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(session->evlist); 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(perf_session__env(rep->session))); 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 if (sort_order || !field_order) { 533 ret += fprintf(fp, "\n# Sort order : %s", 534 sort_order ? : default_mem_sort_order); 535 } 536 } else 537 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 538 539 if (socked_id > -1) 540 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 541 542 return ret + fprintf(fp, "\n#\n"); 543 } 544 545 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 546 { 547 struct evsel *pos; 548 int i = 0, ret; 549 550 evlist__for_each_entry(evlist, pos) { 551 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 552 rep->min_percent, pos, 553 perf_session__env(rep->session)); 554 if (ret != 0) 555 return ret; 556 } 557 558 return 0; 559 } 560 561 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 562 { 563 struct evsel *pos; 564 int i = 0; 565 566 if (!quiet) { 567 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 568 evlist->stats.total_lost_samples); 569 } 570 571 evlist__for_each_entry(evlist, pos) { 572 struct hists *hists = evsel__hists(pos); 573 const char *evname = evsel__name(pos); 574 575 i++; 576 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 577 continue; 578 579 if (rep->skip_empty && !hists->stats.nr_samples) 580 continue; 581 582 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 583 584 if (rep->total_cycles_mode) { 585 char *buf; 586 587 if (!annotation_br_cntr_abbr_list(&buf, pos, true)) { 588 fprintf(stdout, "%s", buf); 589 fprintf(stdout, "#\n"); 590 free(buf); 591 } 592 report__browse_block_hists(&rep->block_reports[i - 1].hist, 593 rep->min_percent, pos, NULL); 594 continue; 595 } 596 597 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 598 !(symbol_conf.use_callchain || 599 symbol_conf.show_branchflag_count)); 600 fprintf(stdout, "\n\n"); 601 } 602 603 if (!quiet) 604 fprintf(stdout, "#\n# (%s)\n#\n", help); 605 606 if (rep->show_threads) { 607 bool style = !strcmp(rep->pretty_printing_style, "raw"); 608 perf_read_values_display(stdout, &rep->show_threads_values, 609 style); 610 perf_read_values_destroy(&rep->show_threads_values); 611 } 612 613 if (sort__mode == SORT_MODE__BRANCH) 614 branch_type_stat_display(stdout, &rep->brtype_stat); 615 616 return 0; 617 } 618 619 static void report__warn_kptr_restrict(const struct report *rep) 620 { 621 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 622 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 623 624 if (evlist__exclude_kernel(rep->session->evlist)) 625 return; 626 627 if (kernel_map == NULL || 628 (dso__hit(map__dso(kernel_map)) && 629 (kernel_kmap->ref_reloc_sym == NULL || 630 kernel_kmap->ref_reloc_sym->addr == 0))) { 631 const char *desc = 632 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 633 "can't be resolved."; 634 635 if (kernel_map && map__has_symbols(kernel_map)) { 636 desc = "If some relocation was applied (e.g. " 637 "kexec) symbols may be misresolved."; 638 } 639 640 ui__warning( 641 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 642 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 643 "Samples in kernel modules can't be resolved as well.\n\n", 644 desc); 645 } 646 } 647 648 static int report__gtk_browse_hists(struct report *rep, const char *help) 649 { 650 int (*hist_browser)(struct evlist *evlist, const char *help, 651 struct hist_browser_timer *timer, float min_pcnt); 652 653 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 654 655 if (hist_browser == NULL) { 656 ui__error("GTK browser not found!\n"); 657 return -1; 658 } 659 660 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 661 } 662 663 static int report__browse_hists(struct report *rep) 664 { 665 int ret; 666 struct perf_session *session = rep->session; 667 struct evlist *evlist = session->evlist; 668 char *help = NULL, *path = NULL; 669 670 path = system_path(TIPDIR); 671 if (perf_tip(&help, path) || help == NULL) { 672 /* fallback for people who don't install perf ;-) */ 673 free(path); 674 path = system_path(DOCDIR); 675 if (perf_tip(&help, path) || help == NULL) 676 help = strdup("Cannot load tips.txt file, please install perf!"); 677 } 678 free(path); 679 680 switch (use_browser) { 681 case 1: 682 if (rep->total_cycles_mode) { 683 ret = evlist__tui_block_hists_browse(evlist, rep); 684 break; 685 } 686 687 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 688 perf_session__env(session), true); 689 /* 690 * Usually "ret" is the last pressed key, and we only 691 * care if the key notifies us to switch data file. 692 */ 693 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 694 ret = 0; 695 break; 696 case 2: 697 ret = report__gtk_browse_hists(rep, help); 698 break; 699 default: 700 ret = evlist__tty_browse_hists(evlist, rep, help); 701 break; 702 } 703 free(help); 704 return ret; 705 } 706 707 static int report__collapse_hists(struct report *rep) 708 { 709 struct perf_session *session = rep->session; 710 struct evlist *evlist = session->evlist; 711 struct ui_progress prog; 712 struct evsel *pos; 713 int ret = 0; 714 715 /* 716 * The pipe data needs to setup hierarchy hpp formats now, because it 717 * cannot know about evsels in the data before reading the data. The 718 * normal file data saves the event (attribute) info in the header 719 * section, but pipe does not have the luxury. 