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/drv_configs.h"
28 #include "util/evlist.h"
29 #include "util/evsel.h"
30 #include "util/machine.h"
31 #include "util/session.h"
32 #include "util/symbol.h"
33 #include "util/thread.h"
34 #include "util/thread_map.h"
36 #include "util/util.h"
37 #include <linux/rbtree.h>
38 #include <subcmd/parse-options.h>
39 #include "util/parse-events.h"
40 #include "util/cpumap.h"
41 #include "util/xyarray.h"
42 #include "util/sort.h"
43 #include "util/term.h"
44 #include "util/intlist.h"
45 #include "util/parse-branch-options.h"
46 #include "arch/common.h"
48 #include "util/debug.h"
63 #include <sys/syscall.h>
64 #include <sys/ioctl.h>
66 #include <sys/prctl.h>
69 #include <sys/utsname.h>
72 #include <linux/stringify.h>
73 #include <linux/time64.h>
74 #include <linux/types.h>
76 #include "sane_ctype.h"
78 static volatile int done;
80 #define HEADER_LINE_NR 5
82 static void perf_top__update_print_entries(struct perf_top *top)
84 top->print_entries = top->winsize.ws_row - HEADER_LINE_NR;
87 static void perf_top__sig_winch(int sig __maybe_unused,
88 siginfo_t *info __maybe_unused, void *arg)
90 struct perf_top *top = arg;
92 get_term_dimensions(&top->winsize);
93 perf_top__update_print_entries(top);
96 static int perf_top__parse_source(struct perf_top *top, struct hist_entry *he)
99 struct annotation *notes;
103 if (!he || !he->ms.sym)
110 * We can't annotate with just /proc/kallsyms
112 if (map->dso->symtab_type == DSO_BINARY_TYPE__KALLSYMS &&
113 !dso__is_kcore(map->dso)) {
114 pr_err("Can't annotate %s: No vmlinux file was found in the "
115 "path\n", sym->name);
120 notes = symbol__annotation(sym);
121 if (notes->src != NULL) {
122 pthread_mutex_lock(¬es->lock);
126 pthread_mutex_lock(¬es->lock);
128 if (symbol__alloc_hist(sym) < 0) {
129 pthread_mutex_unlock(¬es->lock);
130 pr_err("Not enough memory for annotating '%s' symbol!\n",
136 err = symbol__disassemble(sym, map, NULL, 0);
139 top->sym_filter_entry = he;
142 symbol__strerror_disassemble(sym, map, err, msg, sizeof(msg));
143 pr_err("Couldn't annotate %s: %s\n", sym->name, msg);
146 pthread_mutex_unlock(¬es->lock);
150 static void __zero_source_counters(struct hist_entry *he)
152 struct symbol *sym = he->ms.sym;
153 symbol__annotate_zero_histograms(sym);
156 static void ui__warn_map_erange(struct map *map, struct symbol *sym, u64 ip)
159 int err = uname(&uts);
161 ui__warning("Out of bounds address found:\n\n"
162 "Addr: %" PRIx64 "\n"
164 "Map: %" PRIx64 "-%" PRIx64 "\n"
165 "Symbol: %" PRIx64 "-%" PRIx64 " %c %s\n"
169 "Not all samples will be on the annotation output.\n\n"
170 "Please report to linux-kernel@vger.kernel.org\n",
171 ip, map->dso->long_name, dso__symtab_origin(map->dso),
172 map->start, map->end, sym->start, sym->end,
173 sym->binding == STB_GLOBAL ? 'g' :
174 sym->binding == STB_LOCAL ? 'l' : 'w', sym->name,
175 err ? "[unknown]" : uts.machine,
176 err ? "[unknown]" : uts.release, perf_version_string);
177 if (use_browser <= 0)
180 map->erange_warned = true;
183 static void perf_top__record_precise_ip(struct perf_top *top,
184 struct hist_entry *he,
187 struct annotation *notes;
188 struct symbol *sym = he->ms.sym;
191 if (sym == NULL || (use_browser == 0 &&
192 (top->sym_filter_entry == NULL ||
193 top->sym_filter_entry->ms.sym != sym)))
196 notes = symbol__annotation(sym);
198 if (pthread_mutex_trylock(¬es->lock))
201 err = hist_entry__inc_addr_samples(he, counter, ip);
203 pthread_mutex_unlock(¬es->lock);
207 * This function is now called with he->hists->lock held.
208 * Release it before going to sleep.
