1 /* 2 * builtin-report.c 3 * 4 * Builtin report command: Analyze the perf.data input file, 5 * look up and read DSOs and symbol information and display 6 * a histogram of results, along various sorting keys. 7 */ 8 #include "builtin.h" 9 10 #include "util/util.h" 11 #include "util/cache.h" 12 13 #include "util/annotate.h" 14 #include "util/color.h" 15 #include <linux/list.h> 16 #include <linux/rbtree.h> 17 #include "util/symbol.h" 18 #include "util/callchain.h" 19 #include "util/strlist.h" 20 #include "util/values.h" 21 22 #include "perf.h" 23 #include "util/debug.h" 24 #include "util/evlist.h" 25 #include "util/evsel.h" 26 #include "util/header.h" 27 #include "util/session.h" 28 #include "util/tool.h" 29 30 #include "util/parse-options.h" 31 #include "util/parse-events.h" 32 33 #include "util/thread.h" 34 #include "util/sort.h" 35 #include "util/hist.h" 36 #include "util/data.h" 37 #include "arch/common.h" 38 39 #include <dlfcn.h> 40 #include <linux/bitmap.h> 41 42 struct report { 43 struct perf_tool tool; 44 struct perf_session *session; 45 bool force, use_tui, use_gtk, use_stdio; 46 bool hide_unresolved; 47 bool dont_use_callchains; 48 bool show_full_info; 49 bool show_threads; 50 bool inverted_callchain; 51 bool mem_mode; 52 bool header; 53 bool header_only; 54 int max_stack; 55 struct perf_read_values show_threads_values; 56 const char *pretty_printing_style; 57 const char *cpu_list; 58 const char *symbol_filter_str; 59 float min_percent; 60 u64 nr_entries; 61 u64 queue_size; 62 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 63 }; 64 65 static int report__config(const char *var, const char *value, void *cb) 66 { 67 struct report *rep = cb; 68 69 if (!strcmp(var, "report.group")) { 70 symbol_conf.event_group = perf_config_bool(var, value); 71 return 0; 72 } 73 if (!strcmp(var, "report.percent-limit")) { 74 rep->min_percent = strtof(value, NULL); 75 return 0; 76 } 77 if (!strcmp(var, "report.children")) { 78 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 79 return 0; 80 } 81 if (!strcmp(var, "report.queue-size")) { 82 rep->queue_size = perf_config_u64(var, value); 83 return 0; 84 } 85 86 return perf_default_config(var, value, cb); 87 } 88 89 static int hist_iter__report_callback(struct hist_entry_iter *iter, 90 struct addr_location *al, bool single, 91 void *arg) 92 { 93 int err = 0; 94 struct report *rep = arg; 95 struct hist_entry *he = iter->he; 96 struct perf_evsel *evsel = iter->evsel; 97 struct mem_info *mi; 98 struct branch_info *bi; 99 100 if (!ui__has_annotation()) 101 return 0; 102 103 if (sort__mode == SORT_MODE__BRANCH) { 104 bi = he->branch_info; 105 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx); 106 if (err) 107 goto out; 108 109 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx); 110 111 } else if (rep->mem_mode) { 112 mi = he->mem_info; 113 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx); 114 if (err) 115 goto out; 116 117 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 118 119 } else if (symbol_conf.cumulate_callchain) { 120 if (single) 121 err = hist_entry__inc_addr_samples(he, evsel->idx, 122 al->addr); 123 } else { 124 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 125 } 126 127 out: 128 return err; 129 } 130 131 static int process_sample_event(struct perf_tool *tool, 132 union perf_event *event, 133 struct perf_sample *sample, 134 struct perf_evsel *evsel, 135 struct machine *machine) 136 { 137 struct report *rep = container_of(tool, struct report, tool); 138 struct addr_location al; 139 struct hist_entry_iter iter = { 140 .hide_unresolved = rep->hide_unresolved, 141 .add_entry_cb = hist_iter__report_callback, 142 }; 143 int ret; 144 145 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) { 146 pr_debug("problem processing %d event, skipping it.