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 #include "util/cache.h"
  12
  13 #include "util/annotate.h"
  14 #include "util/color.h"
  15 #include <linux/list.h>
  16 #include <linux/rbtree.h>
  17 #include "util/symbol.h"
  18 #include "util/callchain.h"
  19 #include "util/strlist.h"
  20 #include "util/values.h"
  21
  22 #include "perf.h"
  23 #include "util/debug.h"
  24 #include "util/evlist.h"
  25 #include "util/evsel.h"
  26 #include "util/header.h"
  27 #include "util/session.h"
  28 #include "util/tool.h"
  29
  30 #include "util/parse-options.h"
  31 #include "util/parse-events.h"
  32
  33 #include "util/thread.h"
  34 #include "util/sort.h"
  35 #include "util/hist.h"
  36 #include "util/data.h"
  37 #include "arch/common.h"
  38
  39 #include <dlfcn.h>
  40 #include <linux/bitmap.h>
  41
  42 struct report {
  43         struct perf_tool        tool;
  44         struct perf_session     *session;
  45         bool                    force, use_tui, use_gtk, use_stdio;
  46         bool                    hide_unresolved;
  47         bool                    dont_use_callchains;
  48         bool                    show_full_info;
  49         bool                    show_threads;
  50         bool                    inverted_callchain;
  51         bool                    mem_mode;
  52         bool                    header;
  53         bool                    header_only;
  54         int                     max_stack;
  55         struct perf_read_values show_threads_values;
  56         const char              *pretty_printing_style;
  57         const char              *cpu_list;
  58         const char              *symbol_filter_str;
  59         float                   min_percent;
  60         u64                     nr_entries;
  61         u64                     queue_size;
  62         DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
  63 };
  64
  65 static int report__config(const char *var, const char *value, void *cb)
  66 {
  67         struct report *rep = cb;
  68
  69         if (!strcmp(var, "report.group")) {
  70                 symbol_conf.event_group = perf_config_bool(var, value);
  71                 return 0;
  72         }
  73         if (!strcmp(var, "report.percent-limit")) {
  74                 rep->min_percent = strtof(value, NULL);
  75                 return 0;
  76         }
  77         if (!strcmp(var, "report.children")) {
  78                 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
  79                 return 0;
  80         }
  81         if (!strcmp(var, "report.queue-size")) {
  82                 rep->queue_size = perf_config_u64(var, value);
  83                 return 0;
  84         }
  85
  86         return perf_default_config(var, value, cb);
  87 }
  88
  89 static void report__inc_stats(struct report *rep, struct hist_entry *he)
  90 {
  91         /*
  92          * The @he is either of a newly created one or an existing one
  93          * merging current sample.  We only want to count a new one so
  94          * checking ->nr_events being 1.
  95          */
  96         if (he->stat.nr_events == 1)
  97                 rep->nr_entries++;
  98 }
  99
 100 static int hist_iter__report_callback(struct hist_entry_iter *iter,
 101                                       struct addr_location *al, bool single,
 102                                       void *arg)
 103 {
 104         int err = 0;
 105         struct report *rep = arg;
 106         struct hist_entry *he = iter->he;
 107         struct perf_evsel *evsel = iter->evsel;
 108         struct mem_info *mi;
 109         struct branch_info *bi;
 110
 111         report__inc_stats(rep, he);
 112
 113         if (!ui__has_annotation())
 114                 return 0;
 115
 116         if (sort__mode == SORT_MODE__BRANCH) {
 117                 bi = he->branch_info;
 118                 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
 119                 if (err)
 120                         goto out;
 121
 122                 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
 123
 124         } else if (rep->mem_mode) {
 125                 mi = he->mem_info;
 126                 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
 127                 if (err)
 128                         goto out;
 129
 130                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
 131
 132         } else if (symbol_conf.cumulate_callchain) {
 133                 if (single)
 134                         err = hist_entry__inc_addr_samples(he, evsel->idx,
 135                                                            al->addr);
 136         } else {
 137                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
 138         }
 139
 140 out:
 141         return err;
 142 }
 143
 144 static int process_sample_event(struct perf_tool *tool,
 145                                 union perf_event *event,
 146                                 struct perf_sample *sample,
 147                                 struct perf_evsel *evsel,
 148                                 struct machine *machine)
 149 {
 150         struct report *rep = container_of(tool, struct report, tool);
 151         struct addr_location al;
 152         struct hist_entry_iter iter = {
 153                 .hide_unresolved = rep->hide_unresolved,
 154                 .add_entry_cb = hist_iter__report_callback,
 155         };
 156         int ret;
 157
 158         if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
 159                 pr_debug("problem processing %d event, skipping it.\n",
