Merge remote-tracking branches 'spi/topic/rspi', 'spi/topic/sc18is602', 'spi/topic...
[cascardo/linux.git] / 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/config.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 <subcmd/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 "util/parse-branch-options.h"
44 #include "arch/common.h"
45
46 #include "util/debug.h"
47
48 #include <assert.h>
49 #include <elf.h>
50 #include <fcntl.h>
51
52 #include <stdio.h>
53 #include <termios.h>
54 #include <unistd.h>
55 #include <inttypes.h>
56
57 #include <errno.h>
58 #include <time.h>
59 #include <sched.h>
60
61 #include <sys/syscall.h>
62 #include <sys/ioctl.h>
63 #include <poll.h>
64 #include <sys/prctl.h>
65 #include <sys/wait.h>
66 #include <sys/uio.h>
67 #include <sys/utsname.h>
68 #include <sys/mman.h>
69
70 #include <linux/stringify.h>
71 #include <linux/types.h>
72
73 static volatile int done;
74
75 #define HEADER_LINE_NR  5
76
77 static void perf_top__update_print_entries(struct perf_top *top)
78 {
79         top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
80 }
81
82 static void perf_top__sig_winch(int sig __maybe_unused,
83                                 siginfo_t *info __maybe_unused, void *arg)
84 {
85         struct perf_top *top = arg;
86
87         get_term_dimensions(&top->winsize);
88         perf_top__update_print_entries(top);
89 }
90
91 static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
92 {
93         struct symbol *sym;
94         struct annotation *notes;
95         struct map *map;
96         int err = -1;
97
98         if (!he || !he->ms.sym)
99                 return -1;
100
101         sym = he->ms.sym;
102         map = he->ms.map;
103
104         /*
105          * We can't annotate with just /proc/kallsyms
106          */
107         if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
108             !dso__is_kcore(map->dso)) {
109                 pr_err("Can't annotate %s: No vmlinux file was found in the "
110                        "path\n", sym->name);
111                 sleep(1);
112                 return -1;
113         }
114
115         notes = symbol__annotation(sym);
116         if (notes->src != NULL) {
117                 pthread_mutex_lock(&notes->lock);
118                 goto out_assign;
119         }
120
121         pthread_mutex_lock(&notes->lock);
122
123         if (symbol__alloc_hist(sym) < 0) {
124                 pthread_mutex_unlock(&notes->lock);
125                 pr_err("Not enough memory for annotating '%s' symbol!\n",
126                        sym->name);
127                 sleep(1);
128                 return err;
129         }
130
131         err = symbol__disassemble(sym, map, 0);
132         if (err == 0) {
133 out_assign:
134                 top->sym_filter_entry = he;
135         } else {
136                 char msg[BUFSIZ];
137                 symbol__strerror_disassemble(sym, map, err, msg, sizeof(msg));
138                 pr_err("Couldn't annotate %s: %s\n", sym->name, msg);
139         }
140
141         pthread_mutex_unlock(&notes->lock);
142         return err;
143 }
144
145 static void __zero_source_counters(struct hist_entry *he)
146 {
147         struct symbol *sym = he->ms.sym;
148         symbol__annotate_zero_histograms(sym);
149 }
150
151 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
152 {
153         struct utsname uts;
154         int err = uname(&uts);
155
156         ui__warning("Out of bounds address found:\n\n"
157                     "Addr:   %" PRIx64 "\n"
158                     "DSO:    %s %c\n"
159                     "Map:    %" PRIx64 "-%" PRIx64 "\n"
160                     "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
161                     "Arch:   %s\n"
162                     "Kernel: %s\n"
163                     "Tools:  %s\n\n"
164                     "Not all samples will be on the annotation output.\n\n"
165                     "Please report to linux-kernel@vger.kernel.org\n",
166                     ip, map->dso->long_name, dso__symtab_origin(map->dso),
167                     map->start, map->end, sym->start, sym->end,
168                     sym->binding == STB_GLOBAL ? 'g' :
169                     sym->binding == STB_LOCAL  ? 'l' : 'w', sym->name,
170                     err ? "[unknown]" : uts.machine,
171                     err ? "[unknown]" : uts.release, perf_version_string);
172         if (use_browser <= 0)
173                 sleep(5);
174
175         map->erange_warned = true;
176 }
177
178 static void perf_top__record_precise_ip(struct perf_top *top,
179                                         struct hist_entry *he,
180                                         int counter, u64 ip)
181 {
182         struct annotation *notes;
183         struct symbol *sym = he->ms.sym;
184         int err = 0;
185
186         if (sym == NULL || (use_browser == 0 &&
187                             (top->sym_filter_entry == NULL ||
188                              top->sym_filter_entry->ms.sym != sym)))
189                 return;
190
191         notes = symbol__annotation(sym);
192
193         if (pthread_mutex_trylock(&notes->lock))
194                 return;
195
196         err = hist_entry__inc_addr_samples(he, counter, ip);
197
198         pthread_mutex_unlock(&notes->lock);
199
200         if (unlikely(err)) {
201                 /*
202                  * This function is now called with he->hists->lock held.
203                  * Release it before going to sleep.
