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 true); 535 /* 536 * Usually "ret" is the last pressed key, and we only 537 * care if the key notifies us to switch data file. 538 */ 539 if (ret != K_SWITCH_INPUT_DATA) 540 ret = 0; 541 break; 542 case 2: 543 ret = report__gtk_browse_hists(rep, help); 544 break; 545 default: 546 ret = perf_evlist__tty_browse_hists(evlist, rep, help); 547 break; 548 } 549 550 return ret; 551 } 552 553 static int report__collapse_hists(struct report *rep) 554 { 555 struct ui_progress prog; 556 struct perf_evsel *pos; 557 int ret = 0; 558 559 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 560 561 evlist__for_each_entry(rep->session->evlist, pos) { 562 struct hists *hists = evsel__hists(pos); 563 564 if (pos->idx == 0) 565 hists->symbol_filter_str = rep->symbol_filter_str; 566 567 hists->socket_filter = rep->socket_filter; 568 569 ret = hists__collapse_resort(hists, &prog); 570 if (ret < 0) 571 break; 572 573 /* Non-group events are considered as leader */ 574 if (symbol_conf.event_group && 575 !perf_evsel__is_group_leader(pos)) { 576 struct hists *leader_hists = evsel__hists(pos->leader); 577 578 hists__match(leader_hists, hists); 579 hists__link(leader_hists, hists); 580 } 581 } 582 583 ui_progress__finish(); 584 return ret; 585 } 586 587 static void report__output_resort(struct report *rep) 588 { 589 struct ui_progress prog; 590 struct perf_evsel *pos; 591 592 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 593 594 evlist__for_each_entry(rep->session->evlist, pos) 595 perf_evsel__output_resort(pos, &prog); 596 597 ui_progress__finish(); 598 } 599 600 static void stats_setup(struct report *rep) 601 { 602 memset(&rep->tool, 0, sizeof(rep->tool)); 603 rep->tool.no_warn = true; 604 } 605 606 static int stats_print(struct report *rep) 607 { 608 struct perf_session *session = rep->session; 609 610 perf_session__fprintf_nr_events(session, stdout); 611 return 0; 612 } 613 614 static void tasks_setup(struct report *rep) 615 { 616 memset(&rep->tool, 0, sizeof(rep->tool)); 617 if (rep->mmaps_mode) { 618 rep->tool.mmap = perf_event__process_mmap; 619 rep->tool.mmap2 = perf_event__process_mmap2; 620 } 621 rep->tool.comm = perf_event__process_comm; 622 rep->tool.exit = perf_event__process_exit; 623 rep->tool.fork = perf_event__process_fork; 624 rep->tool.no_warn = true; 625 } 626 627 struct task { 628 struct thread *thread; 629 struct list_head list; 630 struct list_head children; 631 }; 632 633 static struct task *tasks_list(struct task *task, struct machine *machine) 634 { 635 struct thread *parent_thread, *thread = task->thread; 636 struct task *parent_task; 637 638 /* Already listed. */ 639 if (!list_empty(&task->list)) 640 return NULL; 641 642 /* Last one in the chain. */ 643 if (thread->ppid == -1) 644 return task; 645 646 parent_thread = machine__find_thread(machine, -1, thread->ppid); 647 if (!parent_thread) 648 return ERR_PTR(-ENOENT); 649 650 parent_task = thread__priv(parent_thread); 651 list_add_tail(&task->list, &parent_task->children); 652 return tasks_list(parent_task, machine); 653 } 654 655 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 656 { 657 size_t printed = 0; 658 struct rb_node *nd; 659 660 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) { 661 struct map *map = rb_entry(nd, struct map, rb_node); 662 663 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 664 indent, "", map->start, map->end, 665 map->prot & PROT_READ ? 'r' : '-', 666 map->prot & PROT_WRITE ? 'w' : '-', 667 map->prot & PROT_EXEC ? 'x' : '-', 668 map->flags & MAP_SHARED ? 's' : 'p', 669 map->pgoff, 670 map->ino, map->dso->name); 671 } 672 673 return printed; 674 } 675 676 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp) 677 { 678 int printed = 0, i; 679 for (i = 0; i < MAP__NR_TYPES; ++i) 680 printed += maps__fprintf_task(&mg->maps[i], indent, fp); 681 return printed; 682 } 683 684 static void task__print_level(struct task *task, FILE *fp, int level) 685 { 686 struct thread *thread = task->thread; 687 struct task *child; 688 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 689 thread->pid_, thread->tid, thread->ppid, 690 level, ""); 691 692 fprintf(fp, "%s\n", thread__comm_str(thread)); 693 694 map_groups__fprintf_task(thread->mg, comm_indent, fp); 695 696 if (!list_empty(&task->children)) { 697 list_for_each_entry(child, &task->children, list) 698 task__print_level(child, fp, level + 1); 699 } 700 } 701 702 static int tasks_print(struct report *rep, FILE *fp) 703 { 704 struct perf_session *session = rep->session; 705 struct machine *machine = &session->machines.host; 706 struct task *tasks, *task; 707 unsigned int nr = 0, itask = 0, i; 708 struct rb_node *nd; 709 LIST_HEAD(list); 710 711 /* 712 * No locking needed while accessing machine->threads, 713 * because --tasks is single threaded command. 