720 */ 721 if (perf_data__is_pipe(session->data)) { 722 if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) { 723 ui__error("Failed to setup hierarchy output formats\n"); 724 return -1; 725 } 726 } 727 728 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 729 730 evlist__for_each_entry(rep->session->evlist, pos) { 731 struct hists *hists = evsel__hists(pos); 732 733 if (pos->core.idx == 0) 734 hists->symbol_filter_str = rep->symbol_filter_str; 735 736 hists->socket_filter = rep->socket_filter; 737 738 ret = hists__collapse_resort(hists, &prog); 739 if (ret < 0) 740 break; 741 742 /* Non-group events are considered as leader */ 743 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 744 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 745 746 hists__match(leader_hists, hists); 747 hists__link(leader_hists, hists); 748 } 749 } 750 751 ui_progress__finish(); 752 return ret; 753 } 754 755 static int hists__resort_cb(struct hist_entry *he, void *arg) 756 { 757 struct report *rep = arg; 758 struct symbol *sym = he->ms.sym; 759 760 if (rep->symbol_ipc && sym && !sym->annotate2) { 761 struct evsel *evsel = hists_to_evsel(he->hists); 762 763 symbol__annotate2(&he->ms, evsel, NULL); 764 } 765 766 return 0; 767 } 768 769 static void report__output_resort(struct report *rep) 770 { 771 struct ui_progress prog; 772 struct evsel *pos; 773 774 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 775 776 evlist__for_each_entry(rep->session->evlist, pos) { 777 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 778 } 779 780 ui_progress__finish(); 781 } 782 783 static int count_sample_event(const struct perf_tool *tool __maybe_unused, 784 union perf_event *event __maybe_unused, 785 struct perf_sample *sample __maybe_unused, 786 struct evsel *evsel, 787 struct machine *machine __maybe_unused) 788 { 789 struct hists *hists = evsel__hists(evsel); 790 791 hists__inc_nr_events(hists); 792 return 0; 793 } 794 795 static int count_lost_samples_event(const struct perf_tool *tool, 796 union perf_event *event, 797 struct perf_sample *sample, 798 struct machine *machine __maybe_unused) 799 { 800 struct report *rep = container_of(tool, struct report, tool); 801 struct evsel *evsel; 802 803 evsel = evlist__id2evsel(rep->session->evlist, sample->id); 804 if (evsel) { 805 struct hists *hists = evsel__hists(evsel); 806 u32 count = event->lost_samples.lost; 807 808 if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF) 809 hists__inc_nr_dropped_samples(hists, count); 810 else 811 hists__inc_nr_lost_samples(hists, count); 812 } 813 return 0; 814 } 815 816 static int process_attr(const struct perf_tool *tool __maybe_unused, 817 union perf_event *event, 818 struct evlist **pevlist); 819 820 static void stats_setup(struct report *rep) 821 { 822 perf_tool__init(&rep->tool, /*ordered_events=*/false); 823 rep->tool.attr = process_attr; 824 rep->tool.sample = count_sample_event; 825 rep->tool.lost_samples = count_lost_samples_event; 826 rep->tool.event_update = perf_event__process_event_update; 827 rep->tool.no_warn = true; 828 } 829 830 static int stats_print(struct report *rep) 831 { 832 struct perf_session *session = rep->session; 833 834 perf_session__fprintf_nr_events(session, stdout); 835 evlist__fprintf_nr_events(session->evlist, stdout); 836 return 0; 837 } 838 839 static void tasks_setup(struct report *rep) 840 { 841 perf_tool__init(&rep->tool, /*ordered_events=*/true); 842 if (rep->mmaps_mode) { 843 rep->tool.mmap = perf_event__process_mmap; 844 rep->tool.mmap2 = perf_event__process_mmap2; 845 } 846 rep->tool.attr = process_attr; 847 rep->tool.comm = perf_event__process_comm; 848 rep->tool.exit = perf_event__process_exit; 849 rep->tool.fork = perf_event__process_fork; 850 rep->tool.no_warn = true; 851 } 852 853 struct maps__fprintf_task_args { 854 int indent; 855 FILE *fp; 856 size_t printed; 857 }; 858 859 static int maps__fprintf_task_cb(struct map *map, void *data) 860 { 861 struct maps__fprintf_task_args *args = data; 862 const struct dso *dso = map__dso(map); 863 u32 prot = map__prot(map); 864 const struct dso_id *dso_id = dso__id_const(dso); 865 int ret; 866 char buf[SBUILD_ID_SIZE]; 867 868 if (dso_id->mmap2_valid) 869 snprintf(buf, sizeof(buf), "%" PRIu64, dso_id->ino); 870 else 871 build_id__snprintf(&dso_id->build_id, buf, sizeof(buf)); 872 873 ret = fprintf(args->fp, 874 "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %s %s\n", 875 args->indent, "", map__start(map), map__end(map), 876 prot & PROT_READ ? 'r' : '-', 877 prot & PROT_WRITE ? 'w' : '-', 878 prot & PROT_EXEC ? 'x' : '-', 879 map__flags(map) ? 's' : 'p', 880 map__pgoff(map), 881 buf, dso__name(dso)); 882 883 if (ret < 0) 884 return ret; 885 886 args->printed += ret; 887 return 0; 888 } 889 890 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 891 { 892 struct maps__fprintf_task_args args = { 893 .indent = indent, 894 .fp = fp, 895 .printed = 0, 896 }; 897 898 maps__for_each_map(maps, maps__fprintf_task_cb, &args); 899 900 return args.printed; 901 } 902 903 static int thread_level(struct machine *machine, const struct thread *thread) 904 { 905 struct thread *parent_thread; 906 int res; 907 908 if (thread__tid(thread) <= 0) 909 return 0; 910 911 if (thread__ppid(thread) <= 0) 912 return 1; 913 914 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread)); 915 if (!parent_thread) { 916 pr_err("Missing parent thread of %d\n", thread__tid(thread)); 917 return 0; 918 } 919 res = 1 + thread_level(machine, parent_thread); 920 thread__put(parent_thread); 921 return res; 922 } 923 924 static void task__print_level(struct machine *machine, struct