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>
20 #include "util/symbol.h"
21 #include "util/callchain.h"
22 #include "util/values.h"
25 #include "util/debug.h"
26 #include "util/evlist.h"
27 #include "util/evsel.h"
28 #include "util/header.h"
29 #include "util/session.h"
30 #include "util/tool.h"
32 #include <subcmd/parse-options.h>
33 #include <subcmd/exec-cmd.h>
34 #include "util/parse-events.h"
36 #include "util/thread.h"
37 #include "util/sort.h"
38 #include "util/hist.h"
39 #include "util/data.h"
40 #include "arch/common.h"
41 #include "util/time-utils.h"
42 #include "util/auxtrace.h"
43 #include "util/units.h"
44 #include "util/branch.h"
51 #include <linux/bitmap.h>
52 #include <linux/stringify.h>
53 #include <sys/types.h>
56 #include <linux/mman.h>
59 struct perf_tool tool;
60 struct perf_session *session;
61 bool use_tui, use_gtk, use_stdio;
64 bool inverted_callchain;
71 bool nonany_branch_mode;
74 struct perf_read_values show_threads_values;
75 struct annotation_options annotation_opts;
76 const char *pretty_printing_style;
78 const char *symbol_filter_str;
80 struct perf_time_interval *ptime_range;
87 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
88 struct branch_type_stat brtype_stat;
92 static int report__config(const char *var, const char *value, void *cb)
94 struct report *rep = cb;
96 if (!strcmp(var, "report.group")) {
97 symbol_conf.event_group = perf_config_bool(var, value);
100 if (!strcmp(var, "report.percent-limit")) {
101 double pcnt = strtof(value, NULL);
103 rep->min_percent = pcnt;
104 callchain_param.min_percent = pcnt;
107 if (!strcmp(var, "report.children")) {
108 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
111 if (!strcmp(var, "report.queue-size"))
112 return perf_config_u64(&rep->queue_size, var, value);
114 if (!strcmp(var, "report.sort_order")) {
115 default_sort_order = strdup(value);
122 static int hist_iter__report_callback(struct hist_entry_iter *iter,
123 struct addr_location *al, bool single,
127 struct report *rep = arg;
128 struct hist_entry *he = iter->he;
129 struct perf_evsel *evsel = iter->evsel;
130 struct perf_sample *sample = iter->sample;
132 struct branch_info *bi;
134 if (!ui__has_annotation() && !rep->symbol_ipc)
137 hist__account_cycles(sample->branch_stack, al, sample,
138 rep->nonany_branch_mode);
140 if (sort__mode == SORT_MODE__BRANCH) {
141 bi = he->branch_info;
142 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
146 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
148 } else if (rep->mem_mode) {
150 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
154 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
156 } else if (symbol_conf.cumulate_callchain) {
158 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
160 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
167 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
168 struct addr_location *al __maybe_unused,
169 bool single __maybe_unused,
172 struct hist_entry *he = iter->he;
173 struct report *rep = arg;
174 struct branch_info *bi;
175 struct perf_sample *sample = iter->sample;
176 struct perf_evsel *evsel = iter->evsel;
179 if (!ui__has_annotation() && !rep->symbol_ipc)
182 hist__account_cycles(sample->branch_stack, al, sample,
183 rep->nonany_branch_mode);
185 bi = he->branch_info;
186 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
190 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
192 branch_type_count(&rep->brtype_stat, &bi->flags,
193 bi->from.addr, bi->to.addr);
199 static void setup_forced_leader(struct report *report,
200 struct perf_evlist *evlist)
202 if (report->group_set)
203 perf_evlist__force_leader(evlist);
206 static int process_feature_event(struct perf_session *session,
207 union perf_event *event)
209 struct report *rep = container_of(session->tool, struct report, tool);
211 if (event->feat.feat_id < HEADER_LAST_FEATURE)
212 return perf_event__process_feature(session, event);
214 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
215 pr_err("failed: wrong feature ID: %" PRIu64 "\n",
216 event->feat.feat_id);
221 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
222 * means all features are received, now we can force the
225 setup_forced_leader(rep, session->evlist);
229 static int process_sample_event(struct perf_tool *tool,
230 union perf_event *event,
231 struct perf_sample *sample,
232 struct perf_evsel *evsel,
233 struct machine *machine)
235 struct report *rep = container_of(tool, struct report, tool);
236 struct addr_location al;
237 struct hist_entry_iter iter = {
240 .hide_unresolved = symbol_conf.hide_unresolved,
241 .add_entry_cb = hist_iter__report_callback,
245 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
250 if (machine__resolve(machine, &al, sample) < 0) {
251 pr_debug("problem processing %d event, skipping it.\n",
256 if (symbol_conf.hide_unresolved && al.sym == NULL)
259 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
262 if (sort__mode == SORT_MODE__BRANCH) {
264 * A non-synthesized event might not have a branch stack if
265 * branch stacks have been synthesized (using itrace options).
