4 * Builtin top command: Display a continuously updated profile of
5 * any workload, CPU or specific PID.
7 * Copyright (C) 2008, Red Hat Inc, Ingo Molnar <mingo@redhat.com>
8 * 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com>
10 * Improvements and fixes by:
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>
18 * Released under the GPL v2. (and only v2, not any later version)
24 #include "util/annotate.h"
25 #include "util/cache.h"
26 #include "util/color.h"
27 #include "util/evlist.h"
28 #include "util/evsel.h"
29 #include "util/machine.h"
30 #include "util/session.h"
31 #include "util/symbol.h"
32 #include "util/thread.h"
33 #include "util/thread_map.h"
35 #include "util/util.h"
36 #include <linux/rbtree.h>
37 #include "util/parse-options.h"
38 #include "util/parse-events.h"
39 #include "util/cpumap.h"
40 #include "util/xyarray.h"
41 #include "util/sort.h"
42 #include "util/intlist.h"
43 #include "arch/common.h"
45 #include "util/debug.h"
60 #include <sys/syscall.h>
61 #include <sys/ioctl.h>
63 #include <sys/prctl.h>
66 #include <sys/utsname.h>
69 #include <linux/unistd.h>
70 #include <linux/types.h>
72 static volatile int done
;
74 #define HEADER_LINE_NR 5
76 static void perf_top__update_print_entries(struct perf_top
*top
)
78 top
->print_entries
= top
->winsize
.ws_row
- HEADER_LINE_NR
;
81 static void perf_top__sig_winch(int sig __maybe_unused
,
82 siginfo_t
*info __maybe_unused
, void *arg
)
84 struct perf_top
*top
= arg
;
86 get_term_dimensions(&top
->winsize
);
87 perf_top__update_print_entries(top
);
90 static int perf_top__parse_source(struct perf_top
*top
, struct hist_entry
*he
)
93 struct annotation
*notes
;
97 if (!he
|| !he
->ms
.sym
)
104 * We can't annotate with just /proc/kallsyms
106 if (map
->dso
->symtab_type
== DSO_BINARY_TYPE__KALLSYMS
&&
107 !dso__is_kcore(map
->dso
)) {
108 pr_err("Can't annotate %s: No vmlinux file was found in the "
109 "path\n", sym
->name
);
114 notes
= symbol__annotation(sym
);
115 if (notes
->src
!= NULL
) {
116 pthread_mutex_lock(¬es
->lock
);
120 pthread_mutex_lock(¬es
->lock
);
122 if (symbol__alloc_hist(sym
) < 0) {
123 pthread_mutex_unlock(¬es
->lock
);
124 pr_err("Not enough memory for annotating '%s' symbol!\n",
130 err
= symbol__annotate(sym
, map
, 0);
133 top
->sym_filter_entry
= he
;
136 pthread_mutex_unlock(¬es
->lock
);
140 static void __zero_source_counters(struct hist_entry
*he
)
142 struct symbol
*sym
= he
->ms
.sym
;
143 symbol__annotate_zero_histograms(sym
);
146 static void ui__warn_map_erange(struct map
*map
, struct symbol
*sym
, u64 ip
)
149 int err
= uname(&uts
);
151 ui__warning("Out of bounds address found:\n\n"
152 "Addr: %" PRIx64
"\n"
154 "Map: %" PRIx64
"-%" PRIx64
"\n"
155 "Symbol: %" PRIx64
"-%" PRIx64
" %c %s\n"
159 "Not all samples will be on the annotation output.\n\n"
160 "Please report to linux-kernel@vger.kernel.org\n",
161 ip
, map
->dso
->long_name
, dso__symtab_origin(map
->dso
),
162 map
->start
, map
->end
, sym
->start
, sym
->end
,
163 sym
->binding
== STB_GLOBAL
? 'g' :
164 sym
->binding
== STB_LOCAL
? 'l' : 'w', sym
->name
,
165 err
? "[unknown]" : uts
.machine
,
166 err
? "[unknown]" : uts
.release
, perf_version_string
);
167 if (use_browser
<= 0)
170 map
->erange_warned
= true;
173 static void perf_top__record_precise_ip(struct perf_top
*top
,
174 struct hist_entry
*he
,
