builtin-report.c (47921B)
1// SPDX-License-Identifier: GPL-2.0 2/* 3 * builtin-report.c 4 * 5 * Builtin report command: Analyze the perf.data input file, 6 * look up and read DSOs and symbol information and display 7 * a histogram of results, along various sorting keys. 8 */ 9#include "builtin.h" 10 11#include "util/config.h" 12 13#include "util/annotate.h" 14#include "util/color.h" 15#include "util/dso.h" 16#include <linux/list.h> 17#include <linux/rbtree.h> 18#include <linux/err.h> 19#include <linux/zalloc.h> 20#include "util/map.h" 21#include "util/symbol.h" 22#include "util/map_symbol.h" 23#include "util/mem-events.h" 24#include "util/branch.h" 25#include "util/callchain.h" 26#include "util/values.h" 27 28#include "perf.h" 29#include "util/debug.h" 30#include "util/evlist.h" 31#include "util/evsel.h" 32#include "util/evswitch.h" 33#include "util/header.h" 34#include "util/session.h" 35#include "util/srcline.h" 36#include "util/tool.h" 37 38#include <subcmd/parse-options.h> 39#include <subcmd/exec-cmd.h> 40#include "util/parse-events.h" 41 42#include "util/thread.h" 43#include "util/sort.h" 44#include "util/hist.h" 45#include "util/data.h" 46#include "arch/common.h" 47#include "util/time-utils.h" 48#include "util/auxtrace.h" 49#include "util/units.h" 50#include "util/util.h" // perf_tip() 51#include "ui/ui.h" 52#include "ui/progress.h" 53#include "util/block-info.h" 54 55#include <dlfcn.h> 56#include <errno.h> 57#include <inttypes.h> 58#include <regex.h> 59#include <linux/ctype.h> 60#include <signal.h> 61#include <linux/bitmap.h> 62#include <linux/string.h> 63#include <linux/stringify.h> 64#include <linux/time64.h> 65#include <sys/types.h> 66#include <sys/stat.h> 67#include <unistd.h> 68#include <linux/mman.h> 69 70struct report { 71 struct perf_tool tool; 72 struct perf_session *session; 73 struct evswitch evswitch; 74#ifdef HAVE_SLANG_SUPPORT 75 bool use_tui; 76#endif 77 bool use_gtk; 78 bool use_stdio; 79 bool show_full_info; 80 bool show_threads; 81 bool inverted_callchain; 82 bool mem_mode; 83 bool stats_mode; 84 bool tasks_mode; 85 bool mmaps_mode; 86 bool header; 87 bool header_only; 88 bool nonany_branch_mode; 89 bool group_set; 90 bool stitch_lbr; 91 bool disable_order; 92 bool skip_empty; 93 int max_stack; 94 struct perf_read_values show_threads_values; 95 struct annotation_options annotation_opts; 96 const char *pretty_printing_style; 97 const char *cpu_list; 98 const char *symbol_filter_str; 99 const char *time_str; 100 struct perf_time_interval *ptime_range; 101 int range_size; 102 int range_num; 103 float min_percent; 104 u64 nr_entries; 105 u64 queue_size; 106 u64 total_cycles; 107 int socket_filter; 108 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 109 struct branch_type_stat brtype_stat; 110 bool symbol_ipc; 111 bool total_cycles_mode; 112 struct block_report *block_reports; 113 int nr_block_reports; 114}; 115 116static int report__config(const char *var, const char *value, void *cb) 117{ 118 struct report *rep = cb; 119 120 if (!strcmp(var, "report.group")) { 121 symbol_conf.event_group = perf_config_bool(var, value); 122 return 0; 123 } 124 if (!strcmp(var, "report.percent-limit")) { 125 double pcnt = strtof(value, NULL); 126 127 rep->min_percent = pcnt; 128 callchain_param.min_percent = pcnt; 129 return 0; 130 } 131 if (!strcmp(var, "report.children")) { 132 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 133 return 0; 134 } 135 if (!strcmp(var, "report.queue-size")) 136 return perf_config_u64(&rep->queue_size, var, value); 137 138 if (!strcmp(var, "report.sort_order")) { 139 default_sort_order = strdup(value); 140 return 0; 141 } 142 143 if (!strcmp(var, "report.skip-empty")) { 144 rep->skip_empty = perf_config_bool(var, value); 145 return 0; 146 } 147 148 return 0; 149} 150 151static int hist_iter__report_callback(struct hist_entry_iter *iter, 152 struct addr_location *al, bool single, 153 void *arg) 154{ 155 int err = 0; 156 struct report *rep = arg; 157 struct hist_entry *he = iter->he; 158 struct evsel *evsel = iter->evsel; 159 struct perf_sample *sample = iter->sample; 160 struct mem_info *mi; 161 struct branch_info *bi; 162 163 if (!ui__has_annotation() && !rep->symbol_ipc) 164 return 0; 165 166 if (sort__mode == SORT_MODE__BRANCH) { 167 bi = he->branch_info; 168 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 169 if (err) 170 goto out; 171 172 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 173 174 } else if (rep->mem_mode) { 175 mi = he->mem_info; 176 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel); 177 if (err) 178 goto out; 179 180 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 181 182 } else if (symbol_conf.cumulate_callchain) { 183 if (single) 184 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 185 } else { 186 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 187 } 188 189out: 190 return err; 191} 192 193static int hist_iter__branch_callback(struct hist_entry_iter *iter, 194 struct addr_location *al __maybe_unused, 195 bool single __maybe_unused, 196 void *arg) 197{ 198 struct hist_entry *he = iter->he; 199 struct report *rep = arg; 200 struct branch_info *bi = he->branch_info; 201 struct perf_sample *sample = iter->sample; 202 struct evsel *evsel = iter->evsel; 203 int err; 204 205 branch_type_count(&rep->brtype_stat, &bi->flags, 206 bi->from.addr, bi->to.addr); 207 208 if (!ui__has_annotation() && !rep->symbol_ipc) 209 return 0; 210 211 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 212 if (err) 213 goto out; 214 215 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 216 217out: 218 return err; 219} 220 221static void setup_forced_leader(struct report *report, 222 struct evlist *evlist) 223{ 224 if (report->group_set) 225 evlist__force_leader(evlist); 226} 227 228static int process_feature_event(struct perf_session *session, 229 union perf_event *event) 230{ 231 struct report *rep = container_of(session->tool, struct report, tool); 232 233 if (event->feat.feat_id < HEADER_LAST_FEATURE) 234 return perf_event__process_feature(session, event); 235 236 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 237 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 238 event->feat.feat_id); 239 return -1; 240 } else if (rep->header_only) { 241 session_done = 1; 242 } 243 244 /* 245 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 246 * means all features are received, now we can force the 247 * group if needed. 