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 u64 total_samples; 116 u64 singlethreaded_samples; 117 int socket_filter; 118 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 119 struct branch_type_stat brtype_stat; 120 bool symbol_ipc; 121 bool total_cycles_mode; 122 struct block_report *block_reports; 123 int nr_block_reports; 124 }; 125 126 static int report__config(const char *var, const char *value, void *cb) 127 { 128 struct report *rep = cb; 129 130 if (!strcmp(var, "report.group")) { 131 symbol_conf.event_group = perf_config_bool(var, value); 132 return 0; 133 } 134 if (!strcmp(var, "report.percent-limit")) { 135 double pcnt = strtof(value, NULL); 136 137 rep->min_percent = pcnt; 138 callchain_param.min_percent = pcnt; 139 return 0; 140 } 141 if (!strcmp(var, "report.children")) { 142 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 143 return 0; 144 } 145 if (!strcmp(var, "report.queue-size")) 146 return perf_config_u64(&rep->queue_size, var, value); 147 148 if (!strcmp(var, "report.sort_order")) { 149 default_sort_order = strdup(value); 150 if (!default_sort_order) { 151 pr_err("Not enough memory for report.sort_order\n"); 152 return -1; 153 } 154 return 0; 155 } 156 157 if (!strcmp(var, "report.skip-empty")) { 158 rep->skip_empty = perf_config_bool(var, value); 159 return 0; 160 } 161 162 pr_debug("%s variable unknown, ignoring...", var); 163 return 0; 164 } 165 166 static int hist_iter__report_callback(struct hist_entry_iter *iter, 167 struct addr_location *al, bool single, 168 void *arg) 169 { 170 int err = 0; 171 struct report *rep = arg; 172 struct hist_entry *he = iter->he; 173 struct evsel *evsel = iter->evsel; 174 struct perf_sample *sample = iter->sample; 175 struct mem_info *mi; 176 struct branch_info *bi; 177 178 if (!ui__has_annotation() && !rep->symbol_ipc) 179 return 0; 180 181 if (sort__mode == SORT_MODE__BRANCH) { 182 bi = he->branch_info; 183 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 184 if (err) 185 goto out; 186 187 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 188 189 } else if (rep->mem_mode) { 190 mi = he->mem_info; 191 err = addr_map_symbol__inc_samples(mem_info__daddr(mi), sample, evsel); 192 if (err) 193 goto out; 194 195 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 196 197 } else if (symbol_conf.cumulate_callchain) { 198 if (single) 199 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 200 } else { 201 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 202 } 203 204 out: 205 return err; 206 } 207 208 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 209 struct addr_location *al __maybe_unused, 210 bool single __maybe_unused, 211 void *arg) 212 { 213 struct hist_entry *he = iter->he; 214 struct report *rep = arg; 215 struct branch_info *bi = he->branch_info; 216 struct perf_sample *sample = iter->sample; 217 struct evsel *evsel = iter->evsel; 218 int err; 219 220 branch_type_count(&rep->brtype_stat, &bi->flags, 221 bi->from.addr, bi->to.addr); 222 223 if (!ui__has_annotation() && !rep->symbol_ipc) 224 return 0; 225 226 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 227 if (err) 228 goto out; 229 230 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 231 232 out: 233 return err; 234 } 235 236 static void setup_forced_leader(struct report *report, 237 struct evlist *evlist) 238 { 239 if (report->group_set) 240 evlist__force_leader(evlist); 241 } 242 243 static int process_feature_event(struct perf_session *session, 244 union perf_event *event) 245 { 246 struct report *rep = container_of(session->tool, struct report, tool); 247 248 if (event->feat.feat_id < HEADER_LAST_FEATURE) 249 return perf_event__process_feature(session, event); 250 251 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 252 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 253 event->feat.feat_id); 254 return -1; 255 } else if (rep->header_only) { 256 session_done = 1; 257 } 258 259 /* 260 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 261 * means all features are received, now we can force the 262 * group if needed. 263 */ 264 setup_forced_leader(rep, session->evlist); 265 return 0; 266 } 267 268 static int process_sample_event(const struct perf_tool *tool, 269 union perf_event *event, 270 struct perf_sample *sample, 271 struct evsel *evsel, 272 struct machine *machine) 273 { 274 struct report *rep = container_of(tool, struct report, tool); 275 struct addr_location al; 276 struct hist_entry_iter iter = { 277 .evsel = evsel, 278 .sample = sample, 279 .hide_unresolved = symbol_conf.hide_unresolved, 280 .add_entry_cb = hist_iter__report_callback, 281 }; 282 int ret = 0; 283 284 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 285 sample->time)) { 286 return 0; 287 } 288 289 if (evswitch__discard(&rep->evswitch, evsel)) 290 return 0; 291 292 addr_location__init(&al); 293 if (machine__resolve(machine, &al, sample) < 0) { 294 pr_debug("problem processing %d event, skipping it.\n", 295 event->header.type); 296 ret = -1; 297 goto out_put; 298 } 299 300 if (rep->stitch_lbr) 301 thread__set_lbr_stitch_enable(al.thread, true); 302 303 if (symbol_conf.hide_unresolved && al.sym == NULL) 304 goto out_put; 305 306 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 307 goto out_put; 308 309 if (sort__mode == SORT_MODE__BRANCH) { 310 /* 311 * A non-synthesized event might not have a branch stack if 312 * branch stacks have been synthesized (using itrace options). 313 */ 314 if (!sample->branch_stack) 315 goto out_put; 316 317 iter.add_entry_cb = hist_iter__branch_callback; 318 iter.ops = &hist_iter_branch; 319 } else if (rep->mem_mode) { 320 iter.ops = &hist_iter_mem; 321 } else if (symbol_conf.cumulate_callchain) { 322 iter.ops = &hist_iter_cumulative; 323 } else { 324 iter.ops = &hist_iter_normal; 325 } 326 327 if (al.map != NULL) 328 dso__set_hit(map__dso(al.map)); 329 330 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 331 hist__account_cycles(sample->branch_stack, &al, sample, 332 rep->nonany_branch_mode, 333 &rep->total_cycles, evsel); 334 } 335 336 rep->total_samples++; 337 if (al.parallelism == 1) 338 rep->singlethreaded_samples++; 339 340 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 341 if (ret < 0) 342 pr_debug("problem adding hist entry, skipping event\n"); 343 out_put: 344 addr_location__exit(&al); 345 return ret; 346 } 347 348 static int process_read_event(const struct perf_tool *tool, 349 union perf_event *event, 350 struct perf_sample *sample __maybe_unused, 351 struct evsel *evsel, 352 struct machine *machine __maybe_unused) 353 { 354 struct report *rep = container_of(tool, struct report, tool); 355 356 if (rep->show_threads) { 357 int err = perf_read_values_add_value(&rep->show_threads_values, 358 event->read.pid, event->read.tid, 359 evsel, 360 event->read.value); 361 362 if (err) 363 return err; 364 } 365 366 return 0; 367 } 368 369 /* For pipe mode, sample_type is not currently set */ 370 static int report__setup_sample_type(struct report *rep) 371 { 372 struct perf_session *session = rep->session; 373 u64 sample_type = evlist__combined_sample_type(session->evlist); 374 bool is_pipe = perf_data__is_pipe(session->data); 375 struct evsel *evsel; 376 377 if (session->itrace_synth_opts->callchain || 378 session->itrace_synth_opts->add_callchain || 379 (!is_pipe && 380 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 381 !session->itrace_synth_opts->set)) 382 sample_type |= PERF_SAMPLE_CALLCHAIN; 383 384 if (session->itrace_synth_opts->last_branch || 385 session->itrace_synth_opts->add_last_branch) 386 sample_type |= PERF_SAMPLE_BRANCH_STACK; 387 388 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 389 if (perf_hpp_list.parent) { 390 ui__error("Selected --sort parent, but no " 391 "callchain data. Did you call " 392 "'perf record' without -g?