tools/perf/builtin-report.c

   1 /*
   2  * builtin-report.c
   3  *
   4  * Builtin report command: Analyze the perf.data input file,
   5  * look up and read DSOs and symbol information and display
   6  * a histogram of results, along various sorting keys.
   7  */
   8 #include "builtin.h"
   9
  10 #include "util/util.h"
  11
  12 #include "util/color.h"
  13 #include "util/list.h"
  14 #include "util/cache.h"
  15 #include "util/rbtree.h"
  16 #include "util/symbol.h"
  17 #include "util/string.h"
  18
  19 #include "perf.h"
  20
  21 #include "util/parse-options.h"
  22 #include "util/parse-events.h"
  23
  24 #define SHOW_KERNEL     1
  25 #define SHOW_USER       2
  26 #define SHOW_HV         4
  27
  28 static char             const *input_name = "perf.data";
  29 static char             *vmlinux = NULL;
  30
  31 static char             default_sort_order[] = "comm,dso";
  32 static char             *sort_order = default_sort_order;
  33
  34 static int              input;
  35 static int              show_mask = SHOW_KERNEL | SHOW_USER | SHOW_HV;
  36
  37 static int              dump_trace = 0;
  38 #define dprintf(x...)   do { if (dump_trace) printf(x); } while (0)
  39 #define cdprintf(x...)  do { if (dump_trace) color_fprintf(stdout, color, x); } while (0)
  40
  41 static int              verbose;
  42 static int              full_paths;
  43
  44 static unsigned long    page_size;
  45 static unsigned long    mmap_window = 32;
  46
  47 struct ip_chain_event {
  48         __u16 nr;
  49         __u16 hv;
  50         __u16 kernel;
  51         __u16 user;
  52         __u64 ips[];
  53 };
  54
  55 struct ip_event {
  56         struct perf_event_header header;
  57         __u64 ip;
  58         __u32 pid, tid;
  59         unsigned char __more_data[];
  60 };
  61
  62 struct mmap_event {
  63         struct perf_event_header header;
  64         __u32 pid, tid;
  65         __u64 start;
  66         __u64 len;
  67         __u64 pgoff;
  68         char filename[PATH_MAX];
  69 };
  70
  71 struct comm_event {
  72         struct perf_event_header header;
  73         __u32 pid, tid;
  74         char comm[16];
  75 };
  76
  77 struct fork_event {
  78         struct perf_event_header header;
  79         __u32 pid, ppid;
  80 };
  81
  82 struct period_event {
  83         struct perf_event_header header;
  84         __u64 time;
  85         __u64 id;
  86         __u64 sample_period;
  87 };
  88
  89 typedef union event_union {
  90         struct perf_event_header        header;
  91         struct ip_event                 ip;
  92         struct mmap_event               mmap;
  93         struct comm_event               comm;
  94         struct fork_event               fork;
  95         struct period_event             period;
  96 } event_t;
  97
  98 static LIST_HEAD(dsos);
  99 static struct dso *kernel_dso;
 100 static struct dso *vdso;
 101
 102 static void dsos__add(struct dso *dso)
 103 {
 104         list_add_tail(&dso->node, &dsos);
 105 }
 106
 107 static struct dso *dsos__find(const char *name)
 108 {
 109         struct dso *pos;
 110
 111         list_for_each_entry(pos, &dsos, node)
 112                 if (strcmp(pos->name, name) == 0)
 113                         return pos;
 114         return NULL;
 115 }
 116
 117 static struct dso *dsos__findnew(const char *name)
 118 {
 119         struct dso *dso = dsos__find(name);
 120         int nr;
 121
 122         if (dso)
 123                 return dso;
 124
 125         dso = dso__new(name, 0);
 126         if (!dso)
 127                 goto out_delete_dso;
 128
 129         nr = dso__load(dso, NULL, verbose);
 130         if (nr < 0) {
 131                 if (verbose)
 132                         fprintf(stderr, "Failed to open: %s\n", name);
 133                 goto out_delete_dso;
 134         }
 135         if (!nr && verbose) {
 136                 fprintf(stderr,
 137                 "No symbols found in: %s, maybe install a debug package?\n",
 138                                 name);
 139         }
 140
 141         dsos__add(dso);
 142
 143         return dso;
 144
 145 out_delete_dso:
 146         dso__delete(dso);
 147         return NULL;
 148 }
 149
 150 static void dsos__fprintf(FILE *fp)
 151 {
 152         struct dso *pos;
 153
 154         list_for_each_entry(pos, &dsos, node)
 155                 dso__fprintf(pos, fp);
 156 }
 157
 158 static struct symbol *vdso__find_symbol(struct dso *dso, __u64 ip)
 159 {
 160         return dso__find_symbol(kernel_dso, ip);
 161 }
 162
 163 static int load_kernel(void)
 164 {
 165         int err;
 166
 167         kernel_dso = dso__new("[kernel]", 0);
 168         if (!kernel_dso)
 169                 return -1;
 170
 171         err = dso__load_kernel(kernel_dso, vmlinux, NULL, verbose);
 172         if (err) {
 173                 dso__delete(kernel_dso);
 174                 kernel_dso = NULL;
 175         } else
 176                 dsos__add(kernel_dso);
 177
 178         vdso = dso__new("[vdso]", 0);
 179         if (!vdso)
 180                 return -1;
 181
 182         vdso->find_symbol = vdso__find_symbol;
 183
 184         dsos__add(vdso);
 185
 186         return err;
 187 }
 188
 189 static char __cwd[PATH_MAX];
 190 static char *cwd = __cwd;
 191 static int cwdlen;
 192
 193 static int strcommon(const char *pathname)
 194 {
 195         int n = 0;