248 */ 249 setup_forced_leader(rep, session->evlist); 250 return 0; 251} 252 253static int process_sample_event(struct perf_tool *tool, 254 union perf_event *event, 255 struct perf_sample *sample, 256 struct evsel *evsel, 257 struct machine *machine) 258{ 259 struct report *rep = container_of(tool, struct report, tool); 260 struct addr_location al; 261 struct hist_entry_iter iter = { 262 .evsel = evsel, 263 .sample = sample, 264 .hide_unresolved = symbol_conf.hide_unresolved, 265 .add_entry_cb = hist_iter__report_callback, 266 }; 267 int ret = 0; 268 269 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 270 sample->time)) { 271 return 0; 272 } 273 274 if (evswitch__discard(&rep->evswitch, evsel)) 275 return 0; 276 277 if (machine__resolve(machine, &al, sample) < 0) { 278 pr_debug("problem processing %d event, skipping it.\n", 279 event->header.type); 280 return -1; 281 } 282 283 if (rep->stitch_lbr) 284 al.thread->lbr_stitch_enable = true; 285 286 if (symbol_conf.hide_unresolved && al.sym == NULL) 287 goto out_put; 288 289 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 290 goto out_put; 291 292 if (sort__mode == SORT_MODE__BRANCH) { 293 /* 294 * A non-synthesized event might not have a branch stack if 295 * branch stacks have been synthesized (using itrace options). 296 */ 297 if (!sample->branch_stack) 298 goto out_put; 299 300 iter.add_entry_cb = hist_iter__branch_callback; 301 iter.ops = &hist_iter_branch; 302 } else if (rep->mem_mode) { 303 iter.ops = &hist_iter_mem; 304 } else if (symbol_conf.cumulate_callchain) { 305 iter.ops = &hist_iter_cumulative; 306 } else { 307 iter.ops = &hist_iter_normal; 308 } 309 310 if (al.map != NULL) 311 al.map->dso->hit = 1; 312 313 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 314 hist__account_cycles(sample->branch_stack, &al, sample, 315 rep->nonany_branch_mode, 316 &rep->total_cycles); 317 } 318 319 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 320 if (ret < 0) 321 pr_debug("problem adding hist entry, skipping event\n"); 322out_put: 323 addr_location__put(&al); 324 return ret; 325} 326 327static int process_read_event(struct perf_tool *tool, 328 union perf_event *event, 329 struct perf_sample *sample __maybe_unused, 330 struct evsel *evsel, 331 struct machine *machine __maybe_unused) 332{ 333 struct report *rep = container_of(tool, struct report, tool); 334 335 if (rep->show_threads) { 336 const char *name = evsel__name(evsel); 337 int err = perf_read_values_add_value(&rep->show_threads_values, 338 event->read.pid, event->read.tid, 339 evsel->core.idx, 340 name, 341 event->read.value); 342 343 if (err) 344 return err; 345 } 346 347 return 0; 348} 349 350/* For pipe mode, sample_type is not currently set */ 351static int report__setup_sample_type(struct report *rep) 352{ 353 struct perf_session *session = rep->session; 354 u64 sample_type = evlist__combined_sample_type(session->evlist); 355 bool is_pipe = perf_data__is_pipe(session->data); 356 struct evsel *evsel; 357 358 if (session->itrace_synth_opts->callchain || 359 session->itrace_synth_opts->add_callchain || 360 (!is_pipe && 361 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 362 !session->itrace_synth_opts->set)) 363 sample_type |= PERF_SAMPLE_CALLCHAIN; 364 365 if (session->itrace_synth_opts->last_branch || 366 session->itrace_synth_opts->add_last_branch) 367 sample_type |= PERF_SAMPLE_BRANCH_STACK; 368 369 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 370 if (perf_hpp_list.parent) { 371 ui__error("Selected --sort parent, but no " 372 "callchain data. Did you call " 373 "'perf record' without -g?\n"); 374 return -EINVAL; 375 } 376 if (symbol_conf.use_callchain && 377 !symbol_conf.show_branchflag_count) { 378 ui__error("Selected -g or --branch-history.\n" 379 "But no callchain or branch data.\n" 380 "Did you call 'perf record' without -g or -b?\n"); 381 return -1; 382 } 383 } else if (!callchain_param.enabled && 384 callchain_param.mode != CHAIN_NONE && 385 !symbol_conf.use_callchain) { 386 symbol_conf.use_callchain = true; 387 if (callchain_register_param(&callchain_param) < 0) { 388 ui__error("Can't register callchain params.\n"); 389 return -EINVAL; 390 } 391 } 392 393 if (symbol_conf.cumulate_callchain) { 394 /* Silently ignore if callchain is missing */ 395 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 396 symbol_conf.cumulate_callchain = false; 397 perf_hpp__cancel_cumulate(); 398 } 399 } 400 401 if (sort__mode == SORT_MODE__BRANCH) { 402 if (!is_pipe && 403 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 404 ui__error("Selected -b but no branch data. " 405 "Did you call perf record without -b?\n"); 406 return -1; 407 } 408 } 409 410 if (sort__mode == SORT_MODE__MEMORY) { 411 /* 412 * FIXUP: prior to kernel 5.18, Arm SPE missed to set 413 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward 414 * compatibility, set the bit if it's an old perf data file. 415 */ 416 evlist__for_each_entry(session->evlist, evsel) { 417 if (strstr(evsel->name, "arm_spe") && 418 !(sample_type & PERF_SAMPLE_DATA_SRC)) { 419 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC; 420 sample_type |= PERF_SAMPLE_DATA_SRC; 421 } 422 } 423 424 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 425 ui__error("Selected --mem-mode but no mem data. " 426 "Did you call perf record without -d?\n"); 427 return -1; 428 } 429 } 430 431 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env)); 432 433 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 434 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 435 "Please apply --call-graph lbr when recording.\n"); 436 rep->stitch_lbr = false; 437 } 438 439 /* ??? handle more cases than just ANY? */ 440 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 441 rep->nonany_branch_mode = true; 442 443#if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT) 444 if (dwarf_callchain_users) { 445 ui__warning("Please install libunwind or libdw " 446 "development packages during the perf build.