1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * builtin-report.c 4 * 5 * Builtin report command: Analyze the perf.data input file, 6 * look up and read DSOs and symbol information and display 7 * a histogram of results, along various sorting keys. 8 */ 9 #include "builtin.h" 10 11 #include "util/config.h" 12 13 #include "util/annotate.h" 14 #include "util/color.h" 15 #include "util/dso.h" 16 #include <linux/list.h> 17 #include <linux/rbtree.h> 18 #include <linux/err.h> 19 #include <linux/zalloc.h> 20 #include "util/map.h" 21 #include "util/symbol.h" 22 #include "util/map_symbol.h" 23 #include "util/mem-events.h" 24 #include "util/branch.h" 25 #include "util/callchain.h" 26 #include "util/values.h" 27 28 #include "perf.h" 29 #include "util/debug.h" 30 #include "util/evlist.h" 31 #include "util/evsel.h" 32 #include "util/evswitch.h" 33 #include "util/header.h" 34 #include "util/session.h" 35 #include "util/srcline.h" 36 #include "util/tool.h" 37 38 #include <subcmd/parse-options.h> 39 #include <subcmd/exec-cmd.h> 40 #include "util/parse-events.h" 41 42 #include "util/thread.h" 43 #include "util/sort.h" 44 #include "util/hist.h" 45 #include "util/data.h" 46 #include "arch/common.h" 47 #include "util/time-utils.h" 48 #include "util/auxtrace.h" 49 #include "util/units.h" 50 #include "util/util.h" // perf_tip() 51 #include "ui/ui.h" 52 #include "ui/progress.h" 53 #include "util/block-info.h" 54 55 #include <dlfcn.h> 56 #include <errno.h> 57 #include <inttypes.h> 58 #include <regex.h> 59 #include <linux/ctype.h> 60 #include <signal.h> 61 #include <linux/bitmap.h> 62 #include <linux/string.h> 63 #include <linux/stringify.h> 64 #include <linux/time64.h> 65 #include <sys/types.h> 66 #include <sys/stat.h> 67 #include <unistd.h> 68 #include <linux/mman.h> 69 70 #ifdef HAVE_LIBTRACEEVENT 71 #include <traceevent/event-parse.h> 72 #endif 73 74 struct report { 75 struct perf_tool tool; 76 struct perf_session *session; 77 struct evswitch evswitch; 78 #ifdef HAVE_SLANG_SUPPORT 79 bool use_tui; 80 #endif 81 #ifdef HAVE_GTK2_SUPPORT 82 bool use_gtk; 83 #endif 84 bool use_stdio; 85 bool show_full_info; 86 bool show_threads; 87 bool inverted_callchain; 88 bool mem_mode; 89 bool stats_mode; 90 bool tasks_mode; 91 bool mmaps_mode; 92 bool header; 93 bool header_only; 94 bool nonany_branch_mode; 95 bool group_set; 96 bool stitch_lbr; 97 bool disable_order; 98 bool skip_empty; 99 int max_stack; 100 struct perf_read_values show_threads_values; 101 struct annotation_options annotation_opts; 102 const char *pretty_printing_style; 103 const char *cpu_list; 104 const char *symbol_filter_str; 105 const char *time_str; 106 struct perf_time_interval *ptime_range; 107 int range_size; 108 int range_num; 109 float min_percent; 110 u64 nr_entries; 111 u64 queue_size; 112 u64 total_cycles; 113 int socket_filter; 114 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 115 struct branch_type_stat brtype_stat; 116 bool symbol_ipc; 117 bool total_cycles_mode; 118 struct block_report *block_reports; 119 int nr_block_reports; 120 }; 121 122 static int report__config(const char *var, const char *value, void *cb) 123 { 124 struct report *rep = cb; 125 126 if (!strcmp(var, "report.group")) { 127 symbol_conf.event_group = perf_config_bool(var, value); 128 return 0; 129 } 130 if (!strcmp(var, "report.percent-limit")) { 131 double pcnt = strtof(value, NULL); 132 133 rep->min_percent = pcnt; 134 callchain_param.min_percent = pcnt; 135 return 0; 136 } 137 if (!strcmp(var, "report.children")) { 138 symbol_conf.cumulate_callchain = perf_config_bool(var, value); 139 return 0; 140 } 141 if (!strcmp(var, "report.queue-size")) 142 return perf_config_u64(&rep->queue_size, var, value); 143 144 if (!strcmp(var, "report.sort_order")) { 145 default_sort_order = strdup(value); 146 if (!default_sort_order) { 147 pr_err("Not enough memory for report.sort_order\n"); 148 return -1; 149 } 150 return 0; 151 } 152 153 if (!strcmp(var, "report.skip-empty")) { 154 rep->skip_empty = perf_config_bool(var, value); 155 return 0; 156 } 157 158 pr_debug("%s variable unknown, ignoring...", var); 159 return 0; 160 } 161 162 static int hist_iter__report_callback(struct hist_entry_iter *iter, 163 struct addr_location *al, bool single, 164 void *arg) 165 { 166 int err = 0; 167 struct report *rep = arg; 168 struct hist_entry *he = iter->he; 169 struct evsel *evsel = iter->evsel; 170 struct perf_sample *sample = iter->sample; 171 struct mem_info *mi; 172 struct branch_info *bi; 173 174 if (!ui__has_annotation() && !rep->symbol_ipc) 175 return 0; 176 177 if (sort__mode == SORT_MODE__BRANCH) { 178 bi = he->branch_info; 179 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 180 if (err) 181 goto out; 182 183 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 184 185 } else if (rep->mem_mode) { 186 mi = he->mem_info; 187 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel); 188 if (err) 189 goto out; 190 191 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 192 193 } else if (symbol_conf.cumulate_callchain) { 194 if (single) 195 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 196 } else { 197 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr); 198 } 199 200 out: 201 return err; 202 } 203 204 static int hist_iter__branch_callback(struct hist_entry_iter *iter, 205 struct addr_location *al __maybe_unused, 206 bool single __maybe_unused, 207 void *arg) 208 { 209 struct hist_entry *he = iter->he; 210 struct report *rep = arg; 211 struct branch_info *bi = he->branch_info; 212 struct perf_sample *sample = iter->sample; 213 struct evsel *evsel = iter->evsel; 214 int err; 215 216 branch_type_count(&rep->brtype_stat, &bi->flags, 217 bi->from.addr, bi->to.addr); 218 219 if (!ui__has_annotation() && !rep->symbol_ipc) 220 return 0; 221 222 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel); 223 if (err) 224 goto out; 225 226 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel); 227 228 out: 229 return err; 230 } 231 232 static void setup_forced_leader(struct report *report, 233 struct evlist *evlist) 234 { 235 if (report->group_set) 236 evlist__force_leader(evlist); 237 } 238 239 static int process_feature_event(struct perf_session *session, 240 union perf_event *event) 241 { 242 struct report *rep = container_of(session->tool, struct report, tool); 243 244 if (event->feat.feat_id < HEADER_LAST_FEATURE) 245 return perf_event__process_feature(session, event); 246 247 if (event->feat.feat_id != HEADER_LAST_FEATURE) { 248 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n", 249 event->feat.feat_id); 250 return -1; 251 } else if (rep->header_only) { 252 session_done = 1; 253 } 254 255 /* 256 * (feat_id = HEADER_LAST_FEATURE) is the end marker which 257 * means all features are received, now we can force the 258 * group if needed. 