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/util.h" 12 #include "util/config.h" 13 14 #include "util/annotate.h" 15 #include "util/color.h" 16 #include <linux/list.h> 17 #include <linux/rbtree.h> 18 #include <linux/err.h> 19 #include "util/symbol.h" 20 #include "util/callchain.h" 21 #include "util/values.h" 22 23 #include "perf.h" 24 #include "util/debug.h" 25 #include "util/evlist.h" 26 #include "util/evsel.h" 27 #include "util/header.h" 28 #include "util/session.h" 29 #include "util/tool.h" 30 31 #include <subcmd/parse-options.h> 32 #include <subcmd/exec-cmd.h> 33 #include "util/parse-events.h" 34 35 #include "util/thread.h" 36 #include "util/sort.h" 37 #include "util/hist.h" 38 #include "util/data.h" 39 #include "arch/common.h" 40 #include "util/time-utils.h" 41 #include "util/auxtrace.h" 42 #include "util/units.h" 43 #include "util/branch.h" 44 45 #include <dlfcn.h> 46 #include <errno.h> 47 #include <inttypes.h> 48 #include <regex.h> 49 #include <signal.h> 50 #include <linux/bitmap.h> 51 #include <linux/stringify.h> 52 #include <sys/types.h> 53 #include <sys/stat.h> 54 #include <unistd.h> 55 #include <linux/mman.h> 56 57 struct report { 58 struct perf_tool tool; 59 struct perf_session *session; 60 bool use_tui, use_gtk, use_stdio; 61 bool show_full_info; 62 bool show_threads; 63 bool inverted_callchain; 64 bool mem_mode; 65 bool stats_mode; 66 bool tasks_mode; 67 bool mmaps_mode; 68 bool header; 69 bool header_only; 70 bool nonany_branch_mode; 71 int max_stack; 72 struct perf_read_values show_threads_values; 73 const char *pretty_printing_style; 74 const char *cpu_list; 75 const char *symbol_filter_str; 76 const char *time_str; 77 struct perf_time_interval *ptime_range; 78 int range_size; 79 int range_num; 80 float min_percent; 81 u64 nr_entries; 82 u64 queue_size; 83 int socket_filter; 84 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 85 struct branch_type_stat brtype_stat; 86 }; 87 88 static int report__config(const char *var, const char *value, void *cb) 89 { 90 struct report *rep = cb; 91 92 if (!strcmp(var, "report.group")) { 93 symbol_conf.event_group = perf_config_bool(var, value); 94 return 0; 95 } 96 if (!strcmp(var, "report.percent-limit")) { 97 double pcnt = strtof(value, NULL); 98 99 rep->min_percent = pcnt; 100 callchain_param.min_percent = pcnt; 101 return 0; 102 } 103 if (!strcmp(var, "report.children")) { 104 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 105 return 0; 106 } 107 if (!strcmp(var, "report.queue-size")) 108 return perf_config_u64(&rep->queue_size, var, value); 109 110 if (!strcmp(var, "report.sort_order")) { 111 default_sort_order = strdup(value); 112 return 0; 113 } 114 115 return 0; 116 } 117 118 static int hist_iter__report_callback(struct hist_entry_iter *iter, 119 struct addr_location *al, bool single, 120 void *arg) 121 { 122 int err = 0; 123 struct report *rep = arg; 124 struct hist_entry *he = iter->he; 125 struct perf_evsel *evsel = iter->evsel; 126 struct perf_sample *sample = iter->sample; 127 struct mem_info *mi; 128 struct branch_info *bi; 129 130 if (!ui__has_annotation()) 131 return 0; 132 133 hist__account_cycles(sample->branch_stack, al, sample, 134 rep->nonany_branch_mode); 135 136 if (sort__mode == SORT_MODE__BRANCH) { 137 bi = he->branch_info; 138 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel->idx); 139 if (err) 140 goto out; 141 142 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel->idx); 143 144 } else if (rep->mem_mode) { 145 mi = he->mem_info; 146 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel->idx); 147 if (err) 148 goto out; 149 150 err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr); 151 152 } else if (symbol_conf.cumulate_callchain) { 153 if (single) 154 err = hist_entry__inc_addr_samples(he, sample, evsel->idx, 155 al->addr); 156 } else { 157 err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr); 158 } 159 160 out: 161 return err; 162 } 163 164 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 165 struct addr_location *al __maybe_unused, 166 bool single __maybe_unused, 167 void *arg) 168 { 169 struct hist_entry *he = iter->he; 170 struct report *rep = arg; 171 struct branch_info *bi; 172 struct perf_sample *sample = iter->sample; 173 struct perf_evsel *evsel = iter->evsel; 174 int err; 175 176 if (!ui__has_annotation()) 177 return 0; 178 179 hist__account_cycles(sample->branch_stack, al, sample, 180 rep->nonany_branch_mode); 181 182 bi = he->branch_info; 183 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel->idx); 184 if (err) 185 goto out; 186 187 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel->idx); 188 189 branch_type_count(&rep->brtype_stat, &bi->flags, 190 bi->from.addr, bi->to.addr); 191 192 out: 193 return err; 194 } 195 196 static int process_sample_event(struct perf_tool *tool, 197 union perf_event *event, 198 struct perf_sample *sample, 199 struct perf_evsel *evsel, 200 struct machine *machine) 201 { 202 struct report *rep = container_of(tool, struct report, tool); 203 struct addr_location al; 204 struct hist_entry_iter iter = { 205 .evsel = evsel, 206 .sample = sample, 207 .hide_unresolved = symbol_conf.hide_unresolved, 208 .add_entry_cb = hist_iter__report_callback, 209 }; 210 int ret = 0; 211 212 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 213 sample->time)) { 214 return 0; 215 } 216 217 if (machine__resolve(machine, &al, sample) < 0) { 218 pr_debug("problem processing %d event, skipping it.