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/util.h"
12 #include "util/config.h"
14 #include "util/annotate.h"
15 #include "util/color.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
19 #include "util/symbol.h"
20 #include "util/callchain.h"
21 #include "util/values.h"
24 #include "util/debug.h"
25 #include "util/evlist.h"
26 #include "util/evsel.h"
27 #include "util/header.h"
28 #include "util/session.h"
29 #include "util/tool.h"
31 #include <subcmd/parse-options.h>
32 #include <subcmd/exec-cmd.h>
33 #include "util/parse-events.h"
35 #include "util/thread.h"
36 #include "util/sort.h"
37 #include "util/hist.h"
38 #include "util/data.h"
39 #include "arch/common.h"
40 #include "util/time-utils.h"
41 #include "util/auxtrace.h"
42 #include "util/units.h"
43 #include "util/branch.h"
50 #include <linux/bitmap.h>
51 #include <linux/stringify.h>
52 #include <sys/types.h>
55 #include <linux/mman.h>
58 struct perf_tool tool;
59 struct perf_session *session;
60 bool use_tui, use_gtk, use_stdio;
63 bool inverted_callchain;
70 bool nonany_branch_mode;
73 struct perf_read_values show_threads_values;
74 struct annotation_options annotation_opts;
75 const char *pretty_printing_style;
77 const char *symbol_filter_str;
79 struct perf_time_interval *ptime_range;
86 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
87 struct branch_type_stat brtype_stat;
90 static int report__config(const char *var, const char *value, void *cb)
92 struct report *rep = cb;
94 if (!strcmp(var, "report.group")) {
95 symbol_conf.event_group = perf_config_bool(var, value);
98 if (!strcmp(var, "report.percent-limit")) {
99 double pcnt = strtof(value, NULL);
101 rep->min_percent = pcnt;
102 callchain_param.min_percent = pcnt;
105 if (!strcmp(var, "report.children")) {
106 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
109 if (!strcmp(var, "report.queue-size"))
110 return perf_config_u64(&rep->queue_size, var, value);
112 if (!strcmp(var, "report.sort_order")) {
113 default_sort_order = strdup(value);
120 static int hist_iter__report_callback(struct hist_entry_iter *iter,
121 struct addr_location *al, bool single,
125 struct report *rep = arg;
126 struct hist_entry *he = iter->he;
127 struct perf_evsel *evsel = iter->evsel;
128 struct perf_sample *sample = iter->sample;
130 struct branch_info *bi;
132 if (!ui__has_annotation())
135 hist__account_cycles(sample->branch_stack, al, sample,
136 rep->nonany_branch_mode);
138 if (sort__mode == SORT_MODE__BRANCH) {
139 bi = he->branch_info;
140 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
144 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
146 } else if (rep->mem_mode) {
148 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
152 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
154 } else if (symbol_conf.cumulate_callchain) {
156 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
158 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
165 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
166 struct addr_location *al __maybe_unused,
167 bool single __maybe_unused,
170 struct hist_entry *he = iter->he;
171 struct report *rep = arg;
172 struct branch_info *bi;
173 struct perf_sample *sample = iter->sample;
174 struct perf_evsel *evsel = iter->evsel;
177 if (!ui__has_annotation())
180 hist__account_cycles(sample->branch_stack, al, sample,
181 rep->nonany_branch_mode);
183 bi = he->branch_info;
184 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
188 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
190 branch_type_count(&rep->brtype_stat, &bi->flags,
191 bi->from.addr, bi->to.addr);
197 static void setup_forced_leader(struct report *report,
198 struct perf_evlist *evlist)
200 if (report->group_set)
201 perf_evlist__force_leader(evlist);
204 static int process_feature_event(struct perf_session *session,
205 union perf_event *event)
207 struct report *rep = container_of(session->tool, struct report, tool);
209 if (event->feat.feat_id < HEADER_LAST_FEATURE)
210 return perf_event__process_feature(session, event);
212 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
213 pr_err("failed: wrong feature ID: %" PRIu64 "\n",
214 event->feat.feat_id);
219 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
220 * means all features are received, now we can force the
223 setup_forced_leader(rep, session->evlist);
227 static int process_sample_event(struct perf_tool *tool,
228 union perf_event *event,
229 struct perf_sample *sample,
230 struct perf_evsel *evsel,
231 struct machine *machine)
233 struct report *rep = container_of(tool, struct report, tool);
234 struct addr_location al;
235 struct hist_entry_iter iter = {
238 .hide_unresolved = symbol_conf.hide_unresolved,
239 .add_entry_cb = hist_iter__report_callback,
243 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
248 if (machine__resolve(machine, &al, sample) < 0) {
249 pr_debug("problem processing %d event, skipping it.\n",
254 if (symbol_conf.hide_unresolved && al.sym == NULL)
257 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
260 if (sort__mode == SORT_MODE__BRANCH) {
262 * A non-synthesized event might not have a branch stack if
263 * branch stacks have been synthesized (using itrace options).