204                  */
205                 pthread_mutex_unlock(&he->hists->lock);
206
207                 if (err == -ERANGE && !he->ms.map->erange_warned)
208                         ui__warn_map_erange(he->ms.map, sym, ip);
209                 else if (err == -ENOMEM) {
210                         pr_err("Not enough memory for annotating '%s' symbol!\n",
211                                sym->name);
212                         sleep(1);
213                 }
214
215                 pthread_mutex_lock(&he->hists->lock);
216         }
217 }
218
219 static void perf_top__show_details(struct perf_top *top)
220 {
221         struct hist_entry *he = top->sym_filter_entry;
222         struct annotation *notes;
223         struct symbol *symbol;
224         int more;
225
226         if (!he)
227                 return;
228
229         symbol = he->ms.sym;
230         notes = symbol__annotation(symbol);
231
232         pthread_mutex_lock(&notes->lock);
233
234         if (notes->src == NULL)
235                 goto out_unlock;
236
237         printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
238         printf("  Events  Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
239
240         more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
241                                        0, top->sym_pcnt_filter, top->print_entries, 4);
242
243         if (top->evlist->enabled) {
244                 if (top->zero)
245                         symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
246                 else
247                         symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
248         }
249         if (more != 0)
250                 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
251 out_unlock:
252         pthread_mutex_unlock(&notes->lock);
253 }
254
255 static void perf_top__print_sym_table(struct perf_top *top)
256 {
257         char bf[160];
258         int printed = 0;
259         const int win_width = top->winsize.ws_col - 1;
260         struct perf_evsel *evsel = top->sym_evsel;
261         struct hists *hists = evsel__hists(evsel);
262
263         puts(CONSOLE_CLEAR);
264
265         perf_top__header_snprintf(top, bf, sizeof(bf));
266         printf("%s\n", bf);
267
268         perf_top__reset_sample_counters(top);
269
270         printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
271
272         if (hists->stats.nr_lost_warned !=
273             hists->stats.nr_events[PERF_RECORD_LOST]) {
274                 hists->stats.nr_lost_warned =
275                               hists->stats.nr_events[PERF_RECORD_LOST];
276                 color_fprintf(stdout, PERF_COLOR_RED,
277                               "WARNING: LOST %d chunks, Check IO/CPU overload",
278                               hists->stats.nr_lost_warned);
279                 ++printed;
280         }
281
282         if (top->sym_filter_entry) {
283                 perf_top__show_details(top);
284                 return;
285         }
286
287         if (top->evlist->enabled) {
288                 if (top->zero) {
289                         hists__delete_entries(hists);
290                 } else {
291                         hists__decay_entries(hists, top->hide_user_symbols,
292                                              top->hide_kernel_symbols);
293                 }
294         }
295
296         hists__collapse_resort(hists, NULL);
297         perf_evsel__output_resort(evsel, NULL);
298
299         hists__output_recalc_col_len(hists, top->print_entries - printed);
300         putchar('\n');
301         hists__fprintf(hists, false, top->print_entries - printed, win_width,
302                        top->min_percent, stdout, symbol_conf.use_callchain);
303 }
304
305 static void prompt_integer(int *target, const char *msg)
306 {
307         char *buf = malloc(0), *p;
308         size_t dummy = 0;
309         int tmp;
310
311         fprintf(stdout, "\n%s: ", msg);
312         if (getline(&buf, &dummy, stdin) < 0)
313                 return;
314
315         p = strchr(buf, '\n');
316         if (p)
317                 *p = 0;
318
319         p = buf;
320         while(*p) {
321                 if (!isdigit(*p))
322                         goto out_free;
323                 p++;
324         }
325         tmp = strtoul(buf, NULL, 10);
326         *target = tmp;
327 out_free:
328         free(buf);
329 }
330
331 static void prompt_percent(int *target, const char *msg)
332 {
333         int tmp = 0;
334
335         prompt_integer(&tmp, msg);
336         if (tmp >= 0 && tmp <= 100)
337                 *target = tmp;
338 }
339
340 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
341 {
342         char *buf = malloc(0), *p;
343         struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
344         struct hists *hists = evsel__hists(top->sym_evsel);
345         struct rb_node *next;
346         size_t dummy = 0;
347
348         /* zero counters of active symbol */
349         if (syme) {
350                 __zero_source_counters(syme);
351                 top->sym_filter_entry = NULL;
352         }
353
354         fprintf(stdout, "\n%s: ", msg);
355         if (getline(&buf, &dummy, stdin) < 0)
356                 goto out_free;
357
358         p = strchr(buf, '\n');
359         if (p)
360                 *p = 0;
361
362         next = rb_first(&hists->entries);
363         while (next) {
364                 n = rb_entry(next, struct hist_entry, rb_node);
365                 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
366                         found = n;
367                         break;
368                 }
369                 next = rb_next(&n->rb_node);
370         }
371
372         if (!found) {
373                 fprintf(stderr, "Sorry, %s is not active.\n", buf);
374                 sleep(1);
375         } else
376                 perf_top__parse_source(top, found);
377
378 out_free:
379         free(buf);
380 }
381
382 static void perf_top__print_mapped_keys(struct perf_top *top)
383 {
384         char *name = NULL;
385
386         if (top->sym_filter_entry) {
387                 struct symbol *sym = top->sym_filter_entry->ms.sym;
388                 name = sym->name;
389         }
390
391         fprintf(stdout, "\nMapped keys:\n");
392         fprintf(stdout, "\t[d]     display refresh delay.             \t(%d)\n", top->delay_secs);
393         fprintf(stdout, "\t[e]     display entries (lines).           \t(%d)\n", top->print_entries);
394
395         if (top->evlist->nr_entries > 1)
396                 fprintf(stdout, "\t[E]     active event counter.              \t(%s)\n", perf_evsel__name(top->sym_evsel));
397
398         fprintf(stdout, "\t[f]     profile display filter (count).    \t(%d)\n", top->count_filter);
399
400         fprintf(stdout, "\t[F]     annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
401         fprintf(stdout, "\t[s]     annotate symbol.                   \t(%s)\n", name?: "NULL");
402         fprintf(stdout, "\t[S]     stop annotation.\n");
403
404         fprintf(stdout,
405                 "\t[K]     hide kernel_symbols symbols.     \t(%s)\n",
406                 top->hide_kernel_symbols ? "yes" : "no");
407         fprintf(stdout,
408                 "\t[U]     hide user symbols.               \t(%s)\n",
409                 top->hide_user_symbols ? "yes" : "no");
