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/map_symbol.h"
23 #include "util/mem-events.h"
24 #include "util/branch.h"
25 #include "util/callchain.h"
26 #include "util/values.h"
29 #include "util/debug.h"
30 #include "util/evlist.h"
31 #include "util/evsel.h"
32 #include "util/evswitch.h"
33 #include "util/header.h"
34 #include "util/session.h"
35 #include "util/srcline.h"
36 #include "util/tool.h"
38 #include <subcmd/parse-options.h>
39 #include <subcmd/exec-cmd.h>
40 #include "util/parse-events.h"
42 #include "util/thread.h"
43 #include "util/sort.h"
44 #include "util/hist.h"
45 #include "util/data.h"
46 #include "arch/common.h"
47 #include "util/time-utils.h"
48 #include "util/auxtrace.h"
49 #include "util/units.h"
50 #include "util/branch.h"
51 #include "util/util.h" // perf_tip()
53 #include "ui/progress.h"
54 #include "util/block-info.h"
60 #include <linux/ctype.h>
62 #include <linux/bitmap.h>
63 #include <linux/string.h>
64 #include <linux/stringify.h>
65 #include <linux/time64.h>
66 #include <sys/types.h>
69 #include <linux/mman.h>
72 struct perf_tool tool;
73 struct perf_session *session;
74 struct evswitch evswitch;
75 bool use_tui, use_gtk, use_stdio;
78 bool inverted_callchain;
85 bool nonany_branch_mode;
88 struct perf_read_values show_threads_values;
89 struct annotation_options annotation_opts;
90 const char *pretty_printing_style;
92 const char *symbol_filter_str;
94 struct perf_time_interval *ptime_range;
102 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
103 struct branch_type_stat brtype_stat;
105 bool total_cycles_mode;
106 struct block_report *block_reports;
109 static int report__config(const char *var, const char *value, void *cb)
111 struct report *rep = cb;
113 if (!strcmp(var, "report.group")) {
114 symbol_conf.event_group = perf_config_bool(var, value);
117 if (!strcmp(var, "report.percent-limit")) {
118 double pcnt = strtof(value, NULL);
120 rep->min_percent = pcnt;
121 callchain_param.min_percent = pcnt;
124 if (!strcmp(var, "report.children")) {
125 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
128 if (!strcmp(var, "report.queue-size"))
129 return perf_config_u64(&rep->queue_size, var, value);
131 if (!strcmp(var, "report.sort_order")) {
132 default_sort_order = strdup(value);
139 static int hist_iter__report_callback(struct hist_entry_iter *iter,
140 struct addr_location *al, bool single,
144 struct report *rep = arg;
145 struct hist_entry *he = iter->he;
146 struct evsel *evsel = iter->evsel;
147 struct perf_sample *sample = iter->sample;
149 struct branch_info *bi;
151 if (!ui__has_annotation() && !rep->symbol_ipc)
154 if (sort__mode == SORT_MODE__BRANCH) {
155 bi = he->branch_info;
156 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
160 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
162 } else if (rep->mem_mode) {
164 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
168 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
170 } else if (symbol_conf.cumulate_callchain) {
172 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
174 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
181 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
182 struct addr_location *al __maybe_unused,
183 bool single __maybe_unused,
186 struct hist_entry *he = iter->he;
187 struct report *rep = arg;
188 struct branch_info *bi;
189 struct perf_sample *sample = iter->sample;
190 struct evsel *evsel = iter->evsel;
193 if (!ui__has_annotation() && !rep->symbol_ipc)
196 bi = he->branch_info;
197 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
201 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
203 branch_type_count(&rep->brtype_stat, &bi->flags,
204 bi->from.addr, bi->to.addr);
210 static void setup_forced_leader(struct report *report,
211 struct evlist *evlist)
213 if (report->group_set)
214 perf_evlist__force_leader(evlist);
217 static int process_feature_event(struct perf_session *session,
218 union perf_event *event)
220 struct report *rep = container_of(session->tool, struct report, tool);
222 if (event->feat.feat_id < HEADER_LAST_FEATURE)
223 return perf_event__process_feature(session, event);
225 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
226 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
227 event->feat.feat_id);
232 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
233 * means all features are received, now we can force the
236 setup_forced_leader(rep, session->evlist);
240 static int process_sample_event(struct perf_tool *tool,
241 union perf_event *event,
242 struct perf_sample *sample,
244 struct machine *machine)
246 struct report *rep = container_of(tool, struct report, tool);
247 struct addr_location al;
248 struct hist_entry_iter iter = {
251 .hide_unresolved = symbol_conf.hide_unresolved,
252 .add_entry_cb = hist_iter__report_callback,
256 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
261 if (evswitch__discard(&rep->evswitch, evsel))
264 if (machine__resolve(machine, &al, sample) < 0) {
265 pr_debug("problem processing %d event, skipping it.\n",
270 if (symbol_conf.hide_unresolved && al.sym == NULL)
273 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
276 if (sort__mode == SORT_MODE__BRANCH) {
278 * A non-synthesized event might not have a branch stack if
279 * branch stacks have been synthesized (using itrace options).
