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 #ifdef HAVE_SLANG_SUPPORT 75 bool use_tui; 76 #endif 77 bool use_gtk; 78 bool use_stdio; 79 bool show_full_info; 80 bool show_threads; 81 bool inverted_callchain; 82 bool mem_mode; 83 bool stats_mode; 84 bool tasks_mode; 85 bool mmaps_mode; 86 bool header; 87 bool header_only; 88 bool nonany_branch_mode; 89 bool group_set; 90 bool stitch_lbr; 91 bool disable_order; 92 bool skip_empty; 93 int max_stack; 94 struct perf_read_values show_threads_values; 95 struct annotation_options annotation_opts; 96 const char *pretty_printing_style; 97 const char *cpu_list; 98 const char *symbol_filter_str; 99 const char *time_str; 100 struct perf_time_interval *ptime_range; 101 int range_size; 102 int range_num; 103 float min_percent; 104 u64 nr_entries; 105 u64 queue_size; 106 u64 total_cycles; 107 int socket_filter; 108 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 109 struct branch_type_stat brtype_stat; 110 bool symbol_ipc; 111 bool total_cycles_mode; 112 struct block_report *block_reports; 113 int nr_block_reports; 114 }; 115 116 static int report__config(const char *var, const char *value, void *cb) 117 { 118 struct report *rep = cb; 119 120 if (!strcmp(var, "report.group")) { 121 symbol_conf.event_group = perf_config_bool(var, value); 122 return 0; 123 } 124 if (!strcmp(var, "report.percent-limit")) { 125 double pcnt = strtof(value, NULL); 126 127 rep->min_percent = pcnt; 128 callchain_param.min_percent = pcnt; 129 return 0; 130 } 131 if (!strcmp(var, "report.children")) { 132 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 133 return 0; 134 } 135 if (!strcmp(var, "report.queue-size")) 136 return perf_config_u64(&rep->queue_size, var, value); 137 138 if (!strcmp(var, "report.sort_order")) { 139 default_sort_order = strdup(value); 140 return 0; 141 } 142 143 if (!strcmp(var, "report.skip-empty")) { 144 rep->skip_empty = perf_config_bool(var, value); 145 return 0; 146 } 147 148 return 0; 149 } 150 151 static int hist_iter__report_callback(struct hist_entry_iter *iter, 152 struct addr_location *al, bool single, 153 void *arg) 154 { 155 int err = 0; 156 struct report *rep = arg; 157 struct hist_entry *he = iter->he; 158 struct evsel *evsel = iter->evsel; 159 struct perf_sample *sample = iter->sample; 160 struct mem_info *mi; 161 struct branch_info *bi; 162 163 if (!ui__has_annotation() && !rep->symbol_ipc) 164 return 0; 165 166 if (sort__mode == SORT_MODE__BRANCH) { 167 bi = he->branch_info; 168 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 169 if (err) 170 goto out; 171 172 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 173 174 } else if (rep->mem_mode) { 175 mi = he->mem_info; 176 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel); 177 if (err) 178 goto out; 179 180 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 181 182 } else if (symbol_conf.cumulate_callchain) { 183 if (single) 184 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 185 } else { 186 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 187 } 188 189 out: 190 return err; 191 } 192 193 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 194 struct addr_location *al __maybe_unused, 195 bool single __maybe_unused, 196 void *arg) 197 { 198 struct hist_entry *he = iter->he; 199 struct report *rep = arg; 200 struct branch_info *bi = he->branch_info; 201 struct perf_sample *sample = iter->sample; 202 struct evsel *evsel = iter->evsel; 203 int err; 204 205 branch_type_count(&rep->brtype_stat, &bi->flags, 206 bi->from.addr, bi->to.addr); 207 208 if (!ui__has_annotation() && !rep->symbol_ipc) 209 return 0; 210 211 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 212 if (err) 213 goto out; 214 215 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 216 217 out: 218 return err; 219 } 220 221 static void setup_forced_leader(struct report *report, 222 struct evlist *evlist) 223 { 224 if (report->group_set) 225 evlist__force_leader(evlist); 226 } 227 228 static int process_feature_event(struct perf_session *session, 229 union perf_event *event) 230 { 231 struct report *rep = container_of(session->tool, struct report, tool); 232 233 if (event->feat.feat_id < HEADER_LAST_FEATURE) 234 return perf_event__process_feature(session, event); 235 236 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 237 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 238 event->feat.feat_id); 239 return -1; 240 } else if (rep->header_only) { 241 session_done = 1; 242 } 243 244 /* 245 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 246 * means all features are received, now we can force the 247 * group if needed. 248 */ 249 setup_forced_leader(rep, session->evlist); 250 return 0; 251 } 252 253 static int process_sample_event(struct perf_tool *tool, 254 union perf_event *event, 255 struct perf_sample *sample, 256 struct evsel *evsel, 257 struct machine *machine) 258 { 259 struct report *rep = container_of(tool, struct report, tool); 260 struct addr_location al; 261 struct hist_entry_iter iter = { 262 .evsel = evsel, 263 .sample = sample, 264 .hide_unresolved = symbol_conf.hide_unresolved, 265 .add_entry_cb = hist_iter__report_callback, 266 }; 267 int ret = 0; 268 269 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 270 sample->time)) { 271 return 0; 272 } 273 274 if (evswitch__discard(&rep->evswitch, evsel)) 275 return 0; 276 277 if (machine__resolve(machine, &al, sample) < 0) { 278 pr_debug("problem processing %d event, skipping it.\n", 279 event->header.type); 280 return -1; 281 } 282 283 if (rep->stitch_lbr) 284 al.thread->lbr_stitch_enable = true; 285 286 if (symbol_conf.hide_unresolved && al.sym == NULL) 287 goto out_put; 288 289 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 290 goto out_put; 291 292 if (sort__mode == SORT_MODE__BRANCH) { 293 /* 294 * A non-synthesized event might not have a branch stack if 295 * branch stacks have been synthesized (using itrace options). 