| 1 | /* |
| 2 | * builtin-top.c |
| 3 | * |
| 4 | * Builtin top command: Display a continuously updated profile of |
| 5 | * any workload, CPU or specific PID. |
| 6 | * |
| 7 | * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com> |
| 8 | * 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com> |
| 9 | * |
| 10 | * Improvements and fixes by: |
| 11 | * |
| 12 | * Arjan van de Ven <arjan@linux.intel.com> |
| 13 | * Yanmin Zhang <yanmin.zhang@intel.com> |
| 14 | * Wu Fengguang <fengguang.wu@intel.com> |
| 15 | * Mike Galbraith <efault@gmx.de> |
| 16 | * Paul Mackerras <paulus@samba.org> |
| 17 | * |
| 18 | * Released under the GPL v2. (and only v2, not any later version) |
| 19 | */ |
| 20 | #include "builtin.h" |
| 21 | |
| 22 | #include "perf.h" |
| 23 | |
| 24 | #include "util/annotate.h" |
| 25 | #include "util/config.h" |
| 26 | #include "util/color.h" |
| 27 | #include "util/drv_configs.h" |
| 28 | #include "util/evlist.h" |
| 29 | #include "util/evsel.h" |
| 30 | #include "util/event.h" |
| 31 | #include "util/machine.h" |
| 32 | #include "util/session.h" |
| 33 | #include "util/symbol.h" |
| 34 | #include "util/thread.h" |
| 35 | #include "util/thread_map.h" |
| 36 | #include "util/top.h" |
| 37 | #include <linux/rbtree.h> |
| 38 | #include <subcmd/parse-options.h> |
| 39 | #include "util/parse-events.h" |
| 40 | #include "util/cpumap.h" |
| 41 | #include "util/xyarray.h" |
| 42 | #include "util/sort.h" |
| 43 | #include "util/term.h" |
| 44 | #include "util/intlist.h" |
| 45 | #include "util/parse-branch-options.h" |
| 46 | #include "arch/common.h" |
| 47 | |
| 48 | #include "util/debug.h" |
| 49 | |
| 50 | #include <assert.h> |
| 51 | #include <elf.h> |
| 52 | #include <fcntl.h> |
| 53 | |
| 54 | #include <stdio.h> |
| 55 | #include <termios.h> |
| 56 | #include <unistd.h> |
| 57 | #include <inttypes.h> |
| 58 | |
| 59 | #include <errno.h> |
| 60 | #include <time.h> |
| 61 | #include <sched.h> |
| 62 | #include <signal.h> |
| 63 | |
| 64 | #include <sys/syscall.h> |
| 65 | #include <sys/ioctl.h> |
| 66 | #include <poll.h> |
| 67 | #include <sys/prctl.h> |
| 68 | #include <sys/wait.h> |
| 69 | #include <sys/uio.h> |
| 70 | #include <sys/utsname.h> |
| 71 | #include <sys/mman.h> |
| 72 | |
| 73 | #include <linux/stringify.h> |
| 74 | #include <linux/time64.h> |
| 75 | #include <linux/types.h> |
| 76 | |
| 77 | #include "sane_ctype.h" |
| 78 | |
| 79 | static volatile int done; |
| 80 | static volatile int resize; |
| 81 | |
| 82 | #define HEADER_LINE_NR 5 |
| 83 | |
| 84 | static void perf_top__update_print_entries(struct perf_top *top) |
| 85 | { |
| 86 | top->print_entries = top->winsize.ws_row - HEADER_LINE_NR; |
| 87 | } |
| 88 | |
| 89 | static void winch_sig(int sig __maybe_unused) |
| 90 | { |
| 91 | resize = 1; |
| 92 | } |
| 93 | |
| 94 | static void perf_top__resize(struct perf_top *top) |
| 95 | { |
| 96 | get_term_dimensions(&top->winsize); |
| 97 | perf_top__update_print_entries(top); |
| 98 | } |
| 99 | |
| 100 | static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he) |
| 101 | { |
| 102 | struct perf_evsel *evsel = hists_to_evsel(he->hists); |
| 103 | struct symbol *sym; |
| 104 | struct annotation *notes; |
| 105 | struct map *map; |
| 106 | int err = -1; |
| 107 | |
| 108 | if (!he || !he->ms.sym) |
| 109 | return -1; |
| 110 | |
| 111 | sym = he->ms.sym; |
| 112 | map = he->ms.map; |
| 113 | |
| 114 | /* |
| 115 | * We can't annotate with just /proc/kallsyms |
| 116 | */ |
| 117 | if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS && |
| 118 | !dso__is_kcore(map->dso)) { |
| 119 | pr_err("Can't annotate %s: No vmlinux file was found in the " |
| 120 | "path\n", sym->name); |
| 121 | sleep(1); |
| 122 | return -1; |
| 123 | } |
| 124 | |
| 125 | notes = symbol__annotation(sym); |
| 126 | if (notes->src != NULL) { |
| 127 | pthread_mutex_lock(¬es->lock); |
| 128 | goto out_assign; |
| 129 | } |
| 130 | |
| 131 | pthread_mutex_lock(¬es->lock); |
| 132 | |
| 133 | if (symbol__alloc_hist(sym) < 0) { |
| 134 | pthread_mutex_unlock(¬es->lock); |
| 135 | pr_err("Not enough memory for annotating '%s' symbol!\n", |
| 136 | sym->name); |
| 137 | sleep(1); |
| 138 | return err; |
| 139 | } |
| 140 | |
| 141 | err = symbol__annotate(sym, map, evsel, 0, NULL); |
| 142 | if (err == 0) { |
| 143 | out_assign: |
| 144 | top->sym_filter_entry = he; |
| 145 | } else { |
| 146 | char msg[BUFSIZ]; |
| 147 | symbol__strerror_disassemble(sym, map, err, msg, sizeof(msg)); |
| 148 | pr_err("Couldn't annotate %s: %s\n", sym->name, msg); |
| 149 | } |
| 150 | |
| 151 | pthread_mutex_unlock(¬es->lock); |
| 152 | return err; |
| 153 | } |
| 154 | |
| 155 | static void __zero_source_counters(struct hist_entry *he) |
| 156 | { |
| 157 | struct symbol *sym = he->ms.sym; |
| 158 | symbol__annotate_zero_histograms(sym); |
| 159 | } |
| 160 | |
| 161 | static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip) |
| 162 | { |
| 163 | struct utsname uts; |
| 164 | int err = uname(&uts); |
| 165 | |
| 166 | ui__warning("Out of bounds address found:\n\n" |
| 167 | "Addr: %" PRIx64 "\n" |
| 168 | "DSO: %s %c\n" |
| 169 | "Map: %" PRIx64 "-%" PRIx64 "\n" |
| 170 | "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n" |
| 171 | "Arch: %s\n" |
| 172 | "Kernel: %s\n" |
| 173 | "Tools: %s\n\n" |
| 174 | "Not all samples will be on the annotation output.\n\n" |
| 175 | "Please report to linux-kernel@vger.kernel.org\n", |
| 176 | ip, map->dso->long_name, dso__symtab_origin(map->dso), |
| 177 | map->start, map->end, sym->start, sym->end, |
| 178 | sym->binding == STB_GLOBAL ? 'g' : |
| 179 | sym->binding == STB_LOCAL ? 'l' : 'w', sym->name, |
| 180 | err ? "[unknown]" : uts.machine, |
| 181 | err ? "[unknown]" : uts.release, perf_version_string); |
| 182 | if (use_browser <= 0) |
| 183 | sleep(5); |
| 184 | |
| 185 | map->erange_warned = true; |
| 186 | } |
| 187 | |
| 188 | static void perf_top__record_precise_ip(struct perf_top *top, |
| 189 | struct hist_entry *he, |
| 190 | struct perf_sample *sample, |
| 191 | int counter, u64 ip) |
| 192 | { |
| 193 | struct annotation *notes; |
| 194 | struct symbol *sym = he->ms.sym; |
| 195 | int err = 0; |
| 196 | |
| 197 | if (sym == NULL || (use_browser == 0 && |
| 198 | (top->sym_filter_entry == NULL || |
| 199 | top->sym_filter_entry->ms.sym != sym))) |
| 200 | return; |
| 201 | |
| 202 | notes = symbol__annotation(sym); |
| 203 | |
| 204 | if (pthread_mutex_trylock(¬es->lock)) |
| 205 | return; |
| 206 | |
| 207 | err = hist_entry__inc_addr_samples(he, sample, counter, ip); |
| 208 | |
| 209 | pthread_mutex_unlock(¬es->lock); |
| 210 | |