210 pthread_mutex_unlock(&he->hists->lock);
212 if (err == -ERANGE && !he->ms.map->erange_warned)
213 ui__warn_map_erange(he->ms.map, sym, ip);
214 else if (err == -ENOMEM) {
215 pr_err("Not enough memory for annotating '%s' symbol!\n",
220 pthread_mutex_lock(&he->hists->lock);
224 static void perf_top__show_details(struct perf_top *top)
226 struct hist_entry *he = top->sym_filter_entry;
227 struct annotation *notes;
228 struct symbol *symbol;
235 notes = symbol__annotation(symbol);
237 pthread_mutex_lock(¬es->lock);
239 if (notes->src == NULL)
242 printf("Showing %s for %s\n", perf_evsel__name(top->sym_evsel), symbol->name);
243 printf(" Events Pcnt (>=%d%%)\n", top->sym_pcnt_filter);
245 more = symbol__annotate_printf(symbol, he->ms.map, top->sym_evsel,
246 0, top->sym_pcnt_filter, top->print_entries, 4);
248 if (top->evlist->enabled) {
250 symbol__annotate_zero_histogram(symbol, top->sym_evsel->idx);
252 symbol__annotate_decay_histogram(symbol, top->sym_evsel->idx);
255 printf("%d lines not displayed, maybe increase display entries [e]\n", more);
257 pthread_mutex_unlock(¬es->lock);
260 static void perf_top__print_sym_table(struct perf_top *top)
264 const int win_width = top->winsize.ws_col - 1;
265 struct perf_evsel *evsel = top->sym_evsel;
266 struct hists *hists = evsel__hists(evsel);
270 perf_top__header_snprintf(top, bf, sizeof(bf));
273 perf_top__reset_sample_counters(top);
275 printf("%-*.*s\n", win_width, win_width, graph_dotted_line);
277 if (hists->stats.nr_lost_warned !=
278 hists->stats.nr_events[PERF_RECORD_LOST]) {
279 hists->stats.nr_lost_warned =
280 hists->stats.nr_events[PERF_RECORD_LOST];
281 color_fprintf(stdout, PERF_COLOR_RED,
282 "WARNING: LOST %d chunks, Check IO/CPU overload",
283 hists->stats.nr_lost_warned);
287 if (top->sym_filter_entry) {
288 perf_top__show_details(top);
292 if (top->evlist->enabled) {
294 hists__delete_entries(hists);
296 hists__decay_entries(hists, top->hide_user_symbols,
297 top->hide_kernel_symbols);
301 hists__collapse_resort(hists, NULL);
302 perf_evsel__output_resort(evsel, NULL);
304 hists__output_recalc_col_len(hists, top->print_entries - printed);
306 hists__fprintf(hists, false, top->print_entries - printed, win_width,
307 top->min_percent, stdout, symbol_conf.use_callchain);
310 static void prompt_integer(int *target, const char *msg)
312 char *buf = malloc(0), *p;
316 fprintf(stdout, "\n%s: ", msg);
317 if (getline(&buf, &dummy, stdin) < 0)
320 p = strchr(buf, '\n');
330 tmp = strtoul(buf, NULL, 10);
336 static void prompt_percent(int *target, const char *msg)
340 prompt_integer(&tmp, msg);
341 if (tmp >= 0 && tmp <= 100)
345 static void perf_top__prompt_symbol(struct perf_top *top, const char *msg)
347 char *buf = malloc(0), *p;
348 struct hist_entry *syme = top->sym_filter_entry, *n, *found = NULL;
349 struct hists *hists = evsel__hists(top->sym_evsel);
350 struct rb_node *next;
353 /* zero counters of active symbol */
355 __zero_source_counters(syme);
356 top->sym_filter_entry = NULL;
359 fprintf(stdout, "\n%s: ", msg);
360 if (getline(&buf, &dummy, stdin) < 0)
363 p = strchr(buf, '\n');
367 next = rb_first(&hists->entries);
369 n = rb_entry(next, struct hist_entry, rb_node);
370 if (n->ms.sym && !strcmp(buf, n->ms.sym->name)) {
374 next = rb_next(&n->rb_node);
378 fprintf(stderr, "Sorry, %s is not active.\n", buf);
381 perf_top__parse_source(top, found);
387 static void perf_top__print_mapped_keys(struct perf_top *top)
391 if (top->sym_filter_entry) {
392 struct symbol *sym = top->sym_filter_entry->ms.sym;
396 fprintf(stdout, "\nMapped keys:\n");