714 */ 715 716 /* Count all the threads. */ 717 for (i = 0; i < THREADS__TABLE_SIZE; i++) 718 nr += machine->threads[i].nr; 719 720 tasks = malloc(sizeof(*tasks) * nr); 721 if (!tasks) 722 return -ENOMEM; 723 724 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 725 struct threads *threads = &machine->threads[i]; 726 727 for (nd = rb_first(&threads->entries); nd; nd = rb_next(nd)) { 728 task = tasks + itask++; 729 730 task->thread = rb_entry(nd, struct thread, rb_node); 731 INIT_LIST_HEAD(&task->children); 732 INIT_LIST_HEAD(&task->list); 733 thread__set_priv(task->thread, task); 734 } 735 } 736 737 /* 738 * Iterate every task down to the unprocessed parent 739 * and link all in task children list. Task with no 740 * parent is added into 'list'. 741 */ 742 for (itask = 0; itask < nr; itask++) { 743 task = tasks + itask; 744 745 if (!list_empty(&task->list)) 746 continue; 747 748 task = tasks_list(task, machine); 749 if (IS_ERR(task)) { 750 pr_err("Error: failed to process tasks\n"); 751 free(tasks); 752 return PTR_ERR(task); 753 } 754 755 if (task) 756 list_add_tail(&task->list, &list); 757 } 758 759 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 760 761 list_for_each_entry(task, &list, list) 762 task__print_level(task, fp, 0); 763 764 free(tasks); 765 return 0; 766 } 767 768 static int __cmd_report(struct report *rep) 769 { 770 int ret; 771 struct perf_session *session = rep->session; 772 struct perf_evsel *pos; 773 struct perf_data *data = session->data; 774 775 signal(SIGINT, sig_handler); 776 777 if (rep->cpu_list) { 778 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 779 rep->cpu_bitmap); 780 if (ret) { 781 ui__error("failed to set cpu bitmap\n"); 782 return ret; 783 } 784 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 785 } 786 787 if (rep->show_threads) { 788 ret = perf_read_values_init(&rep->show_threads_values); 789 if (ret) 790 return ret; 791 } 792 793 ret = report__setup_sample_type(rep); 794 if (ret) { 795 /* report__setup_sample_type() already showed error message */ 796 return ret; 797 } 798 799 if (rep->stats_mode) 800 stats_setup(rep); 801 802 if (rep->tasks_mode) 803 tasks_setup(rep); 804 805 ret = perf_session__process_events(session); 806 if (ret) { 807 ui__error("failed to process sample\n"); 808 return ret; 809 } 810 811 if (rep->stats_mode) 812 return stats_print(rep); 813 814 if (rep->tasks_mode) 815 return tasks_print(rep, stdout); 816 817 report__warn_kptr_restrict(rep); 818 819 evlist__for_each_entry(session->evlist, pos) 820 rep->nr_entries += evsel__hists(pos)->nr_entries; 821 822 if (use_browser == 0) { 823 if (verbose > 3) 824 perf_session__fprintf(session, stdout); 825 826 if (verbose > 2) 827 perf_session__fprintf_dsos(session, stdout); 828 829 if (dump_trace) { 830 perf_session__fprintf_nr_events(session, stdout); 831 perf_evlist__fprintf_nr_events(session->evlist, stdout); 832 return 0; 833 } 834 } 835 836 ret = report__collapse_hists(rep); 837 if (ret) { 838 ui__error("failed to process hist entry\n"); 839 return ret; 840 } 841 842 if (session_done()) 843 return 0; 844 845 /* 846 * recalculate number of entries after collapsing since it 847 * might be changed during the collapse phase. 848 */ 849 rep->nr_entries = 0; 850 evlist__for_each_entry(session->evlist, pos) 851 rep->nr_entries += evsel__hists(pos)->nr_entries; 852 853 if (rep->nr_entries == 0) { 854 ui__error("The %s file has no samples!