tools/perf/builtin-top.c

   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/cache.h"
  26 #include "util/color.h"
  27 #include "util/evlist.h"
  28 #include "util/evsel.h"
  29 #include "util/machine.h"
  30 #include "util/session.h"
  31 #include "util/symbol.h"
  32 #include "util/thread.h"
  33 #include "util/thread_map.h"
  34 #include "util/top.h"
  35 #include "util/util.h"
  36 #include <linux/rbtree.h>
  37 #include "util/parse-options.h"
  38 #include "util/parse-events.h"
  39 #include "util/cpumap.h"
  40 #include "util/xyarray.h"
  41 #include "util/sort.h"
  42 #include "util/intlist.h"
  43 #include "arch/common.h"
  44
  45 #include "util/debug.h"
  46
  47 #include <assert.h>
  48 #include <elf.h>
  49 #include <fcntl.h>
  50
  51 #include <stdio.h>
  52 #include <termios.h>
  53 #include <unistd.h>
  54 #include <inttypes.h>
  55
  56 #include <errno.h>
  57 #include <time.h>
  58 #include <sched.h>
  59
  60 #include <sys/syscall.h>
  61 #include <sys/ioctl.h>
  62 #include <sys/poll.h>
  63 #include <sys/prctl.h>
  64 #include <sys/wait.h>
  65 #include <sys/uio.h>
  66 #include <sys/utsname.h>
  67 #include <sys/mman.h>
  68
  69 #include <linux/unistd.h>
  70 #include <linux/types.h>
  71
  72 static volatile int done;
  73
  74 #define HEADER_LINE_NR  5
  75
  76 static void perf_top__update_print_entries(struct perf_top *top)
  77 {
  78         top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
  79 }
  80
  81 static void perf_top__sig_winch(int sig __maybe_unused,
  82                                 siginfo_t *info __maybe_unused, void *arg)
  83 {
  84         struct perf_top *top = arg;
  85
  86         get_term_dimensions(&top->winsize);
  87         perf_top__update_print_entries(top);
  88 }
  89
  90 static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
  91 {
  92         struct symbol *sym;
  93         struct annotation *notes;
  94         struct map *map;
  95         int err = -1;
  96
  97         if (!he || !he->ms.sym)
  98                 return -1;
  99
 100         sym = he->ms.sym;
 101         map = he->ms.map;
 102
 103         /*
 104          * We can't annotate with just /proc/kallsyms
 105          */
 106         if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
 107             !dso__is_kcore(map->dso)) {
 108                 pr_err("Can't annotate %s: No vmlinux file was found in the "
 109                        "path\n", sym->name);
 110                 sleep(1);
 111                 return -1;
 112         }
 113
 114         notes = symbol__annotation(sym);
 115         if (notes->src != NULL) {
 116                 pthread_mutex_lock(&notes->lock);
 117                 goto out_assign;
 118         }
 119
 120         pthread_mutex_lock(&notes->lock);
 121
 122         if (symbol__alloc_hist(sym) < 0) {
 123                 pthread_mutex_unlock(&notes->lock);
 124                 pr_err("Not enough memory for annotating '%s' symbol!\n",
 125                        sym->name);
 126                 sleep(1);
 127                 return err;
 128         }
 129
 130         err = symbol__annotate(sym, map, 0);
 131         if (err == 0) {
 132 out_assign:
 133                 top->sym_filter_entry = he;
 134         }
 135
 136         pthread_mutex_unlock(&notes->lock);
 137         return err;
 138 }
 139
 140 static void __zero_source_counters(struct hist_entry *he)
 141 {
 142         struct symbol *sym = he->ms.sym;
 143         symbol__annotate_zero_histograms(sym);
 144 }
 145
 146 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
 147 {
 148         struct utsname uts;
 149         int err = uname(&uts);
 150
 151         ui__warning("Out of bounds address found:\n\n"
 152                     "Addr:   %" PRIx64 "\n"
 153                     "DSO:    %s %c\n"
 154                     "Map:    %" PRIx64 "-%" PRIx64 "\n"
 155                     "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
 156                     "Arch:   %s\n"
 157                     "Kernel: %s\n"
 158                     "Tools:  %s\n\n"
 159                     "Not all samples will be on the annotation output.\n\n"
 160                     "Please report to linux-kernel@vger.kernel.org\n",
 161                     ip, map->dso->long_name, dso__symtab_origin(map->dso),
 162                     map->start, map->end, sym->start, sym->end,
 163                     sym->binding == STB_GLOBAL ? 'g' :
 164                     sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
 165                     err ? "[unknown]" : uts.machine,
 166                     err ? "[unknown]" : uts.release, perf_version_string);
 167         if (use_browser <= 0)
 168                 sleep(5);
 169
 170         map->erange_warned = true;
 171 }
 172
 173 static void perf_top__record_precise_ip(struct perf_top *top,
 174                                         struct hist_entry *he,
 175                                         int counter, u64 ip)
 176 {
 177         struct annotation *notes;
 178         struct symbol *sym;
 179         int err;
 180
 181         if (he == NULL || he->ms.sym == NULL ||
 182             ((top->sym_filter_entry == NULL ||
 183               top->sym_filter_entry->ms.sym != he->ms.sym) && use_browser != 1))
 184                 return;
 185
 186         sym = he->ms.sym;
 187         notes = symbol__annotation(sym);
 188
 189         if (pthread_mutex_trylock(&notes->lock))
 190                 return;
 191
