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/mem-info.h" 35 #include "util/session.h" 36 #include "util/srcline.h" 37 #include "util/tool.h" 38 39 #include <subcmd/parse-options.h> 40 #include <subcmd/exec-cmd.h> 41 #include "util/parse-events.h" 42 43 #include "util/thread.h" 44 #include "util/sort.h" 45 #include "util/hist.h" 46 #include "util/data.h" 47 #include "arch/common.h" 48 #include "util/time-utils.h" 49 #include "util/auxtrace.h" 50 #include "util/units.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/list_sort.h> 64 #include <linux/string.h> 65 #include <linux/stringify.h> 66 #include <linux/time64.h> 67 #include <sys/types.h> 68 #include <sys/stat.h> 69 #include <unistd.h> 70 #include <linux/mman.h> 71 72 #ifdef HAVE_LIBTRACEEVENT 73 #include <event-parse.h> 74 #endif 75 76 struct report { 77 struct perf_tool tool; 78 struct perf_session *session; 79 struct evswitch evswitch; 80 #ifdef HAVE_SLANG_SUPPORT 81 bool use_tui; 82 #endif 83 #ifdef HAVE_GTK2_SUPPORT 84 bool use_gtk; 85 #endif 86 bool use_stdio; 87 bool show_full_info; 88 bool show_threads; 89 bool inverted_callchain; 90 bool mem_mode; 91 bool stats_mode; 92 bool tasks_mode; 93 bool mmaps_mode; 94 bool header; 95 bool header_only; 96 bool nonany_branch_mode; 97 bool group_set; 98 bool stitch_lbr; 99 bool disable_order; 100 bool skip_empty; 101 bool data_type; 102 int max_stack; 103 struct perf_read_values show_threads_values; 104 const char *pretty_printing_style; 105 const char *cpu_list; 106 const char *symbol_filter_str; 107 const char *time_str; 108 struct perf_time_interval *ptime_range; 109 int range_size; 110 int range_num; 111 float min_percent; 112 u64 nr_entries; 113 u64 queue_size; 114 u64 total_cycles; 115 int socket_filter; 116 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 117 struct branch_type_stat brtype_stat; 118 bool symbol_ipc; 119 bool total_cycles_mode; 120 struct block_report *block_reports; 121 int nr_block_reports; 122 }; 123 124 static int report__config(const char *var, const char *value, void *cb) 125 { 126 struct report *rep = cb; 127 128 if (!strcmp(var, "report.group")) { 129 symbol_conf.event_group = perf_config_bool(var, value); 130 return 0; 131 } 132 if (!strcmp(var, "report.percent-limit")) { 133 double pcnt = strtof(value, NULL); 134 135 rep->min_percent = pcnt; 136 callchain_param.min_percent = pcnt; 137 return 0; 138 } 139 if (!strcmp(var, "report.children")) { 140 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 141 return 0; 142 } 143 if (!strcmp(var, "report.queue-size")) 144 return perf_config_u64(&rep->queue_size, var, value); 145 146 if (!strcmp(var, "report.sort_order")) { 147 default_sort_order = strdup(value); 148 if (!default_sort_order) { 149 pr_err("Not enough memory for report.sort_order\n"); 150 return -1; 151 } 152 return 0; 153 } 154 155 if (!strcmp(var, "report.skip-empty")) { 156 rep->skip_empty = perf_config_bool(var, value); 157 return 0; 158 } 159 160 pr_debug("%s variable unknown, ignoring...", var); 161 return 0; 162 } 163 164 static int hist_iter__report_callback(struct hist_entry_iter *iter, 165 struct addr_location *al, bool single, 166 void *arg) 167 { 168 int err = 0; 169 struct report *rep = arg; 170 struct hist_entry *he = iter->he; 171 struct evsel *evsel = iter->evsel; 172 struct perf_sample *sample = iter->sample; 173 struct mem_info *mi; 174 struct branch_info *bi; 175 176 if (!ui__has_annotation() && !rep->symbol_ipc) 177 return 0; 178 179 if (sort__mode == SORT_MODE__BRANCH) { 180 bi = he->branch_info; 181 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 182 if (err) 183 goto out; 184 185 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 186 187 } else if (rep->mem_mode) { 188 mi = he->mem_info; 189 err = addr_map_symbol__inc_samples(mem_info__daddr(mi), sample, evsel); 190 if (err) 191 goto out; 192 193 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 194 195 } else if (symbol_conf.cumulate_callchain) { 196 if (single) 197 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 198 } else { 199 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 200 } 201 202 out: 203 return err; 204 } 205 206 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 207 struct addr_location *al __maybe_unused, 208 bool single __maybe_unused, 209 void *arg) 210 { 211 struct hist_entry *he = iter->he; 212 struct report *rep = arg; 213 struct branch_info *bi = he->branch_info; 214 struct perf_sample *sample = iter->sample; 215 struct evsel *evsel = iter->evsel; 216 int err; 217 218 branch_type_count(&rep->brtype_stat, &bi->flags, 219 bi->from.addr, bi->to.addr); 220 221 if (!ui__has_annotation() && !rep->symbol_ipc) 222 return 0; 223 224 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 225 if (err) 226 goto out; 227 228 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 229 230 out: 231 return err; 232 } 233 234 static void setup_forced_leader(struct report *report, 235 struct evlist *evlist) 236 { 237 if (report->group_set) 238 evlist__force_leader(evlist); 239 } 240 241 static int process_feature_event(struct perf_session *session, 242 union perf_event *event) 243 { 244 struct report *rep = container_of(session->tool, struct report, tool); 245 246 if (event->feat.feat_id < HEADER_LAST_FEATURE) 247 return perf_event__process_feature(session, event); 248 249 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 250 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 251 event->feat.feat_id); 252 return -1; 253 } else if (rep->header_only) { 254 session_done = 1; 255 } 256 257 /* 258 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 259 * means all features are received, now we can force the 260 * group if needed. 261 */ 262 setup_forced_leader(rep, session->evlist); 263 return 0; 264 } 265 266 static int process_sample_event(const struct perf_tool *tool, 267 union perf_event *event, 268 struct perf_sample *sample, 269 struct evsel *evsel, 270 struct machine *machine) 271 { 272 struct report *rep = container_of(tool, struct report, tool); 273 struct addr_location al; 274 struct hist_entry_iter iter = { 275 .evsel = evsel, 276 .sample = sample, 277 .hide_unresolved = symbol_conf.hide_unresolved, 278 .add_entry_cb = hist_iter__report_callback, 279 }; 280 int ret = 0; 281 282 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 283 sample->time)) { 284 return 0; 285 } 286 287 if (evswitch__discard(&rep->evswitch, evsel)) 288 return 0; 289 290 addr_location__init(&al); 291 if (machine__resolve(machine, &al, sample) < 0) { 292 pr_debug("problem processing %d event, skipping it.\n", 293 event->header.type); 294 ret = -1; 295 goto out_put; 296 } 297 298 if (rep->stitch_lbr) 299 thread__set_lbr_stitch_enable(al.thread, true); 300 301 if (symbol_conf.hide_unresolved && al.sym == NULL) 302 goto out_put; 303 304 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 305 goto out_put; 306 307 if (sort__mode == SORT_MODE__BRANCH) { 308 /* 309 * A non-synthesized event might not have a branch stack if 310 * branch stacks have been synthesized (using itrace options). 