| 211 | if (unlikely(err)) { |
| 212 | /* |
| 213 | * This function is now called with he->hists->lock held. |
| 214 | * Release it before going to sleep. |
| 215 | */ |
| 216 | pthread_mutex_unlock(&he->hists->lock); |
| 217 | |
| 218 | if (err == -ERANGE && !he->ms.map->erange_warned) |
| 219 | ui__warn_map_erange(he->ms.map, sym, ip); |
| 220 | else if (err == -ENOMEM) { |
| 221 | pr_err("Not enough memory for annotating '%s' symbol!\n", |
| 222 | sym->name); |
| 223 | sleep(1); |
| 224 | } |
| 225 | |
| 226 | pthread_mutex_lock(&he->hists->lock); |
| 227 | } |
| 228 | } |
| 229 | |
| 230 | static void perf_top__show_details(struct perf_top *top) |
| 231 | { |
| 232 | struct hist_entry *he = top->sym_filter_entry; |
| 233 | struct perf_evsel *evsel = hists_to_evsel(he->hists); |
| 234 | struct annotation *notes; |
| 235 | struct symbol *symbol; |
| 236 | int more; |
| 237 | |
| 238 | if (!he) |
| 239 | return; |
| 240 | |
| 241 | symbol = he->ms.sym; |
| 242 | notes = symbol__annotation(symbol); |
| 243 | |
| 244 | pthread_mutex_lock(¬es->lock); |
| 245 | |
| 246 | symbol__calc_percent(symbol, evsel); |
| 247 | |
| 248 | if (notes->src == NULL) |
| 249 | goto out_unlock; |
| 250 | |
| 251 | printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name); |
| 252 | printf(" Events Pcnt (>=%d%%)\n", top->sym_pcnt_filter); |
| 253 | |
| 254 | more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel, |
| 255 | 0, top->sym_pcnt_filter, top->print_entries, 4); |
| 256 | |
| 257 | if (top->evlist->enabled) { |
| 258 | if (top->zero) |
| 259 | symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx); |
| 260 | else |
| 261 | symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx); |
| 262 | } |
| 263 | if (more != 0) |
| 264 | printf("%d lines not displayed, maybe increase display entries [e]\n", more); |
| 265 | out_unlock: |
| 266 | pthread_mutex_unlock(¬es->lock); |
| 267 | } |
| 268 | |
| 269 | static void perf_top__print_sym_table(struct perf_top *top) |
| 270 | { |
| 271 | char bf[160]; |
| 272 | int printed = 0; |
| 273 | const int win_width = top->winsize.ws_col - 1; |
| 274 | struct perf_evsel *evsel = top->sym_evsel; |
| 275 | struct hists *hists = evsel__hists(evsel); |
| 276 | |
| 277 | puts(CONSOLE_CLEAR); |
| 278 | |
| 279 | perf_top__header_snprintf(top, bf, sizeof(bf)); |
| 280 | printf("%s\n", bf); |
| 281 | |
| 282 | perf_top__reset_sample_counters(top); |
| 283 | |
| 284 | printf("%-*.*s\n", win_width, win_width, graph_dotted_line); |
| 285 | |
| 286 | if (!top->record_opts.overwrite && |
| 287 | (hists->stats.nr_lost_warned != |
| 288 | hists->stats.nr_events[PERF_RECORD_LOST])) { |
| 289 | hists->stats.nr_lost_warned = |
| 290 | hists->stats.nr_events[PERF_RECORD_LOST]; |
| 291 | color_fprintf(stdout, PERF_COLOR_RED, |
| 292 | "WARNING: LOST %d chunks, Check IO/CPU overload", |
| 293 | hists->stats.nr_lost_warned); |
| 294 | ++printed; |
| 295 | } |
| 296 | |
| 297 | if (top->sym_filter_entry) { |
| 298 | perf_top__show_details(top); |
| 299 | return; |
| 300 | } |
| 301 | |
| 302 | if (top->evlist->enabled) { |
| 303 | if (top->zero) { |
| 304 | hists__delete_entries(hists); |
| 305 | } else { |
| 306 | hists__decay_entries(hists, top->hide_user_symbols, |
| 307 | top->hide_kernel_symbols); |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | hists__collapse_resort(hists, NULL); |
| 312 | perf_evsel__output_resort(evsel, NULL); |
| 313 | |
| 314 | hists__output_recalc_col_len(hists, top->print_entries - printed); |
| 315 | putchar('\n'); |
| 316 | hists__fprintf(hists, false, top->print_entries - printed, win_width, |
| 317 | top->min_percent, stdout, symbol_conf.use_callchain); |
| 318 | } |
| 319 | |
| 320 | static void prompt_integer(int *target, const char *msg) |
| 321 | { |
| 322 | char *buf = malloc(0), *p; |
| 323 | size_t dummy = 0; |
| 324 | int tmp; |
| 325 | |
| 326 | fprintf(stdout, "\n%s: ", msg); |
| 327 | if (getline(&buf, &dummy, stdin) < 0) |
| 328 | return; |
| 329 | |
| 330 | p = strchr(buf, '\n'); |
| 331 | if (p) |
| 332 | *p = 0; |
| 333 | |
| 334 | p = buf; |
| 335 | while(*p) { |
| 336 | if (!isdigit(*p)) |
| 337 | goto out_free; |
| 338 | p++; |
| 339 | } |
| 340 | tmp = strtoul(buf, NULL, 10); |
| 341 | *target = tmp; |
| 342 | out_free: |
| 343 | free(buf); |
| 344 | } |
| 345 | |
| 346 | static void prompt_percent(int *target, const char *msg) |
| 347 | { |
| 348 | int tmp = 0; |
| 349 | |
| 350 | prompt_integer(&tmp, msg); |
| 351 | if (tmp >= 0 && tmp <= 100) |
| 352 | *target = tmp; |
| 353 | } |
| 354 | |
| 355 | static void perf_top__prompt_symbol(struct perf_top *top, const char *msg) |
| 356 | { |
| 357 | char *buf = malloc(0), *p; |
| 358 | struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL; |
| 359 | struct hists *hists = evsel__hists(top->sym_evsel); |
| 360 | struct rb_node *next; |
| 361 | size_t dummy = 0; |
| 362 | |
| 363 | /* zero counters of active symbol */ |
| 364 | if (syme) { |
| 365 | __zero_source_counters(syme); |
| 366 | top->sym_filter_entry = NULL; |
| 367 | } |
| 368 | |
| 369 | fprintf(stdout, "\n%s: ", msg); |
| 370 | if (getline(&buf, &dummy, stdin) < 0) |
| 371 | goto out_free; |
| 372 | |
| 373 | p = strchr(buf, '\n'); |
| 374 | if (p) |
| 375 | *p = 0; |
| 376 | |
| 377 | next = rb_first(&hists->entries); |
| 378 | while (next) { |
| 379 | n = rb_entry(next, struct hist_entry, rb_node); |
| 380 | if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) { |
| 381 | found = n; |
| 382 | break; |
| 383 | } |
| 384 | next = rb_next(&n->rb_node); |
| 385 | } |
| 386 | |
| 387 | if (!found) { |
| 388 | fprintf(stderr, "Sorry, %s is not active.\n", buf); |
| 389 | sleep(1); |
| 390 | } else |
| 391 | perf_top__parse_source(top, found); |
| 392 | |
| 393 | out_free: |
| 394 | free(buf); |
| 395 | } |
| 396 | |
| 397 | static void perf_top__print_mapped_keys(struct perf_top *top) |
| 398 | { |
| 399 | char *name = NULL; |
| 400 | |
| 401 | if (top->sym_filter_entry) { |
| 402 | struct symbol *sym = top->sym_filter_entry->ms.sym; |
| 403 | name = sym->name; |
| 404 | } |
| 405 | |
| 406 | fprintf(stdout, "\nMapped keys:\n"); |
| 407 | fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top->delay_secs); |
| 408 | fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top->print_entries); |
| 409 | |
| 410 | if (top->evlist->nr_entries > 1) |
| 411 | fprintf(stdout, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top->sym_evsel)); |
| 412 | |
| 413 | fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top->count_filter); |
| 414 | |
| 415 | fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter); |
