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 pr_err("Can't annotate %s: No vmlinux file was found in the "
108 "path\n", sym
->name
);
113 notes
= symbol__annotation(sym
);
114 if (notes
->src
!= NULL
) {
115 pthread_mutex_lock(¬es
->lock
);
119 pthread_mutex_lock(¬es
->lock
);
121 if (symbol__alloc_hist(sym
) < 0) {
122 pthread_mutex_unlock(¬es
->lock
);
123 pr_err("Not enough memory for annotating '%s' symbol!\n",
129 err
= symbol__annotate(sym
, map
, 0);
132 top
->sym_filter_entry
= he
;
135 pthread_mutex_unlock(¬es
->lock
);
139 static void __zero_source_counters(struct hist_entry
*he
)
141 struct symbol
*sym
= he
->ms
.sym
;
142 symbol__annotate_zero_histograms(sym
);
145 static void ui__warn_map_erange(struct map
*map
, struct symbol
*sym
, u64 ip
)
148 int err
= uname(&uts
);
150 ui__warning("Out of bounds address found:\n\n"
151 "Addr: %" PRIx64
"\n"
153 "Map: %" PRIx64
"-%" PRIx64
"\n"
154 "Symbol: %" PRIx64
"-%" PRIx64
" %c %s\n"
158 "Not all samples will be on the annotation output.\n\n"
159 "Please report to linux-kernel@vger.kernel.org\n",
160 ip
, map
->dso
->long_name
, dso__symtab_origin(map
->dso
),
161 map
->start
, map
->end
, sym
->start
, sym
->end
,
162 sym
->binding
== STB_GLOBAL
? 'g' :
163 sym
->binding
== STB_LOCAL
? 'l' : 'w', sym
->name
,
164 err
? "[unknown]" : uts
.machine
,
165 err
? "[unknown]" : uts
.release
, perf_version_string
);
166 if (use_browser
<= 0)
169 map
->erange_warned
= true;
172 static void perf_top__record_precise_ip(struct perf_top
*top
,
173 struct hist_entry
*he
,
176 struct annotation
*notes
;
180 if (he
== NULL
|| he
->ms
.sym
== NULL
||
181 ((top
->sym_filter_entry
== NULL
||
182 top
->sym_filter_entry
->ms
.sym
!= he
->ms
.sym
) && use_browser
!= 1))
186 notes
= symbol__annotation(sym
);
188 if (pthread_mutex_trylock(¬es
->lock
))
191 if (notes
->src
== NULL
&& symbol__alloc_hist(sym
) < 0) {
192 pthread_mutex_unlock(¬es
->lock
);
193 pr_err("Not enough memory for annotating '%s' symbol!\n",
199 ip
= he
->ms
.map
->map_ip(he
->ms
.map
, ip
);
200 err
= symbol__inc_addr_samples(sym
, he
->ms
.map
, counter
, ip
);
202 pthread_mutex_unlock(¬es
->lock
);
204 if (err
== -ERANGE
&& !he
->ms
.map
->erange_warned
)
205 ui__warn_map_erange(he
->ms
.map
, sym
, ip
);
208 static void perf_top__show_details(struct perf_top
*top
)
210 struct hist_entry
*he
= top
->sym_filter_entry
;
211 struct annotation
*notes
;
212 struct symbol
*symbol
;
219 notes
= symbol__annotation(symbol
);
221 pthread_mutex_lock(¬es
->lock
);
223 if (notes
->src
== NULL
)
226 printf("Showing %s for %s\n", perf_evsel__name(top
->sym_evsel
), symbol
->name
);
227 printf(" Events Pcnt (>=%d%%)\n", top
->sym_pcnt_filter
);
229 more
= symbol__annotate_printf(symbol
, he
->ms
.map
, top
->sym_evsel
,
230 0, top
->sym_pcnt_filter
, top
->print_entries
, 4);
232 symbol__annotate_zero_histogram(symbol
, top
->sym_evsel
->idx
);
234 symbol__annotate_decay_histogram(symbol
, top
->sym_evsel
->idx
);
236 printf("%d lines not displayed, maybe increase display entries [e]\n", more
);
238 pthread_mutex_unlock(¬es
->lock
);
241 static const char CONSOLE_CLEAR
[] = "\e[H\e[2J";
243 static struct hist_entry
*perf_evsel__add_hist_entry(struct perf_evsel
*evsel
,
244 struct addr_location
*al
,
245 struct perf_sample
*sample
)
