Merge tag 'perf-core-for-mingo' of git://git.kernel.org/pub/scm/linux/kernel/git...
[firefly-linux-kernel-4.4.55.git] / tools / perf / builtin-report.c
1 /*
2  * builtin-report.c
3  *
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"
9
10 #include "util/util.h"
11 #include "util/cache.h"
12
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"
21
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"
29
30 #include "util/parse-options.h"
31 #include "util/parse-events.h"
32
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"
38
39 #include "util/auxtrace.h"
40
41 #include <dlfcn.h>
42 #include <linux/bitmap.h>
43
44 struct report {
45         struct perf_tool        tool;
46         struct perf_session     *session;
47         bool                    force, use_tui, use_gtk, use_stdio;
48         bool                    hide_unresolved;
49         bool                    dont_use_callchains;
50         bool                    show_full_info;
51         bool                    show_threads;
52         bool                    inverted_callchain;
53         bool                    mem_mode;
54         bool                    header;
55         bool                    header_only;
56         bool                    nonany_branch_mode;
57         int                     max_stack;
58         struct perf_read_values show_threads_values;
59         const char              *pretty_printing_style;
60         const char              *cpu_list;
61         const char              *symbol_filter_str;
62         float                   min_percent;
63         u64                     nr_entries;
64         u64                     queue_size;
65         DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
66 };
67
68 static int report__config(const char *var, const char *value, void *cb)
69 {
70         struct report *rep = cb;
71
72         if (!strcmp(var, "report.group")) {
73                 symbol_conf.event_group = perf_config_bool(var, value);
74                 return 0;
75         }
76         if (!strcmp(var, "report.percent-limit")) {
77                 rep->min_percent = strtof(value, NULL);
78                 return 0;
79         }
80         if (!strcmp(var, "report.children")) {
81                 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
82                 return 0;
83         }
84         if (!strcmp(var, "report.queue-size")) {
85                 rep->queue_size = perf_config_u64(var, value);
86                 return 0;
87         }
88
89         return perf_default_config(var, value, cb);
90 }
91
92 static int hist_iter__report_callback(struct hist_entry_iter *iter,
93                                       struct addr_location *al, bool single,
94                                       void *arg)
95 {
96         int err = 0;
97         struct report *rep = arg;
98         struct hist_entry *he = iter->he;
99         struct perf_evsel *evsel = iter->evsel;
100         struct mem_info *mi;
101         struct branch_info *bi;
102
103         if (!ui__has_annotation())
104                 return 0;
105
106         hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
107                              rep->nonany_branch_mode);
108
109         if (sort__mode == SORT_MODE__BRANCH) {
110                 bi = he->branch_info;
111                 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
112                 if (err)
113                         goto out;
114
115                 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
116
117         } else if (rep->mem_mode) {
118                 mi = he->mem_info;
119                 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
120                 if (err)
121                         goto out;
122
123                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
124
125         } else if (symbol_conf.cumulate_callchain) {
126                 if (single)
127                         err = hist_entry__inc_addr_samples(he, evsel->idx,
128                                                            al->addr);
129         } else {
130                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
131         }
132
133 out:
134         return err;
135 }
136
137 static int process_sample_event(struct perf_tool *tool,
138                                 union perf_event *event,
139                                 struct perf_sample *sample,
140                                 struct perf_evsel *evsel,
141                                 struct machine *machine)
142 {
143         struct report *rep = container_of(tool, struct report, tool);
144         struct addr_location al;
145         struct hist_entry_iter iter = {
146                 .evsel                  = evsel,
147                 .sample                 = sample,
148                 .hide_unresolved        = rep->hide_unresolved,
149                 .add_entry_cb           = hist_iter__report_callback,
150         };
151         int ret = 0;
152
153         if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
154                 pr_debug("problem processing %d event, skipping it.\n",
155                          event->header.type);
156                 return -1;
157         }
158
159         if (rep->hide_unresolved && al.sym == NULL)
160                 goto out_put;
161
162         if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
163                 goto out_put;
164
165         if (sort__mode == SORT_MODE__BRANCH)
166                 iter.ops = &hist_iter_branch;
167         else if (rep->mem_mode)
168                 iter.ops = &hist_iter_mem;
169         else if (symbol_conf.cumulate_callchain)
170                 iter.ops = &hist_iter_cumulative;
171         else
172                 iter.ops = &hist_iter_normal;
173
174         if (al.map != NULL)
175                 al.map->dso->hit = 1;
176
177         ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
178         if (ret < 0)
179                 pr_debug("problem adding hist entry, skipping event\n");
180 out_put:
181         addr_location__put(&al);
182         return ret;
183 }
184
185 static int process_read_event(struct perf_tool *tool,
186                               union perf_event *event,
187                               struct perf_sample *sample __maybe_unused,
188                               struct perf_evsel *evsel,
189                               struct machine *machine __maybe_unused)
190 {
191         struct report *rep = container_of(tool, struct report, tool);
192
193         if (rep->show_threads) {
194                 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
195                 perf_read_values_add_value(&rep->show_threads_values,