 196
 197         while (pathname[n] == cwd[n] && n < cwdlen)
 198                 ++n;
 199
 200         return n;
 201 }
 202
 203 struct map {
 204         struct list_head node;
 205         __u64    start;
 206         __u64    end;
 207         __u64    pgoff;
 208         __u64    (*map_ip)(struct map *, __u64);
 209         struct dso       *dso;
 210 };
 211
 212 static __u64 map__map_ip(struct map *map, __u64 ip)
 213 {
 214         return ip - map->start + map->pgoff;
 215 }
 216
 217 static __u64 vdso__map_ip(struct map *map, __u64 ip)
 218 {
 219         return ip;
 220 }
 221
 222 static inline int is_anon_memory(const char *filename)
 223 {
 224      return strcmp(filename, "//anon") == 0;
 225 }
 226
 227 static struct map *map__new(struct mmap_event *event)
 228 {
 229         struct map *self = malloc(sizeof(*self));
 230
 231         if (self != NULL) {
 232                 const char *filename = event->filename;
 233                 char newfilename[PATH_MAX];
 234                 int anon;
 235
 236                 if (cwd) {
 237                         int n = strcommon(filename);
 238
 239                         if (n == cwdlen) {
 240                                 snprintf(newfilename, sizeof(newfilename),
 241                                          ".%s", filename + n);
 242                                 filename = newfilename;
 243                         }
 244                 }
 245
 246                 anon = is_anon_memory(filename);
 247
 248                 if (anon) {
 249                         snprintf(newfilename, sizeof(newfilename), "/tmp/perf-%d.map", event->pid);
 250                         filename = newfilename;
 251                 }
 252
 253                 self->start = event->start;
 254                 self->end   = event->start + event->len;
 255                 self->pgoff = event->pgoff;
 256
 257                 self->dso = dsos__findnew(filename);
 258                 if (self->dso == NULL)
 259                         goto out_delete;
 260
 261                 if (self->dso == vdso || anon)
 262                         self->map_ip = vdso__map_ip;
 263                 else
 264                         self->map_ip = map__map_ip;
 265         }
 266         return self;
 267 out_delete:
 268         free(self);
 269         return NULL;
 270 }
 271
 272 static struct map *map__clone(struct map *self)
 273 {
 274         struct map *map = malloc(sizeof(*self));
 275
 276         if (!map)
 277                 return NULL;
 278
 279         memcpy(map, self, sizeof(*self));
 280
 281         return map;
 282 }
 283
 284 static int map__overlap(struct map *l, struct map *r)
 285 {
 286         if (l->start > r->start) {
 287                 struct map *t = l;
 288                 l = r;
 289                 r = t;
 290         }
 291
 292         if (l->end > r->start)
 293                 return 1;
 294
 295         return 0;
 296 }
 297
 298 static size_t map__fprintf(struct map *self, FILE *fp)
 299 {
 300         return fprintf(fp, " %Lx-%Lx %Lx %s\n",
 301                        self->start, self->end, self->pgoff, self->dso->name);
 302 }
 303
 304
 305 struct thread {
 306         struct rb_node   rb_node;
 307         struct list_head maps;
 308         pid_t            pid;
 309         char             *comm;
 310 };
 311
 312 static struct thread *thread__new(pid_t pid)
 313 {
 314         struct thread *self = malloc(sizeof(*self));
 315
 316         if (self != NULL) {
 317                 self->pid = pid;
 318                 self->comm = malloc(32);
 319                 if (self->comm)
 320                         snprintf(self->comm, 32, ":%d", self->pid);
 321                 INIT_LIST_HEAD(&self->maps);
 322         }
 323
 324         return self;
 325 }
 326
 327 static int thread__set_comm(struct thread *self, const char *comm)
 328 {
 329         if (self->comm)
 330                 free(self->comm);
 331         self->comm = strdup(comm);
 332         return self->comm ? 0 : -ENOMEM;
 333 }
 334
 335 static size_t thread__fprintf(struct thread *self, FILE *fp)
 336 {
 337         struct map *pos;
 338         size_t ret = fprintf(fp, "Thread %d %s\n", self->pid, self->comm);
 339
 340         list_for_each_entry(pos, &self->maps, node)
 341                 ret += map__fprintf(pos, fp);
 342
 343         return ret;
 344 }
 345
 346
 347 static struct rb_root threads;
 348 static struct thread *last_match;
 349
 350 static struct thread *threads__findnew(pid_t pid)
 351 {
 352         struct rb_node **p = &threads.rb_node;
 353         struct rb_node *parent = NULL;
 354         struct thread *th;
 355
 356         /*
 357          * Font-end cache - PID lookups come in blocks,
 358          * so most of the time we dont have to look up
 359          * the full rbtree:
 360          */
 361         if (last_match && last_match->pid == pid)
 362                 return last_match;
 363
 364         while (*p != NULL) {
 365                 parent = *p;
 366                 th = rb_entry(parent, struct thread, rb_node);
 367
 368                 if (th->pid == pid) {
 369                         last_match = th;
 370                         return th;
 371                 }
 372
 373                 if (pid < th->pid)
 374                         p = &(*p)->rb_left;
 375                 else
 376                         p = &(*p)->rb_right;
