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 ip
= he
->ms
.map
->map_ip(he
->ms
.map
, ip
);
194 if (ui__has_annotation())
195 err
= hist_entry__inc_addr_samples(he
, counter
, ip
);
197 pthread_mutex_unlock(¬es
->lock
);
200 * This function is now called with he->hists->lock held.
201 * Release it before going to sleep.
203 pthread_mutex_unlock(&he
->hists
->lock
);
205 if (err
== -ERANGE
&& !he
->ms
.map
->erange_warned
)
206 ui__warn_map_erange(he
->ms
.map
, sym
, ip
);
207 else if (err
== -ENOMEM
) {
208 pr_err("Not enough memory for annotating '%s' symbol!\n",
213 pthread_mutex_lock(&he
->hists
->lock
);
216 static void perf_top__show_details(struct perf_top
*top
)
218 struct hist_entry
*he
= top
->sym_filter_entry
;
219 struct annotation
*notes
;
220 struct symbol
*symbol
;
227 notes
= symbol__annotation(symbol
);
229 pthread_mutex_lock(¬es
->lock
);
231 if (notes
->src
== NULL
)
234 printf("Showing %s for %s\n", perf_evsel__name(top
->sym_evsel
), symbol
->name
);
235 printf(" Events Pcnt (>=%d%%)\n", top
->sym_pcnt_filter
);
237 more
= symbol__annotate_printf(symbol
, he
->ms
.map
, top
->sym_evsel
,
238 0, top
->sym_pcnt_filter
, top
->print_entries
, 4);
240 symbol__annotate_zero_histogram(symbol
, top
->sym_evsel
->idx
);
242 symbol__annotate_decay_histogram(symbol
, top
->sym_evsel
->idx
);
244 printf("%d lines not displayed, maybe increase display entries [e]\n", more
);
246 pthread_mutex_unlock(¬es
->lock
);
249 static void perf_top__print_sym_table(struct perf_top
*top
)
253 const int win_width
= top
->winsize
.ws_col
- 1;
257 perf_top__header_snprintf(top
, bf
, sizeof(bf
));
260 perf_top__reset_sample_counters(top
);
262 printf("%-*.*s\n", win_width
, win_width
, graph_dotted_line
);
264 if (top
->sym_evsel
->hists
.stats
.nr_lost_warned
!=
265 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
]) {
266 top
->sym_evsel
->hists
.stats
.nr_lost_warned
=
267 top
->sym_evsel
->hists
.stats
.nr_events
[PERF_RECORD_LOST
];
268 color_fprintf(stdout
, PERF_COLOR_RED
,
269 "WARNING: LOST %d chunks, Check IO/CPU overload",
270 top
->sym_evsel
->hists
.stats
.nr_lost_warned
);
274 if (top
->sym_filter_entry
) {
275 perf_top__show_details(top
);
280 hists__delete_entries(&top
->sym_evsel
->hists
);
282 hists__decay_entries(&top
->sym_evsel
->hists
,
283 top
->hide_user_symbols
,
284 top
->hide_kernel_symbols
);
287 hists__collapse_resort(&top
->sym_evsel
->hists
, NULL
);
288 hists__output_resort(&top
->sym_evsel
->hists
);
290 hists__output_recalc_col_len(&top
->sym_evsel
->hists
,
291 top
->print_entries
- printed
);
293 hists__fprintf(&top
->sym_evsel
->hists
, false,
294 top
->print_entries
- printed
, win_width
,
295 top
->min_percent
, stdout
);
298 static void prompt_integer(int *target
, const char *msg
)
300 char *buf
= malloc(0), *p
;
304 fprintf(stdout
, "\n%s: ", msg
);
305 if (getline(&buf
, &dummy
, stdin
) < 0)
308 p
= strchr(buf
, '\n');
318 tmp
= strtoul(buf
, NULL
, 10);
324 static void prompt_percent(int *target
, const char *msg
)
328 prompt_integer(&tmp
, msg
);
329 if (tmp