296 */ 297 if (!sample->branch_stack) 298 goto out_put; 299 300 iter.add_entry_cb = hist_iter__branch_callback; 301 iter.ops = &hist_iter_branch; 302 } else if (rep->mem_mode) { 303 iter.ops = &hist_iter_mem; 304 } else if (symbol_conf.cumulate_callchain) { 305 iter.ops = &hist_iter_cumulative; 306 } else { 307 iter.ops = &hist_iter_normal; 308 } 309 310 if (al.map != NULL) 311 al.map->dso->hit = 1; 312 313 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 314 hist__account_cycles(sample->branch_stack, &al, sample, 315 rep->nonany_branch_mode, 316 &rep->total_cycles); 317 } 318 319 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 320 if (ret < 0) 321 pr_debug("problem adding hist entry, skipping event\n"); 322 out_put: 323 addr_location__put(&al); 324 return ret; 325 } 326 327 static int process_read_event(struct perf_tool *tool, 328 union perf_event *event, 329 struct perf_sample *sample __maybe_unused, 330 struct evsel *evsel, 331 struct machine *machine __maybe_unused) 332 { 333 struct report *rep = container_of(tool, struct report, tool); 334 335 if (rep->show_threads) { 336 const char *name = evsel__name(evsel); 337 int err = perf_read_values_add_value(&rep->show_threads_values, 338 event->read.pid, event->read.tid, 339 evsel->core.idx, 340 name, 341 event->read.value); 342 343 if (err) 344 return err; 345 } 346 347 return 0; 348 } 349 350 /* For pipe mode, sample_type is not currently set */ 351 static int report__setup_sample_type(struct report *rep) 352 { 353 struct perf_session *session = rep->session; 354 u64 sample_type = evlist__combined_sample_type(session->evlist); 355 bool is_pipe = perf_data__is_pipe(session->data); 356 357 if (session->itrace_synth_opts->callchain || 358 session->itrace_synth_opts->add_callchain || 359 (!is_pipe && 360 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 361 !session->itrace_synth_opts->set)) 362 sample_type |= PERF_SAMPLE_CALLCHAIN; 363 364 if (session->itrace_synth_opts->last_branch || 365 session->itrace_synth_opts->add_last_branch) 366 sample_type |= PERF_SAMPLE_BRANCH_STACK; 367 368 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 369 if (perf_hpp_list.parent) { 370 ui__error("Selected --sort parent, but no " 371 "callchain data. Did you call " 372 "'perf record' without -g?\n"); 373 return -EINVAL; 374 } 375 if (symbol_conf.use_callchain && 376 !symbol_conf.show_branchflag_count) { 377 ui__error("Selected -g or --branch-history.\n" 378 "But no callchain or branch data.\n" 379 "Did you call 'perf record' without -g or -b?\n"); 380 return -1; 381 } 382 } else if (!callchain_param.enabled && 383 callchain_param.mode != CHAIN_NONE && 384 !symbol_conf.use_callchain) { 385 symbol_conf.use_callchain = true; 386 if (callchain_register_param(&callchain_param) < 0) { 387 ui__error("Can't register callchain params.\n"); 388 return -EINVAL; 389 } 390 } 391 392 if (symbol_conf.cumulate_callchain) { 393 /* Silently ignore if callchain is missing */ 394 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 395 symbol_conf.cumulate_callchain = false; 396 perf_hpp__cancel_cumulate(); 397 } 398 } 399 400 if (sort__mode == SORT_MODE__BRANCH) { 401 if (!is_pipe && 402 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 403 ui__error("Selected -b but no branch data. " 404 "Did you call perf record without -b?\n"); 405 return -1; 406 } 407 } 408 409 if (sort__mode == SORT_MODE__MEMORY) { 410 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 411 ui__error("Selected --mem-mode but no mem data. " 412 "Did you call perf record without -d?\n"); 413 return -1; 414 } 415 } 416 417 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env)); 418 419 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 420 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 421 "Please apply --call-graph lbr when recording.\n"); 422 rep->stitch_lbr = false; 423 } 424 425 /* ??? handle more cases than just ANY? */ 426 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 427 rep->nonany_branch_mode = true; 428 429 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT) 430 if (dwarf_callchain_users) { 431 ui__warning("Please install libunwind or libdw " 432 "development packages during the perf build.\n"); 433 } 434 #endif 435 436 return 0; 437 } 438 439 static void sig_handler(int sig __maybe_unused) 440 { 441 session_done = 1; 442 } 443 444 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 445 const char *evname, FILE *fp) 446 { 447 size_t ret; 448 char unit; 449 unsigned long nr_samples = hists->stats.nr_samples; 450 u64 nr_events = hists->stats.total_period; 451 struct evsel *evsel = hists_to_evsel(hists); 452 char buf[512]; 453 size_t size = sizeof(buf); 454 int socked_id = hists->socket_filter; 455 456 if (quiet) 457 return 0; 458 459 if (symbol_conf.filter_relative) { 460 nr_samples = hists->stats.nr_non_filtered_samples; 461 nr_events = hists->stats.total_non_filtered_period; 462 } 463 464 if (evsel__is_group_event(evsel)) { 465 struct evsel *pos; 466 467 evsel__group_desc(evsel, buf, size); 468 evname = buf; 469 470 for_each_group_member(pos, evsel) { 471 const struct hists *pos_hists = evsel__hists(pos); 472 473 if (symbol_conf.filter_relative) { 474 nr_samples += pos_hists->stats.nr_non_filtered_samples; 475 nr_events += pos_hists->stats.total_non_filtered_period; 476 } else { 477 nr_samples += pos_hists->stats.nr_samples; 478 nr_events += pos_hists->stats.total_period; 479 } 480 } 481 } 482 483 nr_samples = convert_unit(nr_samples, &unit); 484 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 485 if (evname != NULL) { 486 ret += fprintf(fp, " of event%s '%s'", 487 evsel->core.nr_members > 1 ? "s" : "", evname); 488 } 489 490 if (rep->time_str) 491 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 492 493 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 494 ret += fprintf(fp, ", show reference callgraph"); 495 } 496 497 if (rep->mem_mode) { 498 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 499 