\n", 147 event->header.type); 148 return -1; 149 } 150 151 if (rep->hide_unresolved && al.sym == NULL) 152 return 0; 153 154 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 155 return 0; 156 157 if (sort__mode == SORT_MODE__BRANCH) 158 iter.ops = &hist_iter_branch; 159 else if (rep->mem_mode) 160 iter.ops = &hist_iter_mem; 161 else if (symbol_conf.cumulate_callchain) 162 iter.ops = &hist_iter_cumulative; 163 else 164 iter.ops = &hist_iter_normal; 165 166 if (al.map != NULL) 167 al.map->dso->hit = 1; 168 169 ret = hist_entry_iter__add(&iter, &al, evsel, sample, rep->max_stack, 170 rep); 171 if (ret < 0) 172 pr_debug("problem adding hist entry, skipping event\n"); 173 174 return ret; 175 } 176 177 static int process_read_event(struct perf_tool *tool, 178 union perf_event *event, 179 struct perf_sample *sample __maybe_unused, 180 struct perf_evsel *evsel, 181 struct machine *machine __maybe_unused) 182 { 183 struct report *rep = container_of(tool, struct report, tool); 184 185 if (rep->show_threads) { 186 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 187 perf_read_values_add_value(&rep->show_threads_values, 188 event->read.pid, event->read.tid, 189 event->read.id, 190 name, 191 event->read.value); 192 } 193 194 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 195 evsel ? perf_evsel__name(evsel) : "FAIL", 196 event->read.value); 197 198 return 0; 199 } 200 201 /* For pipe mode, sample_type is not currently set */ 202 static int report__setup_sample_type(struct report *rep) 203 { 204 struct perf_session *session = rep->session; 205 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 206 bool is_pipe = perf_data_file__is_pipe(session->file); 207 208 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 209 if (sort__has_parent) { 210 ui__error("Selected --sort parent, but no " 211 "callchain data. Did you call " 212 "'perf record' without -g?\n"); 213 return -EINVAL; 214 } 215 if (symbol_conf.use_callchain) { 216 ui__error("Selected -g or --branch-history but no " 217 "callchain data. Did\n" 218 "you call 'perf record' without -g?\n"); 219 return -1; 220 } 221 } else if (!rep->dont_use_callchains && 222 callchain_param.mode != CHAIN_NONE && 223 !symbol_conf.use_callchain) { 224 symbol_conf.use_callchain = true; 225 if (callchain_register_param(&callchain_param) < 0) { 226 ui__error("Can't register callchain params.\n"); 227 return -EINVAL; 228 } 229 } 230 231 if (symbol_conf.cumulate_callchain) { 232 /* Silently ignore if callchain is missing */ 233 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 234 symbol_conf.cumulate_callchain = false; 235 perf_hpp__cancel_cumulate(); 236 } 237 } 238 239 if (sort__mode == SORT_MODE__BRANCH) { 240 if (!is_pipe && 241 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 242 ui__error("Selected -b but no branch data. " 243 "Did you call perf record without -b?\n"); 244 return -1; 245 } 246 } 247 248 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) { 249 if ((sample_type & PERF_SAMPLE_REGS_USER) && 250 (sample_type & PERF_SAMPLE_STACK_USER)) 251 callchain_param.record_mode = CALLCHAIN_DWARF; 252 else 253 callchain_param.record_mode = CALLCHAIN_FP; 254 } 255 return 0; 256 } 257 258 static void sig_handler(int sig __maybe_unused) 259 { 260 session_done = 1; 261 } 262 263 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 264 const char *evname, FILE *fp) 265 { 266 size_t ret; 267 char unit; 268 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 269 u64 nr_events = hists->stats.total_period; 270 struct perf_evsel *evsel = hists_to_evsel(hists); 271 char buf[512]; 272 size_t size = sizeof(buf); 273 274 if (symbol_conf.filter_relative) { 275 nr_samples = hists->stats.nr_non_filtered_samples; 276 nr_events = hists->stats.total_non_filtered_period; 277 } 278 279 if (perf_evsel__is_group_event(evsel)) { 280 struct perf_evsel *pos; 281 282 perf_evsel__group_desc(evsel, buf, size); 283 evname = buf; 284 285 for_each_group_member(pos, evsel) { 286 const struct hists *pos_hists = evsel__hists(pos); 287 288 if (symbol_conf.filter_relative) { 289 nr_samples += pos_hists->stats.nr_non_filtered_samples; 290 nr_events += pos_hists->stats.total_non_filtered_period; 291 } else { 292 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE]; 293 nr_events += pos_hists->stats.total_period; 294 } 295 } 296 } 297 298 nr_samples = convert_unit(nr_samples, &unit); 299 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 300 if (evname != NULL) 301 ret += fprintf(fp, " of event '%s'", evname); 302 303 if (rep->mem_mode) { 304 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 305 ret += fprintf(fp, "\n# Sort order : %s", sort_order); 306 } else 307 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 308 return ret + fprintf(fp, "\n#\n"); 309 } 310 311 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 312 struct report *rep, 313 const