }
1155
1156         return &ppi->io_stats;
1157 }
1158
1159 static char *get_dev_name(struct per_dev_info *pdi, char *buffer, int size)
1160 {
1161         if (pdi->name)
1162                 snprintf(buffer, size, "%s", pdi->name);
1163         else
1164                 snprintf(buffer, size, "%d,%d",MAJOR(pdi->dev),MINOR(pdi->dev));
1165         return buffer;
1166 }
1167
1168 static void check_time(struct per_dev_info *pdi, struct blk_io_trace *bit)
1169 {
1170         unsigned long long this = bit->time;
1171         unsigned long long last = pdi->last_reported_time;
1172
1173         pdi->backwards = (this < last) ? 'B' : ' ';
1174         pdi->last_reported_time = this;
1175 }
1176
1177 static inline void __account_m(struct io_stats *ios, struct blk_io_trace *t,
1178                                int rw)
1179 {
1180         if (rw) {
1181                 ios->mwrites++;
1182                 ios->mwrite_kb += t_kb(t);
1183         } else {
1184                 ios->mreads++;
1185                 ios->mread_kb += t_kb(t);
1186         }
1187 }
1188
1189 static inline void account_m(struct blk_io_trace *t, struct per_cpu_info *pci,
1190                              int rw)
1191 {
1192         __account_m(&pci->io_stats, t, rw);
1193
1194         if (per_process_stats) {
1195                 struct io_stats *ios = find_process_io_stats(t->pid);
1196
1197                 __account_m(ios, t, rw);
1198         }
1199 }
1200
1201 static inline void __account_pc_queue(struct io_stats *ios,
1202                                       struct blk_io_trace *t, int rw)
1203 {
1204         if (rw) {
1205                 ios->qwrites_pc++;
1206                 ios->qwrite_kb_pc += t_kb(t);
1207         } else {
1208                 ios->qreads_pc++;
1209                 ios->qread_kb += t_kb(t);
1210         }
1211 }
1212
1213 static inline void account_pc_queue(struct blk_io_trace *t,
1214                                     struct per_cpu_info *pci, int rw)
1215 {
1216         __account_pc_queue(&pci->io_stats, t, rw);
1217
1218         if (per_process_stats) {
1219                 struct io_stats *ios = find_process_io_stats(t->pid);
1220
1221                 __account_pc_queue(ios, t, rw);
1222         }
1223 }
1224
1225 static inline void __account_pc_issue(struct io_stats *ios, int rw,
1226                                       unsigned int bytes)
1227 {
1228         if (rw) {
1229                 ios->iwrites_pc++;
1230                 ios->iwrite_kb_pc += bytes >> 10;
1231         } else {
1232                 ios->ireads_pc++;
1233                 ios->iread_kb_pc += bytes >> 10;
1234         }
1235 }
1236
1237 static inline void account_pc_issue(struct blk_io_trace *t,
1238                                     struct per_cpu_info *pci, int rw)
1239 {
1240         __account_pc_issue(&pci->io_stats, rw, t->bytes);
1241
1242         if (per_process_stats) {
1243                 struct io_stats *ios = find_process_io_stats(t->pid);
1244
1245                 __account_pc_issue(ios, rw, t->bytes);
1246         }
1247 }
1248
1249 static inline void __account_pc_requeue(struct io_stats *ios,
1250                                         struct blk_io_trace *t, int rw)
1251 {
1252         if (rw) {
1253                 ios->wrqueue_pc++;
1254                 ios->iwrite_kb_pc -= t_kb(t);
1255         } else {
1256                 ios->rrqueue_pc++;
1257                 ios->iread_kb_pc -= t_kb(t);
1258         }
1259 }
1260
1261 static inline void account_pc_requeue(struct blk_io_trace *t,
1262                                       struct per_cpu_info *pci, int rw)
1263 {
1264         __account_pc_requeue(&pci->io_stats, t, rw);
1265
1266         if (per_process_stats) {
1267                 struct io_stats *ios = find_process_io_stats(t->pid);
1268
1269                 __account_pc_requeue(ios, t, rw);
1270         }
1271 }
1272
1273 static inline void __account_pc_c(struct io_stats *ios, int rw)
1274 {
1275         if (rw)
1276                 ios->cwrites_pc++;
1277         else
1278                 ios->creads_pc++;
1279 }
1280
1281 static inline void account_pc_c(struct blk_io_trace *t,
1282                                 struct per_cpu_info *pci, int rw)
1283 {
1284         __account_pc_c(&pci->io_stats, rw);
1285
1286         if (per_process_stats) {
1287                 struct io_stats *ios = find_process_io_stats(t->pid);
1288
1289                 __account_pc_c(ios, rw);
1290         }
1291 }
1292
1293 static inline void __account_queue(struct io_stats *ios, struct blk_io_trace *t,
1294                                    int rw)
1295 {
1296         if (rw) {
1297                 ios->qwrites++;
1298                 ios->qwrite_kb += t_kb(t);
1299         } else {
1300                 ios->qreads++;
1301                 ios->qread_kb += t_kb(t);
1302         }
1303 }
1304
1305 static inline void account_queue(struct blk_io_trace *t,
1306                                  struct per_cpu_info *pci, int rw)
1307 {
1308         __account_queue(&pci->io_stats, t, rw);
1309
1310         if (per_process_stats) {
1311                 struct io_stats *ios = find_process_io_stats(t->pid);
1312
1313                 __account_queue(ios, t, rw);
1314         }
1315 }
1316
1317 static inline void __account_c(struct io_stats *ios, int rw, int bytes)
1318 {
1319         if (rw) {
1320                 ios->cwrites++;
1321                 ios->cwrite_kb += bytes >> 10;
1322         } else {
1323                 ios->creads++;
1324                 ios->cread_kb += bytes >> 10;
1325         }
1326 }
1327
1328 static inline void account_c(struct blk_io_trace *t, struct per_cpu_info *pci,
1329                              int rw, int bytes)
1330 {
1331         __account_c(&pci->io_stats, rw, bytes);
1332
1333         if (per_process_stats) {
1334                 struct io_stats *ios = find_process_io_stats(t->pid);
1335
1336                 __account_c(ios, rw, bytes);
1337         }
1338 }
1339
1340 static inline void __account_issue(struct io_stats *ios, int rw,
1341                                    unsigned int bytes)
1342 {
1343         if (rw) {
1344                 ios->iwrites++;
1345                 ios->iwrite_kb += bytes >> 10;
1346         } else {
1347                 ios->ireads++;
1348                 ios->iread_kb += bytes >> 10;
1349         }
1350 }
1351
1352 static inline void account_issue(struct blk_io_trace *t,
1353                                  struct per_cpu_info *pci, int rw)
1354 {
1355         __account_issue(&pci->io_stats, rw, t->bytes);
1356
1357         if (per_process_stats) {
1358                 struct io_stats *ios = find_process_io_stats(t->pid);
1359
1360                 __account_issue(ios, rw, t->bytes);
1361         }
1362 }
1363
1364 static inline void __account_unplug(struct io_stats *ios, int timer)
1365 {
1366         if (timer)
1367                 ios->timer_unplugs++;
1368         else
1369                 ios->io_unplugs++;
1370 }
1371
1372 static inline void account_unplug(struct blk_io_trace *t,
1373                                   struct per_cpu_info *pci, int timer)
1374 {
1375         __account_unplug(&pci->io_stats, timer);
1376