259 */ 260 setup_forced_leader(rep, session->evlist); 261 return 0; 262 } 263 264 static int process_sample_event(struct perf_tool *tool, 265 union perf_event *event, 266 struct perf_sample *sample, 267 struct evsel *evsel, 268 struct machine *machine) 269 { 270 struct report *rep = container_of(tool, struct report, tool); 271 struct addr_location al; 272 struct hist_entry_iter iter = { 273 .evsel = evsel, 274 .sample = sample, 275 .hide_unresolved = symbol_conf.hide_unresolved, 276 .add_entry_cb = hist_iter__report_callback, 277 }; 278 int ret = 0; 279 280 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num, 281 sample->time)) { 282 return 0; 283 } 284 285 if (evswitch__discard(&rep->evswitch, evsel)) 286 return 0; 287 288 addr_location__init(&al); 289 if (machine__resolve(machine, &al, sample) < 0) { 290 pr_debug("problem processing %d event, skipping it.\n", 291 event->header.type); 292 ret = -1; 293 goto out_put; 294 } 295 296 if (rep->stitch_lbr) 297 thread__set_lbr_stitch_enable(al.thread, true); 298 299 if (symbol_conf.hide_unresolved && al.sym == NULL) 300 goto out_put; 301 302 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap)) 303 goto out_put; 304 305 if (sort__mode == SORT_MODE__BRANCH) { 306 /* 307 * A non-synthesized event might not have a branch stack if 308 * branch stacks have been synthesized (using itrace options). 309 */ 310 if (!sample->branch_stack) 311 goto out_put; 312 313 iter.add_entry_cb = hist_iter__branch_callback; 314 iter.ops = &hist_iter_branch; 315 } else if (rep->mem_mode) { 316 iter.ops = &hist_iter_mem; 317 } else if (symbol_conf.cumulate_callchain) { 318 iter.ops = &hist_iter_cumulative; 319 } else { 320 iter.ops = &hist_iter_normal; 321 } 322 323 if (al.map != NULL) 324 map__dso(al.map)->hit = 1; 325 326 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) { 327 hist__account_cycles(sample->branch_stack, &al, sample, 328 rep->nonany_branch_mode, 329 &rep->total_cycles); 330 } 331 332 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep); 333 if (ret < 0) 334 pr_debug("problem adding hist entry, skipping event\n"); 335 out_put: 336 addr_location__exit(&al); 337 return ret; 338 } 339 340 static int process_read_event(struct perf_tool *tool, 341 union perf_event *event, 342 struct perf_sample *sample __maybe_unused, 343 struct evsel *evsel, 344 struct machine *machine __maybe_unused) 345 { 346 struct report *rep = container_of(tool, struct report, tool); 347 348 if (rep->show_threads) { 349 const char *name = evsel__name(evsel); 350 int err = perf_read_values_add_value(&rep->show_threads_values, 351 event->read.pid, event->read.tid, 352 evsel->core.idx, 353 name, 354 event->read.value); 355 356 if (err) 357 return err; 358 } 359 360 return 0; 361 } 362 363 /* For pipe mode, sample_type is not currently set */ 364 static int report__setup_sample_type(struct report *rep) 365 { 366 struct perf_session *session = rep->session; 367 u64 sample_type = evlist__combined_sample_type(session->evlist); 368 bool is_pipe = perf_data__is_pipe(session->data); 369 struct evsel *evsel; 370 371 if (session->itrace_synth_opts->callchain || 372 session->itrace_synth_opts->add_callchain || 373 (!is_pipe && 374 perf_header__has_feat(&session->header, HEADER_AUXTRACE) && 375 !session->itrace_synth_opts->set)) 376 sample_type |= PERF_SAMPLE_CALLCHAIN; 377 378 if (session->itrace_synth_opts->last_branch || 379 session->itrace_synth_opts->add_last_branch) 380 sample_type |= PERF_SAMPLE_BRANCH_STACK; 381 382 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) { 383 if (perf_hpp_list.parent) { 384 ui__error("Selected --sort parent, but no " 385 "callchain data. Did you call " 386 "'perf record' without -g?\n"); 387 return -EINVAL; 388 } 389 if (symbol_conf.use_callchain && 390 !symbol_conf.show_branchflag_count) { 391 ui__error("Selected -g or --branch-history.\n" 392 "But no callchain or branch data.\n" 393 "Did you call 'perf record' without -g or -b?\n"); 394 return -1; 395 } 396 } else if (!callchain_param.enabled && 397 callchain_param.mode != CHAIN_NONE && 398 !symbol_conf.use_callchain) { 399 symbol_conf.use_callchain = true; 400 if (callchain_register_param(&callchain_param) < 0) { 401 ui__error("Can't register callchain params.\n"); 402 return -EINVAL; 403 } 404 } 405 406 if (symbol_conf.cumulate_callchain) { 407 /* Silently ignore if callchain is missing */ 408 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) { 409 symbol_conf.cumulate_callchain = false; 410 perf_hpp__cancel_cumulate(); 411 } 412 } 413 414 if (sort__mode == SORT_MODE__BRANCH) { 415 if (!is_pipe && 416 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) { 417 ui__error("Selected -b but no branch data. " 418 "Did you call perf record without -b?\n"); 419 return -1; 420 } 421 } 422 423 if (sort__mode == SORT_MODE__MEMORY) { 424 /* 425 * FIXUP: prior to kernel 5.18, Arm SPE missed to set 426 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward 427 * compatibility, set the bit if it's an old perf data file. 428 */ 429 evlist__for_each_entry(session->evlist, evsel) { 430 if (strstr(evsel->name, "arm_spe") && 431 !(sample_type & PERF_SAMPLE_DATA_SRC)) { 432 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC; 433 sample_type |= PERF_SAMPLE_DATA_SRC; 434 } 435 } 436 437 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) { 438 ui__error("Selected --mem-mode but no mem data. " 439 "Did you call perf record without -d?\n"); 440 return -1; 441 } 442 } 443 444 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env)); 445 446 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) { 447 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n" 448 "Please apply --call-graph lbr when recording.\n"); 449 rep->stitch_lbr = false; 450 } 451 452 /* ??? handle more cases than just ANY? */ 453 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY)) 454 rep->nonany_branch_mode = true; 455 456 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT) 457 if (dwarf_callchain_users) { 458 ui__warning("Please install libunwind or libdw " 459 "development packages during the perf build.