267 if (!sample->branch_stack)
270 iter.add_entry_cb = hist_iter__branch_callback;
271 iter.ops = &hist_iter_branch;
272 } else if (rep->mem_mode) {
273 iter.ops = &hist_iter_mem;
274 } else if (symbol_conf.cumulate_callchain) {
275 iter.ops = &hist_iter_cumulative;
277 iter.ops = &hist_iter_normal;
281 al.map->dso->hit = 1;
283 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
285 pr_debug("problem adding hist entry, skipping event\n");
287 addr_location__put(&al);
291 static int process_read_event(struct perf_tool *tool,
292 union perf_event *event,
293 struct perf_sample *sample __maybe_unused,
294 struct perf_evsel *evsel,
295 struct machine *machine __maybe_unused)
297 struct report *rep = container_of(tool, struct report, tool);
299 if (rep->show_threads) {
300 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
301 int err = perf_read_values_add_value(&rep->show_threads_values,
302 event->read.pid, event->read.tid,
314 /* For pipe mode, sample_type is not currently set */
315 static int report__setup_sample_type(struct report *rep)
317 struct perf_session *session = rep->session;
318 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
319 bool is_pipe = perf_data__is_pipe(session->data);
321 if (session->itrace_synth_opts->callchain ||
323 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
324 !session->itrace_synth_opts->set))
325 sample_type |= PERF_SAMPLE_CALLCHAIN;
327 if (session->itrace_synth_opts->last_branch)
328 sample_type |= PERF_SAMPLE_BRANCH_STACK;
330 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
331 if (perf_hpp_list.parent) {
332 ui__error("Selected --sort parent, but no "
333 "callchain data. Did you call "
334 "'perf record' without -g?\n");
337 if (symbol_conf.use_callchain &&
338 !symbol_conf.show_branchflag_count) {
339 ui__error("Selected -g or --branch-history.\n"
340 "But no callchain or branch data.\n"
341 "Did you call 'perf record' without -g or -b?\n");
344 } else if (!callchain_param.enabled &&
345 callchain_param.mode != CHAIN_NONE &&
346 !symbol_conf.use_callchain) {
347 symbol_conf.use_callchain = true;
348 if (callchain_register_param(&callchain_param) < 0) {
349 ui__error("Can't register callchain params.\n");
354 if (symbol_conf.cumulate_callchain) {
355 /* Silently ignore if callchain is missing */
356 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
357 symbol_conf.cumulate_callchain = false;
358 perf_hpp__cancel_cumulate();
362 if (sort__mode == SORT_MODE__BRANCH) {
364 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
365 ui__error("Selected -b but no branch data. "
366 "Did you call perf record without -b?\n");
371 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
372 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
373 (sample_type & PERF_SAMPLE_STACK_USER)) {
374 callchain_param.record_mode = CALLCHAIN_DWARF;
375 dwarf_callchain_users = true;
376 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
377 callchain_param.record_mode = CALLCHAIN_LBR;
379 callchain_param.record_mode = CALLCHAIN_FP;
382 /* ??? handle more cases than just ANY? */
383 if (!(perf_evlist__combined_branch_type(session->evlist) &
384 PERF_SAMPLE_BRANCH_ANY))
385 rep->nonany_branch_mode = true;
390 static void sig_handler(int sig __maybe_unused)
395 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
396 const char *evname, FILE *fp)
400 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
401 u64 nr_events = hists->stats.total_period;
402 struct perf_evsel *evsel = hists_to_evsel(hists);
404 size_t size = sizeof(buf);
405 int socked_id = hists->socket_filter;
410 if (symbol_conf.filter_relative) {
411 nr_samples = hists->stats.nr_non_filtered_samples;
412 nr_events = hists->stats.total_non_filtered_period;
415 if (perf_evsel__is_group_event(evsel)) {
416 struct perf_evsel *pos;
418 perf_evsel__group_desc(evsel, buf, size);