281 if (!sample->branch_stack)
284 iter.add_entry_cb = hist_iter__branch_callback;
285 iter.ops = &hist_iter_branch;
286 } else if (rep->mem_mode) {
287 iter.ops = &hist_iter_mem;
288 } else if (symbol_conf.cumulate_callchain) {
289 iter.ops = &hist_iter_cumulative;
291 iter.ops = &hist_iter_normal;
295 al.map->dso->hit = 1;
297 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
298 hist__account_cycles(sample->branch_stack, &al, sample,
299 rep->nonany_branch_mode,
303 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
305 pr_debug("problem adding hist entry, skipping event\n");
307 addr_location__put(&al);
311 static int process_read_event(struct perf_tool *tool,
312 union perf_event *event,
313 struct perf_sample *sample __maybe_unused,
315 struct machine *machine __maybe_unused)
317 struct report *rep = container_of(tool, struct report, tool);
319 if (rep->show_threads) {
320 const char *name = perf_evsel__name(evsel);
321 int err = perf_read_values_add_value(&rep->show_threads_values,
322 event->read.pid, event->read.tid,
334 /* For pipe mode, sample_type is not currently set */
335 static int report__setup_sample_type(struct report *rep)
337 struct perf_session *session = rep->session;
338 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
339 bool is_pipe = perf_data__is_pipe(session->data);
341 if (session->itrace_synth_opts->callchain ||
343 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
344 !session->itrace_synth_opts->set))
345 sample_type |= PERF_SAMPLE_CALLCHAIN;
347 if (session->itrace_synth_opts->last_branch)
348 sample_type |= PERF_SAMPLE_BRANCH_STACK;
350 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
351 if (perf_hpp_list.parent) {
352 ui__error("Selected --sort parent, but no "
353 "callchain data. Did you call "
354 "'perf record' without -g?\n");
357 if (symbol_conf.use_callchain &&
358 !symbol_conf.show_branchflag_count) {
359 ui__error("Selected -g or --branch-history.\n"
360 "But no callchain or branch data.\n"
361 "Did you call 'perf record' without -g or -b?\n");
364 } else if (!callchain_param.enabled &&
365 callchain_param.mode != CHAIN_NONE &&
366 !symbol_conf.use_callchain) {
367 symbol_conf.use_callchain = true;
368 if (callchain_register_param(&callchain_param) < 0) {
369 ui__error("Can't register callchain params.\n");
374 if (symbol_conf.cumulate_callchain) {
375 /* Silently ignore if callchain is missing */
376 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
377 symbol_conf.cumulate_callchain = false;
378 perf_hpp__cancel_cumulate();
382 if (sort__mode == SORT_MODE__BRANCH) {
384 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
385 ui__error("Selected -b but no branch data. "
386 "Did you call perf record without -b?\n");
391 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
392 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
393 (sample_type & PERF_SAMPLE_STACK_USER)) {
394 callchain_param.record_mode = CALLCHAIN_DWARF;
395 dwarf_callchain_users = true;
396 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
397 callchain_param.record_mode = CALLCHAIN_LBR;
399 callchain_param.record_mode = CALLCHAIN_FP;
402 /* ??? handle more cases than just ANY? */
403 if (!(perf_evlist__combined_branch_type(session->evlist) &
404 PERF_SAMPLE_BRANCH_ANY))
405 rep->nonany_branch_mode = true;
407 #ifndef HAVE_LIBUNWIND_SUPPORT
408 if (dwarf_callchain_users) {
409 ui__warning("Please install libunwind development packages "
410 "during the perf build.\n");
417 static void sig_handler(int sig __maybe_unused)
422 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
423 const char *evname, FILE *fp)
427 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
428 u64 nr_events = hists->stats.total_period;
429 struct evsel *evsel = hists_to_evsel(hists);
431 size_t size = sizeof(buf);
432 int socked_id = hists->socket_filter;
437 if (symbol_conf.filter_relative) {
438 nr_samples = hists->stats.nr_non_filtered_samples;
439 nr_events = hists->stats.total_non_filtered_period;
442 if (perf_evsel__is_group_event(evsel)) {
445 perf_evsel__group_desc(evsel, buf, size);
448 for_each_group_member(pos, evsel) {
449 const struct hists *pos_hists = evsel__hists(pos);