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 500 } else 501 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 502 503 if (socked_id > -1) 504 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 505 506 return ret + fprintf(fp, "\n#\n"); 507 } 508 509 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 510 { 511 struct evsel *pos; 512 int i = 0, ret; 513 514 evlist__for_each_entry(evlist, pos) { 515 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 516 rep->min_percent, pos, 517 &rep->session->header.env, 518 &rep->annotation_opts); 519 if (ret != 0) 520 return ret; 521 } 522 523 return 0; 524 } 525 526 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 527 { 528 struct evsel *pos; 529 int i = 0; 530 531 if (!quiet) { 532 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 533 evlist->stats.total_lost_samples); 534 } 535 536 evlist__for_each_entry(evlist, pos) { 537 struct hists *hists = evsel__hists(pos); 538 const char *evname = evsel__name(pos); 539 540 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 541 continue; 542 543 if (rep->skip_empty && !hists->stats.nr_samples) 544 continue; 545 546 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 547 548 if (rep->total_cycles_mode) { 549 report__browse_block_hists(&rep->block_reports[i++].hist, 550 rep->min_percent, pos, 551 NULL, NULL); 552 continue; 553 } 554 555 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 556 !(symbol_conf.use_callchain || 557 symbol_conf.show_branchflag_count)); 558 fprintf(stdout, "\n\n"); 559 } 560 561 if (!quiet) 562 fprintf(stdout, "#\n# (%s)\n#\n", help); 563 564 if (rep->show_threads) { 565 bool style = !strcmp(rep->pretty_printing_style, "raw"); 566 perf_read_values_display(stdout, &rep->show_threads_values, 567 style); 568 perf_read_values_destroy(&rep->show_threads_values); 569 } 570 571 if (sort__mode == SORT_MODE__BRANCH) 572 branch_type_stat_display(stdout, &rep->brtype_stat); 573 574 return 0; 575 } 576 577 static void report__warn_kptr_restrict(const struct report *rep) 578 { 579 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 580 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 581 582 if (evlist__exclude_kernel(rep->session->evlist)) 583 return; 584 585 if (kernel_map == NULL || 586 (kernel_map->dso->hit && 587 (kernel_kmap->ref_reloc_sym == NULL || 588 kernel_kmap->ref_reloc_sym->addr == 0))) { 589 const char *desc = 590 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 591 "can't be resolved."; 592 593 if (kernel_map && map__has_symbols(kernel_map)) { 594 desc = "If some relocation was applied (e.g. " 595 "kexec) symbols may be misresolved."; 596 } 597 598 ui__warning( 599 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 600 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 601 "Samples in kernel modules can't be resolved as well.\n\n", 602 desc); 603 } 604 } 605 606 static int report__gtk_browse_hists(struct report *rep, const char *help) 607 { 608 int (*hist_browser)(struct evlist *evlist, const char *help, 609 struct hist_browser_timer *timer, float min_pcnt); 610 611 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 612 613 if (hist_browser == NULL) { 614 ui__error("GTK browser not found!\n"); 615 return -1; 616 } 617 618 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 619 } 620 621 static int report__browse_hists(struct report *rep) 622 { 623 int ret; 624 struct perf_session *session = rep->session; 625 struct evlist *evlist = session->evlist; 626 char *help = NULL, *path = NULL; 627 628 path = system_path(TIPDIR); 629 if (perf_tip(&help, path) || help == NULL) { 630 /* fallback for people who don't install perf ;-) */ 631 free(path); 632 path = system_path(DOCDIR); 633 if (perf_tip(&help, path) || help == NULL) 634 help = strdup("Cannot load tips.txt file, please install perf!"); 635 } 636 free(path); 637 638 switch (use_browser) { 639 case 1: 640 if (rep->total_cycles_mode) { 641 ret = evlist__tui_block_hists_browse(evlist, rep); 642 break; 643 } 644 645 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 646 &session->header.env, true, &rep->annotation_opts); 647 /* 648 * Usually "ret" is the last pressed key, and we only 649 * care if the key notifies us to switch data file. 650 */ 651 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 652 ret = 0; 653 break; 654 case 2: 655 ret = report__gtk_browse_hists(rep, help); 656 break; 657 default: 658 ret = evlist__tty_browse_hists(evlist, rep, help); 659 break; 660 } 661 free(help); 662 return ret; 663 } 664 665 static int report__collapse_hists(struct report *rep) 666 { 667 struct ui_progress prog; 668 struct evsel *pos; 669 int ret = 0; 670 671 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 672 673 evlist__for_each_entry(rep->session->evlist, pos) { 674 struct hists *hists = evsel__hists(pos); 675 676 if (pos->core.idx == 0) 677 hists->symbol_filter_str = rep->symbol_filter_str; 678 679 hists->socket_filter = rep->socket_filter; 680 681 ret = hists__collapse_resort(hists, &prog); 682 if (ret < 0) 683 break; 684 685 /* Non-group events are considered as leader */ 686 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 687 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 688 689 hists__match(leader_hists, hists); 690 hists__link(leader_hists, hists); 691 } 692 } 693 694 ui_progress__finish(); 695 return ret; 696 } 697 698 static int hists__resort_cb(struct hist_entry *he, void *arg) 699 { 700 struct report *rep = arg; 701 struct symbol *sym = he->ms.sym; 702 703 if (rep->symbol_ipc && sym && !sym->annotate2) { 704 struct evsel *evsel = hists_to_evsel(he->hists); 705 706 symbol__annotate2(&he->ms, evsel, 707 &annotation__default_options, NULL); 708 } 709 710 return 0; 711 } 712 713 static void report__output_resort(struct report *rep) 714 { 715 struct ui_progress prog; 716 struct evsel *pos; 717 718 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 719 720 evlist__for_each_entry(rep->session->evlist, pos) { 721 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 