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 "arch/common.h" 37 38 #include <linux/bitmap.h> 39 40 struct perf_report { 41 struct perf_tool tool; 42 struct perf_session *session; 43 bool force, use_tui, use_gtk, use_stdio; 44 bool hide_unresolved; 45 bool dont_use_callchains; 46 bool show_full_info; 47 bool show_threads; 48 bool inverted_callchain; 49 bool mem_mode; 50 struct perf_read_values show_threads_values; 51 const char *pretty_printing_style; 52 symbol_filter_t annotate_init; 53 const char *cpu_list; 54 const char *symbol_filter_str; 55 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 56 }; 57 58 static int perf_report_config(const char *var, const char *value, void *cb) 59 { 60 if (!strcmp(var, "report.group")) { 61 symbol_conf.event_group = perf_config_bool(var, value); 62 return 0; 63 } 64 65 return perf_default_config(var, value, cb); 66 } 67 68 static int perf_report__add_mem_hist_entry(struct perf_tool *tool, 69 struct addr_location *al, 70 struct perf_sample *sample, 71 struct perf_evsel *evsel, 72 struct machine *machine, 73 union perf_event *event) 74 { 75 struct perf_report *rep = container_of(tool, struct perf_report, tool); 76 struct symbol *parent = NULL; 77 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 78 int err = 0; 79 struct hist_entry *he; 80 struct mem_info *mi, *mx; 81 uint64_t cost; 82 83 if ((sort__has_parent || symbol_conf.use_callchain) && 84 sample->callchain) { 85 err = machine__resolve_callchain(machine, evsel, al->thread, 86 sample, &parent); 87 if (err) 88 return err; 89 } 90 91 mi = machine__resolve_mem(machine, al->thread, sample, cpumode); 92 if (!mi) 93 return -ENOMEM; 94 95 if (rep->hide_unresolved && !al->sym) 96 return 0; 97 98 cost = sample->weight; 99 if (!cost) 100 cost = 1; 101 102 /* 103 * must pass period=weight in order to get the correct 104 * sorting from hists__collapse_resort() which is solely 105 * based on periods. We want sorting be done on nr_events * weight 106 * and this is indirectly achieved by passing period=weight here 107 * and the he_stat__add_period() function. 108 */ 109 he = __hists__add_mem_entry(&evsel->hists, al, parent, mi, cost, cost); 110 if (!he) 111 return -ENOMEM; 112 113 /* 114 * In the TUI browser, we are doing integrated annotation, 115 * so we don't allocate the extra space needed because the stdio 116 * code will not use it. 117 */ 118 if (sort__has_sym && he->ms.sym && use_browser > 0) { 119 struct annotation *notes = symbol__annotation(he->ms.sym); 120 121 assert(evsel != NULL); 122 123 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0) 124 goto out; 125 126 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 127 if (err) 128 goto out; 129 } 130 131 if (sort__has_sym && he->mem_info->daddr.sym && use_browser > 0) { 132 struct annotation *notes; 133 134 mx = he->mem_info; 135 136 notes = symbol__annotation(mx->daddr.sym); 137 if (notes->src == NULL && symbol__alloc_hist(mx->daddr.sym) < 0) 138 goto out; 139 140 err = symbol__inc_addr_samples(mx->daddr.sym, 141 mx->daddr.map, 142 evsel->idx, 143 mx->daddr.al_addr); 144 if (err) 145 goto out; 146 } 147 148 evsel->hists.stats.total_period += cost; 149 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 150 err = 0; 151 152 if (symbol_conf.use_callchain) { 153 err = callchain_append(he->callchain, 154 &callchain_cursor, 155 sample->period); 156 } 157 out: 158 return err; 159 } 160 161 static int perf_report__add_branch_hist_entry(struct perf_tool *tool, 162 struct addr_location *al, 163 struct perf_sample *sample, 164 struct perf_evsel *evsel, 165 struct machine *machine) 166 { 167 struct perf_report *rep = container_of(tool, struct perf_report, tool); 168 struct symbol *parent = NULL; 169 int err = 0; 170 unsigned i; 171 struct hist_entry *he; 172 struct branch_info *bi, *bx; 173 174 if ((sort__has_parent || symbol_conf.use_callchain) 175 && sample->callchain) { 176 err = machine__resolve_callchain(machine, evsel, al->thread, 177 sample, &parent); 178 if (err) 179 return err; 180 } 181 182 bi = machine__resolve_bstack(machine, al->thread, 183 sample->branch_stack); 184 if (!bi) 185 return -ENOMEM; 186 187 for (i = 0; i < sample->branch_stack->nr; i++) { 188 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym)) 189 continue; 190 /* 191 * The report shows the percentage of total branches captured 192 * and not events sampled. Thus we use a pseudo period of 1. 