\n", 219 event->header.type); 220 return -1; 221 } 222 223 if (symbol_conf.hide_unresolved && al.sym == NULL) 224 goto out_put; 225 226 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 227 goto out_put; 228 229 if (sort__mode == SORT_MODE__BRANCH) { 230 /* 231 * A non-synthesized event might not have a branch stack if 232 * branch stacks have been synthesized (using itrace options). 233 */ 234 if (!sample->branch_stack) 235 goto out_put; 236 237 iter.add_entry_cb = hist_iter__branch_callback; 238 iter.ops = &hist_iter_branch; 239 } else if (rep->mem_mode) { 240 iter.ops = &hist_iter_mem; 241 } else if (symbol_conf.cumulate_callchain) { 242 iter.ops = &hist_iter_cumulative; 243 } else { 244 iter.ops = &hist_iter_normal; 245 } 246 247 if (al.map != NULL) 248 al.map->dso->hit = 1; 249 250 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 251 if (ret < 0) 252 pr_debug("problem adding hist entry, skipping event\n"); 253 out_put: 254 addr_location__put(&al); 255 return ret; 256 } 257 258 static int process_read_event(struct perf_tool *tool, 259 union perf_event *event, 260 struct perf_sample *sample __maybe_unused, 261 struct perf_evsel *evsel, 262 struct machine *machine __maybe_unused) 263 { 264 struct report *rep = container_of(tool, struct report, tool); 265 266 if (rep->show_threads) { 267 const char *name = evsel ? perf_evsel__name(evsel) : "unknown"; 268 int err = perf_read_values_add_value(&rep->show_threads_values, 269 event->read.pid, event->read.tid, 270 evsel->idx, 271 name, 272 event->read.value); 273 274 if (err) 275 return err; 276 } 277 278 return 0; 279 } 280 281 /* For pipe mode, sample_type is not currently set */ 282 static int report__setup_sample_type(struct report *rep) 283 { 284 struct perf_session *session = rep->session; 285 u64 sample_type = perf_evlist__combined_sample_type(session->evlist); 286 bool is_pipe = perf_data__is_pipe(session->data); 287 288 if (session->itrace_synth_opts->callchain || 289 (!is_pipe && 290 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 291 !session->itrace_synth_opts->set)) 292 sample_type |= PERF_SAMPLE_CALLCHAIN; 293 294 if (session->itrace_synth_opts->last_branch) 295 sample_type |= PERF_SAMPLE_BRANCH_STACK; 296 297 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 298 if (perf_hpp_list.parent) { 299 ui__error("Selected --sort parent, but no " 300 "callchain data. Did you call " 301 "'perf record' without -g?\n"); 302 return -EINVAL; 303 } 304 if (symbol_conf.use_callchain && 305 !symbol_conf.show_branchflag_count) { 306 ui__error("Selected -g or --branch-history.\n" 307 "But no callchain or branch data.\n" 308 "Did you call 'perf record' without -g or -b?\n"); 309 return -1; 310 } 311 } else if (!callchain_param.enabled && 312 callchain_param.mode != CHAIN_NONE && 313 !symbol_conf.use_callchain) { 314 symbol_conf.use_callchain = true; 315 if (callchain_register_param(&callchain_param) < 0) { 316 ui__error("Can't register callchain params.\n"); 317 return -EINVAL; 318 } 319 } 320 321 if (symbol_conf.cumulate_callchain) { 322 /* Silently ignore if callchain is missing */ 323 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 324 symbol_conf.cumulate_callchain = false; 325 perf_hpp__cancel_cumulate(); 326 } 327 } 328 329 if (sort__mode == SORT_MODE__BRANCH) { 330 if (!is_pipe && 331 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 332 ui__error("Selected -b but no branch data. " 333 "Did you call perf record without -b?\n"); 334 return -1; 335 } 336 } 337 338 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) { 339 if ((sample_type & PERF_SAMPLE_REGS_USER) && 340 (sample_type & PERF_SAMPLE_STACK_USER)) { 341 callchain_param.record_mode = CALLCHAIN_DWARF; 342 dwarf_callchain_users = true; 343 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK) 344 callchain_param.record_mode = CALLCHAIN_LBR; 345 else 346 callchain_param.record_mode = CALLCHAIN_FP; 347 } 348 349 /* ??? handle more cases than just ANY? */ 350 if (!