451 if (symbol_conf.filter_relative) {
452 nr_samples += pos_hists->stats.nr_non_filtered_samples;
453 nr_events += pos_hists->stats.total_non_filtered_period;
455 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
456 nr_events += pos_hists->stats.total_period;
461 nr_samples = convert_unit(nr_samples, &unit);
462 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
463 if (evname != NULL) {
464 ret += fprintf(fp, " of event%s '%s'",
465 evsel->core.nr_members > 1 ? "s" : "", evname);
469 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
471 if (symbol_conf.show_ref_callgraph &&
472 strstr(evname, "call-graph=no")) {
473 ret += fprintf(fp, ", show reference callgraph");
477 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
478 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
480 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
483 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
485 return ret + fprintf(fp, "\n#\n");
488 static int perf_evlist__tui_block_hists_browse(struct evlist *evlist,
494 evlist__for_each_entry(evlist, pos) {
495 ret = report__browse_block_hists(&rep->block_reports[i++].hist,
496 rep->min_percent, pos);
504 static int perf_evlist__tty_browse_hists(struct evlist *evlist,
512 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
513 evlist->stats.total_lost_samples);
516 evlist__for_each_entry(evlist, pos) {
517 struct hists *hists = evsel__hists(pos);
518 const char *evname = perf_evsel__name(pos);
520 if (symbol_conf.event_group &&
521 !perf_evsel__is_group_leader(pos))
524 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
526 if (rep->total_cycles_mode) {
527 report__browse_block_hists(&rep->block_reports[i++].hist,
528 rep->min_percent, pos);
532 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
533 !(symbol_conf.use_callchain ||
534 symbol_conf.show_branchflag_count));
535 fprintf(stdout, "\n\n");
539 fprintf(stdout, "#\n# (%s)\n#\n", help);
541 if (rep->show_threads) {
542 bool style = !strcmp(rep->pretty_printing_style, "raw");
543 perf_read_values_display(stdout, &rep->show_threads_values,
545 perf_read_values_destroy(&rep->show_threads_values);
548 if (sort__mode == SORT_MODE__BRANCH)
549 branch_type_stat_display(stdout, &rep->brtype_stat);
554 static void report__warn_kptr_restrict(const struct report *rep)
556 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
557 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
559 if (perf_evlist__exclude_kernel(rep->session->evlist))
562 if (kernel_map == NULL ||
563 (kernel_map->dso->hit &&
564 (kernel_kmap->ref_reloc_sym == NULL ||
565 kernel_kmap->ref_reloc_sym->addr == 0))) {
567 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
568 "can't be resolved.";
570 if (kernel_map && map__has_symbols(kernel_map)) {
571 desc = "If some relocation was applied (e.g. "
572 "kexec) symbols may be misresolved.";
576 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
577 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
578 "Samples in kernel modules can't be resolved as well.\n\n",
583 static int report__gtk_browse_hists(struct report *rep, const char *help)
585 int (*hist_browser)(struct evlist *evlist, const char *help,
586 struct hist_browser_timer *timer, float min_pcnt);
588 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
590 if (hist_browser == NULL) {
591 ui__error("GTK browser not found!\n");
595 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
598 static int report__browse_hists(struct report *rep)
601 struct perf_session *session = rep->session;
602 struct evlist *evlist = session->evlist;
603 const char *help = perf_tip(system_path(TIPDIR));
606 /* fallback for people who don't install perf ;-) */
607 help = perf_tip(DOCDIR);
609 help = "Cannot load tips.txt file, please install perf!";
612 switch (use_browser) {
614 if (rep->total_cycles_mode) {
615 ret = perf_evlist__tui_block_hists_browse(evlist, rep);
619 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
621 &session->header.env,
622 true, &rep->annotation_opts);
624 * Usually "ret" is the last pressed key, and we only
625 * care if the key notifies us to switch data file.