397 fprintf(stdout, "\t[d] display refresh delay. \t(%d)\n", top->delay_secs);
398 fprintf(stdout, "\t[e] display entries (lines). \t(%d)\n", top->print_entries);
400 if (top->evlist->nr_entries > 1)
401 fprintf(stdout, "\t[E] active event counter. \t(%s)\n", perf_evsel__name(top->sym_evsel));
403 fprintf(stdout, "\t[f] profile display filter (count). \t(%d)\n", top->count_filter);
405 fprintf(stdout, "\t[F] annotate display filter (percent). \t(%d%%)\n", top->sym_pcnt_filter);
406 fprintf(stdout, "\t[s] annotate symbol. \t(%s)\n", name?: "NULL");
407 fprintf(stdout, "\t[S] stop annotation.\n");
410 "\t[K] hide kernel_symbols symbols. \t(%s)\n",
411 top->hide_kernel_symbols ? "yes" : "no");
413 "\t[U] hide user symbols. \t(%s)\n",
414 top->hide_user_symbols ? "yes" : "no");
415 fprintf(stdout, "\t[z] toggle sample zeroing. \t(%d)\n", top->zero ? 1 : 0);
416 fprintf(stdout, "\t[qQ] quit.\n");
419 static int perf_top__key_mapped(struct perf_top *top, int c)
435 return top->evlist->nr_entries > 1 ? 1 : 0;
443 static bool perf_top__handle_keypress(struct perf_top *top, int c)
447 if (!perf_top__key_mapped(top, c)) {
448 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
451 perf_top__print_mapped_keys(top);
452 fprintf(stdout, "\nEnter selection, or unmapped key to continue: ");
455 set_term_quiet_input(&save);
457 poll(&stdin_poll, 1, -1);
460 tcsetattr(0, TCSAFLUSH, &save);
461 if (!perf_top__key_mapped(top, c))
467 prompt_integer(&top->delay_secs, "Enter display delay");
468 if (top->delay_secs < 1)
472 prompt_integer(&top->print_entries, "Enter display entries (lines)");
473 if (top->print_entries == 0) {
474 struct sigaction act = {
475 .sa_sigaction = perf_top__sig_winch,
476 .sa_flags = SA_SIGINFO,
478 perf_top__sig_winch(SIGWINCH, NULL, top);
479 sigaction(SIGWINCH, &act, NULL);
481 signal(SIGWINCH, SIG_DFL);
485 if (top->evlist->nr_entries > 1) {
486 /* Select 0 as the default event: */
489 fprintf(stderr, "\nAvailable events:");
491 evlist__for_each_entry(top->evlist, top->sym_evsel)
492 fprintf(stderr, "\n\t%d %s", top->sym_evsel->idx, perf_evsel__name(top->sym_evsel));
494 prompt_integer(&counter, "Enter details event counter");
496 if (counter >= top->evlist->nr_entries) {
497 top->sym_evsel = perf_evlist__first(top->evlist);
498 fprintf(stderr, "Sorry, no such event, using %s.\n", perf_evsel__name(top->sym_evsel));
502 evlist__for_each_entry(top->evlist, top->sym_evsel)
503 if (top->sym_evsel->idx == counter)
506 top->sym_evsel = perf_evlist__first(top->evlist);
509 prompt_integer(&top->count_filter, "Enter display event count filter");
512 prompt_percent(&top->sym_pcnt_filter,
513 "Enter details display event filter (percent)");
516 top->hide_kernel_symbols = !top->hide_kernel_symbols;
520 printf("exiting.\n");
521 if (top->dump_symtab)
522 perf_session__fprintf_dsos(top->session, stderr);
526 perf_top__prompt_symbol(top, "Enter details symbol");
529 if (!top->sym_filter_entry)
532 struct hist_entry *syme = top->sym_filter_entry;
534 top->sym_filter_entry = NULL;
535 __zero_source_counters(syme);
539 top->hide_user_symbols = !top->hide_user_symbols;
542 top->zero = !top->zero;
551 static void perf_top__sort_new_samples(void *arg)
553 struct perf_top *t = arg;
554 struct perf_evsel *evsel = t->sym_evsel;
557 perf_top__reset_sample_counters(t);
559 if (t->evlist->selected != NULL)
560 t->sym_evsel = t->evlist->selected;
562 hists = evsel__hists(evsel);
564 if (t->evlist->enabled) {
566 hists__delete_entries(hists);
568 hists__decay_entries(hists, t->hide_user_symbols,
569 t->hide_kernel_symbols);
573 hists__collapse_resort(hists, NULL);
574 perf_evsel__output_resort(evsel, NULL);
