Loading...
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 *
9 * Improvements and fixes by:
10 *
11 * Arjan van de Ven <arjan@linux.intel.com>
12 * Yanmin Zhang <yanmin.zhang@intel.com>
13 * Wu Fengguang <fengguang.wu@intel.com>
14 * Mike Galbraith <efault@gmx.de>
15 * Paul Mackerras <paulus@samba.org>
16 *
17 * Released under the GPL v2. (and only v2, not any later version)
18 */
19#include "builtin.h"
20
21#include "perf.h"
22
23#include "util/annotate.h"
24#include "util/cache.h"
25#include "util/color.h"
26#include "util/evlist.h"
27#include "util/evsel.h"
28#include "util/session.h"
29#include "util/symbol.h"
30#include "util/thread.h"
31#include "util/thread_map.h"
32#include "util/top.h"
33#include "util/util.h"
34#include <linux/rbtree.h>
35#include "util/parse-options.h"
36#include "util/parse-events.h"
37#include "util/cpumap.h"
38#include "util/xyarray.h"
39
40#include "util/debug.h"
41
42#include <assert.h>
43#include <fcntl.h>
44
45#include <stdio.h>
46#include <termios.h>
47#include <unistd.h>
48#include <inttypes.h>
49
50#include <errno.h>
51#include <time.h>
52#include <sched.h>
53
54#include <sys/syscall.h>
55#include <sys/ioctl.h>
56#include <sys/poll.h>
57#include <sys/prctl.h>
58#include <sys/wait.h>
59#include <sys/uio.h>
60#include <sys/mman.h>
61
62#include <linux/unistd.h>
63#include <linux/types.h>
64
65static struct perf_top top = {
66 .count_filter = 5,
67 .delay_secs = 2,
68 .display_weighted = -1,
69 .target_pid = -1,
70 .target_tid = -1,
71 .active_symbols = LIST_HEAD_INIT(top.active_symbols),
72 .active_symbols_lock = PTHREAD_MUTEX_INITIALIZER,
73 .active_symbols_cond = PTHREAD_COND_INITIALIZER,
74 .freq = 1000, /* 1 KHz */
75};
76
77static bool system_wide = false;
78
79static bool use_tui, use_stdio;
80
81static int default_interval = 0;
82
83static bool kptr_restrict_warned;
84static bool vmlinux_warned;
85static bool inherit = false;
86static int realtime_prio = 0;
87static bool group = false;
88static unsigned int page_size;
89static unsigned int mmap_pages = 128;
90
91static bool dump_symtab = false;
92
93static struct winsize winsize;
94
95static const char *sym_filter = NULL;
96struct sym_entry *sym_filter_entry_sched = NULL;
97static int sym_pcnt_filter = 5;
98
99/*
100 * Source functions
101 */
102
103void get_term_dimensions(struct winsize *ws)
104{
105 char *s = getenv("LINES");
106
107 if (s != NULL) {
108 ws->ws_row = atoi(s);
109 s = getenv("COLUMNS");
110 if (s != NULL) {
111 ws->ws_col = atoi(s);
112 if (ws->ws_row && ws->ws_col)
113 return;
114 }
115 }
116#ifdef TIOCGWINSZ
117 if (ioctl(1, TIOCGWINSZ, ws) == 0 &&
118 ws->ws_row && ws->ws_col)
119 return;
120#endif
121 ws->ws_row = 25;
122 ws->ws_col = 80;
123}
124
125static void update_print_entries(struct winsize *ws)
126{
127 top.print_entries = ws->ws_row;
128
129 if (top.print_entries > 9)
130 top.print_entries -= 9;
131}
132
133static void sig_winch_handler(int sig __used)
134{
135 get_term_dimensions(&winsize);
136 update_print_entries(&winsize);
137}
138
139static int parse_source(struct sym_entry *syme)
140{
141 struct symbol *sym;
142 struct annotation *notes;
143 struct map *map;
144 int err = -1;
145
146 if (!syme)
147 return -1;
148
149 sym = sym_entry__symbol(syme);
150 map = syme->map;
151
152 /*
153 * We can't annotate with just /proc/kallsyms
154 */
155 if (map->dso->symtab_type == SYMTAB__KALLSYMS) {
156 pr_err("Can't annotate %s: No vmlinux file was found in the "
157 "path\n", sym->name);
158 sleep(1);
159 return -1;
160 }
161
162 notes = symbol__annotation(sym);
163 if (notes->src != NULL) {
164 pthread_mutex_lock(¬es->lock);
165 goto out_assign;
166 }
167
168 pthread_mutex_lock(¬es->lock);
169
170 if (symbol__alloc_hist(sym, top.evlist->nr_entries) < 0) {
171 pthread_mutex_unlock(¬es->lock);
172 pr_err("Not enough memory for annotating '%s' symbol!\n",
173 sym->name);
174 sleep(1);
175 return err;
176 }
177
178 err = symbol__annotate(sym, syme->map, 0);
179 if (err == 0) {
180out_assign:
181 top.sym_filter_entry = syme;
182 }
183
184 pthread_mutex_unlock(¬es->lock);
185 return err;
186}
187
188static void __zero_source_counters(struct sym_entry *syme)
189{
190 struct symbol *sym = sym_entry__symbol(syme);
191 symbol__annotate_zero_histograms(sym);
192}
193
194static void record_precise_ip(struct sym_entry *syme, struct map *map,
195 int counter, u64 ip)
196{
197 struct annotation *notes;
198 struct symbol *sym;
199
200 if (syme != top.sym_filter_entry)
201 return;
202
203 sym = sym_entry__symbol(syme);
204 notes = symbol__annotation(sym);
205
206 if (pthread_mutex_trylock(¬es->lock))
207 return;
208
209 ip = map->map_ip(map, ip);
210 symbol__inc_addr_samples(sym, map, counter, ip);
211
212 pthread_mutex_unlock(¬es->lock);
213}
214
215static void show_details(struct sym_entry *syme)
216{
217 struct annotation *notes;
218 struct symbol *symbol;
219 int more;
220
221 if (!syme)
222 return;
223
224 symbol = sym_entry__symbol(syme);
225 notes = symbol__annotation(symbol);
226
227 pthread_mutex_lock(¬es->lock);
228
229 if (notes->src == NULL)
230 goto out_unlock;
231
232 printf("Showing %s for %s\n", event_name(top.sym_evsel), symbol->name);
233 printf(" Events Pcnt (>=%d%%)\n", sym_pcnt_filter);
234
235 more = symbol__annotate_printf(symbol, syme->map, top.sym_evsel->idx,
236 0, sym_pcnt_filter, top.print_entries, 4);
237 if (top.zero)
238 symbol__annotate_zero_histogram(symbol, top.sym_evsel->idx);
239 else
240 symbol__annotate_decay_histogram(symbol, top.sym_evsel->idx);
241 if (more != 0)
242 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
243out_unlock:
244 pthread_mutex_unlock(¬es->lock);
245}
246
247static const char CONSOLE_CLEAR[] = "[H[2J";
248
249static void __list_insert_active_sym(struct sym_entry *syme)
250{
251 list_add(&syme->node, &top.active_symbols);
252}
253
254static void print_sym_table(struct perf_session *session)
255{
256 char bf[160];
257 int printed = 0;
258 struct rb_node *nd;
259 struct sym_entry *syme;
260 struct rb_root tmp = RB_ROOT;
261 const int win_width = winsize.ws_col - 1;
262 int sym_width, dso_width, dso_short_width;
263 float sum_ksamples = perf_top__decay_samples(&top, &tmp);
264
265 puts(CONSOLE_CLEAR);
266
267 perf_top__header_snprintf(&top, bf, sizeof(bf));
268 printf("%s\n", bf);
269
270 perf_top__reset_sample_counters(&top);
271
272 printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
273
274 if (session->hists.stats.total_lost != 0) {
275 color_fprintf(stdout, PERF_COLOR_RED, "WARNING:");
276 printf(" LOST %" PRIu64 " events, Check IO/CPU overload\n",
277 session->hists.stats.total_lost);
278 }
279
280 if (top.sym_filter_entry) {
281 show_details(top.sym_filter_entry);
282 return;
283 }
284
285 perf_top__find_widths(&top, &tmp, &dso_width, &dso_short_width,
286 &sym_width);
287
288 if (sym_width + dso_width > winsize.ws_col - 29) {
289 dso_width = dso_short_width;
290 if (sym_width + dso_width > winsize.ws_col - 29)
291 sym_width = winsize.ws_col - dso_width - 29;
292 }
293 putchar('\n');
294 if (top.evlist->nr_entries == 1)
295 printf(" samples pcnt");
296 else
297 printf(" weight samples pcnt");
298
299 if (verbose)
300 printf(" RIP ");
301 printf(" %-*.*s DSO\n", sym_width, sym_width, "function");
302 printf(" %s _______ _____",
303 top.evlist->nr_entries == 1 ? " " : "______");
304 if (verbose)
305 printf(" ________________");
306 printf(" %-*.*s", sym_width, sym_width, graph_line);
307 printf(" %-*.*s", dso_width, dso_width, graph_line);
308 puts("\n");
309
310 for (nd = rb_first(&tmp); nd; nd = rb_next(nd)) {
311 struct symbol *sym;
312 double pcnt;
313
314 syme = rb_entry(nd, struct sym_entry, rb_node);
315 sym = sym_entry__symbol(syme);
316 if (++printed > top.print_entries ||
317 (int)syme->snap_count < top.count_filter)
318 continue;
319
320 pcnt = 100.0 - (100.0 * ((sum_ksamples - syme->snap_count) /
321 sum_ksamples));
322
323 if (top.evlist->nr_entries == 1 || !top.display_weighted)
324 printf("%20.2f ", syme->weight);
325 else
326 printf("%9.1f %10ld ", syme->weight, syme->snap_count);
327
328 percent_color_fprintf(stdout, "%4.1f%%", pcnt);
329 if (verbose)
330 printf(" %016" PRIx64, sym->start);
331 printf(" %-*.*s", sym_width, sym_width, sym->name);
332 printf(" %-*.*s\n", dso_width, dso_width,
333 dso_width >= syme->map->dso->long_name_len ?
