1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * builtin-report.c 4 * 5 * Builtin report command: Analyze the perf.data input file, 6 * look up and read DSOs and symbol information and display 7 * a histogram of results, along various sorting keys. 8 */ 9 #include "builtin.h" 10 11 #include "util/config.h" 12 13 #include "util/annotate.h" 14 #include "util/color.h" 15 #include "util/dso.h" 16 #include <linux/list.h> 17 #include <linux/rbtree.h> 18 #include <linux/err.h> 19 #include <linux/zalloc.h> 20 #include "util/map.h" 21 #include "util/symbol.h" 22 #include "util/map_symbol.h" 23 #include "util/mem-events.h" 24 #include "util/branch.h" 25 #include "util/callchain.h" 26 #include "util/values.h" 27 28 #include "perf.h" 29 #include "util/debug.h" 30 #include "util/evlist.h" 31 #include "util/evsel.h" 32 #include "util/evswitch.h" 33 #include "util/header.h" 34 #include "util/session.h" 35 #include "util/srcline.h" 36 #include "util/tool.h" 37 38 #include <subcmd/parse-options.h> 39 #include <subcmd/exec-cmd.h> 40 #include "util/parse-events.h" 41 42 #include "util/thread.h" 43 #include "util/sort.h" 44 #include "util/hist.h" 45 #include "util/data.h" 46 #include "arch/common.h" 47 #include "util/time-utils.h" 48 #include "util/auxtrace.h" 49 #include "util/units.h" 50 #include "util/util.h" // perf_tip() 51 #include "ui/ui.h" 52 #include "ui/progress.h" 53 #include "util/block-info.h" 54 55 #include <dlfcn.h> 56 #include <errno.h> 57 #include <inttypes.h> 58 #include <regex.h> 59 #include <linux/ctype.h> 60 #include <signal.h> 61 #include <linux/bitmap.h> 62 #include <linux/string.h> 63 #include <linux/stringify.h> 64 #include <linux/time64.h> 65 #include <sys/types.h> 66 #include <sys/stat.h> 67 #include <unistd.h> 68 #include <linux/mman.h> 69 70 struct report { 71 struct perf_tool tool; 72 struct perf_session *session; 73 struct evswitch evswitch; 74 bool use_tui, use_gtk, use_stdio; 75 bool show_full_info; 76 bool show_threads; 77 bool inverted_callchain; 78 bool mem_mode; 79 bool stats_mode; 80 bool tasks_mode; 81 bool mmaps_mode; 82 bool header; 83 bool header_only; 84 bool nonany_branch_mode; 85 bool group_set; 86 bool stitch_lbr; 87 bool disable_order; 88 bool skip_empty; 89 int max_stack; 90 struct perf_read_values show_threads_values; 91 struct annotation_options annotation_opts; 92 const char *pretty_printing_style; 93 const char *cpu_list; 94 const char *symbol_filter_str; 95 const char *time_str; 96 struct perf_time_interval *ptime_range; 97 int range_size; 98 int range_num; 99 float min_percent; 100 u64 nr_entries; 101 u64 queue_size; 102 u64 total_cycles; 103 int socket_filter; 104 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 105 struct branch_type_stat brtype_stat; 106 bool symbol_ipc; 107 bool total_cycles_mode; 108 struct block_report *block_reports; 109 int nr_block_reports; 110 }; 111 112 static int report__config(const char *var, const char *value, void *cb) 113 { 114 struct report *rep = cb; 115 116 if (!strcmp(var, "report.group")) { 117 symbol_conf.event_group = perf_config_bool(var, value); 118 return 0; 119 } 120 if (!strcmp(var, "report.percent-limit")) { 121 double pcnt = strtof(value, NULL); 122 123 rep->min_percent = pcnt; 124 callchain_param.min_percent = pcnt; 125 return 0; 126 } 127 if (!strcmp(var, "report.children")) { 128 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 129 return 0; 130 } 131 if (!strcmp(var, "report.queue-size")) 132 return perf_config_u64(&rep->queue_size, var, value); 133 134 if (!strcmp(var, "report.sort_order")) { 135 default_sort_order = strdup(value); 136 return 0; 137 } 138 139 if (!strcmp(var, "report.skip-empty")) { 140 rep->skip_empty = perf_config_bool(var, value); 141 return 0; 142 } 143 144 return 0; 145 } 146 147 static int hist_iter__report_callback(struct hist_entry_iter *iter, 148 struct addr_location *al, bool single, 149 void *arg) 150 { 151 int err = 0; 152 struct report *rep = arg; 153 struct hist_entry *he = iter->he; 154 struct evsel *evsel = iter->evsel; 155 struct perf_sample *sample = iter->sample; 156 struct mem_info *mi; 157 struct branch_info *bi; 158 159 if (!ui__has_annotation() && !rep->symbol_ipc) 160 return 0; 161 162 if (sort__mode == SORT_MODE__BRANCH) { 163 bi = he->branch_info; 164 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 165 if (err) 166 goto out; 167 168 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 169 170 } else if (rep->mem_mode) { 171 mi = he->mem_info; 172 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel); 173 if (err) 174 goto out; 175 176 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 177 178 } else if (symbol_conf.cumulate_callchain) { 179 if (single) 180 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 181 } else { 182 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 183 } 184 185 out: 186 return err; 187 } 188 189 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 190 struct addr_location *al __maybe_unused, 191 bool single __maybe_unused, 192 void *arg) 193 { 194 struct hist_entry *he = iter->he; 195 struct report *rep = arg; 196 struct branch_info *bi = he->branch_info; 197 struct perf_sample *sample = iter->sample; 198 struct evsel *evsel = iter->evsel; 199 int err; 200 201 branch_type_count(&rep->brtype_stat, &bi->flags, 202 bi->from.addr, bi->to.addr); 203 204 if (!ui__has_annotation() && !rep->symbol_ipc) 205 return 0; 206 207 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 208 if (err) 209 goto out; 210 211 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 212 213 out: 214 return err; 215 } 216 217 static void setup_forced_leader(struct report *report, 218 struct evlist *evlist) 219 { 220 if (report->group_set) 221 evlist__force_leader(evlist); 222 } 223 224 static int process_feature_event(struct perf_session *session, 225 union perf_event *event) 226 { 227 struct report *rep = container_of(session->tool, struct report, tool); 228 229 if (event->feat.feat_id < HEADER_LAST_FEATURE) 230 return perf_event__process_feature(session, event); 231 232 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 233 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 234 event->feat.feat_id); 235 return -1; 236 } else if (rep->header_only) { 237 session_done = 1; 238 } 239 240 /* 241 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 242 * means all features are received, now we can force the 243 * group if needed. 