4 * Builtin report command: Analyze the perf.data input file,
5 * look up and read DSOs and symbol information and display
6 * a histogram of results, along various sorting keys.
10 #include "util/util.h"
11 #include "util/cache.h"
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include <linux/list.h>
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28 #include "util/tool.h"
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
36 #include "util/data.h"
37 #include "arch/common.h"
40 #include <linux/bitmap.h>
43 struct perf_tool tool;
44 struct perf_session *session;
45 bool force, use_tui, use_gtk, use_stdio;
47 bool dont_use_callchains;
50 bool inverted_callchain;
55 struct perf_read_values show_threads_values;
56 const char *pretty_printing_style;
58 const char *symbol_filter_str;
62 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
65 static int report__config(const char *var, const char *value, void *cb)
67 struct report *rep = cb;
69 if (!strcmp(var, "report.group")) {
70 symbol_conf.event_group = perf_config_bool(var, value);
73 if (!strcmp(var, "report.percent-limit")) {
74 rep->min_percent = strtof(value, NULL);
77 if (!strcmp(var, "report.children")) {
78 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
81 if (!strcmp(var, "report.queue-size")) {
82 rep->queue_size = perf_config_u64(var, value);
86 return perf_default_config(var, value, cb);
89 static void report__inc_stats(struct report *rep, struct hist_entry *he)
92 * The @he is either of a newly created one or an existing one
93 * merging current sample. We only want to count a new one so
94 * checking ->nr_events being 1.
96 if (he->stat.nr_events == 1)
100 static int hist_iter__report_callback(struct hist_entry_iter *iter,
101 struct addr_location *al, bool single,
105 struct report *rep = arg;
106 struct hist_entry *he = iter->he;
107 struct perf_evsel *evsel = iter->evsel;
109 struct branch_info *bi;
111 report__inc_stats(rep, he);
113 if (!ui__has_annotation())
116 if (sort__mode == SORT_MODE__BRANCH) {
117 bi = he->branch_info;
118 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
122 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
124 } else if (rep->mem_mode) {
126 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
130 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
132 } else if (symbol_conf.cumulate_callchain) {
134 err = hist_entry__inc_addr_samples(he, evsel->idx,
137 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
144 static int process_sample_event(struct perf_tool *tool,
145 union perf_event *event,
146 struct perf_sample *sample,
147 struct perf_evsel *evsel,
148 struct machine *machine)
150 struct report *rep = container_of(tool, struct report, tool);
151 struct addr_location al;
152 struct hist_entry_iter iter = {
153 .hide_unresolved = rep->hide_unresolved,
154 .add_entry_cb = hist_iter__report_callback,
158 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
159 pr_debug("problem processing %d event, skipping it.\n",
164 if (rep->hide_unresolved && al.sym == NULL)
167 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
170 if (sort__mode == SORT_MODE__BRANCH)
171 iter.ops = &hist_iter_branch;
172 else if (rep->mem_mode)
173 iter.ops = &hist_iter_mem;
174 else if (symbol_conf.cumulate_callchain)
175 iter.ops = &hist_iter_cumulative;
177 iter.ops = &hist_iter_normal;
180 al.map->dso->hit = 1;
182 ret = hist_entry_iter__add(&iter, &al, evsel, sample, rep->max_stack,
185 pr_debug("problem adding hist entry, skipping event\n");
190 static int process_read_event(struct perf_tool *tool,
191 union perf_event *event,
192 struct perf_sample *sample __maybe_unused,
193 struct perf_evsel *evsel,
194 struct machine *machine __maybe_unused)
196 struct report *rep = container_of(tool, struct report, tool);
198 if (rep->show_threads) {
199 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
200 perf_read_values_add_value(&rep->show_threads_values,
201 event->read.pid, event->read.tid,
207 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
208 evsel ? perf_evsel__name(evsel) : "FAIL",
214 /* For pipe mode, sample_type is not currently set */
215 static int report__setup_sample_type(struct report *rep)
217 struct perf_session *session = rep->session;
218 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
219 bool is_pipe = perf_data_file__is_pipe(session->file);
221 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
222 if (sort__has_parent) {
223 ui__error("Selected --sort parent, but no "
224 "callchain data. Did you call "
225 "'perf record' without -g?\n");