177 struct annotation
*notes
;
181 if (he
== NULL
|| he
->ms
.sym
== NULL
||
182 ((top
->sym_filter_entry
== NULL
||
183 top
->sym_filter_entry
->ms
.sym
!= he
->ms
.sym
) && use_browser
!= 1))
187 notes
= symbol__annotation(sym
);
189 if (pthread_mutex_trylock(¬es
->lock
))
192 if (notes
->src
== NULL
&& symbol__alloc_hist(sym
) < 0) {
193 pthread_mutex_unlock(¬es
->lock
);
194 pr_err("Not enough memory for annotating '%s' symbol!\n",
200 ip
= he
->ms
.map
->map_ip(he
->ms
.map
, ip
);
201 err
= symbol__inc_addr_samples(sym
, he
->ms
.map
, counter
, ip
);
203 pthread_mutex_unlock(¬es
->lock
);
205 if (err
== -ERANGE
&& !he
->ms
.map
->erange_warned
)
206 ui__warn_map_erange(he
->ms
.map
, sym
, ip
);
209 static void perf_top__show_details(struct perf_top
*top
)
211 struct hist_entry
*he
= top
->sym_filter_entry
;
212 struct annotation
*notes
;
213 struct symbol
*symbol
;
220 notes
= symbol__annotation(symbol
);
222 pthread_mutex_lock(¬es
->lock
);
224 if (notes
->src
== NULL
)
227 printf("Showing %s for %s\n", perf_evsel__name(top
->sym_evsel
), symbol
->name
);
228 printf(" Events Pcnt (>=%d%%)\n", top
->sym_pcnt_filter
);
230 more
= symbol__annotate_printf(symbol
, he
->ms
.map
, top
->sym_evsel
,
231 0, top
->sym_pcnt_filter
, top
->print_entries
, 4);
233 symbol__annotate_zero_histogram(symbol
, top
->sym_evsel
->idx
);
235 symbol__annotate_decay_histogram(symbol
, top
->sym_evsel
->idx
);
237 printf("%d lines not displayed, maybe increase display entries [e]\n", more
);
239 pthread_mutex_unlock(¬es
->lock
);
242 static struct hist_entry
*perf_evsel__add_hist_entry(struct perf_evsel
*evsel
,
243 struct addr_location
*al
,
244 struct perf_sample
*sample
)
246 struct hist_entry
*he
;
248 pthread_mutex_lock(&evsel
->hists
.lock
);
249 he
= __hists__add_entry(&evsel
->hists
, al
, NULL
, sample
->period
,
250 sample
->weight
, sample
->transaction
);
251 pthread_mutex_unlock(&evsel
->hists
.lock
);
255 hists__inc_nr_events(&evsel
->hists
, PERF_RECORD_SAMPLE
);
259 static void perf_top__print_sym_table(struct perf_top
*top
)
263 const int win_width
= top
->winsize
.ws_col
- 1;
267 perf_top__header_snprintf(top
, bf
, sizeof(bf
));
270 perf_top__reset_sample_counters(top
);
272 printf("%-*.*s\n", win_width
, win_width
, graph_dotted_line
);
274 if (top
->sym_evsel
->hists
.stats
.nr_lost_warned
!=
275 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
]) {
276 top
->sym_evsel
->hists
.stats
.nr_lost_warned
=
277 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
];
278 color_fprintf(stdout
, PERF_COLOR_RED
,
279 "WARNING: LOST %d chunks, Check IO/CPU overload",
280 top
->sym_evsel
->hists
.stats
.nr_lost_warned
);
284 if (top
->sym_filter_entry
) {
285 perf_top__show_details(top
);
289 hists__collapse_resort(&top
->sym_evsel
->hists
, NULL
);
290 hists__output_resort(&top
->sym_evsel
->hists
);
291 hists__decay_entries(&top
->sym_evsel
->hists
,
292 top
->hide_user_symbols
,
293 top
->hide_kernel_symbols
);
294 hists__output_recalc_col_len(&top
->sym_evsel
->hists
,
295 top
->print_entries
- printed
);
297 hists__fprintf(&top
->sym_evsel
->hists
, false,
298 top
->print_entries
- printed
, win_width
,
299 top
->min_percent
, stdout
);
302 static void prompt_integer(int *target
, const char *msg
)
304 char *buf
= malloc(0), *p
;
308 fprintf(stdout
, "\n%s: ", msg
);
309 if (getline(&buf