thread *thread, FILE *fp) 925 { 926 int level = thread_level(machine, thread); 927 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 928 thread__pid(thread), thread__tid(thread), 929 thread__ppid(thread), level, ""); 930 931 fprintf(fp, "%s\n", thread__comm_str(thread)); 932 933 maps__fprintf_task(thread__maps(thread), comm_indent, fp); 934 } 935 936 /* 937 * Sort two thread list nodes such that they form a tree. The first node is the 938 * root of the tree, its children are ordered numerically after it. If a child 939 * has children itself then they appear immediately after their parent. For 940 * example, the 4 threads in the order they'd appear in the list: 941 * - init with a TID 1 and a parent of 0 942 * - systemd with a TID 3000 and a parent of init/1 943 * - systemd child thread with TID 4000, the parent is 3000 944 * - NetworkManager is a child of init with a TID of 3500. 945 */ 946 static int task_list_cmp(void *priv, const struct list_head *la, const struct list_head *lb) 947 { 948 struct machine *machine = priv; 949 struct thread_list *task_a = list_entry(la, struct thread_list, list); 950 struct thread_list *task_b = list_entry(lb, struct thread_list, list); 951 struct thread *a = task_a->thread; 952 struct thread *b = task_b->thread; 953 int level_a, level_b, res; 954 955 /* Same thread? */ 956 if (thread__tid(a) == thread__tid(b)) 957 return 0; 958 959 /* Compare a and b to root. */ 960 if (thread__tid(a) == 0) 961 return -1; 962 963 if (thread__tid(b) == 0) 964 return 1; 965 966 /* If parents match sort by tid. */ 967 if (thread__ppid(a) == thread__ppid(b)) 968 return thread__tid(a) < thread__tid(b) ? -1 : 1; 969 970 /* 971 * Find a and b such that if they are a child of each other a and b's 972 * tid's match, otherwise a and b have a common parent and distinct 973 * tid's to sort by. First make the depths of the threads match. 974 */ 975 level_a = thread_level(machine, a); 976 level_b = thread_level(machine, b); 977 a = thread__get(a); 978 b = thread__get(b); 979 for (int i = level_a; i > level_b; i--) { 980 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(a)); 981 982 thread__put(a); 983 if (!parent) { 984 pr_err("Missing parent thread of %d\n", thread__tid(a)); 985 thread__put(b); 986 return -1; 987 } 988 a = parent; 989 } 990 for (int i = level_b; i > level_a; i--) { 991 struct thread *parent = machine__find_thread(machine, -1, thread__ppid(b)); 992 993 thread__put(b); 994 if (!parent) { 995 pr_err("Missing parent thread of %d\n", thread__tid(b)); 996 thread__put(a); 997 return 1; 998 } 999 b = parent; 1000 } 1001 /* Search up to a common parent. */ 1002 while (thread__ppid(a) != thread__ppid(b)) { 1003 struct thread *parent; 1004 1005 parent = machine__find_thread(machine, -1, thread__ppid(a)); 1006 thread__put(a); 1007 if (!parent) 1008 pr_err("Missing parent thread of %d\n", thread__tid(a)); 1009 a = parent; 1010 parent = machine__find_thread(machine, -1, thread__ppid(b)); 1011 thread__put(b); 1012 if (!parent) 1013 pr_err("Missing parent thread of %d\n", thread__tid(b)); 1014 b = parent; 1015 if (!a || !b) { 1016 /* Handle missing parent (unexpected) with some sanity. */ 1017 thread__put(a); 1018 thread__put(b); 1019 return !a && !b ? 0 : (!a ? -1 : 1); 1020 } 1021 } 1022 if (thread__tid(a) == thread__tid(b)) { 1023 /* a is a child of b or vice-versa, deeper levels appear later. */ 1024 res = level_a < level_b ? -1 : (level_a > level_b ? 1 : 0); 1025 } else { 1026 /* Sort by tid now the parent is the same. */ 1027 res = thread__tid(a) < thread__tid(b) ? -1 : 1; 1028 } 1029 thread__put(a); 1030 thread__put(b); 1031 return res; 1032 } 1033 1034 static int tasks_print(struct report *rep, FILE *fp) 1035 { 1036 struct machine *machine = &rep->session->machines.host; 1037 LIST_HEAD(tasks); 1038 int ret; 1039 1040 ret = machine__thread_list(machine, &tasks); 1041 if (!ret) { 1042 struct thread_list *task; 1043 1044 list_sort(machine, &tasks, task_list_cmp); 1045 1046 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 1047 1048 list_for_each_entry(task, &tasks, list) 1049 task__print_level(machine, task->thread, fp); 1050 } 1051 thread_list__delete(&tasks); 1052 return ret; 1053 } 1054 1055 static int __cmd_report(struct report *rep) 1056 { 1057 int ret; 1058 struct perf_session *session = rep->session; 1059 struct evsel *pos; 1060 struct perf_data *data = session->data; 1061 1062 signal(SIGINT, sig_handler); 1063 1064 if (rep->cpu_list) { 1065 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 1066 rep->cpu_bitmap); 1067 if (ret) { 1068 ui__error("failed to set cpu bitmap\n"); 1069 return ret; 1070 } 1071 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 1072 } 1073 1074 if (rep->show_threads) { 1075 ret = perf_read_values_init(&rep->show_threads_values); 1076 if (ret) 1077 return ret; 1078 } 1079 1080 ret = report__setup_sample_type(rep); 1081 if (ret) { 1082 /* report__setup_sample_type() already showed error message */ 1083 return ret; 1084 } 1085 1086 if (rep->stats_mode) 1087 stats_setup(rep); 1088 1089 if (rep->tasks_mode) 1090 tasks_setup(rep); 1091 1092 ret = perf_session__process_events(session); 1093 if (ret) { 1094 ui__error("failed to process sample\n"); 1095 return ret; 1096 } 1097 1098 /* Don't show Latency column for non-parallel profiles by default. */ 1099 if (!symbol_conf.prefer_latency && rep->total_samples && 1100 rep->singlethreaded_samples * 100 / rep->total_samples >= 99) 1101 perf_hpp__cancel_latency(session->evlist); 1102 1103 evlist__check_mem_load_aux(session->evlist); 