Merge tag 'arc-4.5-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/vgupta/arc
[cascardo/linux.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 <subcmd/parse-options.h>
31 #include <subcmd/exec-cmd.h>
32 #include "util/parse-events.h"
33
34 #include "util/thread.h"
35 #include "util/sort.h"
36 #include "util/hist.h"
37 #include "util/data.h"
38 #include "arch/common.h"
39
40 #include "util/auxtrace.h"
41
42 #include <dlfcn.h>
43 #include <linux/bitmap.h>
44
45 struct report {
46         struct perf_tool        tool;
47         struct perf_session     *session;
48         bool                    use_tui, use_gtk, use_stdio;
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         int                     socket_filter;
66         DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
67 };
68
69 static int report__config(const char *var, const char *value, void *cb)
70 {
71         struct report *rep = cb;
72
73         if (!strcmp(var, "report.group")) {
74                 symbol_conf.event_group = perf_config_bool(var, value);
75                 return 0;
76         }
77         if (!strcmp(var, "report.percent-limit")) {
78                 rep->min_percent = strtof(value, NULL);
79                 return 0;
80         }
81         if (!strcmp(var, "report.children")) {
82                 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
83                 return 0;
84         }
85         if (!strcmp(var, "report.queue-size")) {
86                 rep->queue_size = perf_config_u64(var, value);
87                 return 0;
88         }
89
90         return perf_default_config(var, value, cb);
91 }
92
93 static int hist_iter__report_callback(struct hist_entry_iter *iter,
94                                       struct addr_location *al, bool single,
95                                       void *arg)
96 {
97         int err = 0;
98         struct report *rep = arg;
99         struct hist_entry *he = iter->he;
100         struct perf_evsel *evsel = iter->evsel;
101         struct mem_info *mi;
102         struct branch_info *bi;
103
104         if (!ui__has_annotation())
105                 return 0;
106
107         hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
108                              rep->nonany_branch_mode);
109
110         if (sort__mode == SORT_MODE__BRANCH) {
111                 bi = he->branch_info;
112                 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
113                 if (err)
114                         goto out;
115
116                 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
117
118         } else if (rep->mem_mode) {
119                 mi = he->mem_info;
120                 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
121                 if (err)
122                         goto out;
123
124                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
125
126         } else if (symbol_conf.cumulate_callchain) {
127                 if (single)
128                         err = hist_entry__inc_addr_samples(he, evsel->idx,
129                                                            al->addr);
130         } else {
131                 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
132         }
133
134 out:
135         return err;
136 }
137
138 static int process_sample_event(struct perf_tool *tool,
139                                 union perf_event *event,
140                                 struct perf_sample *sample,
141                                 struct perf_evsel *evsel,
142                                 struct machine *machine)
143 {
144         struct report *rep = container_of(tool, struct report, tool);
145         struct addr_location al;
146         struct hist_entry_iter iter = {
147                 .evsel                  = evsel,
148                 .sample                 = sample,
149                 .hide_unresolved        = symbol_conf.hide_unresolved,
150                 .add_entry_cb           = hist_iter__report_callback,
151         };
152         int ret = 0;
153
154         if (perf_event__preprocess_sample(event, machine, &al, sample) < 0) {
155                 pr_debug("problem processing %d event, skipping it.\n",
156                          event->header.type);
157                 return -1;
158         }
159
160         if (symbol_conf.hide_unresolved && al.sym == NULL)
161                 goto out_put;
162
163         if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
164                 goto out_put;
165
166         if (sort__mode == SORT_MODE__BRANCH) {
167                 /*
168                  * A non-synthesized event might not have a branch stack if
169                  * branch stacks have been synthesized (using itrace options).