421 for_each_group_member(pos, evsel) {
422 const struct hists *pos_hists = evsel__hists(pos);
424 if (symbol_conf.filter_relative) {
425 nr_samples += pos_hists->stats.nr_non_filtered_samples;
426 nr_events += pos_hists->stats.total_non_filtered_period;
428 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
429 nr_events += pos_hists->stats.total_period;
434 nr_samples = convert_unit(nr_samples, &unit);
435 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
436 if (evname != NULL) {
437 ret += fprintf(fp, " of event%s '%s'",
438 evsel->nr_members > 1 ? "s" : "", evname);
442 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
444 if (symbol_conf.show_ref_callgraph &&
445 strstr(evname, "call-graph=no")) {
446 ret += fprintf(fp, ", show reference callgraph");
450 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
451 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
453 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
456 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
458 return ret + fprintf(fp, "\n#\n");
461 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
465 struct perf_evsel *pos;
468 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
469 evlist->stats.total_lost_samples);
472 evlist__for_each_entry(evlist, pos) {
473 struct hists *hists = evsel__hists(pos);
474 const char *evname = perf_evsel__name(pos);
476 if (symbol_conf.event_group &&
477 !perf_evsel__is_group_leader(pos))
480 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
481 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
482 !(symbol_conf.use_callchain ||
483 symbol_conf.show_branchflag_count));
484 fprintf(stdout, "\n\n");
488 fprintf(stdout, "#\n# (%s)\n#\n", help);
490 if (rep->show_threads) {
491 bool style = !strcmp(rep->pretty_printing_style, "raw");
492 perf_read_values_display(stdout, &rep->show_threads_values,
494 perf_read_values_destroy(&rep->show_threads_values);
497 if (sort__mode == SORT_MODE__BRANCH)
498 branch_type_stat_display(stdout, &rep->brtype_stat);
503 static void report__warn_kptr_restrict(const struct report *rep)
505 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
506 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
508 if (perf_evlist__exclude_kernel(rep->session->evlist))
511 if (kernel_map == NULL ||
512 (kernel_map->dso->hit &&
513 (kernel_kmap->ref_reloc_sym == NULL ||
514 kernel_kmap->ref_reloc_sym->addr == 0))) {
516 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
517 "can't be resolved.";
519 if (kernel_map && map__has_symbols(kernel_map)) {
520 desc = "If some relocation was applied (e.g. "
521 "kexec) symbols may be misresolved.";
525 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
526 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
527 "Samples in kernel modules can't be resolved as well.\n\n",
532 static int report__gtk_browse_hists(struct report *rep, const char *help)
534 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
535 struct hist_browser_timer *timer, float min_pcnt);
537 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
539 if (hist_browser == NULL) {
540 ui__error("GTK browser not found!\n");
544 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
547 static int report__browse_hists(struct report *rep)
550 struct perf_session *session = rep->session;
551 struct perf_evlist *evlist = session->evlist;
552 const char *help = perf_tip(system_path(TIPDIR));
555 /* fallback for people who don't install perf ;-) */
556 help = perf_tip(DOCDIR);
558 help = "Cannot load tips.txt file, please install perf!";
561 switch (use_browser) {
563 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
565 &session->header.env,
566 true, &rep->annotation_opts);
568 * Usually "ret" is the last pressed key, and we only
569 * care if the key notifies us to switch data file.