265 if (!sample->branch_stack)
268 iter.add_entry_cb = hist_iter__branch_callback;
269 iter.ops = &hist_iter_branch;
270 } else if (rep->mem_mode) {
271 iter.ops = &hist_iter_mem;
272 } else if (symbol_conf.cumulate_callchain) {
273 iter.ops = &hist_iter_cumulative;
275 iter.ops = &hist_iter_normal;
279 al.map->dso->hit = 1;
281 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
283 pr_debug("problem adding hist entry, skipping event\n");
285 addr_location__put(&al);
289 static int process_read_event(struct perf_tool *tool,
290 union perf_event *event,
291 struct perf_sample *sample __maybe_unused,
292 struct perf_evsel *evsel,
293 struct machine *machine __maybe_unused)
295 struct report *rep = container_of(tool, struct report, tool);
297 if (rep->show_threads) {
298 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
299 int err = perf_read_values_add_value(&rep->show_threads_values,
300 event->read.pid, event->read.tid,
312 /* For pipe mode, sample_type is not currently set */
313 static int report__setup_sample_type(struct report *rep)
315 struct perf_session *session = rep->session;
316 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
317 bool is_pipe = perf_data__is_pipe(session->data);
319 if (session->itrace_synth_opts->callchain ||
321 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
322 !session->itrace_synth_opts->set))
323 sample_type |= PERF_SAMPLE_CALLCHAIN;
325 if (session->itrace_synth_opts->last_branch)
326 sample_type |= PERF_SAMPLE_BRANCH_STACK;
328 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
329 if (perf_hpp_list.parent) {
330 ui__error("Selected --sort parent, but no "
331 "callchain data. Did you call "
332 "'perf record' without -g?\n");
335 if (symbol_conf.use_callchain &&
336 !symbol_conf.show_branchflag_count) {
337 ui__error("Selected -g or --branch-history.\n"
338 "But no callchain or branch data.\n"
339 "Did you call 'perf record' without -g or -b?\n");
342 } else if (!callchain_param.enabled &&
343 callchain_param.mode != CHAIN_NONE &&
344 !symbol_conf.use_callchain) {
345 symbol_conf.use_callchain = true;
346 if (callchain_register_param(&callchain_param) < 0) {
347 ui__error("Can't register callchain params.\n");
352 if (symbol_conf.cumulate_callchain) {
353 /* Silently ignore if callchain is missing */
354 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
355 symbol_conf.cumulate_callchain = false;
356 perf_hpp__cancel_cumulate();
360 if (sort__mode == SORT_MODE__BRANCH) {
362 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
363 ui__error("Selected -b but no branch data. "
364 "Did you call perf record without -b?\n");
369 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
370 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
371 (sample_type & PERF_SAMPLE_STACK_USER)) {
372 callchain_param.record_mode = CALLCHAIN_DWARF;
373 dwarf_callchain_users = true;
374 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
375 callchain_param.record_mode = CALLCHAIN_LBR;
377 callchain_param.record_mode = CALLCHAIN_FP;
380 /* ??? handle more cases than just ANY? */
381 if (!(perf_evlist__combined_branch_type(session->evlist) &
382 PERF_SAMPLE_BRANCH_ANY))
383 rep->nonany_branch_mode = true;
388 static void sig_handler(int sig __maybe_unused)
393 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
394 const char *evname, FILE *fp)
398 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
399 u64 nr_events = hists->stats.total_period;
400 struct perf_evsel *evsel = hists_to_evsel(hists);
402 size_t size = sizeof(buf);
403 int socked_id = hists->socket_filter;
408 if (symbol_conf.filter_relative) {
409 nr_samples = hists->stats.nr_non_filtered_samples;
410 nr_events = hists->stats.total_non_filtered_period;
413 if (perf_evsel__is_group_event(evsel)) {
414 struct perf_evsel *pos;
416 perf_evsel__group_desc(evsel, buf, size);