577 static void *display_thread_tui(void *arg)
579 struct perf_evsel *pos;
580 struct perf_top *top = arg;
581 const char *help = "For a higher level overview, try: perf top --sort comm,dso";
582 struct hist_browser_timer hbt = {
583 .timer = perf_top__sort_new_samples,
585 .refresh = top->delay_secs,
588 perf_top__sort_new_samples(top);
591 * Initialize the uid_filter_str, in the future the TUI will allow
592 * Zooming in/out UIDs. For now juse use whatever the user passed
595 evlist__for_each_entry(top->evlist, pos) {
596 struct hists *hists = evsel__hists(pos);
597 hists->uid_filter_str = top->record_opts.target.uid_str;
600 perf_evlist__tui_browse_hists(top->evlist, help, &hbt,
602 &top->session->header.env);
608 static void display_sig(int sig __maybe_unused)
613 static void display_setup_sig(void)
615 signal(SIGSEGV, sighandler_dump_stack);
616 signal(SIGFPE, sighandler_dump_stack);
617 signal(SIGINT, display_sig);
618 signal(SIGQUIT, display_sig);
619 signal(SIGTERM, display_sig);
622 static void *display_thread(void *arg)
624 struct pollfd stdin_poll = { .fd = 0, .events = POLLIN };
626 struct perf_top *top = arg;
630 pthread__unblock_sigwinch();
632 delay_msecs = top->delay_secs * MSEC_PER_SEC;
633 set_term_quiet_input(&save);
638 perf_top__print_sym_table(top);
640 * Either timeout expired or we got an EINTR due to SIGWINCH,
641 * refresh screen in both cases.
643 switch (poll(&stdin_poll, 1, delay_msecs)) {
652 tcsetattr(0, TCSAFLUSH, &save);
654 if (perf_top__handle_keypress(top, c))
660 tcsetattr(0, TCSAFLUSH, &save);
664 static int hist_iter__top_callback(struct hist_entry_iter *iter,
665 struct addr_location *al, bool single,
668 struct perf_top *top = arg;
669 struct hist_entry *he = iter->he;
670 struct perf_evsel *evsel = iter->evsel;
672 if (perf_hpp_list.sym && single)
673 perf_top__record_precise_ip(top, he, evsel->idx, al->addr);
675 hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
676 !(top->record_opts.branch_stack & PERF_SAMPLE_BRANCH_ANY));
680 static void perf_event__process_sample(struct perf_tool *tool,
681 const union perf_event *event,
682 struct perf_evsel *evsel,
683 struct perf_sample *sample,
684 struct machine *machine)
686 struct perf_top *top = container_of(tool, struct perf_top, tool);
687 struct addr_location al;
690 if (!machine && perf_guest) {
691 static struct intlist *seen;
694 seen = intlist__new(NULL);
696 if (!intlist__has_entry(seen, sample->pid)) {
697 pr_err("Can't find guest [%d]'s kernel information\n",
699 intlist__add(seen, sample->pid);
705 pr_err("%u unprocessable samples recorded.\r",
706 top->session->evlist->stats.nr_unprocessable_samples++);
710 if (event->header.misc & PERF_RECORD_MISC_EXACT_IP)
711 top->exact_samples++;
713 if (machine__resolve(machine, &al, sample) < 0)
716 if (!machine->kptr_restrict_warned &&
717 symbol_conf.kptr_restrict &&
718 al.cpumode == PERF_RECORD_MISC_KERNEL) {
720 "Kernel address maps (/proc/{kallsyms,modules}) are restricted.\n\n"
721 "Check /proc/sys/kernel/kptr_restrict.\n\n"
722 "Kernel%s samples will not be resolved.\n",
723 al.map && !RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION]) ?
725 if (use_browser <= 0)
727 machine->kptr_restrict_warned = true;
730 if (al.sym == NULL) {
731 const char *msg = "Kernel samples will not be resolved.\n";
733 * As we do lazy loading of symtabs we only will know if the
734 * specified vmlinux file is invalid when we actually have a
735 * hit in kernel space and then try to load it. So if we get
736 * here and there are _no_ symbols in the DSO backing the
737 * kernel map, bail out.