334 syme->map->dso->long_name :
335 syme->map->dso->short_name);
336 }
337}
338
339static void prompt_integer(int *target, const char *msg)
340{
341 char *buf = malloc(0), *p;
342 size_t dummy = 0;
343 int tmp;
344
345 fprintf(stdout, "\n%s: ", msg);
346 if (getline(&buf, &dummy, stdin) < 0)
347 return;
348
349 p = strchr(buf, '\n');
350 if (p)
351 *p = 0;
352
353 p = buf;
354 while(*p) {
355 if (!isdigit(*p))
356 goto out_free;
357 p++;
358 }
359 tmp = strtoul(buf, NULL, 10);
360 *target = tmp;
361out_free:
362 free(buf);
363}
364
365static void prompt_percent(int *target, const char *msg)
366{
367 int tmp = 0;
368
369 prompt_integer(&tmp, msg);
370 if (tmp >= 0 && tmp <= 100)
371 *target = tmp;
372}
373
374static void prompt_symbol(struct sym_entry **target, const char *msg)
375{
376 char *buf = malloc(0), *p;
377 struct sym_entry *syme = *target, *n, *found = NULL;
378 size_t dummy = 0;
379
380 /* zero counters of active symbol */
381 if (syme) {
382 __zero_source_counters(syme);
383 *target = NULL;
384 }
385
386 fprintf(stdout, "\n%s: ", msg);
387 if (getline(&buf, &dummy, stdin) < 0)
388 goto out_free;
389
390 p = strchr(buf, '\n');
391 if (p)
392 *p = 0;
393
394 pthread_mutex_lock(&top.active_symbols_lock);
395 syme = list_entry(top.active_symbols.next, struct sym_entry, node);
396 pthread_mutex_unlock(&top.active_symbols_lock);
397
398 list_for_each_entry_safe_from(syme, n, &top.active_symbols, node) {
399 struct symbol *sym = sym_entry__symbol(syme);
400
401 if (!strcmp(buf, sym->name)) {
402 found = syme;
403 break;
404 }
405 }
406
407 if (!found) {
408 fprintf(stderr, "Sorry, %s is not active.\n", buf);
409 sleep(1);
410 return;
411 } else
412 parse_source(found);
413
414out_free:
415 free(buf);
416}
417
418static void print_mapped_keys(void)
419{
420 char *name = NULL;
421
422 if (top.sym_filter_entry) {
423 struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
424 name = sym->name;
425 }
426
427 fprintf(stdout, "\nMapped keys:\n");
428 fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top.delay_secs);
429 fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top.print_entries);
430
431 if (top.evlist->nr_entries > 1)
432 fprintf(stdout, "\t[E] active event counter. \t(%s)\n", event_name(top.sym_evsel));
433
434 fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top.count_filter);
435
436 fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", sym_pcnt_filter);
437 fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL");
438 fprintf(stdout, "\t[S] stop annotation.\n");
439
440 if (top.evlist->nr_entries > 1)
441 fprintf(stdout, "\t[w] toggle display weighted/count[E]r. \t(%d)\n", top.display_weighted ? 1 : 0);
442
443 fprintf(stdout,
444 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
445 top.hide_kernel_symbols ? "yes" : "no");
446 fprintf(stdout,
447 "\t[U] hide user symbols. \t(%s)\n",
448 top.hide_user_symbols ? "yes" : "no");
449 fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top.zero ? 1 : 0);
450 fprintf(stdout, "\t[qQ] quit.\n");
451}
452
453static int key_mapped(int c)
454{
455 switch (c) {
456 case 'd':
457 case 'e':
458 case 'f':
459 case 'z':
460 case 'q':
461 case 'Q':
462 case 'K':
463 case 'U':
464 case 'F':
465 case 's':
466 case 'S':
467 return 1;
468 case 'E':
469 case 'w':
470 return top.evlist->nr_entries > 1 ? 1 : 0;
471 default:
472 break;
473 }
474
475 return 0;
476}
477
478static void handle_keypress(struct perf_session *session, int c)
479{
480 if (!key_mapped(c)) {
481 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
482 struct termios tc, save;
483
484 print_mapped_keys();
485 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
486 fflush(stdout);
487
488 tcgetattr(0, &save);
489 tc = save;
490 tc.c_lflag &= ~(ICANON | ECHO);
491 tc.c_cc[VMIN] = 0;
492 tc.c_cc[VTIME] = 0;
493 tcsetattr(0, TCSANOW, &tc);
494
495 poll(&stdin_poll, 1, -1);
496 c = getc(stdin);
497
498 tcsetattr(0, TCSAFLUSH, &save);
499 if (!key_mapped(c))
500 return;
501 }
502
503 switch (c) {
504 case 'd':
505 prompt_integer(&top.delay_secs, "Enter display delay");
506 if (top.delay_secs < 1)
507 top.delay_secs = 1;
508 break;
509 case 'e':
510 prompt_integer(&top.print_entries, "Enter display entries (lines)");
511 if (top.print_entries == 0) {
512 sig_winch_handler(SIGWINCH);
513 signal(SIGWINCH, sig_winch_handler);
514 } else
515 signal(SIGWINCH, SIG_DFL);
516 break;
517 case 'E':
518 if (top.evlist->nr_entries > 1) {
519 /* Select 0 as the default event: */
520 int counter = 0;
521
522 fprintf(stderr, "\nAvailable events:");
523
524 list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
525 fprintf(stderr, "\n\t%d %s", top.sym_evsel->idx, event_name(top.sym_evsel));
526
527 prompt_integer(&counter, "Enter details event counter");
528
529 if (counter >= top.evlist->nr_entries) {
530 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
531 fprintf(stderr, "Sorry, no such event, using %s.\n", event_name(top.sym_evsel));
532 sleep(1);
533 break;
534 }
535 list_for_each_entry(top.sym_evsel, &top.evlist->entries, node)
536 if (top.sym_evsel->idx == counter)
537 break;
538 } else
539 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
540 break;
541 case 'f':
542 prompt_integer(&top.count_filter, "Enter display event count filter");
543 break;
544 case 'F':
545 prompt_percent(&sym_pcnt_filter, "Enter details display event filter (percent)");
546 break;
547 case 'K':
548 top.hide_kernel_symbols = !top.hide_kernel_symbols;
549 break;
550 case 'q':
551 case 'Q':
552 printf("exiting.\n");
553 if (dump_symtab)
554 perf_session__fprintf_dsos(session, stderr);
555 exit(0);
556 case 's':
557 prompt_symbol(&top.sym_filter_entry, "Enter details symbol");
558 break;
559 case 'S':
560 if (!top.sym_filter_entry)
561 break;
562 else {
563 struct sym_entry *syme = top.sym_filter_entry;
564
565 top.sym_filter_entry = NULL;
566 __zero_source_counters(syme);
567 }
568 break;
569 case 'U':
570 top.hide_user_symbols = !top.hide_user_symbols;
571 break;
572 case 'w':
573 top.display_weighted = ~top.display_weighted;
574 break;
575 case 'z':
576 top.zero = !top.zero;
577 break;
578 default:
579 break;
580 }
581}
582
583static void *display_thread_tui(void *arg __used)
584{
585 int err = 0;
586 pthread_mutex_lock(&top.active_symbols_lock);
587 while (list_empty(&top.active_symbols)) {
588 err = pthread_cond_wait(&top.active_symbols_cond,
589 &top.active_symbols_lock);
590 if (err)
591 break;
592 }
593 pthread_mutex_unlock(&top.active_symbols_lock);
594 if (!err)
595 perf_top__tui_browser(&top);
596 exit_browser(0);
597 exit(0);
598 return NULL;
599}
600
601static void *display_thread(void *arg __used)
602{
603 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
604 struct termios tc, save;
605 int delay_msecs, c;
606 struct perf_session *session = (struct perf_session *) arg;
607
608 tcgetattr(0, &save);
609 tc = save;
610 tc.c_lflag &= ~(ICANON | ECHO);
611 tc.c_cc[VMIN] = 0;
612 tc.c_cc[VTIME] = 0;
613
614repeat:
615 delay_msecs = top.delay_secs * 1000;
616 tcsetattr(0, TCSANOW, &tc);
617 /* trash return*/
618 getc(stdin);
619
620 do {
621 print_sym_table(session);
622 } while (!poll(&stdin_poll, 1, delay_msecs) == 1);
623
624 c = getc(stdin);
625 tcsetattr(0, TCSAFLUSH, &save);
626
627 handle_keypress(session, c);
628 goto repeat;
629
630 return NULL;
631}
632
633/* Tag samples to be skipped. */
634static const char *skip_symbols[] = {
635 "default_idle",
636 "native_safe_halt",
637 "cpu_idle",
638 "enter_idle",
639 "exit_idle",
640 "mwait_idle",
641 "mwait_idle_with_hints",
642 "poll_idle",
643 "ppc64_runlatch_off",
644 "pseries_dedicated_idle_sleep",
645 NULL
646};
647
648static int symbol_filter(struct map *map, struct symbol *sym)
649{
650 struct sym_entry *syme;
651 const char *name = sym->name;
652 int i;
653
654 /*
655 * ppc64 uses function descriptors and appends a '.' to the
656 * start of every instruction address. Remove it.