170                  */
171                 if (!sample->branch_stack)
172                         goto out_put;
173                 iter.ops = &hist_iter_branch;
174         } else if (rep->mem_mode) {
175                 iter.ops = &hist_iter_mem;
176         } else if (symbol_conf.cumulate_callchain) {
177                 iter.ops = &hist_iter_cumulative;
178         } else {
179                 iter.ops = &hist_iter_normal;
180         }
181
182         if (al.map != NULL)
183                 al.map->dso->hit = 1;
184
185         ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
186         if (ret < 0)
187                 pr_debug("problem adding hist entry, skipping event\n");
188 out_put:
189         addr_location__put(&al);
190         return ret;
191 }
192
193 static int process_read_event(struct perf_tool *tool,
194                               union perf_event *event,
195                               struct perf_sample *sample __maybe_unused,
196                               struct perf_evsel *evsel,
197                               struct machine *machine __maybe_unused)
198 {
199         struct report *rep = container_of(tool, struct report, tool);
200
201         if (rep->show_threads) {
202                 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
203                 perf_read_values_add_value(&rep->show_threads_values,
204                                            event->read.pid, event->read.tid,
205                                            event->read.id,
206                                            name,
207                                            event->read.value);
208         }
209
210         dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
211                     evsel ? perf_evsel__name(evsel) : "FAIL",
212                     event->read.value);
213
214         return 0;
215 }
216
217 /* For pipe mode, sample_type is not currently set */
218 static int report__setup_sample_type(struct report *rep)
219 {
220         struct perf_session *session = rep->session;
221         u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
222         bool is_pipe = perf_data_file__is_pipe(session->file);
223
224         if (session->itrace_synth_opts->callchain ||
225             (!is_pipe &&
226              perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
227              !session->itrace_synth_opts->set))
228                 sample_type |= PERF_SAMPLE_CALLCHAIN;
229
230         if (session->itrace_synth_opts->last_branch)
231                 sample_type |= PERF_SAMPLE_BRANCH_STACK;
232
233         if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
234                 if (sort__has_parent) {
235                         ui__error("Selected --sort parent, but no "
236                                     "callchain data. Did you call "
237                                     "'perf record' without -g?\n");
238                         return -EINVAL;
239                 }
240                 if (symbol_conf.use_callchain) {
241                         ui__error("Selected -g or --branch-history but no "
242                                   "callchain data. Did\n"
243                                   "you call 'perf record' without -g?\n");
244                         return -1;
245                 }
246         } else if (!rep->dont_use_callchains &&
247                    callchain_param.mode != CHAIN_NONE &&
248                    !symbol_conf.use_callchain) {
249                         symbol_conf.use_callchain = true;
250                         if (callchain_register_param(&callchain_param) < 0) {
251                                 ui__error("Can't register callchain params.\n");
252                                 return -EINVAL;
253                         }
254         }
255
256         if (symbol_conf.cumulate_callchain) {
257                 /* Silently ignore if callchain is missing */
258                 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
259                         symbol_conf.cumulate_callchain = false;
260                         perf_hpp__cancel_cumulate();
261                 }
262         }
263
264         if (sort__mode == SORT_MODE__BRANCH) {
265                 if (!is_pipe &&
266                     !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
267                         ui__error("Selected -b but no branch data. "
268                                   "Did you call perf record without -b?\n");
269                         return -1;
270                 }
271         }
272
273         if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
274                 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
275                     (sample_type & PERF_SAMPLE_STACK_USER))
276                         callchain_param.record_mode = CALLCHAIN_DWARF;
277                 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
278                         callchain_param.record_mode = CALLCHAIN_LBR;
279                 else
280                         callchain_param.record_mode = CALLCHAIN_FP;
281         }
282
283         /* ??? handle more cases than just ANY? */
284         if (!(perf_evlist__combined_branch_type(session->evlist) &
285                                 PERF_SAMPLE_BRANCH_ANY))
286                 rep->nonany_branch_mode = true;
287
288         return 0;
289 }
290
291 static void sig_handler(int sig __maybe_unused)
292 {
293         session_done = 1;
294 }
295
296 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
297                                               const char *evname, FILE *fp)
298 {
299         size_t ret;
300         char unit;
301         unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
302         u64 nr_events = hists->stats.total_period;
303         struct perf_evsel *evsel = hists_to_evsel(hists);
304         char buf[512];
305         size_t size = sizeof(buf);
306         int socked_id = hists->socket_filter;
307
308         if (symbol_conf.filter_relative) {
309                 nr_samples = hists->stats.nr_non_filtered_samples;
310                 nr_events = hists->stats.total_non_filtered_period;
311         }
312
313         if (perf_evsel__is_group_event(evsel)) {
314                 struct perf_evsel *pos;
315
316                 perf_evsel__group_desc(evsel, buf, size);
317                 evname = buf;
318
319                 for_each_group_member(pos, evsel) {
320                         const struct hists *pos_hists = evsel__hists(pos);
321
322                         if (symbol_conf.filter_relative) {
323                                 nr_samples += pos_hists->stats.nr_non_filtered_samples;
324                                 nr_events += pos_hists->stats.total_non_filtered_period;
325                         } else {
326                                 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
327                                 nr_events += pos_hists->stats.total_period;