1377         if (per_process_stats) {
1378                 struct io_stats *ios = find_process_io_stats(t->pid);
1379
1380                 __account_unplug(ios, timer);
1381         }
1382 }
1383
1384 static inline void __account_requeue(struct io_stats *ios,
1385                                      struct blk_io_trace *t, int rw)
1386 {
1387         if (rw) {
1388                 ios->wrqueue++;
1389                 ios->iwrite_kb -= t_kb(t);
1390         } else {
1391                 ios->rrqueue++;
1392                 ios->iread_kb -= t_kb(t);
1393         }
1394 }
1395
1396 static inline void account_requeue(struct blk_io_trace *t,
1397                                    struct per_cpu_info *pci, int rw)
1398 {
1399         __account_requeue(&pci->io_stats, t, rw);
1400
1401         if (per_process_stats) {
1402                 struct io_stats *ios = find_process_io_stats(t->pid);
1403
1404                 __account_requeue(ios, t, rw);
1405         }
1406 }
1407
1408 static void log_complete(struct per_dev_info *pdi, struct per_cpu_info *pci,
1409                          struct blk_io_trace *t, char *act)
1410 {
1411         process_fmt(act, pci, t, log_track_complete(pdi, t), 0, NULL);
1412 }
1413
1414 static void log_insert(struct per_dev_info *pdi, struct per_cpu_info *pci,
1415                        struct blk_io_trace *t, char *act)
1416 {
1417         process_fmt(act, pci, t, log_track_insert(pdi, t), 0, NULL);
1418 }
1419
1420 static void log_queue(struct per_cpu_info *pci, struct blk_io_trace *t,
1421                       char *act)
1422 {
1423         process_fmt(act, pci, t, -1, 0, NULL);
1424 }
1425
1426 static void log_issue(struct per_dev_info *pdi, struct per_cpu_info *pci,
1427                       struct blk_io_trace *t, char *act)
1428 {
1429         process_fmt(act, pci, t, log_track_issue(pdi, t), 0, NULL);
1430 }
1431
1432 static void log_merge(struct per_dev_info *pdi, struct per_cpu_info *pci,
1433                       struct blk_io_trace *t, char *act)
1434 {
1435         if (act[0] == 'F')
1436                 log_track_frontmerge(pdi, t);
1437
1438         process_fmt(act, pci, t, -1ULL, 0, NULL);
1439 }
1440
1441 static void log_action(struct per_cpu_info *pci, struct blk_io_trace *t,
1442                         char *act)
1443 {
1444         process_fmt(act, pci, t, -1ULL, 0, NULL);
1445 }
1446
1447 static void log_generic(struct per_cpu_info *pci, struct blk_io_trace *t,
1448                         char *act)
1449 {
1450         process_fmt(act, pci, t, -1ULL, 0, NULL);
1451 }
1452
1453 static void log_unplug(struct per_cpu_info *pci, struct blk_io_trace *t,
1454                       char *act)
1455 {
1456         process_fmt(act, pci, t, -1ULL, 0, NULL);
1457 }
1458
1459 static void log_split(struct per_cpu_info *pci, struct blk_io_trace *t,
1460                       char *act)
1461 {
1462         process_fmt(act, pci, t, -1ULL, 0, NULL);
1463 }
1464
1465 static void log_pc(struct per_cpu_info *pci, struct blk_io_trace *t, char *act)
1466 {
1467         unsigned char *buf = (unsigned char *) t + sizeof(*t);
1468
1469         process_fmt(act, pci, t, -1ULL, t->pdu_len, buf);
1470 }
1471
1472 static void dump_trace_pc(struct blk_io_trace *t, struct per_dev_info *pdi,
1473                           struct per_cpu_info *pci)
1474 {
1475         int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1476         int act = t->action & 0xffff;
1477
1478         switch (act) {
1479                 case __BLK_TA_QUEUE:
1480                         log_generic(pci, t, "Q");
1481                         account_pc_queue(t, pci, w);
1482                         break;
1483                 case __BLK_TA_GETRQ:
1484                         log_generic(pci, t, "G");
1485                         break;
1486                 case __BLK_TA_SLEEPRQ:
1487                         log_generic(pci, t, "S");
1488                         break;
1489                 case __BLK_TA_REQUEUE:
1490                         /*
1491                          * can happen if we miss traces, don't let it go
1492                          * below zero
1493                          */
1494                         if (pdi->cur_depth[w])
1495                                 pdi->cur_depth[w]--;
1496                         account_pc_requeue(t, pci, w);
1497                         log_generic(pci, t, "R");
1498                         break;
1499                 case __BLK_TA_ISSUE:
1500                         account_pc_issue(t, pci, w);
1501                         pdi->cur_depth[w]++;
1502                         if (pdi->cur_depth[w] > pdi->max_depth[w])
1503                                 pdi->max_depth[w] = pdi->cur_depth[w];
1504                         log_pc(pci, t, "D");
1505                         break;
1506                 case __BLK_TA_COMPLETE:
1507                         if (pdi->cur_depth[w])
1508                                 pdi->cur_depth[w]--;
1509                         log_pc(pci, t, "C");
1510                         account_pc_c(t, pci, w);
1511                         break;
1512                 case __BLK_TA_INSERT:
1513                         log_pc(pci, t, "I");
1514                         break;
1515                 default:
1516                         fprintf(stderr, "Bad pc action %x\n", act);
1517                         break;
1518         }
1519 }
1520
1521 static void dump_trace_fs(struct blk_io_trace *t, struct per_dev_info *pdi,
1522                           struct per_cpu_info *pci)
1523 {
1524         int w = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
1525         int act = t->action & 0xffff;
1526
1527         switch (act) {
1528                 case __BLK_TA_QUEUE:
1529                         log_track_queue(pdi, t);
1530                         account_queue(t, pci, w);
1531                         log_queue(pci, t, "Q");
1532                         break;
1533                 case __BLK_TA_INSERT:
1534                         log_insert(pdi, pci, t, "I");
1535                         break;
1536                 case __BLK_TA_BACKMERGE:
1537                         account_m(t, pci, w);
1538                         log_merge(pdi, pci, t, "M");
1539                         break;
1540                 case __BLK_TA_FRONTMERGE:
1541                         account_m(t, pci, w);
1542                         log_merge(pdi, pci, t, "F");
1543                         break;
1544                 case __BLK_TA_GETRQ:
1545                         log_track_getrq(pdi, t);
1546                         log_generic(pci, t, "G");
1547                         break;
1548                 case __BLK_TA_SLEEPRQ:
1549                         log_generic(pci, t, "S");
1550                         break;
1551                 case __BLK_TA_REQUEUE:
1552                         /*
1553                          * can happen if we miss traces, don't let it go
1554                          * below zero
1555                          */
1556                         if (pdi->cur_depth[w])
1557                                 pdi->cur_depth[w]--;
1558                         account_requeue(t, pci, w);