722 } 723 724 ui_progress__finish(); 725 } 726 727 static int count_sample_event(struct perf_tool *tool __maybe_unused, 728 union perf_event *event __maybe_unused, 729 struct perf_sample *sample __maybe_unused, 730 struct evsel *evsel, 731 struct machine *machine __maybe_unused) 732 { 733 struct hists *hists = evsel__hists(evsel); 734 735 hists__inc_nr_events(hists); 736 return 0; 737 } 738 739 static int process_attr(struct perf_tool *tool __maybe_unused, 740 union perf_event *event, 741 struct evlist **pevlist); 742 743 static void stats_setup(struct report *rep) 744 { 745 memset(&rep->tool, 0, sizeof(rep->tool)); 746 rep->tool.attr = process_attr; 747 rep->tool.sample = count_sample_event; 748 rep->tool.no_warn = true; 749 } 750 751 static int stats_print(struct report *rep) 752 { 753 struct perf_session *session = rep->session; 754 755 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty); 756 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty); 757 return 0; 758 } 759 760 static void tasks_setup(struct report *rep) 761 { 762 memset(&rep->tool, 0, sizeof(rep->tool)); 763 rep->tool.ordered_events = true; 764 if (rep->mmaps_mode) { 765 rep->tool.mmap = perf_event__process_mmap; 766 rep->tool.mmap2 = perf_event__process_mmap2; 767 } 768 rep->tool.attr = process_attr; 769 rep->tool.comm = perf_event__process_comm; 770 rep->tool.exit = perf_event__process_exit; 771 rep->tool.fork = perf_event__process_fork; 772 rep->tool.no_warn = true; 773 } 774 775 struct task { 776 struct thread *thread; 777 struct list_head list; 778 struct list_head children; 779 }; 780 781 static struct task *tasks_list(struct task *task, struct machine *machine) 782 { 783 struct thread *parent_thread, *thread = task->thread; 784 struct task *parent_task; 785 786 /* Already listed. */ 787 if (!list_empty(&task->list)) 788 return NULL; 789 790 /* Last one in the chain. */ 791 if (thread->ppid == -1) 792 return task; 793 794 parent_thread = machine__find_thread(machine, -1, thread->ppid); 795 if (!parent_thread) 796 return ERR_PTR(-ENOENT); 797 798 parent_task = thread__priv(parent_thread); 799 list_add_tail(&task->list, &parent_task->children); 800 return tasks_list(parent_task, machine); 801 } 802 803 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 804 { 805 size_t printed = 0; 806 struct map *map; 807 808 maps__for_each_entry(maps, map) { 809 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 810 indent, "", map->start, map->end, 811 map->prot & PROT_READ ? 'r' : '-', 812 map->prot & PROT_WRITE ? 'w' : '-', 813 map->prot & PROT_EXEC ? 'x' : '-', 814 map->flags & MAP_SHARED ? 's' : 'p', 815 map->pgoff, 816 map->dso->id.ino, map->dso->name); 817 } 818 819 return printed; 820 } 821 822 static void task__print_level(struct task *task, FILE *fp, int level) 823 { 824 struct thread *thread = task->thread; 825 struct task *child; 826 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 827 thread->pid_, thread->tid, thread->ppid, 828 level, ""); 829 830 fprintf(fp, "%s\n", thread__comm_str(thread)); 831 832 maps__fprintf_task(thread->maps, comm_indent, fp); 833 834 if (!list_empty(&task->children)) { 835 list_for_each_entry(child, &task->children, list) 836 task__print_level(child, fp, level + 1); 837 } 838 } 839 840 static int tasks_print(struct report *rep, FILE *fp) 841 { 842 struct perf_session *session = rep->session; 843 struct machine *machine = &session->machines.host; 844 struct task *tasks, *task; 845 unsigned int nr = 0, itask = 0, i; 846 struct rb_node *nd; 847 LIST_HEAD(list); 848 849 /* 850 * No locking needed while accessing machine->threads, 851 * because --tasks is single threaded command. 852 */ 853 854 /* Count all the threads. */ 855 for (i = 0; i < THREADS__TABLE_SIZE; i++) 856 nr += machine->threads[i].nr; 857 858 tasks = malloc(sizeof(*tasks) * nr); 859 if (!tasks) 860 return -ENOMEM; 861 862 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 863 struct threads *threads = &machine->threads[i]; 864 865 for (nd = rb_first_cached(&threads->entries); nd; 866 nd = rb_next(nd)) { 867 task = tasks + itask++; 868 869 task->thread = rb_entry(nd, struct thread, rb_node); 870 INIT_LIST_HEAD(&task->children); 871 INIT_LIST_HEAD(&task->list); 872 thread__set_priv(task->thread, task); 873 } 874 } 875 876 /* 877 * Iterate every task down to the unprocessed parent 878 * and link all in task children list. Task with no 879 * parent is added into 'list'. 880 */ 881 for (itask = 0; itask < nr; itask++) { 882 task = tasks + itask; 883 884 if (!list_empty(&task->list)) 885 continue; 886 887 task = tasks_list(task, machine); 888 if (IS_ERR(task)) { 889 pr_err("Error: failed to process tasks\n"); 890 free(tasks); 891 return PTR_ERR(task); 892 } 893 894 if (task) 895 list_add_tail(&task->list, &list); 896 } 897 898 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 899 900 list_for_each_entry(task, &list, list) 901 task__print_level(task, fp, 0); 902 903 free(tasks); 904 return 0; 905 } 906 907 static int __cmd_report(struct report *rep) 908 { 909 int ret; 910 struct perf_session *session = rep->session; 911 struct evsel *pos; 912 struct perf_data *data = session->data; 913 914 signal(SIGINT, sig_handler); 915 916 if (rep->cpu_list) { 917 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 918 rep->cpu_bitmap); 919 if (ret) { 920 ui__error("failed to set cpu bitmap\n"); 921 return ret; 922 } 923 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 924 } 925 926 if (rep->show_threads) { 927 ret = perf_read_values_init(&rep->show_threads_values); 928 if (ret) 929 return ret; 930 } 931 932 ret = report__setup_sample_type(rep); 933 if (ret) { 934 /* report__setup_sample_type() already showed error message */ 935 return ret; 936 } 937 938 if (rep->stats_mode) 939 stats_setup(rep); 940 941 if (rep->tasks_mode) 942 tasks_setup(rep); 943 944 ret = perf_session__process_events(session); 945 if (ret) { 946 ui__error("failed to process sample\n"); 947 