244 */ 245 setup_forced_leader(rep, session->evlist); 246 return 0; 247 } 248 249 static int process_sample_event(struct perf_tool *tool, 250 union perf_event *event, 251 struct perf_sample *sample, 252 struct evsel *evsel, 253 struct machine *machine) 254 { 255 struct report *rep = container_of(tool, struct report, tool); 256 struct addr_location al; 257 struct hist_entry_iter iter = { 258 .evsel = evsel, 259 .sample = sample, 260 .hide_unresolved = symbol_conf.hide_unresolved, 261 .add_entry_cb = hist_iter__report_callback, 262 }; 263 int ret = 0; 264 265 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 266 sample->time)) { 267 return 0; 268 } 269 270 if (evswitch__discard(&rep->evswitch, evsel)) 271 return 0; 272 273 if (machine__resolve(machine, &al, sample) < 0) { 274 pr_debug("problem processing %d event, skipping it.\n", 275 event->header.type); 276 return -1; 277 } 278 279 if (rep->stitch_lbr) 280 al.thread->lbr_stitch_enable = true; 281 282 if (symbol_conf.hide_unresolved && al.sym == NULL) 283 goto out_put; 284 285 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 286 goto out_put; 287 288 if (sort__mode == SORT_MODE__BRANCH) { 289 /* 290 * A non-synthesized event might not have a branch stack if 291 * branch stacks have been synthesized (using itrace options). 292 */ 293 if (!sample->branch_stack) 294 goto out_put; 295 296 iter.add_entry_cb = hist_iter__branch_callback; 297 iter.ops = &hist_iter_branch; 298 } else if (rep->mem_mode) { 299 iter.ops = &hist_iter_mem; 300 } else if (symbol_conf.cumulate_callchain) { 301 iter.ops = &hist_iter_cumulative; 302 } else { 303 iter.ops = &hist_iter_normal; 304 } 305 306 if (al.map != NULL) 307 al.map->dso->hit = 1; 308 309 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 310 hist__account_cycles(sample->branch_stack, &al, sample, 311 rep->nonany_branch_mode, 312 &rep->total_cycles); 313 } 314 315 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 316 if (ret < 0) 317 pr_debug("problem adding hist entry, skipping event\n"); 318 out_put: 319 addr_location__put(&al); 320 return ret; 321 } 322 323 static int process_read_event(struct perf_tool *tool, 324 union perf_event *event, 325 struct perf_sample *sample __maybe_unused, 326 struct evsel *evsel, 327 struct machine *machine __maybe_unused) 328 { 329 struct report *rep = container_of(tool, struct report, tool); 330 331 if (rep->show_threads) { 332 const char *name = evsel__name(evsel); 333 int err = perf_read_values_add_value(&rep->show_threads_values, 334 event->read.pid, event->read.tid, 335 evsel->core.idx, 336 name, 337 event->read.value); 338 339 if (err) 340 return err; 341 } 342 343 return 0; 344 } 345 346 /* For pipe mode, sample_type is not currently set */ 347 static int report__setup_sample_type(struct report *rep) 348 { 349 struct perf_session *session = rep->session; 350 u64 sample_type = evlist__combined_sample_type(session->evlist); 351 bool is_pipe = perf_data__is_pipe(session->data); 352 353 if (session->itrace_synth_opts->callchain || 354 session->itrace_synth_opts->add_callchain || 355 (!is_pipe && 356 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 357 !session->itrace_synth_opts->set)) 358 sample_type |= PERF_SAMPLE_CALLCHAIN; 359 360 if (session->itrace_synth_opts->last_branch || 361 session->itrace_synth_opts->add_last_branch) 362 sample_type |= PERF_SAMPLE_BRANCH_STACK; 363 364 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 365 if (perf_hpp_list.parent) { 366 ui__error("Selected --sort parent, but no " 367 "callchain data. Did you call " 368 "'perf record' without -g?\n"); 369 return -EINVAL; 370 } 371 if (symbol_conf.use_callchain && 372 !symbol_conf.show_branchflag_count) { 373 ui__error("Selected -g or --branch-history.\n" 374 "But no callchain or branch data.\n" 375 "Did you call 'perf record' without -g or -b?\n"); 376 return -1; 377 } 378 } else if (!callchain_param.enabled && 379 callchain_param.mode != CHAIN_NONE && 380 !symbol_conf.use_callchain) { 381 symbol_conf.use_callchain = true; 382 if (callchain_register_param(&callchain_param) < 0) { 383 ui__error("Can't register callchain params.\n"); 384 return -EINVAL; 385 } 386 } 387 388 if (symbol_conf.cumulate_callchain) { 389 /* Silently ignore if callchain is missing */ 390 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 391 symbol_conf.cumulate_callchain = false; 392 perf_hpp__cancel_cumulate(); 393 } 394 } 395 396 if (sort__mode == SORT_MODE__BRANCH) { 397 if (!is_pipe && 398 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 399 ui__error("Selected -b but no branch data. " 400 "Did you call perf record without -b?\n"); 401 return -1; 402 } 403 } 404 405 if (sort__mode == SORT_MODE__MEMORY) { 406 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 407 ui__error("Selected --mem-mode but no mem data. " 408 "Did you call perf record without -d?\n"); 409 return -1; 410 } 411 } 412 413 callchain_param_setup(sample_type); 414 415 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 416 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 417 "Please apply --call-graph lbr when recording.\n"); 418 rep->stitch_lbr = false; 419 } 420 421 /* ??? handle more cases than just ANY? */ 422 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 423 rep->nonany_branch_mode = true; 424 425 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT) 426 if (dwarf_callchain_users) { 427 ui__warning("Please install libunwind or libdw " 428 "development packages during the perf build.