\n"); 447 } 448#endif 449 450 return 0; 451} 452 453static void sig_handler(int sig __maybe_unused) 454{ 455 session_done = 1; 456} 457 458static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 459 const char *evname, FILE *fp) 460{ 461 size_t ret; 462 char unit; 463 unsigned long nr_samples = hists->stats.nr_samples; 464 u64 nr_events = hists->stats.total_period; 465 struct evsel *evsel = hists_to_evsel(hists); 466 char buf[512]; 467 size_t size = sizeof(buf); 468 int socked_id = hists->socket_filter; 469 470 if (quiet) 471 return 0; 472 473 if (symbol_conf.filter_relative) { 474 nr_samples = hists->stats.nr_non_filtered_samples; 475 nr_events = hists->stats.total_non_filtered_period; 476 } 477 478 if (evsel__is_group_event(evsel)) { 479 struct evsel *pos; 480 481 evsel__group_desc(evsel, buf, size); 482 evname = buf; 483 484 for_each_group_member(pos, evsel) { 485 const struct hists *pos_hists = evsel__hists(pos); 486 487 if (symbol_conf.filter_relative) { 488 nr_samples += pos_hists->stats.nr_non_filtered_samples; 489 nr_events += pos_hists->stats.total_non_filtered_period; 490 } else { 491 nr_samples += pos_hists->stats.nr_samples; 492 nr_events += pos_hists->stats.total_period; 493 } 494 } 495 } 496 497 nr_samples = convert_unit(nr_samples, &unit); 498 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 499 if (evname != NULL) { 500 ret += fprintf(fp, " of event%s '%s'", 501 evsel->core.nr_members > 1 ? "s" : "", evname); 502 } 503 504 if (rep->time_str) 505 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 506 507 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 508 ret += fprintf(fp, ", show reference callgraph"); 509 } 510 511 if (rep->mem_mode) { 512 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 513 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 514 } else 515 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 516 517 if (socked_id > -1) 518 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 519 520 return ret + fprintf(fp, "\n#\n"); 521} 522 523static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 524{ 525 struct evsel *pos; 526 int i = 0, ret; 527 528 evlist__for_each_entry(evlist, pos) { 529 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 530 rep->min_percent, pos, 531 &rep->session->header.env, 532 &rep->annotation_opts); 533 if (ret != 0) 534 return ret; 535 } 536 537 return 0; 538} 539 540static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 541{ 542 struct evsel *pos; 543 int i = 0; 544 545 if (!quiet) { 546 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 547 evlist->stats.total_lost_samples); 548 } 549 550 evlist__for_each_entry(evlist, pos) { 551 struct hists *hists = evsel__hists(pos); 552 const char *evname = evsel__name(pos); 553 554 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 555 continue; 556 557 if (rep->skip_empty && !hists->stats.nr_samples) 558 continue; 559 560 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 561 562 if (rep->total_cycles_mode) { 563 report__browse_block_hists(&rep->block_reports[i++].hist, 564 rep->min_percent, pos, 565 NULL, NULL); 566 continue; 567 } 568 569 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 570 !(symbol_conf.use_callchain || 571 symbol_conf.show_branchflag_count)); 572 fprintf(stdout, "\n\n"); 573 } 574 575 if (!quiet) 576 fprintf(stdout, "#\n# (%s)\n#\n", help); 577 578 if (rep->show_threads) { 579 bool style = !strcmp(rep->pretty_printing_style, "raw"); 580 perf_read_values_display(stdout, &rep->show_threads_values, 581 style); 582 perf_read_values_destroy(&rep->show_threads_values); 583 } 584 585 if (sort__mode == SORT_MODE__BRANCH) 586 branch_type_stat_display(stdout, &rep->brtype_stat); 587 588 return 0; 589} 590 591static void report__warn_kptr_restrict(const struct report *rep) 592{ 593 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 594 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 595 596 if (evlist__exclude_kernel(rep->session->evlist)) 597 return; 598 599 if (kernel_map == NULL || 600 (kernel_map->dso->hit && 601 (kernel_kmap->ref_reloc_sym == NULL || 602 kernel_kmap->ref_reloc_sym->addr == 0))) { 603 const char *desc = 604 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 605 "can't be resolved."; 606 607 if (kernel_map && map__has_symbols(kernel_map)) { 608 desc = "If some relocation was applied (e.g. " 609 "kexec) symbols may be misresolved."; 610 } 611 612 ui__warning( 613"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 614"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 615"Samples in kernel modules can't be resolved as well.\n\n", 616 desc); 617 } 618} 619 620static int report__gtk_browse_hists(struct report *rep, const char *help) 621{ 622 int (*hist_browser)(struct evlist *evlist, const char *help, 623 struct hist_browser_timer *timer, float min_pcnt); 624 625 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 626 627 if (hist_browser == NULL) { 628 ui__error("GTK browser not found!\n"); 629 return -1; 630 } 631 632 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 633} 634 635static int report__browse_hists(struct report *rep) 636{ 637 int ret; 638 struct perf_session *session = rep->session; 639 struct evlist *evlist = session->evlist; 640 char *help = NULL, *path = NULL; 641 642 path = system_path(TIPDIR); 643 if (perf_tip(&help, path) || help == NULL) { 644 /* fallback for people who don't install perf ;-) */ 645 free(path); 646 path = system_path(DOCDIR); 647 if (perf_tip(&help, path) || help == NULL) 648 help = strdup("Cannot load tips.txt file, please install perf!"); 649 } 650 free(path); 651 652 switch (use_browser) { 653 case 1: 654 if (rep->total_cycles_mode) { 655 ret = evlist__tui_block_hists_browse(evlist, rep); 656 break; 657 } 658 659 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 660 &session->header.env, true, &rep->annotation_opts); 661 /* 662 * Usually "ret" is the last pressed key, and we only 663 * care if the key notifies us to switch data file. 664 */ 665 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 666 ret = 0; 667 break; 668 case 2: 669 ret = report__gtk_browse_hists(rep, help); 670 break; 671 default: 672 ret = evlist__tty_browse_hists(evlist, rep, help); 673 break; 674 } 675 free(help); 676 return ret; 677} 678 679static int report__collapse_hists(struct report *rep) 680{ 681 struct ui_progress prog; 682 struct evsel *pos; 683 int ret = 0; 684 685 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 686 687 evlist__for_each_entry(rep->session->evlist, pos) { 688 struct hists *hists = evsel__hists(pos); 689 690 if (pos->core.idx == 0) 691 hists->symbol_filter_str = rep->symbol_filter_str; 692 693 hists->socket_filter = rep->socket_filter; 694 695 ret = hists__collapse_resort(hists, &prog); 696 if (ret < 0) 697 break; 698 699 /* Non-group events are considered as leader */ 700 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 701 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 702 703 hists__match(leader_hists, hists); 704 hists__link(leader_hists, hists); 705 } 706 } 707 708 ui_progress__finish(); 709 return ret; 710} 711 712static int hists__resort_cb(struct hist_entry *he, void *arg) 713{ 714 struct report *rep = arg; 715 struct symbol *sym = he->ms.sym; 716 717 if (rep->symbol_ipc && sym && !sym->annotate2) { 718 struct evsel *evsel = hists_to_evsel(he->hists); 719 720 symbol__annotate2(&he->ms, evsel, 721 &annotation__default_options, NULL); 722 } 723 724 return 0; 725} 726 727static void report__output_resort(struct report *rep) 728{ 729 struct ui_progress prog; 730 struct evsel *pos; 731 732 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 733 734 evlist__for_each_entry(rep->session->evlist, pos) { 735 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 736 } 737 738 ui_progress__finish(); 739} 740 741static int count_sample_event(struct perf_tool *tool __maybe_unused, 742 union perf_event *event __maybe_unused, 743 struct perf_sample *sample __maybe_unused, 744 struct evsel *evsel, 745 struct machine *machine __maybe_unused) 746{ 747 struct hists *hists = evsel__hists(evsel); 748 749 hists__inc_nr_events(hists); 750 return 0; 751} 752 753static int process_attr(struct perf_tool *tool __maybe_unused, 754 union perf_event *event, 755 struct evlist **pevlist); 756 757static void stats_setup(struct report *rep) 758{ 759 memset(&rep->tool, 0, sizeof(rep->tool)); 760 rep->tool.attr = process_attr; 761 rep->tool.sample = count_sample_event; 762 rep->tool.no_warn = true; 763} 764 765static int stats_print(struct report *rep) 766{ 767 struct perf_session *session = rep->session; 768 769 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty); 770 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty); 771 return 0; 772} 773 774static void tasks_setup(struct report *rep) 775{ 776 memset(&rep->tool, 0, sizeof(rep->tool)); 777 rep->tool.ordered_events = true; 778 if (rep->mmaps_mode) { 779 rep->tool.mmap = perf_event__process_mmap; 780 rep->tool.mmap2 = perf_event__process_mmap2; 781 } 782 rep->tool.attr = process_attr; 783 rep->tool.comm = perf_event__process_comm; 784 rep->tool.exit = perf_event__process_exit; 785 rep->tool.fork = perf_event__process_fork; 786 rep->tool.no_warn = true; 787} 788 789struct task { 790 struct thread *thread; 791 struct list_head list; 792 struct list_head children; 793}; 794 795static struct task *tasks_list(struct task *task, struct machine *machine) 796{ 797 struct thread *parent_thread, *thread = task->thread; 798 struct task *parent_task; 799 800 /* Already listed. */ 801 if (!list_empty(&task->list)) 802 return NULL; 803 804 /* Last one in the chain. */ 805 if (thread->ppid == -1) 806 return task; 807 808 parent_thread = machine__find_thread(machine, -1, thread->ppid); 809 if (!parent_thread) 810 return ERR_PTR(-ENOENT); 811 812 parent_task = thread__priv(parent_thread); 813 list_add_tail(&task->list, &parent_task->children); 814 return tasks_list(parent_task, machine); 815} 816 817static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 818{ 819 size_t printed = 0; 820 struct map *map; 821 822 maps__for_each_entry(maps, map) { 823 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 824 indent, "", map->start, map->end, 825 map->prot & PROT_READ ? 'r' : '-', 826 map->prot & PROT_WRITE ? 'w' : '-', 827 map->prot & PROT_EXEC ? 'x' : '-', 828 map->flags & MAP_SHARED ? 's' : 'p', 829 map->pgoff, 830 map->dso->id.ino, map->dso->name); 831 } 832 833 return printed; 834} 835 836static void task__print_level(struct task *task, FILE *fp, int level) 837{ 838 struct thread *thread = task->thread; 839 struct task *child; 840 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 841 thread->pid_, thread->tid, thread->ppid, 842 level, ""); 843 844 fprintf(fp, "%s\n", thread__comm_str(thread)); 845 846 maps__fprintf_task(thread->maps, comm_indent, fp); 847 848 if (!list_empty(&task->children)) { 849 list_for_each_entry(child, &task->children, list) 850 task__print_level(child, fp, level + 1); 851 } 852} 853 854static int tasks_print(struct report *rep, FILE *fp) 855{ 856 struct perf_session *session = rep->session; 857 struct machine *machine = &session->machines.host; 858 struct task *tasks, *task; 859 unsigned int nr = 0, itask = 0, i; 860 struct rb_node *nd; 861 LIST_HEAD(list); 862 863 /* 864 * No locking needed while accessing machine->threads, 865 * because --tasks is single threaded command. 866 */ 867 868 /* Count all the threads. */ 869 for (i = 0; i < THREADS__TABLE_SIZE; i++) 870 nr += machine->threads[i].nr; 871 872 tasks = malloc(sizeof(*tasks) * nr); 873 if (!tasks) 874 return -ENOMEM; 875 876 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 877 struct threads *threads = &machine->threads[i]; 878 879 for (nd = rb_first_cached(&threads->entries); nd; 880 nd = rb_next(nd)) { 881 task = tasks + itask++; 882 883 task->thread = rb_entry(nd, struct thread, rb_node); 884 INIT_LIST_HEAD(&task->children); 885 INIT_LIST_HEAD(&task->list); 886 thread__set_priv(task->thread, task); 887 } 888 } 889 890 /* 891 * Iterate every task down to the unprocessed parent 892 * and link all in task children list. Task with no 893 * parent is added into 'list'. 