char *help) 314 { 315 struct perf_evsel *pos; 316 317 evlist__for_each(evlist, pos) { 318 struct hists *hists = evsel__hists(pos); 319 const char *evname = perf_evsel__name(pos); 320 321 if (symbol_conf.event_group && 322 !perf_evsel__is_group_leader(pos)) 323 continue; 324 325 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 326 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout); 327 fprintf(stdout, "\n\n"); 328 } 329 330 if (sort_order == default_sort_order && 331 parent_pattern == default_parent_pattern) { 332 fprintf(stdout, "#\n# (%s)\n#\n", help); 333 334 if (rep->show_threads) { 335 bool style = !strcmp(rep->pretty_printing_style, "raw"); 336 perf_read_values_display(stdout, &rep->show_threads_values, 337 style); 338 perf_read_values_destroy(&rep->show_threads_values); 339 } 340 } 341 342 return 0; 343 } 344 345 static void report__warn_kptr_restrict(const struct report *rep) 346 { 347 struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION]; 348 struct kmap *kernel_kmap = map__kmap(kernel_map); 349 350 if (kernel_map == NULL || 351 (kernel_map->dso->hit && 352 (kernel_kmap->ref_reloc_sym == NULL || 353 kernel_kmap->ref_reloc_sym->addr == 0))) { 354 const char *desc = 355 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 356 "can't be resolved."; 357 358 if (kernel_map) { 359 const struct dso *kdso = kernel_map->dso; 360 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 361 desc = "If some relocation was applied (e.g. " 362 "kexec) symbols may be misresolved."; 363 } 364 } 365 366 ui__warning( 367 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 368 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 369 "Samples in kernel modules can't be resolved as well.\n\n", 370 desc); 371 } 372 } 373 374 static int report__gtk_browse_hists(struct report *rep, const char *help) 375 { 376 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 377 struct hist_browser_timer *timer, float min_pcnt); 378 379 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 380 381 if (hist_browser == NULL) { 382 ui__error("GTK browser not found!\n"); 383 return -1; 384 } 385 386 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 387 } 388 389 static int report__browse_hists(struct report *rep) 390 { 391 int ret; 392 struct perf_session *session = rep->session; 393 struct perf_evlist *evlist = session->evlist; 394 const char *help = "For a higher level overview, try: perf report --sort comm,dso"; 395 396 switch (use_browser) { 397 case 1: 398 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 399 rep->min_percent, 400 &session->header.env); 401 /* 402 * Usually "ret" is the last pressed key, and we only 403 * care if the key notifies us to switch data file. 404 */ 405 if (ret != K_SWITCH_INPUT_DATA) 406 ret = 0; 407 break; 408 case 2: 409 ret = report__gtk_browse_hists(rep, help); 410 break; 411 default: 412 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 413 break; 414 } 415 416 return ret; 417 } 418 419 static void report__collapse_hists(struct report *rep) 420 { 421 struct ui_progress prog; 422 struct perf_evsel *pos; 423 424 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 425 426 evlist__for_each(rep->session->evlist, pos) { 427 struct hists *hists = evsel__hists(pos); 428 429 if (pos->idx == 0) 430 hists->symbol_filter_str = rep->symbol_filter_str; 431 432 hists__collapse_resort(hists, &prog); 433 434 /* Non-group events are considered as leader */ 435 if (symbol_conf.event_group && 436 !perf_evsel__is_group_leader(pos)) { 437 struct hists *leader_hists = evsel__hists(pos->leader); 438 439 hists__match(leader_hists, hists); 440 hists__link(leader_hists, hists); 441 } 442 } 443 444 ui_progress__finish(); 445 } 446 447 static void report__output_resort(struct report *rep) 448 { 449 struct ui_progress prog; 450 struct perf_evsel *pos; 451 452 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 453 454 evlist__for_each(rep->session->evlist, pos) 455 hists__output_resort(evsel__hists(pos), &prog); 456 457 ui_progress__finish(); 458 } 459 460 static int __cmd_report(struct report *rep) 461 { 462 int ret; 463 struct perf_session *session = rep->session; 464 struct perf_evsel *pos; 465 struct perf_data_file *file = session->file; 466 467 signal(SIGINT, sig_handler); 468 469 if (rep->cpu_list) { 470 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 