419 for_each_group_member(pos, evsel) {
420 const struct hists *pos_hists = evsel__hists(pos);
422 if (symbol_conf.filter_relative) {
423 nr_samples += pos_hists->stats.nr_non_filtered_samples;
424 nr_events += pos_hists->stats.total_non_filtered_period;
426 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
427 nr_events += pos_hists->stats.total_period;
432 nr_samples = convert_unit(nr_samples, &unit);
433 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
434 if (evname != NULL) {
435 ret += fprintf(fp, " of event%s '%s'",
436 evsel->nr_members > 1 ? "s" : "", evname);
440 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
442 if (symbol_conf.show_ref_callgraph &&
443 strstr(evname, "call-graph=no")) {
444 ret += fprintf(fp, ", show reference callgraph");
448 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
449 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
451 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
454 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
456 return ret + fprintf(fp, "\n#\n");
459 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
463 struct perf_evsel *pos;
466 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
467 evlist->stats.total_lost_samples);
470 evlist__for_each_entry(evlist, pos) {
471 struct hists *hists = evsel__hists(pos);
472 const char *evname = perf_evsel__name(pos);
474 if (symbol_conf.event_group &&
475 !perf_evsel__is_group_leader(pos))
478 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
479 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
480 !(symbol_conf.use_callchain ||
481 symbol_conf.show_branchflag_count));
482 fprintf(stdout, "\n\n");
486 fprintf(stdout, "#\n# (%s)\n#\n", help);
488 if (rep->show_threads) {
489 bool style = !strcmp(rep->pretty_printing_style, "raw");
490 perf_read_values_display(stdout, &rep->show_threads_values,
492 perf_read_values_destroy(&rep->show_threads_values);
495 if (sort__mode == SORT_MODE__BRANCH)
496 branch_type_stat_display(stdout, &rep->brtype_stat);
501 static void report__warn_kptr_restrict(const struct report *rep)
503 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
504 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
506 if (perf_evlist__exclude_kernel(rep->session->evlist))
509 if (kernel_map == NULL ||
510 (kernel_map->dso->hit &&
511 (kernel_kmap->ref_reloc_sym == NULL ||
512 kernel_kmap->ref_reloc_sym->addr == 0))) {
514 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
515 "can't be resolved.";
517 if (kernel_map && map__has_symbols(kernel_map)) {
518 desc = "If some relocation was applied (e.g. "
519 "kexec) symbols may be misresolved.";
523 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
524 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
525 "Samples in kernel modules can't be resolved as well.\n\n",
530 static int report__gtk_browse_hists(struct report *rep, const char *help)
532 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
533 struct hist_browser_timer *timer, float min_pcnt);
535 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
537 if (hist_browser == NULL) {
538 ui__error("GTK browser not found!\n");
542 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
545 static int report__browse_hists(struct report *rep)
548 struct perf_session *session = rep->session;
549 struct perf_evlist *evlist = session->evlist;
550 const char *help = perf_tip(system_path(TIPDIR));
553 /* fallback for people who don't install perf ;-) */
554 help = perf_tip(DOCDIR);
556 help = "Cannot load tips.txt file, please install perf!";
559 switch (use_browser) {
561 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
563 &session->header.env,
564 true, &rep->annotation_opts);
566 * Usually "ret" is the last pressed key, and we only
567 * care if the key notifies us to switch data file.
