1 /* 2 * builtin-report.c 3 * 4 * Builtin report command: Analyze the perf.data input file, 5 * look up and read DSOs and symbol information and display 6 * a histogram of results, along various sorting keys. 7 */ 8 #include "builtin.h" 9 10 #include "util/util.h" 11 #include "util/cache.h" 12 13 #include "util/annotate.h" 14 #include "util/color.h" 15 #include <linux/list.h> 16 #include <linux/rbtree.h> 17 #include "util/symbol.h" 18 #include "util/callchain.h" 19 #include "util/strlist.h" 20 #include "util/values.h" 21 22 #include "perf.h" 23 #include "util/debug.h" 24 #include "util/evlist.h" 25 #include "util/evsel.h" 26 #include "util/header.h" 27 #include "util/session.h" 28 #include "util/tool.h" 29 30 #include "util/parse-options.h" 31 #include "util/parse-events.h" 32 33 #include "util/thread.h" 34 #include "util/sort.h" 35 #include "util/hist.h" 36 #include "util/data.h" 37 #include "arch/common.h" 38 39 #include <dlfcn.h> 40 #include <linux/bitmap.h> 41 42 struct report { 43 struct perf_tool tool; 44 struct perf_session *session; 45 bool force, use_tui, use_gtk, use_stdio; 46 bool hide_unresolved; 47 bool dont_use_callchains; 48 bool show_full_info; 49 bool show_threads; 50 bool inverted_callchain; 51 bool mem_mode; 52 bool header; 53 bool header_only; 54 int max_stack; 55 struct perf_read_values show_threads_values; 56 const char *pretty_printing_style; 57 const char *cpu_list; 58 const char *symbol_filter_str; 59 float min_percent; 60 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 61 }; 62 63 static int report__config(const char *var, const char *value, void *cb) 64 { 65 if (!strcmp(var, "report.group")) { 66 symbol_conf.event_group = perf_config_bool(var, value); 67 return 0; 68 } 69 if (!strcmp(var, "report.percent-limit")) { 70 struct report *rep = cb; 71 rep->min_percent = strtof(value, NULL); 72 return 0; 73 } 74 75 return perf_default_config(var, value, cb); 76 } 77 78 static int report__add_mem_hist_entry(struct perf_tool *tool, struct addr_location *al, 79 struct perf_sample *sample, struct perf_evsel *evsel, 80 union perf_event *event) 81 { 82 struct report *rep = container_of(tool, struct report, tool); 83 struct symbol *parent = NULL; 84 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 85 struct hist_entry *he; 86 struct mem_info *mi, *mx; 87 uint64_t cost; 88 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack); 89 90 if (err) 91 return err; 92 93 mi = machine__resolve_mem(al->machine, al->thread, sample, cpumode); 94 if (!mi) 95 return -ENOMEM; 96 97 if (rep->hide_unresolved && !al->sym) 98 return 0; 99 100 cost = sample->weight; 101 if (!cost) 102 cost = 1; 103 104 /* 105 * must pass period=weight in order to get the correct 106 * sorting from hists__collapse_resort() which is solely 107 * based on periods. We want sorting be done on nr_events * weight 108 * and this is indirectly achieved by passing period=weight here 109 * and the he_stat__add_period() function. 110 */ 111 he = __hists__add_entry(&evsel->hists, al, parent, NULL, mi, 112 cost, cost, 0); 113 if (!he) 114 return -ENOMEM; 115 116 if (ui__has_annotation()) { 117 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 118 if (err) 119 goto out; 120 121 mx = he->mem_info; 122 err = addr_map_symbol__inc_samples(&mx->daddr, evsel->idx); 123 if (err) 124 goto out; 125 } 126 127 evsel->hists.stats.total_period += cost; 128 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 129 err = hist_entry__append_callchain(he, sample); 130 out: 131 return err; 132 } 133 134 static int report__add_branch_hist_entry(struct perf_tool *tool, struct addr_location *al, 135 struct perf_sample *sample, struct perf_evsel *evsel) 136 { 137 struct report *rep = container_of(tool, struct report, tool); 138 struct symbol *parent = NULL; 139 unsigned i; 140 struct hist_entry *he; 141 struct branch_info *bi, *bx; 142 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack); 143 144 if (err) 145 return err; 146 147 bi = machine__resolve_bstack(al->machine, al->thread, 148 sample->branch_stack); 149 if (!bi) 150 return -ENOMEM; 151 152 for (i = 0; i < sample->branch_stack->nr; i++) { 153 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym)) 154 continue; 155 156 err = -ENOMEM; 157 158 /* overwrite the 'al' to branch-to info */ 159 al->map = bi[i].to.map; 160 al->sym = bi[i].to.sym; 161 al->addr = bi[i].to.addr; 162 /* 163 * The report shows the percentage of total branches captured 164 * and not events sampled. Thus we use a pseudo period of 1. 