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