~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/tools/perf/builtin-top.c

Version: ~ [ linux-6.1-rc7 ] ~ [ linux-6.0.10 ] ~ [ linux-5.19.17 ] ~ [ linux-5.18.19 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.80 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.156 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.225 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.267 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.300 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.334 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 // SPDX-License-Identifier: GPL-2.0-only
  2 /*
  3  * builtin-top.c
  4  *
  5  * Builtin top command: Display a continuously updated profile of
  6  * any workload, CPU or specific PID.
  7  *
  8  * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
  9  *               2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
 10  *
 11  * Improvements and fixes by:
 12  *
 13  *   Arjan van de Ven <arjan@linux.intel.com>
 14  *   Yanmin Zhang <yanmin.zhang@intel.com>
 15  *   Wu Fengguang <fengguang.wu@intel.com>
 16  *   Mike Galbraith <efault@gmx.de>
 17  *   Paul Mackerras <paulus@samba.org>
 18  */
 19 #include "builtin.h"
 20 
 21 #include "perf.h"
 22 
 23 #include "util/annotate.h"
 24 #include "util/bpf-event.h"
 25 #include "util/config.h"
 26 #include "util/color.h"
 27 #include "util/dso.h"
 28 #include "util/evlist.h"
 29 #include "util/evsel.h"
 30 #include "util/evsel_config.h"
 31 #include "util/event.h"
 32 #include "util/machine.h"
 33 #include "util/map.h"
 34 #include "util/mmap.h"
 35 #include "util/session.h"
 36 #include "util/symbol.h"
 37 #include "util/synthetic-events.h"
 38 #include "util/top.h"
 39 #include "util/util.h"
 40 #include <linux/rbtree.h>
 41 #include <subcmd/parse-options.h>
 42 #include "util/parse-events.h"
 43 #include "util/callchain.h"
 44 #include "util/cpumap.h"
 45 #include "util/sort.h"
 46 #include "util/string2.h"
 47 #include "util/term.h"
 48 #include "util/intlist.h"
 49 #include "util/parse-branch-options.h"
 50 #include "arch/common.h"
 51 #include "ui/ui.h"
 52 
 53 #include "util/debug.h"
 54 #include "util/ordered-events.h"
 55 
 56 #include <assert.h>
 57 #include <elf.h>
 58 #include <fcntl.h>
 59 
 60 #include <stdio.h>
 61 #include <termios.h>
 62 #include <unistd.h>
 63 #include <inttypes.h>
 64 
 65 #include <errno.h>
 66 #include <time.h>
 67 #include <sched.h>
 68 #include <signal.h>
 69 
 70 #include <sys/syscall.h>
 71 #include <sys/ioctl.h>
 72 #include <poll.h>
 73 #include <sys/prctl.h>
 74 #include <sys/wait.h>
 75 #include <sys/uio.h>
 76 #include <sys/utsname.h>
 77 #include <sys/mman.h>
 78 
 79 #include <linux/stringify.h>
 80 #include <linux/time64.h>
 81 #include <linux/types.h>
 82 #include <linux/err.h>
 83 
 84 #include <linux/ctype.h>
 85 #include <perf/mmap.h>
 86 
 87 static volatile int done;
 88 static volatile int resize;
 89 
 90 #define HEADER_LINE_NR  5
 91 
 92 static void perf_top__update_print_entries(struct perf_top *top)
 93 {
 94         top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
 95 }
 96 
 97 static void winch_sig(int sig __maybe_unused)
 98 {
 99         resize = 1;
100 }
101 
102 static void perf_top__resize(struct perf_top *top)
103 {
104         get_term_dimensions(&top->winsize);
105         perf_top__update_print_entries(top);
106 }
107 
108 static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
109 {
110         struct evsel *evsel;
111         struct symbol *sym;
112         struct annotation *notes;
113         struct map *map;
114         int err = -1;
115 
116         if (!he || !he->ms.sym)
117                 return -1;
118 
119         evsel = hists_to_evsel(he->hists);
120 
121         sym = he->ms.sym;
122         map = he->ms.map;
123 
124         /*
125          * We can't annotate with just /proc/kallsyms
126          */
127         if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
128             !dso__is_kcore(map->dso)) {
129                 pr_err("Can't annotate %s: No vmlinux file was found in the "
130                        "path\n", sym->name);
131                 sleep(1);
132                 return -1;
133         }
134 
135         notes = symbol__annotation(sym);
136         pthread_mutex_lock(&notes->lock);
137 
138         if (!symbol__hists(sym, top->evlist->core.nr_entries)) {
139                 pthread_mutex_unlock(&notes->lock);
140                 pr_err("Not enough memory for annotating '%s' symbol!\n",
141                        sym->name);
142                 sleep(1);
143                 return err;
144         }
145 
146         err = symbol__annotate(&he->ms, evsel, &top->annotation_opts, NULL);
147         if (err == 0) {
148                 top->sym_filter_entry = he;
149         } else {
150                 char msg[BUFSIZ];
151                 symbol__strerror_disassemble(&he->ms, err, msg, sizeof(msg));
152                 pr_err("Couldn't annotate %s: %s\n", sym->name, msg);
153         }
154 
155         pthread_mutex_unlock(&notes->lock);
156         return err;
157 }
158 
159 static void __zero_source_counters(struct hist_entry *he)
160 {
161         struct symbol *sym = he->ms.sym;
162         symbol__annotate_zero_histograms(sym);
163 }
164 
165 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
166 {
167         struct utsname uts;
168         int err = uname(&uts);
169 
170         ui__warning("Out of bounds address found:\n\n"
171                     "Addr:   %" PRIx64 "\n"
172                     "DSO:    %s %c\n"
173                     "Map:    %" PRIx64 "-%" PRIx64 "\n"
174                     "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
175                     "Arch:   %s\n"
176                     "Kernel: %s\n"
177                     "Tools:  %s\n\n"
178                     "Not all samples will be on the annotation output.\n\n"
179                     "Please report to linux-kernel@vger.kernel.org\n",
180                     ip, map->dso->long_name, dso__symtab_origin(map->dso),
181                     map->start, map->end, sym->start, sym->end,
182                     sym->binding == STB_GLOBAL ? 'g' :
183                     sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
184                     err ? "[unknown]" : uts.machine,
185                     err ? "[unknown]" : uts.release, perf_version_string);
186         if (use_browser <= 0)
187                 sleep(5);
188 
189         map->erange_warned = true;
190 }
191 
192 static void perf_top__record_precise_ip(struct perf_top *top,
193                                         struct hist_entry *he,
194                                         struct perf_sample *sample,
195                                         struct evsel *evsel, u64 ip)
196 {
197         struct annotation *notes;
198         struct symbol *sym = he->ms.sym;
199         int err = 0;
200 
201         if (sym == NULL || (use_browser == 0 &&
202                             (top->sym_filter_entry == NULL ||
203                              top->sym_filter_entry->ms.sym != sym)))
204                 return;
205 
206         notes = symbol__annotation(sym);
207 
208         if (pthread_mutex_trylock(&notes->lock))
209                 return;
210 
211         err = hist_entry__inc_addr_samples(he, sample, evsel, ip);
212 
213         pthread_mutex_unlock(&notes->lock);
214 
215         if (unlikely(err)) {
216                 /*
217                  * This function is now called with he->hists->lock held.
218                  * Release it before going to sleep.