return ret; 948 } 949 950 evlist__check_mem_load_aux(session->evlist); 951 952 if (rep->stats_mode) 953 return stats_print(rep); 954 955 if (rep->tasks_mode) 956 return tasks_print(rep, stdout); 957 958 report__warn_kptr_restrict(rep); 959 960 evlist__for_each_entry(session->evlist, pos) 961 rep->nr_entries += evsel__hists(pos)->nr_entries; 962 963 if (use_browser == 0) { 964 if (verbose > 3) 965 perf_session__fprintf(session, stdout); 966 967 if (verbose > 2) 968 perf_session__fprintf_dsos(session, stdout); 969 970 if (dump_trace) { 971 perf_session__fprintf_nr_events(session, stdout, 972 rep->skip_empty); 973 evlist__fprintf_nr_events(session->evlist, stdout, 974 rep->skip_empty); 975 return 0; 976 } 977 } 978 979 ret = report__collapse_hists(rep); 980 if (ret) { 981 ui__error("failed to process hist entry\n"); 982 return ret; 983 } 984 985 if (session_done()) 986 return 0; 987 988 /* 989 * recalculate number of entries after collapsing since it 990 * might be changed during the collapse phase. 991 */ 992 rep->nr_entries = 0; 993 evlist__for_each_entry(session->evlist, pos) 994 rep->nr_entries += evsel__hists(pos)->nr_entries; 995 996 if (rep->nr_entries == 0) { 997 ui__error("The %s data has no samples!\n", data->path); 998 return 0; 999 } 1000 1001 report__output_resort(rep); 1002 1003 if (rep->total_cycles_mode) { 1004 int block_hpps[6] = { 1005 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1006 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1007 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1008 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1009 PERF_HPP_REPORT__BLOCK_RANGE, 1010 PERF_HPP_REPORT__BLOCK_DSO, 1011 }; 1012 1013 rep->block_reports = block_info__create_report(session->evlist, 1014 rep->total_cycles, 1015 block_hpps, 6, 1016 &rep->nr_block_reports); 1017 if (!rep->block_reports) 1018 return -1; 1019 } 1020 1021 return report__browse_hists(rep); 1022 } 1023 1024 static int 1025 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1026 { 1027 struct callchain_param *callchain = opt->value; 1028 1029 callchain->enabled = !unset; 1030 /* 1031 * --no-call-graph 1032 */ 1033 if (unset) { 1034 symbol_conf.use_callchain = false; 1035 callchain->mode = CHAIN_NONE; 1036 return 0; 1037 } 1038 1039 return parse_callchain_report_opt(arg); 1040 } 1041 1042 static int 1043 parse_time_quantum(const struct option *opt, const char *arg, 1044 int unset __maybe_unused) 1045 { 1046 unsigned long *time_q = opt->value; 1047 char *end; 1048 1049 *time_q = strtoul(arg, &end, 0); 1050 if (end == arg) 1051 goto parse_err; 1052 if (*time_q == 0) { 1053 pr_err("time quantum cannot be 0"); 1054 return -1; 1055 } 1056 end = skip_spaces(end); 1057 if (*end == 0) 1058 return 0; 1059 if (!strcmp(end, "s")) { 1060 *time_q *= NSEC_PER_SEC; 1061 return 0; 1062 } 1063 if (!strcmp(end, "ms")) { 1064 *time_q *= NSEC_PER_MSEC; 1065 return 0; 1066 } 1067 if (!strcmp(end, "us")) { 1068 *time_q *= NSEC_PER_USEC; 1069 return 0; 1070 } 1071 if (!strcmp(end, "ns")) 1072 return 0; 1073 parse_err: 1074 pr_err("Cannot parse time quantum `%s'\n", arg); 1075 return -1; 1076 } 1077 1078 int 1079 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1080 const char *arg, int unset __maybe_unused) 1081 { 1082 if (arg) { 1083 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1084 if (err) { 1085 char buf[BUFSIZ]; 1086 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1087 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1088 return -1; 1089 } 1090 have_ignore_callees = 1; 1091 } 1092 1093 return 0; 1094 } 1095 1096 static int 1097 parse_branch_mode(const struct option *opt, 1098 const char *str __maybe_unused, int unset) 1099 { 1100 int *branch_mode = opt->value; 1101 1102 *branch_mode = !unset; 1103 return 0; 1104 } 1105 1106 static int 1107 parse_percent_limit(const struct option *opt, const char *str, 1108 int unset __maybe_unused) 1109 { 1110 struct report *rep = opt->value; 1111 double pcnt = strtof(str, NULL); 1112 1113 rep->min_percent = pcnt; 1114 callchain_param.min_percent = pcnt; 1115 return 0; 1116 } 1117 1118 static int process_attr(struct perf_tool *tool __maybe_unused, 1119 union perf_event *event, 1120 struct evlist **pevlist) 1121 { 1122 u64 sample_type; 1123 int err; 1124 1125 err = perf_event__process_attr(tool, event, pevlist); 1126 if (err) 1127 return err; 1128 1129 /* 1130 * Check if we need to enable callchains based 1131 * on events sample_type. 1132 */ 1133 sample_type = evlist__combined_sample_type(*pevlist); 1134 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env)); 1135 return 0; 1136 } 1137 1138 int cmd_report(int argc, const char **argv) 1139 { 1140 struct perf_session *session; 1141 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1142 struct stat st; 1143 bool has_br_stack = false; 1144 int branch_mode = -1; 1145 int last_key = 0; 1146 bool branch_call_mode = false; 1147 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1148 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1149 CALLCHAIN_REPORT_HELP 1150 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1151 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1152 const char * const report_usage[] = { 1153 "perf report [<options>]", 1154 NULL 1155 }; 1156 struct report report = { 1157 .tool = { 1158 .sample = process_sample_event, 1159 .mmap = perf_event__process_mmap, 1160 .mmap2 = perf_event__process_mmap2, 1161 .comm = perf_event__process_comm, 1162 .namespaces = perf_event__process_namespaces, 1163 .cgroup = perf_event__process_cgroup, 1164 .exit = perf_event__process_exit, 1165 .fork = perf_event__process_fork, 1166 .lost = perf_event__process_lost, 1167 .read = process_read_event, 1168 .attr = process_attr, 1169 .tracing_data = perf_event__process_tracing_data, 1170 .build_id = perf_event__process_build_id, 1171 .id_index = perf_event__process_id_index, 1172 .auxtrace_info = perf_event__process_auxtrace_info, 1173 .auxtrace = perf_event__process_auxtrace, 