perf tools: Account entry stats when it's added to the output tree
[linux-2.6/btrfs-unstable.git] / tools / perf / builtin-report.c
blob89c95289fd5102b3b1257d91e4b603d212986347
1 /*
2 * builtin-report.c
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.
7 */
8 #include "builtin.h"
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"
22 #include "perf.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"
39 #include <dlfcn.h>
40 #include <linux/bitmap.h>
42 struct report {
43 struct perf_tool tool;
44 struct perf_session *session;
45 bool force, use_tui, use_gtk, use_stdio;
46 bool hide_unresolved;
47 bool dont_use_callchains;
48 bool show_full_info;
49 bool show_threads;
50 bool inverted_callchain;
51 bool mem_mode;
52 bool header;
53 bool header_only;
54 int max_stack;
55 struct perf_read_values show_threads_values;
56 const char *pretty_printing_style;
57 const char *cpu_list;
58 const char *symbol_filter_str;
59 float min_percent;
60 u64 nr_entries;
61 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
64 static int report__config(const char *var, const char *value, void *cb)
66 if (!strcmp(var, "report.group")) {
67 symbol_conf.event_group = perf_config_bool(var, value);
68 return 0;
70 if (!strcmp(var, "report.percent-limit")) {
71 struct report *rep = cb;
72 rep->min_percent = strtof(value, NULL);
73 return 0;
76 return perf_default_config(var, value, cb);
79 static void report__inc_stats(struct report *rep, struct hist_entry *he)
82 * The @he is either of a newly created one or an existing one
83 * merging current sample. We only want to count a new one so
84 * checking ->nr_events being 1.
86 if (he->stat.nr_events == 1)
87 rep->nr_entries++;
90 * Only counts number of samples at this stage as it's more
91 * natural to do it here and non-sample events are also
92 * counted in perf_session_deliver_event(). The dump_trace
93 * requires this info is ready before going to the output tree.
95 hists__inc_nr_events(he->hists, PERF_RECORD_SAMPLE);
96 if (!he->filtered)
97 he->hists->stats.nr_non_filtered_samples++;
100 static int report__add_mem_hist_entry(struct report *rep, struct addr_location *al,
101 struct perf_sample *sample, struct perf_evsel *evsel)
103 struct symbol *parent = NULL;
104 struct hist_entry *he;
105 struct mem_info *mi, *mx;
106 uint64_t cost;
107 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
109 if (err)
110 return err;
112 mi = sample__resolve_mem(sample, al);
113 if (!mi)
114 return -ENOMEM;
116 if (rep->hide_unresolved && !al->sym)
117 return 0;
119 cost = sample->weight;
120 if (!cost)
121 cost = 1;
124 * must pass period=weight in order to get the correct
125 * sorting from hists__collapse_resort() which is solely
126 * based on periods. We want sorting be done on nr_events * weight
127 * and this is indirectly achieved by passing period=weight here
128 * and the he_stat__add_period() function.
130 he = __hists__add_entry(&evsel->hists, al, parent, NULL, mi,
131 cost, cost, 0);
132 if (!he)
133 return -ENOMEM;
135 if (ui__has_annotation()) {
136 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
137 if (err)
138 goto out;
140 mx = he->mem_info;
141 err = addr_map_symbol__inc_samples(&mx->daddr, evsel->idx);
142 if (err)
143 goto out;
146 report__inc_stats(rep, he);
148 err = hist_entry__append_callchain(he, sample);
149 out:
150 return err;
153 static int report__add_branch_hist_entry(struct report *rep, struct addr_location *al,
154 struct perf_sample *sample, struct perf_evsel *evsel)
156 struct symbol *parent = NULL;
157 unsigned i;
158 struct hist_entry *he;
159 struct branch_info *bi, *bx;
160 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
162 if (err)
163 return err;
165 bi = sample__resolve_bstack(sample, al);
166 if (!bi)
167 return -ENOMEM;
169 for (i = 0; i < sample->branch_stack->nr; i++) {
170 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
171 continue;
173 err = -ENOMEM;
175 /* overwrite the 'al' to branch-to info */
176 al->map = bi[i].to.map;
177 al->sym = bi[i].to.sym;
178 al->addr = bi[i].to.addr;
180 * The report shows the percentage of total branches captured
181 * and not events sampled. Thus we use a pseudo period of 1.
