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 #include "util/time-utils.h" 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 const char *time_str; 63 struct perf_time_interval ptime; 64 float min_percent; 65 u64 nr_entries; 66 u64 queue_size; 67 int socket_filter; 68 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 69 }; 70 71 static int report__config(const char *var, const char *value, void *cb) 72 { 73 struct report *rep = cb; 74 75 if (!strcmp(var, "report.group")) { 76 symbol_conf.event_group = perf_config_bool(var, value); 77 return 0; 78 } 79 if (!strcmp(var, "report.percent-limit")) { 80 double pcnt = strtof(value, NULL); 81 82 rep->min_percent = pcnt; 83 callchain_param.min_percent = pcnt; 84 return 0; 85 } 86 if (!strcmp(var, "report.children")) { 87 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 88 return 0; 89 } 90 if (!strcmp(var, "report.queue-size")) { 91 rep->queue_size = perf_config_u64(var, value); 92 return 0; 93 } 94 if (!strcmp(var, "report.sort_order")) { 95 default_sort_order = strdup(value); 96 return 0; 97 } 98 99 return 0; 100 } 101 102 static int hist_iter__report_callback(struct hist_entry_iter *iter, 103 struct addr_location *al, bool single, 104 void *arg) 105 { 106 int err = 0; 107 struct report *rep = arg; 108 struct hist_entry *he = iter->he; 109 struct perf_evsel *evsel = iter->evsel; 110 struct mem_info *mi; 111 struct branch_info *bi; 112 113 if (!ui__has_annotation()) 114 return 0; 115 116 hist__account_cycles(iter->sample->branch_stack, al, iter->sample, 117 rep->nonany_branch_mode); 118 119 if (sort__mode == SORT_MODE__BRANCH) { 120 bi = he->branch_info; 121 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx); 122 if (err) 123 goto out; 124 125 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx); 126 127 } else if (rep->mem_mode) { 128 mi = he->mem_info; 129 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx); 130 if (err) 131 goto out; 132 133 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 134 135 } else if (symbol_conf.cumulate_callchain) { 136 if (single) 137 err = hist_entry__inc_addr_samples(he, evsel->idx, 138 al->addr); 139 } else { 140 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 141 } 142 143 out: 144 return err; 145 } 146 147 static int process_sample_event(struct perf_tool *tool, 148 union perf_event *event, 149 struct perf_sample *sample, 150 struct perf_evsel *evsel, 151 struct machine *machine) 152 { 153 struct report *rep = container_of(tool, struct report, tool); 154 struct addr_location al; 155 struct hist_entry_iter iter = { 156 .evsel = evsel, 157 .sample = sample, 158 .hide_unresolved = symbol_conf.hide_unresolved, 159 .add_entry_cb = hist_iter__report_callback, 160 }; 161 int ret = 0; 162 163 if (perf_time__skip_sample(&rep->ptime, sample->time)) 164 return 0; 165 166 if (machine__resolve(machine, &al, sample) < 0) { 167 pr_debug("problem processing %d event, skipping it.\n", 168 event->header.type); 169 return -1; 170 } 171 172 if (symbol_conf.hide_unresolved && al.sym == NULL) 173 goto out_put; 174 175 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 176 goto out_put; 177 178 if (sort__mode == SORT_MODE__BRANCH) { 179 /* 180 * A non-synthesized event might not have a branch stack if 181 * branch stacks have been synthesized (using itrace options). 182 */ 183 if (!sample->branch_stack) 184 goto out_put; 185 iter.ops = &hist_iter_branch; 186 } else if (rep->mem_mode) { 187 iter.ops = &hist_iter_mem; 188 } else if (symbol_conf.cumulate_callchain) { 189 iter.ops = &hist_iter_cumulative; 190 } else { 191 iter.ops = &hist_iter_normal; 192 } 193 194 if (al.map != NULL) 195 al.map->dso->hit = 1; 196 197 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 198 if (ret < 0) 199 pr_debug("problem adding hist entry, skipping event\n"); 200 out_put: 201 addr_location__put(&al); 202 return ret; 203 } 204 205 static int process_read_event(struct perf_tool *tool, 206 union perf_event *event, 207 struct perf_sample *sample __maybe_unused, 208 struct perf_evsel *evsel, 209 struct machine *machine __maybe_unused) 210 { 211 struct report *rep = container_of(tool, struct report, tool); 212 213 if (rep->show_threads) { 214 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 215 int err = perf_read_values_add_value(&rep->show_threads_values, 