 377         }
 378
 379         th = thread__new(pid);
 380         if (th != NULL) {
 381                 rb_link_node(&th->rb_node, parent, p);
 382                 rb_insert_color(&th->rb_node, &threads);
 383                 last_match = th;
 384         }
 385
 386         return th;
 387 }
 388
 389 static void thread__insert_map(struct thread *self, struct map *map)
 390 {
 391         struct map *pos, *tmp;
 392
 393         list_for_each_entry_safe(pos, tmp, &self->maps, node) {
 394                 if (map__overlap(pos, map)) {
 395                         list_del_init(&pos->node);
 396                         /* XXX leaks dsos */
 397                         free(pos);
 398                 }
 399         }
 400
 401         list_add_tail(&map->node, &self->maps);
 402 }
 403
 404 static int thread__fork(struct thread *self, struct thread *parent)
 405 {
 406         struct map *map;
 407
 408         if (self->comm)
 409                 free(self->comm);
 410         self->comm = strdup(parent->comm);
 411         if (!self->comm)
 412                 return -ENOMEM;
 413
 414         list_for_each_entry(map, &parent->maps, node) {
 415                 struct map *new = map__clone(map);
 416                 if (!new)
 417                         return -ENOMEM;
 418                 thread__insert_map(self, new);
 419         }
 420
 421         return 0;
 422 }
 423
 424 static struct map *thread__find_map(struct thread *self, __u64 ip)
 425 {
 426         struct map *pos;
 427
 428         if (self == NULL)
 429                 return NULL;
 430
 431         list_for_each_entry(pos, &self->maps, node)
 432                 if (ip >= pos->start && ip <= pos->end)
 433                         return pos;
 434
 435         return NULL;
 436 }
 437
 438 static size_t threads__fprintf(FILE *fp)
 439 {
 440         size_t ret = 0;
 441         struct rb_node *nd;
 442
 443         for (nd = rb_first(&threads); nd; nd = rb_next(nd)) {
 444                 struct thread *pos = rb_entry(nd, struct thread, rb_node);
 445
 446                 ret += thread__fprintf(pos, fp);
 447         }
 448
 449         return ret;
 450 }
 451
 452 /*
 453  * histogram, sorted on item, collects counts
 454  */
 455
 456 static struct rb_root hist;
 457
 458 struct hist_entry {
 459         struct rb_node   rb_node;
 460
 461         struct thread    *thread;
 462         struct map       *map;
 463         struct dso       *dso;
 464         struct symbol    *sym;
 465         __u64            ip;
 466         char             level;
 467
 468         __u64            count;
 469 };
 470
 471 /*
 472  * configurable sorting bits
 473  */
 474
 475 struct sort_entry {
 476         struct list_head list;
 477
 478         char *header;
 479
 480         int64_t (*cmp)(struct hist_entry *, struct hist_entry *);
 481         int64_t (*collapse)(struct hist_entry *, struct hist_entry *);
 482         size_t  (*print)(FILE *fp, struct hist_entry *);
 483 };
 484
 485 /* --sort pid */
 486
 487 static int64_t
 488 sort__thread_cmp(struct hist_entry *left, struct hist_entry *right)
 489 {
 490         return right->thread->pid - left->thread->pid;
 491 }
 492
 493 static size_t
 494 sort__thread_print(FILE *fp, struct hist_entry *self)
 495 {
 496         return fprintf(fp, "%16s:%5d", self->thread->comm ?: "", self->thread->pid);
 497 }
 498
 499 static struct sort_entry sort_thread = {
 500         .header = "         Command:  Pid",
 501         .cmp    = sort__thread_cmp,
 502         .print  = sort__thread_print,
 503 };
 504
 505 /* --sort comm */
 506
 507 static int64_t
 508 sort__comm_cmp(struct hist_entry *left, struct hist_entry *right)
 509 {
 510         return right->thread->pid - left->thread->pid;
 511 }
 512
 513 static int64_t
 514 sort__comm_collapse(struct hist_entry *left, struct hist_entry *right)
 515 {
 516         char *comm_l = left->thread->comm;
 517         char *comm_r = right->thread->comm;
 518
 519         if (!comm_l || !comm_r) {
 520                 if (!comm_l && !comm_r)
 521                         return 0;
 522                 else if (!comm_l)
 523                         return -1;
 524                 else
 525                         return 1;
 526         }
 527
 528         return strcmp(comm_l, comm_r);
 529 }
 530
 531 static size_t
 532 sort__comm_print(FILE *fp, struct hist_entry *self)
 533 {
 534         return fprintf(fp, "%16s", self->thread->comm);
 535 }
 536
 537 static struct sort_entry sort_comm = {
 538         .header         = "         Command",
 539         .cmp            = sort__comm_cmp,
 540         .collapse       = sort__comm_collapse,
 541         .print          = sort__comm_print,
 542 };
 543
 544 /* --sort dso */
 545
 546 static int64_t
 547 sort__dso_cmp(struct hist_entry *left, struct hist_entry *right)
 548 {
 549         struct dso *dso_l = left->dso;
 550         struct dso *dso_r = right->dso;
 551
 552         if (!dso_l || !dso_r) {
 553                 if (!dso_l && !dso_r)
 554                         return 0;
 555                 else if (!dso_l)
 556                         return -1;
 557                 else
 558                         return 1;
 559         }
 560
 561         return strcmp(dso_l->name, dso_r->name);
 562 }
 563
 564 static size_t
 565 sort__dso_print(FILE *fp, struct hist_entry *self)
 566 {
 567         if (self->dso)
