1 // SPDX-License-Identifier: GPL-2.0
5 * Builtin report command: Analyze the perf.data input file,
6 * look up and read DSOs and symbol information and display
7 * a histogram of results, along various sorting keys.
11 #include "util/config.h"
13 #include "util/annotate.h"
14 #include "util/color.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
19 #include <linux/zalloc.h>
21 #include "util/symbol.h"
22 #include "util/callchain.h"
23 #include "util/values.h"
26 #include "util/debug.h"
27 #include "util/evlist.h"
28 #include "util/evsel.h"
29 #include "util/evswitch.h"
30 #include "util/header.h"
31 #include "util/session.h"
32 #include "util/srcline.h"
33 #include "util/tool.h"
35 #include <subcmd/parse-options.h>
36 #include <subcmd/exec-cmd.h>
37 #include "util/parse-events.h"
39 #include "util/thread.h"
40 #include "util/sort.h"
41 #include "util/hist.h"
42 #include "util/data.h"
43 #include "arch/common.h"
44 #include "util/time-utils.h"
45 #include "util/auxtrace.h"
46 #include "util/units.h"
47 #include "util/branch.h"
48 #include "util/util.h"
50 #include "ui/progress.h"
56 #include <linux/ctype.h>
58 #include <linux/bitmap.h>
59 #include <linux/string.h>
60 #include <linux/stringify.h>
61 #include <linux/time64.h>
62 #include <sys/types.h>
65 #include <linux/mman.h>
68 struct perf_tool tool;
69 struct perf_session *session;
70 struct evswitch evswitch;
71 bool use_tui, use_gtk, use_stdio;
74 bool inverted_callchain;
81 bool nonany_branch_mode;
84 struct perf_read_values show_threads_values;
85 struct annotation_options annotation_opts;
86 const char *pretty_printing_style;
88 const char *symbol_filter_str;
90 struct perf_time_interval *ptime_range;
97 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
98 struct branch_type_stat brtype_stat;
102 static int report__config(const char *var, const char *value, void *cb)
104 struct report *rep = cb;
106 if (!strcmp(var, "report.group")) {
107 symbol_conf.event_group = perf_config_bool(var, value);
110 if (!strcmp(var, "report.percent-limit")) {
111 double pcnt = strtof(value, NULL);
113 rep->min_percent = pcnt;
114 callchain_param.min_percent = pcnt;
117 if (!strcmp(var, "report.children")) {
118 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
121 if (!strcmp(var, "report.queue-size"))
122 return perf_config_u64(&rep->queue_size, var, value);
124 if (!strcmp(var, "report.sort_order")) {
125 default_sort_order = strdup(value);
132 static int hist_iter__report_callback(struct hist_entry_iter *iter,
133 struct addr_location *al, bool single,
137 struct report *rep = arg;
138 struct hist_entry *he = iter->he;
139 struct evsel *evsel = iter->evsel;
140 struct perf_sample *sample = iter->sample;
142 struct branch_info *bi;
144 if (!ui__has_annotation() && !rep->symbol_ipc)
147 if (sort__mode == SORT_MODE__BRANCH) {
148 bi = he->branch_info;
149 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
153 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
155 } else if (rep->mem_mode) {
157 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
161 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
163 } else if (symbol_conf.cumulate_callchain) {
165 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
167 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
174 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
175 struct addr_location *al __maybe_unused,
176 bool single __maybe_unused,
179 struct hist_entry *he = iter->he;
180 struct report *rep = arg;
181 struct branch_info *bi;
182 struct perf_sample *sample = iter->sample;
183 struct evsel *evsel = iter->evsel;
186 if (!ui__has_annotation() && !rep->symbol_ipc)
189 bi = he->branch_info;
190 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
194 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
196 branch_type_count(&rep->brtype_stat, &bi->flags,
197 bi->from.addr, bi->to.addr);
203 static void setup_forced_leader(struct report *report,
204 struct evlist *evlist)
206 if (report->group_set)
207 perf_evlist__force_leader(evlist);
210 static int process_feature_event(struct perf_session *session,
211 union perf_event *event)
213 struct report *rep = container_of(session->tool, struct report, tool);
215 if (event->feat.feat_id < HEADER_LAST_FEATURE)
216 return perf_event__process_feature(session, event);
218 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
219 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
220 event->feat.feat_id);
225 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
226 * means all features are received, now we can force the
229 setup_forced_leader(rep, session->evlist);
233 static int process_sample_event(struct perf_tool *tool,
234 union perf_event *event,
235 struct perf_sample *sample,
237 struct machine *machine)
239 struct report *rep = container_of(tool, struct report, tool);
240 struct addr_location al;
241 struct hist_entry_iter iter = {
244 .hide_unresolved = symbol_conf.hide_unresolved,
245 .add_entry_cb = hist_iter__report_callback,
249 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
254 if (evswitch__discard(&rep->evswitch, evsel))
257 if (machine__resolve(machine, &al, sample) < 0) {
258 pr_debug("problem processing %d event, skipping it.\n",
263 if (symbol_conf.hide_unresolved && al.sym == NULL)
266 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
269 if (sort__mode == SORT_MODE__BRANCH) {
271 * A non-synthesized event might not have a branch stack if
272 * branch stacks have been synthesized (using itrace options).