569 if (ret != K_SWITCH_INPUT_DATA)
573 ret = report__gtk_browse_hists(rep, help);
576 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
583 static int report__collapse_hists(struct report *rep)
585 struct ui_progress prog;
586 struct perf_evsel *pos;
589 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
591 evlist__for_each_entry(rep->session->evlist, pos) {
592 struct hists *hists = evsel__hists(pos);
595 hists->symbol_filter_str = rep->symbol_filter_str;
597 hists->socket_filter = rep->socket_filter;
599 ret = hists__collapse_resort(hists, &prog);
603 /* Non-group events are considered as leader */
604 if (symbol_conf.event_group &&
605 !perf_evsel__is_group_leader(pos)) {
606 struct hists *leader_hists = evsel__hists(pos->leader);
608 hists__match(leader_hists, hists);
609 hists__link(leader_hists, hists);
613 ui_progress__finish();
617 static void report__output_resort(struct report *rep)
619 struct ui_progress prog;
620 struct perf_evsel *pos;
622 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
624 evlist__for_each_entry(rep->session->evlist, pos)
625 perf_evsel__output_resort(pos, &prog);
627 ui_progress__finish();
630 static void stats_setup(struct report *rep)
632 memset(&rep->tool, 0, sizeof(rep->tool));
633 rep->tool.no_warn = true;
636 static int stats_print(struct report *rep)
638 struct perf_session *session = rep->session;
640 perf_session__fprintf_nr_events(session, stdout);
644 static void tasks_setup(struct report *rep)
646 memset(&rep->tool, 0, sizeof(rep->tool));
647 rep->tool.ordered_events = true;
648 if (rep->mmaps_mode) {
649 rep->tool.mmap = perf_event__process_mmap;
650 rep->tool.mmap2 = perf_event__process_mmap2;
652 rep->tool.comm = perf_event__process_comm;
653 rep->tool.exit = perf_event__process_exit;
654 rep->tool.fork = perf_event__process_fork;
655 rep->tool.no_warn = true;
659 struct thread *thread;
660 struct list_head list;
661 struct list_head children;
664 static struct task *tasks_list(struct task *task, struct machine *machine)
666 struct thread *parent_thread, *thread = task->thread;
667 struct task *parent_task;
669 /* Already listed. */
670 if (!list_empty(&task->list))
673 /* Last one in the chain. */
674 if (thread->ppid == -1)
677 parent_thread = machine__find_thread(machine, -1, thread->ppid);
679 return ERR_PTR(-ENOENT);
681 parent_task = thread__priv(parent_thread);
682 list_add_tail(&task->list, &parent_task->children);
683 return tasks_list(parent_task, machine);
686 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
691 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) {
692 struct map *map = rb_entry(nd, struct map, rb_node);
694 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
695 indent, "", map->start, map->end,
696 map->prot & PROT_READ ? 'r' : '-',
697 map->prot & PROT_WRITE ? 'w' : '-',
698 map->prot & PROT_EXEC ? 'x' : '-',
699 map->flags & MAP_SHARED ? 's' : 'p',
701 map->ino, map->dso->name);
707 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp)
709 return maps__fprintf_task(&mg->maps, indent, fp);
712 static void task__print_level(struct task *task, FILE *fp, int level)
714 struct thread *thread = task->thread;
716 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
717 thread->pid_, thread->tid, thread->ppid,
720 fprintf(fp, "%s\n", thread__comm_str(thread));
722 map_groups__fprintf_task(thread->mg, comm_indent, fp);
724 if (!list_empty(&task->children)) {
725 list_for_each_entry(child, &task->children, list)
726 task__print_level(child, fp, level + 1);
730 static int tasks_print(struct report *rep, FILE *fp)
732 struct perf_session *session = rep->session;
733 struct machine *machine = &session->machines.host;
734 struct task *tasks, *task;
735 unsigned int nr = 0, itask = 0, i;
740 * No locking needed while accessing machine->threads,
741 * because --tasks is single threaded command.