193 */ 194 he = __hists__add_branch_entry(&evsel->hists, al, parent, 195 &bi[i], 1, 1); 196 if (he) { 197 struct annotation *notes; 198 err = -ENOMEM; 199 bx = he->branch_info; 200 if (bx->from.sym && use_browser == 1 && sort__has_sym) { 201 notes = symbol__annotation(bx->from.sym); 202 if (!notes->src 203 && symbol__alloc_hist(bx->from.sym) < 0) 204 goto out; 205 206 err = symbol__inc_addr_samples(bx->from.sym, 207 bx->from.map, 208 evsel->idx, 209 bx->from.al_addr); 210 if (err) 211 goto out; 212 } 213 214 if (bx->to.sym && use_browser == 1 && sort__has_sym) { 215 notes = symbol__annotation(bx->to.sym); 216 if (!notes->src 217 && symbol__alloc_hist(bx->to.sym) < 0) 218 goto out; 219 220 err = symbol__inc_addr_samples(bx->to.sym, 221 bx->to.map, 222 evsel->idx, 223 bx->to.al_addr); 224 if (err) 225 goto out; 226 } 227 evsel->hists.stats.total_period += 1; 228 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 229 err = 0; 230 } else 231 return -ENOMEM; 232 } 233 out: 234 return err; 235 } 236 237 static int perf_evsel__add_hist_entry(struct perf_evsel *evsel, 238 struct addr_location *al, 239 struct perf_sample *sample, 240 struct machine *machine) 241 { 242 struct symbol *parent = NULL; 243 int err = 0; 244 struct hist_entry *he; 245 246 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) { 247 err = machine__resolve_callchain(machine, evsel, al->thread, 248 sample, &parent); 249 if (err) 250 return err; 251 } 252 253 he = __hists__add_entry(&evsel->hists, al, parent, sample->period, 254 sample->weight); 255 if (he == NULL) 256 return -ENOMEM; 257 258 if (symbol_conf.use_callchain) { 259 err = callchain_append(he->callchain, 260 &callchain_cursor, 261 sample->period); 262 if (err) 263 return err; 264 } 265 /* 266 * Only in the TUI browser we are doing integrated annotation, 267 * so we don't allocated the extra space needed because the stdio 268 * code will not use it. 269 */ 270 if (he->ms.sym != NULL && use_browser == 1 && sort__has_sym) { 271 struct annotation *notes = symbol__annotation(he->ms.sym); 272 273 assert(evsel != NULL); 274 275 err = -ENOMEM; 276 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0) 277 goto out; 278 279 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 280 } 281 282 evsel->hists.stats.total_period += sample->period; 283 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 284 out: 285 return err; 286 } 287 288 289 static int process_sample_event(struct perf_tool *tool, 290 union perf_event *event, 291 struct perf_sample *sample, 292 struct perf_evsel *evsel, 293 struct machine *machine) 294 { 295 struct perf_report *rep = container_of(tool, struct perf_report, tool); 296 struct addr_location al; 297 298 if (perf_event__preprocess_sample(event, machine, &al, sample, 299 rep->annotate_init) < 0) { 300 fprintf(stderr, "problem processing %d event, skipping it.\n", 301 event->header.type); 302 return -1; 303 } 304 305 if (al.filtered || (rep->hide_unresolved && al.sym == NULL)) 306 return 0; 307 308 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 309 return 0; 310 311 if (sort__branch_mode == 1) { 312 if (perf_report__add_branch_hist_entry(tool, &al, sample, 313 evsel, machine)) { 314 pr_debug("problem adding lbr entry, skipping event\n"); 315 return -1; 316 } 317 } else if (rep->mem_mode == 1) { 318 if (perf_report__add_mem_hist_entry(tool, &al, sample, 319 evsel, machine, event)) { 320 pr_debug("problem adding mem entry, skipping event\n"); 321 return -1; 322 } 