, &dummy
, stdin
) < 0)
312 p
= strchr(buf
, '\n');
322 tmp
= strtoul(buf
, NULL
, 10);
328 static void prompt_percent(int *target
, const char *msg
)
332 prompt_integer(&tmp
, msg
);
333 if (tmp
>= 0 && tmp
<= 100)
337 static void perf_top__prompt_symbol(struct perf_top
*top
, const char *msg
)
339 char *buf
= malloc(0), *p
;
340 struct hist_entry
*syme
= top
->sym_filter_entry
, *n
, *found
= NULL
;
341 struct rb_node
*next
;
344 /* zero counters of active symbol */
346 __zero_source_counters(syme
);
347 top
->sym_filter_entry
= NULL
;
350 fprintf(stdout
, "\n%s: ", msg
);
351 if (getline(&buf
, &dummy
, stdin
) < 0)
354 p
= strchr(buf
, '\n');
358 next
= rb_first(&top
->sym_evsel
->hists
.entries
);
360 n
= rb_entry(next
, struct hist_entry
, rb_node
);
361 if (n
->ms
.sym
&& !strcmp(buf
, n
->ms
.sym
->name
)) {
365 next
= rb_next(&n
->rb_node
);
369 fprintf(stderr
, "Sorry, %s is not active.\n", buf
);
372 perf_top__parse_source(top
, found
);
378 static void perf_top__print_mapped_keys(struct perf_top
*top
)
382 if (top
->sym_filter_entry
) {
383 struct symbol
*sym
= top
->sym_filter_entry
->ms
.sym
;
387 fprintf(stdout
, "\nMapped keys:\n");
388 fprintf(stdout
, "\t[d] display refresh delay. \t(%d)\n", top
->delay_secs
);
389 fprintf(stdout
, "\t[e] display entries (lines). \t(%d)\n", top
->print_entries
);
391 if (top
->evlist
->nr_entries
> 1)
392 fprintf(stdout
, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top
->sym_evsel
));
394 fprintf(stdout
, "\t[f] profile display filter (count). \t(%d)\n", top
->count_filter
);
396 fprintf(stdout
, "\t[F] annotate display filter (percent). \t(%d%%)\n", top
->sym_pcnt_filter
);
397 fprintf(stdout
, "\t[s] annotate symbol. \t(%s)\n", name
?: "NULL");
398 fprintf(stdout
, "\t[S] stop annotation.\n");
401 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
402 top
->hide_kernel_symbols
? "yes" : "no");
404 "\t[U] hide user symbols. \t(%s)\n",
405 top
->hide_user_symbols
? "yes" : "no");
406 fprintf(stdout
, "\t[z] toggle sample zeroing. \t(%d)\n", top
->zero
? 1 : 0);
407 fprintf(stdout
, "\t[qQ] quit.\n");
410 static int perf_top__key_mapped(struct perf_top
*top
, int c
)
426 return top
->evlist
->nr_entries
> 1 ? 1 : 0;
434 static bool perf_top__handle_keypress(struct perf_top
*top
, int c
)
438 if (!perf_top__key_mapped(top
, c
)) {
439 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
440 struct termios tc
, save
;
442 perf_top__print_mapped_keys(top
);
443 fprintf(stdout
, "\nEnter selection, or unmapped key to continue: ");
448 tc
.c_lflag
&= ~(ICANON
| ECHO
);
451 tcsetattr(0, TCSANOW
, &tc
);
453 poll(&stdin_poll
, 1, -1);
456 tcsetattr(0, TCSAFLUSH
, &save
);
457 if (!perf_top__key_mapped(top
, c
))
463 prompt_integer(&top
->delay_secs
, "Enter display delay");
464 if (top
->delay_secs
< 1)
468 prompt_integer(&top
->print_entries
, "Enter display entries (lines)");
469 if (top
->print_entries
== 0) {
470 struct sigaction act
= {
471 .sa_sigaction
= perf_top__sig_winch
,
472 .sa_flags
= SA_SIGINFO
,
474 perf_top__sig_winch(SIGWINCH
, NULL
, top
);
475 sigaction(SIGWINCH
, &act
, NULL
);
477 signal(SIGWINCH
, SIG_DFL
);
481 if (top
->evlist
->nr_entries
> 1) {
482 /* Select 0 as the default event: */
485 fprintf(stderr
, "\nAvailable events:");
487 list_for_each_entry(top
->sym_evsel
, &top
->evlist
->entries
, node
)