274 if (!sample->branch_stack)
277 iter.add_entry_cb = hist_iter__branch_callback;
278 iter.ops = &hist_iter_branch;
279 } else if (rep->mem_mode) {
280 iter.ops = &hist_iter_mem;
281 } else if (symbol_conf.cumulate_callchain) {
282 iter.ops = &hist_iter_cumulative;
284 iter.ops = &hist_iter_normal;
288 al.map->dso->hit = 1;
290 if (ui__has_annotation() || rep->symbol_ipc) {
291 hist__account_cycles(sample->branch_stack, &al, sample,
292 rep->nonany_branch_mode);
295 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
297 pr_debug("problem adding hist entry, skipping event\n");
299 addr_location__put(&al);
303 static int process_read_event(struct perf_tool *tool,
304 union perf_event *event,
305 struct perf_sample *sample __maybe_unused,
307 struct machine *machine __maybe_unused)
309 struct report *rep = container_of(tool, struct report, tool);
311 if (rep->show_threads) {
312 const char *name = perf_evsel__name(evsel);
313 int err = perf_read_values_add_value(&rep->show_threads_values,
314 event->read.pid, event->read.tid,
326 /* For pipe mode, sample_type is not currently set */
327 static int report__setup_sample_type(struct report *rep)
329 struct perf_session *session = rep->session;
330 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
331 bool is_pipe = perf_data__is_pipe(session->data);
333 if (session->itrace_synth_opts->callchain ||
335 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
336 !session->itrace_synth_opts->set))
337 sample_type |= PERF_SAMPLE_CALLCHAIN;
339 if (session->itrace_synth_opts->last_branch)
340 sample_type |= PERF_SAMPLE_BRANCH_STACK;
342 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
343 if (perf_hpp_list.parent) {
344 ui__error("Selected --sort parent, but no "
345 "callchain data. Did you call "
346 "'perf record' without -g?\n");
349 if (symbol_conf.use_callchain &&
350 !symbol_conf.show_branchflag_count) {
351 ui__error("Selected -g or --branch-history.\n"
352 "But no callchain or branch data.\n"
353 "Did you call 'perf record' without -g or -b?\n");
356 } else if (!callchain_param.enabled &&
357 callchain_param.mode != CHAIN_NONE &&
358 !symbol_conf.use_callchain) {
359 symbol_conf.use_callchain = true;
360 if (callchain_register_param(&callchain_param) < 0) {
361 ui__error("Can't register callchain params.\n");
366 if (symbol_conf.cumulate_callchain) {
367 /* Silently ignore if callchain is missing */
368 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
369 symbol_conf.cumulate_callchain = false;
370 perf_hpp__cancel_cumulate();
374 if (sort__mode == SORT_MODE__BRANCH) {
376 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
377 ui__error("Selected -b but no branch data. "
378 "Did you call perf record without -b?\n");
383 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
384 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
385 (sample_type & PERF_SAMPLE_STACK_USER)) {
386 callchain_param.record_mode = CALLCHAIN_DWARF;
387 dwarf_callchain_users = true;
388 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
389 callchain_param.record_mode = CALLCHAIN_LBR;
391 callchain_param.record_mode = CALLCHAIN_FP;
394 /* ??? handle more cases than just ANY? */
395 if (!(perf_evlist__combined_branch_type(session->evlist) &
396 PERF_SAMPLE_BRANCH_ANY))
397 rep->nonany_branch_mode = true;
402 static void sig_handler(int sig __maybe_unused)
407 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
408 const char *evname, FILE *fp)
412 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
413 u64 nr_events = hists->stats.total_period;
414 struct evsel *evsel = hists_to_evsel(hists);
416 size_t size = sizeof(buf);
417 int socked_id = hists->socket_filter;
422 if (symbol_conf.filter_relative) {
423 nr_samples = hists->stats.nr_non_filtered_samples;
424 nr_events = hists->stats.total_non_filtered_period;
427 if (perf_evsel__is_group_event(evsel)) {