| 416 | fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL"); |
| 417 | fprintf(stdout, "\t[S] stop annotation.\n"); |
| 418 | |
| 419 | fprintf(stdout, |
| 420 | "\t[K] hide kernel symbols. \t(%s)\n", |
| 421 | top->hide_kernel_symbols ? "yes" : "no"); |
| 422 | fprintf(stdout, |
| 423 | "\t[U] hide user symbols. \t(%s)\n", |
| 424 | top->hide_user_symbols ? "yes" : "no"); |
| 425 | fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top->zero ? 1 : 0); |
| 426 | fprintf(stdout, "\t[qQ] quit.\n"); |
| 427 | } |
| 428 | |
| 429 | static int perf_top__key_mapped(struct perf_top *top, int c) |
| 430 | { |
| 431 | switch (c) { |
| 432 | case 'd': |
| 433 | case 'e': |
| 434 | case 'f': |
| 435 | case 'z': |
| 436 | case 'q': |
| 437 | case 'Q': |
| 438 | case 'K': |
| 439 | case 'U': |
| 440 | case 'F': |
| 441 | case 's': |
| 442 | case 'S': |
| 443 | return 1; |
| 444 | case 'E': |
| 445 | return top->evlist->nr_entries > 1 ? 1 : 0; |
| 446 | default: |
| 447 | break; |
| 448 | } |
| 449 | |
| 450 | return 0; |
| 451 | } |
| 452 | |
| 453 | static bool perf_top__handle_keypress(struct perf_top *top, int c) |
| 454 | { |
| 455 | bool ret = true; |
| 456 | |
| 457 | if (!perf_top__key_mapped(top, c)) { |
| 458 | struct pollfd stdin_poll = { .fd = 0, .events = POLLIN }; |
| 459 | struct termios save; |
| 460 | |
| 461 | perf_top__print_mapped_keys(top); |
| 462 | fprintf(stdout, "\nEnter selection, or unmapped key to continue: "); |
| 463 | fflush(stdout); |
| 464 | |
| 465 | set_term_quiet_input(&save); |
| 466 | |
| 467 | poll(&stdin_poll, 1, -1); |
| 468 | c = getc(stdin); |
| 469 | |
| 470 | tcsetattr(0, TCSAFLUSH, &save); |
| 471 | if (!perf_top__key_mapped(top, c)) |
| 472 | return ret; |
| 473 | } |
| 474 | |
| 475 | switch (c) { |
| 476 | case 'd': |
| 477 | prompt_integer(&top->delay_secs, "Enter display delay"); |
| 478 | if (top->delay_secs < 1) |
| 479 | top->delay_secs = 1; |
| 480 | break; |
| 481 | case 'e': |
| 482 | prompt_integer(&top->print_entries, "Enter display entries (lines)"); |
| 483 | if (top->print_entries == 0) { |
| 484 | perf_top__resize(top); |
| 485 | signal(SIGWINCH, winch_sig); |
| 486 | } else { |
| 487 | signal(SIGWINCH, SIG_DFL); |
| 488 | } |
| 489 | break; |
| 490 | case 'E': |
| 491 | if (top->evlist->nr_entries > 1) { |
| 492 | /* Select 0 as the default event: */ |
| 493 | int counter = 0; |
| 494 | |
| 495 | fprintf(stderr, "\nAvailable events:"); |
| 496 | |
| 497 | evlist__for_each_entry(top->evlist, top->sym_evsel) |
| 498 | fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel)); |
| 499 | |
| 500 | prompt_integer(&counter, "Enter details event counter"); |
| 501 | |
| 502 | if (counter >= top->evlist->nr_entries) { |
| 503 | top->sym_evsel = perf_evlist__first(top->evlist); |
| 504 | fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel)); |
| 505 | sleep(1); |
| 506 | break; |
| 507 | } |
| 508 | evlist__for_each_entry(top->evlist, top->sym_evsel) |
| 509 | if (top->sym_evsel->idx == counter) |
| 510 | break; |
| 511 | } else |
| 512 | top->sym_evsel = perf_evlist__first(top->evlist); |
| 513 | break; |
| 514 | case 'f': |
| 515 | prompt_integer(&top->count_filter, "Enter display event count filter"); |
| 516 | break; |
| 517 | case 'F': |
| 518 | prompt_percent(&top->sym_pcnt_filter, |
| 519 | "Enter details display event filter (percent)"); |
| 520 | break; |
| 521 | case 'K': |
| 522 | top->hide_kernel_symbols = !top->hide_kernel_symbols; |
| 523 | break; |
| 524 | case 'q': |
| 525 | case 'Q': |
| 526 | printf("exiting.\n"); |
| 527 | if (top->dump_symtab) |
| 528 | perf_session__fprintf_dsos(top->session, stderr); |
| 529 | ret = false; |
| 530 | break; |
| 531 | case 's': |
| 532 | perf_top__prompt_symbol(top, "Enter details symbol"); |
| 533 | break; |
| 534 | case 'S': |
| 535 | if (!top->sym_filter_entry) |
| 536 | break; |
| 537 | else { |
| 538 | struct hist_entry *syme = top->sym_filter_entry; |
| 539 | |
| 540 | top->sym_filter_entry = NULL; |
| 541 | __zero_source_counters(syme); |
| 542 | } |
| 543 | break; |
| 544 | case 'U': |
| 545 | top->hide_user_symbols = !top->hide_user_symbols; |
| 546 | break; |
| 547 | case 'z': |
| 548 | top->zero = !top->zero; |
| 549 | break; |
| 550 | default: |
| 551 | break; |
| 552 | } |
| 553 | |
| 554 | return ret; |
| 555 | } |
| 556 | |
| 557 | static void perf_top__sort_new_samples(void *arg) |
| 558 | { |
| 559 | struct perf_top *t = arg; |
| 560 | struct perf_evsel *evsel = t->sym_evsel; |
| 561 | struct hists *hists; |
| 562 | |
| 563 | perf_top__reset_sample_counters(t); |
| 564 | |
| 565 | if (t->evlist->selected != NULL) |
| 566 | t->sym_evsel = t->evlist->selected; |
| 567 | |
| 568 | hists = evsel__hists(evsel); |
| 569 | |
| 570 | if (t->evlist->enabled) { |
| 571 | if (t->zero) { |
| 572 | hists__delete_entries(hists); |
| 573 | } else { |
| 574 | hists__decay_entries(hists, t->hide_user_symbols, |
| 575 | t->hide_kernel_symbols); |
| 576 | } |
| 577 | } |
| 578 | |
| 579 | hists__collapse_resort(hists, NULL); |
| 580 | perf_evsel__output_resort(evsel, NULL); |
| 581 | } |
| 582 | |
| 583 | static void *display_thread_tui(void *arg) |
| 584 | { |
| 585 | struct perf_evsel *pos; |
| 586 | struct perf_top *top = arg; |
| 587 | const char *help = "For a higher level overview, try: perf top --sort comm,dso"; |
| 588 | struct hist_browser_timer hbt = { |
| 589 | .timer = perf_top__sort_new_samples, |
| 590 | .arg = top, |
| 591 | .refresh = top->delay_secs, |
| 592 | }; |
| 593 | |
| 594 | /* In order to read symbols from other namespaces perf to needs to call |
| 595 | * setns(2). This isn't permitted if the struct_fs has multiple users. |
| 596 | * unshare(2) the fs so that we may continue to setns into namespaces |
| 597 | * that we're observing. |
| 598 | */ |
| 599 | unshare(CLONE_FS); |
| 600 | |
| 601 | perf_top__sort_new_samples(top); |
| 602 | |
| 603 | /* |
| 604 | * Initialize the uid_filter_str, in the future the TUI will allow |
| 605 | * Zooming in/out UIDs. For now juse use whatever the user passed |
| 606 | * via --uid. |
| 607 | */ |
| 608 | evlist__for_each_entry(top->evlist, pos) { |
| 609 | struct hists *hists = evsel__hists(pos); |
| 610 | hists->uid_filter_str = top->record_opts.target.uid_str; |
| 611 | } |
| 612 | |
| 613 | perf_evlist__tui_browse_hists(top->evlist, help, &hbt, |
| 614 | top->min_percent, |
| 615 | &top->session->header.env, |
| 616 | !top->record_opts.overwrite); |
| 617 | |
| 618 | done = 1; |
| 619 | return NULL; |
| 620 | } |
| 621 | |
| 622 | static void display_sig(int sig __maybe_unused) |
| 623 | { |
| 624 | done = 1; |
| 625 | } |
| 626 | |