1104 1105 if (rep->stats_mode) 1106 return stats_print(rep); 1107 1108 if (rep->tasks_mode) 1109 return tasks_print(rep, stdout); 1110 1111 report__warn_kptr_restrict(rep); 1112 1113 evlist__for_each_entry(session->evlist, pos) 1114 rep->nr_entries += evsel__hists(pos)->nr_entries; 1115 1116 if (use_browser == 0) { 1117 if (verbose > 3) 1118 perf_session__fprintf(session, stdout); 1119 1120 if (verbose > 2) 1121 perf_session__fprintf_dsos(session, stdout); 1122 1123 if (dump_trace) { 1124 stats_print(rep); 1125 return 0; 1126 } 1127 } 1128 1129 ret = report__collapse_hists(rep); 1130 if (ret) { 1131 ui__error("failed to process hist entry\n"); 1132 return ret; 1133 } 1134 1135 if (session_done()) 1136 return 0; 1137 1138 /* 1139 * recalculate number of entries after collapsing since it 1140 * might be changed during the collapse phase. 1141 */ 1142 rep->nr_entries = 0; 1143 evlist__for_each_entry(session->evlist, pos) 1144 rep->nr_entries += evsel__hists(pos)->nr_entries; 1145 1146 if (rep->nr_entries == 0) { 1147 ui__error("The %s data has no samples!\n", data->path); 1148 return 0; 1149 } 1150 1151 report__output_resort(rep); 1152 1153 if (rep->total_cycles_mode) { 1154 int nr_hpps = 4; 1155 int block_hpps[PERF_HPP_REPORT__BLOCK_MAX_INDEX] = { 1156 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1157 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1158 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1159 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1160 }; 1161 1162 if (session->evlist->nr_br_cntr > 0) 1163 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_BRANCH_COUNTER; 1164 1165 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_RANGE; 1166 block_hpps[nr_hpps++] = PERF_HPP_REPORT__BLOCK_DSO; 1167 1168 rep->block_reports = block_info__create_report(session->evlist, 1169 rep->total_cycles, 1170 block_hpps, nr_hpps, 1171 &rep->nr_block_reports); 1172 if (!rep->block_reports) 1173 return -1; 1174 } 1175 1176 return report__browse_hists(rep); 1177 } 1178 1179 static int 1180 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1181 { 1182 struct callchain_param *callchain = opt->value; 1183 1184 callchain->enabled = !unset; 1185 /* 1186 * --no-call-graph 1187 */ 1188 if (unset) { 1189 symbol_conf.use_callchain = false; 1190 callchain->mode = CHAIN_NONE; 1191 return 0; 1192 } 1193 1194 return parse_callchain_report_opt(arg); 1195 } 1196 1197 static int 1198 parse_time_quantum(const struct option *opt, const char *arg, 1199 int unset __maybe_unused) 1200 { 1201 unsigned long *time_q = opt->value; 1202 char *end; 1203 1204 *time_q = strtoul(arg, &end, 0); 1205 if (end == arg) 1206 goto parse_err; 1207 if (*time_q == 0) { 1208 pr_err("time quantum cannot be 0"); 1209 return -1; 1210 } 1211 end = skip_spaces(end); 1212 if (*end == 0) 1213 return 0; 1214 if (!strcmp(end, "s")) { 1215 *time_q *= NSEC_PER_SEC; 1216 return 0; 1217 } 1218 if (!strcmp(end, "ms")) { 1219 *time_q *= NSEC_PER_MSEC; 1220 return 0; 1221 } 1222 if (!strcmp(end, "us")) { 1223 *time_q *= NSEC_PER_USEC; 1224 return 0; 1225 } 1226 if (!strcmp(end, "ns")) 1227 return 0; 1228 parse_err: 1229 pr_err("Cannot parse time quantum `%s'\n", arg); 1230 return -1; 1231 } 1232 1233 int 1234 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1235 const char *arg, int unset __maybe_unused) 1236 { 1237 if (arg) { 1238 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1239 if (err) { 1240 char buf[BUFSIZ]; 1241 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1242 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1243 return -1; 1244 } 1245 have_ignore_callees = 1; 1246 } 1247 1248 return 0; 1249 } 1250 1251 static int 1252 parse_branch_mode(const struct option *opt, 1253 const char *str __maybe_unused, int unset) 1254 { 1255 int *branch_mode = opt->value; 1256 1257 *branch_mode = !unset; 1258 return 0; 1259 } 1260 1261 static int 1262 parse_percent_limit(const struct option *opt, const char *str, 1263 int unset __maybe_unused) 1264 { 1265 struct report *rep = opt->value; 1266 double pcnt = strtof(str, NULL); 1267 1268 rep->min_percent = pcnt; 1269 callchain_param.min_percent = pcnt; 1270 return 0; 1271 } 1272 1273 static int process_attr(const struct perf_tool *tool __maybe_unused, 1274 union perf_event *event, 1275 struct evlist **pevlist) 1276 { 1277 struct perf_session *session; 1278 struct perf_env *env; 1279 u64 sample_type; 1280 int err; 1281 1282 err = perf_event__process_attr(tool, event, pevlist); 1283 if (err) 1284 return err; 1285 1286 /* 1287 * Check if we need to enable callchains based 1288 * on events sample_type. 1289 */ 1290 sample_type = evlist__combined_sample_type(*pevlist); 1291 session = (*pevlist)->session; 1292 env = perf_session__env(session); 1293 callchain_param_setup(sample_type, perf_env__arch(env)); 1294 return 0; 1295 } 1296 1297 #define CALLCHAIN_BRANCH_SORT_ORDER \ 1298 "srcline,symbol,dso,callchain_branch_predicted," \ 1299 "callchain_branch_abort,callchain_branch_cycles" 1300 1301 int cmd_report(int argc, const char **argv) 1302 { 1303 struct perf_session *session; 1304 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1305 struct stat st; 1306 bool has_br_stack = false; 1307 int branch_mode = -1; 1308 int last_key = 0; 1309 bool branch_call_mode = false; 1310 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1311 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1312 CALLCHAIN_REPORT_HELP 1313 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1314 