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 <subcmd/parse-options.h> 31 #include <subcmd/exec-cmd.h> 32 #include "util/parse-events.h" 33 34 #include "util/thread.h" 35 #include "util/sort.h" 36 #include "util/hist.h" 37 #include "util/data.h" 38 #include "arch/common.h" 39 40 #include "util/auxtrace.h" 41 42 #include <dlfcn.h> 43 #include <linux/bitmap.h> 44 45 struct report { 46 struct perf_tool tool; 47 struct perf_session *session; 48 bool use_tui, use_gtk, use_stdio; 49 bool dont_use_callchains; 50 bool show_full_info; 51 bool show_threads; 52 bool inverted_callchain; 53 bool mem_mode; 54 bool header; 55 bool header_only; 56 bool nonany_branch_mode; 57 int max_stack; 58 struct perf_read_values show_threads_values; 59 const char *pretty_printing_style; 60 const char *cpu_list; 61 const char *symbol_filter_str; 62 float min_percent; 63 u64 nr_entries; 64 u64 queue_size; 65 int socket_filter; 66 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 67 }; 68 69 static int report__config(const char *var, const char *value, void *cb) 70 { 71 struct report *rep = cb; 72 73 if (!strcmp(var, "report.group")) { 74 symbol_conf.event_group = perf_config_bool(var, value); 75 return 0; 76 } 77 if (!strcmp(var, "report.percent-limit")) { 78 double pcnt = strtof(value, NULL); 79 80 rep->min_percent = pcnt; 81 callchain_param.min_percent = pcnt; 82 return 0; 83 } 84 if (!strcmp(var, "report.children")) { 85 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 86 return 0; 87 } 88 if (!strcmp(var, "report.queue-size")) { 89 rep->queue_size = perf_config_u64(var, value); 90 return 0; 91 } 92 93 return perf_default_config(var, value, cb); 94 } 95 96 static int hist_iter__report_callback(struct hist_entry_iter *iter, 97 struct addr_location *al, bool single, 98 void *arg) 99 { 100 int err = 0; 101 struct report *rep = arg; 102 struct hist_entry *he = iter->he; 103 struct perf_evsel *evsel = iter->evsel; 104 struct mem_info *mi; 105 struct branch_info *bi; 106 107 if (!ui__has_annotation()) 108 return 0; 109 110 hist__account_cycles(iter->sample->branch_stack, al, iter->sample, 111 rep->nonany_branch_mode); 112 113 if (sort__mode == SORT_MODE__BRANCH) { 114 bi = he->branch_info; 115 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx); 116 if (err) 117 goto out; 118 119 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx); 120 121 } else if (rep->mem_mode) { 122 mi = he->mem_info; 123 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx); 124 if (err) 125 goto out; 126 127 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 128 129 } else if (symbol_conf.cumulate_callchain) { 130 if (single) 131 err = hist_entry__inc_addr_samples(he, evsel->idx, 132 al->addr); 133 } else { 134 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 135 } 136 137 out: 138 return err; 139 } 140 141 static int process_sample_event(struct perf_tool *tool, 142 union perf_event *event, 143 struct perf_sample *sample, 144 struct perf_evsel *evsel, 145 struct machine *machine) 146 { 147 struct report *rep = container_of(tool, struct report, tool); 148 struct addr_location al; 149 struct hist_entry_iter iter = { 150 .evsel = evsel, 151 .sample = sample, 152 .hide_unresolved = symbol_conf.hide_unresolved, 153 .add_entry_cb = hist_iter__report_callback, 154 }; 155 int ret = 0; 156 157 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) { 158 pr_debug("problem processing %d event, skipping it.\n", 159 event->header.type); 160 return -1; 161 } 162 163 if (symbol_conf.hide_unresolved && al.sym == NULL) 164 goto out_put; 165 166 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 167 goto out_put; 168 169 if (sort__mode == SORT_MODE__BRANCH) { 170 /* 171 * A non-synthesized event might not have a branch stack if 172 * branch stacks have been synthesized (using itrace options). 