627 if (ret != K_SWITCH_INPUT_DATA)
631 ret = report__gtk_browse_hists(rep, help);
634 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
641 static int report__collapse_hists(struct report *rep)
643 struct ui_progress prog;
647 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
649 evlist__for_each_entry(rep->session->evlist, pos) {
650 struct hists *hists = evsel__hists(pos);
653 hists->symbol_filter_str = rep->symbol_filter_str;
655 hists->socket_filter = rep->socket_filter;
657 ret = hists__collapse_resort(hists, &prog);
661 /* Non-group events are considered as leader */
662 if (symbol_conf.event_group &&
663 !perf_evsel__is_group_leader(pos)) {
664 struct hists *leader_hists = evsel__hists(pos->leader);
666 hists__match(leader_hists, hists);
667 hists__link(leader_hists, hists);
671 ui_progress__finish();
675 static int hists__resort_cb(struct hist_entry *he, void *arg)
677 struct report *rep = arg;
678 struct symbol *sym = he->ms.sym;
680 if (rep->symbol_ipc && sym && !sym->annotate2) {
681 struct evsel *evsel = hists_to_evsel(he->hists);
683 symbol__annotate2(&he->ms, evsel,
684 &annotation__default_options, NULL);
690 static void report__output_resort(struct report *rep)
692 struct ui_progress prog;
695 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
697 evlist__for_each_entry(rep->session->evlist, pos) {
698 perf_evsel__output_resort_cb(pos, &prog,
699 hists__resort_cb, rep);
702 ui_progress__finish();
705 static void stats_setup(struct report *rep)
707 memset(&rep->tool, 0, sizeof(rep->tool));
708 rep->tool.no_warn = true;
711 static int stats_print(struct report *rep)
713 struct perf_session *session = rep->session;
715 perf_session__fprintf_nr_events(session, stdout);
719 static void tasks_setup(struct report *rep)
721 memset(&rep->tool, 0, sizeof(rep->tool));
722 rep->tool.ordered_events = true;
723 if (rep->mmaps_mode) {
724 rep->tool.mmap = perf_event__process_mmap;
725 rep->tool.mmap2 = perf_event__process_mmap2;
727 rep->tool.comm = perf_event__process_comm;
728 rep->tool.exit = perf_event__process_exit;
729 rep->tool.fork = perf_event__process_fork;
730 rep->tool.no_warn = true;
734 struct thread *thread;
735 struct list_head list;
736 struct list_head children;
739 static struct task *tasks_list(struct task *task, struct machine *machine)
741 struct thread *parent_thread, *thread = task->thread;
742 struct task *parent_task;
744 /* Already listed. */
745 if (!list_empty(&task->list))
748 /* Last one in the chain. */
749 if (thread->ppid == -1)
752 parent_thread = machine__find_thread(machine, -1, thread->ppid);
754 return ERR_PTR(-ENOENT);
756 parent_task = thread__priv(parent_thread);
757 list_add_tail(&task->list, &parent_task->children);
758 return tasks_list(parent_task, machine);
761 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
766 maps__for_each_entry(maps, map) {
767 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
768 indent, "", map->start, map->end,
769 map->prot & PROT_READ ? 'r' : '-',
770 map->prot & PROT_WRITE ? 'w' : '-',
771 map->prot & PROT_EXEC ? 'x' : '-',
772 map->flags & MAP_SHARED ? 's' : 'p',
774 map->dso->id.ino, map->dso->name);
780 static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp)
782 return maps__fprintf_task(&mg->maps, indent, fp);
785 static void task__print_level(struct task *task, FILE *fp, int level)
787 struct thread *thread = task->thread;
789 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
790 thread->pid_, thread->tid, thread->ppid,
793 fprintf(fp, "%s\n", thread__comm_str(thread));
795 map_groups__fprintf_task(thread->mg, comm_indent, fp);
797 if (!list_empty(&task->children)) {
798 list_for_each_entry(child, &task->children, list)
799 task__print_level(child, fp, level + 1);
803 static int tasks_print(struct report *rep, FILE *fp)
805 struct perf_session *session = rep->session;
806 struct machine *machine = &session->machines.host;
807 struct task *tasks, *task;
808 unsigned int nr = 0, itask = 0, i;
813 * No locking needed while accessing machine->threads,
814 * because --tasks is single threaded command.