>= 0 && tmp
<= 100)
333 static void perf_top__prompt_symbol(struct perf_top
*top
, const char *msg
)
335 char *buf
= malloc(0), *p
;
336 struct hist_entry
*syme
= top
->sym_filter_entry
, *n
, *found
= NULL
;
337 struct rb_node
*next
;
340 /* zero counters of active symbol */
342 __zero_source_counters(syme
);
343 top
->sym_filter_entry
= NULL
;
346 fprintf(stdout
, "\n%s: ", msg
);
347 if (getline(&buf
, &dummy
, stdin
) < 0)
350 p
= strchr(buf
, '\n');
354 next
= rb_first(&top
->sym_evsel
->hists
.entries
);
356 n
= rb_entry(next
, struct hist_entry
, rb_node
);
357 if (n
->ms
.sym
&& !strcmp(buf
, n
->ms
.sym
->name
)) {
361 next
= rb_next(&n
->rb_node
);
365 fprintf(stderr
, "Sorry, %s is not active.\n", buf
);
368 perf_top__parse_source(top
, found
);
374 static void perf_top__print_mapped_keys(struct perf_top
*top
)
378 if (top
->sym_filter_entry
) {
379 struct symbol
*sym
= top
->sym_filter_entry
->ms
.sym
;
383 fprintf(stdout
, "\nMapped keys:\n");
384 fprintf(stdout
, "\t[d] display refresh delay. \t(%d)\n", top
->delay_secs
);
385 fprintf(stdout
, "\t[e] display entries (lines). \t(%d)\n", top
->print_entries
);
387 if (top
->evlist
->nr_entries
> 1)
388 fprintf(stdout
, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top
->sym_evsel
));
390 fprintf(stdout
, "\t[f] profile display filter (count). \t(%d)\n", top
->count_filter
);
392 fprintf(stdout
, "\t[F] annotate display filter (percent). \t(%d%%)\n", top
->sym_pcnt_filter
);
393 fprintf(stdout
, "\t[s] annotate symbol. \t(%s)\n", name
?: "NULL");
394 fprintf(stdout
, "\t[S] stop annotation.\n");
397 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
398 top
->hide_kernel_symbols
? "yes" : "no");
400 "\t[U] hide user symbols. \t(%s)\n",
401 top
->hide_user_symbols
? "yes" : "no");
402 fprintf(stdout
, "\t[z] toggle sample zeroing. \t(%d)\n", top
->zero
? 1 : 0);
403 fprintf(stdout
, "\t[qQ] quit.\n");
406 static int perf_top__key_mapped(struct perf_top
*top
, int c
)
422 return top
->evlist
->nr_entries
> 1 ? 1 : 0;
430 static bool perf_top__handle_keypress(struct perf_top
*top
, int c
)
434 if (!perf_top__key_mapped(top
, c
)) {
435 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
436 struct termios tc
, save
;
438 perf_top__print_mapped_keys(top
);
439 fprintf(stdout
, "\nEnter selection, or unmapped key to continue: ");
444 tc
.c_lflag
&= ~(ICANON
| ECHO
);
447 tcsetattr(0, TCSANOW
, &tc
);
449 poll(&stdin_poll
, 1, -1);
452 tcsetattr(0, TCSAFLUSH
, &save
);
453 if (!perf_top__key_mapped(top
, c
))
459 prompt_integer(&top
->delay_secs
, "Enter display delay");
460 if (top
->delay_secs
< 1)
464 prompt_integer(&top
->print_entries
, "Enter display entries (lines)");
465 if (top
->print_entries
== 0) {
466 struct sigaction act
= {
467 .sa_sigaction
= perf_top__sig_winch
,
468 .sa_flags
= SA_SIGINFO
,
470 perf_top__sig_winch(SIGWINCH
, NULL
, top
);
471 sigaction(SIGWINCH
, &act
, NULL
);
473 signal(SIGWINCH
, SIG_DFL
);
477 if (top
->evlist
->nr_entries
> 1) {
478 /* Select 0 as the default event: */
481 fprintf(stderr
, "\nAvailable events:");