430 perf_evsel__group_desc(evsel, buf, size);
433 for_each_group_member(pos, evsel) {
434 const struct hists *pos_hists = evsel__hists(pos);
436 if (symbol_conf.filter_relative) {
437 nr_samples += pos_hists->stats.nr_non_filtered_samples;
438 nr_events += pos_hists->stats.total_non_filtered_period;
440 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
441 nr_events += pos_hists->stats.total_period;
446 nr_samples = convert_unit(nr_samples, &unit);
447 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
448 if (evname != NULL) {
449 ret += fprintf(fp, " of event%s '%s'",
450 evsel->core.nr_members > 1 ? "s" : "", evname);
454 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
456 if (symbol_conf.show_ref_callgraph &&
457 strstr(evname, "call-graph=no")) {
458 ret += fprintf(fp, ", show reference callgraph");
462 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
463 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
465 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
468 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
470 return ret + fprintf(fp, "\n#\n");
473 static int perf_evlist__tty_browse_hists(struct evlist *evlist,
480 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
481 evlist->stats.total_lost_samples);
484 evlist__for_each_entry(evlist, pos) {
485 struct hists *hists = evsel__hists(pos);
486 const char *evname = perf_evsel__name(pos);
488 if (symbol_conf.event_group &&
489 !perf_evsel__is_group_leader(pos))
492 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
493 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
494 !(symbol_conf.use_callchain ||
495 symbol_conf.show_branchflag_count));
496 fprintf(stdout, "\n\n");
500 fprintf(stdout, "#\n# (%s)\n#\n", help);
502 if (rep->show_threads) {
503 bool style = !strcmp(rep->pretty_printing_style, "raw");
504 perf_read_values_display(stdout, &rep->show_threads_values,
506 perf_read_values_destroy(&rep->show_threads_values);
509 if (sort__mode == SORT_MODE__BRANCH)
510 branch_type_stat_display(stdout, &rep->brtype_stat);
515 static void report__warn_kptr_restrict(const struct report *rep)
517 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
518 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
520 if (perf_evlist__exclude_kernel(rep->session->evlist))
523 if (kernel_map == NULL ||
524 (kernel_map->dso->hit &&
525 (kernel_kmap->ref_reloc_sym == NULL ||
526 kernel_kmap->ref_reloc_sym->addr == 0))) {
528 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
529 "can't be resolved.";
531 if (kernel_map && map__has_symbols(kernel_map)) {
532 desc = "If some relocation was applied (e.g. "
533 "kexec) symbols may be misresolved.";
537 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
538 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
539 "Samples in kernel modules can't be resolved as well.\n\n",
544 static int report__gtk_browse_hists(struct report *rep, const char *help)
546 int (*hist_browser)(struct evlist *evlist, const char *help,
547 struct hist_browser_timer *timer, float min_pcnt);
549 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
551 if (hist_browser == NULL) {
552 ui__error("GTK browser not found!\n");
556 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
559 static int report__browse_hists(struct report *rep)
562 struct perf_session *session = rep->session;
563 struct evlist *evlist = session->evlist;
564 const char *help = perf_tip(system_path(TIPDIR));
567 /* fallback for people who don't install perf ;-) */
568 help = perf_tip(DOCDIR);
570 help = "Cannot load tips.txt file, please install perf!";
573 switch (use_browser) {
575 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
577 &session->header.env,
578 true, &rep->annotation_opts);
580 * Usually "ret" is the last pressed key, and we only
581 * care if the key notifies us to switch data file.
