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                 pr_err("Can't annotate %s: No vmlinux file was found in the "
 108                        "path\n", sym->name);
 109                 sleep(1);
 110                 return -1;
 111         }
 112
 113         notes = symbol__annotation(sym);
 114         if (notes->src != NULL) {
 115                 pthread_mutex_lock(&notes->lock);
 116                 goto out_assign;
 117         }
 118
 119         pthread_mutex_lock(&notes->lock);
 120
 121         if (symbol__alloc_hist(sym) < 0) {
 122                 pthread_mutex_unlock(&notes->lock);
 123                 pr_err("Not enough memory for annotating '%s' symbol!\n",
 124                        sym->name);
 125                 sleep(1);
 126                 return err;
 127         }
 128
 129         err = symbol__annotate(sym, map, 0);
 130         if (err == 0) {
 131 out_assign:
 132                 top->sym_filter_entry = he;
 133         }
 134
 135         pthread_mutex_unlock(&notes->lock);
 136         return err;
 137 }
 138
 139 static void __zero_source_counters(struct hist_entry *he)
 140 {
 141         struct symbol *sym = he->ms.sym;
 142         symbol__annotate_zero_histograms(sym);
 143 }
 144
 145 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
 146 {
 147         struct utsname uts;
 148         int err = uname(&uts);
 149
 150         ui__warning("Out of bounds address found:\n\n"
 151                     "Addr:   %" PRIx64 "\n"
 152                     "DSO:    %s %c\n"
 153                     "Map:    %" PRIx64 "-%" PRIx64 "\n"
 154                     "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
 155                     "Arch:   %s\n"
 156                     "Kernel: %s\n"
 157                     "Tools:  %s\n\n"
 158                     "Not all samples will be on the annotation output.\n\n"
 159                     "Please report to linux-kernel@vger.kernel.org\n",
 160                     ip, map->dso->long_name, dso__symtab_origin(map->dso),
 161                     map->start, map->end, sym->start, sym->end,
 162                     sym->binding == STB_GLOBAL ? 'g' :
 163                     sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
 164                     err ? "[unknown]" : uts.machine,
 165                     err ? "[unknown]" : uts.release, perf_version_string);
 166         if (use_browser <= 0)
 167                 sleep(5);
 168
 169         map->erange_warned = true;
 170 }
 171
 172 static void perf_top__record_precise_ip(struct perf_top *top,
 173                                         struct hist_entry *he,
 174                                         int counter, u64 ip)
 175 {
 176         struct annotation *notes;
 177         struct symbol *sym;
 178         int err;
 179
 180         if (he == NULL || he->ms.sym == NULL ||
 181             ((top->sym_filter_entry == NULL ||
 182               top->sym_filter_entry->ms.sym != he->ms.sym) && use_browser != 1))
 183                 return;
 184
 185         sym = he->ms.sym;
 186         notes = symbol__annotation(sym);
 187
 188         if (pthread_mutex_trylock(&notes->lock))
 189                 return;
 190
 191         if (notes->src == NULL && symbol__alloc_hist(sym) < 0) {
 192                 pthread_mutex_unlock(&notes->lock);
 193                 pr_err("Not enough memory for annotating '%s' symbol!\n",
 194                        sym->name);
 195                 sleep(1);
 196                 return;
 197         }
 198
 199         ip = he->ms.map->map_ip(he->ms.map, ip);
 200         err = symbol__inc_addr_samples(sym, he->ms.map, counter, ip);
 201
 202         pthread_mutex_unlock(&notes->lock);
 203
 204         if (err == -ERANGE && !he->ms.map->erange_warned)
 205                 ui__warn_map_erange(he->ms.map, sym, ip);
 206 }
 207
 208 static void perf_top__show_details(struct perf_top *top)
 209 {
 210         struct hist_entry *he = top->sym_filter_entry;
 211         struct annotation *notes;
 212         struct symbol *symbol;
 213         int more;
 214
 215         if (!he)
 216                 return;
 217
 218         symbol = he->ms.sym;
 219         notes = symbol__annotation(symbol);
 220
 221         pthread_mutex_lock(&notes->lock);
 222
 223         if (notes->src == NULL)
 224                 goto out_unlock;
 225
 226         printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
 227         printf("  Events  Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
 228
 229         more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
 230                                        0, top->sym_pcnt_filter, top->print_entries, 4);
 231         if (top->zero)
 232                 symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
 233         else
 234                 symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
 235         if (more != 0)
 236                 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
 237 out_unlock:
 238         pthread_mutex_unlock(&notes->lock);
 239 }
 240
 241 static const char               CONSOLE_CLEAR[] = "\e[H\e[2J";
 242
 243 static struct hist_entry *perf_evsel__add_hist_entry(struct perf_evsel *evsel,
 244                                                      struct addr_location *al,