483 evlist__for_each(top
->evlist
, top
->sym_evsel
)
484 fprintf(stderr
, "\n\t%d %s", top
->sym_evsel
->idx
, perf_evsel__name(top
->sym_evsel
));
486 prompt_integer(&counter
, "Enter details event counter");
488 if (counter
>= top
->evlist
->nr_entries
) {
489 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
490 fprintf(stderr
, "Sorry, no such event, using %s.\n", perf_evsel__name(top
->sym_evsel
));
494 evlist__for_each(top
->evlist
, top
->sym_evsel
)
495 if (top
->sym_evsel
->idx
== counter
)
498 top
->sym_evsel
= perf_evlist__first(top
->evlist
);
501 prompt_integer(&top
->count_filter
, "Enter display event count filter");
504 prompt_percent(&top
->sym_pcnt_filter
,
505 "Enter details display event filter (percent)");
508 top
->hide_kernel_symbols
= !top
->hide_kernel_symbols
;
512 printf("exiting.\n");
513 if (top
->dump_symtab
)
514 perf_session__fprintf_dsos(top
->session
, stderr
);
518 perf_top__prompt_symbol(top
, "Enter details symbol");
521 if (!top
->sym_filter_entry
)
524 struct hist_entry
*syme
= top
->sym_filter_entry
;
526 top
->sym_filter_entry
= NULL
;
527 __zero_source_counters(syme
);
531 top
->hide_user_symbols
= !top
->hide_user_symbols
;
534 top
->zero
= !top
->zero
;
543 static void perf_top__sort_new_samples(void *arg
)
545 struct perf_top
*t
= arg
;
546 perf_top__reset_sample_counters(t
);
548 if (t
->evlist
->selected
!= NULL
)
549 t
->sym_evsel
= t
->evlist
->selected
;
552 hists__delete_entries(&t
->sym_evsel
->hists
);
554 hists__decay_entries(&t
->sym_evsel
->hists
,
555 t
->hide_user_symbols
,
556 t
->hide_kernel_symbols
);
559 hists__collapse_resort(&t
->sym_evsel
->hists
, NULL
);
560 hists__output_resort(&t
->sym_evsel
->hists
);
563 static void *display_thread_tui(void *arg
)
565 struct perf_evsel
*pos
;
566 struct perf_top
*top
= arg
;
567 const char *help
= "For a higher level overview, try: perf top --sort comm,dso";
568 struct hist_browser_timer hbt
= {
569 .timer
= perf_top__sort_new_samples
,
571 .refresh
= top
->delay_secs
,
574 perf_top__sort_new_samples(top
);
577 * Initialize the uid_filter_str, in the future the TUI will allow
578 * Zooming in/out UIDs. For now juse use whatever the user passed
581 evlist__for_each(top
->evlist
, pos
)
582 pos
->hists
.uid_filter_str
= top
->record_opts
.target
.uid_str
;
584 perf_evlist__tui_browse_hists(top
->evlist
, help
, &hbt
, top
->min_percent
,
585 &top
->session
->header
.env
);
591 static void display_sig(int sig __maybe_unused
)
596 static void display_setup_sig(void)
598 signal(SIGSEGV
, display_sig
);
599 signal(SIGFPE
, display_sig
);
600 signal(SIGINT
, display_sig
);
601 signal(SIGQUIT
, display_sig
);
602 signal(SIGTERM
, display_sig
);
605 static void *display_thread(void *arg
)
607 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
609 struct perf_top
*top
= arg
;
613 pthread__unblock_sigwinch();
615 delay_msecs
= top
->delay_secs
* 1000;
616 set_term_quiet_input(&save
);
621 perf_top__print_sym_table(top
);
623 * Either timeout expired or we got an EINTR due to SIGWINCH,
624 * refresh screen in both cases.