894 */ 895 for (itask = 0; itask < nr; itask++) { 896 task = tasks + itask; 897 898 if (!list_empty(&task->list)) 899 continue; 900 901 task = tasks_list(task, machine); 902 if (IS_ERR(task)) { 903 pr_err("Error: failed to process tasks\n"); 904 free(tasks); 905 return PTR_ERR(task); 906 } 907 908 if (task) 909 list_add_tail(&task->list, &list); 910 } 911 912 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 913 914 list_for_each_entry(task, &list, list) 915 task__print_level(task, fp, 0); 916 917 free(tasks); 918 return 0; 919} 920 921static int __cmd_report(struct report *rep) 922{ 923 int ret; 924 struct perf_session *session = rep->session; 925 struct evsel *pos; 926 struct perf_data *data = session->data; 927 928 signal(SIGINT, sig_handler); 929 930 if (rep->cpu_list) { 931 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 932 rep->cpu_bitmap); 933 if (ret) { 934 ui__error("failed to set cpu bitmap\n"); 935 return ret; 936 } 937 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 938 } 939 940 if (rep->show_threads) { 941 ret = perf_read_values_init(&rep->show_threads_values); 942 if (ret) 943 return ret; 944 } 945 946 ret = report__setup_sample_type(rep); 947 if (ret) { 948 /* report__setup_sample_type() already showed error message */ 949 return ret; 950 } 951 952 if (rep->stats_mode) 953 stats_setup(rep); 954 955 if (rep->tasks_mode) 956 tasks_setup(rep); 957 958 ret = perf_session__process_events(session); 959 if (ret) { 960 ui__error("failed to process sample\n"); 961 return ret; 962 } 963 964 evlist__check_mem_load_aux(session->evlist); 965 966 if (rep->stats_mode) 967 return stats_print(rep); 968 969 if (rep->tasks_mode) 970 return tasks_print(rep, stdout); 971 972 report__warn_kptr_restrict(rep); 973 974 evlist__for_each_entry(session->evlist, pos) 975 rep->nr_entries += evsel__hists(pos)->nr_entries; 976 977 if (use_browser == 0) { 978 if (verbose > 3) 979 perf_session__fprintf(session, stdout); 980 981 if (verbose > 2) 982 perf_session__fprintf_dsos(session, stdout); 983 984 if (dump_trace) { 985 perf_session__fprintf_nr_events(session, stdout, 986 rep->skip_empty); 987 evlist__fprintf_nr_events(session->evlist, stdout, 988 rep->skip_empty); 989 return 0; 990 } 991 } 992 993 ret = report__collapse_hists(rep); 994 if (ret) { 995 ui__error("failed to process hist entry\n"); 996 return ret; 997 } 998 999 if (session_done()) 1000 return 0; 1001 1002 /* 1003 * recalculate number of entries after collapsing since it 1004 * might be changed during the collapse phase. 1005 */ 1006 rep->nr_entries = 0; 1007 evlist__for_each_entry(session->evlist, pos) 1008 rep->nr_entries += evsel__hists(pos)->nr_entries; 1009 1010 if (rep->nr_entries == 0) { 1011 ui__error("The %s data has no samples!\n", data->path); 1012 return 0; 1013 } 1014 1015 report__output_resort(rep); 1016 1017 if (rep->total_cycles_mode) { 1018 int block_hpps[6] = { 1019 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1020 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1021 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1022 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1023 PERF_HPP_REPORT__BLOCK_RANGE, 1024 PERF_HPP_REPORT__BLOCK_DSO, 1025 }; 1026 1027 rep->block_reports = block_info__create_report(session->evlist, 1028 rep->total_cycles, 1029 block_hpps, 6, 1030 &rep->nr_block_reports); 1031 if (!rep->block_reports) 1032 return -1; 1033 } 1034 1035 return report__browse_hists(rep); 1036} 1037 1038static int 1039report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1040{ 1041 struct callchain_param *callchain = opt->value; 1042 1043 callchain->enabled = !unset; 1044 /* 1045 * --no-call-graph 1046 */ 1047 if (unset) { 1048 symbol_conf.use_callchain = false; 1049 callchain->mode = CHAIN_NONE; 1050 return 0; 1051 } 1052 1053 return parse_callchain_report_opt(arg); 1054} 1055 1056static int 1057parse_time_quantum(const struct option *opt, const char *arg, 1058 int unset __maybe_unused) 1059{ 1060 unsigned long *time_q = opt->value; 1061 char *end; 1062 1063 *time_q = strtoul(arg, &end, 0); 1064 if (end == arg) 1065 goto parse_err; 1066 if (*time_q == 0) { 1067 pr_err("time quantum cannot be 0"); 1068 return -1; 1069 } 1070 end = skip_spaces(end); 1071 if (*end == 0) 1072 return 0; 1073 if (!strcmp(end, "s")) { 1074 *time_q *= NSEC_PER_SEC; 1075 return 0; 1076 } 1077 if (!strcmp(end, "ms")) { 1078 *time_q *= NSEC_PER_MSEC; 1079 return 0; 1080 } 1081 if (!strcmp(end, "us")) { 1082 *time_q *= NSEC_PER_USEC; 1083 return 0; 1084 } 1085 if (!strcmp(end, "ns")) 1086 return 0; 1087parse_err: 1088 pr_err("Cannot parse time quantum `%s'\n", arg); 1089 return -1; 1090} 1091 1092int 1093report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1094 const char *arg, int unset __maybe_unused) 1095{ 1096 if (arg) { 1097 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1098 if (err) { 1099 char buf[BUFSIZ]; 1100 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1101 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1102 return -1; 1103 } 1104 have_ignore_callees = 1; 1105 } 1106 1107 return 0; 1108} 1109 1110static int 1111parse_branch_mode(const struct option *opt, 1112 const char *str __maybe_unused, int unset) 1113{ 1114 int *branch_mode = opt->value; 1115 1116 *branch_mode = !unset; 1117 return 0; 1118} 1119 1120static int 1121parse_percent_limit(const struct option *opt, const char *str, 1122 int unset __maybe_unused) 1123{ 1124 struct report *rep = opt->value; 1125 double pcnt = strtof(str, NULL); 1126 1127 rep->min_percent = pcnt; 1128 callchain_param.min_percent = pcnt; 1129 return 0; 1130} 1131 1132static int process_attr(struct perf_tool *tool __maybe_unused, 1133 union perf_event *event, 1134 struct evlist **pevlist) 1135{ 1136 u64 sample_type; 1137 int err; 1138 1139 err = perf_event__process_attr(tool, event, pevlist); 1140 if (err) 1141 return err; 1142 1143 /* 1144 * Check if we need to enable callchains based 1145 * on events sample_type. 1146 */ 1147 sample_type = evlist__combined_sample_type(*pevlist); 1148 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env)); 1149 return 0; 1150} 1151 1152int cmd_report(int argc, const char **argv) 1153{ 1154 struct perf_session *session; 1155 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1156 struct stat st; 1157 bool has_br_stack = false; 1158 int branch_mode = -1; 1159 int last_key = 0; 1160 bool branch_call_mode = false; 1161#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1162 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1163 CALLCHAIN_REPORT_HELP 1164 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1165 