 245                                                      struct perf_sample *sample)
 246 {
 247         struct hist_entry *he;
 248
 249         pthread_mutex_lock(&evsel->hists.lock);
 250         he = __hists__add_entry(&evsel->hists, al, NULL, sample->period,
 251                                 sample->weight);
 252         pthread_mutex_unlock(&evsel->hists.lock);
 253
 254         if (he == NULL)
 255                 return NULL;
 256
 257         hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
 258         return he;
 259 }
 260
 261 static void perf_top__print_sym_table(struct perf_top *top)
 262 {
 263         char bf[160];
 264         int printed = 0;
 265         const int win_width = top->winsize.ws_col - 1;
 266
 267         puts(CONSOLE_CLEAR);
 268
 269         perf_top__header_snprintf(top, bf, sizeof(bf));
 270         printf("%s\n", bf);
 271
 272         perf_top__reset_sample_counters(top);
 273
 274         printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
 275
 276         if (top->sym_evsel->hists.stats.nr_lost_warned !=
 277             top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST]) {
 278                 top->sym_evsel->hists.stats.nr_lost_warned =
 279                         top->sym_evsel->hists.stats.nr_events[PERF_RECORD_LOST];
 280                 color_fprintf(stdout, PERF_COLOR_RED,
 281                               "WARNING: LOST %d chunks, Check IO/CPU overload",
 282                               top->sym_evsel->hists.stats.nr_lost_warned);
 283                 ++printed;
 284         }
 285
 286         if (top->sym_filter_entry) {
 287                 perf_top__show_details(top);
 288                 return;
 289         }
 290
 291         hists__collapse_resort(&top->sym_evsel->hists);
 292         hists__output_resort(&top->sym_evsel->hists);
 293         hists__decay_entries(&top->sym_evsel->hists,
 294                              top->hide_user_symbols,
 295                              top->hide_kernel_symbols);
 296         hists__output_recalc_col_len(&top->sym_evsel->hists,
 297                                      top->print_entries - printed);
 298         putchar('\n');
 299         hists__fprintf(&top->sym_evsel->hists, false,
 300                        top->print_entries - printed, win_width,
 301                        top->min_percent, stdout);
 302 }
 303
 304 static void prompt_integer(int *target, const char *msg)
 305 {
 306         char *buf = malloc(0), *p;
 307         size_t dummy = 0;
 308         int tmp;
 309
 310         fprintf(stdout, "\n%s: ", msg);
 311         if (getline(&buf, &dummy, stdin) < 0)
 312                 return;
 313
 314         p = strchr(buf, '\n');
 315         if (p)
 316                 *p = 0;
 317
 318         p = buf;
 319         while(*p) {
 320                 if (!isdigit(*p))
 321                         goto out_free;
 322                 p++;
 323         }
 324         tmp = strtoul(buf, NULL, 10);
 325         *target = tmp;
 326 out_free:
 327         free(buf);
 328 }
 329
 330 static void prompt_percent(int *target, const char *msg)
 331 {
 332         int tmp = 0;
 333
 334         prompt_integer(&tmp, msg);
 335         if (tmp >= 0 && tmp <= 100)
 336                 *target = tmp;
 337 }
 338
 339 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
 340 {
 341         char *buf = malloc(0), *p;
 342         struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
 343         struct rb_node *next;
 344         size_t dummy = 0;
 345
 346         /* zero counters of active symbol */
 347         if (syme) {
 348                 __zero_source_counters(syme);
 349                 top->sym_filter_entry = NULL;
 350         }
 351
 352         fprintf(stdout, "\n%s: ", msg);
 353         if (getline(&buf, &dummy, stdin) < 0)
 354                 goto out_free;
 355
 356         p = strchr(buf, '\n');
 357         if (p)
 358                 *p = 0;
 359
 360         next = rb_first(&top->sym_evsel->hists.entries);
 361         while (next) {
 362                 n = rb_entry(next, struct hist_entry, rb_node);
 363                 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
 364                         found = n;
 365                         break;
 366                 }
 367                 next = rb_next(&n->rb_node);
 368         }
 369
 370         if (!found) {
 371                 fprintf(stderr, "Sorry, %s is not active.\n", buf);
 372                 sleep(1);
 373         } else
 374                 perf_top__parse_source(top, found);
 375
 376 out_free:
 377         free(buf);
 378 }
 379
 380 static void perf_top__print_mapped_keys(struct perf_top *top)
 381 {
 382         char *name = NULL;
 383
 384         if (top->sym_filter_entry) {
 385                 struct symbol *sym = top->sym_filter_entry->ms.sym;
 386                 name = sym->name;
 387         }
 388
 389         fprintf(stdout, "\nMapped keys:\n");
 390         fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
 391         fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
 392
 393         if (top->evlist->nr_entries > 1)
 394                 fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
 395
 396         fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
 397
 398         fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