228 if (symbol_conf.use_callchain) {
229 ui__error("Selected -g but no callchain data. Did "
230 "you call 'perf record' without -g?\n");
233 } else if (!rep->dont_use_callchains &&
234 callchain_param.mode != CHAIN_NONE &&
235 !symbol_conf.use_callchain) {
236 symbol_conf.use_callchain = true;
237 if (callchain_register_param(&callchain_param) < 0) {
238 ui__error("Can't register callchain params.\n");
243 if (symbol_conf.cumulate_callchain) {
244 /* Silently ignore if callchain is missing */
245 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
246 symbol_conf.cumulate_callchain = false;
247 perf_hpp__cancel_cumulate();
251 if (sort__mode == SORT_MODE__BRANCH) {
253 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
254 ui__error("Selected -b but no branch data. "
255 "Did you call perf record without -b?\n");
263 static void sig_handler(int sig __maybe_unused)
268 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
269 const char *evname, FILE *fp)
273 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
274 u64 nr_events = hists->stats.total_period;
275 struct perf_evsel *evsel = hists_to_evsel(hists);
277 size_t size = sizeof(buf);
279 if (symbol_conf.filter_relative) {
280 nr_samples = hists->stats.nr_non_filtered_samples;
281 nr_events = hists->stats.total_non_filtered_period;
284 if (perf_evsel__is_group_event(evsel)) {
285 struct perf_evsel *pos;
287 perf_evsel__group_desc(evsel, buf, size);
290 for_each_group_member(pos, evsel) {
291 const struct hists *pos_hists = evsel__hists(pos);
293 if (symbol_conf.filter_relative) {
294 nr_samples += pos_hists->stats.nr_non_filtered_samples;
295 nr_events += pos_hists->stats.total_non_filtered_period;
297 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
298 nr_events += pos_hists->stats.total_period;
303 nr_samples = convert_unit(nr_samples, &unit);
304 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
306 ret += fprintf(fp, " of event '%s'", evname);
309 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
310 ret += fprintf(fp, "\n# Sort order : %s", sort_order);
312 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
313 return ret + fprintf(fp, "\n#\n");
316 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
320 struct perf_evsel *pos;
322 evlist__for_each(evlist, pos) {
323 struct hists *hists = evsel__hists(pos);
324 const char *evname = perf_evsel__name(pos);
326 if (symbol_conf.event_group &&
327 !perf_evsel__is_group_leader(pos))
330 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
331 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
332 fprintf(stdout, "\n\n");
335 if (sort_order == default_sort_order &&
336 parent_pattern == default_parent_pattern) {
337 fprintf(stdout, "#\n# (%s)\n#\n", help);
339 if (rep->show_threads) {
340 bool style = !strcmp(rep->pretty_printing_style, "raw");
341 perf_read_values_display(stdout, &rep->show_threads_values,
343 perf_read_values_destroy(&rep->show_threads_values);
350 static void report__warn_kptr_restrict(const struct report *rep)
352 struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION];
353 struct kmap *kernel_kmap = map__kmap(kernel_map);
355 if (kernel_map == NULL ||
356 (kernel_map->dso->hit &&
357 (kernel_kmap->ref_reloc_sym == NULL ||
358 kernel_kmap->ref_reloc_sym->addr == 0))) {
360 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
361 "can't be resolved.";
364 const struct dso *kdso = kernel_map->dso;
365 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
366 desc = "If some relocation was applied (e.g. "
367 "kexec) symbols may be misresolved.";
372 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
373 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
374 "Samples in kernel modules can't be resolved as well.\n\n",
379 static int report__gtk_browse_hists(struct report *rep, const char *help)
381 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
382 struct hist_browser_timer *timer, float min_pcnt);
384 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
386 if (hist_browser == NULL) {
387 ui__error("GTK browser not found!\n");
391 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
394 static int report__browse_hists(struct report *rep)
397 struct perf_session *session = rep->session;
398 struct perf_evlist *evlist = session->evlist;
399 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
401 switch (use_browser) {
403 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
405 &session->header.env);
407 * Usually "ret" is the last pressed key, and we only
408 * care if the key notifies us to switch data file.