1174 .event_update = perf_event__process_event_update, 1175 .feature = process_feature_event, 1176 .ordered_events = true, 1177 .ordering_requires_timestamps = true, 1178 }, 1179 .max_stack = PERF_MAX_STACK_DEPTH, 1180 .pretty_printing_style = "normal", 1181 .socket_filter = -1, 1182 .annotation_opts = annotation__default_options, 1183 .skip_empty = true, 1184 }; 1185 char *sort_order_help = sort_help("sort by key(s):"); 1186 char *field_order_help = sort_help("output field(s): overhead period sample "); 1187 const struct option options[] = { 1188 OPT_STRING('i', "input", &input_name, "file", 1189 "input file name"), 1190 OPT_INCR('v', "verbose", &verbose, 1191 "be more verbose (show symbol address, etc)"), 1192 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"), 1193 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1194 "dump raw trace in ASCII"), 1195 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1196 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1197 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1198 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1199 "file", "vmlinux pathname"), 1200 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1201 "don't load vmlinux even if found"), 1202 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1203 "file", "kallsyms pathname"), 1204 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1205 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1206 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1207 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1208 "Show a column with the number of samples"), 1209 OPT_BOOLEAN('T', "threads", &report.show_threads, 1210 "Show per-thread event counters"), 1211 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1212 "pretty printing style key: normal raw"), 1213 #ifdef HAVE_SLANG_SUPPORT 1214 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1215 #endif 1216 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1217 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1218 "Use the stdio interface"), 1219 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1220 OPT_BOOLEAN(0, "header-only", &report.header_only, 1221 "Show only data header."), 1222 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1223 sort_order_help), 1224 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1225 field_order_help), 1226 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1227 "Show sample percentage for different cpu modes"), 1228 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1229 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1230 OPT_STRING('p', "parent", &parent_pattern, "regex", 1231 "regex filter to identify parent, see: '--sort parent'"), 1232 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1233 "Only display entries with parent-match"), 1234 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1235 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1236 report_callchain_help, &report_parse_callchain_opt, 1237 callchain_default_opt), 1238 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1239 "Accumulate callchains of children and show total overhead as well. " 1240 "Enabled by default, use --no-children to disable."), 1241 OPT_INTEGER(0, "max-stack", &report.max_stack, 1242 "Set the maximum stack depth when parsing the callchain, " 1243 "anything beyond the specified depth will be ignored. " 1244 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1245 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1246 "alias for inverted call graph"), 1247 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1248 "ignore callees of these functions in call graphs", 1249 report_parse_ignore_callees_opt), 1250 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1251 "only consider symbols in these dsos"), 1252 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1253 "only consider symbols in these comms"), 1254 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1255 "only consider symbols in these pids"), 1256 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1257 "only consider symbols in these tids"), 1258 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1259 "only consider these symbols"), 1260 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1261 "only show symbols that (partially) match with this filter"), 1262 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1263 "width[,width...]", 1264 "don't try to adjust column width, use these fixed values"), 1265 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1266 "separator for columns, no spaces will be added between " 1267 "columns '.' is reserved."), 1268 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1269 "Only display entries resolved to a symbol"), 1270 OPT_CALLBACK(0, "symfs", NULL, "directory", 1271 "Look for files with symbols relative to this directory", 1272 symbol__config_symfs), 1273 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1274 "list of cpus to profile"), 1275 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1276 "Display extended information about perf.data file"), 1277 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src, 1278 "Interleave source code with assembly code (default)"), 1279 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw, 1280 "Display raw encoding of assembly instructions (default)"), 1281 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style", 1282 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1283 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix", 1284 "Add prefix to source file path names in programs (with --prefix-strip)"), 1285 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N", 1286 "Strip first N entries of source file path name in programs (with --prefix)"), 1287 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1288 "Show a column with the sum of periods"), 1289 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1290 "Show event group information