t/btrace2fio.c

   1 #include <stdio.h>
   2 #include <stdio.h>
   3 #include <unistd.h>
   4 #include <inttypes.h>
   5 #include <assert.h>
   6
   7 #include "../io_ddir.h"
   8 #include "../flist.h"
   9 #include "../hash.h"
  10 #include "../fifo.h"
  11 #include "../blktrace_api.h"
  12 #include "../os/os.h"
  13 #include "../log.h"
  14 #include "../lib/linux-dev-lookup.h"
  15
  16 #define TRACE_FIFO_SIZE 8192
  17
  18 static unsigned int rt_threshold = 1000000;
  19 static unsigned int ios_threshold = 10;
  20 static int output_ascii = 1;
  21 static char *filename;
  22
  23 struct bs {
  24         unsigned int bs;
  25         unsigned int nr;
  26         int merges;
  27 };
  28
  29 struct trace_file {
  30         char *name;
  31         int major, minor;
  32 };
  33
  34 struct btrace_out {
  35         unsigned long ios[DDIR_RWDIR_CNT];
  36         unsigned long rw_bs[DDIR_RWDIR_CNT];
  37         unsigned long merges[DDIR_RWDIR_CNT];
  38
  39         uint64_t last_end[DDIR_RWDIR_CNT];
  40         uint64_t seq[DDIR_RWDIR_CNT];
  41
  42         struct bs *bs[DDIR_RWDIR_CNT];
  43         unsigned int nr_bs[DDIR_RWDIR_CNT];
  44
  45         int inflight;
  46         unsigned int depth;
  47         uint64_t first_ttime;
  48         uint64_t last_ttime;
  49         uint64_t kb;
  50
  51         uint64_t start_delay;
  52 };
  53
  54 struct btrace_pid {
  55         struct flist_head hash_list;
  56         struct flist_head pid_list;
  57         pid_t pid;
  58
  59         struct trace_file *files;
  60         int nr_files;
  61         unsigned int last_major, last_minor;
  62
  63         struct btrace_out o;
  64 };
  65
  66 struct inflight {
  67         struct flist_head list;
  68         struct btrace_pid *p;
  69         uint64_t end_sector;
  70 };
  71
  72 #define PID_HASH_BITS   10
  73 #define PID_HASH_SIZE   (1U << PID_HASH_BITS)
  74
  75 static struct flist_head pid_hash[PID_HASH_SIZE];
  76 static FLIST_HEAD(pid_list);
  77
  78 #define INFLIGHT_HASH_BITS      8
  79 #define INFLIGHT_HASH_SIZE      (1U << INFLIGHT_HASH_BITS)
  80 static struct flist_head inflight_hash[INFLIGHT_HASH_SIZE];
  81
  82 static uint64_t first_ttime = -1ULL;
  83
  84 static struct inflight *inflight_find(uint64_t sector)
  85 {
  86         struct flist_head *inflight_list;
  87         struct flist_head *e;
  88
  89         inflight_list = &inflight_hash[hash_long(sector, INFLIGHT_HASH_BITS)];
  90
  91         flist_for_each(e, inflight_list) {
  92                 struct inflight *i = flist_entry(e, struct inflight, list);
  93
  94                 if (i->end_sector == sector)
  95                         return i;
  96         }
  97
  98         return NULL;
  99 }
 100
 101 static void inflight_remove(struct inflight *i)
 102 {
 103         struct btrace_out *o = &i->p->o;
 104
 105         o->inflight--;
 106         assert(o->inflight >= 0);
 107         flist_del(&i->list);
 108         free(i);
 109 }
 110
 111 static void __inflight_add(struct inflight *i)
 112 {
 113         struct flist_head *list;
 114
 115         list = &inflight_hash[hash_long(i->end_sector, INFLIGHT_HASH_BITS)];
 116         flist_add_tail(&i->list, list);
 117 }
 118
 119 static void inflight_add(struct btrace_pid *p, uint64_t sector, uint32_t len)
 120 {
 121         struct btrace_out *o = &p->o;
 122         struct inflight *i;
 123
 124         i = calloc(1, sizeof(*i));
 125         i->p = p;
 126         o->inflight++;
 127         o->depth = max((int) o->depth, o->inflight);
 128         i->end_sector = sector + (len >> 9);
 129         __inflight_add(i);
 130 }
 131
 132 static void inflight_merge(struct inflight *i, int rw, unsigned int size)
 133 {
 134         i->p->o.merges[rw]++;
 135         if (size) {
 136                 i->end_sector += (size >> 9);
 137                 flist_del(&i->list);
