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