739 * We may never get here, for instance, if we use -K/
740 * --hide-kernel-symbols, even if the user specifies an
741 * invalid --vmlinux ;-)
743 if (!machine->kptr_restrict_warned && !top->vmlinux_warned &&
744 al.map == machine->vmlinux_maps[MAP__FUNCTION] &&
745 RB_EMPTY_ROOT(&al.map->dso->symbols[MAP__FUNCTION])) {
746 if (symbol_conf.vmlinux_name) {
748 dso__strerror_load(al.map->dso, serr, sizeof(serr));
749 ui__warning("The %s file can't be used: %s\n%s",
750 symbol_conf.vmlinux_name, serr, msg);
752 ui__warning("A vmlinux file was not found.\n%s",
756 if (use_browser <= 0)
758 top->vmlinux_warned = true;
762 if (al.sym == NULL || !al.sym->idle) {
763 struct hists *hists = evsel__hists(evsel);
764 struct hist_entry_iter iter = {
767 .add_entry_cb = hist_iter__top_callback,
770 if (symbol_conf.cumulate_callchain)
771 iter.ops = &hist_iter_cumulative;
773 iter.ops = &hist_iter_normal;
775 pthread_mutex_lock(&hists->lock);
777 err = hist_entry_iter__add(&iter, &al, top->max_stack, top);
779 pr_err("Problem incrementing symbol period, skipping event\n");
781 pthread_mutex_unlock(&hists->lock);
784 addr_location__put(&al);
787 static void perf_top__mmap_read_idx(struct perf_top *top, int idx)
789 struct perf_sample sample;
790 struct perf_evsel *evsel;
791 struct perf_session *session = top->session;
792 union perf_event *event;
793 struct machine *machine;
796 while ((event = perf_evlist__mmap_read(top->evlist, idx)) != NULL) {
797 ret = perf_evlist__parse_sample(top->evlist, event, &sample);
799 pr_err("Can't parse sample, err = %d\n", ret);
803 evsel = perf_evlist__id2evsel(session->evlist, sample.id);
804 assert(evsel != NULL);
806 if (event->header.type == PERF_RECORD_SAMPLE)
809 switch (sample.cpumode) {
810 case PERF_RECORD_MISC_USER:
812 if (top->hide_user_symbols)
814 machine = &session->machines.host;
816 case PERF_RECORD_MISC_KERNEL:
817 ++top->kernel_samples;
818 if (top->hide_kernel_symbols)
820 machine = &session->machines.host;
822 case PERF_RECORD_MISC_GUEST_KERNEL:
823 ++top->guest_kernel_samples;
824 machine = perf_session__find_machine(session,
827 case PERF_RECORD_MISC_GUEST_USER:
828 ++top->guest_us_samples;
830 * TODO: we don't process guest user from host side
831 * except simple counting.
835 if (event->header.type == PERF_RECORD_SAMPLE)
837 machine = &session->machines.host;
842 if (event->header.type == PERF_RECORD_SAMPLE) {
843 perf_event__process_sample(&top->tool, event, evsel,
845 } else if (event->header.type < PERF_RECORD_MAX) {
846 hists__inc_nr_events(evsel__hists(evsel), event->header.type);
847 machine__process_event(machine, event, &sample);
849 ++session->evlist->stats.nr_unknown_events;
851 perf_evlist__mmap_consume(top->evlist, idx);
855 static void perf_top__mmap_read(struct perf_top *top)
859 for (i = 0; i < top->evlist->nr_mmaps; i++)
860 perf_top__mmap_read_idx(top, i);
863 static int perf_top__start_counters(struct perf_top *top)
866 struct perf_evsel *counter;
867 struct perf_evlist *evlist = top->evlist;
868 struct record_opts *opts = &top->record_opts;
870 perf_evlist__config(evlist, opts, &callchain_param);
872 evlist__for_each_entry(evlist, counter) {
874 if (perf_evsel__open(counter, top->evlist->cpus,
875 top->evlist->threads) < 0) {
876 if (perf_evsel__fallback(counter, errno, msg, sizeof(msg))) {
878 ui__warning("%s\n", msg);
882 perf_evsel__open_strerror(counter, &opts->target,
883 errno, msg, sizeof(msg));
884 ui__error("%s\n", msg);
889 if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
890 ui__error("Failed to mmap with %d (%s)\n",
891 errno, str_error_r(errno, msg, sizeof(msg)));
901 static int callchain_param__setup_sample_type(struct callchain_param *callchain)
903 if (!perf_hpp_list.sym) {
904 if (callchain->enabled) {
905 ui__error("Selected -g but \"sym\" not present in --sort/-s.");
908 } else if (callchain->mode != CHAIN_NONE) {
909 if (callchain_register_param(callchain) < 0) {
910 ui__error("Can't register callchain params.\n");
918 static int __cmd_top(struct perf_top *top)
921 struct perf_evsel *pos;
922 struct perf_evsel_config_term *err_term;