196                                            event->read.pid, event->read.tid,
197                                            event->read.id,
198                                            name,
199                                            event->read.value);
200         }
201
202         dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
203                     evsel ? perf_evsel__name(evsel) : "FAIL",
204                     event->read.value);
205
206         return 0;
207 }
208
209 /* For pipe mode, sample_type is not currently set */
210 static int report__setup_sample_type(struct report *rep)
211 {
212         struct perf_session *session = rep->session;
213         u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
214         bool is_pipe = perf_data_file__is_pipe(session->file);
215
216         if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
217                 if (sort__has_parent) {
218                         ui__error("Selected --sort parent, but no "
219                                     "callchain data. Did you call "
220                                     "'perf record' without -g?\n");
221                         return -EINVAL;
222                 }
223                 if (symbol_conf.use_callchain) {
224                         ui__error("Selected -g or --branch-history but no "
225                                   "callchain data. Did\n"
226                                   "you call 'perf record' without -g?\n");
227                         return -1;
228                 }
229         } else if (!rep->dont_use_callchains &&
230                    callchain_param.mode != CHAIN_NONE &&
231                    !symbol_conf.use_callchain) {
232                         symbol_conf.use_callchain = true;
233                         if (callchain_register_param(&callchain_param) < 0) {
234                                 ui__error("Can't register callchain params.\n");
235                                 return -EINVAL;
236                         }
237         }
238
239         if (symbol_conf.cumulate_callchain) {
240                 /* Silently ignore if callchain is missing */
241                 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
242                         symbol_conf.cumulate_callchain = false;
243                         perf_hpp__cancel_cumulate();
244                 }
245         }
246
247         if (sort__mode == SORT_MODE__BRANCH) {
248                 if (!is_pipe &&
249                     !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
250                         ui__error("Selected -b but no branch data. "
251                                   "Did you call perf record without -b?\n");
252                         return -1;
253                 }
254         }
255
256         if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
257                 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
258                     (sample_type & PERF_SAMPLE_STACK_USER))
259                         callchain_param.record_mode = CALLCHAIN_DWARF;
260                 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
261                         callchain_param.record_mode = CALLCHAIN_LBR;
262                 else
263                         callchain_param.record_mode = CALLCHAIN_FP;
264         }
265
266         /* ??? handle more cases than just ANY? */
267         if (!(perf_evlist__combined_branch_type(session->evlist) &
268                                 PERF_SAMPLE_BRANCH_ANY))
269                 rep->nonany_branch_mode = true;
270
271         return 0;
272 }
273
274 static void sig_handler(int sig __maybe_unused)
275 {
276         session_done = 1;
277 }
278
279 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
280                                               const char *evname, FILE *fp)
281 {
282         size_t ret;
283         char unit;
284         unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
285         u64 nr_events = hists->stats.total_period;
286         struct perf_evsel *evsel = hists_to_evsel(hists);
287         char buf[512];
288         size_t size = sizeof(buf);
289
290         if (symbol_conf.filter_relative) {
291                 nr_samples = hists->stats.nr_non_filtered_samples;
292                 nr_events = hists->stats.total_non_filtered_period;
293         }
294
295         if (perf_evsel__is_group_event(evsel)) {
296                 struct perf_evsel *pos;
297
298                 perf_evsel__group_desc(evsel, buf, size);
299                 evname = buf;
300
301                 for_each_group_member(pos, evsel) {
302                         const struct hists *pos_hists = evsel__hists(pos);
303
304                         if (symbol_conf.filter_relative) {
305                                 nr_samples += pos_hists->stats.nr_non_filtered_samples;
306                                 nr_events += pos_hists->stats.total_non_filtered_period;
307                         } else {
308                                 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
309                                 nr_events += pos_hists->stats.total_period;
310                         }
311                 }
312         }
313
314         nr_samples = convert_unit(nr_samples, &unit);
315         ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
316         if (evname != NULL)
317                 ret += fprintf(fp, " of event '%s'", evname);
318
319         if (rep->mem_mode) {
320                 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
321                 ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
322         } else
323                 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
324         return ret + fprintf(fp, "\n#\n");
325 }
326
327 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
328                                          struct report *rep,
329                                          const char *help)
330 {
331         struct perf_evsel *pos;
332
333         fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
334         evlist__for_each(evlist, pos) {
335                 struct hists *hists = evsel__hists(pos);
336                 const char *evname = perf_evsel__name(pos);
337
338                 if (symbol_conf.event_group &&
339                     !perf_evsel__is_group_leader(pos))
340                         continue;
341
342                 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
343                 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
344                 fprintf(stdout, "\n\n");
345         }
346
347         if (sort_order == NULL &&
348             parent_pattern == default_parent_pattern)
349                 fprintf(stdout, "#\n# (%s)\n#\n", help);