626 switch (poll(&stdin_poll
, 1, delay_msecs
)) {
635 tcsetattr(0, TCSAFLUSH
, &save
);
637 if (perf_top__handle_keypress(top
, c
))
643 tcsetattr(0, TCSAFLUSH
, &save
);
647 static int symbol_filter(struct map
*map
, struct symbol
*sym
)
649 const char *name
= sym
->name
;
651 if (!map
->dso
->kernel
)
654 * ppc64 uses function descriptors and appends a '.' to the
655 * start of every instruction address. Remove it.
660 if (!strcmp(name
, "_text") ||
661 !strcmp(name
, "_etext") ||
662 !strcmp(name
, "_sinittext") ||
663 !strncmp("init_module", name
, 11) ||
664 !strncmp("cleanup_module", name
, 14) ||
665 strstr(name
, "_text_start") ||
666 strstr(name
, "_text_end"))
669 if (symbol__is_idle(sym
))
675 static int hist_iter__top_callback(struct hist_entry_iter
*iter
,
676 struct addr_location
*al
, bool single
,
679 struct perf_top
*top
= arg
;
680 struct hist_entry
*he
= iter
->he
;
681 struct perf_evsel
*evsel
= iter
->evsel
;
683 if (sort__has_sym
&& single
) {
687 ip
= al
->map
->unmap_ip(al
->map
, ip
);
689 perf_top__record_precise_ip(top
, he
, evsel
->idx
, ip
);
695 static void perf_event__process_sample(struct perf_tool
*tool
,
696 const union perf_event
*event
,
697 struct perf_evsel
*evsel
,
698 struct perf_sample
*sample
,
699 struct machine
*machine
)
701 struct perf_top
*top
= container_of(tool
, struct perf_top
, tool
);
702 struct addr_location al
;
705 if (!machine
&& perf_guest
) {
706 static struct intlist
*seen
;
709 seen
= intlist__new(NULL
);
711 if (!intlist__has_entry(seen
, sample
->pid
)) {
712 pr_err("Can't find guest [%d]'s kernel information\n",
714 intlist__add(seen
, sample
->pid
);
720 pr_err("%u unprocessable samples recorded.\r",
721 top
->session
->stats
.nr_unprocessable_samples
++);
725 if (event
->header
.misc
& PERF_RECORD_MISC_EXACT_IP
)
726 top
->exact_samples
++;
728 if (perf_event__preprocess_sample(event
, machine
, &al
, sample
) < 0)
731 if (!top
->kptr_restrict_warned
&&
732 symbol_conf
.kptr_restrict
&&
733 al
.cpumode
== PERF_RECORD_MISC_KERNEL
) {
735 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
736 "Check /proc/sys/kernel/kptr_restrict.\n\n"
737 "Kernel%s samples will not be resolved.\n",
738 !RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
]) ?
740 if (use_browser
<= 0)
742 top
->kptr_restrict_warned
= true;
745 if (al
.sym
== NULL
) {
746 const char *msg
= "Kernel samples will not be resolved.\n";
748 * As we do lazy loading of symtabs we only will know if the
749 * specified vmlinux file is invalid when we actually have a
750 * hit in kernel space and then try to load it. So if we get
751 * here and there are _no_ symbols in the DSO backing the
752 * kernel map, bail out.