923 struct perf_evlist *evlist = top->evlist;
924 struct record_opts *opts = &top->record_opts;
928 top->session = perf_session__new(NULL, false, NULL);
929 if (top->session == NULL)
933 ret = perf_env__lookup_objdump(&top->session->header.env);
938 ret = callchain_param__setup_sample_type(&callchain_param);
942 if (perf_session__register_idle_thread(top->session) < 0)
945 machine__synthesize_threads(&top->session->machines.host, &opts->target,
946 top->evlist->threads, false, opts->proc_map_timeout);
948 if (perf_hpp_list.socket) {
949 ret = perf_env__read_cpu_topology_map(&perf_env);
951 goto out_err_cpu_topo;
954 ret = perf_top__start_counters(top);
958 ret = perf_evlist__apply_drv_configs(evlist, &pos, &err_term);
960 error("failed to set config \"%s\" on event %s with %d (%s)\n",
961 err_term->val.drv_cfg, perf_evsel__name(pos), errno,
962 str_error_r(errno, msg, sizeof(msg)));
966 top->session->evlist = top->evlist;
967 perf_session__set_id_hdr_size(top->session);
970 * When perf is starting the traced process, all the events (apart from
971 * group members) have enable_on_exec=1 set, so don't spoil it by
972 * prematurely enabling them.
974 * XXX 'top' still doesn't start workloads like record, trace, but should,
975 * so leave the check here.
977 if (!target__none(&opts->target))
978 perf_evlist__enable(top->evlist);
980 /* Wait for a minimal set of events before starting the snapshot */
981 perf_evlist__poll(top->evlist, 100);
983 perf_top__mmap_read(top);
986 if (pthread_create(&thread, NULL, (use_browser > 0 ? display_thread_tui :
987 display_thread), top)) {
988 ui__error("Could not create display thread.\n");
992 if (top->realtime_prio) {
993 struct sched_param param;
995 param.sched_priority = top->realtime_prio;
996 if (sched_setscheduler(0, SCHED_FIFO, ¶m)) {
997 ui__error("Could not set realtime priority.\n");
1003 u64 hits = top->samples;
1005 perf_top__mmap_read(top);
1007 if (hits == top->samples)
1008 ret = perf_evlist__poll(top->evlist, 100);
1013 pthread_join(thread, NULL);
1015 perf_session__delete(top->session);
1016 top->session = NULL;
1021 char errbuf[BUFSIZ];
1022 const char *err = str_error_r(-ret, errbuf, sizeof(errbuf));
1024 ui__error("Could not read the CPU topology map: %s\n", err);
1030 callchain_opt(const struct option *opt, const char *arg, int unset)
1032 symbol_conf.use_callchain = true;
1033 return record_callchain_opt(opt, arg, unset);
1037 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1039 struct callchain_param *callchain = opt->value;
1041 callchain->enabled = !unset;
1042 callchain->record_mode = CALLCHAIN_FP;
1048 symbol_conf.use_callchain = false;
1049 callchain->record_mode = CALLCHAIN_NONE;
1053 return parse_callchain_top_opt(arg);
1056 static int perf_top_config(const char *var, const char *value, void *cb __maybe_unused)
1058 if (!strcmp(var, "top.call-graph"))
1059 var = "call-graph.record-mode"; /* fall-through */
1060 if (!strcmp(var, "top.children")) {
1061 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
1069 parse_percent_limit(const struct option *opt, const char *arg,
1070 int unset __maybe_unused)
1072 struct perf_top *top = opt->value;
1074 top->min_percent = strtof(arg, NULL);
1078 const char top_callchain_help[] = CALLCHAIN_RECORD_HELP CALLCHAIN_REPORT_HELP
1079 "\n\t\t\t\tDefault: fp,graph,0.5,caller,function";
1081 int cmd_top(int argc, const char **argv)
1083 char errbuf[BUFSIZ];
1084 struct perf_top top = {
1088 .mmap_pages = UINT_MAX,
1089 .user_freq = UINT_MAX,
1090 .user_interval = ULLONG_MAX,
1091 .freq = 4000, /* 4 KHz */
1095 .proc_map_timeout = 500,
1097 .max_stack = sysctl_perf_event_max_stack,
1098 .sym_pcnt_filter = 5,
1100 struct record_opts *opts = &top.record_opts;
1101 struct target *target = &opts->target;
1102 const struct option options[] = {
1103 OPT_CALLBACK('e', "event", &top.evlist, "event",
1104 "event selector. use 'perf list' to list available events",
1105 parse_events_option),
1106 OPT_U64('c', "count", &opts->user_interval, "event period to sample"),
1107 OPT_STRING('p', "pid", &target->pid, "pid",