165 */ 166 he = __hists__add_entry(&evsel->hists, al, parent, &bi[i], NULL, 167 1, 1, 0); 168 if (he) { 169 if (ui__has_annotation()) { 170 bx = he->branch_info; 171 err = addr_map_symbol__inc_samples(&bx->from, 172 evsel->idx); 173 if (err) 174 goto out; 175 176 err = addr_map_symbol__inc_samples(&bx->to, 177 evsel->idx); 178 if (err) 179 goto out; 180 } 181 182 evsel->hists.stats.total_period += 1; 183 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 184 } else 185 goto out; 186 } 187 err = 0; 188 out: 189 free(bi); 190 return err; 191 } 192 193 static int report__add_hist_entry(struct perf_tool *tool, struct perf_evsel *evsel, 194 struct addr_location *al, struct perf_sample *sample) 195 { 196 struct report *rep = container_of(tool, struct report, tool); 197 struct symbol *parent = NULL; 198 struct hist_entry *he; 199 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack); 200 201 if (err) 202 return err; 203 204 he = __hists__add_entry(&evsel->hists, al, parent, NULL, NULL, 205 sample->period, sample->weight, 206 sample->transaction); 207 if (he == NULL) 208 return -ENOMEM; 209 210 err = hist_entry__append_callchain(he, sample); 211 if (err) 212 goto out; 213 214 if (ui__has_annotation()) 215 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 216 217 evsel->hists.stats.total_period += sample->period; 218 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 219 out: 220 return err; 221 } 222 223 224 static int process_sample_event(struct perf_tool *tool, 225 union perf_event *event, 226 struct perf_sample *sample, 227 struct perf_evsel *evsel, 228 struct machine *machine) 229 { 230 struct report *rep = container_of(tool, struct report, tool); 231 struct addr_location al; 232 int ret; 233 234 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) { 235 pr_debug("problem processing %d event, skipping it.\n", 236 event->header.type); 237 return -1; 238 } 239 240 if (al.filtered || (rep->hide_unresolved && al.sym == NULL)) 241 return 0; 242 243 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 244 return 0; 245 246 if (sort__mode == SORT_MODE__BRANCH) { 247 ret = report__add_branch_hist_entry(tool, &al, sample, evsel); 248 if (ret < 0) 249 pr_debug("problem adding lbr entry, skipping event\n"); 250 } else if (rep->mem_mode == 1) { 251 ret = report__add_mem_hist_entry(tool, &al, sample, evsel, event); 252 if (ret < 0) 253 pr_debug("problem adding mem entry, skipping event\n"); 254 } else { 255 if (al.map != NULL) 256 al.map->dso->hit = 1; 257 258 ret = report__add_hist_entry(tool, evsel, &al, sample); 259 if (ret < 0) 260 pr_debug("problem incrementing symbol period, skipping event\n"); 261 } 262 return ret; 263 } 264 265 static int process_read_event(struct perf_tool *tool, 266 union perf_event *event, 267 struct perf_sample *sample __maybe_unused, 268 struct perf_evsel *evsel, 269 struct machine *machine __maybe_unused) 270 { 271 struct report *rep = container_of(tool, struct report, tool); 272 273 if (rep->show_threads) { 274 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 275 perf_read_values_add_value(&rep->show_threads_values, 276 event->read.pid, event->read.tid, 277 event->read.id, 278 name, 279 event->read.value); 280 } 281 282 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 283 evsel ? perf_evsel__name(evsel) : "FAIL", 284 event->read.value); 285 286 return 0; 287 } 288 289 /* For pipe mode, sample_type is not currently set */ 290 static int report__setup_sample_type(struct report *rep) 291 { 292 struct perf_session *session = rep->session; 293 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 294 bool is_pipe = perf_data_file__is_pipe(session->file); 295 296 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 297 if (sort__has_parent) { 298 ui__error("Selected --sort parent, but no " 299 "callchain data. Did you call " 300 "'perf record' without -g?