410         fprintf(stdout, "\t[z]     toggle sample zeroing.             \t(%d)\n", top->zero ? 1 : 0);
411         fprintf(stdout, "\t[qQ]    quit.\n");
412 }
413
414 static int perf_top__key_mapped(struct perf_top *top, int c)
415 {
416         switch (c) {
417                 case 'd':
418                 case 'e':
419                 case 'f':
420                 case 'z':
421                 case 'q':
422                 case 'Q':
423                 case 'K':
424                 case 'U':
425                 case 'F':
426                 case 's':
427                 case 'S':
428                         return 1;
429                 case 'E':
430                         return top->evlist->nr_entries > 1 ? 1 : 0;
431                 default:
432                         break;
433         }
434
435         return 0;
436 }
437
438 static bool perf_top__handle_keypress(struct perf_top *top, int c)
439 {
440         bool ret = true;
441
442         if (!perf_top__key_mapped(top, c)) {
443                 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
444                 struct termios save;
445
446                 perf_top__print_mapped_keys(top);
447                 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
448                 fflush(stdout);
449
450                 set_term_quiet_input(&save);
451
452                 poll(&stdin_poll, 1, -1);
453                 c = getc(stdin);
454
455                 tcsetattr(0, TCSAFLUSH, &save);
456                 if (!perf_top__key_mapped(top, c))
457                         return ret;
458         }
459
460         switch (c) {
461                 case 'd':
462                         prompt_integer(&top->delay_secs, "Enter display delay");
463                         if (top->delay_secs < 1)
464                                 top->delay_secs = 1;
465                         break;
466                 case 'e':
467                         prompt_integer(&top->print_entries, "Enter display entries (lines)");
468                         if (top->print_entries == 0) {
469                                 struct sigaction act = {
470                                         .sa_sigaction = perf_top__sig_winch,
471                                         .sa_flags     = SA_SIGINFO,
472                                 };
473                                 perf_top__sig_winch(SIGWINCH, NULL, top);
474                                 sigaction(SIGWINCH, &act, NULL);
475                         } else {
476                                 signal(SIGWINCH, SIG_DFL);
477                         }
478                         break;
479                 case 'E':
480                         if (top->evlist->nr_entries > 1) {
481                                 /* Select 0 as the default event: */
482                                 int counter = 0;
483
484                                 fprintf(stderr, "\nAvailable events:");
485
486                                 evlist__for_each_entry(top->evlist, top->sym_evsel)
487                                         fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
488
489                                 prompt_integer(&counter, "Enter details event counter");
490
491                                 if (counter >= top->evlist->nr_entries) {
492                                         top->sym_evsel = perf_evlist__first(top->evlist);
493                                         fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
494                                         sleep(1);
495                                         break;
496                                 }
497                                 evlist__for_each_entry(top->evlist, top->sym_evsel)
498                                         if (top->sym_evsel->idx == counter)
499                                                 break;
500                         } else
501                                 top->sym_evsel = perf_evlist__first(top->evlist);
502                         break;
503                 case 'f':
504                         prompt_integer(&top->count_filter, "Enter display event count filter");
505                         break;
506                 case 'F':
507                         prompt_percent(&top->sym_pcnt_filter,
508                                        "Enter details display event filter (percent)");
509                         break;
510                 case 'K':
511                         top->hide_kernel_symbols = !top->hide_kernel_symbols;
512                         break;
513                 case 'q':
514                 case 'Q':
515                         printf("exiting.\n");
516                         if (top->dump_symtab)
517                                 perf_session__fprintf_dsos(top->session, stderr);
518                         ret = false;
519                         break;
520                 case 's':
521                         perf_top__prompt_symbol(top, "Enter details symbol");
522                         break;
523                 case 'S':
524                         if (!top->sym_filter_entry)
525                                 break;
526                         else {
527                                 struct hist_entry *syme = top->sym_filter_entry;
528
529                                 top->sym_filter_entry = NULL;
530                                 __zero_source_counters(syme);
531                         }
532                         break;
533                 case 'U':
534                         top->hide_user_symbols = !top->hide_user_symbols;
535                         break;
536                 case 'z':
537                         top->zero = !top->zero;
538                         break;
539                 default:
540                         break;
541         }
542
543         return ret;
544 }
545
546 static void perf_top__sort_new_samples(void *arg)
547 {
548         struct perf_top *t = arg;
549         struct perf_evsel *evsel = t->sym_evsel;
550         struct hists *hists;
551
552         perf_top__reset_sample_counters(t);
553
554         if (t->evlist->selected != NULL)
555                 t->sym_evsel = t->evlist->selected;
556
557         hists = evsel__hists(evsel);
558
559         if (t->evlist->enabled) {
560                 if (t->zero) {
561                         hists__delete_entries(hists);
562                 } else {
563                         hists__decay_entries(hists, t->hide_user_symbols,
564                                              t->hide_kernel_symbols);
565                 }
566         }
567
568         hists__collapse_resort(hists, NULL);
569         perf_evsel__output_resort(evsel, NULL);
570 }
571
572 static void *display_thread_tui(void *arg)
573 {
574         struct perf_evsel *pos;
575         struct perf_top *top = arg;
576         const char *help = "For a higher level overview, try: perf top --sort comm,dso";
577         struct hist_browser_timer hbt = {
578                 .timer          = perf_top__sort_new_samples,
579                 .arg            = top,
580                 .refresh        = top->delay_secs,
581         };
582
583         perf_top__sort_new_samples(top);
584
585         /*
586          * Initialize the uid_filter_str, in the future the TUI will allow
587          * Zooming in/out UIDs. For now juse use whatever the user passed
588          * via --uid.