1559                         log_queue(pci, t, "R");
1560                         break;
1561                 case __BLK_TA_ISSUE:
1562                         account_issue(t, pci, w);
1563                         pdi->cur_depth[w]++;
1564                         if (pdi->cur_depth[w] > pdi->max_depth[w])
1565                                 pdi->max_depth[w] = pdi->cur_depth[w];
1566                         log_issue(pdi, pci, t, "D");
1567                         break;
1568                 case __BLK_TA_COMPLETE:
1569                         if (pdi->cur_depth[w])
1570                                 pdi->cur_depth[w]--;
1571                         account_c(t, pci, w, t->bytes);
1572                         log_complete(pdi, pci, t, "C");
1573                         break;
1574                 case __BLK_TA_PLUG:
1575                         log_action(pci, t, "P");
1576                         break;
1577                 case __BLK_TA_UNPLUG_IO:
1578                         account_unplug(t, pci, 0);
1579                         log_unplug(pci, t, "U");
1580                         break;
1581                 case __BLK_TA_UNPLUG_TIMER:
1582                         account_unplug(t, pci, 1);
1583                         log_unplug(pci, t, "UT");
1584                         break;
1585                 case __BLK_TA_SPLIT:
1586                         log_split(pci, t, "X");
1587                         break;
1588                 case __BLK_TA_BOUNCE:
1589                         log_generic(pci, t, "B");
1590                         break;
1591                 case __BLK_TA_REMAP:
1592                         log_generic(pci, t, "A");
1593                         break;
1594                 case __BLK_TA_DRV_DATA:
1595                         have_drv_data = 1;
1596                         /* dump to binary file only */
1597                         break;
1598                 default:
1599                         fprintf(stderr, "Bad fs action %x\n", t->action);
1600                         break;
1601         }
1602 }
1603
1604 static void dump_trace(struct blk_io_trace *t, struct per_cpu_info *pci,
1605                        struct per_dev_info *pdi)
1606 {
1607         if (text_output) {
1608                 if (t->action == BLK_TN_MESSAGE)
1609                         handle_notify(t);
1610                 else if (t->action & BLK_TC_ACT(BLK_TC_PC))
1611                         dump_trace_pc(t, pdi, pci);
1612                 else
1613                         dump_trace_fs(t, pdi, pci);
1614         }
1615
1616         if (!pdi->events)
1617                 pdi->first_reported_time = t->time;
1618
1619         pdi->events++;
1620
1621         if (bin_output_msgs ||
1622                             !(t->action & BLK_TC_ACT(BLK_TC_NOTIFY) &&
1623                               t->action == BLK_TN_MESSAGE))
1624                 output_binary(t, sizeof(*t) + t->pdu_len);
1625 }
1626
1627 /*
1628  * print in a proper way, not too small and not too big. if more than
1629  * 1000,000K, turn into M and so on
1630  */
1631 static char *size_cnv(char *dst, unsigned long long num, int in_kb)
1632 {
1633         char suff[] = { '\0', 'K', 'M', 'G', 'P' };
1634         unsigned int i = 0;
1635
1636         if (in_kb)
1637                 i++;
1638
1639         while (num > 1000 * 1000ULL && (i < sizeof(suff) - 1)) {
1640                 i++;
1641                 num /= 1000;
1642         }
1643
1644         sprintf(dst, "%'8Lu%c", num, suff[i]);
1645         return dst;
1646 }
1647
1648 static void dump_io_stats(struct per_dev_info *pdi, struct io_stats *ios,
1649                           char *msg)
1650 {
1651         static char x[256], y[256];
1652
1653         fprintf(ofp, "%s\n", msg);
1654
1655         fprintf(ofp, " Reads Queued:    %s, %siB\t", size_cnv(x, ios->qreads, 0), size_cnv(y, ios->qread_kb, 1));
1656         fprintf(ofp, " Writes Queued:    %s, %siB\n", size_cnv(x, ios->qwrites, 0), size_cnv(y, ios->qwrite_kb, 1));
1657         fprintf(ofp, " Read Dispatches: %s, %siB\t", size_cnv(x, ios->ireads, 0), size_cnv(y, ios->iread_kb, 1));
1658         fprintf(ofp, " Write Dispatches: %s, %siB\n", size_cnv(x, ios->iwrites, 0), size_cnv(y, ios->iwrite_kb, 1));
1659         fprintf(ofp, " Reads Requeued:  %s\t\t", size_cnv(x, ios->rrqueue, 0));
1660         fprintf(ofp, " Writes Requeued:  %s\n", size_cnv(x, ios->wrqueue, 0));
1661         fprintf(ofp, " Reads Completed: %s, %siB\t", size_cnv(x, ios->creads, 0), size_cnv(y, ios->cread_kb, 1));
1662         fprintf(ofp, " Writes Completed: %s, %siB\n", size_cnv(x, ios->cwrites, 0), size_cnv(y, ios->cwrite_kb, 1));
1663         fprintf(ofp, " Read Merges:     %s, %siB\t", size_cnv(x, ios->mreads, 0), size_cnv(y, ios->mread_kb, 1));
1664         fprintf(ofp, " Write Merges:     %s, %siB\n", size_cnv(x, ios->mwrites, 0), size_cnv(y, ios->mwrite_kb, 1));
1665         if (pdi) {
1666                 fprintf(ofp, " Read depth:      %'8u%8c\t", pdi->max_depth[0], ' ');
1667                 fprintf(ofp, " Write depth:      %'8u\n", pdi->max_depth[1]);
1668         }
1669         if (ios->qreads_pc || ios->qwrites_pc || ios->ireads_pc || ios->iwrites_pc ||
1670             ios->rrqueue_pc || ios->wrqueue_pc || ios->creads_pc || ios->cwrites_pc) {
1671                 fprintf(ofp, " PC Reads Queued: %s, %siB\t", size_cnv(x, ios->qreads_pc, 0), size_cnv(y, ios->qread_kb_pc, 1));
1672                 fprintf(ofp, " PC Writes Queued: %s, %siB\n", size_cnv(x, ios->qwrites_pc, 0), size_cnv(y, ios->qwrite_kb_pc, 1));
1673                 fprintf(ofp, " PC Read Disp.:   %s, %siB\t", size_cnv(x, ios->ireads_pc, 0), size_cnv(y, ios->iread_kb_pc, 1));
1674                 fprintf(ofp, " PC Write Disp.:   %s, %siB\n", size_cnv(x, ios->iwrites_pc, 0), size_cnv(y, ios->iwrite_kb_pc, 1));
1675                 fprintf(ofp, " PC Reads Req.:   %s\t\t", size_cnv(x, ios->rrqueue_pc, 0));
1676                 fprintf(ofp, " PC Writes Req.:   %s\n", size_cnv(x, ios->wrqueue_pc, 0));
1677                 fprintf(ofp, " PC Reads Compl.: %s\t\t", size_cnv(x, ios->creads_pc, 0));
1678                 fprintf(ofp, " PC Writes Compl.: %s\n", size_cnv(x, ios->cwrites, 0));
1679         }
1680         fprintf(ofp, " IO unplugs:      %'8lu%8c\t", ios->io_unplugs, ' ');
1681         fprintf(ofp, " Timer unplugs:    %'8lu\n", ios->timer_unplugs);
1682 }
1683
1684 static void dump_wait_stats(struct per_process_info *ppi)
1685 {
1686         unsigned long rawait = ppi->longest_allocation_wait[0] / 1000;
1687         unsigned long rdwait = ppi->longest_dispatch_wait[0] / 1000;
1688         unsigned long rcwait = ppi->longest_completion_wait[0] / 1000;
1689         unsigned long wawait = ppi->longest_allocation_wait[1] / 1000;
1690         unsigned long wdwait = ppi->longest_dispatch_wait[1] / 1000;
1691         unsigned long wcwait = ppi->longest_completion_wait[1] / 1000;
1692
1693         fprintf(ofp, " Allocation wait: %'8lu%8c\t", rawait, ' ');
1694         fprintf(ofp, " Allocation wait:  %'8lu\n", wawait);
1695         fprintf(ofp, " Dispatch wait:   %'8lu%8c\t", rdwait, ' ');
1696         fprintf(ofp, " Dispatch wait:    %'8lu\n", wdwait);