657 */
658 if (name[0] == '.')
659 name++;
660
661 if (!strcmp(name, "_text") ||
662 !strcmp(name, "_etext") ||
663 !strcmp(name, "_sinittext") ||
664 !strncmp("init_module", name, 11) ||
665 !strncmp("cleanup_module", name, 14) ||
666 strstr(name, "_text_start") ||
667 strstr(name, "_text_end"))
668 return 1;
669
670 syme = symbol__priv(sym);
671 syme->map = map;
672 symbol__annotate_init(map, sym);
673
674 if (!top.sym_filter_entry && sym_filter && !strcmp(name, sym_filter)) {
675 /* schedule initial sym_filter_entry setup */
676 sym_filter_entry_sched = syme;
677 sym_filter = NULL;
678 }
679
680 for (i = 0; skip_symbols[i]; i++) {
681 if (!strcmp(skip_symbols[i], name)) {
682 sym->ignore = true;
683 break;
684 }
685 }
686
687 return 0;
688}
689
690static void perf_event__process_sample(const union perf_event *event,
691 struct perf_sample *sample,
692 struct perf_session *session)
693{
694 u64 ip = event->ip.ip;
695 struct sym_entry *syme;
696 struct addr_location al;
697 struct machine *machine;
698 u8 origin = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
699
700 ++top.samples;
701
702 switch (origin) {
703 case PERF_RECORD_MISC_USER:
704 ++top.us_samples;
705 if (top.hide_user_symbols)
706 return;
707 machine = perf_session__find_host_machine(session);
708 break;
709 case PERF_RECORD_MISC_KERNEL:
710 ++top.kernel_samples;
711 if (top.hide_kernel_symbols)
712 return;
713 machine = perf_session__find_host_machine(session);
714 break;
715 case PERF_RECORD_MISC_GUEST_KERNEL:
716 ++top.guest_kernel_samples;
717 machine = perf_session__find_machine(session, event->ip.pid);
718 break;
719 case PERF_RECORD_MISC_GUEST_USER:
720 ++top.guest_us_samples;
721 /*
722 * TODO: we don't process guest user from host side
723 * except simple counting.
724 */
725 return;
726 default:
727 return;
728 }
729
730 if (!machine && perf_guest) {
731 pr_err("Can't find guest [%d]'s kernel information\n",
732 event->ip.pid);
733 return;
734 }
735
736 if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
737 top.exact_samples++;
738
739 if (perf_event__preprocess_sample(event, session, &al, sample,
740 symbol_filter) < 0 ||
741 al.filtered)
742 return;
743
744 if (!kptr_restrict_warned &&
745 symbol_conf.kptr_restrict &&
746 al.cpumode == PERF_RECORD_MISC_KERNEL) {
747 ui__warning(
748"Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
749"Check /proc/sys/kernel/kptr_restrict.\n\n"
750"Kernel%s samples will not be resolved.\n",
751 !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
752 " modules" : "");
753 if (use_browser <= 0)
754 sleep(5);
755 kptr_restrict_warned = true;
756 }
757
758 if (al.sym == NULL) {
759 const char *msg = "Kernel samples will not be resolved.\n";
760 /*
761 * As we do lazy loading of symtabs we only will know if the
762 * specified vmlinux file is invalid when we actually have a
763 * hit in kernel space and then try to load it. So if we get
764 * here and there are _no_ symbols in the DSO backing the
765 * kernel map, bail out.
766 *
767 * We may never get here, for instance, if we use -K/
768 * --hide-kernel-symbols, even if the user specifies an
769 * invalid --vmlinux ;-)
770 */
771 if (!kptr_restrict_warned && !vmlinux_warned &&
772 al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
773 RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
774 if (symbol_conf.vmlinux_name) {
775 ui__warning("The %s file can't be used.\n%s",
776 symbol_conf.vmlinux_name, msg);
777 } else {
778 ui__warning("A vmlinux file was not found.\n%s",
779 msg);
780 }
781
782 if (use_browser <= 0)
783 sleep(5);
784 vmlinux_warned = true;
785 }
786
787 return;
788 }
789
790 /* let's see, whether we need to install initial sym_filter_entry */
791 if (sym_filter_entry_sched) {
792 top.sym_filter_entry = sym_filter_entry_sched;
793 sym_filter_entry_sched = NULL;
794 if (parse_source(top.sym_filter_entry) < 0) {
795 struct symbol *sym = sym_entry__symbol(top.sym_filter_entry);
796
797 pr_err("Can't annotate %s", sym->name);
798 if (top.sym_filter_entry->map->dso->symtab_type == SYMTAB__KALLSYMS) {
799 pr_err(": No vmlinux file was found in the path:\n");
800 machine__fprintf_vmlinux_path(machine, stderr);
801 } else
802 pr_err(".\n");
803 exit(1);
804 }
805 }
806
807 syme = symbol__priv(al.sym);
808 if (!al.sym->ignore) {
809 struct perf_evsel *evsel;
810
811 evsel = perf_evlist__id2evsel(top.evlist, sample->id);
812 assert(evsel != NULL);
813 syme->count[evsel->idx]++;
814 record_precise_ip(syme, al.map, evsel->idx, ip);
815 pthread_mutex_lock(&top.active_symbols_lock);
816 if (list_empty(&syme->node) || !syme->node.next) {
817 static bool first = true;
818 __list_insert_active_sym(syme);
819 if (first) {
820 pthread_cond_broadcast(&top.active_symbols_cond);
821 first = false;
822 }
823 }
824 pthread_mutex_unlock(&top.active_symbols_lock);
825 }
826}
827
828static void perf_session__mmap_read_idx(struct perf_session *self, int idx)
829{
830 struct perf_sample sample;
831 union perf_event *event;
832 int ret;
833
834 while ((event = perf_evlist__mmap_read(top.evlist, idx)) != NULL) {
835 ret = perf_session__parse_sample(self, event, &sample);
836 if (ret) {
837 pr_err("Can't parse sample, err = %d\n", ret);
838 continue;
839 }
840
841 if (event->header.type == PERF_RECORD_SAMPLE)
842 perf_event__process_sample(event, &sample, self);
843 else
844 perf_event__process(event, &sample, self);
845 }
846}
847
848static void perf_session__mmap_read(struct perf_session *self)
849{
850 int i;
851
852 for (i = 0; i < top.evlist->nr_mmaps; i++)
853 perf_session__mmap_read_idx(self, i);
854}
855
856static void start_counters(struct perf_evlist *evlist)
857{
858 struct perf_evsel *counter;
859
860 list_for_each_entry(counter, &evlist->entries, node) {
861 struct perf_event_attr *attr = &counter->attr;
862
863 attr->sample_type = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
864
865 if (top.freq) {
866 attr->sample_type |= PERF_SAMPLE_PERIOD;
867 attr->freq = 1;
868 attr->sample_freq = top.freq;
869 }
870
871 if (evlist->nr_entries > 1) {
872 attr->sample_type |= PERF_SAMPLE_ID;
873 attr->read_format |= PERF_FORMAT_ID;
874 }
875
876 attr->mmap = 1;
877 attr->inherit = inherit;
878try_again:
879 if (perf_evsel__open(counter, top.evlist->cpus,
880 top.evlist->threads, group) < 0) {
881 int err = errno;
882
883 if (err == EPERM || err == EACCES) {
884 ui__warning_paranoid();
885 goto out_err;
886 }
887 /*
888 * If it's cycles then fall back to hrtimer
889 * based cpu-clock-tick sw counter, which
890 * is always available even if no PMU support:
891 */
892 if (attr->type == PERF_TYPE_HARDWARE &&
893 attr->config == PERF_COUNT_HW_CPU_CYCLES) {
894 if (verbose)
895 ui__warning("Cycles event not supported,\n"
896 "trying to fall back to cpu-clock-ticks\n");
897
898 attr->type = PERF_TYPE_SOFTWARE;
899 attr->config = PERF_COUNT_SW_CPU_CLOCK;
900 goto try_again;
901 }
902
903 if (err == ENOENT) {
904 ui__warning("The %s event is not supported.\n",
905 event_name(counter));
906 goto out_err;
907 }
908
909 ui__warning("The sys_perf_event_open() syscall "
910 "returned with %d (%s). /bin/dmesg "
911 "may provide additional information.\n"
912 "No CONFIG_PERF_EVENTS=y kernel support "
913 "configured?\n", err, strerror(err));
914 goto out_err;
915 }
916 }
917
918 if (perf_evlist__mmap(evlist, mmap_pages, false) < 0) {
919 ui__warning("Failed to mmap with %d (%s)\n",
920 errno, strerror(errno));
921 goto out_err;
922 }
923
924 return;
925
926out_err:
927 exit_browser(0);
928 exit(0);
929}
930
931static int __cmd_top(void)
932{
933 pthread_t thread;
934 int ret __used;
935 /*
936 * FIXME: perf_session__new should allow passing a O_MMAP, so that all this
937 * mmap reading, etc is encapsulated in it. Use O_WRONLY for now.