 192         ip = he->ms.map->map_ip(he->ms.map, ip);
 193         err = hist_entry__inc_addr_samples(he, counter, ip);
 194
 195         pthread_mutex_unlock(&notes->lock);
 196
 197         if (err == -ERANGE && !he->ms.map->erange_warned)
 198                 ui__warn_map_erange(he->ms.map, sym, ip);
 199         else if (err == -ENOMEM) {
 200                 pr_err("Not enough memory for annotating '%s' symbol!\n",
 201                        sym->name);
 202                 sleep(1);
 203         }
 204 }
 205
 206 static void perf_top__show_details(struct perf_top *top)
 207 {
 208         struct hist_entry *he = top->sym_filter_entry;
 209         struct annotation *notes;
 210         struct symbol *symbol;
 211         int more;
 212
 213         if (!he)
 214                 return;
 215
 216         symbol = he->ms.sym;
 217         notes = symbol__annotation(symbol);
 218
 219         pthread_mutex_lock(&notes->lock);
 220
 221         if (notes->src == NULL)
 222                 goto out_unlock;
 223
 224         printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
 225         printf("  Events  Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
 226
 227         more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
 228                                        0, top->sym_pcnt_filter, top->print_entries, 4);
 229         if (top->zero)
 230                 symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
 231         else
 232                 symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
 233         if (more != 0)
 234                 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
 235 out_unlock:
 236         pthread_mutex_unlock(&notes->lock);
 237 }
 238
 239 static struct hist_entry *perf_evsel__add_hist_entry(struct perf_evsel *evsel,
 240                                                      struct addr_location *al,
 241                                                      struct perf_sample *sample)
 242 {
 243         struct hist_entry *he;
 244
 245         pthread_mutex_lock(&evsel->hists.lock);
 246         he = __hists__add_entry(&evsel->hists, al, NULL, NULL, NULL,
 247                                 sample->period, sample->weight,
 248                                 sample->transaction);
 249         pthread_mutex_unlock(&evsel->hists.lock);
 250         if (he == NULL)
 251                 return NULL;
 252
 253         hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
 254         return he;
 255 }
 256
 257 static void perf_top__print_sym_table(struct perf_top *top)
 258 {
 259         char bf[160];
 260         int printed = 0;
 261         const int win_width = top->winsize.ws_col - 1;
 262
 263         puts(CONSOLE_CLEAR);
 264
 265         perf_top__header_snprintf(top, bf, sizeof(bf));
 266         printf("%s\n", bf);
 267
 268         perf_top__reset_sample_counters(top);
 269
 270         printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
 271
 272         if (top->sym_evsel->hists.stats.nr_lost_warned !=
 273             top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST]) {
 274                 top->sym_evsel->hists.stats.nr_lost_warned =
 275                         top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST];
 276                 color_fprintf(stdout, PERF_COLOR_RED,
 277                               "WARNING: LOST %d chunks, Check IO/CPU overload",
 278                               top->sym_evsel->hists.stats.nr_lost_warned);
 279                 ++printed;
 280         }
 281
 282         if (top->sym_filter_entry) {
 283                 perf_top__show_details(top);
 284                 return;
 285         }
 286
 287         hists__collapse_resort(&top->sym_evsel->hists, NULL);
 288         hists__output_resort(&top->sym_evsel->hists);
 289         hists__decay_entries(&top->sym_evsel->hists,
 290                              top->hide_user_symbols,
 291                              top->hide_kernel_symbols);
 292         hists__output_recalc_col_len(&top->sym_evsel->hists,
 293                                      top->print_entries - printed);
 294         putchar('\n');
 295         hists__fprintf(&top->sym_evsel->hists, false,
 296                        top->print_entries - printed, win_width,
 297                        top->min_percent, stdout);
 298 }
 299
 300 static void prompt_integer(int *target, const char *msg)
 301 {
 302         char *buf = malloc(0), *p;
 303         size_t dummy = 0;
 304         int tmp;
 305
 306         fprintf(stdout, "\n%s: ", msg);
 307         if (getline(&buf, &dummy, stdin) < 0)
 308                 return;
 309
 310         p = strchr(buf, '\n');
 311         if (p)
 312                 *p = 0;
 313
 314         p = buf;
 315         while(*p) {
 316                 if (!isdigit(*p))
 317                         goto out_free;
 318                 p++;
 319         }
 320         tmp = strtoul(buf, NULL, 10);
 321         *target = tmp;
 322 out_free:
 323         free(buf);
 324 }
 325
 326 static void prompt_percent(int *target, const char *msg)
 327 {
 328         int tmp = 0;
 329
 330         prompt_integer(&tmp, msg);
 331         if (tmp >= 0 && tmp <= 100)
 332                 *target = tmp;
 333 }
 334
 335 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
 336 {
 337         char *buf = malloc(0), *p;
 338         struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
 339         struct rb_node *next;