\n"); 393 return -EINVAL; 394 } 395 if (symbol_conf.use_callchain && 396 !symbol_conf.show_branchflag_count) { 397 ui__error("Selected -g or --branch-history.\n" 398 "But no callchain or branch data.\n" 399 "Did you call 'perf record' without -g or -b?\n"); 400 return -1; 401 } 402 } else if (!callchain_param.enabled && 403 callchain_param.mode != CHAIN_NONE && 404 !symbol_conf.use_callchain) { 405 symbol_conf.use_callchain = true; 406 if (callchain_register_param(&callchain_param) < 0) { 407 ui__error("Can't register callchain params.\n"); 408 return -EINVAL; 409 } 410 } 411 412 if (symbol_conf.cumulate_callchain) { 413 /* Silently ignore if callchain is missing */ 414 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 415 symbol_conf.cumulate_callchain = false; 416 perf_hpp__cancel_cumulate(); 417 } 418 } 419 420 if (sort__mode == SORT_MODE__BRANCH) { 421 if (!is_pipe && 422 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 423 ui__error("Selected -b but no branch data. " 424 "Did you call perf record without -b?\n"); 425 return -1; 426 } 427 } 428 429 if (sort__mode == SORT_MODE__MEMORY) { 430 /* 431 * FIXUP: prior to kernel 5.18, Arm SPE missed to set 432 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward 433 * compatibility, set the bit if it's an old perf data file. 434 */ 435 evlist__for_each_entry(session->evlist, evsel) { 436 if (strstr(evsel__name(evsel), "arm_spe") && 437 !(sample_type & PERF_SAMPLE_DATA_SRC)) { 438 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC; 439 sample_type |= PERF_SAMPLE_DATA_SRC; 440 } 441 } 442 443 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 444 ui__error("Selected --mem-mode but no mem data. " 445 "Did you call perf record without -d?\n"); 446 return -1; 447 } 448 } 449 450 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env)); 451 452 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 453 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 454 "Please apply --call-graph lbr when recording.\n"); 455 rep->stitch_lbr = false; 456 } 457 458 /* ??? handle more cases than just ANY? */ 459 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 460 rep->nonany_branch_mode = true; 461 462 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_LIBDW_SUPPORT) 463 if (dwarf_callchain_users) { 464 ui__warning("Please install libunwind or libdw " 465 "development packages during the perf build.\n"); 466 } 467 #endif 468 469 return 0; 470 } 471 472 static void sig_handler(int sig __maybe_unused) 473 { 474 session_done = 1; 475 } 476 477 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 478 const char *evname, FILE *fp) 479 { 480 size_t ret; 481 char unit; 482 unsigned long nr_samples = hists->stats.nr_samples; 483 u64 nr_events = hists->stats.total_period; 484 struct evsel *evsel = hists_to_evsel(hists); 485 char buf[512]; 486 size_t size = sizeof(buf); 487 int socked_id = hists->socket_filter; 488 489 if (quiet) 490 return 0; 491 492 if (symbol_conf.filter_relative) { 493 nr_samples = hists->stats.nr_non_filtered_samples; 494 nr_events = hists->stats.total_non_filtered_period; 495 } 496 497 if (evsel__is_group_event(evsel)) { 498 struct evsel *pos; 499 500 evsel__group_desc(evsel, buf, size); 501 evname = buf; 502 503 for_each_group_member(pos, evsel) { 504 const struct hists *pos_hists = evsel__hists(pos); 505 506 if (symbol_conf.filter_relative) { 507 nr_samples += pos_hists->stats.nr_non_filtered_samples; 508 nr_events += pos_hists->stats.total_non_filtered_period; 509 } else { 510 nr_samples += pos_hists->stats.nr_samples; 511 nr_events += pos_hists->stats.total_period; 512 } 513 } 514 } 515 516 nr_samples = convert_unit(nr_samples, &unit); 517 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 518 if (evname != NULL) { 519 ret += fprintf(fp, " of event%s '%s'", 520 evsel->core.nr_members > 1 ? "s" : "", evname); 521 } 522 523 if (rep->time_str) 524 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 525 526 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 527 ret += fprintf(fp, ", show reference callgraph"); 528 } 529 530 if (rep->mem_mode) { 531 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 532 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 533 } else 534 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 535 536 if (socked_id > -1) 537 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 538 539 return ret + fprintf(fp, "\n#\n"); 540 } 541 542 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 543 { 544 struct evsel *pos; 545 int i = 0, ret; 546 547 evlist__for_each_entry(evlist, pos) { 548 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 549 rep->min_percent, pos, 550 &rep->session->header.env); 551 if (ret != 0) 552 return ret; 553 } 554 555 return 0; 556 } 557 558 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 559 { 560 struct evsel *pos; 561 int i = 0; 562 563 if (!quiet) { 564 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 565 evlist->stats.total_lost_samples); 566 } 567 568 evlist__for_each_entry(evlist, pos) { 569 struct hists *hists = evsel__hists(pos); 570 const char *evname = evsel__name(pos); 571 572 i++; 573 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 574 continue; 575 576 if (rep->skip_empty && !hists->stats.nr_samples) 577 continue; 578 579 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 580 581 if (rep->total_cycles_mode) { 582 char *buf; 583 584 if (!annotation_br_cntr_abbr_list(&buf, pos, true)) { 585 fprintf(stdout, "%s", buf); 586 fprintf(stdout, "#\n"); 587 free(buf); 588 } 589 report__browse_block_hists(&rep->block_reports[i - 1].hist, 590 rep->min_percent, pos, NULL); 591 continue; 592 } 593 594 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 595 !