744 /* Count all the threads. */
745 for (i = 0; i < THREADS__TABLE_SIZE; i++)
746 nr += machine->threads[i].nr;
748 tasks = malloc(sizeof(*tasks) * nr);
752 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
753 struct threads *threads = &machine->threads[i];
755 for (nd = rb_first(&threads->entries); nd; nd = rb_next(nd)) {
756 task = tasks + itask++;
758 task->thread = rb_entry(nd, struct thread, rb_node);
759 INIT_LIST_HEAD(&task->children);
760 INIT_LIST_HEAD(&task->list);
761 thread__set_priv(task->thread, task);
766 * Iterate every task down to the unprocessed parent
767 * and link all in task children list. Task with no
768 * parent is added into 'list'.
770 for (itask = 0; itask < nr; itask++) {
771 task = tasks + itask;
773 if (!list_empty(&task->list))
776 task = tasks_list(task, machine);
778 pr_err("Error: failed to process tasks\n");
780 return PTR_ERR(task);
784 list_add_tail(&task->list, &list);
787 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
789 list_for_each_entry(task, &list, list)
790 task__print_level(task, fp, 0);
796 static int __cmd_report(struct report *rep)
799 struct perf_session *session = rep->session;
800 struct perf_evsel *pos;
801 struct perf_data *data = session->data;
803 signal(SIGINT, sig_handler);
806 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
809 ui__error("failed to set cpu bitmap\n");
812 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
815 if (rep->show_threads) {
816 ret = perf_read_values_init(&rep->show_threads_values);
821 ret = report__setup_sample_type(rep);
823 /* report__setup_sample_type() already showed error message */
833 ret = perf_session__process_events(session);
835 ui__error("failed to process sample\n");
840 return stats_print(rep);
843 return tasks_print(rep, stdout);
845 report__warn_kptr_restrict(rep);
847 evlist__for_each_entry(session->evlist, pos)
848 rep->nr_entries += evsel__hists(pos)->nr_entries;
850 if (use_browser == 0) {
852 perf_session__fprintf(session, stdout);
855 perf_session__fprintf_dsos(session, stdout);
858 perf_session__fprintf_nr_events(session, stdout);
859 perf_evlist__fprintf_nr_events(session->evlist, stdout);
864 ret = report__collapse_hists(rep);
866 ui__error("failed to process hist entry\n");
874 * recalculate number of entries after collapsing since it
875 * might be changed during the collapse phase.
