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/config.h"
26 #include "util/color.h"
27 #include "util/evlist.h"
28 #include "util/evsel.h"
29 #include "util/machine.h"
30 #include "util/session.h"
31 #include "util/symbol.h"
32 #include "util/thread.h"
33 #include "util/thread_map.h"
35 #include "util/util.h"
36 #include <linux/rbtree.h>
37 #include <subcmd/parse-options.h>
38 #include "util/parse-events.h"
39 #include "util/cpumap.h"
40 #include "util/xyarray.h"
41 #include "util/sort.h"
42 #include "util/intlist.h"
43 #include "util/parse-branch-options.h"
44 #include "arch/common.h"
46 #include "util/debug.h"
61 #include <sys/syscall.h>
62 #include <sys/ioctl.h>
64 #include <sys/prctl.h>
67 #include <sys/utsname.h>
70 #include <linux/stringify.h>
71 #include <linux/time64.h>
72 #include <linux/types.h>
74 static volatile int done
;
76 #define HEADER_LINE_NR 5
78 static void perf_top__update_print_entries(struct perf_top
*top
)
80 top
->print_entries
= top
->winsize
.ws_row
- HEADER_LINE_NR
;
83 static void perf_top__sig_winch(int sig __maybe_unused
,
84 siginfo_t
*info __maybe_unused
, void *arg
)
86 struct perf_top
*top
= arg
;
88 get_term_dimensions(&top
->winsize
);
89 perf_top__update_print_entries(top
);
92 static int perf_top__parse_source(struct perf_top
*top
, struct hist_entry
*he
)
95 struct annotation
*notes
;
99 if (!he
|| !he
->ms
.sym
)
106 * We can't annotate with just /proc/kallsyms
108 if (map
->dso
->symtab_type
== DSO_BINARY_TYPE__KALLSYMS
&&
109 !dso__is_kcore(map
->dso
)) {
110 pr_err("Can't annotate %s: No vmlinux file was found in the "
111 "path\n", sym
->name
);
116 notes
= symbol__annotation(sym
);
117 if (notes
->src
!= NULL
) {
118 pthread_mutex_lock(¬es
->lock
);
122 pthread_mutex_lock(¬es
->lock
);
124 if (symbol__alloc_hist(sym
) < 0) {
125 pthread_mutex_unlock(¬es
->lock
);
126 pr_err("Not enough memory for annotating '%s' symbol!\n",
132 err
= symbol__disassemble(sym
, map
, 0);
135 top
->sym_filter_entry
= he
;
138 symbol__strerror_disassemble(sym
, map
, err
, msg
, sizeof(msg
));
139 pr_err("Couldn't annotate %s: %s\n", sym
->name
, msg
);
142 pthread_mutex_unlock(¬es
->lock
);
146 static void __zero_source_counters(struct hist_entry
*he
)
148 struct symbol
*sym
= he
->ms
.sym
;
149 symbol__annotate_zero_histograms(sym
);
152 static void ui__warn_map_erange(struct map
*map
, struct symbol
*sym
, u64 ip
)
155 int err
= uname(&uts
);
157 ui__warning("Out of bounds address found:\n\n"
158 "Addr: %" PRIx64
"\n"
160 "Map: %" PRIx64
"-%" PRIx64
"\n"
161 "Symbol: %" PRIx64
"-%" PRIx64
" %c %s\n"
165 "Not all samples will be on the annotation output.\n\n"
166 "Please report to linux-kernel@vger.kernel.org\n",
167 ip
, map
->dso
->long_name
, dso__symtab_origin(map
->dso
),
168 map
->start
, map
->end
, sym
->start
, sym
->end
,
169 sym
->binding
== STB_GLOBAL
? 'g' :
170 sym
->binding
== STB_LOCAL
? 'l' : 'w', sym
->name
,
171 err
? "[unknown]" : uts
.machine
,
172 err
? "[unknown]" : uts
.release
, perf_version_string
);
173 if (use_browser
<= 0)
176 map
->erange_warned
= true;
179 static void perf_top__record_precise_ip(struct perf_top
*top
,
180 struct hist_entry
*he
,
183 struct annotation
*notes
;
184 struct symbol
*sym
= he
->ms
.sym
;
187 if (sym
== NULL
|| (use_browser
== 0 &&
188 (top
->sym_filter_entry
== NULL
||
189 top
->sym_filter_entry
->ms
.sym
!= sym
)))
192 notes
= symbol__annotation(sym
);
194 if (pthread_mutex_trylock(¬es
->lock
))
197 err
= hist_entry__inc_addr_samples(he
, counter
, ip
);
199 pthread_mutex_unlock(¬es
->lock
);
203 * This function is now called with he->hists->lock held.