471 rep->cpu_bitmap); 472 if (ret) 473 return ret; 474 } 475 476 if (rep->show_threads) 477 perf_read_values_init(&rep->show_threads_values); 478 479 ret = report__setup_sample_type(rep); 480 if (ret) 481 return ret; 482 483 ret = perf_session__process_events(session, &rep->tool); 484 if (ret) 485 return ret; 486 487 report__warn_kptr_restrict(rep); 488 489 evlist__for_each(session->evlist, pos) 490 rep->nr_entries += evsel__hists(pos)->nr_entries; 491 492 if (use_browser == 0) { 493 if (verbose > 3) 494 perf_session__fprintf(session, stdout); 495 496 if (verbose > 2) 497 perf_session__fprintf_dsos(session, stdout); 498 499 if (dump_trace) { 500 perf_session__fprintf_nr_events(session, stdout); 501 perf_evlist__fprintf_nr_events(session->evlist, stdout); 502 return 0; 503 } 504 } 505 506 report__collapse_hists(rep); 507 508 if (session_done()) 509 return 0; 510 511 /* 512 * recalculate number of entries after collapsing since it 513 * might be changed during the collapse phase. 514 */ 515 rep->nr_entries = 0; 516 evlist__for_each(session->evlist, pos) 517 rep->nr_entries += evsel__hists(pos)->nr_entries; 518 519 if (rep->nr_entries == 0) { 520 ui__error("The %s file has no samples!\n", file->path); 521 return 0; 522 } 523 524 report__output_resort(rep); 525 526 return report__browse_hists(rep); 527 } 528 529 static int 530 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 531 { 532 struct report *rep = (struct report *)opt->value; 533 534 /* 535 * --no-call-graph 536 */ 537 if (unset) { 538 rep->dont_use_callchains = true; 539 return 0; 540 } 541 542 return parse_callchain_report_opt(arg); 543 } 544 545 int 546 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 547 const char *arg, int unset __maybe_unused) 548 { 549 if (arg) { 550 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 551 if (err) { 552 char buf[BUFSIZ]; 553 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 554 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 555 return -1; 556 } 557 have_ignore_callees = 1; 558 } 559 560 return 0; 561 } 562 563 static int 564 parse_branch_mode(const struct option *opt __maybe_unused, 565 const char *str __maybe_unused, int unset) 566 { 567 int *branch_mode = opt->value; 568 569 *branch_mode = !unset; 570 return 0; 571 } 572 573 static int 574 parse_percent_limit(const struct option *opt, const char *str, 575 int unset __maybe_unused) 576 { 577 struct report *rep = opt->value; 578 579 rep->min_percent = strtof(str, NULL); 580 return 0; 581 } 582 583 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused) 584 { 585 struct perf_session *session; 586 struct stat st; 587 bool has_br_stack = false; 588 int branch_mode = -1; 589 bool branch_call_mode = false; 590 char callchain_default_opt[] = "fractal,0.5,callee"; 591 const char * const report_usage[] = { 592 "perf report [<options>]", 593 NULL 594 }; 595 struct report report = { 596 .tool = { 597 .sample = process_sample_event, 598 .mmap = perf_event__process_mmap, 599 .mmap2 = perf_event__process_mmap2, 600 .comm = perf_event__process_comm, 601 .exit = perf_event__process_exit, 602 .fork = perf_event__process_fork, 603 .lost = perf_event__process_lost, 604 .read = process_read_event, 605 .attr = perf_event__process_attr, 606 .tracing_data = perf_event__process_tracing_data, 607 .build_id = perf_event__process_build_id, 608 .ordered_events = true, 609 .ordering_requires_timestamps = true, 610 }, 611 .max_stack = PERF_MAX_STACK_DEPTH, 612 .pretty_printing_style = "normal", 613 }; 614 const struct option options[] = { 615 OPT_STRING('i', "input", &input_name, "file", 616 "input file name"), 617 OPT_INCR('v', "verbose", &verbose, 618 "be more verbose (show symbol address, etc)"), 619 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 620 "dump raw trace in ASCII"), 621 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 622 "file", "vmlinux pathname"), 623 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 624 "file", "kallsyms pathname"), 625 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"), 626 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 627 "load module symbols - WARNING: use only with -k and LIVE kernel"), 628 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 629 "Show a column with the number of samples"), 630 