589          */
590         evlist__for_each_entry(top->evlist, pos) {
591                 struct hists *hists = evsel__hists(pos);
592                 hists->uid_filter_str = top->record_opts.target.uid_str;
593         }
594
595         perf_evlist__tui_browse_hists(top->evlist, help, &hbt,
596                                       top->min_percent,
597                                       &top->session->header.env);
598
599         done = 1;
600         return NULL;
601 }
602
603 static void display_sig(int sig __maybe_unused)
604 {
605         done = 1;
606 }
607
608 static void display_setup_sig(void)
609 {
610         signal(SIGSEGV, sighandler_dump_stack);
611         signal(SIGFPE, sighandler_dump_stack);
612         signal(SIGINT,  display_sig);
613         signal(SIGQUIT, display_sig);
614         signal(SIGTERM, display_sig);
615 }
616
617 static void *display_thread(void *arg)
618 {
619         struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
620         struct termios save;
621         struct perf_top *top = arg;
622         int delay_msecs, c;
623
624         display_setup_sig();
625         pthread__unblock_sigwinch();
626 repeat:
627         delay_msecs = top->delay_secs * 1000;
628         set_term_quiet_input(&save);
629         /* trash return*/
630         getc(stdin);
631
632         while (!done) {
633                 perf_top__print_sym_table(top);
634                 /*
635                  * Either timeout expired or we got an EINTR due to SIGWINCH,
636                  * refresh screen in both cases.
637                  */
638                 switch (poll(&stdin_poll, 1, delay_msecs)) {
639                 case 0:
640                         continue;
641                 case -1:
642                         if (errno == EINTR)
643                                 continue;
644                         /* Fall trhu */
645                 default:
646                         c = getc(stdin);
647                         tcsetattr(0, TCSAFLUSH, &save);
648
649                         if (perf_top__handle_keypress(top, c))
650                                 goto repeat;
651                         done = 1;
652                 }
653         }
654
655         tcsetattr(0, TCSAFLUSH, &save);
656         return NULL;
657 }
658
659 static int symbol_filter(struct map *map, struct symbol *sym)
660 {
661         const char *name = sym->name;
662
663         if (!__map__is_kernel(map))
664                 return 0;
665         /*
666          * ppc64 uses function descriptors and appends a '.' to the
667          * start of every instruction address. Remove it.
668          */
669         if (name[0] == '.')
