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