754 * We may never get here, for instance, if we use -K/
755 * --hide-kernel-symbols, even if the user specifies an
756 * invalid --vmlinux ;-)
758 if (!top
->kptr_restrict_warned
&& !top
->vmlinux_warned
&&
759 al
.map
== machine
->vmlinux_maps
[MAP__FUNCTION
] &&
760 RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
])) {
761 if (symbol_conf
.vmlinux_name
) {
762 ui__warning("The %s file can't be used.\n%s",
763 symbol_conf
.vmlinux_name
, msg
);
765 ui__warning("A vmlinux file was not found.\n%s",
769 if (use_browser
<= 0)
771 top
->vmlinux_warned
= true;
775 if (al
.sym
== NULL
|| !al
.sym
->ignore
) {
776 struct hist_entry_iter iter
= {
777 .add_entry_cb
= hist_iter__top_callback
,
780 if (symbol_conf
.cumulate_callchain
)
781 iter
.ops
= &hist_iter_cumulative
;
783 iter
.ops
= &hist_iter_normal
;
785 pthread_mutex_lock(&evsel
->hists
.lock
);
787 err
= hist_entry_iter__add(&iter
, &al
, evsel
, sample
,
788 top
->max_stack
, top
);
790 pr_err("Problem incrementing symbol period, skipping event\n");
792 pthread_mutex_unlock(&evsel
->hists
.lock
);
798 static void perf_top__mmap_read_idx(struct perf_top
*top
, int idx
)
800 struct perf_sample sample
;
801 struct perf_evsel
*evsel
;
802 struct perf_session
*session
= top
->session
;
803 union perf_event
*event
;
804 struct machine
*machine
;
808 while ((event
= perf_evlist__mmap_read(top
->evlist
, idx
)) != NULL
) {
809 ret
= perf_evlist__parse_sample(top
->evlist
, event
, &sample
);
811 pr_err("Can't parse sample, err = %d\n", ret
);
815 evsel
= perf_evlist__id2evsel(session
->evlist
, sample
.id
);
816 assert(evsel
!= NULL
);
818 origin
= event
->header
.misc
& PERF_RECORD_MISC_CPUMODE_MASK
;
820 if (event
->header
.type
== PERF_RECORD_SAMPLE
)
824 case PERF_RECORD_MISC_USER
:
826 if (top
->hide_user_symbols
)
828 machine
= &session
->machines
.host
;
830 case PERF_RECORD_MISC_KERNEL
:
831 ++top
->kernel_samples
;
832 if (top
->hide_kernel_symbols
)
834 machine
= &session
->machines
.host
;
836 case PERF_RECORD_MISC_GUEST_KERNEL
:
837 ++top
->guest_kernel_samples
;
838 machine
= perf_session__find_machine(session
,
841 case PERF_RECORD_MISC_GUEST_USER
:
842 ++top
->guest_us_samples
;
844 * TODO: we don't process guest user from host side
845 * except simple counting.
853 if (event
->header
.type
== PERF_RECORD_SAMPLE
) {
854 perf_event__process_sample(&top
->tool
, event
, evsel
,
856 } else if (event
->header
.type
< PERF_RECORD_MAX
) {
857 hists__inc_nr_events(&evsel
->hists
, event
->header
.type
);
858 machine__process_event(machine
, event
, &sample
);
860 ++session
->stats
.nr_unknown_events
;
862 perf_evlist__mmap_consume(top
->evlist
, idx
);
866 static void perf_top__mmap_read(struct perf_top
*top
)
870 for (i
= 0; i
< top
->evlist
->nr_mmaps
; i
++)
871 perf_top__mmap_read_idx(top
, i
);
874 static int perf_top__start_counters(struct perf_top
*top
)
877 struct perf_evsel
*counter
;
878 struct perf_evlist
*evlist
= top
->evlist
;
879 struct record_opts
*opts
= &top
->record_opts
;
881 perf_evlist__config(evlist
, opts
);
883 evlist__for_each(evlist
, counter
) {
885 if (perf_evsel__open(counter
, top
->evlist
->cpus
,
886 top
->evlist
->threads
) < 0) {
887 if (perf_evsel__fallback(counter
, errno
, msg
, sizeof(msg
))) {
889 ui__warning("%s\n", msg
);
893 perf_evsel__open_strerror(counter
, &opts
->target
,
894 errno
, msg
, sizeof(msg
));
895 ui__error("%s\n", msg
);
900 if (perf_evlist__mmap(evlist
, opts
->mmap_pages
, false) < 0) {
901 ui__error("Failed to mmap with %d (%s)\n",
902 errno
, strerror(errno
));
