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/config.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 #include <linux/stringify.h> 45 46 struct report { 47 struct perf_tool tool; 48 struct perf_session *session; 49 bool use_tui, use_gtk, use_stdio; 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 if (!strcmp(var, "report.sort_order")) { 93 default_sort_order = strdup(value); 94 return 0; 95 } 96 97 return 0; 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 if (!ui__has_annotation()) 112 return 0; 113 114 hist__account_cycles(iter->sample->branch_stack, al, iter->sample, 115 rep->nonany_branch_mode); 116 117 if (sort__mode == SORT_MODE__BRANCH) { 118 bi = he->branch_info; 119 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx); 120 if (err) 121 goto out; 122 123 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx); 124 125 } else if (rep->mem_mode) { 126 mi = he->mem_info; 127 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx); 128 if (err) 129 goto out; 130 131 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 132 133 } else if (symbol_conf.cumulate_callchain) { 134 if (single) 135 err = hist_entry__inc_addr_samples(he, evsel->idx, 136 al->addr); 137 } else { 138 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 139 } 140 141 out: 142 return err; 143 } 144 145 static int process_sample_event(struct perf_tool *tool, 146 union perf_event *event, 147 struct perf_sample *sample, 148 struct perf_evsel *evsel, 149 struct machine *machine) 150 { 151 struct report *rep = container_of(tool, struct report, tool); 152 struct addr_location al; 153 struct hist_entry_iter iter = { 154 .evsel = evsel, 155 .sample = sample, 156 .hide_unresolved = symbol_conf.hide_unresolved, 157 .add_entry_cb = hist_iter__report_callback, 158 }; 159 int ret = 0; 160 161 if (machine__resolve(machine, &al, sample) < 0) { 162 pr_debug("problem processing %d event, skipping it.\n", 163 event->header.type); 164 return -1; 165 } 166 167 if (symbol_conf.hide_unresolved && al.sym == NULL) 168 goto out_put; 169 170 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 171 goto out_put; 172 173 if (sort__mode == SORT_MODE__BRANCH) { 174 /* 175 * A non-synthesized event might not have a branch stack if 176 * branch stacks have been synthesized (using itrace options). 177 */ 178 if (!sample->branch_stack) 179 goto out_put; 180 iter.ops = &hist_iter_branch; 181 } else if (rep->mem_mode) { 182 iter.ops = &hist_iter_mem; 183 } else if (symbol_conf.cumulate_callchain) { 184 iter.ops = &hist_iter_cumulative; 185 } else { 186 iter.ops = &hist_iter_normal; 187 } 188 189 if (al.map != NULL) 190 al.map->dso->hit = 1; 191 192 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 193 if (ret < 0) 194 pr_debug("problem adding hist entry, skipping event\n"); 195 out_put: 196 addr_location__put(&al); 197 return ret; 198 } 199 200 static int process_read_event(struct perf_tool *tool, 201 union perf_event *event, 202 struct perf_sample *sample __maybe_unused, 203 struct perf_evsel *evsel, 204 struct machine *machine __maybe_unused) 205 { 206 struct report *rep = container_of(tool, struct report, tool); 207 208 if (rep->show_threads) { 209 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 210 int err = perf_read_values_add_value(&rep->show_threads_values, 211 event->read.pid, event->read.tid, 212 event->read.id, 213 name, 214 event->read.value); 215 216 if (err) 217 return err; 218 } 219 220 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 221 evsel ? perf_evsel__name(evsel) : "FAIL", 222 event->read.value); 223 224 return 0; 225 } 226 227 /* For pipe mode, sample_type is not currently set */ 228 static int report__setup_sample_type(struct report *rep) 229 { 230 struct perf_session *session = rep->session; 231 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 232 bool is_pipe = perf_data_file__is_pipe(session->file); 233 234 if (session->itrace_synth_opts->callchain || 235 (!is_pipe && 236 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 237 !session->itrace_synth_opts->set)) 238 sample_type |= PERF_SAMPLE_CALLCHAIN; 239 240 if (session->itrace_synth_opts->last_branch) 241 sample_type |= PERF_SAMPLE_BRANCH_STACK; 242 243 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 244 if (perf_hpp_list.parent) { 245 ui__error("Selected --sort parent, but no " 246 "callchain data. Did you call " 247 "'perf record' without -g?\n"); 248 return -EINVAL; 249 } 250 if (symbol_conf.use_callchain) { 251 ui__error("Selected -g or --branch-history but no " 252 "callchain data. Did\n" 253 "you call 'perf record' without -g?\n"); 254 return -1; 255 } 256 } else if (!callchain_param.enabled && 257 callchain_param.mode != CHAIN_NONE && 258 !symbol_conf.use_callchain) { 259 symbol_conf.use_callchain = true; 260 if (callchain_register_param(&callchain_param) < 0) { 261 ui__error("Can't register callchain params.