(symbol_conf.use_callchain || 596 symbol_conf.show_branchflag_count)); 597 fprintf(stdout, "\n\n"); 598 } 599 600 if (!quiet) 601 fprintf(stdout, "#\n# (%s)\n#\n", help); 602 603 if (rep->show_threads) { 604 bool style = !strcmp(rep->pretty_printing_style, "raw"); 605 perf_read_values_display(stdout, &rep->show_threads_values, 606 style); 607 perf_read_values_destroy(&rep->show_threads_values); 608 } 609 610 if (sort__mode == SORT_MODE__BRANCH) 611 branch_type_stat_display(stdout, &rep->brtype_stat); 612 613 return 0; 614 } 615 616 static void report__warn_kptr_restrict(const struct report *rep) 617 { 618 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 619 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 620 621 if (evlist__exclude_kernel(rep->session->evlist)) 622 return; 623 624 if (kernel_map == NULL || 625 (dso__hit(map__dso(kernel_map)) && 626 (kernel_kmap->ref_reloc_sym == NULL || 627 kernel_kmap->ref_reloc_sym->addr == 0))) { 628 const char *desc = 629 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 630 "can't be resolved."; 631 632 if (kernel_map && map__has_symbols(kernel_map)) { 633 desc = "If some relocation was applied (e.g. " 634 "kexec) symbols may be misresolved."; 635 } 636 637 ui__warning( 638 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 639 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 640 "Samples in kernel modules can't be resolved as well.\n\n", 641 desc); 642 } 643 } 644 645 static int report__gtk_browse_hists(struct report *rep, const char *help) 646 { 647 int (*hist_browser)(struct evlist *evlist, const char *help, 648 struct hist_browser_timer *timer, float min_pcnt); 649 650 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 651 652 if (hist_browser == NULL) { 653 ui__error("GTK browser not found!\n"); 654 return -1; 655 } 656 657 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 658 } 659 660 static int report__browse_hists(struct report *rep) 661 { 662 int ret; 663 struct perf_session *session = rep->session; 664 struct evlist *evlist = session->evlist; 665 char *help = NULL, *path = NULL; 666 667 path = system_path(TIPDIR); 668 if (perf_tip(&help, path) || help == NULL) { 669 /* fallback for people who don't install perf ;-) */ 670 free(path); 671 path = system_path(DOCDIR); 672 if (perf_tip(&help, path) || help == NULL) 673 help = strdup("Cannot load tips.txt file, please install perf!"); 674 } 675 free(path); 676 677 switch (use_browser) { 678 case 1: 679 if (rep->total_cycles_mode) { 680 ret = evlist__tui_block_hists_browse(evlist, rep); 681 break; 682 } 683 684 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 685 &session->header.env, true); 686 /* 687 * Usually "ret" is the last pressed key, and we only 688 * care if the key notifies us to switch data file. 689 */ 690 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 691 ret = 0; 692 break; 693 case 2: 694 ret = report__gtk_browse_hists(rep, help); 695 break; 696 default: 697 ret = evlist__tty_browse_hists(evlist, rep, help); 698 break; 699 } 700 free(help); 701 return ret; 702 } 703 704 static int report__collapse_hists(struct report *rep) 705 { 706 struct perf_session *session = rep->session; 707 struct evlist *evlist = session->evlist; 708 struct ui_progress prog; 709 struct evsel *pos; 710 int ret = 0; 711 712 /* 713 * The pipe data needs to setup hierarchy hpp formats now, because it 714 * cannot know about evsels in the data before reading the data. The 715 * normal file data saves the event (attribute) info in the header 716 * section, but pipe does not have the luxury. 717 */ 718 if (perf_data__is_pipe(session->data)) { 719 if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) { 720 ui__error("Failed to setup hierarchy output formats\n"); 721 return -1; 722 } 723 } 724 725 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 726 727 evlist__for_each_entry(rep->session->evlist, pos) { 728 struct hists *hists = evsel__hists(pos); 729 730 if (pos->core.idx == 0) 731 hists->symbol_filter_str = rep->symbol_filter_str; 732 733 hists->socket_filter = rep->socket_filter; 734 735 ret = hists__collapse_resort(hists, &prog); 736 if (ret < 0) 737 break; 738 739 /* Non-group events are considered as leader */ 740 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 741 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 742 743 hists__match(leader_hists, hists); 744 hists__link(leader_hists, hists); 745 } 746 } 747 748 ui_progress__finish(); 749 return ret; 750 } 751 752 static int hists__resort_cb(struct hist_entry *he, void *arg) 753 { 754 struct report *rep = arg; 755 struct symbol *sym = he->ms.sym; 756 757 if (rep->symbol_ipc && sym && !sym->annotate2) { 758 struct evsel *evsel = hists_to_evsel(he->hists); 759 760 symbol__annotate2(&he->ms, evsel, NULL); 761 } 762 763 return 0; 764 } 765 766 static void report__output_resort(struct report *rep) 767 { 768 struct ui_progress prog; 769 struct evsel *pos; 770 771 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 772 773 evlist__for_each_entry(rep->session->evlist, pos) { 774 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 775 } 776 777 ui_progress__finish(); 778 } 779 780 static int count_sample_event(const struct perf_tool *tool __maybe_unused, 781 union perf_event *event __maybe_unused, 782 struct perf_sample *sample __maybe_unused, 783 struct evsel *evsel, 784 struct machine *machine __maybe_unused) 785 { 786 struct hists *hists = evsel__hists(evsel); 787 788 hists__inc_nr_events(hists); 789 return 0; 790 } 791 792 static int count_lost_samples_event(const struct perf_tool *tool, 793 union perf_event *event, 794 struct perf_sample *sample, 795 struct machine *machine __maybe_unused) 796 { 797 struct report *rep = container_of(tool, struct report, tool); 798 struct evsel *evsel; 799 800 evsel = evlist__id2evsel(rep->session->evlist, sample->id); 801 if (evsel) { 802 struct hists *hists = evsel__hists(evsel); 803 u32 count = event->lost_samples.lost; 804 805 if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF) 806 hists__inc_nr_dropped_samples(hists, count); 807 else 808 hists__inc_nr_lost_samples(hists, count); 809 } 810 return 0; 811 } 812 813 static int process_attr(const struct perf_tool *tool __maybe_unused, 814 union perf_event *event, 815 struct evlist **pevlist); 816 817 static void stats_setup(struct report *rep) 818 { 819 perf_tool__init(&rep->tool, /*ordered_events=*/false); 820 rep->tool.attr = process_attr; 821 rep->tool.sample = count_sample_event; 822 rep->tool.lost_samples = count_lost_samples_event; 823 rep->tool.event_update = perf_event__process_event_update; 824 rep->tool.no_warn = true; 825 } 826 827 static int stats_print(struct report *rep) 828 { 829 struct perf_session *session = rep->session; 830 831 perf_session__fprintf_nr_events(session, stdout); 832 evlist__fprintf_nr_events(session->evlist, stdout); 833 return 0; 834 } 835 836 static void tasks_setup(struct report *rep) 837 { 838 perf_tool__init(&rep->tool, /*ordered_events=*/true); 839 if (rep->mmaps_mode) { 840 rep->tool.mmap = perf_event__process_mmap; 841 rep->tool.mmap2 = perf_event__process_mmap2; 842 } 843 rep->tool.attr = process_attr; 844 rep->tool.comm = perf_event__process_comm; 845 rep->tool.exit = perf_event__process_exit; 846 rep->tool.fork = perf_event__process_fork; 847 rep->tool.no_warn = true; 848 } 849 850 struct maps__fprintf_task_args { 851 int indent; 852 FILE *fp; 853 size_t printed; 854 }; 855 856 static int maps__fprintf_task_cb(struct map *map, void *data) 857 { 858 struct maps__fprintf_task_args *args = data; 859 const struct dso *dso = map__dso(map); 860 u32 prot = map__prot(map); 861 int ret; 862 863 ret = fprintf(args->fp, 864 "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 865 args->indent, "", map__start(map), map__end(map), 866 prot & PROT_READ ? 