912 static int perf_top__setup_sample_type(struct perf_top
*top __maybe_unused
)
914 if (!sort__has_sym
) {
915 if (symbol_conf
.use_callchain
) {
916 ui__error("Selected -g but \"sym\" not present in --sort/-s.");
919 } else if (callchain_param
.mode
!= CHAIN_NONE
) {
920 if (callchain_register_param(&callchain_param
) < 0) {
921 ui__error("Can't register callchain params.\n");
929 static int __cmd_top(struct perf_top
*top
)
931 struct record_opts
*opts
= &top
->record_opts
;
935 top
->session
= perf_session__new(NULL
, false, NULL
);
936 if (top
->session
== NULL
)
939 machines__set_symbol_filter(&top
->session
->machines
, symbol_filter
);
942 ret
= perf_session_env__lookup_objdump(&top
->session
->header
.env
);
947 ret
= perf_top__setup_sample_type(top
);
951 machine__synthesize_threads(&top
->session
->machines
.host
, &opts
->target
,
952 top
->evlist
->threads
, false);
953 ret
= perf_top__start_counters(top
);
957 top
->session
->evlist
= top
->evlist
;
958 perf_session__set_id_hdr_size(top
->session
);
961 * When perf is starting the traced process, all the events (apart from
962 * group members) have enable_on_exec=1 set, so don't spoil it by
963 * prematurely enabling them.
965 * XXX 'top' still doesn't start workloads like record, trace, but should,
966 * so leave the check here.
968 if (!target__none(&opts
->target
))
969 perf_evlist__enable(top
->evlist
);
971 /* Wait for a minimal set of events before starting the snapshot */
972 poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
974 perf_top__mmap_read(top
);
977 if (pthread_create(&thread
, NULL
, (use_browser
> 0 ? display_thread_tui
:
978 display_thread
), top
)) {
979 ui__error("Could not create display thread.\n");
983 if (top
->realtime_prio
) {
984 struct sched_param param
;
986 param
.sched_priority
= top
->realtime_prio
;
987 if (sched_setscheduler(0, SCHED_FIFO
, ¶m
)) {
988 ui__error("Could not set realtime priority.\n");
994 u64 hits
= top
->samples
;
996 perf_top__mmap_read(top
);
998 if (hits
== top
->samples
)
999 ret
= poll(top
->evlist
->pollfd
, top
->evlist
->nr_fds
, 100);
1004 pthread_join(thread
, NULL
);
1006 perf_session__delete(top
->session
);
1007 top
->session
= NULL
;
1013 callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1015 symbol_conf
.use_callchain
= true;
1016 return record_callchain_opt(opt
, arg
, unset
);
1020 parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1022 symbol_conf
.use_callchain
= true;
1023 return record_parse_callchain_opt(opt
, arg
, unset
);
1026 static int perf_top_config(const char *var
, const char *value
, void *cb
)
1028 struct perf_top
*top
= cb
;
1030 if (!strcmp(var
, "top.call-graph"))
1031 return record_parse_callchain(value
, &top
->record_opts
);
1032 if (!strcmp(var
, "top.children")) {
1033 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
1037 return perf_default_config(var
, value
, cb
);
1041 parse_percent_limit(const struct option
*opt
, const char *arg
,
1042 int unset __maybe_unused
)
1044 struct perf_top
*top
= opt
->value
;
1046 top
->min_percent
= strtof(arg
, NULL
);
1050 int cmd_top(int argc
, const char **argv
, const char *prefix __maybe_unused
)
1053 char errbuf
[BUFSIZ
];
1054 struct perf_top top
= {
1058 .mmap_pages
= UINT_MAX
,
1059 .user_freq
= UINT_MAX
,
1060 .user_interval
= ULLONG_MAX
,
1061 .freq
= 4000, /* 4 KHz */
1066 .max_stack
= PERF_MAX_STACK_DEPTH
,
1067 .sym_pcnt_filter
= 5,
1069 struct record_opts
*opts
= &top
.record_opts
;
1070 struct target
*target
= &opts
->target
;
1071 const struct option options
[] = {
1072 OPT_CALLBACK('e', "event", &top