\n"); 460 } 461 #endif 462 463 return 0; 464 } 465 466 static void sig_handler(int sig __maybe_unused) 467 { 468 session_done = 1; 469 } 470 471 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep, 472 const char *evname, FILE *fp) 473 { 474 size_t ret; 475 char unit; 476 unsigned long nr_samples = hists->stats.nr_samples; 477 u64 nr_events = hists->stats.total_period; 478 struct evsel *evsel = hists_to_evsel(hists); 479 char buf[512]; 480 size_t size = sizeof(buf); 481 int socked_id = hists->socket_filter; 482 483 if (quiet) 484 return 0; 485 486 if (symbol_conf.filter_relative) { 487 nr_samples = hists->stats.nr_non_filtered_samples; 488 nr_events = hists->stats.total_non_filtered_period; 489 } 490 491 if (evsel__is_group_event(evsel)) { 492 struct evsel *pos; 493 494 evsel__group_desc(evsel, buf, size); 495 evname = buf; 496 497 for_each_group_member(pos, evsel) { 498 const struct hists *pos_hists = evsel__hists(pos); 499 500 if (symbol_conf.filter_relative) { 501 nr_samples += pos_hists->stats.nr_non_filtered_samples; 502 nr_events += pos_hists->stats.total_non_filtered_period; 503 } else { 504 nr_samples += pos_hists->stats.nr_samples; 505 nr_events += pos_hists->stats.total_period; 506 } 507 } 508 } 509 510 nr_samples = convert_unit(nr_samples, &unit); 511 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit); 512 if (evname != NULL) { 513 ret += fprintf(fp, " of event%s '%s'", 514 evsel->core.nr_members > 1 ? "s" : "", evname); 515 } 516 517 if (rep->time_str) 518 ret += fprintf(fp, " (time slices: %s)", rep->time_str); 519 520 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) { 521 ret += fprintf(fp, ", show reference callgraph"); 522 } 523 524 if (rep->mem_mode) { 525 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events); 526 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order); 527 } else 528 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events); 529 530 if (socked_id > -1) 531 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id); 532 533 return ret + fprintf(fp, "\n#\n"); 534 } 535 536 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep) 537 { 538 struct evsel *pos; 539 int i = 0, ret; 540 541 evlist__for_each_entry(evlist, pos) { 542 ret = report__browse_block_hists(&rep->block_reports[i++].hist, 543 rep->min_percent, pos, 544 &rep->session->header.env, 545 &rep->annotation_opts); 546 if (ret != 0) 547 return ret; 548 } 549 550 return 0; 551 } 552 553 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help) 554 { 555 struct evsel *pos; 556 int i = 0; 557 558 if (!quiet) { 559 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", 560 evlist->stats.total_lost_samples); 561 } 562 563 evlist__for_each_entry(evlist, pos) { 564 struct hists *hists = evsel__hists(pos); 565 const char *evname = evsel__name(pos); 566 567 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) 568 continue; 569 570 if (rep->skip_empty && !hists->stats.nr_samples) 571 continue; 572 573 hists__fprintf_nr_sample_events(hists, rep, evname, stdout); 574 575 if (rep->total_cycles_mode) { 576 report__browse_block_hists(&rep->block_reports[i++].hist, 577 rep->min_percent, pos, 578 NULL, NULL); 579 continue; 580 } 581 582 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout, 583 !(symbol_conf.use_callchain || 584 symbol_conf.show_branchflag_count)); 585 fprintf(stdout, "\n\n"); 586 } 587 588 if (!quiet) 589 fprintf(stdout, "#\n# (%s)\n#\n", help); 590 591 if (rep->show_threads) { 592 bool style = !strcmp(rep->pretty_printing_style, "raw"); 593 perf_read_values_display(stdout, &rep->show_threads_values, 594 style); 595 perf_read_values_destroy(&rep->show_threads_values); 596 } 597 598 if (sort__mode == SORT_MODE__BRANCH) 599 branch_type_stat_display(stdout, &rep->brtype_stat); 600 601 return 0; 602 } 603 604 static void report__warn_kptr_restrict(const struct report *rep) 605 { 606 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host); 607 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL; 608 609 if (evlist__exclude_kernel(rep->session->evlist)) 610 return; 611 612 if (kernel_map == NULL || 613 (map__dso(kernel_map)->hit && 614 (kernel_kmap->ref_reloc_sym == NULL || 615 kernel_kmap->ref_reloc_sym->addr == 0))) { 616 const char *desc = 617 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 618 "can't be resolved."; 619 620 if (kernel_map && map__has_symbols(kernel_map)) { 621 desc = "If some relocation was applied (e.g. " 622 "kexec) symbols may be misresolved."; 623 } 624 625 ui__warning( 626 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 627 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 628 "Samples in kernel modules can't be resolved as well.\n\n", 629 desc); 630 } 631 } 632 633 static int report__gtk_browse_hists(struct report *rep, const char *help) 634 { 635 int (*hist_browser)(struct evlist *evlist, const char *help, 636 struct hist_browser_timer *timer, float min_pcnt); 637 638 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists"); 639 640 if (hist_browser == NULL) { 641 ui__error("GTK browser not found!\n"); 642 return -1; 643 } 644 645 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent); 646 } 647 648 static int report__browse_hists(struct report *rep) 649 { 650 int ret; 651 struct perf_session *session = rep->session; 652 struct evlist *evlist = session->evlist; 653 char *help = NULL, *path = NULL; 654 655 path = system_path(TIPDIR); 656 if (perf_tip(&help, path) || help == NULL) { 657 /* fallback for people who don't install perf ;-) */ 658 free(path); 659 path = system_path(DOCDIR); 660 if (perf_tip(&help, path) || help == NULL) 661 help = strdup("Cannot load tips.txt file, please install perf!"); 662 } 663 free(path); 664 665 switch (use_browser) { 666 case 1: 667 if (rep->total_cycles_mode) { 668 ret = evlist__tui_block_hists_browse(evlist, rep); 669 break; 670 } 671 672 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent, 673 &session->header.env, true, &rep->annotation_opts); 674 /* 675 * Usually "ret" is the last pressed key, and we only 676 * care if the key notifies us to switch data file. 677 */ 678 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD) 679 ret = 0; 680 break; 681 case 2: 682 ret = report__gtk_browse_hists(rep, help); 683 break; 684 default: 685 ret = evlist__tty_browse_hists(evlist, rep, help); 686 break; 687 } 688 free(help); 689 return ret; 690 } 691 692 static int report__collapse_hists(struct report *rep) 693 { 694 struct perf_session *session = rep->session; 695 struct evlist *evlist = session->evlist; 696 struct ui_progress prog; 697 struct evsel *pos; 698 int ret = 0; 699 700 /* 701 * The pipe data needs to setup hierarchy hpp formats now, because it 702 * cannot know about evsels in the data before reading the data. The 703 * normal file data saves the event (attribute) info in the header 704 * section, but pipe does not have the luxury. 