670                 name++;
671
672         if (!strcmp(name, "_text") ||
673             !strcmp(name, "_etext") ||
674             !strcmp(name, "_sinittext") ||
675             !strncmp("init_module", name, 11) ||
676             !strncmp("cleanup_module", name, 14) ||
677             strstr(name, "_text_start") ||
678             strstr(name, "_text_end"))
679                 return 1;
680
681         if (symbol__is_idle(sym))
682                 sym->ignore = true;
683
684         return 0;
685 }
686
687 static int hist_iter__top_callback(struct hist_entry_iter *iter,
688                                    struct addr_location *al, bool single,
689                                    void *arg)
690 {
691         struct perf_top *top = arg;
692         struct hist_entry *he = iter->he;
693         struct perf_evsel *evsel = iter->evsel;
694
695         if (perf_hpp_list.sym && single)
696                 perf_top__record_precise_ip(top, he, evsel->idx, al->addr);
697
698         hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
699                      !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY));
700         return 0;
701 }
702
703 static void perf_event__process_sample(struct perf_tool *tool,
704                                        const union perf_event *event,
705                                        struct perf_evsel *evsel,
706                                        struct perf_sample *sample,
707                                        struct machine *machine)
708 {
709         struct perf_top *top = container_of(tool, struct perf_top, tool);
710         struct addr_location al;
711         int err;
712
713         if (!machine && perf_guest) {
714                 static struct intlist *seen;
715
716                 if (!seen)
717                         seen = intlist__new(NULL);
718
719                 if (!intlist__has_entry(seen, sample->pid)) {
720                         pr_err("Can't find guest [%d]'s kernel information\n",
721                                 sample->pid);
722                         intlist__add(seen, sample->pid);
723                 }
724                 return;
725         }
726
727         if (!machine) {
728                 pr_err("%u unprocessable samples recorded.\r",
729                        top->session->evlist->stats.nr_unprocessable_samples++);
730                 return;
731         }
732
733         if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
734                 top->exact_samples++;
735
736         if (machine__resolve(machine, &al, sample) < 0)
737                 return;
738
739         if (!machine->kptr_restrict_warned &&
740             symbol_conf.kptr_restrict &&
741             al.cpumode == PERF_RECORD_MISC_KERNEL) {
742                 ui__warning(
743 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
744 "Check /proc/sys/kernel/kptr_restrict.\n\n"
745 "Kernel%s samples will not be resolved.\n",
746                           al.map && !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
747                           " modules" : "");
748                 if (use_browser <= 0)
749                         sleep(5);
750                 machine->kptr_restrict_warned = true;
751         }
752
753         if (al.sym == NULL) {
754                 const char *msg = "Kernel samples will not be resolved.\n";
755                 /*
756                  * As we do lazy loading of symtabs we only will know if the
757                  * specified vmlinux file is invalid when we actually have a
758                  * hit in kernel space and then try to load it. So if we get
759                  * here and there are _no_ symbols in the DSO backing the
760                  * kernel map, bail out.
761                  *
762                  * We may never get here, for instance, if we use -K/
763                  * --hide-kernel-symbols, even if the user specifies an
764                  * invalid --vmlinux ;-)
765                  */
766                 if (!machine->kptr_restrict_warned && !top->vmlinux_warned &&
767                     al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
768                     RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
769                         if (symbol_conf.vmlinux_name) {
770                                 char serr[256];
771                                 dso__strerror_load(al.map->dso, serr, sizeof(serr));
772                                 ui__warning("The %s file can't be used: %s\n%s",
773                                             symbol_conf.vmlinux_name, serr, msg);
774                         } else {
775                                 ui__warning("A vmlinux file was not found.\n%s",
776                                             msg);
777                         }
778
779                         if (use_browser <= 0)
780                                 sleep(5);
781                         top->vmlinux_warned = true;
782                 }
783         }
784
785         if (al.sym == NULL || !al.sym->ignore) {
786                 struct hists *hists = evsel__hists(evsel);
787                 struct hist_entry_iter iter = {
788                         .evsel          = evsel,
789                         .sample         = sample,
790                         .add_entry_cb   = hist_iter__top_callback,
791                 };
792
793                 if (symbol_conf.cumulate_callchain)
794                         iter.ops = &hist_iter_cumulative;
795                 else
796                         iter.ops = &hist_iter_normal;
797
798                 pthread_mutex_lock(&hists->lock);
799
800                 err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
801                 if (err < 0)
802                         pr_err("Problem incrementing symbol period, skipping event\n");
803
804                 pthread_mutex_unlock(&hists->lock);
805         }
806
807         addr_location__put(&al);
808 }
809
810 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
811 {
812         struct perf_sample sample;
813         struct perf_evsel *evsel;
814         struct perf_session *session = top->session;
815         union perf_event *event;
816         struct machine *machine;
817         int ret;
818
819         while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
820                 ret = perf_evlist__parse_sample(top->evlist, event, &sample);
821                 if (ret) {
822                         pr_err("Can't parse sample, err = %d\n", ret);
823                         goto next_event;
824                 }
825
826                 evsel = perf_evlist__id2evsel(session->evlist, sample.id);
827                 assert(evsel != NULL);
828
829                 if (event->header.type == PERF_RECORD_SAMPLE)
830                         ++top->samples;
831
832                 switch (sample.cpumode) {
833                 case PERF_RECORD_MISC_USER:
834                         ++top->us_samples;
835                         if (top->hide_user_symbols)
836                                 goto next_event;
837                         machine = &session->machines.host;
838                         break;
839                 case PERF_RECORD_MISC_KERNEL:
840                         ++top->kernel_samples;
841                         if (top->hide_kernel_symbols)
842                                 goto next_event;
843                         machine = &session->machines.host;
844                         break;
845                 case PERF_RECORD_MISC_GUEST_KERNEL:
846                         ++top->guest_kernel_samples;
847                         machine = perf_session__find_machine(session,
848                                                              sample.pid);
849                         break;
850                 case PERF_RECORD_MISC_GUEST_USER:
851                         ++top->guest_us_samples;
852                         /*
853                          * TODO: we don't process guest user from host side
854                          * except simple counting.