OPT_BOOLEAN('T', "threads", &report.show_threads, 631 "Show per-thread event counters"), 632 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 633 "pretty printing style key: normal raw"), 634 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 635 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 636 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 637 "Use the stdio interface"), 638 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 639 OPT_BOOLEAN(0, "header-only", &report.header_only, 640 "Show only data header."), 641 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 642 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 643 " Please refer the man page for the complete list."), 644 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 645 "output field(s): overhead, period, sample plus all of sort keys"), 646 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 647 "Show sample percentage for different cpu modes"), 648 OPT_STRING('p', "parent", &parent_pattern, "regex", 649 "regex filter to identify parent, see: '--sort parent'"), 650 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 651 "Only display entries with parent-match"), 652 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order[,branch]", 653 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address), add branches. " 654 "Default: fractal,0.5,callee,function", &report_parse_callchain_opt, callchain_default_opt), 655 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 656 "Accumulate callchains of children and show total overhead as well"), 657 OPT_INTEGER(0, "max-stack", &report.max_stack, 658 "Set the maximum stack depth when parsing the callchain, " 659 "anything beyond the specified depth will be ignored. " 660 "Default: " __stringify(PERF_MAX_STACK_DEPTH)), 661 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 662 "alias for inverted call graph"), 663 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 664 "ignore callees of these functions in call graphs", 665 report_parse_ignore_callees_opt), 666 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 667 "only consider symbols in these dsos"), 668 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 669 "only consider symbols in these comms"), 670 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 671 "only consider these symbols"), 672 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 673 "only show symbols that (partially) match with this filter"), 674 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 675 "width[,width...]", 676 "don't try to adjust column width, use these fixed values"), 677 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator", 678 "separator for columns, no spaces will be added between " 679 "columns '.' is reserved."), 680 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved, 681 "Only display entries resolved to a symbol"), 682 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 683 "Look for files with symbols relative to this directory"), 684 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 685 "list of cpus to profile"), 686 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 687 "Display extended information about perf.data file"), 688 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 689 "Interleave source code with assembly code (default)"), 690 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 691 "Display raw encoding of assembly instructions (default)"), 692 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 693 "Specify disassembler style (e.g. -M intel for intel syntax)"), 694 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 695 "Show a column with the sum of periods"), 696 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 697 "Show event group information together"), 698 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 699 "use branch records for per branch histogram filling", 700 parse_branch_mode), 701 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 702 "add last branch records to call history"), 703 OPT_STRING(0, "objdump", &objdump_path, "path", 704 "objdump binary to use for disassembly and annotations"), 705 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 706 "Disable symbol demangling"), 707 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 708 "Enable kernel symbol demangling"), 709 