.evlist
, "event",
1073 "event selector. use 'perf list' to list available events",
1074 parse_events_option
),
1075 OPT_U64('c', "count", &opts
->user_interval
, "event period to sample"),
1076 OPT_STRING('p', "pid", &target
->pid
, "pid",
1077 "profile events on existing process id"),
1078 OPT_STRING('t', "tid", &target
->tid
, "tid",
1079 "profile events on existing thread id"),
1080 OPT_BOOLEAN('a', "all-cpus", &target
->system_wide
,
1081 "system-wide collection from all CPUs"),
1082 OPT_STRING('C', "cpu", &target
->cpu_list
, "cpu",
1083 "list of cpus to monitor"),
1084 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1085 "file", "vmlinux pathname"),
1086 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1087 "don't load vmlinux even if found"),
1088 OPT_BOOLEAN('K', "hide_kernel_symbols", &top
.hide_kernel_symbols
,
1089 "hide kernel symbols"),
1090 OPT_CALLBACK('m', "mmap-pages", &opts
->mmap_pages
, "pages",
1091 "number of mmap data pages",
1092 perf_evlist__parse_mmap_pages
),
1093 OPT_INTEGER('r', "realtime", &top
.realtime_prio
,
1094 "collect data with this RT SCHED_FIFO priority"),
1095 OPT_INTEGER('d', "delay", &top
.delay_secs
,
1096 "number of seconds to delay between refreshes"),
1097 OPT_BOOLEAN('D', "dump-symtab", &top
.dump_symtab
,
1098 "dump the symbol table used for profiling"),
1099 OPT_INTEGER('f', "count-filter", &top
.count_filter
,
1100 "only display functions with more events than this"),
1101 OPT_BOOLEAN(0, "group", &opts
->group
,
1102 "put the counters into a counter group"),
1103 OPT_BOOLEAN('i', "no-inherit", &opts
->no_inherit
,
1104 "child tasks do not inherit counters"),
1105 OPT_STRING(0, "sym-annotate", &top
.sym_filter
, "symbol name",
1106 "symbol to annotate"),
1107 OPT_BOOLEAN('z', "zero", &top
.zero
, "zero history across updates"),
1108 OPT_UINTEGER('F', "freq", &opts
->user_freq
, "profile at this frequency"),
1109 OPT_INTEGER('E', "entries", &top
.print_entries
,
1110 "display this many functions"),
1111 OPT_BOOLEAN('U', "hide_user_symbols", &top
.hide_user_symbols
,
1112 "hide user symbols"),
1113 OPT_BOOLEAN(0, "tui", &top
.use_tui
, "Use the TUI interface"),
1114 OPT_BOOLEAN(0, "stdio", &top
.use_stdio
, "Use the stdio interface"),
1115 OPT_INCR('v', "verbose", &verbose
,
1116 "be more verbose (show counter open errors, etc)"),
1117 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1118 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1119 " Please refer the man page for the complete list."),
1120 OPT_STRING(0, "fields", &field_order
, "key[,keys...]",
1121 "output field(s): overhead, period, sample plus all of sort keys"),
1122 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1123 "Show a column with the number of samples"),
1124 OPT_CALLBACK_NOOPT('g', NULL
, &top
.record_opts
,
1125 NULL
, "enables call-graph recording",
1127 OPT_CALLBACK(0, "call-graph", &top
.record_opts
,
1128 "mode[,dump_size]", record_callchain_help
,
1129 &parse_callchain_opt
),
1130 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1131 "Accumulate callchains of children and show total overhead as well"),
1132 OPT_INTEGER(0, "max-stack", &top
.max_stack
,
1133 "Set the maximum stack depth when parsing the callchain. "
1134 "Default: " __stringify(PERF_MAX_STACK_DEPTH
)),
1135 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1136 "ignore callees of these functions in call graphs",
1137 report_parse_ignore_callees_opt
),
1138 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1139 "Show a column with the sum of periods"),