247 struct hist_entry
*he
;
249 pthread_mutex_lock(&evsel
->hists
.lock
);
250 he
= __hists__add_entry(&evsel
->hists
, al
, NULL
, sample
->period
,
252 pthread_mutex_unlock(&evsel
->hists
.lock
);
257 hists__inc_nr_events(&evsel
->hists
, PERF_RECORD_SAMPLE
);
261 static void perf_top__print_sym_table(struct perf_top
*top
)
265 const int win_width
= top
->winsize
.ws_col
- 1;
269 perf_top__header_snprintf(top
, bf
, sizeof(bf
));
272 perf_top__reset_sample_counters(top
);
274 printf("%-*.*s\n", win_width
, win_width
, graph_dotted_line
);
276 if (top
->sym_evsel
->hists
.stats
.nr_lost_warned
!=
277 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
]) {
278 top
->sym_evsel
->hists
.stats
.nr_lost_warned
=
279 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
];
280 color_fprintf(stdout
, PERF_COLOR_RED
,
281 "WARNING: LOST %d chunks, Check IO/CPU overload",
282 top
->sym_evsel
->hists
.stats
.nr_lost_warned
);
286 if (top
->sym_filter_entry
) {
287 perf_top__show_details(top
);
291 hists__collapse_resort(&top
->sym_evsel
->hists
);
292 hists__output_resort(&top
->sym_evsel
->hists
);
293 hists__decay_entries(&top
->sym_evsel
->hists
,
294 top
->hide_user_symbols
,
295 top
->hide_kernel_symbols
);
296 hists__output_recalc_col_len(&top
->sym_evsel
->hists
,
297 top
->print_entries
- printed
);
299 hists__fprintf(&top
->sym_evsel
->hists
, false,
300 top
->print_entries
- printed
, win_width
,
301 top
->min_percent
, stdout
);
304 static void prompt_integer(int *target
, const char *msg
)
306 char *buf
= malloc(0), *p
;
310 fprintf(stdout
, "\n%s: ", msg
);
311 if (getline(&buf
, &dummy
, stdin
) < 0)
314 p
= strchr(buf
, '\n');
324 tmp
= strtoul(buf
, NULL
, 10);
330 static void prompt_percent(int *target
, const char *msg
)
334 prompt_integer(&tmp
, msg
);
335 if (tmp
>= 0 && tmp
<= 100)
339 static void perf_top__prompt_symbol(struct perf_top
*top
, const char *msg
)
341 char *buf
= malloc(0), *p
;
342 struct hist_entry
*syme
= top
->sym_filter_entry
, *n
, *found
= NULL
;
343 struct rb_node
*next
;
346 /* zero counters of active symbol */
348 __zero_source_counters(syme
);
349 top
->sym_filter_entry
= NULL
;
352 fprintf(stdout
, "\n%s: ", msg
);
353 if (getline(&buf
, &dummy
, stdin
) < 0)
356 p
= strchr(buf
, '\n');
360 next
= rb_first(&top
->sym_evsel
->hists
.entries
);
362 n
= rb_entry(next
, struct hist_entry
, rb_node
);
363 if (n
->ms
.sym
&& !strcmp(buf
, n
->ms
.sym
->name
)) {
367 next
= rb_next(&n
->rb_node
);
371 fprintf(stderr
, "Sorry, %s is not active.\n", buf
);
374 perf_top__parse_source(top
, found
);
380 static void perf_top__print_mapped_keys(struct perf_top
*top
)
384 if (top
->sym_filter_entry
) {
385 struct symbol
*sym
= top
->sym_filter_entry
->ms
.sym
;
389 fprintf(stdout
, "\nMapped keys:\n");
390 fprintf(stdout
, "\t[d] display refresh delay. \t(%d)\n", top
->delay_secs
);
391 fprintf(stdout
, "\t[e] display entries (lines). \t(%d)\n", top
->print_entries
);
393 if (top
->evlist
->nr_entries
> 1)
394 fprintf(stdout
, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top
->sym_evsel
));
396 fprintf(stdout
, "\t[f] profile display filter (count). \t(%d)\n", top
->count_filter
);
398 fprintf(stdout
, "\t[F] annotate display filter (percent). \t(%d%%)\n", top
->sym_pcnt_filter
);
399 fprintf(stdout