219                  */
220                 pthread_mutex_unlock(&he->hists->lock);
221 
222                 if (err == -ERANGE && !he->ms.map->erange_warned)
223                         ui__warn_map_erange(he->ms.map, sym, ip);
224                 else if (err == -ENOMEM) {
225                         pr_err("Not enough memory for annotating '%s' symbol!\n",
226                                sym->name);
227                         sleep(1);
228                 }
229 
230                 pthread_mutex_lock(&he->hists->lock);
231         }
232 }
233 
234 static void perf_top__show_details(struct perf_top *top)
235 {
236         struct hist_entry *he = top->sym_filter_entry;
237         struct evsel *evsel;
238         struct annotation *notes;
239         struct symbol *symbol;
240         int more;
241 
242         if (!he)
243                 return;
244 
245         evsel = hists_to_evsel(he->hists);
246 
247         symbol = he->ms.sym;
248         notes = symbol__annotation(symbol);
249 
250         pthread_mutex_lock(&notes->lock);
251 
252         symbol__calc_percent(symbol, evsel);
253 
254         if (notes->src == NULL)
255                 goto out_unlock;
256 
257         printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
258         printf("  Events  Pcnt (>=%d%%)\n", top->annotation_opts.min_pcnt);
259 
260         more = symbol__annotate_printf(&he->ms, top->sym_evsel, &top->annotation_opts);
261 
262         if (top->evlist->enabled) {
263                 if (top->zero)
264                         symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
265                 else
266                         symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
267         }
268         if (more != 0)
269                 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
270 out_unlock:
271         pthread_mutex_unlock(&notes->lock);
272 }
273 
274 static void perf_top__resort_hists(struct perf_top *t)
275 {
276         struct evlist *evlist = t->evlist;
277         struct evsel *pos;
278 
279         evlist__for_each_entry(evlist, pos) {
280                 struct hists *hists = evsel__hists(pos);
281 
282                 /*
283                  * unlink existing entries so that they can be linked
284                  * in a correct order in hists__match() below.
285                  */
286                 hists__unlink(hists);
287 
288                 if (evlist->enabled) {
289                         if (t->zero) {
290                                 hists__delete_entries(hists);
291                         } else {
292                                 hists__decay_entries(hists, t->hide_user_symbols,
293                                                      t->hide_kernel_symbols);
294                         }
295                 }
296 
297                 hists__collapse_resort(hists, NULL);
298 
299                 /* Non-group events are considered as leader */
300                 if (symbol_conf.event_group &&
301                     !perf_evsel__is_group_leader(pos)) {
302                         struct hists *leader_hists = evsel__hists(pos->leader);
303 
304                         hists__match(leader_hists, hists);
305                         hists__link(leader_hists, hists);
306                 }
307         }
308 
309         evlist__for_each_entry(evlist, pos) {
310                 perf_evsel__output_resort(pos, NULL);
311         }
312 }
313 
314 static void perf_top__print_sym_table(struct perf_top *top)
315 {
316         char bf[160];
317         int printed = 0;
318         const int win_width = top->winsize.ws_col - 1;
319         struct evsel *evsel = top->sym_evsel;
320         struct hists *hists = evsel__hists(evsel);
321 
322         puts(CONSOLE_CLEAR);
323 
324         perf_top__header_snprintf(top, bf, sizeof(bf));
325         printf("%s\n", bf);
326 
327         printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
328 
329         if (!top->record_opts.overwrite &&
330             (hists->stats.nr_lost_warned !=
331             hists->stats.nr_events[PERF_RECORD_LOST])) {
332                 hists->stats.nr_lost_warned =
333                               hists->stats.nr_events[PERF_RECORD_LOST];
334                 color_fprintf(stdout, PERF_COLOR_RED,
335                               "WARNING: LOST %d chunks, Check IO/CPU overload",
336                               hists->stats.nr_lost_warned);
337                 ++printed;
338         }
339 
340         if (top->sym_filter_entry) {
341                 perf_top__show_details(top);
342                 return;
343         }
344 
345         perf_top__resort_hists(top);
346 
347         hists__output_recalc_col_len(hists, top->print_entries - printed);
348         putchar('\n');
349         hists__fprintf(hists, false, top->print_entries - printed, win_width,
350                        top->min_percent, stdout, !symbol_conf.use_callchain);
351 }
352 
353 static void prompt_integer(int *target, const char *msg)
354 {
355         char *buf = malloc(0), *p;
356         size_t dummy = 0;
357         int tmp;
358 
359         fprintf(stdout, "\n%s: ", msg);
360         if (getline(&buf, &dummy, stdin) < 0)
361                 return;
362 
363         p = strchr(buf, '\n');
364         if (p)
365                 *p = 0;
366 
367         p = buf;
368         while(*p) {
369                 if (!isdigit(*p))
370                         goto out_free;
371                 p++;
372         }
373         tmp = strtoul(buf, NULL, 10);
374         *target = tmp;
375 out_free:
376         free(buf);
377 }
378 
379 static void prompt_percent(int *target, const char *msg)
380 {
381         int tmp = 0;
382 
383         prompt_integer(&tmp, msg);
384         if (tmp >= 0 && tmp <= 100)
385                 *target = tmp;
386 }
387 
388 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
389 {
390         char *buf = malloc(0), *p;
391         struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
392         struct hists *hists = evsel__hists(top->sym_evsel);
393         struct rb_node *next;
394         size_t dummy = 0;
395 
396         /* zero counters of active symbol */
397         if (syme) {
398                 __zero_source_counters(syme);
399                 top->sym_filter_entry = NULL;
400         }
401 
402         fprintf(stdout, "\n%s: ", msg);
403         if (getline(&buf, &dummy, stdin) < 0)
404                 goto out_free;
405 
406         p = strchr(buf, '\n');
407         if (p)
408                 *p = 0;
409 
410         next = rb_first_cached(&hists->entries);
411         while (next) {
412                 n = rb_entry(next, struct hist_entry, rb_node);
413                 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
414                         found = n;
415                         break;
416                 }
417                 next = rb_next(&n->rb_node);
418         }
419 
420         if (!found) {
421                 fprintf(stderr, "Sorry, %s is not active.\n", buf);
422                 sleep(1);
423         } else
424                 perf_top__parse_source(top, found);
425 
426 out_free:
427         free(buf);
428 }
429 
430 static void perf_top__print_mapped_keys(struct perf_top *top)
431 {
432         char *name = NULL;
433 
434         if (top->sym_filter_entry) {
435                 struct symbol *sym = top->sym_filter_entry->ms.sym;
436                 name = sym->name;
437         }
438 
439         fprintf(stdout, "\nMapped keys:\n");
440         fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
441         fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
442 
443         if (top->evlist->core.nr_entries > 1)
444                 fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
445 
446         fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
447 
448         fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->annotation_opts.min_pcnt);
449         fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
450         fprintf(stdout, "\t[S]     stop annotation.\n");
451 
452         fprintf(stdout,
453                 "\t[K]     hide kernel symbols.             \t(%s)\n",
454                 top->hide_kernel_symbols ? "yes" : "no");
455         fprintf(stdout,
456                 "\t[U]     hide user symbols.               \t(%s)\n",
457                 top->hide_user_symbols ? "yes" : "no");
458         fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
459         fprintf(stdout, "\t[qQ]    quit.\n");
460 }
461 
462 static int perf_top__key_mapped(struct perf_top *top, int c)
463 {
464         switch (c) {
465                 case 'd':
466                 case 'e':
467                 case 'f':
468                 case 'z':
469                 case 'q':
470                 case 'Q':
471                 case 'K':
472                 case 'U':