 340         size_t dummy = 0;
 341
 342         /* zero counters of active symbol */
 343         if (syme) {
 344                 __zero_source_counters(syme);
 345                 top->sym_filter_entry = NULL;
 346         }
 347
 348         fprintf(stdout, "\n%s: ", msg);
 349         if (getline(&buf, &dummy, stdin) < 0)
 350                 goto out_free;
 351
 352         p = strchr(buf, '\n');
 353         if (p)
 354                 *p = 0;
 355
 356         next = rb_first(&top->sym_evsel->hists.entries);
 357         while (next) {
 358                 n = rb_entry(next, struct hist_entry, rb_node);
 359                 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
 360                         found = n;
 361                         break;
 362                 }
 363                 next = rb_next(&n->rb_node);
 364         }
 365
 366         if (!found) {
 367                 fprintf(stderr, "Sorry, %s is not active.\n", buf);
 368                 sleep(1);
 369         } else
 370                 perf_top__parse_source(top, found);
 371
 372 out_free:
 373         free(buf);
 374 }
 375
 376 static void perf_top__print_mapped_keys(struct perf_top *top)
 377 {
 378         char *name = NULL;
 379
 380         if (top->sym_filter_entry) {
 381                 struct symbol *sym = top->sym_filter_entry->ms.sym;
 382                 name = sym->name;
 383         }
 384
 385         fprintf(stdout, "\nMapped keys:\n");
 386         fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
 387         fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
 388
 389         if (top->evlist->nr_entries > 1)
 390                 fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
 391
 392         fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
 393
 394         fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
 395         fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
 396         fprintf(stdout, "\t[S]     stop annotation.\n");
 397
 398         fprintf(stdout,
 399                 "\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
 400                 top->hide_kernel_symbols ? "yes" : "no");
 401         fprintf(stdout,
 402                 "\t[U]     hide user symbols.               \t(%s)\n",
 403                 top->hide_user_symbols ? "yes" : "no");
 404         fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
 405         fprintf(stdout, "\t[qQ]    quit.\n");
 406 }
 407
 408 static int perf_top__key_mapped(struct perf_top *top, int c)
 409 {
 410         switch (c) {
 411                 case 'd':
 412                 case 'e':
 413                 case 'f':
 414                 case 'z':
 415                 case 'q':
 416                 case 'Q':
 417                 case 'K':
 418                 case 'U':
 419                 case 'F':
 420                 case 's':
 421                 case 'S':
 422                         return 1;
 423                 case 'E':
 424                         return top->evlist->nr_entries > 1 ? 1 : 0;
 425                 default:
 426                         break;
 427         }
 428
 429         return 0;
 430 }
 431
 432 static bool perf_top__handle_keypress(struct perf_top *top, int c)
 433 {
 434         bool ret = true;
 435
 436         if (!perf_top__key_mapped(top, c)) {
 437                 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 438                 struct termios tc, save;
 439
 440                 perf_top__print_mapped_keys(top);
 441                 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
 442                 fflush(stdout);
 443
 444                 tcgetattr(0, &save);
 445                 tc = save;
 446                 tc.c_lflag &= ~(ICANON | ECHO);
 447                 tc.c_cc[VMIN] = 0;
 448                 tc.c_cc[VTIME] = 0;
 449                 tcsetattr(0, TCSANOW, &tc);
 450
 451                 poll(&stdin_poll, 1, -1);
 452                 c = getc(stdin);
 453
 454                 tcsetattr(0, TCSAFLUSH, &save);
 455                 if (!perf_top__key_mapped(top, c))
 456                         return ret;
 457         }
 458
 459         switch (c) {
 460                 case 'd':
 461                         prompt_integer(&top->delay_secs, "Enter display delay");
 462                         if (top->delay_secs < 1)
 463                                 top->delay_secs = 1;
 464                         break;
 465                 case 'e':
 466                         prompt_integer(&top->print_entries, "Enter display entries (lines)");
 467                         if (top->print_entries == 0) {
 468                                 struct sigaction act = {
 469                                         .sa_sigaction = perf_top__sig_winch,
 470                                         .sa_flags     = SA_SIGINFO,
 471                                 };
 472                                 perf_top__sig_winch(SIGWINCH, NULL, top);
 473                                 sigaction(SIGWINCH, &act, NULL);
 474                         } else {
 475                                 signal(SIGWINCH, SIG_DFL);
 476                         }
 477                         break;
 478                 case 'E':
 479                         if (top->evlist->nr_entries > 1) {
 480                                 /* Select 0 as the default event: */
 481                                 int counter = 0;
 482
 483                                 fprintf(stderr, "\nAvailable events:");