878 evlist__for_each_entry(session->evlist, pos)
879 rep->nr_entries += evsel__hists(pos)->nr_entries;
881 if (rep->nr_entries == 0) {
882 ui__error("The %s file has no samples!\n", data->file.path);
886 report__output_resort(rep);
888 return report__browse_hists(rep);
892 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
894 struct callchain_param *callchain = opt->value;
896 callchain->enabled = !unset;
901 symbol_conf.use_callchain = false;
902 callchain->mode = CHAIN_NONE;
906 return parse_callchain_report_opt(arg);
910 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
911 const char *arg, int unset __maybe_unused)
914 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
917 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
918 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
921 have_ignore_callees = 1;
928 parse_branch_mode(const struct option *opt,
929 const char *str __maybe_unused, int unset)
931 int *branch_mode = opt->value;
933 *branch_mode = !unset;
938 parse_percent_limit(const struct option *opt, const char *str,
939 int unset __maybe_unused)
941 struct report *rep = opt->value;
942 double pcnt = strtof(str, NULL);
944 rep->min_percent = pcnt;
945 callchain_param.min_percent = pcnt;
949 int cmd_report(int argc, const char **argv)
951 struct perf_session *session;
952 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
954 bool has_br_stack = false;
955 int branch_mode = -1;
956 bool branch_call_mode = false;
957 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
958 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
959 CALLCHAIN_REPORT_HELP
960 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
961 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
962 const char * const report_usage[] = {
963 "perf report [<options>]",
966 struct report report = {
968 .sample = process_sample_event,
969 .mmap = perf_event__process_mmap,
970 .mmap2 = perf_event__process_mmap2,
971 .comm = perf_event__process_comm,
972 .namespaces = perf_event__process_namespaces,
973 .exit = perf_event__process_exit,
974 .fork = perf_event__process_fork,
975 .lost = perf_event__process_lost,
976 .read = process_read_event,
977 .attr = perf_event__process_attr,
978 .tracing_data = perf_event__process_tracing_data,
979 .build_id = perf_event__process_build_id,
980 .id_index = perf_event__process_id_index,
981 .auxtrace_info = perf_event__process_auxtrace_info,
982 .auxtrace = perf_event__process_auxtrace,
983 .feature = process_feature_event,
984 .ordered_events = true,
985 .ordering_requires_timestamps = true,
987 .max_stack = PERF_MAX_STACK_DEPTH,
988 .pretty_printing_style = "normal",
990 .annotation_opts = annotation__default_options,
992 const struct option options[] = {
993 OPT_STRING('i', "input", &input_name, "file",
995 OPT_INCR('v', "verbose", &verbose,
996 "be more verbose (show symbol address, etc)"),
997 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
998 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
999 "dump raw trace in ASCII"),
1000 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1001 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1002 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1003 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1004 "file", "vmlinux pathname"),
1005 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1006 "don't load vmlinux even if found"),
1007 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1008 "file", "kallsyms pathname"),
1009 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1010 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1011 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1012 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1013 "Show a column with the number of samples"),
1014 OPT_BOOLEAN('T', "threads", &report.show_threads,
1015 "Show per-thread event counters"),
1016 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1017 "pretty printing style key: normal raw"),
1018 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1019 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1020 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1021 "Use the stdio interface"),
1022 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1023 OPT_BOOLEAN(0, "header-only", &report.header_only,
1024 "Show only data header."),
1025 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1026 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1027 " Please refer the man page for the complete list."),
1028 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1029 "output field(s): overhead, period, sample plus all of sort keys"),
1030 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1031 "Show sample percentage for different cpu modes"),
1032 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1033 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1034 OPT_STRING('p', "parent", &parent_pattern, "regex",
1035 "regex filter to identify parent, see: '--sort parent'"),
1036 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1037 "Only display entries with parent-match"),
1038 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1039 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1040 report_callchain_help, &report_parse_callchain_opt,
1041 callchain_default_opt),
1042 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1043 "Accumulate callchains of children and show total overhead as well"),
1044 OPT_INTEGER(0, "max-stack", &report.max_stack,
1045 "Set the maximum stack depth when parsing the callchain, "
1046 "anything beyond the specified depth will be ignored. "
1047 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1048 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1049 "alias for inverted call graph"),
1050 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1051 "ignore callees of these functions in call graphs",
1052 report_parse_ignore_callees_opt),
1053 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1054 "only consider symbols in these dsos"),