488 fprintf(stderr
, "\n\t%d %s", top
->sym_evsel
->idx
, perf_evsel__name(top
->sym_evsel
));
490 prompt_integer(&counter
, "Enter details event counter");
492 if (counter
>= top
->evlist
->nr_entries
) {
493 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
494 fprintf(stderr
, "Sorry, no such event, using %s.\n", perf_evsel__name(top
->sym_evsel
));
498 list_for_each_entry(top
->sym_evsel
, &top
->evlist
->entries
, node
)
499 if (top
->sym_evsel
->idx
== counter
)
502 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
505 prompt_integer(&top
->count_filter
, "Enter display event count filter");
508 prompt_percent(&top
->sym_pcnt_filter
,
509 "Enter details display event filter (percent)");
512 top
->hide_kernel_symbols
= !top
->hide_kernel_symbols
;
516 printf("exiting.\n");
517 if (top
->dump_symtab
)
518 perf_session__fprintf_dsos(top
->session
, stderr
);
522 perf_top__prompt_symbol(top
, "Enter details symbol");
525 if (!top
->sym_filter_entry
)
528 struct hist_entry
*syme
= top
->sym_filter_entry
;
530 top
->sym_filter_entry
= NULL
;
531 __zero_source_counters(syme
);
535 top
->hide_user_symbols
= !top
->hide_user_symbols
;
538 top
->zero
= !top
->zero
;
547 static void perf_top__sort_new_samples(void *arg
)
549 struct perf_top
*t
= arg
;
550 perf_top__reset_sample_counters(t
);
552 if (t
->evlist
->selected
!= NULL
)
553 t
->sym_evsel
= t
->evlist
->selected
;
555 hists__collapse_resort(&t
->sym_evsel
->hists
, NULL
);
556 hists__output_resort(&t
->sym_evsel
->hists
);
557 hists__decay_entries(&t
->sym_evsel
->hists
,
558 t
->hide_user_symbols
,
559 t
->hide_kernel_symbols
);
562 static void *display_thread_tui(void *arg
)
564 struct perf_evsel
*pos
;
565 struct perf_top
*top
= arg
;
566 const char *help
= "For a higher level overview, try: perf top --sort comm,dso";
567 struct hist_browser_timer hbt
= {
568 .timer
= perf_top__sort_new_samples
,
570 .refresh
= top
->delay_secs
,
573 perf_top__sort_new_samples(top
);
576 * Initialize the uid_filter_str, in the future the TUI will allow
577 * Zooming in/out UIDs. For now juse use whatever the user passed
580 list_for_each_entry(pos
, &top
->evlist
->entries
, node
)
581 pos
->hists
.uid_filter_str
= top
->record_opts
.target
.uid_str
;
583 perf_evlist__tui_browse_hists(top
->evlist
, help
, &hbt
, top
->min_percent
,
584 &top
->session
->header
.env
);
590 static void *display_thread(void *arg
)
592 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
593 struct termios tc
, save
;
594 struct perf_top
*top
= arg
;
599 tc
.c_lflag
&= ~(ICANON
| ECHO
);
603 pthread__unblock_sigwinch();
605 delay_msecs
= top
->delay_secs
* 1000;
606 tcsetattr(0, TCSANOW
, &tc
);
611 perf_top__print_sym_table(top
);
613 * Either timeout expired or we got an EINTR due to SIGWINCH,
614 * refresh screen in both cases.
616 switch (poll(&stdin_poll
, 1, delay_msecs
)) {
625 tcsetattr(0, TCSAFLUSH
, &save
);
627 if (perf_top__handle_keypress(top
, c
))
636 /* Tag samples to be skipped. */
637 static const char *skip_symbols
[] = {
645 "mwait_idle_with_hints",
647 "ppc64_runlatch_off",
648 "pseries_dedicated_idle_sleep",
652 static int symbol_filter(struct map
*map __maybe_unused
, struct symbol
*sym
)
654 const char *name
= sym
->name
;
658 * ppc64 uses function descriptors and appends a '.' to the
659 * start of every instruction address. Remove it.