 568                 return fprintf(fp, "%-25s", self->dso->name);
 569
 570         return fprintf(fp, "%016llx         ", (__u64)self->ip);
 571 }
 572
 573 static struct sort_entry sort_dso = {
 574         .header = "Shared Object            ",
 575         .cmp    = sort__dso_cmp,
 576         .print  = sort__dso_print,
 577 };
 578
 579 /* --sort symbol */
 580
 581 static int64_t
 582 sort__sym_cmp(struct hist_entry *left, struct hist_entry *right)
 583 {
 584         __u64 ip_l, ip_r;
 585
 586         if (left->sym == right->sym)
 587                 return 0;
 588
 589         ip_l = left->sym ? left->sym->start : left->ip;
 590         ip_r = right->sym ? right->sym->start : right->ip;
 591
 592         return (int64_t)(ip_r - ip_l);
 593 }
 594
 595 static size_t
 596 sort__sym_print(FILE *fp, struct hist_entry *self)
 597 {
 598         size_t ret = 0;
 599
 600         if (verbose)
 601                 ret += fprintf(fp, "%#018llx  ", (__u64)self->ip);
 602
 603         if (self->sym) {
 604                 ret += fprintf(fp, "[%c] %s",
 605                         self->dso == kernel_dso ? 'k' : '.', self->sym->name);
 606         } else {
 607                 ret += fprintf(fp, "%#016llx", (__u64)self->ip);
 608         }
 609
 610         return ret;
 611 }
 612
 613 static struct sort_entry sort_sym = {
 614         .header = "Symbol",
 615         .cmp    = sort__sym_cmp,
 616         .print  = sort__sym_print,
 617 };
 618
 619 static int sort__need_collapse = 0;
 620
 621 struct sort_dimension {
 622         char                    *name;
 623         struct sort_entry       *entry;
 624         int                     taken;
 625 };
 626
 627 static struct sort_dimension sort_dimensions[] = {
 628         { .name = "pid",        .entry = &sort_thread,  },
 629         { .name = "comm",       .entry = &sort_comm,    },
 630         { .name = "dso",        .entry = &sort_dso,     },
 631         { .name = "symbol",     .entry = &sort_sym,     },
 632 };
 633
 634 static LIST_HEAD(hist_entry__sort_list);
 635
 636 static int sort_dimension__add(char *tok)
 637 {
 638         int i;
 639
 640         for (i = 0; i < ARRAY_SIZE(sort_dimensions); i++) {
 641                 struct sort_dimension *sd = &sort_dimensions[i];
 642
 643                 if (sd->taken)
 644                         continue;
 645
 646                 if (strncasecmp(tok, sd->name, strlen(tok)))
 647                         continue;
 648
 649                 if (sd->entry->collapse)
 650                         sort__need_collapse = 1;
 651
 652                 list_add_tail(&sd->entry->list, &hist_entry__sort_list);
 653                 sd->taken = 1;
 654
 655                 return 0;
 656         }
 657
 658         return -ESRCH;
 659 }
 660
 661 static int64_t
 662 hist_entry__cmp(struct hist_entry *left, struct hist_entry *right)
 663 {
 664         struct sort_entry *se;
 665         int64_t cmp = 0;
 666
 667         list_for_each_entry(se, &hist_entry__sort_list, list) {
 668                 cmp = se->cmp(left, right);
 669                 if (cmp)
 670                         break;
 671         }
 672
 673         return cmp;
 674 }
 675
 676 static int64_t
 677 hist_entry__collapse(struct hist_entry *left, struct hist_entry *right)
 678 {
 679         struct sort_entry *se;
 680         int64_t cmp = 0;
 681
 682         list_for_each_entry(se, &hist_entry__sort_list, list) {
 683                 int64_t (*f)(struct hist_entry *, struct hist_entry *);
 684
 685                 f = se->collapse ?: se->cmp;
 686
 687                 cmp = f(left, right);
 688                 if (cmp)
 689                         break;
 690         }
 691
 692         return cmp;
 693 }
 694
 695 static size_t
 696 hist_entry__fprintf(FILE *fp, struct hist_entry *self, __u64 total_samples)
 697 {
 698         struct sort_entry *se;
 699         size_t ret;
 700
 701         if (total_samples) {
 702                 double percent = self->count * 100.0 / total_samples;
 703                 char *color = PERF_COLOR_NORMAL;
 704
 705                 /*
 706                  * We color high-overhead entries in red, mid-overhead
 707                  * entries in green - and keep the low overhead places
 708                  * normal:
 709                  */
 710                 if (percent >= 5.0) {
 711                         color = PERF_COLOR_RED;
 712                 } else {
 713                         if (percent >= 0.5)
 714                                 color = PERF_COLOR_GREEN;
 715                 }
 716
 717                 ret = color_fprintf(fp, color, "   %6.2f%%",
 718                                 (self->count * 100.0) / total_samples);
 719         } else
 720                 ret = fprintf(fp, "%12Ld ", self->count);
 721
 722         list_for_each_entry(se, &hist_entry__sort_list, list) {
 723                 fprintf(fp, "  ");
 724                 ret += se->print(fp, self);
 725         }
 726
 727         ret += fprintf(fp, "\n");
 728
 729         return ret;
 730 }
 731
 732 /*
 733  * collect histogram counts
 734  */
 735
 736 static int
 737 hist_entry__add(struct thread *thread, struct map *map, struct dso *dso,
 738                 struct symbol *sym, __u64 ip, char level, __u64 count)
 739 {
 740         struct rb_node **p = &hist.rb_node;
 741         struct rb_node *parent = NULL;
 742         struct hist_entry *he;