\n"); 262 return -EINVAL; 263 } 264 } 265 266 if (symbol_conf.cumulate_callchain) { 267 /* Silently ignore if callchain is missing */ 268 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 269 symbol_conf.cumulate_callchain = false; 270 perf_hpp__cancel_cumulate(); 271 } 272 } 273 274 if (sort__mode == SORT_MODE__BRANCH) { 275 if (!is_pipe && 276 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 277 ui__error("Selected -b but no branch data. " 278 "Did you call perf record without -b?\n"); 279 return -1; 280 } 281 } 282 283 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) { 284 if ((sample_type & PERF_SAMPLE_REGS_USER) && 285 (sample_type & PERF_SAMPLE_STACK_USER)) 286 callchain_param.record_mode = CALLCHAIN_DWARF; 287 else if (sample_type & PERF_SAMPLE_BRANCH_STACK) 288 callchain_param.record_mode = CALLCHAIN_LBR; 289 else 290 callchain_param.record_mode = CALLCHAIN_FP; 291 } 292 293 /* ??? handle more cases than just ANY? */ 294 if (!(perf_evlist__combined_branch_type(session->evlist) & 295 PERF_SAMPLE_BRANCH_ANY)) 296 rep->nonany_branch_mode = true; 297 298 return 0; 299 } 300 301 static void sig_handler(int sig __maybe_unused) 302 { 303 session_done = 1; 304 } 305 306 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 307 const char *evname, FILE *fp) 308 { 309 size_t ret; 310 char unit; 311 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 312 u64 nr_events = hists->stats.total_period; 313 struct perf_evsel *evsel = hists_to_evsel(hists); 314 char buf[512]; 315 size_t size = sizeof(buf); 316 int socked_id = hists->socket_filter; 317 318 if (symbol_conf.filter_relative) { 319 nr_samples = hists->stats.nr_non_filtered_samples; 320 nr_events = hists->stats.total_non_filtered_period; 321 } 322 323 if (perf_evsel__is_group_event(evsel)) { 324 struct perf_evsel *pos; 325 326 perf_evsel__group_desc(evsel, buf, size); 327 evname = buf; 328 329 for_each_group_member(pos, evsel) { 330 const struct hists *pos_hists = evsel__hists(pos); 331 332 if (symbol_conf.filter_relative) { 333 nr_samples += pos_hists->stats.nr_non_filtered_samples; 334 nr_events += pos_hists->stats.total_non_filtered_period; 335 } else { 336 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE]; 337 nr_events += pos_hists->stats.total_period; 338 } 339 } 340 } 341 342 nr_samples = convert_unit(nr_samples, &unit); 343 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 344 if (evname != NULL) 345 ret += fprintf(fp, " of event '%s'", evname); 346 347 if (symbol_conf.show_ref_callgraph && 348 strstr(evname, "call-graph=no")) { 349 ret += fprintf(fp, ", show reference callgraph"); 350 } 351 352 if (rep->mem_mode) { 353 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 354 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 355 } else 356 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 357 358 if (socked_id > -1) 359 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 360 361 return ret + fprintf(fp, "\n#\n"); 362 } 363 364 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 365 struct report *rep, 366 const char *help) 367 { 368 struct perf_evsel *pos; 369 370 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples); 371 evlist__for_each_entry(evlist, pos) { 372 struct hists *hists = evsel__hists(pos); 373 const char *evname = perf_evsel__name(pos); 374 375 if (symbol_conf.event_group && 376 !perf_evsel__is_group_leader(pos)) 377 continue; 378 379 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 380 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout, 381 symbol_conf.use_callchain); 382 fprintf(stdout, "\n\n"); 383 } 384 385 if (sort_order == NULL && 386 parent_pattern == default_parent_pattern) 387 fprintf(stdout, "#\n# (%s)\n#\n", help); 388 389 if (rep->show_threads) { 390 bool style = !strcmp(rep->pretty_printing_style, "raw"); 391 perf_read_values_display(stdout, &rep->show_threads_values, 392 style); 393 perf_read_values_destroy(&rep->show_threads_values); 394 } 395 396 return 0; 397 } 398 399 static void report__warn_kptr_restrict(const struct report *rep) 400 { 401 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 402 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 403 404 if (kernel_map == NULL || 405 (kernel_map->dso->hit && 406 (kernel_kmap->ref_reloc_sym == NULL || 407 kernel_kmap->ref_reloc_sym->addr == 0))) { 408 const char *desc = 409 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 410 "can't be resolved."; 411 412 if (kernel_map) { 413 const struct dso *kdso = kernel_map->dso; 414 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 415 desc = "If some relocation was applied (e.g. " 416 "kexec) symbols may be misresolved."; 417 } 418 } 419 420 ui__warning( 421 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 422 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 423 "Samples in kernel modules can't be resolved as well.