183 he = __hists__add_entry(&evsel->hists, al, parent, &bi[i], NULL,
184 1, 1, 0);
185 if (he) {
186 if (ui__has_annotation()) {
187 bx = he->branch_info;
188 err = addr_map_symbol__inc_samples(&bx->from,
189 evsel->idx);
190 if (err)
191 goto out;
193 err = addr_map_symbol__inc_samples(&bx->to,
194 evsel->idx);
195 if (err)
196 goto out;
198 report__inc_stats(rep, he);
199 } else
200 goto out;
202 err = 0;
203 out:
204 free(bi);
205 return err;
208 static int report__add_hist_entry(struct report *rep, struct perf_evsel *evsel,
209 struct addr_location *al, struct perf_sample *sample)
211 struct symbol *parent = NULL;
212 struct hist_entry *he;
213 int err = sample__resolve_callchain(sample, &parent, evsel, al, rep->max_stack);
215 if (err)
216 return err;
218 he = __hists__add_entry(&evsel->hists, al, parent, NULL, NULL,
219 sample->period, sample->weight,
220 sample->transaction);
221 if (he == NULL)
222 return -ENOMEM;
224 err = hist_entry__append_callchain(he, sample);
225 if (err)
226 goto out;
228 if (ui__has_annotation())
229 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
231 report__inc_stats(rep, he);
233 out:
234 return err;
238 static int process_sample_event(struct perf_tool *tool,
239 union perf_event *event,
240 struct perf_sample *sample,
241 struct perf_evsel *evsel,
242 struct machine *machine)
244 struct report *rep = container_of(tool, struct report, tool);
245 struct addr_location al;
246 int ret;
248 if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
249 pr_debug("problem processing %d event, skipping it.\n",
250 event->header.type);
251 return -1;
254 if (rep->hide_unresolved && al.sym == NULL)
255 return 0;
257 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
258 return 0;
260 if (sort__mode == SORT_MODE__BRANCH) {
261 ret = report__add_branch_hist_entry(rep, &al, sample, evsel);
262 if (ret < 0)
263 pr_debug("problem adding lbr entry, skipping event\n");
264 } else if (rep->mem_mode == 1) {
265 ret = report__add_mem_hist_entry(rep, &al, sample, evsel);
266 if (ret < 0)
267 pr_debug("problem adding mem entry, skipping event\n");
268 } else {
269 if (al.map != NULL)
270 al.map->dso->hit = 1;
272 ret = report__add_hist_entry(rep, evsel, &al, sample);
273 if (ret < 0)
274 pr_debug("problem incrementing symbol period, skipping event\n");
276 return ret;
279 static int process_read_event(struct perf_tool *tool,
280 union perf_event *event,
281 struct perf_sample *sample __maybe_unused,
282 struct perf_evsel *evsel,
283 struct machine *machine __maybe_unused)
285 struct report *rep = container_of(tool, struct report, tool);
287 if (rep->show_threads) {
288 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
289 perf_read_values_add_value(&rep->show_threads_values,
290 event->read.pid, event->read.tid,
291 event->read.id,
292 name,
293 event->read.value);
296 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
297 evsel ? perf_evsel__name(evsel) : "FAIL",
298 event->read.value);
300 return 0;
303 /* For pipe mode, sample_type is not currently set */
304 static int report__setup_sample_type(struct report *rep)
306 struct perf_session *session = rep->session;
307 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
308 bool is_pipe = perf_data_file__is_pipe(session->file);
310 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
311 if (sort__has_parent) {
312 ui__error("Selected --sort parent, but no "
313 "callchain data. Did you call "
314 "'perf record' without -g?\n");
315 return -EINVAL;
317 if (symbol_conf.use_callchain) {
318 ui__error("Selected -g but no callchain data. Did "
319 "you call 'perf record' without -g?\n");
320 return -1;
322 } else if (!rep->dont_use_callchains &&
323 callchain_param.mode != CHAIN_NONE &&
324 !symbol_conf.use_callchain) {
325 symbol_conf.use_callchain = true;
326 if (callchain_register_param(&callchain_param) < 0) {
327 ui__error("Can't register callchain params.\n");
328 return -EINVAL;
332 if (sort__mode == SORT_MODE__BRANCH) {
333 if (!is_pipe &&
334 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
335 ui__error("Selected -b but no branch data. "