\n", data->file.path); 855 return 0; 856 } 857 858 report__output_resort(rep); 859 860 return report__browse_hists(rep); 861 } 862 863 static int 864 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 865 { 866 struct callchain_param *callchain = opt->value; 867 868 callchain->enabled = !unset; 869 /* 870 * --no-call-graph 871 */ 872 if (unset) { 873 symbol_conf.use_callchain = false; 874 callchain->mode = CHAIN_NONE; 875 return 0; 876 } 877 878 return parse_callchain_report_opt(arg); 879 } 880 881 int 882 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 883 const char *arg, int unset __maybe_unused) 884 { 885 if (arg) { 886 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 887 if (err) { 888 char buf[BUFSIZ]; 889 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 890 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 891 return -1; 892 } 893 have_ignore_callees = 1; 894 } 895 896 return 0; 897 } 898 899 static int 900 parse_branch_mode(const struct option *opt, 901 const char *str __maybe_unused, int unset) 902 { 903 int *branch_mode = opt->value; 904 905 *branch_mode = !unset; 906 return 0; 907 } 908 909 static int 910 parse_percent_limit(const struct option *opt, const char *str, 911 int unset __maybe_unused) 912 { 913 struct report *rep = opt->value; 914 double pcnt = strtof(str, NULL); 915 916 rep->min_percent = pcnt; 917 callchain_param.min_percent = pcnt; 918 return 0; 919 } 920 921 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 922 923 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 924 CALLCHAIN_REPORT_HELP 925 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 926 927 int cmd_report(int argc, const char **argv) 928 { 929 struct perf_session *session; 930 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 931 struct stat st; 932 bool has_br_stack = false; 933 int branch_mode = -1; 934 bool branch_call_mode = false; 935 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 936 const char * const report_usage[] = { 937 "perf report [<options>]", 938 NULL 939 }; 940 struct report report = { 941 .tool = { 942 .sample = process_sample_event, 943 .mmap = perf_event__process_mmap, 944 .mmap2 = perf_event__process_mmap2, 945 .comm = perf_event__process_comm, 946 .namespaces = perf_event__process_namespaces, 947 .exit = perf_event__process_exit, 948 .fork = perf_event__process_fork, 949 .lost = perf_event__process_lost, 950 .read = process_read_event, 951 .attr = perf_event__process_attr, 952 .tracing_data = perf_event__process_tracing_data, 953 .build_id = perf_event__process_build_id, 954 .id_index = perf_event__process_id_index, 955 .auxtrace_info = perf_event__process_auxtrace_info, 956 .auxtrace = perf_event__process_auxtrace, 957 .feature = perf_event__process_feature, 958 .ordered_events = true, 959 .ordering_requires_timestamps = true, 960 }, 961 .max_stack = PERF_MAX_STACK_DEPTH, 962 .pretty_printing_style = "normal", 963 .socket_filter = -1, 964 }; 965 const struct option options[] = { 966 OPT_STRING('i', "input", &input_name, "file", 967 "input file name"), 968 OPT_INCR('v', "verbose", &verbose, 969 "be more verbose (show symbol address, etc)"), 970 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"), 971 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 972 "dump raw trace in ASCII"), 973 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 974 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 975 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 976 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 977 "file", "vmlinux pathname"), 978 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 979 "file", "kallsyms pathname"), 980 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 981 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 982 "load module symbols - WARNING: use only with -k and LIVE kernel"), 983 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 984 "Show a column with the number of samples"), 985 OPT_BOOLEAN('T', "threads", &report.show_threads, 986 "Show per-thread event counters"), 987 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 988 "pretty printing style key: normal raw"), 989 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 990 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 991 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 992 "Use the stdio interface"), 993 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 994 OPT_BOOLEAN(0, "header-only", &report.header_only, 995 "Show only data header."), 996 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 997 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." 