1108 "profile events on existing process id"),
1109 OPT_STRING('t', "tid", &target->tid, "tid",
1110 "profile events on existing thread id"),
1111 OPT_BOOLEAN('a', "all-cpus", &target->system_wide,
1112 "system-wide collection from all CPUs"),
1113 OPT_STRING('C', "cpu", &target->cpu_list, "cpu",
1114 "list of cpus to monitor"),
1115 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1116 "file", "vmlinux pathname"),
1117 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1118 "don't load vmlinux even if found"),
1119 OPT_BOOLEAN('K', "hide_kernel_symbols", &top.hide_kernel_symbols,
1120 "hide kernel symbols"),
1121 OPT_CALLBACK('m', "mmap-pages", &opts->mmap_pages, "pages",
1122 "number of mmap data pages",
1123 perf_evlist__parse_mmap_pages),
1124 OPT_INTEGER('r', "realtime", &top.realtime_prio,
1125 "collect data with this RT SCHED_FIFO priority"),
1126 OPT_INTEGER('d', "delay", &top.delay_secs,
1127 "number of seconds to delay between refreshes"),
1128 OPT_BOOLEAN('D', "dump-symtab", &top.dump_symtab,
1129 "dump the symbol table used for profiling"),
1130 OPT_INTEGER('f', "count-filter", &top.count_filter,
1131 "only display functions with more events than this"),
1132 OPT_BOOLEAN(0, "group", &opts->group,
1133 "put the counters into a counter group"),
1134 OPT_BOOLEAN('i', "no-inherit", &opts->no_inherit,
1135 "child tasks do not inherit counters"),
1136 OPT_STRING(0, "sym-annotate", &top.sym_filter, "symbol name",
1137 "symbol to annotate"),
1138 OPT_BOOLEAN('z', "zero", &top.zero, "zero history across updates"),
1139 OPT_UINTEGER('F', "freq", &opts->user_freq, "profile at this frequency"),
1140 OPT_INTEGER('E', "entries", &top.print_entries,
1141 "display this many functions"),
1142 OPT_BOOLEAN('U', "hide_user_symbols", &top.hide_user_symbols,
1143 "hide user symbols"),
1144 OPT_BOOLEAN(0, "tui", &top.use_tui, "Use the TUI interface"),
1145 OPT_BOOLEAN(0, "stdio", &top.use_stdio, "Use the stdio interface"),
1146 OPT_INCR('v', "verbose", &verbose,
1147 "be more verbose (show counter open errors, etc)"),
1148 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1149 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1150 " Please refer the man page for the complete list."),
1151 OPT_STRING(0, "fields", &field_order, "key[,keys...]",
1152 "output field(s): overhead, period, sample plus all of sort keys"),
1153 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1154 "Show a column with the number of samples"),
1155 OPT_CALLBACK_NOOPT('g', NULL, &callchain_param,
1156 NULL, "enables call-graph recording and display",
1158 OPT_CALLBACK(0, "call-graph", &callchain_param,
1159 "record_mode[,record_size],print_type,threshold[,print_limit],order,sort_key[,branch]",
1160 top_callchain_help, &parse_callchain_opt),
1161 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1162 "Accumulate callchains of children and show total overhead as well"),
1163 OPT_INTEGER(0, "max-stack", &top.max_stack,
1164 "Set the maximum stack depth when parsing the callchain. "
1165 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1166 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1167 "ignore callees of these functions in call graphs",
1168 report_parse_ignore_callees_opt),
1169 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1170 "Show a column with the sum of periods"),
1171 OPT_STRING(0, "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1172 "only consider symbols in these dsos"),
1173 OPT_STRING(0, "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1174 "only consider symbols in these comms"),
1175 OPT_STRING(0, "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1176 "only consider these symbols"),
1177 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1178 "Interleave source code with assembly code (default)"),
1179 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1180 "Display raw encoding of assembly instructions (default)"),
1181 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1182 "Enable kernel symbol demangling"),
1183 OPT_STRING(0, "objdump", &objdump_path, "path",
1184 "objdump binary to use for disassembly and annotations"),
1185 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1186 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1187 OPT_STRING('u', "uid", &target->uid_str, "user", "user to profile"),
1188 OPT_CALLBACK(0, "percent-limit", &top, "percent",
1189 "Don't show entries under that percent", parse_percent_limit),
1190 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1191 "How to display percentage of filtered entries", parse_filter_percentage),
1192 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1194 "don't try to adjust column width, use these fixed values"),
1195 OPT_UINTEGER(0, "proc-map-timeout", &opts->proc_map_timeout,
1196 "per thread proc mmap processing timeout in ms"),
1197 OPT_CALLBACK_NOOPT('b', "branch-any", &opts->branch_stack,
1198 "branch any", "sample any taken branches",
1199 parse_branch_stack),
1200 OPT_CALLBACK('j', "branch-filter", &opts->branch_stack,
1201 "branch filter mask", "branch stack filter modes",
1202 parse_branch_stack),
1203 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1204 "Show raw trace event output (do not use print fmt or plugins)"),
1205 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1206 "Show entries in a hierarchy"),
1209 const char * const top_usage[] = {
1210 "perf top [<options>]",
1213 int status = hists__init();
1218 top.evlist = perf_evlist__new();
1219 if (top.evlist == NULL)
1222 status = perf_config(perf_top_config, &top);
1226 argc = parse_options(argc, argv, options, top_usage, 0);
1228 usage_with_options(top_usage, options);
1230 if (!top.evlist->nr_entries &&
1231 perf_evlist__add_default(top.evlist) < 0) {
1232 pr_err("Not enough memory for event selector list\n");
1233 goto out_delete_evlist;
1236 if (symbol_conf.report_hierarchy) {
1237 /* disable incompatible options */
1238 symbol_conf.event_group = false;
1239 symbol_conf.cumulate_callchain = false;
1242 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1243 parse_options_usage(top_usage, options, "fields", 0);
1244 parse_options_usage(NULL, options, "hierarchy", 0);
1245 goto out_delete_evlist;
1249 sort__mode = SORT_MODE__TOP;
1250 /* display thread wants entries to be collapsed in a different tree */
1251 perf_hpp_list.need_collapse = 1;
1255 else if (top.use_tui)
1258 setup_browser(false);
1260 if (setup_sorting(top.evlist) < 0) {
1262 parse_options_usage(top_usage, options, "s", 1);
1264 parse_options_usage(sort_order ? NULL : top_usage,
1265 options, "fields", 0);
1266 goto out_delete_evlist;
1269 status = target__validate(target);
1271 target__strerror(target, status, errbuf, BUFSIZ);
1272 ui__warning("%s\n", errbuf);
1275 status = target__parse_uid(target);
1277 int saved_errno = errno;
1279 target__strerror(target, status, errbuf, BUFSIZ);
1280 ui__error("%s\n", errbuf);
1282 status = -saved_errno;
1283 goto out_delete_evlist;
1286 if (target__none(target))
1287 target->system_wide = true;
1289 if (perf_evlist__create_maps(top.evlist, target) < 0) {
1290 ui__error("Couldn't create thread/CPU maps: %s\n",
1291 errno == ENOENT ? "No such process" : str_error_r(errno, errbuf, sizeof(errbuf)));
1292 goto out_delete_evlist;
1295 symbol_conf.nr_events = top.evlist->nr_entries;
1297 if (top.delay_secs < 1)
1300 if (record_opts__config(opts)) {
1302 goto out_delete_evlist;
1305 top.sym_evsel = perf_evlist__first(top.evlist);
1307 if (!callchain_param.enabled) {
1308 symbol_conf.cumulate_callchain = false;
1309 perf_hpp__cancel_cumulate();
1312 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1313 callchain_param.order = ORDER_CALLER;
1315 status = symbol__annotation_init();
1317 goto out_delete_evlist;
1319 symbol_conf.try_vmlinux_path = (symbol_conf.vmlinux_name == NULL);
1320 if (symbol__init(NULL) < 0)
1323 sort__setup_elide(stdout);
1325 get_term_dimensions(&top.winsize);
1326 if (top.print_entries == 0) {
1327 struct sigaction act = {
1328 .sa_sigaction = perf_top__sig_winch,
1329 .sa_flags = SA_SIGINFO,
1331 perf_top__update_print_entries(&top);
1332 sigaction(SIGWINCH, &act, NULL);
1335 status = __cmd_top(&top);
1338 perf_evlist__delete(top.evlist);