\n"); 301 return -EINVAL; 302 } 303 if (symbol_conf.use_callchain) { 304 ui__error("Selected -g but no callchain data. Did " 305 "you call 'perf record' without -g?\n"); 306 return -1; 307 } 308 } else if (!rep->dont_use_callchains && 309 callchain_param.mode != CHAIN_NONE && 310 !symbol_conf.use_callchain) { 311 symbol_conf.use_callchain = true; 312 if (callchain_register_param(&callchain_param) < 0) { 313 ui__error("Can't register callchain params.\n"); 314 return -EINVAL; 315 } 316 } 317 318 if (sort__mode == SORT_MODE__BRANCH) { 319 if (!is_pipe && 320 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 321 ui__error("Selected -b but no branch data. " 322 "Did you call perf record without -b?\n"); 323 return -1; 324 } 325 } 326 327 return 0; 328 } 329 330 static void sig_handler(int sig __maybe_unused) 331 { 332 session_done = 1; 333 } 334 335 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 336 const char *evname, FILE *fp) 337 { 338 size_t ret; 339 char unit; 340 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 341 u64 nr_events = hists->stats.total_period; 342 struct perf_evsel *evsel = hists_to_evsel(hists); 343 char buf[512]; 344 size_t size = sizeof(buf); 345 346 if (perf_evsel__is_group_event(evsel)) { 347 struct perf_evsel *pos; 348 349 perf_evsel__group_desc(evsel, buf, size); 350 evname = buf; 351 352 for_each_group_member(pos, evsel) { 353 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE]; 354 nr_events += pos->hists.stats.total_period; 355 } 356 } 357 358 nr_samples = convert_unit(nr_samples, &unit); 359 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 360 if (evname != NULL) 361 ret += fprintf(fp, " of event '%s'", evname); 362 363 if (rep->mem_mode) { 364 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 365 ret += fprintf(fp, "\n# Sort order : %s", sort_order); 366 } else 367 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 368 return ret + fprintf(fp, "\n#\n"); 369 } 370 371 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 372 struct report *rep, 373 const char *help) 374 { 375 struct perf_evsel *pos; 376 377 evlist__for_each(evlist, pos) { 378 struct hists *hists = &pos->hists; 379 const char *evname = perf_evsel__name(pos); 380 381 if (symbol_conf.event_group && 382 !perf_evsel__is_group_leader(pos)) 383 continue; 384 385 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 386 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout); 387 fprintf(stdout, "\n\n"); 388 } 389 390 if (sort_order == default_sort_order && 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 } 401 402 return 0; 403 } 404 405 static void report__warn_kptr_restrict(const struct report *rep) 406 { 407 struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION]; 408 struct kmap *kernel_kmap = map__kmap(kernel_map); 409 410 if (kernel_map == NULL || 411 (kernel_map->dso->hit && 412 (kernel_kmap->ref_reloc_sym == NULL || 413 kernel_kmap->ref_reloc_sym->addr == 0))) { 414 const char *desc = 415 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 416 "can't be resolved."; 417 418 if (kernel_map) { 419 const struct dso *kdso = kernel_map->dso; 420 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 421 desc = "If some relocation was applied (e.g. " 422 "kexec) symbols may be misresolved."; 423 } 424 } 425 426 ui__warning( 427 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 428 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 429 "Samples in kernel modules can't be resolved as well.\n\n", 430 desc); 431 } 432 } 433 434 static int report__gtk_browse_hists(struct report *rep, const char *help) 435 { 436 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 437 struct hist_browser_timer *timer, float min_pcnt); 438 439 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 440 441 if (hist_browser == NULL) { 442 ui__error("GTK browser not found!