204 * Release it before going to sleep.
206 pthread_mutex_unlock(&he
->hists
->lock
);
208 if (err
== -ERANGE
&& !he
->ms
.map
->erange_warned
)
209 ui__warn_map_erange(he
->ms
.map
, sym
, ip
);
210 else if (err
== -ENOMEM
) {
211 pr_err("Not enough memory for annotating '%s' symbol!\n",
216 pthread_mutex_lock(&he
->hists
->lock
);
220 static void perf_top__show_details(struct perf_top
*top
)
222 struct hist_entry
*he
= top
->sym_filter_entry
;
223 struct annotation
*notes
;
224 struct symbol
*symbol
;
231 notes
= symbol__annotation(symbol
);
233 pthread_mutex_lock(¬es
->lock
);
235 if (notes
->src
== NULL
)
238 printf("Showing %s for %s\n", perf_evsel__name(top
->sym_evsel
), symbol
->name
);
239 printf(" Events Pcnt (>=%d%%)\n", top
->sym_pcnt_filter
);
241 more
= symbol__annotate_printf(symbol
, he
->ms
.map
, top
->sym_evsel
,
242 0, top
->sym_pcnt_filter
, top
->print_entries
, 4);
244 if (top
->evlist
->enabled
) {
246 symbol__annotate_zero_histogram(symbol
, top
->sym_evsel
->idx
);
248 symbol__annotate_decay_histogram(symbol
, top
->sym_evsel
->idx
);
251 printf("%d lines not displayed, maybe increase display entries [e]\n", more
);
253 pthread_mutex_unlock(¬es
->lock
);
256 static void perf_top__print_sym_table(struct perf_top
*top
)
260 const int win_width
= top
->winsize
.ws_col
- 1;
261 struct perf_evsel
*evsel
= top
->sym_evsel
;
262 struct hists
*hists
= evsel__hists(evsel
);
266 perf_top__header_snprintf(top
, bf
, sizeof(bf
));
269 perf_top__reset_sample_counters(top
);
271 printf("%-*.*s\n", win_width
, win_width
, graph_dotted_line
);
273 if (hists
->stats
.nr_lost_warned
!=
274 hists
->stats
.nr_events
[PERF_RECORD_LOST
]) {
275 hists
->stats
.nr_lost_warned
=
276 hists
->stats
.nr_events
[PERF_RECORD_LOST
];
277 color_fprintf(stdout
, PERF_COLOR_RED
,
278 "WARNING: LOST %d chunks, Check IO/CPU overload",
279 hists
->stats
.nr_lost_warned
);
283 if (top
->sym_filter_entry
) {
284 perf_top__show_details(top
);
288 if (top
->evlist
->enabled
) {
290 hists__delete_entries(hists
);
292 hists__decay_entries(hists
, top
->hide_user_symbols
,
293 top
->hide_kernel_symbols
);
297 hists__collapse_resort(hists
, NULL
);
298 perf_evsel__output_resort(evsel
, NULL
);
300 hists__output_recalc_col_len(hists
, top
->print_entries
- printed
);
302 hists__fprintf(hists
, false, top
->print_entries
- printed
, win_width
,
303 top
->min_percent
, stdout
, symbol_conf
.use_callchain
);
306 static void prompt_integer(int *target
, const char *msg
)
308 char *buf
= malloc(0), *p
;
312 fprintf(stdout
, "\n%s: ", msg
);
313 if (getline(&buf
, &dummy
, stdin
) < 0)
316 p
= strchr(buf
, '\n');
326 tmp
= strtoul(buf
, NULL
, 10);
332 static void prompt_percent(int *target
, const char *msg
)
336 prompt_integer(&tmp
, msg
);
337 if (tmp
>= 0 && tmp
<= 100)
341 static void perf_top__prompt_symbol(struct perf_top
*top
, const char *msg
)
343 char *buf
= malloc(0), *p
;
344 struct hist_entry
*syme
= top
->sym_filter_entry
, *n
, *found
= NULL
;
345 struct hists
*hists
= evsel__hists(top
->sym_evsel
);
346 struct rb_node
*next
;
349 /* zero counters of active symbol */
351 __zero_source_counters(syme
);
352 top
->sym_filter_entry
= NULL
;
355 fprintf(stdout
, "\n%s: ", msg
);
356 if (getline(&buf
, &dummy
, stdin
) < 0)
359 p
= strchr(buf
, '\n');
363 next
= rb_first(&hists
->entries
);
365 n
= rb_entry(next
, struct hist_entry
, rb_node
);
366 if (n
->ms
.sym
&& !strcmp(buf
, n
->ms
.sym
->name
)) {
370 next
= rb_next(&n
->rb_node
);
374 fprintf(stderr
, "Sorry, %s is not active.\n", buf
);
377 perf_top__parse_source(top
, found
);
383 static void perf_top__print_mapped_keys(struct perf_top
*top
)