(perf_evlist__combined_branch_type(session->evlist) & 351 PERF_SAMPLE_BRANCH_ANY)) 352 rep->nonany_branch_mode = true; 353 354 return 0; 355 } 356 357 static void sig_handler(int sig __maybe_unused) 358 { 359 session_done = 1; 360 } 361 362 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 363 const char *evname, FILE *fp) 364 { 365 size_t ret; 366 char unit; 367 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE]; 368 u64 nr_events = hists->stats.total_period; 369 struct perf_evsel *evsel = hists_to_evsel(hists); 370 char buf[512]; 371 size_t size = sizeof(buf); 372 int socked_id = hists->socket_filter; 373 374 if (quiet) 375 return 0; 376 377 if (symbol_conf.filter_relative) { 378 nr_samples = hists->stats.nr_non_filtered_samples; 379 nr_events = hists->stats.total_non_filtered_period; 380 } 381 382 if (perf_evsel__is_group_event(evsel)) { 383 struct perf_evsel *pos; 384 385 perf_evsel__group_desc(evsel, buf, size); 386 evname = buf; 387 388 for_each_group_member(pos, evsel) { 389 const struct hists *pos_hists = evsel__hists(pos); 390 391 if (symbol_conf.filter_relative) { 392 nr_samples += pos_hists->stats.nr_non_filtered_samples; 393 nr_events += pos_hists->stats.total_non_filtered_period; 394 } else { 395 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE]; 396 nr_events += pos_hists->stats.total_period; 397 } 398 } 399 } 400 401 nr_samples = convert_unit(nr_samples, &unit); 402 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 403 if (evname != NULL) 404 ret += fprintf(fp, " of event '%s'", evname); 405 406 if (rep->time_str) 407 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 408 409 if (symbol_conf.show_ref_callgraph && 410 strstr(evname, "call-graph=no")) { 411 ret += fprintf(fp, ", show reference callgraph"); 412 } 413 414 if (rep->mem_mode) { 415 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 416 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 417 } else 418 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 419 420 if (socked_id > -1) 421 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 422 423 return ret + fprintf(fp, "\n#\n"); 424 } 425 426 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 427 struct report *rep, 428 const char *help) 429 { 430 struct perf_evsel *pos; 431 432 if (!quiet) { 433 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 434 evlist->stats.total_lost_samples); 435 } 436 437 evlist__for_each_entry(evlist, pos) { 438 struct hists *hists = evsel__hists(pos); 439 const char *evname = perf_evsel__name(pos); 440 441 if (symbol_conf.event_group && 442 !perf_evsel__is_group_leader(pos)) 443 continue; 444 445 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 446 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 447 symbol_conf.use_callchain || 448 symbol_conf.show_branchflag_count); 449 fprintf(stdout, "\n\n"); 450 } 451 452 if (!quiet) 453 fprintf(stdout, "#\n# (%s)\n#\n", help); 454 455 if (rep->show_threads) { 456 bool style = !strcmp(rep->pretty_printing_style, "raw"); 457 perf_read_values_display(stdout, &rep->show_threads_values, 458 style); 459 perf_read_values_destroy(&rep->show_threads_values); 460 } 461 462 if (sort__mode == SORT_MODE__BRANCH) 463 branch_type_stat_display(stdout, &rep->brtype_stat); 464 465 return 0; 466 } 467 468 static void report__warn_kptr_restrict(const struct report *rep) 469 { 470 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 471 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 472 473 if (perf_evlist__exclude_kernel(rep->session->evlist)) 474 return; 475 476 if (kernel_map == NULL || 477 (kernel_map->dso->hit && 478 (kernel_kmap->ref_reloc_sym == NULL || 479 kernel_kmap->ref_reloc_sym->addr == 0))) { 480 const char *desc = 481 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 482 "can't be resolved."; 483 484 if (kernel_map) { 485 const struct dso *kdso = kernel_map->dso; 486 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) { 487 desc = "If some relocation was applied (e.g. " 488 "kexec) symbols may be misresolved."; 489 } 490 } 491 492 ui__warning( 493 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 494 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 495 "Samples in kernel modules can't be resolved as well.\n\n", 496 desc); 497 } 498 } 499 500 static int report__gtk_browse_hists(struct report *rep, const char *help) 501 { 502 int (*hist_browser)(struct perf_evlist *evlist, const char *help, 503 struct hist_browser_timer *timer, float min_pcnt); 504 505 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists"); 506 507 if (hist_browser == NULL) { 508 ui__error("GTK browser not found!\n"); 509 return -1; 510 } 511 512 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 513 } 514 515 static int report__browse_hists(struct report *rep) 516 { 517 int ret; 518 struct perf_session *session = rep->session; 519 struct perf_evlist *evlist = session->evlist; 520 const char *help = perf_tip(system_path(TIPDIR)); 521 522 if (help == NULL) { 523 /* fallback for people who don't install perf ;-) */ 524 help = perf_tip(DOCDIR); 525 if (help == NULL) 526 help = "Cannot load tips.txt file, please install perf!"; 527 } 528 529 switch (use_browser) { 530 case 1: 531 ret = perf_evlist__tui_browse_hists(evlist, help, NULL, 532 rep->min_percent, 533 &session->header.env); 534 /* 535 * Usually "ret" is the last pressed key, and we only 536 * care if the key notifies us to switch data file. 537 */ 538 if (ret != K_SWITCH_INPUT_DATA) 539 ret = 0; 540 break; 541 case 2: 542 ret = report__gtk_browse_hists(rep, help); 543 break; 544 default: 545 