 138                 __inflight_add(i);
 139         }
 140 }
 141
 142 /*
 143  * fifo refill frontend, to avoid reading data in trace sized bites
 144  */
 145 static int refill_fifo(struct fifo *fifo, int fd)
 146 {
 147         char buf[TRACE_FIFO_SIZE];
 148         unsigned int total;
 149         int ret;
 150
 151         total = sizeof(buf);
 152         if (total > fifo_room(fifo))
 153                 total = fifo_room(fifo);
 154
 155         ret = read(fd, buf, total);
 156         if (ret < 0) {
 157                 perror("read refill");
 158                 return -1;
 159         }
 160
 161         if (ret > 0)
 162                 ret = fifo_put(fifo, buf, ret);
 163
 164         return ret;
 165 }
 166
 167 /*
 168  * Retrieve 'len' bytes from the fifo, refilling if necessary.
 169  */
 170 static int trace_fifo_get(struct fifo *fifo, int fd, void *buf,
 171                           unsigned int len)
 172 {
 173         if (fifo_len(fifo) < len) {
 174                 int ret = refill_fifo(fifo, fd);
 175
 176                 if (ret < 0)
 177                         return ret;
 178         }
 179
 180         return fifo_get(fifo, buf, len);
 181 }
 182
 183 /*
 184  * Just discard the pdu by seeking past it.
 185  */
 186 static int discard_pdu(struct fifo *fifo, int fd, struct blk_io_trace *t)
 187 {
 188         if (t->pdu_len == 0)
 189                 return 0;
 190
 191         return trace_fifo_get(fifo, fd, NULL, t->pdu_len);
 192 }
 193
 194 static void handle_trace_notify(struct blk_io_trace *t)
 195 {
 196         switch (t->action) {
 197         case BLK_TN_PROCESS:
 198                 //printf("got process notify: %x, %d\n", t->action, t->pid);
 199                 break;
 200         case BLK_TN_TIMESTAMP:
 201                 //printf("got timestamp notify: %x, %d\n", t->action, t->pid);
 202                 break;
 203         case BLK_TN_MESSAGE:
 204                 break;
 205         default:
 206                 log_err("unknown trace act %x\n", t->action);
 207                 break;
 208         }
 209 }
 210
 211 static void __add_bs(struct btrace_out *o, unsigned int len, int rw)
 212 {
 213         o->bs[rw] = realloc(o->bs[rw], (o->nr_bs[rw] + 1) * sizeof(struct bs));
 214         o->bs[rw][o->nr_bs[rw]].bs = len;
 215         o->bs[rw][o->nr_bs[rw]].nr = 1;
 216         o->nr_bs[rw]++;
 217 }
 218
 219 static void add_bs(struct btrace_out *o, unsigned int len, int rw)
 220 {
 221         struct bs *bs = o->bs[rw];
 222         int i;
 223
 224         if (!o->nr_bs[rw]) {
 225                 __add_bs(o, len, rw);
 226                 return;
 227         }
 228
 229         for (i = 0; i < o->nr_bs[rw]; i++) {
 230                 if (bs[i].bs == len) {
 231                         bs[i].nr++;
 232                         return;
 233                 }
 234         }
 235
 236         __add_bs(o, len, rw);
 237 }
 238
 239 #define FMINORBITS      20
 240 #define FMINORMASK      ((1U << FMINORBITS) - 1)
 241 #define FMAJOR(dev)     ((unsigned int) ((dev) >> FMINORBITS))
 242 #define FMINOR(dev)     ((unsigned int) ((dev) & FMINORMASK))
 243
 244 static void btrace_add_file(struct btrace_pid *p, uint32_t devno)
 245 {
 246         unsigned int maj = FMAJOR(devno);
 247         unsigned int min = FMINOR(devno);
 248         struct trace_file *f;
 249         unsigned int i;
 250         char dev[256];
 251
 252         if (filename)
 253                 return;
 254         if (p->last_major == maj && p->last_minor == min)
 255                 return;
 256
 257         p->last_major = maj;
 258         p->last_minor = min;
 259
 260         /*
 261          * check for this file in our list
 262          */
 263         for (i = 0; i < p->nr_files; i++) {
 264                 f = &p->files[i];
 265
 266                 if (f->major == maj && f->minor == min)
 267                         return;