\n"); 429 } 430 #endif 431 432 return 0; 433 } 434 435 static void sig_handler(int sig __maybe_unused) 436 { 437 session_done = 1; 438 } 439 440 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 441 const char *evname, FILE *fp) 442 { 443 size_t ret; 444 char unit; 445 unsigned long nr_samples = hists->stats.nr_samples; 446 u64 nr_events = hists->stats.total_period; 447 struct evsel *evsel = hists_to_evsel(hists); 448 char buf[512]; 449 size_t size = sizeof(buf); 450 int socked_id = hists->socket_filter; 451 452 if (quiet) 453 return 0; 454 455 if (symbol_conf.filter_relative) { 456 nr_samples = hists->stats.nr_non_filtered_samples; 457 nr_events = hists->stats.total_non_filtered_period; 458 } 459 460 if (evsel__is_group_event(evsel)) { 461 struct evsel *pos; 462 463 evsel__group_desc(evsel, buf, size); 464 evname = buf; 465 466 for_each_group_member(pos, evsel) { 467 const struct hists *pos_hists = evsel__hists(pos); 468 469 if (symbol_conf.filter_relative) { 470 nr_samples += pos_hists->stats.nr_non_filtered_samples; 471 nr_events += pos_hists->stats.total_non_filtered_period; 472 } else { 473 nr_samples += pos_hists->stats.nr_samples; 474 nr_events += pos_hists->stats.total_period; 475 } 476 } 477 } 478 479 nr_samples = convert_unit(nr_samples, &unit); 480 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 481 if (evname != NULL) { 482 ret += fprintf(fp, " of event%s '%s'", 483 evsel->core.nr_members > 1 ? "s" : "", evname); 484 } 485 486 if (rep->time_str) 487 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 488 489 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 490 ret += fprintf(fp, ", show reference callgraph"); 491 } 492 493 if (rep->mem_mode) { 494 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 495 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 496 } else 497 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 498 499 if (socked_id > -1) 500 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 501 502 return ret + fprintf(fp, "\n#\n"); 503 } 504 505 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 506 { 507 struct evsel *pos; 508 int i = 0, ret; 509 510 evlist__for_each_entry(evlist, pos) { 511 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 512 rep->min_percent, pos, 513 &rep->session->header.env, 514 &rep->annotation_opts); 515 if (ret != 0) 516 return ret; 517 } 518 519 return 0; 520 } 521 522 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 523 { 524 struct evsel *pos; 525 int i = 0; 526 527 if (!quiet) { 528 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 529 evlist->stats.total_lost_samples); 530 } 531 532 evlist__for_each_entry(evlist, pos) { 533 struct hists *hists = evsel__hists(pos); 534 const char *evname = evsel__name(pos); 535 536 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 537 continue; 538 539 if (rep->skip_empty && !hists->stats.nr_samples) 540 continue; 541 542 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 543 544 if (rep->total_cycles_mode) { 545 report__browse_block_hists(&rep->block_reports[i++].hist, 546 rep->min_percent, pos, 547 NULL, NULL); 548 continue; 549 } 550 551 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 552 !(symbol_conf.use_callchain || 553 symbol_conf.show_branchflag_count)); 554 fprintf(stdout, "\n\n"); 555 } 556 557 if (!quiet) 558 fprintf(stdout, "#\n# (%s)\n#\n", help); 559 560 if (rep->show_threads) { 561 bool style = !strcmp(rep->pretty_printing_style, "raw"); 562 perf_read_values_display(stdout, &rep->show_threads_values, 563 style); 564 perf_read_values_destroy(&rep->show_threads_values); 565 } 566 567 if (sort__mode == SORT_MODE__BRANCH) 568 branch_type_stat_display(stdout, &rep->brtype_stat); 569 570 return 0; 571 } 572 573 static void report__warn_kptr_restrict(const struct report *rep) 574 { 575 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 576 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 577 578 if (evlist__exclude_kernel(rep->session->evlist)) 579 return; 580 581 if (kernel_map == NULL || 582 (kernel_map->dso->hit && 583 (kernel_kmap->ref_reloc_sym == NULL || 584 kernel_kmap->ref_reloc_sym->addr == 0))) { 585 const char *desc = 586 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 587 "can't be resolved."; 588 589 if (kernel_map && map__has_symbols(kernel_map)) { 590 desc = "If some relocation was applied (e.g. " 591 "kexec) symbols may be misresolved."; 592 } 593 594 ui__warning( 595 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 596 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 597 "Samples in kernel modules can't be resolved as well.\n\n", 598 desc); 599 } 600 } 601 602 static int report__gtk_browse_hists(struct report *rep, const char *help) 603 { 604 int (*hist_browser)(struct evlist *evlist, const char *help, 605 struct hist_browser_timer *timer, float min_pcnt); 606 607 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 608 609 if (hist_browser == NULL) { 610 ui__error("GTK browser not found!\n"); 611 return -1; 612 } 613 614 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 615 } 616 617 static int report__browse_hists(struct report *rep) 618 { 619 int ret; 620 struct perf_session *session = rep->session; 621 struct evlist *evlist = session->evlist; 622 const char *help = perf_tip(system_path(TIPDIR)); 623 624 if (help == NULL) { 625 /* fallback for people who don't install perf ;-) */ 626 help = perf_tip(DOCDIR); 627 if (help == NULL) 628 help = "Cannot load tips.txt file, please install perf!"; 629 } 630 631 switch (use_browser) { 632 case 1: 633 if (rep->total_cycles_mode) { 634 ret = evlist__tui_block_hists_browse(evlist, rep); 635 break; 636 } 637 638 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 639 &session->header.env, true, &rep->annotation_opts); 640 /* 641 * Usually "ret" is the last pressed key, and we only 642 * care if the key notifies us to switch data file. 643 */ 644 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 645 ret = 0; 646 break; 647 case 2: 648 ret = report__gtk_browse_hists(rep, help); 649 break; 650 default: 651 ret = evlist__tty_browse_hists(evlist, rep, help); 652 break; 653 } 654 655 return ret; 656 } 657 658 static int report__collapse_hists(struct report *rep) 659 { 660 struct ui_progress prog; 661 struct evsel *pos; 662 int ret = 0; 663 664 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 665 666 evlist__for_each_entry(rep->session->evlist, pos) { 667 struct hists *hists = evsel__hists(pos); 668 669 if (pos->core.idx == 0) 670 hists->symbol_filter_str = rep->symbol_filter_str; 671 672 hists->socket_filter = rep->socket_filter; 673 674 ret = hists__collapse_resort(hists, &prog); 675 if (ret < 0) 676 break; 677 678 /* Non-group events are considered as leader */ 679 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 680 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 681 682 hists__match(leader_hists, hists); 683 hists__link(leader_hists, hists); 684 } 685 } 686 687 ui_progress__finish(); 688 return ret; 689 } 690 691 static int hists__resort_cb(struct hist_entry *he, void *arg) 692 { 693 struct report *rep = arg; 694 struct symbol *sym = he->ms.sym; 695 696 if (rep->symbol_ipc && sym && !sym->annotate2) { 697 struct evsel *evsel = hists_to_evsel(he->hists); 698 699 symbol__annotate2(&he->ms, evsel, 700 &annotation__default_options, NULL); 701 } 702 703 return 0; 704 } 705 706 static void report__output_resort(struct report *rep) 707 { 708 struct ui_progress prog; 709 struct evsel *pos; 710 711 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 712 713 evlist__for_each_entry(rep->session->evlist, pos) { 714 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 715 } 716 717 ui_progress__finish(); 718 } 719 720 static int count_sample_event(struct perf_tool *tool __maybe_unused, 721 union perf_event *event __maybe_unused, 722 struct perf_sample *sample __maybe_unused, 723 struct evsel *evsel, 724 struct machine *machine __maybe_unused) 725 { 726 struct hists *hists = evsel__hists(evsel); 727 728 hists__inc_nr_events(hists); 729 return 0; 730 } 731 732 static int process_attr(struct perf_tool *tool __maybe_unused, 733 union perf_event *event, 734 struct evlist **pevlist); 735 736 static void stats_setup(struct report *rep) 737 { 738 memset(&rep->tool, 0, sizeof(rep->tool)); 739 rep->tool.attr = process_attr; 740 rep->tool.sample = count_sample_event; 741 rep->tool.no_warn = true; 742 } 743 744 static int stats_print(struct report *rep) 745 { 746 struct perf_session *session = rep->session; 747 748 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty); 749 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty); 750 return 0; 751 } 752 753 static void tasks_setup(struct report *rep) 754 { 755 memset(&rep->tool, 0, sizeof(rep->tool)); 756 rep->tool.ordered_events = true; 757 if (rep->mmaps_mode) { 758 rep->tool.mmap = perf_event__process_mmap; 759 rep->tool.mmap2 = perf_event__process_mmap2; 760 } 761 rep->tool.attr = process_attr; 762 rep->tool.comm = perf_event__process_comm; 763 rep->tool.exit = perf_event__process_exit; 764 rep->tool.fork = perf_event__process_fork; 765 rep->tool.no_warn = true; 766 } 767 768 struct task { 769 struct thread *thread; 770 struct list_head list; 771 struct list_head children; 772 }; 773 774 static struct task *tasks_list(struct task *task, struct machine *machine) 775 { 776 struct thread *parent_thread, *thread = task->thread; 777 struct task *parent_task; 778 779 /* Already listed. */ 780 if (!list_empty(&task->list)) 781 return NULL; 782 783 /* Last one in the chain. */ 784 if (thread->ppid == -1) 785 return task; 786 787 parent_thread = machine__find_thread(machine, -1, thread->ppid); 788 if (!parent_thread) 789 return ERR_PTR(-ENOENT); 790 791 parent_task = thread__priv(parent_thread); 792 list_add_tail(&task->list, &parent_task->children); 793 return tasks_list(parent_task, machine); 794 } 795 796 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 797 { 798 size_t printed = 0; 799 struct map *map; 800 801 maps__for_each_entry(maps, map) { 802 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 803 indent, "", map->start, map->end, 804 map->prot & PROT_READ ? 'r' : '-', 805 map->prot & PROT_WRITE ? 'w' : '-', 806 map->prot & PROT_EXEC ? 'x' : '-', 807 map->flags & MAP_SHARED ? 's' : 'p', 808 map->pgoff, 809 map->dso->id.ino, map->dso->name); 810 } 811 812 return printed; 813 } 814 815 static void task__print_level(struct task *task, FILE *fp, int level) 816 { 817 struct thread *thread = task->thread; 818 struct task *child; 819 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 820 thread->pid_, thread->tid, thread->ppid, 821 level, ""); 822 823 fprintf(fp, "%s\n", thread__comm_str(thread)); 824 825 maps__fprintf_task(thread->maps, comm_indent, fp); 826 827 if (!list_empty(&task->children)) { 828 list_for_each_entry(child, &task->children, list) 829 task__print_level(child, fp, level + 1); 830 } 831 } 832 833 static int tasks_print(struct report *rep, FILE *fp) 834 { 835 struct perf_session *session = rep->session; 836 struct machine *machine = &session->machines.host; 837 struct task *tasks, *task; 838 unsigned int nr = 0, itask = 0, i; 839 struct rb_node *nd; 840 LIST_HEAD(list); 841 842 /* 843 * No locking needed while accessing machine->threads, 844 * because --tasks is single threaded command. 845 */ 846 847 /* Count all the threads. */ 848 for (i = 0; i < THREADS__TABLE_SIZE; i++) 849 nr += machine->threads[i].nr; 850 851 tasks = malloc(sizeof(*tasks) * nr); 852 if (!tasks) 853 return -ENOMEM; 854 855 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 856 struct threads *threads = &machine->threads[i]; 857 858 for (nd = rb_first_cached(&threads->entries); nd; 859 nd = rb_next(nd)) { 860 task = tasks + itask++; 861 862 task->thread = rb_entry(nd, struct thread, rb_node); 863 INIT_LIST_HEAD(&task->children); 864 INIT_LIST_HEAD(&task->list); 865 thread__set_priv(task->thread, task); 866 } 867 } 868 869 /* 870 * Iterate every task down to the unprocessed parent 871 * and link all in task children list. Task with no 872 * parent is added into 'list'. 