311 */ 312 if (!sample->branch_stack) 313 goto out_put; 314 315 iter.add_entry_cb = hist_iter__branch_callback; 316 iter.ops = &hist_iter_branch; 317 } else if (rep->mem_mode) { 318 iter.ops = &hist_iter_mem; 319 } else if (symbol_conf.cumulate_callchain) { 320 iter.ops = &hist_iter_cumulative; 321 } else { 322 iter.ops = &hist_iter_normal; 323 } 324 325 if (al.map != NULL) 326 dso__set_hit(map__dso(al.map)); 327 328 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 329 hist__account_cycles(sample->branch_stack, &al, sample, 330 rep->nonany_branch_mode, 331 &rep->total_cycles, evsel); 332 } 333 334 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 335 if (ret < 0) 336 pr_debug("problem adding hist entry, skipping event\n"); 337 out_put: 338 addr_location__exit(&al); 339 return ret; 340 } 341 342 static int process_read_event(const struct perf_tool *tool, 343 union perf_event *event, 344 struct perf_sample *sample __maybe_unused, 345 struct evsel *evsel, 346 struct machine *machine __maybe_unused) 347 { 348 struct report *rep = container_of(tool, struct report, tool); 349 350 if (rep->show_threads) { 351 int err = perf_read_values_add_value(&rep->show_threads_values, 352 event->read.pid, event->read.tid, 353 evsel, 354 event->read.value); 355 356 if (err) 357 return err; 358 } 359 360 return 0; 361 } 362 363 /* For pipe mode, sample_type is not currently set */ 364 static int report__setup_sample_type(struct report *rep) 365 { 366 struct perf_session *session = rep->session; 367 u64 sample_type = evlist__combined_sample_type(session->evlist); 368 bool is_pipe = perf_data__is_pipe(session->data); 369 struct evsel *evsel; 370 371 if (session->itrace_synth_opts->callchain || 372 session->itrace_synth_opts->add_callchain || 373 (!is_pipe && 374 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 375 !session->itrace_synth_opts->set)) 376 sample_type |= PERF_SAMPLE_CALLCHAIN; 377 378 if (session->itrace_synth_opts->last_branch || 379 session->itrace_synth_opts->add_last_branch) 380 sample_type |= PERF_SAMPLE_BRANCH_STACK; 381 382 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 383 if (perf_hpp_list.parent) { 384 ui__error("Selected --sort parent, but no " 385 "callchain data. Did you call " 386 "'perf record' without -g?\n"); 387 return -EINVAL; 388 } 389 if (symbol_conf.use_callchain && 390 !symbol_conf.show_branchflag_count) { 391 ui__error("Selected -g or --branch-history.\n" 392 "But no callchain or branch data.\n" 393 "Did you call 'perf record' without -g or -b?\n"); 394 return -1; 395 } 396 } else if (!callchain_param.enabled && 397 callchain_param.mode != CHAIN_NONE && 398 !symbol_conf.use_callchain) { 399 symbol_conf.use_callchain = true; 400 if (callchain_register_param(&callchain_param) < 0) { 401 ui__error("Can't register callchain params.\n"); 402 return -EINVAL; 403 } 404 } 405 406 if (symbol_conf.cumulate_callchain) { 407 /* Silently ignore if callchain is missing */ 408 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 409 symbol_conf.cumulate_callchain = false; 410 perf_hpp__cancel_cumulate(); 411 } 412 } 413 414 if (sort__mode == SORT_MODE__BRANCH) { 415 if (!is_pipe && 416 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 417 ui__error("Selected -b but no branch data. " 418 "Did you call perf record without -b?\n"); 419 return -1; 420 } 421 } 422 423 if (sort__mode == SORT_MODE__MEMORY) { 424 /* 425 * FIXUP: prior to kernel 5.18, Arm SPE missed to set 426 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward 427 * compatibility, set the bit if it's an old perf data file. 428 */ 429 evlist__for_each_entry(session->evlist, evsel) { 430 if (strstr(evsel__name(evsel), "arm_spe") && 431 !(sample_type & PERF_SAMPLE_DATA_SRC)) { 432 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC; 433 sample_type |= PERF_SAMPLE_DATA_SRC; 434 } 435 } 436 437 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 438 ui__error("Selected --mem-mode but no mem data. " 439 "Did you call perf record without -d?\n"); 440 return -1; 441 } 442 } 443 444 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env)); 445 446 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 447 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 448 "Please apply --call-graph lbr when recording.\n"); 449 rep->stitch_lbr = false; 450 } 451 452 /* ??? handle more cases than just ANY? */ 453 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 454 rep->nonany_branch_mode = true; 455 456 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_LIBDW_SUPPORT) 457 if (dwarf_callchain_users) { 458 ui__warning("Please install libunwind or libdw " 459 "development packages during the perf build.\n"); 460 } 461 #endif 462 463 return 0; 464 } 465 466 static void sig_handler(int sig __maybe_unused) 467 { 468 session_done = 1; 469 } 470 471 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 472 const char *evname, FILE *fp) 473 { 474 size_t ret; 475 char unit; 476 unsigned long nr_samples = hists->stats.nr_samples; 477 u64 nr_events = hists->stats.total_period; 478 struct evsel *evsel = hists_to_evsel(hists); 479 char buf[512]; 480 size_t size = sizeof(buf); 481 int socked_id = hists->socket_filter; 482 483 if (quiet) 484 return 0; 485 486 if (symbol_conf.filter_relative) { 487 nr_samples = hists->stats.nr_non_filtered_samples; 488 nr_events = hists->stats.total_non_filtered_period; 489 } 490 491 if (evsel__is_group_event(evsel)) { 492 struct evsel *pos; 493 494 evsel__group_desc(evsel, buf, size); 495 evname = buf; 496 497 for_each_group_member(pos, evsel) { 498 const struct hists *pos_hists = evsel__hists(pos); 499 500 if (symbol_conf.filter_relative) { 501 nr_samples += pos_hists->stats.nr_non_filtered_samples; 502 nr_events += pos_hists->stats.total_non_filtered_period; 503 } else { 504 nr_samples += pos_hists->stats.nr_samples; 505 nr_events += pos_hists->stats.total_period; 506 } 507 } 508 } 509 510 nr_samples = convert_unit(nr_samples, &unit); 511 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 512 if (evname != NULL) { 513 ret += fprintf(fp, " of event%s '%s'", 514 evsel->core.nr_members > 1 ? "s" : "", evname); 515 } 516 517 if (rep->time_str) 518 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 519 520 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 521 ret += fprintf(fp, ", show reference callgraph"); 522 } 523 524 if (rep->mem_mode) { 525 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 526 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 527 } else 528 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 529 530 if (socked_id > -1) 531 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 532 533 return ret + fprintf(fp, "\n#\n"); 534 } 535 536 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 