 160                          event->header.type);
 161                 return -1;
 162         }
 163
 164         if (rep->hide_unresolved && al.sym == NULL)
 165                 return 0;
 166
 167         if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
 168                 return 0;
 169
 170         if (sort__mode == SORT_MODE__BRANCH)
 171                 iter.ops = &hist_iter_branch;
 172         else if (rep->mem_mode)
 173                 iter.ops = &hist_iter_mem;
 174         else if (symbol_conf.cumulate_callchain)
 175                 iter.ops = &hist_iter_cumulative;
 176         else
 177                 iter.ops = &hist_iter_normal;
 178
 179         if (al.map != NULL)
 180                 al.map->dso->hit = 1;
 181
 182         ret = hist_entry_iter__add(&iter, &al, evsel, sample, rep->max_stack,
 183                                    rep);
 184         if (ret < 0)
 185                 pr_debug("problem adding hist entry, skipping event\n");
 186
 187         return ret;
 188 }
 189
 190 static int process_read_event(struct perf_tool *tool,
 191                               union perf_event *event,
 192                               struct perf_sample *sample __maybe_unused,
 193                               struct perf_evsel *evsel,
 194                               struct machine *machine __maybe_unused)
 195 {
 196         struct report *rep = container_of(tool, struct report, tool);
 197
 198         if (rep->show_threads) {
 199                 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
 200                 perf_read_values_add_value(&rep->show_threads_values,
 201                                            event->read.pid, event->read.tid,
 202                                            event->read.id,
 203                                            name,
 204                                            event->read.value);
 205         }
 206
 207         dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
 208                     evsel ? perf_evsel__name(evsel) : "FAIL",
 209                     event->read.value);
 210
 211         return 0;
 212 }
 213
 214 /* For pipe mode, sample_type is not currently set */
 215 static int report__setup_sample_type(struct report *rep)
 216 {
 217         struct perf_session *session = rep->session;
 218         u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
 219         bool is_pipe = perf_data_file__is_pipe(session->file);
 220
 221         if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
 222                 if (sort__has_parent) {
 223                         ui__error("Selected --sort parent, but no "
 224                                     "callchain data. Did you call "
 225                                     "'perf record' without -g?\n");
 226                         return -EINVAL;
 227                 }
 228                 if (symbol_conf.use_callchain) {
 229                         ui__error("Selected -g or --branch-history but no "
 230                                   "callchain data. Did\n"
 231                                   "you call 'perf record' without -g?\n");
 232                         return -1;
 233                 }
 234         } else if (!rep->dont_use_callchains &&
 235                    callchain_param.mode != CHAIN_NONE &&
 236                    !symbol_conf.use_callchain) {
 237                         symbol_conf.use_callchain = true;
 238                         if (callchain_register_param(&callchain_param) < 0) {
 239                                 ui__error("Can't register callchain params.\n");
 240                                 return -EINVAL;
 241                         }
 242         }
 243
 244         if (symbol_conf.cumulate_callchain) {
 245                 /* Silently ignore if callchain is missing */
 246                 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
 247                         symbol_conf.cumulate_callchain = false;
 248                         perf_hpp__cancel_cumulate();
 249                 }
 250         }
 251
 252         if (sort__mode == SORT_MODE__BRANCH) {
 253                 if (!is_pipe &&
 254                     !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
 255                         ui__error("Selected -b but no branch data. "
 256                                   "Did you call perf record without -b?\n");
 257                         return -1;
 258                 }
 259         }
 260
 261         if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
 262                 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
 263                     (sample_type & PERF_SAMPLE_STACK_USER))
 264                         callchain_param.record_mode = CALLCHAIN_DWARF;
 265                 else
 266                         callchain_param.record_mode = CALLCHAIN_FP;
 267         }
 268         return 0;
 269 }
 270
 271 static void sig_handler(int sig __maybe_unused)
 272 {
 273         session_done = 1;
 274 }
 275
 276 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
 277                                               const char *evname, FILE *fp)
 278 {
 279         size_t ret;
 280         char unit;
 281         unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
 282         u64 nr_events = hists->stats.total_period;
 283         struct perf_evsel *evsel = hists_to_evsel(hists);
 284         char buf[512];
 285         size_t size = sizeof(buf);
 286
 287         if (symbol_conf.filter_relative) {
