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