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1315 const char * const report_usage[] = { 1316 "perf report [<options>]", 1317 NULL 1318 }; 1319 struct report report = { 1320 .max_stack = PERF_MAX_STACK_DEPTH, 1321 .pretty_printing_style = "normal", 1322 .socket_filter = -1, 1323 .skip_empty = true, 1324 }; 1325 char *sort_order_help = sort_help("sort by key(s):", SORT_MODE__NORMAL); 1326 char *field_order_help = sort_help("output field(s):", SORT_MODE__NORMAL); 1327 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL; 1328 const struct option options[] = { 1329 OPT_STRING('i', "input", &input_name, "file", 1330 "input file name"), 1331 OPT_INCR('v', "verbose", &verbose, 1332 "be more verbose (show symbol address, etc)"), 1333 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"), 1334 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1335 "dump raw trace in ASCII"), 1336 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1337 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1338 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1339 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1340 "file", "vmlinux pathname"), 1341 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1342 "don't load vmlinux even if found"), 1343 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1344 "file", "kallsyms pathname"), 1345 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1346 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1347 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1348 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1349 "Show a column with the number of samples"), 1350 OPT_BOOLEAN('T', "threads", &report.show_threads, 1351 "Show per-thread event counters"), 1352 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1353 "pretty printing style key: normal raw"), 1354 #ifdef HAVE_SLANG_SUPPORT 1355 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1356 #endif 1357 #ifdef HAVE_GTK2_SUPPORT 1358 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1359 #endif 1360 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1361 "Use the stdio interface"), 1362 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1363 OPT_BOOLEAN(0, "header-only", &report.header_only, 1364 "Show only data header."), 1365 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1366 sort_order_help), 1367 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1368 field_order_help), 1369 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1370 "Show sample percentage for different cpu modes"), 1371 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1372 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1373 OPT_STRING('p', "parent", &parent_pattern, "regex", 1374 "regex filter to identify parent, see: '--sort parent'"), 1375 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1376 "Only display entries with parent-match"), 1377 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1378 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1379 report_callchain_help, &report_parse_callchain_opt, 1380 callchain_default_opt), 1381 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1382 "Accumulate callchains of children and show total overhead as well. " 1383 "Enabled by default, use --no-children to disable."), 1384 OPT_INTEGER(0, "max-stack", &report.max_stack, 1385 "Set the maximum stack depth when parsing the callchain, " 1386 "anything beyond the specified depth will be ignored. " 1387 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1388 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1389 "alias for inverted call graph"), 1390 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1391 "ignore callees of these functions in call graphs", 1392 report_parse_ignore_callees_opt), 1393 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1394 "only consider symbols in these dsos"), 1395 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1396 "only consider symbols in these comms"), 1397 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1398 "only consider symbols in these pids"), 1399 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1400 "only consider symbols in these tids"), 1401 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1402 "only consider these symbols"), 1403 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1404 "only show symbols that (partially) match with this filter"), 1405 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1406 "width[,width...]", 1407 "don't try to adjust column width, use these fixed values"), 1408 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1409 "separator for columns, no spaces will be added between " 1410 "columns '.' is reserved."), 1411 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1412 "Only display entries resolved to a symbol"), 1413 OPT_CALLBACK(0, "symfs", NULL, "directory", 1414 "Look for files with symbols relative to this directory", 1415 symbol__config_symfs), 1416 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1417 "list of cpus to profile"), 1418 OPT_STRING(0, "parallelism", &symbol_conf.parallelism_list_str, "parallelism", 1419 "only consider these parallelism levels (cpu set format)"), 1420 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1421 "Display extended information about perf.data file"), 1422 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src, 1423 "Interleave source code with assembly code (default)"), 1424 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw, 1425 "Display raw encoding of assembly