664 if (!strcmp(name
, "_text") ||
665 !strcmp(name
, "_etext") ||
666 !strcmp(name
, "_sinittext") ||
667 !strncmp("init_module", name
, 11) ||
668 !strncmp("cleanup_module", name
, 14) ||
669 strstr(name
, "_text_start") ||
670 strstr(name
, "_text_end"))
673 for (i
= 0; skip_symbols
[i
]; i
++) {
674 if (!strcmp(skip_symbols
[i
], name
)) {
683 static void perf_event__process_sample(struct perf_tool
*tool
,
684 const union perf_event
*event
,
685 struct perf_evsel
*evsel
,
686 struct perf_sample
*sample
,
687 struct machine
*machine
)
689 struct perf_top
*top
= container_of(tool
, struct perf_top
, tool
);
690 struct symbol
*parent
= NULL
;
692 struct addr_location al
;
695 if (!machine
&& perf_guest
) {
696 static struct intlist
*seen
;
699 seen
= intlist__new(NULL
);
701 if (!intlist__has_entry(seen
, sample
->pid
)) {
702 pr_err("Can't find guest [%d]'s kernel information\n",
704 intlist__add(seen
, sample
->pid
);
710 pr_err("%u unprocessable samples recorded.\r",
711 top
->session
->stats
.nr_unprocessable_samples
++);
715 if (event
->header
.misc
& PERF_RECORD_MISC_EXACT_IP
)
716 top
->exact_samples
++;
718 if (perf_event__preprocess_sample(event
, machine
, &al
, sample
) < 0 ||
722 if (!top
->kptr_restrict_warned
&&
723 symbol_conf
.kptr_restrict
&&
724 al
.cpumode
== PERF_RECORD_MISC_KERNEL
) {
726 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
727 "Check /proc/sys/kernel/kptr_restrict.\n\n"
728 "Kernel%s samples will not be resolved.\n",
729 !RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
]) ?
731 if (use_browser
<= 0)
733 top
->kptr_restrict_warned
= true;
736 if (al
.sym
== NULL
) {
737 const char *msg
= "Kernel samples will not be resolved.\n";
739 * As we do lazy loading of symtabs we only will know if the
740 * specified vmlinux file is invalid when we actually have a
741 * hit in kernel space and then try to load it. So if we get
742 * here and there are _no_ symbols in the DSO backing the
743 * kernel map, bail out.
745 * We may never get here, for instance, if we use -K/
746 * --hide-kernel-symbols, even if the user specifies an
747 * invalid --vmlinux ;-)
749 if (!top
->kptr_restrict_warned
&& !top
->vmlinux_warned
&&
750 al
.map
== machine
->vmlinux_maps
[MAP__FUNCTION
] &&
751 RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
])) {
752 if (symbol_conf
.vmlinux_name
) {
753 ui__warning("The %s file can't be used.\n%s",
754 symbol_conf
.vmlinux_name
, msg
);
756 ui__warning("A vmlinux file was not found.\n%s",
760 if (use_browser
<= 0)
762 top
->vmlinux_warned
= true;
766 if (al
.sym
== NULL
|| !al
.sym
->ignore
) {
767 struct hist_entry
*he
;
769 if ((sort__has_parent
|| symbol_conf
.use_callchain
) &&
771 err
= machine__resolve_callchain(machine
, evsel
,
779 he
= perf_evsel__add_hist_entry(evsel
, &al
, sample
);
781 pr_err("Problem incrementing symbol period, skipping event\n");
785 if (symbol_conf
.use_callchain
) {
786 err
= callchain_append(he
->callchain
, &callchain_cursor
,
793 perf_top__record_precise_ip(top
, he
, evsel
->idx
, ip
);
799 static void perf_top__mmap_read_idx(struct perf_top
*top
, int idx
)
801 struct perf_sample sample
;
802 struct perf_evsel
*evsel
;
803 struct perf_session
*session
= top
->session
;
804 union perf_event
*event
;
805 struct machine
*machine
;
809 while ((event
= perf_evlist__mmap_read(top
->evlist
, idx
)) != NULL
) {
810 ret
= perf_evlist__parse_sample(top
->evlist
, event
, &sample
);
812 pr_err("Can't parse sample, err = %d\n", ret
);
816 evsel
= perf_evlist__id2evsel(session
->evlist
, sample
.id
);
817 assert(evsel
!= NULL
);
819 origin
= event
->header
.misc
& PERF_RECORD_MISC_CPUMODE_MASK
;
821 if (event
->header
.type
== PERF_RECORD_SAMPLE
)
825 case PERF_RECORD_MISC_USER
:
827 if (top
->hide_user_symbols
)
829 machine
= &session
->machines
.host
;
831 case PERF_RECORD_MISC_KERNEL
:
832 ++top
->kernel_samples
;
833 if (top
->hide_kernel_symbols
)
835 machine
= &session
->machines
.host
;
837 case PERF_RECORD_MISC_GUEST_KERNEL
:
838 ++top
->guest_kernel_samples
;
839 machine
= perf_session__find_machine(session
,
842 case PERF_RECORD_MISC_GUEST_USER
:
843 ++top
->guest_us_samples
;
845 * TODO: we don't process guest user from host side
846 * except simple counting.