, "\t[s] annotate symbol. \t(%s)\n", name
?: "NULL");
400 fprintf(stdout
, "\t[S] stop annotation.\n");
403 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
404 top
->hide_kernel_symbols
? "yes" : "no");
406 "\t[U] hide user symbols. \t(%s)\n",
407 top
->hide_user_symbols
? "yes" : "no");
408 fprintf(stdout
, "\t[z] toggle sample zeroing. \t(%d)\n", top
->zero
? 1 : 0);
409 fprintf(stdout
, "\t[qQ] quit.\n");
412 static int perf_top__key_mapped(struct perf_top
*top
, int c
)
428 return top
->evlist
->nr_entries
> 1 ? 1 : 0;
436 static bool perf_top__handle_keypress(struct perf_top
*top
, int c
)
440 if (!perf_top__key_mapped(top
, c
)) {
441 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
442 struct termios tc
, save
;
444 perf_top__print_mapped_keys(top
);
445 fprintf(stdout
, "\nEnter selection, or unmapped key to continue: ");
450 tc
.c_lflag
&= ~(ICANON
| ECHO
);
453 tcsetattr(0, TCSANOW
, &tc
);
455 poll(&stdin_poll
, 1, -1);
458 tcsetattr(0, TCSAFLUSH
, &save
);
459 if (!perf_top__key_mapped(top
, c
))
465 prompt_integer(&top
->delay_secs
, "Enter display delay");
466 if (top
->delay_secs
< 1)
470 prompt_integer(&top
->print_entries
, "Enter display entries (lines)");
471 if (top
->print_entries
== 0) {
472 struct sigaction act
= {
473 .sa_sigaction
= perf_top__sig_winch
,
474 .sa_flags
= SA_SIGINFO
,
476 perf_top__sig_winch(SIGWINCH
, NULL
, top
);
477 sigaction(SIGWINCH
, &act
, NULL
);
479 signal(SIGWINCH
, SIG_DFL
);
483 if (top
->evlist
->nr_entries
> 1) {
484 /* Select 0 as the default event: */
487 fprintf(stderr
, "\nAvailable events:");
489 list_for_each_entry(top
->sym_evsel
, &top
->evlist
->entries
, node
)
490 fprintf(stderr
, "\n\t%d %s", top
->sym_evsel
->idx
, perf_evsel__name(top
->sym_evsel
));
492 prompt_integer(&counter
, "Enter details event counter");
494 if (counter
>= top
->evlist
->nr_entries
) {
495 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
496 fprintf(stderr
, "Sorry, no such event, using %s.\n", perf_evsel__name(top
->sym_evsel
));
500 list_for_each_entry(top
->sym_evsel
, &top
->evlist
->entries
, node
)
501 if (top
->sym_evsel
->idx
== counter
)
504 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
507 prompt_integer(&top
->count_filter
, "Enter display event count filter");
510 prompt_percent(&top
->sym_pcnt_filter
,
511 "Enter details display event filter (percent)");
514 top
->hide_kernel_symbols
= !top
->hide_kernel_symbols
;
518 printf("exiting.\n");
519 if (top
->dump_symtab
)
520 perf_session__fprintf_dsos(top
->session
, stderr
);
524 perf_top__prompt_symbol(top
, "Enter details symbol");
527 if (!top
->sym_filter_entry
)
530 struct hist_entry
*syme
= top
->sym_filter_entry
;
532 top
->sym_filter_entry
= NULL
;
533 __zero_source_counters(syme
);
537 top
->hide_user_symbols
= !top
->hide_user_symbols
;
540 top
->zero
= !top
->zero
;
549 static void perf_top__sort_new_samples(void *arg
)
551 struct perf_top
*t
= arg
;
552 perf_top__reset_sample_counters(t
);
554 if (t
->evlist
->selected
!= NULL
)
555 t
->sym_evsel
= t
->evlist
->selected
;
557 hists__collapse_resort(&t
->sym_evsel
->hists
);
558 hists__output_resort(&t
->sym_evsel
->hists
);
559 hists__decay_entries(&t
->sym_evsel
->hists
,
560 t
->hide_user_symbols
,
561 t
->hide_kernel_symbols
);
564 static void *display_thread_tui(void *arg