\n\n", 424 desc); 425 } 426 } 427 428 static int report__gtk_browse_hists(struct report *rep, const char *help) 429 { 430 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 431 struct hist_browser_timer *timer, float min_pcnt); 432 433 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 434 435 if (hist_browser == NULL) { 436 ui__error("GTK browser not found!\n"); 437 return -1; 438 } 439 440 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 441 } 442 443 static int report__browse_hists(struct report *rep) 444 { 445 int ret; 446 struct perf_session *session = rep->session; 447 struct perf_evlist *evlist = session->evlist; 448 const char *help = perf_tip(system_path(TIPDIR)); 449 450 if (help == NULL) { 451 /* fallback for people who don't install perf ;-) */ 452 help = perf_tip(DOCDIR); 453 if (help == NULL) 454 help = "Cannot load tips.txt file, please install perf!"; 455 } 456 457 switch (use_browser) { 458 case 1: 459 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 460 rep->min_percent, 461 &session->header.env); 462 /* 463 * Usually "ret" is the last pressed key, and we only 464 * care if the key notifies us to switch data file. 465 */ 466 if (ret != K_SWITCH_INPUT_DATA) 467 ret = 0; 468 break; 469 case 2: 470 ret = report__gtk_browse_hists(rep, help); 471 break; 472 default: 473 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 474 break; 475 } 476 477 return ret; 478 } 479 480 static int report__collapse_hists(struct report *rep) 481 { 482 struct ui_progress prog; 483 struct perf_evsel *pos; 484 int ret = 0; 485 486 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 487 488 evlist__for_each_entry(rep->session->evlist, pos) { 489 struct hists *hists = evsel__hists(pos); 490 491 if (pos->idx == 0) 492 hists->symbol_filter_str = rep->symbol_filter_str; 493 494 hists->socket_filter = rep->socket_filter; 495 496 ret = hists__collapse_resort(hists, &prog); 497 if (ret < 0) 498 break; 499 500 /* Non-group events are considered as leader */ 501 if (symbol_conf.event_group && 502 !perf_evsel__is_group_leader(pos)) { 503 struct hists *leader_hists = evsel__hists(pos->leader); 504 505 hists__match(leader_hists, hists); 506 hists__link(leader_hists, hists); 507 } 508 } 509 510 ui_progress__finish(); 511 return ret; 512 } 513 514 static void report__output_resort(struct report *rep) 515 { 516 struct ui_progress prog; 517 struct perf_evsel *pos; 518 519 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 520 521 evlist__for_each_entry(rep->session->evlist, pos) 522 perf_evsel__output_resort(pos, &prog); 523 524 ui_progress__finish(); 525 } 526 527 static int __cmd_report(struct report *rep) 528 { 529 int ret; 530 struct perf_session *session = rep->session; 531 struct perf_evsel *pos; 532 struct perf_data_file *file = session->file; 533 534 signal(SIGINT, sig_handler); 535 536 if (rep->cpu_list) { 537 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 538 rep->cpu_bitmap); 539 if (ret) { 540 ui__error("failed to set cpu bitmap\n"); 541 return ret; 542 } 543 } 544 545 if (rep->show_threads) { 546 ret = perf_read_values_init(&rep->show_threads_values); 547 if (ret) 548 return ret; 549 } 550 551 ret = report__setup_sample_type(rep); 552 if (ret) { 553 /* report__setup_sample_type() already showed error message */ 554 return ret; 555 } 556 557 ret = perf_session__process_events(session); 558 if (ret) { 559 ui__error("failed to process sample\n"); 560 return ret; 561 } 562 563 report__warn_kptr_restrict(rep); 564 565 evlist__for_each_entry(session->evlist, pos) 566 rep->nr_entries += evsel__hists(pos)->nr_entries; 567 568 if (use_browser == 0) { 569 if (verbose > 3) 570 perf_session__fprintf(session, stdout); 571 572 if (verbose > 2) 573 perf_session__fprintf_dsos(session, stdout); 574 575 if (dump_trace) { 576 perf_session__fprintf_nr_events(session, stdout); 577 perf_evlist__fprintf_nr_events(session->evlist, stdout); 578 return 0; 579 } 580 } 581 582 ret = report__collapse_hists(rep); 583 if (ret) { 584 ui__error("failed to process hist entry\n"); 585 return ret; 586 } 587 588 if (session_done()) 589 return 0; 590 591 /* 592 * recalculate number of entries after collapsing since it 593 * might be changed during the collapse phase. 594 */ 595 rep->nr_entries = 0; 596 evlist__for_each_entry(session->evlist, pos) 597 rep->nr_entries += evsel__hists(pos)->nr_entries; 598 599 if (rep->nr_entries == 0) { 600 ui__error("The %s file has no samples!