\n"); 443 return -1; 444 } 445 446 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 447 } 448 449 static int report__browse_hists(struct report *rep) 450 { 451 int ret; 452 struct perf_session *session = rep->session; 453 struct perf_evlist *evlist = session->evlist; 454 const char *help = "For a higher level overview, try: perf report --sort comm,dso"; 455 456 switch (use_browser) { 457 case 1: 458 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 459 rep->min_percent, 460 &session->header.env); 461 /* 462 * Usually "ret" is the last pressed key, and we only 463 * care if the key notifies us to switch data file. 464 */ 465 if (ret != K_SWITCH_INPUT_DATA) 466 ret = 0; 467 break; 468 case 2: 469 ret = report__gtk_browse_hists(rep, help); 470 break; 471 default: 472 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 473 break; 474 } 475 476 return ret; 477 } 478 479 static u64 report__collapse_hists(struct report *rep) 480 { 481 struct ui_progress prog; 482 struct perf_evsel *pos; 483 u64 nr_samples = 0; 484 /* 485 * Count number of histogram entries to use when showing progress, 486 * reusing nr_samples variable. 487 */ 488 evlist__for_each(rep->session->evlist, pos) 489 nr_samples += pos->hists.nr_entries; 490 491 ui_progress__init(&prog, nr_samples, "Merging related events..."); 492 /* 493 * Count total number of samples, will be used to check if this 494 * session had any. 495 */ 496 nr_samples = 0; 497 498 evlist__for_each(rep->session->evlist, pos) { 499 struct hists *hists = &pos->hists; 500 501 if (pos->idx == 0) 502 hists->symbol_filter_str = rep->symbol_filter_str; 503 504 hists__collapse_resort(hists, &prog); 505 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE]; 506 507 /* Non-group events are considered as leader */ 508 if (symbol_conf.event_group && 509 !perf_evsel__is_group_leader(pos)) { 510 struct hists *leader_hists = &pos->leader->hists; 511 512 hists__match(leader_hists, hists); 513 hists__link(leader_hists, hists); 514 } 515 } 516 517 ui_progress__finish(); 518 519 return nr_samples; 520 } 521 522 static int __cmd_report(struct report *rep) 523 { 524 int ret; 525 u64 nr_samples; 526 struct perf_session *session = rep->session; 527 struct perf_evsel *pos; 528 struct perf_data_file *file = session->file; 529 530 signal(SIGINT, sig_handler); 531 532 if (rep->cpu_list) { 533 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 534 rep->cpu_bitmap); 535 if (ret) 536 return ret; 537 } 538 539 if (rep->show_threads) 540 perf_read_values_init(&rep->show_threads_values); 541 542 ret = report__setup_sample_type(rep); 543 if (ret) 544 return ret; 545 546 ret = perf_session__process_events(session, &rep->tool); 547 if (ret) 548 return ret; 549 550 report__warn_kptr_restrict(rep); 551 552 if (use_browser == 0) { 553 if (verbose > 3) 554 perf_session__fprintf(session, stdout); 555 556 if (verbose > 2) 557 perf_session__fprintf_dsos(session, stdout); 558 559 if (dump_trace) { 560 perf_session__fprintf_nr_events(session, stdout); 561 return 0; 562 } 563 } 564 565 nr_samples = report__collapse_hists(rep); 566 567 if (session_done()) 568 return 0; 569 570 if (nr_samples == 0) { 571 ui__error("The %s file has no samples!\n", file->path); 572 return 0; 573 } 574 575 evlist__for_each(session->evlist, pos) 576 hists__output_resort(&pos->hists); 577 578 return report__browse_hists(rep); 579 } 580 581 static int 582 parse_callchain_opt(const struct option *opt, const char *arg, int unset) 583 { 584 struct report *rep = (struct report *)opt->value; 585 char *tok, *tok2; 586 char *endptr; 587 588 /* 589 * --no-call-graph 590 */ 591 if (unset) { 592 rep->dont_use_callchains = true; 593 return 0; 594 } 595 596 symbol_conf.use_callchain = true; 597 598 if (!arg) 599 return 0; 600 601 tok = strtok((char *)arg, ","); 602 if (!tok) 603 return -1; 604 605 /* get the output mode */ 606 if (!strncmp(tok, "graph", strlen(arg))) 607 callchain_param.mode = CHAIN_GRAPH_ABS; 608 