537 { 538 struct evsel *pos; 539 int i = 0, ret; 540 541 evlist__for_each_entry(evlist, pos) { 542 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 543 rep->min_percent, pos, 544 &rep->session->header.env); 545 if (ret != 0) 546 return ret; 547 } 548 549 return 0; 550 } 551 552 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 553 { 554 struct evsel *pos; 555 int i = 0; 556 557 if (!quiet) { 558 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 559 evlist->stats.total_lost_samples); 560 } 561 562 evlist__for_each_entry(evlist, pos) { 563 struct hists *hists = evsel__hists(pos); 564 const char *evname = evsel__name(pos); 565 566 i++; 567 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 568 continue; 569 570 if (rep->skip_empty && !hists->stats.nr_samples) 571 continue; 572 573 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 574 575 if (rep->total_cycles_mode) { 576 char *buf; 577 578 if (!annotation_br_cntr_abbr_list(&buf, pos, true)) { 579 fprintf(stdout, "%s", buf); 580 fprintf(stdout, "#\n"); 581 free(buf); 582 } 583 report__browse_block_hists(&rep->block_reports[i - 1].hist, 584 rep->min_percent, pos, NULL); 585 continue; 586 } 587 588 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 589 !(symbol_conf.use_callchain || 590 symbol_conf.show_branchflag_count)); 591 fprintf(stdout, "\n\n"); 592 } 593 594 if (!quiet) 595 fprintf(stdout, "#\n# (%s)\n#\n", help); 596 597 if (rep->show_threads) { 598 bool style = !strcmp(rep->pretty_printing_style, "raw"); 599 perf_read_values_display(stdout, &rep->show_threads_values, 600 style); 601 perf_read_values_destroy(&rep->show_threads_values); 602 } 603 604 if (sort__mode == SORT_MODE__BRANCH) 605 branch_type_stat_display(stdout, &rep->brtype_stat); 606 607 return 0; 608 } 609 610 static void report__warn_kptr_restrict(const struct report *rep) 611 { 612 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 613 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 614 615 if (evlist__exclude_kernel(rep->session->evlist)) 616 return; 617 618 if (kernel_map == NULL || 619 (dso__hit(map__dso(kernel_map)) && 620 (kernel_kmap->ref_reloc_sym == NULL || 621 kernel_kmap->ref_reloc_sym->addr == 0))) { 622 const char *desc = 623 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 624 "can't be resolved."; 625 626 if (kernel_map && map__has_symbols(kernel_map)) { 627 desc = "If some relocation was applied (e.g. " 628 "kexec) symbols may be misresolved."; 629 } 630 631 ui__warning( 632 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 633 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 634 "Samples in kernel modules can't be resolved as well.\n\n", 635 desc); 636 } 637 } 638 639 static int report__gtk_browse_hists(struct report *rep, const char *help) 640 { 641 int (*hist_browser)(struct evlist *evlist, const char *help, 642 struct hist_browser_timer *timer, float min_pcnt); 643 644 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 645 646 if (hist_browser == NULL) { 647 ui__error("GTK browser not found!\n"); 648 return -1; 649 } 650 651 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 652 } 653 654 static int report__browse_hists(struct report *rep) 655 { 656 int ret; 657 struct perf_session *session = rep->session; 658 struct evlist *evlist = session->evlist; 659 char *help = NULL, *path = NULL; 660 661 path = system_path(TIPDIR); 662 if (perf_tip(&help, path) || help == NULL) { 663 /* fallback for people who don't install perf ;-) */ 664 free(path); 665 path = system_path(DOCDIR); 666 if (perf_tip(&help, path) || help == NULL) 667 help = strdup("Cannot load tips.txt file, please install perf!"); 668 } 669 free(path); 670 671 switch (use_browser) { 672 case 1: 673 if (rep->total_cycles_mode) { 674 ret = evlist__tui_block_hists_browse(evlist, rep); 675 break; 676 } 677 678 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 679 &session->header.env, true); 680 /* 681 * Usually "ret" is the last pressed key, and we only 682 * care if the key notifies us to switch data file. 683 */ 684 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 685 ret = 0; 686 break; 687 case 2: 688 ret = report__gtk_browse_hists(rep, help); 689 break; 690 default: 691 ret = evlist__tty_browse_hists(evlist, rep, help); 692 break; 693 } 694 free(help); 695 return ret; 696 } 697 698 static int report__collapse_hists(struct report *rep) 699 { 700 struct perf_session *session = rep->session; 701 struct evlist *evlist = session->evlist; 702 struct ui_progress prog; 703 struct evsel *pos; 704 int ret = 0; 705 706 /* 707 * The pipe data needs to setup hierarchy hpp formats now, because it 708 * cannot know about evsels in the data before reading the data. The 709 * normal file data saves the event (attribute) info in the header 710 * section, but pipe does not have the luxury. 711 */ 712 if (perf_data__is_pipe(session->data)) { 713 if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) { 714 ui__error("Failed to setup hierarchy output formats\n"); 715 return -1; 716 } 717 } 718 719 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 720 721 evlist__for_each_entry(rep->session->evlist, pos) { 722 struct hists *hists = evsel__hists(pos); 723 724 if (pos->core.idx == 0) 725 hists->symbol_filter_str = rep->symbol_filter_str; 726 727 hists->socket_filter = rep->socket_filter; 728 729 ret = hists__collapse_resort(hists, &prog); 730 if (ret < 0) 731 break; 732 733 /* Non-group events are considered as leader */ 734 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 735 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 736 737 hists__match(leader_hists, hists); 738 hists__link(leader_hists, hists); 739 } 740 } 741 742 ui_progress__finish(); 743 return ret; 744 } 745 746 static int hists__resort_cb(struct hist_entry *he, void *arg) 747 { 748 struct report *rep = arg; 749 struct symbol *sym = he->ms.sym; 750 751 if (rep->symbol_ipc && sym && !sym->annotate2) { 752 struct evsel *evsel = hists_to_evsel(he->hists); 753 754 symbol__annotate2(&he->ms, evsel, NULL); 755 } 756 757 return 0; 758 } 759 760 static void report__output_resort(struct report *rep) 761 { 762 struct ui_progress prog; 763 struct evsel *pos; 764 765 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 766 767 evlist__for_each_entry(rep->session->evlist, pos) { 768 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 769 } 770 771 ui_progress__finish(); 772 } 773 774 static int count_sample_event(const struct perf_tool *tool __maybe_unused, 775 union perf_event *event __maybe_unused, 776 struct perf_sample *sample __maybe_unused, 777 struct evsel *evsel, 778 struct machine *machine __maybe_unused) 779 { 780 struct hists *hists = evsel__hists(evsel); 781 782 hists__inc_nr_events(hists); 783 return 0; 784 } 785 786 static int count_lost_samples_event(const struct perf_tool *tool, 787 union perf_event *event, 788 struct perf_sample *sample, 789 struct