\n", file->path); 601 return 0; 602 } 603 604 report__output_resort(rep); 605 606 return report__browse_hists(rep); 607 } 608 609 static int 610 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 611 { 612 struct callchain_param *callchain = opt->value; 613 614 callchain->enabled = !unset; 615 /* 616 * --no-call-graph 617 */ 618 if (unset) { 619 symbol_conf.use_callchain = false; 620 callchain->mode = CHAIN_NONE; 621 return 0; 622 } 623 624 return parse_callchain_report_opt(arg); 625 } 626 627 int 628 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 629 const char *arg, int unset __maybe_unused) 630 { 631 if (arg) { 632 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 633 if (err) { 634 char buf[BUFSIZ]; 635 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 636 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 637 return -1; 638 } 639 have_ignore_callees = 1; 640 } 641 642 return 0; 643 } 644 645 static int 646 parse_branch_mode(const struct option *opt __maybe_unused, 647 const char *str __maybe_unused, int unset) 648 { 649 int *branch_mode = opt->value; 650 651 *branch_mode = !unset; 652 return 0; 653 } 654 655 static int 656 parse_percent_limit(const struct option *opt, const char *str, 657 int unset __maybe_unused) 658 { 659 struct report *rep = opt->value; 660 double pcnt = strtof(str, NULL); 661 662 rep->min_percent = pcnt; 663 callchain_param.min_percent = pcnt; 664 return 0; 665 } 666 667 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 668 669 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 670 CALLCHAIN_REPORT_HELP 671 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 672 673 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 674 { 675 struct perf_session *session; 676 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 677 struct stat st; 678 bool has_br_stack = false; 679 int branch_mode = -1; 680 bool branch_call_mode = false; 681 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 682 const char * const report_usage[] = { 683 "perf report [<options>]", 684 NULL 685 }; 686 struct report report = { 687 .tool = { 688 .sample = process_sample_event, 689 .mmap = perf_event__process_mmap, 690 .mmap2 = perf_event__process_mmap2, 691 .comm = perf_event__process_comm, 692 .exit = perf_event__process_exit, 693 .fork = perf_event__process_fork, 694 .lost = perf_event__process_lost, 695 .read = process_read_event, 696 .attr = perf_event__process_attr, 697 .tracing_data = perf_event__process_tracing_data, 698 .build_id = perf_event__process_build_id, 699 .id_index = perf_event__process_id_index, 700 .auxtrace_info = perf_event__process_auxtrace_info, 701 .auxtrace = perf_event__process_auxtrace, 702 .ordered_events = true, 703 .ordering_requires_timestamps = true, 704 }, 705 .max_stack = PERF_MAX_STACK_DEPTH, 706 .pretty_printing_style = "normal", 707 .socket_filter = -1, 708 }; 709 const struct option options[] = { 710 OPT_STRING('i', "input", &input_name, "file", 711 "input file name"), 712 OPT_INCR('v', "verbose", &verbose, 713 "be more verbose (show symbol address, etc)"), 714 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 715 "dump raw trace in ASCII"), 716 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 717 "file", "vmlinux pathname"), 718 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 719 "file", "kallsyms pathname"), 720 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 721 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 722 "load module symbols - WARNING: use only with -k and LIVE kernel"), 723 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 724 "Show a column with the number of samples"), 725 OPT_BOOLEAN('T', "threads", &report.show_threads, 726 "Show per-thread event counters"), 727 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 728 "pretty printing style key: normal raw"), 729 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 730 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 731 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 732 "Use the stdio interface"), 733 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 734 OPT_BOOLEAN(0, "header-only", &report.header_only, 735 "Show only data header."), 736 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 737 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 738 " Please refer the man page for the complete list."), 739 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 740 "output field(s): overhead, period, sample plus all of sort keys"), 741 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 742 "Show sample percentage for different cpu modes"), 743 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 744 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 745 OPT_STRING('p', "parent", &parent_pattern, "regex", 746 "regex filter to identify parent, see: '--sort parent'"), 747 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 748 "Only display entries with parent-match"), 749 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 750 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 751 report_callchain_help, &report_parse_callchain_opt, 752 callchain_default_opt), 753 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 754 "Accumulate callchains of children and show total overhead as well"), 755 OPT_INTEGER(0, "max-stack", &report.max_stack, 756 "Set the maximum stack depth when parsing the callchain, " 757 "anything beyond the specified depth will be ignored. " 758 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 759 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 760 "alias for inverted call graph"), 761 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 762 "ignore callees of these functions in call graphs", 763 report_parse_ignore_callees_opt), 764 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 765 "only consider symbols in these dsos"), 766 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 767 "only consider symbols in these comms"), 768 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 769 "only consider symbols in these pids"), 770 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 771 "only consider symbols in these tids"), 772 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 773 "only consider these symbols"), 774 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 775 "only show symbols that (partially) match with this filter"), 776 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 777 "width[,width...]", 778 "don't try to adjust column width, use these fixed values"), 779 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 780 "separator for columns, no spaces will be added between " 781 "columns '.' is reserved."), 782 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 783 "Only display entries resolved to a symbol"), 784 OPT_CALLBACK(0, "symfs", NULL, "directory", 785 "Look for files with symbols relative to this directory", 786 symbol__config_symfs), 787 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 788 "list of cpus to profile"), 789 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 790 "Display extended information about perf.data file"), 791 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 792 "Interleave source code with assembly code (default)"), 793 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 794 "Display raw encoding of assembly instructions (default)"), 795 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 796 "Specify disassembler style (e.g. -M intel for intel syntax)"), 797 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 798 "Show a column with the sum of periods"), 799 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 800 "Show event group information together"), 801 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 802 "use branch records for per branch histogram filling", 803 parse_branch_mode), 804 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 805 "add last branch records to call history"), 806 OPT_STRING(0, "objdump", &objdump_path, "path", 807 "objdump binary to use for disassembly and annotations"), 808 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 809 "Disable symbol demangling"), 810 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 811 "Enable kernel symbol demangling"), 812 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 813 OPT_CALLBACK(0, "percent-limit", &report, "percent", 814 "Don't show entries under that percent", parse_percent_limit), 815 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 816 "how to display percentage of filtered entries", parse_filter_percentage), 817 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 818 "Instruction Tracing options", 819 itrace_parse_synth_opts), 820 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 821 "Show full source file name path for source lines"), 822 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 823 "Show callgraph from reference event"), 824 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 825 "only show processor socket that match with this filter"), 826 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 827 "Show raw trace event output (do not use print fmt or plugins)"), 828 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 829 "Show entries in a hierarchy"), 830 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 831 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 832 stdio__config_color, "always"), 833 OPT_END() 834 }; 835 struct perf_data_file file = { 836 .mode = PERF_DATA_MODE_READ, 837 }; 838 int ret = hists__init(); 839 840 if (ret < 0) 841 return ret; 842 843 perf_config(report__config, &report); 844 845 argc = parse_options(argc, argv, options, report_usage, 0); 846 if (argc) { 847 /* 848 * Special case: if there's an argument left then assume that 849 * it's a symbol filter: 