 743         struct hist_entry entry = {
 744                 .thread = thread,
 745                 .map    = map,
 746                 .dso    = dso,
 747                 .sym    = sym,
 748                 .ip     = ip,
 749                 .level  = level,
 750                 .count  = count,
 751         };
 752         int cmp;
 753
 754         while (*p != NULL) {
 755                 parent = *p;
 756                 he = rb_entry(parent, struct hist_entry, rb_node);
 757
 758                 cmp = hist_entry__cmp(&entry, he);
 759
 760                 if (!cmp) {
 761                         he->count += count;
 762                         return 0;
 763                 }
 764
 765                 if (cmp < 0)
 766                         p = &(*p)->rb_left;
 767                 else
 768                         p = &(*p)->rb_right;
 769         }
 770
 771         he = malloc(sizeof(*he));
 772         if (!he)
 773                 return -ENOMEM;
 774         *he = entry;
 775         rb_link_node(&he->rb_node, parent, p);
 776         rb_insert_color(&he->rb_node, &hist);
 777
 778         return 0;
 779 }
 780
 781 static void hist_entry__free(struct hist_entry *he)
 782 {
 783         free(he);
 784 }
 785
 786 /*
 787  * collapse the histogram
 788  */
 789
 790 static struct rb_root collapse_hists;
 791
 792 static void collapse__insert_entry(struct hist_entry *he)
 793 {
 794         struct rb_node **p = &collapse_hists.rb_node;
 795         struct rb_node *parent = NULL;
 796         struct hist_entry *iter;
 797         int64_t cmp;
 798
 799         while (*p != NULL) {
 800                 parent = *p;
 801                 iter = rb_entry(parent, struct hist_entry, rb_node);
 802
 803                 cmp = hist_entry__collapse(iter, he);
 804
 805                 if (!cmp) {
 806                         iter->count += he->count;
 807                         hist_entry__free(he);
 808                         return;
 809                 }
 810
 811                 if (cmp < 0)
 812                         p = &(*p)->rb_left;
 813                 else
 814                         p = &(*p)->rb_right;
 815         }
 816
 817         rb_link_node(&he->rb_node, parent, p);
 818         rb_insert_color(&he->rb_node, &collapse_hists);
 819 }
 820
 821 static void collapse__resort(void)
 822 {
 823         struct rb_node *next;
 824         struct hist_entry *n;
 825
 826         if (!sort__need_collapse)
 827                 return;
 828
 829         next = rb_first(&hist);
 830         while (next) {
 831                 n = rb_entry(next, struct hist_entry, rb_node);
 832                 next = rb_next(&n->rb_node);
 833
 834                 rb_erase(&n->rb_node, &hist);
 835                 collapse__insert_entry(n);
 836         }
 837 }
 838
 839 /*
 840  * reverse the map, sort on count.
 841  */
 842
 843 static struct rb_root output_hists;
 844
 845 static void output__insert_entry(struct hist_entry *he)
 846 {
 847         struct rb_node **p = &output_hists.rb_node;
 848         struct rb_node *parent = NULL;
 849         struct hist_entry *iter;
 850
 851         while (*p != NULL) {
 852                 parent = *p;
 853                 iter = rb_entry(parent, struct hist_entry, rb_node);
 854
 855                 if (he->count > iter->count)
 856                         p = &(*p)->rb_left;
 857                 else
 858                         p = &(*p)->rb_right;
 859         }
 860
 861         rb_link_node(&he->rb_node, parent, p);
 862         rb_insert_color(&he->rb_node, &output_hists);
 863 }
 864
 865 static void output__resort(void)
 866 {
 867         struct rb_node *next;
 868         struct hist_entry *n;
 869         struct rb_root *tree = &hist;
 870
 871         if (sort__need_collapse)
 872                 tree = &collapse_hists;
 873
 874         next = rb_first(tree);
 875
 876         while (next) {
 877                 n = rb_entry(next, struct hist_entry, rb_node);
 878                 next = rb_next(&n->rb_node);
 879
 880                 rb_erase(&n->rb_node, tree);
 881                 output__insert_entry(n);
 882         }
 883 }
 884
 885 static size_t output__fprintf(FILE *fp, __u64 total_samples)
 886 {
 887         struct hist_entry *pos;
 888         struct sort_entry *se;
 889         struct rb_node *nd;
 890         size_t ret = 0;
 891
 892         fprintf(fp, "\n");
 893         fprintf(fp, "#\n");
 894         fprintf(fp, "# (%Ld samples)\n", (__u64)total_samples);
 895         fprintf(fp, "#\n");
 896
 897         fprintf(fp, "# Overhead");
 898         list_for_each_entry(se, &hist_entry__sort_list, list)
 899                 fprintf(fp, "  %s", se->header);
 900         fprintf(fp, "\n");
 901
 902         fprintf(fp, "# ........");
 903         list_for_each_entry(se, &hist_entry__sort_list, list) {
 904                 int i;
 905
 906                 fprintf(fp, "  ");
 907                 for (i = 0; i < strlen(se->header); i++)
 908                         fprintf(fp, ".");
 909         }
 910         fprintf(fp, "\n");
 911
 912         fprintf(fp, "#\n");
 913
 914         for (nd = rb_first(&output_hists); nd; nd = rb_next(nd)) {
 915                 pos = rb_entry(nd, struct hist_entry, rb_node);
 916                 ret += hist_entry__fprintf(fp, pos, total_samples);
 917         }
 918
 919         if (!strcmp(sort_order, default_sort_order)) {