387 if (top
->sym_filter_entry
) {
388 struct symbol
*sym
= top
->sym_filter_entry
->ms
.sym
;
392 fprintf(stdout
, "\nMapped keys:\n");
393 fprintf(stdout
, "\t[d] display refresh delay. \t(%d)\n", top
->delay_secs
);
394 fprintf(stdout
, "\t[e] display entries (lines). \t(%d)\n", top
->print_entries
);
396 if (top
->evlist
->nr_entries
> 1)
397 fprintf(stdout
, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top
->sym_evsel
));
399 fprintf(stdout
, "\t[f] profile display filter (count). \t(%d)\n", top
->count_filter
);
401 fprintf(stdout
, "\t[F] annotate display filter (percent). \t(%d%%)\n", top
->sym_pcnt_filter
);
402 fprintf(stdout
, "\t[s] annotate symbol. \t(%s)\n", name
?: "NULL");
403 fprintf(stdout
, "\t[S] stop annotation.\n");
406 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
407 top
->hide_kernel_symbols
? "yes" : "no");
409 "\t[U] hide user symbols. \t(%s)\n",
410 top
->hide_user_symbols
? "yes" : "no");
411 fprintf(stdout
, "\t[z] toggle sample zeroing. \t(%d)\n", top
->zero
? 1 : 0);
412 fprintf(stdout
, "\t[qQ] quit.\n");
415 static int perf_top__key_mapped(struct perf_top
*top
, int c
)
431 return top
->evlist
->nr_entries
> 1 ? 1 : 0;
439 static bool perf_top__handle_keypress(struct perf_top
*top
, int c
)
443 if (!perf_top__key_mapped(top
, c
)) {
444 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
447 perf_top__print_mapped_keys(top
);
448 fprintf(stdout
, "\nEnter selection, or unmapped key to continue: ");
451 set_term_quiet_input(&save
);
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 evlist__for_each_entry(top
->evlist
, top
->sym_evsel
)
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 evlist__for_each_entry(top
->evlist
, top
->sym_evsel
)
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 struct perf_evsel
*evsel
= t
->sym_evsel
;
553 perf_top__reset_sample_counters(t
);
555 if (t
->evlist
->selected
!= NULL
)
556 t
->sym_evsel
= t
->evlist
->selected
;
558 hists
= evsel__hists(evsel
);
560 if (t
->evlist
->enabled
) {
562 hists__delete_entries(hists
);
564 hists__decay_entries(hists
, t
->hide_user_symbols
,
565 t
->hide_kernel_symbols
);
569 hists__collapse_resort(hists
, NULL
);
570 perf_evsel__output_resort(evsel
, NULL
);
573 static void *display_thread_tui(void *arg
)
575 struct perf_evsel
*pos
;
576 struct perf_top
*top
= arg
;
577 const char *help
= "For a higher level overview, try: perf top --sort comm,dso";
578 struct hist_browser_timer hbt
= {
579 .timer
= perf_top__sort_new_samples
,
581 .refresh
= top
->delay_secs
,
584 perf_top__sort_new_samples(top
);
587 * Initialize the uid_filter_str, in the future the TUI will allow
588 * Zooming in/out UIDs. For now juse use whatever the user passed
591 evlist__for_each_entry(top
->evlist
, pos
) {
592 struct hists
*hists
= evsel__hists(pos
);
593 hists
->uid_filter_str
= top
->record_opts
.target
.uid_str
;
596 perf_evlist__tui_browse_hists(top
->evlist
, help
, &hbt
,
598 &top
->session
->header
.env
);
604 static void display_sig(int sig __maybe_unused
)
609 static void display_setup_sig(void)
611 signal(SIGSEGV
, sighandler_dump_stack
);
612 signal(SIGFPE
, sighandler_dump_stack
);
613 signal(SIGINT
, display_sig
);
614 signal(SIGQUIT
, display_sig
);
615 signal(SIGTERM
, display_sig
);
618 static void *display_thread(void *arg
)
620 struct pollfd stdin_poll
= { .fd
= 0, .events
= POLLIN
};
622 struct perf_top
*top
= arg
;
626 pthread__unblock_sigwinch();
628 delay_msecs
= top
->delay_secs
* MSEC_PER_SEC
;
629 set_term_quiet_input(&save
);
634 perf_top__print_sym_table(top
);
636 * Either timeout expired or we got an EINTR due to SIGWINCH,
637 * refresh screen in both cases.