473                 case 'F':
474                 case 's':
475                 case 'S':
476                         return 1;
477                 case 'E':
478                         return top->evlist->core.nr_entries > 1 ? 1 : 0;
479                 default:
480                         break;
481         }
482 
483         return 0;
484 }
485 
486 static bool perf_top__handle_keypress(struct perf_top *top, int c)
487 {
488         bool ret = true;
489 
490         if (!perf_top__key_mapped(top, c)) {
491                 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
492                 struct termios save;
493 
494                 perf_top__print_mapped_keys(top);
495                 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
496                 fflush(stdout);
497 
498                 set_term_quiet_input(&save);
499 
500                 poll(&stdin_poll, 1, -1);
501                 c = getc(stdin);
502 
503                 tcsetattr(0, TCSAFLUSH, &save);
504                 if (!perf_top__key_mapped(top, c))
505                         return ret;
506         }
507 
508         switch (c) {
509                 case 'd':
510                         prompt_integer(&top->delay_secs, "Enter display delay");
511                         if (top->delay_secs < 1)
512                                 top->delay_secs = 1;
513                         break;
514                 case 'e':
515                         prompt_integer(&top->print_entries, "Enter display entries (lines)");
516                         if (top->print_entries == 0) {
517                                 perf_top__resize(top);
518                                 signal(SIGWINCH, winch_sig);
519                         } else {
520                                 signal(SIGWINCH, SIG_DFL);
521                         }
522                         break;
523                 case 'E':
524                         if (top->evlist->core.nr_entries > 1) {
525                                 /* Select 0 as the default event: */
526                                 int counter = 0;
527 
528                                 fprintf(stderr, "\nAvailable events:");
529 
530                                 evlist__for_each_entry(top->evlist, top->sym_evsel)
531                                         fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
532 
533                                 prompt_integer(&counter, "Enter details event counter");
534 
535                                 if (counter >= top->evlist->core.nr_entries) {
536                                         top->sym_evsel = evlist__first(top->evlist);
537                                         fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
538                                         sleep(1);
539                                         break;
540                                 }
541                                 evlist__for_each_entry(top->evlist, top->sym_evsel)
542                                         if (top->sym_evsel->idx == counter)
543                                                 break;
544                         } else
545                                 top->sym_evsel = evlist__first(top->evlist);
546                         break;
547                 case 'f':
548                         prompt_integer(&top->count_filter, "Enter display event count filter");
549                         break;
550                 case 'F':
551                         prompt_percent(&top->annotation_opts.min_pcnt,
552                                        "Enter details display event filter (percent)");
553                         break;
554                 case 'K':
555                         top->hide_kernel_symbols = !top->hide_kernel_symbols;
556                         break;
557                 case 'q':
558                 case 'Q':
559                         printf("exiting.\n");
560                         if (top->dump_symtab)
561                                 perf_session__fprintf_dsos(top->session, stderr);
562                         ret = false;
563                         break;
564                 case 's':
565                         perf_top__prompt_symbol(top, "Enter details symbol");
566                         break;
567                 case 'S':
568                         if (!top->sym_filter_entry)
569                                 break;
570                         else {
571                                 struct hist_entry *syme = top->sym_filter_entry;
572 
573                                 top->sym_filter_entry = NULL;
574                                 __zero_source_counters(syme);
575                         }
576                         break;
577                 case 'U':
578                         top->hide_user_symbols = !top->hide_user_symbols;
579                         break;
580                 case 'z':
581                         top->zero = !top->zero;
582                         break;
583                 default:
584                         break;
585         }
586 
587         return ret;
588 }
589 
590 static void perf_top__sort_new_samples(void *arg)
591 {
592         struct perf_top *t = arg;
593 
594         if (t->evlist->selected != NULL)
595                 t->sym_evsel = t->evlist->selected;
596 
597         perf_top__resort_hists(t);
598 
599         if (t->lost || t->drop)
600                 pr_warning("Too slow to read ring buffer (change period (-c/-F) or limit CPUs (-C)\n");
601 }
602 
603 static void stop_top(void)
604 {
605         session_done = 1;
606         done = 1;
607 }
608 
609 static void *display_thread_tui(void *arg)
610 {
611         struct evsel *pos;
612         struct perf_top *top = arg;
613         const char *help = "For a higher level overview, try: perf top --sort comm,dso";
614         struct hist_browser_timer hbt = {
615                 .timer          = perf_top__sort_new_samples,
616                 .arg            = top,
617                 .refresh        = top->delay_secs,
618         };
619 
620         /* In order to read symbols from other namespaces perf to  needs to call
621          * setns(2).  This isn't permitted if the struct_fs has multiple users.
622          * unshare(2) the fs so that we may continue to setns into namespaces
623          * that we're observing.
624          */
625         unshare(CLONE_FS);
626 
627         prctl(PR_SET_NAME, "perf-top-UI", 0, 0, 0);
628 
629         perf_top__sort_new_samples(top);
630 
631         /*
632          * Initialize the uid_filter_str, in the future the TUI will allow
633          * Zooming in/out UIDs. For now just use whatever the user passed
634          * via --uid.
635          */
636         evlist__for_each_entry(top->evlist, pos) {
637                 struct hists *hists = evsel__hists(pos);
638                 hists->uid_filter_str = top->record_opts.target.uid_str;
639         }
640 
641         perf_evlist__tui_browse_hists(top->evlist, help, &hbt,
642                                       top->min_percent,
643                                       &top->session->header.env,
644                                       !top->record_opts.overwrite,
645                                       &top->annotation_opts);
646 
647         stop_top();
648         return NULL;
649 }
650 
651 static void display_sig(int sig __maybe_unused)
652 {
653         stop_top();
654 }
655 
656 static void display_setup_sig(void)
657 {
658         signal(SIGSEGV, sighandler_dump_stack);
659         signal(SIGFPE, sighandler_dump_stack);
660         signal(SIGINT,  display_sig);
661         signal(SIGQUIT, display_sig);
662         signal(SIGTERM, display_sig);
663 }
664 
665 static void *display_thread(void *arg)
666 {
667         struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
668         struct termios save;
669         struct perf_top *top = arg;
670         int delay_msecs, c;
671 
672         /* In order to read symbols from other namespaces perf to  needs to call
673          * setns(2).  This isn't permitted if the struct_fs has multiple users.
674          * unshare(2) the fs so that we may continue to setns into namespaces
675          * that we're observing.
676          */
677         unshare(CLONE_FS);
678 
679         prctl(PR_SET_NAME, "perf-top-UI", 0, 0, 0);
680 
681         display_setup_sig();
682         pthread__unblock_sigwinch();
683 repeat:
684         delay_msecs = top->delay_secs * MSEC_PER_SEC;
685         set_term_quiet_input(&save);
686         /* trash return*/
687         clearerr(stdin);
688         if (poll(&stdin_poll, 1, 0) > 0)
689                 getc(stdin);
690 
691         while (!done) {
692                 perf_top__print_sym_table(top);
693                 /*
694                  * Either timeout expired or we got an EINTR due to SIGWINCH,
695                  * refresh screen in both cases.