350
351         if (rep->show_threads) {
352                 bool style = !strcmp(rep->pretty_printing_style, "raw");
353                 perf_read_values_display(stdout, &rep->show_threads_values,
354                                          style);
355                 perf_read_values_destroy(&rep->show_threads_values);
356         }
357
358         return 0;
359 }
360
361 static void report__warn_kptr_restrict(const struct report *rep)
362 {
363         struct map *kernel_map = rep->session->machines.host.vmlinux_maps[MAP__FUNCTION];
364         struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
365
366         if (kernel_map == NULL ||
367             (kernel_map->dso->hit &&
368              (kernel_kmap->ref_reloc_sym == NULL ||
369               kernel_kmap->ref_reloc_sym->addr == 0))) {
370                 const char *desc =
371                     "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
372                     "can't be resolved.";
373
374                 if (kernel_map) {
375                         const struct dso *kdso = kernel_map->dso;
376                         if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
377                                 desc = "If some relocation was applied (e.g. "
378                                        "kexec) symbols may be misresolved.";
379                         }
380                 }
381
382                 ui__warning(
383 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
384 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
385 "Samples in kernel modules can't be resolved as well.\n\n",
386                 desc);
387         }
388 }
389
390 static int report__gtk_browse_hists(struct report *rep, const char *help)
391 {
392         int (*hist_browser)(struct perf_evlist *evlist, const char *help,
393                             struct hist_browser_timer *timer, float min_pcnt);
394
395         hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
396
397         if (hist_browser == NULL) {
398                 ui__error("GTK browser not found!\n");
399                 return -1;
400         }
401
402         return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
403 }
404
405 static int report__browse_hists(struct report *rep)
406 {
407         int ret;
408         struct perf_session *session = rep->session;
409         struct perf_evlist *evlist = session->evlist;
410         const char *help = "For a higher level overview, try: perf report --sort comm,dso";
411
412         switch (use_browser) {
413         case 1:
414                 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
415                                                     rep->min_percent,
416                                                     &session->header.env);
417                 /*
418                  * Usually "ret" is the last pressed key, and we only
419                  * care if the key notifies us to switch data file.
420                  */
421                 if (ret != K_SWITCH_INPUT_DATA)
422                         ret = 0;
423                 break;
424         case 2:
425                 ret = report__gtk_browse_hists(rep, help);
426                 break;
427         default:
428                 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
429                 break;
430         }
431
432         return ret;
433 }
434
435 static void report__collapse_hists(struct report *rep)
436 {
437         struct ui_progress prog;
438         struct perf_evsel *pos;
439
440         ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
441
442         evlist__for_each(rep->session->evlist, pos) {
443                 struct hists *hists = evsel__hists(pos);
444
445                 if (pos->idx == 0)
446                         hists->symbol_filter_str = rep->symbol_filter_str;
447
448                 hists__collapse_resort(hists, &prog);
449
450                 /* Non-group events are considered as leader */
451                 if (symbol_conf.event_group &&
452                     !perf_evsel__is_group_leader(pos)) {
453                         struct hists *leader_hists = evsel__hists(pos->leader);
454
455                         hists__match(leader_hists, hists);
456                         hists__link(leader_hists, hists);
457                 }
458         }
459
460         ui_progress__finish();
461 }
462
463 static void report__output_resort(struct report *rep)
464 {
465         struct ui_progress prog;
466         struct perf_evsel *pos;
467
468         ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
469
470         evlist__for_each(rep->session->evlist, pos)
471                 hists__output_resort(evsel__hists(pos), &prog);
472
473         ui_progress__finish();
474 }
475
476 static int __cmd_report(struct report *rep)
477 {
478         int ret;
479         struct perf_session *session = rep->session;
480         struct perf_evsel *pos;
481         struct perf_data_file *file = session->file;
482
483         signal(SIGINT, sig_handler);
484
485         if (rep->cpu_list) {
486                 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
487                                                rep->cpu_bitmap);
488                 if (ret)
489                         return ret;
490         }
491
492         if (rep->show_threads)
493                 perf_read_values_init(&rep->show_threads_values);
494
495         ret = report__setup_sample_type(rep);
496         if (ret)
497                 return ret;
498
499         ret = perf_session__process_events(session);
500         if (ret)
501                 return ret;
502
503         report__warn_kptr_restrict(rep);
504
505         evlist__for_each(session->evlist, pos)
506                 rep->nr_entries += evsel__hists(pos)->nr_entries;
507
508         if (use_browser == 0) {
509                 if (verbose > 3)
510                         perf_session__fprintf(session, stdout);
511
512                 if (verbose > 2)
513                         perf_session__fprintf_dsos(session, stdout);
514
515                 if (dump_trace) {
516                         perf_session__fprintf_nr_events(session, stdout);
517                         perf_evlist__fprintf_nr_events(session->evlist, stdout);
518                         return 0;
519                 }
520         }
521
522         report__collapse_hists(rep);
523
524         if (session_done())
525                 return 0;
526
527         /*
528          * recalculate number of entries after collapsing since it
529          * might be changed during the collapse phase.
530          */
531         rep->nr_entries = 0;
532         evlist__for_each(session->evlist, pos)