| 627 | static void display_setup_sig(void) |
| 628 | { |
| 629 | signal(SIGSEGV, sighandler_dump_stack); |
| 630 | signal(SIGFPE, sighandler_dump_stack); |
| 631 | signal(SIGINT, display_sig); |
| 632 | signal(SIGQUIT, display_sig); |
| 633 | signal(SIGTERM, display_sig); |
| 634 | } |
| 635 | |
| 636 | static void *display_thread(void *arg) |
| 637 | { |
| 638 | struct pollfd stdin_poll = { .fd = 0, .events = POLLIN }; |
| 639 | struct termios save; |
| 640 | struct perf_top *top = arg; |
| 641 | int delay_msecs, c; |
| 642 | |
| 643 | /* In order to read symbols from other namespaces perf to needs to call |
| 644 | * setns(2). This isn't permitted if the struct_fs has multiple users. |
| 645 | * unshare(2) the fs so that we may continue to setns into namespaces |
| 646 | * that we're observing. |
| 647 | */ |
| 648 | unshare(CLONE_FS); |
| 649 | |
| 650 | display_setup_sig(); |
| 651 | pthread__unblock_sigwinch(); |
| 652 | repeat: |
| 653 | delay_msecs = top->delay_secs * MSEC_PER_SEC; |
| 654 | set_term_quiet_input(&save); |
| 655 | /* trash return*/ |
| 656 | getc(stdin); |
| 657 | |
| 658 | while (!done) { |
| 659 | perf_top__print_sym_table(top); |
| 660 | /* |
| 661 | * Either timeout expired or we got an EINTR due to SIGWINCH, |
| 662 | * refresh screen in both cases. |
| 663 | */ |
| 664 | switch (poll(&stdin_poll, 1, delay_msecs)) { |
| 665 | case 0: |
| 666 | continue; |
| 667 | case -1: |
| 668 | if (errno == EINTR) |
| 669 | continue; |
| 670 | __fallthrough; |
| 671 | default: |
| 672 | c = getc(stdin); |
| 673 | tcsetattr(0, TCSAFLUSH, &save); |
| 674 | |
| 675 | if (perf_top__handle_keypress(top, c)) |
| 676 | goto repeat; |
| 677 | done = 1; |
| 678 | } |
| 679 | } |
| 680 | |
| 681 | tcsetattr(0, TCSAFLUSH, &save); |
| 682 | return NULL; |
| 683 | } |
| 684 | |
| 685 | static int hist_iter__top_callback(struct hist_entry_iter *iter, |
| 686 | struct addr_location *al, bool single, |
| 687 | void *arg) |
| 688 | { |
| 689 | struct perf_top *top = arg; |
| 690 | struct hist_entry *he = iter->he; |
| 691 | struct perf_evsel *evsel = iter->evsel; |
| 692 | |
| 693 | if (perf_hpp_list.sym && single) |
| 694 | perf_top__record_precise_ip(top, he, iter->sample, evsel->idx, al->addr); |
| 695 | |
| 696 | hist__account_cycles(iter->sample->branch_stack, al, iter->sample, |
| 697 | !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY)); |
| 698 | return 0; |
| 699 | } |
| 700 | |
| 701 | static void perf_event__process_sample(struct perf_tool *tool, |
| 702 | const union perf_event *event, |
| 703 | struct perf_evsel *evsel, |
| 704 | struct perf_sample *sample, |
| 705 | struct machine *machine) |
| 706 | { |
| 707 | struct perf_top *top = container_of(tool, struct perf_top, tool); |
| 708 | struct addr_location al; |
| 709 | int err; |
| 710 | |
| 711 | if (!machine && perf_guest) { |
| 712 | static struct intlist *seen; |
| 713 | |
| 714 | if (!seen) |
| 715 | seen = intlist__new(NULL); |
| 716 | |
| 717 | if (!intlist__has_entry(seen, sample->pid)) { |
| 718 | pr_err("Can't find guest [%d]'s kernel information\n", |
| 719 | sample->pid); |
| 720 | intlist__add(seen, sample->pid); |
| 721 | } |
| 722 | return; |
| 723 | } |
| 724 | |
| 725 | if (!machine) { |
| 726 | pr_err("%u unprocessable samples recorded.\r", |
| 727 | top->session->evlist->stats.nr_unprocessable_samples++); |
| 728 | return; |
| 729 | } |
| 730 | |
| 731 | if (event->header.misc & PERF_RECORD_MISC_EXACT_IP) |
| 732 | top->exact_samples++; |
| 733 | |
| 734 | if (machine__resolve(machine, &al, sample) < 0) |
| 735 | return; |
| 736 | |
| 737 | if (!machine->kptr_restrict_warned && |
| 738 | symbol_conf.kptr_restrict && |
| 739 | al.cpumode == PERF_RECORD_MISC_KERNEL) { |
| 740 | if (!perf_evlist__exclude_kernel(top->session->evlist)) { |
| 741 | ui__warning( |
| 742 | "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n" |
| 743 | "Check /proc/sys/kernel/kptr_restrict.\n\n" |
| 744 | "Kernel%s samples will not be resolved.\n", |
| 745 | al.map && !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ? |
| 746 | " modules" : ""); |
| 747 | if (use_browser <= 0) |
| 748 | sleep(5); |
| 749 | } |
| 750 | machine->kptr_restrict_warned = true; |
| 751 | } |
| 752 | |
| 753 | if (al.sym == NULL) { |
| 754 | const char *msg = "Kernel samples will not be resolved.\n"; |
| 755 | /* |
| 756 | * As we do lazy loading of symtabs we only will know if the |
| 757 | * specified vmlinux file is invalid when we actually have a |
| 758 | * hit in kernel space and then try to load it. So if we get |
| 759 | * here and there are _no_ symbols in the DSO backing the |
| 760 | * kernel map, bail out. |
| 761 | * |
| 762 | * We may never get here, for instance, if we use -K/ |
| 763 | * --hide-kernel-symbols, even if the user specifies an |
| 764 | * invalid --vmlinux ;-) |
| 765 | */ |
| 766 | if (!machine->kptr_restrict_warned && !top->vmlinux_warned && |
| 767 | al.map == machine->vmlinux_maps[MAP__FUNCTION] && |
| 768 | RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) { |
| 769 | if (symbol_conf.vmlinux_name) { |
| 770 | char serr[256]; |
| 771 | dso__strerror_load(al.map->dso, serr, sizeof(serr)); |
| 772 | ui__warning("The %s file can't be used: %s\n%s", |
| 773 | symbol_conf.vmlinux_name, serr, msg); |
| 774 | } else { |
| 775 | ui__warning("A vmlinux file was not found.\n%s", |
| 776 | msg); |
| 777 | } |
| 778 | |
| 779 | if (use_browser <= 0) |
| 780 | sleep(5); |
| 781 | top->vmlinux_warned = true; |
| 782 | } |
| 783 | } |
| 784 | |
| 785 | if (al.sym == NULL || !al.sym->idle) { |
| 786 | struct hists *hists = evsel__hists(evsel); |
| 787 | struct hist_entry_iter iter = { |
| 788 | .evsel = evsel, |
| 789 | .sample = sample, |
| 790 | .add_entry_cb = hist_iter__top_callback, |
| 791 | }; |
| 792 | |
| 793 | if (symbol_conf.cumulate_callchain) |
| 794 | iter.ops = &hist_iter_cumulative; |
| 795 | else |
| 796 | iter.ops = &hist_iter_normal; |
| 797 | |
| 798 | pthread_mutex_lock(&hists->lock); |
| 799 | |
| 800 | err = hist_entry_iter__add(&iter, &al, top->max_stack, top); |
| 801 | if (err < 0) |
| 802 | pr_err("Problem incrementing symbol period, skipping event\n"); |
| 803 | |
| 804 | pthread_mutex_unlock(&hists->lock); |
| 805 | } |
| 806 | |
| 807 | addr_location__put(&al); |
| 808 | } |
| 809 | |
| 810 | static void perf_top__mmap_read_idx(struct perf_top *top, int idx) |
| 811 | { |
| 812 | struct record_opts *opts = &top->record_opts; |
| 813 | struct perf_evlist *evlist = top->evlist; |
| 814 | struct perf_sample sample; |
| 815 | struct perf_evsel *evsel; |
| 816 | struct perf_mmap *md; |
| 817 | struct perf_session *session = top->session; |