 288                 nr_samples = hists->stats.nr_non_filtered_samples;
 289                 nr_events = hists->stats.total_non_filtered_period;
 290         }
 291
 292         if (perf_evsel__is_group_event(evsel)) {
 293                 struct perf_evsel *pos;
 294
 295                 perf_evsel__group_desc(evsel, buf, size);
 296                 evname = buf;
 297
 298                 for_each_group_member(pos, evsel) {
 299                         const struct hists *pos_hists = evsel__hists(pos);
 300
 301                         if (symbol_conf.filter_relative) {
 302                                 nr_samples += pos_hists->stats.nr_non_filtered_samples;
 303                                 nr_events += pos_hists->stats.total_non_filtered_period;
 304                         } else {
 305                                 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
 306                                 nr_events += pos_hists->stats.total_period;
 307                         }
 308                 }
 309         }
 310
 311         nr_samples = convert_unit(nr_samples, &unit);
 312         ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
 313         if (evname != NULL)
 314                 ret += fprintf(fp, " of event '%s'", evname);
 315
 316         if (rep->mem_mode) {
 317                 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
 318                 ret += fprintf(fp, "\n# Sort order   : %s", sort_order);
 319         } else
 320                 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
 321         return ret + fprintf(fp, "\n#\n");
 322 }
 323
 324 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
 325                                          struct report *rep,
 326                                          const char *help)
 327 {
 328         struct perf_evsel *pos;
 329
 330         evlist__for_each(evlist, pos) {
 331                 struct hists *hists = evsel__hists(pos);
 332                 const char *evname = perf_evsel__name(pos);
 333
 334                 if (symbol_conf.event_group &&
 335                     !perf_evsel__is_group_leader(pos))
 336                         continue;
 337
 338                 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
 339                 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
 340                 fprintf(stdout, "\n\n");
 341         }
 342
 343         if (sort_order == default_sort_order &&
 344             parent_pattern == default_parent_pattern) {
 345                 fprintf(stdout, "#\n# (%s)\n#\n", help);
 346
 347                 if (rep->show_threads) {
 348                         bool style = !strcmp(rep->pretty_printing_style, "raw");
 349                         perf_read_values_display(stdout, &rep->show_threads_values,
 350                                                  style);
 351                         perf_read_values_destroy(&rep->show_threads_values);
 352                 }
 353         }
 354
 355         return 0;
 356 }
 357
 358 static void report__warn_kptr_restrict(const struct report *rep)
 359 {
 360         struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION];
 361         struct kmap *kernel_kmap = map__kmap(kernel_map);
 362
 363         if (kernel_map == NULL ||
 364             (kernel_map->dso->hit &&
 365              (kernel_kmap->ref_reloc_sym == NULL ||
 366               kernel_kmap->ref_reloc_sym->addr == 0))) {
 367                 const char *desc =
 368                     "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
 369                     "can't be resolved.";
 370
 371                 if (kernel_map) {
 372                         const struct dso *kdso = kernel_map->dso;
 373                         if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
 374                                 desc = "If some relocation was applied (e.g. "
 375                                        "kexec) symbols may be misresolved.";
 376                         }
 377                 }
 378
 379                 ui__warning(
 380 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
 381 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
 382 "Samples in kernel modules can't be resolved as well.\n\n",
 383                 desc);
 384         }
 385 }
 386
 387 static int report__gtk_browse_hists(struct report *rep, const char *help)
 388 {
 389         int (*hist_browser)(struct perf_evlist *evlist, const char *help,
 390                             struct hist_browser_timer *timer, float min_pcnt);
 391
 392         hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
 393
 394         if (hist_browser == NULL) {
 395                 ui__error("GTK browser not found!\n");
 396                 return -1;
 397         }
 398
 399         return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
 400 }
 401
 402 static int report__browse_hists(struct report *rep)
 403 {
 404         int ret;
 405         struct perf_session *session = rep->session;
 406         struct perf_evlist *evlist = session->evlist;
 407         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
 408
 409         switch (use_browser) {
 410         case 1:
 411                 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
 412                                                     rep->min_percent,
 413                                                     &session->header.env);
 414                 /*
 415                  * Usually "ret" is the last pressed key, and we only
 416                  * care if the key notifies us to switch data file.