 484
 485                                 evlist__for_each(top->evlist, top->sym_evsel)
 486                                         fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
 487
 488                                 prompt_integer(&counter, "Enter details event counter");
 489
 490                                 if (counter >= top->evlist->nr_entries) {
 491                                         top->sym_evsel = perf_evlist__first(top->evlist);
 492                                         fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
 493                                         sleep(1);
 494                                         break;
 495                                 }
 496                                 evlist__for_each(top->evlist, top->sym_evsel)
 497                                         if (top->sym_evsel->idx == counter)
 498                                                 break;
 499                         } else
 500                                 top->sym_evsel = perf_evlist__first(top->evlist);
 501                         break;
 502                 case 'f':
 503                         prompt_integer(&top->count_filter, "Enter display event count filter");
 504                         break;
 505                 case 'F':
 506                         prompt_percent(&top->sym_pcnt_filter,
 507                                        "Enter details display event filter (percent)");
 508                         break;
 509                 case 'K':
 510                         top->hide_kernel_symbols = !top->hide_kernel_symbols;
 511                         break;
 512                 case 'q':
 513                 case 'Q':
 514                         printf("exiting.\n");
 515                         if (top->dump_symtab)
 516                                 perf_session__fprintf_dsos(top->session, stderr);
 517                         ret = false;
 518                         break;
 519                 case 's':
 520                         perf_top__prompt_symbol(top, "Enter details symbol");
 521                         break;
 522                 case 'S':
 523                         if (!top->sym_filter_entry)
 524                                 break;
 525                         else {
 526                                 struct hist_entry *syme = top->sym_filter_entry;
 527
 528                                 top->sym_filter_entry = NULL;
 529                                 __zero_source_counters(syme);
 530                         }
 531                         break;
 532                 case 'U':
 533                         top->hide_user_symbols = !top->hide_user_symbols;
 534                         break;
 535                 case 'z':
 536                         top->zero = !top->zero;
 537                         break;
 538                 default:
 539                         break;
 540         }
 541
 542         return ret;
 543 }
 544
 545 static void perf_top__sort_new_samples(void *arg)
 546 {
 547         struct perf_top *t = arg;
 548         perf_top__reset_sample_counters(t);
 549
 550         if (t->evlist->selected != NULL)
 551                 t->sym_evsel = t->evlist->selected;
 552
 553         hists__collapse_resort(&t->sym_evsel->hists, NULL);
 554         hists__output_resort(&t->sym_evsel->hists);
 555         hists__decay_entries(&t->sym_evsel->hists,
 556                              t->hide_user_symbols,
 557                              t->hide_kernel_symbols);
 558 }
 559
 560 static void *display_thread_tui(void *arg)
 561 {
 562         struct perf_evsel *pos;
 563         struct perf_top *top = arg;
 564         const char *help = "For a higher level overview, try: perf top --sort comm,dso";
 565         struct hist_browser_timer hbt = {
 566                 .timer          = perf_top__sort_new_samples,
 567                 .arg            = top,
 568                 .refresh        = top->delay_secs,
 569         };
 570
 571         perf_top__sort_new_samples(top);
 572
 573         /*
 574          * Initialize the uid_filter_str, in the future the TUI will allow
 575          * Zooming in/out UIDs. For now juse use whatever the user passed
 576          * via --uid.
 577          */
 578         evlist__for_each(top->evlist, pos)
 579                 pos->hists.uid_filter_str = top->record_opts.target.uid_str;
 580
 581         perf_evlist__tui_browse_hists(top->evlist, help, &hbt, top->min_percent,
 582                                       &top->session->header.env);
 583
 584         done = 1;
 585         return NULL;
 586 }
 587
 588 static void *display_thread(void *arg)
 589 {
 590         struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 591         struct termios tc, save;
 592         struct perf_top *top = arg;
 593         int delay_msecs, c;
 594
 595         tcgetattr(0, &save);
 596         tc = save;
 597         tc.c_lflag &= ~(ICANON | ECHO);
 598         tc.c_cc[VMIN] = 0;
 599         tc.c_cc[VTIME] = 0;
 600
 601         pthread__unblock_sigwinch();
 602 repeat:
 603         delay_msecs = top->delay_secs * 1000;
 604         tcsetattr(0, TCSANOW, &tc);
 605         /* trash return*/
 606         getc(stdin);
 607
 608         while (!done) {
 609                 perf_top__print_sym_table(top);
 610                 /*
 611                  * Either timeout expired or we got an EINTR due to SIGWINCH,
 612                  * refresh screen in both cases.
