1 /* 2 * builtin-report.c 3 * 4 * Builtin report command: Analyze the perf.data input file, 5 * look up and read DSOs and symbol information and display 6 * a histogram of results, along various sorting keys. 7 */ 8 #include "builtin.h" 9 10 #include "util/util.h" 11 12 #include "util/annotate.h" 13 #include "util/color.h" 14 #include <linux/list.h> 15 #include "util/cache.h" 16 #include <linux/rbtree.h> 17 #include "util/symbol.h" 18 #include "util/callchain.h" 19 #include "util/strlist.h" 20 #include "util/values.h" 21 22 #include "perf.h" 23 #include "util/debug.h" 24 #include "util/evlist.h" 25 #include "util/evsel.h" 26 #include "util/header.h" 27 #include "util/session.h" 28 29 #include "util/parse-options.h" 30 #include "util/parse-events.h" 31 32 #include "util/thread.h" 33 #include "util/sort.h" 34 #include "util/hist.h" 35 36 #include <linux/bitmap.h> 37 38 static char const *input_name = "perf.data"; 39 40 static bool force, use_tui, use_stdio; 41 static bool hide_unresolved; 42 static bool dont_use_callchains; 43 44 static bool show_threads; 45 static struct perf_read_values show_threads_values; 46 47 static const char default_pretty_printing_style[] = "normal"; 48 static const char *pretty_printing_style = default_pretty_printing_style; 49 50 static char callchain_default_opt[] = "fractal,0.5,callee"; 51 static bool inverted_callchain; 52 static symbol_filter_t annotate_init; 53 54 static const char *cpu_list; 55 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS); 56 57 static int perf_session__add_hist_entry(struct perf_session *session, 58 struct addr_location *al, 59 struct perf_sample *sample, 60 struct perf_evsel *evsel) 61 { 62 struct symbol *parent = NULL; 63 int err = 0; 64 struct hist_entry *he; 65 66 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) { 67 err = perf_session__resolve_callchain(session, al->thread, 68 sample->callchain, &parent); 69 if (err) 70 return err; 71 } 72 73 he = __hists__add_entry(&evsel->hists, al, parent, sample->period); 74 if (he == NULL) 75 return -ENOMEM; 76 77 if (symbol_conf.use_callchain) { 78 err = callchain_append(he->callchain, &session->callchain_cursor, 79 sample->period); 80 if (err) 81 return err; 82 } 83 /* 84 * Only in the newt browser we are doing integrated annotation, 85 * so we don't allocated the extra space needed because the stdio 86 * code will not use it. 87 */ 88 if (al->sym != NULL && use_browser > 0) { 89 struct annotation *notes = symbol__annotation(he->ms.sym); 90 91 assert(evsel != NULL); 92 93 err = -ENOMEM; 94 if (notes->src == NULL && 95 symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0) 96 goto out; 97 98 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr); 99 } 100 101 evsel->hists.stats.total_period += sample->period; 102 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 103 out: 104 return err; 105 } 106 107 108 static int process_sample_event(union perf_event *event, 109 struct perf_sample *sample, 110 struct perf_evsel *evsel, 111 struct perf_session *session) 112 { 113 struct addr_location al; 114 115 if (perf_event__preprocess_sample(event, session, &al, sample, 116 annotate_init) < 0) { 117 fprintf(stderr, "problem processing %d event, skipping it.\n", 118 event->header.type); 119 return -1; 120 } 121 122 if (al.filtered || (hide_unresolved && al.sym == NULL)) 123 return 0; 124 125 if (cpu_list && !test_bit(sample->cpu, cpu_bitmap)) 126 return 0; 127 128 if (al.map != NULL) 129 al.map->dso->hit = 1; 130 131 if (perf_session__add_hist_entry(session, &al, sample, evsel)) { 132 pr_debug("problem incrementing symbol period, skipping event\n"); 133 return -1; 134 } 135 136 return 0; 137 } 138 139 static int process_read_event(union perf_event *event, 140 struct perf_sample *sample __used, 141 struct perf_session *session) 142 { 143 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist, 144 event->read.id); 145 if (show_threads) { 146 const char *name = evsel ? event_name(evsel) : "unknown"; 147 perf_read_values_add_value(&show_threads_values, 148 event->read.pid, event->read.tid, 149 event->read.id, 150 name, 151 event->read.value); 152 } 153 154 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid, 155 evsel ? event_name(evsel) : "FAIL", 156 event->read.value); 157 158 return 0; 159 } 160 161 static int perf_session__setup_sample_type(struct perf_session *self) 162 { 163 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) { 164 if (sort__has_parent) { 165 ui__warning("Selected --sort parent, but no " 166 "callchain data. Did you call " 167 "'perf record' without -g?