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1166 const char * const report_usage[] = { 1167 "perf report [<options>]", 1168 NULL 1169 }; 1170 struct report report = { 1171 .tool = { 1172 .sample = process_sample_event, 1173 .mmap = perf_event__process_mmap, 1174 .mmap2 = perf_event__process_mmap2, 1175 .comm = perf_event__process_comm, 1176 .namespaces = perf_event__process_namespaces, 1177 .cgroup = perf_event__process_cgroup, 1178 .exit = perf_event__process_exit, 1179 .fork = perf_event__process_fork, 1180 .lost = perf_event__process_lost, 1181 .read = process_read_event, 1182 .attr = process_attr, 1183 .tracing_data = perf_event__process_tracing_data, 1184 .build_id = perf_event__process_build_id, 1185 .id_index = perf_event__process_id_index, 1186 .auxtrace_info = perf_event__process_auxtrace_info, 1187 .auxtrace = perf_event__process_auxtrace, 1188 .event_update = perf_event__process_event_update, 1189 .feature = process_feature_event, 1190 .ordered_events = true, 1191 .ordering_requires_timestamps = true, 1192 }, 1193 .max_stack = PERF_MAX_STACK_DEPTH, 1194 .pretty_printing_style = "normal", 1195 .socket_filter = -1, 1196 .annotation_opts = annotation__default_options, 1197 .skip_empty = true, 1198 }; 1199 char *sort_order_help = sort_help("sort by key(s):"); 1200 char *field_order_help = sort_help("output field(s): overhead period sample "); 1201 const struct option options[] = { 1202 OPT_STRING('i', "input", &input_name, "file", 1203 "input file name"), 1204 OPT_INCR('v', "verbose", &verbose, 1205 "be more verbose (show symbol address, etc)"), 1206 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"), 1207 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1208 "dump raw trace in ASCII"), 1209 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1210 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1211 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1212 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1213 "file", "vmlinux pathname"), 1214 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1215 "don't load vmlinux even if found"), 1216 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1217 "file", "kallsyms pathname"), 1218 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1219 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1220 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1221 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1222 "Show a column with the number of samples"), 1223 OPT_BOOLEAN('T', "threads", &report.show_threads, 1224 "Show per-thread event counters"), 1225 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1226 "pretty printing style key: normal raw"), 1227#ifdef HAVE_SLANG_SUPPORT 1228 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1229#endif 1230 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1231 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1232 "Use the stdio interface"), 1233 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1234 OPT_BOOLEAN(0, "header-only", &report.header_only, 1235 "Show only data header."), 1236 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1237 sort_order_help), 1238 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1239 field_order_help), 1240 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1241 "Show sample percentage for different cpu modes"), 1242 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1243 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1244 OPT_STRING('p', "parent", &parent_pattern, "regex", 1245 "regex filter to identify parent, see: '--sort parent'"), 1246 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1247 "Only display entries with parent-match"), 1248 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1249 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1250 report_callchain_help, &report_parse_callchain_opt, 1251 callchain_default_opt), 1252 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1253 "Accumulate callchains of children and show total overhead as well. " 1254 "Enabled by default, use --no-children to disable."), 1255 OPT_INTEGER(0, "max-stack", &report.max_stack, 1256 "Set the maximum stack depth when parsing the callchain, " 1257 "anything beyond the specified depth will be ignored. " 1258 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1259 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1260 "alias for inverted call graph"), 1261 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1262 "ignore callees of these functions in call graphs", 1263 report_parse_ignore_callees_opt), 1264 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1265 "only consider symbols in these dsos"), 1266 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1267 "only consider symbols in these comms"), 1268 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1269 "only consider symbols in these pids"), 1270 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1271 "only consider symbols in these tids"), 1272 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1273 "only consider these symbols"), 1274 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1275 "only show symbols that (partially) match with this filter"), 1276 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1277 "width[,width...]", 1278 "don't try to adjust column width, use these fixed values"), 1279 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1280 "separator for columns, no spaces will be added between " 1281 "columns '.' is reserved."), 1282 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1283 "Only display entries resolved to a symbol"), 1284 OPT_CALLBACK(0, "symfs", NULL, "directory", 1285 "Look for files with symbols relative to this directory", 1286 symbol__config_symfs), 1287 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1288 "list of cpus to profile"), 1289 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1290 "Display extended information about perf.data file"), 1291 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src, 1292 "Interleave source code with assembly code (default)"), 1293 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw, 1294 "Display raw encoding of assembly instructions (default)"), 1295 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style", 