 399         fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
 400         fprintf(stdout, "\t[S]     stop annotation.\n");
 401
 402         fprintf(stdout,
 403                 "\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
 404                 top->hide_kernel_symbols ? "yes" : "no");
 405         fprintf(stdout,
 406                 "\t[U]     hide user symbols.               \t(%s)\n",
 407                 top->hide_user_symbols ? "yes" : "no");
 408         fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
 409         fprintf(stdout, "\t[qQ]    quit.\n");
 410 }
 411
 412 static int perf_top__key_mapped(struct perf_top *top, int c)
 413 {
 414         switch (c) {
 415                 case 'd':
 416                 case 'e':
 417                 case 'f':
 418                 case 'z':
 419                 case 'q':
 420                 case 'Q':
 421                 case 'K':
 422                 case 'U':
 423                 case 'F':
 424                 case 's':
 425                 case 'S':
 426                         return 1;
 427                 case 'E':
 428                         return top->evlist->nr_entries > 1 ? 1 : 0;
 429                 default:
 430                         break;
 431         }
 432
 433         return 0;
 434 }
 435
 436 static bool perf_top__handle_keypress(struct perf_top *top, int c)
 437 {
 438         bool ret = true;
 439
 440         if (!perf_top__key_mapped(top, c)) {
 441                 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 442                 struct termios tc, save;
 443
 444                 perf_top__print_mapped_keys(top);
 445                 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
 446                 fflush(stdout);
 447
 448                 tcgetattr(0, &save);
 449                 tc = save;
 450                 tc.c_lflag &= ~(ICANON | ECHO);
 451                 tc.c_cc[VMIN] = 0;
 452                 tc.c_cc[VTIME] = 0;
 453                 tcsetattr(0, TCSANOW, &tc);
 454
 455                 poll(&stdin_poll, 1, -1);
 456                 c = getc(stdin);
 457
 458                 tcsetattr(0, TCSAFLUSH, &save);
 459                 if (!perf_top__key_mapped(top, c))
 460                         return ret;
 461         }
 462
 463         switch (c) {
 464                 case 'd':
 465                         prompt_integer(&top->delay_secs, "Enter display delay");
 466                         if (top->delay_secs < 1)
 467                                 top->delay_secs = 1;
 468                         break;
 469                 case 'e':
 470                         prompt_integer(&top->print_entries, "Enter display entries (lines)");
 471                         if (top->print_entries == 0) {
 472                                 struct sigaction act = {
 473                                         .sa_sigaction = perf_top__sig_winch,
 474                                         .sa_flags     = SA_SIGINFO,
 475                                 };
 476                                 perf_top__sig_winch(SIGWINCH, NULL, top);
 477                                 sigaction(SIGWINCH, &act, NULL);
 478                         } else {
 479                                 signal(SIGWINCH, SIG_DFL);
 480                         }
 481                         break;
 482                 case 'E':
 483                         if (top->evlist->nr_entries > 1) {
 484                                 /* Select 0 as the default event: */
 485                                 int counter = 0;
 486
 487                                 fprintf(stderr, "\nAvailable events:");
 488
 489                                 list_for_each_entry(top->sym_evsel, &top->evlist->entries, node)
 490                                         fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
 491
 492                                 prompt_integer(&counter, "Enter details event counter");
 493
 494                                 if (counter >= top->evlist->nr_entries) {
 495                                         top->sym_evsel = perf_evlist__first(top->evlist);
 496                                         fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
 497                                         sleep(1);
 498                                         break;
 499                                 }
 500                                 list_for_each_entry(top->sym_evsel, &top->evlist->entries, node)
 501                                         if (top->sym_evsel->idx == counter)
 502                                                 break;
 503                         } else
 504                                 top->sym_evsel = perf_evlist__first(top->evlist);
 505                         break;
 506                 case 'f':
 507                         prompt_integer(&top->count_filter, "Enter display event count filter");
 508                         break;
 509                 case 'F':
 510                         prompt_percent(&top->sym_pcnt_filter,
 511                                        "Enter details display event filter (percent)");
 512                         break;
 513                 case 'K':
 514                         top->hide_kernel_symbols = !top->hide_kernel_symbols;
 515                         break;
 516                 case 'q':