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 546 break; 547 } 548 549 return ret; 550 } 551 552 static int report__collapse_hists(struct report *rep) 553 { 554 struct ui_progress prog; 555 struct perf_evsel *pos; 556 int ret = 0; 557 558 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 559 560 evlist__for_each_entry(rep->session->evlist, pos) { 561 struct hists *hists = evsel__hists(pos); 562 563 if (pos->idx == 0) 564 hists->symbol_filter_str = rep->symbol_filter_str; 565 566 hists->socket_filter = rep->socket_filter; 567 568 ret = hists__collapse_resort(hists, &prog); 569 if (ret < 0) 570 break; 571 572 /* Non-group events are considered as leader */ 573 if (symbol_conf.event_group && 574 !perf_evsel__is_group_leader(pos)) { 575 struct hists *leader_hists = evsel__hists(pos->leader); 576 577 hists__match(leader_hists, hists); 578 hists__link(leader_hists, hists); 579 } 580 } 581 582 ui_progress__finish(); 583 return ret; 584 } 585 586 static void report__output_resort(struct report *rep) 587 { 588 struct ui_progress prog; 589 struct perf_evsel *pos; 590 591 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 592 593 evlist__for_each_entry(rep->session->evlist, pos) 594 perf_evsel__output_resort(pos, &prog); 595 596 ui_progress__finish(); 597 } 598 599 static void stats_setup(struct report *rep) 600 { 601 memset(&rep->tool, 0, sizeof(rep->tool)); 602 rep->tool.no_warn = true; 603 } 604 605 static int stats_print(struct report *rep) 606 { 607 struct perf_session *session = rep->session; 608 609 perf_session__fprintf_nr_events(session, stdout); 610 return 0; 611 } 612 613 static void tasks_setup(struct report *rep) 614 { 615 memset(&rep->tool, 0, sizeof(rep->tool)); 616 if (rep->mmaps_mode) { 617 rep->tool.mmap = perf_event__process_mmap; 618 rep->tool.mmap2 = perf_event__process_mmap2; 619 } 620 rep->tool.comm = perf_event__process_comm; 621 rep->tool.exit = perf_event__process_exit; 622 rep->tool.fork = perf_event__process_fork; 623 rep->tool.no_warn = true; 624 } 625 626 struct task { 627 struct thread *thread; 628 struct list_head list; 629 struct list_head children; 630 }; 631 632 static struct task *tasks_list(struct task *task, struct machine *machine) 633 { 634 struct thread *parent_thread, *thread = task->thread; 635 struct task *parent_task; 636 637 /* Already listed. */ 638 if (!list_empty(&task->list)) 639 return NULL; 640 641 /* Last one in the chain. */ 642 if (thread->ppid == -1) 643 return task; 644 645 parent_thread = machine__find_thread(machine, -1, thread->ppid); 646 if (!parent_thread) 647 return ERR_PTR(-ENOENT); 648 649 parent_task = thread__priv(parent_thread); 650 list_add_tail(&task->list, &parent_task->children); 651 return tasks_list(parent_task, machine); 652 } 653 654 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 655 { 656 size_t printed = 0; 657 struct rb_node *nd; 658 659 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) { 660 struct map *map = rb_entry(nd, struct map, rb_node); 661 662 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 663 indent, "", map->start, map->end, 664 map->prot & PROT_READ ? 'r' : '-', 665 map->prot & PROT_WRITE ? 'w' : '-', 666 map->prot & PROT_EXEC ? 'x' : '-', 667 map->flags & MAP_SHARED ? 's' : 'p', 668 map->pgoff, 669 map->ino, map->dso->name); 670 } 671 672 return printed; 673 } 674 675 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp) 676 { 677 int printed = 0, i; 678 for (i = 0; i < MAP__NR_TYPES; ++i) 679 printed += maps__fprintf_task(&mg->maps[i], indent, fp); 680 return printed; 681 } 682 683 static void task__print_level(struct task *task, FILE *fp, int level) 684 { 685 struct thread *thread = task->thread; 686 struct task *child; 687 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 688 thread->pid_, thread->tid, thread->ppid, 689 level, ""); 690 691 fprintf(fp, "%s\n", thread__comm_str(thread)); 692 693 map_groups__fprintf_task(thread->mg, comm_indent, fp); 694 695 if (!list_empty(&task->children)) { 696 list_for_each_entry(child, &task->children, list) 697 task__print_level(child, fp, level + 1); 698 } 699 } 700 701 static int tasks_print(struct report *rep, FILE *fp) 702 { 703 struct perf_session *session = rep->session; 704 struct machine *machine = &session->machines.host; 705 struct task *tasks, *task; 706 unsigned int nr = 0, itask = 0, i; 707 struct rb_node *nd; 708 LIST_HEAD(list); 709 710 /* 711 * No locking needed while accessing machine->threads, 712 * because --tasks is single threaded command. 713 */ 714 715 /* Count all the threads. */ 716 for (i = 0; i < THREADS__TABLE_SIZE; i++) 717 nr += machine->threads[i].nr; 718 719 tasks = malloc(sizeof(*tasks) * nr); 720 if (!tasks) 721 return -ENOMEM; 722 723 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 724 struct threads *threads = &machine->threads[i]; 725 726 for (nd = rb_first(&threads->entries); nd; nd = rb_next(nd)) { 727 task = tasks + itask++; 728 729 task->thread = rb_entry(nd, struct thread, rb_node); 730 INIT_LIST_HEAD(&task->children); 731 INIT_LIST_HEAD(&task->list); 732 thread__set_priv(task->thread, task); 733 } 734 } 735 736 /* 737 * Iterate every task down to the unprocessed parent 738 * and link all in task children list. Task with no 739 * parent is added into 'list'. 