696                  */
697                 switch (poll(&stdin_poll, 1, delay_msecs)) {
698                 case 0:
699                         continue;
700                 case -1:
701                         if (errno == EINTR)
702                                 continue;
703                         __fallthrough;
704                 default:
705                         c = getc(stdin);
706                         tcsetattr(0, TCSAFLUSH, &save);
707 
708                         if (perf_top__handle_keypress(top, c))
709                                 goto repeat;
710                         stop_top();
711                 }
712         }
713 
714         tcsetattr(0, TCSAFLUSH, &save);
715         return NULL;
716 }
717 
718 static int hist_iter__top_callback(struct hist_entry_iter *iter,
719                                    struct addr_location *al, bool single,
720                                    void *arg)
721 {
722         struct perf_top *top = arg;
723         struct hist_entry *he = iter->he;
724         struct evsel *evsel = iter->evsel;
725 
726         if (perf_hpp_list.sym && single)
727                 perf_top__record_precise_ip(top, he, iter->sample, evsel, al->addr);
728 
729         hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
730                      !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY),
731                      NULL);
732         return 0;
733 }
734 
735 static void perf_event__process_sample(struct perf_tool *tool,
736                                        const union perf_event *event,
737                                        struct evsel *evsel,
738                                        struct perf_sample *sample,
739                                        struct machine *machine)
740 {
741         struct perf_top *top = container_of(tool, struct perf_top, tool);
742         struct addr_location al;
743         int err;
744 
745         if (!machine && perf_guest) {
746                 static struct intlist *seen;
747 
748                 if (!seen)
749                         seen = intlist__new(NULL);
750 
751                 if (!intlist__has_entry(seen, sample->pid)) {
752                         pr_err("Can't find guest [%d]'s kernel information\n",
753                                 sample->pid);
754                         intlist__add(seen, sample->pid);
755                 }
756                 return;
757         }
758 
759         if (!machine) {
760                 pr_err("%u unprocessable samples recorded.\r",
761                        top->session->evlist->stats.nr_unprocessable_samples++);
762                 return;
763         }
764 
765         if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
766                 top->exact_samples++;
767 
768         if (machine__resolve(machine, &al, sample) < 0)
769                 return;
770 
771         if (!machine->kptr_restrict_warned &&
772             symbol_conf.kptr_restrict &&
773             al.cpumode == PERF_RECORD_MISC_KERNEL) {
774                 if (!perf_evlist__exclude_kernel(top->session->evlist)) {
775                         ui__warning(
776 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
777 "Check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n"
778 "Kernel%s samples will not be resolved.\n",
779                           al.map && map__has_symbols(al.map) ?
780                           " modules" : "");
781                         if (use_browser <= 0)
782                                 sleep(5);
783                 }
784                 machine->kptr_restrict_warned = true;
785         }
786 
787         if (al.sym == NULL && al.map != NULL) {
788                 const char *msg = "Kernel samples will not be resolved.\n";
789                 /*
790                  * As we do lazy loading of symtabs we only will know if the
791                  * specified vmlinux file is invalid when we actually have a
792                  * hit in kernel space and then try to load it. So if we get
793                  * here and there are _no_ symbols in the DSO backing the
794                  * kernel map, bail out.
795                  *
796                  * We may never get here, for instance, if we use -K/
797                  * --hide-kernel-symbols, even if the user specifies an
798                  * invalid --vmlinux ;-)
799                  */
800                 if (!machine->kptr_restrict_warned && !top->vmlinux_warned &&
801                     __map__is_kernel(al.map) && map__has_symbols(al.map)) {
802                         if (symbol_conf.vmlinux_name) {
803                                 char serr[256];
804                                 dso__strerror_load(al.map->dso, serr, sizeof(serr));
805                                 ui__warning("The %s file can't be used: %s\n%s",
806                                             symbol_conf.vmlinux_name, serr, msg);
807                         } else {
808                                 ui__warning("A vmlinux file was not found.\n%s",
809                                             msg);
810                         }
811 
812                         if (use_browser <= 0)
813                                 sleep(5);
814                         top->vmlinux_warned = true;
815                 }
816         }
817 
818         if (al.sym == NULL || !al.sym->idle) {
819                 struct hists *hists = evsel__hists(evsel);
820                 struct hist_entry_iter iter = {
821                         .evsel          = evsel,
822                         .sample         = sample,
823                         .add_entry_cb   = hist_iter__top_callback,
824                 };
825 
826                 if (symbol_conf.cumulate_callchain)
827                         iter.ops = &hist_iter_cumulative;
828                 else
829                         iter.ops = &hist_iter_normal;
830 
831                 pthread_mutex_lock(&hists->lock);
832 
833                 err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
834                 if (err < 0)
835                         pr_err("Problem incrementing symbol period, skipping event\n");
836 
837                 pthread_mutex_unlock(&hists->lock);
838         }
839 
840         addr_location__put(&al);
841 }
842 
843 static void
844 perf_top__process_lost(struct perf_top *top, union perf_event *event,
845                        struct evsel *evsel)
846 {
847         struct hists *hists = evsel__hists(evsel);
848 
849         top->lost += event->lost.lost;
850         top->lost_total += event->lost.lost;
851         hists->stats.total_lost += event->lost.lost;
852 }
853 
854 static void
855 perf_top__process_lost_samples(struct perf_top *top,
856                                union perf_event *event,
857                                struct evsel *evsel)
858 {
859         struct hists *hists = evsel__hists(evsel);
860 
861         top->lost += event->lost_samples.lost;
862         top->lost_total += event->lost_samples.lost;
863         hists->stats.total_lost_samples += event->lost_samples.lost;
864 }
865 
866 static u64 last_timestamp;
867 
868 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
869 {
870         struct record_opts *opts = &top->record_opts;
871         struct evlist *evlist = top->evlist;
872         struct mmap *md;
873         union perf_event *event;
874 
875         md = opts->overwrite ? &evlist->overwrite_mmap[idx] : &evlist->mmap[idx];
876         if (perf_mmap__read_init(&md->core) < 0)
877                 return;
878 
879         while ((event = perf_mmap__read_event(&md->core)) != NULL) {
880                 int ret;
881 
882                 ret = perf_evlist__parse_sample_timestamp(evlist, event, &last_timestamp);
883                 if (ret && ret != -1)
884                         break;
885 
886                 ret = ordered_events__queue(top->qe.in, event, last_timestamp, 0);
887                 if (ret)
888                         break;
889 
890                 perf_mmap__consume(&md->core);
891 
892                 if (top->qe.rotate) {
893                         pthread_mutex_lock(&top->qe.mutex);
894                         top->qe.rotate = false;
895                         pthread_cond_signal(&top->qe.cond);
896                         pthread_mutex_unlock(&top->qe.mutex);
897                 }
898         }
899 
900         perf_mmap__read_done(&md->core);
901 }
902 
903 static void perf_top__mmap_read(struct perf_top *top)
904 {
905         bool overwrite = top->record_opts.overwrite;
906         struct evlist *evlist = top->evlist;
907         int i;
908 
909         if (overwrite)
910                 perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_DATA_PENDING);
911 
912         for (i = 0; i < top->evlist->core.nr_mmaps; i++)
913                 perf_top__mmap_read_idx(top, i);
914 
915         if (overwrite) {
916                 perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_EMPTY);
917                 perf_evlist__toggle_bkw_mmap(evlist, BKW_MMAP_RUNNING);
918         }
919 }
920 
921 /*
922  * Check per-event overwrite term.