998 " Please refer the man page for the complete list."), 999 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1000 "output field(s): overhead, period, sample plus all of sort keys"), 1001 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1002 "Show sample percentage for different cpu modes"), 1003 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1004 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1005 OPT_STRING('p', "parent", &parent_pattern, "regex", 1006 "regex filter to identify parent, see: '--sort parent'"), 1007 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1008 "Only display entries with parent-match"), 1009 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1010 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1011 report_callchain_help, &report_parse_callchain_opt, 1012 callchain_default_opt), 1013 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1014 "Accumulate callchains of children and show total overhead as well"), 1015 OPT_INTEGER(0, "max-stack", &report.max_stack, 1016 "Set the maximum stack depth when parsing the callchain, " 1017 "anything beyond the specified depth will be ignored. " 1018 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1019 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1020 "alias for inverted call graph"), 1021 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1022 "ignore callees of these functions in call graphs", 1023 report_parse_ignore_callees_opt), 1024 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1025 "only consider symbols in these dsos"), 1026 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1027 "only consider symbols in these comms"), 1028 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1029 "only consider symbols in these pids"), 1030 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1031 "only consider symbols in these tids"), 1032 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1033 "only consider these symbols"), 1034 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1035 "only show symbols that (partially) match with this filter"), 1036 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1037 "width[,width...]", 1038 "don't try to adjust column width, use these fixed values"), 1039 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1040 "separator for columns, no spaces will be added between " 1041 "columns '.' is reserved."), 1042 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1043 "Only display entries resolved to a symbol"), 1044 OPT_CALLBACK(0, "symfs", NULL, "directory", 1045 "Look for files with symbols relative to this directory", 1046 symbol__config_symfs), 1047 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1048 "list of cpus to profile"), 1049 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1050 "Display extended information about perf.data file"), 1051 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, 1052 "Interleave source code with assembly code (default)"), 1053 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, 1054 "Display raw encoding of assembly instructions (default)"), 1055 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1056 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1057 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1058 "Show a column with the sum of periods"), 1059 OPT_BOOLEAN(0, "group", &symbol_conf.event_group, 1060 "Show event group information together"), 1061 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1062 "use branch records for per branch histogram filling", 1063 parse_branch_mode), 1064 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1065 "add last branch records to call history"), 1066 OPT_STRING(0, "objdump", &objdump_path, "path", 1067 "objdump binary to use for disassembly and annotations"), 1068 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1069 "Disable symbol demangling"), 1070 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1071 "Enable kernel symbol demangling"), 1072 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1073 