 268         }
 269
 270         strcpy(dev, "/dev");
 271         if (!blktrace_lookup_device(NULL, dev, maj, min)) {
 272                 log_err("fio: failed to find device %u/%u\n", maj, min);
 273                 return;
 274         }
 275
 276         p->files = realloc(p->files, (p->nr_files + 1) * sizeof(*f));
 277         f = &p->files[p->nr_files];
 278         f->name = strdup(dev);
 279         f->major = maj;
 280         f->minor = min;
 281         p->nr_files++;
 282 }
 283
 284 static void handle_trace_discard(struct blk_io_trace *t, struct btrace_pid *p)
 285 {
 286         struct btrace_out *o = &p->o;
 287
 288         btrace_add_file(p, t->device);
 289
 290         if (o->first_ttime == -1ULL)
 291                 o->first_ttime = t->time;
 292
 293         o->ios[DDIR_TRIM]++;
 294         add_bs(o, t->bytes, DDIR_TRIM);
 295 }
 296
 297 static void handle_trace_fs(struct blk_io_trace *t, struct btrace_pid *p)
 298 {
 299         struct btrace_out *o = &p->o;
 300         int rw;
 301
 302         btrace_add_file(p, t->device);
 303
 304         first_ttime = min(first_ttime, (uint64_t) t->time);
 305
 306         if (o->first_ttime == -1ULL)
 307                 o->first_ttime = t->time;
 308
 309         rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
 310
 311         add_bs(o, t->bytes, rw);
 312         o->ios[rw]++;
 313
 314         if (t->sector == o->last_end[rw] || o->last_end[rw] == -1ULL)
 315                 o->seq[rw]++;
 316
 317         o->last_end[rw] = t->sector + (t->bytes >> 9);
 318 }
 319
 320 static void handle_queue_trace(struct blk_io_trace *t, struct btrace_pid *p)
 321 {
 322         if (t->action & BLK_TC_ACT(BLK_TC_NOTIFY))
 323                 handle_trace_notify(t);
 324         else if (t->action & BLK_TC_ACT(BLK_TC_DISCARD))
 325                 handle_trace_discard(t, p);
 326         else
 327                 handle_trace_fs(t, p);
 328 }
 329
 330 static void handle_trace(struct blk_io_trace *t, struct btrace_pid *p)
 331 {
 332         unsigned int act = t->action & 0xffff;
 333
 334         if (act == __BLK_TA_QUEUE) {
 335                 inflight_add(p, t->sector, t->bytes);
 336                 handle_queue_trace(t, p);
 337         } else if (act == __BLK_TA_REQUEUE) {
 338                 p->o.inflight--;
 339         } else if (act == __BLK_TA_BACKMERGE) {
 340                 struct inflight *i;
 341
 342                 i = inflight_find(t->sector + (t->bytes >> 9));
 343                 if (i)
 344                         inflight_remove(i);
 345
 346                 i = inflight_find(t->sector);
 347                 if (i) {
 348                         int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
 349
 350                         inflight_merge(i, rw, t->bytes);
 351                 }
 352         } else if (act == __BLK_TA_FRONTMERGE) {
 353                 struct inflight *i;
 354
 355                 i = inflight_find(t->sector + (t->bytes >> 9));
 356                 if (i)
 357                         inflight_remove(i);
 358
 359                 i = inflight_find(t->sector);
 360                 if (i) {
 361                         int rw = (t->action & BLK_TC_ACT(BLK_TC_WRITE)) != 0;
 362
 363                         inflight_merge(i, rw, 0);
 364                 }
 365         } else if (act == __BLK_TA_COMPLETE) {
 366                 struct inflight *i;
 367
 368                 i = inflight_find(t->sector + (t->bytes >> 9));
 369                 if (i) {
 370                         i->p->o.kb += (t->bytes >> 10);
 371                         inflight_remove(i);
 372                 }
 373         }
 374 }
 375
 376 static void byteswap_trace(struct blk_io_trace *t)
 377 {
 378         t->magic = fio_swap32(t->magic);
 379         t->sequence = fio_swap32(t->sequence);
 380         t->time = fio_swap64(t->time);
 381         t->sector = fio_swap64(t->sector);