 517                 case 'Q':
 518                         printf("exiting.\n");
 519                         if (top->dump_symtab)
 520                                 perf_session__fprintf_dsos(top->session, stderr);
 521                         ret = false;
 522                         break;
 523                 case 's':
 524                         perf_top__prompt_symbol(top, "Enter details symbol");
 525                         break;
 526                 case 'S':
 527                         if (!top->sym_filter_entry)
 528                                 break;
 529                         else {
 530                                 struct hist_entry *syme = top->sym_filter_entry;
 531
 532                                 top->sym_filter_entry = NULL;
 533                                 __zero_source_counters(syme);
 534                         }
 535                         break;
 536                 case 'U':
 537                         top->hide_user_symbols = !top->hide_user_symbols;
 538                         break;
 539                 case 'z':
 540                         top->zero = !top->zero;
 541                         break;
 542                 default:
 543                         break;
 544         }
 545
 546         return ret;
 547 }
 548
 549 static void perf_top__sort_new_samples(void *arg)
 550 {
 551         struct perf_top *t = arg;
 552         perf_top__reset_sample_counters(t);
 553
 554         if (t->evlist->selected != NULL)
 555                 t->sym_evsel = t->evlist->selected;
 556
 557         hists__collapse_resort(&t->sym_evsel->hists);
 558         hists__output_resort(&t->sym_evsel->hists);
 559         hists__decay_entries(&t->sym_evsel->hists,
 560                              t->hide_user_symbols,
 561                              t->hide_kernel_symbols);
 562 }
 563
 564 static void *display_thread_tui(void *arg)
 565 {
 566         struct perf_evsel *pos;
 567         struct perf_top *top = arg;
 568         const char *help = "For a higher level overview, try: perf top --sort comm,dso";
 569         struct hist_browser_timer hbt = {
 570                 .timer          = perf_top__sort_new_samples,
 571                 .arg            = top,
 572                 .refresh        = top->delay_secs,
 573         };
 574
 575         perf_top__sort_new_samples(top);
 576
 577         /*
 578          * Initialize the uid_filter_str, in the future the TUI will allow
 579          * Zooming in/out UIDs. For now juse use whatever the user passed
 580          * via --uid.
 581          */
 582         list_for_each_entry(pos, &top->evlist->entries, node)
 583                 pos->hists.uid_filter_str = top->record_opts.target.uid_str;
 584
 585         perf_evlist__tui_browse_hists(top->evlist, help, &hbt, top->min_percent,
 586                                       &top->session->header.env);
 587
 588         done = 1;
 589         return NULL;
 590 }
 591
 592 static void *display_thread(void *arg)
 593 {
 594         struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
 595         struct termios tc, save;
 596         struct perf_top *top = arg;
 597         int delay_msecs, c;
 598
 599         tcgetattr(0, &save);
 600         tc = save;
 601         tc.c_lflag &= ~(ICANON | ECHO);
 602         tc.c_cc[VMIN] = 0;
 603         tc.c_cc[VTIME] = 0;
 604
 605         pthread__unblock_sigwinch();
 606 repeat:
 607         delay_msecs = top->delay_secs * 1000;
 608         tcsetattr(0, TCSANOW, &tc);
 609         /* trash return*/
 610         getc(stdin);
 611
 612         while (!done) {
 613                 perf_top__print_sym_table(top);
 614                 /*
 615                  * Either timeout expired or we got an EINTR due to SIGWINCH,
 616                  * refresh screen in both cases.
 617                  */
 618                 switch (poll(&stdin_poll, 1, delay_msecs)) {
 619                 case 0:
 620                         continue;
 621                 case -1:
 622                         if (errno == EINTR)
 623                                 continue;
 624                         /* Fall trhu */
 625                 default:
 626                         c = getc(stdin);
 627                         tcsetattr(0, TCSAFLUSH, &save);
 628
 629                         if (perf_top__handle_keypress(top, c))
 630                                 goto repeat;
 631                         done = 1;
 632                 }
 633         }
 634
 635         return NULL;
 636 }
 637
 638 /* Tag samples to be skipped. */
 639 static const char *skip_symbols[] = {
 640         "intel_idle",
 641         "default_idle",
 642         "native_safe_halt",
 643         "cpu_idle",
 644         "enter_idle",
 645         "exit_idle",
 646         "mwait_idle",
 647         "mwait_idle_with_hints",
 648         "poll_idle",
 649         "ppc64_runlatch_off",
 650         "pseries_dedicated_idle_sleep",
 651         NULL
 652 };
 653
 654 static int symbol_filter(struct map *map __maybe_unused, struct symbol *sym)
 655 {
 656         const char *name = sym->name;
 657         int i;
 658
 659         /*
 660          * ppc64 uses function descriptors and appends a '.' to the
 661          * start of every instruction address. Remove it.
 662          */
 663         if (name[0] == '.')