173 */ 174 if (!sample->branch_stack) 175 goto out_put; 176 iter.ops = &hist_iter_branch; 177 } else if (rep->mem_mode) { 178 iter.ops = &hist_iter_mem; 179 } else if (symbol_conf.cumulate_callchain) { 180 iter.ops = &hist_iter_cumulative; 181 } else { 182 iter.ops = &hist_iter_normal; 183 } 184 185 if (al.map != NULL) 186 al.map->dso->hit = 1; 187 188 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 189 if (ret < 0) 190 pr_debug("problem adding hist entry, skipping event\n"); 191 out_put: 192 addr_location__put(&al); 193 return ret; 194 } 195 196 static int process_read_event(struct perf_tool *tool, 197 union perf_event *event, 198 struct perf_sample *sample __maybe_unused, 199 struct perf_evsel *evsel, 200 struct machine *machine __maybe_unused) 201 { 202 struct report *rep = container_of(tool, struct report, tool); 203 204 if (rep->show_threads) { 205 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 206 perf_read_values_add_value(&rep->show_threads_values, 207 event->read.pid, event->read.tid, 208 event->read.id, 209 name, 210 event->read.value); 211 } 212 213 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 214 evsel ? perf_evsel__name(evsel) : "FAIL", 215 event->read.value); 216 217 return 0; 218 } 219 220 /* For pipe mode, sample_type is not currently set */ 221 static int report__setup_sample_type(struct report *rep) 222 { 223 struct perf_session *session = rep->session; 224 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 225 bool is_pipe = perf_data_file__is_pipe(session->file); 226 227 if (session->itrace_synth_opts->callchain || 228 (!is_pipe && 229 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 230 !session->itrace_synth_opts->set)) 231 sample_type |= PERF_SAMPLE_CALLCHAIN; 232 233 if (session->itrace_synth_opts->last_branch) 234 sample_type |= PERF_SAMPLE_BRANCH_STACK; 235 236 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 237 if (sort__has_parent) { 238 ui__error("Selected --sort parent, but no " 239 "callchain data. Did you call " 240 "'perf record' without -g?\n"); 241 return -EINVAL; 242 } 243 if (symbol_conf.use_callchain) { 244 ui__error("Selected -g or --branch-history but no " 245 "callchain data. Did\n" 246 "you call 'perf record' without -g?\n"); 247 return -1; 248 } 249 } else if (!rep->dont_use_callchains && 250 callchain_param.mode != CHAIN_NONE && 251 !symbol_conf.use_callchain) { 252 symbol_conf.use_callchain = true; 253 if (callchain_register_param(&callchain_param) < 0) { 254 ui__error("Can't register callchain params.\n"); 255 return -EINVAL; 256 } 257 } 258 259 if (symbol_conf.cumulate_callchain) { 260 /* Silently ignore if callchain is missing */ 261 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 262 symbol_conf.cumulate_callchain = false; 263 perf_hpp__cancel_cumulate(); 264 } 265 } 266 267 if (sort__mode == SORT_MODE__BRANCH) { 268 if (!is_pipe && 269 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 270 ui__error("Selected -b but no branch data. " 271 "Did you call perf record without -b?\n"); 272 return -1; 273 } 274 } 275 276 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) { 277 if ((sample_type & PERF_SAMPLE_REGS_USER) && 278 (sample_type & PERF_SAMPLE_STACK_USER)) 279 callchain_param.record_mode = CALLCHAIN_DWARF; 280 else if (sample_type & PERF_SAMPLE_BRANCH_STACK) 281 callchain_param.record_mode = CALLCHAIN_LBR; 282 else 283 callchain_param.record_mode = CALLCHAIN_FP; 284 } 285 286 /* ??? handle more cases than just ANY? */ 287 if (!(perf_evlist__combined_branch_type(session->evlist) & 288 PERF_SAMPLE_BRANCH_ANY)) 289 rep->nonany_branch_mode = true; 290 291 return 0; 292 } 293 294 static void sig_handler(int sig __maybe_unused) 295 { 296 session_done = 1; 297 } 298 299 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 300 const char *evname, FILE *fp) 301 { 302 size_t ret; 303 char unit; 304 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 305 u64 nr_events = hists->stats.total_period; 306 struct perf_evsel *evsel = hists_to_evsel(hists); 307 char buf[512]; 308 size_t size = sizeof(buf); 309 int socked_id = hists->socket_filter; 310 311 if (symbol_conf.filter_relative) { 312 nr_samples = hists->stats.nr_non_filtered_samples; 313 nr_events = hists->stats.total_non_filtered_period; 314 } 315 316 if (perf_evsel__is_group_event(evsel)) { 317 struct perf_evsel *pos; 318 319 perf_evsel__group_desc(evsel, buf, size); 320 evname = buf; 321 322 for_each_group_member(pos, evsel) { 323 const struct hists *pos_hists = evsel__hists(pos); 324 325 if (symbol_conf.filter_relative) { 