 613                  */
 614                 switch (poll(&stdin_poll, 1, delay_msecs)) {
 615                 case 0:
 616                         continue;
 617                 case -1:
 618                         if (errno == EINTR)
 619                                 continue;
 620                         /* Fall trhu */
 621                 default:
 622                         c = getc(stdin);
 623                         tcsetattr(0, TCSAFLUSH, &save);
 624
 625                         if (perf_top__handle_keypress(top, c))
 626                                 goto repeat;
 627                         done = 1;
 628                 }
 629         }
 630
 631         return NULL;
 632 }
 633
 634 static int symbol_filter(struct map *map __maybe_unused, struct symbol *sym)
 635 {
 636         const char *name = sym->name;
 637
 638         /*
 639          * ppc64 uses function descriptors and appends a '.' to the
 640          * start of every instruction address. Remove it.
 641          */
 642         if (name[0] == '.')
 643                 name++;
 644
 645         if (!strcmp(name, "_text") ||
 646             !strcmp(name, "_etext") ||
 647             !strcmp(name, "_sinittext") ||
 648             !strncmp("init_module", name, 11) ||
 649             !strncmp("cleanup_module", name, 14) ||
 650             strstr(name, "_text_start") ||
 651             strstr(name, "_text_end"))
 652                 return 1;
 653
 654         if (symbol__is_idle(sym))
 655                 sym->ignore = true;
 656
 657         return 0;
 658 }
 659
 660 static void perf_event__process_sample(struct perf_tool *tool,
 661                                        const union perf_event *event,
 662                                        struct perf_evsel *evsel,
 663                                        struct perf_sample *sample,
 664                                        struct machine *machine)
 665 {
 666         struct perf_top *top = container_of(tool, struct perf_top, tool);
 667         struct symbol *parent = NULL;
 668         u64 ip = sample->ip;
 669         struct addr_location al;
 670         int err;
 671
 672         if (!machine && perf_guest) {
 673                 static struct intlist *seen;
 674
 675                 if (!seen)
 676                         seen = intlist__new(NULL);
 677
 678                 if (!intlist__has_entry(seen, sample->pid)) {
 679                         pr_err("Can't find guest [%d]'s kernel information\n",
 680                                 sample->pid);
 681                         intlist__add(seen, sample->pid);
 682                 }
 683                 return;
 684         }
 685
 686         if (!machine) {
 687                 pr_err("%u unprocessable samples recorded.\r",
 688                        top->session->stats.nr_unprocessable_samples++);
 689                 return;
 690         }
 691
 692         if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
 693                 top->exact_samples++;
 694
 695         if (perf_event__preprocess_sample(event, machine, &al, sample) < 0 ||
 696             al.filtered)
 697                 return;
 698
 699         if (!top->kptr_restrict_warned &&
 700             symbol_conf.kptr_restrict &&
 701             al.cpumode == PERF_RECORD_MISC_KERNEL) {
 702                 ui__warning(
 703 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
 704 "Check /proc/sys/kernel/kptr_restrict.\n\n"
 705 "Kernel%s samples will not be resolved.\n",
 706                           !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
 707                           " modules" : "");
 708                 if (use_browser <= 0)
 709                         sleep(5);
 710                 top->kptr_restrict_warned = true;
 711         }
 712
 713         if (al.sym == NULL) {
 714                 const char *msg = "Kernel samples will not be resolved.\n";
 715                 /*
 716                  * As we do lazy loading of symtabs we only will know if the
 717                  * specified vmlinux file is invalid when we actually have a
 718                  * hit in kernel space and then try to load it. So if we get
 719                  * here and there are _no_ symbols in the DSO backing the
 720                  * kernel map, bail out.
 721                  *
 722                  * We may never get here, for instance, if we use -K/
 723                  * --hide-kernel-symbols, even if the user specifies an
 724                  * invalid --vmlinux ;-)
 725                  */
 726                 if (!top->kptr_restrict_warned && !top->vmlinux_warned &&
 727                     al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
 728                     RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
 729                         if (symbol_conf.vmlinux_name) {
 730                                 ui__warning("The %s file can't be used.\n%s",
 731                                             symbol_conf.vmlinux_name, msg);
 732                         } else {
 733                                 ui__warning("A vmlinux file was not found.\n%s",
 734                                             msg);
 735                         }
 736
 737                         if (use_browser <= 0)
 738                                 sleep(5);
 739                         top->vmlinux_warned = true;
 740                 }
 741         }
 742
 743         if (al.sym == NULL || !al.sym->ignore) {
 744                 struct hist_entry *he;
 745
 746                 err = sample__resolve_callchain(sample, &parent, evsel, &al,
 747                                                 top->max_stack);
 748                 if (err)
 749                         return;
 750
 751                 he = perf_evsel__add_hist_entry(evsel, &al, sample);
 752                 if (he == NULL) {
 753                         pr_err("Problem incrementing symbol period, skipping event\n");
 754                         return;
 755                 }
 756
 757                 err = hist_entry__append_callchain(he, sample);
 758                 if (err)
 759                         return;
 760
 761                 if (sort__has_sym)
 762                         perf_top__record_precise_ip(top, he, evsel->idx, ip);
 763         }
 764
 765         return;
 766 }
 767
 768 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
 769 {
 770         struct perf_sample sample;
 771         struct perf_evsel *evsel;
 772         struct perf_session *session = top->session;