\n"); 168 return -EINVAL; 169 } 170 if (symbol_conf.use_callchain) { 171 ui__warning("Selected -g but no callchain data. Did " 172 "you call 'perf record' without -g?\n"); 173 return -1; 174 } 175 } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE && 176 !symbol_conf.use_callchain) { 177 symbol_conf.use_callchain = true; 178 if (callchain_register_param(&callchain_param) < 0) { 179 ui__warning("Can't register callchain " 180 "params.\n"); 181 return -EINVAL; 182 } 183 } 184 185 return 0; 186 } 187 188 static struct perf_event_ops event_ops = { 189 .sample = process_sample_event, 190 .mmap = perf_event__process_mmap, 191 .comm = perf_event__process_comm, 192 .exit = perf_event__process_task, 193 .fork = perf_event__process_task, 194 .lost = perf_event__process_lost, 195 .read = process_read_event, 196 .attr = perf_event__process_attr, 197 .event_type = perf_event__process_event_type, 198 .tracing_data = perf_event__process_tracing_data, 199 .build_id = perf_event__process_build_id, 200 .ordered_samples = true, 201 .ordering_requires_timestamps = true, 202 }; 203 204 extern volatile int session_done; 205 206 static void sig_handler(int sig __used) 207 { 208 session_done = 1; 209 } 210 211 static size_t hists__fprintf_nr_sample_events(struct hists *self, 212 const char *evname, FILE *fp) 213 { 214 size_t ret; 215 char unit; 216 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE]; 217 218 nr_events = convert_unit(nr_events, &unit); 219 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit); 220 if (evname != NULL) 221 ret += fprintf(fp, " %s", evname); 222 return ret + fprintf(fp, "\n#\n"); 223 } 224 225 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist, 226 const char *help) 227 { 228 struct perf_evsel *pos; 229 230 list_for_each_entry(pos, &evlist->entries, node) { 231 struct hists *hists = &pos->hists; 232 const char *evname = event_name(pos); 233 234 hists__fprintf_nr_sample_events(hists, evname, stdout); 235 hists__fprintf(hists, NULL, false, true, 0, 0, stdout); 236 fprintf(stdout, "\n\n"); 237 } 238 239 if (sort_order == default_sort_order && 240 parent_pattern == default_parent_pattern) { 241 fprintf(stdout, "#\n# (%s)\n#\n", help); 242 243 if (show_threads) { 244 bool style = !strcmp(pretty_printing_style, "raw"); 245 perf_read_values_display(stdout, &show_threads_values, 246 style); 247 perf_read_values_destroy(&show_threads_values); 248 } 249 } 250 251 return 0; 252 } 253 254 static int __cmd_report(void) 255 { 256 int ret = -EINVAL; 257 u64 nr_samples; 258 struct perf_session *session; 259 struct perf_evsel *pos; 260 struct map *kernel_map; 261 struct kmap *kernel_kmap; 262 const char *help = "For a higher level overview, try: perf report --sort comm,dso"; 263 264 signal(SIGINT, sig_handler); 265 266 session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops); 267 if (session == NULL) 268 return -ENOMEM; 269 270 if (cpu_list) { 271 ret = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap); 272 if (ret) 273 goto out_delete; 274 } 275 276 if (show_threads) 277 perf_read_values_init(&show_threads_values); 278 279 ret = perf_session__setup_sample_type(session); 280 if (ret) 281 goto out_delete; 282 283 ret = perf_session__process_events(session, &event_ops); 284 if (ret) 285 goto out_delete; 286 287 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION]; 288 kernel_kmap = map__kmap(kernel_map); 289 if (kernel_map == NULL || 290 (kernel_map->dso->hit && 291 (kernel_kmap->ref_reloc_sym == NULL || 292 kernel_kmap->ref_reloc_sym->addr == 0))) { 293 const struct dso *kdso = kernel_map->dso; 294 295 ui__warning( 296 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n" 297 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n" 298 "Samples in kernel modules can't be resolved as well.