1697         fprintf(ofp, " Completion wait: %'8lu%8c\t", rcwait, ' ');
1698         fprintf(ofp, " Completion wait:  %'8lu\n", wcwait);
1699 }
1700
1701 static int ppi_name_compare(const void *p1, const void *p2)
1702 {
1703         struct per_process_info *ppi1 = *((struct per_process_info **) p1);
1704         struct per_process_info *ppi2 = *((struct per_process_info **) p2);
1705         int res;
1706
1707         res = strverscmp(ppi1->ppm->comm, ppi2->ppm->comm);
1708         if (!res)
1709                 res = ppi1->ppm->pid > ppi2->ppm->pid;
1710
1711         return res;
1712 }
1713
1714 static void sort_process_list(void)
1715 {
1716         struct per_process_info **ppis;
1717         struct per_process_info *ppi;
1718         int i = 0;
1719
1720         ppis = malloc(ppi_list_entries * sizeof(struct per_process_info *));
1721
1722         ppi = ppi_list;
1723         while (ppi) {
1724                 ppis[i++] = ppi;
1725                 ppi = ppi->list_next;
1726         }
1727
1728         qsort(ppis, ppi_list_entries, sizeof(ppi), ppi_name_compare);
1729
1730         i = ppi_list_entries - 1;
1731         ppi_list = NULL;
1732         while (i >= 0) {
1733                 ppi = ppis[i];
1734
1735                 ppi->list_next = ppi_list;
1736                 ppi_list = ppi;
1737                 i--;
1738         }
1739
1740         free(ppis);
1741 }
1742
1743 static void show_process_stats(void)
1744 {
1745         struct per_process_info *ppi;
1746
1747         sort_process_list();
1748
1749         ppi = ppi_list;
1750         while (ppi) {
1751                 struct process_pid_map *ppm = ppi->ppm;
1752                 char name[64];
1753
1754                 if (ppi->more_than_one)
1755                         sprintf(name, "%s (%u, ...)", ppm->comm, ppm->pid);
1756                 else
1757                         sprintf(name, "%s (%u)", ppm->comm, ppm->pid);
1758
1759                 dump_io_stats(NULL, &ppi->io_stats, name);
1760                 dump_wait_stats(ppi);
1761                 ppi = ppi->list_next;
1762         }
1763
1764         fprintf(ofp, "\n");
1765 }
1766
1767 static void show_device_and_cpu_stats(void)
1768 {
1769         struct per_dev_info *pdi;
1770         struct per_cpu_info *pci;
1771         struct io_stats total, *ios;
1772         unsigned long long rrate, wrate, msec;
1773         int i, j, pci_events;
1774         char line[3 + 8/*cpu*/ + 2 + 32/*dev*/ + 3];
1775         char name[32];
1776         double ratio;
1777
1778         for (pdi = devices, i = 0; i < ndevices; i++, pdi++) {
1779
1780                 memset(&total, 0, sizeof(total));
1781                 pci_events = 0;
1782
1783                 if (i > 0)
1784                         fprintf(ofp, "\n");
1785
1786                 for (pci = pdi->cpus, j = 0; j < pdi->ncpus; j++, pci++) {
1787                         if (!pci->nelems)
1788                                 continue;
1789
1790                         ios = &pci->io_stats;
1791                         total.qreads += ios->qreads;
1792                         total.qwrites += ios->qwrites;
1793                         total.creads += ios->creads;
1794                         total.cwrites += ios->cwrites;
1795                         total.mreads += ios->mreads;
1796                         total.mwrites += ios->mwrites;
1797                         total.ireads += ios->ireads;
1798                         total.iwrites += ios->iwrites;
1799                         total.rrqueue += ios->rrqueue;
1800                         total.wrqueue += ios->wrqueue;
1801                         total.qread_kb += ios->qread_kb;
1802                         total.qwrite_kb += ios->qwrite_kb;
1803                         total.cread_kb += ios->cread_kb;
1804                         total.cwrite_kb += ios->cwrite_kb;
1805                         total.iread_kb += ios->iread_kb;
1806                         total.iwrite_kb += ios->iwrite_kb;
1807                         total.mread_kb += ios->mread_kb;
1808                         total.mwrite_kb += ios->mwrite_kb;
1809
1810                         total.qreads_pc += ios->qreads_pc;
1811                         total.qwrites_pc += ios->qwrites_pc;
1812                         total.creads_pc += ios->creads_pc;
1813                         total.cwrites_pc += ios->cwrites_pc;
1814                         total.ireads_pc += ios->ireads_pc;
1815                         total.iwrites_pc += ios->iwrites_pc;
1816                         total.rrqueue_pc += ios->rrqueue_pc;
1817                         total.wrqueue_pc += ios->wrqueue_pc;
1818                         total.qread_kb_pc += ios->qread_kb_pc;
1819                         total.qwrite_kb_pc += ios->qwrite_kb_pc;
1820                         total.iread_kb_pc += ios->iread_kb_pc;
1821                         total.iwrite_kb_pc += ios->iwrite_kb_pc;
1822
1823                         total.timer_unplugs += ios->timer_unplugs;
1824                         total.io_unplugs += ios->io_unplugs;
1825
1826                         snprintf(line, sizeof(line) - 1, "CPU%d (%s):",
1827                                  j, get_dev_name(pdi, name, sizeof(name)));
1828                         dump_io_stats(pdi, ios, line);
1829                         pci_events++;
1830                 }
1831
1832                 if (pci_events > 1) {
1833                         fprintf(ofp, "\n");
1834                         snprintf(line, sizeof(line) - 1, "Total (%s):",
1835                                  get_dev_name(pdi, name, sizeof(name)));
1836                         dump_io_stats(NULL, &total, line);
1837                 }
1838
1839                 wrate = rrate = 0;
1840                 msec = (pdi->last_reported_time - pdi->first_reported_time) / 1000000;
1841                 if (msec) {
1842                         rrate = 1000 * total.cread_kb / msec;
1843                         wrate = 1000 * total.cwrite_kb / msec;
1844                 }
1845
1846                 fprintf(ofp, "\nThroughput (R/W): %'LuKiB/s / %'LuKiB/s\n",
1847                         rrate, wrate);
1848                 fprintf(ofp, "Events (%s): %'Lu entries\n",
1849                         get_dev_name(pdi, line, sizeof(line)), pdi->events);
1850
1851                 collect_pdi_skips(pdi);
1852                 if (!pdi->skips && !pdi->events)
1853                         ratio = 0.0;
1854                 else
1855                         ratio = 100.0 * ((double)pdi->seq_skips /
1856                                         (double)(pdi->events + pdi->seq_skips));
1857                 fprintf(ofp, "Skips: %'lu forward (%'llu - %5.1lf%%)\n",
1858                         pdi->skips, pdi->seq_skips, ratio);
1859         }
1860 }
1861
1862 static void find_genesis(void)
1863 {
1864         struct trace *t = trace_list;
1865
1866         genesis_time = -1ULL;
1867         while (t != NULL) {
1868                 if (t->bit->time < genesis_time)
1869                         genesis_time = t->bit->time;
1870
1871                 t = t->next;
1872         }
1873
1874         /* The time stamp record will usually be the first
1875          * record in the trace, but not always.
