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