)
566 struct perf_evsel
*pos
;
567 struct perf_top
*top
= arg
;
568 const char *help
= "For a higher level overview, try: perf top --sort comm,dso";
569 struct hist_browser_timer hbt
= {
570 .timer
= perf_top__sort_new_samples
,
572 .refresh
= top
->delay_secs
,
575 perf_top__sort_new_samples(top
);
578 * Initialize the uid_filter_str, in the future the TUI will allow
579 * Zooming in/out UIDs. For now juse use whatever the user passed
582 list_for_each_entry(pos
, &top
->evlist
->entries
, node
)
583 pos
->hists
.uid_filter_str
= top
->record_opts
.target
.uid_str
;
585 perf_evlist__tui_browse_hists(top
->evlist
, help
, &hbt
, top
->min_percent
,
586 &top
->session
->header
.env
);
592 static void *display_thread(void *arg
)
594 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
595 struct termios tc
, save
;
596 struct perf_top
*top
= arg
;
601 tc
.c_lflag
&= ~(ICANON
| ECHO
);
605 pthread__unblock_sigwinch();
607 delay_msecs
= top
->delay_secs
* 1000;
608 tcsetattr(0, TCSANOW
, &tc
);
613 perf_top__print_sym_table(top
);
615 * Either timeout expired or we got an EINTR due to SIGWINCH,
616 * refresh screen in both cases.
618 switch (poll(&stdin_poll
, 1, delay_msecs
)) {
627 tcsetattr(0, TCSAFLUSH
, &save
);
629 if (perf_top__handle_keypress(top
, c
))
638 /* Tag samples to be skipped. */
639 static const char *skip_symbols
[] = {
647 "mwait_idle_with_hints",
649 "ppc64_runlatch_off",
650 "pseries_dedicated_idle_sleep",
654 static int symbol_filter(struct map
*map __maybe_unused
, struct symbol
*sym
)
656 const char *name
= sym
->name
;
660 * ppc64 uses function descriptors and appends a '.' to the
661 * start of every instruction address. Remove it.
666 if (!strcmp(name
, "_text") ||
667 !strcmp(name
, "_etext") ||
668 !strcmp(name
, "_sinittext") ||
669 !strncmp("init_module", name
, 11) ||
670 !strncmp("cleanup_module", name
, 14) ||
671 strstr(name
, "_text_start") ||
672 strstr(name
, "_text_end"))
675 for (i
= 0; skip_symbols
[i
]; i
++) {
676 if (!strcmp(skip_symbols
[i
], name
)) {
685 static void perf_event__process_sample(struct perf_tool
*tool
,
686 const union perf_event
*event
,
687 struct perf_evsel
*evsel
,
688 struct perf_sample
*sample
,
689 struct machine
*machine
)
691 struct perf_top
*top
= container_of(tool
, struct perf_top
, tool
);
692 struct symbol
*parent
= NULL
;
693 u64 ip
= event
->ip
.ip
;
694 struct addr_location al
;
697 if (!machine
&& perf_guest
) {
698 static struct intlist
*seen
;
701 seen
= intlist__new(NULL
);
703 if (!intlist__has_entry(seen
, event
->ip
.pid
)) {
704 pr_err("Can't find guest [%d]'s kernel information\n",
706 intlist__add(seen
, event
->ip
.pid
);
712 pr_err("%u unprocessable samples recorded.\r",
713 top
->session
->stats
.nr_unprocessable_samples
++);
717 if (event
->header
.misc
& PERF_RECORD_MISC_EXACT_IP
)
718 top
->exact_samples
++;
720 if (perf_event__preprocess_sample(event
, machine
, &al
, sample
,
721 symbol_filter
) < 0 ||
725 if (!top
->kptr_restrict_warned
&&
726 symbol_conf
.kptr_restrict
&&
727 al
.cpumode
== PERF_RECORD_MISC_KERNEL
) {
729 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
730 "Check /proc/sys/kernel/kptr_restrict.\n\n"
731 "Kernel%s samples will not be resolved.\n",
732 !RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
]) ?