873 */ 874 for (itask = 0; itask < nr; itask++) { 875 task = tasks + itask; 876 877 if (!list_empty(&task->list)) 878 continue; 879 880 task = tasks_list(task, machine); 881 if (IS_ERR(task)) { 882 pr_err("Error: failed to process tasks\n"); 883 free(tasks); 884 return PTR_ERR(task); 885 } 886 887 if (task) 888 list_add_tail(&task->list, &list); 889 } 890 891 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 892 893 list_for_each_entry(task, &list, list) 894 task__print_level(task, fp, 0); 895 896 free(tasks); 897 return 0; 898 } 899 900 static int __cmd_report(struct report *rep) 901 { 902 int ret; 903 struct perf_session *session = rep->session; 904 struct evsel *pos; 905 struct perf_data *data = session->data; 906 907 signal(SIGINT, sig_handler); 908 909 if (rep->cpu_list) { 910 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 911 rep->cpu_bitmap); 912 if (ret) { 913 ui__error("failed to set cpu bitmap\n"); 914 return ret; 915 } 916 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 917 } 918 919 if (rep->show_threads) { 920 ret = perf_read_values_init(&rep->show_threads_values); 921 if (ret) 922 return ret; 923 } 924 925 ret = report__setup_sample_type(rep); 926 if (ret) { 927 /* report__setup_sample_type() already showed error message */ 928 return ret; 929 } 930 931 if (rep->stats_mode) 932 stats_setup(rep); 933 934 if (rep->tasks_mode) 935 tasks_setup(rep); 936 937 ret = perf_session__process_events(session); 938 if (ret) { 939 ui__error("failed to process sample\n"); 940 return ret; 941 } 942 943 evlist__check_mem_load_aux(session->evlist); 944 945 if (rep->stats_mode) 946 return stats_print(rep); 947 948 if (rep->tasks_mode) 949 return tasks_print(rep, stdout); 950 951 report__warn_kptr_restrict(rep); 952 953 evlist__for_each_entry(session->evlist, pos) 954 rep->nr_entries += evsel__hists(pos)->nr_entries; 955 956 if (use_browser == 0) { 957 if (verbose > 3) 958 perf_session__fprintf(session, stdout); 959 960 if (verbose > 2) 961 perf_session__fprintf_dsos(session, stdout); 962 963 if (dump_trace) { 964 perf_session__fprintf_nr_events(session, stdout, 965 rep->skip_empty); 966 evlist__fprintf_nr_events(session->evlist, stdout, 967 rep->skip_empty); 968 return 0; 969 } 970 } 971 972 ret = report__collapse_hists(rep); 973 if (ret) { 974 ui__error("failed to process hist entry\n"); 975 return ret; 976 } 977 978 if (session_done()) 979 return 0; 980 981 /* 982 * recalculate number of entries after collapsing since it 983 * might be changed during the collapse phase. 984 */ 985 rep->nr_entries = 0; 986 evlist__for_each_entry(session->evlist, pos) 987 rep->nr_entries += evsel__hists(pos)->nr_entries; 988 989 if (rep->nr_entries == 0) { 990 ui__error("The %s data has no samples!\n", data->path); 991 return 0; 992 } 993 994 report__output_resort(rep); 995 996 if (rep->total_cycles_mode) { 997 int block_hpps[6] = { 998 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 999 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1000 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1001 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1002 PERF_HPP_REPORT__BLOCK_RANGE, 1003 PERF_HPP_REPORT__BLOCK_DSO, 1004 }; 1005 1006 rep->block_reports = block_info__create_report(session->evlist, 1007 rep->total_cycles, 1008 block_hpps, 6, 1009 &rep->nr_block_reports); 1010 if (!rep->block_reports) 1011 return -1; 1012 } 1013 1014 return report__browse_hists(rep); 1015 } 1016 1017 static int 1018 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1019 { 1020 struct callchain_param *callchain = opt->value; 1021 1022 callchain->enabled = !unset; 1023 /* 1024 * --no-call-graph 1025 */ 1026 if (unset) { 1027 symbol_conf.use_callchain = false; 1028 callchain->mode = CHAIN_NONE; 1029 return 0; 1030 } 1031 1032 return parse_callchain_report_opt(arg); 1033 } 1034 1035 static int 1036 parse_time_quantum(const struct option *opt, const char *arg, 1037 int unset __maybe_unused) 1038 { 1039 unsigned long *time_q = opt->value; 1040 char *end; 1041 1042 *time_q = strtoul(arg, &end, 0); 1043 if (end == arg) 1044 goto parse_err; 1045 if (*time_q == 0) { 1046 pr_err("time quantum cannot be 0"); 1047 return -1; 1048 } 1049 end = skip_spaces(end); 1050 if (*end == 0) 1051 return 0; 1052 if (!strcmp(end, "s")) { 1053 *time_q *= NSEC_PER_SEC; 1054 return 0; 1055 } 1056 if (!strcmp(end, "ms")) { 1057 *time_q *= NSEC_PER_MSEC; 1058 return 0; 1059 } 1060 if (!strcmp(end, "us")) { 1061 *time_q *= NSEC_PER_USEC; 1062 return 0; 1063 } 1064 if (!strcmp(end, "ns")) 1065 return 0; 1066 parse_err: 1067 pr_err("Cannot parse time quantum `%s'\n", arg); 1068 return -1; 1069 } 1070 1071 int 1072 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1073 const char *arg, int unset __maybe_unused) 1074 { 1075 if (arg) { 1076 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1077 if (err) { 1078 char buf[BUFSIZ]; 1079 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1080 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1081 return -1; 1082 } 1083 have_ignore_callees = 1; 1084 } 1085 1086 return 0; 1087 } 1088 1089 static int 1090 parse_branch_mode(const struct option *opt, 1091 const char *str __maybe_unused, int unset) 1092 { 1093 int *branch_mode = opt->value; 1094 1095 *branch_mode = !unset; 1096 return 0; 1097 } 1098 1099 static int 1100 parse_percent_limit(const struct option *opt, const char *str, 1101 int unset __maybe_unused) 1102 { 1103 struct report *rep = opt->value; 1104 double pcnt = strtof(str, NULL); 1105 1106 rep->min_percent = pcnt; 1107 callchain_param.min_percent = pcnt; 1108 return 0; 1109 } 1110 1111 static int process_attr(struct perf_tool *tool __maybe_unused, 1112 union perf_event *event, 1113 struct evlist **pevlist) 1114 { 1115 u64 sample_type; 1116 int err; 1117 1118 err = perf_event__process_attr(tool, event, pevlist); 1119 if (err) 1120 return err; 1121 1122 /* 1123 * Check if we need to enable callchains based 1124 * on events sample_type. 1125 */ 1126 sample_type = evlist__combined_sample_type(*pevlist); 1127 callchain_param_setup(sample_type); 1128 return 0; 1129 } 1130 1131 int cmd_report(int argc, const char **argv) 1132 { 1133 struct perf_session *session; 1134 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1135 struct stat st; 1136 bool has_br_stack = false; 1137 int branch_mode = -1; 1138 int last_key = 0; 1139 bool branch_call_mode = false; 1140 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1141 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1142 CALLCHAIN_REPORT_HELP 1143 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1144 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1145 const char * const report_usage[] = { 1146 "perf report [<options>]", 1147 NULL 1148 }; 1149 struct report report = { 1150 .tool = { 1151 .sample = process_sample_event, 1152 .mmap = perf_event__process_mmap, 1153 .mmap2 = perf_event__process_mmap2, 1154 .comm = perf_event__process_comm, 1155 .namespaces = perf_event__process_namespaces, 1156 .cgroup = perf_event__process_cgroup, 1157 .exit = perf_event__process_exit, 1158 .fork = perf_event__process_fork, 1159 .lost = perf_event__process_lost, 1160 .read = process_read_event, 1161 .attr = process_attr, 1162 .tracing_data = perf_event__process_tracing_data, 1163 .build_id = perf_event__process_build_id, 1164 .id_index = perf_event__process_id_index, 1165 .auxtrace_info = perf_event__process_auxtrace_info, 1166 .auxtrace = perf_event__process_auxtrace, 1167 .event_update = perf_event__process_event_update, 1168 .feature = process_feature_event, 1169 .ordered_events = true, 1170 .ordering_requires_timestamps = true, 1171 }, 1172 .max_stack = PERF_MAX_STACK_DEPTH, 1173 .pretty_printing_style = "normal", 1174 .socket_filter = -1, 1175 .annotation_opts = annotation__default_options, 1176 .skip_empty = true, 1177 }; 1178 char *sort_order_help = sort_help("sort by key(s):"); 1179 char *field_order_help = sort_help("output field(s): overhead period sample "); 1180 const struct option options[] = { 1181 OPT_STRING('i', "input", &input_name, "file", 1182 "input file name"), 1183 OPT_INCR('v', "verbose", &verbose, 1184 "be more verbose (show symbol address, etc)"), 1185 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"), 1186 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1187 "dump raw trace in ASCII"), 1188 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1189 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1190 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1191 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1192 "file", "vmlinux pathname"), 1193 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1194 "don't load vmlinux even if found"), 1195 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1196 "file", "kallsyms pathname"), 1197 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1198 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1199 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1200 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1201 "Show a column with the number of samples"), 1202 OPT_BOOLEAN('T', "threads", &report.show_threads, 1203 "Show per-thread event counters"), 1204 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1205 "pretty printing style key: normal raw"), 1206 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1207 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1208 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1209 "Use the stdio interface"), 1210 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1211 OPT_BOOLEAN(0, "header-only", &report.header_only, 1212 "Show only data header."), 1213 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1214 sort_order_help), 1215 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1216 field_order_help), 1217 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1218 "Show sample percentage for different cpu modes"), 1219 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1220 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1221 OPT_STRING('p', "parent", &parent_pattern, "regex", 1222 "regex filter to identify parent, see: '--sort parent'"), 1223 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1224 "Only display entries with parent-match"), 1225 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1226 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1227 report_callchain_help, &report_parse_callchain_opt, 1228 callchain_default_opt), 1229 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1230 "Accumulate callchains of children and show total overhead as well. " 1231 "Enabled by default, use --no-children to disable."), 1232 OPT_INTEGER(0, "max-stack", &report.max_stack, 1233 "Set the maximum stack depth when parsing the callchain, " 1234 "anything beyond the specified depth will be ignored. " 1235 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1236 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1237 "alias for inverted call graph"), 1238 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1239 "ignore callees of these functions in call graphs", 1240 report_parse_ignore_callees_opt), 1241 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1242 "only consider symbols in these dsos"), 1243 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1244 "only consider symbols in these comms"), 1245 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1246 "only consider symbols in these pids"), 1247 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1248 "only consider symbols in these tids"), 1249 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1250 "only consider these symbols"), 1251 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1252 "only show symbols that (partially) match with this filter"), 1253 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1254 "width[,width...]", 1255 "don't try to adjust column width, use these fixed values"), 1256 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1257 "separator for columns, no spaces will be added between " 1258 "columns '.' is reserved."), 1259 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1260 "Only display entries resolved to a symbol"), 1261 OPT_CALLBACK(0, "symfs", NULL, "directory", 1262 "Look for files with symbols relative to this directory", 1263 symbol__config_symfs), 1264 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1265 "list of cpus to profile"), 1266 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1267 "Display extended information about perf.data file"), 1268 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src, 1269 "Interleave source code with assembly code (default)"), 1270 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw, 1271 "Display raw encoding of assembly instructions (default)"), 1272 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style", 1273 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1274 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix", 