 417                  */
 418                 if (ret != K_SWITCH_INPUT_DATA)
 419                         ret = 0;
 420                 break;
 421         case 2:
 422                 ret = report__gtk_browse_hists(rep, help);
 423                 break;
 424         default:
 425                 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
 426                 break;
 427         }
 428
 429         return ret;
 430 }
 431
 432 static void report__collapse_hists(struct report *rep)
 433 {
 434         struct ui_progress prog;
 435         struct perf_evsel *pos;
 436
 437         ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
 438
 439         evlist__for_each(rep->session->evlist, pos) {
 440                 struct hists *hists = evsel__hists(pos);
 441
 442                 if (pos->idx == 0)
 443                         hists->symbol_filter_str = rep->symbol_filter_str;
 444
 445                 hists__collapse_resort(hists, &prog);
 446
 447                 /* Non-group events are considered as leader */
 448                 if (symbol_conf.event_group &&
 449                     !perf_evsel__is_group_leader(pos)) {
 450                         struct hists *leader_hists = evsel__hists(pos->leader);
 451
 452                         hists__match(leader_hists, hists);
 453                         hists__link(leader_hists, hists);
 454                 }
 455         }
 456
 457         ui_progress__finish();
 458 }
 459
 460 static void report__output_resort(struct report *rep)
 461 {
 462         struct ui_progress prog;
 463         struct perf_evsel *pos;
 464
 465         ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
 466
 467         evlist__for_each(rep->session->evlist, pos)
 468                 hists__output_resort(evsel__hists(pos), &prog);
 469
 470         ui_progress__finish();
 471 }
 472
 473 static int __cmd_report(struct report *rep)
 474 {
 475         int ret;
 476         struct perf_session *session = rep->session;
 477         struct perf_evsel *pos;
 478         struct perf_data_file *file = session->file;
 479
 480         signal(SIGINT, sig_handler);
 481
 482         if (rep->cpu_list) {
 483                 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
 484                                                rep->cpu_bitmap);
 485                 if (ret)
 486                         return ret;
 487         }
 488
 489         if (rep->show_threads)
 490                 perf_read_values_init(&rep->show_threads_values);
 491
 492         ret = report__setup_sample_type(rep);
 493         if (ret)
 494                 return ret;
 495
 496         ret = perf_session__process_events(session, &rep->tool);
 497         if (ret)
 498                 return ret;
 499
 500         report__warn_kptr_restrict(rep);
 501
 502         if (use_browser == 0) {
 503                 if (verbose > 3)
 504                         perf_session__fprintf(session, stdout);
 505
 506                 if (verbose > 2)
 507                         perf_session__fprintf_dsos(session, stdout);
 508
 509                 if (dump_trace) {
 510                         perf_session__fprintf_nr_events(session, stdout);
 511                         perf_evlist__fprintf_nr_events(session->evlist, stdout);
 512                         return 0;
 513                 }
 514         }
 515
 516         report__collapse_hists(rep);
 517
 518         if (session_done())
 519                 return 0;
 520
 521         /*
 522          * recalculate number of entries after collapsing since it
 523          * might be changed during the collapse phase.