326 nr_samples += pos_hists->stats.nr_non_filtered_samples; 327 nr_events += pos_hists->stats.total_non_filtered_period; 328 } else { 329 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE]; 330 nr_events += pos_hists->stats.total_period; 331 } 332 } 333 } 334 335 nr_samples = convert_unit(nr_samples, &unit); 336 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 337 if (evname != NULL) 338 ret += fprintf(fp, " of event '%s'", evname); 339 340 if (symbol_conf.show_ref_callgraph && 341 strstr(evname, "call-graph=no")) { 342 ret += fprintf(fp, ", show reference callgraph"); 343 } 344 345 if (rep->mem_mode) { 346 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 347 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 348 } else 349 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 350 351 if (socked_id > -1) 352 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 353 354 return ret + fprintf(fp, "\n#\n"); 355 } 356 357 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 358 struct report *rep, 359 const char *help) 360 { 361 struct perf_evsel *pos; 362 363 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples); 364 evlist__for_each(evlist, pos) { 365 struct hists *hists = evsel__hists(pos); 366 const char *evname = perf_evsel__name(pos); 367 368 if (symbol_conf.event_group && 369 !perf_evsel__is_group_leader(pos)) 370 continue; 371 372 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 373 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout); 374 fprintf(stdout, "\n\n"); 375 } 376 377 if (sort_order == NULL && 378 parent_pattern == default_parent_pattern) 379 fprintf(stdout, "#\n# (%s)\n#\n", help); 380 381 if (rep->show_threads) { 382 bool style = !strcmp(rep->pretty_printing_style, "raw"); 383 perf_read_values_display(stdout, &rep->show_threads_values, 384 style); 385 perf_read_values_destroy(&rep->show_threads_values); 386 } 387 388 return 0; 389 } 390 391 static void report__warn_kptr_restrict(const struct report *rep) 392 { 393 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 394 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 395 396 if (kernel_map == NULL || 397 (kernel_map->dso->hit && 398 (kernel_kmap->ref_reloc_sym == NULL || 399 kernel_kmap->ref_reloc_sym->addr == 0))) { 400 const char *desc = 401 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 402 "can't be resolved."; 403 404 if (kernel_map) { 405 const struct dso *kdso = kernel_map->dso; 406 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 407 desc = "If some relocation was applied (e.g. " 408 "kexec) symbols may be misresolved."; 409 } 410 } 411 412 ui__warning( 413 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 414 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 415 "Samples in kernel modules can't be resolved as well.\n\n", 416 desc); 417 } 418 } 419 420 static int report__gtk_browse_hists(struct report *rep, const char *help) 421 { 422 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 423 struct hist_browser_timer *timer, float min_pcnt); 424 425 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 426 427 if (hist_browser == NULL) { 428 ui__error("GTK browser not found!\n"); 429 return -1; 430 } 431 432 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 433 } 434 435 static int report__browse_hists(struct report *rep) 436 { 437 int ret; 438 struct perf_session *session = rep->session; 439 struct perf_evlist *evlist = session->evlist; 440 const char *help = perf_tip(system_path(TIPDIR)); 441 442 if (help == NULL) { 443 /* fallback for people who don't install perf ;-) */ 444 help = perf_tip(DOCDIR); 445 if (help == NULL) 446 help = "Cannot load tips.txt file, please install perf!"; 447 } 448 449 switch (use_browser) { 450 case 1: 451 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 452 rep->min_percent, 453 &session->header.env); 454 /* 455 * Usually "ret" is the last pressed key, and we only 456 * care if the key notifies us to switch data file. 457 */ 458 if (ret != K_SWITCH_INPUT_DATA) 459 ret = 0; 460 break; 461 case 2: 462 ret = report__gtk_browse_hists(rep, help); 463 break; 464 default: 465 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 466 break; 