639 switch (poll(&stdin_poll
, 1, delay_msecs
)) {
648 tcsetattr(0, TCSAFLUSH
, &save
);
650 if (perf_top__handle_keypress(top
, c
))
656 tcsetattr(0, TCSAFLUSH
, &save
);
660 static int hist_iter__top_callback(struct hist_entry_iter
*iter
,
661 struct addr_location
*al
, bool single
,
664 struct perf_top
*top
= arg
;
665 struct hist_entry
*he
= iter
->he
;
666 struct perf_evsel
*evsel
= iter
->evsel
;
668 if (perf_hpp_list
.sym
&& single
)
669 perf_top__record_precise_ip(top
, he
, evsel
->idx
, al
->addr
);
671 hist__account_cycles(iter
->sample
->branch_stack
, al
, iter
->sample
,
672 !(top
->record_opts
.branch_stack
& PERF_SAMPLE_BRANCH_ANY
));
676 static void perf_event__process_sample(struct perf_tool
*tool
,
677 const union perf_event
*event
,
678 struct perf_evsel
*evsel
,
679 struct perf_sample
*sample
,
680 struct machine
*machine
)
682 struct perf_top
*top
= container_of(tool
, struct perf_top
, tool
);
683 struct addr_location al
;
686 if (!machine
&& perf_guest
) {
687 static struct intlist
*seen
;
690 seen
= intlist__new(NULL
);
692 if (!intlist__has_entry(seen
, sample
->pid
)) {
693 pr_err("Can't find guest [%d]'s kernel information\n",
695 intlist__add(seen
, sample
->pid
);
701 pr_err("%u unprocessable samples recorded.\r",
702 top
->session
->evlist
->stats
.nr_unprocessable_samples
++);
706 if (event
->header
.misc
& PERF_RECORD_MISC_EXACT_IP
)
707 top
->exact_samples
++;
709 if (machine__resolve(machine
, &al
, sample
) < 0)
712 if (!machine
->kptr_restrict_warned
&&
713 symbol_conf
.kptr_restrict
&&
714 al
.cpumode
== PERF_RECORD_MISC_KERNEL
) {
716 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
717 "Check /proc/sys/kernel/kptr_restrict.\n\n"
718 "Kernel%s samples will not be resolved.\n",
719 al
.map
&& !RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
]) ?
721 if (use_browser
<= 0)
723 machine
->kptr_restrict_warned
= true;
726 if (al
.sym
== NULL
) {
727 const char *msg
= "Kernel samples will not be resolved.\n";
729 * As we do lazy loading of symtabs we only will know if the
730 * specified vmlinux file is invalid when we actually have a
731 * hit in kernel space and then try to load it. So if we get
732 * here and there are _no_ symbols in the DSO backing the
733 * kernel map, bail out.
735 * We may never get here, for instance, if we use -K/
736 * --hide-kernel-symbols, even if the user specifies an
737 * invalid --vmlinux ;-)
739 if (!machine
->kptr_restrict_warned
&& !top
->vmlinux_warned
&&
740 al
.map
== machine
->vmlinux_maps
[MAP__FUNCTION
] &&
741 RB_EMPTY_ROOT(&al
.map
->dso
->symbols
[MAP__FUNCTION
])) {
742 if (symbol_conf
.vmlinux_name
) {
744 dso__strerror_load(al
.map
->dso
, serr
, sizeof(serr
));
745 ui__warning("The %s file can't be used: %s\n%s",
746 symbol_conf
.vmlinux_name
, serr
, msg
);
748 ui__warning("A vmlinux file was not found.\n%s",
752 if (use_browser
<= 0)
754 top
->vmlinux_warned
= true;
758 if (al
.sym
== NULL
|| !al
.sym
->idle
) {
759 struct hists
*hists
= evsel__hists(evsel
);
760 struct hist_entry_iter iter
= {
763 .add_entry_cb
= hist_iter__top_callback
,
766 if (symbol_conf
.cumulate_callchain
)
767 iter
.ops
= &hist_iter_cumulative
;
769 iter
.ops
= &hist_iter_normal
;
771 pthread_mutex_lock(&hists
->lock
);
773 err
= hist_entry_iter__add(&iter
, &al
, top
->max_stack
, top
);
775 pr_err("Problem incrementing symbol period, skipping event\n");
777 pthread_mutex_unlock(&hists
->lock
);
780 addr_location__put(&al
);
783 static void perf_top__mmap_read_idx(struct perf_top
*top
, int idx
)
785 struct perf_sample sample
;
786 struct perf_evsel
*evsel
;
787 struct perf_session
*session
= top
->session
;
788 union perf_event
*event
;
789 struct machine
*machine
;
792 while ((event
= perf_evlist__mmap_read(top
->evlist
, idx
)) != NULL
) {
793 ret
= perf_evlist__parse_sample(top
->evlist
, event
, &sample
);
795 pr_err("Can't parse sample, err = %d\n", ret
);
799 evsel
= perf_evlist__id2evsel(session
->evlist
, sample
.id
);
800 assert(evsel
!= NULL
);
802 if (event
->header
.type
== PERF_RECORD_SAMPLE
)
805 switch (sample
.cpumode
) {
806 case PERF_RECORD_MISC_USER
:
808 if (top
->hide_user_symbols
)
810 machine
= &session
->machines
.host
;
812 case PERF_RECORD_MISC_KERNEL
:
813 ++top
->kernel_samples
;
814 if (top
->hide_kernel_symbols
)
816 machine
= &session
->machines
.host
;
818 case PERF_RECORD_MISC_GUEST_KERNEL
:
819 ++top
->guest_kernel_samples
;
820 machine
= perf_session__find_machine(session
,
823 case PERF_RECORD_MISC_GUEST_USER
:
824 ++top
->guest_us_samples
;
826 * TODO: we don't process guest user from host side
827 * except simple counting.