583 if (ret != K_SWITCH_INPUT_DATA)
587 ret = report__gtk_browse_hists(rep, help);
590 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
597 static int report__collapse_hists(struct report *rep)
599 struct ui_progress prog;
603 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
605 evlist__for_each_entry(rep->session->evlist, pos) {
606 struct hists *hists = evsel__hists(pos);
609 hists->symbol_filter_str = rep->symbol_filter_str;
611 hists->socket_filter = rep->socket_filter;
613 ret = hists__collapse_resort(hists, &prog);
617 /* Non-group events are considered as leader */
618 if (symbol_conf.event_group &&
619 !perf_evsel__is_group_leader(pos)) {
620 struct hists *leader_hists = evsel__hists(pos->leader);
622 hists__match(leader_hists, hists);
623 hists__link(leader_hists, hists);
627 ui_progress__finish();
631 static int hists__resort_cb(struct hist_entry *he, void *arg)
633 struct report *rep = arg;
634 struct symbol *sym = he->ms.sym;
636 if (rep->symbol_ipc && sym && !sym->annotate2) {
637 struct evsel *evsel = hists_to_evsel(he->hists);
639 symbol__annotate2(sym, he->ms.map, evsel,
640 &annotation__default_options, NULL);
646 static void report__output_resort(struct report *rep)
648 struct ui_progress prog;
651 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
653 evlist__for_each_entry(rep->session->evlist, pos) {
654 perf_evsel__output_resort_cb(pos, &prog,
655 hists__resort_cb, rep);
658 ui_progress__finish();
661 static void stats_setup(struct report *rep)
663 memset(&rep->tool, 0, sizeof(rep->tool));
664 rep->tool.no_warn = true;
667 static int stats_print(struct report *rep)
669 struct perf_session *session = rep->session;
671 perf_session__fprintf_nr_events(session, stdout);
675 static void tasks_setup(struct report *rep)
677 memset(&rep->tool, 0, sizeof(rep->tool));
678 rep->tool.ordered_events = true;
679 if (rep->mmaps_mode) {
680 rep->tool.mmap = perf_event__process_mmap;
681 rep->tool.mmap2 = perf_event__process_mmap2;
683 rep->tool.comm = perf_event__process_comm;
684 rep->tool.exit = perf_event__process_exit;
685 rep->tool.fork = perf_event__process_fork;
686 rep->tool.no_warn = true;
690 struct thread *thread;
691 struct list_head list;
692 struct list_head children;
695 static struct task *tasks_list(struct task *task, struct machine *machine)
697 struct thread *parent_thread, *thread = task->thread;
698 struct task *parent_task;
700 /* Already listed. */
701 if (!list_empty(&task->list))
704 /* Last one in the chain. */
705 if (thread->ppid == -1)
708 parent_thread = machine__find_thread(machine, -1, thread->ppid);
710 return ERR_PTR(-ENOENT);
712 parent_task = thread__priv(parent_thread);
713 list_add_tail(&task->list, &parent_task->children);
714 return tasks_list(parent_task, machine);
717 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
722 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) {
723 struct map *map = rb_entry(nd, struct map, rb_node);
725 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
726 indent, "", map->start, map->end,
727 map->prot & PROT_READ ? 'r' : '-',
728 map->prot & PROT_WRITE ? 'w' : '-',
729 map->prot & PROT_EXEC ? 'x' : '-',
730 map->flags & MAP_SHARED ? 's' : 'p',
732 map->ino, map->dso->name);
738 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp)
740 return maps__fprintf_task(&mg->maps, indent, fp);
743 static void task__print_level(struct task *task, FILE *fp, int level)
745 struct thread *thread = task->thread;
747 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
748 thread->pid_, thread->tid, thread->ppid,
751 fprintf(fp, "%s\n", thread__comm_str(thread));
753 map_groups__fprintf_task(thread->mg, comm_indent, fp);
755 if (!list_empty(&task->children)) {
756 list_for_each_entry(child, &task->children, list)
757 task__print_level(child, fp, level + 1);
761 static int tasks_print(struct report *rep, FILE *fp)
763 struct perf_session *session = rep->session;
764 struct machine *machine = &session->machines.host;
765 struct task *tasks, *task;
766 unsigned int nr = 0, itask = 0, i;
771 * No locking needed while accessing machine->threads,
772 * because --tasks is single threaded command.