938 */
939 struct perf_session *session = perf_session__new(NULL, O_WRONLY, false, false, NULL);
940 if (session == NULL)
941 return -ENOMEM;
942
943 if (top.target_tid != -1)
944 perf_event__synthesize_thread_map(top.evlist->threads,
945 perf_event__process, session);
946 else
947 perf_event__synthesize_threads(perf_event__process, session);
948
949 start_counters(top.evlist);
950 session->evlist = top.evlist;
951 perf_session__update_sample_type(session);
952
953 /* Wait for a minimal set of events before starting the snapshot */
954 poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
955
956 perf_session__mmap_read(session);
957
958 if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
959 display_thread), session)) {
960 printf("Could not create display thread.\n");
961 exit(-1);
962 }
963
964 if (realtime_prio) {
965 struct sched_param param;
966
967 param.sched_priority = realtime_prio;
968 if (sched_setscheduler(0, SCHED_FIFO, ¶m)) {
969 printf("Could not set realtime priority.\n");
970 exit(-1);
971 }
972 }
973
974 while (1) {
975 u64 hits = top.samples;
976
977 perf_session__mmap_read(session);
978
979 if (hits == top.samples)
980 ret = poll(top.evlist->pollfd, top.evlist->nr_fds, 100);
981 }
982
983 return 0;
984}
985
986static const char * const top_usage[] = {
987 "perf top [<options>]",
988 NULL
989};
990
991static const struct option options[] = {
992 OPT_CALLBACK('e', "event", &top.evlist, "event",
993 "event selector. use 'perf list' to list available events",
994 parse_events_option),
995 OPT_INTEGER('c', "count", &default_interval,
996 "event period to sample"),
997 OPT_INTEGER('p', "pid", &top.target_pid,
998 "profile events on existing process id"),
999 OPT_INTEGER('t', "tid", &top.target_tid,
1000 "profile events on existing thread id"),
1001 OPT_BOOLEAN('a', "all-cpus", &system_wide,
1002 "system-wide collection from all CPUs"),
1003 OPT_STRING('C', "cpu", &top.cpu_list, "cpu",
1004 "list of cpus to monitor"),
1005 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1006 "file", "vmlinux pathname"),
1007 OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1008 "hide kernel symbols"),
1009 OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
1010 OPT_INTEGER('r', "realtime", &realtime_prio,
1011 "collect data with this RT SCHED_FIFO priority"),
1012 OPT_INTEGER('d', "delay", &top.delay_secs,
1013 "number of seconds to delay between refreshes"),
1014 OPT_BOOLEAN('D', "dump-symtab", &dump_symtab,
1015 "dump the symbol table used for profiling"),
1016 OPT_INTEGER('f', "count-filter", &top.count_filter,
1017 "only display functions with more events than this"),
1018 OPT_BOOLEAN('g', "group", &group,
1019 "put the counters into a counter group"),
1020 OPT_BOOLEAN('i', "inherit", &inherit,
1021 "child tasks inherit counters"),
1022 OPT_STRING('s', "sym-annotate", &sym_filter, "symbol name",
1023 "symbol to annotate"),
1024 OPT_BOOLEAN('z', "zero", &top.zero,
1025 "zero history across updates"),
1026 OPT_INTEGER('F', "freq", &top.freq,
1027 "profile at this frequency"),
1028 OPT_INTEGER('E', "entries", &top.print_entries,
1029 "display this many functions"),
1030 OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1031 "hide user symbols"),
1032 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
1033 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
1034 OPT_INCR('v', "verbose", &verbose,
1035 "be more verbose (show counter open errors, etc)"),
1036 OPT_END()
1037};
1038
1039int cmd_top(int argc, const char **argv, const char *prefix __used)
1040{
1041 struct perf_evsel *pos;
1042 int status = -ENOMEM;
1043
1044 top.evlist = perf_evlist__new(NULL, NULL);
1045 if (top.evlist == NULL)
1046 return -ENOMEM;
1047
1048 page_size = sysconf(_SC_PAGE_SIZE);
1049
1050 argc = parse_options(argc, argv, options, top_usage, 0);
1051 if (argc)
1052 usage_with_options(top_usage, options);
1053
1054 /*
1055 * XXX For now start disabled, only using TUI if explicitely asked for.
1056 * Change that when handle_keys equivalent gets written, live annotation
1057 * done, etc.
1058 */
1059 use_browser = 0;
1060
1061 if (use_stdio)
1062 use_browser = 0;
1063 else if (use_tui)
1064 use_browser = 1;
1065
1066 setup_browser(false);
1067
1068 /* CPU and PID are mutually exclusive */
1069 if (top.target_tid > 0 && top.cpu_list) {
1070 printf("WARNING: PID switch overriding CPU\n");
1071 sleep(1);
1072 top.cpu_list = NULL;
1073 }
1074
1075 if (top.target_pid != -1)
1076 top.target_tid = top.target_pid;
1077
1078 if (perf_evlist__create_maps(top.evlist, top.target_pid,
1079 top.target_tid, top.cpu_list) < 0)
1080 usage_with_options(top_usage, options);
1081
1082 if (!top.evlist->nr_entries &&
1083 perf_evlist__add_default(top.evlist) < 0) {
1084 pr_err("Not enough memory for event selector list\n");
1085 return -ENOMEM;
1086 }
1087
1088 if (top.delay_secs < 1)
1089 top.delay_secs = 1;
1090
1091 /*
1092 * User specified count overrides default frequency.
1093 */
1094 if (default_interval)
1095 top.freq = 0;
1096 else if (top.freq) {
1097 default_interval = top.freq;
1098 } else {
1099 fprintf(stderr, "frequency and count are zero, aborting\n");
1100 exit(EXIT_FAILURE);
1101 }
1102
1103 list_for_each_entry(pos, &top.evlist->entries, node) {
1104 if (perf_evsel__alloc_fd(pos, top.evlist->cpus->nr,
1105 top.evlist->threads->nr) < 0)
1106 goto out_free_fd;
1107 /*
1108 * Fill in the ones not specifically initialized via -c:
1109 */
1110 if (pos->attr.sample_period)
1111 continue;
1112
1113 pos->attr.sample_period = default_interval;
1114 }
1115
1116 if (perf_evlist__alloc_pollfd(top.evlist) < 0 ||
1117 perf_evlist__alloc_mmap(top.evlist) < 0)
1118 goto out_free_fd;
1119
1120 top.sym_evsel = list_entry(top.evlist->entries.next, struct perf_evsel, node);
1121
1122 symbol_conf.priv_size = (sizeof(struct sym_entry) + sizeof(struct annotation) +
1123 (top.evlist->nr_entries + 1) * sizeof(unsigned long));
1124
1125 symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1126 if (symbol__init() < 0)
1127 return -1;
1128
1129 get_term_dimensions(&winsize);
1130 if (top.print_entries == 0) {
1131 update_print_entries(&winsize);
1132 signal(SIGWINCH, sig_winch_handler);
1133 }
1134
1135 status = __cmd_top();
1136out_free_fd:
1137 perf_evlist__delete(top.evlist);
1138
1139 return status;
1140}
1/*
2 * builtin-top.c
3 *
4 * Builtin top command: Display a continuously updated profile of
5 * any workload, CPU or specific PID.