467 } 468 469 return ret; 470 } 471 472 static void report__collapse_hists(struct report *rep) 473 { 474 struct ui_progress prog; 475 struct perf_evsel *pos; 476 477 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 478 479 evlist__for_each(rep->session->evlist, pos) { 480 struct hists *hists = evsel__hists(pos); 481 482 if (pos->idx == 0) 483 hists->symbol_filter_str = rep->symbol_filter_str; 484 485 hists->socket_filter = rep->socket_filter; 486 487 hists__collapse_resort(hists, &prog); 488 489 /* Non-group events are considered as leader */ 490 if (symbol_conf.event_group && 491 !perf_evsel__is_group_leader(pos)) { 492 struct hists *leader_hists = evsel__hists(pos->leader); 493 494 hists__match(leader_hists, hists); 495 hists__link(leader_hists, hists); 496 } 497 } 498 499 ui_progress__finish(); 500 } 501 502 static void report__output_resort(struct report *rep) 503 { 504 struct ui_progress prog; 505 struct perf_evsel *pos; 506 507 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 508 509 evlist__for_each(rep->session->evlist, pos) 510 perf_evsel__output_resort(pos, &prog); 511 512 ui_progress__finish(); 513 } 514 515 static int __cmd_report(struct report *rep) 516 { 517 int ret; 518 struct perf_session *session = rep->session; 519 struct perf_evsel *pos; 520 struct perf_data_file *file = session->file; 521 522 signal(SIGINT, sig_handler); 523 524 if (rep->cpu_list) { 525 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 526 rep->cpu_bitmap); 527 if (ret) { 528 ui__error("failed to set cpu bitmap\n"); 529 return ret; 530 } 531 } 532 533 if (rep->show_threads) 534 perf_read_values_init(&rep->show_threads_values); 535 536 ret = report__setup_sample_type(rep); 537 if (ret) { 538 /* report__setup_sample_type() already showed error message */ 539 return ret; 540 } 541 542 ret = perf_session__process_events(session); 543 if (ret) { 544 ui__error("failed to process sample\n"); 545 return ret; 546 } 547 548 report__warn_kptr_restrict(rep); 549 550 evlist__for_each(session->evlist, pos) 551 rep->nr_entries += evsel__hists(pos)->nr_entries; 552 553 if (use_browser == 0) { 554 if (verbose > 3) 555 perf_session__fprintf(session, stdout); 556 557 if (verbose > 2) 558 perf_session__fprintf_dsos(session, stdout); 559 560 if (dump_trace) { 561 perf_session__fprintf_nr_events(session, stdout); 562 perf_evlist__fprintf_nr_events(session->evlist, stdout); 563 return 0; 564 } 565 } 566 567 report__collapse_hists(rep); 568 569 if (session_done()) 570 return 0; 571 572 /* 573 * recalculate number of entries after collapsing since it 574 * might be changed during the collapse phase. 575 */ 576 rep->nr_entries = 0; 577 evlist__for_each(session->evlist, pos) 578 rep->nr_entries += evsel__hists(pos)->nr_entries; 579 580 if (rep->nr_entries == 0) { 581 ui__error("The %s file has no samples!\n", file->path); 582 return 0; 583 } 584 585 report__output_resort(rep); 586 587 return report__browse_hists(rep); 588 } 589 590 static int 591 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 592 { 593 struct report *rep = (struct report *)opt->value; 594 595 /* 596 * --no-call-graph 597 */ 598 if (unset) { 599 rep->dont_use_callchains = true; 600 return 0; 601 } 602 603 return parse_callchain_report_opt(arg); 604 } 605 606 int 607 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 608 const char *arg, int unset __maybe_unused) 609 { 610 if (arg) { 611 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 612 if (err) { 613 char buf[BUFSIZ]; 614 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 615 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 616 return -1; 617 } 618 have_ignore_callees = 1; 619 } 620 621 return 0; 622 } 623 624 static int 625 parse_branch_mode(const struct option *opt __maybe_unused, 626 const char *str __maybe_unused, int unset) 627 { 628 int *branch_mode = opt->value; 629 630 *branch_mode = !unset; 631 return 0; 632 } 633 634 static int 635 parse_percent_limit(const struct option *opt, const char *str, 636 int unset __maybe_unused) 637 { 638 struct report *rep = opt->value; 639 double pcnt = strtof(str, NULL); 640 641 rep->min_percent = pcnt; 642 callchain_param.min_percent = pcnt; 643 return 0; 644 } 645 646 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 647 648 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 649 CALLCHAIN_REPORT_HELP 650 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 651 652 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 653 { 654 struct perf_session *session; 655 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 656 struct stat st; 657 bool has_br_stack = false; 658 int branch_mode = -1; 659 bool branch_call_mode = false; 660 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 661 const char * const report_usage[] = { 662 "perf report [<options>]", 663 NULL 664 }; 665 struct report report = { 666 .tool = { 667 .sample = process_sample_event, 668 .mmap = perf_event__process_mmap, 669 .mmap2 = perf_event__process_mmap2, 670 .comm = perf_event__process_comm, 671 .exit = perf_event__process_exit, 672 .fork = perf_event__process_fork, 673 .lost = perf_event__process_lost, 674 .read = process_read_event, 675 .attr = perf_event__process_attr, 676 .tracing_data = perf_event__process_tracing_data, 677 .build_id = perf_event__process_build_id, 678 .id_index = perf_event__process_id_index, 679 .auxtrace_info = perf_event__process_auxtrace_info, 680 .auxtrace = perf_event__process_auxtrace, 681 .ordered_events = true, 682 .ordering_requires_timestamps = true, 683 }, 684 .max_stack = PERF_MAX_STACK_DEPTH, 685 .pretty_printing_style = "normal", 686 .socket_filter = -1, 687 }; 688 const struct option options[] = { 689 OPT_STRING('i', "input", &input_name, "file", 690 "input file name"), 691 OPT_INCR('v', "verbose", &verbose, 692 "be more verbose (show symbol address, etc)"), 693 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 694 "dump raw trace in ASCII"), 695 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 696 "file", "vmlinux pathname"), 697 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 698 "file", "kallsyms pathname"), 699 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 700 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 701 "load module symbols - WARNING: use only with -k and LIVE kernel"), 702 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 703 "Show a column with the number of samples"), 704 OPT_BOOLEAN('T', "threads", &report.show_threads, 705 "Show per-thread event counters"), 706 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 707 "pretty printing style key: normal raw"), 708 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 709 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 710 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 711 "Use the stdio interface"), 712 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 713 OPT_BOOLEAN(0, "header-only", &report.header_only, 714 "Show only data header."), 715 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 716 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 717 " Please refer the man page for the complete list."), 718 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 719 "output field(s): overhead, period, sample plus all of sort keys"), 720 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 721 "Show sample percentage for different cpu modes"), 722 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 723 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 724 OPT_STRING('p', "parent", &parent_pattern, "regex", 725 "regex filter to identify parent, see: '--sort parent'"), 726 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 727 "Only display entries with parent-match"), 728 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, 729 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 730 report_callchain_help, &report_parse_callchain_opt, 731 callchain_default_opt), 732 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 733 "Accumulate callchains of children and show total overhead as well"), 734 OPT_INTEGER(0, "max-stack", &report.max_stack, 735 "Set the maximum stack depth when parsing the callchain, " 736 "anything beyond the specified depth will be ignored. " 737 "Default: " __stringify(PERF_MAX_STACK_DEPTH)), 738 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 739 "alias for inverted call graph"), 740 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 741 "ignore callees of these functions in call graphs", 742 report_parse_ignore_callees_opt), 743 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 744 "only consider symbols in these dsos"), 745 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 746 "only consider symbols in these comms"), 747 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 748 "only consider symbols in these pids"), 749 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 750 "only consider symbols in these tids"), 751 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 752 "only consider these symbols"), 753 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 754 "only show symbols that (partially) match with this filter"), 755 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 756 "width[,width...]", 757 "don't try to adjust column width, use these fixed values"), 758 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 759 "separator for columns, no spaces will be added between " 760 "columns '.' is reserved."), 761 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 762 "Only display entries resolved to a symbol"), 763 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 764 "Look for files with symbols relative to this directory"), 765 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 766 "list of cpus to profile"), 767 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 768 "Display extended information about perf.data file"), 769 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 770 "Interleave source code with assembly code (default)"), 771 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 772 "Display raw encoding of assembly instructions (default)"), 773 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 774 "Specify disassembler style (e.g. -M intel for intel syntax)"), 775 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 776 "Show a column with the sum of periods"), 777 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 778 "Show event group information together"), 779 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 780 "use branch records for per branch histogram filling", 781 parse_branch_mode), 782 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 783 "add last branch records to call history"), 784 OPT_STRING(0, "objdump", &objdump_path, "path", 785 "objdump binary to use for disassembly and annotations"), 786 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 787 "Disable symbol demangling"), 788 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 789 "Enable kernel symbol demangling"), 790 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 791 OPT_CALLBACK(0, "percent-limit", &report, "percent", 792 "Don't show entries under that percent", parse_percent_limit), 793 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 794 "how to display percentage of filtered entries", parse_filter_percentage), 795 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 796 "Instruction Tracing options", 797 itrace_parse_synth_opts), 798 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 799 "Show full source file name path for source lines"), 800 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 801 "Show callgraph from reference event"), 802 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 803 "only show processor socket that match with this filter"), 804 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 805 "Show raw trace event output (do not use print fmt or plugins)"), 806 OPT_END() 807 }; 808 struct perf_data_file file = { 809 .mode = PERF_DATA_MODE_READ, 810 }; 811 int ret = hists__init(); 812 813 if (ret < 0) 814 return ret; 815 816 perf_config(report__config, &report); 817 818 argc = parse_options(argc, argv, options, report_usage, 0); 819 if (argc) { 820 /* 821 * Special case: if there's an argument left then assume that 822 * it's a symbol filter: 823 */ 824 if (argc > 1) 825 usage_with_options(report_usage, options); 826 827 report.symbol_filter_str = argv[0]; 828 } 829 830 if (symbol_conf.vmlinux_name && 831 access(symbol_conf.vmlinux_name, R_OK)) { 832 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 833 return -EINVAL; 834 } 835 if (symbol_conf.kallsyms_name && 836 access(symbol_conf.kallsyms_name, R_OK)) { 837 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 838 return -EINVAL; 839 } 840 841 if (report.use_stdio) 842 use_browser = 0; 843 else if (report.use_tui) 844 use_browser = 1; 845 else if (report.use_gtk) 846 use_browser = 2; 847 848 if (report.inverted_callchain) 849 callchain_param.order = ORDER_CALLER; 850 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 851 callchain_param.order = ORDER_CALLER; 852 853 if (itrace_synth_opts.callchain && 854 (int)itrace_synth_opts.callchain_sz > report.max_stack) 855 report.max_stack = itrace_synth_opts.callchain_sz; 856 857 if (!input_name || !strlen(input_name)) { 858 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 859 input_name = "-"; 860 else 861 input_name = "perf.data"; 862 } 863 864 file.path = input_name; 865 file.force = symbol_conf.force; 866 867 repeat: 868 session = perf_session__new(&file, false, &report.tool); 869 if (session == NULL) 870 return -1; 871 872 if (report.queue_size) { 873 ordered_events__set_alloc_size(&session->ordered_events, 874 report.queue_size); 875 } 876 877 session->itrace_synth_opts = &itrace_synth_opts; 878 879 report.session = session; 880 881 has_br_stack = perf_header__has_feat(&session->header, 882 HEADER_BRANCH_STACK); 883 884 if (itrace_synth_opts.last_branch) 885 has_br_stack = true; 886 887 /* 888 * Branch mode is a tristate: 889 * -1 means default, so decide based on the file having branch data. 890 * 0/1 means the user chose a mode. 891 */ 892 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 893 !branch_call_mode) { 894 sort__mode = SORT_MODE__BRANCH; 895 symbol_conf.cumulate_callchain = false; 896 } 897 if (branch_call_mode) { 898 callchain_param.key = CCKEY_ADDRESS; 899 callchain_param.branch_callstack = 1; 900 symbol_conf.use_callchain = true; 901 callchain_register_param(&callchain_param); 902 if (sort_order == NULL) 903 sort_order = "srcline,symbol,dso"; 904 } 905 906 if (report.mem_mode) { 907 if (sort__mode == SORT_MODE__BRANCH) { 908 pr_err("branch and mem mode incompatible\n"); 909 goto error; 910 } 911 sort__mode = SORT_MODE__MEMORY; 912 symbol_conf.cumulate_callchain = false; 913 } 914 915 /* Force tty output for header output and per-thread stat. */ 916 if (report.header || report.header_only || report.show_threads) 917 use_browser = 0; 918 919 if (strcmp(input_name, "-") != 0) 920 setup_browser(true); 921 else 922 use_browser = 0; 923 924 if (setup_sorting(session->evlist) < 0) { 925 if (sort_order) 926 parse_options_usage(report_usage, options, "s", 1); 927 if (field_order) 928 parse_options_usage(sort_order ? NULL : report_usage, 929 options, "F", 1); 930 goto error; 931 } 932 933 if (report.header || report.header_only) { 934 perf_session__fprintf_info(session, stdout, 935 report.show_full_info); 936 if (report.header_only) { 937 ret = 0; 938 goto error; 939 } 940 } else if (use_browser == 0) { 941 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 942 stdout); 943 } 944 945 /* 946 * Only in the TUI browser we are doing integrated annotation, 947 * so don't allocate extra space that won't be used in the stdio 948 * implementation. 949 */ 950 if (ui__has_annotation()) { 951 symbol_conf.priv_size = sizeof(struct annotation); 952 machines__set_symbol_filter(&session->machines, 953 symbol__annotate_init); 954 /* 955 * For searching by name on the "Browse map details". 956 * providing it only in verbose mode not to bloat too 957 * much struct symbol. 958 */ 959 if (verbose) { 960 /* 961 * XXX: Need to provide a less kludgy way to ask for 962 * more space per symbol, the u32 is for the index on 963 * the ui browser. 964 * See symbol__browser_index. 965 */ 966 symbol_conf.priv_size += sizeof(u32); 967 symbol_conf.sort_by_name = true; 968 } 969 } 970 971 if (symbol__init(&session->header.env) < 0) 972 goto error; 973 974 sort__setup_elide(stdout); 975 976 ret = __cmd_report(&report); 977 if (ret == K_SWITCH_INPUT_DATA) { 978 perf_session__delete(session); 979 goto repeat; 980 } else 981 ret = 0; 982 983 error: 984 perf_session__delete(session); 985 return ret; 986 } 987