323 } else { 324 if (al.map != NULL) 325 al.map->dso->hit = 1; 326 327 if (perf_evsel__add_hist_entry(evsel, &al, sample, machine)) { 328 pr_debug("problem incrementing symbol period, skipping event\n"); 329 return -1; 330 } 331 } 332 return 0; 333 } 334 335 static int process_read_event(struct perf_tool *tool, 336 union perf_event *event, 337 struct perf_sample *sample __maybe_unused, 338 struct perf_evsel *evsel, 339 struct machine *machine __maybe_unused) 340 { 341 struct perf_report *rep = container_of(tool, struct perf_report, tool); 342 343 if (rep->show_threads) { 344 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 345 perf_read_values_add_value(&rep->show_threads_values, 346 event->read.pid, event->read.tid, 347 event->read.id, 348 name, 349 event->read.value); 350 } 351 352 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 353 evsel ? perf_evsel__name(evsel) : "FAIL", 354 event->read.value); 355 356 return 0; 357 } 358 359 /* For pipe mode, sample_type is not currently set */ 360 static int perf_report__setup_sample_type(struct perf_report *rep) 361 { 362 struct perf_session *self = rep->session; 363 u64 sample_type = perf_evlist__sample_type(self->evlist); 364 365 if (!self->fd_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 366 if (sort__has_parent) { 367 ui__error("Selected --sort parent, but no " 368 "callchain data. Did you call " 369 "'perf record' without -g?\n"); 370 return -EINVAL; 371 } 372 if (symbol_conf.use_callchain) { 373 ui__error("Selected -g but no callchain data. Did " 374 "you call 'perf record' without -g?\n"); 375 return -1; 376 } 377 } else if (!rep->dont_use_callchains && 378 callchain_param.mode != CHAIN_NONE && 379 !symbol_conf.use_callchain) { 380 symbol_conf.use_callchain = true; 381 if (callchain_register_param(&callchain_param) < 0) { 382 ui__error("Can't register callchain params.\n"); 383 return -EINVAL; 384 } 385 } 386 387 if (sort__branch_mode == 1) { 388 if (!self->fd_pipe && 389 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 390 ui__error("Selected -b but no branch data. " 391 "Did you call perf record without -b?\n"); 392 return -1; 393 } 394 } 395 396 return 0; 397 } 398 399 extern volatile int session_done; 400 401 static void sig_handler(int sig __maybe_unused) 402 { 403 session_done = 1; 404 } 405 406 static size_t hists__fprintf_nr_sample_events(struct perf_report *rep, 407 struct hists *self, 408 const char *evname, FILE *fp) 409 { 410 size_t ret; 411 char unit; 412 unsigned long nr_samples = self->stats.nr_events[PERF_RECORD_SAMPLE]; 413 u64 nr_events = self->stats.total_period; 414 struct perf_evsel *evsel = hists_to_evsel(self); 415 char buf[512]; 416 size_t size = sizeof(buf); 417 418 if (perf_evsel__is_group_event(evsel)) { 419 struct perf_evsel *pos; 420 421 perf_evsel__group_desc(evsel, buf, size); 422 evname = buf; 423 424 for_each_group_member(pos, evsel) { 425 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE]; 426 nr_events += pos->hists.stats.total_period; 427 } 428 } 429 430 nr_samples = convert_unit(nr_samples, &unit); 431 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 432 if (evname != NULL) 433 ret += fprintf(fp, " of event '%s'", evname); 434 435 if (rep->mem_mode) { 436 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 437 ret += fprintf(fp, "\n# Sort order : %s", sort_order); 438 } else 439 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 440 return ret + fprintf(fp, "\n#\n"); 441 } 442 443 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 444 struct perf_report *rep, 445 const char *help) 446 { 447 struct perf_evsel *pos; 448 449 list_for_each_entry(pos, &evlist->entries, node) { 450 struct hists *hists = &pos->hists; 451 const char *evname = perf_evsel__name(pos); 452 453 if (symbol_conf.event_group && 454 !perf_evsel__is_group_leader(pos)) 455 continue; 456 457 hists__fprintf_nr_sample_events(rep, hists, evname, stdout); 458 hists__fprintf(hists, true, 