831 if (event
->header
.type
== PERF_RECORD_SAMPLE
)
833 machine
= &session
->machines
.host
;
838 if (event
->header
.type
== PERF_RECORD_SAMPLE
) {
839 perf_event__process_sample(&top
->tool
, event
, evsel
,
841 } else if (event
->header
.type
< PERF_RECORD_MAX
) {
842 hists__inc_nr_events(evsel__hists(evsel
), event
->header
.type
);
843 machine__process_event(machine
, event
, &sample
);
845 ++session
->evlist
->stats
.nr_unknown_events
;
847 perf_evlist__mmap_consume(top
->evlist
, idx
);
851 static void perf_top__mmap_read(struct perf_top
*top
)
855 for (i
= 0; i
< top
->evlist
->nr_mmaps
; i
++)
856 perf_top__mmap_read_idx(top
, i
);
859 static int perf_top__start_counters(struct perf_top
*top
)
862 struct perf_evsel
*counter
;
863 struct perf_evlist
*evlist
= top
->evlist
;
864 struct record_opts
*opts
= &top
->record_opts
;
866 perf_evlist__config(evlist
, opts
, &callchain_param
);
868 evlist__for_each_entry(evlist
, counter
) {
870 if (perf_evsel__open(counter
, top
->evlist
->cpus
,
871 top
->evlist
->threads
) < 0) {
872 if (perf_evsel__fallback(counter
, errno
, msg
, sizeof(msg
))) {
874 ui__warning("%s\n", msg
);
878 perf_evsel__open_strerror(counter
, &opts
->target
,
879 errno
, msg
, sizeof(msg
));
880 ui__error("%s\n", msg
);
885 if (perf_evlist__mmap(evlist
, opts
->mmap_pages
, false) < 0) {
886 ui__error("Failed to mmap with %d (%s)\n",
887 errno
, str_error_r(errno
, msg
, sizeof(msg
)));
897 static int callchain_param__setup_sample_type(struct callchain_param
*callchain
)
899 if (!perf_hpp_list
.sym
) {
900 if (callchain
->enabled
) {
901 ui__error("Selected -g but \"sym\" not present in --sort/-s.");
904 } else if (callchain
->mode
!= CHAIN_NONE
) {
905 if (callchain_register_param(callchain
) < 0) {
906 ui__error("Can't register callchain params.\n");
914 static int __cmd_top(struct perf_top
*top
)
916 struct record_opts
*opts
= &top
->record_opts
;
920 top
->session
= perf_session__new(NULL
, false, NULL
);
921 if (top
->session
== NULL
)
925 ret
= perf_env__lookup_objdump(&top
->session
->header
.env
);
930 ret
= callchain_param__setup_sample_type(&callchain_param
);
934 if (perf_session__register_idle_thread(top
->session
) < 0)
937 machine__synthesize_threads(&top
->session
->machines
.host
, &opts
->target
,
938 top
->evlist
->threads
, false, opts
->proc_map_timeout
);
940 if (perf_hpp_list
.socket
) {
941 ret
= perf_env__read_cpu_topology_map(&perf_env
);
943 goto out_err_cpu_topo
;
946 ret
= perf_top__start_counters(top
);
950 top
->session
->evlist
= top
->evlist
;
951 perf_session__set_id_hdr_size(top
->session
);
954 * When perf is starting the traced process, all the events (apart from
955 * group members) have enable_on_exec=1 set, so don't spoil it by
956 * prematurely enabling them.