6 *
7 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
8 * 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
9 *
10 * Improvements and fixes by:
11 *
12 * Arjan van de Ven <arjan@linux.intel.com>
13 * Yanmin Zhang <yanmin.zhang@intel.com>
14 * Wu Fengguang <fengguang.wu@intel.com>
15 * Mike Galbraith <efault@gmx.de>
16 * Paul Mackerras <paulus@samba.org>
17 *
18 * Released under the GPL v2. (and only v2, not any later version)
19 */
20#include "builtin.h"
21
22#include "perf.h"
23
24#include "util/annotate.h"
25#include "util/config.h"
26#include "util/color.h"
27#include "util/drv_configs.h"
28#include "util/evlist.h"
29#include "util/evsel.h"
30#include "util/machine.h"
31#include "util/session.h"
32#include "util/symbol.h"
33#include "util/thread.h"
34#include "util/thread_map.h"
35#include "util/top.h"
36#include "util/util.h"
37#include <linux/rbtree.h>
38#include <subcmd/parse-options.h>
39#include "util/parse-events.h"
40#include "util/cpumap.h"
41#include "util/xyarray.h"
42#include "util/sort.h"
43#include "util/intlist.h"
44#include "util/parse-branch-options.h"
45#include "arch/common.h"
46
47#include "util/debug.h"
48
49#include <assert.h>
50#include <elf.h>
51#include <fcntl.h>
52
53#include <stdio.h>
54#include <termios.h>
55#include <unistd.h>
56#include <inttypes.h>
57
58#include <errno.h>
59#include <time.h>
60#include <sched.h>
61
62#include <sys/syscall.h>
63#include <sys/ioctl.h>
64#include <poll.h>
65#include <sys/prctl.h>
66#include <sys/wait.h>
67#include <sys/uio.h>
68#include <sys/utsname.h>
69#include <sys/mman.h>
70
71#include <linux/stringify.h>
72#include <linux/time64.h>
73#include <linux/types.h>
74
75static volatile int done;
76
77#define HEADER_LINE_NR 5
78
79static void perf_top__update_print_entries(struct perf_top *top)
80{
81 top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
82}
83
84static void perf_top__sig_winch(int sig __maybe_unused,
85 siginfo_t *info __maybe_unused, void *arg)
86{
87 struct perf_top *top = arg;
88
89 get_term_dimensions(&top->winsize);
90 perf_top__update_print_entries(top);
91}
92
93static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
94{
95 struct symbol *sym;
96 struct annotation *notes;
97 struct map *map;
98 int err = -1;
99
100 if (!he || !he->ms.sym)
101 return -1;
102
103 sym = he->ms.sym;
104 map = he->ms.map;
105
106 /*
107 * We can't annotate with just /proc/kallsyms
108 */
109 if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
110 !dso__is_kcore(map->dso)) {
111 pr_err("Can't annotate %s: No vmlinux file was found in the "
112 "path\n", sym->name);
113 sleep(1);
114 return -1;
115 }
116
117 notes = symbol__annotation(sym);
118 if (notes->src != NULL) {
119 pthread_mutex_lock(¬es->lock);
120 goto out_assign;
121 }
122
123 pthread_mutex_lock(¬es->lock);
124
125 if (symbol__alloc_hist(sym) < 0) {
126 pthread_mutex_unlock(¬es->lock);
127 pr_err("Not enough memory for annotating '%s' symbol!\n",
128 sym->name);
129 sleep(1);
130 return err;
131 }
132
133 err = symbol__disassemble(sym, map, NULL, 0);
134 if (err == 0) {
135out_assign:
136 top->sym_filter_entry = he;
137 } else {
138 char msg[BUFSIZ];
139 symbol__strerror_disassemble(sym, map, err, msg, sizeof(msg));
140 pr_err("Couldn't annotate %s: %s\n", sym->name, msg);
141 }
142
143 pthread_mutex_unlock(¬es->lock);
144 return err;
145}
146
147static void __zero_source_counters(struct hist_entry *he)
148{
149 struct symbol *sym = he->ms.sym;
150 symbol__annotate_zero_histograms(sym);
151}
152
153static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
154{
155 struct utsname uts;
156 int err = uname(&uts);
157
158 ui__warning("Out of bounds address found:\n\n"
159 "Addr: %" PRIx64 "\n"
160 "DSO: %s %c\n"
161 "Map: %" PRIx64 "-%" PRIx64 "\n"
162 "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
163 "Arch: %s\n"
164 "Kernel: %s\n"
165 "Tools: %s\n\n"
166 "Not all samples will be on the annotation output.\n\n"
167 "Please report to linux-kernel@vger.kernel.org\n",
168 ip, map->dso->long_name, dso__symtab_origin(map->dso),
169 map->start, map->end, sym->start, sym->end,
170 sym->binding == STB_GLOBAL ? 'g' :
171 sym->binding == STB_LOCAL ? 'l' : 'w', sym->name,
172 err ? "[unknown]" : uts.machine,
173 err ? "[unknown]" : uts.release, perf_version_string);
174 if (use_browser <= 0)
175 sleep(5);
176
177 map->erange_warned = true;
178}
179
180static void perf_top__record_precise_ip(struct perf_top *top,
181 struct hist_entry *he,
182 int counter, u64 ip)
183{
184 struct annotation *notes;
185 struct symbol *sym = he->ms.sym;
186 int err = 0;
187
188 if (sym == NULL || (use_browser == 0 &&
189 (top->sym_filter_entry == NULL ||
190 top->sym_filter_entry->ms.sym != sym)))
191 return;
192
193 notes = symbol__annotation(sym);
194
195 if (pthread_mutex_trylock(¬es->lock))
196 return;
197
198 err = hist_entry__inc_addr_samples(he, counter, ip);
199
200 pthread_mutex_unlock(¬es->lock);
201
202 if (unlikely(err)) {
203 /*
204 * This function is now called with he->hists->lock held.
205 * Release it before going to sleep.
206 */
207 pthread_mutex_unlock(&he->hists->lock);
208
209 if (err == -ERANGE && !he->ms.map->erange_warned)
210 ui__warn_map_erange(he->ms.map, sym, ip);
211 else if (err == -ENOMEM) {
212 pr_err("Not enough memory for annotating '%s' symbol!\n",
213 sym->name);
214 sleep(1);
215 }
216
217 pthread_mutex_lock(&he->hists->lock);
218 }
219}
220
221static void perf_top__show_details(struct perf_top *top)
222{
223 struct hist_entry *he = top->sym_filter_entry;
224 struct annotation *notes;
225 struct symbol *symbol;
226 int more;
227
228 if (!he)
229 return;
230
231 symbol = he->ms.sym;
232 notes = symbol__annotation(symbol);
233
234 pthread_mutex_lock(¬es->lock);
235
236 if (notes->src == NULL)
237 goto out_unlock;
238
239 printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
240 printf(" Events Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
241
242 more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
243 0, top->sym_pcnt_filter, top->print_entries, 4);
244
245 if (top->evlist->enabled) {
246 if (top->zero)
247 symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
248 else
249 symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
250 }
251 if (more != 0)
252 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
253out_unlock:
254 pthread_mutex_unlock(¬es->lock);
255}
256
257static void perf_top__print_sym_table(struct perf_top *top)
258{
259 char bf[160];
260 int printed = 0;
261 const int win_width = top->winsize.ws_col - 1;
262 struct perf_evsel *evsel = top->sym_evsel;
263 struct hists *hists = evsel__hists(evsel);
264
265 puts(CONSOLE_CLEAR);
266
267 perf_top__header_snprintf(top, bf, sizeof(bf));
268 printf("%s\n", bf);
269
270 perf_top__reset_sample_counters(top);
271
272 printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
273
274 if (hists->stats.nr_lost_warned !=
275 hists->stats.nr_events[PERF_RECORD_LOST]) {
276 hists->stats.nr_lost_warned =
277 hists->stats.nr_events[PERF_RECORD_LOST];
278 color_fprintf(stdout, PERF_COLOR_RED,
279 "WARNING: LOST %d chunks, Check IO/CPU overload",
280 hists->stats.nr_lost_warned);
281 ++printed;
282 }
283
284 if (top->sym_filter_entry) {
285 perf_top__show_details(top);
286 return;
287 }
288
289 if (top->evlist->enabled) {
290 if (top->zero) {
291 hists__delete_entries(hists);
292 } else {
293 hists__decay_entries(hists, top->hide_user_symbols,
294 top->hide_kernel_symbols);
295 }
296 }
297
298 hists__collapse_resort(hists, NULL);
299 perf_evsel__output_resort(evsel, NULL);
300
301 hists__output_recalc_col_len(hists, top->print_entries - printed);
302 putchar('\n');
303 hists__fprintf(hists, false, top->print_entries - printed, win_width,
304 top->min_percent, stdout, symbol_conf.use_callchain);
305}
306
307static void prompt_integer(int *target, const char *msg)
308{
309 char *buf = malloc(0), *p;
310 size_t dummy = 0;
311 int tmp;
312
313 fprintf(stdout, "\n%s: ", msg);
314 if (getline(&buf, &dummy, stdin) < 0)
315 return;
316
317 p = strchr(buf, '\n');
318 if (p)
319 *p = 0;
320
321 p = buf;
322 while(*p) {
323 if (!isdigit(*p))
324 goto out_free;
325 p++;
326 }
327 tmp = strtoul(buf, NULL, 10);
328 *target = tmp;
329out_free:
330 free(buf);
331}
332
333static void prompt_percent(int *target, const char *msg)
334{
335 int tmp = 0;
336
337 prompt_integer(&tmp, msg);
338 if (tmp >= 0 && tmp <= 100)
339 *target = tmp;
340}
341
342static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
343{
344 char *buf = malloc(0), *p;
345 struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
346 struct hists *hists = evsel__hists(top->sym_evsel);
347 struct rb_node *next;
348 size_t dummy = 0;
349
350 /* zero counters of active symbol */
351 if (syme) {
352 __zero_source_counters(syme);
353 top->sym_filter_entry = NULL;
354 }
355
356 fprintf(stdout, "\n%s: ", msg);
357 if (getline(&buf, &dummy, stdin) < 0)
358 goto out_free;
359
360 p = strchr(buf, '\n');
361 if (p)
362 *p = 0;
363
364 next = rb_first(&hists->entries);
365 while (next) {
366 n = rb_entry(next, struct hist_entry, rb_node);
367 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
368 found = n;
369 break;
370 }
371 next = rb_next(&n->rb_node);
372 }
373
374 if (!found) {
375 fprintf(stderr, "Sorry, %s is not active.\n", buf);
376 sleep(1);
377 } else
378 perf_top__parse_source(top, found);