 664                 name++;
 665
 666         if (!strcmp(name, "_text") ||
 667             !strcmp(name, "_etext") ||
 668             !strcmp(name, "_sinittext") ||
 669             !strncmp("init_module", name, 11) ||
 670             !strncmp("cleanup_module", name, 14) ||
 671             strstr(name, "_text_start") ||
 672             strstr(name, "_text_end"))
 673                 return 1;
 674
 675         for (i = 0; skip_symbols[i]; i++) {
 676                 if (!strcmp(skip_symbols[i], name)) {
 677                         sym->ignore = true;
 678                         break;
 679                 }
 680         }
 681
 682         return 0;
 683 }
 684
 685 static void perf_event__process_sample(struct perf_tool *tool,
 686                                        const union perf_event *event,
 687                                        struct perf_evsel *evsel,
 688                                        struct perf_sample *sample,
 689                                        struct machine *machine)
 690 {
 691         struct perf_top *top = container_of(tool, struct perf_top, tool);
 692         struct symbol *parent = NULL;
 693         u64 ip = event->ip.ip;
 694         struct addr_location al;
 695         int err;
 696
 697         if (!machine && perf_guest) {
 698                 static struct intlist *seen;
 699
 700                 if (!seen)
 701                         seen = intlist__new(NULL);
 702
 703                 if (!intlist__has_entry(seen, event->ip.pid)) {
 704                         pr_err("Can't find guest [%d]'s kernel information\n",
 705                                 event->ip.pid);
 706                         intlist__add(seen, event->ip.pid);
 707                 }
 708                 return;
 709         }
 710
 711         if (!machine) {
 712                 pr_err("%u unprocessable samples recorded.\r",
 713                        top->session->stats.nr_unprocessable_samples++);
 714                 return;
 715         }
 716
 717         if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
 718                 top->exact_samples++;
 719
 720         if (perf_event__preprocess_sample(event, machine, &al, sample,
 721                                           symbol_filter) < 0 ||
 722             al.filtered)
 723                 return;
 724
 725         if (!top->kptr_restrict_warned &&
 726             symbol_conf.kptr_restrict &&
 727             al.cpumode == PERF_RECORD_MISC_KERNEL) {
 728                 ui__warning(
 729 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
 730 "Check /proc/sys/kernel/kptr_restrict.\n\n"
 731 "Kernel%s samples will not be resolved.\n",
 732                           !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
 733                           " modules" : "");
 734                 if (use_browser <= 0)
 735                         sleep(5);
 736                 top->kptr_restrict_warned = true;
 737         }
 738
 739         if (al.sym == NULL) {
 740                 const char *msg = "Kernel samples will not be resolved.\n";
 741                 /*
 742                  * As we do lazy loading of symtabs we only will know if the
 743                  * specified vmlinux file is invalid when we actually have a
 744                  * hit in kernel space and then try to load it. So if we get
 745                  * here and there are _no_ symbols in the DSO backing the
 746                  * kernel map, bail out.
 747                  *
 748                  * We may never get here, for instance, if we use -K/
 749                  * --hide-kernel-symbols, even if the user specifies an
 750                  * invalid --vmlinux ;-)
 751                  */
 752                 if (!top->kptr_restrict_warned && !top->vmlinux_warned &&
 753                     al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
 754                     RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
 755                         if (symbol_conf.vmlinux_name) {
 756                                 ui__warning("The %s file can't be used.\n%s",
 757                                             symbol_conf.vmlinux_name, msg);
 758                         } else {
 759                                 ui__warning("A vmlinux file was not found.\n%s",
 760                                             msg);
 761                         }
 762
 763                         if (use_browser <= 0)
 764                                 sleep(5);
 765                         top->vmlinux_warned = true;
 766                 }
 767         }
 768
 769         if (al.sym == NULL || !al.sym->ignore) {
 770                 struct hist_entry *he;
 771
 772                 if ((sort__has_parent || symbol_conf.use_callchain) &&
 773                     sample->callchain) {
 774                         err = machine__resolve_callchain(machine, evsel,
 775                                                          al.thread, sample,
 776                                                          &parent);
 777
 778                         if (err)
 779                                 return;
 780                 }
 781
 782                 he = perf_evsel__add_hist_entry(evsel, &al, sample);
 783                 if (he == NULL) {
 784                         pr_err("Problem incrementing symbol period, skipping event\n");
 785                         return;
 786                 }
 787
 788                 if (symbol_conf.use_callchain) {
 789                         err = callchain_append(he->callchain, &callchain_cursor,
 790                                                sample->period);
 791                         if (err)
 792                                 return;
 793                 }
 794
 795                 if (sort__has_sym)
 796                         perf_top__record_precise_ip(top, he, evsel->idx, ip);
 797         }
 798
 799         return;
 800 }
 801
 802 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
 803 {
 804         struct perf_sample sample;
 805         struct perf_evsel *evsel;
 806         struct perf_session *session = top->session;
 807         union perf_event *event;
 808         struct machine *machine;
 809         u8 origin;
 810         int ret;
 811
 812         while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
 813                 ret = perf_evlist__parse_sample(top->evlist, event, &sample);
 814                 if (ret) {
 815                         pr_err("Can't parse sample, err = %d\n", ret);
 816                         continue;
 817                 }
 818
 819                 evsel = perf_evlist__id2evsel(session->evlist, sample.id);
 820                 assert(evsel != NULL);
 821
 822                 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
 823
 824                 if (event->header.type == PERF_RECORD_SAMPLE)
 825                         ++top->samples;
 826
 827                 switch (origin) {
 828                 case PERF_RECORD_MISC_USER:
 829                         ++top->us_samples;
 830                         if (top->hide_user_symbols)
 831                                 continue;
 832                         machine = &session->machines.host;
 833                         break;
 834                 case PERF_RECORD_MISC_KERNEL:
 835                         ++top->kernel_samples;
 836                         if (top->hide_kernel_symbols)
 837                                 continue;
 838                         machine = &session->machines.host;
 839                         break;
 840                 case PERF_RECORD_MISC_GUEST_KERNEL:
 841                         ++top->guest_kernel_samples;
 842                         machine = perf_session__find_machine(session, event->ip.pid);
 843                         break;
 844                 case PERF_RECORD_MISC_GUEST_USER:
 845                         ++top->guest_us_samples;
 846                         /*
 847                          * TODO: we don't process guest user from host side
 848                          * except simple counting.