854 if (event
->header
.type
== PERF_RECORD_SAMPLE
) {
855 perf_event__process_sample(&top
->tool
, event
, evsel
,
857 } else if (event
->header
.type
< PERF_RECORD_MAX
) {
858 hists__inc_nr_events(&evsel
->hists
, event
->header
.type
);
859 machine__process_event(machine
, event
, &sample
);
861 ++session
->stats
.nr_unknown_events
;
863 perf_evlist__mmap_consume(top
->evlist
, idx
);
867 static void perf_top__mmap_read(struct perf_top
*top
)
871 for (i
= 0; i
< top
->evlist
->nr_mmaps
; i
++)
872 perf_top__mmap_read_idx(top
, i
);
875 static int perf_top__start_counters(struct perf_top
*top
)
878 struct perf_evsel
*counter
;
879 struct perf_evlist
*evlist
= top
->evlist
;
880 struct perf_record_opts
*opts
= &top
->record_opts
;
882 perf_evlist__config(evlist
, opts
);
884 list_for_each_entry(counter
, &evlist
->entries
, node
) {
886 if (perf_evsel__open(counter
, top
->evlist
->cpus
,
887 top
->evlist
->threads
) < 0) {
888 if (perf_evsel__fallback(counter
, errno
, msg
, sizeof(msg
))) {
890 ui__warning("%s\n", msg
);
894 perf_evsel__open_strerror(counter
, &opts
->target
,
895 errno
, msg
, sizeof(msg
));
896 ui__error("%s\n", msg
);
901 if (perf_evlist__mmap(evlist
, opts
->mmap_pages
, false) < 0) {
902 ui__error("Failed to mmap with %d (%s)\n",
903 errno
, strerror(errno
));
913 static int perf_top__setup_sample_type(struct perf_top
*top __maybe_unused
)
915 if (!sort__has_sym
) {
916 if (symbol_conf
.use_callchain
) {
917 ui__error("Selected -g but \"sym\" not present in --sort/-s.");
920 } else if (callchain_param
.mode
!= CHAIN_NONE
) {
921 if (callchain_register_param(&callchain_param
) < 0) {
922 ui__error("Can't register callchain params.\n");
930 static int __cmd_top(struct perf_top
*top
)
932 struct perf_record_opts
*opts
= &top
->record_opts
;
936 top
->session
= perf_session__new(NULL
, false, NULL
);
937 if (top
->session
== NULL
)
940 machines__set_symbol_filter(&top
->session
->machines
, symbol_filter
);
943 ret
= perf_session_env__lookup_objdump(&top
->session
->header
.env
);
948 ret
= perf_top__setup_sample_type(top
);
952 if (perf_target__has_task(&opts
->target
))
953 perf_event__synthesize_thread_map(&top
->tool
, top
->evlist
->threads
,
955 &top
->session
->machines
.host
);
957 perf_event__synthesize_threads(&top
->tool
, perf_event__process
,
958 &top
->session
->machines
.host
);
960 ret
= perf_top__start_counters(top
);
964 top
->session
->evlist
= top
->evlist
;
965 perf_session__set_id_hdr_size(top
->session
);
968 * When perf is starting the traced process, all the events (apart from
969 * group members) have enable_on_exec=1 set, so don't spoil it by
970 * prematurely enabling them.
972 * XXX 'top' still doesn't start workloads like record, trace, but should,
973 * so leave the check here.