775 /* Count all the threads. */
776 for (i = 0; i < THREADS__TABLE_SIZE; i++)
777 nr += machine->threads[i].nr;
779 tasks = malloc(sizeof(*tasks) * nr);
783 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
784 struct threads *threads = &machine->threads[i];
786 for (nd = rb_first_cached(&threads->entries); nd;
788 task = tasks + itask++;
790 task->thread = rb_entry(nd, struct thread, rb_node);
791 INIT_LIST_HEAD(&task->children);
792 INIT_LIST_HEAD(&task->list);
793 thread__set_priv(task->thread, task);
798 * Iterate every task down to the unprocessed parent
799 * and link all in task children list. Task with no
800 * parent is added into 'list'.
802 for (itask = 0; itask < nr; itask++) {
803 task = tasks + itask;
805 if (!list_empty(&task->list))
808 task = tasks_list(task, machine);
810 pr_err("Error: failed to process tasks\n");
812 return PTR_ERR(task);
816 list_add_tail(&task->list, &list);
819 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
821 list_for_each_entry(task, &list, list)
822 task__print_level(task, fp, 0);
828 static int __cmd_report(struct report *rep)
831 struct perf_session *session = rep->session;
833 struct perf_data *data = session->data;
835 signal(SIGINT, sig_handler);
838 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
841 ui__error("failed to set cpu bitmap\n");
844 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
847 if (rep->show_threads) {
848 ret = perf_read_values_init(&rep->show_threads_values);
853 ret = report__setup_sample_type(rep);
855 /* report__setup_sample_type() already showed error message */
865 ret = perf_session__process_events(session);
867 ui__error("failed to process sample\n");
872 return stats_print(rep);
875 return tasks_print(rep, stdout);
877 report__warn_kptr_restrict(rep);
879 evlist__for_each_entry(session->evlist, pos)
880 rep->nr_entries += evsel__hists(pos)->nr_entries;
882 if (use_browser == 0) {
884 perf_session__fprintf(session, stdout);
887 perf_session__fprintf_dsos(session, stdout);
890 perf_session__fprintf_nr_events(session, stdout);
891 perf_evlist__fprintf_nr_events(session->evlist, stdout);
896 ret = report__collapse_hists(rep);
898 ui__error("failed to process hist entry\n");
906 * recalculate number of entries after collapsing since it
907 * might be changed during the collapse phase.
910 evlist__for_each_entry(session->evlist, pos)
911 rep->nr_entries += evsel__hists(pos)->nr_entries;
913 if (rep->nr_entries == 0) {
914 ui__error("The %s data has no samples!\n", data->path);
918 report__output_resort(rep);
920 return report__browse_hists(rep);
924 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
926 struct callchain_param *callchain = opt->value;
928 callchain->enabled = !unset;
933 symbol_conf.use_callchain = false;
934 callchain->mode = CHAIN_NONE;
938 return parse_callchain_report_opt(arg);
942 parse_time_quantum(const struct option *opt, const char *arg,
943 int unset __maybe_unused)
945 unsigned long *time_q = opt->value;
948 *time_q = strtoul(arg, &end, 0);
952 pr_err("time quantum cannot be 0");
955 end = skip_spaces(end);
958 if (!strcmp(end, "s")) {
959 *time_q *= NSEC_PER_SEC;
962 if (!strcmp(end, "ms")) {
963 *time_q *= NSEC_PER_MSEC;
966 if (!strcmp(end, "us")) {
967 *time_q *= NSEC_PER_USEC;
970 if (!strcmp(end, "ns"))
973 pr_err("Cannot parse time quantum `%s'\n", arg);
978 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
979 const char *arg, int unset __maybe_unused)
982 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
985 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
986 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
989 have_ignore_callees = 1;
996 parse_branch_mode(const struct option *opt,
997 const char *str __maybe_unused, int unset)
999 int *branch_mode = opt->value;
1001 *branch_mode = !unset;
1006 parse_percent_limit(const struct option *opt, const char *str,
1007 int unset __maybe_unused)
1009 struct report *rep = opt->value;
1010 double pcnt = strtof(str, NULL);
1012 rep->min_percent = pcnt;
1013 callchain_param.min_percent = pcnt;
1017 int cmd_report(int argc, const char **argv)
1019 struct perf_session *session;
1020 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1022 bool has_br_stack = false;
1023 int branch_mode = -1;
1024 bool branch_call_mode = false;
1025 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1026 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1027 CALLCHAIN_REPORT_HELP
1028 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1029 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1030 const char * const report_usage[] = {
1031 "perf report [<options>]",
1034 struct report report = {
1036 .sample = process_sample_event,
1037 .mmap = perf_event__process_mmap,
1038 .mmap2 = perf_event__process_mmap2,
1039 .comm = perf_event__process_comm,
1040 .namespaces = perf_event__process_namespaces,
1041 .exit = perf_event__process_exit,
1042 .fork = perf_event__process_fork,
1043 .lost = perf_event__process_lost,
1044 .read = process_read_event,
1045 .attr = perf_event__process_attr,