817 /* Count all the threads. */
818 for (i = 0; i < THREADS__TABLE_SIZE; i++)
819 nr += machine->threads[i].nr;
821 tasks = malloc(sizeof(*tasks) * nr);
825 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
826 struct threads *threads = &machine->threads[i];
828 for (nd = rb_first_cached(&threads->entries); nd;
830 task = tasks + itask++;
832 task->thread = rb_entry(nd, struct thread, rb_node);
833 INIT_LIST_HEAD(&task->children);
834 INIT_LIST_HEAD(&task->list);
835 thread__set_priv(task->thread, task);
840 * Iterate every task down to the unprocessed parent
841 * and link all in task children list. Task with no
842 * parent is added into 'list'.
844 for (itask = 0; itask < nr; itask++) {
845 task = tasks + itask;
847 if (!list_empty(&task->list))
850 task = tasks_list(task, machine);
852 pr_err("Error: failed to process tasks\n");
854 return PTR_ERR(task);
858 list_add_tail(&task->list, &list);
861 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
863 list_for_each_entry(task, &list, list)
864 task__print_level(task, fp, 0);
870 static int __cmd_report(struct report *rep)
873 struct perf_session *session = rep->session;
875 struct perf_data *data = session->data;
877 signal(SIGINT, sig_handler);
880 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
883 ui__error("failed to set cpu bitmap\n");
886 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
889 if (rep->show_threads) {
890 ret = perf_read_values_init(&rep->show_threads_values);
895 ret = report__setup_sample_type(rep);
897 /* report__setup_sample_type() already showed error message */
907 ret = perf_session__process_events(session);
909 ui__error("failed to process sample\n");
914 return stats_print(rep);
917 return tasks_print(rep, stdout);
919 report__warn_kptr_restrict(rep);
921 evlist__for_each_entry(session->evlist, pos)
922 rep->nr_entries += evsel__hists(pos)->nr_entries;
924 if (use_browser == 0) {
926 perf_session__fprintf(session, stdout);
929 perf_session__fprintf_dsos(session, stdout);
932 perf_session__fprintf_nr_events(session, stdout);
933 perf_evlist__fprintf_nr_events(session->evlist, stdout);
938 ret = report__collapse_hists(rep);
940 ui__error("failed to process hist entry\n");
948 * recalculate number of entries after collapsing since it
949 * might be changed during the collapse phase.
952 evlist__for_each_entry(session->evlist, pos)
953 rep->nr_entries += evsel__hists(pos)->nr_entries;
955 if (rep->nr_entries == 0) {
956 ui__error("The %s data has no samples!\n", data->path);
960 report__output_resort(rep);
962 if (rep->total_cycles_mode) {
963 rep->block_reports = block_info__create_report(session->evlist,
965 if (!rep->block_reports)
969 return report__browse_hists(rep);
973 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
975 struct callchain_param *callchain = opt->value;
977 callchain->enabled = !unset;
982 symbol_conf.use_callchain = false;
983 callchain->mode = CHAIN_NONE;
987 return parse_callchain_report_opt(arg);
991 parse_time_quantum(const struct option *opt, const char *arg,
992 int unset __maybe_unused)
994 unsigned long *time_q = opt->value;
997 *time_q = strtoul(arg, &end, 0);
1001 pr_err("time quantum cannot be 0");
1004 end = skip_spaces(end);
1007 if (!strcmp(end, "s")) {
1008 *time_q *= NSEC_PER_SEC;
1011 if (!strcmp(end, "ms")) {
1012 *time_q *= NSEC_PER_MSEC;
1015 if (!strcmp(end, "us")) {
1016 *time_q *= NSEC_PER_USEC;
1019 if (!strcmp(end, "ns"))
1022 pr_err("Cannot parse time quantum `%s'\n", arg);
1027 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1028 const char *arg, int unset __maybe_unused)
1031 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1034 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1035 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1038 have_ignore_callees = 1;
1045 parse_branch_mode(const struct option *opt,
1046 const char *str __maybe_unused, int unset)
1048 int *branch_mode = opt->value;
1050 *branch_mode = !unset;
1055 parse_percent_limit(const struct option *opt, const char *str,
1056 int unset __maybe_unused)
1058 struct report *rep = opt->value;
1059 double pcnt = strtof(str, NULL);
1061 rep->min_percent = pcnt;
1062 callchain_param.min_percent = pcnt;
1066 int cmd_report(int argc, const char **argv)
1068 struct perf_session *session;
1069 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1071 bool has_br_stack = false;
1072 int branch_mode = -1;
1073 bool branch_call_mode = false;
1074 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1075 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1076 CALLCHAIN_REPORT_HELP
1077 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1078 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1079 const char * const report_usage[] = {
1080 "perf report [<options>]",
1083 struct report report = {
1085 .sample = process_sample_event,
1086 .mmap = perf_event__process_mmap,
1087 .mmap2 = perf_event__process_mmap2,
1088 .comm = perf_event__process_comm,
1089 .namespaces = perf_event__process_namespaces,
1090 .exit = perf_event__process_exit,
1091 .fork = perf_event__process_fork,
1092 .lost = perf_event__process_lost,
1093 .read = process_read_event,
1094 .attr = perf_event__process_attr,
1095 .tracing_data = perf_event__process_tracing_data,