| 818 | union perf_event *event; |
| 819 | struct machine *machine; |
| 820 | u64 end, start; |
| 821 | int ret; |
| 822 | |
| 823 | md = opts->overwrite ? &evlist->overwrite_mmap[idx] : &evlist->mmap[idx]; |
| 824 | if (perf_mmap__read_init(md, opts->overwrite, &start, &end) < 0) |
| 825 | return; |
| 826 | |
| 827 | while ((event = perf_mmap__read_event(md, opts->overwrite, &start, end)) != NULL) { |
| 828 | ret = perf_evlist__parse_sample(evlist, event, &sample); |
| 829 | if (ret) { |
| 830 | pr_err("Can't parse sample, err = %d\n", ret); |
| 831 | goto next_event; |
| 832 | } |
| 833 | |
| 834 | evsel = perf_evlist__id2evsel(session->evlist, sample.id); |
| 835 | assert(evsel != NULL); |
| 836 | |
| 837 | if (event->header.type == PERF_RECORD_SAMPLE) |
| 838 | ++top->samples; |
| 839 | |
| 840 | switch (sample.cpumode) { |
| 841 | case PERF_RECORD_MISC_USER: |
| 842 | ++top->us_samples; |
| 843 | if (top->hide_user_symbols) |
| 844 | goto next_event; |
| 845 | machine = &session->machines.host; |
| 846 | break; |
| 847 | case PERF_RECORD_MISC_KERNEL: |
| 848 | ++top->kernel_samples; |
| 849 | if (top->hide_kernel_symbols) |
| 850 | goto next_event; |
| 851 | machine = &session->machines.host; |
| 852 | break; |
| 853 | case PERF_RECORD_MISC_GUEST_KERNEL: |
| 854 | ++top->guest_kernel_samples; |
| 855 | machine = perf_session__find_machine(session, |
| 856 | sample.pid); |
| 857 | break; |
| 858 | case PERF_RECORD_MISC_GUEST_USER: |
| 859 | ++top->guest_us_samples; |
| 860 | /* |
| 861 | * TODO: we don't process guest user from host side |
| 862 | * except simple counting. |
| 863 | */ |
| 864 | goto next_event; |
| 865 | default: |
| 866 | if (event->header.type == PERF_RECORD_SAMPLE) |
| 867 | goto next_event; |
| 868 | machine = &session->machines.host; |
| 869 | break; |
| 870 | } |
| 871 | |
| 872 | |
| 873 | if (event->header.type == PERF_RECORD_SAMPLE) { |
| 874 | perf_event__process_sample(&top->tool, event, evsel, |
| 875 | &sample, machine); |
| 876 | } else if (event->header.type < PERF_RECORD_MAX) { |
| 877 | hists__inc_nr_events(evsel__hists(evsel), event->header.type); |
| 878 | machine__process_event(machine, event, &sample); |
| 879 | } else |
| 880 | ++session->evlist->stats.nr_unknown_events; |
| 881 | next_event: |
| 882 | perf_mmap__consume(md); |
| 883 | } |
| 884 | |
| 885 | perf_mmap__read_done(md); |
| 886 | } |
| 887 | |
| 888 | static void perf_top__mmap_read(struct perf_top *top) |
| 889 | { |
| 890 | bool overwrite = top->record_opts.overwrite; |
| 891 | struct perf_evlist *evlist = top->evlist; |
| 892 | unsigned long long start, end; |
| 893 | int i; |
| 894 | |
| 895 | start = rdclock(); |
| 896 | if (overwrite) |
| 897 | perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_DATA_PENDING); |
| 898 | |
| 899 | for (i = 0; i < top->evlist->nr_mmaps; i++) |
| 900 | perf_top__mmap_read_idx(top, i); |
| 901 | |
| 902 | if (overwrite) { |
| 903 | perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_EMPTY); |
| 904 | perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_RUNNING); |
| 905 | } |
| 906 | end = rdclock(); |
| 907 | |
| 908 | if ((end - start) > (unsigned long long)top->delay_secs * NSEC_PER_SEC) |
| 909 | ui__warning("Too slow to read ring buffer.\n" |
| 910 | "Please try increasing the period (-c) or\n" |
| 911 | "decreasing the freq (-F) or\n" |
| 912 | "limiting the number of CPUs (-C)\n"); |
| 913 | } |
| 914 | |
| 915 | /* |
| 916 | * Check per-event overwrite term. |
| 917 | * perf top should support consistent term for all events. |
| 918 | * - All events don't have per-event term |
| 919 | * E.g. "cpu/cpu-cycles/,cpu/instructions/" |
| 920 | * Nothing change, return 0. |
| 921 | * - All events have same per-event term |
| 922 | * E.g. "cpu/cpu-cycles,no-overwrite/,cpu/instructions,no-overwrite/ |
| 923 | * Using the per-event setting to replace the opts->overwrite if |
| 924 | * they are different, then return 0. |
| 925 | * - Events have different per-event term |
| 926 | * E.g. "cpu/cpu-cycles,overwrite/,cpu/instructions,no-overwrite/" |
| 927 | * Return -1 |
| 928 | * - Some of the event set per-event term, but some not. |
| 929 | * E.g. "cpu/cpu-cycles/,cpu/instructions,no-overwrite/" |
| 930 | * Return -1 |
| 931 | */ |
| 932 | static int perf_top__overwrite_check(struct perf_top *top) |
| 933 | { |
| 934 | struct record_opts *opts = &top->record_opts; |
| 935 | struct perf_evlist *evlist = top->evlist; |
| 936 | struct perf_evsel_config_term *term; |
| 937 | struct list_head *config_terms; |
| 938 | struct perf_evsel *evsel; |
| 939 | int set, overwrite = -1; |
| 940 | |
| 941 | evlist__for_each_entry(evlist, evsel) { |
| 942 | set = -1; |
| 943 | config_terms = &evsel->config_terms; |
| 944 | list_for_each_entry(term, config_terms, list) { |
| 945 | if (term->type == PERF_EVSEL__CONFIG_TERM_OVERWRITE) |
| 946 | set = term->val.overwrite ? 1 : 0; |
| 947 | } |
| 948 | |
| 949 | /* no term for current and previous event (likely) */ |
| 950 | if ((overwrite < 0) && (set < 0)) |
| 951 | continue; |
| 952 | |
| 953 | /* has term for both current and previous event, compare */ |
| 954 | if ((overwrite >= 0) && (set >= 0) && (overwrite != set)) |
| 955 | return -1; |
| 956 | |
| 957 | /* no term for current event but has term for previous one */ |
| 958 | if ((overwrite >= 0) && (set < 0)) |
| 959 | return -1; |
| 960 | |
| 961 | /* has term for current event */ |
| 962 | if ((overwrite < 0) && (set >= 0)) { |
| 963 | /* if it's first event, set overwrite */ |
| 964 | if (evsel == perf_evlist__first(evlist)) |
| 965 | overwrite = set; |
| 966 | else |
| 967 | return -1; |
| 968 | } |
| 969 | } |
| 970 | |
| 971 | if ((overwrite >= 0) && (opts->overwrite != overwrite)) |
| 972 | opts->overwrite = overwrite; |
| 973 | |
| 974 | return 0; |
| 975 | } |
| 976 | |
| 977 | static int perf_top_overwrite_fallback(struct perf_top *top, |
| 978 | struct perf_evsel *evsel) |
| 979 | { |
| 980 | struct record_opts *opts = &top->record_opts; |
| 981 | struct perf_evlist *evlist = top->evlist; |
| 982 | struct perf_evsel *counter; |
| 983 | |
| 984 | if (!opts->overwrite) |
| 985 | return 0; |
| 986 | |
| 987 | /* only fall back when first event fails */ |
| 988 | if (evsel != perf_evlist__first(evlist)) |
| 989 | return 0; |
| 990 | |
| 991 | evlist__for_each_entry(evlist, counter) |
| 992 | counter->attr.write_backward = false; |
| 993 | opts->overwrite = false; |
| 994 | pr_debug2("fall back to non-overwrite mode\n"); |
| 995 | return 1; |
| 996 | } |
| 997 | |
| 998 | static int perf_top__start_counters(struct perf_top *top) |
| 999 | { |