 524          */
 525         rep->nr_entries = 0;
 526         evlist__for_each(session->evlist, pos)
 527                 rep->nr_entries += evsel__hists(pos)->nr_entries;
 528
 529         if (rep->nr_entries == 0) {
 530                 ui__error("The %s file has no samples!\n", file->path);
 531                 return 0;
 532         }
 533
 534         report__output_resort(rep);
 535
 536         return report__browse_hists(rep);
 537 }
 538
 539 static int
 540 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
 541 {
 542         struct report *rep = (struct report *)opt->value;
 543
 544         /*
 545          * --no-call-graph
 546          */
 547         if (unset) {
 548                 rep->dont_use_callchains = true;
 549                 return 0;
 550         }
 551
 552         return parse_callchain_report_opt(arg);
 553 }
 554
 555 int
 556 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
 557                                 const char *arg, int unset __maybe_unused)
 558 {
 559         if (arg) {
 560                 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
 561                 if (err) {
 562                         char buf[BUFSIZ];
 563                         regerror(err, &ignore_callees_regex, buf, sizeof(buf));
 564                         pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
 565                         return -1;
 566                 }
 567                 have_ignore_callees = 1;
 568         }
 569
 570         return 0;
 571 }
 572
 573 static int
 574 parse_branch_mode(const struct option *opt __maybe_unused,
 575                   const char *str __maybe_unused, int unset)
 576 {
 577         int *branch_mode = opt->value;
 578
 579         *branch_mode = !unset;
 580         return 0;
 581 }
 582
 583 static int
 584 parse_percent_limit(const struct option *opt, const char *str,
 585                     int unset __maybe_unused)
 586 {
 587         struct report *rep = opt->value;
 588
 589         rep->min_percent = strtof(str, NULL);
 590         return 0;
 591 }
 592
 593 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
 594 {
 595         struct perf_session *session;
 596         struct stat st;
 597         bool has_br_stack = false;
 598         int branch_mode = -1;
 599         bool branch_call_mode = false;
 600         char callchain_default_opt[] = "fractal,0.5,callee";
 601         const char * const report_usage[] = {
 602                 "perf report [<options>]",
 603                 NULL
 604         };
 605         struct report report = {
 606                 .tool = {
 607                         .sample          = process_sample_event,
 608                         .mmap            = perf_event__process_mmap,
 609                         .mmap2           = perf_event__process_mmap2,
 610                         .comm            = perf_event__process_comm,
 611                         .exit            = perf_event__process_exit,
 612                         .fork            = perf_event__process_fork,
 613                         .lost            = perf_event__process_lost,
 614                         .read            = process_read_event,
 615                         .attr            = perf_event__process_attr,
 616                         .tracing_data    = perf_event__process_tracing_data,
 617                         .build_id        = perf_event__process_build_id,
 618                         .ordered_events  = true,
 619                         .ordering_requires_timestamps = true,
 620                 },
 621                 .max_stack               = PERF_MAX_STACK_DEPTH,
 622                 .pretty_printing_style   = "normal",
 623         };
 624         const struct option options[] = {
 625         OPT_STRING('i', "input", &input_name, "file",
 626                     "input file name"),
 627         OPT_INCR('v', "verbose", &verbose,
 628                     "be more verbose (show symbol address, etc)"),
 629         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 630                     "dump raw trace in ASCII"),
 631         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
 632                    "file", "vmlinux pathname"),
 633         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
 634                    "file", "kallsyms pathname"),
 635         OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
 636         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
 637                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
 638         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
 639                     "Show a column with the number of samples"),
 640         OPT_BOOLEAN('T', "threads", &report.show_threads,
 641                     "Show per-thread event counters"),
 642         OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
 643                    "pretty printing style key: normal raw"),
 644         OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
 645         OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
 646         OPT_BOOLEAN(0, "stdio", &report.use_stdio,
 647                     "Use the stdio interface"),
 648         OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
 649         OPT_BOOLEAN(0, "header-only", &report.header_only,
 650                     "Show only data header."),
 651         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 652                    "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
 653                    " Please refer the man page for the complete list."),
 654         OPT_STRING('F', "fields", &field_order, "key[,keys...]",
 655                    "output field(s): overhead, period, sample plus all of sort keys"),
 656         OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
 657                     "Show sample percentage for different cpu modes"),
 658         OPT_STRING('p', "parent", &parent_pattern, "regex",
 659                    "regex filter to identify parent, see: '--sort parent'"),
 660         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
 661                     "Only display entries with parent-match"),