machine *machine __maybe_unused) 790 { 791 struct report *rep = container_of(tool, struct report, tool); 792 struct evsel *evsel; 793 794 evsel = evlist__id2evsel(rep->session->evlist, sample->id); 795 if (evsel) { 796 struct hists *hists = evsel__hists(evsel); 797 u32 count = event->lost_samples.lost; 798 799 if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF) 800 hists__inc_nr_dropped_samples(hists, count); 801 else 802 hists__inc_nr_lost_samples(hists, count); 803 } 804 return 0; 805 } 806 807 static int process_attr(const struct perf_tool *tool __maybe_unused, 808 union perf_event *event, 809 struct evlist **pevlist); 810 811 static void stats_setup(struct report *rep) 812 { 813 perf_tool__init(&rep->tool, /*ordered_events=*/false); 814 rep->tool.attr = process_attr; 815 rep->tool.sample = count_sample_event; 816 rep->tool.lost_samples = count_lost_samples_event; 817 rep->tool.event_update = perf_event__process_event_update; 818 rep->tool.no_warn = true; 819 } 820 821 static int stats_print(struct report *rep) 822 { 823 struct perf_session *session = rep->session; 824 825 perf_session__fprintf_nr_events(session, stdout); 826 evlist__fprintf_nr_events(session->evlist, stdout); 827 return 0; 828 } 829 830 static void tasks_setup(struct report *rep) 831 { 832 perf_tool__init(&rep->tool, /*ordered_events=*/true); 833 if (rep->mmaps_mode) { 834 rep->tool.mmap = perf_event__process_mmap; 835 rep->tool.mmap2 = perf_event__process_mmap2; 836 } 837 rep->tool.attr = process_attr; 838 rep->tool.comm = perf_event__process_comm; 839 rep->tool.exit = perf_event__process_exit; 840 rep->tool.fork = perf_event__process_fork; 841 rep->tool.no_warn = true; 842 } 843 844 struct maps__fprintf_task_args { 845 int indent; 846 FILE *fp; 847 size_t printed; 848 }; 849 850 static int maps__fprintf_task_cb(struct map *map, void *data) 851 { 852 struct maps__fprintf_task_args *args = data; 853 const struct dso *dso = map__dso(map); 854 u32 prot = map__prot(map); 855 int ret; 856 857 ret = fprintf(args->fp, 858 "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 859 args->indent, "", map__start(map), map__end(map), 860 prot & PROT_READ ? 'r' : '-', 861 prot & PROT_WRITE ? 'w' : '-', 862 prot & PROT_EXEC ? 'x' : '-', 863 map__flags(map) ? 's' : 'p', 864 map__pgoff(map), 865 dso__id_const(dso)->ino, dso__name(dso)); 866 867 if (ret < 0) 868 return ret; 869 870 args->printed += ret; 871 return 0; 872 } 873 874 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 875 { 876 struct maps__fprintf_task_args args = { 877 .indent = indent, 878 .fp = fp, 879 .printed = 0, 880 }; 881 882 maps__for_each_map(maps, maps__fprintf_task_cb, &args); 883 884 return args.printed; 885 } 886 887 static int thread_level(struct machine *machine, const struct thread *thread) 888 { 889 struct thread *parent_thread; 890 int res; 891 892 if (thread__tid(thread) <= 0) 893 return 0; 894 895 if (thread__ppid(thread) <= 0) 896 return 1; 897 898 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread)); 899 if (!parent_thread) { 900 pr_err("Missing parent thread of %d\n", thread__tid(thread)); 901 return 0; 902 } 903 res = 1 + thread_level(machine, parent_thread); 904 thread__put(parent_thread); 905 return res; 906 } 907 908 static void task__print_level(struct machine *machine, struct thread *thread, FILE *fp) 909 { 910 int level = thread_level(machine, thread); 911 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 912 thread__pid(thread), thread__tid(thread), 913 thread__ppid(thread), level, ""); 914 915 fprintf(fp, "%s\n", thread__comm_str(thread)); 916 917 maps__fprintf_task(thread__maps(thread), comm_indent, fp); 918 } 919 920 /* 921 * Sort two thread list nodes such that they form a tree. The first node is the 922 * root of the tree, its children are ordered numerically after it. If a child 923 * has children itself then they appear immediately after their parent. For 924 * example, the 4 threads in the order they'd appear in the list: 925 * - init with a TID 1 and a parent of 0 926 * - systemd with a TID 3000 and a parent of init/1 927 * - systemd child thread with TID 4000, the parent is 3000 928 * - NetworkManager is a child of init with a TID of 3500. 929 */ 930 static int task_list_cmp(void *priv, const struct list_head *la, const struct list_head *lb) 931 { 932 struct machine *machine = priv; 933 struct thread_list *task_a = list_entry(la, struct thread_list, list); 934 struct thread_list *task_b = list_entry(lb, struct thread_list, list); 935 struct thread *a = task_a->thread; 936 struct thread *b = task_b->thread; 937 int level_a, level_b, res; 938 939 /* Same thread? */ 940 if (thread__tid(a) == thread__tid(b)) 941 return 0; 942 943 /* Compare a and b to root. */ 944 if (thread__tid(a) == 0) 945 return -1; 946 947 if (thread__tid(b) == 0) 948 return 1; 949 950 /* If parents match sort by tid. */ 951 if (thread__ppid(a) == thread__ppid(b)) 952 return thread__tid(a) < thread__tid(b) ? -1 : 1; 953 954 /* 955 * Find a and b such that if they are a child of each other a and b's 956 * tid's match, otherwise a and b have a common parent and distinct 957 * tid's to sort by. First make the depths of the threads match. 958 */ 959 level_a = thread_level(machine, a); 960 level_b = thread_level(machine, b); 961 a = thread__get(a); 962 b = thread__get(b); 963 for (int i = level_a; i > level_b; i--) { 964 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(a)); 965 966 thread__put(a); 967 if (!parent) { 968 pr_err("Missing parent thread of %d\n", thread__tid(a)); 969 thread__put(b); 970 return -1; 971 } 972 a = parent; 973 } 974 for (int i = level_b; i > level_a; i--) { 975 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(b)); 976 977 thread__put(b); 978 if (!parent) { 979 pr_err("Missing parent thread of %d\n", thread__tid(b)); 980 thread__put(a); 981 return 1; 982 } 983 b = parent; 984 } 985 /* Search up to a common parent. */ 986 while (thread__ppid(a) != thread__ppid(b)) { 987 struct thread *parent; 988 989 parent = machine__find_thread(machine, -1, thread__ppid(a)); 990 thread__put(a); 991 if (!parent) 992 pr_err("Missing parent thread of %d\n", thread__tid(a)); 993 a = parent; 994 parent = machine__find_thread(machine, -1, thread__ppid(b)); 995 thread__put(b); 996 if (!parent) 997 pr_err("Missing parent thread of %d\n", thread__tid(b)); 998 b = parent; 999 if (!a || !b) { 1000 /* Handle missing parent (unexpected) with some sanity. */ 1001 thread__put(a); 1002 thread__put(b); 1003 return !a && !b ? 0 : (!a ? -1 : 1); 1004 } 1005 } 1006 if (thread__tid(a) == thread__tid(b)) { 1007 /* a is a child of b or vice-versa, deeper levels appear later. */ 1008 res = level_a < level_b ? -1 : (level_a > level_b ? 