336 "Did you call perf record without -b?\n");
337 return -1;
341 return 0;
344 static void sig_handler(int sig __maybe_unused)
346 session_done = 1;
349 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
350 const char *evname, FILE *fp)
352 size_t ret;
353 char unit;
354 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
355 u64 nr_events = hists->stats.total_period;
356 struct perf_evsel *evsel = hists_to_evsel(hists);
357 char buf[512];
358 size_t size = sizeof(buf);
360 if (symbol_conf.filter_relative) {
361 nr_samples = hists->stats.nr_non_filtered_samples;
362 nr_events = hists->stats.total_non_filtered_period;
365 if (perf_evsel__is_group_event(evsel)) {
366 struct perf_evsel *pos;
368 perf_evsel__group_desc(evsel, buf, size);
369 evname = buf;
371 for_each_group_member(pos, evsel) {
372 if (symbol_conf.filter_relative) {
373 nr_samples += pos->hists.stats.nr_non_filtered_samples;
374 nr_events += pos->hists.stats.total_non_filtered_period;
375 } else {
376 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
377 nr_events += pos->hists.stats.total_period;
382 nr_samples = convert_unit(nr_samples, &unit);
383 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
384 if (evname != NULL)
385 ret += fprintf(fp, " of event '%s'", evname);
387 if (rep->mem_mode) {
388 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
389 ret += fprintf(fp, "\n# Sort order : %s", sort_order);
390 } else
391 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
392 return ret + fprintf(fp, "\n#\n");
395 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
396 struct report *rep,
397 const char *help)
399 struct perf_evsel *pos;
401 evlist__for_each(evlist, pos) {
402 struct hists *hists = &pos->hists;
403 const char *evname = perf_evsel__name(pos);
405 if (symbol_conf.event_group &&
406 !perf_evsel__is_group_leader(pos))
407 continue;
409 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
410 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
411 fprintf(stdout, "\n\n");
414 if (sort_order == default_sort_order &&
415 parent_pattern == default_parent_pattern) {
416 fprintf(stdout, "#\n# (%s)\n#\n", help);
418 if (rep->show_threads) {
419 bool style = !strcmp(rep->pretty_printing_style, "raw");
420 perf_read_values_display(stdout, &rep->show_threads_values,
421 style);
422 perf_read_values_destroy(&rep->show_threads_values);
426 return 0;
429 static void report__warn_kptr_restrict(const struct report *rep)
431 struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION];
432 struct kmap *kernel_kmap = map__kmap(kernel_map);
434 if (kernel_map == NULL ||
435 (kernel_map->dso->hit &&
436 (kernel_kmap->ref_reloc_sym == NULL ||
437 kernel_kmap->ref_reloc_sym->addr == 0))) {
438 const char *desc =
439 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
440 "can't be resolved.";
442 if (kernel_map) {
443 const struct dso *kdso = kernel_map->dso;
444 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
445 desc = "If some relocation was applied (e.g. "
446 "kexec) symbols may be misresolved.";
450 ui__warning(
451 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
452 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
453 "Samples in kernel modules can't be resolved as well.\n\n",
454 desc);
458 static int report__gtk_browse_hists(struct report *rep, const char *help)
460 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
461 struct hist_browser_timer *timer, float min_pcnt);
463 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
465 if (hist_browser == NULL) {
466 ui__error("GTK browser not found!\n");
467 return -1;
470 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
473 static int report__browse_hists(struct report *rep)
475 int ret;
476 struct perf_session *session = rep->session;
477 struct perf_evlist *evlist = session->evlist;
478 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
480 switch (use_browser) {
481 case 1:
482 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
483 rep->min_percent,
484 &session->header.env);
486 * Usually "ret" is the last pressed key, and we only
487 * care if the key notifies us to switch data file.