216 event->read.pid, event->read.tid, 217 event->read.id, 218 name, 219 event->read.value); 220 221 if (err) 222 return err; 223 } 224 225 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 226 evsel ? perf_evsel__name(evsel) : "FAIL", 227 event->read.value); 228 229 return 0; 230 } 231 232 /* For pipe mode, sample_type is not currently set */ 233 static int report__setup_sample_type(struct report *rep) 234 { 235 struct perf_session *session = rep->session; 236 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 237 bool is_pipe = perf_data_file__is_pipe(session->file); 238 239 if (session->itrace_synth_opts->callchain || 240 (!is_pipe && 241 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 242 !session->itrace_synth_opts->set)) 243 sample_type |= PERF_SAMPLE_CALLCHAIN; 244 245 if (session->itrace_synth_opts->last_branch) 246 sample_type |= PERF_SAMPLE_BRANCH_STACK; 247 248 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 249 if (perf_hpp_list.parent) { 250 ui__error("Selected --sort parent, but no " 251 "callchain data. Did you call " 252 "'perf record' without -g?\n"); 253 return -EINVAL; 254 } 255 if (symbol_conf.use_callchain) { 256 ui__error("Selected -g or --branch-history but no " 257 "callchain data. Did\n" 258 "you call 'perf record' without -g?\n"); 259 return -1; 260 } 261 } else if (!callchain_param.enabled && 262 callchain_param.mode != CHAIN_NONE && 263 !symbol_conf.use_callchain) { 264 symbol_conf.use_callchain = true; 265 if (callchain_register_param(&callchain_param) < 0) { 266 ui__error("Can't register callchain params.\n"); 267 return -EINVAL; 268 } 269 } 270 271 if (symbol_conf.cumulate_callchain) { 272 /* Silently ignore if callchain is missing */ 273 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 274 symbol_conf.cumulate_callchain = false; 275 perf_hpp__cancel_cumulate(); 276 } 277 } 278 279 if (sort__mode == SORT_MODE__BRANCH) { 280 if (!is_pipe && 281 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 282 ui__error("Selected -b but no branch data. " 283 "Did you call perf record without -b?\n"); 284 return -1; 285 } 286 } 287 288 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) { 289 if ((sample_type & PERF_SAMPLE_REGS_USER) && 290 (sample_type & PERF_SAMPLE_STACK_USER)) 291 callchain_param.record_mode = CALLCHAIN_DWARF; 292 else if (sample_type & PERF_SAMPLE_BRANCH_STACK) 293 callchain_param.record_mode = CALLCHAIN_LBR; 294 else 295 callchain_param.record_mode = CALLCHAIN_FP; 296 } 297 298 /* ??? handle more cases than just ANY? */ 299 if (!(perf_evlist__combined_branch_type(session->evlist) & 300 PERF_SAMPLE_BRANCH_ANY)) 301 rep->nonany_branch_mode = true; 302 303 return 0; 304 } 305 306 static void sig_handler(int sig __maybe_unused) 307 { 308 session_done = 1; 309 } 310 311 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 312 const char *evname, FILE *fp) 313 { 314 size_t ret; 315 char unit; 316 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 317 u64 nr_events = hists->stats.total_period; 318 struct perf_evsel *evsel = hists_to_evsel(hists); 319 char buf[512]; 320 size_t size = sizeof(buf); 321 int socked_id = hists->socket_filter; 322 323 if (symbol_conf.filter_relative) { 324 nr_samples = hists->stats.nr_non_filtered_samples; 325 nr_events = hists->stats.total_non_filtered_period; 326 } 327 328 if (perf_evsel__is_group_event(evsel)) { 329 struct perf_evsel *pos; 330 331 perf_evsel__group_desc(evsel, buf, size); 332 evname = buf; 333 334 for_each_group_member(pos, evsel) { 335 const struct hists *pos_hists = evsel__hists(pos); 336 337 if (symbol_conf.filter_relative) { 338 nr_samples += pos_hists->stats.nr_non_filtered_samples; 339 nr_events += pos_hists->stats.total_non_filtered_period; 340 } else { 341 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE]; 342 nr_events += pos_hists->stats.total_period; 343 } 344 } 345 } 346 347 nr_samples = convert_unit(nr_samples, &unit); 348 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 349 if (evname != NULL) 350 ret += fprintf(fp, " of event '%s'", evname); 351 352 if (symbol_conf.show_ref_callgraph && 353 strstr(evname, "call-graph=no")) { 354 ret += fprintf(fp, ", show reference callgraph"); 355 } 356 357 if (rep->mem_mode) { 358 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 359 