571 if (ret != K_SWITCH_INPUT_DATA)
575 ret = report__gtk_browse_hists(rep, help);
578 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
585 static int report__collapse_hists(struct report *rep)
587 struct ui_progress prog;
588 struct perf_evsel *pos;
591 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
593 evlist__for_each_entry(rep->session->evlist, pos) {
594 struct hists *hists = evsel__hists(pos);
597 hists->symbol_filter_str = rep->symbol_filter_str;
599 hists->socket_filter = rep->socket_filter;
601 ret = hists__collapse_resort(hists, &prog);
605 /* Non-group events are considered as leader */
606 if (symbol_conf.event_group &&
607 !perf_evsel__is_group_leader(pos)) {
608 struct hists *leader_hists = evsel__hists(pos->leader);
610 hists__match(leader_hists, hists);
611 hists__link(leader_hists, hists);
615 ui_progress__finish();
619 static int hists__resort_cb(struct hist_entry *he, void *arg)
621 struct report *rep = arg;
622 struct symbol *sym = he->ms.sym;
624 if (rep->symbol_ipc && sym && !sym->annotate2) {
625 struct perf_evsel *evsel = hists_to_evsel(he->hists);
627 symbol__annotate2(sym, he->ms.map, evsel,
628 &annotation__default_options, NULL);
634 static void report__output_resort(struct report *rep)
636 struct ui_progress prog;
637 struct perf_evsel *pos;
639 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
641 evlist__for_each_entry(rep->session->evlist, pos) {
642 perf_evsel__output_resort_cb(pos, &prog,
643 hists__resort_cb, rep);
646 ui_progress__finish();
649 static void stats_setup(struct report *rep)
651 memset(&rep->tool, 0, sizeof(rep->tool));
652 rep->tool.no_warn = true;
655 static int stats_print(struct report *rep)
657 struct perf_session *session = rep->session;
659 perf_session__fprintf_nr_events(session, stdout);
663 static void tasks_setup(struct report *rep)
665 memset(&rep->tool, 0, sizeof(rep->tool));
666 rep->tool.ordered_events = true;
667 if (rep->mmaps_mode) {
668 rep->tool.mmap = perf_event__process_mmap;
669 rep->tool.mmap2 = perf_event__process_mmap2;
671 rep->tool.comm = perf_event__process_comm;
672 rep->tool.exit = perf_event__process_exit;
673 rep->tool.fork = perf_event__process_fork;
674 rep->tool.no_warn = true;
678 struct thread *thread;
679 struct list_head list;
680 struct list_head children;
683 static struct task *tasks_list(struct task *task, struct machine *machine)
685 struct thread *parent_thread, *thread = task->thread;
686 struct task *parent_task;
688 /* Already listed. */
689 if (!list_empty(&task->list))
692 /* Last one in the chain. */
693 if (thread->ppid == -1)
696 parent_thread = machine__find_thread(machine, -1, thread->ppid);
698 return ERR_PTR(-ENOENT);
700 parent_task = thread__priv(parent_thread);
701 list_add_tail(&task->list, &parent_task->children);
702 return tasks_list(parent_task, machine);
705 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
710 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) {
711 struct map *map = rb_entry(nd, struct map, rb_node);
713 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
714 indent, "", map->start, map->end,
715 map->prot & PROT_READ ? 'r' : '-',
716 map->prot & PROT_WRITE ? 'w' : '-',
717 map->prot & PROT_EXEC ? 'x' : '-',
718 map->flags & MAP_SHARED ? 's' : 'p',
720 map->ino, map->dso->name);
726 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp)
728 return maps__fprintf_task(&mg->maps, indent, fp);
731 static void task__print_level(struct task *task, FILE *fp, int level)
733 struct thread *thread = task->thread;
735 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
736 thread->pid_, thread->tid, thread->ppid,
739 fprintf(fp, "%s\n", thread__comm_str(thread));
741 map_groups__fprintf_task(thread->mg, comm_indent, fp);
743 if (!list_empty(&task->children)) {
744 list_for_each_entry(child, &task->children, list)
745 task__print_level(child, fp, level + 1);
749 static int tasks_print(struct report *rep, FILE *fp)
751 struct perf_session *session = rep->session;
752 struct machine *machine = &session->machines.host;
753 struct task *tasks, *task;
754 unsigned int nr = 0, itask = 0, i;
759 * No locking needed while accessing machine->threads,
760 * because --tasks is single threaded command.