 382         t->bytes = fio_swap32(t->bytes);
 383         t->action = fio_swap32(t->action);
 384         t->pid = fio_swap32(t->pid);
 385         t->device = fio_swap32(t->device);
 386         t->cpu = fio_swap32(t->cpu);
 387         t->error = fio_swap16(t->error);
 388         t->pdu_len = fio_swap16(t->pdu_len);
 389 }
 390
 391 static struct btrace_pid *pid_hash_find(pid_t pid, struct flist_head *list)
 392 {
 393         struct flist_head *e;
 394         struct btrace_pid *p;
 395
 396         flist_for_each(e, list) {
 397                 p = flist_entry(e, struct btrace_pid, hash_list);
 398                 if (p->pid == pid)
 399                         return p;
 400         }
 401
 402         return NULL;
 403 }
 404
 405 static struct btrace_pid *pid_hash_get(pid_t pid)
 406 {
 407         struct flist_head *hash_list;
 408         struct btrace_pid *p;
 409
 410         hash_list = &pid_hash[hash_long(pid, PID_HASH_BITS)];
 411
 412         p = pid_hash_find(pid, hash_list);
 413         if (!p) {
 414                 int i;
 415
 416                 p = calloc(1, sizeof(*p));
 417                 p->o.first_ttime = -1ULL;
 418                 p->o.last_ttime = -1ULL;
 419
 420                 for (i = 0; i < DDIR_RWDIR_CNT; i++)
 421                         p->o.last_end[i] = -1ULL;
 422
 423                 p->pid = pid;
 424                 flist_add_tail(&p->hash_list, hash_list);
 425                 flist_add_tail(&p->pid_list, &pid_list);
 426         }
 427
 428         return p;
 429 }
 430
 431 /*
 432  * Load a blktrace file by reading all the blk_io_trace entries, and storing
 433  * them as io_pieces like the fio text version would do.
 434  */
 435 static int load_blktrace(const char *filename, int need_swap)
 436 {
 437         struct btrace_pid *p;
 438         unsigned long traces;
 439         struct blk_io_trace t;
 440         struct fifo *fifo;
 441         int fd;
 442
 443         fd = open(filename, O_RDONLY);
 444         if (fd < 0) {
 445                 perror("open trace file\n");
 446                 return 1;
 447         }
 448
 449         fifo = fifo_alloc(TRACE_FIFO_SIZE);
 450
 451         traces = 0;
 452         do {
 453                 int ret = trace_fifo_get(fifo, fd, &t, sizeof(t));
 454
 455                 if (ret < 0)
 456                         goto err;
 457                 else if (!ret)
 458                         break;
 459                 else if (ret < (int) sizeof(t)) {
 460                         log_err("fio: short fifo get\n");
 461                         break;
 462                 }
 463
 464                 if (need_swap)
 465                         byteswap_trace(&t);
 466
 467                 if ((t.magic & 0xffffff00) != BLK_IO_TRACE_MAGIC) {
 468                         log_err("fio: bad magic in blktrace data: %x\n", t.magic);
 469                         goto err;
 470                 }
 471                 if ((t.magic & 0xff) != BLK_IO_TRACE_VERSION) {
 472                         log_err("fio: bad blktrace version %d\n", t.magic & 0xff);
 473                         goto err;
 474                 }
 475                 ret = discard_pdu(fifo, fd, &t);
 476                 if (ret < 0) {
 477                         log_err("blktrace lseek\n");
 478                         goto err;
 479                 } else if (t.pdu_len != ret) {
 480                         log_err("fio: discarded %d of %d\n", ret, t.pdu_len);
 481                         goto err;
 482                 }
 483
 484                 p = pid_hash_get(t.pid);
 485                 handle_trace(&t, p);
 486                 p->o.last_ttime = t.time;
 487                 traces++;
 488         } while (1);
 489
 490         fifo_free(fifo);
 491         close(fd);
 492
 493         if (output_ascii)
 494                 printf("Traces loaded: %lu\n", traces);