379
380out_free:
381 free(buf);
382}
383
384static void perf_top__print_mapped_keys(struct perf_top *top)
385{
386 char *name = NULL;
387
388 if (top->sym_filter_entry) {
389 struct symbol *sym = top->sym_filter_entry->ms.sym;
390 name = sym->name;
391 }
392
393 fprintf(stdout, "\nMapped keys:\n");
394 fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top->delay_secs);
395 fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top->print_entries);
396
397 if (top->evlist->nr_entries > 1)
398 fprintf(stdout, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top->sym_evsel));
399
400 fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top->count_filter);
401
402 fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
403 fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL");
404 fprintf(stdout, "\t[S] stop annotation.\n");
405
406 fprintf(stdout,
407 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
408 top->hide_kernel_symbols ? "yes" : "no");
409 fprintf(stdout,
410 "\t[U] hide user symbols. \t(%s)\n",
411 top->hide_user_symbols ? "yes" : "no");
412 fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top->zero ? 1 : 0);
413 fprintf(stdout, "\t[qQ] quit.\n");
414}
415
416static int perf_top__key_mapped(struct perf_top *top, int c)
417{
418 switch (c) {
419 case 'd':
420 case 'e':
421 case 'f':
422 case 'z':
423 case 'q':
424 case 'Q':
425 case 'K':
426 case 'U':
427 case 'F':
428 case 's':
429 case 'S':
430 return 1;
431 case 'E':
432 return top->evlist->nr_entries > 1 ? 1 : 0;
433 default:
434 break;
435 }
436
437 return 0;
438}
439
440static bool perf_top__handle_keypress(struct perf_top *top, int c)
441{
442 bool ret = true;
443
444 if (!perf_top__key_mapped(top, c)) {
445 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
446 struct termios save;
447
448 perf_top__print_mapped_keys(top);
449 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
450 fflush(stdout);
451
452 set_term_quiet_input(&save);
453
454 poll(&stdin_poll, 1, -1);
455 c = getc(stdin);
456
457 tcsetattr(0, TCSAFLUSH, &save);
458 if (!perf_top__key_mapped(top, c))
459 return ret;
460 }
461
462 switch (c) {
463 case 'd':
464 prompt_integer(&top->delay_secs, "Enter display delay");
465 if (top->delay_secs < 1)
466 top->delay_secs = 1;
467 break;
468 case 'e':
469 prompt_integer(&top->print_entries, "Enter display entries (lines)");
470 if (top->print_entries == 0) {
471 struct sigaction act = {
472 .sa_sigaction = perf_top__sig_winch,
473 .sa_flags = SA_SIGINFO,
474 };
475 perf_top__sig_winch(SIGWINCH, NULL, top);
476 sigaction(SIGWINCH, &act, NULL);
477 } else {
478 signal(SIGWINCH, SIG_DFL);
479 }
480 break;
481 case 'E':
482 if (top->evlist->nr_entries > 1) {
483 /* Select 0 as the default event: */
484 int counter = 0;
485
486 fprintf(stderr, "\nAvailable events:");
487
488 evlist__for_each_entry(top->evlist, top->sym_evsel)
489 fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
490
491 prompt_integer(&counter, "Enter details event counter");
492
493 if (counter >= top->evlist->nr_entries) {
494 top->sym_evsel = perf_evlist__first(top->evlist);
495 fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
496 sleep(1);
497 break;
498 }
499 evlist__for_each_entry(top->evlist, top->sym_evsel)
500 if (top->sym_evsel->idx == counter)
501 break;
502 } else
503 top->sym_evsel = perf_evlist__first(top->evlist);
504 break;
505 case 'f':
506 prompt_integer(&top->count_filter, "Enter display event count filter");
507 break;
508 case 'F':
509 prompt_percent(&top->sym_pcnt_filter,
510 "Enter details display event filter (percent)");
511 break;
512 case 'K':
513 top->hide_kernel_symbols = !top->hide_kernel_symbols;
514 break;
515 case 'q':
516 case 'Q':
517 printf("exiting.\n");
518 if (top->dump_symtab)
519 perf_session__fprintf_dsos(top->session, stderr);
520 ret = false;
521 break;
522 case 's':
523 perf_top__prompt_symbol(top, "Enter details symbol");
524 break;
525 case 'S':
526 if (!top->sym_filter_entry)
527 break;
528 else {
529 struct hist_entry *syme = top->sym_filter_entry;
530
531 top->sym_filter_entry = NULL;
532 __zero_source_counters(syme);
533 }
534 break;
535 case 'U':
536 top->hide_user_symbols = !top->hide_user_symbols;
537 break;
538 case 'z':
539 top->zero = !top->zero;
540 break;
541 default:
542 break;
543 }
544
545 return ret;
546}
547
548static void perf_top__sort_new_samples(void *arg)
549{
550 struct perf_top *t = arg;
551 struct perf_evsel *evsel = t->sym_evsel;
552 struct hists *hists;
553
554 perf_top__reset_sample_counters(t);
555
556 if (t->evlist->selected != NULL)
557 t->sym_evsel = t->evlist->selected;
558
559 hists = evsel__hists(evsel);
560
561 if (t->evlist->enabled) {
562 if (t->zero) {
563 hists__delete_entries(hists);
564 } else {
565 hists__decay_entries(hists, t->hide_user_symbols,
566 t->hide_kernel_symbols);
567 }
568 }
569
570 hists__collapse_resort(hists, NULL);
571 perf_evsel__output_resort(evsel, NULL);
572}
573
574static void *display_thread_tui(void *arg)
575{
576 struct perf_evsel *pos;
577 struct perf_top *top = arg;
578 const char *help = "For a higher level overview, try: perf top --sort comm,dso";
579 struct hist_browser_timer hbt = {
580 .timer = perf_top__sort_new_samples,
581 .arg = top,
582 .refresh = top->delay_secs,
583 };
584
585 perf_top__sort_new_samples(top);
586
587 /*
588 * Initialize the uid_filter_str, in the future the TUI will allow
589 * Zooming in/out UIDs. For now juse use whatever the user passed
590 * via --uid.
591 */
592 evlist__for_each_entry(top->evlist, pos) {
593 struct hists *hists = evsel__hists(pos);
594 hists->uid_filter_str = top->record_opts.target.uid_str;
595 }
596
597 perf_evlist__tui_browse_hists(top->evlist, help, &hbt,
598 top->min_percent,
599 &top->session->header.env);
600
601 done = 1;
602 return NULL;
603}
604
605static void display_sig(int sig __maybe_unused)
606{
607 done = 1;
608}
609
610static void display_setup_sig(void)
611{
612 signal(SIGSEGV, sighandler_dump_stack);
613 signal(SIGFPE, sighandler_dump_stack);
614 signal(SIGINT, display_sig);
615 signal(SIGQUIT, display_sig);
616 signal(SIGTERM, display_sig);
617}
618
619static void *display_thread(void *arg)
620{
621 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
622 struct termios save;
623 struct perf_top *top = arg;
624 int delay_msecs, c;
625
626 display_setup_sig();
627 pthread__unblock_sigwinch();
628repeat:
629 delay_msecs = top->delay_secs * MSEC_PER_SEC;
630 set_term_quiet_input(&save);
631 /* trash return*/
632 getc(stdin);
633
634 while (!done) {
635 perf_top__print_sym_table(top);
636 /*
637 * Either timeout expired or we got an EINTR due to SIGWINCH,
638 * refresh screen in both cases.
639 */
640 switch (poll(&stdin_poll, 1, delay_msecs)) {
641 case 0:
642 continue;
643 case -1:
644 if (errno == EINTR)
645 continue;
646 /* Fall trhu */
647 default:
648 c = getc(stdin);
649 tcsetattr(0, TCSAFLUSH, &save);
650
651 if (perf_top__handle_keypress(top, c))
652 goto repeat;
653 done = 1;
654 }
655 }
656
657 tcsetattr(0, TCSAFLUSH, &save);
658 return NULL;
659}
660
661static int hist_iter__top_callback(struct hist_entry_iter *iter,
662 struct addr_location *al, bool single,
663 void *arg)
664{
665 struct perf_top *top = arg;
666 struct hist_entry *he = iter->he;
667 struct perf_evsel *evsel = iter->evsel;
668
669 if (perf_hpp_list.sym && single)
670 perf_top__record_precise_ip(top, he, evsel->idx, al->addr);
671
672 hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
673 !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY));
674 return 0;
675}
676
677static void perf_event__process_sample(struct perf_tool *tool,
678 const union perf_event *event,
679 struct perf_evsel *evsel,
680 struct perf_sample *sample,
681 struct machine *machine)
682{
683 struct perf_top *top = container_of(tool, struct perf_top, tool);
684 struct addr_location al;
685 int err;
686
687 if (!machine && perf_guest) {
688 static struct intlist *seen;
689
690 if (!seen)
691 seen = intlist__new(NULL);
692
693 if (!intlist__has_entry(seen, sample->pid)) {
694 pr_err("Can't find guest [%d]'s kernel information\n",
695 sample->pid);
696 intlist__add(seen, sample->pid);
697 }
698 return;
699 }
700
701 if (!machine) {
702 pr_err("%u unprocessable samples recorded.\r",
703 top->session->evlist->stats.nr_unprocessable_samples++);
704 return;
705 }
706
707 if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
708 top->exact_samples++;
709
710 if (machine__resolve(machine, &al, sample) < 0)
711 return;
712
713 if (!machine->kptr_restrict_warned &&
714 symbol_conf.kptr_restrict &&
715 al.cpumode == PERF_RECORD_MISC_KERNEL) {
716 ui__warning(
717"Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
718"Check /proc/sys/kernel/kptr_restrict.\n\n"
719"Kernel%s samples will not be resolved.\n",
720 al.map && !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
721 " modules" : "");
722 if (use_browser <= 0)
723 sleep(5);
724 machine->kptr_restrict_warned = true;
725 }
726
727 if (al.sym == NULL) {
728 const char *msg = "Kernel samples will not be resolved.\n";
729 /*
730 * As we do lazy loading of symtabs we only will know if the
731 * specified vmlinux file is invalid when we actually have a
732 * hit in kernel space and then try to load it. So if we get
733 * here and there are _no_ symbols in the DSO backing the
734 * kernel map, bail out.