\n\n", 299 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ? 300 "As no suitable kallsyms nor vmlinux was found, kernel samples\n" 301 "can't be resolved." : 302 "If some relocation was applied (e.g. kexec) symbols may be misresolved."); 303 } 304 305 if (dump_trace) { 306 perf_session__fprintf_nr_events(session, stdout); 307 goto out_delete; 308 } 309 310 if (verbose > 3) 311 perf_session__fprintf(session, stdout); 312 313 if (verbose > 2) 314 perf_session__fprintf_dsos(session, stdout); 315 316 nr_samples = 0; 317 list_for_each_entry(pos, &session->evlist->entries, node) { 318 struct hists *hists = &pos->hists; 319 320 hists__collapse_resort(hists); 321 hists__output_resort(hists); 322 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE]; 323 } 324 325 if (nr_samples == 0) { 326 ui__warning("The %s file has no samples!\n", input_name); 327 goto out_delete; 328 } 329 330 if (use_browser > 0) { 331 perf_evlist__tui_browse_hists(session->evlist, help, 332 NULL, NULL, 0); 333 } else 334 perf_evlist__tty_browse_hists(session->evlist, help); 335 336 out_delete: 337 /* 338 * Speed up the exit process, for large files this can 339 * take quite a while. 340 * 341 * XXX Enable this when using valgrind or if we ever 342 * librarize this command. 343 * 344 * Also experiment with obstacks to see how much speed 345 * up we'll get here. 346 * 347 * perf_session__delete(session); 348 */ 349 return ret; 350 } 351 352 static int 353 parse_callchain_opt(const struct option *opt __used, const char *arg, 354 int unset) 355 { 356 char *tok, *tok2; 357 char *endptr; 358 359 /* 360 * --no-call-graph 361 */ 362 if (unset) { 363 dont_use_callchains = true; 364 return 0; 365 } 366 367 symbol_conf.use_callchain = true; 368 369 if (!arg) 370 return 0; 371 372 tok = strtok((char *)arg, ","); 373 if (!tok) 374 return -1; 375 376 /* get the output mode */ 377 if (!strncmp(tok, "graph", strlen(arg))) 378 callchain_param.mode = CHAIN_GRAPH_ABS; 379 380 else if (!strncmp(tok, "flat", strlen(arg))) 381 callchain_param.mode = CHAIN_FLAT; 382 383 else if (!strncmp(tok, "fractal", strlen(arg))) 384 callchain_param.mode = CHAIN_GRAPH_REL; 385 386 else if (!strncmp(tok, "none", strlen(arg))) { 387 callchain_param.mode = CHAIN_NONE; 388 symbol_conf.use_callchain = false; 389 390 return 0; 391 } 392 393 else 394 return -1; 395 396 /* get the min percentage */ 397 tok = strtok(NULL, ","); 398 if (!tok) 399 goto setup; 400 401 callchain_param.min_percent = strtod(tok, &endptr); 402 if (tok == endptr) 403 return -1; 404 405 /* get the print limit */ 406 tok2 = strtok(NULL, ","); 407 if (!tok2) 408 goto setup; 409 410 if (tok2[0] != 'c') { 411 callchain_param.print_limit = strtod(tok2, &endptr); 412 tok2 = strtok(NULL, ","); 413 if (!tok2) 414 goto setup; 415 } 416 417 /* get the call chain order */ 418 if (!strcmp(tok2, "caller")) 419 callchain_param.order = ORDER_CALLER; 420 else if (!strcmp(tok2, "callee")) 421 callchain_param.order = ORDER_CALLEE; 422 else 423 return -1; 424 setup: 425 if (callchain_register_param(&callchain_param) < 0) { 426 fprintf(stderr, "Can't register callchain params\n"); 427 return -1; 428 } 429 return 0; 430 } 431 432 static const char * const report_usage[] = { 433 "perf report [<options>] <command>", 434 NULL 435 }; 436 437 static const struct option options[] = { 438 OPT_STRING('i', "input", &input_name, "file", 439 "input file name"), 440 OPT_INCR('v', "verbose", &verbose, 441 "be more verbose (show symbol address, etc)"), 442 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 443 "dump raw trace in ASCII"), 444 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 445 "file", "vmlinux pathname"), 446 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 447 "file", "kallsyms pathname"), 448 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"), 449 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules, 450 "load module symbols - WARNING: use only with -k and LIVE kernel"), 451 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, 452 "Show a column with the number of samples"), 453 OPT_BOOLEAN('T', "threads", &show_threads, 454 "Show