1055 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1056 "only consider symbols in these comms"),
1057 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1058 "only consider symbols in these pids"),
1059 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1060 "only consider symbols in these tids"),
1061 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1062 "only consider these symbols"),
1063 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1064 "only show symbols that (partially) match with this filter"),
1065 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1067 "don't try to adjust column width, use these fixed values"),
1068 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1069 "separator for columns, no spaces will be added between "
1070 "columns '.' is reserved."),
1071 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1072 "Only display entries resolved to a symbol"),
1073 OPT_CALLBACK(0, "symfs", NULL, "directory",
1074 "Look for files with symbols relative to this directory",
1075 symbol__config_symfs),
1076 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1077 "list of cpus to profile"),
1078 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1079 "Display extended information about perf.data file"),
1080 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1081 "Interleave source code with assembly code (default)"),
1082 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1083 "Display raw encoding of assembly instructions (default)"),
1084 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1085 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1086 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1087 "Show a column with the sum of periods"),
1088 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1089 "Show event group information together"),
1090 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1091 "use branch records for per branch histogram filling",
1093 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1094 "add last branch records to call history"),
1095 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1096 "objdump binary to use for disassembly and annotations"),
1097 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1098 "Disable symbol demangling"),
1099 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1100 "Enable kernel symbol demangling"),
1101 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1102 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1103 "Don't show entries under that percent", parse_percent_limit),
1104 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1105 "how to display percentage of filtered entries", parse_filter_percentage),
1106 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1107 "Instruction Tracing options\n" ITRACE_HELP,
1108 itrace_parse_synth_opts),
1109 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1110 "Show full source file name path for source lines"),
1111 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1112 "Show callgraph from reference event"),
1113 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1114 "only show processor socket that match with this filter"),
1115 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1116 "Show raw trace event output (do not use print fmt or plugins)"),
1117 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1118 "Show entries in a hierarchy"),
1119 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1120 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1121 stdio__config_color, "always"),
1122 OPT_STRING(0, "time", &report.time_str, "str",
1123 "Time span of interest (start,stop)"),
1124 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1125 "Show inline function"),
1126 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1127 "Set percent type local/global-period/hits",
1128 annotate_parse_percent_type),
1131 struct perf_data data = {
1132 .mode = PERF_DATA_MODE_READ,
1134 int ret = hists__init();
1139 ret = perf_config(report__config, &report);
1143 argc = parse_options(argc, argv, options, report_usage, 0);
1146 * Special case: if there's an argument left then assume that
1147 * it's a symbol filter:
1150 usage_with_options(report_usage, options);
1152 report.symbol_filter_str = argv[0];
1155 if (report.mmaps_mode)
1156 report.tasks_mode = true;
1159 perf_quiet_option();
1161 if (symbol_conf.vmlinux_name &&
1162 access(symbol_conf.vmlinux_name, R_OK)) {
1163 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1166 if (symbol_conf.kallsyms_name &&
1167 access(symbol_conf.kallsyms_name, R_OK)) {
1168 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1172 if (report.inverted_callchain)
1173 callchain_param.order = ORDER_CALLER;
1174 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1175 callchain_param.order = ORDER_CALLER;
1177 if (itrace_synth_opts.callchain &&
1178 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1179 report.max_stack = itrace_synth_opts.callchain_sz;
1181 if (!input_name || !strlen(input_name)) {
1182 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1185 input_name = "perf.data";
1188 data.file.path = input_name;
1189 data.force = symbol_conf.force;
1192 session = perf_session__new(&data, false, &report.tool);
1193 if (session == NULL)
1196 if (report.queue_size) {
1197 ordered_events__set_alloc_size(&session->ordered_events,
1201 session->itrace_synth_opts = &itrace_synth_opts;
1203 report.session = session;
1205 has_br_stack = perf_header__has_feat(&session->header,
1206 HEADER_BRANCH_STACK);
1208 setup_forced_leader(&report, session->evlist);
1210 if (itrace_synth_opts.last_branch)
1211 has_br_stack = true;
1213 if (has_br_stack && branch_call_mode)
1214 symbol_conf.show_branchflag_count = true;
1216 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1219 * Branch mode is a tristate:
1220 * -1 means default, so decide based on the file having branch data.
1221 * 0/1 means the user chose a mode.