958 * XXX 'top' still doesn't start workloads like record, trace, but should,
959 * so leave the check here.
961 if (!target__none(&opts
->target
))
962 perf_evlist__enable(top
->evlist
);
964 /* Wait for a minimal set of events before starting the snapshot */
965 perf_evlist__poll(top
->evlist
, 100);
967 perf_top__mmap_read(top
);
970 if (pthread_create(&thread
, NULL
, (use_browser
> 0 ? display_thread_tui
:
971 display_thread
), top
)) {
972 ui__error("Could not create display thread.\n");
976 if (top
->realtime_prio
) {
977 struct sched_param param
;
979 param
.sched_priority
= top
->realtime_prio
;
980 if (sched_setscheduler(0, SCHED_FIFO
, ¶m
)) {
981 ui__error("Could not set realtime priority.\n");
987 u64 hits
= top
->samples
;
989 perf_top__mmap_read(top
);
991 if (hits
== top
->samples
)
992 ret
= perf_evlist__poll(top
->evlist
, 100);
997 pthread_join(thread
, NULL
);
999 perf_session__delete(top
->session
);
1000 top
->session
= NULL
;
1005 char errbuf
[BUFSIZ
];
1006 const char *err
= str_error_r(-ret
, errbuf
, sizeof(errbuf
));
1008 ui__error("Could not read the CPU topology map: %s\n", err
);
1014 callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1016 symbol_conf
.use_callchain
= true;
1017 return record_callchain_opt(opt
, arg
, unset
);
1021 parse_callchain_opt(const struct option
*opt
, const char *arg
, int unset
)
1023 struct callchain_param
*callchain
= opt
->value
;
1025 callchain
->enabled
= !unset
;
1026 callchain
->record_mode
= CALLCHAIN_FP
;
1032 symbol_conf
.use_callchain
= false;
1033 callchain
->record_mode
= CALLCHAIN_NONE
;
1037 return parse_callchain_top_opt(arg
);
1040 static int perf_top_config(const char *var
, const char *value
, void *cb __maybe_unused
)
1042 if (!strcmp(var
, "top.call-graph"))
1043 var
= "call-graph.record-mode"; /* fall-through */
1044 if (!strcmp(var
, "top.children")) {
1045 symbol_conf
.cumulate_callchain
= perf_config_bool(var
, value
);
1053 parse_percent_limit(const struct option
*opt
, const char *arg
,
1054 int unset __maybe_unused
)
1056 struct perf_top
*top
= opt
->value
;
1058 top
->min_percent
= strtof(arg
, NULL
);
1062 const char top_callchain_help
[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1063 "\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1065 int cmd_top(int argc
, const char **argv
, const char *prefix __maybe_unused
)
1067 char errbuf
[BUFSIZ
];
1068 struct perf_top top
= {
1072 .mmap_pages
= UINT_MAX
,
1073 .user_freq
= UINT_MAX
,
1074 .user_interval
= ULLONG_MAX
,
1075 .freq
= 4000, /* 4 KHz */
1079 .proc_map_timeout
= 500,
1081 .max_stack
= sysctl_perf_event_max_stack
,
1082 .sym_pcnt_filter
= 5,
1084 struct record_opts
*opts
= &top
.record_opts
;
1085 struct target
*target
= &opts
->target
;
1086 const struct option options
[] = {
1087 OPT_CALLBACK('e', "event", &top
.evlist
, "event",
1088 "event selector. use 'perf list' to list available events",
1089 parse_events_option
),
1090 OPT_U64('c', "count", &opts
->user_interval
, "event period to sample"),
1091 OPT_STRING('p', "pid", &target
->pid
, "pid",
1092 "profile events on existing process id"),
1093 OPT_STRING('t', "tid", &target
->tid
, "tid",
1094 "profile events on existing thread id"),
1095 OPT_BOOLEAN('a', "all-cpus", &target
->system_wide
,
1096 "system-wide collection from all CPUs"),
1097 OPT_STRING('C', "cpu", &target
->cpu_list
, "cpu",
1098 "list of cpus to monitor"),
1099 OPT_STRING('k', "vmlinux", &symbol_conf
.vmlinux_name
,
1100 "file", "vmlinux pathname"),
1101 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf
.ignore_vmlinux
,
1102 "don't load vmlinux even if found"),
1103 OPT_BOOLEAN('K', "hide_kernel_symbols", &top
.hide_kernel_symbols
,
1104 "hide kernel symbols"),
1105 OPT_CALLBACK('m', "mmap-pages", &opts
->mmap_pages
, "pages",
1106 "number of mmap data pages",
1107 perf_evlist__parse_mmap_pages