1140 OPT_STRING(0, "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1141 "only consider symbols in these dsos"),
1142 OPT_STRING(0, "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1143 "only consider symbols in these comms"),
1144 OPT_STRING(0, "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1145 "only consider these symbols"),
1146 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
1147 "Interleave source code with assembly code (default)"),
1148 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
1149 "Display raw encoding of assembly instructions (default)"),
1150 OPT_STRING(0, "objdump", &objdump_path
, "path",
1151 "objdump binary to use for disassembly and annotations"),
1152 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
1153 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1154 OPT_STRING('u', "uid", &target
->uid_str
, "user", "user to profile"),
1155 OPT_CALLBACK(0, "percent-limit", &top
, "percent",
1156 "Don't show entries under that percent", parse_percent_limit
),
1157 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1158 "How to display percentage of filtered entries", parse_filter_percentage
),
1159 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1161 "don't try to adjust column width, use these fixed values"),
1164 const char * const top_usage
[] = {
1165 "perf top [<options>]",
1169 top
.evlist
= perf_evlist__new();
1170 if (top
.evlist
== NULL
)
1173 perf_config(perf_top_config
, &top
);
1175 argc
= parse_options(argc
, argv
, options
, top_usage
, 0);
1177 usage_with_options(top_usage
, options
);
1179 sort__mode
= SORT_MODE__TOP
;
1180 /* display thread wants entries to be collapsed in a different tree */
1181 sort__need_collapse
= 1;
1183 if (setup_sorting() < 0) {
1185 parse_options_usage(top_usage
, options
, "s", 1);
1187 parse_options_usage(sort_order
? NULL
: top_usage
,
1188 options
, "fields", 0);
1189 goto out_delete_evlist
;
1194 else if (top
.use_tui
)
1197 setup_browser(false);
1199 status
= target__validate(target
);
1201 target__strerror(target
, status
, errbuf
, BUFSIZ
);
1202 ui__warning("%s\n", errbuf
);
1205 status
= target__parse_uid(target
);
1207 int saved_errno
= errno
;
1209 target__strerror(target
, status
, errbuf
, BUFSIZ
);
1210 ui__error("%s\n", errbuf
);
1212 status
= -saved_errno
;
1213 goto out_delete_evlist
;
1216 if (target__none(target
))
1217 target
->system_wide
= true;
1219 if (perf_evlist__create_maps(top
.evlist
, target
) < 0)
1220 usage_with_options(top_usage
, options
);
1222 if (!top
.evlist
->nr_entries
&&
1223 perf_evlist__add_default(top
.evlist
) < 0) {
1224 ui__error("Not enough memory for event selector list\n");
1225 goto out_delete_evlist
;
1228 symbol_conf
.nr_events
= top
.evlist
->nr_entries
;
1230 if (top
.delay_secs
< 1)
1233 if (record_opts__config(opts
)) {
1235 goto out_delete_evlist
;
1238 top
.sym_evsel
= perf_evlist__first(top
.evlist
);
1240 if (!symbol_conf
.use_callchain
) {
1241 symbol_conf
.cumulate_callchain
= false;
1242 perf_hpp__cancel_cumulate();
1245 symbol_conf
.priv_size
= sizeof(struct annotation
);
1247 symbol_conf
.try_vmlinux_path
= (symbol_conf
.vmlinux_name
== NULL
);
1248 if (symbol__init(NULL
) < 0)
1251 sort__setup_elide(stdout
);
1253 get_term_dimensions(&top
.winsize
);
1254 if (top
.print_entries
== 0) {
1255 struct sigaction act
= {
1256 .sa_sigaction
= perf_top__sig_winch
,
1257 .sa_flags
= SA_SIGINFO
,
1259 perf_top__update_print_entries(&top
);
1260 sigaction(SIGWINCH
, &act
, NULL
);
1263 status
= __cmd_top(&top
);
1266 perf_evlist__delete(top
.evlist
);