923  * perf top should support consistent term for all events.
924  * - All events don't have per-event term
925  *   E.g. "cpu/cpu-cycles/,cpu/instructions/"
926  *   Nothing change, return 0.
927  * - All events have same per-event term
928  *   E.g. "cpu/cpu-cycles,no-overwrite/,cpu/instructions,no-overwrite/
929  *   Using the per-event setting to replace the opts->overwrite if
930  *   they are different, then return 0.
931  * - Events have different per-event term
932  *   E.g. "cpu/cpu-cycles,overwrite/,cpu/instructions,no-overwrite/"
933  *   Return -1
934  * - Some of the event set per-event term, but some not.
935  *   E.g. "cpu/cpu-cycles/,cpu/instructions,no-overwrite/"
936  *   Return -1
937  */
938 static int perf_top__overwrite_check(struct perf_top *top)
939 {
940         struct record_opts *opts = &top->record_opts;
941         struct evlist *evlist = top->evlist;
942         struct perf_evsel_config_term *term;
943         struct list_head *config_terms;
944         struct evsel *evsel;
945         int set, overwrite = -1;
946 
947         evlist__for_each_entry(evlist, evsel) {
948                 set = -1;
949                 config_terms = &evsel->config_terms;
950                 list_for_each_entry(term, config_terms, list) {
951                         if (term->type == PERF_EVSEL__CONFIG_TERM_OVERWRITE)
952                                 set = term->val.overwrite ? 1 : 0;
953                 }
954 
955                 /* no term for current and previous event (likely) */
956                 if ((overwrite < 0) && (set < 0))
957                         continue;
958 
959                 /* has term for both current and previous event, compare */
960                 if ((overwrite >= 0) && (set >= 0) && (overwrite != set))
961                         return -1;
962 
963                 /* no term for current event but has term for previous one */
964                 if ((overwrite >= 0) && (set < 0))
965                         return -1;
966 
967                 /* has term for current event */
968                 if ((overwrite < 0) && (set >= 0)) {
969                         /* if it's first event, set overwrite */
970                         if (evsel == evlist__first(evlist))
971                                 overwrite = set;
972                         else
973                                 return -1;
974                 }
975         }
976 
977         if ((overwrite >= 0) && (opts->overwrite != overwrite))
978                 opts->overwrite = overwrite;
979 
980         return 0;
981 }
982 
983 static int perf_top_overwrite_fallback(struct perf_top *top,
984                                        struct evsel *evsel)
985 {
986         struct record_opts *opts = &top->record_opts;
987         struct evlist *evlist = top->evlist;
988         struct evsel *counter;
989 
990         if (!opts->overwrite)
991                 return 0;
992 
993         /* only fall back when first event fails */
994         if (evsel != evlist__first(evlist))
995                 return 0;
996 
997         evlist__for_each_entry(evlist, counter)
998                 counter->core.attr.write_backward = false;
999         opts->overwrite = false;
1000         pr_debug2("fall back to non-overwrite mode\n");
1001         return 1;
1002 }
1003 
1004 static int perf_top__start_counters(struct perf_top *top)
1005 {
1006         char msg[BUFSIZ];
1007         struct evsel *counter;
1008         struct evlist *evlist = top->evlist;
1009         struct record_opts *opts = &top->record_opts;
1010 
1011         if (perf_top__overwrite_check(top)) {
1012                 ui__error("perf top only support consistent per-event "
1013                           "overwrite setting for all events\n");
1014                 goto out_err;
1015         }
1016 
1017         perf_evlist__config(evlist, opts, &callchain_param);
1018 
1019         evlist__for_each_entry(evlist, counter) {
1020 try_again:
1021                 if (evsel__open(counter, top->evlist->core.cpus,
1022                                      top->evlist->core.threads) < 0) {
1023 
1024                         /*
1025                          * Specially handle overwrite fall back.
1026                          * Because perf top is the only tool which has
1027                          * overwrite mode by default, support
1028                          * both overwrite and non-overwrite mode, and
1029                          * require consistent mode for all events.
1030                          *
1031                          * May move it to generic code with more tools
1032                          * have similar attribute.
1033                          */
1034                         if (perf_missing_features.write_backward &&
1035                             perf_top_overwrite_fallback(top, counter))
1036                                 goto try_again;
1037 
1038                         if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
1039                                 if (verbose > 0)
1040                                         ui__warning("%s\n", msg);
1041                                 goto try_again;
1042                         }
1043 
1044                         perf_evsel__open_strerror(counter, &opts->target,
1045                                                   errno, msg, sizeof(msg));
1046                         ui__error("%s\n", msg);
1047                         goto out_err;
1048                 }
1049         }
1050 
1051         if (evlist__mmap(evlist, opts->mmap_pages) < 0) {
1052                 ui__error("Failed to mmap with %d (%s)\n",
1053                             errno, str_error_r(errno, msg, sizeof(msg)));
1054                 goto out_err;
1055         }
1056 
1057         return 0;
1058 
1059 out_err:
1060         return -1;
1061 }
1062 
1063 static int callchain_param__setup_sample_type(struct callchain_param *callchain)
1064 {
1065         if (callchain->mode != CHAIN_NONE) {
1066                 if (callchain_register_param(callchain) < 0) {
1067                         ui__error("Can't register callchain params.\n");
1068                         return -EINVAL;
1069                 }
1070         }
1071 
1072         return 0;
1073 }
1074 
1075 static struct ordered_events *rotate_queues(struct perf_top *top)
1076 {
1077         struct ordered_events *in = top->qe.in;
1078 
1079         if (top->qe.in == &top->qe.data[1])
1080                 top->qe.in = &top->qe.data[0];
1081         else
1082                 top->qe.in = &top->qe.data[1];
1083 
1084         return in;
1085 }
1086 
1087 static void *process_thread(void *arg)
1088 {
1089         struct perf_top *top = arg;
1090 
1091         while (!done) {
1092                 struct ordered_events *out, *in = top->qe.in;
1093 
1094                 if (!in->nr_events) {
1095                         usleep(100);
1096                         continue;
1097                 }
1098 
1099                 out = rotate_queues(top);
1100 
1101                 pthread_mutex_lock(&top->qe.mutex);
1102                 top->qe.rotate = true;
1103                 pthread_cond_wait(&top->qe.cond, &top->qe.mutex);
1104                 pthread_mutex_unlock(&top->qe.mutex);
1105 
1106                 if (ordered_events__flush(out, OE_FLUSH__TOP))
1107                         pr_err("failed to process events\n");
1108         }
1109 
1110         return NULL;
1111 }
1112 
1113 /*
1114  * Allow only 'top->delay_secs' seconds behind samples.