 495
 496         return 0;
 497 err:
 498         close(fd);
 499         fifo_free(fifo);
 500         return 1;
 501 }
 502
 503 static int bs_cmp(const void *ba, const void *bb)
 504 {
 505         const struct bs *bsa = ba;
 506         const struct bs *bsb = bb;
 507
 508         return bsb->nr - bsa->nr;
 509 }
 510
 511 static void __output_p_ascii(struct btrace_pid *p, unsigned long *ios)
 512 {
 513         const char *msg[] = { "reads", "writes", "trims" };
 514         struct btrace_out *o = &p->o;
 515         unsigned long total, usec;
 516         int i, j;
 517
 518         printf("[pid:\t%u]\n", p->pid);
 519
 520         total = ddir_rw_sum(o->ios);
 521         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 522                 float perc;
 523
 524                 if (!o->ios[i])
 525                         continue;
 526
 527                 ios[i] += o->ios[i] + o->merges[i];
 528                 printf("%s\n", msg[i]);
 529                 perc = ((float) o->ios[i] * 100.0) / (float) total;
 530                 printf("\tios:    %lu (perc=%3.2f%%)\n", o->ios[i], perc);
 531                 perc = ((float) o->merges[i] * 100.0) / (float) total;
 532                 printf("\tmerges: %lu (perc=%3.2f%%)\n", o->merges[i], perc);
 533                 perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
 534                 printf("\tseq:    %lu (perc=%3.2f%%)\n", o->seq[i], perc);
 535
 536                 for (j = 0; j < o->nr_bs[i]; j++) {
 537                         struct bs *bs = &o->bs[i][j];
 538
 539                         perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
 540                         printf("\tbs=%u, perc=%3.2f%%\n", bs->bs, perc);
 541                 }
 542         }
 543
 544         printf("depth:\t%u\n", o->depth);
 545         usec = (o->last_ttime - o->first_ttime) / 1000ULL;
 546         printf("usec:\t%lu (delay=%llu)\n", usec, (unsigned long long) o->start_delay);
 547         printf("rate:\t%.2fKB/sec\n", ((float) o->kb * 1000.0) / ((float) usec / 1000.0));
 548
 549         printf("files:\t");
 550         for (i = 0; i < p->nr_files; i++)
 551                 printf("%s,", p->files[i].name);
 552         printf("\n");
 553
 554         printf("\n");
 555 }
 556
 557 static int __output_p_fio(struct btrace_pid *p, unsigned long *ios)
 558 {
 559         struct btrace_out *o = &p->o;
 560         unsigned long total;
 561         unsigned long long time;
 562         float perc;
 563         int i, j;
 564
 565         if ((o->ios[0] + o->ios[1]) && o->ios[2]) {
 566                 log_err("fio: trace has both read/write and trim\n");
 567                 return 1;
 568         }
 569
 570         printf("[pid%u]\n", p->pid);
 571         printf("direct=1\n");
 572         if (o->depth == 1)
 573                 printf("ioengine=sync\n");
 574         else
 575                 printf("ioengine=libaio\niodepth=%u\n", o->depth);
 576
 577         if (o->ios[0] && !o->ios[1])
 578                 printf("rw=randread\n");
 579         else if (!o->ios[0] && o->ios[1])
 580                 printf("rw=randwrite\n");
 581         else if (o->ios[2])
 582                 printf("rw=randtrim\n");
 583         else {
 584                 printf("rw=randrw\n");
 585                 total = ddir_rw_sum(o->ios);
 586                 perc = ((float) o->ios[0] * 100.0) / (float) total;
 587                 printf("rwmixread=%u\n", (int) (perc + 0.99));
 588         }
 589
 590         printf("percentage_random=");
 591         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 592                 if (o->seq[i] && o->ios[i]) {
 593                         perc = ((float) o->seq[i] * 100.0) / (float) o->ios[i];
 594                         if (perc >= 99.0)
 595                                 perc = 100.0;
 596                 } else
 597                         perc = 100.0;