1 : 0); 1009 } else { 1010 /* Sort by tid now the parent is the same. */ 1011 res = thread__tid(a) < thread__tid(b) ? -1 : 1; 1012 } 1013 thread__put(a); 1014 thread__put(b); 1015 return res; 1016 } 1017 1018 static int tasks_print(struct report *rep, FILE *fp) 1019 { 1020 struct machine *machine = &rep->session->machines.host; 1021 LIST_HEAD(tasks); 1022 int ret; 1023 1024 ret = machine__thread_list(machine, &tasks); 1025 if (!ret) { 1026 struct thread_list *task; 1027 1028 list_sort(machine, &tasks, task_list_cmp); 1029 1030 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 1031 1032 list_for_each_entry(task, &tasks, list) 1033 task__print_level(machine, task->thread, fp); 1034 } 1035 thread_list__delete(&tasks); 1036 return ret; 1037 } 1038 1039 static int __cmd_report(struct report *rep) 1040 { 1041 int ret; 1042 struct perf_session *session = rep->session; 1043 struct evsel *pos; 1044 struct perf_data *data = session->data; 1045 1046 signal(SIGINT, sig_handler); 1047 1048 if (rep->cpu_list) { 1049 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 1050 rep->cpu_bitmap); 1051 if (ret) { 1052 ui__error("failed to set cpu bitmap\n"); 1053 return ret; 1054 } 1055 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 1056 } 1057 1058 if (rep->show_threads) { 1059 ret = perf_read_values_init(&rep->show_threads_values); 1060 if (ret) 1061 return ret; 1062 } 1063 1064 ret = report__setup_sample_type(rep); 1065 if (ret) { 1066 /* report__setup_sample_type() already showed error message */ 1067 return ret; 1068 } 1069 1070 if (rep->stats_mode) 1071 stats_setup(rep); 1072 1073 if (rep->tasks_mode) 1074 tasks_setup(rep); 1075 1076 ret = perf_session__process_events(session); 1077 if (ret) { 1078 ui__error("failed to process sample\n"); 1079 return ret; 1080 } 1081 1082 evlist__check_mem_load_aux(session->evlist); 1083 1084 if (rep->stats_mode) 1085 return stats_print(rep); 1086 1087 if (rep->tasks_mode) 1088 return tasks_print(rep, stdout); 1089 1090 report__warn_kptr_restrict(rep); 1091 1092 evlist__for_each_entry(session->evlist, pos) 1093 rep->nr_entries += evsel__hists(pos)->nr_entries; 1094 1095 if (use_browser == 0) { 1096 if (verbose > 3) 1097 perf_session__fprintf(session, stdout); 1098 1099 if (verbose > 2) 1100 perf_session__fprintf_dsos(session, stdout); 1101 1102 if (dump_trace) { 1103 stats_print(rep); 1104 return 0; 1105 } 1106 } 1107 1108 ret = report__collapse_hists(rep); 1109 if (ret) { 1110 ui__error("failed to process hist entry\n"); 1111 return ret; 1112 } 1113 1114 if (session_done()) 1115 return 0; 1116 1117 /* 1118 * recalculate number of entries after collapsing since it 1119 * might be changed during the collapse phase. 1120 */ 1121 rep->nr_entries = 0; 1122 evlist__for_each_entry(session->evlist, pos) 1123 rep->nr_entries += evsel__hists(pos)->nr_entries; 1124 1125 if (rep->nr_entries == 0) { 1126 ui__error("The %s data has no samples!\n", data->path); 1127 return 0; 1128 } 1129 1130 report__output_resort(rep); 1131 1132 if (rep->total_cycles_mode) { 1133 int nr_hpps = 4; 1134 int block_hpps[PERF_HPP_REPORT__BLOCK_MAX_INDEX] = { 1135 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1136 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1137 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1138 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1139 }; 1140 1141 if (session->evlist->nr_br_cntr > 0) 1142 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER; 1143 1144 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_RANGE; 1145 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_DSO; 1146 1147 rep->block_reports = block_info__create_report(session->evlist, 1148 rep->total_cycles, 1149 block_hpps, nr_hpps, 1150 &rep->nr_block_reports); 1151 if (!rep->block_reports) 1152 return -1; 1153 } 1154 1155 return report__browse_hists(rep); 1156 } 1157 1158 static int 1159 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1160 { 1161 struct callchain_param *callchain = opt->value; 1162 1163 callchain->enabled = !unset; 1164 /* 1165 * --no-call-graph 1166 */ 1167 if (unset) { 1168 symbol_conf.use_callchain = false; 1169 callchain->mode = CHAIN_NONE; 1170 return 0; 1171 } 1172 1173 return parse_callchain_report_opt(arg); 1174 } 1175 1176 static int 1177 parse_time_quantum(const struct option *opt, const char *arg, 1178 int unset __maybe_unused) 1179 { 1180 unsigned long *time_q = opt->value; 1181 char *end; 1182 1183 *time_q = strtoul(arg, &end, 0); 1184 if (end == arg) 1185 goto parse_err; 1186 if (*time_q == 0) { 1187 pr_err("time quantum cannot be 0"); 1188 return -1; 1189 } 1190 end = skip_spaces(end); 1191 if (*end == 0) 1192 return 0; 1193 if (!strcmp(end, "s")) { 1194 *time_q *= NSEC_PER_SEC; 1195 return 0; 1196 } 1197 if (!strcmp(end, "ms")) { 1198 *time_q *= NSEC_PER_MSEC; 1199 return 0; 1200 } 1201 if (!strcmp(end, "us")) { 1202 *time_q *= NSEC_PER_USEC; 1203 return 0; 1204 } 1205 if (!strcmp(end, "ns")) 1206 return 0; 1207 parse_err: 1208 pr_err("Cannot parse time quantum `%s'\n", arg); 1209 return -1; 1210 } 1211 1212 int 1213 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1214 const char *arg, int unset __maybe_unused) 1215 { 1216 if (arg) { 1217 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1218 if (err) { 1219 char buf[BUFSIZ]; 1220 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1221 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1222 return -1; 1223 } 1224 have_ignore_callees = 1; 1225 } 1226 1227 return 0; 1228 } 1229 1230 static int 1231 parse_branch_mode(const struct option *opt, 1232 const char *str __maybe_unused, int unset) 1233 { 1234 int *branch_mode = opt->value; 1235 1236 *branch_mode = !unset; 1237 return 0; 1238 } 1239 1240 static int 1241 parse_percent_limit(const struct option *opt, const char *str, 1242 int unset __maybe_unused) 1243 { 1244 struct report *rep = opt->value; 1245 double pcnt = strtof(str, NULL); 1246 1247 rep->min_percent = pcnt; 1248 callchain_param.min_percent = pcnt; 1249 return 0; 1250 } 1251 1252 static int process_attr(const struct perf_tool *tool __maybe_unused, 1253 union perf_event *event, 1254 struct evlist **pevlist) 1255 { 1256 u64 sample_type; 1257 int err; 1258 1259 err = perf_event__process_attr(tool, event, pevlist); 1260 if (err) 1261 return err; 1262 1263 /* 1264 * Check if we need to enable callchains based 1265 * on events sample_type. 1266 */ 1267 sample_type = evlist__combined_sample_type(*pevlist); 1268 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env)); 1269 return 0; 1270 } 1271 1272 #define CALLCHAIN_BRANCH_SORT_ORDER \ 1273 "srcline,symbol,dso,callchain_branch_predicted," \ 1274 "callchain_branch_abort,callchain_branch_cycles" 1275 1276 int cmd_report(int argc, const char **argv) 1277 { 1278 struct perf_session *session; 1279 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1280 struct stat st; 1281 bool has_br_stack = false; 1282 int branch_mode = -1; 1283 int last_key = 0; 1284 bool branch_call_mode = false; 1285 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1286 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1287 