609 else if (!strncmp(tok, "flat", strlen(arg))) 610 callchain_param.mode = CHAIN_FLAT; 611 612 else if (!strncmp(tok, "fractal", strlen(arg))) 613 callchain_param.mode = CHAIN_GRAPH_REL; 614 615 else if (!strncmp(tok, "none", strlen(arg))) { 616 callchain_param.mode = CHAIN_NONE; 617 symbol_conf.use_callchain = false; 618 619 return 0; 620 } 621 622 else 623 return -1; 624 625 /* get the min percentage */ 626 tok = strtok(NULL, ","); 627 if (!tok) 628 goto setup; 629 630 callchain_param.min_percent = strtod(tok, &endptr); 631 if (tok == endptr) 632 return -1; 633 634 /* get the print limit */ 635 tok2 = strtok(NULL, ","); 636 if (!tok2) 637 goto setup; 638 639 if (tok2[0] != 'c') { 640 callchain_param.print_limit = strtoul(tok2, &endptr, 0); 641 tok2 = strtok(NULL, ","); 642 if (!tok2) 643 goto setup; 644 } 645 646 /* get the call chain order */ 647 if (!strncmp(tok2, "caller", strlen("caller"))) 648 callchain_param.order = ORDER_CALLER; 649 else if (!strncmp(tok2, "callee", strlen("callee"))) 650 callchain_param.order = ORDER_CALLEE; 651 else 652 return -1; 653 654 /* Get the sort key */ 655 tok2 = strtok(NULL, ","); 656 if (!tok2) 657 goto setup; 658 if (!strncmp(tok2, "function", strlen("function"))) 659 callchain_param.key = CCKEY_FUNCTION; 660 else if (!strncmp(tok2, "address", strlen("address"))) 661 callchain_param.key = CCKEY_ADDRESS; 662 else 663 return -1; 664 setup: 665 if (callchain_register_param(&callchain_param) < 0) { 666 pr_err("Can't register callchain params\n"); 667 return -1; 668 } 669 return 0; 670 } 671 672 int 673 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 674 const char *arg, int unset __maybe_unused) 675 { 676 if (arg) { 677 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 678 if (err) { 679 char buf[BUFSIZ]; 680 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 681 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 682 return -1; 683 } 684 have_ignore_callees = 1; 685 } 686 687 return 0; 688 } 689 690 static int 691 parse_branch_mode(const struct option *opt __maybe_unused, 692 const char *str __maybe_unused, int unset) 693 { 694 int *branch_mode = opt->value; 695 696 *branch_mode = !unset; 697 return 0; 698 } 699 700 static int 701 parse_percent_limit(const struct option *opt, const char *str, 702 int unset __maybe_unused) 703 { 704 struct report *rep = opt->value; 705 706 rep->min_percent = strtof(str, NULL); 707 return 0; 708 } 709 710 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 711 { 712 struct perf_session *session; 713 struct stat st; 714 bool has_br_stack = false; 715 int branch_mode = -1; 716 int ret = -1; 717 char callchain_default_opt[] = "fractal,0.5,callee"; 718 const char * const report_usage[] = { 719 "perf report [<options>]", 720 NULL 721 }; 722 struct report report = { 723 .tool = { 724 .sample = process_sample_event, 725 .mmap = perf_event__process_mmap, 726 .mmap2 = perf_event__process_mmap2, 727 .comm = perf_event__process_comm, 728 .exit = perf_event__process_exit, 729 .fork = perf_event__process_fork, 730 .lost = perf_event__process_lost, 731 .read = process_read_event, 732 .attr = perf_event__process_attr, 733 .tracing_data = perf_event__process_tracing_data, 734 .build_id = perf_event__process_build_id, 735 .ordered_samples = true, 736 .ordering_requires_timestamps = true, 737 }, 738 .max_stack = PERF_MAX_STACK_DEPTH, 739 .pretty_printing_style = "normal", 740 }; 741 const struct option options[] = { 742 OPT_STRING('i', "input", &input_name, "file", 743 "input file name"), 744 OPT_INCR('v', "verbose", &verbose, 745 "be more verbose (show symbol address, etc)"), 746 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 747 "dump raw trace in ASCII"), 748 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 749 "file", "vmlinux pathname"), 750 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 751 "file", "kallsyms pathname"), 752 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"), 753 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 754 "load module symbols - WARNING: use only