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