1876          */
1877         if (start_timestamp
1878          && start_timestamp != genesis_time) {
1879                 long delta = genesis_time - start_timestamp;
1880
1881                 abs_start_time.tv_sec  += SECONDS(delta);
1882                 abs_start_time.tv_nsec += NANO_SECONDS(delta);
1883                 if (abs_start_time.tv_nsec < 0) {
1884                         abs_start_time.tv_nsec += 1000000000;
1885                         abs_start_time.tv_sec -= 1;
1886                 } else
1887                 if (abs_start_time.tv_nsec > 1000000000) {
1888                         abs_start_time.tv_nsec -= 1000000000;
1889                         abs_start_time.tv_sec += 1;
1890                 }
1891         }
1892 }
1893
1894 static inline int check_stopwatch(struct blk_io_trace *bit)
1895 {
1896         if (bit->time < stopwatch_end &&
1897             bit->time >= stopwatch_start)
1898                 return 0;
1899
1900         return 1;
1901 }
1902
1903 /*
1904  * return youngest entry read
1905  */
1906 static int sort_entries(unsigned long long *youngest)
1907 {
1908         struct per_dev_info *pdi = NULL;
1909         struct per_cpu_info *pci = NULL;
1910         struct trace *t;
1911
1912         if (!genesis_time)
1913                 find_genesis();
1914
1915         *youngest = 0;
1916         while ((t = trace_list) != NULL) {
1917                 struct blk_io_trace *bit = t->bit;
1918
1919                 trace_list = t->next;
1920
1921                 bit->time -= genesis_time;
1922
1923                 if (bit->time < *youngest || !*youngest)
1924                         *youngest = bit->time;
1925
1926                 if (!pdi || pdi->dev != bit->device) {
1927                         pdi = get_dev_info(bit->device);
1928                         pci = NULL;
1929                 }
1930
1931                 if (!pci || pci->cpu != bit->cpu)
1932                         pci = get_cpu_info(pdi, bit->cpu);
1933
1934                 if (bit->sequence < pci->smallest_seq_read)
1935                         pci->smallest_seq_read = bit->sequence;
1936
1937                 if (check_stopwatch(bit)) {
1938                         bit_free(bit);
1939                         t_free(t);
1940                         continue;
1941                 }
1942
1943                 if (trace_rb_insert_sort(t))
1944                         return -1;
1945         }
1946
1947         return 0;
1948 }
1949
1950 /*
1951  * to continue, we must have traces from all online cpus in the tree
1952  */
1953 static int check_cpu_map(struct per_dev_info *pdi)
1954 {
1955         unsigned long *cpu_map;
1956         struct rb_node *n;
1957         struct trace *__t;
1958         unsigned int i;
1959         int ret, cpu;
1960
1961         /*
1962          * create a map of the cpus we have traces for
1963          */
1964         cpu_map = malloc(pdi->cpu_map_max / sizeof(long));
1965         n = rb_first(&rb_sort_root);
1966         while (n) {
1967                 __t = rb_entry(n, struct trace, rb_node);
1968                 cpu = __t->bit->cpu;
1969
1970                 cpu_map[CPU_IDX(cpu)] |= (1UL << CPU_BIT(cpu));
1971                 n = rb_next(n);
1972         }
1973
1974         /*
1975          * we can't continue if pdi->cpu_map has entries set that we don't
1976          * have in the sort rbtree. the opposite is not a problem, though
1977          */
1978         ret = 0;
1979         for (i = 0; i < pdi->cpu_map_max / CPUS_PER_LONG; i++) {
1980                 if (pdi->cpu_map[i] & ~(cpu_map[i])) {
1981                         ret = 1;
1982                         break;
1983                 }
1984         }
1985
1986         free(cpu_map);
1987         return ret;
1988 }
1989
1990 static int check_sequence(struct per_dev_info *pdi, struct trace *t, int force)
1991 {
1992         struct blk_io_trace *bit = t->bit;
1993         unsigned long expected_sequence;
1994         struct per_cpu_info *pci;
1995         struct trace *__t;
1996
1997         pci = get_cpu_info(pdi, bit->cpu);
1998         expected_sequence = pci->last_sequence + 1;
1999
2000         if (!expected_sequence) {
2001                 /*
2002                  * 1 should be the first entry, just allow it
2003                  */
2004                 if (bit->sequence == 1)
2005                         return 0;
2006                 if (bit->sequence == pci->smallest_seq_read)
2007                         return 0;
2008
2009                 return check_cpu_map(pdi);
2010         }
2011
2012         if (bit->sequence == expected_sequence)
2013                 return 0;
2014
2015         /*
2016          * we may not have seen that sequence yet. if we are not doing
2017          * the final run, break and wait for more entries.
2018          */
2019         if (expected_sequence < pci->smallest_seq_read) {
2020                 __t = trace_rb_find_last(pdi, pci, expected_sequence);
2021                 if (!__t)
2022                         goto skip;
2023
2024                 __put_trace_last(pdi, __t);
2025                 return 0;
2026         } else if (!force) {
2027                 return 1;
2028         } else {
2029 skip:
2030                 if (check_current_skips(pci, bit->sequence))
2031                         return 0;
2032
2033                 if (expected_sequence < bit->sequence)
2034                         insert_skip(pci, expected_sequence, bit->sequence - 1);
2035                 return 0;
2036         }
2037 }
2038
2039 static void show_entries_rb(int force)
2040 {
2041         struct per_dev_info *pdi = NULL;
2042         struct per_cpu_info *pci = NULL;
2043         struct blk_io_trace *bit;
2044         struct rb_node *n;
2045         struct trace *t;
2046
2047         while ((n = rb_first(&rb_sort_root)) != NULL) {
2048                 if (is_done() && !force && !pipeline)
2049                         break;
2050
2051                 t = rb_entry(n, struct trace, rb_node);
2052                 bit = t->bit;
2053
2054                 if (read_sequence - t->read_sequence < 1 && !force)
2055                         break;
2056
2057                 if (!pdi || pdi->dev != bit->device) {
2058                         pdi = get_dev_info(bit->device);
2059                         pci = NULL;
2060                 }
2061
2062                 if (!pdi) {
2063                         fprintf(stderr, "Unknown device ID? (%d,%d)\n",
2064                                 MAJOR(bit->device), MINOR(bit->device));
2065                         break;
2066                 }
2067
2068                 if (check_sequence(pdi, t, force))
2069                         break;
2070
2071                 if (!force && bit->time > last_allowed_time)
2072                         break;
2073
2074                 check_time(pdi, bit);
2075
2076                 if (!pci || pci->cpu != bit->cpu)
2077                         pci = get_cpu_info(pdi, bit->cpu);
2078
2079                 pci->last_sequence = bit->sequence;
2080
2081                 pci->nelems++;
2082
2083                 if (bit->action & (act_mask << BLK_TC_SHIFT))
2084                         dump_trace(bit, pci, pdi);
2085
2086                 put_trace(pdi, t);
2087         }
2088 }
2089
2090 static int read_data(int fd, void *buffer, int bytes, int block, int *fdblock)
2091 {
2092         int ret, bytes_left, fl;
2093         void *p;
2094
2095         if (block != *fdblock) {
2096                 fl = fcntl(fd, F_GETFL);
2097
2098                 if (!block) {
2099                         *fdblock = 0;
2100                         fcntl(fd, F_SETFL, fl | O_NONBLOCK);
2101                 } else {
2102                         *fdblock = 1;
2103                         fcntl(fd, F_SETFL, fl & ~O_NONBLOCK);
2104                 }
2105         }
2106
2107         bytes_left = bytes;