 920                 fprintf(fp, "#\n");
 921                 fprintf(fp, "# (For more details, try: perf report --sort comm,dso,symbol)\n");
 922                 fprintf(fp, "#\n");
 923         }
 924         fprintf(fp, "\n");
 925
 926         return ret;
 927 }
 928
 929 static void register_idle_thread(void)
 930 {
 931         struct thread *thread = threads__findnew(0);
 932
 933         if (thread == NULL ||
 934                         thread__set_comm(thread, "[idle]")) {
 935                 fprintf(stderr, "problem inserting idle task.\n");
 936                 exit(-1);
 937         }
 938 }
 939
 940 static unsigned long total = 0,
 941                      total_mmap = 0,
 942                      total_comm = 0,
 943                      total_fork = 0,
 944                      total_unknown = 0;
 945
 946 static int
 947 process_overflow_event(event_t *event, unsigned long offset, unsigned long head)
 948 {
 949         char level;
 950         int show = 0;
 951         struct dso *dso = NULL;
 952         struct thread *thread = threads__findnew(event->ip.pid);
 953         __u64 ip = event->ip.ip;
 954         __u64 period = 1;
 955         struct map *map = NULL;
 956         void *more_data = event->ip.__more_data;
 957         struct ip_chain_event *chain;
 958
 959         if (event->header.type & PERF_SAMPLE_PERIOD) {
 960                 period = *(__u64 *)more_data;
 961                 more_data += sizeof(__u64);
 962         }
 963
 964         dprintf("%p [%p]: PERF_EVENT (IP, %d): %d: %p period: %Ld\n",
 965                 (void *)(offset + head),
 966                 (void *)(long)(event->header.size),
 967                 event->header.misc,
 968                 event->ip.pid,
 969                 (void *)(long)ip,
 970                 (long long)period);
 971
 972         if (event->header.type & PERF_SAMPLE_CALLCHAIN) {
 973                 int i;
 974
 975                 chain = (void *)more_data;
 976
 977                 if (dump_trace) {
 978                         dprintf("... chain: u:%d, k:%d, nr:%d\n",
 979                                 chain->user,
 980                                 chain->kernel,
 981                                 chain->nr);
 982
 983                         for (i = 0; i < chain->nr; i++)
 984                                 dprintf("..... %2d: %p\n", i, (void *)chain->ips[i]);
 985                 }
 986         }
 987
 988         dprintf(" ... thread: %s:%d\n", thread->comm, thread->pid);
 989
 990         if (thread == NULL) {
 991                 fprintf(stderr, "problem processing %d event, skipping it.\n",
 992                         event->header.type);
 993                 return -1;
 994         }
 995
 996         if (event->header.misc & PERF_EVENT_MISC_KERNEL) {
 997                 show = SHOW_KERNEL;
 998                 level = 'k';
 999
1000                 dso = kernel_dso;
1001
1002                 dprintf(" ...... dso: %s\n", dso->name);
1003
1004         } else if (event->header.misc & PERF_EVENT_MISC_USER) {
1005
1006                 show = SHOW_USER;
1007                 level = '.';
1008
1009                 map = thread__find_map(thread, ip);
1010                 if (map != NULL) {
1011                         ip = map->map_ip(map, ip);
1012                         dso = map->dso;
1013                 } else {
1014                         /*
1015                          * If this is outside of all known maps,
1016                          * and is a negative address, try to look it
1017                          * up in the kernel dso, as it might be a
1018                          * vsyscall (which executes in user-mode):
1019                          */
1020                         if ((long long)ip < 0)
1021                                 dso = kernel_dso;
1022                 }
1023                 dprintf(" ...... dso: %s\n", dso ? dso->name : "<not found>");
1024
1025         } else {
1026                 show = SHOW_HV;
1027                 level = 'H';
1028                 dprintf(" ...... dso: [hypervisor]\n");
1029         }
1030
1031         if (show & show_mask) {
1032                 struct symbol *sym = NULL;
1033
1034                 if (dso)
1035                         sym = dso->find_symbol(dso, ip);
1036
1037                 if (hist_entry__add(thread, map, dso, sym, ip, level, period)) {
1038                         fprintf(stderr,
1039                 "problem incrementing symbol count, skipping event\n");
1040                         return -1;
1041                 }
1042         }
1043         total += period;
1044
1045         return 0;
1046 }
1047
1048 static int
1049 process_mmap_event(event_t *event, unsigned long offset, unsigned long head)
1050 {
1051         struct thread *thread = threads__findnew(event->mmap.pid);
1052         struct map *map = map__new(&event->mmap);
1053
1054         dprintf("%p [%p]: PERF_EVENT_MMAP %d: [%p(%p) @ %p]: %s\n",
1055                 (void *)(offset + head),
1056                 (void *)(long)(event->header.size),
1057                 event->mmap.pid,
1058                 (void *)(long)event->mmap.start,
1059                 (void *)(long)event->mmap.len,
1060                 (void *)(long)event->mmap.pgoff,
1061                 event->mmap.filename);
1062
1063         if (thread == NULL || map == NULL) {