| 1000 | char msg[BUFSIZ]; |
| 1001 | struct perf_evsel *counter; |
| 1002 | struct perf_evlist *evlist = top->evlist; |
| 1003 | struct record_opts *opts = &top->record_opts; |
| 1004 | |
| 1005 | if (perf_top__overwrite_check(top)) { |
| 1006 | ui__error("perf top only support consistent per-event " |
| 1007 | "overwrite setting for all events\n"); |
| 1008 | goto out_err; |
| 1009 | } |
| 1010 | |
| 1011 | perf_evlist__config(evlist, opts, &callchain_param); |
| 1012 | |
| 1013 | evlist__for_each_entry(evlist, counter) { |
| 1014 | try_again: |
| 1015 | if (perf_evsel__open(counter, top->evlist->cpus, |
| 1016 | top->evlist->threads) < 0) { |
| 1017 | |
| 1018 | /* |
| 1019 | * Specially handle overwrite fall back. |
| 1020 | * Because perf top is the only tool which has |
| 1021 | * overwrite mode by default, support |
| 1022 | * both overwrite and non-overwrite mode, and |
| 1023 | * require consistent mode for all events. |
| 1024 | * |
| 1025 | * May move it to generic code with more tools |
| 1026 | * have similar attribute. |
| 1027 | */ |
| 1028 | if (perf_missing_features.write_backward && |
| 1029 | perf_top_overwrite_fallback(top, counter)) |
| 1030 | goto try_again; |
| 1031 | |
| 1032 | if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) { |
| 1033 | if (verbose > 0) |
| 1034 | ui__warning("%s\n", msg); |
| 1035 | goto try_again; |
| 1036 | } |
| 1037 | |
| 1038 | perf_evsel__open_strerror(counter, &opts->target, |
| 1039 | errno, msg, sizeof(msg)); |
| 1040 | ui__error("%s\n", msg); |
| 1041 | goto out_err; |
| 1042 | } |
| 1043 | } |
| 1044 | |
| 1045 | if (perf_evlist__mmap(evlist, opts->mmap_pages) < 0) { |
| 1046 | ui__error("Failed to mmap with %d (%s)\n", |
| 1047 | errno, str_error_r(errno, msg, sizeof(msg))); |
| 1048 | goto out_err; |
| 1049 | } |
| 1050 | |
| 1051 | return 0; |
| 1052 | |
| 1053 | out_err: |
| 1054 | return -1; |
| 1055 | } |
| 1056 | |
| 1057 | static int callchain_param__setup_sample_type(struct callchain_param *callchain) |
| 1058 | { |
| 1059 | if (!perf_hpp_list.sym) { |
| 1060 | if (callchain->enabled) { |
| 1061 | ui__error("Selected -g but \"sym\" not present in --sort/-s."); |
| 1062 | return -EINVAL; |
| 1063 | } |
| 1064 | } else if (callchain->mode != CHAIN_NONE) { |
| 1065 | if (callchain_register_param(callchain) < 0) { |
| 1066 | ui__error("Can't register callchain params.\n"); |
| 1067 | return -EINVAL; |
| 1068 | } |
| 1069 | } |
| 1070 | |
| 1071 | return 0; |
| 1072 | } |
| 1073 | |
| 1074 | static int __cmd_top(struct perf_top *top) |
| 1075 | { |
| 1076 | char msg[512]; |
| 1077 | struct perf_evsel *pos; |
| 1078 | struct perf_evsel_config_term *err_term; |
| 1079 | struct perf_evlist *evlist = top->evlist; |
| 1080 | struct record_opts *opts = &top->record_opts; |
| 1081 | pthread_t thread; |
| 1082 | int ret; |
| 1083 | |
| 1084 | top->session = perf_session__new(NULL, false, NULL); |
| 1085 | if (top->session == NULL) |
| 1086 | return -1; |
| 1087 | |
| 1088 | if (!objdump_path) { |
| 1089 | ret = perf_env__lookup_objdump(&top->session->header.env); |
| 1090 | if (ret) |
| 1091 | goto out_delete; |
| 1092 | } |
| 1093 | |
| 1094 | ret = callchain_param__setup_sample_type(&callchain_param); |
| 1095 | if (ret) |
| 1096 | goto out_delete; |
| 1097 | |
| 1098 | if (perf_session__register_idle_thread(top->session) < 0) |
| 1099 | goto out_delete; |
| 1100 | |
| 1101 | if (top->nr_threads_synthesize > 1) |
| 1102 | perf_set_multithreaded(); |
| 1103 | |
| 1104 | machine__synthesize_threads(&top->session->machines.host, &opts->target, |
| 1105 | top->evlist->threads, false, |
| 1106 | opts->proc_map_timeout, |
| 1107 | top->nr_threads_synthesize); |
| 1108 | |
| 1109 | if (top->nr_threads_synthesize > 1) |
| 1110 | perf_set_singlethreaded(); |
| 1111 | |
| 1112 | if (perf_hpp_list.socket) { |
| 1113 | ret = perf_env__read_cpu_topology_map(&perf_env); |
| 1114 | if (ret < 0) |
| 1115 | goto out_err_cpu_topo; |
| 1116 | } |
| 1117 | |
| 1118 | ret = perf_top__start_counters(top); |
| 1119 | if (ret) |
| 1120 | goto out_delete; |
| 1121 | |
| 1122 | ret = perf_evlist__apply_drv_configs(evlist, &pos, &err_term); |
| 1123 | if (ret) { |
| 1124 | pr_err("failed to set config \"%s\" on event %s with %d (%s)\n", |
| 1125 | err_term->val.drv_cfg, perf_evsel__name(pos), errno, |
| 1126 | str_error_r(errno, msg, sizeof(msg))); |
| 1127 | goto out_delete; |
| 1128 | } |
| 1129 | |
| 1130 | top->session->evlist = top->evlist; |
| 1131 | perf_session__set_id_hdr_size(top->session); |
| 1132 | |
| 1133 | /* |
| 1134 | * When perf is starting the traced process, all the events (apart from |
| 1135 | * group members) have enable_on_exec=1 set, so don't spoil it by |
| 1136 | * prematurely enabling them. |
| 1137 | * |
| 1138 | * XXX 'top' still doesn't start workloads like record, trace, but should, |
| 1139 | * so leave the check here. |
| 1140 | */ |
| 1141 | if (!target__none(&opts->target)) |
| 1142 | perf_evlist__enable(top->evlist); |
| 1143 | |
| 1144 | /* Wait for a minimal set of events before starting the snapshot */ |
| 1145 | perf_evlist__poll(top->evlist, 100); |
| 1146 | |
| 1147 | perf_top__mmap_read(top); |
| 1148 | |
| 1149 | ret = -1; |
| 1150 | if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui : |
| 1151 | display_thread), top)) { |
| 1152 | ui__error("Could not create display thread.\n"); |
| 1153 | goto out_delete; |
| 1154 | } |
| 1155 | |
| 1156 | if (top->realtime_prio) { |
| 1157 | struct sched_param param; |
| 1158 | |
| 1159 | param.sched_priority = top->realtime_prio; |
| 1160 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { |
| 1161 | ui__error("Could not set realtime priority.\n"); |
| 1162 | goto out_join; |
| 1163 | } |
| 1164 | } |
| 1165 | |
| 1166 | while (!done) { |
| 1167 | u64 hits = top->samples; |
| 1168 | |
| 1169 | perf_top__mmap_read(top); |
| 1170 | |
| 1171 | if (opts->overwrite || (hits == top->samples)) |
| 1172 | ret = perf_evlist__poll(top->evlist, 100); |
| 1173 | |
| 1174 | if (resize) { |
| 1175 | perf_top__resize(top); |
| 1176 | resize = 0; |
| 1177 | } |
| 1178 | } |
| 1179 | |
| 1180 | ret = 0; |
| 1181 | out_join: |
| 1182 | pthread_join(thread, NULL); |
| 1183 | out_delete: |
| 1184 | perf_session__delete(top->session); |
| 1185 | top->session = NULL; |
| 1186 | |
| 1187 | return ret; |
| 1188 | |
| 1189 | out_err_cpu_topo: { |
| 1190 | char errbuf[BUFSIZ]; |
| 1191 | const char *err = str_error_r(-ret, errbuf, sizeof(errbuf)); |
| 1192 | |
| 1193 | ui__error("Could not read the CPU topology map: %s\n", err); |
| 1194 | goto out_delete; |
| 1195 | } |
| 1196 | } |
| 1197 | |