together"), 1291 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1292 "Sort the output by the event at the index n in group. " 1293 "If n is invalid, sort by the first event. " 1294 "WARNING: should be used on grouped events."), 1295 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1296 "use branch records for per branch histogram filling", 1297 parse_branch_mode), 1298 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1299 "add last branch records to call history"), 1300 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path", 1301 "objdump binary to use for disassembly and annotations"), 1302 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1303 "Disable symbol demangling"), 1304 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1305 "Enable kernel symbol demangling"), 1306 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1307 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1308 "Number of samples to save per histogram entry for individual browsing"), 1309 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1310 "Don't show entries under that percent", parse_percent_limit), 1311 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1312 "how to display percentage of filtered entries", parse_filter_percentage), 1313 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1314 "Instruction Tracing options\n" ITRACE_HELP, 1315 itrace_parse_synth_opts), 1316 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1317 "Show full source file name path for source lines"), 1318 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1319 "Show callgraph from reference event"), 1320 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1321 "Enable LBR callgraph stitching approach"), 1322 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1323 "only show processor socket that match with this filter"), 1324 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1325 "Show raw trace event output (do not use print fmt or plugins)"), 1326 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1327 "Show entries in a hierarchy"), 1328 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1329 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1330 stdio__config_color, "always"), 1331 OPT_STRING(0, "time", &report.time_str, "str", 1332 "Time span of interest (start,stop)"), 1333 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1334 "Show inline function"), 1335 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period", 1336 "Set percent type local/global-period/hits", 1337 annotate_parse_percent_type), 1338 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1339 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1340 "Set time quantum for time sort key (default 100ms)", 1341 parse_time_quantum), 1342 OPTS_EVSWITCH(&report.evswitch), 1343 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1344 "Sort all blocks by 'Sampled Cycles%'"), 1345 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1346 "Disable raw trace ordering"), 1347 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1348 "Do not display empty (or dummy) events in the output"), 1349 OPT_END() 1350 }; 1351 struct perf_data data = { 1352 .mode = PERF_DATA_MODE_READ, 1353 }; 1354 int ret = hists__init(); 1355 char sort_tmp[128]; 1356 1357 if (ret < 0) 1358 goto exit; 1359 1360 ret = perf_config(report__config, &report); 1361 if (ret) 1362 goto exit; 1363 1364 argc = parse_options(argc, argv, options, report_usage, 0); 1365 if (argc) { 1366 /* 1367 * Special case: if there's an argument left then assume that 1368 * it's a symbol filter: 1369 */ 1370 if (argc > 1) 1371 usage_with_options(report_usage, options); 1372 1373 report.symbol_filter_str = argv[0]; 1374 } 1375 1376 if (annotate_check_args(&report.annotation_opts) < 0) { 1377 ret = -EINVAL; 1378 goto exit; 1379 } 1380 1381 if (report.mmaps_mode) 1382 report.tasks_mode = true; 1383 1384 if (dump_trace && report.disable_order) 1385 report.tool.ordered_events = false; 1386 1387 if (quiet) 1388 perf_quiet_option(); 1389 1390 ret = symbol__validate_sym_arguments(); 1391 if (ret) 1392 goto exit; 1393 1394 if (report.inverted_callchain) 1395 callchain_param.order = ORDER_CALLER; 1396 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1397 callchain_param.order = ORDER_CALLER; 1398 1399 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1400 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1401 report.max_stack = itrace_synth_opts.callchain_sz; 1402 1403 if (!input_name || !strlen(input_name)) { 1404 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1405 input_name = "-"; 1406 else 1407 input_name = "perf.data"; 1408 } 1409 1410 data.path = input_name; 1411 data.force = symbol_conf.force; 1412 1413 repeat: 1414 session = perf_session__new(&data, &report.tool); 1415 if (IS_ERR(session)) { 1416 ret = PTR_ERR(session); 1417 goto exit; 1418 } 1419 1420 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1421 if (ret) 1422 goto exit; 1423 1424 if (zstd_init(&(session->zstd_data), 0) < 0) 1425 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1426 1427 if (report.queue_size) { 1428 ordered_events__set_alloc_size(&session->ordered_events, 1429 report.queue_size); 1430 } 1431 1432 session->itrace_synth_opts = &itrace_synth_opts; 1433 1434 report.session = session; 1435 1436 has_br_stack = perf_header__has_feat(&session->header, 1437 HEADER_BRANCH_STACK); 1438 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1439 has_br_stack = false; 1440 1441 setup_forced_leader(&report, session->evlist); 1442 1443 if (symbol_conf.group_sort_idx && !session->evlist->core.nr_groups) { 1444 parse_options_usage(NULL, options, "group-sort-idx", 0); 1445 ret = -EINVAL; 1446 goto error; 1447 } 1448 1449 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1450 has_br_stack = true; 1451 1452 if (has_br_stack && branch_call_mode) 1453 symbol_conf.show_branchflag_count = true; 1454 1455 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1456 