705 */ 706 if (perf_data__is_pipe(session->data)) { 707 if (perf_hpp__setup_hists_formats(&perf_hpp_list, evlist) < 0) { 708 ui__error("Failed to setup hierarchy output formats\n"); 709 return -1; 710 } 711 } 712 713 ui_progress__init(&prog, rep->nr_entries, "Merging related events..."); 714 715 evlist__for_each_entry(rep->session->evlist, pos) { 716 struct hists *hists = evsel__hists(pos); 717 718 if (pos->core.idx == 0) 719 hists->symbol_filter_str = rep->symbol_filter_str; 720 721 hists->socket_filter = rep->socket_filter; 722 723 ret = hists__collapse_resort(hists, &prog); 724 if (ret < 0) 725 break; 726 727 /* Non-group events are considered as leader */ 728 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) { 729 struct hists *leader_hists = evsel__hists(evsel__leader(pos)); 730 731 hists__match(leader_hists, hists); 732 hists__link(leader_hists, hists); 733 } 734 } 735 736 ui_progress__finish(); 737 return ret; 738 } 739 740 static int hists__resort_cb(struct hist_entry *he, void *arg) 741 { 742 struct report *rep = arg; 743 struct symbol *sym = he->ms.sym; 744 745 if (rep->symbol_ipc && sym && !sym->annotate2) { 746 struct evsel *evsel = hists_to_evsel(he->hists); 747 748 symbol__annotate2(&he->ms, evsel, &rep->annotation_opts, NULL); 749 } 750 751 return 0; 752 } 753 754 static void report__output_resort(struct report *rep) 755 { 756 struct ui_progress prog; 757 struct evsel *pos; 758 759 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output..."); 760 761 evlist__for_each_entry(rep->session->evlist, pos) { 762 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep); 763 } 764 765 ui_progress__finish(); 766 } 767 768 static int count_sample_event(struct perf_tool *tool __maybe_unused, 769 union perf_event *event __maybe_unused, 770 struct perf_sample *sample __maybe_unused, 771 struct evsel *evsel, 772 struct machine *machine __maybe_unused) 773 { 774 struct hists *hists = evsel__hists(evsel); 775 776 hists__inc_nr_events(hists); 777 return 0; 778 } 779 780 static int count_lost_samples_event(struct perf_tool *tool, 781 union perf_event *event, 782 struct perf_sample *sample, 783 struct machine *machine __maybe_unused) 784 { 785 struct report *rep = container_of(tool, struct report, tool); 786 struct evsel *evsel; 787 788 evsel = evlist__id2evsel(rep->session->evlist, sample->id); 789 if (evsel) { 790 hists__inc_nr_lost_samples(evsel__hists(evsel), 791 event->lost_samples.lost); 792 } 793 return 0; 794 } 795 796 static int process_attr(struct perf_tool *tool __maybe_unused, 797 union perf_event *event, 798 struct evlist **pevlist); 799 800 static void stats_setup(struct report *rep) 801 { 802 memset(&rep->tool, 0, sizeof(rep->tool)); 803 rep->tool.attr = process_attr; 804 rep->tool.sample = count_sample_event; 805 rep->tool.lost_samples = count_lost_samples_event; 806 rep->tool.no_warn = true; 807 } 808 809 static int stats_print(struct report *rep) 810 { 811 struct perf_session *session = rep->session; 812 813 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty); 814 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty); 815 return 0; 816 } 817 818 static void tasks_setup(struct report *rep) 819 { 820 memset(&rep->tool, 0, sizeof(rep->tool)); 821 rep->tool.ordered_events = true; 822 if (rep->mmaps_mode) { 823 rep->tool.mmap = perf_event__process_mmap; 824 rep->tool.mmap2 = perf_event__process_mmap2; 825 } 826 rep->tool.attr = process_attr; 827 rep->tool.comm = perf_event__process_comm; 828 rep->tool.exit = perf_event__process_exit; 829 rep->tool.fork = perf_event__process_fork; 830 rep->tool.no_warn = true; 831 } 832 833 struct task { 834 struct thread *thread; 835 struct list_head list; 836 struct list_head children; 837 }; 838 839 static struct task *tasks_list(struct task *task, struct machine *machine) 840 { 841 struct thread *parent_thread, *thread = task->thread; 842 struct task *parent_task; 843 844 /* Already listed. */ 845 if (!list_empty(&task->list)) 846 return NULL; 847 848 /* Last one in the chain. */ 849 if (thread__ppid(thread) == -1) 850 return task; 851 852 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread)); 853 if (!parent_thread) 854 return ERR_PTR(-ENOENT); 855 856 parent_task = thread__priv(parent_thread); 857 thread__put(parent_thread); 858 list_add_tail(&task->list, &parent_task->children); 859 return tasks_list(parent_task, machine); 860 } 861 862 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp) 863 { 864 size_t printed = 0; 865 struct map_rb_node *rb_node; 866 867 maps__for_each_entry(maps, rb_node) { 868 struct map *map = rb_node->map; 869 const struct dso *dso = map__dso(map); 870 u32 prot = map__prot(map); 871 872 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n", 873 indent, "", map__start(map), map__end(map), 874 prot & PROT_READ ? 'r' : '-', 875 prot & PROT_WRITE ? 'w' : '-', 876 prot & PROT_EXEC ? 'x' : '-', 877 map__flags(map) ? 's' : 'p', 878 map__pgoff(map), 879 dso->id.ino, dso->name); 880 } 881 882 return printed; 883 } 884 885 static void task__print_level(struct task *task, FILE *fp, int level) 886 { 887 struct thread *thread = task->thread; 888 struct task *child; 889 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s", 890 thread__pid(thread), thread__tid(thread), 891 thread__ppid(thread), level, ""); 892 893 fprintf(fp, "%s\n", thread__comm_str(thread)); 894 895 maps__fprintf_task(thread__maps(thread), comm_indent, fp); 896 897 if (!list_empty(&task->children)) { 898 list_for_each_entry(child, &task->children, list) 899 task__print_level(child, fp, level + 1); 900 } 901 } 902 903 static int tasks_print(struct report *rep, FILE *fp) 904 { 905 struct perf_session *session = rep->session; 906 struct machine *machine = &session->machines.host; 907 struct task *tasks, *task; 908 unsigned int nr = 0, itask = 0, i; 909 struct rb_node *nd; 910 LIST_HEAD(list); 911 912 /* 913 * No locking needed while accessing machine->threads, 914 * because --tasks is single threaded command. 915 */ 916 917 /* Count all the threads. */ 918 for (i = 0; i < THREADS__TABLE_SIZE; i++) 919 nr += machine->threads[i].nr; 920 921 tasks = malloc(sizeof(*tasks) * nr); 922 if (!tasks) 923 return -ENOMEM; 924 925 for (i = 0; i < THREADS__TABLE_SIZE; i++) { 926 struct threads *threads = &machine->threads[i]; 927 928 for (nd = rb_first_cached(&threads->entries); nd; 929 nd = rb_next(nd)) { 930 task = tasks + itask++; 931 932 task->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread; 933 INIT_LIST_HEAD(&task->children); 934 INIT_LIST_HEAD(&task->list); 935 thread__set_priv(task->thread, task); 936 } 937 } 938 939 /* 940 * Iterate every task down to the unprocessed parent 941 * and link all in task children list. Task with no 942 * parent is added into 'list'. 