735 *
736 * We may never get here, for instance, if we use -K/
737 * --hide-kernel-symbols, even if the user specifies an
738 * invalid --vmlinux ;-)
739 */
740 if (!machine->kptr_restrict_warned && !top->vmlinux_warned &&
741 al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
742 RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
743 if (symbol_conf.vmlinux_name) {
744 char serr[256];
745 dso__strerror_load(al.map->dso, serr, sizeof(serr));
746 ui__warning("The %s file can't be used: %s\n%s",
747 symbol_conf.vmlinux_name, serr, msg);
748 } else {
749 ui__warning("A vmlinux file was not found.\n%s",
750 msg);
751 }
752
753 if (use_browser <= 0)
754 sleep(5);
755 top->vmlinux_warned = true;
756 }
757 }
758
759 if (al.sym == NULL || !al.sym->idle) {
760 struct hists *hists = evsel__hists(evsel);
761 struct hist_entry_iter iter = {
762 .evsel = evsel,
763 .sample = sample,
764 .add_entry_cb = hist_iter__top_callback,
765 };
766
767 if (symbol_conf.cumulate_callchain)
768 iter.ops = &hist_iter_cumulative;
769 else
770 iter.ops = &hist_iter_normal;
771
772 pthread_mutex_lock(&hists->lock);
773
774 err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
775 if (err < 0)
776 pr_err("Problem incrementing symbol period, skipping event\n");
777
778 pthread_mutex_unlock(&hists->lock);
779 }
780
781 addr_location__put(&al);
782}
783
784static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
785{
786 struct perf_sample sample;
787 struct perf_evsel *evsel;
788 struct perf_session *session = top->session;
789 union perf_event *event;
790 struct machine *machine;
791 int ret;
792
793 while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
794 ret = perf_evlist__parse_sample(top->evlist, event, &sample);
795 if (ret) {
796 pr_err("Can't parse sample, err = %d\n", ret);
797 goto next_event;
798 }
799
800 evsel = perf_evlist__id2evsel(session->evlist, sample.id);
801 assert(evsel != NULL);
802
803 if (event->header.type == PERF_RECORD_SAMPLE)
804 ++top->samples;
805
806 switch (sample.cpumode) {
807 case PERF_RECORD_MISC_USER:
808 ++top->us_samples;
809 if (top->hide_user_symbols)
810 goto next_event;
811 machine = &session->machines.host;
812 break;
813 case PERF_RECORD_MISC_KERNEL:
814 ++top->kernel_samples;
815 if (top->hide_kernel_symbols)
816 goto next_event;
817 machine = &session->machines.host;
818 break;
819 case PERF_RECORD_MISC_GUEST_KERNEL:
820 ++top->guest_kernel_samples;
821 machine = perf_session__find_machine(session,
822 sample.pid);
823 break;
824 case PERF_RECORD_MISC_GUEST_USER:
825 ++top->guest_us_samples;
826 /*
827 * TODO: we don't process guest user from host side
828 * except simple counting.
829 */
830 goto next_event;
831 default:
832 if (event->header.type == PERF_RECORD_SAMPLE)
833 goto next_event;
834 machine = &session->machines.host;
835 break;
836 }
837
838
839 if (event->header.type == PERF_RECORD_SAMPLE) {
840 perf_event__process_sample(&top->tool, event, evsel,
841 &sample, machine);
842 } else if (event->header.type < PERF_RECORD_MAX) {
843 hists__inc_nr_events(evsel__hists(evsel), event->header.type);
844 machine__process_event(machine, event, &sample);
845 } else
846 ++session->evlist->stats.nr_unknown_events;
847next_event:
848 perf_evlist__mmap_consume(top->evlist, idx);
849 }
850}
851
852static void perf_top__mmap_read(struct perf_top *top)
853{
854 int i;
855
856 for (i = 0; i < top->evlist->nr_mmaps; i++)
857 perf_top__mmap_read_idx(top, i);
858}
859
860static int perf_top__start_counters(struct perf_top *top)
861{
862 char msg[512];
863 struct perf_evsel *counter;
864 struct perf_evlist *evlist = top->evlist;
865 struct record_opts *opts = &top->record_opts;
866
867 perf_evlist__config(evlist, opts, &callchain_param);
868
869 evlist__for_each_entry(evlist, counter) {
870try_again:
871 if (perf_evsel__open(counter, top->evlist->cpus,
872 top->evlist->threads) < 0) {
873 if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
874 if (verbose)
875 ui__warning("%s\n", msg);
876 goto try_again;
877 }
878
879 perf_evsel__open_strerror(counter, &opts->target,
880 errno, msg, sizeof(msg));
881 ui__error("%s\n", msg);
882 goto out_err;
883 }
884 }
885
886 if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
887 ui__error("Failed to mmap with %d (%s)\n",
888 errno, str_error_r(errno, msg, sizeof(msg)));
889 goto out_err;
890 }
891
892 return 0;
893
894out_err:
895 return -1;
896}
897
898static int callchain_param__setup_sample_type(struct callchain_param *callchain)
899{
900 if (!perf_hpp_list.sym) {
901 if (callchain->enabled) {
902 ui__error("Selected -g but \"sym\" not present in --sort/-s.");
903 return -EINVAL;
904 }
905 } else if (callchain->mode != CHAIN_NONE) {
906 if (callchain_register_param(callchain) < 0) {
907 ui__error("Can't register callchain params.\n");
908 return -EINVAL;
909 }
910 }
911
912 return 0;
913}
914
915static int __cmd_top(struct perf_top *top)
916{
917 char msg[512];
918 struct perf_evsel *pos;
919 struct perf_evsel_config_term *err_term;
920 struct perf_evlist *evlist = top->evlist;
921 struct record_opts *opts = &top->record_opts;
922 pthread_t thread;
923 int ret;
924
925 top->session = perf_session__new(NULL, false, NULL);
926 if (top->session == NULL)
927 return -1;
928
929 if (!objdump_path) {
930 ret = perf_env__lookup_objdump(&top->session->header.env);
931 if (ret)
932 goto out_delete;
933 }
934
935 ret = callchain_param__setup_sample_type(&callchain_param);
936 if (ret)
937 goto out_delete;
938
939 if (perf_session__register_idle_thread(top->session) < 0)
940 goto out_delete;
941
942 machine__synthesize_threads(&top->session->machines.host, &opts->target,
943 top->evlist->threads, false, opts->proc_map_timeout);
944
945 if (perf_hpp_list.socket) {
946 ret = perf_env__read_cpu_topology_map(&perf_env);
947 if (ret < 0)
948 goto out_err_cpu_topo;
949 }
950
951 ret = perf_top__start_counters(top);
952 if (ret)
953 goto out_delete;
954
955 ret = perf_evlist__apply_drv_configs(evlist, &pos, &err_term);
956 if (ret) {
957 error("failed to set config \"%s\" on event %s with %d (%s)\n",
958 err_term->val.drv_cfg, perf_evsel__name(pos), errno,
959 str_error_r(errno, msg, sizeof(msg)));
960 goto out_delete;
961 }
962
963 top->session->evlist = top->evlist;
964 perf_session__set_id_hdr_size(top->session);
965
966 /*
967 * When perf is starting the traced process, all the events (apart from
968 * group members) have enable_on_exec=1 set, so don't spoil it by
969 * prematurely enabling them.
970 *
971 * XXX 'top' still doesn't start workloads like record, trace, but should,
972 * so leave the check here.