| 1198 | static int |
| 1199 | callchain_opt(const struct option *opt, const char *arg, int unset) |
| 1200 | { |
| 1201 | symbol_conf.use_callchain = true; |
| 1202 | return record_callchain_opt(opt, arg, unset); |
| 1203 | } |
| 1204 | |
| 1205 | static int |
| 1206 | parse_callchain_opt(const struct option *opt, const char *arg, int unset) |
| 1207 | { |
| 1208 | struct callchain_param *callchain = opt->value; |
| 1209 | |
| 1210 | callchain->enabled = !unset; |
| 1211 | callchain->record_mode = CALLCHAIN_FP; |
| 1212 | |
| 1213 | /* |
| 1214 | * --no-call-graph |
| 1215 | */ |
| 1216 | if (unset) { |
| 1217 | symbol_conf.use_callchain = false; |
| 1218 | callchain->record_mode = CALLCHAIN_NONE; |
| 1219 | return 0; |
| 1220 | } |
| 1221 | |
| 1222 | return parse_callchain_top_opt(arg); |
| 1223 | } |
| 1224 | |
| 1225 | static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused) |
| 1226 | { |
| 1227 | if (!strcmp(var, "top.call-graph")) |
| 1228 | var = "call-graph.record-mode"; /* fall-through */ |
| 1229 | if (!strcmp(var, "top.children")) { |
| 1230 | symbol_conf.cumulate_callchain = perf_config_bool(var, value); |
| 1231 | return 0; |
| 1232 | } |
| 1233 | |
| 1234 | return 0; |
| 1235 | } |
| 1236 | |
| 1237 | static int |
| 1238 | parse_percent_limit(const struct option *opt, const char *arg, |
| 1239 | int unset __maybe_unused) |
| 1240 | { |
| 1241 | struct perf_top *top = opt->value; |
| 1242 | |
| 1243 | top->min_percent = strtof(arg, NULL); |
| 1244 | return 0; |
| 1245 | } |
| 1246 | |
| 1247 | const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP |
| 1248 | "\n\t\t\t\tDefault: fp,graph,0.5,caller,function"; |
| 1249 | |
| 1250 | int cmd_top(int argc, const char **argv) |
| 1251 | { |
| 1252 | char errbuf[BUFSIZ]; |
| 1253 | struct perf_top top = { |
| 1254 | .count_filter = 5, |
| 1255 | .delay_secs = 2, |
| 1256 | .record_opts = { |
| 1257 | .mmap_pages = UINT_MAX, |
| 1258 | .user_freq = UINT_MAX, |
| 1259 | .user_interval = ULLONG_MAX, |
| 1260 | .freq = 4000, /* 4 KHz */ |
| 1261 | .target = { |
| 1262 | .uses_mmap = true, |
| 1263 | }, |
| 1264 | .proc_map_timeout = 500, |
| 1265 | .overwrite = 1, |
| 1266 | }, |
| 1267 | .max_stack = sysctl_perf_event_max_stack, |
| 1268 | .sym_pcnt_filter = 5, |
| 1269 | .nr_threads_synthesize = UINT_MAX, |
| 1270 | }; |
| 1271 | struct record_opts *opts = &top.record_opts; |
| 1272 | struct target *target = &opts->target; |
| 1273 | const struct option options[] = { |
| 1274 | OPT_CALLBACK('e', "event", &top.evlist, "event", |
| 1275 | "event selector. use 'perf list' to list available events", |
| 1276 | parse_events_option), |
| 1277 | OPT_U64('c', "count", &opts->user_interval, "event period to sample"), |
| 1278 | OPT_STRING('p', "pid", &target->pid, "pid", |
| 1279 | "profile events on existing process id"), |
| 1280 | OPT_STRING('t', "tid", &target->tid, "tid", |
| 1281 | "profile events on existing thread id"), |
| 1282 | OPT_BOOLEAN('a', "all-cpus", &target->system_wide, |
| 1283 | "system-wide collection from all CPUs"), |
| 1284 | OPT_STRING('C', "cpu", &target->cpu_list, "cpu", |
| 1285 | "list of cpus to monitor"), |
| 1286 | OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, |
| 1287 | "file", "vmlinux pathname"), |
| 1288 | OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux, |
| 1289 | "don't load vmlinux even if found"), |
| 1290 | OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols, |
| 1291 | "hide kernel symbols"), |
| 1292 | OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages", |
| 1293 | "number of mmap data pages", |
| 1294 | perf_evlist__parse_mmap_pages), |
| 1295 | OPT_INTEGER('r', "realtime", &top.realtime_prio, |
| 1296 | "collect data with this RT SCHED_FIFO priority"), |
| 1297 | OPT_INTEGER('d', "delay", &top.delay_secs, |
| 1298 | "number of seconds to delay between refreshes"), |
| 1299 | OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab, |
| 1300 | "dump the symbol table used for profiling"), |
| 1301 | OPT_INTEGER('f', "count-filter", &top.count_filter, |
| 1302 | "only display functions with more events than this"), |
| 1303 | OPT_BOOLEAN(0, "group", &opts->group, |
| 1304 | "put the counters into a counter group"), |
| 1305 | OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit, |
| 1306 | "child tasks do not inherit counters"), |
| 1307 | OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name", |
| 1308 | "symbol to annotate"), |
| 1309 | OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"), |
| 1310 | OPT_CALLBACK('F', "freq", &top.record_opts, "freq or 'max'", |
| 1311 | "profile at this frequency", |
| 1312 | record__parse_freq), |
| 1313 | OPT_INTEGER('E', "entries", &top.print_entries, |
| 1314 | "display this many functions"), |
| 1315 | OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols, |
| 1316 | "hide user symbols"), |
| 1317 | OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"), |
| 1318 | OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"), |
| 1319 | OPT_INCR('v', "verbose", &verbose, |
| 1320 | "be more verbose (show counter open errors, etc)"), |
| 1321 | OPT_STRING('s', "sort", &sort_order, "key[,key2...]", |
| 1322 | "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..." |
| 1323 | " Please refer the man page for the complete list."), |
| 1324 | OPT_STRING(0, "fields", &field_order, "key[,keys...]", |
| 1325 | "output field(s): overhead, period, sample plus all of sort keys"), |
| 1326 | OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples, |
| 1327 | "Show a column with the number of samples"), |
| 1328 | OPT_CALLBACK_NOOPT('g', NULL, &callchain_param, |
| 1329 | NULL, "enables call-graph recording and display", |
| 1330 | &callchain_opt), |
| 1331 | OPT_CALLBACK(0, "call-graph", &callchain_param, |
| 1332 | "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]", |
| 1333 | top_callchain_help, &parse_callchain_opt), |
| 1334 | OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain, |
| 1335 | "Accumulate callchains of children and show total overhead as well"), |
| 1336 | OPT_INTEGER(0, "max-stack", &top.max_stack, |
| 1337 | "Set the maximum stack depth when parsing the callchain. " |
| 1338 | "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)), |
| 1339 | OPT_CALLBACK(0, "ignore-callees", NULL, "regex", |
| 1340 | "ignore callees of these functions in call graphs", |
| 1341 | report_parse_ignore_callees_opt), |
| 1342 | OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period, |
| 1343 | "Show a column with the sum of periods"), |
| 1344 | OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]", |