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 710 OPT_CALLBACK(0, "percent-limit", &report, "percent", 711 "Don't show entries under that percent", parse_percent_limit), 712 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 713 "how to display percentage of filtered entries", parse_filter_percentage), 714 OPT_END() 715 }; 716 struct perf_data_file file = { 717 .mode = PERF_DATA_MODE_READ, 718 }; 719 int ret = hists__init(); 720 721 if (ret < 0) 722 return ret; 723 724 perf_config(report__config, &report); 725 726 argc = parse_options(argc, argv, options, report_usage, 0); 727 728 if (report.use_stdio) 729 use_browser = 0; 730 else if (report.use_tui) 731 use_browser = 1; 732 else if (report.use_gtk) 733 use_browser = 2; 734 735 if (report.inverted_callchain) 736 callchain_param.order = ORDER_CALLER; 737 738 if (!input_name || !strlen(input_name)) { 739 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 740 input_name = "-"; 741 else 742 input_name = "perf.data"; 743 } 744 745 file.path = input_name; 746 file.force = report.force; 747 748 repeat: 749 session = perf_session__new(&file, false, &report.tool); 750 if (session == NULL) 751 return -1; 752 753 if (report.queue_size) { 754 ordered_events__set_alloc_size(&session->ordered_events, 755 report.queue_size); 756 } 757 758 report.session = session; 759 760 has_br_stack = perf_header__has_feat(&session->header, 761 HEADER_BRANCH_STACK); 762 763 /* 764 * Branch mode is a tristate: 765 * -1 means default, so decide based on the file having branch data. 766 * 0/1 means the user chose a mode. 767 */ 768 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 769 branch_call_mode == -1) { 770 sort__mode = SORT_MODE__BRANCH; 771 symbol_conf.cumulate_callchain = false; 772 } 773 if (branch_call_mode) { 774 callchain_param.key = CCKEY_ADDRESS; 775 callchain_param.branch_callstack = 1; 776 symbol_conf.use_callchain = true; 777 callchain_register_param(&callchain_param); 778 if (sort_order == NULL) 779 sort_order = "srcline,symbol,dso"; 780 } 781 782 if (report.mem_mode) { 783 if (sort__mode == SORT_MODE__BRANCH) { 784 pr_err("branch and mem mode incompatible\n"); 785 goto error; 786 } 787 sort__mode = SORT_MODE__MEMORY; 788 symbol_conf.cumulate_callchain = false; 789 } 790 791 if (setup_sorting() < 0) { 792 if (sort_order) 793 parse_options_usage(report_usage, options, "s", 1); 794 if (field_order) 795 parse_options_usage(sort_order ? NULL : report_usage, 796 options, "F", 1); 797 goto error; 798 } 799 800 /* Force tty output for header output. */ 801 if (report.header || report.header_only) 802 use_browser = 0; 803 804 if (strcmp(input_name, "-") != 0) 805 setup_browser(true); 806 else 807 use_browser = 0; 808 809 if (report.header || report.header_only) { 810 perf_session__fprintf_info(session, stdout, 811 report.show_full_info); 812 if (report.header_only) 813 return 0; 814 } else if (use_browser == 0) { 815 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 816 stdout); 817 } 818 819 /* 820 * Only in the TUI browser we are doing integrated annotation, 821 * so don't allocate extra space that won't be used in the stdio 822 * implementation. 823 */ 824 if (ui__has_annotation()) { 825 symbol_conf.priv_size = sizeof(struct annotation); 826 machines__set_symbol_filter(&session->machines, 827 symbol__annotate_init); 828 /* 829 * For searching by name on the "Browse map details". 830 * providing it only in verbose mode not to bloat too 831 * much struct symbol. 832 */ 833 if (verbose) { 834 /* 835 * XXX: Need to provide a less kludgy way to ask for 836 * more space per symbol, the u32 is for the index on 837 * the ui browser. 838 * See symbol__browser_index. 839 */ 840 symbol_conf.priv_size += sizeof(u32); 841 symbol_conf.sort_by_name = true; 842 } 843 } 844 845 if (symbol__init(&session->header.env) < 0) 846 goto error; 847 848 if (argc) { 849 /* 850 * Special case: if there's an argument left then assume that 851 * it's a symbol filter: 852 */ 853 if (argc > 1) 854 usage_with_options(report_usage, options); 855 856 report.symbol_filter_str = argv[0]; 857 } 858 859 sort__setup_elide(stdout); 860 861 ret = __cmd_report(&report); 862 if (ret == K_SWITCH_INPUT_DATA) { 863 perf_session__delete(session); 864 goto repeat; 865 } else 866 ret = 0; 867 868 error: 869 perf_session__delete(session); 870 return ret; 871 } 872