489 if (ret != K_SWITCH_INPUT_DATA)
490 ret = 0;
491 break;
492 case 2:
493 ret = report__gtk_browse_hists(rep, help);
494 break;
495 default:
496 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
497 break;
500 return ret;
503 static void report__collapse_hists(struct report *rep)
505 struct ui_progress prog;
506 struct perf_evsel *pos;
508 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
510 evlist__for_each(rep->session->evlist, pos) {
511 struct hists *hists = &pos->hists;
513 if (pos->idx == 0)
514 hists->symbol_filter_str = rep->symbol_filter_str;
516 hists__collapse_resort(hists, &prog);
518 /* Non-group events are considered as leader */
519 if (symbol_conf.event_group &&
520 !perf_evsel__is_group_leader(pos)) {
521 struct hists *leader_hists = &pos->leader->hists;
523 hists__match(leader_hists, hists);
524 hists__link(leader_hists, hists);
528 ui_progress__finish();
531 static int __cmd_report(struct report *rep)
533 int ret;
534 struct perf_session *session = rep->session;
535 struct perf_evsel *pos;
536 struct perf_data_file *file = session->file;
538 signal(SIGINT, sig_handler);
540 if (rep->cpu_list) {
541 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
542 rep->cpu_bitmap);
543 if (ret)
544 return ret;
547 if (rep->show_threads)
548 perf_read_values_init(&rep->show_threads_values);
550 ret = report__setup_sample_type(rep);
551 if (ret)
552 return ret;
554 ret = perf_session__process_events(session, &rep->tool);
555 if (ret)
556 return ret;
558 report__warn_kptr_restrict(rep);
560 if (use_browser == 0) {
561 if (verbose > 3)
562 perf_session__fprintf(session, stdout);
564 if (verbose > 2)
565 perf_session__fprintf_dsos(session, stdout);
567 if (dump_trace) {
568 perf_session__fprintf_nr_events(session, stdout);
569 return 0;
573 report__collapse_hists(rep);
575 if (session_done())
576 return 0;
578 if (rep->nr_entries == 0) {
579 ui__error("The %s file has no samples!\n", file->path);
580 return 0;
583 evlist__for_each(session->evlist, pos)
584 hists__output_resort(&pos->hists);
586 return report__browse_hists(rep);
589 static int
590 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
592 struct report *rep = (struct report *)opt->value;
595 * --no-call-graph
597 if (unset) {
598 rep->dont_use_callchains = true;
599 return 0;
602 return parse_callchain_report_opt(arg);
606 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
607 const char *arg, int unset __maybe_unused)
609 if (arg) {
610 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
611 if (err) {
612 char buf[BUFSIZ];
613 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
614 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
615 return -1;
617 have_ignore_callees = 1;
620 return 0;
623 static int
624 parse_branch_mode(const struct option *opt __maybe_unused,
625 const char *str __maybe_unused, int unset)
627 int *branch_mode = opt->value;
629 *branch_mode = !unset;
630 return 0;
633 static int
634 parse_percent_limit(const struct option *opt, const char *str,
635 int unset __maybe_unused)
637 struct report *rep = opt->value;
639 rep->min_percent = strtof(str, NULL);
640 return 0;
643 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
645 struct perf_session *session;
646 struct stat st;
647 bool has_br_stack = false;
648 int branch_mode = -1;
649 int ret = -1;
650 char callchain_default_opt[] = "fractal,0.5,callee";
651 const char * const report_usage[] = {
652 "perf report [<options>]",
653 NULL
655 struct report report = {
656 .tool = {
657 .sample = process_sample_event,
658 .mmap = perf_event__process_mmap,
659 .mmap2 = perf_event__process_mmap2,
660 .comm = perf_event__process_comm,
661 .exit = perf_event__process_exit,
662 .fork = perf_event__process_fork,
663 .lost = perf_event__process_lost,
664 .read = process_read_event,
665 .attr = perf_event__process_attr,
666 .tracing_data = perf_event__process_tracing_data,
667 .build_id = perf_event__process_build_id,
668 .ordered_samples = true,
669 .ordering_requires_timestamps = true,
671 .max_stack = PERF_MAX_STACK_DEPTH,
672 .pretty_printing_style = "normal",
674 const struct option options[] = {
675 OPT_STRING('i', "input", &input_name, "file",
676 "input file name"),