 849                          */
 850                         /* Fall thru */
 851                 default:
 852                         continue;
 853                 }
 854
 855
 856                 if (event->header.type == PERF_RECORD_SAMPLE) {
 857                         perf_event__process_sample(&top->tool, event, evsel,
 858                                                    &sample, machine);
 859                 } else if (event->header.type < PERF_RECORD_MAX) {
 860                         hists__inc_nr_events(&evsel->hists, event->header.type);
 861                         machine__process_event(machine, event);
 862                 } else
 863                         ++session->stats.nr_unknown_events;
 864         }
 865 }
 866
 867 static void perf_top__mmap_read(struct perf_top *top)
 868 {
 869         int i;
 870
 871         for (i = 0; i < top->evlist->nr_mmaps; i++)
 872                 perf_top__mmap_read_idx(top, i);
 873 }
 874
 875 static int perf_top__start_counters(struct perf_top *top)
 876 {
 877         char msg[512];
 878         struct perf_evsel *counter;
 879         struct perf_evlist *evlist = top->evlist;
 880         struct perf_record_opts *opts = &top->record_opts;
 881
 882         perf_evlist__config(evlist, opts);
 883
 884         list_for_each_entry(counter, &evlist->entries, node) {
 885 try_again:
 886                 if (perf_evsel__open(counter, top->evlist->cpus,
 887                                      top->evlist->threads) < 0) {
 888                         if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
 889                                 if (verbose)
 890                                         ui__warning("%s\n", msg);
 891                                 goto try_again;
 892                         }
 893
 894                         perf_evsel__open_strerror(counter, &opts->target,
 895                                                   errno, msg, sizeof(msg));
 896                         ui__error("%s\n", msg);
 897                         goto out_err;
 898                 }
 899         }
 900
 901         if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
 902                 ui__error("Failed to mmap with %d (%s)\n",
 903                             errno, strerror(errno));
 904                 goto out_err;
 905         }
 906
 907         return 0;
 908
 909 out_err:
 910         return -1;
 911 }
 912
 913 static int perf_top__setup_sample_type(struct perf_top *top __maybe_unused)
 914 {
 915         if (!sort__has_sym) {
 916                 if (symbol_conf.use_callchain) {
 917                         ui__error("Selected -g but \"sym\" not present in --sort/-s.");
 918                         return -EINVAL;
 919                 }
 920         } else if (callchain_param.mode != CHAIN_NONE) {
 921                 if (callchain_register_param(&callchain_param) < 0) {
 922                         ui__error("Can't register callchain params.\n");
 923                         return -EINVAL;
 924                 }
 925         }
 926
 927         return 0;
 928 }
 929
 930 static int __cmd_top(struct perf_top *top)
 931 {
 932         struct perf_record_opts *opts = &top->record_opts;
 933         pthread_t thread;
 934         int ret;
 935         /*
 936          * FIXME: perf_session__new should allow passing a O_MMAP, so that all this
 937          * mmap reading, etc is encapsulated in it. Use O_WRONLY for now.
 938          */
 939         top->session = perf_session__new(NULL, O_WRONLY, false, false, NULL);
 940         if (top->session == NULL)
 941                 return -ENOMEM;
 942
 943         if (!objdump_path) {
 944                 ret = perf_session_env__lookup_objdump(&top->session->header.env);
 945                 if (ret)
 946                         goto out_delete;
 947         }
 948
 949         ret = perf_top__setup_sample_type(top);
 950         if (ret)
 951                 goto out_delete;
 952
 953         if (perf_target__has_task(&opts->target))
 954                 perf_event__synthesize_thread_map(&top->tool, top->evlist->threads,
 955                                                   perf_event__process,
 956                                                   &top->session->machines.host);
 957         else
 958                 perf_event__synthesize_threads(&top->tool, perf_event__process,
 959                                                &top->session->machines.host);
 960
 961         ret = perf_top__start_counters(top);
 962         if (ret)
 963                 goto out_delete;
 964
 965         top->session->evlist = top->evlist;
 966         perf_session__set_id_hdr_size(top->session);
 967
 968         /*
 969          * When perf is starting the traced process, all the events (apart from
 970          * group members) have enable_on_exec=1 set, so don't spoil it by
 971          * prematurely enabling them.