1046 .tracing_data = perf_event__process_tracing_data,
1047 .build_id = perf_event__process_build_id,
1048 .id_index = perf_event__process_id_index,
1049 .auxtrace_info = perf_event__process_auxtrace_info,
1050 .auxtrace = perf_event__process_auxtrace,
1051 .event_update = perf_event__process_event_update,
1052 .feature = process_feature_event,
1053 .ordered_events = true,
1054 .ordering_requires_timestamps = true,
1056 .max_stack = PERF_MAX_STACK_DEPTH,
1057 .pretty_printing_style = "normal",
1058 .socket_filter = -1,
1059 .annotation_opts = annotation__default_options,
1061 const struct option options[] = {
1062 OPT_STRING('i', "input", &input_name, "file",
1064 OPT_INCR('v', "verbose", &verbose,
1065 "be more verbose (show symbol address, etc)"),
1066 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1067 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1068 "dump raw trace in ASCII"),
1069 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1070 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1071 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1072 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1073 "file", "vmlinux pathname"),
1074 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1075 "don't load vmlinux even if found"),
1076 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1077 "file", "kallsyms pathname"),
1078 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1079 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1080 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1081 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1082 "Show a column with the number of samples"),
1083 OPT_BOOLEAN('T', "threads", &report.show_threads,
1084 "Show per-thread event counters"),
1085 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1086 "pretty printing style key: normal raw"),
1087 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1088 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1089 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1090 "Use the stdio interface"),
1091 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1092 OPT_BOOLEAN(0, "header-only", &report.header_only,
1093 "Show only data header."),
1094 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1095 sort_help("sort by key(s):")),
1096 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1097 sort_help("output field(s): overhead period sample ")),
1098 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1099 "Show sample percentage for different cpu modes"),
1100 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1101 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1102 OPT_STRING('p', "parent", &parent_pattern, "regex",
1103 "regex filter to identify parent, see: '--sort parent'"),
1104 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1105 "Only display entries with parent-match"),
1106 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1107 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1108 report_callchain_help, &report_parse_callchain_opt,
1109 callchain_default_opt),
1110 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1111 "Accumulate callchains of children and show total overhead as well"),
1112 OPT_INTEGER(0, "max-stack", &report.max_stack,
1113 "Set the maximum stack depth when parsing the callchain, "
1114 "anything beyond the specified depth will be ignored. "
1115 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1116 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1117 "alias for inverted call graph"),
1118 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1119 "ignore callees of these functions in call graphs",
1120 report_parse_ignore_callees_opt),
1121 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1122 "only consider symbols in these dsos"),
1123 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1124 "only consider symbols in these comms"),
1125 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1126 "only consider symbols in these pids"),
1127 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1128 "only consider symbols in these tids"),
1129 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1130 "only consider these symbols"),
1131 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1132 "only show symbols that (partially) match with this filter"),
1133 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1135 "don't try to adjust column width, use these fixed values"),
1136 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1137 "separator for columns, no spaces will be added between "
1138 "columns '.' is reserved."),
1139 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1140 "Only display entries resolved to a symbol"),
1141 OPT_CALLBACK(0, "symfs", NULL, "directory",
1142 "Look for files with symbols relative to this directory",
1143 symbol__config_symfs),
1144 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1145 "list of cpus to profile"),
1146 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1147 "Display extended information about perf.data file"),