1296 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1297 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix", 1298 "Add prefix to source file path names in programs (with --prefix-strip)"), 1299 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N", 1300 "Strip first N entries of source file path name in programs (with --prefix)"), 1301 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1302 "Show a column with the sum of periods"), 1303 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1304 "Show event group information together"), 1305 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1306 "Sort the output by the event at the index n in group. " 1307 "If n is invalid, sort by the first event. " 1308 "WARNING: should be used on grouped events."), 1309 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1310 "use branch records for per branch histogram filling", 1311 parse_branch_mode), 1312 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1313 "add last branch records to call history"), 1314 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path", 1315 "objdump binary to use for disassembly and annotations"), 1316 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1317 "Disable symbol demangling"), 1318 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1319 "Enable kernel symbol demangling"), 1320 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1321 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1322 "Number of samples to save per histogram entry for individual browsing"), 1323 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1324 "Don't show entries under that percent", parse_percent_limit), 1325 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1326 "how to display percentage of filtered entries", parse_filter_percentage), 1327 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1328 "Instruction Tracing options\n" ITRACE_HELP, 1329 itrace_parse_synth_opts), 1330 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1331 "Show full source file name path for source lines"), 1332 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1333 "Show callgraph from reference event"), 1334 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1335 "Enable LBR callgraph stitching approach"), 1336 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1337 "only show processor socket that match with this filter"), 1338 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1339 "Show raw trace event output (do not use print fmt or plugins)"), 1340 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1341 "Show entries in a hierarchy"), 1342 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1343 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1344 stdio__config_color, "always"), 1345 OPT_STRING(0, "time", &report.time_str, "str", 1346 "Time span of interest (start,stop)"), 1347 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1348 "Show inline function"), 1349 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period", 1350 "Set percent type local/global-period/hits", 1351 annotate_parse_percent_type), 1352 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1353 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1354 "Set time quantum for time sort key (default 100ms)", 1355 parse_time_quantum), 1356 OPTS_EVSWITCH(&report.evswitch), 1357 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1358 "Sort all blocks by 'Sampled Cycles%'"), 1359 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1360 "Disable raw trace ordering"), 1361 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1362 "Do not display empty (or dummy) events in the output"), 1363 OPT_END() 1364 }; 1365 struct perf_data data = { 1366 .mode = PERF_DATA_MODE_READ, 1367 }; 1368 int ret = hists__init(); 1369 char sort_tmp[128]; 1370 1371 if (ret < 0) 1372 goto exit; 1373 1374 ret = perf_config(report__config, &report); 1375 if (ret) 1376 goto exit; 1377 1378 argc = parse_options(argc, argv, options, report_usage, 0); 1379 if (argc) { 1380 /* 1381 * Special case: if there's an argument left then assume that 1382 * it's a symbol filter: 1383 */ 1384 if (argc > 1) 1385 usage_with_options(report_usage, options); 1386 1387 report.symbol_filter_str = argv[0]; 1388 } 1389 1390 if (annotate_check_args(&report.annotation_opts) < 0) { 1391 ret = -EINVAL; 1392 goto exit; 1393 } 1394 1395 if (report.mmaps_mode) 1396 report.tasks_mode = true; 1397 1398 if (dump_trace && report.disable_order) 1399 report.tool.ordered_events = false; 1400 1401 if (quiet) 1402 perf_quiet_option(); 1403 1404 ret = symbol__validate_sym_arguments(); 1405 if (ret) 1406 goto exit; 1407 1408 if (report.inverted_callchain) 1409 callchain_param.order = ORDER_CALLER; 1410 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1411 callchain_param.order = ORDER_CALLER; 1412 1413 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1414 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1415 report.max_stack = itrace_synth_opts.callchain_sz; 1416 1417 if (!input_name || !strlen(input_name)) { 1418 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1419 input_name = "-"; 1420 else 1421 input_name = "perf.data"; 1422 } 1423 1424 data.path = input_name; 1425 data.force = symbol_conf.force; 1426 1427repeat: 1428 session = perf_session__new(&data, &report.tool); 1429 if (IS_ERR(session)) { 1430 ret = PTR_ERR(session); 1431 goto exit; 1432 } 1433 1434 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1435 if (ret) 1436 goto exit; 1437 1438 if (zstd_init(&(session->zstd_data), 0) < 0) 1439 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1440 1441 if (report.queue_size) { 1442 ordered_events__set_alloc_size(&session->ordered_events, 1443 report.queue_size); 1444 } 1445 1446 session->itrace_synth_opts = &itrace_synth_opts; 1447 1448 report.session = session; 1449 1450 has_br_stack = perf_header__has_feat(&session->header, 1451 HEADER_BRANCH_STACK); 1452 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1453 has_br_stack = false; 1454 1455 setup_forced_leader(&report, session->evlist); 1456 1457 if (symbol_conf.group_sort_idx && !session->evlist->core.nr_groups) { 1458 parse_options_usage(NULL, options, "group-sort-idx", 0); 