740 */ 741 for (itask = 0; itask < nr; itask++) { 742 task = tasks + itask; 743 744 if (!list_empty(&task->list)) 745 continue; 746 747 task = tasks_list(task, machine); 748 if (IS_ERR(task)) { 749 pr_err("Error: failed to process tasks\n"); 750 free(tasks); 751 return PTR_ERR(task); 752 } 753 754 if (task) 755 list_add_tail(&task->list, &list); 756 } 757 758 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 759 760 list_for_each_entry(task, &list, list) 761 task__print_level(task, fp, 0); 762 763 free(tasks); 764 return 0; 765 } 766 767 static int __cmd_report(struct report *rep) 768 { 769 int ret; 770 struct perf_session *session = rep->session; 771 struct perf_evsel *pos; 772 struct perf_data *data = session->data; 773 774 signal(SIGINT, sig_handler); 775 776 if (rep->cpu_list) { 777 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 778 rep->cpu_bitmap); 779 if (ret) { 780 ui__error("failed to set cpu bitmap\n"); 781 return ret; 782 } 783 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 784 } 785 786 if (rep->show_threads) { 787 ret = perf_read_values_init(&rep->show_threads_values); 788 if (ret) 789 return ret; 790 } 791 792 ret = report__setup_sample_type(rep); 793 if (ret) { 794 /* report__setup_sample_type() already showed error message */ 795 return ret; 796 } 797 798 if (rep->stats_mode) 799 stats_setup(rep); 800 801 if (rep->tasks_mode) 802 tasks_setup(rep); 803 804 ret = perf_session__process_events(session); 805 if (ret) { 806 ui__error("failed to process sample\n"); 807 return ret; 808 } 809 810 if (rep->stats_mode) 811 return stats_print(rep); 812 813 if (rep->tasks_mode) 814 return tasks_print(rep, stdout); 815 816 report__warn_kptr_restrict(rep); 817 818 evlist__for_each_entry(session->evlist, pos) 819 rep->nr_entries += evsel__hists(pos)->nr_entries; 820 821 if (use_browser == 0) { 822 if (verbose > 3) 823 perf_session__fprintf(session, stdout); 824 825 if (verbose > 2) 826 perf_session__fprintf_dsos(session, stdout); 827 828 if (dump_trace) { 829 perf_session__fprintf_nr_events(session, stdout); 830 perf_evlist__fprintf_nr_events(session->evlist, stdout); 831 return 0; 832 } 833 } 834 835 ret = report__collapse_hists(rep); 836 if (ret) { 837 ui__error("failed to process hist entry\n"); 838 return ret; 839 } 840 841 if (session_done()) 842 return 0; 843 844 /* 845 * recalculate number of entries after collapsing since it 846 * might be changed during the collapse phase. 847 */ 848 rep->nr_entries = 0; 849 evlist__for_each_entry(session->evlist, pos) 850 rep->nr_entries += evsel__hists(pos)->nr_entries; 851 852 if (rep->nr_entries == 0) { 853 ui__error("The %s file has no samples!\n", data->file.path); 854 return 0; 855 } 856 857 report__output_resort(rep); 858 859 return report__browse_hists(rep); 860 } 861 862 static int 863 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 864 { 865 struct callchain_param *callchain = opt->value; 866 867 callchain->enabled = !unset; 868 /* 869 * --no-call-graph 870 */ 871 if (unset) { 872 symbol_conf.use_callchain = false; 873 callchain->mode = CHAIN_NONE; 874 return 0; 875 } 876 877 return parse_callchain_report_opt(arg); 878 } 879 880 int 881 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 882 const char *arg, int unset __maybe_unused) 883 { 884 if (arg) { 885 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 886 if (err) { 887 char buf[BUFSIZ]; 888 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 889 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 890 return -1; 891 } 892 have_ignore_callees = 1; 893 } 894 895 return 0; 896 } 897 898 static int 899 parse_branch_mode(const struct option *opt, 900 const char *str __maybe_unused, int unset) 901 { 902 int *branch_mode = opt->value; 903 904 *branch_mode = !unset; 905 return 0; 906 } 907 908 static int 909 parse_percent_limit(const struct option *opt, const char *str, 910 int unset __maybe_unused) 911 { 912 struct report *rep = opt->value; 913 double pcnt = strtof(str, NULL); 914 915 rep->min_percent = pcnt; 916 callchain_param.min_percent = pcnt; 917 return 0; 918 } 919 920 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 921 922 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 923 CALLCHAIN_REPORT_HELP 924 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 925 926 int cmd_report(int argc, const char **argv) 927 { 928 struct perf_session *session; 929 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 930 struct stat st; 931 bool has_br_stack = false; 932 int branch_mode = -1; 933 bool branch_call_mode = false; 934 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 935 const char * const report_usage[] = { 936 "perf report [<options>]", 937 NULL 938 }; 939 struct report report = { 940 .tool = { 