 972          *
 973          * XXX 'top' still doesn't start workloads like record, trace, but should,
 974          * so leave the check here.
 975          */
 976         if (!perf_target__none(&opts->target))
 977                 perf_evlist__enable(top->evlist);
 978
 979         /* Wait for a minimal set of events before starting the snapshot */
 980         poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
 981
 982         perf_top__mmap_read(top);
 983
 984         ret = -1;
 985         if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
 986                                                             display_thread), top)) {
 987                 ui__error("Could not create display thread.\n");
 988                 goto out_delete;
 989         }
 990
 991         if (top->realtime_prio) {
 992                 struct sched_param param;
 993
 994                 param.sched_priority = top->realtime_prio;
 995                 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
 996                         ui__error("Could not set realtime priority.\n");
 997                         goto out_delete;
 998                 }
 999         }
1000
1001         while (!done) {
1002                 u64 hits = top->samples;
1003
1004                 perf_top__mmap_read(top);
1005
1006                 if (hits == top->samples)
1007                         ret = poll(top->evlist->pollfd, top->evlist->nr_fds, 100);
1008         }
1009
1010         ret = 0;
1011 out_delete:
1012         perf_session__delete(top->session);
1013         top->session = NULL;
1014
1015         return ret;
1016 }
1017
1018 static int
1019 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1020 {
1021         /*
1022          * --no-call-graph
1023          */
1024         if (unset)
1025                 return 0;
1026
1027         symbol_conf.use_callchain = true;
1028
1029         return record_parse_callchain_opt(opt, arg, unset);
1030 }
1031
1032 static int
1033 parse_percent_limit(const struct option *opt, const char *arg,
1034                     int unset __maybe_unused)
1035 {
1036         struct perf_top *top = opt->value;
1037
1038         top->min_percent = strtof(arg, NULL);
1039         return 0;
1040 }
1041
1042 int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1043 {
1044         int status;
1045         char errbuf[BUFSIZ];
1046         struct perf_top top = {
1047                 .count_filter        = 5,
1048                 .delay_secs          = 2,
1049                 .record_opts = {
1050                         .mmap_pages     = UINT_MAX,
1051                         .user_freq      = UINT_MAX,
1052                         .user_interval  = ULLONG_MAX,
1053                         .freq           = 4000, /* 4 KHz */
1054                         .target              = {
1055                                 .uses_mmap   = true,
1056                         },
1057                 },
1058                 .sym_pcnt_filter     = 5,
1059         };
1060         struct perf_record_opts *opts = &top.record_opts;
1061         struct perf_target *target = &opts->target;
1062         const struct option options[] = {
1063         OPT_CALLBACK('e', "event", &top.evlist, "event",
1064                      "event selector. use 'perf list' to list available events",
1065                      parse_events_option),
1066         OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1067         OPT_STRING('p', "pid", &target->pid, "pid",
1068                     "profile events on existing process id"),
1069         OPT_STRING('t', "tid", &target->tid, "tid",
1070                     "profile events on existing thread id"),
1071         OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1072                             "system-wide collection from all CPUs"),
1073         OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1074                     "list of cpus to monitor"),
1075         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1076                    "file", "vmlinux pathname"),
1077         OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1078                     "hide kernel symbols"),
1079         OPT_UINTEGER('m', "mmap-pages", &opts->mmap_pages,
1080                      "number of mmap data pages"),
1081         OPT_INTEGER('r', "realtime", &top.realtime_prio,
1082                     "collect data with this RT SCHED_FIFO priority"),
1083         OPT_INTEGER('d', "delay", &top.delay_secs,
1084                     "number of seconds to delay between refreshes"),
1085         OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1086                             "dump the symbol table used for profiling"),
1087         OPT_INTEGER('f', "count-filter", &top.count_filter,
1088                     "only display functions with more events than this"),
1089         OPT_BOOLEAN('g', "group", &opts->group,
1090                             "put the counters into a counter group"),
1091         OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1092                     "child tasks do not inherit counters"),
1093         OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1094                     "symbol to annotate"),
1095         OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1096         OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1097         OPT_INTEGER('E', "entries", &top.print_entries,
1098                     "display this many functions"),
1099         OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1100                     "hide user symbols"),
1101         OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1102         OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1103         OPT_INCR('v', "verbose", &verbose,