),
1108 OPT_INTEGER('r', "realtime", &top
.realtime_prio
,
1109 "collect data with this RT SCHED_FIFO priority"),
1110 OPT_INTEGER('d', "delay", &top
.delay_secs
,
1111 "number of seconds to delay between refreshes"),
1112 OPT_BOOLEAN('D', "dump-symtab", &top
.dump_symtab
,
1113 "dump the symbol table used for profiling"),
1114 OPT_INTEGER('f', "count-filter", &top
.count_filter
,
1115 "only display functions with more events than this"),
1116 OPT_BOOLEAN(0, "group", &opts
->group
,
1117 "put the counters into a counter group"),
1118 OPT_BOOLEAN('i', "no-inherit", &opts
->no_inherit
,
1119 "child tasks do not inherit counters"),
1120 OPT_STRING(0, "sym-annotate", &top
.sym_filter
, "symbol name",
1121 "symbol to annotate"),
1122 OPT_BOOLEAN('z', "zero", &top
.zero
, "zero history across updates"),
1123 OPT_UINTEGER('F', "freq", &opts
->user_freq
, "profile at this frequency"),
1124 OPT_INTEGER('E', "entries", &top
.print_entries
,
1125 "display this many functions"),
1126 OPT_BOOLEAN('U', "hide_user_symbols", &top
.hide_user_symbols
,
1127 "hide user symbols"),
1128 OPT_BOOLEAN(0, "tui", &top
.use_tui
, "Use the TUI interface"),
1129 OPT_BOOLEAN(0, "stdio", &top
.use_stdio
, "Use the stdio interface"),
1130 OPT_INCR('v', "verbose", &verbose
,
1131 "be more verbose (show counter open errors, etc)"),
1132 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1133 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1134 " Please refer the man page for the complete list."),
1135 OPT_STRING(0, "fields", &field_order
, "key[,keys...]",
1136 "output field(s): overhead, period, sample plus all of sort keys"),
1137 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf
.show_nr_samples
,
1138 "Show a column with the number of samples"),
1139 OPT_CALLBACK_NOOPT('g', NULL
, &callchain_param
,
1140 NULL
, "enables call-graph recording and display",
1142 OPT_CALLBACK(0, "call-graph", &callchain_param
,
1143 "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1144 top_callchain_help
, &parse_callchain_opt
),
1145 OPT_BOOLEAN(0, "children", &symbol_conf
.cumulate_callchain
,
1146 "Accumulate callchains of children and show total overhead as well"),
1147 OPT_INTEGER(0, "max-stack", &top
.max_stack
,
1148 "Set the maximum stack depth when parsing the callchain. "
1149 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH
)),
1150 OPT_CALLBACK(0, "ignore-callees", NULL
, "regex",
1151 "ignore callees of these functions in call graphs",
1152 report_parse_ignore_callees_opt
),
1153 OPT_BOOLEAN(0, "show-total-period", &symbol_conf
.show_total_period
,
1154 "Show a column with the sum of periods"),
1155 OPT_STRING(0, "dsos", &symbol_conf
.dso_list_str
, "dso[,dso...]",
1156 "only consider symbols in these dsos"),
1157 OPT_STRING(0, "comms", &symbol_conf
.comm_list_str
, "comm[,comm...]",
1158 "only consider symbols in these comms"),
1159 OPT_STRING(0, "symbols", &symbol_conf
.sym_list_str
, "symbol[,symbol...]",
1160 "only consider these symbols"),
1161 OPT_BOOLEAN(0, "source", &symbol_conf
.annotate_src
,
1162 "Interleave source code with assembly code (default)"),
1163 OPT_BOOLEAN(0, "asm-raw", &symbol_conf
.annotate_asm_raw
,
1164 "Display raw encoding of assembly instructions (default)"),
1165 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf
.demangle_kernel
,
1166 "Enable kernel symbol demangling"),
1167 OPT_STRING(0, "objdump", &objdump_path
, "path",
1168 "objdump binary to use for disassembly and annotations"),
1169 OPT_STRING('M', "disassembler-style", &disassembler_style
, "disassembler style",
1170 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1171 OPT_STRING('u', "uid", &target
->uid_str
, "user", "user to profile"),
1172 OPT_CALLBACK(0, "percent-limit", &top
, "percent",
1173 "Don't show entries under that percent", parse_percent_limit
),
1174 OPT_CALLBACK(0, "percentage", NULL