CALLCHAIN_REPORT_HELP 1288 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1289 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1290 const char * const report_usage[] = { 1291 "perf report [<options>]", 1292 NULL 1293 }; 1294 struct report report = { 1295 .max_stack = PERF_MAX_STACK_DEPTH, 1296 .pretty_printing_style = "normal", 1297 .socket_filter = -1, 1298 .skip_empty = true, 1299 }; 1300 char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__NORMAL); 1301 char *field_order_help = sort_help("output field(s):", SORT_MODE__NORMAL); 1302 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL; 1303 const struct option options[] = { 1304 OPT_STRING('i', "input", &input_name, "file", 1305 "input file name"), 1306 OPT_INCR('v', "verbose", &verbose, 1307 "be more verbose (show symbol address, etc)"), 1308 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"), 1309 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1310 "dump raw trace in ASCII"), 1311 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1312 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1313 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1314 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1315 "file", "vmlinux pathname"), 1316 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1317 "don't load vmlinux even if found"), 1318 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1319 "file", "kallsyms pathname"), 1320 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1321 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1322 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1323 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1324 "Show a column with the number of samples"), 1325 OPT_BOOLEAN('T', "threads", &report.show_threads, 1326 "Show per-thread event counters"), 1327 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1328 "pretty printing style key: normal raw"), 1329 #ifdef HAVE_SLANG_SUPPORT 1330 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1331 #endif 1332 #ifdef HAVE_GTK2_SUPPORT 1333 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1334 #endif 1335 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1336 "Use the stdio interface"), 1337 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1338 OPT_BOOLEAN(0, "header-only", &report.header_only, 1339 "Show only data header."), 1340 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1341 sort_order_help), 1342 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1343 field_order_help), 1344 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1345 "Show sample percentage for different cpu modes"), 1346 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1347 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1348 OPT_STRING('p', "parent", &parent_pattern, "regex", 1349 "regex filter to identify parent, see: '--sort parent'"), 1350 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1351 "Only display entries with parent-match"), 1352 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1353 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1354 report_callchain_help, &report_parse_callchain_opt, 1355 callchain_default_opt), 1356 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1357 "Accumulate callchains of children and show total overhead as well. " 1358 "Enabled by default, use --no-children to disable."), 1359 OPT_INTEGER(0, "max-stack", &report.max_stack, 1360 "Set the maximum stack depth when parsing the callchain, " 1361 "anything beyond the specified depth will be ignored. " 1362 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1363 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1364 "alias for inverted call graph"), 1365 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1366 "ignore callees of these functions in call graphs", 1367 report_parse_ignore_callees_opt), 1368 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1369 "only consider symbols in these dsos"), 1370 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1371 "only consider symbols in these comms"), 1372 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1373 "only consider symbols in these pids"), 1374 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1375 "only consider symbols in these tids"), 1376 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1377 "only consider these symbols"), 1378 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1379 "only show symbols that (partially) match with this filter"), 1380 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1381 "width[,width...]", 1382 "don't try to adjust column width, use these fixed values"), 1383 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1384 "separator for columns, no spaces will be added between " 1385 "columns '.' is reserved."), 1386 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1387 "Only display entries resolved to a symbol"), 1388 OPT_CALLBACK(0, "symfs", NULL, "directory", 1389 "Look for files with symbols relative to this directory", 1390 symbol__config_symfs), 1391 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1392 "list of cpus to profile"), 1393 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1394 "Display extended information about perf.data file"), 1395 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src, 1396 "Interleave source code with assembly code (default)"), 1397 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw, 1398 "Display raw encoding of assembly instructions (default)"), 1399 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1400 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1401 OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix", 1402 "Add prefix to source file path names in programs (with --prefix-strip)"), 1403 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N", 1404 "Strip first N entries of source file path name in programs (with --prefix)"), 1405 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1406 "Show a column with the sum of periods"), 1407 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1408 "Show event group information together"), 1409 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1410 "Sort the output by the event at the index n in group. " 1411 "If n is invalid, sort by the first event. " 1412 "WARNING: should be used on grouped events."), 1413 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1414 "use branch records for per branch histogram filling", 1415 parse_branch_mode), 1416 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1417 "add last branch records to call history"), 1418 OPT_STRING(0, "objdump", &objdump_path, "path", 1419 "objdump binary to use for disassembly and annotations"), 1420 OPT_STRING(0, "addr2line", &addr2line_path, "path", 1421 "addr2line binary to use for line numbers"), 1422 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1423 "Symbol demangling. Enabled by default, use --no-demangle to disable."), 1424 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1425 "Enable kernel symbol demangling"), 1426 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1427 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1428 "Number of samples to save per histogram entry for individual browsing"), 1429 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1430 "Don't show entries under that percent", parse_percent_limit), 1431 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1432 "how to display percentage of filtered entries", parse_filter_percentage), 1433 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1434 "Instruction Tracing options\n" ITRACE_HELP, 1435 itrace_parse_synth_opts), 1436 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1437 "Show full source file name path for source lines"), 1438 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1439 "Show callgraph from reference event"), 1440 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1441 "Enable LBR callgraph stitching approach"), 1442 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1443 "only show processor socket that match with this filter"), 1444 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1445 "Show raw trace event output (do not use print fmt or plugins)"), 1446 OPT_BOOLEAN('H', "hierarchy", &symbol_conf.report_hierarchy, 1447 "Show entries in a hierarchy"), 1448 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1449 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1450 stdio__config_color, "always"), 1451 OPT_STRING(0, "time", &report.time_str, "str", 1452 "Time span of interest (start,stop)"), 1453 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1454 "Show inline function"), 1455 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period", 1456 "Set percent type local/global-period/hits", 1457 annotate_parse_percent_type), 1458 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1459 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1460 "Set time quantum for time sort key (default 100ms)", 1461 parse_time_quantum), 1462 OPTS_EVSWITCH(&report.evswitch), 1463 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1464 "Sort all blocks by 'Sampled Cycles%'"), 1465 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1466 "Disable raw trace ordering"), 1467 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1468 "Do not display empty (or dummy) events in the output"), 1469 OPT_END() 1470 }; 1471 struct perf_data data = { 1472 .mode = PERF_DATA_MODE_READ, 1473 }; 1474 int ret = hists__init(); 1475 char sort_tmp[128]; 1476 bool ordered_events = true; 1477 1478 if (ret < 0) 1479 goto exit; 1480 1481 /* 1482 * tasks_mode require access to exited threads to list those that are in 1483 * the data file. Off-cpu events are synthesized after other events and 1484 * reference exited threads. 1485 */ 1486 symbol_conf.keep_exited_threads = true; 1487 1488 annotation_options__init(); 1489 1490 ret = perf_config(report__config, &report); 1491 if (ret) 1492 goto exit; 1493 1494 argc = parse_options(argc, argv, options, report_usage, 0); 1495 if (argc) { 1496 /* 1497 * Special case: if there's an argument left then assume that 1498 * it's a symbol filter: 1499 */ 1500 if (argc > 1) 1501 usage_with_options(report_usage, options); 1502 1503 report.symbol_filter_str = argv[0]; 1504 } 1505 1506 if (disassembler_style) { 1507 annotate_opts.disassembler_style = strdup(disassembler_style); 1508 if (!annotate_opts.disassembler_style) 1509 return -ENOMEM; 1510 } 1511 if (objdump_path) { 1512 annotate_opts.objdump_path = strdup(objdump_path); 1513 if (!annotate_opts.objdump_path) 1514 return -ENOMEM; 1515 } 1516 if (addr2line_path) { 1517 symbol_conf.addr2line_path = strdup(addr2line_path); 1518 if (!symbol_conf.addr2line_path) 1519 return -ENOMEM; 1520 } 1521 1522 if (annotate_check_args() < 0) { 1523 ret = -EINVAL; 1524 goto exit; 1525 } 1526 1527 if (report.mmaps_mode) 1528 report.tasks_mode = true; 1529 1530 if (dump_trace && report.disable_order) 1531 ordered_events = false; 1532 1533 if (quiet) 1534 perf_quiet_option(); 1535 1536 ret = symbol__validate_sym_arguments(); 1537 if (ret) 1538 goto exit; 1539 1540 if (report.inverted_callchain) 1541 callchain_param.order = ORDER_CALLER; 1542 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1543 callchain_param.order = ORDER_CALLER; 1544 1545 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1546 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1547 report.max_stack = itrace_synth_opts.callchain_sz; 1548 1549 if (!input_name || !strlen(input_name)) { 1550 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1551 input_name = "-"; 1552 else 1553 input_name = "perf.data"; 1554 } 1555 1556 data.path = input_name; 1557 data.force = symbol_conf.force; 1558 1559 symbol_conf.skip_empty = report.skip_empty; 1560 1561 repeat: 1562 perf_tool__init(&report.tool, ordered_events); 1563 report.tool.sample = process_sample_event; 1564 report.tool.mmap = perf_event__process_mmap; 1565 report.tool.mmap2 = perf_event__process_mmap2; 1566 report.tool.comm = perf_event__process_comm; 1567 report.tool.namespaces = perf_event__process_namespaces; 1568 report.tool.cgroup = perf_event__process_cgroup; 1569 report.tool.exit = perf_event__process_exit; 1570 report.tool.fork = perf_event__process_fork; 1571 report.tool.lost = perf_event__process_lost; 1572 report.tool.read = process_read_event; 1573 report.tool.attr = process_attr; 1574 #ifdef HAVE_LIBTRACEEVENT 1575 report.tool.tracing_data = perf_event__process_tracing_data; 1576 #endif 1577 report.tool.build_id = perf_event__process_build_id; 1578 report.tool.id_index = perf_event__process_id_index; 1579 report.tool.auxtrace_info = perf_event__process_auxtrace_info; 1580 report.tool.auxtrace = perf_event__process_auxtrace; 1581 report.tool.event_update = perf_event__process_event_update; 1582 report.tool.feature = process_feature_event; 1583 report.tool.ordering_requires_timestamps = true; 1584 1585 session = perf_session__new(&data, &report.tool); 1586 if (IS_ERR(session)) { 1587 ret = PTR_ERR(session); 1588 goto exit; 1589 } 1590 1591 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1592 if (ret) 1593 goto exit; 1594 1595 if (zstd_init(&(session->zstd_data), 0) < 0) 1596 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1597 1598 if (report.queue_size) { 1599 ordered_events__set_alloc_size(&session->ordered_events, 1600 report.queue_size); 1601 } 1602 1603 session->itrace_synth_opts = &itrace_synth_opts; 1604 1605 report.session = session; 1606 1607 has_br_stack = perf_header__has_feat(&session->header, 1608 HEADER_BRANCH_STACK); 1609 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1610 has_br_stack = false; 1611 1612 setup_forced_leader(&report, session->evlist); 1613 1614 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) { 1615 