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 360 } else 361 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 362 363 if (socked_id > -1) 364 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 365 366 return ret + fprintf(fp, "\n#\n"); 367 } 368 369 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 370 struct report *rep, 371 const char *help) 372 { 373 struct perf_evsel *pos; 374 375 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples); 376 evlist__for_each_entry(evlist, pos) { 377 struct hists *hists = evsel__hists(pos); 378 const char *evname = perf_evsel__name(pos); 379 380 if (symbol_conf.event_group && 381 !perf_evsel__is_group_leader(pos)) 382 continue; 383 384 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 385 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout, 386 symbol_conf.use_callchain); 387 fprintf(stdout, "\n\n"); 388 } 389 390 if (sort_order == NULL && 391 parent_pattern == default_parent_pattern) 392 fprintf(stdout, "#\n# (%s)\n#\n", help); 393 394 if (rep->show_threads) { 395 bool style = !strcmp(rep->pretty_printing_style, "raw"); 396 perf_read_values_display(stdout, &rep->show_threads_values, 397 style); 398 perf_read_values_destroy(&rep->show_threads_values); 399 } 400 401 return 0; 402 } 403 404 static void report__warn_kptr_restrict(const struct report *rep) 405 { 406 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 407 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 408 409 if (kernel_map == NULL || 410 (kernel_map->dso->hit && 411 (kernel_kmap->ref_reloc_sym == NULL || 412 kernel_kmap->ref_reloc_sym->addr == 0))) { 413 const char *desc = 414 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 415 "can't be resolved."; 416 417 if (kernel_map) { 418 const struct dso *kdso = kernel_map->dso; 419 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 420 desc = "If some relocation was applied (e.g. " 421 "kexec) symbols may be misresolved."; 422 } 423 } 424 425 ui__warning( 426 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 427 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 428 "Samples in kernel modules can't be resolved as well.\n\n", 429 desc); 430 } 431 } 432 433 static int report__gtk_browse_hists(struct report *rep, const char *help) 434 { 435 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 436 struct hist_browser_timer *timer, float min_pcnt); 437 438 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 439 440 if (hist_browser == NULL) { 441 ui__error("GTK browser not found!\n"); 442 return -1; 443 } 444 445 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 446 } 447 448 static int report__browse_hists(struct report *rep) 449 { 450 int ret; 451 struct perf_session *session = rep->session; 452 struct perf_evlist *evlist = session->evlist; 453 const char *help = perf_tip(system_path(TIPDIR)); 454 455 if (help == NULL) { 456 /* fallback for people who don't install perf ;-) */ 457 help = perf_tip(DOCDIR); 458 if (help == NULL) 459 help = "Cannot load tips.txt file, please install perf!"; 460 } 461 462 switch (use_browser) { 463 case 1: 464 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 465 rep->min_percent, 466 &session->header.env); 467 /* 468 * Usually "ret" is the last pressed key, and we only 469 * care if the key notifies us to switch data file. 470 */ 471 if (ret != K_SWITCH_INPUT_DATA) 472 ret = 0; 473 break; 474 case 2: 475 ret = report__gtk_browse_hists(rep, help); 476 break; 477 default: 478 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 479 break; 480 } 481 482 return ret; 483 } 484 485 static int report__collapse_hists(struct report *rep) 486 { 487 struct ui_progress prog; 488 struct perf_evsel *pos; 489 int ret = 0; 490 491 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 492 493 evlist__for_each_entry(rep->session->evlist, pos) { 494 struct hists *hists = evsel__hists(pos); 495 496 if (pos->idx == 0) 497 hists->symbol_filter_str = rep->symbol_filter_str; 498 499 hists->socket_filter = rep->socket_filter; 500 501 ret = hists__collapse_resort(hists, &prog); 502 if (ret < 0) 503 break; 504 505 /* Non-group events are considered as leader */ 506 if (symbol_conf.event_group && 507 !perf_evsel__is_group_leader(pos)) { 508 struct hists *leader_hists = evsel__hists(pos->leader); 509 510 