 662         OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order[,branch]",
 663                      "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address), add branches. "
 664                      "Default: fractal,0.5,callee,function", &report_parse_callchain_opt, callchain_default_opt),
 665         OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
 666                     "Accumulate callchains of children and show total overhead as well"),
 667         OPT_INTEGER(0, "max-stack", &report.max_stack,
 668                     "Set the maximum stack depth when parsing the callchain, "
 669                     "anything beyond the specified depth will be ignored. "
 670                     "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
 671         OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
 672                     "alias for inverted call graph"),
 673         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
 674                    "ignore callees of these functions in call graphs",
 675                    report_parse_ignore_callees_opt),
 676         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
 677                    "only consider symbols in these dsos"),
 678         OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
 679                    "only consider symbols in these comms"),
 680         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
 681                    "only consider these symbols"),
 682         OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
 683                    "only show symbols that (partially) match with this filter"),
 684         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
 685                    "width[,width...]",
 686                    "don't try to adjust column width, use these fixed values"),
 687         OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
 688                    "separator for columns, no spaces will be added between "
 689                    "columns '.' is reserved."),
 690         OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
 691                     "Only display entries resolved to a symbol"),
 692         OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
 693                     "Look for files with symbols relative to this directory"),
 694         OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
 695                    "list of cpus to profile"),
 696         OPT_BOOLEAN('I', "show-info", &report.show_full_info,
 697                     "Display extended information about perf.data file"),
 698         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
 699                     "Interleave source code with assembly code (default)"),
 700         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
 701                     "Display raw encoding of assembly instructions (default)"),
 702         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
 703                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
 704         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
 705                     "Show a column with the sum of periods"),
 706         OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
 707                     "Show event group information together"),
 708         OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
 709                     "use branch records for per branch histogram filling",
 710                     parse_branch_mode),
 711         OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
 712                     "add last branch records to call history"),
 713         OPT_STRING(0, "objdump", &objdump_path, "path",
 714                    "objdump binary to use for disassembly and annotations"),
 715         OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
 716                     "Disable symbol demangling"),
 717         OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
 718                     "Enable kernel symbol demangling"),
 719         OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
 720         OPT_CALLBACK(0, "percent-limit", &report, "percent",
 721                      "Don't show entries under that percent", parse_percent_limit),
 722         OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
 723                      "how to display percentage of filtered entries", parse_filter_percentage),
 724         OPT_END()
 725         };
 726         struct perf_data_file file = {
 727                 .mode  = PERF_DATA_MODE_READ,
 728         };
 729         int ret = hists__init();
 730
 731         if (ret < 0)
 732                 return ret;
 733
 734         perf_config(report__config, &report);
 735
 736         argc = parse_options(argc, argv, options, report_usage, 0);
 737
 738         if (report.use_stdio)
 739                 use_browser = 0;
 740         else if (report.use_tui)
 741                 use_browser = 1;
 742         else if (report.use_gtk)
 743                 use_browser = 2;
 744
 745         if (report.inverted_callchain)
 746                 callchain_param.order = ORDER_CALLER;
 747
 748         if (!input_name || !strlen(input_name)) {
 749                 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
 750                         input_name = "-";
 751                 else
 752                         input_name = "perf.data";
 753         }
 754
 755         file.path  = input_name;
 756         file.force = report.force;
 757
 758 repeat:
 759         session = perf_session__new(&file, false, &report.tool);
 760         if (session == NULL)
 761                 return -1;
 762
 763         if (report.queue_size) {
 764                 ordered_events__set_alloc_size(&session->ordered_events,
 765                                                report.queue_size);
 766         }
 767
 768         report.session = session;
 769
 770         has_br_stack = perf_header__has_feat(&session->header,
 771                                              HEADER_BRANCH_STACK);
 772
 773         /*
 774          * Branch mode is a tristate:
 775          * -1 means default, so decide based on the file having branch data.