1148 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1149 "Interleave source code with assembly code (default)"),
1150 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1151 "Display raw encoding of assembly instructions (default)"),
1152 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1153 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1154 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1155 "Show a column with the sum of periods"),
1156 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1157 "Show event group information together"),
1158 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1159 "use branch records for per branch histogram filling",
1161 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1162 "add last branch records to call history"),
1163 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1164 "objdump binary to use for disassembly and annotations"),
1165 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1166 "Disable symbol demangling"),
1167 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1168 "Enable kernel symbol demangling"),
1169 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1170 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1171 "Number of samples to save per histogram entry for individual browsing"),
1172 OPT_CALLBACK(0, "percent-limit", &report, "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_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1177 "Instruction Tracing options\n" ITRACE_HELP,
1178 itrace_parse_synth_opts),
1179 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1180 "Show full source file name path for source lines"),
1181 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1182 "Show callgraph from reference event"),
1183 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1184 "only show processor socket that match with this filter"),
1185 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1186 "Show raw trace event output (do not use print fmt or plugins)"),
1187 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1188 "Show entries in a hierarchy"),
1189 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1190 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1191 stdio__config_color, "always"),
1192 OPT_STRING(0, "time", &report.time_str, "str",
1193 "Time span of interest (start,stop)"),
1194 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1195 "Show inline function"),
1196 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1197 "Set percent type local/global-period/hits",
1198 annotate_parse_percent_type),
1199 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1200 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1201 "Set time quantum for time sort key (default 100ms)",
1202 parse_time_quantum),
1203 OPTS_EVSWITCH(&report.evswitch),
1206 struct perf_data data = {
1207 .mode = PERF_DATA_MODE_READ,
1209 int ret = hists__init();
1215 ret = perf_config(report__config, &report);
1219 argc = parse_options(argc, argv, options, report_usage, 0);
1222 * Special case: if there's an argument left then assume that
1223 * it's a symbol filter:
1226 usage_with_options(report_usage, options);
1228 report.symbol_filter_str = argv[0];
1231 if (report.mmaps_mode)
1232 report.tasks_mode = true;
1235 perf_quiet_option();
1237 if (symbol_conf.vmlinux_name &&
1238 access(symbol_conf.vmlinux_name, R_OK)) {
1239 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1242 if (symbol_conf.kallsyms_name &&
1243 access(symbol_conf.kallsyms_name, R_OK)) {
1244 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1248 if (report.inverted_callchain)
1249 callchain_param.order = ORDER_CALLER;
1250 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1251 callchain_param.order = ORDER_CALLER;
1253 if (itrace_synth_opts.callchain &&
1254 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1255 report.max_stack = itrace_synth_opts.callchain_sz;
1257 if (!input_name || !strlen(input_name)) {
1258 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1261 input_name = "perf.data";
1264 data.path = input_name;
1265 data.force = symbol_conf.force;
1268 session = perf_session__new(&data, false, &report.tool);
1269 if (session == NULL)
1272 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1276 if (zstd_init(&(session->zstd_data), 0) < 0)
1277 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1279 if (report.queue_size) {
1280 ordered_events__set_alloc_size(&session->ordered_events,
1284 session->itrace_synth_opts = &itrace_synth_opts;
1286 report.session = session;
1288 has_br_stack = perf_header__has_feat(&session->header,
1289 HEADER_BRANCH_STACK);
1290 if (perf_evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1291 has_br_stack = false;
1293 setup_forced_leader(&report, session->evlist);
1295 if (itrace_synth_opts.last_branch)
1296 has_br_stack = true;
1298 if (has_br_stack && branch_call_mode)
1299 symbol_conf.show_branchflag_count = true;
1301 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1304 * Branch mode is a tristate:
1305 * -1 means default, so decide based on the file having branch data.