per-thread event counters"), 455 OPT_STRING(0, "pretty", &pretty_printing_style, "key", 456 "pretty printing style key: normal raw"), 457 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"), 458 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"), 459 OPT_STRING('s', "sort", &sort_order, "key[,key2...]", 460 "sort by key(s): pid, comm, dso, symbol, parent"), 461 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization, 462 "Show sample percentage for different cpu modes"), 463 OPT_STRING('p', "parent", &parent_pattern, "regex", 464 "regex filter to identify parent, see: '--sort parent'"), 465 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other, 466 "Only display entries with parent-match"), 467 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent, call_order", 468 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. " 469 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt), 470 OPT_BOOLEAN('G', "inverted", &inverted_callchain, "alias for inverted call graph"), 471 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", 472 "only consider symbols in these dsos"), 473 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]", 474 "only consider symbols in these comms"), 475 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", 476 "only consider these symbols"), 477 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, 478 "width[,width...]", 479 "don't try to adjust column width, use these fixed values"), 480 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator", 481 "separator for columns, no spaces will be added between " 482 "columns '.' is reserved."), 483 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved, 484 "Only display entries resolved to a symbol"), 485 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 486 "Look for files with symbols relative to this directory"), 487 OPT_STRING('c', "cpu", &cpu_list, "cpu", "list of cpus to profile"), 488 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", 489 "Specify disassembler style (e.g. -M intel for intel syntax)"), 490 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, 491 "Show a column with the sum of periods"), 492 OPT_END() 493 }; 494 495 int cmd_report(int argc, const char **argv, const char *prefix __used) 496 { 497 argc = parse_options(argc, argv, options, report_usage, 0); 498 499 if (use_stdio) 500 use_browser = 0; 501 else if (use_tui) 502 use_browser = 1; 503 504 if (inverted_callchain) 505 callchain_param.order = ORDER_CALLER; 506 507 if (strcmp(input_name, "-") != 0) 508 setup_browser(true); 509 else 510 use_browser = 0; 511 /* 512 * Only in the newt browser we are doing integrated annotation, 513 * so don't allocate extra space that won't be used in the stdio 514 * implementation. 515 */ 516 if (use_browser > 0) { 517 symbol_conf.priv_size = sizeof(struct annotation); 518 annotate_init = symbol__annotate_init; 519 /* 520 * For searching by name on the "Browse map details". 521 * providing it only in verbose mode not to bloat too 522 * much struct symbol. 523 */ 524 if (verbose) { 525 /* 526 * XXX: Need to provide a less kludgy way to ask for 527 * more space per symbol, the u32 is for the index on 528 * the ui browser. 529 * See symbol__browser_index. 530 */ 531 symbol_conf.priv_size += sizeof(u32); 532 symbol_conf.sort_by_name = true; 533 } 534 } 535 536 if (symbol__init() < 0) 537 return -1; 538 539 setup_sorting(report_usage, options); 540 541 if (parent_pattern != default_parent_pattern) { 542 if (sort_dimension__add("parent") < 0) 543 return -1; 544 545 /* 546 * Only show the parent fields if we explicitly 547 * sort that way. If we only use parent machinery 548 * for filtering, we don't want it. 549 */ 550 if (!strstr(sort_order, "parent")) 551 sort_parent.elide = 1; 552 } else 553 symbol_conf.exclude_other = false; 554 555 /* 556 * Any (unrecognized) arguments left? 557 */ 558 if (argc) 559 usage_with_options(report_usage, options); 560 561 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout); 562 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout); 563 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout); 564 565 return __cmd_report(); 566 } 567