instructions (default)"), 1426 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1427 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1428 OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix", 1429 "Add prefix to source file path names in programs (with --prefix-strip)"), 1430 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N", 1431 "Strip first N entries of source file path name in programs (with --prefix)"), 1432 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1433 "Show a column with the sum of periods"), 1434 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1435 "Show event group information together"), 1436 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1437 "Sort the output by the event at the index n in group. " 1438 "If n is invalid, sort by the first event. " 1439 "WARNING: should be used on grouped events."), 1440 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1441 "use branch records for per branch histogram filling", 1442 parse_branch_mode), 1443 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1444 "add last branch records to call history"), 1445 OPT_STRING(0, "objdump", &objdump_path, "path", 1446 "objdump binary to use for disassembly and annotations"), 1447 OPT_STRING(0, "addr2line", &addr2line_path, "path", 1448 "addr2line binary to use for line numbers"), 1449 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1450 "Symbol demangling. Enabled by default, use --no-demangle to disable."), 1451 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1452 "Enable kernel symbol demangling"), 1453 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1454 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1455 "Number of samples to save per histogram entry for individual browsing"), 1456 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1457 "Don't show entries under that percent", parse_percent_limit), 1458 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1459 "how to display percentage of filtered entries", parse_filter_percentage), 1460 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1461 "Instruction Tracing options\n" ITRACE_HELP, 1462 itrace_parse_synth_opts), 1463 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1464 "Show full source file name path for source lines"), 1465 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1466 "Show callgraph from reference event"), 1467 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1468 "Enable LBR callgraph stitching approach"), 1469 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1470 "only show processor socket that match with this filter"), 1471 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1472 "Show raw trace event output (do not use print fmt or plugins)"), 1473 OPT_BOOLEAN('H', "hierarchy", &symbol_conf.report_hierarchy, 1474 "Show entries in a hierarchy"), 1475 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1476 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1477 stdio__config_color, "always"), 1478 OPT_STRING(0, "time", &report.time_str, "str", 1479 "Time span of interest (start,stop)"), 1480 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1481 "Show inline function"), 1482 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period", 1483 "Set percent type local/global-period/hits", 1484 annotate_parse_percent_type), 1485 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1486 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1487 "Set time quantum for time sort key (default 100ms)", 1488 parse_time_quantum), 1489 OPTS_EVSWITCH(&report.evswitch), 1490 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1491 "Sort all blocks by 'Sampled Cycles%'"), 1492 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1493 "Disable raw trace ordering"), 1494 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1495 "Do not display empty (or dummy) events in the output"), 1496 OPT_BOOLEAN(0, "latency", &symbol_conf.prefer_latency, 1497 "Show latency-centric profile rather than the default\n" 1498 "\t\t\t CPU-consumption-centric profile\n" 1499 "\t\t\t (requires perf record --latency flag)."), 1500 OPT_END() 1501 }; 1502 struct perf_data data = { 1503 .mode = PERF_DATA_MODE_READ, 1504 }; 1505 int ret = hists__init(); 1506 char sort_tmp[128]; 1507 bool ordered_events = true; 1508 1509 if (ret < 0) 1510 goto exit; 1511 1512 /* 1513 * tasks_mode require access to exited threads to list those that are in 1514 * the data file. Off-cpu events are synthesized after other events and 1515 * reference exited threads. 1516 */ 1517 symbol_conf.keep_exited_threads = true; 1518 1519 annotation_options__init(); 1520 1521 ret = perf_config(report__config, &report); 1522 if (ret) 1523 goto exit; 1524 1525 argc = parse_options(argc, argv, options, report_usage, 0); 1526 if (argc) { 1527 /* 1528 * Special case: if there's an argument left then assume that 1529 * it's a symbol filter: 1530 */ 1531 if (argc > 1) 1532 usage_with_options(report_usage, options); 1533 1534 report.symbol_filter_str = argv[0]; 1535 } 1536 1537 if (disassembler_style) { 1538 annotate_opts.disassembler_style = strdup(disassembler_style); 1539 if (!annotate_opts.disassembler_style) 1540 return -ENOMEM; 1541 } 1542 if (objdump_path) { 1543 annotate_opts.objdump_path = strdup(objdump_path); 1544 if (!annotate_opts.objdump_path) 1545 return -ENOMEM; 1546 } 1547 if (addr2line_path) { 1548 symbol_conf.addr2line_path = strdup(addr2line_path); 1549 if (!symbol_conf.addr2line_path) 1550 return -ENOMEM; 1551 } 1552 1553 