943 */ 944 for (itask = 0; itask < nr; itask++) { 945 task = tasks + itask; 946 947 if (!list_empty(&task->list)) 948 continue; 949 950 task = tasks_list(task, machine); 951 if (IS_ERR(task)) { 952 pr_err("Error: failed to process tasks\n"); 953 free(tasks); 954 return PTR_ERR(task); 955 } 956 957 if (task) 958 list_add_tail(&task->list, &list); 959 } 960 961 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm"); 962 963 list_for_each_entry(task, &list, list) 964 task__print_level(task, fp, 0); 965 966 free(tasks); 967 return 0; 968 } 969 970 static int __cmd_report(struct report *rep) 971 { 972 int ret; 973 struct perf_session *session = rep->session; 974 struct evsel *pos; 975 struct perf_data *data = session->data; 976 977 signal(SIGINT, sig_handler); 978 979 if (rep->cpu_list) { 980 ret = perf_session__cpu_bitmap(session, rep->cpu_list, 981 rep->cpu_bitmap); 982 if (ret) { 983 ui__error("failed to set cpu bitmap\n"); 984 return ret; 985 } 986 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap; 987 } 988 989 if (rep->show_threads) { 990 ret = perf_read_values_init(&rep->show_threads_values); 991 if (ret) 992 return ret; 993 } 994 995 ret = report__setup_sample_type(rep); 996 if (ret) { 997 /* report__setup_sample_type() already showed error message */ 998 return ret; 999 } 1000 1001 if (rep->stats_mode) 1002 stats_setup(rep); 1003 1004 if (rep->tasks_mode) 1005 tasks_setup(rep); 1006 1007 ret = perf_session__process_events(session); 1008 if (ret) { 1009 ui__error("failed to process sample\n"); 1010 return ret; 1011 } 1012 1013 evlist__check_mem_load_aux(session->evlist); 1014 1015 if (rep->stats_mode) 1016 return stats_print(rep); 1017 1018 if (rep->tasks_mode) 1019 return tasks_print(rep, stdout); 1020 1021 report__warn_kptr_restrict(rep); 1022 1023 evlist__for_each_entry(session->evlist, pos) 1024 rep->nr_entries += evsel__hists(pos)->nr_entries; 1025 1026 if (use_browser == 0) { 1027 if (verbose > 3) 1028 perf_session__fprintf(session, stdout); 1029 1030 if (verbose > 2) 1031 perf_session__fprintf_dsos(session, stdout); 1032 1033 if (dump_trace) { 1034 perf_session__fprintf_nr_events(session, stdout, 1035 rep->skip_empty); 1036 evlist__fprintf_nr_events(session->evlist, stdout, 1037 rep->skip_empty); 1038 return 0; 1039 } 1040 } 1041 1042 ret = report__collapse_hists(rep); 1043 if (ret) { 1044 ui__error("failed to process hist entry\n"); 1045 return ret; 1046 } 1047 1048 if (session_done()) 1049 return 0; 1050 1051 /* 1052 * recalculate number of entries after collapsing since it 1053 * might be changed during the collapse phase. 1054 */ 1055 rep->nr_entries = 0; 1056 evlist__for_each_entry(session->evlist, pos) 1057 rep->nr_entries += evsel__hists(pos)->nr_entries; 1058 1059 if (rep->nr_entries == 0) { 1060 ui__error("The %s data has no samples!\n", data->path); 1061 return 0; 1062 } 1063 1064 report__output_resort(rep); 1065 1066 if (rep->total_cycles_mode) { 1067 int block_hpps[6] = { 1068 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT, 1069 PERF_HPP_REPORT__BLOCK_LBR_CYCLES, 1070 PERF_HPP_REPORT__BLOCK_CYCLES_PCT, 1071 PERF_HPP_REPORT__BLOCK_AVG_CYCLES, 1072 PERF_HPP_REPORT__BLOCK_RANGE, 1073 PERF_HPP_REPORT__BLOCK_DSO, 1074 }; 1075 1076 rep->block_reports = block_info__create_report(session->evlist, 1077 rep->total_cycles, 1078 block_hpps, 6, 1079 &rep->nr_block_reports); 1080 if (!rep->block_reports) 1081 return -1; 1082 } 1083 1084 return report__browse_hists(rep); 1085 } 1086 1087 static int 1088 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset) 1089 { 1090 struct callchain_param *callchain = opt->value; 1091 1092 callchain->enabled = !unset; 1093 /* 1094 * --no-call-graph 1095 */ 1096 if (unset) { 1097 symbol_conf.use_callchain = false; 1098 callchain->mode = CHAIN_NONE; 1099 return 0; 1100 } 1101 1102 return parse_callchain_report_opt(arg); 1103 } 1104 1105 static int 1106 parse_time_quantum(const struct option *opt, const char *arg, 1107 int unset __maybe_unused) 1108 { 1109 unsigned long *time_q = opt->value; 1110 char *end; 1111 1112 *time_q = strtoul(arg, &end, 0); 1113 if (end == arg) 1114 goto parse_err; 1115 if (*time_q == 0) { 1116 pr_err("time quantum cannot be 0"); 1117 return -1; 1118 } 1119 end = skip_spaces(end); 1120 if (*end == 0) 1121 return 0; 1122 if (!strcmp(end, "s")) { 1123 *time_q *= NSEC_PER_SEC; 1124 return 0; 1125 } 1126 if (!strcmp(end, "ms")) { 1127 *time_q *= NSEC_PER_MSEC; 1128 return 0; 1129 } 1130 if (!strcmp(end, "us")) { 1131 *time_q *= NSEC_PER_USEC; 1132 return 0; 1133 } 1134 if (!strcmp(end, "ns")) 1135 return 0; 1136 parse_err: 1137 pr_err("Cannot parse time quantum `%s'\n", arg); 1138 return -1; 1139 } 1140 1141 int 1142 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused, 1143 const char *arg, int unset __maybe_unused) 1144 { 1145 if (arg) { 1146 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED); 1147 if (err) { 1148 char buf[BUFSIZ]; 1149 regerror(err, &ignore_callees_regex, buf, sizeof(buf)); 1150 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf); 1151 return -1; 1152 } 1153 have_ignore_callees = 1; 1154 } 1155 1156 return 0; 1157 } 1158 1159 static int 1160 parse_branch_mode(const struct option *opt, 1161 const char *str __maybe_unused, int unset) 1162 { 1163 int *branch_mode = opt->value; 1164 1165 *branch_mode = !unset; 1166 return 0; 1167 } 1168 1169 static int 1170 parse_percent_limit(const struct option *opt, const char *str, 1171 int unset __maybe_unused) 1172 { 1173 struct report *rep = opt->value; 1174 double pcnt = strtof(str, NULL); 1175 1176 rep->min_percent = pcnt; 1177 callchain_param.min_percent = pcnt; 1178 return 0; 1179 } 1180 1181 static int process_attr(struct perf_tool *tool __maybe_unused, 1182 union perf_event *event, 1183 struct evlist **pevlist) 1184 { 1185 u64 sample_type; 1186 int err; 1187 1188 err = perf_event__process_attr(tool, event, pevlist); 1189 if (err) 1190 return err; 1191 1192 /* 1193 * Check if we need to enable callchains based 1194 * on events sample_type. 1195 */ 1196 sample_type = evlist__combined_sample_type(*pevlist); 1197 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env)); 1198 return 0; 1199 } 1200 1201 int cmd_report(int argc, const char **argv) 1202 { 1203 struct perf_session *session; 1204 struct itrace_synth_opts itrace_synth_opts = { .set = 0, }; 1205 struct stat st; 1206 bool has_br_stack = false; 1207 int branch_mode = -1; 1208 int last_key = 0; 1209 bool branch_call_mode = false; 1210 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent" 1211 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n" 1212 CALLCHAIN_REPORT_HELP 1213 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT; 1214 