1064                 dprintf("problem processing PERF_EVENT_MMAP, skipping event.\n");
1065                 return 0;
1066         }
1067
1068         thread__insert_map(thread, map);
1069         total_mmap++;
1070
1071         return 0;
1072 }
1073
1074 static int
1075 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
1076 {
1077         struct thread *thread = threads__findnew(event->comm.pid);
1078
1079         dprintf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
1080                 (void *)(offset + head),
1081                 (void *)(long)(event->header.size),
1082                 event->comm.comm, event->comm.pid);
1083
1084         if (thread == NULL ||
1085             thread__set_comm(thread, event->comm.comm)) {
1086                 dprintf("problem processing PERF_EVENT_COMM, skipping event.\n");
1087                 return -1;
1088         }
1089         total_comm++;
1090
1091         return 0;
1092 }
1093
1094 static int
1095 process_fork_event(event_t *event, unsigned long offset, unsigned long head)
1096 {
1097         struct thread *thread = threads__findnew(event->fork.pid);
1098         struct thread *parent = threads__findnew(event->fork.ppid);
1099
1100         dprintf("%p [%p]: PERF_EVENT_FORK: %d:%d\n",
1101                 (void *)(offset + head),
1102                 (void *)(long)(event->header.size),
1103                 event->fork.pid, event->fork.ppid);
1104
1105         if (!thread || !parent || thread__fork(thread, parent)) {
1106                 dprintf("problem processing PERF_EVENT_FORK, skipping event.\n");
1107                 return -1;
1108         }
1109         total_fork++;
1110
1111         return 0;
1112 }
1113
1114 static int
1115 process_period_event(event_t *event, unsigned long offset, unsigned long head)
1116 {
1117         dprintf("%p [%p]: PERF_EVENT_PERIOD: time:%Ld, id:%Ld: period:%Ld\n",
1118                 (void *)(offset + head),
1119                 (void *)(long)(event->header.size),
1120                 event->period.time,
1121                 event->period.id,
1122                 event->period.sample_period);
1123
1124         return 0;
1125 }
1126
1127 static void trace_event(event_t *event)
1128 {
1129         unsigned char *raw_event = (void *)event;
1130         char *color = PERF_COLOR_BLUE;
1131         int i, j;
1132
1133         if (!dump_trace)
1134                 return;
1135
1136         dprintf(".");
1137         cdprintf("\n. ... raw event: size %d bytes\n", event->header.size);
1138
1139         for (i = 0; i < event->header.size; i++) {
1140                 if ((i & 15) == 0) {
1141                         dprintf(".");
1142                         cdprintf("  %04x: ", i);
1143                 }
1144
1145                 cdprintf(" %02x", raw_event[i]);
1146
1147                 if (((i & 15) == 15) || i == event->header.size-1) {
1148                         cdprintf("  ");
1149                         for (j = 0; j < 15-(i & 15); j++)
1150                                 cdprintf("   ");
1151                         for (j = 0; j < (i & 15); j++) {
1152                                 if (isprint(raw_event[i-15+j]))
1153                                         cdprintf("%c", raw_event[i-15+j]);
1154                                 else
1155                                         cdprintf(".");
1156                         }
1157                         cdprintf("\n");
1158                 }
1159         }
1160         dprintf(".\n");
1161 }
1162
1163 static int
1164 process_event(event_t *event, unsigned long offset, unsigned long head)
1165 {
1166         trace_event(event);
1167
1168         if (event->header.misc & PERF_EVENT_MISC_OVERFLOW)
1169                 return process_overflow_event(event, offset, head);
1170
1171         switch (event->header.type) {
1172         case PERF_EVENT_MMAP:
1173                 return process_mmap_event(event, offset, head);
1174
1175         case PERF_EVENT_COMM:
1176                 return process_comm_event(event, offset, head);
1177
1178         case PERF_EVENT_FORK:
1179                 return process_fork_event(event, offset, head);
1180
1181         case PERF_EVENT_PERIOD:
1182                 return process_period_event(event, offset, head);
1183         /*
1184          * We dont process them right now but they are fine:
1185          */
1186
1187         case PERF_EVENT_THROTTLE:
1188         case PERF_EVENT_UNTHROTTLE:
1189                 return 0;
1190
1191         default:
1192                 return -1;
1193         }
1194
1195         return 0;
1196 }
1197
1198 static int __cmd_report(void)
1199 {
1200         int ret, rc = EXIT_FAILURE;
1201         unsigned long offset = 0;
1202         unsigned long head = 0;
1203         struct stat stat;
1204         event_t *event;
1205         uint32_t size;
1206         char *buf;
1207
1208         register_idle_thread();
1209
1210         input = open(input_name, O_RDONLY);
1211         if (input < 0) {
1212                 fprintf(stderr, " failed to open file: %s", input_name);
1213                 if (!strcmp(input_name, "perf.data"))
1214                         fprintf(stderr, "  (try 'perf record' first)");
1215                 fprintf(stderr, "\n");
1216                 exit(-1);
1217         }
1218
1219         ret = fstat(input, &stat);
1220         if (ret < 0) {
1221                 perror("failed to stat file");
1222                 exit(-1);