1104                     "be more verbose (show counter open errors, etc)"),
1105         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1106                    "sort by key(s): pid, comm, dso, symbol, parent, weight, local_weight"),
1107         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1108                     "Show a column with the number of samples"),
1109         OPT_CALLBACK_DEFAULT('G', "call-graph", &top.record_opts,
1110                              "mode[,dump_size]", record_callchain_help,
1111                              &parse_callchain_opt, "fp"),
1112         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1113                     "Show a column with the sum of periods"),
1114         OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1115                    "only consider symbols in these dsos"),
1116         OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1117                    "only consider symbols in these comms"),
1118         OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1119                    "only consider these symbols"),
1120         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1121                     "Interleave source code with assembly code (default)"),
1122         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1123                     "Display raw encoding of assembly instructions (default)"),
1124         OPT_STRING(0, "objdump", &objdump_path, "path",
1125                     "objdump binary to use for disassembly and annotations"),
1126         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1127                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
1128         OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1129         OPT_CALLBACK(0, "percent-limit", &top, "percent",
1130                      "Don't show entries under that percent", parse_percent_limit),
1131         OPT_END()
1132         };
1133         const char * const top_usage[] = {
1134                 "perf top [<options>]",
1135                 NULL
1136         };
1137
1138         top.evlist = perf_evlist__new();
1139         if (top.evlist == NULL)
1140                 return -ENOMEM;
1141
1142         argc = parse_options(argc, argv, options, top_usage, 0);
1143         if (argc)
1144                 usage_with_options(top_usage, options);
1145
1146         if (sort_order == default_sort_order)
1147                 sort_order = "dso,symbol";
1148
1149         if (setup_sorting() < 0)
1150                 usage_with_options(top_usage, options);
1151
1152         /* display thread wants entries to be collapsed in a different tree */
1153         sort__need_collapse = 1;
1154
1155         if (top.use_stdio)
1156                 use_browser = 0;
1157         else if (top.use_tui)
1158                 use_browser = 1;
1159
1160         setup_browser(false);
1161
1162         status = perf_target__validate(target);
1163         if (status) {
1164                 perf_target__strerror(target, status, errbuf, BUFSIZ);
1165                 ui__warning("%s", errbuf);
1166         }
1167
1168         status = perf_target__parse_uid(target);
1169         if (status) {
1170                 int saved_errno = errno;
1171
1172                 perf_target__strerror(target, status, errbuf, BUFSIZ);
1173                 ui__error("%s", errbuf);
1174
1175                 status = -saved_errno;
1176                 goto out_delete_evlist;
1177         }
1178
1179         if (perf_target__none(target))
1180                 target->system_wide = true;
1181
1182         if (perf_evlist__create_maps(top.evlist, target) < 0)
1183                 usage_with_options(top_usage, options);
1184
1185         if (!top.evlist->nr_entries &&
1186             perf_evlist__add_default(top.evlist) < 0) {
1187                 ui__error("Not enough memory for event selector list\n");
1188                 goto out_delete_maps;
1189         }
1190
1191         symbol_conf.nr_events = top.evlist->nr_entries;
1192
1193         if (top.delay_secs < 1)
1194                 top.delay_secs = 1;
1195
1196         if (opts->user_interval != ULLONG_MAX)
1197                 opts->default_interval = opts->user_interval;
1198         if (opts->user_freq != UINT_MAX)
1199                 opts->freq = opts->user_freq;
1200
1201         /*
1202          * User specified count overrides default frequency.
1203          */
1204         if (opts->default_interval)
1205                 opts->freq = 0;
1206         else if (opts->freq) {
1207                 opts->default_interval = opts->freq;
1208         } else {
1209                 ui__error("frequency and count are zero, aborting\n");
1210                 status = -EINVAL;
1211                 goto out_delete_maps;
1212         }
1213
1214         top.sym_evsel = perf_evlist__first(top.evlist);
1215
1216         symbol_conf.priv_size = sizeof(struct annotation);
1217
1218         symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1219         if (symbol__init() < 0)
1220                 return -1;
1221
1222         sort__setup_elide(stdout);
1223
1224         get_term_dimensions(&top.winsize);
1225         if (top.print_entries == 0) {
1226                 struct sigaction act = {
1227                         .sa_sigaction = perf_top__sig_winch,
1228                         .sa_flags     = SA_SIGINFO,
1229                 };
1230                 perf_top__update_print_entries(&top);
1231                 sigaction(SIGWINCH, &act, NULL);
1232         }
1233
1234         status = __cmd_top(&top);
1235
1236 out_delete_maps:
1237         perf_evlist__delete_maps(top.evlist);
1238 out_delete_evlist:
1239         perf_evlist__delete(top.evlist);
1240
1241         return status;
1242 }