 773         union perf_event *event;
 774         struct machine *machine;
 775         u8 origin;
 776         int ret;
 777
 778         while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
 779                 ret = perf_evlist__parse_sample(top->evlist, event, &sample);
 780                 if (ret) {
 781                         pr_err("Can't parse sample, err = %d\n", ret);
 782                         goto next_event;
 783                 }
 784
 785                 evsel = perf_evlist__id2evsel(session->evlist, sample.id);
 786                 assert(evsel != NULL);
 787
 788                 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 789
 790                 if (event->header.type == PERF_RECORD_SAMPLE)
 791                         ++top->samples;
 792
 793                 switch (origin) {
 794                 case PERF_RECORD_MISC_USER:
 795                         ++top->us_samples;
 796                         if (top->hide_user_symbols)
 797                                 goto next_event;
 798                         machine = &session->machines.host;
 799                         break;
 800                 case PERF_RECORD_MISC_KERNEL:
 801                         ++top->kernel_samples;
 802                         if (top->hide_kernel_symbols)
 803                                 goto next_event;
 804                         machine = &session->machines.host;
 805                         break;
 806                 case PERF_RECORD_MISC_GUEST_KERNEL:
 807                         ++top->guest_kernel_samples;
 808                         machine = perf_session__find_machine(session,
 809                                                              sample.pid);
 810                         break;
 811                 case PERF_RECORD_MISC_GUEST_USER:
 812                         ++top->guest_us_samples;
 813                         /*
 814                          * TODO: we don't process guest user from host side
 815                          * except simple counting.
 816                          */
 817                         /* Fall thru */
 818                 default:
 819                         goto next_event;
 820                 }
 821
 822
 823                 if (event->header.type == PERF_RECORD_SAMPLE) {
 824                         perf_event__process_sample(&top->tool, event, evsel,
 825                                                    &sample, machine);
 826                 } else if (event->header.type < PERF_RECORD_MAX) {
 827                         hists__inc_nr_events(&evsel->hists, event->header.type);
 828                         machine__process_event(machine, event, &sample);
 829                 } else
 830                         ++session->stats.nr_unknown_events;
 831 next_event:
 832                 perf_evlist__mmap_consume(top->evlist, idx);
 833         }
 834 }
 835
 836 static void perf_top__mmap_read(struct perf_top *top)
 837 {
 838         int i;
 839
 840         for (i = 0; i < top->evlist->nr_mmaps; i++)
 841                 perf_top__mmap_read_idx(top, i);
 842 }
 843
 844 static int perf_top__start_counters(struct perf_top *top)
 845 {
 846         char msg[512];
 847         struct perf_evsel *counter;
 848         struct perf_evlist *evlist = top->evlist;
 849         struct record_opts *opts = &top->record_opts;
 850
 851         perf_evlist__config(evlist, opts);
 852
 853         evlist__for_each(evlist, counter) {
 854 try_again:
 855                 if (perf_evsel__open(counter, top->evlist->cpus,
 856                                      top->evlist->threads) < 0) {
 857                         if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
 858                                 if (verbose)
 859                                         ui__warning("%s\n", msg);
 860                                 goto try_again;
 861                         }
 862
 863                         perf_evsel__open_strerror(counter, &opts->target,
 864                                                   errno, msg, sizeof(msg));
 865                         ui__error("%s\n", msg);
 866                         goto out_err;
 867                 }
 868         }
 869
 870         if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
 871                 ui__error("Failed to mmap with %d (%s)\n",
 872                             errno, strerror(errno));
 873                 goto out_err;
 874         }
 875
 876         return 0;
 877
 878 out_err:
 879         return -1;
 880 }
 881
 882 static int perf_top__setup_sample_type(struct perf_top *top __maybe_unused)
 883 {
 884         if (!sort__has_sym) {
 885                 if (symbol_conf.use_callchain) {
 886                         ui__error("Selected -g but \"sym\" not present in --sort/-s.");
 887                         return -EINVAL;
 888                 }
 889         } else if (callchain_param.mode != CHAIN_NONE) {
 890                 if (callchain_register_param(&callchain_param) < 0) {
 891                         ui__error("Can't register callchain params.\n");
 892                         return -EINVAL;
 893                 }
 894         }
 895
 896         return 0;
 897 }
 898
 899 static int __cmd_top(struct perf_top *top)
 900 {
 901         struct record_opts *opts = &top->record_opts;
 902         pthread_t thread;
 903         int ret;
 904
 905         top->session = perf_session__new(NULL, false, NULL);
 906         if (top->session == NULL)
 907                 return -ENOMEM;
 908
 909         machines__set_symbol_filter(&top->session->machines, symbol_filter);
 910
 911         if (!objdump_path) {
 912                 ret = perf_session_env__lookup_objdump(&top->session->header.env);
 913                 if (ret)
 914                         goto out_delete;
 915         }
 916
 917         ret = perf_top__setup_sample_type(top);
 918         if (ret)
 919                 goto out_delete;
 920
 921         machine__synthesize_threads(&top->session->machines.host, &opts->target,
 922                                     top->evlist->threads, false);
 923         ret = perf_top__start_counters(top);
 924         if (ret)
 925                 goto out_delete;
 926
 927         top->session->evlist = top->evlist;
 928         perf_session__set_id_hdr_size(top->session);
 929
 930         /*
 931          * When perf is starting the traced process, all the events (apart from
 932          * group members) have enable_on_exec=1 set, so don't spoil it by
 933          * prematurely enabling them.