533                 rep->nr_entries += evsel__hists(pos)->nr_entries;
534
535         if (rep->nr_entries == 0) {
536                 ui__error("The %s file has no samples!\n", file->path);
537                 return 0;
538         }
539
540         report__output_resort(rep);
541
542         return report__browse_hists(rep);
543 }
544
545 static int
546 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
547 {
548         struct report *rep = (struct report *)opt->value;
549
550         /*
551          * --no-call-graph
552          */
553         if (unset) {
554                 rep->dont_use_callchains = true;
555                 return 0;
556         }
557
558         return parse_callchain_report_opt(arg);
559 }
560
561 int
562 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
563                                 const char *arg, int unset __maybe_unused)
564 {
565         if (arg) {
566                 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
567                 if (err) {
568                         char buf[BUFSIZ];
569                         regerror(err, &ignore_callees_regex, buf, sizeof(buf));
570                         pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
571                         return -1;
572                 }
573                 have_ignore_callees = 1;
574         }
575
576         return 0;
577 }
578
579 static int
580 parse_branch_mode(const struct option *opt __maybe_unused,
581                   const char *str __maybe_unused, int unset)
582 {
583         int *branch_mode = opt->value;
584
585         *branch_mode = !unset;
586         return 0;
587 }
588
589 static int
590 parse_percent_limit(const struct option *opt, const char *str,
591                     int unset __maybe_unused)
592 {
593         struct report *rep = opt->value;
594
595         rep->min_percent = strtof(str, NULL);
596         return 0;
597 }
598
599 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
600 {
601         struct perf_session *session;
602         struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
603         struct stat st;
604         bool has_br_stack = false;
605         int branch_mode = -1;
606         bool branch_call_mode = false;
607         char callchain_default_opt[] = "fractal,0.5,callee";
608         const char * const report_usage[] = {
609                 "perf report [<options>]",
610                 NULL
611         };
612         struct report report = {
613                 .tool = {
614                         .sample          = process_sample_event,
615                         .mmap            = perf_event__process_mmap,
616                         .mmap2           = perf_event__process_mmap2,
617                         .comm            = perf_event__process_comm,
618                         .exit            = perf_event__process_exit,
619                         .fork            = perf_event__process_fork,
620                         .lost            = perf_event__process_lost,
621                         .read            = process_read_event,
622                         .attr            = perf_event__process_attr,
623                         .tracing_data    = perf_event__process_tracing_data,
624                         .build_id        = perf_event__process_build_id,
625                         .id_index        = perf_event__process_id_index,
626                         .auxtrace_info   = perf_event__process_auxtrace_info,
627                         .auxtrace        = perf_event__process_auxtrace,
628                         .ordered_events  = true,
629                         .ordering_requires_timestamps = true,
630                 },
631                 .max_stack               = PERF_MAX_STACK_DEPTH,
632                 .pretty_printing_style   = "normal",
633         };
634         const struct option options[] = {
635         OPT_STRING('i', "input", &input_name, "file",
636                     "input file name"),
637         OPT_INCR('v', "verbose", &verbose,
638                     "be more verbose (show symbol address, etc)"),
639         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
640                     "dump raw trace in ASCII"),
641         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
642                    "file", "vmlinux pathname"),
643         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
644                    "file", "kallsyms pathname"),
645         OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
646         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
647                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
648         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
649                     "Show a column with the number of samples"),
650         OPT_BOOLEAN('T', "threads", &report.show_threads,
651                     "Show per-thread event counters"),
652         OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
653                    "pretty printing style key: normal raw"),
654         OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
655         OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
656         OPT_BOOLEAN(0, "stdio", &report.use_stdio,
657                     "Use the stdio interface"),
658         OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
659         OPT_BOOLEAN(0, "header-only", &report.header_only,
660                     "Show only data header."),
661         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
662                    "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
663                    " Please refer the man page for the complete list."),
664         OPT_STRING('F', "fields", &field_order, "key[,keys...]",
665                    "output field(s): overhead, period, sample plus all of sort keys"),
666         OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
667                     "Show sample percentage for different cpu modes"),
668         OPT_STRING('p', "parent", &parent_pattern, "regex",
669                    "regex filter to identify parent, see: '--sort parent'"),
670         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
671                     "Only display entries with parent-match"),
672         OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order[,branch]",
673                      "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address), add branches. "
674                      "Default: fractal,0.5,callee,function", &report_parse_callchain_opt, callchain_default_opt),
675         OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
676                     "Accumulate callchains of children and show total overhead as well"),