1115  */
1116 static int should_drop(struct ordered_event *qevent, struct perf_top *top)
1117 {
1118         union perf_event *event = qevent->event;
1119         u64 delay_timestamp;
1120 
1121         if (event->header.type != PERF_RECORD_SAMPLE)
1122                 return false;
1123 
1124         delay_timestamp = qevent->timestamp + top->delay_secs * NSEC_PER_SEC;
1125         return delay_timestamp < last_timestamp;
1126 }
1127 
1128 static int deliver_event(struct ordered_events *qe,
1129                          struct ordered_event *qevent)
1130 {
1131         struct perf_top *top = qe->data;
1132         struct evlist *evlist = top->evlist;
1133         struct perf_session *session = top->session;
1134         union perf_event *event = qevent->event;
1135         struct perf_sample sample;
1136         struct evsel *evsel;
1137         struct machine *machine;
1138         int ret = -1;
1139 
1140         if (should_drop(qevent, top)) {
1141                 top->drop++;
1142                 top->drop_total++;
1143                 return 0;
1144         }
1145 
1146         ret = perf_evlist__parse_sample(evlist, event, &sample);
1147         if (ret) {
1148                 pr_err("Can't parse sample, err = %d\n", ret);
1149                 goto next_event;
1150         }
1151 
1152         evsel = perf_evlist__id2evsel(session->evlist, sample.id);
1153         assert(evsel != NULL);
1154 
1155         if (event->header.type == PERF_RECORD_SAMPLE) {
1156                 if (evswitch__discard(&top->evswitch, evsel))
1157                         return 0;
1158                 ++top->samples;
1159         }
1160 
1161         switch (sample.cpumode) {
1162         case PERF_RECORD_MISC_USER:
1163                 ++top->us_samples;
1164                 if (top->hide_user_symbols)
1165                         goto next_event;
1166                 machine = &session->machines.host;
1167                 break;
1168         case PERF_RECORD_MISC_KERNEL:
1169                 ++top->kernel_samples;
1170                 if (top->hide_kernel_symbols)
1171                         goto next_event;
1172                 machine = &session->machines.host;
1173                 break;
1174         case PERF_RECORD_MISC_GUEST_KERNEL:
1175                 ++top->guest_kernel_samples;
1176                 machine = perf_session__find_machine(session,
1177                                                      sample.pid);
1178                 break;
1179         case PERF_RECORD_MISC_GUEST_USER:
1180                 ++top->guest_us_samples;
1181                 /*
1182                  * TODO: we don't process guest user from host side
1183                  * except simple counting.
1184                  */
1185                 goto next_event;
1186         default:
1187                 if (event->header.type == PERF_RECORD_SAMPLE)
1188                         goto next_event;
1189                 machine = &session->machines.host;
1190                 break;
1191         }
1192 
1193         if (event->header.type == PERF_RECORD_SAMPLE) {
1194                 perf_event__process_sample(&top->tool, event, evsel,
1195                                            &sample, machine);
1196         } else if (event->header.type == PERF_RECORD_LOST) {
1197                 perf_top__process_lost(top, event, evsel);
1198         } else if (event->header.type == PERF_RECORD_LOST_SAMPLES) {
1199                 perf_top__process_lost_samples(top, event, evsel);
1200         } else if (event->header.type < PERF_RECORD_MAX) {
1201                 hists__inc_nr_events(evsel__hists(evsel), event->header.type);
1202                 machine__process_event(machine, event, &sample);
1203         } else
1204                 ++session->evlist->stats.nr_unknown_events;
1205 
1206         ret = 0;
1207 next_event:
1208         return ret;
1209 }
1210 
1211 static void init_process_thread(struct perf_top *top)
1212 {
1213         ordered_events__init(&top->qe.data[0], deliver_event, top);
1214         ordered_events__init(&top->qe.data[1], deliver_event, top);
1215         ordered_events__set_copy_on_queue(&top->qe.data[0], true);
1216         ordered_events__set_copy_on_queue(&top->qe.data[1], true);
1217         top->qe.in = &top->qe.data[0];
1218         pthread_mutex_init(&top->qe.mutex, NULL);
1219         pthread_cond_init(&top->qe.cond, NULL);
1220 }
1221 
1222 static int __cmd_top(struct perf_top *top)
1223 {
1224         struct record_opts *opts = &top->record_opts;
1225         pthread_t thread, thread_process;
1226         int ret;
1227 
1228         if (!top->annotation_opts.objdump_path) {
1229                 ret = perf_env__lookup_objdump(&top->session->header.env,
1230                                                &top->annotation_opts.objdump_path);
1231                 if (ret)
1232                         return ret;
1233         }
1234 
1235         ret = callchain_param__setup_sample_type(&callchain_param);
1236         if (ret)
1237                 return ret;
1238 
1239         if (perf_session__register_idle_thread(top->session) < 0)
1240                 return ret;
1241 
1242         if (top->nr_threads_synthesize > 1)
1243                 perf_set_multithreaded();
1244 
1245         init_process_thread(top);
1246 
1247         if (opts->record_namespaces)
1248                 top->tool.namespace_events = true;
1249 
1250         ret = perf_event__synthesize_bpf_events(top->session, perf_event__process,
1251                                                 &top->session->machines.host,
1252                                                 &top->record_opts);
1253         if (ret < 0)
1254                 pr_debug("Couldn't synthesize BPF events: Pre-existing BPF programs won't have symbols resolved.\n");
1255 
1256         machine__synthesize_threads(&top->session->machines.host, &opts->target,
1257                                     top->evlist->core.threads, false,
1258                                     top->nr_threads_synthesize);
1259 
1260         if (top->nr_threads_synthesize > 1)
1261                 perf_set_singlethreaded();
1262 
1263         if (perf_hpp_list.socket) {
1264                 ret = perf_env__read_cpu_topology_map(&perf_env);
1265                 if (ret < 0) {
1266                         char errbuf[BUFSIZ];
1267                         const char *err = str_error_r(-ret, errbuf, sizeof(errbuf));
1268 
1269                         ui__error("Could not read the CPU topology map: %s\n", err);
1270                         return ret;
1271                 }
1272         }
1273 
1274         ret = perf_top__start_counters(top);
1275         if (ret)
1276                 return ret;
1277 
1278         top->session->evlist = top->evlist;
1279         perf_session__set_id_hdr_size(top->session);
1280 
1281         /*
1282          * When perf is starting the traced process, all the events (apart from
1283          * group members) have enable_on_exec=1 set, so don't spoil it by
1284          * prematurely enabling them.
1285          *
1286          * XXX 'top' still doesn't start workloads like record, trace, but should,
1287          * so leave the check here.