 598
 599                 if (i)
 600                         printf(",");
 601                 perc = 100.0 - perc;
 602                 printf("%u", (int) perc);
 603         }
 604         printf("\n");
 605
 606         printf("filename=");
 607         for (i = 0; i < p->nr_files; i++) {
 608                 if (i)
 609                         printf(":");
 610                 printf("%s", p->files[i].name);
 611         }
 612         printf("\n");
 613
 614         printf("startdelay=%llus\n", o->start_delay / 1000000ULL);
 615
 616         time = o->last_ttime - o->first_ttime;
 617         time = (time + 1000000000ULL - 1) / 1000000000ULL;
 618         printf("runtime=%llus\n", time);
 619
 620         printf("bssplit=");
 621         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 622
 623                 if (i && o->nr_bs[i - 1] && o->nr_bs[i])
 624                         printf(",");
 625
 626                 for (j = 0; j < o->nr_bs[i]; j++) {
 627                         struct bs *bs = &o->bs[i][j];
 628
 629                         perc = (((float) bs->nr * 100.0) / (float) o->ios[i]);
 630                         if (perc < 1.00)
 631                                 continue;
 632                         if (j)
 633                                 printf(":");
 634                         if (j + 1 == o->nr_bs[i])
 635                                 printf("%u/", bs->bs);
 636                         else
 637                                 printf("%u/%u", bs->bs, (int) perc);
 638                 }
 639         }
 640         printf("\n\n");
 641
 642         return 0;
 643 }
 644
 645 static int __output_p(struct btrace_pid *p, unsigned long *ios)
 646 {
 647         struct btrace_out *o = &p->o;
 648         int i, ret = 0;
 649
 650         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
 651                 if (o->nr_bs[i] <= 1)
 652                         continue;
 653                 qsort(o->bs[i], o->nr_bs[i], sizeof(struct bs), bs_cmp);
 654         }
 655
 656         if (filename) {
 657                 p->files = malloc(sizeof(struct trace_file));
 658                 p->nr_files++;
 659                 p->files[0].name = filename;
 660         }
 661
 662         if (output_ascii)
 663                 __output_p_ascii(p, ios);
 664         else
 665                 ret = __output_p_fio(p, ios);
 666
 667         return ret;
 668 }
 669
 670 static int prune_entry(struct btrace_out *o)
 671 {
 672         uint64_t time;
 673
 674         if (ddir_rw_sum(o->ios) < ios_threshold)
 675                 return 1;
 676
 677         time = (o->last_ttime - o->first_ttime) / 1000ULL;
 678         if (time < rt_threshold)
 679                 return 1;
 680
 681         return 0;
 682 }
 683
 684 static int entry_cmp(void *priv, struct flist_head *a, struct flist_head *b)
 685 {
 686         struct btrace_pid *pa = flist_entry(a, struct btrace_pid, pid_list);
 687         struct btrace_pid *pb = flist_entry(b, struct btrace_pid, pid_list);
 688
 689         return ddir_rw_sum(pb->o.ios) - ddir_rw_sum(pa->o.ios);
 690 }
 691
 692 static void free_p(struct btrace_pid *p)
 693 {
 694         struct btrace_out *o = &p->o;
 695         int i;
 696
 697         for (i = 0; i < p->nr_files; i++) {
 698                 if (p->files[i].name && p->files[i].name != filename)
 699                         free(p->files[i].name);
 700         }
 701
 702         for (i = 0; i < DDIR_RWDIR_CNT; i++)
 703                 free(o->bs[i]);
 704
 705         free(p->files);
 706         flist_del(&p->pid_list);
 707         flist_del(&p->hash_list);
 708         free(p);
 709 }
 710
 711 static int output_p(void)
 712 {
 713         unsigned long ios[DDIR_RWDIR_CNT];
 714         struct flist_head *e, *tmp;
 715         int ret = 0;
 716
 717         flist_for_each_safe(e, tmp, &pid_list) {
 718                 struct btrace_pid *p;