1457 /* 1458 * Branch mode is a tristate: 1459 * -1 means default, so decide based on the file having branch data. 1460 * 0/1 means the user chose a mode. 1461 */ 1462 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1463 !branch_call_mode) { 1464 sort__mode = SORT_MODE__BRANCH; 1465 symbol_conf.cumulate_callchain = false; 1466 } 1467 if (branch_call_mode) { 1468 callchain_param.key = CCKEY_ADDRESS; 1469 callchain_param.branch_callstack = true; 1470 symbol_conf.use_callchain = true; 1471 callchain_register_param(&callchain_param); 1472 if (sort_order == NULL) 1473 sort_order = "srcline,symbol,dso"; 1474 } 1475 1476 if (report.mem_mode) { 1477 if (sort__mode == SORT_MODE__BRANCH) { 1478 pr_err("branch and mem mode incompatible\n"); 1479 goto error; 1480 } 1481 sort__mode = SORT_MODE__MEMORY; 1482 symbol_conf.cumulate_callchain = false; 1483 } 1484 1485 if (symbol_conf.report_hierarchy) { 1486 /* disable incompatible options */ 1487 symbol_conf.cumulate_callchain = false; 1488 1489 if (field_order) { 1490 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1491 parse_options_usage(report_usage, options, "F", 1); 1492 parse_options_usage(NULL, options, "hierarchy", 0); 1493 goto error; 1494 } 1495 1496 perf_hpp_list.need_collapse = true; 1497 } 1498 1499 if (report.use_stdio) 1500 use_browser = 0; 1501 #ifdef HAVE_SLANG_SUPPORT 1502 else if (report.use_tui) 1503 use_browser = 1; 1504 #endif 1505 else if (report.use_gtk) 1506 use_browser = 2; 1507 1508 /* Force tty output for header output and per-thread stat. */ 1509 if (report.header || report.header_only || report.show_threads) 1510 use_browser = 0; 1511 if (report.header || report.header_only) 1512 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1513 if (report.show_full_info) 1514 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1515 if (report.stats_mode || report.tasks_mode) 1516 use_browser = 0; 1517 if (report.stats_mode && report.tasks_mode) { 1518 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1519 goto error; 1520 } 1521 1522 if (report.total_cycles_mode) { 1523 if (sort__mode != SORT_MODE__BRANCH) 1524 report.total_cycles_mode = false; 1525 else 1526 sort_order = NULL; 1527 } 1528 1529 if (strcmp(input_name, "-") != 0) 1530 setup_browser(true); 1531 else 1532 use_browser = 0; 1533 1534 if (sort_order && strstr(sort_order, "ipc")) { 1535 parse_options_usage(report_usage, options, "s", 1); 1536 goto error; 1537 } 1538 1539 if (sort_order && strstr(sort_order, "symbol")) { 1540 if (sort__mode == SORT_MODE__BRANCH) { 1541 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1542 sort_order, "ipc_lbr"); 1543 report.symbol_ipc = true; 1544 } else { 1545 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1546 sort_order, "ipc_null"); 1547 } 1548 1549 sort_order = sort_tmp; 1550 } 1551 1552 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1553 (setup_sorting(session->evlist) < 0)) { 1554 if (sort_order) 1555 parse_options_usage(report_usage, options, "s", 1); 1556 if (field_order) 1557 parse_options_usage(sort_order ? NULL : report_usage, 1558 options, "F", 1); 1559 goto error; 1560 } 1561 1562 if ((report.header || report.header_only) && !quiet) { 1563 perf_session__fprintf_info(session, stdout, 1564 report.show_full_info); 1565 if (report.header_only) { 1566 if (data.is_pipe) { 1567 /* 1568 * we need to process first few records 1569 * which contains PERF_RECORD_HEADER_FEATURE. 1570 */ 1571 perf_session__process_events(session); 1572 } 1573 ret = 0; 1574 goto error; 1575 } 1576 } else if (use_browser == 0 && !quiet && 1577 !report.stats_mode && !report.tasks_mode) { 1578 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1579 stdout); 1580 } 1581 1582 /* 1583 * Only in the TUI browser we are doing integrated annotation, 1584 * so don't allocate extra space that won't be used in the stdio 1585 * implementation. 1586 */ 1587 if (ui__has_annotation() || report.symbol_ipc || 1588 report.total_cycles_mode) { 1589 ret = symbol__annotation_init(); 1590 if (ret < 0) 1591 goto error; 1592 /* 1593 * For searching by name on the "Browse map details". 1594 * providing it only in verbose mode not to bloat too 1595 * much struct symbol. 1596 */ 1597 if (verbose > 0) { 1598 /* 1599 * XXX: Need to provide a less kludgy way to ask for 1600 * more space per symbol, the u32 is for the index on 1601 * the ui browser. 1602 * See symbol__browser_index. 1603 */ 1604 symbol_conf.priv_size += sizeof(u32); 1605 symbol_conf.sort_by_name = true; 1606 } 1607 annotation_config__init(&report.annotation_opts); 1608 } 1609 1610 if (symbol__init(&session->header.env) < 0) 1611 goto error; 1612 1613 if (report.time_str) { 1614 ret = perf_time__parse_for_ranges(report.time_str, session, 1615 &report.ptime_range, 1616 &report.range_size, 1617 &report.range_num); 1618 if (ret < 0) 1619 goto error; 1620 1621 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1622 report.ptime_range, 1623 report.range_num); 1624 } 1625 1626 if (session->tevent.pevent && 1627 tep_set_function_resolver(session->tevent.pevent, 1628 machine__resolve_kernel_addr, 1629 &session->machines.host) < 0) { 1630 pr_err("%s: failed to set libtraceevent function resolver\n", 1631 __func__); 1632 return -1; 1633 } 1634 1635 sort__setup_elide(stdout); 1636 1637 ret = __cmd_report(&report); 1638 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1639 perf_session__delete(session); 1640 last_key = K_SWITCH_INPUT_DATA; 1641 goto repeat; 1642 } else 1643 ret = 0; 1644 1645 error: 1646 if (report.ptime_range) { 1647 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1648 zfree(&report.ptime_range); 1649 } 1650 1651 if (report.block_reports) { 1652 block_info__free_report(report.block_reports, 1653 report.nr_block_reports); 1654 report.block_reports = NULL; 1655 } 1656 1657 zstd_fini(&(session->zstd_data)); 1658 perf_session__delete(session); 1659 exit: 1660 free(sort_order_help); 1661 free(field_order_help); 1662 return ret; 1663 } 1664