1096 .build_id = perf_event__process_build_id,
1097 .id_index = perf_event__process_id_index,
1098 .auxtrace_info = perf_event__process_auxtrace_info,
1099 .auxtrace = perf_event__process_auxtrace,
1100 .event_update = perf_event__process_event_update,
1101 .feature = process_feature_event,
1102 .ordered_events = true,
1103 .ordering_requires_timestamps = true,
1105 .max_stack = PERF_MAX_STACK_DEPTH,
1106 .pretty_printing_style = "normal",
1107 .socket_filter = -1,
1108 .annotation_opts = annotation__default_options,
1110 const struct option options[] = {
1111 OPT_STRING('i', "input", &input_name, "file",
1113 OPT_INCR('v', "verbose", &verbose,
1114 "be more verbose (show symbol address, etc)"),
1115 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1116 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1117 "dump raw trace in ASCII"),
1118 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1119 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1120 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1121 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1122 "file", "vmlinux pathname"),
1123 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1124 "don't load vmlinux even if found"),
1125 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1126 "file", "kallsyms pathname"),
1127 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1128 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1129 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1130 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1131 "Show a column with the number of samples"),
1132 OPT_BOOLEAN('T', "threads", &report.show_threads,
1133 "Show per-thread event counters"),
1134 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1135 "pretty printing style key: normal raw"),
1136 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1137 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1138 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1139 "Use the stdio interface"),
1140 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1141 OPT_BOOLEAN(0, "header-only", &report.header_only,
1142 "Show only data header."),
1143 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1144 sort_help("sort by key(s):")),
1145 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1146 sort_help("output field(s): overhead period sample ")),
1147 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1148 "Show sample percentage for different cpu modes"),
1149 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1150 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1151 OPT_STRING('p', "parent", &parent_pattern, "regex",
1152 "regex filter to identify parent, see: '--sort parent'"),
1153 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1154 "Only display entries with parent-match"),
1155 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1156 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1157 report_callchain_help, &report_parse_callchain_opt,
1158 callchain_default_opt),
1159 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1160 "Accumulate callchains of children and show total overhead as well"),
1161 OPT_INTEGER(0, "max-stack", &report.max_stack,
1162 "Set the maximum stack depth when parsing the callchain, "
1163 "anything beyond the specified depth will be ignored. "
1164 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1165 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1166 "alias for inverted call graph"),
1167 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1168 "ignore callees of these functions in call graphs",
1169 report_parse_ignore_callees_opt),
1170 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1171 "only consider symbols in these dsos"),
1172 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1173 "only consider symbols in these comms"),
1174 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1175 "only consider symbols in these pids"),
1176 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1177 "only consider symbols in these tids"),
1178 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1179 "only consider these symbols"),
1180 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1181 "only show symbols that (partially) match with this filter"),
1182 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1184 "don't try to adjust column width, use these fixed values"),
1185 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1186 "separator for columns, no spaces will be added between "
1187 "columns '.' is reserved."),
1188 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1189 "Only display entries resolved to a symbol"),
1190 OPT_CALLBACK(0, "symfs", NULL, "directory",
1191 "Look for files with symbols relative to this directory",
1192 symbol__config_symfs),
1193 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1194 "list of cpus to profile"),
1195 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1196 "Display extended information about perf.data file"),
1197 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1198 "Interleave source code with assembly code (default)"),
1199 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1200 "Display raw encoding of assembly instructions (default)"),