'r' : '-', 867 prot & PROT_WRITE ? 'w' : '-', 868 prot & PROT_EXEC ? 'x' : '-', 869 map__flags(map) ? 's' : 'p', 870 map__pgoff(map), 871 dso__id_const(dso)->ino, dso__name(dso)); 872 873 if (ret < 0) 874 return ret; 875 876 args->printed += ret; 877 return 0; 878 } 879 880 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 881 { 882 struct maps__fprintf_task_args args = { 883 .indent = indent, 884 .fp = fp, 885 .printed = 0, 886 }; 887 888 maps__for_each_map(maps, maps__fprintf_task_cb, &args); 889 890 return args.printed; 891 } 892 893 static int thread_level(struct machine *machine, const struct thread *thread) 894 { 895 struct thread *parent_thread; 896 int res; 897 898 if (thread__tid(thread) <= 0) 899 return 0; 900 901 if (thread__ppid(thread) <= 0) 902 return 1; 903 904 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread)); 905 if (!parent_thread) { 906 pr_err("Missing parent thread of %d\n", thread__tid(thread)); 907 return 0; 908 } 909 res = 1 + thread_level(machine, parent_thread); 910 thread__put(parent_thread); 911 return res; 912 } 913 914 static void task__print_level(struct machine *machine, struct thread *thread, FILE *fp) 915 { 916 int level = thread_level(machine, thread); 917 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 918 thread__pid(thread), thread__tid(thread), 919 thread__ppid(thread), level, ""); 920 921 fprintf(fp, "%s\n", thread__comm_str(thread)); 922 923 maps__fprintf_task(thread__maps(thread), comm_indent, fp); 924 } 925 926 /* 927 * Sort two thread list nodes such that they form a tree. The first node is the 928 * root of the tree, its children are ordered numerically after it. If a child 929 * has children itself then they appear immediately after their parent. For 930 * example, the 4 threads in the order they'd appear in the list: 931 * - init with a TID 1 and a parent of 0 932 * - systemd with a TID 3000 and a parent of init/1 933 * - systemd child thread with TID 4000, the parent is 3000 934 * - NetworkManager is a child of init with a TID of 3500. 935 */ 936 static int task_list_cmp(void *priv, const struct list_head *la, const struct list_head *lb) 937 { 938 struct machine *machine = priv; 939 struct thread_list *task_a = list_entry(la, struct thread_list, list); 940 struct thread_list *task_b = list_entry(lb, struct thread_list, list); 941 struct thread *a = task_a->thread; 942 struct thread *b = task_b->thread; 943 int level_a, level_b, res; 944 945 /* Same thread? */ 946 if (thread__tid(a) == thread__tid(b)) 947 return 0; 948 949 /* Compare a and b to root. */ 950 if (thread__tid(a) == 0) 951 return -1; 952 953 if (thread__tid(b) == 0) 954 return 1; 955 956 /* If parents match sort by tid. */ 957 if (thread__ppid(a) == thread__ppid(b)) 958 return thread__tid(a) < thread__tid(b) ? -1 : 1; 959 960 /* 961 * Find a and b such that if they are a child of each other a and b's 962 * tid's match, otherwise a and b have a common parent and distinct 963 * tid's to sort by. First make the depths of the threads match. 964 */ 965 level_a = thread_level(machine, a); 966 level_b = thread_level(machine, b); 967 a = thread__get(a); 968 b = thread__get(b); 969 for (int i = level_a; i > level_b; i--) { 970 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(a)); 971 972 thread__put(a); 973 if (!parent) { 974 pr_err("Missing parent thread of %d\n", thread__tid(a)); 975 thread__put(b); 976 return -1; 977 } 978 a = parent; 979 } 980 for (int i = level_b; i > level_a; i--) { 981 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(b)); 982 983 thread__put(b); 984 if (!parent) { 985 pr_err("Missing parent thread of %d\n", thread__tid(b)); 986 thread__put(a); 987 return 1; 988 } 989 b = parent; 990 } 991 /* Search up to a common parent. */ 992 while (thread__ppid(a) != thread__ppid(b)) { 993 struct thread *parent; 994 995 parent = machine__find_thread(machine, -1, thread__ppid(a)); 996 thread__put(a); 997 if (!parent) 998 pr_err("Missing parent thread of %d\n", thread__tid(a)); 999 a = parent; 1000 parent = machine__find_thread(machine, -1, thread__ppid(b)); 1001 thread__put(b); 1002 if (!parent) 1003 pr_err("Missing parent thread of %d\n", thread__tid(b)); 1004 b = parent; 1005 if (!a || !b) { 1006 /* Handle missing parent (unexpected) with some sanity. */ 1007 thread__put(a); 1008 thread__put(b); 1009 return !a && !b ? 0 : (!a ? -1 : 1); 1010 } 1011 } 1012 if (thread__tid(a) == thread__tid(b)) { 1013 /* a is a child of b or vice-versa, deeper levels appear later. */ 1014 res = level_a < level_b ? -1 : (level_a > level_b ? 1 : 0); 1015 } else { 1016 /* Sort by tid now the parent is the same. */ 1017 res = thread__tid(a) < thread__tid(b) ? -1 : 1; 1018 } 1019 thread__put(a); 1020 thread__put(b); 1021 return res; 1022 } 1023 1024 static int tasks_print(struct report *rep, FILE *fp) 1025 { 1026 struct machine *machine = &rep->session->machines.host; 1027 LIST_HEAD(tasks); 1028 int ret; 1029 1030 ret = machine__thread_list(machine, &tasks); 1031 if (!ret) { 1032 struct thread_list *task; 1033 1034 list_sort(machine, &tasks, task_list_cmp); 1035 1036 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 1037 1038 list_for_each_entry(task, &tasks, list) 1039 task__print_level(machine, task->thread, fp); 1040 } 1041 thread_list__delete(&tasks); 1042 return ret; 1043 } 1044 1045 static int __cmd_report(struct report *rep) 1046 { 1047 int ret; 1048 struct perf_session *session = rep->session; 1049 struct evsel *pos; 1050 struct perf_data *data = session->data; 1051 1052 signal(SIGINT, sig_handler); 1053 1054 if (rep->cpu_list) { 1055 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 1056 rep->cpu_bitmap); 1057 if (ret) { 1058 ui__error("failed to set cpu bitmap\n"); 1059 return ret; 1060 } 1061 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 1062 } 1063 1064 if (rep->show_threads) { 1065 ret = perf_read_values_init(&rep->show_threads_values); 1066 if (ret) 1067 return ret; 1068 } 1069 1070 ret = report__setup_sample_type(rep); 1071 if (ret) { 1072 /* report__setup_sample_type() already showed error message */ 1073 return ret; 1074 } 1075 1076 if (rep->stats_mode) 1077 stats_setup(rep); 1078 1079 if (rep->tasks_mode) 1080 tasks_setup(rep); 1081 1082 ret = perf_session__process_events(session); 1083 if (ret) { 1084 ui__error("failed to process sample\n"); 1085 return ret; 1086 } 1087 1088 /* Don't show Latency column for non-parallel profiles by default. */ 