677         OPT_INTEGER(0, "max-stack", &report.max_stack,
678                     "Set the maximum stack depth when parsing the callchain, "
679                     "anything beyond the specified depth will be ignored. "
680                     "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
681         OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
682                     "alias for inverted call graph"),
683         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
684                    "ignore callees of these functions in call graphs",
685                    report_parse_ignore_callees_opt),
686         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
687                    "only consider symbols in these dsos"),
688         OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
689                    "only consider symbols in these comms"),
690         OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
691                    "only consider symbols in these pids"),
692         OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
693                    "only consider symbols in these tids"),
694         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
695                    "only consider these symbols"),
696         OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
697                    "only show symbols that (partially) match with this filter"),
698         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
699                    "width[,width...]",
700                    "don't try to adjust column width, use these fixed values"),
701         OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
702                    "separator for columns, no spaces will be added between "
703                    "columns '.' is reserved."),
704         OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
705                     "Only display entries resolved to a symbol"),
706         OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
707                     "Look for files with symbols relative to this directory"),
708         OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
709                    "list of cpus to profile"),
710         OPT_BOOLEAN('I', "show-info", &report.show_full_info,
711                     "Display extended information about perf.data file"),
712         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
713                     "Interleave source code with assembly code (default)"),
714         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
715                     "Display raw encoding of assembly instructions (default)"),
716         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
717                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
718         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
719                     "Show a column with the sum of periods"),
720         OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
721                     "Show event group information together"),
722         OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
723                     "use branch records for per branch histogram filling",
724                     parse_branch_mode),
725         OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
726                     "add last branch records to call history"),
727         OPT_STRING(0, "objdump", &objdump_path, "path",
728                    "objdump binary to use for disassembly and annotations"),
729         OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
730                     "Disable symbol demangling"),
731         OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
732                     "Enable kernel symbol demangling"),
733         OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
734         OPT_CALLBACK(0, "percent-limit", &report, "percent",
735                      "Don't show entries under that percent", parse_percent_limit),
736         OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
737                      "how to display percentage of filtered entries", parse_filter_percentage),
738         OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
739                             "Instruction Tracing options",
740                             itrace_parse_synth_opts),
741         OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
742                         "Show full source file name path for source lines"),
743         OPT_END()
744         };
745         struct perf_data_file file = {
746                 .mode  = PERF_DATA_MODE_READ,
747         };
748         int ret = hists__init();
749
750         if (ret < 0)
751                 return ret;
752
753         perf_config(report__config, &report);
754
755         argc = parse_options(argc, argv, options, report_usage, 0);
756
757         if (symbol_conf.vmlinux_name &&
758             access(symbol_conf.vmlinux_name, R_OK)) {
759                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
760                 return -EINVAL;
761         }
762         if (symbol_conf.kallsyms_name &&
763             access(symbol_conf.kallsyms_name, R_OK)) {
764                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
765                 return -EINVAL;
766         }
767
768         if (report.use_stdio)
769                 use_browser = 0;
770         else if (report.use_tui)
771                 use_browser = 1;
772         else if (report.use_gtk)
773                 use_browser = 2;
774
775         if (report.inverted_callchain)
776                 callchain_param.order = ORDER_CALLER;
777
778         if (!input_name || !strlen(input_name)) {
779                 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
780                         input_name = "-";
781                 else
782                         input_name = "perf.data";
783         }
784
785         file.path  = input_name;
786         file.force = report.force;
787
788 repeat:
789         session = perf_session__new(&file, false, &report.tool);
790         if (session == NULL)
791                 return -1;
792
793         if (report.queue_size) {
794                 ordered_events__set_alloc_size(&session->ordered_events,
795                                                report.queue_size);
796         }
797
798         session->itrace_synth_opts = &itrace_synth_opts;
799
800         report.session = session;
801
802         has_br_stack = perf_header__has_feat(&session->header,
803                                              HEADER_BRANCH_STACK);
804
805         /*
806          * Branch mode is a tristate:
807          * -1 means default, so decide based on the file having branch data.