855                          */
856                         goto next_event;
857                 default:
858                         if (event->header.type == PERF_RECORD_SAMPLE)
859                                 goto next_event;
860                         machine = &session->machines.host;
861                         break;
862                 }
863
864
865                 if (event->header.type == PERF_RECORD_SAMPLE) {
866                         perf_event__process_sample(&top->tool, event, evsel,
867                                                    &sample, machine);
868                 } else if (event->header.type < PERF_RECORD_MAX) {
869                         hists__inc_nr_events(evsel__hists(evsel), event->header.type);
870                         machine__process_event(machine, event, &sample);
871                 } else
872                         ++session->evlist->stats.nr_unknown_events;
873 next_event:
874                 perf_evlist__mmap_consume(top->evlist, idx);
875         }
876 }
877
878 static void perf_top__mmap_read(struct perf_top *top)
879 {
880         int i;
881
882         for (i = 0; i < top->evlist->nr_mmaps; i++)
883                 perf_top__mmap_read_idx(top, i);
884 }
885
886 static int perf_top__start_counters(struct perf_top *top)
887 {
888         char msg[512];
889         struct perf_evsel *counter;
890         struct perf_evlist *evlist = top->evlist;
891         struct record_opts *opts = &top->record_opts;
892
893         perf_evlist__config(evlist, opts, &callchain_param);
894
895         evlist__for_each_entry(evlist, counter) {
896 try_again:
897                 if (perf_evsel__open(counter, top->evlist->cpus,
898                                      top->evlist->threads) < 0) {
899                         if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
900                                 if (verbose)
901                                         ui__warning("%s\n", msg);
902                                 goto try_again;
903                         }
904
905                         perf_evsel__open_strerror(counter, &opts->target,
906                                                   errno, msg, sizeof(msg));
907                         ui__error("%s\n", msg);
908                         goto out_err;
909                 }
910         }
911
912         if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
913                 ui__error("Failed to mmap with %d (%s)\n",
914                             errno, str_error_r(errno, msg, sizeof(msg)));
915                 goto out_err;
916         }
917
918         return 0;
919
920 out_err:
921         return -1;
922 }
923
924 static int callchain_param__setup_sample_type(struct callchain_param *callchain)
925 {
926         if (!perf_hpp_list.sym) {
927                 if (callchain->enabled) {
928                         ui__error("Selected -g but \"sym\" not present in --sort/-s.");
929                         return -EINVAL;
930                 }
931         } else if (callchain->mode != CHAIN_NONE) {
932                 if (callchain_register_param(callchain) < 0) {
933                         ui__error("Can't register callchain params.\n");
934                         return -EINVAL;
935                 }
936         }
937
938         return 0;
939 }
940
941 static int __cmd_top(struct perf_top *top)
942 {
943         struct record_opts *opts = &top->record_opts;
944         pthread_t thread;
945         int ret;
946
947         top->session = perf_session__new(NULL, false, NULL);
948         if (top->session == NULL)
949                 return -1;
950
951         machines__set_symbol_filter(&top->session->machines, symbol_filter);
952
953         if (!objdump_path) {
954                 ret = perf_env__lookup_objdump(&top->session->header.env);
955                 if (ret)
956                         goto out_delete;
957         }
958
959         ret = callchain_param__setup_sample_type(&callchain_param);
960         if (ret)
961                 goto out_delete;
962
963         if (perf_session__register_idle_thread(top->session) < 0)
964                 goto out_delete;
965
966         machine__synthesize_threads(&top->session->machines.host, &opts->target,
967                                     top->evlist->threads, false, opts->proc_map_timeout);
968
969         if (perf_hpp_list.socket) {
970                 ret = perf_env__read_cpu_topology_map(&perf_env);
971                 if (ret < 0)
972                         goto out_err_cpu_topo;
973         }
974
975         ret = perf_top__start_counters(top);
976         if (ret)
977                 goto out_delete;
978
979         top->session->evlist = top->evlist;
980         perf_session__set_id_hdr_size(top->session);
981
982         /*
983          * When perf is starting the traced process, all the events (apart from
984          * group members) have enable_on_exec=1 set, so don't spoil it by
985          * prematurely enabling them.
986          *
987          * XXX 'top' still doesn't start workloads like record, trace, but should,
988          * so leave the check here.