975 if (!perf_target__none(&opts
->target
))
976 perf_evlist__enable(top
->evlist
);
978 /* Wait for a minimal set of events before starting the snapshot */
979 poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
981 perf_top__mmap_read(top
);
984 if (pthread_create(&thread
, NULL
, (use_browser
> 0 ? display_thread_tui
:
985 display_thread
), top
)) {
986 ui__error("Could not create display thread.\n");
990 if (top
->realtime_prio
) {
991 struct sched_param param
;
993 param
.sched_priority
= top
->realtime_prio
;
994 if (sched_setscheduler(0, SCHED_FIFO
, ¶m
)) {
995 ui__error("Could not set realtime priority.\n");
1001 u64 hits
= top
->samples
;
1003 perf_top__mmap_read(top
);
1005 if (hits
== top
->samples
)
1006 ret
= poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
1011 perf_session__delete(top
->session
);
1012 top
->session
= NULL
;
1018 callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1020 symbol_conf
.use_callchain
= true;
1021 return record_callchain_opt(opt
, arg
, unset
);
1025 parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1027 symbol_conf
.use_callchain
= true;
1028 return record_parse_callchain_opt(opt
, arg
, unset
);
1032 parse_percent_limit(const struct option
*opt
, const char *arg
,
1033 int unset __maybe_unused
)
1035 struct perf_top
*top
= opt
->value
;
1037 top
->min_percent
= strtof(arg
, NULL
);
1041 int cmd_top(int argc
, const char **argv
, const char *prefix __maybe_unused
)
1044 char errbuf
[BUFSIZ
];
1045 struct perf_top top
= {
1049 .mmap_pages
= UINT_MAX
,
1050 .user_freq
= UINT_MAX
,
1051 .user_interval
= ULLONG_MAX
,
1052 .freq
= 4000, /* 4 KHz */
1057 .max_stack
= PERF_MAX_STACK_DEPTH
,
1058 .sym_pcnt_filter
= 5,
1060 struct perf_record_opts
*opts
= &top
.record_opts
;
1061 struct perf_target
*target
= &opts
->target
;
1062 const struct option options
[] = {
1063 OPT_CALLBACK('e', "event", &top
.evlist
, "event",
1064 "event selector. use 'perf list' to list available events",
1065 parse_events_option
),
1066 OPT_U64('c', "count", &opts
->user_interval
, "event period to sample"),
1067 OPT_STRING('p', "pid", &target
->pid
, "pid",
1068 "profile events on existing process id"),
1069 OPT_STRING('t', "tid", &target
->tid
, "tid",
1070 "profile events on existing thread id"),
1071 OPT_BOOLEAN('a', "all-cpus", &target
->system_wide
,
1072 "system-wide collection from all CPUs"),
1073 OPT_STRING('C', "cpu", &target
->cpu_list
, "cpu",
1074 "list of cpus to monitor"),
1075 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1076 "file", "vmlinux pathname"),
1077 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1078 "don't load vmlinux even if found"),
1079 OPT_BOOLEAN('K', "hide_kernel_symbols", &top
.hide_kernel_symbols
,
1080 "hide kernel symbols"),
1081 OPT_CALLBACK('m', "mmap-pages", &opts
->mmap_pages
, "pages",
1082 "number of mmap data pages",
1083 perf_evlist__parse_mmap_pages
),
1084 OPT_INTEGER('r', "realtime", &top
.realtime_prio
,
1085 "collect data with this RT SCHED_FIFO priority"),
1086 OPT_INTEGER('d', "delay", &top
.delay_secs
,
1087 "number of seconds to delay between refreshes"),
1088 OPT_BOOLEAN('D', "dump-symtab", &top
.dump_symtab
,
1089 "dump the symbol table used for profiling"),
1090 OPT_INTEGER('f', "count-filter", &top
.count_filter
,
1091 "only display functions with more events than this"),
1092 OPT_BOOLEAN('g', "group", &opts
->group
,
1093 "put the counters into a counter group"),
1094 OPT_BOOLEAN('i', "no-inherit", &opts
->no_inherit
,
1095 "child tasks do not inherit counters"),
1096 OPT_STRING(0, "sym-annotate", &top
.sym_filter
, "symbol name",
1097 "symbol to annotate"),
1098 OPT_BOOLEAN('z', "zero", &top
.zero
, "zero history across updates"),
1099 OPT_UINTEGER('F', "freq", &opts
->user_freq
, "profile at this frequency"),
1100 OPT_INTEGER('E', "entries", &top
.print_entries
,
1101 "display this many functions"),
1102 OPT_BOOLEAN('U', "hide_user_symbols", &top
.hide_user_symbols
,
1103 "hide user symbols"),
1104 OPT_BOOLEAN(0, "tui", &top
.use_tui
, "Use the TUI interface"),
1105 OPT_BOOLEAN(0, "stdio", &top
.use_stdio
, "Use the stdio interface"),
1106 OPT_INCR('v', "verbose", &verbose
,
1107 "be more verbose (show counter open errors, etc)"),
1108 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1109 "sort by key(s): pid, comm, dso, symbol, parent, weight, local_weight,"