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT; 1215 const char * const report_usage[] = { 1216 "perf report [<options>]", 1217 NULL 1218 }; 1219 struct report report = { 1220 .tool = { 1221 .sample = process_sample_event, 1222 .mmap = perf_event__process_mmap, 1223 .mmap2 = perf_event__process_mmap2, 1224 .comm = perf_event__process_comm, 1225 .namespaces = perf_event__process_namespaces, 1226 .cgroup = perf_event__process_cgroup, 1227 .exit = perf_event__process_exit, 1228 .fork = perf_event__process_fork, 1229 .lost = perf_event__process_lost, 1230 .read = process_read_event, 1231 .attr = process_attr, 1232 #ifdef HAVE_LIBTRACEEVENT 1233 .tracing_data = perf_event__process_tracing_data, 1234 #endif 1235 .build_id = perf_event__process_build_id, 1236 .id_index = perf_event__process_id_index, 1237 .auxtrace_info = perf_event__process_auxtrace_info, 1238 .auxtrace = perf_event__process_auxtrace, 1239 .event_update = perf_event__process_event_update, 1240 .feature = process_feature_event, 1241 .ordered_events = true, 1242 .ordering_requires_timestamps = true, 1243 }, 1244 .max_stack = PERF_MAX_STACK_DEPTH, 1245 .pretty_printing_style = "normal", 1246 .socket_filter = -1, 1247 .skip_empty = true, 1248 }; 1249 char *sort_order_help = sort_help("sort by key(s):"); 1250 char *field_order_help = sort_help("output field(s): overhead period sample "); 1251 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL; 1252 const struct option options[] = { 1253 OPT_STRING('i', "input", &input_name, "file", 1254 "input file name"), 1255 OPT_INCR('v', "verbose", &verbose, 1256 "be more verbose (show symbol address, etc)"), 1257 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"), 1258 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1259 "dump raw trace in ASCII"), 1260 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"), 1261 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"), 1262 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"), 1263 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 1264 "file", "vmlinux pathname"), 1265 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, 1266 "don't load vmlinux even if found"), 1267 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 1268 "file", "kallsyms pathname"), 1269 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"), 1270 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 1271 "load module symbols - WARNING: use only with -k and LIVE kernel"), 1272 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 1273 "Show a column with the number of samples"), 1274 OPT_BOOLEAN('T', "threads", &report.show_threads, 1275 "Show per-thread event counters"), 1276 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key", 1277 "pretty printing style key: normal raw"), 1278 #ifdef HAVE_SLANG_SUPPORT 1279 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"), 1280 #endif 1281 #ifdef HAVE_GTK2_SUPPORT 1282 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"), 1283 #endif 1284 OPT_BOOLEAN(0, "stdio", &report.use_stdio, 1285 "Use the stdio interface"), 1286 OPT_BOOLEAN(0, "header", &report.header, "Show data header."), 1287 OPT_BOOLEAN(0, "header-only", &report.header_only, 1288 "Show only data header."), 1289 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 1290 sort_order_help), 1291 OPT_STRING('F', "fields", &field_order, "key[,keys...]", 1292 field_order_help), 1293 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization, 1294 "Show sample percentage for different cpu modes"), 1295 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 1296 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN), 1297 OPT_STRING('p', "parent", &parent_pattern, "regex", 1298 "regex filter to identify parent, see: '--sort parent'"), 1299 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 1300 "Only display entries with parent-match"), 1301 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param, 1302 "print_type,threshold[,print_limit],order,sort_key[,branch],value", 1303 report_callchain_help, &report_parse_callchain_opt, 1304 callchain_default_opt), 1305 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, 1306 "Accumulate callchains of children and show total overhead as well. " 1307 "Enabled by default, use --no-children to disable."), 1308 OPT_INTEGER(0, "max-stack", &report.max_stack, 1309 "Set the maximum stack depth when parsing the callchain, " 1310 "anything beyond the specified depth will be ignored. " 1311 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), 1312 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain, 1313 "alias for inverted call graph"), 1314 OPT_CALLBACK(0, "ignore-callees", NULL, "regex", 1315 "ignore callees of these functions in call graphs", 1316 report_parse_ignore_callees_opt), 1317 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 1318 "only consider symbols in these dsos"), 1319 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 1320 "only consider symbols in these comms"), 1321 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 1322 "only consider symbols in these pids"), 1323 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 1324 "only consider symbols in these tids"), 1325 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 1326 "only consider these symbols"), 1327 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter", 1328 "only show symbols that (partially) match with this filter"), 1329 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 1330 "width[,width...]", 1331 "don't try to adjust column width, use these fixed values"), 1332 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator", 1333 "separator for columns, no spaces will be added between " 1334 "columns '.' is reserved."), 1335 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved, 1336 "Only display entries resolved to a symbol"), 1337 OPT_CALLBACK(0, "symfs", NULL, "directory", 1338 "Look for files with symbols relative to this directory", 1339 symbol__config_symfs), 1340 OPT_STRING('C', "cpu", &report.cpu_list, "cpu", 1341 "list of cpus to profile"), 1342 OPT_BOOLEAN('I', "show-info", &report.show_full_info, 1343 "Display extended information about perf.data file"), 1344 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src, 1345 "Interleave source code with assembly code (default)"), 1346 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw, 1347 "Display raw encoding of assembly instructions (default)"), 1348 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 1349 "Specify disassembler style (e.g. -M intel for intel syntax)"), 1350 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix", 1351 "Add prefix to source