410 if (ret != K_SWITCH_INPUT_DATA)
414 ret = report__gtk_browse_hists(rep, help);
417 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
424 static void report__collapse_hists(struct report *rep)
426 struct ui_progress prog;
427 struct perf_evsel *pos;
429 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
431 evlist__for_each(rep->session->evlist, pos) {
432 struct hists *hists = evsel__hists(pos);
435 hists->symbol_filter_str = rep->symbol_filter_str;
437 hists__collapse_resort(hists, &prog);
439 /* Non-group events are considered as leader */
440 if (symbol_conf.event_group &&
441 !perf_evsel__is_group_leader(pos)) {
442 struct hists *leader_hists = evsel__hists(pos->leader);
444 hists__match(leader_hists, hists);
445 hists__link(leader_hists, hists);
449 ui_progress__finish();
452 static int __cmd_report(struct report *rep)
455 struct perf_session *session = rep->session;
456 struct perf_evsel *pos;
457 struct perf_data_file *file = session->file;
459 signal(SIGINT, sig_handler);
462 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
468 if (rep->show_threads)
469 perf_read_values_init(&rep->show_threads_values);
471 ret = report__setup_sample_type(rep);
475 ret = perf_session__process_events(session, &rep->tool);
479 report__warn_kptr_restrict(rep);
481 if (use_browser == 0) {
483 perf_session__fprintf(session, stdout);
486 perf_session__fprintf_dsos(session, stdout);
489 perf_session__fprintf_nr_events(session, stdout);
494 report__collapse_hists(rep);
499 if (rep->nr_entries == 0) {
500 ui__error("The %s file has no samples!\n", file->path);
504 evlist__for_each(session->evlist, pos)
505 hists__output_resort(evsel__hists(pos));
507 return report__browse_hists(rep);
511 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
513 struct report *rep = (struct report *)opt->value;
519 rep->dont_use_callchains = true;
523 return parse_callchain_report_opt(arg);
527 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
528 const char *arg, int unset __maybe_unused)
531 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
534 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
535 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
538 have_ignore_callees = 1;
545 parse_branch_mode(const struct option *opt __maybe_unused,
546 const char *str __maybe_unused, int unset)
548 int *branch_mode = opt->value;
550 *branch_mode = !unset;
555 parse_percent_limit(const struct option *opt, const char *str,
556 int unset __maybe_unused)
558 struct report *rep = opt->value;
560 rep->min_percent = strtof(str, NULL);
564 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
566 struct perf_session *session;
568 bool has_br_stack = false;
569 int branch_mode = -1;
571 char callchain_default_opt[] = "fractal,0.5,callee";
572 const char * const report_usage[] = {
573 "perf report [<options>]",
576 struct report report = {
578 .sample = process_sample_event,
579 .mmap = perf_event__process_mmap,
580 .mmap2 = perf_event__process_mmap2,
581 .comm = perf_event__process_comm,
582 .exit = perf_event__process_exit,
583 .fork = perf_event__process_fork,
584 .lost = perf_event__process_lost,
585 .read = process_read_event,
586 .attr = perf_event__process_attr,
587 .tracing_data = perf_event__process_tracing_data,
588 .build_id = perf_event__process_build_id,
589 .ordered_events = true,
590 .ordering_requires_timestamps = true,
592 .max_stack = PERF_MAX_STACK_DEPTH,
593 .pretty_printing_style = "normal",
595 const struct option options[] = {
596 OPT_STRING('i', "input", &input_name, "file",
598 OPT_INCR('v', "verbose", &verbose,
599 "be more verbose (show symbol address, etc)"),
600 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
601 "dump raw trace in ASCII"),
602 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
603 "file", "vmlinux pathname"),
604 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
605 "file", "kallsyms pathname"),
606 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
607 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
608 "load module symbols - WARNING: use only with -k and LIVE kernel"),
609 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
610 "Show a column with the number of samples"),
611 OPT_BOOLEAN('T', "threads", &report.show_threads,
612 "Show per-thread event counters"),
613 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
614 "pretty printing style key: normal raw"),
615 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
616 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
617 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
618 "Use the stdio interface"),
619 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
620 OPT_BOOLEAN(0, "header-only", &report.header_only,
621 "Show only data header."),
622 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
623 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
624 " Please refer the man page for the complete list."),
625 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
626 "output field(s): overhead, period, sample plus all of sort keys"),
627 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
628 "Show sample percentage for different cpu modes"),
629 OPT_STRING('p', "parent", &parent_pattern, "regex",
630 "regex filter to identify parent, see: '--sort parent'"),
631 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
632 "Only display entries with parent-match"),
633 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
634 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
635 "Default: fractal,0.5,callee,function", &report_parse_callchain_opt, callchain_default_opt),
636 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
637 "Accumulate callchains of children and show total overhead as well"),