1110 " abort, in_tx, transaction"),
1111 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1112 "Show a column with the number of samples"),
1113 OPT_CALLBACK_NOOPT('G', NULL
, &top
.record_opts
,
1114 NULL
, "enables call-graph recording",
1116 OPT_CALLBACK(0, "call-graph", &top
.record_opts
,
1117 "mode[,dump_size]", record_callchain_help
,
1118 &parse_callchain_opt
),
1119 OPT_INTEGER(0, "max-stack", &top
.max_stack
,
1120 "Set the maximum stack depth when parsing the callchain. "
1121 "Default: " __stringify(PERF_MAX_STACK_DEPTH
)),
1122 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1123 "ignore callees of these functions in call graphs",
1124 report_parse_ignore_callees_opt
),
1125 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1126 "Show a column with the sum of periods"),
1127 OPT_STRING(0, "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1128 "only consider symbols in these dsos"),
1129 OPT_STRING(0, "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1130 "only consider symbols in these comms"),
1131 OPT_STRING(0, "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1132 "only consider these symbols"),
1133 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
1134 "Interleave source code with assembly code (default)"),
1135 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
1136 "Display raw encoding of assembly instructions (default)"),
1137 OPT_STRING(0, "objdump", &objdump_path
, "path",
1138 "objdump binary to use for disassembly and annotations"),
1139 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
1140 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1141 OPT_STRING('u', "uid", &target
->uid_str
, "user", "user to profile"),
1142 OPT_CALLBACK(0, "percent-limit", &top
, "percent",
1143 "Don't show entries under that percent", parse_percent_limit
),
1146 const char * const top_usage
[] = {
1147 "perf top [<options>]",
1151 top
.evlist
= perf_evlist__new();
1152 if (top
.evlist
== NULL
)
1155 argc
= parse_options(argc
, argv
, options
, top_usage
, 0);
1157 usage_with_options(top_usage
, options
);
1159 if (sort_order
== default_sort_order
)
1160 sort_order
= "dso,symbol";
1162 if (setup_sorting() < 0)
1163 usage_with_options(top_usage
, options
);
1165 /* display thread wants entries to be collapsed in a different tree */
1166 sort__need_collapse
= 1;
1170 else if (top
.use_tui
)
1173 setup_browser(false);
1175 status
= perf_target__validate(target
);
1177 perf_target__strerror(target
, status
, errbuf
, BUFSIZ
);
1178 ui__warning("%s", errbuf
);
1181 status
= perf_target__parse_uid(target
);
1183 int saved_errno
= errno
;
1185 perf_target__strerror(target
, status
, errbuf
, BUFSIZ
);
1186 ui__error("%s", errbuf
);
1188 status
= -saved_errno
;
1189 goto out_delete_evlist
;
1192 if (perf_target__none(target
))
1193 target
->system_wide
= true;
1195 if (perf_evlist__create_maps(top
.evlist
, target
) < 0)
1196 usage_with_options(top_usage
, options
);
1198 if (!top
.evlist
->nr_entries
&&
1199 perf_evlist__add_default(top
.evlist
) < 0) {
1200 ui__error("Not enough memory for event selector list\n");
1201 goto out_delete_maps
;
1204 symbol_conf
.nr_events
= top
.evlist
->nr_entries
;
1206 if (top
.delay_secs
< 1)
1209 if (opts
->user_interval
!= ULLONG_MAX
)
1210 opts
->default_interval
= opts
->user_interval
;
1211 if (opts
->user_freq
!= UINT_MAX
)
1212 opts
->freq
= opts
->user_freq
;
1215 * User specified count overrides default frequency.
1217 if (opts
->default_interval
)
1219 else if (opts
->freq
) {
1220 opts
->default_interval
= opts
->freq
;
1222 ui__error("frequency and count are zero, aborting\n");
1224 goto out_delete_maps
;
1227 top
.sym_evsel
= perf_evlist__first(top
.evlist
);
1229 symbol_conf
.priv_size
= sizeof(struct annotation
);
1231 symbol_conf
.try_vmlinux_path
= (symbol_conf
.vmlinux_name
== NULL
);
1232 if (symbol__init() < 0)
1235 sort__setup_elide(stdout
);
1237 get_term_dimensions(&top
.winsize
);
1238 if (top
.print_entries
== 0) {
1239 struct sigaction act
= {
1240 .sa_sigaction
= perf_top__sig_winch
,
1241 .sa_flags
= SA_SIGINFO
,
1243 perf_top__update_print_entries(&top
);
1244 sigaction(SIGWINCH
, &act
, NULL
);
1247 status
= __cmd_top(&top
);
1250 perf_evlist__delete_maps(top
.evlist
);
1252 perf_evlist__delete(top
.evlist
);