1223         }
1224
1225         if (!stat.st_size) {
1226                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1227                 exit(0);
1228         }
1229
1230         if (load_kernel() < 0) {
1231                 perror("failed to load kernel symbols");
1232                 return EXIT_FAILURE;
1233         }
1234
1235         if (!full_paths) {
1236                 if (getcwd(__cwd, sizeof(__cwd)) == NULL) {
1237                         perror("failed to get the current directory");
1238                         return EXIT_FAILURE;
1239                 }
1240                 cwdlen = strlen(cwd);
1241         } else {
1242                 cwd = NULL;
1243                 cwdlen = 0;
1244         }
1245 remap:
1246         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1247                            MAP_SHARED, input, offset);
1248         if (buf == MAP_FAILED) {
1249                 perror("failed to mmap file");
1250                 exit(-1);
1251         }
1252
1253 more:
1254         event = (event_t *)(buf + head);
1255
1256         size = event->header.size;
1257         if (!size)
1258                 size = 8;
1259
1260         if (head + event->header.size >= page_size * mmap_window) {
1261                 unsigned long shift = page_size * (head / page_size);
1262                 int ret;
1263
1264                 ret = munmap(buf, page_size * mmap_window);
1265                 assert(ret == 0);
1266
1267                 offset += shift;
1268                 head -= shift;
1269                 goto remap;
1270         }
1271
1272         size = event->header.size;
1273
1274         dprintf("\n%p [%p]: event: %d\n",
1275                         (void *)(offset + head),
1276                         (void *)(long)event->header.size,
1277                         event->header.type);
1278
1279         if (!size || process_event(event, offset, head) < 0) {
1280
1281                 dprintf("%p [%p]: skipping unknown header type: %d\n",
1282                         (void *)(offset + head),
1283                         (void *)(long)(event->header.size),
1284                         event->header.type);
1285
1286                 total_unknown++;
1287
1288                 /*
1289                  * assume we lost track of the stream, check alignment, and
1290                  * increment a single u64 in the hope to catch on again 'soon'.
1291                  */
1292
1293                 if (unlikely(head & 7))
1294                         head &= ~7ULL;
1295
1296                 size = 8;
1297         }
1298
1299         head += size;
1300
1301         if (offset + head < stat.st_size)
1302                 goto more;
1303
1304         rc = EXIT_SUCCESS;
1305         close(input);
1306
1307         dprintf("      IP events: %10ld\n", total);
1308         dprintf("    mmap events: %10ld\n", total_mmap);
1309         dprintf("    comm events: %10ld\n", total_comm);
1310         dprintf("    fork events: %10ld\n", total_fork);
1311         dprintf(" unknown events: %10ld\n", total_unknown);
1312
1313         if (dump_trace)
1314                 return 0;
1315
1316         if (verbose >= 3)
1317                 threads__fprintf(stdout);
1318
1319         if (verbose >= 2)
1320                 dsos__fprintf(stdout);
1321
1322         collapse__resort();
1323         output__resort();
1324         output__fprintf(stdout, total);
1325
1326         return rc;
1327 }
1328
1329 static const char * const report_usage[] = {
1330         "perf report [<options>] <command>",
1331         NULL
1332 };
1333
1334 static const struct option options[] = {
1335         OPT_STRING('i', "input", &input_name, "file",
1336                     "input file name"),
1337         OPT_BOOLEAN('v', "verbose", &verbose,
1338                     "be more verbose (show symbol address, etc)"),
1339         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1340                     "dump raw trace in ASCII"),
1341         OPT_STRING('k', "vmlinux", &vmlinux, "file", "vmlinux pathname"),
1342         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1343                    "sort by key(s): pid, comm, dso, symbol. Default: pid,symbol"),
1344         OPT_BOOLEAN('P', "full-paths", &full_paths,
1345                     "Don't shorten the pathnames taking into account the cwd"),
1346         OPT_END()
1347 };
1348
1349 static void setup_sorting(void)
1350 {
1351         char *tmp, *tok, *str = strdup(sort_order);
1352
1353         for (tok = strtok_r(str, ", ", &tmp);
1354                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1355                 if (sort_dimension__add(tok) < 0) {
1356                         error("Unknown --sort key: `%s'", tok);
1357                         usage_with_options(report_usage, options);
1358                 }
1359         }
1360
1361         free(str);
1362 }
1363
1364 int cmd_report(int argc, const char **argv, const char *prefix)
1365 {
1366         symbol__init();
1367
1368         page_size = getpagesize();
1369
1370         argc = parse_options(argc, argv, options, report_usage, 0);
1371
1372         setup_sorting();
1373
1374         /*
1375          * Any (unrecognized) arguments left?
1376          */
1377         if (argc)
1378                 usage_with_options(report_usage, options);
1379
1380         setup_pager();
1381
1382         return __cmd_report();
1383 }