file path names in programs (with --prefix-strip)"), 1352 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N", 1353 "Strip first N entries of source file path name in programs (with --prefix)"), 1354 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 1355 "Show a column with the sum of periods"), 1356 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set, 1357 "Show event group information together"), 1358 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx, 1359 "Sort the output by the event at the index n in group. " 1360 "If n is invalid, sort by the first event. " 1361 "WARNING: should be used on grouped events."), 1362 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "", 1363 "use branch records for per branch histogram filling", 1364 parse_branch_mode), 1365 OPT_BOOLEAN(0, "branch-history", &branch_call_mode, 1366 "add last branch records to call history"), 1367 OPT_STRING(0, "objdump", &objdump_path, "path", 1368 "objdump binary to use for disassembly and annotations"), 1369 OPT_STRING(0, "addr2line", &addr2line_path, "path", 1370 "addr2line binary to use for line numbers"), 1371 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle, 1372 "Disable symbol demangling"), 1373 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, 1374 "Enable kernel symbol demangling"), 1375 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"), 1376 OPT_INTEGER(0, "samples", &symbol_conf.res_sample, 1377 "Number of samples to save per histogram entry for individual browsing"), 1378 OPT_CALLBACK(0, "percent-limit", &report, "percent", 1379 "Don't show entries under that percent", parse_percent_limit), 1380 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", 1381 "how to display percentage of filtered entries", parse_filter_percentage), 1382 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts", 1383 "Instruction Tracing options\n" ITRACE_HELP, 1384 itrace_parse_synth_opts), 1385 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename, 1386 "Show full source file name path for source lines"), 1387 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph, 1388 "Show callgraph from reference event"), 1389 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr, 1390 "Enable LBR callgraph stitching approach"), 1391 OPT_INTEGER(0, "socket-filter", &report.socket_filter, 1392 "only show processor socket that match with this filter"), 1393 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, 1394 "Show raw trace event output (do not use print fmt or plugins)"), 1395 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, 1396 "Show entries in a hierarchy"), 1397 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode", 1398 "'always' (default), 'never' or 'auto' only applicable to --stdio mode", 1399 stdio__config_color, "always"), 1400 OPT_STRING(0, "time", &report.time_str, "str", 1401 "Time span of interest (start,stop)"), 1402 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name, 1403 "Show inline function"), 1404 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period", 1405 "Set percent type local/global-period/hits", 1406 annotate_parse_percent_type), 1407 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"), 1408 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)", 1409 "Set time quantum for time sort key (default 100ms)", 1410 parse_time_quantum), 1411 OPTS_EVSWITCH(&report.evswitch), 1412 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode, 1413 "Sort all blocks by 'Sampled Cycles%'"), 1414 OPT_BOOLEAN(0, "disable-order", &report.disable_order, 1415 "Disable raw trace ordering"), 1416 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty, 1417 "Do not display empty (or dummy) events in the output"), 1418 OPT_END() 1419 }; 1420 struct perf_data data = { 1421 .mode = PERF_DATA_MODE_READ, 1422 }; 1423 int ret = hists__init(); 1424 char sort_tmp[128]; 1425 1426 if (ret < 0) 1427 goto exit; 1428 1429 annotation_options__init(&report.annotation_opts); 1430 1431 ret = perf_config(report__config, &report); 1432 if (ret) 1433 goto exit; 1434 1435 argc = parse_options(argc, argv, options, report_usage, 0); 1436 if (argc) { 1437 /* 1438 * Special case: if there's an argument left then assume that 1439 * it's a symbol filter: 1440 */ 1441 if (argc > 1) 1442 usage_with_options(report_usage, options); 1443 1444 report.symbol_filter_str = argv[0]; 1445 } 1446 1447 if (disassembler_style) { 1448 report.annotation_opts.disassembler_style = strdup(disassembler_style); 1449 if (!report.annotation_opts.disassembler_style) 1450 return -ENOMEM; 1451 } 1452 if (objdump_path) { 1453 report.annotation_opts.objdump_path = strdup(objdump_path); 1454 if (!report.annotation_opts.objdump_path) 1455 return -ENOMEM; 1456 } 1457 if (addr2line_path) { 1458 symbol_conf.addr2line_path = strdup(addr2line_path); 1459 if (!symbol_conf.addr2line_path) 1460 return -ENOMEM; 1461 } 1462 1463 if (annotate_check_args(&report.annotation_opts) < 0) { 1464 ret = -EINVAL; 1465 goto exit; 1466 } 1467 1468 if (report.mmaps_mode) 1469 report.tasks_mode = true; 1470 1471 if (dump_trace && report.disable_order) 1472 report.tool.ordered_events = false; 1473 1474 if (quiet) 1475 perf_quiet_option(); 1476 1477 ret = symbol__validate_sym_arguments(); 1478 if (ret) 1479 goto exit; 1480 1481 if (report.inverted_callchain) 1482 callchain_param.order = ORDER_CALLER; 1483 if (symbol_conf.cumulate_callchain && !callchain_param.order_set) 1484 callchain_param.order = ORDER_CALLER; 1485 1486 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) && 1487 (int)itrace_synth_opts.callchain_sz > report.max_stack) 1488 report.max_stack = itrace_synth_opts.callchain_sz; 1489 1490 if (!input_name || !strlen(input_name)) { 1491 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode)) 1492 input_name = "-"; 1493 else 1494 input_name = "perf.data"; 1495 } 1496 1497 data.path = input_name; 1498 data.force = symbol_conf.force; 1499 1500 repeat: 1501 session = perf_session__new(&data, &report.tool); 1502 if (IS_ERR(session)) { 1503 ret = PTR_ERR(session); 1504 goto exit; 1505 } 1506 1507 ret = evswitch__init(&report.evswitch, session->evlist, stderr); 1508 if (ret) 1509 goto exit; 1510 1511 if (zstd_init(&(session->zstd_data), 0) < 0) 1512 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n"); 1513 1514 if (report.queue_size) { 1515 ordered_events__set_alloc_size(&session->ordered_events, 1516 report.queue_size); 1517 } 1518 1519 session->itrace_synth_opts = &itrace_synth_opts; 1520 1521 report.session = session; 1522 1523 has_br_stack = perf_header__has_feat(&session->header, 1524 HEADER_BRANCH_STACK); 1525 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER) 