hists__match(leader_hists, hists); 511 hists__link(leader_hists, hists); 512 } 513 } 514 515 ui_progress__finish(); 516 return ret; 517 } 518 519 static void report__output_resort(struct report *rep) 520 { 521 struct ui_progress prog; 522 struct perf_evsel *pos; 523 524 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 525 526 evlist__for_each_entry(rep->session->evlist, pos) 527 perf_evsel__output_resort(pos, &prog); 528 529 ui_progress__finish(); 530 } 531 532 static int __cmd_report(struct report *rep) 533 { 534 int ret; 535 struct perf_session *session = rep->session; 536 struct perf_evsel *pos; 537 struct perf_data_file *file = session->file; 538 539 signal(SIGINT, sig_handler); 540 541 if (rep->cpu_list) { 542 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 543 rep->cpu_bitmap); 544 if (ret) { 545 ui__error("failed to set cpu bitmap\n"); 546 return ret; 547 } 548 } 549 550 if (rep->show_threads) { 551 ret = perf_read_values_init(&rep->show_threads_values); 552 if (ret) 553 return ret; 554 } 555 556 ret = report__setup_sample_type(rep); 557 if (ret) { 558 /* report__setup_sample_type() already showed error message */ 559 return ret; 560 } 561 562 ret = perf_session__process_events(session); 563 if (ret) { 564 ui__error("failed to process sample\n"); 565 return ret; 566 } 567 568 report__warn_kptr_restrict(rep); 569 570 evlist__for_each_entry(session->evlist, pos) 571 rep->nr_entries += evsel__hists(pos)->nr_entries; 572 573 if (use_browser == 0) { 574 if (verbose > 3) 575 perf_session__fprintf(session, stdout); 576 577 if (verbose > 2) 578 perf_session__fprintf_dsos(session, stdout); 579 580 if (dump_trace) { 581 perf_session__fprintf_nr_events(session, stdout); 582 perf_evlist__fprintf_nr_events(session->evlist, stdout); 583 return 0; 584 } 585 } 586 587 ret = report__collapse_hists(rep); 588 if (ret) { 589 ui__error("failed to process hist entry\n"); 590 return ret; 591 } 592 593 if (session_done()) 594 return 0; 595 596 /* 597 * recalculate number of entries after collapsing since it 598 * might be changed during the collapse phase. 599 */ 600 rep->nr_entries = 0; 601 evlist__for_each_entry(session->evlist, pos) 602 rep->nr_entries += evsel__hists(pos)->nr_entries; 603 604 if (rep->nr_entries == 0) { 605 ui__error("The %s file has no samples!\n", file->path); 606 return 0; 607 } 608 609 report__output_resort(rep); 610 611 return report__browse_hists(rep); 612 } 613 614 static int 615 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 616 { 617 struct callchain_param *callchain = opt->value; 618 619 callchain->enabled = !unset; 620 /* 621 * --no-call-graph 622 */ 623 if (unset) { 624 symbol_conf.use_callchain = false; 625 callchain->mode = CHAIN_NONE; 626 return 0; 627 } 628 629 return parse_callchain_report_opt(arg); 630 } 631 632 int 633 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 634 const char *arg, int unset __maybe_unused) 635 { 636 if (arg) { 637 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 638 if (err) { 639 char buf[BUFSIZ]; 640 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 641 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 642 return -1; 643 } 644 have_ignore_callees = 1; 645 } 646 647 return 0; 648 } 649 650 static int 651 parse_branch_mode(const struct option *opt, 652 const char *str __maybe_unused, int unset) 653 { 654 int *branch_mode = opt->value; 655 656 *branch_mode = !unset; 657 return 0; 658 } 659 660 static int 661 parse_percent_limit(const struct option *opt, const char *str, 662 int unset __maybe_unused) 663 { 664 struct report *rep = opt->value; 665 double pcnt = strtof(str, NULL); 666 667 rep->min_percent = pcnt; 668 callchain_param.min_percent = pcnt; 669 return 0; 670 } 671 672 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 673 674 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 675 CALLCHAIN_REPORT_HELP 676 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 677 678 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 679 { 680 struct perf_session *session; 681 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 682 struct stat st; 683 bool has_br_stack = false; 684 int branch_mode = -1; 685 bool branch_call_mode = false; 686 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 687 const char * const report_usage[] = { 688 "perf report [<options>]", 689 NULL 690 }; 691 struct report report = { 692 .tool = { 693 .sample = process_sample_event, 694 .mmap = perf_event__process_mmap, 695 .mmap2 = perf_event__process_mmap2, 696 .comm = perf_event__process_comm, 697 .exit = perf_event__process_exit, 698 .fork = perf_event__process_fork, 699 .lost = perf_event__process_lost, 700 .read = process_read_event, 701 .attr = perf_event__process_attr, 702 .tracing_data = perf_event__process_tracing_data, 703 .build_id = perf_event__process_build_id, 704 .id_index = perf_event__process_id_index, 705 .auxtrace_info = perf_event__process_auxtrace_info, 706 .auxtrace = perf_event__process_auxtrace, 707 .ordered_events = true, 708 .ordering_requires_timestamps = true, 709 }, 710 .max_stack = PERF_MAX_STACK_DEPTH, 711 .pretty_printing_style = "normal", 712 .socket_filter = -1, 713 }; 714 const struct option options[] = { 715 OPT_STRING('i', "input", &input_name, "file", 716 "input file name"), 717 OPT_INCR('v', "verbose", &verbose, 718 "be more verbose (show symbol address, etc)"), 719 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 720 "dump raw trace in ASCII"), 721 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 722 "file", "vmlinux pathname"), 723 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 724 "file", "kallsyms pathname"), 725 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 726 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 727 "load module symbols - WARNING: use only with -k and LIVE kernel"), 728 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 729 "Show a column with the number of samples"), 730 OPT_BOOLEAN('T', "threads", &report.show_threads, 731 "Show per-thread event counters"), 732 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 733 "pretty printing style key: normal raw"), 734 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 735 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 736 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 737 "Use the stdio interface"), 738 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 739 OPT_BOOLEAN(0, "header-only", &report.header_only, 740 "Show only data header."), 741 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 742 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 743 " Please refer the man page for the complete list."), 744 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 745 "output field(s): overhead, period, sample plus all of sort keys"), 746 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 747 "Show sample percentage for different cpu modes"), 748 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 749 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 750 OPT_STRING('p', "parent", &parent_pattern, "regex", 751 "regex filter to identify parent, see: '--sort parent'"), 752 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 753 "Only display entries with parent-match"), 754 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 755 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 756 report_callchain_help, &report_parse_callchain_opt, 757 callchain_default_opt), 758 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 759 "Accumulate callchains of children and show total overhead as well"), 760 OPT_INTEGER(0, "max-stack", &report.max_stack, 761 "Set the maximum stack depth when parsing the callchain, " 762 "anything beyond the specified depth will be ignored. " 763 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 764 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 765 "alias for inverted call graph"), 766 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 767 "ignore callees of these functions in call graphs", 768 report_parse_ignore_callees_opt), 769 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 770 "only consider symbols in these dsos"), 771 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 772 "only consider symbols in these comms"), 773 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 774 "only consider symbols in these pids"), 775 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 776 "only consider symbols in these tids"), 777 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 778 "only consider these symbols"), 779 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 780 "only show