with -k and LIVE kernel"), 755 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 756 "Show a column with the number of samples"), 757 OPT_BOOLEAN('T', "threads", &report.show_threads, 758 "Show per-thread event counters"), 759 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 760 "pretty printing style key: normal raw"), 761 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 762 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 763 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 764 "Use the stdio interface"), 765 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 766 OPT_BOOLEAN(0, "header-only", &report.header_only, 767 "Show only data header."), 768 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 769 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline," 770 " dso_to, dso_from, symbol_to, symbol_from, mispredict," 771 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, " 772 "snoop, locked, abort, in_tx, transaction"), 773 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 774 "Show sample percentage for different cpu modes"), 775 OPT_STRING('p', "parent", &parent_pattern, "regex", 776 "regex filter to identify parent, see: '--sort parent'"), 777 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 778 "Only display entries with parent-match"), 779 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order", 780 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). " 781 "Default: fractal,0.5,callee,function", &parse_callchain_opt, callchain_default_opt), 782 OPT_INTEGER(0, "max-stack", &report.max_stack, 783 "Set the maximum stack depth when parsing the callchain, " 784 "anything beyond the specified depth will be ignored. " 785 "Default: " __stringify(PERF_MAX_STACK_DEPTH)), 786 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 787 "alias for inverted call graph"), 788 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 789 "ignore callees of these functions in call graphs", 790 report_parse_ignore_callees_opt), 791 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 792 "only consider symbols in these dsos"), 793 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 794 "only consider symbols in these comms"), 795 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 796 "only consider these symbols"), 797 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 798 "only show symbols that (partially) match with this filter"), 799 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 800 "width[,width...]", 801 "don't try to adjust column width, use these fixed values"), 802 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator", 803 "separator for columns, no spaces will be added between " 804 "columns '.' is reserved."), 805 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved, 806 "Only display entries resolved to a symbol"), 807 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 808 "Look for files with symbols relative to this directory"), 809 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 810 "list of cpus to profile"), 811 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 812 "Display extended information about perf.data file"), 813 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 814 "Interleave source code with assembly code (default)"), 815 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 816 "Display raw encoding of assembly instructions (default)"), 817 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 818 "Specify disassembler style (e.g. -M intel for intel syntax)"), 819 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 820 "Show a column with the sum of periods"), 821 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 822 "Show event group information together"), 823 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 824 "use branch records for histogram filling", parse_branch_mode), 825 OPT_STRING(0, "objdump", &objdump_path, "path", 826 "objdump binary to use for disassembly and annotations"), 827 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 