973 */
974 if (!target__none(&opts->target))
975 perf_evlist__enable(top->evlist);
976
977 /* Wait for a minimal set of events before starting the snapshot */
978 perf_evlist__poll(top->evlist, 100);
979
980 perf_top__mmap_read(top);
981
982 ret = -1;
983 if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
984 display_thread), top)) {
985 ui__error("Could not create display thread.\n");
986 goto out_delete;
987 }
988
989 if (top->realtime_prio) {
990 struct sched_param param;
991
992 param.sched_priority = top->realtime_prio;
993 if (sched_setscheduler(0, SCHED_FIFO, ¶m)) {
994 ui__error("Could not set realtime priority.\n");
995 goto out_join;
996 }
997 }
998
999 while (!done) {
1000 u64 hits = top->samples;
1001
1002 perf_top__mmap_read(top);
1003
1004 if (hits == top->samples)
1005 ret = perf_evlist__poll(top->evlist, 100);
1006 }
1007
1008 ret = 0;
1009out_join:
1010 pthread_join(thread, NULL);
1011out_delete:
1012 perf_session__delete(top->session);
1013 top->session = NULL;
1014
1015 return ret;
1016
1017out_err_cpu_topo: {
1018 char errbuf[BUFSIZ];
1019 const char *err = str_error_r(-ret, errbuf, sizeof(errbuf));
1020
1021 ui__error("Could not read the CPU topology map: %s\n", err);
1022 goto out_delete;
1023}
1024}
1025
1026static int
1027callchain_opt(const struct option *opt, const char *arg, int unset)
1028{
1029 symbol_conf.use_callchain = true;
1030 return record_callchain_opt(opt, arg, unset);
1031}
1032
1033static int
1034parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1035{
1036 struct callchain_param *callchain = opt->value;
1037
1038 callchain->enabled = !unset;
1039 callchain->record_mode = CALLCHAIN_FP;
1040
1041 /*
1042 * --no-call-graph
1043 */
1044 if (unset) {
1045 symbol_conf.use_callchain = false;
1046 callchain->record_mode = CALLCHAIN_NONE;
1047 return 0;
1048 }
1049
1050 return parse_callchain_top_opt(arg);
1051}
1052
1053static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1054{
1055 if (!strcmp(var, "top.call-graph"))
1056 var = "call-graph.record-mode"; /* fall-through */
1057 if (!strcmp(var, "top.children")) {
1058 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
1059 return 0;
1060 }
1061
1062 return 0;
1063}
1064
1065static int
1066parse_percent_limit(const struct option *opt, const char *arg,
1067 int unset __maybe_unused)
1068{
1069 struct perf_top *top = opt->value;
1070
1071 top->min_percent = strtof(arg, NULL);
1072 return 0;
1073}
1074
1075const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1076 "\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1077
1078int cmd_top(int argc, const char **argv, const char *prefix __maybe_unused)
1079{
1080 char errbuf[BUFSIZ];
1081 struct perf_top top = {
1082 .count_filter = 5,
1083 .delay_secs = 2,
1084 .record_opts = {
1085 .mmap_pages = UINT_MAX,
1086 .user_freq = UINT_MAX,
1087 .user_interval = ULLONG_MAX,
1088 .freq = 4000, /* 4 KHz */
1089 .target = {
1090 .uses_mmap = true,
1091 },
1092 .proc_map_timeout = 500,
1093 },
1094 .max_stack = sysctl_perf_event_max_stack,
1095 .sym_pcnt_filter = 5,
1096 };
1097 struct record_opts *opts = &top.record_opts;
1098 struct target *target = &opts->target;
1099 const struct option options[] = {
1100 OPT_CALLBACK('e', "event", &top.evlist, "event",
1101 "event selector. use 'perf list' to list available events",
1102 parse_events_option),
1103 OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1104 OPT_STRING('p', "pid", &target->pid, "pid",
1105 "profile events on existing process id"),
1106 OPT_STRING('t', "tid", &target->tid, "tid",
1107 "profile events on existing thread id"),
1108 OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1109 "system-wide collection from all CPUs"),
1110 OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1111 "list of cpus to monitor"),
1112 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1113 "file", "vmlinux pathname"),
1114 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1115 "don't load vmlinux even if found"),
1116 OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1117 "hide kernel symbols"),
1118 OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1119 "number of mmap data pages",
1120 perf_evlist__parse_mmap_pages),
1121 OPT_INTEGER('r', "realtime", &top.realtime_prio,
1122 "collect data with this RT SCHED_FIFO priority"),
1123 OPT_INTEGER('d', "delay", &top.delay_secs,
1124 "number of seconds to delay between refreshes"),
1125 OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1126 "dump the symbol table used for profiling"),
1127 OPT_INTEGER('f', "count-filter", &top.count_filter,
1128 "only display functions with more events than this"),
1129 OPT_BOOLEAN(0, "group", &opts->group,
1130 "put the counters into a counter group"),
1131 OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1132 "child tasks do not inherit counters"),
1133 OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1134 "symbol to annotate"),
1135 OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1136 OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1137 OPT_INTEGER('E', "entries", &top.print_entries,
1138 "display this many functions"),
1139 OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1140 "hide user symbols"),
1141 OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1142 OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1143 OPT_INCR('v', "verbose", &verbose,
1144 "be more verbose (show counter open errors, etc)"),
1145 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1146 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1147 " Please refer the man page for the complete list."),
1148 OPT_STRING(0, "fields", &field_order, "key[,keys...]",
1149 "output field(s): overhead, period, sample plus all of sort keys"),
1150 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1151 "Show a column with the number of samples"),
1152 OPT_CALLBACK_NOOPT('g', NULL, &callchain_param,
1153 NULL, "enables call-graph recording and display",
1154 &callchain_opt),
1155 OPT_CALLBACK(0, "call-graph", &callchain_param,
1156 "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1157 top_callchain_help, &parse_callchain_opt),
1158 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1159 "Accumulate callchains of children and show total overhead as well"),
1160 OPT_INTEGER(0, "max-stack", &top.max_stack,
1161 "Set the maximum stack depth when parsing the callchain. "
1162 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1163 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1164 "ignore callees of these functions in call graphs",
1165 report_parse_ignore_callees_opt),
1166 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1167 "Show a column with the sum of periods"),
1168 OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1169 "only consider symbols in these dsos"),
1170 OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1171 "only consider symbols in these comms"),
1172 OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1173 "only consider these symbols"),
1174 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1175 "Interleave source code with assembly code (default)"),
1176 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1177 "Display raw encoding of assembly instructions (default)"),
1178 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1179 "Enable kernel symbol demangling"),
1180 OPT_STRING(0, "objdump", &objdump_path, "path",
1181 "objdump binary to use for disassembly and annotations"),
1182 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1183 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1184 OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1185 OPT_CALLBACK(0, "percent-limit", &top, "percent",
1186 "Don't show entries under that percent", parse_percent_limit),
1187 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1188 "How to display percentage of filtered entries", parse_filter_percentage),
1189 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1190 "width[,width...]",
1191 "don't try to adjust column width, use these fixed values"),
1192 OPT_UINTEGER(0, "proc-map-timeout", &opts->proc_map_timeout,
1193 "per thread proc mmap processing timeout in ms"),
1194 OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
1195 "branch any", "sample any taken branches",
1196 parse_branch_stack),
1197 OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
1198 "branch filter mask", "branch stack filter modes",
1199 parse_branch_stack),
1200 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1201 "Show raw trace event output (do not use print fmt or plugins)"),
1202 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1203 "Show entries in a hierarchy"),
1204 OPT_END()
1205 };
1206 const char * const top_usage[] = {
1207 "perf top [<options>]",
1208 NULL
1209 };
1210 int status = hists__init();
1211
1212 if (status < 0)
1213 return status;
1214
1215 top.evlist = perf_evlist__new();
1216 if (top.evlist == NULL)
1217 return -ENOMEM;
1218
1219 perf_config(perf_top_config, &top);
1220
1221 argc = parse_options(argc, argv, options, top_usage, 0);
1222 if (argc)
1223 usage_with_options(top_usage, options);
1224
1225 if (!top.evlist->nr_entries &&
1226 perf_evlist__add_default(top.evlist) < 0) {
1227 pr_err("Not enough memory for event selector list\n");
1228 goto out_delete_evlist;
1229 }
1230
1231 if (symbol_conf.report_hierarchy) {
1232 /* disable incompatible options */
1233 symbol_conf.event_group = false;
1234 symbol_conf.cumulate_callchain = false;
1235
1236 if (field_order) {
1237 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1238 parse_options_usage(top_usage, options, "fields", 0);
1239 parse_options_usage(NULL, options, "hierarchy", 0);
1240 goto out_delete_evlist;
1241 }
1242 }
1243
1244 sort__mode = SORT_MODE__TOP;
1245 /* display thread wants entries to be collapsed in a different tree */
1246 perf_hpp_list.need_collapse = 1;
1247
1248 if (top.use_stdio)
1249 use_browser = 0;
1250 else if (top.use_tui)
1251 use_browser = 1;
1252
1253 setup_browser(false);
1254
1255 if (setup_sorting(top.evlist) < 0) {
1256 if (sort_order)
1257 parse_options_usage(top_usage, options, "s", 1);
1258 if (field_order)
1259 parse_options_usage(sort_order ? NULL : top_usage,
1260 options, "fields", 0);
1261 goto out_delete_evlist;
1262 }
1263
1264 status = target__validate(target);
1265 if (status) {
1266 target__strerror(target, status, errbuf, BUFSIZ);
1267 ui__warning("%s\n", errbuf);
1268 }
1269
1270 status = target__parse_uid(target);
1271 if (status) {
1272 int saved_errno = errno;
1273
1274 target__strerror(target, status, errbuf, BUFSIZ);
1275 ui__error("%s\n", errbuf);
1276
1277 status = -saved_errno;
1278 goto out_delete_evlist;
1279 }
1280
1281 if (target__none(target))
1282 target->system_wide = true;
1283
1284 if (perf_evlist__create_maps(top.evlist, target) < 0) {
1285 ui__error("Couldn't create thread/CPU maps: %s\n",
1286 errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf)));
1287 goto out_delete_evlist;
1288 }
1289
1290 symbol_conf.nr_events = top.evlist->nr_entries;
1291
1292 if (top.delay_secs < 1)
1293 top.delay_secs = 1;
1294
1295 if (record_opts__config(opts)) {
1296 status = -EINVAL;
1297 goto out_delete_evlist;
1298 }
1299
1300 top.sym_evsel = perf_evlist__first(top.evlist);
1301
1302 if (!callchain_param.enabled) {
1303 symbol_conf.cumulate_callchain = false;
1304 perf_hpp__cancel_cumulate();
1305 }
1306
1307 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1308 callchain_param.order = ORDER_CALLER;
1309
1310 status = symbol__annotation_init();
1311 if (status < 0)
1312 goto out_delete_evlist;
1313
1314 symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1315 if (symbol__init(NULL) < 0)
1316 return -1;
1317
1318 sort__setup_elide(stdout);
1319
1320 get_term_dimensions(&top.winsize);
1321 if (top.print_entries == 0) {
1322 struct sigaction act = {
1323 .sa_sigaction = perf_top__sig_winch,
1324 .sa_flags = SA_SIGINFO,
1325 };
1326 perf_top__update_print_entries(&top);
1327 sigaction(SIGWINCH, &act, NULL);
1328 }
1329
1330 status = __cmd_top(&top);
1331
1332out_delete_evlist:
1333 perf_evlist__delete(top.evlist);
1334
1335 return status;
1336}