symbols that (partially) match with this filter"), 781 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 782 "width[,width...]", 783 "don't try to adjust column width, use these fixed values"), 784 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 785 "separator for columns, no spaces will be added between " 786 "columns '.' is reserved."), 787 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 788 "Only display entries resolved to a symbol"), 789 OPT_CALLBACK(0, "symfs", NULL, "directory", 790 "Look for files with symbols relative to this directory", 791 symbol__config_symfs), 792 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 793 "list of cpus to profile"), 794 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 795 "Display extended information about perf.data file"), 796 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 797 "Interleave source code with assembly code (default)"), 798 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 799 "Display raw encoding of assembly instructions (default)"), 800 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 801 "Specify disassembler style (e.g. -M intel for intel syntax)"), 802 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 803 "Show a column with the sum of periods"), 804 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 805 "Show event group information together"), 806 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 807 "use branch records for per branch histogram filling", 808 parse_branch_mode), 809 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 810 "add last branch records to call history"), 811 OPT_STRING(0, "objdump", &objdump_path, "path", 812 "objdump binary to use for disassembly and annotations"), 813 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 814 "Disable symbol demangling"), 815 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 816 "Enable kernel symbol demangling"), 817 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 818 OPT_CALLBACK(0, "percent-limit", &report, "percent", 819 "Don't show entries under that percent", parse_percent_limit), 820 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 821 "how to display percentage of filtered entries", parse_filter_percentage), 822 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 823 "Instruction Tracing options", 824 itrace_parse_synth_opts), 825 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 826 "Show full source file name path for source lines"), 827 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 828 "Show callgraph from reference event"), 829 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 830 "only show processor socket that match with this filter"), 831 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 832 "Show raw trace event output (do not use print fmt or plugins)"), 833 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 834 "Show entries in a hierarchy"), 835 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 836 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 837 stdio__config_color, "always"), 838 OPT_STRING(0, "time", &report.time_str, "str", 839 "Time span of interest (start,stop)"), 840 OPT_END() 841 }; 842 struct perf_data_file file = { 843 .mode = PERF_DATA_MODE_READ, 844 }; 845 int ret = hists__init(); 846 847 if (ret < 0) 848 return ret; 849 850 perf_config(report__config, &report); 851 852 argc = parse_options(argc, argv, options, report_usage, 0); 853 if (argc) { 854 /* 855 * Special case: if there's an argument left then assume that 856 * it's a symbol filter: 857 */ 858 if (argc > 1) 859 usage_with_options(report_usage, options); 860 861 report.symbol_filter_str = argv[0]; 862 } 863 864 if (symbol_conf.vmlinux_name && 865 access(symbol_conf.vmlinux_name, R_OK)) { 866 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 867 return -EINVAL; 868 } 869 if (symbol_conf.kallsyms_name && 870 access(symbol_conf.kallsyms_name, R_OK)) { 871 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 872 return -EINVAL; 873 } 874 875 if (report.use_stdio) 876 use_browser = 0; 877 else if (report.use_tui) 878 use_browser = 1; 879 else if (report.use_gtk) 880 use_browser = 2; 881 882 if (report.inverted_callchain) 883 callchain_param.order = ORDER_CALLER; 884 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 