0, 0, stdout); 459 fprintf(stdout, "\n\n"); 460 } 461 462 if (sort_order == default_sort_order && 463 parent_pattern == default_parent_pattern) { 464 fprintf(stdout, "#\n# (%s)\n#\n", help); 465 466 if (rep->show_threads) { 467 bool style = !strcmp(rep->pretty_printing_style, "raw"); 468 perf_read_values_display(stdout, &rep->show_threads_values, 469 style); 470 perf_read_values_destroy(&rep->show_threads_values); 471 } 472 } 473 474 return 0; 475 } 476 477 static int __cmd_report(struct perf_report *rep) 478 { 479 int ret = -EINVAL; 480 u64 nr_samples; 481 struct perf_session *session = rep->session; 482 struct perf_evsel *pos; 483 struct map *kernel_map; 484 struct kmap *kernel_kmap; 485 const char *help = "For a higher level overview, try: perf report --sort comm,dso"; 486 487 signal(SIGINT, sig_handler); 488 489 if (rep->cpu_list) { 490 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 491 rep->cpu_bitmap); 492 if (ret) 493 goto out_delete; 494 } 495 496 if (use_browser <= 0) 497 perf_session__fprintf_info(session, stdout, rep->show_full_info); 498 499 if (rep->show_threads) 500 perf_read_values_init(&rep->show_threads_values); 501 502 ret = perf_report__setup_sample_type(rep); 503 if (ret) 504 goto out_delete; 505 506 ret = perf_session__process_events(session, &rep->tool); 507 if (ret) 508 goto out_delete; 509 510 kernel_map = session->machines.host.vmlinux_maps[MAP__FUNCTION]; 511 kernel_kmap = map__kmap(kernel_map); 512 if (kernel_map == NULL || 513 (kernel_map->dso->hit && 514 (kernel_kmap->ref_reloc_sym == NULL || 515 kernel_kmap->ref_reloc_sym->addr == 0))) { 516 const char *desc = 517 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 518 "can't be resolved."; 519 520 if (kernel_map) { 521 const struct dso *kdso = kernel_map->dso; 522 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 523 desc = "If some relocation was applied (e.g. " 524 "kexec) symbols may be misresolved."; 525 } 526 } 527 528 ui__warning( 529 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 530 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 531 "Samples in kernel modules can't be resolved as well.\n\n", 532 desc); 533 } 534 535 if (verbose > 3) 536 perf_session__fprintf(session, stdout); 537 538 if (verbose > 2) 539 perf_session__fprintf_dsos(session, stdout); 540 541 if (dump_trace) { 542 perf_session__fprintf_nr_events(session, stdout); 543 goto out_delete; 544 } 545 546 nr_samples = 0; 547 list_for_each_entry(pos, &session->evlist->entries, node) { 548 struct hists *hists = &pos->hists; 549 550 if (pos->idx == 0) 551 hists->symbol_filter_str = rep->symbol_filter_str; 552 553 hists__collapse_resort(hists); 554 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE]; 555 556 /* Non-group events are considered as leader */ 557 if (symbol_conf.event_group && 558 !perf_evsel__is_group_leader(pos)) { 559 struct hists *leader_hists = &pos->leader->hists; 560 561 hists__match(leader_hists, hists); 562 hists__link(leader_hists, hists); 563 } 564 } 565 566 if (nr_samples == 0) { 567 ui__error("The %s file has no samples!\n", session->filename); 568 goto out_delete; 569 } 570 571 list_for_each_entry(pos, &session->evlist->entries, node) 572 hists__output_resort(&pos->hists); 573 574 if (use_browser > 0) { 575 if (use_browser == 1) { 576 ret = perf_evlist__tui_browse_hists(session->evlist, 577 help, 578 NULL, 579 &session->header.env); 580 /* 581 * Usually "ret" is the last pressed key, and we only 582 * care if the key notifies us to switch data file. 583 */ 584 if (ret != K_SWITCH_INPUT_DATA) 585 ret = 0; 586 587 } else if (use_browser == 2) { 588 perf_evlist__gtk_browse_hists(session->evlist, help, 589 NULL); 590 } 591 } else 592 perf_evlist__tty_browse_hists(session->evlist, rep, help); 593 594 out_delete: 595 /* 596 * Speed up the exit process, for large files this can 597 * take quite a while. 598 * 599 * XXX Enable this when using valgrind or if we ever 600 * librarize this command. 