941 .sample = process_sample_event, 942 .mmap = perf_event__process_mmap, 943 .mmap2 = perf_event__process_mmap2, 944 .comm = perf_event__process_comm, 945 .namespaces = perf_event__process_namespaces, 946 .exit = perf_event__process_exit, 947 .fork = perf_event__process_fork, 948 .lost = perf_event__process_lost, 949 .read = process_read_event, 950 .attr = perf_event__process_attr, 951 .tracing_data = perf_event__process_tracing_data, 952 .build_id = perf_event__process_build_id, 953 .id_index = perf_event__process_id_index, 954 .auxtrace_info = perf_event__process_auxtrace_info, 955 .auxtrace = perf_event__process_auxtrace, 956 .feature = perf_event__process_feature, 957 .ordered_events = true, 958 .ordering_requires_timestamps = true, 959 }, 960 .max_stack = PERF_MAX_STACK_DEPTH, 961 .pretty_printing_style = "normal", 962 .socket_filter = -1, 963 }; 964 const struct option options[] = { 965 OPT_STRING('i', "input", &input_name, "file", 966 "input file name"), 967 OPT_INCR('v', "verbose", &verbose, 968 "be more verbose (show symbol address, etc)"), 969 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"), 970 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 971 "dump raw trace in ASCII"), 972 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 973 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 974 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 975 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 976 "file", "vmlinux pathname"), 977 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 978 "file", "kallsyms pathname"), 979 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 980 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 981 "load module symbols - WARNING: use only with -k and LIVE kernel"), 982 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 983 "Show a column with the number of samples"), 984 OPT_BOOLEAN('T', "threads", &report.show_threads, 985 "Show per-thread event counters"), 986 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 987 "pretty printing style key: normal raw"), 988 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 989 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 990 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 991 "Use the stdio interface"), 992 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 993 OPT_BOOLEAN(0, "header-only", &report.header_only, 994 "Show only data header."), 995 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 996 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 997 " Please refer the man page for the complete list."), 998 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 999 "output field(s): overhead, period, sample plus all of sort keys"), 1000 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1001 "Show sample percentage for different cpu modes"), 1002 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1003 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1004 OPT_STRING('p', "parent", &parent_pattern, "regex", 1005 "regex filter to identify parent, see: '--sort parent'"), 1006 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1007 "Only display entries with parent-match"), 1008 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1009 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1010 report_callchain_help, &report_parse_callchain_opt, 1011 callchain_default_opt), 1012 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1013 "Accumulate callchains of children and show total overhead as well"), 1014 OPT_INTEGER(0, "max-stack", &report.max_stack, 1015 "Set the maximum stack depth when parsing the callchain, " 1016 "anything beyond the specified depth will be ignored. " 1017 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1018 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1019 "alias for inverted call graph"), 1020 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1021 "ignore callees of these functions in call graphs", 1022 report_parse_ignore_callees_opt), 1023 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1024 "only consider symbols in these dsos"), 1025 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1026 "only consider symbols in these comms"), 1027 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1028 "only consider symbols in these pids"), 1029 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1030 "only consider symbols in these tids"), 1031 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1032 "only consider these symbols"), 1033 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1034 "only show symbols that (partially) match with this filter"), 1035 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1036 "width[,width...]", 1037 "don't try to adjust column width, use these fixed values"), 1038 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1039 "separator for columns, no spaces will be added between " 1040 "columns '.' is reserved."), 1041 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1042 "Only