763 /* Count all the threads. */
764 for (i = 0; i < THREADS__TABLE_SIZE; i++)
765 nr += machine->threads[i].nr;
767 tasks = malloc(sizeof(*tasks) * nr);
771 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
772 struct threads *threads = &machine->threads[i];
774 for (nd = rb_first_cached(&threads->entries); nd;
776 task = tasks + itask++;
778 task->thread = rb_entry(nd, struct thread, rb_node);
779 INIT_LIST_HEAD(&task->children);
780 INIT_LIST_HEAD(&task->list);
781 thread__set_priv(task->thread, task);
786 * Iterate every task down to the unprocessed parent
787 * and link all in task children list. Task with no
788 * parent is added into 'list'.
790 for (itask = 0; itask < nr; itask++) {
791 task = tasks + itask;
793 if (!list_empty(&task->list))
796 task = tasks_list(task, machine);
798 pr_err("Error: failed to process tasks\n");
800 return PTR_ERR(task);
804 list_add_tail(&task->list, &list);
807 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
809 list_for_each_entry(task, &list, list)
810 task__print_level(task, fp, 0);
816 static int __cmd_report(struct report *rep)
819 struct perf_session *session = rep->session;
820 struct perf_evsel *pos;
821 struct perf_data *data = session->data;
823 signal(SIGINT, sig_handler);
826 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
829 ui__error("failed to set cpu bitmap\n");
832 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
835 if (rep->show_threads) {
836 ret = perf_read_values_init(&rep->show_threads_values);
841 ret = report__setup_sample_type(rep);
843 /* report__setup_sample_type() already showed error message */
853 ret = perf_session__process_events(session);
855 ui__error("failed to process sample\n");
860 return stats_print(rep);
863 return tasks_print(rep, stdout);
865 report__warn_kptr_restrict(rep);
867 evlist__for_each_entry(session->evlist, pos)
868 rep->nr_entries += evsel__hists(pos)->nr_entries;
870 if (use_browser == 0) {
872 perf_session__fprintf(session, stdout);
875 perf_session__fprintf_dsos(session, stdout);
878 perf_session__fprintf_nr_events(session, stdout);
879 perf_evlist__fprintf_nr_events(session->evlist, stdout);
884 ret = report__collapse_hists(rep);
886 ui__error("failed to process hist entry\n");
894 * recalculate number of entries after collapsing since it
895 * might be changed during the collapse phase.
898 evlist__for_each_entry(session->evlist, pos)
899 rep->nr_entries += evsel__hists(pos)->nr_entries;
901 if (rep->nr_entries == 0) {
902 ui__error("The %s file has no samples!\n", data->file.path);
906 report__output_resort(rep);
908 return report__browse_hists(rep);
912 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
914 struct callchain_param *callchain = opt->value;
916 callchain->enabled = !unset;
921 symbol_conf.use_callchain = false;
922 callchain->mode = CHAIN_NONE;
926 return parse_callchain_report_opt(arg);
930 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
931 const char *arg, int unset __maybe_unused)
934 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
937 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
938 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
941 have_ignore_callees = 1;
948 parse_branch_mode(const struct option *opt,
949 const char *str __maybe_unused, int unset)
951 int *branch_mode = opt->value;
953 *branch_mode = !unset;
958 parse_percent_limit(const struct option *opt, const char *str,
959 int unset __maybe_unused)
961 struct report *rep = opt->value;
962 double pcnt = strtof(str, NULL);
964 rep->min_percent = pcnt;
965 callchain_param.min_percent = pcnt;
969 int cmd_report(int argc, const char **argv)
971 struct perf_session *session;
972 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
974 bool has_br_stack = false;
975 int branch_mode = -1;
976 bool branch_call_mode = false;
977 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
978 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
979 CALLCHAIN_REPORT_HELP
980 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
981 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
982 const char * const report_usage[] = {
983 "perf report [<options>]",
986 struct report report = {
988 .sample = process_sample_event,
989 .mmap = perf_event__process_mmap,
990 .mmap2 = perf_event__process_mmap2,
991 .comm = perf_event__process_comm,
992 .namespaces = perf_event__process_namespaces,
993 .exit = perf_event__process_exit,
994 .fork = perf_event__process_fork,
995 .lost = perf_event__process_lost,
996 .read = process_read_event,
997 .attr = perf_event__process_attr,
998 .tracing_data = perf_event__process_tracing_data,
999 .build_id = perf_event__process_build_id,
1000 .id_index = perf_event__process_id_index,
1001 .auxtrace_info = perf_event__process_auxtrace_info,
1002 .auxtrace = perf_event__process_auxtrace,