601 * 602 * Also experiment with obstacks to see how much speed 603 * up we'll get here. 604 * 605 * perf_session__delete(session); 606 */ 607 return ret; 608 } 609 610 static int 611 parse_callchain_opt(const struct option *opt, const char *arg, int unset) 612 { 613 struct perf_report *rep = (struct perf_report *)opt->value; 614 char *tok, *tok2; 615 char *endptr; 616 617 /* 618 * --no-call-graph 619 */ 620 if (unset) { 621 rep->dont_use_callchains = true; 622 return 0; 623 } 624 625 symbol_conf.use_callchain = true; 626 627 if (!arg) 628 return 0; 629 630 tok = strtok((char *)arg, ","); 631 if (!tok) 632 return -1; 633 634 /* get the output mode */ 635 if (!strncmp(tok, "graph", strlen(arg))) 636 callchain_param.mode = CHAIN_GRAPH_ABS; 637 638 else if (!strncmp(tok, "flat", strlen(arg))) 639 callchain_param.mode = CHAIN_FLAT; 640 641 else if (!strncmp(tok, "fractal", strlen(arg))) 642 callchain_param.mode = CHAIN_GRAPH_REL; 643 644 else if (!strncmp(tok, "none", strlen(arg))) { 645 callchain_param.mode = CHAIN_NONE; 646 symbol_conf.use_callchain = false; 647 648 return 0; 649 } 650 651 else 652 return -1; 653 654 /* get the min percentage */ 655 tok = strtok(NULL, ","); 656 if (!tok) 657 goto setup; 658 659 callchain_param.min_percent = strtod(tok, &endptr); 660 if (tok == endptr) 661 return -1; 662 663 /* get the print limit */ 664 tok2 = strtok(NULL, ","); 665 if (!tok2) 666 goto setup; 667 668 if (tok2[0] != 'c') { 669 callchain_param.print_limit = strtoul(tok2, &endptr, 0); 670 tok2 = strtok(NULL, ","); 671 if (!tok2) 672 goto setup; 673 } 674 675 /* get the call chain order */ 676 if (!strcmp(tok2, "caller")) 677 callchain_param.order = ORDER_CALLER; 678 else if (!strcmp(tok2, "callee")) 679 callchain_param.order = ORDER_CALLEE; 680 else 681 return -1; 682 setup: 683 if (callchain_register_param(&callchain_param) < 0) { 684 fprintf(stderr, "Can't register callchain params\n"); 685 return -1; 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 sort__branch_mode = !unset; 695 return 0; 696 } 697 698 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 699 { 700 struct perf_session *session; 701 struct stat st; 702 bool has_br_stack = false; 703 int ret = -1; 704 char callchain_default_opt[] = "fractal,0.5,callee"; 705 const char * const report_usage[] = { 706 "perf report [<options>]", 707 NULL 708 }; 709 struct perf_report report = { 710 .tool = { 711 .sample = process_sample_event, 712 .mmap = perf_event__process_mmap, 713 .comm = perf_event__process_comm, 714 .exit = perf_event__process_exit, 715 .fork = perf_event__process_fork, 716 .lost = perf_event__process_lost, 717 .read = process_read_event, 718 .attr = perf_event__process_attr, 719 .event_type = perf_event__process_event_type, 720 .tracing_data = perf_event__process_tracing_data, 721 .build_id = perf_event__process_build_id, 722 .ordered_samples = true, 723 .ordering_requires_timestamps = true, 724 }, 725 .pretty_printing_style = "normal", 726 }; 727 const struct option options[] = { 728 OPT_STRING('i', "input", &input_name, "file", 729 "input file name"), 730 OPT_INCR('v', "verbose", &verbose, 731 "be more verbose (show symbol address, etc)"), 732 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 733 "dump raw trace in ASCII"), 734 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 735 "file", "vmlinux pathname"), 736 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 737 "file", "kallsyms pathname"), 738 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"), 739 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 740 "load module symbols - WARNING: use only with -k and LIVE kernel"), 741 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 742 "Show a column with the number of samples"), 743 OPT_BOOLEAN('T', "threads", &report.show_threads, 744 "Show per-thread event counters"), 745 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 746 "pretty printing