 934          *
 935          * XXX 'top' still doesn't start workloads like record, trace, but should,
 936          * so leave the check here.
 937          */
 938         if (!target__none(&opts->target))
 939                 perf_evlist__enable(top->evlist);
 940
 941         /* Wait for a minimal set of events before starting the snapshot */
 942         poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
 943
 944         perf_top__mmap_read(top);
 945
 946         ret = -1;
 947         if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
 948                                                             display_thread), top)) {
 949                 ui__error("Could not create display thread.\n");
 950                 goto out_delete;
 951         }
 952
 953         if (top->realtime_prio) {
 954                 struct sched_param param;
 955
 956                 param.sched_priority = top->realtime_prio;
 957                 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
 958                         ui__error("Could not set realtime priority.\n");
 959                         goto out_delete;
 960                 }
 961         }
 962
 963         while (!done) {
 964                 u64 hits = top->samples;
 965
 966                 perf_top__mmap_read(top);
 967
 968                 if (hits == top->samples)
 969                         ret = poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
 970         }
 971
 972         ret = 0;
 973 out_delete:
 974         perf_session__delete(top->session);
 975         top->session = NULL;
 976
 977         return ret;
 978 }
 979
 980 static int
 981 callchain_opt(const struct option *opt, const char *arg, int unset)
 982 {
 983         symbol_conf.use_callchain = true;
 984         return record_callchain_opt(opt, arg, unset);
 985 }
 986
 987 static int
 988 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
 989 {
 990         symbol_conf.use_callchain = true;
 991         return record_parse_callchain_opt(opt, arg, unset);
 992 }
 993
 994 static int
 995 parse_percent_limit(const struct option *opt, const char *arg,
 996                     int unset __maybe_unused)
 997 {
 998         struct perf_top *top = opt->value;
 999
1000         top->min_percent = strtof(arg, NULL);
1001         return 0;
1002 }
1003
1004 int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1005 {
1006         int status = -1;
1007         char errbuf[BUFSIZ];
1008         struct perf_top top = {
1009                 .count_filter        = 5,
1010                 .delay_secs          = 2,
1011                 .record_opts = {
1012                         .mmap_pages     = UINT_MAX,
1013                         .user_freq      = UINT_MAX,
1014                         .user_interval  = ULLONG_MAX,
1015                         .freq           = 4000, /* 4 KHz */
1016                         .target         = {
1017                                 .uses_mmap   = true,
1018                         },
1019                 },
1020                 .max_stack           = PERF_MAX_STACK_DEPTH,
1021                 .sym_pcnt_filter     = 5,
1022         };
1023         struct record_opts *opts = &top.record_opts;
1024         struct target *target = &opts->target;
1025         const struct option options[] = {
1026         OPT_CALLBACK('e', "event", &top.evlist, "event",
1027                      "event selector. use 'perf list' to list available events",
1028                      parse_events_option),
1029         OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1030         OPT_STRING('p', "pid", &target->pid, "pid",
1031                     "profile events on existing process id"),
1032         OPT_STRING('t', "tid", &target->tid, "tid",
1033                     "profile events on existing thread id"),
1034         OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1035                             "system-wide collection from all CPUs"),
1036         OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1037                     "list of cpus to monitor"),
1038         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1039                    "file", "vmlinux pathname"),
1040         OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1041                     "don't load vmlinux even if found"),
1042         OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1043                     "hide kernel symbols"),
1044         OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1045                      "number of mmap data pages",
1046                      perf_evlist__parse_mmap_pages),
1047         OPT_INTEGER('r', "realtime", &top.realtime_prio,
1048                     "collect data with this RT SCHED_FIFO priority"),
1049         OPT_INTEGER('d', "delay", &top.delay_secs,
1050                     "number of seconds to delay between refreshes"),
1051         OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1052                             "dump the symbol table used for profiling"),
1053         OPT_INTEGER('f', "count-filter", &top.count_filter,
1054                     "only display functions with more events than this"),
1055         OPT_BOOLEAN(0, "group", &opts->group,
1056                             "put the counters into a counter group"),
1057         OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1058                     "child tasks do not inherit counters"),
1059         OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1060                     "symbol to annotate"),
1061         OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1062         OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1063         OPT_INTEGER('E', "entries", &top.print_entries,
1064                     "display this many functions"),
1065         OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1066                     "hide user symbols"),