1089 if (!symbol_conf.prefer_latency && rep->total_samples && 1090 rep->singlethreaded_samples * 100 / rep->total_samples >= 99) 1091 perf_hpp__cancel_latency(); 1092 1093 evlist__check_mem_load_aux(session->evlist); 1094 1095 if (rep->stats_mode) 1096 return stats_print(rep); 1097 1098 if (rep->tasks_mode) 1099 return tasks_print(rep, stdout); 1100 1101 report__warn_kptr_restrict(rep); 1102 1103 evlist__for_each_entry(session->evlist, pos) 1104 rep->nr_entries += evsel__hists(pos)->nr_entries; 1105 1106 if (use_browser == 0) { 1107 if (verbose > 3) 1108 perf_session__fprintf(session, stdout); 1109 1110 if (verbose > 2) 1111 perf_session__fprintf_dsos(session, stdout); 1112 1113 if (dump_trace) { 1114 stats_print(rep); 1115 return 0; 1116 } 1117 } 1118 1119 ret = report__collapse_hists(rep); 1120 if (ret) { 1121 ui__error("failed to process hist entry\n"); 1122 return ret; 1123 } 1124 1125 if (session_done()) 1126 return 0; 1127 1128 /* 1129 * recalculate number of entries after collapsing since it 1130 * might be changed during the collapse phase. 1131 */ 1132 rep->nr_entries = 0; 1133 evlist__for_each_entry(session->evlist, pos) 1134 rep->nr_entries += evsel__hists(pos)->nr_entries; 1135 1136 if (rep->nr_entries == 0) { 1137 ui__error("The %s data has no samples!\n", data->path); 1138 return 0; 1139 } 1140 1141 report__output_resort(rep); 1142 1143 if (rep->total_cycles_mode) { 1144 int nr_hpps = 4; 1145 int block_hpps[PERF_HPP_REPORT__BLOCK_MAX_INDEX] = { 1146 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1147 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1148 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1149 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1150 }; 1151 1152 if (session->evlist->nr_br_cntr > 0) 1153 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER; 1154 1155 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_RANGE; 1156 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_DSO; 1157 1158 rep->block_reports = block_info__create_report(session->evlist, 1159 rep->total_cycles, 1160 block_hpps, nr_hpps, 1161 &rep->nr_block_reports); 1162 if (!rep->block_reports) 1163 return -1; 1164 } 1165 1166 return report__browse_hists(rep); 1167 } 1168 1169 static int 1170 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1171 { 1172 struct callchain_param *callchain = opt->value; 1173 1174 callchain->enabled = !unset; 1175 /* 1176 * --no-call-graph 1177 */ 1178 if (unset) { 1179 symbol_conf.use_callchain = false; 1180 callchain->mode = CHAIN_NONE; 1181 return 0; 1182 } 1183 1184 return parse_callchain_report_opt(arg); 1185 } 1186 1187 static int 1188 parse_time_quantum(const struct option *opt, const char *arg, 1189 int unset __maybe_unused) 1190 { 1191 unsigned long *time_q = opt->value; 1192 char *end; 1193 1194 *time_q = strtoul(arg, &end, 0); 1195 if (end == arg) 1196 goto parse_err; 1197 if (*time_q == 0) { 1198 pr_err("time quantum cannot be 0"); 1199 return -1; 1200 } 1201 end = skip_spaces(end); 1202 if (*end == 0) 1203 return 0; 1204 if (!strcmp(end, "s")) { 1205 *time_q *= NSEC_PER_SEC; 1206 return 0; 1207 } 1208 if (!strcmp(end, "ms")) { 1209 *time_q *= NSEC_PER_MSEC; 1210 return 0; 1211 } 1212 if (!strcmp(end, "us")) { 1213 *time_q *= NSEC_PER_USEC; 1214 return 0; 1215 } 1216 if (!strcmp(end, "ns")) 1217 return 0; 1218 parse_err: 1219 pr_err("Cannot parse time quantum `%s'\n", arg); 1220 return -1; 1221 } 1222 1223 int 1224 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1225 const char *arg, int unset __maybe_unused) 1226 { 1227 if (arg) { 1228 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1229 if (err) { 1230 char buf[BUFSIZ]; 1231 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1232 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1233 return -1; 1234 } 1235 have_ignore_callees = 1; 1236 } 1237 1238 return 0; 1239 } 1240 1241 static int 1242 parse_branch_mode(const struct option *opt, 1243 const char *str __maybe_unused, int unset) 1244 { 1245 int *branch_mode = opt->value; 1246 1247 *branch_mode = !unset; 1248 return 0; 1249 } 1250 1251 static int 1252 parse_percent_limit(const struct option *opt, const char *str, 1253 int unset __maybe_unused) 1254 { 1255 struct report *rep = opt->value; 1256 double pcnt = strtof(str, NULL); 1257 1258 rep->min_percent = pcnt; 1259 callchain_param.min_percent = pcnt; 1260 return 0; 1261 } 1262 1263 static int process_attr(const struct perf_tool *tool __maybe_unused, 1264 union perf_event *event, 1265 struct evlist **pevlist) 1266 { 1267 u64 sample_type; 1268 int err; 1269 1270 err = perf_event__process_attr(tool, event, pevlist); 1271 if (err) 1272 return err; 1273 1274 /* 1275 * Check if we need to enable callchains based 1276 * on events sample_type. 1277 */ 1278 sample_type = evlist__combined_sample_type(*pevlist); 1279 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env)); 1280 return 0; 1281 } 1282 1283 #define CALLCHAIN_BRANCH_SORT_ORDER \ 1284 "srcline,symbol,dso,callchain_branch_predicted," \ 1285 "callchain_branch_abort,callchain_branch_cycles" 1286 1287 int cmd_report(int argc, const char **argv) 1288 { 1289 struct perf_session *session; 1290 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1291 struct stat st; 1292 bool has_br_stack = false; 1293 int branch_mode = -1; 1294 int last_key = 0; 1295 bool branch_call_mode = false; 1296 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1297 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1298 CALLCHAIN_REPORT_HELP 1299 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1300 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1301 const char * const report_usage[] = { 1302 "perf report [<options>]", 1303 NULL 1304 }; 1305 struct report report = { 1306 .max_stack = PERF_MAX_STACK_DEPTH, 1307 .pretty_printing_style = "normal", 1308 .socket_filter = -1, 1309 .skip_empty = true, 1310 }; 1311 char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__NORMAL); 1312 char *field_order_help = sort_help("output field(s):", SORT_MODE__NORMAL); 1313 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL; 1314 const struct option options[] = { 1315 OPT_STRING('i', "input", &input_name, "file", 1316 "input file name"), 1317 OPT_INCR('v', "verbose", &verbose, 1318 "be more verbose (show symbol address, etc)"), 1319 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"), 1320 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1321 "dump raw trace in ASCII"), 1322 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1323 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1324 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1325 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1326 "file", "vmlinux pathname"), 1327 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1328 "don't load vmlinux even if found"), 1329 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1330 "file", "kallsyms pathname"), 1331 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1332 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1333 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1334 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1335 "Show a column with the number of samples"), 1336 OPT_BOOLEAN('T', "threads", &report.show_threads, 1337 "Show per-thread event counters"), 1338 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1339 "pretty printing style key: normal raw"), 1340 #ifdef HAVE_SLANG_SUPPORT 1341 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1342 #endif 1343 #ifdef HAVE_GTK2_SUPPORT 1344 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1345 #endif 1346 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1347 "Use the stdio interface"), 1348 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1349 OPT_BOOLEAN(0, "header-only", &report.header_only, 1350 "Show only data header."), 1351 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1352 sort_order_help), 1353 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1354 field_order_help), 1355 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1356 "Show sample percentage for different cpu modes"), 1357 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1358 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1359 OPT_STRING('p', "parent", &parent_pattern, "regex", 1360 "regex filter to identify parent, see: '--sort parent'"), 1361 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1362 "Only display entries with parent-match"), 1363 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1364 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1365 report_callchain_help, &report_parse_callchain_opt, 1366 callchain_default_opt), 1367 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1368 "Accumulate callchains of children and show total overhead as well. " 1369 "Enabled by default, use --no-children to disable."), 1370 OPT_INTEGER(0, "max-stack", &report.max_stack, 1371 "Set the maximum stack depth when parsing the callchain, " 1372 "anything beyond the specified depth will be ignored. " 1373 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1374 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1375 "alias for inverted call graph"), 1376 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1377 "ignore callees of these functions in call graphs", 1378 report_parse_ignore_callees_opt), 1379 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1380 "only consider symbols in these dsos"), 1381 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1382 "only consider symbols in these comms"), 1383 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1384 "only consider symbols in these pids"), 1385 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1386 "only consider symbols in these tids"), 1387 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1388 "only consider these symbols"), 1389 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1390 "only show symbols that (partially) match with this filter"), 1391 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1392 "width[,width...]", 1393 "don't try to adjust column width, use these fixed values"), 1394 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1395 "separator for columns, no spaces will be added between " 1396 "columns '.' is reserved."), 1397 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1398 "Only display entries resolved to a symbol"), 1399 OPT_CALLBACK(0, "symfs", NULL, "directory", 1400 "Look for files with symbols relative to this directory", 1401 symbol__config_symfs), 1402 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1403 "list of cpus to profile"), 1404 OPT_STRING(0, "parallelism", &symbol_conf.parallelism_list_str, "parallelism", 1405 "only consider these parallelism levels (cpu set format)"), 1406 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1407 "Display extended information about perf.data file"), 1408 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src, 1409 "Interleave source code with assembly code (default)"), 1410 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw, 1411 "Display raw encoding of assembly instructions (default)"), 1412 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1413 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1414 OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix", 1415 "Add prefix to source file path names in programs (with --prefix-strip)"), 1416 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N", 1417 "Strip first N entries of source file path name in programs (with --prefix)"), 1418 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1419 "Show a column with the sum of periods"), 1420 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1421 "Show event group information together"), 1422 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1423 "Sort the output by the event at the index n in group. " 1424 "If n is invalid, sort by the first event. " 1425 "WARNING: should be used on grouped events."), 1426 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1427 "use branch records for per branch histogram filling", 1428 parse_branch_mode), 1429 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1430 "add last branch records to call history"), 1431 OPT_STRING(0, "objdump", &objdump_path, "path", 1432 "objdump binary to use for disassembly and annotations"), 1433 OPT_STRING(0, "addr2line", &addr2line_path, "path", 1434 "addr2line binary to use for line numbers"), 1435 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1436 "Symbol demangling. Enabled by default, use --no-demangle to disable."), 1437 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1438 "Enable kernel symbol demangling"), 1439 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1440 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1441 "Number of samples to save per histogram entry for individual browsing"), 1442 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1443 "Don't show entries under that percent", parse_percent_limit), 1444 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1445 "how to display percentage of filtered entries", parse_filter_percentage), 1446 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1447 "Instruction Tracing options\n" ITRACE_HELP, 1448 itrace_parse_synth_opts), 1449 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1450 "Show full source file name path for source lines"), 1451 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1452 "Show callgraph from reference event"), 1453 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1454 "Enable LBR callgraph stitching approach"), 1455 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1456 "only show processor socket that match with this filter"), 1457 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1458 "Show raw trace event output (do not use print fmt or plugins)"), 1459 OPT_BOOLEAN('H', "hierarchy", &symbol_conf.report_hierarchy, 1460 "Show entries in a hierarchy"), 1461 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1462 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1463 stdio__config_color, "always"), 1464 OPT_STRING(0, "time", &report.time_str, "str", 1465 "Time span of interest (start,stop)"), 1466 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1467 "Show inline function"), 1468 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period", 1469 "Set percent type local/global-period/hits", 1470 annotate_parse_percent_type), 1471 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1472 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1473 "Set time quantum for time sort key (default 100ms)", 1474 parse_time_quantum), 1475 OPTS_EVSWITCH(&report.evswitch), 1476 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1477 "Sort all blocks by 'Sampled Cycles%'"), 1478 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1479 "Disable raw trace ordering"), 1480 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1481 "Do not display empty (or dummy) events in the output"), 1482 OPT_BOOLEAN(0, "latency", &symbol_conf.prefer_latency, 1483 "Show latency-centric profile rather than the default\n" 1484 "\t\t\t CPU-consumption-centric profile\n" 1485 "\t\t\t (requires perf record --latency flag)."), 1486 OPT_END() 1487 }; 1488 struct perf_data data = { 1489 .mode = PERF_DATA_MODE_READ, 1490 }; 1491 int ret = hists__init(); 1492 char sort_tmp[128]; 1493 bool ordered_events = true; 1494 1495 if (ret < 0) 1496 goto exit; 1497 1498 /* 1499 * tasks_mode require access to exited threads to list those that are in 1500 * the data file. Off-cpu events are synthesized after other events and 1501 * reference exited threads. 1502 */ 1503 symbol_conf.keep_exited_threads = true; 1504 1505 annotation_options__init(); 1506 1507 ret = perf_config(report__config, &report); 1508 if (ret) 1509 goto exit; 1510 1511 argc = parse_options(argc, argv, options, report_usage, 0); 1512 if (argc) { 1513 /* 1514 * Special case: if there's an argument left then assume that 1515 * it's a symbol filter: 1516 */ 1517 if (argc > 1) 1518 usage_with_options(report_usage, options); 1519 1520 report.symbol_filter_str = argv[0]; 1521 } 1522 1523 if (disassembler_style) { 1524 annotate_opts.disassembler_style = strdup(disassembler_style); 1525 if (!annotate_opts.disassembler_style) 1526 return -ENOMEM; 1527 } 1528 if (objdump_path) { 1529 annotate_opts.objdump_path = strdup(objdump_path); 1530 if (!annotate_opts.objdump_path) 1531 return -ENOMEM; 1532 } 1533 if (addr2line_path) { 1534 symbol_conf.addr2line_path = strdup(addr2line_path); 1535 if (!symbol_conf.addr2line_path) 1536 return -ENOMEM; 1537 } 1538 1539 if (annotate_check_args() < 0) { 1540 ret = -EINVAL; 1541 goto exit; 1542 } 1543 1544 if (report.mmaps_mode) 1545 report.tasks_mode = true; 1546 1547 if (dump_trace && report.disable_order) 1548 ordered_events = false; 1549 1550 if (quiet) 1551 perf_quiet_option(); 1552 1553 ret = symbol__validate_sym_arguments(); 1554 if (ret) 1555 goto exit; 1556 1557 if (report.inverted_callchain) 1558 callchain_param.order = ORDER_CALLER; 1559 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1560 callchain_param.order = ORDER_CALLER; 1561 1562 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1563 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1564 report.max_stack = itrace_synth_opts.callchain_sz; 1565 1566 if (!input_name || !strlen(input_name)) { 1567 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1568 input_name = "-"; 1569 else 1570 input_name = "perf.data"; 1571 } 1572 1573 repeat: 1574 data.path = input_name; 1575 data.force = symbol_conf.force; 1576 1577 symbol_conf.skip_empty = report.skip_empty; 1578 1579 perf_tool__init(&report.tool, ordered_events); 1580 report.tool.sample = process_sample_event; 1581 report.tool.mmap = perf_event__process_mmap; 1582 report.tool.mmap2 = perf_event__process_mmap2; 1583 report.tool.comm = perf_event__process_comm; 1584 report.tool.namespaces = perf_event__process_namespaces; 1585 report.tool.cgroup = perf_event__process_cgroup; 1586 report.tool.exit = perf_event__process_exit; 1587 report.tool.fork = perf_event__process_fork; 1588 report.tool.context_switch = perf_event__process_switch; 1589 report.tool.lost = perf_event__process_lost; 1590 report.tool.read = process_read_event; 1591 report.tool.attr = process_attr; 1592 #ifdef HAVE_LIBTRACEEVENT 1593 report.tool.tracing_data = perf_event__process_tracing_data; 1594 #endif 1595 report.tool.build_id = perf_event__process_build_id; 1596 report.tool.id_index = perf_event__process_id_index; 1597 report.tool.auxtrace_info = perf_event__process_auxtrace_info; 1598 report.tool.auxtrace = perf_event__process_auxtrace; 1599 report.tool.event_update = perf_event__process_event_update; 1600 report.tool.feature = process_feature_event; 1601 report.tool.ordering_requires_timestamps = true; 1602 1603 session = perf_session__new(&data, &report.tool); 1604 if (IS_ERR(session)) { 1605 ret = PTR_ERR(session); 1606 goto exit; 1607 } 1608 1609 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1610 if (ret) 1611 goto exit; 1612 1613 if (zstd_init(&(session->zstd_data), 0) < 0) 1614 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1615 1616 if (report.queue_size) { 1617 ordered_events__set_alloc_size(&session->ordered_events, 1618 report.queue_size); 1619 } 1620 1621 session->itrace_synth_opts = &itrace_synth_opts; 1622 1623 report.session = session; 1624 1625 has_br_stack = perf_header__has_feat(&session->header, 1626 HEADER_BRANCH_STACK); 1627 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1628 has_br_stack = false; 1629 1630 setup_forced_leader(&report, session->evlist); 1631 1632 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) { 1633 parse_options_usage(NULL, options, "group-sort-idx", 0); 1634 ret = -EINVAL; 1635 goto error; 1636 } 1637 1638 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1639 has_br_stack = true; 1640 1641 if (has_br_stack && branch_call_mode) 1642 symbol_conf.show_branchflag_count = true; 1643 1644 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1645 1646 /* 1647 * Branch mode is a tristate: 1648 * -1 means default, so decide based on the file having branch data. 1649 * 0/1 means the user chose a mode. 1650 */ 1651 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1652 !branch_call_mode) { 1653 sort__mode = SORT_MODE__BRANCH; 1654 symbol_conf.cumulate_callchain = false; 1655 } 1656 if (branch_call_mode) { 1657 callchain_param.key = CCKEY_ADDRESS; 1658 callchain_param.branch_callstack = true; 1659 symbol_conf.use_callchain = true; 1660 callchain_register_param(&callchain_param); 1661 if (sort_order == NULL) 1662 sort_order = CALLCHAIN_BRANCH_SORT_ORDER; 1663 } 1664 1665 if (report.mem_mode) { 1666 if (sort__mode == SORT_MODE__BRANCH) { 1667 pr_err("branch and mem mode incompatible\n"); 