1288          */
1289         if (!target__none(&opts->target))
1290                 evlist__enable(top->evlist);
1291 
1292         ret = -1;
1293         if (pthread_create(&thread_process, NULL, process_thread, top)) {
1294                 ui__error("Could not create process thread.\n");
1295                 return ret;
1296         }
1297 
1298         if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
1299                                                             display_thread), top)) {
1300                 ui__error("Could not create display thread.\n");
1301                 goto out_join_thread;
1302         }
1303 
1304         if (top->realtime_prio) {
1305                 struct sched_param param;
1306 
1307                 param.sched_priority = top->realtime_prio;
1308                 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
1309                         ui__error("Could not set realtime priority.\n");
1310                         goto out_join;
1311                 }
1312         }
1313 
1314         /* Wait for a minimal set of events before starting the snapshot */
1315         evlist__poll(top->evlist, 100);
1316 
1317         perf_top__mmap_read(top);
1318 
1319         while (!done) {
1320                 u64 hits = top->samples;
1321 
1322                 perf_top__mmap_read(top);
1323 
1324                 if (opts->overwrite || (hits == top->samples))
1325                         ret = evlist__poll(top->evlist, 100);
1326 
1327                 if (resize) {
1328                         perf_top__resize(top);
1329                         resize = 0;
1330                 }
1331         }
1332 
1333         ret = 0;
1334 out_join:
1335         pthread_join(thread, NULL);
1336 out_join_thread:
1337         pthread_cond_signal(&top->qe.cond);
1338         pthread_join(thread_process, NULL);
1339         return ret;
1340 }
1341 
1342 static int
1343 callchain_opt(const struct option *opt, const char *arg, int unset)
1344 {
1345         symbol_conf.use_callchain = true;
1346         return record_callchain_opt(opt, arg, unset);
1347 }
1348 
1349 static int
1350 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1351 {
1352         struct callchain_param *callchain = opt->value;
1353 
1354         callchain->enabled = !unset;
1355         callchain->record_mode = CALLCHAIN_FP;
1356 
1357         /*
1358          * --no-call-graph
1359          */
1360         if (unset) {
1361                 symbol_conf.use_callchain = false;
1362                 callchain->record_mode = CALLCHAIN_NONE;
1363                 return 0;
1364         }
1365 
1366         return parse_callchain_top_opt(arg);
1367 }
1368 
1369 static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1370 {
1371         if (!strcmp(var, "top.call-graph")) {
1372                 var = "call-graph.record-mode";
1373                 return perf_default_config(var, value, cb);
1374         }
1375         if (!strcmp(var, "top.children")) {
1376                 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
1377                 return 0;
1378         }
1379 
1380         return 0;
1381 }
1382 
1383 static int
1384 parse_percent_limit(const struct option *opt, const char *arg,
1385                     int unset __maybe_unused)
1386 {
1387         struct perf_top *top = opt->value;
1388 
1389         top->min_percent = strtof(arg, NULL);
1390         return 0;
1391 }
1392 
1393 const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1394         "\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1395 
1396 int cmd_top(int argc, const char **argv)
1397 {
1398         char errbuf[BUFSIZ];
1399         struct perf_top top = {
1400                 .count_filter        = 5,
1401                 .delay_secs          = 2,
1402                 .record_opts = {
1403                         .mmap_pages     = UINT_MAX,
1404                         .user_freq      = UINT_MAX,
1405                         .user_interval  = ULLONG_MAX,
1406                         .freq           = 4000, /* 4 KHz */
1407                         .target         = {
1408                                 .uses_mmap   = true,
1409                         },
1410                         /*
1411                          * FIXME: This will lose PERF_RECORD_MMAP and other metadata
1412                          * when we pause, fix that and reenable. Probably using a
1413                          * separate evlist with a dummy event, i.e. a non-overwrite
1414                          * ring buffer just for metadata events, while PERF_RECORD_SAMPLE
1415                          * stays in overwrite mode. -acme
1416                          * */
1417                         .overwrite      = 0,
1418                         .sample_time    = true,
1419                         .sample_time_set = true,
1420                 },
1421                 .max_stack           = sysctl__max_stack(),
1422                 .annotation_opts     = annotation__default_options,
1423                 .nr_threads_synthesize = UINT_MAX,
1424         };
1425         struct record_opts *opts = &top.record_opts;
1426         struct target *target = &opts->target;
1427         const struct option options[] = {
1428         OPT_CALLBACK('e', "event", &top.evlist, "event",
1429                      "event selector. use 'perf list' to list available events",
1430                      parse_events_option),
1431         OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1432         OPT_STRING('p', "pid", &target->pid, "pid",
1433                     "profile events on existing process id"),
1434         OPT_STRING('t', "tid", &target->tid, "tid",
1435                     "profile events on existing thread id"),
1436         OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1437                             "system-wide collection from all CPUs"),
1438         OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1439                     "list of cpus to monitor"),
1440         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1441                    "file", "vmlinux pathname"),
1442         OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1443                     "don't load vmlinux even if found"),
1444         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1445                    "file", "kallsyms pathname"),
1446         OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1447                     "hide kernel symbols"),
1448         OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1449                      "number of mmap data pages",
1450                      perf_evlist__parse_mmap_pages),
1451         OPT_INTEGER('r', "realtime", &top.realtime_prio,
1452                     "collect data with this RT SCHED_FIFO priority"),
1453         OPT_INTEGER('d', "delay", &top.delay_secs,
1454                     "number of seconds to delay between refreshes"),
1455         OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1456                             "dump the symbol table used for profiling"),
1457         OPT_INTEGER('f', "count-filter", &top.count_filter,
1458                     "only display functions with more events than this"),
1459         OPT_BOOLEAN(0, "group", &opts->group,
1460                             "put the counters into a counter group"),
1461         OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1462                     "child tasks do not inherit counters"),
1463         OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1464                     "symbol to annotate"),
1465         OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1466         OPT_CALLBACK('F', "freq", &top.record_opts, "freq or 'max'",
1467                      "profile at this frequency",
1468                       record__parse_freq),
1469         OPT_INTEGER('E', "entries", &top.print_entries,
1470                     "display this many functions"),
1471         OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1472                     "hide user symbols"),
1473         OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1474         OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1475         OPT_INCR('v', "verbose", &verbose,
1476                     "be more verbose (show counter open errors, etc)"),
1477         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1478                    "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1479                    " Please refer the man page for the complete list."),
1480         OPT_STRING(0, "fields", &field_order, "key[,keys...]",
1481                    "output field(s): overhead, period, sample plus all of sort keys"),
1482         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1483                     "Show a column with the number of samples"),
1484         OPT_CALLBACK_NOOPT('g', NULL, &callchain_param,
1485                            NULL, "enables call-graph recording and display",
1486                            &callchain_opt),
1487         OPT_CALLBACK(0, "call-graph", &callchain_param,
1488                      "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1489                      top_callchain_help, &parse_callchain_opt),
1490         OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1491                     "Accumulate callchains of children and show total overhead as well"),
1492         OPT_INTEGER(0, "max-stack", &top.max_stack,
1493                     "Set the maximum stack depth when parsing the callchain. "
1494                     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1495         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1496                    "ignore callees of these functions in call graphs",
1497                    report_parse_ignore_callees_opt),
1498         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1499                     "Show a column with the sum of periods"),
1500         OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1501                    "only consider symbols in these dsos"),
1502         OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1503                    "only consider symbols in these comms"),
1504         OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1505                    "only consider these symbols"),
1506         OPT_BOOLEAN(0, "source", &top.annotation_opts.annotate_src,
1507                     "Interleave source code with assembly code (default)"),
1508         OPT_BOOLEAN(0, "asm-raw", &top.annotation_opts.show_asm_raw,
1509                     "Display raw encoding of assembly instructions (default)"),
1510         OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1511                     "Enable kernel symbol demangling"),
1512         OPT_BOOLEAN(0, "no-bpf-event", &top.record_opts.no_bpf_event, "do not record bpf events"),
1513         OPT_STRING(0, "objdump", &top.annotation_opts.objdump_path, "path",
1514                     "objdump binary to use for disassembly and annotations"),
1515         OPT_STRING('M', "disassembler-style", &top.annotation_opts.disassembler_style, "disassembler style",
1516                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
1517         OPT_STRING(0, "prefix", &top.annotation_opts.prefix, "prefix",
1518                     "Add prefix to source file path names in programs (with --prefix-strip)"),
1519         OPT_STRING(0, "prefix-strip", &top.annotation_opts.prefix_strip, "N",
1520                     "Strip first N entries of source file path name in programs (with --prefix)"),
1521         OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1522         OPT_CALLBACK(0, "percent-limit", &top, "percent",
1523                      "Don't show entries under that percent", parse_percent_limit),
1524         OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1525                      "How to display percentage of filtered entries", parse_filter_percentage),
1526         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1527                    "width[,width...]",
1528                    "don't try to adjust column width, use these fixed values"),
1529         OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout,
1530                         "per thread proc mmap processing timeout in ms"),
1531         OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
1532                      "branch any", "sample any taken branches",
1533                      parse_branch_stack),
1534         OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
1535                      "branch filter mask", "branch stack filter modes",
1536                      parse_branch_stack),
1537         OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1538                     "Show raw trace event output (do not use print fmt or plugins)"),
1539         OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1540                     "Show entries in a hierarchy"),
1541         OPT_BOOLEAN(0, "overwrite", &top.record_opts.overwrite,
1542                     "Use a backward ring buffer, default: no"),
1543         OPT_BOOLEAN(0, "force", &symbol_conf.force, "don't complain, do it"),
1544         OPT_UINTEGER(0, "num-thread-synthesize", &top.nr_threads_synthesize,
1545                         "number of thread to run event synthesize"),
1546         OPT_BOOLEAN(0, "namespaces", &opts->record_namespaces,
1547                     "Record namespaces events"),
1548         OPTS_EVSWITCH(&top.evswitch),
1549         OPT_END()
1550         };
1551         struct evlist *sb_evlist = NULL;
1552         const char * const top_usage[] = {
1553                 "perf top [<options>]",
1554                 NULL
1555         };
1556         int status = hists__init();
1557 
1558         if (status < 0)
1559                 return status;
1560 
1561         top.annotation_opts.min_pcnt = 5;
1562         top.annotation_opts.context  = 4;
1563 
1564         top.evlist = evlist__new();
1565         if (top.evlist == NULL)
1566                 return -ENOMEM;
1567 
1568         status = perf_config(perf_top_config, &top);
1569         if (status)
1570                 return status;
1571         /*
1572          * Since the per arch annotation init routine may need the cpuid, read
1573          * it here, since we are not getting this from the perf.data header.