850 */ 851 if (argc > 1) 852 usage_with_options(report_usage, options); 853 854 report.symbol_filter_str = argv[0]; 855 } 856 857 if (symbol_conf.vmlinux_name && 858 access(symbol_conf.vmlinux_name, R_OK)) { 859 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 860 return -EINVAL; 861 } 862 if (symbol_conf.kallsyms_name && 863 access(symbol_conf.kallsyms_name, R_OK)) { 864 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 865 return -EINVAL; 866 } 867 868 if (report.use_stdio) 869 use_browser = 0; 870 else if (report.use_tui) 871 use_browser = 1; 872 else if (report.use_gtk) 873 use_browser = 2; 874 875 if (report.inverted_callchain) 876 callchain_param.order = ORDER_CALLER; 877 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 878 callchain_param.order = ORDER_CALLER; 879 880 if (itrace_synth_opts.callchain && 881 (int)itrace_synth_opts.callchain_sz > report.max_stack) 882 report.max_stack = itrace_synth_opts.callchain_sz; 883 884 if (!input_name || !strlen(input_name)) { 885 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 886 input_name = "-"; 887 else 888 input_name = "perf.data"; 889 } 890 891 file.path = input_name; 892 file.force = symbol_conf.force; 893 894 repeat: 895 session = perf_session__new(&file, false, &report.tool); 896 if (session == NULL) 897 return -1; 898 899 if (report.queue_size) { 900 ordered_events__set_alloc_size(&session->ordered_events, 901 report.queue_size); 902 } 903 904 session->itrace_synth_opts = &itrace_synth_opts; 905 906 report.session = session; 907 908 has_br_stack = perf_header__has_feat(&session->header, 909 HEADER_BRANCH_STACK); 910 911 if (itrace_synth_opts.last_branch) 912 has_br_stack = true; 913 914 /* 915 * Branch mode is a tristate: 916 * -1 means default, so decide based on the file having branch data. 917 * 0/1 means the user chose a mode. 918 */ 919 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 920 !branch_call_mode) { 921 sort__mode = SORT_MODE__BRANCH; 922 symbol_conf.cumulate_callchain = false; 923 } 924 if (branch_call_mode) { 925 callchain_param.key = CCKEY_ADDRESS; 926 callchain_param.branch_callstack = 1; 927 symbol_conf.use_callchain = true; 928 callchain_register_param(&callchain_param); 929 if (sort_order == NULL) 930 sort_order = "srcline,symbol,dso"; 931 } 932 933 if (report.mem_mode) { 934 if (sort__mode == SORT_MODE__BRANCH) { 935 pr_err("branch and mem mode incompatible\n"); 936 goto error; 937 } 938 sort__mode = SORT_MODE__MEMORY; 939 symbol_conf.cumulate_callchain = false; 940 } 941 942 if (symbol_conf.report_hierarchy) { 943 /* disable incompatible options */ 944 symbol_conf.cumulate_callchain = false; 945 946 if (field_order) { 947 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 948 parse_options_usage(report_usage, options, "F", 1); 949 parse_options_usage(NULL, options, "hierarchy", 0); 950 goto error; 951 } 952 953 perf_hpp_list.need_collapse = true; 954 } 955 956 /* Force tty output for header output and per-thread stat. */ 957 if (report.header || report.header_only || report.show_threads) 958 use_browser = 0; 959 960 if (strcmp(input_name, "-") != 0) 961 setup_browser(true); 962 else 963 use_browser = 0; 964 965 if (setup_sorting(session->evlist) < 0) { 966 if (sort_order) 967 parse_options_usage(report_usage, options, "s", 1); 968 if (field_order) 969 parse_options_usage(sort_order ? NULL : report_usage, 970 options, "F", 1); 971 goto error; 972 } 973 974 if (report.header || report.header_only) { 975 perf_session__fprintf_info(session, stdout, 976 report.show_full_info); 977 if (report.header_only) { 978 ret = 0; 979 goto error; 980 } 981 } else if (use_browser == 0) { 982 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 983 stdout); 984 } 985 986 /* 987 * Only in the TUI browser we are doing integrated annotation, 988 * so don't allocate extra space that won't be used in the stdio 989 * implementation. 990 */ 991 if (ui__has_annotation()) { 992 ret = symbol__annotation_init(); 993 if (ret < 0) 994 goto error; 995 /* 996 * For searching by name on the "Browse map details". 997 * providing it only in verbose mode not to bloat too 998 * much struct symbol. 999 */ 1000 if (verbose) { 1001 /* 1002 * XXX: Need to provide a less kludgy way to ask for 1003 * more space per symbol, the u32 is for the index on 1004 * the ui browser. 1005 * See symbol__browser_index. 1006 */ 1007 symbol_conf.priv_size += sizeof(u32); 1008 symbol_conf.sort_by_name = true; 1009 } 1010 } 1011 1012 if (symbol__init(&session->header.env) < 0) 1013 goto error; 1014 1015 sort__setup_elide(stdout); 1016 1017 ret = __cmd_report(&report); 1018 if (ret == K_SWITCH_INPUT_DATA) { 1019 perf_session__delete(session); 1020 goto repeat; 1021 } else 1022 ret = 0; 1023 1024 error: 1025 perf_session__delete(session); 1026 return ret; 1027 } 1028