1067         OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1068         OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1069         OPT_INCR('v', "verbose", &verbose,
1070                     "be more verbose (show counter open errors, etc)"),
1071         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1072                    "sort by key(s): pid, comm, dso, symbol, parent, weight, local_weight,"
1073                    " abort, in_tx, transaction"),
1074         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1075                     "Show a column with the number of samples"),
1076         OPT_CALLBACK_NOOPT('g', NULL, &top.record_opts,
1077                            NULL, "enables call-graph recording",
1078                            &callchain_opt),
1079         OPT_CALLBACK(0, "call-graph", &top.record_opts,
1080                      "mode[,dump_size]", record_callchain_help,
1081                      &parse_callchain_opt),
1082         OPT_INTEGER(0, "max-stack", &top.max_stack,
1083                     "Set the maximum stack depth when parsing the callchain. "
1084                     "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
1085         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1086                    "ignore callees of these functions in call graphs",
1087                    report_parse_ignore_callees_opt),
1088         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1089                     "Show a column with the sum of periods"),
1090         OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1091                    "only consider symbols in these dsos"),
1092         OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1093                    "only consider symbols in these comms"),
1094         OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1095                    "only consider these symbols"),
1096         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1097                     "Interleave source code with assembly code (default)"),
1098         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1099                     "Display raw encoding of assembly instructions (default)"),
1100         OPT_STRING(0, "objdump", &objdump_path, "path",
1101                     "objdump binary to use for disassembly and annotations"),
1102         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1103                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
1104         OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1105         OPT_CALLBACK(0, "percent-limit", &top, "percent",
1106                      "Don't show entries under that percent", parse_percent_limit),
1107         OPT_END()
1108         };
1109         const char * const top_usage[] = {
1110                 "perf top [<options>]",
1111                 NULL
1112         };
1113
1114         top.evlist = perf_evlist__new();
1115         if (top.evlist == NULL)
1116                 return -ENOMEM;
1117
1118         argc = parse_options(argc, argv, options, top_usage, 0);
1119         if (argc)
1120                 usage_with_options(top_usage, options);
1121
1122         if (sort_order == default_sort_order)
1123                 sort_order = "dso,symbol";
1124
1125         if (setup_sorting() < 0) {
1126                 parse_options_usage(top_usage, options, "s", 1);
1127                 goto out_delete_evlist;
1128         }
1129
1130         /* display thread wants entries to be collapsed in a different tree */
1131         sort__need_collapse = 1;
1132
1133         if (top.use_stdio)
1134                 use_browser = 0;
1135         else if (top.use_tui)
1136                 use_browser = 1;
1137
1138         setup_browser(false);
1139
1140         status = target__validate(target);
1141         if (status) {
1142                 target__strerror(target, status, errbuf, BUFSIZ);
1143                 ui__warning("%s\n", errbuf);
1144         }
1145
1146         status = target__parse_uid(target);
1147         if (status) {
1148                 int saved_errno = errno;
1149
1150                 target__strerror(target, status, errbuf, BUFSIZ);
1151                 ui__error("%s\n", errbuf);
1152
1153                 status = -saved_errno;
1154                 goto out_delete_evlist;
1155         }
1156
1157         if (target__none(target))
1158                 target->system_wide = true;
1159
1160         if (perf_evlist__create_maps(top.evlist, target) < 0)
1161                 usage_with_options(top_usage, options);
1162
1163         if (!top.evlist->nr_entries &&
1164             perf_evlist__add_default(top.evlist) < 0) {
1165                 ui__error("Not enough memory for event selector list\n");
1166                 goto out_delete_evlist;
1167         }
1168
1169         symbol_conf.nr_events = top.evlist->nr_entries;
1170
1171         if (top.delay_secs < 1)
1172                 top.delay_secs = 1;
1173
1174         if (record_opts__config(opts)) {
1175                 status = -EINVAL;
1176                 goto out_delete_evlist;
1177         }
1178
1179         top.sym_evsel = perf_evlist__first(top.evlist);
1180
1181         symbol_conf.priv_size = sizeof(struct annotation);
1182
1183         symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1184         if (symbol__init() < 0)
1185                 return -1;
1186
1187         sort__setup_elide(stdout);
1188
1189         get_term_dimensions(&top.winsize);
1190         if (top.print_entries == 0) {
1191                 struct sigaction act = {
1192                         .sa_sigaction = perf_top__sig_winch,
1193                         .sa_flags     = SA_SIGINFO,
1194                 };
1195                 perf_top__update_print_entries(&top);
1196                 sigaction(SIGWINCH, &act, NULL);
1197         }
1198
1199         status = __cmd_top(&top);
1200
1201 out_delete_evlist:
1202         perf_evlist__delete(top.evlist);
1203
1204         return status;
1205 }