677 OPT_INCR('v', "verbose", &verbose,
678 "be more verbose (show symbol address, etc)"),
679 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
680 "dump raw trace in ASCII"),
681 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
682 "file", "vmlinux pathname"),
683 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
684 "file", "kallsyms pathname"),
685 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
686 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
687 "load module symbols - WARNING: use only with -k and LIVE kernel"),
688 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
689 "Show a column with the number of samples"),
690 OPT_BOOLEAN('T', "threads", &report.show_threads,
691 "Show per-thread event counters"),
692 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
693 "pretty printing style key: normal raw"),
694 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
695 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
696 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
697 "Use the stdio interface"),
698 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
699 OPT_BOOLEAN(0, "header-only", &report.header_only,
700 "Show only data header."),
701 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
702 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
703 " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
704 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
705 "snoop, locked, abort, in_tx, transaction"),
706 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
707 "Show sample percentage for different cpu modes"),
708 OPT_STRING('p', "parent", &parent_pattern, "regex",
709 "regex filter to identify parent, see: '--sort parent'"),
710 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
711 "Only display entries with parent-match"),
712 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
713 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
714 "Default: fractal,0.5,callee,function", &report_parse_callchain_opt, callchain_default_opt),
715 OPT_INTEGER(0, "max-stack", &report.max_stack,
716 "Set the maximum stack depth when parsing the callchain, "
717 "anything beyond the specified depth will be ignored. "
718 "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
719 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
720 "alias for inverted call graph"),
721 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
722 "ignore callees of these functions in call graphs",
723 report_parse_ignore_callees_opt),
724 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
725 "only consider symbols in these dsos"),
726 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
727 "only consider symbols in these comms"),
728 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
729 "only consider these symbols"),
730 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
731 "only show symbols that (partially) match with this filter"),
732 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
733 "width[,width...]",
734 "don't try to adjust column width, use these fixed values"),
735 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
736 "separator for columns, no spaces will be added between "
737 "columns '.' is reserved."),
738 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
739 "Only display entries resolved to a symbol"),
740 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
741 "Look for files with symbols relative to this directory"),
742 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
743 "list of cpus to profile"),
744 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
745 "Display extended information about perf.data file"),
746 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
747 "Interleave source code with assembly code (default)"),
748 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
749 "Display raw encoding of assembly instructions (default)"),
750 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
751 "Specify disassembler style (e.g. -M intel for intel syntax)"),
752 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
753 "Show a column with the sum of periods"),
754 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
755 "Show event group information together"),
756 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
757 "use branch records for histogram filling", parse_branch_mode),
758 OPT_STRING(0, "objdump", &objdump_path, "path",
759 "objdump binary to use for disassembly and annotations"),
760 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
761 "Disable symbol demangling"),
762 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
763 OPT_CALLBACK(0, "percent-limit", &report, "percent",
764 "Don't show entries under that percent", parse_percent_limit),
765 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
766 "how to display percentage of filtered entries", parse_filter_percentage),
767 OPT_END()
769 struct perf_data_file file = {
770 .mode = PERF_DATA_MODE_READ,
773 perf_config(report__config, &report);
775 argc = parse_options(argc, argv, options, report_usage, 0);
777 if (report.use_stdio)
778 use_browser = 0;
779 else if (report.use_tui)
780 use_browser = 1;
781 else if (report.use_gtk)
782 use_browser = 2;
784 if (report.inverted_callchain)
785 callchain_param.order = ORDER_CALLER;
787 if (!input_name || !strlen(input_name)) {
788 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
789 input_name = "-";
790 else
791 input_name = "perf.data";
794 file.path = input_name;
795 file.force = report.force;
797 repeat:
798 session = perf_session__new(&file, false, &report.tool);
799 if (session == NULL)
800 return -ENOMEM;
802 report.session = session;
804 has_br_stack = perf_header__has_feat(&session->header,
805 HEADER_BRANCH_STACK);
807 if (branch_mode == -1 && has_br_stack)
808 sort__mode = SORT_MODE__BRANCH;
810 /* sort__mode could be NORMAL if --no-branch-stack */
811 if (sort__mode == SORT_MODE__BRANCH) {
813 * if no sort_order is provided, then specify
814 * branch-mode specific order
816 if (sort_order == default_sort_order)
817 sort_order = "comm,dso_from,symbol_from,"
818 "dso_to,symbol_to";
821 if (report.mem_mode) {
822 if (sort__mode == SORT_MODE__BRANCH) {
823 pr_err("branch and mem mode incompatible\n");
824 goto error;
826 sort__mode = SORT_MODE__MEMORY;
829 * if no sort_order is provided, then specify
830 * branch-mode specific order
832 if (sort_order == default_sort_order)
833 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
836 if (setup_sorting() < 0) {
837 parse_options_usage(report_usage, options, "s", 1);
838 goto error;
841 if (parent_pattern != default_parent_pattern) {
842 if (sort_dimension__add("parent") < 0)
843 goto error;
846 /* Force tty output for header output. */
847 if (report.header || report.header_only)
848 use_browser = 0;
850 if (strcmp(input_name, "-") != 0)
851 setup_browser(true);
852 else {
853 use_browser = 0;
854 perf_hpp__init();
857 if (report.header || report.header_only) {
858 perf_session__fprintf_info(session, stdout,
859 report.show_full_info);
860 if (report.header_only)
861 return 0;
862 } else if (use_browser == 0) {
863 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
864 stdout);
868 * Only in the TUI browser we are doing integrated annotation,
869 * so don't allocate extra space that won't be used in the stdio
870 * implementation.
872 if (ui__has_annotation()) {
873 symbol_conf.priv_size = sizeof(struct annotation);
874 machines__set_symbol_filter(&session->machines,
875 symbol__annotate_init);
877 * For searching by name on the "Browse map details".
878 * providing it only in verbose mode not to bloat too
879 * much struct symbol.
881 if (verbose) {
883 * XXX: Need to provide a less kludgy way to ask for
884 * more space per symbol, the u32 is for the index on
885 * the ui browser.
886 * See symbol__browser_index.
888 symbol_conf.priv_size += sizeof(u32);
889 symbol_conf.sort_by_name = true;
893 if (symbol__init() < 0)
894 goto error;
896 if (argc) {
898 * Special case: if there's an argument left then assume that
899 * it's a symbol filter:
901 if (argc > 1)
902 usage_with_options(report_usage, options);
904 report.symbol_filter_str = argv[0];
907 sort__setup_elide(stdout);
909 ret = __cmd_report(&report);
910 if (ret == K_SWITCH_INPUT_DATA) {
911 perf_session__delete(session);
912 goto repeat;
913 } else
914 ret = 0;
916 error:
917 perf_session__delete(session);
918 return ret;