1459 ret = -EINVAL; 1460 goto error; 1461 } 1462 1463 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1464 has_br_stack = true; 1465 1466 if (has_br_stack && branch_call_mode) 1467 symbol_conf.show_branchflag_count = true; 1468 1469 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1470 1471 /* 1472 * Branch mode is a tristate: 1473 * -1 means default, so decide based on the file having branch data. 1474 * 0/1 means the user chose a mode. 1475 */ 1476 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1477 !branch_call_mode) { 1478 sort__mode = SORT_MODE__BRANCH; 1479 symbol_conf.cumulate_callchain = false; 1480 } 1481 if (branch_call_mode) { 1482 callchain_param.key = CCKEY_ADDRESS; 1483 callchain_param.branch_callstack = true; 1484 symbol_conf.use_callchain = true; 1485 callchain_register_param(&callchain_param); 1486 if (sort_order == NULL) 1487 sort_order = "srcline,symbol,dso"; 1488 } 1489 1490 if (report.mem_mode) { 1491 if (sort__mode == SORT_MODE__BRANCH) { 1492 pr_err("branch and mem mode incompatible\n"); 1493 goto error; 1494 } 1495 sort__mode = SORT_MODE__MEMORY; 1496 symbol_conf.cumulate_callchain = false; 1497 } 1498 1499 if (symbol_conf.report_hierarchy) { 1500 /* disable incompatible options */ 1501 symbol_conf.cumulate_callchain = false; 1502 1503 if (field_order) { 1504 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1505 parse_options_usage(report_usage, options, "F", 1); 1506 parse_options_usage(NULL, options, "hierarchy", 0); 1507 goto error; 1508 } 1509 1510 perf_hpp_list.need_collapse = true; 1511 } 1512 1513 if (report.use_stdio) 1514 use_browser = 0; 1515#ifdef HAVE_SLANG_SUPPORT 1516 else if (report.use_tui) 1517 use_browser = 1; 1518#endif 1519 else if (report.use_gtk) 1520 use_browser = 2; 1521 1522 /* Force tty output for header output and per-thread stat. */ 1523 if (report.header || report.header_only || report.show_threads) 1524 use_browser = 0; 1525 if (report.header || report.header_only) 1526 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1527 if (report.show_full_info) 1528 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1529 if (report.stats_mode || report.tasks_mode) 1530 use_browser = 0; 1531 if (report.stats_mode && report.tasks_mode) { 1532 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1533 goto error; 1534 } 1535 1536 if (report.total_cycles_mode) { 1537 if (sort__mode != SORT_MODE__BRANCH) 1538 report.total_cycles_mode = false; 1539 else 1540 sort_order = NULL; 1541 } 1542 1543 if (strcmp(input_name, "-") != 0) 1544 setup_browser(true); 1545 else 1546 use_browser = 0; 1547 1548 if (sort_order && strstr(sort_order, "ipc")) { 1549 parse_options_usage(report_usage, options, "s", 1); 1550 goto error; 1551 } 1552 1553 if (sort_order && strstr(sort_order, "symbol")) { 1554 if (sort__mode == SORT_MODE__BRANCH) { 1555 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1556 sort_order, "ipc_lbr"); 1557 report.symbol_ipc = true; 1558 } else { 1559 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1560 sort_order, "ipc_null"); 1561 } 1562 1563 sort_order = sort_tmp; 1564 } 1565 1566 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1567 (setup_sorting(session->evlist) < 0)) { 1568 if (sort_order) 1569 parse_options_usage(report_usage, options, "s", 1); 1570 if (field_order) 1571 parse_options_usage(sort_order ? NULL : report_usage, 1572 options, "F", 1); 1573 goto error; 1574 } 1575 1576 if ((report.header || report.header_only) && !quiet) { 1577 perf_session__fprintf_info(session, stdout, 1578 report.show_full_info); 1579 if (report.header_only) { 1580 if (data.is_pipe) { 1581 /* 1582 * we need to process first few records 1583 * which contains PERF_RECORD_HEADER_FEATURE. 1584 */ 1585 perf_session__process_events(session); 1586 } 1587 ret = 0; 1588 goto error; 1589 } 1590 } else if (use_browser == 0 && !quiet && 1591 !report.stats_mode && !report.tasks_mode) { 1592 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1593 stdout); 1594 } 1595 1596 /* 1597 * Only in the TUI browser we are doing integrated annotation, 1598 * so don't allocate extra space that won't be used in the stdio 1599 * implementation. 1600 */ 1601 if (ui__has_annotation() || report.symbol_ipc || 1602 report.total_cycles_mode) { 1603 ret = symbol__annotation_init(); 1604 if (ret < 0) 1605 goto error; 1606 /* 1607 * For searching by name on the "Browse map details". 1608 * providing it only in verbose mode not to bloat too 1609 * much struct symbol. 1610 */ 1611 if (verbose > 0) { 1612 /* 1613 * XXX: Need to provide a less kludgy way to ask for 1614 * more space per symbol, the u32 is for the index on 1615 * the ui browser. 1616 * See symbol__browser_index. 1617 */ 1618 symbol_conf.priv_size += sizeof(u32); 1619 symbol_conf.sort_by_name = true; 1620 } 1621 annotation_config__init(&report.annotation_opts); 1622 } 1623 1624 if (symbol__init(&session->header.env) < 0) 1625 goto error; 1626 1627 if (report.time_str) { 1628 ret = perf_time__parse_for_ranges(report.time_str, session, 1629 &report.ptime_range, 1630 &report.range_size, 1631 &report.range_num); 1632 if (ret < 0) 1633 goto error; 1634 1635 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1636 report.ptime_range, 1637 report.range_num); 1638 } 1639 1640 if (session->tevent.pevent && 1641 tep_set_function_resolver(session->tevent.pevent, 1642 machine__resolve_kernel_addr, 1643 &session->machines.host) < 0) { 1644 pr_err("%s: failed to set libtraceevent function resolver\n", 1645 __func__); 1646 return -1; 1647 } 1648 1649 sort__setup_elide(stdout); 1650 1651 ret = __cmd_report(&report); 1652 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1653 perf_session__delete(session); 1654 last_key = K_SWITCH_INPUT_DATA; 1655 goto repeat; 1656 } else 1657 ret = 0; 1658 1659error: 1660 if (report.ptime_range) { 1661 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1662 zfree(&report.ptime_range); 1663 } 1664 1665 if (report.block_reports) { 1666 block_info__free_report(report.block_reports, 1667 report.nr_block_reports); 1668 report.block_reports = NULL; 1669 } 1670 1671 zstd_fini(&(session->zstd_data)); 1672 perf_session__delete(session); 1673exit: 1674 free(sort_order_help); 1675 free(field_order_help); 1676 return ret; 1677}