, "relative|absolute",
1175 "How to display percentage of filtered entries", parse_filter_percentage
),
1176 OPT_STRING('w', "column-widths", &symbol_conf
.col_width_list_str
,
1178 "don't try to adjust column width, use these fixed values"),
1179 OPT_UINTEGER(0, "proc-map-timeout", &opts
->proc_map_timeout
,
1180 "per thread proc mmap processing timeout in ms"),
1181 OPT_CALLBACK_NOOPT('b', "branch-any", &opts
->branch_stack
,
1182 "branch any", "sample any taken branches",
1183 parse_branch_stack
),
1184 OPT_CALLBACK('j', "branch-filter", &opts
->branch_stack
,
1185 "branch filter mask", "branch stack filter modes",
1186 parse_branch_stack
),
1187 OPT_BOOLEAN(0, "raw-trace", &symbol_conf
.raw_trace
,
1188 "Show raw trace event output (do not use print fmt or plugins)"),
1189 OPT_BOOLEAN(0, "hierarchy", &symbol_conf
.report_hierarchy
,
1190 "Show entries in a hierarchy"),
1193 const char * const top_usage
[] = {
1194 "perf top [<options>]",
1197 int status
= hists__init();
1202 top
.evlist
= perf_evlist__new();
1203 if (top
.evlist
== NULL
)
1206 perf_config(perf_top_config
, &top
);
1208 argc
= parse_options(argc
, argv
, options
, top_usage
, 0);
1210 usage_with_options(top_usage
, options
);
1212 if (!top
.evlist
->nr_entries
&&
1213 perf_evlist__add_default(top
.evlist
) < 0) {
1214 pr_err("Not enough memory for event selector list\n");
1215 goto out_delete_evlist
;
1218 if (symbol_conf
.report_hierarchy
) {
1219 /* disable incompatible options */
1220 symbol_conf
.event_group
= false;
1221 symbol_conf
.cumulate_callchain
= false;
1224 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1225 parse_options_usage(top_usage
, options
, "fields", 0);
1226 parse_options_usage(NULL
, options
, "hierarchy", 0);
1227 goto out_delete_evlist
;
1231 sort__mode
= SORT_MODE__TOP
;
1232 /* display thread wants entries to be collapsed in a different tree */
1233 perf_hpp_list
.need_collapse
= 1;
1237 else if (top
.use_tui
)
1240 setup_browser(false);
1242 if (setup_sorting(top
.evlist
) < 0) {
1244 parse_options_usage(top_usage
, options
, "s", 1);
1246 parse_options_usage(sort_order
? NULL
: top_usage
,
1247 options
, "fields", 0);
1248 goto out_delete_evlist
;
1251 status
= target__validate(target
);
1253 target__strerror(target
, status
, errbuf
, BUFSIZ
);
1254 ui__warning("%s\n", errbuf
);
1257 status
= target__parse_uid(target
);
1259 int saved_errno
= errno
;
1261 target__strerror(target
, status
, errbuf
, BUFSIZ
);
1262 ui__error("%s\n", errbuf
);
1264 status
= -saved_errno
;
1265 goto out_delete_evlist
;
1268 if (target__none(target
))
1269 target
->system_wide
= true;
1271 if (perf_evlist__create_maps(top
.evlist
, target
) < 0) {
1272 ui__error("Couldn't create thread/CPU maps: %s\n",
1273 errno
== ENOENT
? "No such process" : str_error_r(errno
, errbuf
, sizeof(errbuf
)));
1274 goto out_delete_evlist
;
1277 symbol_conf
.nr_events
= top
.evlist
->nr_entries
;
1279 if (top
.delay_secs
< 1)
1282 if (record_opts__config(opts
)) {
1284 goto out_delete_evlist
;
1287 top
.sym_evsel
= perf_evlist__first(top
.evlist
);
1289 if (!callchain_param
.enabled
) {
1290 symbol_conf
.cumulate_callchain
= false;
1291 perf_hpp__cancel_cumulate();
1294 if (symbol_conf
.cumulate_callchain
&& !callchain_param
.order_set
)
1295 callchain_param
.order
= ORDER_CALLER
;
1297 status
= symbol__annotation_init();
1299 goto out_delete_evlist
;
1301 symbol_conf
.try_vmlinux_path
= (symbol_conf
.vmlinux_name
== NULL
);
1302 if (symbol__init(NULL
) < 0)
1305 sort__setup_elide(stdout
);
1307 get_term_dimensions(&top
.winsize
);
1308 if (top
.print_entries
== 0) {
1309 struct sigaction act
= {
1310 .sa_sigaction
= perf_top__sig_winch
,
1311 .sa_flags
= SA_SIGINFO
,
1313 perf_top__update_print_entries(&top
);
1314 sigaction(SIGWINCH
, &act
, NULL
);
1317 status
= __cmd_top(&top
);
1320 perf_evlist__delete(top
.evlist
);