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1074 "Don't show entries under that percent", parse_percent_limit), 1075 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1076 "how to display percentage of filtered entries", parse_filter_percentage), 1077 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1078 "Instruction Tracing options", 1079 itrace_parse_synth_opts), 1080 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1081 "Show full source file name path for source lines"), 1082 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1083 "Show callgraph from reference event"), 1084 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1085 "only show processor socket that match with this filter"), 1086 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1087 "Show raw trace event output (do not use print fmt or plugins)"), 1088 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1089 "Show entries in a hierarchy"), 1090 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1091 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1092 stdio__config_color, "always"), 1093 OPT_STRING(0, "time", &report.time_str, "str", 1094 "Time span of interest (start,stop)"), 1095 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1096 "Show inline function"), 1097 OPT_END() 1098 }; 1099 struct perf_data data = { 1100 .mode = PERF_DATA_MODE_READ, 1101 }; 1102 int ret = hists__init(); 1103 1104 if (ret < 0) 1105 return ret; 1106 1107 ret = perf_config(report__config, &report); 1108 if (ret) 1109 return ret; 1110 1111 argc = parse_options(argc, argv, options, report_usage, 0); 1112 if (argc) { 1113 /* 1114 * Special case: if there's an argument left then assume that 1115 * it's a symbol filter: 1116 */ 1117 if (argc > 1) 1118 usage_with_options(report_usage, options); 1119 1120 report.symbol_filter_str = argv[0]; 1121 } 1122 1123 if (report.mmaps_mode) 1124 report.tasks_mode = true; 1125 1126 if (quiet) 1127 perf_quiet_option(); 1128 1129 if (symbol_conf.vmlinux_name && 1130 access(symbol_conf.vmlinux_name, R_OK)) { 1131 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name); 1132 return -EINVAL; 1133 } 1134 if (symbol_conf.kallsyms_name && 1135 access(symbol_conf.kallsyms_name, R_OK)) { 1136 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name); 1137 return -EINVAL; 1138 } 1139 1140 if (report.inverted_callchain) 1141 callchain_param.order = ORDER_CALLER; 1142 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1143 callchain_param.order = ORDER_CALLER; 1144 1145 if (itrace_synth_opts.callchain && 1146 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1147 report.max_stack = itrace_synth_opts.callchain_sz; 1148 1149 if (!input_name || !strlen(input_name)) { 1150 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1151 input_name = "-"; 1152 else 1153 input_name = "perf.data"; 1154 } 1155 1156 data.file.path = input_name; 1157 data.force = symbol_conf.force; 1158 1159 repeat: 1160 session = perf_session__new(&data, false, &report.tool); 1161 if (session == NULL) 1162 return -1; 1163 1164 if (report.queue_size) { 1165 ordered_events__set_alloc_size(&session->ordered_events, 1166 report.queue_size); 1167 } 1168 1169 session->itrace_synth_opts = &itrace_synth_opts; 1170 1171 report.session = session; 1172 1173 has_br_stack = perf_header__has_feat(&session->header, 1174 HEADER_BRANCH_STACK); 1175 1176 if (itrace_synth_opts.last_branch) 1177 has_br_stack = true; 1178 1179 if (has_br_stack && branch_call_mode) 1180 symbol_conf.show_branchflag_count = true; 1181 1182 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1183 1184 /* 1185 * Branch mode is a tristate: 1186 * -1 means default, so decide based on the file having branch data. 1187 * 0/1 means the user chose a mode. 1188 */ 1189 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1190 !branch_call_mode) { 1191 sort__mode = SORT_MODE__BRANCH; 1192 symbol_conf.cumulate_callchain = false; 1193 } 1194 if (branch_call_mode) { 1195 callchain_param.key = CCKEY_ADDRESS; 1196 callchain_param.branch_callstack = 1; 1197 symbol_conf.use_callchain = true; 1198 callchain_register_param(&callchain_param); 1199 if (sort_order == NULL) 1200 sort_order = "srcline,symbol,dso"; 