734 if (use_browser
<= 0)
736 top
->kptr_restrict_warned
= true;
739 if (al
.sym
== NULL
) {
740 const char *msg
= "Kernel samples will not be resolved.\n";
742 * As we do lazy loading of symtabs we only will know if the
743 * specified vmlinux file is invalid when we actually have a
744 * hit in kernel space and then try to load it. So if we get
745 * here and there are _no_ symbols in the DSO backing the
746 * kernel map, bail out.
748 * We may never get here, for instance, if we use -K/
749 * --hide-kernel-symbols, even if the user specifies an
750 * invalid --vmlinux ;-)
752 if (!top
->kptr_restrict_warned
&& !top
->vmlinux_warned
&&
753 al
.map
== machine
->vmlinux_maps
[MAP__FUNCTION
] &&
754 RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
])) {
755 if (symbol_conf
.vmlinux_name
) {
756 ui__warning("The %s file can't be used.\n%s",
757 symbol_conf
.vmlinux_name
, msg
);
759 ui__warning("A vmlinux file was not found.\n%s",
763 if (use_browser
<= 0)
765 top
->vmlinux_warned
= true;
769 if (al
.sym
== NULL
|| !al
.sym
->ignore
) {
770 struct hist_entry
*he
;
772 if ((sort__has_parent
|| symbol_conf
.use_callchain
) &&
774 err
= machine__resolve_callchain(machine
, evsel
,
782 he
= perf_evsel__add_hist_entry(evsel
, &al
, sample
);
784 pr_err("Problem incrementing symbol period, skipping event\n");
788 if (symbol_conf
.use_callchain
) {
789 err
= callchain_append(he
->callchain
, &callchain_cursor
,
796 perf_top__record_precise_ip(top
, he
, evsel
->idx
, ip
);
802 static void perf_top__mmap_read_idx(struct perf_top
*top
, int idx
)
804 struct perf_sample sample
;
805 struct perf_evsel
*evsel
;
806 struct perf_session
*session
= top
->session
;
807 union perf_event
*event
;
808 struct machine
*machine
;
812 while ((event
= perf_evlist__mmap_read(top
->evlist
, idx
)) != NULL
) {
813 ret
= perf_evlist__parse_sample(top
->evlist
, event
, &sample
);
815 pr_err("Can't parse sample, err = %d\n", ret
);
819 evsel
= perf_evlist__id2evsel(session
->evlist
, sample
.id
);
820 assert(evsel
!= NULL
);
822 origin
= event
->header
.misc
& PERF_RECORD_MISC_CPUMODE_MASK
;
824 if (event
->header
.type
== PERF_RECORD_SAMPLE
)
828 case PERF_RECORD_MISC_USER
:
830 if (top
->hide_user_symbols
)
832 machine
= &session
->machines
.host
;
834 case PERF_RECORD_MISC_KERNEL
:
835 ++top
->kernel_samples
;
836 if (top
->hide_kernel_symbols
)
838 machine
= &session
->machines
.host
;
840 case PERF_RECORD_MISC_GUEST_KERNEL
:
841 ++top
->guest_kernel_samples
;
842 machine
= perf_session__find_machine(session
, event
->ip
.pid
);
844 case PERF_RECORD_MISC_GUEST_USER
:
845 ++top
->guest_us_samples
;
847 * TODO: we don't process guest user from host side
848 * except simple counting.
856 if (event
->header
.type
== PERF_RECORD_SAMPLE
) {
857 perf_event__process_sample(&top
->tool
, event
, evsel
,
859 } else if (event
->header
.type
< PERF_RECORD_MAX
) {
860 hists__inc_nr_events(&evsel
->hists
, event
->header
.type
);
861 machine__process_event(machine
, event
);
863 ++session
->stats
.nr_unknown_events
;
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 * 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.
939 top
->session
= perf_session__new(NULL
, O_WRONLY
, false, false, NULL
);
940 if (top
->session
== NULL
)
944 ret
= perf_session_env__lookup_objdump(&top
->session
->header
.env
);
949 ret
= perf_top__setup_sample_type(top
);
953 if (perf_target__has_task(&opts
->target
))
954 perf_event__synthesize_thread_map(&top
->tool
, top
->evlist
->threads
,
956 &top
->session
->machines
.host
);
958 perf_event__synthesize_threads(&top
->tool
, perf_event__process
,
959 &top
->session
->machines
.host
);
961 ret
= perf_top__start_counters(top
);
965 top
->session
->evlist
= top
->evlist
;
966 perf_session__set_id_hdr_size(top
->session
);
969 * When perf is starting the traced process, all the events (apart from
970 * group members) have enable_on_exec=1 set, so don't spoil it by
971 * prematurely enabling them.