2108         p = buffer;
2109         while (bytes_left > 0) {
2110                 ret = read(fd, p, bytes_left);
2111                 if (!ret)
2112                         return 1;
2113                 else if (ret < 0) {
2114                         if (errno != EAGAIN) {
2115                                 perror("read");
2116                                 return -1;
2117                         }
2118
2119                         /*
2120                          * never do partial reads. we can return if we
2121                          * didn't read anything and we should not block,
2122                          * otherwise wait for data
2123                          */
2124                         if ((bytes_left == bytes) && !block)
2125                                 return 1;
2126
2127                         usleep(10);
2128                         continue;
2129                 } else {
2130                         p += ret;
2131                         bytes_left -= ret;
2132                 }
2133         }
2134
2135         return 0;
2136 }
2137
2138 static inline __u16 get_pdulen(struct blk_io_trace *bit)
2139 {
2140         if (data_is_native)
2141                 return bit->pdu_len;
2142
2143         return __bswap_16(bit->pdu_len);
2144 }
2145
2146 static inline __u32 get_magic(struct blk_io_trace *bit)
2147 {
2148         if (data_is_native)
2149                 return bit->magic;
2150
2151         return __bswap_32(bit->magic);
2152 }
2153
2154 static int read_events(int fd, int always_block, int *fdblock)
2155 {
2156         struct per_dev_info *pdi = NULL;
2157         unsigned int events = 0;
2158
2159         while (!is_done() && events < rb_batch) {
2160                 struct blk_io_trace *bit;
2161                 struct trace *t;
2162                 int pdu_len, should_block, ret;
2163                 __u32 magic;
2164
2165                 bit = bit_alloc();
2166
2167                 should_block = !events || always_block;
2168
2169                 ret = read_data(fd, bit, sizeof(*bit), should_block, fdblock);
2170                 if (ret) {
2171                         bit_free(bit);
2172                         if (!events && ret < 0)
2173                                 events = ret;
2174                         break;
2175                 }
2176
2177                 /*
2178                  * look at first trace to check whether we need to convert
2179                  * data in the future
2180                  */
2181                 if (data_is_native == -1 && check_data_endianness(bit->magic))
2182                         break;
2183
2184                 magic = get_magic(bit);
2185                 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2186                         fprintf(stderr, "Bad magic %x\n", magic);
2187                         break;
2188                 }
2189
2190                 pdu_len = get_pdulen(bit);
2191                 if (pdu_len) {
2192                         void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2193
2194                         if (read_data(fd, ptr + sizeof(*bit), pdu_len, 1, fdblock)) {
2195                                 bit_free(ptr);
2196                                 break;
2197                         }
2198
2199                         bit = ptr;
2200                 }
2201
2202                 trace_to_cpu(bit);
2203
2204                 if (verify_trace(bit)) {
2205                         bit_free(bit);
2206                         continue;
2207                 }
2208
2209                 /*
2210                  * not a real trace, so grab and handle it here
2211                  */
2212                 if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
2213                         handle_notify(bit);
2214                         output_binary(bit, sizeof(*bit) + bit->pdu_len);
2215                         continue;
2216                 }
2217
2218                 t = t_alloc();
2219                 memset(t, 0, sizeof(*t));
2220                 t->bit = bit;
2221                 t->read_sequence = read_sequence;
2222
2223                 t->next = trace_list;
2224                 trace_list = t;
2225
2226                 if (!pdi || pdi->dev != bit->device)
2227                         pdi = get_dev_info(bit->device);
2228
2229                 if (bit->time > pdi->last_read_time)
2230                         pdi->last_read_time = bit->time;
2231
2232                 events++;
2233         }
2234
2235         return events;
2236 }
2237
2238 /*
2239  * Managing input streams
2240  */
2241
2242 struct ms_stream {
2243         struct ms_stream *next;
2244         struct trace *first, *last;
2245         struct per_dev_info *pdi;
2246         unsigned int cpu;
2247 };
2248
2249 #define MS_HASH(d, c) ((MAJOR(d) & 0xff) ^ (MINOR(d) & 0xff) ^ (cpu & 0xff))
2250
2251 struct ms_stream *ms_head;
2252 struct ms_stream *ms_hash[256];
2253
2254 static void ms_sort(struct ms_stream *msp);
2255 static int ms_prime(struct ms_stream *msp);
2256
2257 static inline struct trace *ms_peek(struct ms_stream *msp)
2258 {
2259         return (msp == NULL) ? NULL : msp->first;
2260 }
2261
2262 static inline __u64 ms_peek_time(struct ms_stream *msp)
2263 {
2264         return ms_peek(msp)->bit->time;
2265 }
2266
2267 static inline void ms_resort(struct ms_stream *msp)
2268 {
2269         if (msp->next && ms_peek_time(msp) > ms_peek_time(msp->next)) {
2270                 ms_head = msp->next;
2271                 msp->next = NULL;
2272                 ms_sort(msp);
2273         }
2274 }
2275
2276 static inline void ms_deq(struct ms_stream *msp)
2277 {
2278         msp->first = msp->first->next;
2279         if (!msp->first) {
2280                 msp->last = NULL;
2281                 if (!ms_prime(msp)) {
2282                         ms_head = msp->next;
2283                         msp->next = NULL;
2284                         return;
2285                 }
2286         }
2287
2288         ms_resort(msp);
2289 }
2290
2291 static void ms_sort(struct ms_stream *msp)
2292 {
2293         __u64 msp_t = ms_peek_time(msp);
2294         struct ms_stream *this_msp = ms_head;
2295
2296         if (this_msp == NULL)
2297                 ms_head = msp;
2298         else if (msp_t < ms_peek_time(this_msp)) {
2299                 msp->next = this_msp;
2300                 ms_head = msp;
2301         }
2302         else {
2303                 while (this_msp->next && ms_peek_time(this_msp->next) < msp_t)
2304                         this_msp = this_msp->next;
2305
2306                 msp->next = this_msp->next;
2307                 this_msp->next = msp;
2308         }
2309 }
2310
2311 static int ms_prime(struct ms_stream *msp)
2312 {
2313         __u32 magic;
2314         unsigned int i;
2315         struct trace *t;
2316         struct per_dev_info *pdi = msp->pdi;
2317         struct per_cpu_info *pci = get_cpu_info(pdi, msp->cpu);
2318         struct blk_io_trace *bit = NULL;
2319         int ret, pdu_len, ndone = 0;
2320
2321         for (i = 0; !is_done() && pci->fd >= 0 && i < rb_batch; i++) {
2322                 bit = bit_alloc();
2323                 ret = read_data(pci->fd, bit, sizeof(*bit), 1, &pci->fdblock);
2324                 if (ret)
2325                         goto err;
2326
2327                 if (data_is_native == -1 && check_data_endianness(bit->magic))
2328                         goto err;
2329
2330                 magic = get_magic(bit);
2331                 if ((magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
2332                         fprintf(stderr, "Bad magic %x\n", magic);
2333                         goto err;
2334
2335                 }
2336
2337                 pdu_len = get_pdulen(bit);
2338                 if (pdu_len) {
2339                         void *ptr = realloc(bit, sizeof(*bit) + pdu_len);