808          * 0/1 means the user chose a mode.
809          */
810         if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
811             !branch_call_mode) {
812                 sort__mode = SORT_MODE__BRANCH;
813                 symbol_conf.cumulate_callchain = false;
814         }
815         if (branch_call_mode) {
816                 callchain_param.key = CCKEY_ADDRESS;
817                 callchain_param.branch_callstack = 1;
818                 symbol_conf.use_callchain = true;
819                 callchain_register_param(&callchain_param);
820                 if (sort_order == NULL)
821                         sort_order = "srcline,symbol,dso";
822         }
823
824         if (report.mem_mode) {
825                 if (sort__mode == SORT_MODE__BRANCH) {
826                         pr_err("branch and mem mode incompatible\n");
827                         goto error;
828                 }
829                 sort__mode = SORT_MODE__MEMORY;
830                 symbol_conf.cumulate_callchain = false;
831         }
832
833         if (setup_sorting() < 0) {
834                 if (sort_order)
835                         parse_options_usage(report_usage, options, "s", 1);
836                 if (field_order)
837                         parse_options_usage(sort_order ? NULL : report_usage,
838                                             options, "F", 1);
839                 goto error;
840         }
841
842         /* Force tty output for header output and per-thread stat. */
843         if (report.header || report.header_only || report.show_threads)
844                 use_browser = 0;
845
846         if (strcmp(input_name, "-") != 0)
847                 setup_browser(true);
848         else
849                 use_browser = 0;
850
851         if (report.header || report.header_only) {
852                 perf_session__fprintf_info(session, stdout,
853                                            report.show_full_info);
854                 if (report.header_only) {
855                         ret = 0;
856                         goto error;
857                 }
858         } else if (use_browser == 0) {
859                 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
860                       stdout);
861         }
862
863         /*
864          * Only in the TUI browser we are doing integrated annotation,
865          * so don't allocate extra space that won't be used in the stdio
866          * implementation.
867          */
868         if (ui__has_annotation()) {
869                 symbol_conf.priv_size = sizeof(struct annotation);
870                 machines__set_symbol_filter(&session->machines,
871                                             symbol__annotate_init);
872                 /*
873                  * For searching by name on the "Browse map details".
874                  * providing it only in verbose mode not to bloat too
875                  * much struct symbol.
876                  */
877                 if (verbose) {
878                         /*
879                          * XXX: Need to provide a less kludgy way to ask for
880                          * more space per symbol, the u32 is for the index on
881                          * the ui browser.
882                          * See symbol__browser_index.
883                          */
884                         symbol_conf.priv_size += sizeof(u32);
885                         symbol_conf.sort_by_name = true;
886                 }
887         }
888
889         if (symbol__init(&session->header.env) < 0)
890                 goto error;
891
892         if (argc) {
893                 /*
894                  * Special case: if there's an argument left then assume that
895                  * it's a symbol filter:
896                  */
897                 if (argc > 1)
898                         usage_with_options(report_usage, options);
899
900                 report.symbol_filter_str = argv[0];
901         }
902
903         sort__setup_elide(stdout);
904
905         ret = __cmd_report(&report);
906         if (ret == K_SWITCH_INPUT_DATA) {
907                 perf_session__delete(session);
908                 goto repeat;
909         } else
910                 ret = 0;
911
912 error:
913         perf_session__delete(session);
914         return ret;
915 }