973 * XXX 'top' still doesn't start workloads like record, trace, but should,
974 * so leave the check here.
976 if (!perf_target__none(&opts
->target
))
977 perf_evlist__enable(top
->evlist
);
979 /* Wait for a minimal set of events before starting the snapshot */
980 poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
982 perf_top__mmap_read(top
);
985 if (pthread_create(&thread
, NULL
, (use_browser
> 0 ? display_thread_tui
:
986 display_thread
), top
)) {
987 ui__error("Could not create display thread.\n");
991 if (top
->realtime_prio
) {
992 struct sched_param param
;
994 param
.sched_priority
= top
->realtime_prio
;
995 if (sched_setscheduler(0, SCHED_FIFO
, ¶m
)) {
996 ui__error("Could not set realtime priority.\n");
1002 u64 hits
= top
->samples
;
1004 perf_top__mmap_read(top
);
1006 if (hits
== top
->samples
)
1007 ret
= poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
1012 perf_session__delete(top
->session
);
1013 top
->session
= NULL
;
1019 parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1027 symbol_conf
.use_callchain
= true;
1029 return record_parse_callchain_opt(opt
, arg
, unset
);
1033 parse_percent_limit(const struct option
*opt
, const char *arg
,
1034 int unset __maybe_unused
)
1036 struct perf_top
*top
= opt
->value
;
1038 top
->min_percent
= strtof(arg
, NULL
);
1042 int cmd_top(int argc
, const char **argv
, const char *prefix __maybe_unused
)
1045 char errbuf
[BUFSIZ
];
1046 struct perf_top top
= {
1050 .mmap_pages
= UINT_MAX
,
1051 .user_freq
= UINT_MAX
,
1052 .user_interval
= ULLONG_MAX
,
1053 .freq
= 4000, /* 4 KHz */
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('K', "hide_kernel_symbols", &top
.hide_kernel_symbols
,
1078 "hide kernel symbols"),
1079 OPT_UINTEGER('m', "mmap-pages", &opts
->mmap_pages
,
1080 "number of mmap data pages"),
1081 OPT_INTEGER('r', "realtime", &top
.realtime_prio
,
1082 "collect data with this RT SCHED_FIFO priority"),
1083 OPT_INTEGER('d', "delay", &top
.delay_secs
,
1084 "number of seconds to delay between refreshes"),
1085 OPT_BOOLEAN('D', "dump-symtab", &top
.dump_symtab
,
1086 "dump the symbol table used for profiling"),
1087 OPT_INTEGER('f', "count-filter", &top
.count_filter
,
1088 "only display functions with more events than this"),
1089 OPT_BOOLEAN('g', "group", &opts
->group
,
1090 "put the counters into a counter group"),
1091 OPT_BOOLEAN('i', "no-inherit", &opts
->no_inherit
,
1092 "child tasks do not inherit counters"),
1093 OPT_STRING(0, "sym-annotate", &top
.sym_filter
, "symbol name",
1094 "symbol to annotate"),
1095 OPT_BOOLEAN('z', "zero", &top
.zero
, "zero history across updates"),
1096 OPT_UINTEGER('F', "freq", &opts
->user_freq
, "profile at this frequency"),
1097 OPT_INTEGER('E', "entries", &top
.print_entries
,
1098 "display this many functions"),
1099 OPT_BOOLEAN('U', "hide_user_symbols", &top
.hide_user_symbols
,
1100 "hide user symbols"),
1101 OPT_BOOLEAN(0, "tui", &top
.use_tui
, "Use the TUI interface"),
1102 OPT_BOOLEAN(0, "stdio", &top
.use_stdio
, "Use the stdio interface"),
1103 OPT_INCR('v', "verbose", &verbose
,
1104 "be more verbose (show counter open errors, etc)"),
1105 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1106 "sort by key(s): pid, comm, dso, symbol, parent, weight, local_weight"),
1107 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1108 "Show a column with the number of samples"),
1109 OPT_CALLBACK_DEFAULT('G', "call-graph", &top
.record_opts
,
1110 "mode[,dump_size]", record_callchain_help
,
1111 &parse_callchain_opt
, "fp"),