1275 "Add prefix to source file path names in programs (with --prefix-strip)"), 1276 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N", 1277 "Strip first N entries of source file path name in programs (with --prefix)"), 1278 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1279 "Show a column with the sum of periods"), 1280 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1281 "Show event group information together"), 1282 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1283 "Sort the output by the event at the index n in group. " 1284 "If n is invalid, sort by the first event. " 1285 "WARNING: should be used on grouped events."), 1286 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1287 "use branch records for per branch histogram filling", 1288 parse_branch_mode), 1289 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1290 "add last branch records to call history"), 1291 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path", 1292 "objdump binary to use for disassembly and annotations"), 1293 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1294 "Disable symbol demangling"), 1295 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1296 "Enable kernel symbol demangling"), 1297 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1298 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1299 "Number of samples to save per histogram entry for individual browsing"), 1300 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1301 "Don't show entries under that percent", parse_percent_limit), 1302 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1303 "how to display percentage of filtered entries", parse_filter_percentage), 1304 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1305 "Instruction Tracing options\n" ITRACE_HELP, 1306 itrace_parse_synth_opts), 1307 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1308 "Show full source file name path for source lines"), 1309 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1310 "Show callgraph from reference event"), 1311 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1312 "Enable LBR callgraph stitching approach"), 1313 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1314 "only show processor socket that match with this filter"), 1315 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1316 "Show raw trace event output (do not use print fmt or plugins)"), 1317 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1318 "Show entries in a hierarchy"), 1319 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1320 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1321 stdio__config_color, "always"), 1322 OPT_STRING(0, "time", &report.time_str, "str", 1323 "Time span of interest (start,stop)"), 1324 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1325 "Show inline function"), 1326 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period", 1327 "Set percent type local/global-period/hits", 1328 annotate_parse_percent_type), 1329 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1330 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1331 "Set time quantum for time sort key (default 100ms)", 1332 parse_time_quantum), 1333 OPTS_EVSWITCH(&report.evswitch), 1334 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1335 "Sort all blocks by 'Sampled Cycles%'"), 1336 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1337 "Disable raw trace ordering"), 1338 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1339 "Do not display empty (or dummy) events in the output"), 1340 OPT_END() 1341 }; 1342 struct perf_data data = { 1343 .mode = PERF_DATA_MODE_READ, 1344 }; 1345 int ret = hists__init(); 1346 char sort_tmp[128]; 1347 1348 if (ret < 0) 1349 goto exit; 1350 1351 ret = perf_config(report__config, &report); 1352 if (ret) 1353 goto exit; 1354 1355 argc = parse_options(argc, argv, options, report_usage, 0); 1356 if (argc) { 1357 /* 1358 * Special case: if there's an argument left then assume that 1359 * it's a symbol filter: 1360 */ 1361 if (argc > 1) 1362 usage_with_options(report_usage, options); 1363 1364 report.symbol_filter_str = argv[0]; 1365 } 1366 1367 if (annotate_check_args(&report.annotation_opts) < 0) { 1368 ret = -EINVAL; 1369 goto exit; 1370 } 1371 1372 if (report.mmaps_mode) 1373 report.tasks_mode = true; 1374 1375 if (dump_trace && report.disable_order) 1376 report.tool.ordered_events = false; 1377 1378 if (quiet) 1379 perf_quiet_option(); 1380 1381 ret = symbol__validate_sym_arguments(); 1382 if (ret) 1383 goto exit; 1384 1385 if (report.inverted_callchain) 1386 callchain_param.order = ORDER_CALLER; 1387 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1388 callchain_param.order = ORDER_CALLER; 1389 1390 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1391 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1392 report.max_stack = itrace_synth_opts.callchain_sz; 1393 1394 if (!input_name || !strlen(input_name)) { 1395 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1396 input_name = "-"; 1397 else 1398 input_name = "perf.data"; 1399 } 1400 1401 data.path = input_name; 1402 data.force = symbol_conf.force; 1403 1404 repeat: 1405 session = perf_session__new(&data, &report.tool); 1406 if (IS_ERR(session)) { 1407 ret = PTR_ERR(session); 1408 goto exit; 1409 } 1410 1411 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1412 if (ret) 1413 goto exit; 1414 1415 if (zstd_init(&(session->zstd_data), 0) < 0) 1416 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1417 1418 if (report.queue_size) { 1419 ordered_events__set_alloc_size(&session->ordered_events, 1420 report.queue_size); 1421 } 1422 1423 session->itrace_synth_opts = &itrace_synth_opts; 1424 1425 report.session = session; 1426 1427 has_br_stack = perf_header__has_feat(&session->header, 1428 HEADER_BRANCH_STACK); 1429 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1430 has_br_stack = false; 1431 1432 setup_forced_leader(&report, session->evlist); 1433 1434 if (symbol_conf.group_sort_idx && !session->evlist->core.nr_groups) { 1435 parse_options_usage(NULL, options, "group-sort-idx", 0); 1436 ret = -EINVAL; 1437 goto error; 1438 } 1439 1440 