638 OPT_INTEGER(0, "max-stack", &report.max_stack,
639 "Set the maximum stack depth when parsing the callchain, "
640 "anything beyond the specified depth will be ignored. "
641 "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
642 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
643 "alias for inverted call graph"),
644 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
645 "ignore callees of these functions in call graphs",
646 report_parse_ignore_callees_opt),
647 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
648 "only consider symbols in these dsos"),
649 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
650 "only consider symbols in these comms"),
651 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
652 "only consider these symbols"),
653 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
654 "only show symbols that (partially) match with this filter"),
655 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
657 "don't try to adjust column width, use these fixed values"),
658 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
659 "separator for columns, no spaces will be added between "
660 "columns '.' is reserved."),
661 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
662 "Only display entries resolved to a symbol"),
663 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
664 "Look for files with symbols relative to this directory"),
665 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
666 "list of cpus to profile"),
667 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
668 "Display extended information about perf.data file"),
669 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
670 "Interleave source code with assembly code (default)"),
671 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
672 "Display raw encoding of assembly instructions (default)"),
673 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
674 "Specify disassembler style (e.g. -M intel for intel syntax)"),
675 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
676 "Show a column with the sum of periods"),
677 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
678 "Show event group information together"),
679 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
680 "use branch records for histogram filling", parse_branch_mode),
681 OPT_STRING(0, "objdump", &objdump_path, "path",
682 "objdump binary to use for disassembly and annotations"),
683 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
684 "Disable symbol demangling"),
685 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
686 "Enable kernel symbol demangling"),
687 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
688 OPT_CALLBACK(0, "percent-limit", &report, "percent",
689 "Don't show entries under that percent", parse_percent_limit),
690 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
691 "how to display percentage of filtered entries", parse_filter_percentage),
694 struct perf_data_file file = {
695 .mode = PERF_DATA_MODE_READ,
698 perf_config(report__config, &report);
700 argc = parse_options(argc, argv, options, report_usage, 0);
702 if (report.use_stdio)
704 else if (report.use_tui)
706 else if (report.use_gtk)
709 if (report.inverted_callchain)
710 callchain_param.order = ORDER_CALLER;
712 if (!input_name || !strlen(input_name)) {
713 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
716 input_name = "perf.data";
719 file.path = input_name;
720 file.force = report.force;
723 session = perf_session__new(&file, false, &report.tool);
727 if (report.queue_size) {
728 ordered_events__set_alloc_size(&session->ordered_events,
732 report.session = session;
734 has_br_stack = perf_header__has_feat(&session->header,
735 HEADER_BRANCH_STACK);
737 if ((branch_mode == -1 && has_br_stack) || branch_mode == 1) {
738 sort__mode = SORT_MODE__BRANCH;
739 symbol_conf.cumulate_callchain = false;
742 if (report.mem_mode) {
743 if (sort__mode == SORT_MODE__BRANCH) {
744 pr_err("branch and mem mode incompatible\n");
747 sort__mode = SORT_MODE__MEMORY;
748 symbol_conf.cumulate_callchain = false;
751 if (setup_sorting() < 0) {
753 parse_options_usage(report_usage, options, "s", 1);
755 parse_options_usage(sort_order ? NULL : report_usage,
760 /* Force tty output for header output. */
761 if (report.header || report.header_only)
764 if (strcmp(input_name, "-") != 0)
769 if (report.header || report.header_only) {
770 perf_session__fprintf_info(session, stdout,
771 report.show_full_info);
772 if (report.header_only)
774 } else if (use_browser == 0) {
775 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
780 * Only in the TUI browser we are doing integrated annotation,
781 * so don't allocate extra space that won't be used in the stdio
784 if (ui__has_annotation()) {
785 symbol_conf.priv_size = sizeof(struct annotation);
786 machines__set_symbol_filter(&session->machines,
787 symbol__annotate_init);
789 * For searching by name on the "Browse map details".
790 * providing it only in verbose mode not to bloat too
791 * much struct symbol.
795 * XXX: Need to provide a less kludgy way to ask for
796 * more space per symbol, the u32 is for the index on
798 * See symbol__browser_index.
800 symbol_conf.priv_size += sizeof(u32);
801 symbol_conf.sort_by_name = true;
805 if (symbol__init(&session->header.env) < 0)
810 * Special case: if there's an argument left then assume that
811 * it's a symbol filter:
814 usage_with_options(report_usage, options);
816 report.symbol_filter_str = argv[0];
819 sort__setup_elide(stdout);
821 ret = __cmd_report(&report);
822 if (ret == K_SWITCH_INPUT_DATA) {
823 perf_session__delete(session);
829 perf_session__delete(session);