2340                         ret = read_data(pci->fd, ptr + sizeof(*bit), pdu_len,
2341                                                              1, &pci->fdblock);
2342                         if (ret) {
2343                                 free(ptr);
2344                                 bit = NULL;
2345                                 goto err;
2346                         }
2347
2348                         bit = ptr;
2349                 }
2350
2351                 trace_to_cpu(bit);
2352                 if (verify_trace(bit))
2353                         goto err;
2354
2355                 if (bit->action & BLK_TC_ACT(BLK_TC_NOTIFY) && bit->action != BLK_TN_MESSAGE) {
2356                         handle_notify(bit);
2357                         output_binary(bit, sizeof(*bit) + bit->pdu_len);
2358                         bit_free(bit);
2359
2360                         i -= 1;
2361                         continue;
2362                 }
2363
2364                 if (bit->time > pdi->last_read_time)
2365                         pdi->last_read_time = bit->time;
2366
2367                 t = t_alloc();
2368                 memset(t, 0, sizeof(*t));
2369                 t->bit = bit;
2370
2371                 if (msp->first == NULL)
2372                         msp->first = msp->last = t;
2373                 else {
2374                         msp->last->next = t;
2375                         msp->last = t;
2376                 }
2377
2378                 ndone++;
2379         }
2380
2381         return ndone;
2382
2383 err:
2384         if (bit) bit_free(bit);
2385
2386         cpu_mark_offline(pdi, pci->cpu);
2387         close(pci->fd);
2388         pci->fd = -1;
2389
2390         return ndone;
2391 }
2392
2393 static struct ms_stream *ms_alloc(struct per_dev_info *pdi, int cpu)
2394 {
2395         struct ms_stream *msp = malloc(sizeof(*msp));
2396
2397         msp->next = NULL;
2398         msp->first = msp->last = NULL;
2399         msp->pdi = pdi;
2400         msp->cpu = cpu;
2401
2402         if (ms_prime(msp))
2403                 ms_sort(msp);
2404
2405         return msp;
2406 }
2407
2408 static int setup_file(struct per_dev_info *pdi, int cpu)
2409 {
2410         int len = 0;
2411         struct stat st;
2412         char *p, *dname;
2413         struct per_cpu_info *pci = get_cpu_info(pdi, cpu);
2414
2415         pci->cpu = cpu;
2416         pci->fdblock = -1;
2417
2418         p = strdup(pdi->name);
2419         dname = dirname(p);
2420         if (strcmp(dname, ".")) {
2421                 input_dir = dname;
2422                 p = strdup(pdi->name);
2423                 strcpy(pdi->name, basename(p));
2424         }
2425         free(p);
2426
2427         if (input_dir)
2428                 len = sprintf(pci->fname, "%s/", input_dir);
2429
2430         snprintf(pci->fname + len, sizeof(pci->fname)-1-len,
2431                  "%s.blktrace.%d", pdi->name, pci->cpu);
2432         if (stat(pci->fname, &st) < 0)
2433                 return 0;
2434         if (!st.st_size)
2435                 return 1;
2436
2437         pci->fd = open(pci->fname, O_RDONLY);
2438         if (pci->fd < 0) {
2439                 perror(pci->fname);
2440                 return 0;
2441         }
2442
2443         printf("Input file %s added\n", pci->fname);
2444         cpu_mark_online(pdi, pci->cpu);
2445
2446         pdi->nfiles++;
2447         ms_alloc(pdi, pci->cpu);
2448
2449         return 1;
2450 }
2451
2452 static int handle(struct ms_stream *msp)
2453 {
2454         struct trace *t;
2455         struct per_dev_info *pdi;
2456         struct per_cpu_info *pci;
2457         struct blk_io_trace *bit;
2458
2459         t = ms_peek(msp);
2460
2461         bit = t->bit;
2462         pdi = msp->pdi;
2463         pci = get_cpu_info(pdi, msp->cpu);
2464         pci->nelems++;
2465         bit->time -= genesis_time;
2466
2467         if (t->bit->time > stopwatch_end)
2468                 return 0;
2469
2470         pdi->last_reported_time = bit->time;
2471         if ((bit->action & (act_mask << BLK_TC_SHIFT))&&
2472             t->bit->time >= stopwatch_start)
2473                 dump_trace(bit, pci, pdi);
2474
2475         ms_deq(msp);
2476
2477         if (text_output)
2478                 trace_rb_insert_last(pdi, t);
2479         else {
2480                 bit_free(t->bit);
2481                 t_free(t);
2482         }
2483
2484         return 1;
2485 }
2486
2487 /*
2488  * Check if we need to sanitize the name. We allow 'foo', or if foo.blktrace.X
2489  * is given, then strip back down to 'foo' to avoid missing files.
2490  */
2491 static int name_fixup(char *name)
2492 {
2493         char *b;
2494
2495         if (!name)
2496                 return 1;
2497
2498         b = strstr(name, ".blktrace.");
2499         if (b)
2500                 *b = '\0';
2501
2502         return 0;
2503 }
2504
2505 static int do_file(void)
2506 {
2507         int i, cpu, ret;
2508         struct per_dev_info *pdi;
2509
2510         /*
2511          * first prepare all files for reading
2512          */
2513         for (i = 0; i < ndevices; i++) {
2514                 pdi = &devices[i];
2515                 ret = name_fixup(pdi->name);
2516                 if (ret)
2517                         return ret;
2518
2519                 for (cpu = 0; setup_file(pdi, cpu); cpu++)
2520                         ;
2521
2522                 if (!cpu) {
2523                         fprintf(stderr,"No input files found for %s\n",
2524                                 pdi->name);
2525                         return 1;
2526                 }
2527         }
2528
2529         /*
2530          * Get the initial time stamp
2531          */
2532         if (ms_head)
2533                 genesis_time = ms_peek_time(ms_head);
2534
2535         /*
2536          * Keep processing traces while any are left
2537          */
2538         while (!is_done() && ms_head && handle(ms_head))
2539                 ;
2540
2541         return 0;
2542 }
2543
2544 static void do_pipe(int fd)
2545 {
2546         unsigned long long youngest;
2547         int events, fdblock;
2548
2549         last_allowed_time = -1ULL;
2550         fdblock = -1;
2551         while ((events = read_events(fd, 0, &fdblock)) > 0) {
2552                 read_sequence++;
2553
2554 #if 0
2555                 smallest_seq_read = -1U;
2556 #endif
2557
2558                 if (sort_entries(&youngest))
2559                         break;
2560
2561                 if (youngest > stopwatch_end)
2562                         break;
2563
2564                 show_entries_rb(0);
2565         }
2566
2567         if (rb_sort_entries)
2568                 show_entries_rb(1);
2569 }
2570
2571 static int do_fifo(void)
2572 {
2573         int fd;
2574
2575         if (!strcmp(pipename, "-"))
2576                 fd = dup(STDIN_FILENO);
2577         else
2578                 fd = open(pipename, O_RDONLY);
2579
2580         if (fd == -1) {
2581                 perror("dup stdin");
2582                 return -1;
2583         }
2584
2585         do_pipe(fd);
2586         close(fd);
2587         return 0;
2588 }
2589
2590 static void show_stats(void)
2591 {
2592         if (!ofp)
2593                 return;
2594         if (stats_printed)
2595                 return;
2596
2597         stats_printed = 1;
2598
2599         if (per_process_stats)
2600                 show_process_stats();
2601
2602         if (per_device_and_cpu_stats)
2603                 show_device_and_cpu_stats();
2604
2605         fflush(ofp);
2606 }
2607
2608 static void handle_sigint(__attribute__((__unused__)) int sig)
2609 {
2610         done = 1;
2611 }
2612
2613 /*
2614  * Extract start and duration times from a string, allowing
2615  * us to specify a time interval of interest within a trace.
2616  * Format: "duration" (start is zero) or "start:duration".