style key: normal raw"), 747 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 748 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 749 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 750 "Use the stdio interface"), 751 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 752 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline," 753 " dso_to, dso_from, symbol_to, symbol_from, mispredict," 754 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, " 755 "snoop, locked"), 756 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 757 "Show sample percentage for different cpu modes"), 758 OPT_STRING('p', "parent", &parent_pattern, "regex", 759 "regex filter to identify parent, see: '--sort parent'"), 760 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 761 "Only display entries with parent-match"), 762 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order", 763 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit and callchain order. " 764 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt), 765 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 766 "alias for inverted call graph"), 767 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 768 "only consider symbols in these dsos"), 769 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 770 "only consider symbols in these comms"), 771 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 772 "only consider these symbols"), 773 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 774 "only show symbols that (partially) match with this filter"), 775 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 776 "width[,width...]", 777 "don't try to adjust column width, use these fixed values"), 778 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator", 779 "separator for columns, no spaces will be added between " 780 "columns '.' is reserved."), 781 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved, 782 "Only display entries resolved to a symbol"), 783 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 784 "Look for files with symbols relative to this directory"), 785 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 786 "list of cpus to profile"), 787 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 788 "Display extended information about perf.data file"), 789 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 790 "Interleave source code with assembly code (default)"), 791 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 792 "Display raw encoding of assembly instructions (default)"), 793 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 794 "Specify disassembler style (e.g. -M intel for intel syntax)"), 795 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 796 "Show a column with the sum of periods"), 797 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 798 "Show event group information together"), 799 OPT_CALLBACK_NOOPT('b', "branch-stack", &sort__branch_mode, "", 800 "use branch records for histogram filling", parse_branch_mode), 801 OPT_STRING(0, "objdump", &objdump_path, "path", 802 "objdump binary to use for disassembly and annotations"), 803 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 804 "Disable symbol demangling"), 805 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 806 OPT_END() 807 }; 808 809 perf_config(perf_report_config, NULL); 810 811 argc = parse_options(argc, argv, options, report_usage, 0); 812 813 if (report.use_stdio) 814 use_browser = 0; 815 else if (report.use_tui) 816 use_browser = 1; 817 else if (report.use_gtk) 818 use_browser = 2; 819 820 if (report.inverted_callchain) 821 callchain_param.order = ORDER_CALLER; 822 823 if (!input_name || !strlen(input_name)) { 824 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 825 input_name = "-"; 826 else 827 input_name = "perf.data"; 828 } 829 830 if (strcmp(input_name, "-") != 0) 831 setup_browser(true); 832 else { 833 use_browser = 0; 834 