1223 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1224 !branch_call_mode) {
1225 sort__mode = SORT_MODE__BRANCH;
1226 symbol_conf.cumulate_callchain = false;
1228 if (branch_call_mode) {
1229 callchain_param.key = CCKEY_ADDRESS;
1230 callchain_param.branch_callstack = 1;
1231 symbol_conf.use_callchain = true;
1232 callchain_register_param(&callchain_param);
1233 if (sort_order == NULL)
1234 sort_order = "srcline,symbol,dso";
1237 if (report.mem_mode) {
1238 if (sort__mode == SORT_MODE__BRANCH) {
1239 pr_err("branch and mem mode incompatible\n");
1242 sort__mode = SORT_MODE__MEMORY;
1243 symbol_conf.cumulate_callchain = false;
1246 if (symbol_conf.report_hierarchy) {
1247 /* disable incompatible options */
1248 symbol_conf.cumulate_callchain = false;
1251 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1252 parse_options_usage(report_usage, options, "F", 1);
1253 parse_options_usage(NULL, options, "hierarchy", 0);
1257 perf_hpp_list.need_collapse = true;
1260 if (report.use_stdio)
1262 else if (report.use_tui)
1264 else if (report.use_gtk)
1267 /* Force tty output for header output and per-thread stat. */
1268 if (report.header || report.header_only || report.show_threads)
1270 if (report.header || report.header_only)
1271 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1272 if (report.show_full_info)
1273 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1274 if (report.stats_mode || report.tasks_mode)
1276 if (report.stats_mode && report.tasks_mode) {
1277 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1281 if (strcmp(input_name, "-") != 0)
1282 setup_browser(true);
1286 if (setup_sorting(session->evlist) < 0) {
1288 parse_options_usage(report_usage, options, "s", 1);
1290 parse_options_usage(sort_order ? NULL : report_usage,
1295 if ((report.header || report.header_only) && !quiet) {
1296 perf_session__fprintf_info(session, stdout,
1297 report.show_full_info);
1298 if (report.header_only) {
1302 } else if (use_browser == 0 && !quiet &&
1303 !report.stats_mode && !report.tasks_mode) {
1304 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1309 * Only in the TUI browser we are doing integrated annotation,
1310 * so don't allocate extra space that won't be used in the stdio
1313 if (ui__has_annotation()) {
1314 ret = symbol__annotation_init();
1318 * For searching by name on the "Browse map details".
1319 * providing it only in verbose mode not to bloat too
1320 * much struct symbol.
1324 * XXX: Need to provide a less kludgy way to ask for
1325 * more space per symbol, the u32 is for the index on
1327 * See symbol__browser_index.
1329 symbol_conf.priv_size += sizeof(u32);
1330 symbol_conf.sort_by_name = true;
1332 annotation_config__init();
1335 if (symbol__init(&session->header.env) < 0)
1338 report.ptime_range = perf_time__range_alloc(report.time_str,
1339 &report.range_size);
1340 if (!report.ptime_range) {
1345 if (perf_time__parse_str(report.ptime_range, report.time_str) != 0) {
1346 if (session->evlist->first_sample_time == 0 &&
1347 session->evlist->last_sample_time == 0) {
1348 pr_err("HINT: no first/last sample time found in perf data.\n"
1349 "Please use latest perf binary to execute 'perf record'\n"
1350 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n");
1355 report.range_num = perf_time__percent_parse_str(
1356 report.ptime_range, report.range_size,
1358 session->evlist->first_sample_time,
1359 session->evlist->last_sample_time);
1361 if (report.range_num < 0) {
1362 pr_err("Invalid time string\n");
1367 report.range_num = 1;
1370 if (session->tevent.pevent &&
1371 tep_set_function_resolver(session->tevent.pevent,
1372 machine__resolve_kernel_addr,
1373 &session->machines.host) < 0) {
1374 pr_err("%s: failed to set libtraceevent function resolver\n",
1379 sort__setup_elide(stdout);
1381 ret = __cmd_report(&report);
1382 if (ret == K_SWITCH_INPUT_DATA) {
1383 perf_session__delete(session);
1389 zfree(&report.ptime_range);
1391 perf_session__delete(session);