2617  */
2618 static int find_stopwatch_interval(char *string)
2619 {
2620         double value;
2621         char *sp;
2622
2623         value = strtod(string, &sp);
2624         if (sp == string) {
2625                 fprintf(stderr,"Invalid stopwatch timer: %s\n", string);
2626                 return 1;
2627         }
2628         if (*sp == ':') {
2629                 stopwatch_start = DOUBLE_TO_NANO_ULL(value);
2630                 string = sp + 1;
2631                 value = strtod(string, &sp);
2632                 if (sp == string || *sp != '\0') {
2633                         fprintf(stderr,"Invalid stopwatch duration time: %s\n",
2634                                 string);
2635                         return 1;
2636                 }
2637         } else if (*sp != '\0') {
2638                 fprintf(stderr,"Invalid stopwatch start timer: %s\n", string);
2639                 return 1;
2640         }
2641         stopwatch_end = DOUBLE_TO_NANO_ULL(value);
2642         if (stopwatch_end <= stopwatch_start) {
2643                 fprintf(stderr, "Invalid stopwatch interval: %Lu -> %Lu\n",
2644                         stopwatch_start, stopwatch_end);
2645                 return 1;
2646         }
2647
2648         return 0;
2649 }
2650
2651 static int is_pipe(const char *str)
2652 {
2653         struct stat st;
2654
2655         if (!strcmp(str, "-"))
2656                 return 1;
2657         if (!stat(str, &st) && S_ISFIFO(st.st_mode))
2658                 return 1;
2659
2660         return 0;
2661 }
2662
2663 #define S_OPTS  "a:A:b:D:d:f:F:hi:o:Oqstw:vVM"
2664 static char usage_str[] =    "\n\n" \
2665         "-i <file>           | --input=<file>\n" \
2666         "[ -a <action field> | --act-mask=<action field> ]\n" \
2667         "[ -A <action mask>  | --set-mask=<action mask> ]\n" \
2668         "[ -b <traces>       | --batch=<traces> ]\n" \
2669         "[ -d <file>         | --dump-binary=<file> ]\n" \
2670         "[ -D <dir>          | --input-directory=<dir> ]\n" \
2671         "[ -f <format>       | --format=<format> ]\n" \
2672         "[ -F <spec>         | --format-spec=<spec> ]\n" \
2673         "[ -h                | --hash-by-name ]\n" \
2674         "[ -o <file>         | --output=<file> ]\n" \
2675         "[ -O                | --no-text-output ]\n" \
2676         "[ -q                | --quiet ]\n" \
2677         "[ -s                | --per-program-stats ]\n" \
2678         "[ -t                | --track-ios ]\n" \
2679         "[ -w <time>         | --stopwatch=<time> ]\n" \
2680         "[ -M                | --no-msgs\n" \
2681         "[ -v                | --verbose ]\n" \
2682         "[ -V                | --version ]\n\n" \
2683         "\t-a Only trace specified actions. See documentation\n" \
2684         "\t-A Give trace mask as a single value. See documentation\n" \
2685         "\t-b stdin read batching\n" \
2686         "\t-d Output file. If specified, binary data is written to file\n" \
2687         "\t-D Directory to prepend to input file names\n" \
2688         "\t-f Output format. Customize the output format. The format field\n" \
2689         "\t   identifies can be found in the documentation\n" \
2690         "\t-F Format specification. Can be found in the documentation\n" \
2691         "\t-h Hash processes by name, not pid\n" \
2692         "\t-i Input file containing trace data, or '-' for stdin\n" \
2693         "\t-o Output file. If not given, output is stdout\n" \
2694         "\t-O Do NOT output text data\n" \
2695         "\t-q Quiet. Don't display any stats at the end of the trace\n" \
2696         "\t-s Show per-program io statistics\n" \
2697         "\t-t Track individual ios. Will tell you the time a request took\n" \
2698         "\t   to get queued, to get dispatched, and to get completed\n" \
2699         "\t-w Only parse data between the given time interval in seconds.\n" \
2700         "\t   If 'start' isn't given, blkparse defaults the start time to 0\n" \
2701         "\t-M Do not output messages to binary file\n" \
2702         "\t-v More verbose for marginal errors\n" \
2703         "\t-V Print program version info\n\n";
2704
2705 static void usage(char *prog)
2706 {
2707         fprintf(stderr, "Usage: %s %s", prog, usage_str);
2708 }
2709
2710 int main(int argc, char *argv[])
2711 {
2712         int i, c, ret, mode;
2713         int act_mask_tmp = 0;
2714         char *ofp_buffer = NULL;
2715         char *bin_ofp_buffer = NULL;
2716
2717         while ((c = getopt_long(argc, argv, S_OPTS, l_opts, NULL)) != -1) {
2718                 switch (c) {
2719                 case 'a':
2720                         i = find_mask_map(optarg);
2721                         if (i < 0) {
2722                                 fprintf(stderr,"Invalid action mask %s\n",
2723                                         optarg);
2724                                 return 1;
2725                         }
2726                         act_mask_tmp |= i;
2727                         break;
2728
2729                 case 'A':
2730                         if ((sscanf(optarg, "%x", &i) != 1) ||
2731                                                         !valid_act_opt(i)) {
2732                                 fprintf(stderr,
2733                                         "Invalid set action mask %s/0x%x\n",
2734                                         optarg, i);
2735                                 return 1;
2736                         }
2737                         act_mask_tmp = i;
2738                         break;
2739                 case 'i':
2740                         if (is_pipe(optarg) && !pipeline) {
2741                                 pipeline = 1;
2742                                 pipename = strdup(optarg);
2743                         } else if (resize_devices(optarg) != 0)
2744                                 return 1;
2745                         break;
2746                 case 'D':
2747                         input_dir = optarg;
2748                         break;
2749                 case 'o':
2750                         output_name = optarg;
2751                         break;
2752                 case 'O':
2753                         text_output = 0;
2754                         break;
2755                 case 'b':
2756                         rb_batch = atoi(optarg);
2757                         if (rb_batch <= 0)
2758                                 rb_batch = RB_BATCH_DEFAULT;
2759                         break;
2760                 case 's':
2761                         per_process_stats = 1;
2762                         break;
2763                 case 't':
2764                         track_ios = 1;
2765                         break;
2766                 case 'q':
2767                         per_device_and_cpu_stats = 0;
2768                         break;
2769                 case 'w':
2770                         if (find_stopwatch_interval(optarg) != 0)
2771                                 return 1;
2772                         break;
2773                 case 'f':
2774                         set_all_format_specs(optarg);
2775                         break;
2776                 case 'F':
2777                         if (add_format_spec(optarg) != 0)
2778                                 return 1;
2779                         break;
2780                 case 'h':
2781                         ppi_hash_by_pid = 0;
2782                         break;
2783                 case 'v':
2784                         verbose++;
2785                         break;
2786                 case 'V':
2787                         printf("%s version %s\n", argv[0], blkparse_version);
2788                         return 0;
2789                 case 'd':
2790                         dump_binary = optarg;
2791                         break;
2792                 case 'M':
2793                         bin_output_msgs = 0;
2794                         break;
2795                 default:
2796                         usage(argv[0]);
2797                         return 1;
2798                 }
2799         }
2800
2801         while (optind < argc) {
2802                 if (is_pipe(argv[optind]) && !pipeline) {
2803                         pipeline = 1;
2804                         pipename = strdup(argv[optind]);
2805                 } else if (resize_devices(argv[optind]) != 0)
2806                         return 1;
2807                 optind++;
2808         }
2809
2810         if (!pipeline && !ndevices) {
2811                 usage(argv[0]);
2812                 return 1;
2813         }
2814
2815         if (act_mask_tmp != 0)
2816                 act_mask = act_mask_tmp;
2817
2818         memset(&rb_sort_root, 0, sizeof(rb_sort_root));
2819
2820         signal(SIGINT, handle_sigint);
2821         signal(SIGHUP, handle_sigint);
2822         signal(SIGTERM, handle_sigint);
2823
2824         setlocale(LC_NUMERIC, "en_US");
2825
2826         if (text_output) {
2827                 if (!output_name) {
2828                         ofp = fdopen(STDOUT_FILENO, "w");
2829                         mode = _IOLBF;
2830                 } else {
2831                         char ofname[128];
2832
2833                         snprintf(ofname, sizeof(ofname) - 1, "%s", output_name);
2834                         ofp = fopen(ofname, "w");
2835                         mode = _IOFBF;
2836                 }
2837
2838                 if (!ofp) {
2839                         perror("fopen");
2840                         return 1;
2841                 }
2842
2843                 ofp_buffer = malloc(4096);
2844                 if (setvbuf(ofp, ofp_buffer, mode, 4096)) {
2845                         perror("setvbuf");
2846                         return 1;
2847                 }
2848         }
2849
2850         if (dump_binary) {
2851                 if (!strcmp(dump_binary, "-"))
2852                         dump_fp = stdout;
2853                 else {
2854                         dump_fp = fopen(dump_binary, "w");
2855                         if (!dump_fp) {
2856                                 perror(dump_binary);
2857                                 dump_binary = NULL;
2858                                 return 1;
2859                         }
2860                 }
2861                 bin_ofp_buffer = malloc(128 * 1024);
2862                 if (setvbuf(dump_fp, bin_ofp_buffer, _IOFBF, 128 * 1024)) {
2863                         perror("setvbuf binary");
2864                         return 1;
2865                 }
2866         }
2867
2868         if (pipeline)
2869                 ret = do_fifo();
2870         else
2871                 ret = do_file();
2872
2873         if (!ret)
2874                 show_stats();
2875
2876         if (have_drv_data && !dump_binary)
2877                 printf("\ndiscarded traces containing low-level device driver "
2878                        "specific data (only available in binary output)\n");
2879
2880         if (ofp_buffer) {
2881                 fflush(ofp);
2882                 free(ofp_buffer);
2883         }
2884         if (bin_ofp_buffer) {
2885                 fflush(dump_fp);
2886                 free(bin_ofp_buffer);
2887         }
2888         return ret;
2889 }