989          */
990         if (!target__none(&opts->target))
991                 perf_evlist__enable(top->evlist);
992
993         /* Wait for a minimal set of events before starting the snapshot */
994         perf_evlist__poll(top->evlist, 100);
995
996         perf_top__mmap_read(top);
997
998         ret = -1;
999         if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
1000                                                             display_thread), top)) {
1001                 ui__error("Could not create display thread.\n");
1002                 goto out_delete;
1003         }
1004
1005         if (top->realtime_prio) {
1006                 struct sched_param param;
1007
1008                 param.sched_priority = top->realtime_prio;
1009                 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
1010                         ui__error("Could not set realtime priority.\n");
1011                         goto out_join;
1012                 }
1013         }
1014
1015         while (!done) {
1016                 u64 hits = top->samples;
1017
1018                 perf_top__mmap_read(top);
1019
1020                 if (hits == top->samples)
1021                         ret = perf_evlist__poll(top->evlist, 100);
1022         }
1023
1024         ret = 0;
1025 out_join:
1026         pthread_join(thread, NULL);
1027 out_delete:
1028         perf_session__delete(top->session);
1029         top->session = NULL;
1030
1031         return ret;
1032
1033 out_err_cpu_topo: {
1034         char errbuf[BUFSIZ];
1035         const char *err = str_error_r(-ret, errbuf, sizeof(errbuf));
1036
1037         ui__error("Could not read the CPU topology map: %s\n", err);
1038         goto out_delete;
1039 }
1040 }
1041
1042 static int
1043 callchain_opt(const struct option *opt, const char *arg, int unset)
1044 {
1045         symbol_conf.use_callchain = true;
1046         return record_callchain_opt(opt, arg, unset);
1047 }
1048
1049 static int
1050 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1051 {
1052         struct callchain_param *callchain = opt->value;
1053
1054         callchain->enabled = !unset;
1055         callchain->record_mode = CALLCHAIN_FP;
1056
1057         /*
1058          * --no-call-graph
1059          */
1060         if (unset) {
1061                 symbol_conf.use_callchain = false;
1062                 callchain->record_mode = CALLCHAIN_NONE;
1063                 return 0;
1064         }
1065
1066         return parse_callchain_top_opt(arg);
1067 }
1068
1069 static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1070 {
1071         if (!strcmp(var, "top.call-graph"))
1072                 var = "call-graph.record-mode"; /* fall-through */
1073         if (!strcmp(var, "top.children")) {
1074                 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
1075                 return 0;
1076         }
1077
1078         return 0;
1079 }
1080
1081 static int
1082 parse_percent_limit(const struct option *opt, const char *arg,
1083                     int unset __maybe_unused)
1084 {
1085         struct perf_top *top = opt->value;
1086
1087         top->min_percent = strtof(arg, NULL);
1088         return 0;
1089 }
1090
1091 const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1092         "\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1093
1094 int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1095 {
1096         char errbuf[BUFSIZ];
1097         struct perf_top top = {
1098                 .count_filter        = 5,
1099                 .delay_secs          = 2,
1100                 .record_opts = {
1101                         .mmap_pages     = UINT_MAX,
1102                         .user_freq      = UINT_MAX,
1103                         .user_interval  = ULLONG_MAX,
1104                         .freq           = 4000, /* 4 KHz */
1105                         .target         = {
1106                                 .uses_mmap   = true,
1107                         },
1108                         .proc_map_timeout    = 500,
1109                 },
1110                 .max_stack           = sysctl_perf_event_max_stack,
1111                 .sym_pcnt_filter     = 5,
1112         };
1113         struct record_opts *opts = &top.record_opts;
1114         struct target *target = &opts->target;
1115         const struct option options[] = {
1116         OPT_CALLBACK('e', "event", &top.evlist, "event",
1117                      "event selector. use 'perf list' to list available events",
1118                      parse_events_option),
1119         OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1120         OPT_STRING('p', "pid", &target->pid, "pid",
1121                     "profile events on existing process id"),
1122         OPT_STRING('t', "tid", &target->tid, "tid",
1123                     "profile events on existing thread id"),
1124         OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1125                             "system-wide collection from all CPUs"),
1126         OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1127                     "list of cpus to monitor"),
1128         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1129                    "file", "vmlinux pathname"),
1130         OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1131                     "don't load vmlinux even if found"),
1132         OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1133                     "hide kernel symbols"),
1134         OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1135                      "number of mmap data pages",
1136                      perf_evlist__parse_mmap_pages),
1137         OPT_INTEGER('r', "realtime", &top.realtime_prio,
1138                     "collect data with this RT SCHED_FIFO priority"),
1139         OPT_INTEGER('d', "delay", &top.delay_secs,
1140                     "number of seconds to delay between refreshes"),
1141         OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1142                             "dump the symbol table used for profiling"),
1143         OPT_INTEGER('f', "count-filter", &top.count_filter,
1144                     "only display functions with more events than this"),
1145         OPT_BOOLEAN(0, "group", &opts->group,
1146                             "put the counters into a counter group"),
1147         OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1148                     "child tasks do not inherit counters"),
1149         OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1150                     "symbol to annotate"),
1151         OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1152         OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1153         OPT_INTEGER('E', "entries", &top.print_entries,
1154                     "display this many functions"),
1155         OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1156                     "hide user symbols"),
1157         OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1158         OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1159         OPT_INCR('v', "verbose", &verbose,
1160                     "be more verbose (show counter open errors, etc)"),
1161         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1162                    "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1163                    " Please refer the man page for the complete list."),
1164         OPT_STRING(0, "fields", &field_order, "key[,keys...]",
1165                    "output field(s): overhead, period, sample plus all of sort keys"),
1166         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1167                     "Show a column with the number of samples"),
1168         OPT_CALLBACK_NOOPT('g', NULL, &callchain_param,
1169                            NULL, "enables call-graph recording and display",
1170                            &callchain_opt),
1171         OPT_CALLBACK(0, "call-graph", &callchain_param,