 776          * 0/1 means the user chose a mode.
 777          */
 778         if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
 779             branch_call_mode == -1) {
 780                 sort__mode = SORT_MODE__BRANCH;
 781                 symbol_conf.cumulate_callchain = false;
 782         }
 783         if (branch_call_mode) {
 784                 callchain_param.key = CCKEY_ADDRESS;
 785                 callchain_param.branch_callstack = 1;
 786                 symbol_conf.use_callchain = true;
 787                 callchain_register_param(&callchain_param);
 788                 if (sort_order == NULL)
 789                         sort_order = "srcline,symbol,dso";
 790         }
 791
 792         if (report.mem_mode) {
 793                 if (sort__mode == SORT_MODE__BRANCH) {
 794                         pr_err("branch and mem mode incompatible\n");
 795                         goto error;
 796                 }
 797                 sort__mode = SORT_MODE__MEMORY;
 798                 symbol_conf.cumulate_callchain = false;
 799         }
 800
 801         if (setup_sorting() < 0) {
 802                 if (sort_order)
 803                         parse_options_usage(report_usage, options, "s", 1);
 804                 if (field_order)
 805                         parse_options_usage(sort_order ? NULL : report_usage,
 806                                             options, "F", 1);
 807                 goto error;
 808         }
 809
 810         /* Force tty output for header output. */
 811         if (report.header || report.header_only)
 812                 use_browser = 0;
 813
 814         if (strcmp(input_name, "-") != 0)
 815                 setup_browser(true);
 816         else
 817                 use_browser = 0;
 818
 819         if (report.header || report.header_only) {
 820                 perf_session__fprintf_info(session, stdout,
 821                                            report.show_full_info);
 822                 if (report.header_only)
 823                         return 0;
 824         } else if (use_browser == 0) {
 825                 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
 826                       stdout);
 827         }
 828
 829         /*
 830          * Only in the TUI browser we are doing integrated annotation,
 831          * so don't allocate extra space that won't be used in the stdio
 832          * implementation.
 833          */
 834         if (ui__has_annotation()) {
 835                 symbol_conf.priv_size = sizeof(struct annotation);
 836                 machines__set_symbol_filter(&session->machines,
 837                                             symbol__annotate_init);
 838                 /*
 839                  * For searching by name on the "Browse map details".
 840                  * providing it only in verbose mode not to bloat too
 841                  * much struct symbol.
 842                  */
 843                 if (verbose) {
 844                         /*
 845                          * XXX: Need to provide a less kludgy way to ask for
 846                          * more space per symbol, the u32 is for the index on
 847                          * the ui browser.
 848                          * See symbol__browser_index.
 849                          */
 850                         symbol_conf.priv_size += sizeof(u32);
 851                         symbol_conf.sort_by_name = true;
 852                 }
 853         }
 854
 855         if (symbol__init(&session->header.env) < 0)
 856                 goto error;
 857
 858         if (argc) {
 859                 /*
 860                  * Special case: if there's an argument left then assume that
 861                  * it's a symbol filter:
 862                  */
 863                 if (argc > 1)
 864                         usage_with_options(report_usage, options);
 865
 866                 report.symbol_filter_str = argv[0];
 867         }
 868
 869         sort__setup_elide(stdout);
 870
 871         ret = __cmd_report(&report);
 872         if (ret == K_SWITCH_INPUT_DATA) {
 873                 perf_session__delete(session);
 874                 goto repeat;
 875         } else
 876                 ret = 0;
 877
 878 error:
 879         perf_session__delete(session);
 880         return ret;
 881 }