if (annotate_check_args() < 0) { 1554 ret = -EINVAL; 1555 goto exit; 1556 } 1557 1558 if (report.mmaps_mode) 1559 report.tasks_mode = true; 1560 1561 if (dump_trace && report.disable_order) 1562 ordered_events = false; 1563 1564 if (quiet) 1565 perf_quiet_option(); 1566 1567 ret = symbol__validate_sym_arguments(); 1568 if (ret) 1569 goto exit; 1570 1571 if (report.inverted_callchain) 1572 callchain_param.order = ORDER_CALLER; 1573 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1574 callchain_param.order = ORDER_CALLER; 1575 1576 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1577 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1578 report.max_stack = itrace_synth_opts.callchain_sz; 1579 1580 if (!input_name || !strlen(input_name)) { 1581 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1582 input_name = "-"; 1583 else 1584 input_name = "perf.data"; 1585 } 1586 1587 repeat: 1588 data.path = input_name; 1589 data.force = symbol_conf.force; 1590 1591 symbol_conf.skip_empty = report.skip_empty; 1592 1593 perf_tool__init(&report.tool, ordered_events); 1594 report.tool.sample = process_sample_event; 1595 report.tool.mmap = perf_event__process_mmap; 1596 report.tool.mmap2 = perf_event__process_mmap2; 1597 report.tool.comm = perf_event__process_comm; 1598 report.tool.namespaces = perf_event__process_namespaces; 1599 report.tool.cgroup = perf_event__process_cgroup; 1600 report.tool.exit = perf_event__process_exit; 1601 report.tool.fork = perf_event__process_fork; 1602 report.tool.context_switch = perf_event__process_switch; 1603 report.tool.lost = perf_event__process_lost; 1604 report.tool.read = process_read_event; 1605 report.tool.attr = process_attr; 1606 #ifdef HAVE_LIBTRACEEVENT 1607 report.tool.tracing_data = perf_event__process_tracing_data; 1608 #endif 1609 report.tool.build_id = perf_event__process_build_id; 1610 report.tool.id_index = perf_event__process_id_index; 1611 report.tool.auxtrace_info = perf_event__process_auxtrace_info; 1612 report.tool.auxtrace = perf_event__process_auxtrace; 1613 report.tool.event_update = perf_event__process_event_update; 1614 report.tool.feature = process_feature_event; 1615 report.tool.ordering_requires_timestamps = true; 1616 1617 session = perf_session__new(&data, &report.tool); 1618 if (IS_ERR(session)) { 1619 ret = PTR_ERR(session); 1620 goto exit; 1621 } 1622 1623 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1624 if (ret) 1625 goto exit; 1626 1627 if (zstd_init(&(session->zstd_data), 0) < 0) 1628 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1629 1630 if (report.queue_size) { 1631 ordered_events__set_alloc_size(&session->ordered_events, 1632 report.queue_size); 1633 } 1634 1635 session->itrace_synth_opts = &itrace_synth_opts; 1636 1637 report.session = session; 1638 1639 has_br_stack = perf_header__has_feat(&session->header, 1640 HEADER_BRANCH_STACK); 1641 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1642 has_br_stack = false; 1643 1644 setup_forced_leader(&report, session->evlist); 1645 1646 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) { 1647 parse_options_usage(NULL, options, "group-sort-idx", 0); 1648 ret = -EINVAL; 1649 goto error; 1650 } 1651 1652 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1653 has_br_stack = true; 1654 1655 if (has_br_stack && branch_call_mode) 1656 symbol_conf.show_branchflag_count = true; 1657 1658 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1659 1660 /* 1661 * Branch mode is a tristate: 1662 * -1 means default, so decide based on the file having branch data. 1663 * 0/1 means the user chose a mode. 1664 */ 1665 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1666 !branch_call_mode) { 1667 sort__mode = SORT_MODE__BRANCH; 1668 symbol_conf.cumulate_callchain = false; 1669 } 1670 if (branch_call_mode) { 1671 callchain_param.key = CCKEY_ADDRESS; 1672 callchain_param.branch_callstack = true; 1673 symbol_conf.use_callchain = true; 1674 callchain_register_param(&callchain_param); 1675 if (sort_order == NULL) 1676 sort_order = CALLCHAIN_BRANCH_SORT_ORDER; 1677 } 1678 1679 if (report.mem_mode) { 1680 if (sort__mode == SORT_MODE__BRANCH) { 1681 pr_err("branch and mem mode incompatible\n"); 1682 goto error; 1683 } 1684 sort__mode = SORT_MODE__MEMORY; 1685 symbol_conf.cumulate_callchain = false; 1686 } 1687 1688 if (symbol_conf.report_hierarchy) { 1689 /* 1690 * The hist entries in hierarchy are added during the collpase 1691 * phase. Let's enable it even if no sort keys require it. 1692 */ 1693 perf_hpp_list.need_collapse = true; 1694 } 1695 1696 if (report.use_stdio) 1697 use_browser = 0; 1698 #ifdef HAVE_SLANG_SUPPORT 1699 else if (report.use_tui) 1700 use_browser = 1; 1701 #endif 1702 #ifdef HAVE_GTK2_SUPPORT 1703 else if (report.use_gtk) 1704 use_browser = 2; 1705 #endif 1706 1707 /* Force tty output for header output and per-thread stat. */ 1708 if (report.header || report.header_only || report.show_threads) 1709 use_browser = 0; 1710 if (report.header || report.header_only) 1711 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1712 if (report.show_full_info) 1713 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1714 if (report.stats_mode || report.tasks_mode) 1715 use_browser = 0; 1716 if (report.stats_mode && report.tasks_mode) { 1717 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1718 goto error; 1719 } 1720 1721 if (report.total_cycles_mode) { 1722 if (sort__mode != SORT_MODE__BRANCH) 1723 report.total_cycles_mode = false; 1724 else 1725 sort_order = NULL; 1726 } 1727 