| 1345 | "only consider symbols in these dsos"), |
| 1346 | OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]", |
| 1347 | "only consider symbols in these comms"), |
| 1348 | OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]", |
| 1349 | "only consider these symbols"), |
| 1350 | OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src, |
| 1351 | "Interleave source code with assembly code (default)"), |
| 1352 | OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw, |
| 1353 | "Display raw encoding of assembly instructions (default)"), |
| 1354 | OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel, |
| 1355 | "Enable kernel symbol demangling"), |
| 1356 | OPT_STRING(0, "objdump", &objdump_path, "path", |
| 1357 | "objdump binary to use for disassembly and annotations"), |
| 1358 | OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style", |
| 1359 | "Specify disassembler style (e.g. -M intel for intel syntax)"), |
| 1360 | OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"), |
| 1361 | OPT_CALLBACK(0, "percent-limit", &top, "percent", |
| 1362 | "Don't show entries under that percent", parse_percent_limit), |
| 1363 | OPT_CALLBACK(0, "percentage", NULL, "relative|absolute", |
| 1364 | "How to display percentage of filtered entries", parse_filter_percentage), |
| 1365 | OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str, |
| 1366 | "width[,width...]", |
| 1367 | "don't try to adjust column width, use these fixed values"), |
| 1368 | OPT_UINTEGER(0, "proc-map-timeout", &opts->proc_map_timeout, |
| 1369 | "per thread proc mmap processing timeout in ms"), |
| 1370 | OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack, |
| 1371 | "branch any", "sample any taken branches", |
| 1372 | parse_branch_stack), |
| 1373 | OPT_CALLBACK('j', "branch-filter", &opts->branch_stack, |
| 1374 | "branch filter mask", "branch stack filter modes", |
| 1375 | parse_branch_stack), |
| 1376 | OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace, |
| 1377 | "Show raw trace event output (do not use print fmt or plugins)"), |
| 1378 | OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy, |
| 1379 | "Show entries in a hierarchy"), |
| 1380 | OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"), |
| 1381 | OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize, |
| 1382 | "number of thread to run event synthesize"), |
| 1383 | OPT_END() |
| 1384 | }; |
| 1385 | const char * const top_usage[] = { |
| 1386 | "perf top [<options>]", |
| 1387 | NULL |
| 1388 | }; |
| 1389 | int status = hists__init(); |
| 1390 | |
| 1391 | if (status < 0) |
| 1392 | return status; |
| 1393 | |
| 1394 | top.evlist = perf_evlist__new(); |
| 1395 | if (top.evlist == NULL) |
| 1396 | return -ENOMEM; |
| 1397 | |
| 1398 | status = perf_config(perf_top_config, &top); |
| 1399 | if (status) |
| 1400 | return status; |
| 1401 | |
| 1402 | argc = parse_options(argc, argv, options, top_usage, 0); |
| 1403 | if (argc) |
| 1404 | usage_with_options(top_usage, options); |
| 1405 | |
| 1406 | if (!top.evlist->nr_entries && |
| 1407 | perf_evlist__add_default(top.evlist) < 0) { |
| 1408 | pr_err("Not enough memory for event selector list\n"); |
| 1409 | goto out_delete_evlist; |
| 1410 | } |
| 1411 | |
| 1412 | if (symbol_conf.report_hierarchy) { |
| 1413 | /* disable incompatible options */ |
| 1414 | symbol_conf.event_group = false; |
| 1415 | symbol_conf.cumulate_callchain = false; |
| 1416 | |
| 1417 | if (field_order) { |
| 1418 | pr_err("Error: --hierarchy and --fields options cannot be used together\n"); |
| 1419 | parse_options_usage(top_usage, options, "fields", 0); |
| 1420 | parse_options_usage(NULL, options, "hierarchy", 0); |
| 1421 | goto out_delete_evlist; |
| 1422 | } |
| 1423 | } |
| 1424 | |
| 1425 | sort__mode = SORT_MODE__TOP; |
| 1426 | /* display thread wants entries to be collapsed in a different tree */ |
| 1427 | perf_hpp_list.need_collapse = 1; |
| 1428 | |
| 1429 | if (top.use_stdio) |
| 1430 | use_browser = 0; |
| 1431 | else if (top.use_tui) |
| 1432 | use_browser = 1; |
| 1433 | |
| 1434 | setup_browser(false); |
| 1435 | |
| 1436 | if (setup_sorting(top.evlist) < 0) { |
| 1437 | if (sort_order) |
| 1438 | parse_options_usage(top_usage, options, "s", 1); |
| 1439 | if (field_order) |
| 1440 | parse_options_usage(sort_order ? NULL : top_usage, |
| 1441 | options, "fields", 0); |
| 1442 | goto out_delete_evlist; |
| 1443 | } |
| 1444 | |
| 1445 | status = target__validate(target); |
| 1446 | if (status) { |
| 1447 | target__strerror(target, status, errbuf, BUFSIZ); |
| 1448 | ui__warning("%s\n", errbuf); |
| 1449 | } |
| 1450 | |
| 1451 | status = target__parse_uid(target); |
| 1452 | if (status) { |
| 1453 | int saved_errno = errno; |
| 1454 | |
| 1455 | target__strerror(target, status, errbuf, BUFSIZ); |
| 1456 | ui__error("%s\n", errbuf); |
| 1457 | |
| 1458 | status = -saved_errno; |
| 1459 | goto out_delete_evlist; |
| 1460 | } |
| 1461 | |
| 1462 | if (target__none(target)) |
| 1463 | target->system_wide = true; |
| 1464 | |
| 1465 | if (perf_evlist__create_maps(top.evlist, target) < 0) { |
| 1466 | ui__error("Couldn't create thread/CPU maps: %s\n", |
| 1467 | errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf))); |
| 1468 | goto out_delete_evlist; |
| 1469 | } |
| 1470 | |
| 1471 | symbol_conf.nr_events = top.evlist->nr_entries; |
| 1472 | |
| 1473 | if (top.delay_secs < 1) |
| 1474 | top.delay_secs = 1; |
| 1475 | |
| 1476 | if (record_opts__config(opts)) { |
| 1477 | status = -EINVAL; |
| 1478 | goto out_delete_evlist; |
| 1479 | } |
| 1480 | |
| 1481 | top.sym_evsel = perf_evlist__first(top.evlist); |
| 1482 | |
| 1483 | if (!callchain_param.enabled) { |
| 1484 | symbol_conf.cumulate_callchain = false; |
| 1485 | perf_hpp__cancel_cumulate(); |
| 1486 | } |
| 1487 | |
| 1488 | if (symbol_conf.cumulate_callchain && !callchain_param.order_set) |
| 1489 | callchain_param.order = ORDER_CALLER; |
| 1490 | |
| 1491 | status = symbol__annotation_init(); |
| 1492 | if (status < 0) |
| 1493 | goto out_delete_evlist; |
| 1494 | |
| 1495 | symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL); |
| 1496 | if (symbol__init(NULL) < 0) |
| 1497 | return -1; |
| 1498 | |
| 1499 | sort__setup_elide(stdout); |
| 1500 | |
| 1501 | get_term_dimensions(&top.winsize); |
| 1502 | if (top.print_entries == 0) { |
| 1503 | perf_top__update_print_entries(&top); |
| 1504 | signal(SIGWINCH, winch_sig); |
| 1505 | } |
| 1506 | |
| 1507 | status = __cmd_top(&top); |
| 1508 | |
| 1509 | out_delete_evlist: |
| 1510 | perf_evlist__delete(top.evlist); |
| 1511 | |
| 1512 | return status; |
| 1513 | } |