1003 .event_update = perf_event__process_event_update,
1004 .feature = process_feature_event,
1005 .ordered_events = true,
1006 .ordering_requires_timestamps = true,
1008 .max_stack = PERF_MAX_STACK_DEPTH,
1009 .pretty_printing_style = "normal",
1010 .socket_filter = -1,
1011 .annotation_opts = annotation__default_options,
1013 const struct option options[] = {
1014 OPT_STRING('i', "input", &input_name, "file",
1016 OPT_INCR('v', "verbose", &verbose,
1017 "be more verbose (show symbol address, etc)"),
1018 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1019 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1020 "dump raw trace in ASCII"),
1021 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1022 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1023 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1024 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1025 "file", "vmlinux pathname"),
1026 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1027 "don't load vmlinux even if found"),
1028 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1029 "file", "kallsyms pathname"),
1030 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1031 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1032 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1033 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1034 "Show a column with the number of samples"),
1035 OPT_BOOLEAN('T', "threads", &report.show_threads,
1036 "Show per-thread event counters"),
1037 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1038 "pretty printing style key: normal raw"),
1039 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1040 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1041 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1042 "Use the stdio interface"),
1043 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1044 OPT_BOOLEAN(0, "header-only", &report.header_only,
1045 "Show only data header."),
1046 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1047 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1048 " Please refer the man page for the complete list."),
1049 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1050 "output field(s): overhead, period, sample plus all of sort keys"),
1051 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1052 "Show sample percentage for different cpu modes"),
1053 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1054 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1055 OPT_STRING('p', "parent", &parent_pattern, "regex",
1056 "regex filter to identify parent, see: '--sort parent'"),
1057 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1058 "Only display entries with parent-match"),
1059 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1060 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1061 report_callchain_help, &report_parse_callchain_opt,
1062 callchain_default_opt),
1063 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1064 "Accumulate callchains of children and show total overhead as well"),
1065 OPT_INTEGER(0, "max-stack", &report.max_stack,
1066 "Set the maximum stack depth when parsing the callchain, "
1067 "anything beyond the specified depth will be ignored. "
1068 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1069 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1070 "alias for inverted call graph"),
1071 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1072 "ignore callees of these functions in call graphs",
1073 report_parse_ignore_callees_opt),
1074 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1075 "only consider symbols in these dsos"),
1076 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1077 "only consider symbols in these comms"),
1078 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1079 "only consider symbols in these pids"),
1080 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1081 "only consider symbols in these tids"),
1082 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1083 "only consider these symbols"),
1084 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1085 "only show symbols that (partially) match with this filter"),
1086 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1088 "don't try to adjust column width, use these fixed values"),
1089 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1090 "separator for columns, no spaces will be added between "
1091 "columns '.' is reserved."),
1092 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1093 "Only display entries resolved to a symbol"),
1094 OPT_CALLBACK(0, "symfs", NULL, "directory",
1095 "Look for files with symbols relative to this directory",
1096 symbol__config_symfs),
1097 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1098 "list of cpus to profile"),