1574          */
1575         status = perf_env__read_cpuid(&perf_env);
1576         if (status) {
1577                 /*
1578                  * Some arches do not provide a get_cpuid(), so just use pr_debug, otherwise
1579                  * warn the user explicitely.
1580                  */
1581                 eprintf(status == ENOSYS ? 1 : 0, verbose,
1582                         "Couldn't read the cpuid for this machine: %s\n",
1583                         str_error_r(errno, errbuf, sizeof(errbuf)));
1584         }
1585         top.evlist->env = &perf_env;
1586 
1587         argc = parse_options(argc, argv, options, top_usage, 0);
1588         if (argc)
1589                 usage_with_options(top_usage, options);
1590 
1591         if (annotate_check_args(&top.annotation_opts) < 0)
1592                 goto out_delete_evlist;
1593 
1594         if (!top.evlist->core.nr_entries &&
1595             perf_evlist__add_default(top.evlist) < 0) {
1596                 pr_err("Not enough memory for event selector list\n");
1597                 goto out_delete_evlist;
1598         }
1599 
1600         status = evswitch__init(&top.evswitch, top.evlist, stderr);
1601         if (status)
1602                 goto out_delete_evlist;
1603 
1604         if (symbol_conf.report_hierarchy) {
1605                 /* disable incompatible options */
1606                 symbol_conf.event_group = false;
1607                 symbol_conf.cumulate_callchain = false;
1608 
1609                 if (field_order) {
1610                         pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1611                         parse_options_usage(top_usage, options, "fields", 0);
1612                         parse_options_usage(NULL, options, "hierarchy", 0);
1613                         goto out_delete_evlist;
1614                 }
1615         }
1616 
1617         if (opts->branch_stack && callchain_param.enabled)
1618                 symbol_conf.show_branchflag_count = true;
1619 
1620         sort__mode = SORT_MODE__TOP;
1621         /* display thread wants entries to be collapsed in a different tree */
1622         perf_hpp_list.need_collapse = 1;
1623 
1624         if (top.use_stdio)
1625                 use_browser = 0;
1626         else if (top.use_tui)
1627                 use_browser = 1;
1628 
1629         setup_browser(false);
1630 
1631         if (setup_sorting(top.evlist) < 0) {
1632                 if (sort_order)
1633                         parse_options_usage(top_usage, options, "s", 1);
1634                 if (field_order)
1635                         parse_options_usage(sort_order ? NULL : top_usage,
1636                                             options, "fields", 0);
1637                 goto out_delete_evlist;
1638         }
1639 
1640         status = target__validate(target);
1641         if (status) {
1642                 target__strerror(target, status, errbuf, BUFSIZ);
1643                 ui__warning("%s\n", errbuf);
1644         }
1645 
1646         status = target__parse_uid(target);
1647         if (status) {
1648                 int saved_errno = errno;
1649 
1650                 target__strerror(target, status, errbuf, BUFSIZ);
1651                 ui__error("%s\n", errbuf);
1652 
1653                 status = -saved_errno;
1654                 goto out_delete_evlist;
1655         }
1656 
1657         if (target__none(target))
1658                 target->system_wide = true;
1659 
1660         if (perf_evlist__create_maps(top.evlist, target) < 0) {
1661                 ui__error("Couldn't create thread/CPU maps: %s\n",
1662                           errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf)));
1663                 goto out_delete_evlist;
1664         }
1665 
1666         if (top.delay_secs < 1)
1667                 top.delay_secs = 1;
1668 
1669         if (record_opts__config(opts)) {
1670                 status = -EINVAL;
1671                 goto out_delete_evlist;
1672         }
1673 
1674         top.sym_evsel = evlist__first(top.evlist);
1675 
1676         if (!callchain_param.enabled) {
1677                 symbol_conf.cumulate_callchain = false;
1678                 perf_hpp__cancel_cumulate();
1679         }
1680 
1681         if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1682                 callchain_param.order = ORDER_CALLER;
1683 
1684         status = symbol__annotation_init();
1685         if (status < 0)
1686                 goto out_delete_evlist;
1687 
1688         annotation_config__init(&top.annotation_opts);
1689 
1690         symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1691         status = symbol__init(NULL);
1692         if (status < 0)
1693                 goto out_delete_evlist;
1694 
1695         sort__setup_elide(stdout);
1696 
1697         get_term_dimensions(&top.winsize);
1698         if (top.print_entries == 0) {
1699                 perf_top__update_print_entries(&top);
1700                 signal(SIGWINCH, winch_sig);
1701         }
1702 
1703         top.session = perf_session__new(NULL, false, NULL);
1704         if (IS_ERR(top.session)) {
1705                 status = PTR_ERR(top.session);
1706                 goto out_delete_evlist;
1707         }
1708 
1709         if (!top.record_opts.no_bpf_event)
1710                 bpf_event__add_sb_event(&sb_evlist, &perf_env);
1711 
1712         if (perf_evlist__start_sb_thread(sb_evlist, target)) {
1713                 pr_debug("Couldn't start the BPF side band thread:\nBPF programs starting from now on won't be annotatable\n");
1714                 opts->no_bpf_event = true;
1715         }
1716 
1717         status = __cmd_top(&top);
1718 
1719         if (!opts->no_bpf_event)
1720                 perf_evlist__stop_sb_thread(sb_evlist);
1721 
1722 out_delete_evlist:
1723         evlist__delete(top.evlist);
1724         perf_session__delete(top.session);
1725 
1726         return status;
1727 }
1728 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp