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;
60 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
63 static int perf_report_config(const char *var, const char *value, void *cb)
65 if (!strcmp(var, "report.group")) {
66 symbol_conf.event_group = perf_config_bool(var, value);
69 if (!strcmp(var, "report.percent-limit")) {
70 struct perf_report *rep = cb;
71 rep->min_percent = strtof(value, NULL);
75 return perf_default_config(var, value, cb);
78 static int perf_report__add_mem_hist_entry(struct perf_tool *tool,
79 struct addr_location *al,
80 struct perf_sample *sample,
81 struct perf_evsel *evsel,
82 struct machine *machine,
83 union perf_event *event)
85 struct perf_report *rep = container_of(tool, struct perf_report, tool);
86 struct symbol *parent = NULL;
87 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
89 struct hist_entry *he;
90 struct mem_info *mi, *mx;
93 if ((sort__has_parent || symbol_conf.use_callchain) &&
95 err = machine__resolve_callchain(machine, evsel, al->thread,
102 mi = machine__resolve_mem(machine, al->thread, sample, cpumode);
106 if (rep->hide_unresolved && !al->sym)
109 cost = sample->weight;
114 * must pass period=weight in order to get the correct
115 * sorting from hists__collapse_resort() which is solely
116 * based on periods. We want sorting be done on nr_events * weight
117 * and this is indirectly achieved by passing period=weight here
118 * and the he_stat__add_period() function.
120 he = __hists__add_entry(&evsel->hists, al, parent, NULL, mi,
126 * In the TUI browser, we are doing integrated annotation,
127 * so we don't allocate the extra space needed because the stdio
128 * code will not use it.
130 if (sort__has_sym && he->ms.sym && use_browser > 0) {
131 struct annotation *notes = symbol__annotation(he->ms.sym);
133 assert(evsel != NULL);
135 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
138 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
143 if (sort__has_sym && he->mem_info->daddr.sym && use_browser > 0) {
144 struct annotation *notes;
148 notes = symbol__annotation(mx->daddr.sym);
149 if (notes->src == NULL && symbol__alloc_hist(mx->daddr.sym) < 0)
152 err = symbol__inc_addr_samples(mx->daddr.sym,
160 evsel->hists.stats.total_period += cost;
161 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
164 if (symbol_conf.use_callchain) {
165 err = callchain_append(he->callchain,
173 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
174 struct addr_location *al,
175 struct perf_sample *sample,
176 struct perf_evsel *evsel,
177 struct machine *machine)
179 struct perf_report *rep = container_of(tool, struct perf_report, tool);
180 struct symbol *parent = NULL;
183 struct hist_entry *he;
184 struct branch_info *bi, *bx;
186 if ((sort__has_parent || symbol_conf.use_callchain)
187 && sample->callchain) {
188 err = machine__resolve_callchain(machine, evsel, al->thread,
195 bi = machine__resolve_bstack(machine, al->thread,
196 sample->branch_stack);
200 for (i = 0; i < sample->branch_stack->nr; i++) {
201 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
206 /* overwrite the 'al' to branch-to info */
207 al->map = bi[i].to.map;
208 al->sym = bi[i].to.sym;
209 al->addr = bi[i].to.addr;
211 * The report shows the percentage of total branches captured
212 * and not events sampled. Thus we use a pseudo period of 1.
214 he = __hists__add_entry(&evsel->hists, al, parent, &bi[i], NULL,
217 struct annotation *notes;
218 bx = he->branch_info;
219 if (bx->from.sym && use_browser == 1 && sort__has_sym) {
220 notes = symbol__annotation(bx->from.sym);
222 && symbol__alloc_hist(bx->from.sym) < 0)
225 err = symbol__inc_addr_samples(bx->from.sym,
233 if (bx->to.sym && use_browser == 1 && sort__has_sym) {
234 notes = symbol__annotation(bx->to.sym);
236 && symbol__alloc_hist(bx->to.sym) < 0)
239 err = symbol__inc_addr_samples(bx->to.sym,
246 evsel->hists.stats.total_period += 1;
247 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
257 static int perf_evsel__add_hist_entry(struct perf_tool *tool,
258 struct perf_evsel *evsel,
259 struct addr_location *al,
260 struct perf_sample *sample,
261 struct machine *machine)
263 struct perf_report *rep = container_of(tool, struct perf_report, tool);
264 struct symbol *parent = NULL;
266 struct hist_entry *he;
268 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
269 err = machine__resolve_callchain(machine, evsel, al->thread,
276 he = __hists__add_entry(&evsel->hists, al, parent, NULL, NULL,
277 sample->period, sample->weight,
278 sample->transaction);
282 if (symbol_conf.use_callchain) {
283 err = callchain_append(he->callchain,
290 * Only in the TUI browser we are doing integrated annotation,
291 * so we don't allocated the extra space needed because the stdio
292 * code will not use it.
294 if (he->ms.sym != NULL && use_browser == 1 && sort__has_sym) {
295 struct annotation *notes = symbol__annotation(he->ms.sym);
297 assert(evsel != NULL);
300 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
303 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
306 evsel->hists.stats.total_period += sample->period;
307 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
313 static int process_sample_event(struct perf_tool *tool,
314 union perf_event *event,
315 struct perf_sample *sample,
316 struct perf_evsel *evsel,
317 struct machine *machine)
319 struct perf_report *rep = container_of(tool, struct perf_report, tool);
320 struct addr_location al;
323 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
324 fprintf(stderr, "problem processing %d event, skipping it.\n",
329 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
332 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
335 if (sort__mode == SORT_MODE__BRANCH) {
336 ret = perf_report__add_branch_hist_entry(tool, &al, sample,
339 pr_debug("problem adding lbr entry, skipping event\n");
340 } else if (rep->mem_mode == 1) {
341 ret = perf_report__add_mem_hist_entry(tool, &al, sample,
342 evsel, machine, event);
344 pr_debug("problem adding mem entry, skipping event\n");
347 al.map->dso->hit = 1;
349 ret = perf_evsel__add_hist_entry(tool, evsel, &al, sample,
352 pr_debug("problem incrementing symbol period, skipping event\n");
357 static int process_read_event(struct perf_tool *tool,
358 union perf_event *event,
359 struct perf_sample *sample __maybe_unused,
360 struct perf_evsel *evsel,
361 struct machine *machine __maybe_unused)
363 struct perf_report *rep = container_of(tool, struct perf_report, tool);
365 if (rep->show_threads) {
366 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
367 perf_read_values_add_value(&rep->show_threads_values,
368 event->read.pid, event->read.tid,
374 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
375 evsel ? perf_evsel__name(evsel) : "FAIL",
381 /* For pipe mode, sample_type is not currently set */
382 static int perf_report__setup_sample_type(struct perf_report *rep)
384 struct perf_session *session = rep->session;
385 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
386 bool is_pipe = perf_data_file__is_pipe(session->file);
388 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
389 if (sort__has_parent) {
390 ui__error("Selected --sort parent, but no "
391 "callchain data. Did you call "
392 "'perf record' without -g?\n");
395 if (symbol_conf.use_callchain) {
396 ui__error("Selected -g but no callchain data. Did "
397 "you call 'perf record' without -g?\n");
400 } else if (!rep->dont_use_callchains &&
401 callchain_param.mode != CHAIN_NONE &&
402 !symbol_conf.use_callchain) {
403 symbol_conf.use_callchain = true;
404 if (callchain_register_param(&callchain_param) < 0) {
405 ui__error("Can't register callchain params.\n");
410 if (sort__mode == SORT_MODE__BRANCH) {
412 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
413 ui__error("Selected -b but no branch data. "
414 "Did you call perf record without -b?\n");
422 static void sig_handler(int sig __maybe_unused)
427 static size_t hists__fprintf_nr_sample_events(struct perf_report *rep,
429 const char *evname, FILE *fp)
433 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
434 u64 nr_events = hists->stats.total_period;
435 struct perf_evsel *evsel = hists_to_evsel(hists);
437 size_t size = sizeof(buf);
439 if (perf_evsel__is_group_event(evsel)) {
440 struct perf_evsel *pos;
442 perf_evsel__group_desc(evsel, buf, size);
445 for_each_group_member(pos, evsel) {
446 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
447 nr_events += pos->hists.stats.total_period;
451 nr_samples = convert_unit(nr_samples, &unit);
452 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
454 ret += fprintf(fp, " of event '%s'", evname);
457 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
458 ret += fprintf(fp, "\n# Sort order : %s", sort_order);
460 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
461 return ret + fprintf(fp, "\n#\n");
464 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
465 struct perf_report *rep,
468 struct perf_evsel *pos;
470 list_for_each_entry(pos, &evlist->entries, node) {
471 struct hists *hists = &pos->hists;
472 const char *evname = perf_evsel__name(pos);
474 if (symbol_conf.event_group &&
475 !perf_evsel__is_group_leader(pos))
478 hists__fprintf_nr_sample_events(rep, hists, evname, stdout);
479 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
480 fprintf(stdout, "\n\n");
483 if (sort_order == default_sort_order &&
484 parent_pattern == default_parent_pattern) {
485 fprintf(stdout, "#\n# (%s)\n#\n", help);
487 if (rep->show_threads) {
488 bool style = !strcmp(rep->pretty_printing_style, "raw");
489 perf_read_values_display(stdout, &rep->show_threads_values,
491 perf_read_values_destroy(&rep->show_threads_values);
498 static int __cmd_report(struct perf_report *rep)
502 struct perf_session *session = rep->session;
503 struct perf_evsel *pos;
504 struct map *kernel_map;
505 struct kmap *kernel_kmap;
506 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
507 struct ui_progress prog;
508 struct perf_data_file *file = session->file;
510 signal(SIGINT, sig_handler);
513 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
519 if (rep->show_threads)
520 perf_read_values_init(&rep->show_threads_values);
522 ret = perf_report__setup_sample_type(rep);
526 ret = perf_session__process_events(session, &rep->tool);
530 kernel_map = session->machines.host.vmlinux_maps[MAP__FUNCTION];
531 kernel_kmap = map__kmap(kernel_map);
532 if (kernel_map == NULL ||
533 (kernel_map->dso->hit &&
534 (kernel_kmap->ref_reloc_sym == NULL ||
535 kernel_kmap->ref_reloc_sym->addr == 0))) {
537 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
538 "can't be resolved.";
541 const struct dso *kdso = kernel_map->dso;
542 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
543 desc = "If some relocation was applied (e.g. "
544 "kexec) symbols may be misresolved.";
549 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
550 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
551 "Samples in kernel modules can't be resolved as well.\n\n",
556 perf_session__fprintf(session, stdout);
559 perf_session__fprintf_dsos(session, stdout);
562 perf_session__fprintf_nr_events(session, stdout);
567 list_for_each_entry(pos, &session->evlist->entries, node)
568 nr_samples += pos->hists.nr_entries;
570 ui_progress__init(&prog, nr_samples, "Merging related events...");
573 list_for_each_entry(pos, &session->evlist->entries, node) {
574 struct hists *hists = &pos->hists;
577 hists->symbol_filter_str = rep->symbol_filter_str;
579 hists__collapse_resort(hists, &prog);
580 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
582 /* Non-group events are considered as leader */
583 if (symbol_conf.event_group &&
584 !perf_evsel__is_group_leader(pos)) {
585 struct hists *leader_hists = &pos->leader->hists;
587 hists__match(leader_hists, hists);
588 hists__link(leader_hists, hists);
591 ui_progress__finish();
596 if (nr_samples == 0) {
597 ui__error("The %s file has no samples!\n", file->path);
601 list_for_each_entry(pos, &session->evlist->entries, node)
602 hists__output_resort(&pos->hists);
604 if (use_browser > 0) {
605 if (use_browser == 1) {
606 ret = perf_evlist__tui_browse_hists(session->evlist,
609 &session->header.env);
611 * Usually "ret" is the last pressed key, and we only
612 * care if the key notifies us to switch data file.
614 if (ret != K_SWITCH_INPUT_DATA)
617 } else if (use_browser == 2) {
618 int (*hist_browser)(struct perf_evlist *,
620 struct hist_browser_timer *,
623 hist_browser = dlsym(perf_gtk_handle,
624 "perf_evlist__gtk_browse_hists");
625 if (hist_browser == NULL) {
626 ui__error("GTK browser not found!\n");
629 hist_browser(session->evlist, help, NULL,
633 perf_evlist__tty_browse_hists(session->evlist, rep, help);
639 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
641 struct perf_report *rep = (struct perf_report *)opt->value;
649 rep->dont_use_callchains = true;
653 symbol_conf.use_callchain = true;
658 tok = strtok((char *)arg, ",");
662 /* get the output mode */
663 if (!strncmp(tok, "graph", strlen(arg)))
664 callchain_param.mode = CHAIN_GRAPH_ABS;
666 else if (!strncmp(tok, "flat", strlen(arg)))
667 callchain_param.mode = CHAIN_FLAT;
669 else if (!strncmp(tok, "fractal", strlen(arg)))
670 callchain_param.mode = CHAIN_GRAPH_REL;
672 else if (!strncmp(tok, "none", strlen(arg))) {
673 callchain_param.mode = CHAIN_NONE;
674 symbol_conf.use_callchain = false;
682 /* get the min percentage */
683 tok = strtok(NULL, ",");
687 callchain_param.min_percent = strtod(tok, &endptr);
691 /* get the print limit */
692 tok2 = strtok(NULL, ",");
696 if (tok2[0] != 'c') {
697 callchain_param.print_limit = strtoul(tok2, &endptr, 0);
698 tok2 = strtok(NULL, ",");
703 /* get the call chain order */
704 if (!strncmp(tok2, "caller", strlen("caller")))
705 callchain_param.order = ORDER_CALLER;
706 else if (!strncmp(tok2, "callee", strlen("callee")))
707 callchain_param.order = ORDER_CALLEE;
711 /* Get the sort key */
712 tok2 = strtok(NULL, ",");
715 if (!strncmp(tok2, "function", strlen("function")))
716 callchain_param.key = CCKEY_FUNCTION;
717 else if (!strncmp(tok2, "address", strlen("address")))
718 callchain_param.key = CCKEY_ADDRESS;
722 if (callchain_register_param(&callchain_param) < 0) {
723 fprintf(stderr, "Can't register callchain params\n");
730 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
731 const char *arg, int unset __maybe_unused)
734 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
737 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
738 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
741 have_ignore_callees = 1;
748 parse_branch_mode(const struct option *opt __maybe_unused,
749 const char *str __maybe_unused, int unset)
751 int *branch_mode = opt->value;
753 *branch_mode = !unset;
758 parse_percent_limit(const struct option *opt, const char *str,
759 int unset __maybe_unused)
761 struct perf_report *rep = opt->value;
763 rep->min_percent = strtof(str, NULL);
767 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
769 struct perf_session *session;
771 bool has_br_stack = false;
772 int branch_mode = -1;
774 char callchain_default_opt[] = "fractal,0.5,callee";
775 const char * const report_usage[] = {
776 "perf report [<options>]",
779 struct perf_report report = {
781 .sample = process_sample_event,
782 .mmap = perf_event__process_mmap,
783 .mmap2 = perf_event__process_mmap2,
784 .comm = perf_event__process_comm,
785 .exit = perf_event__process_exit,
786 .fork = perf_event__process_fork,
787 .lost = perf_event__process_lost,
788 .read = process_read_event,
789 .attr = perf_event__process_attr,
790 .tracing_data = perf_event__process_tracing_data,
791 .build_id = perf_event__process_build_id,
792 .ordered_samples = true,
793 .ordering_requires_timestamps = true,
795 .max_stack = PERF_MAX_STACK_DEPTH,
796 .pretty_printing_style = "normal",
798 const struct option options[] = {
799 OPT_STRING('i', "input", &input_name, "file",
801 OPT_INCR('v', "verbose", &verbose,
802 "be more verbose (show symbol address, etc)"),
803 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
804 "dump raw trace in ASCII"),
805 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
806 "file", "vmlinux pathname"),
807 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
808 "file", "kallsyms pathname"),
809 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
810 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
811 "load module symbols - WARNING: use only with -k and LIVE kernel"),
812 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
813 "Show a column with the number of samples"),
814 OPT_BOOLEAN('T', "threads", &report.show_threads,
815 "Show per-thread event counters"),
816 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
817 "pretty printing style key: normal raw"),
818 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
819 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
820 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
821 "Use the stdio interface"),
822 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
823 OPT_BOOLEAN(0, "header-only", &report.header_only,
824 "Show only data header."),
825 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
826 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
827 " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
828 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
829 "snoop, locked, abort, in_tx, transaction"),
830 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
831 "Show sample percentage for different cpu modes"),
832 OPT_STRING('p', "parent", &parent_pattern, "regex",
833 "regex filter to identify parent, see: '--sort parent'"),
834 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
835 "Only display entries with parent-match"),
836 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
837 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
838 "Default: fractal,0.5,callee,function", &parse_callchain_opt, callchain_default_opt),
839 OPT_INTEGER(0, "max-stack", &report.max_stack,
840 "Set the maximum stack depth when parsing the callchain, "
841 "anything beyond the specified depth will be ignored. "
842 "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
843 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
844 "alias for inverted call graph"),
845 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
846 "ignore callees of these functions in call graphs",
847 report_parse_ignore_callees_opt),
848 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
849 "only consider symbols in these dsos"),
850 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
851 "only consider symbols in these comms"),
852 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
853 "only consider these symbols"),
854 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
855 "only show symbols that (partially) match with this filter"),
856 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
858 "don't try to adjust column width, use these fixed values"),
859 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
860 "separator for columns, no spaces will be added between "
861 "columns '.' is reserved."),
862 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
863 "Only display entries resolved to a symbol"),
864 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
865 "Look for files with symbols relative to this directory"),
866 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
867 "list of cpus to profile"),
868 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
869 "Display extended information about perf.data file"),
870 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
871 "Interleave source code with assembly code (default)"),
872 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
873 "Display raw encoding of assembly instructions (default)"),
874 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
875 "Specify disassembler style (e.g. -M intel for intel syntax)"),
876 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
877 "Show a column with the sum of periods"),
878 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
879 "Show event group information together"),
880 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
881 "use branch records for histogram filling", parse_branch_mode),
882 OPT_STRING(0, "objdump", &objdump_path, "path",
883 "objdump binary to use for disassembly and annotations"),
884 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
885 "Disable symbol demangling"),
886 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
887 OPT_CALLBACK(0, "percent-limit", &report, "percent",
888 "Don't show entries under that percent", parse_percent_limit),
891 struct perf_data_file file = {
892 .mode = PERF_DATA_MODE_READ,
895 perf_config(perf_report_config, &report);
897 argc = parse_options(argc, argv, options, report_usage, 0);
899 if (report.use_stdio)
901 else if (report.use_tui)
903 else if (report.use_gtk)
906 if (report.inverted_callchain)
907 callchain_param.order = ORDER_CALLER;
909 if (!input_name || !strlen(input_name)) {
910 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
913 input_name = "perf.data";
916 file.path = input_name;
917 file.force = report.force;
920 session = perf_session__new(&file, false, &report.tool);
924 report.session = session;
926 has_br_stack = perf_header__has_feat(&session->header,
927 HEADER_BRANCH_STACK);
929 if (branch_mode == -1 && has_br_stack)
930 sort__mode = SORT_MODE__BRANCH;
932 /* sort__mode could be NORMAL if --no-branch-stack */
933 if (sort__mode == SORT_MODE__BRANCH) {
935 * if no sort_order is provided, then specify
936 * branch-mode specific order
938 if (sort_order == default_sort_order)
939 sort_order = "comm,dso_from,symbol_from,"
943 if (report.mem_mode) {
944 if (sort__mode == SORT_MODE__BRANCH) {
945 fprintf(stderr, "branch and mem mode incompatible\n");
948 sort__mode = SORT_MODE__MEMORY;
951 * if no sort_order is provided, then specify
952 * branch-mode specific order
954 if (sort_order == default_sort_order)
955 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
958 if (setup_sorting() < 0) {
959 parse_options_usage(report_usage, options, "s", 1);
963 if (parent_pattern != default_parent_pattern) {
964 if (sort_dimension__add("parent") < 0)
968 /* Force tty output for header output. */
969 if (report.header || report.header_only)
972 if (strcmp(input_name, "-") != 0)
979 if (report.header || report.header_only) {
980 perf_session__fprintf_info(session, stdout,
981 report.show_full_info);
982 if (report.header_only)
984 } else if (use_browser == 0) {
985 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
990 * Only in the TUI browser we are doing integrated annotation,
991 * so don't allocate extra space that won't be used in the stdio
994 if (use_browser == 1 && sort__has_sym) {
995 symbol_conf.priv_size = sizeof(struct annotation);
996 machines__set_symbol_filter(&session->machines,
997 symbol__annotate_init);
999 * For searching by name on the "Browse map details".
1000 * providing it only in verbose mode not to bloat too
1001 * much struct symbol.
1005 * XXX: Need to provide a less kludgy way to ask for
1006 * more space per symbol, the u32 is for the index on
1008 * See symbol__browser_index.
1010 symbol_conf.priv_size += sizeof(u32);
1011 symbol_conf.sort_by_name = true;
1015 if (symbol__init() < 0)
1020 * Special case: if there's an argument left then assume that
1021 * it's a symbol filter:
1024 usage_with_options(report_usage, options);
1026 report.symbol_filter_str = argv[0];
1029 sort__setup_elide(stdout);
1031 ret = __cmd_report(&report);
1032 if (ret == K_SWITCH_INPUT_DATA) {
1033 perf_session__delete(session);
1039 perf_session__delete(session);