1099 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1100 "Display extended information about perf.data file"),
1101 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1102 "Interleave source code with assembly code (default)"),
1103 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1104 "Display raw encoding of assembly instructions (default)"),
1105 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1106 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1107 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1108 "Show a column with the sum of periods"),
1109 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1110 "Show event group information together"),
1111 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1112 "use branch records for per branch histogram filling",
1114 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1115 "add last branch records to call history"),
1116 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1117 "objdump binary to use for disassembly and annotations"),
1118 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1119 "Disable symbol demangling"),
1120 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1121 "Enable kernel symbol demangling"),
1122 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1123 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1124 "Don't show entries under that percent", parse_percent_limit),
1125 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1126 "how to display percentage of filtered entries", parse_filter_percentage),
1127 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1128 "Instruction Tracing options\n" ITRACE_HELP,
1129 itrace_parse_synth_opts),
1130 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1131 "Show full source file name path for source lines"),
1132 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1133 "Show callgraph from reference event"),
1134 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1135 "only show processor socket that match with this filter"),
1136 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1137 "Show raw trace event output (do not use print fmt or plugins)"),
1138 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1139 "Show entries in a hierarchy"),
1140 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1141 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1142 stdio__config_color, "always"),
1143 OPT_STRING(0, "time", &report.time_str, "str",
1144 "Time span of interest (start,stop)"),
1145 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1146 "Show inline function"),
1147 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1148 "Set percent type local/global-period/hits",
1149 annotate_parse_percent_type),
1152 struct perf_data data = {
1153 .mode = PERF_DATA_MODE_READ,
1155 int ret = hists__init();
1161 ret = perf_config(report__config, &report);
1165 argc = parse_options(argc, argv, options, report_usage, 0);
1168 * Special case: if there's an argument left then assume that
1169 * it's a symbol filter:
1172 usage_with_options(report_usage, options);
1174 report.symbol_filter_str = argv[0];
1177 if (report.mmaps_mode)
1178 report.tasks_mode = true;
1181 perf_quiet_option();
1183 if (symbol_conf.vmlinux_name &&
1184 access(symbol_conf.vmlinux_name, R_OK)) {
1185 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1188 if (symbol_conf.kallsyms_name &&
1189 access(symbol_conf.kallsyms_name, R_OK)) {
1190 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1194 if (report.inverted_callchain)
1195 callchain_param.order = ORDER_CALLER;
1196 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1197 callchain_param.order = ORDER_CALLER;
1199 if (itrace_synth_opts.callchain &&
1200 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1201 report.max_stack = itrace_synth_opts.callchain_sz;
1203 if (!input_name || !strlen(input_name)) {
1204 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1207 input_name = "perf.data";
1210 data.file.path = input_name;
1211 data.force = symbol_conf.force;
1214 session = perf_session__new(&data, false, &report.tool);
1215 if (session == NULL)
1218 if (report.queue_size) {
1219 ordered_events__set_alloc_size(&session->ordered_events,
1223 session->itrace_synth_opts = &itrace_synth_opts;
1225 report.session = session;
1227 has_br_stack = perf_header__has_feat(&session->header,
1228 HEADER_BRANCH_STACK);
1230 setup_forced_leader(&report, session->evlist);
1232 if (itrace_synth_opts.last_branch)
1233 has_br_stack = true;
1235 if (has_br_stack && branch_call_mode)
1236 symbol_conf.show_branchflag_count = true;
1238 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1241 * Branch mode is a tristate:
1242 * -1 means default, so decide based on the file having branch data.
1243 * 0/1 means the user chose a mode.