1201 } 1202 1203 if (report.mem_mode) { 1204 if (sort__mode == SORT_MODE__BRANCH) { 1205 pr_err("branch and mem mode incompatible\n"); 1206 goto error; 1207 } 1208 sort__mode = SORT_MODE__MEMORY; 1209 symbol_conf.cumulate_callchain = false; 1210 } 1211 1212 if (symbol_conf.report_hierarchy) { 1213 /* disable incompatible options */ 1214 symbol_conf.cumulate_callchain = false; 1215 1216 if (field_order) { 1217 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1218 parse_options_usage(report_usage, options, "F", 1); 1219 parse_options_usage(NULL, options, "hierarchy", 0); 1220 goto error; 1221 } 1222 1223 perf_hpp_list.need_collapse = true; 1224 } 1225 1226 if (report.use_stdio) 1227 use_browser = 0; 1228 else if (report.use_tui) 1229 use_browser = 1; 1230 else if (report.use_gtk) 1231 use_browser = 2; 1232 1233 /* Force tty output for header output and per-thread stat. */ 1234 if (report.header || report.header_only || report.show_threads) 1235 use_browser = 0; 1236 if (report.header || report.header_only) 1237 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1238 if (report.show_full_info) 1239 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1240 if (report.stats_mode || report.tasks_mode) 1241 use_browser = 0; 1242 if (report.stats_mode && report.tasks_mode) { 1243 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1244 goto error; 1245 } 1246 1247 if (strcmp(input_name, "-") != 0) 1248 setup_browser(true); 1249 else 1250 use_browser = 0; 1251 1252 if (setup_sorting(session->evlist) < 0) { 1253 if (sort_order) 1254 parse_options_usage(report_usage, options, "s", 1); 1255 if (field_order) 1256 parse_options_usage(sort_order ? NULL : report_usage, 1257 options, "F", 1); 1258 goto error; 1259 } 1260 1261 if ((report.header || report.header_only) && !quiet) { 1262 perf_session__fprintf_info(session, stdout, 1263 report.show_full_info); 1264 if (report.header_only) { 1265 ret = 0; 1266 goto error; 1267 } 1268 } else if (use_browser == 0 && !quiet && 1269 !report.stats_mode && !report.tasks_mode) { 1270 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1271 stdout); 1272 } 1273 1274 /* 1275 * Only in the TUI browser we are doing integrated annotation, 1276 * so don't allocate extra space that won't be used in the stdio 1277 * implementation. 1278 */ 1279 if (ui__has_annotation()) { 1280 ret = symbol__annotation_init(); 1281 if (ret < 0) 1282 goto error; 1283 /* 1284 * For searching by name on the "Browse map details". 1285 * providing it only in verbose mode not to bloat too 1286 * much struct symbol. 1287 */ 1288 if (verbose > 0) { 1289 /* 1290 * XXX: Need to provide a less kludgy way to ask for 1291 * more space per symbol, the u32 is for the index on 1292 * the ui browser. 1293 * See symbol__browser_index. 1294 */ 1295 symbol_conf.priv_size += sizeof(u32); 1296 symbol_conf.sort_by_name = true; 1297 } 1298 } 1299 1300 if (symbol__init(&session->header.env) < 0) 1301 goto error; 1302 1303 report.ptime_range = perf_time__range_alloc(report.time_str, 1304 &report.range_size); 1305 if (!report.ptime_range) { 1306 ret = -ENOMEM; 1307 goto error; 1308 } 1309 1310 if (perf_time__parse_str(report.ptime_range, report.time_str) != 0) { 1311 if (session->evlist->first_sample_time == 0 && 1312 session->evlist->last_sample_time == 0) { 1313 pr_err("HINT: no first/last sample time found in perf data.\n" 1314 "Please use latest perf binary to execute 'perf record'\n" 1315 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n"); 1316 ret = -EINVAL; 1317 goto error; 1318 } 1319 1320 report.range_num = perf_time__percent_parse_str( 1321 report.ptime_range, report.range_size, 1322 report.time_str, 1323 session->evlist->first_sample_time, 1324 session->evlist->last_sample_time); 1325 1326 if (report.range_num < 0) { 1327 pr_err("Invalid time string\n"); 1328 ret = -EINVAL; 1329 goto error; 1330 } 1331 } else { 1332 report.range_num = 1; 1333 } 1334 1335 sort__setup_elide(stdout); 1336 1337 ret = __cmd_report(&report); 1338 if (ret == K_SWITCH_INPUT_DATA) { 1339 perf_session__delete(session); 1340 goto repeat; 1341 } else 1342 ret = 0; 1343 1344 error: 1345 zfree(&report.ptime_range); 1346 1347 perf_session__delete(session); 1348 return ret; 1349 } 1350