parse_options_usage(NULL, options, "group-sort-idx", 0); 1616 ret = -EINVAL; 1617 goto error; 1618 } 1619 1620 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1621 has_br_stack = true; 1622 1623 if (has_br_stack && branch_call_mode) 1624 symbol_conf.show_branchflag_count = true; 1625 1626 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1627 1628 /* 1629 * Branch mode is a tristate: 1630 * -1 means default, so decide based on the file having branch data. 1631 * 0/1 means the user chose a mode. 1632 */ 1633 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1634 !branch_call_mode) { 1635 sort__mode = SORT_MODE__BRANCH; 1636 symbol_conf.cumulate_callchain = false; 1637 } 1638 if (branch_call_mode) { 1639 callchain_param.key = CCKEY_ADDRESS; 1640 callchain_param.branch_callstack = true; 1641 symbol_conf.use_callchain = true; 1642 callchain_register_param(&callchain_param); 1643 if (sort_order == NULL) 1644 sort_order = CALLCHAIN_BRANCH_SORT_ORDER; 1645 } 1646 1647 if (report.mem_mode) { 1648 if (sort__mode == SORT_MODE__BRANCH) { 1649 pr_err("branch and mem mode incompatible\n"); 1650 goto error; 1651 } 1652 sort__mode = SORT_MODE__MEMORY; 1653 symbol_conf.cumulate_callchain = false; 1654 } 1655 1656 if (symbol_conf.report_hierarchy) { 1657 /* disable incompatible options */ 1658 symbol_conf.cumulate_callchain = false; 1659 1660 if (field_order) { 1661 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1662 parse_options_usage(report_usage, options, "F", 1); 1663 parse_options_usage(NULL, options, "hierarchy", 0); 1664 goto error; 1665 } 1666 1667 perf_hpp_list.need_collapse = true; 1668 } 1669 1670 if (report.use_stdio) 1671 use_browser = 0; 1672 #ifdef HAVE_SLANG_SUPPORT 1673 else if (report.use_tui) 1674 use_browser = 1; 1675 #endif 1676 #ifdef HAVE_GTK2_SUPPORT 1677 else if (report.use_gtk) 1678 use_browser = 2; 1679 #endif 1680 1681 /* Force tty output for header output and per-thread stat. */ 1682 if (report.header || report.header_only || report.show_threads) 1683 use_browser = 0; 1684 if (report.header || report.header_only) 1685 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1686 if (report.show_full_info) 1687 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1688 if (report.stats_mode || report.tasks_mode) 1689 use_browser = 0; 1690 if (report.stats_mode && report.tasks_mode) { 1691 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1692 goto error; 1693 } 1694 1695 if (report.total_cycles_mode) { 1696 if (sort__mode != SORT_MODE__BRANCH) 1697 report.total_cycles_mode = false; 1698 else 1699 sort_order = NULL; 1700 } 1701 1702 if (sort_order && strstr(sort_order, "type")) { 1703 report.data_type = true; 1704 annotate_opts.annotate_src = false; 1705 1706 #ifndef HAVE_LIBDW_SUPPORT 1707 pr_err("Error: Data type profiling is disabled due to missing DWARF support\n"); 1708 goto error; 1709 #endif 1710 } 1711 1712 if (strcmp(input_name, "-") != 0) 1713 setup_browser(true); 1714 else 1715 use_browser = 0; 1716 1717 if (report.data_type && use_browser == 1) { 1718 symbol_conf.annotate_data_member = true; 1719 symbol_conf.annotate_data_sample = true; 1720 } 1721 1722 if (sort_order && strstr(sort_order, "ipc")) { 1723 parse_options_usage(report_usage, options, "s", 1); 1724 goto error; 1725 } 1726 1727 if (sort_order && strstr(sort_order, "symbol")) { 1728 if (sort__mode == SORT_MODE__BRANCH) { 1729 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1730 sort_order, "ipc_lbr"); 1731 report.symbol_ipc = true; 1732 } else { 1733 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1734 sort_order, "ipc_null"); 1735 } 1736 1737 sort_order = sort_tmp; 1738 } 1739 1740 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1741 (setup_sorting(session->evlist) < 0)) { 1742 if (sort_order) 1743 parse_options_usage(report_usage, options, "s", 1); 1744 if (field_order) 1745 parse_options_usage(sort_order ? NULL : report_usage, 1746 options, "F", 1); 1747 goto error; 1748 } 1749 1750 if ((report.header || report.header_only) && !quiet) { 1751 perf_session__fprintf_info(session, stdout, 1752 report.show_full_info); 1753 if (report.header_only) { 1754 if (data.is_pipe) { 1755 /* 1756 * we need to process first few records 1757 * which contains PERF_RECORD_HEADER_FEATURE. 1758 */ 1759 perf_session__process_events(session); 1760 } 1761 ret = 0; 1762 goto error; 1763 } 1764 } else if (use_browser == 0 && !quiet && 1765 !report.stats_mode && !report.tasks_mode) { 1766 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1767 stdout); 1768 } 1769 1770 /* 1771 * Only in the TUI browser we are doing integrated annotation, 1772 * so don't allocate extra space that won't be used in the stdio 1773 * implementation. 1774 */ 1775 if (ui__has_annotation() || report.symbol_ipc || report.data_type || 1776 report.total_cycles_mode) { 1777 ret = symbol__annotation_init(); 1778 if (ret < 0) 1779 goto error; 1780 /* 1781 * For searching by name on the "Browse map details". 1782 * providing it only in verbose mode not to bloat too 1783 * much struct symbol. 1784 */ 1785 if (verbose > 0) { 1786 /* 1787 * XXX: Need to provide a less kludgy way to ask for 1788 * more space per symbol, the u32 is for the index on 1789 * the ui browser. 1790 * See symbol__browser_index. 1791 */ 1792 symbol_conf.priv_size += sizeof(u32); 1793 } 1794 annotation_config__init(); 1795 } 1796 1797 if (symbol__init(&session->header.env) < 0) 1798 goto error; 1799 1800 if (report.time_str) { 1801 ret = perf_time__parse_for_ranges(report.time_str, session, 1802 &report.ptime_range, 1803 &report.range_size, 1804 &report.range_num); 1805 if (ret < 0) 1806 goto error; 1807 1808 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1809 report.ptime_range, 1810 report.range_num); 1811 } 1812 1813 #ifdef HAVE_LIBTRACEEVENT 1814 if (session->tevent.pevent && 1815 tep_set_function_resolver(session->tevent.pevent, 1816 machine__resolve_kernel_addr, 1817 &session->machines.host) < 0) { 1818 pr_err("%s: failed to set libtraceevent function resolver\n", 1819 __func__); 1820 return -1; 1821 } 1822 #endif 1823 sort__setup_elide(stdout); 1824 1825 ret = __cmd_report(&report); 1826 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1827 perf_session__delete(session); 1828 last_key = K_SWITCH_INPUT_DATA; 1829 goto repeat; 1830 } else 1831 ret = 0; 1832 1833 if (!use_browser && (verbose > 2 || debug_kmaps)) 1834 perf_session__dump_kmaps(session); 1835 error: 1836 if (report.ptime_range) { 1837 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1838 zfree(&report.ptime_range); 1839 } 1840 1841 if (report.block_reports) { 1842 block_info__free_report(report.block_reports, 1843 report.nr_block_reports); 1844 report.block_reports = NULL; 1845 } 1846 1847 zstd_fini(&(session->zstd_data)); 1848 perf_session__delete(session); 1849 exit: 1850 annotation_options__exit(); 1851 free(sort_order_help); 1852 free(field_order_help); 1853 return ret; 1854 } 1855