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1441 has_br_stack = true; 1442 1443 if (has_br_stack && branch_call_mode) 1444 symbol_conf.show_branchflag_count = true; 1445 1446 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1447 1448 /* 1449 * Branch mode is a tristate: 1450 * -1 means default, so decide based on the file having branch data. 1451 * 0/1 means the user chose a mode. 1452 */ 1453 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1454 !branch_call_mode) { 1455 sort__mode = SORT_MODE__BRANCH; 1456 symbol_conf.cumulate_callchain = false; 1457 } 1458 if (branch_call_mode) { 1459 callchain_param.key = CCKEY_ADDRESS; 1460 callchain_param.branch_callstack = true; 1461 symbol_conf.use_callchain = true; 1462 callchain_register_param(&callchain_param); 1463 if (sort_order == NULL) 1464 sort_order = "srcline,symbol,dso"; 1465 } 1466 1467 if (report.mem_mode) { 1468 if (sort__mode == SORT_MODE__BRANCH) { 1469 pr_err("branch and mem mode incompatible\n"); 1470 goto error; 1471 } 1472 sort__mode = SORT_MODE__MEMORY; 1473 symbol_conf.cumulate_callchain = false; 1474 } 1475 1476 if (symbol_conf.report_hierarchy) { 1477 /* disable incompatible options */ 1478 symbol_conf.cumulate_callchain = false; 1479 1480 if (field_order) { 1481 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1482 parse_options_usage(report_usage, options, "F", 1); 1483 parse_options_usage(NULL, options, "hierarchy", 0); 1484 goto error; 1485 } 1486 1487 perf_hpp_list.need_collapse = true; 1488 } 1489 1490 if (report.use_stdio) 1491 use_browser = 0; 1492 else if (report.use_tui) 1493 use_browser = 1; 1494 else if (report.use_gtk) 1495 use_browser = 2; 1496 1497 /* Force tty output for header output and per-thread stat. */ 1498 if (report.header || report.header_only || report.show_threads) 1499 use_browser = 0; 1500 if (report.header || report.header_only) 1501 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1502 if (report.show_full_info) 1503 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1504 if (report.stats_mode || report.tasks_mode) 1505 use_browser = 0; 1506 if (report.stats_mode && report.tasks_mode) { 1507 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1508 goto error; 1509 } 1510 1511 if (report.total_cycles_mode) { 1512 if (sort__mode != SORT_MODE__BRANCH) 1513 report.total_cycles_mode = false; 1514 else 1515 sort_order = NULL; 1516 } 1517 1518 if (strcmp(input_name, "-") != 0) 1519 setup_browser(true); 1520 else 1521 use_browser = 0; 1522 1523 if (sort_order && strstr(sort_order, "ipc")) { 1524 parse_options_usage(report_usage, options, "s", 1); 1525 goto error; 1526 } 1527 1528 if (sort_order && strstr(sort_order, "symbol")) { 1529 if (sort__mode == SORT_MODE__BRANCH) { 1530 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1531 sort_order, "ipc_lbr"); 1532 report.symbol_ipc = true; 1533 } else { 1534 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1535 sort_order, "ipc_null"); 1536 } 1537 1538 sort_order = sort_tmp; 1539 } 1540 1541 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1542 (setup_sorting(session->evlist) < 0)) { 1543 if (sort_order) 1544 parse_options_usage(report_usage, options, "s", 1); 1545 if (field_order) 1546 parse_options_usage(sort_order ? NULL : report_usage, 1547 options, "F", 1); 1548 goto error; 1549 } 1550 1551 if ((report.header || report.header_only) && !quiet) { 1552 perf_session__fprintf_info(session, stdout, 1553 report.show_full_info); 1554 if (report.header_only) { 1555 if (data.is_pipe) { 1556 /* 1557 * we need to process first few records 1558 * which contains PERF_RECORD_HEADER_FEATURE. 1559 */ 1560 perf_session__process_events(session); 1561 } 1562 ret = 0; 1563 goto error; 1564 } 1565 } else if (use_browser == 0 && !quiet && 1566 !report.stats_mode && !report.tasks_mode) { 1567 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1568 stdout); 1569 } 1570 1571 /* 1572 * Only in the TUI browser we are doing integrated annotation, 1573 * so don't allocate extra space that won't be used in the stdio 1574 * implementation. 1575 */ 1576 if (ui__has_annotation() || report.symbol_ipc || 1577 report.total_cycles_mode) { 1578 ret = symbol__annotation_init(); 1579 if (ret < 0) 1580 goto error; 1581 /* 1582 * For searching by name on the "Browse map details". 1583 * providing it only in verbose mode not to bloat too 1584 * much struct symbol. 1585 */ 1586 if (verbose > 0) { 1587 /* 1588 * XXX: Need to provide a less kludgy way to ask for 1589 * more space per symbol, the u32 is for the index on 1590 * the ui browser. 1591 * See symbol__browser_index. 1592 */ 1593 symbol_conf.priv_size += sizeof(u32); 1594 symbol_conf.sort_by_name = true; 1595 } 1596 annotation_config__init(&report.annotation_opts); 1597 } 1598 1599 if (symbol__init(&session->header.env) < 0) 1600 goto error; 1601 1602 if (report.time_str) { 1603 ret = perf_time__parse_for_ranges(report.time_str, session, 1604 &report.ptime_range, 1605 &report.range_size, 1606 &report.range_num); 1607 if (ret < 0) 1608 goto error; 1609 1610 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1611 report.ptime_range, 1612 report.range_num); 1613 } 1614 1615 if (session->tevent.pevent && 1616 tep_set_function_resolver(session->tevent.pevent, 1617 machine__resolve_kernel_addr, 1618 &session->machines.host) < 0) { 1619 pr_err("%s: failed to set libtraceevent function resolver\n", 1620 __func__); 1621 return -1; 1622 } 1623 1624 sort__setup_elide(stdout); 1625 1626 ret = __cmd_report(&report); 1627 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1628 perf_session__delete(session); 1629 last_key = K_SWITCH_INPUT_DATA; 1630 goto repeat; 1631 } else 1632 ret = 0; 1633 1634 error: 1635 if (report.ptime_range) { 1636 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1637 zfree(&report.ptime_range); 1638 } 1639 1640 if (report.block_reports) { 1641 block_info__free_report(report.block_reports, 1642 report.nr_block_reports); 1643 report.block_reports = NULL; 1644 } 1645 1646 zstd_fini(&(session->zstd_data)); 1647 perf_session__delete(session); 1648 exit: 1649 free(sort_order_help); 1650 free(field_order_help); 1651 return ret; 1652 } 1653