1201 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1202 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1203 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1204 "Show a column with the sum of periods"),
1205 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1206 "Show event group information together"),
1207 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1208 "use branch records for per branch histogram filling",
1210 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1211 "add last branch records to call history"),
1212 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1213 "objdump binary to use for disassembly and annotations"),
1214 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1215 "Disable symbol demangling"),
1216 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1217 "Enable kernel symbol demangling"),
1218 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1219 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1220 "Number of samples to save per histogram entry for individual browsing"),
1221 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1222 "Don't show entries under that percent", parse_percent_limit),
1223 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1224 "how to display percentage of filtered entries", parse_filter_percentage),
1225 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1226 "Instruction Tracing options\n" ITRACE_HELP,
1227 itrace_parse_synth_opts),
1228 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1229 "Show full source file name path for source lines"),
1230 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1231 "Show callgraph from reference event"),
1232 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1233 "only show processor socket that match with this filter"),
1234 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1235 "Show raw trace event output (do not use print fmt or plugins)"),
1236 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1237 "Show entries in a hierarchy"),
1238 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1239 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1240 stdio__config_color, "always"),
1241 OPT_STRING(0, "time", &report.time_str, "str",
1242 "Time span of interest (start,stop)"),
1243 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1244 "Show inline function"),
1245 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1246 "Set percent type local/global-period/hits",
1247 annotate_parse_percent_type),
1248 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1249 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1250 "Set time quantum for time sort key (default 100ms)",
1251 parse_time_quantum),
1252 OPTS_EVSWITCH(&report.evswitch),
1253 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1254 "Sort all blocks by 'Sampled Cycles%'"),
1257 struct perf_data data = {
1258 .mode = PERF_DATA_MODE_READ,
1260 int ret = hists__init();
1266 ret = perf_config(report__config, &report);
1270 argc = parse_options(argc, argv, options, report_usage, 0);
1273 * Special case: if there's an argument left then assume that
1274 * it's a symbol filter:
1277 usage_with_options(report_usage, options);
1279 report.symbol_filter_str = argv[0];
1282 if (report.mmaps_mode)
1283 report.tasks_mode = true;
1286 perf_quiet_option();
1288 if (symbol_conf.vmlinux_name &&
1289 access(symbol_conf.vmlinux_name, R_OK)) {
1290 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1293 if (symbol_conf.kallsyms_name &&
1294 access(symbol_conf.kallsyms_name, R_OK)) {
1295 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1299 if (report.inverted_callchain)
1300 callchain_param.order = ORDER_CALLER;
1301 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1302 callchain_param.order = ORDER_CALLER;
1304 if (itrace_synth_opts.callchain &&
1305 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1306 report.max_stack = itrace_synth_opts.callchain_sz;
1308 if (!input_name || !strlen(input_name)) {
1309 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1312 input_name = "perf.data";
1315 data.path = input_name;
1316 data.force = symbol_conf.force;
1319 session = perf_session__new(&data, false, &report.tool);
1320 if (IS_ERR(session))
1321 return PTR_ERR(session);
1323 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1327 if (zstd_init(&(session->zstd_data), 0) < 0)
1328 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1330 if (report.queue_size) {
1331 ordered_events__set_alloc_size(&session->ordered_events,
1335 session->itrace_synth_opts = &itrace_synth_opts;
1337 report.session = session;
1339 has_br_stack = perf_header__has_feat(&session->header,
1340 HEADER_BRANCH_STACK);
1341 if (perf_evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1342 has_br_stack = false;
1344 setup_forced_leader(&report, session->evlist);
1346 if (itrace_synth_opts.last_branch)
1347 has_br_stack = true;
1349 if (has_br_stack && branch_call_mode)
1350 symbol_conf.show_branchflag_count = true;
1352 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1355 * Branch mode is a tristate:
1356 * -1 means default, so decide based on the file having branch data.
1357 * 0/1 means the user chose a mode.