1526 has_br_stack = false; 1527 1528 setup_forced_leader(&report, session->evlist); 1529 1530 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) { 1531 parse_options_usage(NULL, options, "group-sort-idx", 0); 1532 ret = -EINVAL; 1533 goto error; 1534 } 1535 1536 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch) 1537 has_br_stack = true; 1538 1539 if (has_br_stack && branch_call_mode) 1540 symbol_conf.show_branchflag_count = true; 1541 1542 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat)); 1543 1544 /* 1545 * Branch mode is a tristate: 1546 * -1 means default, so decide based on the file having branch data. 1547 * 0/1 means the user chose a mode. 1548 */ 1549 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) && 1550 !branch_call_mode) { 1551 sort__mode = SORT_MODE__BRANCH; 1552 symbol_conf.cumulate_callchain = false; 1553 } 1554 if (branch_call_mode) { 1555 callchain_param.key = CCKEY_ADDRESS; 1556 callchain_param.branch_callstack = true; 1557 symbol_conf.use_callchain = true; 1558 callchain_register_param(&callchain_param); 1559 if (sort_order == NULL) 1560 sort_order = "srcline,symbol,dso"; 1561 } 1562 1563 if (report.mem_mode) { 1564 if (sort__mode == SORT_MODE__BRANCH) { 1565 pr_err("branch and mem mode incompatible\n"); 1566 goto error; 1567 } 1568 sort__mode = SORT_MODE__MEMORY; 1569 symbol_conf.cumulate_callchain = false; 1570 } 1571 1572 if (symbol_conf.report_hierarchy) { 1573 /* disable incompatible options */ 1574 symbol_conf.cumulate_callchain = false; 1575 1576 if (field_order) { 1577 pr_err("Error: --hierarchy and --fields options cannot be used together\n"); 1578 parse_options_usage(report_usage, options, "F", 1); 1579 parse_options_usage(NULL, options, "hierarchy", 0); 1580 goto error; 1581 } 1582 1583 perf_hpp_list.need_collapse = true; 1584 } 1585 1586 if (report.use_stdio) 1587 use_browser = 0; 1588 #ifdef HAVE_SLANG_SUPPORT 1589 else if (report.use_tui) 1590 use_browser = 1; 1591 #endif 1592 #ifdef HAVE_GTK2_SUPPORT 1593 else if (report.use_gtk) 1594 use_browser = 2; 1595 #endif 1596 1597 /* Force tty output for header output and per-thread stat. */ 1598 if (report.header || report.header_only || report.show_threads) 1599 use_browser = 0; 1600 if (report.header || report.header_only) 1601 report.tool.show_feat_hdr = SHOW_FEAT_HEADER; 1602 if (report.show_full_info) 1603 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO; 1604 if (report.stats_mode || report.tasks_mode) 1605 use_browser = 0; 1606 if (report.stats_mode && report.tasks_mode) { 1607 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n"); 1608 goto error; 1609 } 1610 1611 if (report.total_cycles_mode) { 1612 if (sort__mode != SORT_MODE__BRANCH) 1613 report.total_cycles_mode = false; 1614 else 1615 sort_order = NULL; 1616 } 1617 1618 if (strcmp(input_name, "-") != 0) 1619 setup_browser(true); 1620 else 1621 use_browser = 0; 1622 1623 if (sort_order && strstr(sort_order, "ipc")) { 1624 parse_options_usage(report_usage, options, "s", 1); 1625 goto error; 1626 } 1627 1628 if (sort_order && strstr(sort_order, "symbol")) { 1629 if (sort__mode == SORT_MODE__BRANCH) { 1630 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1631 sort_order, "ipc_lbr"); 1632 report.symbol_ipc = true; 1633 } else { 1634 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s", 1635 sort_order, "ipc_null"); 1636 } 1637 1638 sort_order = sort_tmp; 1639 } 1640 1641 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) && 1642 (setup_sorting(session->evlist) < 0)) { 1643 if (sort_order) 1644 parse_options_usage(report_usage, options, "s", 1); 1645 if (field_order) 1646 parse_options_usage(sort_order ? NULL : report_usage, 1647 options, "F", 1); 1648 goto error; 1649 } 1650 1651 if ((report.header || report.header_only) && !quiet) { 1652 perf_session__fprintf_info(session, stdout, 1653 report.show_full_info); 1654 if (report.header_only) { 1655 if (data.is_pipe) { 1656 /* 1657 * we need to process first few records 1658 * which contains PERF_RECORD_HEADER_FEATURE. 1659 */ 1660 perf_session__process_events(session); 1661 } 1662 ret = 0; 1663 goto error; 1664 } 1665 } else if (use_browser == 0 && !quiet && 1666 !report.stats_mode && !report.tasks_mode) { 1667 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n", 1668 stdout); 1669 } 1670 1671 /* 1672 * Only in the TUI browser we are doing integrated annotation, 1673 * so don't allocate extra space that won't be used in the stdio 1674 * implementation. 1675 */ 1676 if (ui__has_annotation() || report.symbol_ipc || 1677 report.total_cycles_mode) { 1678 ret = symbol__annotation_init(); 1679 if (ret < 0) 1680 goto error; 1681 /* 1682 * For searching by name on the "Browse map details". 1683 * providing it only in verbose mode not to bloat too 1684 * much struct symbol. 1685 */ 1686 if (verbose > 0) { 1687 /* 1688 * XXX: Need to provide a less kludgy way to ask for 1689 * more space per symbol, the u32 is for the index on 1690 * the ui browser. 1691 * See symbol__browser_index. 1692 */ 1693 symbol_conf.priv_size += sizeof(u32); 1694 } 1695 annotation_config__init(&report.annotation_opts); 1696 } 1697 1698 if (symbol__init(&session->header.env) < 0) 1699 goto error; 1700 1701 if (report.time_str) { 1702 ret = perf_time__parse_for_ranges(report.time_str, session, 1703 &report.ptime_range, 1704 &report.range_size, 1705 &report.range_num); 1706 if (ret < 0) 1707 goto error; 1708 1709 itrace_synth_opts__set_time_range(&itrace_synth_opts, 1710 report.ptime_range, 1711 report.range_num); 1712 } 1713 1714 #ifdef HAVE_LIBTRACEEVENT 1715 if (session->tevent.pevent && 1716 tep_set_function_resolver(session->tevent.pevent, 1717 machine__resolve_kernel_addr, 1718 &session->machines.host) < 0) { 1719 pr_err("%s: failed to set libtraceevent function resolver\n", 1720 __func__); 1721 return -1; 1722 } 1723 #endif 1724 sort__setup_elide(stdout); 1725 1726 ret = __cmd_report(&report); 1727 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) { 1728 perf_session__delete(session); 1729 last_key = K_SWITCH_INPUT_DATA; 1730 goto repeat; 1731 } else 1732 ret = 0; 1733 1734 error: 1735 if (report.ptime_range) { 1736 itrace_synth_opts__clear_time_range(&itrace_synth_opts); 1737 zfree(&report.ptime_range); 1738 } 1739 1740 if (report.block_reports) { 1741 block_info__free_report(report.block_reports, 1742 report.nr_block_reports); 1743 report.block_reports = NULL; 1744 } 1745 1746 zstd_fini(&(session->zstd_data)); 1747 perf_session__delete(session); 1748 exit: 1749 annotation_options__exit(&report.annotation_opts); 1750 free(sort_order_help); 1751 free(field_order_help); 1752 return ret; 1753 } 1754