885 callchain_param.order = ORDER_CALLER; 886 887 if (itrace_synth_opts.callchain && 888 (int)itrace_synth_opts.callchain_sz > report.max_stack) 889 report.max_stack = itrace_synth_opts.callchain_sz; 890 891 if (!input_name || !strlen(input_name)) { 892 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 893 input_name = "-"; 894 else 895 input_name = "perf.data"; 896 } 897 898 file.path = input_name; 899 file.force = symbol_conf.force; 900 901 repeat: 902 session = perf_session__new(&file, false, &report.tool); 903 if (session == NULL) 904 return -1; 905 906 if (report.queue_size) { 907 ordered_events__set_alloc_size(&session->ordered_events, 908 report.queue_size); 909 } 910 911 session->itrace_synth_opts = &itrace_synth_opts; 912 913 report.session = session; 914 915 has_br_stack = perf_header__has_feat(&session->header, 916 HEADER_BRANCH_STACK); 917 918 if (itrace_synth_opts.last_branch) 919 has_br_stack = true; 920 921 if (has_br_stack && branch_call_mode) 922 symbol_conf.show_branchflag_count = true; 923 924 /* 925 * Branch mode is a tristate: 926 * -1 means default, so decide based on the file having branch data. 927 * 0/1 means the user chose a mode. 928 */ 929 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 930 !branch_call_mode) { 931 sort__mode = SORT_MODE__BRANCH; 932 symbol_conf.cumulate_callchain = false; 933 } 934 if (branch_call_mode) { 935 callchain_param.key = CCKEY_ADDRESS; 936 callchain_param.branch_callstack = 1; 937 symbol_conf.use_callchain = true; 938 callchain_register_param(&callchain_param); 939 if (sort_order == NULL) 940 sort_order = "srcline,symbol,dso"; 941 } 942 943 if (report.mem_mode) { 944 if (sort__mode == SORT_MODE__BRANCH) { 945 pr_err("branch and mem mode incompatible\n"); 946 goto error; 947 } 948 sort__mode = SORT_MODE__MEMORY; 949 symbol_conf.cumulate_callchain = false; 950 } 951 952 if (symbol_conf.report_hierarchy) { 953 /* disable incompatible options */ 954 symbol_conf.cumulate_callchain = false; 955 956 if (field_order) { 957 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 958 parse_options_usage(report_usage, options, "F", 1); 959 parse_options_usage(NULL, options, "hierarchy", 0); 960 goto error; 961 } 962 963 perf_hpp_list.need_collapse = true; 964 } 965 966 /* Force tty output for header output and per-thread stat. */ 967 if (report.header || report.header_only || report.show_threads) 968 use_browser = 0; 969 970 if (strcmp(input_name, "-") != 0) 971 setup_browser(true); 972 else 973 use_browser = 0; 974 975 if (setup_sorting(session->evlist) < 0) { 976 if (sort_order) 977 parse_options_usage(report_usage, options, "s", 1); 978 if (field_order) 979 parse_options_usage(sort_order ? NULL : report_usage, 980 options, "F", 1); 981 goto error; 982 } 983 984 if (report.header || report.header_only) { 985 perf_session__fprintf_info(session, stdout, 986 report.show_full_info); 987 if (report.header_only) { 988 ret = 0; 989 goto error; 990 } 991 } else if (use_browser == 0) { 992 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 993 stdout); 994 } 995 996 /* 997 * Only in the TUI browser we are doing integrated annotation, 998 * so don't allocate extra space that won't be used in the stdio 999 * implementation. 1000 */ 1001 if (ui__has_annotation()) { 1002 ret = symbol__annotation_init(); 1003 if (ret < 0) 1004 goto error; 1005 /* 1006 * For searching by name on the "Browse map details". 1007 * providing it only in verbose mode not to bloat too 1008 * much struct symbol. 1009 */ 1010 if (verbose) { 1011 /* 1012 * XXX: Need to provide a less kludgy way to ask for 1013 * more space per symbol, the u32 is for the index on 1014 * the ui browser. 1015 * See symbol__browser_index. 1016 */ 1017 symbol_conf.priv_size += sizeof(u32); 1018 symbol_conf.sort_by_name = true; 1019 } 1020 } 1021 1022 if (symbol__init(&session->header.env) < 0) 1023 goto error; 1024 1025 if (perf_time__parse_str(&report.ptime, report.time_str) != 0) { 1026 pr_err("Invalid time string\n"); 1027 return -EINVAL; 1028 } 1029 1030 sort__setup_elide(stdout); 1031 1032 ret = __cmd_report(&report); 1033 if (ret == K_SWITCH_INPUT_DATA) { 1034 perf_session__delete(session); 1035 goto repeat; 1036 } else 1037 ret = 0; 1038 1039 error: 1040 perf_session__delete(session); 1041 return ret; 1042 } 1043