1668 goto error; 1669 } 1670 sort__mode = SORT_MODE__MEMORY; 1671 symbol_conf.cumulate_callchain = false; 1672 } 1673 1674 if (symbol_conf.report_hierarchy) { 1675 /* disable incompatible options */ 1676 if (field_order) { 1677 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1678 parse_options_usage(report_usage, options, "F", 1); 1679 parse_options_usage(NULL, options, "hierarchy", 0); 1680 goto error; 1681 } 1682 1683 perf_hpp_list.need_collapse = true; 1684 } 1685 1686 if (report.use_stdio) 1687 use_browser = 0; 1688 #ifdef HAVE_SLANG_SUPPORT 1689 else if (report.use_tui) 1690 use_browser = 1; 1691 #endif 1692 #ifdef HAVE_GTK2_SUPPORT 1693 else if (report.use_gtk) 1694 use_browser = 2; 1695 #endif 1696 1697 /* Force tty output for header output and per-thread stat. */ 1698 if (report.header || report.header_only || report.show_threads) 1699 use_browser = 0; 1700 if (report.header || report.header_only) 1701 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1702 if (report.show_full_info) 1703 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1704 if (report.stats_mode || report.tasks_mode) 1705 use_browser = 0; 1706 if (report.stats_mode && report.tasks_mode) { 1707 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1708 goto error; 1709 } 1710 1711 if (report.total_cycles_mode) { 1712 if (sort__mode != SORT_MODE__BRANCH) 1713 report.total_cycles_mode = false; 1714 else 1715 sort_order = NULL; 1716 } 1717 1718 if (sort_order && strstr(sort_order, "type")) { 1719 report.data_type = true; 1720 annotate_opts.annotate_src = false; 1721 1722 /* disable incompatible options */ 1723 symbol_conf.cumulate_callchain = false; 1724 1725 #ifndef HAVE_LIBDW_SUPPORT 1726 pr_err("Error: Data type profiling is disabled due to missing DWARF support\n"); 1727 goto error; 1728 #endif 1729 } 1730 1731 if (strcmp(input_name, "-") != 0) 1732 setup_browser(true); 1733 else 1734 use_browser = 0; 1735 1736 if (report.data_type && use_browser == 1) { 1737 symbol_conf.annotate_data_member = true; 1738 symbol_conf.annotate_data_sample = true; 1739 } 1740 1741 symbol_conf.enable_latency = true; 1742 if (report.disable_order || !perf_session__has_switch_events(session)) { 1743 if (symbol_conf.parallelism_list_str || 1744 symbol_conf.prefer_latency || 1745 (sort_order && (strstr(sort_order, "latency") || 1746 strstr(sort_order, "parallelism"))) || 1747 (field_order && (strstr(field_order, "latency") || 1748 strstr(field_order, "parallelism")))) { 1749 if (report.disable_order) 1750 ui__error("Use of latency profile or parallelism is incompatible with --disable-order.\n"); 1751 else 1752 ui__error("Use of latency profile or parallelism requires --latency flag during record.\n"); 1753 return -1; 1754 } 1755 /* 1756 * If user did not ask for anything related to 1757 * latency/parallelism explicitly, just don't show it. 1758 */ 1759 symbol_conf.enable_latency = false; 1760 } 1761 1762 if (last_key != K_SWITCH_INPUT_DATA) { 1763 if (sort_order && strstr(sort_order, "ipc")) { 1764 parse_options_usage(report_usage, options, "s", 1); 1765 goto error; 1766 } 1767 1768 if (sort_order && strstr(sort_order, "symbol")) { 1769 if (sort__mode == SORT_MODE__BRANCH) { 1770 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1771 sort_order, "ipc_lbr"); 1772 report.symbol_ipc = true; 1773 } else { 1774 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1775 sort_order, "ipc_null"); 1776 } 1777 1778 sort_order = sort_tmp; 1779 } 1780 } 1781 1782 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1783 (setup_sorting(session->evlist) < 0)) { 1784 if (sort_order) 1785 parse_options_usage(report_usage, options, "s", 1); 1786 if (field_order) 1787 parse_options_usage(sort_order ? NULL : report_usage, 1788 options, "F", 1); 1789 goto error; 1790 } 1791 1792 if ((report.header || report.header_only) && !quiet) { 1793 perf_session__fprintf_info(session, stdout, 1794 report.show_full_info); 1795 if (report.header_only) { 1796 if (data.is_pipe) { 1797 /* 1798 * we need to process first few records 1799 * which contains PERF_RECORD_HEADER_FEATURE. 1800 */ 1801 perf_session__process_events(session); 1802 } 1803 ret = 0; 1804 goto error; 1805 } 1806 } else if (use_browser == 0 && !quiet && 1807 !report.stats_mode && !report.tasks_mode) { 1808 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1809 stdout); 1810 } 1811 1812 /* 1813 * Only in the TUI browser we are doing integrated annotation, 1814 * so don't allocate extra space that won't be used in the stdio 1815 * implementation. 1816 */ 1817 if (ui__has_annotation() || report.symbol_ipc || report.data_type || 1818 report.total_cycles_mode) { 1819 ret = symbol__annotation_init(); 1820 if (ret < 0) 1821 goto error; 1822 /* 1823 * For searching by name on the "Browse map details". 1824 * providing it only in verbose mode not to bloat too 1825 * much struct symbol. 1826 */ 1827 if (verbose > 0) { 1828 /* 1829 * XXX: Need to provide a less kludgy way to ask for 1830 * more space per symbol, the u32 is for the index on 1831 * the ui browser. 1832 * See symbol__browser_index. 1833 */ 1834 symbol_conf.priv_size += sizeof(u32); 1835 } 1836 annotation_config__init(); 1837 } 1838 1839 if (symbol__init(&session->header.env) < 0) 1840 goto error; 1841 1842 if (report.time_str) { 1843 ret = perf_time__parse_for_ranges(report.time_str, session, 1844 &report.ptime_range, 1845 &report.range_size, 1846 &report.range_num); 1847 if (ret < 0) 1848 goto error; 1849 1850 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1851 report.ptime_range, 1852 report.range_num); 1853 } 1854 1855 #ifdef HAVE_LIBTRACEEVENT 1856 if (session->tevent.pevent && 1857 tep_set_function_resolver(session->tevent.pevent, 1858 machine__resolve_kernel_addr, 1859 &session->machines.host) < 0) { 1860 pr_err("%s: failed to set libtraceevent function resolver\n", 1861 __func__); 1862 return -1; 1863 } 1864 #endif 1865 sort__setup_elide(stdout); 1866 1867 ret = __cmd_report(&report); 1868 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1869 perf_session__delete(session); 1870 last_key = K_SWITCH_INPUT_DATA; 1871 /* 1872 * To support switching between data with and without callchains. 1873 * report__setup_sample_type() will update it properly. 1874 */ 1875 symbol_conf.use_callchain = false; 1876 goto repeat; 1877 } else 1878 ret = 0; 1879 1880 if (!use_browser && (verbose > 2 || debug_kmaps)) 1881 perf_session__dump_kmaps(session); 1882 error: 1883 if (report.ptime_range) { 1884 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1885 zfree(&report.ptime_range); 1886 } 1887 1888 if (report.block_reports) { 1889 block_info__free_report(report.block_reports, 1890 report.nr_block_reports); 1891 report.block_reports = NULL; 1892 } 1893 1894 zstd_fini(&(session->zstd_data)); 1895 perf_session__delete(session); 1896 exit: 1897 annotation_options__exit(); 1898 free(sort_order_help); 1899 free(field_order_help); 1900 return ret; 1901 } 1902