 719
 720                 p = flist_entry(e, struct btrace_pid, pid_list);
 721                 if (prune_entry(&p->o)) {
 722                         free_p(p);
 723                         continue;
 724                 }
 725                 p->o.start_delay = (p->o.first_ttime / 1000ULL) - first_ttime;
 726         }
 727
 728         memset(ios, 0, sizeof(ios));
 729
 730         flist_sort(NULL, &pid_list, entry_cmp);
 731
 732         flist_for_each(e, &pid_list) {
 733                 struct btrace_pid *p;
 734
 735                 p = flist_entry(e, struct btrace_pid, pid_list);
 736                 ret |= __output_p(p, ios);
 737         }
 738
 739         if (output_ascii)
 740                 printf("Total: reads=%lu, writes=%lu\n", ios[0], ios[1]);
 741
 742         return ret;
 743 }
 744
 745 static int usage(char *argv[])
 746 {
 747         log_err("%s: <blktrace bin file>\n", argv[0]);
 748         log_err("\t-t\tUsec threshold to ignore task\n");
 749         log_err("\t-n\tNumber IOS threshold to ignore task\n");
 750         log_err("\t-f\tFio job file output\n");
 751         log_err("\t-d\tUse this file/device for replay\n");
 752         return 1;
 753 }
 754
 755 static int trace_needs_swap(const char *trace_file, int *swap)
 756 {
 757         struct blk_io_trace t;
 758         int fd, ret;
 759
 760         *swap = -1;
 761
 762         fd = open(trace_file, O_RDONLY);
 763         if (fd < 0) {
 764                 perror("open");
 765                 return 1;
 766         }
 767
 768         ret = read(fd, &t, sizeof(t));
 769         if (ret < 0) {
 770                 perror("read");
 771                 return 1;
 772         } else if (ret != sizeof(t)) {
 773                 log_err("fio: short read on trace file\n");
 774                 return 1;
 775         }
 776
 777         close(fd);
 778
 779         if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
 780                 *swap = 0;
 781         else {
 782                 /*
 783                  * Maybe it needs to be endian swapped...
 784                  */
 785                 t.magic = fio_swap32(t.magic);
 786                 if ((t.magic & 0xffffff00) == BLK_IO_TRACE_MAGIC)
 787                         *swap = 1;
 788         }
 789
 790         if (*swap == -1) {
 791                 log_err("fio: blktrace appears corrupt\n");
 792                 return 1;
 793         }
 794
 795         return 0;
 796 }
 797
 798 int main(int argc, char *argv[])
 799 {
 800         int need_swap, i, c;
 801
 802         if (argc < 2)
 803                 return usage(argv);
 804
 805         while ((c = getopt(argc, argv, "t:n:fd:")) != -1) {
 806                 switch (c) {
 807                 case 't':
 808                         rt_threshold = atoi(optarg);
 809                         break;
 810                 case 'n':
 811                         ios_threshold = atoi(optarg);
 812                         break;
 813                 case 'f':
 814                         output_ascii = 0;
 815                         break;
 816                 case 'd':
 817                         filename = strdup(optarg);
 818                         break;
 819                 case '?':
 820                 default:
 821                         return usage(argv);
 822                 }
 823         }
 824
 825         if (argc == optind)
 826                 return usage(argv);
 827
 828         if (trace_needs_swap(argv[optind], &need_swap))
 829                 return 1;
 830
 831         for (i = 0; i < PID_HASH_SIZE; i++)
 832                 INIT_FLIST_HEAD(&pid_hash[i]);
 833         for (i = 0; i < INFLIGHT_HASH_SIZE; i++)
 834                 INIT_FLIST_HEAD(&inflight_hash[i]);
 835
 836         load_blktrace(argv[optind], need_swap);
 837         first_ttime /= 1000ULL;
 838
 839         return output_p();
 840 }