perf_hpp__column_enable(PERF_HPP__OVERHEAD); 835 perf_hpp__init(); 836 } 837 838 repeat: 839 session = perf_session__new(input_name, O_RDONLY, 840 report.force, false, &report.tool); 841 if (session == NULL) 842 return -ENOMEM; 843 844 report.session = session; 845 846 has_br_stack = perf_header__has_feat(&session->header, 847 HEADER_BRANCH_STACK); 848 849 if (sort__branch_mode == -1 && has_br_stack) 850 sort__branch_mode = 1; 851 852 /* sort__branch_mode could be 0 if --no-branch-stack */ 853 if (sort__branch_mode == 1) { 854 /* 855 * if no sort_order is provided, then specify 856 * branch-mode specific order 857 */ 858 if (sort_order == default_sort_order) 859 sort_order = "comm,dso_from,symbol_from," 860 "dso_to,symbol_to"; 861 862 } 863 if (report.mem_mode) { 864 if (sort__branch_mode == 1) { 865 fprintf(stderr, "branch and mem mode incompatible\n"); 866 goto error; 867 } 868 /* 869 * if no sort_order is provided, then specify 870 * branch-mode specific order 871 */ 872 if (sort_order == default_sort_order) 873 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked"; 874 } 875 876 if (setup_sorting() < 0) 877 usage_with_options(report_usage, options); 878 879 /* 880 * Only in the TUI browser we are doing integrated annotation, 881 * so don't allocate extra space that won't be used in the stdio 882 * implementation. 883 */ 884 if (use_browser == 1 && sort__has_sym) { 885 symbol_conf.priv_size = sizeof(struct annotation); 886 report.annotate_init = symbol__annotate_init; 887 /* 888 * For searching by name on the "Browse map details". 889 * providing it only in verbose mode not to bloat too 890 * much struct symbol. 891 */ 892 if (verbose) { 893 /* 894 * XXX: Need to provide a less kludgy way to ask for 895 * more space per symbol, the u32 is for the index on 896 * the ui browser. 897 * See symbol__browser_index. 898 */ 899 symbol_conf.priv_size += sizeof(u32); 900 symbol_conf.sort_by_name = true; 901 } 902 } 903 904 if (symbol__init() < 0) 905 goto error; 906 907 if (parent_pattern != default_parent_pattern) { 908 if (sort_dimension__add("parent") < 0) 909 goto error; 910 911 /* 912 * Only show the parent fields if we explicitly 913 * sort that way. If we only use parent machinery 914 * for filtering, we don't want it. 915 */ 916 if (!strstr(sort_order, "parent")) 917 sort_parent.elide = 1; 918 } else 919 symbol_conf.exclude_other = false; 920 921 if (argc) { 922 /* 923 * Special case: if there's an argument left then assume that 924 * it's a symbol filter: 925 */ 926 if (argc > 1) 927 usage_with_options(report_usage, options); 928 929 report.symbol_filter_str = argv[0]; 930 } 931 932 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout); 933 934 if (sort__branch_mode == 1) { 935 sort_entry__setup_elide(&sort_dso_from, symbol_conf.dso_from_list, "dso_from", stdout); 936 sort_entry__setup_elide(&sort_dso_to, symbol_conf.dso_to_list, "dso_to", stdout); 937 sort_entry__setup_elide(&sort_sym_from, symbol_conf.sym_from_list, "sym_from", stdout); 938 sort_entry__setup_elide(&sort_sym_to, symbol_conf.sym_to_list, "sym_to", stdout); 939 } else { 940 if (report.mem_mode) { 941 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "symbol_daddr", stdout); 942 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso_daddr", stdout); 943 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "mem", stdout); 944 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "local_weight", stdout); 945 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "tlb", stdout); 946 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "snoop", stdout); 947 } 948 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout); 949 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout); 950 } 951 952 ret = __cmd_report(&report); 953 if (ret == K_SWITCH_INPUT_DATA) { 954 perf_session__delete(session); 955 goto repeat; 956 } else 957 ret = 0; 958 959 error: 960 perf_session__delete(session); 961 return ret; 962 } 963