1245 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1246 !branch_call_mode) {
1247 sort__mode = SORT_MODE__BRANCH;
1248 symbol_conf.cumulate_callchain = false;
1250 if (branch_call_mode) {
1251 callchain_param.key = CCKEY_ADDRESS;
1252 callchain_param.branch_callstack = 1;
1253 symbol_conf.use_callchain = true;
1254 callchain_register_param(&callchain_param);
1255 if (sort_order == NULL)
1256 sort_order = "srcline,symbol,dso";
1259 if (report.mem_mode) {
1260 if (sort__mode == SORT_MODE__BRANCH) {
1261 pr_err("branch and mem mode incompatible\n");
1264 sort__mode = SORT_MODE__MEMORY;
1265 symbol_conf.cumulate_callchain = false;
1268 if (symbol_conf.report_hierarchy) {
1269 /* disable incompatible options */
1270 symbol_conf.cumulate_callchain = false;
1273 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1274 parse_options_usage(report_usage, options, "F", 1);
1275 parse_options_usage(NULL, options, "hierarchy", 0);
1279 perf_hpp_list.need_collapse = true;
1282 if (report.use_stdio)
1284 else if (report.use_tui)
1286 else if (report.use_gtk)
1289 /* Force tty output for header output and per-thread stat. */
1290 if (report.header || report.header_only || report.show_threads)
1292 if (report.header || report.header_only)
1293 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1294 if (report.show_full_info)
1295 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1296 if (report.stats_mode || report.tasks_mode)
1298 if (report.stats_mode && report.tasks_mode) {
1299 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1303 if (strcmp(input_name, "-") != 0)
1304 setup_browser(true);
1308 if (sort_order && strstr(sort_order, "ipc")) {
1309 parse_options_usage(report_usage, options, "s", 1);
1313 if (sort_order && strstr(sort_order, "symbol")) {
1314 if (sort__mode == SORT_MODE__BRANCH) {
1315 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1316 sort_order, "ipc_lbr");
1317 report.symbol_ipc = true;
1319 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1320 sort_order, "ipc_null");
1323 sort_order = sort_tmp;
1326 if (setup_sorting(session->evlist) < 0) {
1328 parse_options_usage(report_usage, options, "s", 1);
1330 parse_options_usage(sort_order ? NULL : report_usage,
1335 if ((report.header || report.header_only) && !quiet) {
1336 perf_session__fprintf_info(session, stdout,
1337 report.show_full_info);
1338 if (report.header_only) {
1342 } else if (use_browser == 0 && !quiet &&
1343 !report.stats_mode && !report.tasks_mode) {
1344 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1349 * Only in the TUI browser we are doing integrated annotation,
1350 * so don't allocate extra space that won't be used in the stdio
1353 if (ui__has_annotation() || report.symbol_ipc) {
1354 ret = symbol__annotation_init();
1358 * For searching by name on the "Browse map details".
1359 * providing it only in verbose mode not to bloat too
1360 * much struct symbol.
1364 * XXX: Need to provide a less kludgy way to ask for
1365 * more space per symbol, the u32 is for the index on
1367 * See symbol__browser_index.
1369 symbol_conf.priv_size += sizeof(u32);
1370 symbol_conf.sort_by_name = true;
1372 annotation_config__init();
1375 if (symbol__init(&session->header.env) < 0)
1378 report.ptime_range = perf_time__range_alloc(report.time_str,
1379 &report.range_size);
1380 if (!report.ptime_range) {
1385 if (perf_time__parse_str(report.ptime_range, report.time_str) != 0) {
1386 if (session->evlist->first_sample_time == 0 &&
1387 session->evlist->last_sample_time == 0) {
1388 pr_err("HINT: no first/last sample time found in perf data.\n"
1389 "Please use latest perf binary to execute 'perf record'\n"
1390 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n");
1395 report.range_num = perf_time__percent_parse_str(
1396 report.ptime_range, report.range_size,
1398 session->evlist->first_sample_time,
1399 session->evlist->last_sample_time);
1401 if (report.range_num < 0) {
1402 pr_err("Invalid time string\n");
1407 report.range_num = 1;
1410 if (session->tevent.pevent &&
1411 tep_set_function_resolver(session->tevent.pevent,
1412 machine__resolve_kernel_addr,
1413 &session->machines.host) < 0) {
1414 pr_err("%s: failed to set libtraceevent function resolver\n",
1419 sort__setup_elide(stdout);
1421 ret = __cmd_report(&report);
1422 if (ret == K_SWITCH_INPUT_DATA) {
1423 perf_session__delete(session);
1429 zfree(&report.ptime_range);
1431 perf_session__delete(session);