display entries resolved to a symbol"), 1043 OPT_CALLBACK(0, "symfs", NULL, "directory", 1044 "Look for files with symbols relative to this directory", 1045 symbol__config_symfs), 1046 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1047 "list of cpus to profile"), 1048 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1049 "Display extended information about perf.data file"), 1050 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 1051 "Interleave source code with assembly code (default)"), 1052 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 1053 "Display raw encoding of assembly instructions (default)"), 1054 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1055 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1056 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1057 "Show a column with the sum of periods"), 1058 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 1059 "Show event group information together"), 1060 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1061 "use branch records for per branch histogram filling", 1062 parse_branch_mode), 1063 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1064 "add last branch records to call history"), 1065 OPT_STRING(0, "objdump", &objdump_path, "path", 1066 "objdump binary to use for disassembly and annotations"), 1067 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1068 "Disable symbol demangling"), 1069 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1070 "Enable kernel symbol demangling"), 1071 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1072 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1073 "Don't show entries under that percent", parse_percent_limit), 1074 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1075 "how to display percentage of filtered entries", parse_filter_percentage), 1076 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1077 "Instruction Tracing options", 1078 itrace_parse_synth_opts), 1079 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1080 "Show full source file name path for source lines"), 1081 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1082 "Show callgraph from reference event"), 1083 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1084 "only show processor socket that match with this filter"), 1085 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1086 "Show raw trace event output (do not use print fmt or plugins)"), 1087 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1088 "Show entries in a hierarchy"), 1089 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1090 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1091 stdio__config_color, "always"), 1092 OPT_STRING(0, "time", &report.time_str, "str", 1093 "Time span of interest (start,stop)"), 1094 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1095 "Show inline function"), 1096 OPT_END() 1097 }; 1098 struct perf_data data = { 1099 .mode = PERF_DATA_MODE_READ, 1100 }; 1101 int ret = hists__init(); 1102 1103 if (ret < 0) 1104 return ret; 1105 1106 ret = perf_config(report__config, &report); 1107 if (ret) 1108 return ret; 1109 1110 argc = parse_options(argc, argv, options, report_usage, 0); 1111 if (argc) { 1112 /* 1113 * Special case: if there's an argument left then assume that 1114 * it's a symbol filter: 1115 */ 1116 if (argc > 1) 1117 usage_with_options(report_usage, options); 1118 1119 report.symbol_filter_str = argv[0]; 1120 } 1121 1122 if (report.mmaps_mode) 1123 report.tasks_mode = true; 1124 1125 if (quiet) 1126 perf_quiet_option(); 1127 1128 if (symbol_conf.vmlinux_name && 1129 access(symbol_conf.vmlinux_name, R_OK)) { 1130 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 1131 return -EINVAL; 1132 } 1133 if (symbol_conf.kallsyms_name && 1134 access(symbol_conf.kallsyms_name, R_OK)) { 1135 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 1136 return -EINVAL; 1137 } 1138 1139 if (report.inverted_callchain) 1140 callchain_param.order = ORDER_CALLER; 1141 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1142 callchain_param.order = ORDER_CALLER; 1143 1144 if (itrace_synth_opts.callchain && 1145 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1146 report.max_stack = itrace_synth_opts.callchain_sz; 1147 1148 if (!input_name || !strlen(input_name)) { 1149 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1150 input_name = "-"; 1151 else 1152 input_name = "perf.data"; 1153 } 1154 1155 data.file.path = input_name; 1156 data.force = symbol_conf.force; 1157 1158 repeat: 1159 session = perf_session__new(&data, false, &report.tool); 1160 if (session == NULL) 1161 return -1; 1162 1163 if (report.queue_size) { 1164 ordered_events__set_alloc_size(&session->ordered_events, 1165 report.queue_size); 1166 } 1167 1168 session->itrace_synth_opts = &itrace_synth_opts; 1169 1170 report.session = session; 1171 1172 has_br_stack = perf_header__has_feat(&session->header, 1173 HEADER_BRANCH_STACK); 1174 1175 if (itrace_synth_opts.last_branch) 1176 has_br_stack = true; 1177 1178 if (has_br_stack && branch_call_mode) 1179 symbol_conf.show_branchflag_count = true; 1180 1181 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1182 1183 /* 1184 * Branch mode is a tristate: 1185 * -1 means default, so decide based on the file having branch data. 1186 * 0/1 means the user chose a mode. 1187 */ 1188 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1189 !branch_call_mode) { 1190 sort__mode = SORT_MODE__BRANCH; 1191 symbol_conf.cumulate_callchain = false; 1192 } 1193 if (branch_call_mode) { 1194 callchain_param.key = CCKEY_ADDRESS; 1195 callchain_param.branch_callstack = 1; 1196 symbol_conf.use_callchain = true; 1197 callchain_register_param(&callchain_param); 1198 if (sort_order == NULL) 1199 sort_order = "srcline,symbol,dso"; 1200 } 1201 1202 if (report.mem_mode) { 1203 if (sort__mode == SORT_MODE__BRANCH) { 1204 pr_err("branch and mem mode incompatible\n"); 1205 goto error; 1206 } 1207 sort__mode = SORT_MODE__MEMORY; 1208 symbol_conf.cumulate_callchain = false; 1209 } 1210 1211 if (symbol_conf.report_hierarchy) { 1212 /* disable incompatible options */ 1213 symbol_conf.cumulate_callchain = false; 1214 1215 if (field_order) { 1216 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1217 parse_options_usage(report_usage, options, "F", 1); 1218 parse_options_usage(NULL, options, "hierarchy", 0); 1219 goto error; 1220 } 1221 1222 perf_hpp_list.need_collapse = true; 1223 } 1224 1225 if (report.use_stdio) 1226 use_browser = 0; 1227 else if (report.use_tui) 1228 use_browser = 1; 1229 else if (report.use_gtk) 1230 use_browser = 2; 1231 1232 /* Force tty output for header output and per-thread stat. */ 1233 if (report.header || report.header_only || report.show_threads) 1234 use_browser = 0; 1235 if (report.header || report.header_only) 1236 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1237 if (report.show_full_info) 1238 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1239 if (report.stats_mode || report.tasks_mode) 1240 use_browser = 0; 1241 if (report.stats_mode && report.tasks_mode) { 1242 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1243 goto error; 1244 } 1245 1246 if (strcmp(input_name, "-") != 0) 1247 setup_browser(true); 1248 else 1249 use_browser = 0; 1250 1251 if (setup_sorting(session->evlist) < 0) { 1252 if (sort_order) 1253 parse_options_usage(report_usage, options, "s", 1); 1254 if (field_order) 1255 parse_options_usage(sort_order ? NULL : report_usage, 1256 options, "F", 1); 1257 goto error; 1258 } 1259 1260 if ((report.header || report.header_only) && !quiet) { 1261 perf_session__fprintf_info(session, stdout, 1262 report.show_full_info); 1263 if (report.header_only) { 1264 ret = 0; 1265 goto error; 1266 } 1267 } else if (use_browser == 0 && !quiet && 1268 !report.stats_mode && !report.tasks_mode) { 1269 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1270 stdout); 1271 } 1272 1273 /* 1274 * Only in the TUI browser we are doing integrated annotation, 1275 * so don't allocate extra space that won't be used in the stdio 1276 * implementation. 1277 */ 1278 if (ui__has_annotation()) { 1279 ret = symbol__annotation_init(); 1280 if (ret < 0) 1281 goto error; 1282 /* 1283 * For searching by name on the "Browse map details". 1284 * providing it only in verbose mode not to bloat too 1285 * much struct symbol. 1286 */ 1287 if (verbose > 0) { 1288 /* 1289 * XXX: Need to provide a less kludgy way to ask for 1290 * more space per symbol, the u32 is for the index on 1291 * the ui browser. 1292 * See symbol__browser_index. 1293 */ 1294 symbol_conf.priv_size += sizeof(u32); 1295 symbol_conf.sort_by_name = true; 1296 } 1297 } 1298 1299 if (symbol__init(&session->header.env) < 0) 1300 goto error; 1301 1302 report.ptime_range = perf_time__range_alloc(report.time_str, 1303 &report.range_size); 1304 if (!report.ptime_range) { 1305 ret = -ENOMEM; 1306 goto error; 1307 } 1308 1309 if (perf_time__parse_str(report.ptime_range, report.time_str) != 0) { 1310 if (session->evlist->first_sample_time == 0 && 1311 session->evlist->last_sample_time == 0) { 1312 pr_err("HINT: no first/last sample time found in perf data.\n" 1313 "Please use latest perf binary to execute 'perf record'\n" 1314 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n"); 1315 ret = -EINVAL; 1316 goto error; 1317 } 1318 1319 report.range_num = perf_time__percent_parse_str( 1320 report.ptime_range, report.range_size, 1321 report.time_str, 1322 session->evlist->first_sample_time, 1323 session->evlist->last_sample_time); 1324 1325 if (report.range_num < 0) { 1326 pr_err("Invalid time string\n"); 1327 ret = -EINVAL; 1328 goto error; 1329 } 1330 } else { 1331 report.range_num = 1; 1332 } 1333 1334 sort__setup_elide(stdout); 1335 1336 ret = __cmd_report(&report); 1337 if (ret == K_SWITCH_INPUT_DATA) { 1338 perf_session__delete(session); 1339 goto repeat; 1340 } else 1341 ret = 0; 1342 1343 error: 1344 zfree(&report.ptime_range); 1345 1346 perf_session__delete(session); 1347 return ret; 1348 } 1349