1172                      "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1173                      top_callchain_help, &parse_callchain_opt),
1174         OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1175                     "Accumulate callchains of children and show total overhead as well"),
1176         OPT_INTEGER(0, "max-stack", &top.max_stack,
1177                     "Set the maximum stack depth when parsing the callchain. "
1178                     "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1179         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1180                    "ignore callees of these functions in call graphs",
1181                    report_parse_ignore_callees_opt),
1182         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1183                     "Show a column with the sum of periods"),
1184         OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1185                    "only consider symbols in these dsos"),
1186         OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1187                    "only consider symbols in these comms"),
1188         OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1189                    "only consider these symbols"),
1190         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1191                     "Interleave source code with assembly code (default)"),
1192         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1193                     "Display raw encoding of assembly instructions (default)"),
1194         OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1195                     "Enable kernel symbol demangling"),
1196         OPT_STRING(0, "objdump", &objdump_path, "path",
1197                     "objdump binary to use for disassembly and annotations"),
1198         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1199                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
1200         OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1201         OPT_CALLBACK(0, "percent-limit", &top, "percent",
1202                      "Don't show entries under that percent", parse_percent_limit),
1203         OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1204                      "How to display percentage of filtered entries", parse_filter_percentage),
1205         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1206                    "width[,width...]",
1207                    "don't try to adjust column width, use these fixed values"),
1208         OPT_UINTEGER(0, "proc-map-timeout", &opts->proc_map_timeout,
1209                         "per thread proc mmap processing timeout in ms"),
1210         OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
1211                      "branch any", "sample any taken branches",
1212                      parse_branch_stack),
1213         OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
1214                      "branch filter mask", "branch stack filter modes",
1215                      parse_branch_stack),
1216         OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1217                     "Show raw trace event output (do not use print fmt or plugins)"),
1218         OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1219                     "Show entries in a hierarchy"),
1220         OPT_END()
1221         };
1222         const char * const top_usage[] = {
1223                 "perf top [<options>]",
1224                 NULL
1225         };
1226         int status = hists__init();
1227
1228         if (status < 0)
1229                 return status;
1230
1231         top.evlist = perf_evlist__new();
1232         if (top.evlist == NULL)
1233                 return -ENOMEM;
1234
1235         perf_config(perf_top_config, &top);
1236
1237         argc = parse_options(argc, argv, options, top_usage, 0);
1238         if (argc)
1239                 usage_with_options(top_usage, options);
1240
1241         if (!top.evlist->nr_entries &&
1242             perf_evlist__add_default(top.evlist) < 0) {
1243                 pr_err("Not enough memory for event selector list\n");
1244                 goto out_delete_evlist;
1245         }
1246
1247         if (symbol_conf.report_hierarchy) {
1248                 /* disable incompatible options */
1249                 symbol_conf.event_group = false;
1250                 symbol_conf.cumulate_callchain = false;
1251
1252                 if (field_order) {
1253                         pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1254                         parse_options_usage(top_usage, options, "fields", 0);
1255                         parse_options_usage(NULL, options, "hierarchy", 0);
1256                         goto out_delete_evlist;
1257                 }
1258         }
1259
1260         sort__mode = SORT_MODE__TOP;
1261         /* display thread wants entries to be collapsed in a different tree */
1262         perf_hpp_list.need_collapse = 1;
1263
1264         if (top.use_stdio)
1265                 use_browser = 0;
1266         else if (top.use_tui)
1267                 use_browser = 1;
1268
1269         setup_browser(false);
1270
1271         if (setup_sorting(top.evlist) < 0) {
1272                 if (sort_order)
1273                         parse_options_usage(top_usage, options, "s", 1);
1274                 if (field_order)
1275                         parse_options_usage(sort_order ? NULL : top_usage,
1276                                             options, "fields", 0);
1277                 goto out_delete_evlist;
1278         }
1279
1280         status = target__validate(target);
1281         if (status) {
1282                 target__strerror(target, status, errbuf, BUFSIZ);
1283                 ui__warning("%s\n", errbuf);
1284         }
1285
1286         status = target__parse_uid(target);
1287         if (status) {
1288                 int saved_errno = errno;
1289
1290                 target__strerror(target, status, errbuf, BUFSIZ);
1291                 ui__error("%s\n", errbuf);
1292
1293                 status = -saved_errno;
1294                 goto out_delete_evlist;
1295         }
1296
1297         if (target__none(target))
1298                 target->system_wide = true;
1299
1300         if (perf_evlist__create_maps(top.evlist, target) < 0) {
1301                 ui__error("Couldn't create thread/CPU maps: %s\n",
1302                           errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf)));
1303                 goto out_delete_evlist;
1304         }
1305
1306         symbol_conf.nr_events = top.evlist->nr_entries;
1307
1308         if (top.delay_secs < 1)
1309                 top.delay_secs = 1;
1310
1311         if (record_opts__config(opts)) {
1312                 status = -EINVAL;
1313                 goto out_delete_evlist;
1314         }
1315
1316         top.sym_evsel = perf_evlist__first(top.evlist);
1317
1318         if (!callchain_param.enabled) {
1319                 symbol_conf.cumulate_callchain = false;
1320                 perf_hpp__cancel_cumulate();
1321         }
1322
1323         if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1324                 callchain_param.order = ORDER_CALLER;
1325
1326         symbol_conf.priv_size = sizeof(struct annotation);
1327
1328         symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1329         if (symbol__init(NULL) < 0)
1330                 return -1;
1331
1332         sort__setup_elide(stdout);
1333
1334         get_term_dimensions(&top.winsize);
1335         if (top.print_entries == 0) {
1336                 struct sigaction act = {
1337                         .sa_sigaction = perf_top__sig_winch,
1338                         .sa_flags     = SA_SIGINFO,
1339                 };
1340                 perf_top__update_print_entries(&top);
1341                 sigaction(SIGWINCH, &act, NULL);
1342         }
1343
1344         status = __cmd_top(&top);
1345
1346 out_delete_evlist:
1347         perf_evlist__delete(top.evlist);
1348
1349         return status;
1350 }