1306 * 0/1 means the user chose a mode.
1308 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1309 !branch_call_mode) {
1310 sort__mode = SORT_MODE__BRANCH;
1311 symbol_conf.cumulate_callchain = false;
1313 if (branch_call_mode) {
1314 callchain_param.key = CCKEY_ADDRESS;
1315 callchain_param.branch_callstack = 1;
1316 symbol_conf.use_callchain = true;
1317 callchain_register_param(&callchain_param);
1318 if (sort_order == NULL)
1319 sort_order = "srcline,symbol,dso";
1322 if (report.mem_mode) {
1323 if (sort__mode == SORT_MODE__BRANCH) {
1324 pr_err("branch and mem mode incompatible\n");
1327 sort__mode = SORT_MODE__MEMORY;
1328 symbol_conf.cumulate_callchain = false;
1331 if (symbol_conf.report_hierarchy) {
1332 /* disable incompatible options */
1333 symbol_conf.cumulate_callchain = false;
1336 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1337 parse_options_usage(report_usage, options, "F", 1);
1338 parse_options_usage(NULL, options, "hierarchy", 0);
1342 perf_hpp_list.need_collapse = true;
1345 if (report.use_stdio)
1347 else if (report.use_tui)
1349 else if (report.use_gtk)
1352 /* Force tty output for header output and per-thread stat. */
1353 if (report.header || report.header_only || report.show_threads)
1355 if (report.header || report.header_only)
1356 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1357 if (report.show_full_info)
1358 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1359 if (report.stats_mode || report.tasks_mode)
1361 if (report.stats_mode && report.tasks_mode) {
1362 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1366 if (strcmp(input_name, "-") != 0)
1367 setup_browser(true);
1371 if (sort_order && strstr(sort_order, "ipc")) {
1372 parse_options_usage(report_usage, options, "s", 1);
1376 if (sort_order && strstr(sort_order, "symbol")) {
1377 if (sort__mode == SORT_MODE__BRANCH) {
1378 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1379 sort_order, "ipc_lbr");
1380 report.symbol_ipc = true;
1382 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1383 sort_order, "ipc_null");
1386 sort_order = sort_tmp;
1389 if (setup_sorting(session->evlist) < 0) {
1391 parse_options_usage(report_usage, options, "s", 1);
1393 parse_options_usage(sort_order ? NULL : report_usage,
1398 if ((report.header || report.header_only) && !quiet) {
1399 perf_session__fprintf_info(session, stdout,
1400 report.show_full_info);
1401 if (report.header_only) {
1405 } else if (use_browser == 0 && !quiet &&
1406 !report.stats_mode && !report.tasks_mode) {
1407 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1412 * Only in the TUI browser we are doing integrated annotation,
1413 * so don't allocate extra space that won't be used in the stdio
1416 if (ui__has_annotation() || report.symbol_ipc) {
1417 ret = symbol__annotation_init();
1421 * For searching by name on the "Browse map details".
1422 * providing it only in verbose mode not to bloat too
1423 * much struct symbol.
1427 * XXX: Need to provide a less kludgy way to ask for
1428 * more space per symbol, the u32 is for the index on
1430 * See symbol__browser_index.
1432 symbol_conf.priv_size += sizeof(u32);
1433 symbol_conf.sort_by_name = true;
1435 annotation_config__init();
1438 if (symbol__init(&session->header.env) < 0)
1441 if (report.time_str) {
1442 ret = perf_time__parse_for_ranges(report.time_str, session,
1443 &report.ptime_range,
1449 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1454 if (session->tevent.pevent &&
1455 tep_set_function_resolver(session->tevent.pevent,
1456 machine__resolve_kernel_addr,
1457 &session->machines.host) < 0) {
1458 pr_err("%s: failed to set libtraceevent function resolver\n",
1463 sort__setup_elide(stdout);
1465 ret = __cmd_report(&report);
1466 if (ret == K_SWITCH_INPUT_DATA) {
1467 perf_session__delete(session);
1473 if (report.ptime_range) {
1474 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1475 zfree(&report.ptime_range);
1477 zstd_fini(&(session->zstd_data));
1478 perf_session__delete(session);