1728 if (sort_order && strstr(sort_order, "type")) { 1729 report.data_type = true; 1730 annotate_opts.annotate_src = false; 1731 1732 /* disable incompatible options */ 1733 symbol_conf.cumulate_callchain = false; 1734 1735 #ifndef HAVE_LIBDW_SUPPORT 1736 pr_err("Error: Data type profiling is disabled due to missing DWARF support\n"); 1737 goto error; 1738 #endif 1739 } 1740 1741 if (strcmp(input_name, "-") != 0) 1742 setup_browser(true); 1743 else 1744 use_browser = 0; 1745 1746 if (report.data_type && use_browser == 1) { 1747 symbol_conf.annotate_data_member = true; 1748 symbol_conf.annotate_data_sample = true; 1749 } 1750 1751 symbol_conf.enable_latency = true; 1752 if (report.disable_order || !perf_session__has_switch_events(session)) { 1753 if (symbol_conf.parallelism_list_str || 1754 symbol_conf.prefer_latency || 1755 (sort_order && (strstr(sort_order, "latency") || 1756 strstr(sort_order, "parallelism"))) || 1757 (field_order && (strstr(field_order, "latency") || 1758 strstr(field_order, "parallelism")))) { 1759 if (report.disable_order) 1760 ui__error("Use of latency profile or parallelism is incompatible with --disable-order.\n"); 1761 else 1762 ui__error("Use of latency profile or parallelism requires --latency flag during record.\n"); 1763 return -1; 1764 } 1765 /* 1766 * If user did not ask for anything related to 1767 * latency/parallelism explicitly, just don't show it. 1768 */ 1769 symbol_conf.enable_latency = false; 1770 } 1771 1772 if (last_key != K_SWITCH_INPUT_DATA) { 1773 if (sort_order && strstr(sort_order, "ipc")) { 1774 parse_options_usage(report_usage, options, "s", 1); 1775 goto error; 1776 } 1777 1778 if (sort_order && strstr(sort_order, "symbol")) { 1779 if (sort__mode == SORT_MODE__BRANCH) { 1780 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1781 sort_order, "ipc_lbr"); 1782 report.symbol_ipc = true; 1783 } else { 1784 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1785 sort_order, "ipc_null"); 1786 } 1787 1788 sort_order = sort_tmp; 1789 } 1790 } 1791 1792 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1793 (setup_sorting(session->evlist, perf_session__env(session)) < 0)) { 1794 if (sort_order) 1795 parse_options_usage(report_usage, options, "s", 1); 1796 if (field_order) 1797 parse_options_usage(sort_order ? NULL : report_usage, 1798 options, "F", 1); 1799 goto error; 1800 } 1801 1802 if ((report.header || report.header_only) && !quiet) { 1803 perf_session__fprintf_info(session, stdout, 1804 report.show_full_info); 1805 if (report.header_only) { 1806 if (data.is_pipe) { 1807 /* 1808 * we need to process first few records 1809 * which contains PERF_RECORD_HEADER_FEATURE. 1810 */ 1811 perf_session__process_events(session); 1812 } 1813 ret = 0; 1814 goto error; 1815 } 1816 } else if (use_browser == 0 && !quiet && 1817 !report.stats_mode && !report.tasks_mode) { 1818 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1819 stdout); 1820 } 1821 1822 /* 1823 * Only in the TUI browser we are doing integrated annotation, 1824 * so don't allocate extra space that won't be used in the stdio 1825 * implementation. 1826 */ 1827 if (ui__has_annotation() || report.symbol_ipc || report.data_type || 1828 report.total_cycles_mode) { 1829 ret = symbol__annotation_init(); 1830 if (ret < 0) 1831 goto error; 1832 /* 1833 * For searching by name on the "Browse map details". 1834 * providing it only in verbose mode not to bloat too 1835 * much struct symbol. 1836 */ 1837 if (verbose > 0) { 1838 /* 1839 * XXX: Need to provide a less kludgy way to ask for 1840 * more space per symbol, the u32 is for the index on 1841 * the ui browser. 1842 * See symbol__browser_index. 1843 */ 1844 symbol_conf.priv_size += sizeof(u32); 1845 } 1846 annotation_config__init(); 1847 } 1848 1849 if (symbol__init(perf_session__env(session)) < 0) 1850 goto error; 1851 1852 if (report.time_str) { 1853 ret = perf_time__parse_for_ranges(report.time_str, session, 1854 &report.ptime_range, 1855 &report.range_size, 1856 &report.range_num); 1857 if (ret < 0) 1858 goto error; 1859 1860 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1861 report.ptime_range, 1862 report.range_num); 1863 } 1864 1865 #ifdef HAVE_LIBTRACEEVENT 1866 if (session->tevent.pevent && 1867 tep_set_function_resolver(session->tevent.pevent, 1868 machine__resolve_kernel_addr, 1869 &session->machines.host) < 0) { 1870 pr_err("%s: failed to set libtraceevent function resolver\n", 1871 __func__); 1872 return -1; 1873 } 1874 #endif 1875 sort__setup_elide(stdout); 1876 1877 ret = __cmd_report(&report); 1878 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1879 perf_session__delete(session); 1880 last_key = K_SWITCH_INPUT_DATA; 1881 /* 1882 * To support switching between data with and without callchains. 1883 * report__setup_sample_type() will update it properly. 1884 */ 1885 symbol_conf.use_callchain = false; 1886 goto repeat; 1887 } else 1888 ret = 0; 1889 1890 if (!use_browser && (verbose > 2 || debug_kmaps)) 1891 perf_session__dump_kmaps(session); 1892 error: 1893 if (report.ptime_range) { 1894 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1895 zfree(&report.ptime_range); 1896 } 1897 1898 if (report.block_reports) { 1899 block_info__free_report(report.block_reports, 1900 report.nr_block_reports); 1901 report.block_reports = NULL; 1902 } 1903 1904 zstd_fini(&(session->zstd_data)); 1905 perf_session__delete(session); 1906 exit: 1907 annotation_options__exit(); 1908 free(sort_order_help); 1909 free(field_order_help); 1910 return ret; 1911 } 1912