1112 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1113 "Show a column with the sum of periods"),
1114 OPT_STRING(0, "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1115 "only consider symbols in these dsos"),
1116 OPT_STRING(0, "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1117 "only consider symbols in these comms"),
1118 OPT_STRING(0, "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1119 "only consider these symbols"),
1120 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
1121 "Interleave source code with assembly code (default)"),
1122 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
1123 "Display raw encoding of assembly instructions (default)"),
1124 OPT_STRING(0, "objdump", &objdump_path
, "path",
1125 "objdump binary to use for disassembly and annotations"),
1126 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
1127 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1128 OPT_STRING('u', "uid", &target
->uid_str
, "user", "user to profile"),
1129 OPT_CALLBACK(0, "percent-limit", &top
, "percent",
1130 "Don't show entries under that percent", parse_percent_limit
),
1133 const char * const top_usage
[] = {
1134 "perf top [<options>]",
1138 top
.evlist
= perf_evlist__new();
1139 if (top
.evlist
== NULL
)
1142 argc
= parse_options(argc
, argv
, options
, top_usage
, 0);
1144 usage_with_options(top_usage
, options
);
1146 if (sort_order
== default_sort_order
)
1147 sort_order
= "dso,symbol";
1149 if (setup_sorting() < 0)
1150 usage_with_options(top_usage
, options
);
1152 /* display thread wants entries to be collapsed in a different tree */
1153 sort__need_collapse
= 1;
1157 else if (top
.use_tui
)
1160 setup_browser(false);
1162 status
= perf_target__validate(target
);
1164 perf_target__strerror(target
, status
, errbuf
, BUFSIZ
);
1165 ui__warning("%s", errbuf
);
1168 status
= perf_target__parse_uid(target
);
1170 int saved_errno
= errno
;
1172 perf_target__strerror(target
, status
, errbuf
, BUFSIZ
);
1173 ui__error("%s", errbuf
);
1175 status
= -saved_errno
;
1176 goto out_delete_evlist
;
1179 if (perf_target__none(target
))
1180 target
->system_wide
= true;
1182 if (perf_evlist__create_maps(top
.evlist
, target
) < 0)
1183 usage_with_options(top_usage
, options
);
1185 if (!top
.evlist
->nr_entries
&&
1186 perf_evlist__add_default(top
.evlist
) < 0) {
1187 ui__error("Not enough memory for event selector list\n");
1188 goto out_delete_maps
;
1191 symbol_conf
.nr_events
= top
.evlist
->nr_entries
;
1193 if (top
.delay_secs
< 1)
1196 if (opts
->user_interval
!= ULLONG_MAX
)
1197 opts
->default_interval
= opts
->user_interval
;
1198 if (opts
->user_freq
!= UINT_MAX
)
1199 opts
->freq
= opts
->user_freq
;
1202 * User specified count overrides default frequency.
1204 if (opts
->default_interval
)
1206 else if (opts
->freq
) {
1207 opts
->default_interval
= opts
->freq
;
1209 ui__error("frequency and count are zero, aborting\n");
1211 goto out_delete_maps
;
1214 top
.sym_evsel
= perf_evlist__first(top
.evlist
);
1216 symbol_conf
.priv_size
= sizeof(struct annotation
);
1218 symbol_conf
.try_vmlinux_path
= (symbol_conf
.vmlinux_name
== NULL
);
1219 if (symbol__init() < 0)
1222 sort__setup_elide(stdout
);
1224 get_term_dimensions(&top
.winsize
);
1225 if (top
.print_entries
== 0) {
1226 struct sigaction act
= {
1227 .sa_sigaction
= perf_top__sig_winch
,
1228 .sa_flags
= SA_SIGINFO
,
1230 perf_top__update_print_entries(&top
);
1231 sigaction(SIGWINCH
, &act
, NULL
);
1234 status
= __cmd_top(&top
);
1237 perf_evlist__delete_maps(top
.evlist
);
1239 perf_evlist__delete(top
.evlist
);