828 "Disable symbol demangling"), 829 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 830 OPT_CALLBACK(0, "percent-limit", &report, "percent", 831 "Don't show entries under that percent", parse_percent_limit), 832 OPT_END() 833 }; 834 struct perf_data_file file = { 835 .mode = PERF_DATA_MODE_READ, 836 }; 837 838 perf_config(report__config, &report); 839 840 argc = parse_options(argc, argv, options, report_usage, 0); 841 842 if (report.use_stdio) 843 use_browser = 0; 844 else if (report.use_tui) 845 use_browser = 1; 846 else if (report.use_gtk) 847 use_browser = 2; 848 849 if (report.inverted_callchain) 850 callchain_param.order = ORDER_CALLER; 851 852 if (!input_name || !strlen(input_name)) { 853 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 854 input_name = "-"; 855 else 856 input_name = "perf.data"; 857 } 858 859 file.path = input_name; 860 file.force = report.force; 861 862 repeat: 863 session = perf_session__new(&file, false, &report.tool); 864 if (session == NULL) 865 return -ENOMEM; 866 867 report.session = session; 868 869 has_br_stack = perf_header__has_feat(&session->header, 870 HEADER_BRANCH_STACK); 871 872 if (branch_mode == -1 && has_br_stack) 873 sort__mode = SORT_MODE__BRANCH; 874 875 /* sort__mode could be NORMAL if --no-branch-stack */ 876 if (sort__mode == SORT_MODE__BRANCH) { 877 /* 878 * if no sort_order is provided, then specify 879 * branch-mode specific order 880 */ 881 if (sort_order == default_sort_order) 882 sort_order = "comm,dso_from,symbol_from," 883 "dso_to,symbol_to"; 884 885 } 886 if (report.mem_mode) { 887 if (sort__mode == SORT_MODE__BRANCH) { 888 pr_err("branch and mem mode incompatible\n"); 889 goto error; 890 } 891 sort__mode = SORT_MODE__MEMORY; 892 893 /* 894 * if no sort_order is provided, then specify 895 * branch-mode specific order 896 */ 897 if (sort_order == default_sort_order) 898 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked"; 899 } 900 901 if (setup_sorting() < 0) { 902 parse_options_usage(report_usage, options, "s", 1); 903 goto error; 904 } 905 906 if (parent_pattern != default_parent_pattern) { 907 if (sort_dimension__add("parent") < 0) 908 goto error; 909 } 910 911 /* Force tty output for header output. */ 912 if (report.header || report.header_only) 913 use_browser = 0; 914 915 if (strcmp(input_name, "-") != 0) 916 setup_browser(true); 917 else { 918 use_browser = 0; 919 perf_hpp__init(); 920 } 921 922 if (report.header || report.header_only) { 923 perf_session__fprintf_info(session, stdout, 924 report.show_full_info); 925 if (report.header_only) 926 return 0; 927 } else if (use_browser == 0) { 928 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 929 stdout); 930 } 931 932 /* 933 * Only in the TUI browser we are doing integrated annotation, 934 * so don't allocate extra space that won't be used in the stdio 935 * implementation. 936 */ 937 if (use_browser == 1 && sort__has_sym) { 938 symbol_conf.priv_size = sizeof(struct annotation); 939 machines__set_symbol_filter(&session->machines, 940 symbol__annotate_init); 941 /* 942 * For searching by name on the "Browse map details". 943 * providing it only in verbose mode not to bloat too 944 * much struct symbol. 945 */ 946 if (verbose) { 947 /* 948 * XXX: Need to provide a less kludgy way to ask for 949 * more space per symbol, the u32 is for the index on 950 * the ui browser. 951 * See symbol__browser_index. 952 */ 953 symbol_conf.priv_size += sizeof(u32); 954 symbol_conf.sort_by_name = true; 955 } 956 } 957 958 if (symbol__init() < 0) 959 goto error; 960 961 if (argc) { 962 /* 963 * Special case: if there's an argument left then assume that 964 * it's a symbol filter: 965 */ 966 if (argc > 1) 967 usage_with_options(report_usage, options); 968 969 report.symbol_filter_str = argv[0]; 970 } 971 972 sort__setup_elide(stdout); 973 974 ret = __cmd_report(&report); 975 if (ret == K_SWITCH_INPUT_DATA) { 976 perf_session__delete(session); 977 goto repeat; 978 } else 979 ret = 0; 980 981 error: 982 perf_session__delete(session); 983 return ret; 984 } 985