1359 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1360 !branch_call_mode) {
1361 sort__mode = SORT_MODE__BRANCH;
1362 symbol_conf.cumulate_callchain = false;
1364 if (branch_call_mode) {
1365 callchain_param.key = CCKEY_ADDRESS;
1366 callchain_param.branch_callstack = 1;
1367 symbol_conf.use_callchain = true;
1368 callchain_register_param(&callchain_param);
1369 if (sort_order == NULL)
1370 sort_order = "srcline,symbol,dso";
1373 if (report.mem_mode) {
1374 if (sort__mode == SORT_MODE__BRANCH) {
1375 pr_err("branch and mem mode incompatible\n");
1378 sort__mode = SORT_MODE__MEMORY;
1379 symbol_conf.cumulate_callchain = false;
1382 if (symbol_conf.report_hierarchy) {
1383 /* disable incompatible options */
1384 symbol_conf.cumulate_callchain = false;
1387 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1388 parse_options_usage(report_usage, options, "F", 1);
1389 parse_options_usage(NULL, options, "hierarchy", 0);
1393 perf_hpp_list.need_collapse = true;
1396 if (report.use_stdio)
1398 else if (report.use_tui)
1400 else if (report.use_gtk)
1403 /* Force tty output for header output and per-thread stat. */
1404 if (report.header || report.header_only || report.show_threads)
1406 if (report.header || report.header_only)
1407 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1408 if (report.show_full_info)
1409 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1410 if (report.stats_mode || report.tasks_mode)
1412 if (report.stats_mode && report.tasks_mode) {
1413 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1417 if (report.total_cycles_mode) {
1418 if (sort__mode != SORT_MODE__BRANCH)
1419 report.total_cycles_mode = false;
1424 if (strcmp(input_name, "-") != 0)
1425 setup_browser(true);
1429 if (sort_order && strstr(sort_order, "ipc")) {
1430 parse_options_usage(report_usage, options, "s", 1);
1434 if (sort_order && strstr(sort_order, "symbol")) {
1435 if (sort__mode == SORT_MODE__BRANCH) {
1436 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1437 sort_order, "ipc_lbr");
1438 report.symbol_ipc = true;
1440 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1441 sort_order, "ipc_null");
1444 sort_order = sort_tmp;
1447 if (setup_sorting(session->evlist) < 0) {
1449 parse_options_usage(report_usage, options, "s", 1);
1451 parse_options_usage(sort_order ? NULL : report_usage,
1456 if ((report.header || report.header_only) && !quiet) {
1457 perf_session__fprintf_info(session, stdout,
1458 report.show_full_info);
1459 if (report.header_only) {
1463 } else if (use_browser == 0 && !quiet &&
1464 !report.stats_mode && !report.tasks_mode) {
1465 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1470 * Only in the TUI browser we are doing integrated annotation,
1471 * so don't allocate extra space that won't be used in the stdio
1474 if (ui__has_annotation() || report.symbol_ipc ||
1475 report.total_cycles_mode) {
1476 ret = symbol__annotation_init();
1480 * For searching by name on the "Browse map details".
1481 * providing it only in verbose mode not to bloat too
1482 * much struct symbol.
1486 * XXX: Need to provide a less kludgy way to ask for
1487 * more space per symbol, the u32 is for the index on
1489 * See symbol__browser_index.
1491 symbol_conf.priv_size += sizeof(u32);
1492 symbol_conf.sort_by_name = true;
1494 annotation_config__init();
1497 if (symbol__init(&session->header.env) < 0)
1500 if (report.time_str) {
1501 ret = perf_time__parse_for_ranges(report.time_str, session,
1502 &report.ptime_range,
1508 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1513 if (session->tevent.pevent &&
1514 tep_set_function_resolver(session->tevent.pevent,
1515 machine__resolve_kernel_addr,
1516 &session->machines.host) < 0) {
1517 pr_err("%s: failed to set libtraceevent function resolver\n",
1522 sort__setup_elide(stdout);
1524 ret = __cmd_report(&report);
1525 if (ret == K_SWITCH_INPUT_DATA) {
1526 perf_session__delete(session);
1532 if (report.ptime_range) {
1533 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1534 zfree(&report.ptime_range);
1537 if (report.block_reports)
1538 zfree(&report.block_reports);
1540 zstd_fini(&(session->zstd_data));
1541 perf_session__delete(session);