328                         }
329                 }
330         }
331
332         nr_samples = convert_unit(nr_samples, &unit);
333         ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
334         if (evname != NULL)
335                 ret += fprintf(fp, " of event '%s'", evname);
336
337         if (symbol_conf.show_ref_callgraph &&
338             strstr(evname, "call-graph=no")) {
339                 ret += fprintf(fp, ", show reference callgraph");
340         }
341
342         if (rep->mem_mode) {
343                 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
344                 ret += fprintf(fp, "\n# Sort order   : %s", sort_order ? : default_mem_sort_order);
345         } else
346                 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
347
348         if (socked_id > -1)
349                 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
350
351         return ret + fprintf(fp, "\n#\n");
352 }
353
354 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
355                                          struct report *rep,
356                                          const char *help)
357 {
358         struct perf_evsel *pos;
359
360         fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
361         evlist__for_each(evlist, pos) {
362                 struct hists *hists = evsel__hists(pos);
363                 const char *evname = perf_evsel__name(pos);
364
365                 if (symbol_conf.event_group &&
366                     !perf_evsel__is_group_leader(pos))
367                         continue;
368
369                 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
370                 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
371                 fprintf(stdout, "\n\n");
372         }
373
374         if (sort_order == NULL &&
375             parent_pattern == default_parent_pattern)
376                 fprintf(stdout, "#\n# (%s)\n#\n", help);
377
378         if (rep->show_threads) {
379                 bool style = !strcmp(rep->pretty_printing_style, "raw");
380                 perf_read_values_display(stdout, &rep->show_threads_values,
381                                          style);
382                 perf_read_values_destroy(&rep->show_threads_values);
383         }
384
385         return 0;
386 }
387
388 static void report__warn_kptr_restrict(const struct report *rep)
389 {
390         struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
391         struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
392
393         if (kernel_map == NULL ||
394             (kernel_map->dso->hit &&
395              (kernel_kmap->ref_reloc_sym == NULL ||
396               kernel_kmap->ref_reloc_sym->addr == 0))) {
397                 const char *desc =
398                     "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
399                     "can't be resolved.";
400
401                 if (kernel_map) {
402                         const struct dso *kdso = kernel_map->dso;
403                         if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
404                                 desc = "If some relocation was applied (e.g. "
405                                        "kexec) symbols may be misresolved.";
406                         }
407                 }
408
409                 ui__warning(
410 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
411 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
412 "Samples in kernel modules can't be resolved as well.\n\n",
413                 desc);
414         }
415 }
416
417 static int report__gtk_browse_hists(struct report *rep, const char *help)
418 {
419         int (*hist_browser)(struct perf_evlist *evlist, const char *help,
420                             struct hist_browser_timer *timer, float min_pcnt);
421
422         hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
423
424         if (hist_browser == NULL) {
425                 ui__error("GTK browser not found!\n");
426                 return -1;
427         }
428
429         return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
430 }
431
432 static int report__browse_hists(struct report *rep)
433 {
434         int ret;
435         struct perf_session *session = rep->session;
436         struct perf_evlist *evlist = session->evlist;
437         const char *help = perf_tip(system_path(TIPDIR));
438
439         if (help == NULL) {
440                 /* fallback for people who don't install perf ;-) */
441                 help = perf_tip(DOCDIR);
442                 if (help == NULL)
443                         help = "Cannot load tips.txt file, please install perf!";
444         }
445
446         switch (use_browser) {
447         case 1:
448                 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
449                                                     rep->min_percent,
450                                                     &session->header.env);
451                 /*
452                  * Usually "ret" is the last pressed key, and we only
453                  * care if the key notifies us to switch data file.
454                  */
455                 if (ret != K_SWITCH_INPUT_DATA)
456                         ret = 0;
457                 break;
458         case 2:
459                 ret = report__gtk_browse_hists(rep, help);
460                 break;
461         default:
462                 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
463                 break;
464         }
465
466         return ret;
467 }
468
469 static void report__collapse_hists(struct report *rep)
470 {
471         struct ui_progress prog;
472         struct perf_evsel *pos;
473
474         ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
475
476         evlist__for_each(rep->session->evlist, pos) {
477                 struct hists *hists = evsel__hists(pos);
478
479                 if (pos->idx == 0)
480                         hists->symbol_filter_str = rep->symbol_filter_str;
481
482                 hists->socket_filter = rep->socket_filter;
483
484                 hists__collapse_resort(hists, &prog);
485
486                 /* Non-group events are considered as leader */
487                 if (symbol_conf.event_group &&
488                     !perf_evsel__is_group_leader(pos)) {
489                         struct hists *leader_hists = evsel__hists(pos->leader);
490
491                         hists__match(leader_hists, hists);
492                         hists__link(leader_hists, hists);
493                 }
494         }
495
496         ui_progress__finish();
497 }
498
499 static void report__output_resort(struct report *rep)
500 {
501         struct ui_progress prog;
502         struct perf_evsel *pos;
503
504         ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
505
506         evlist__for_each(rep->session->evlist, pos)
507                 hists__output_resort(evsel__hists(pos), &prog);
508
509         ui_progress__finish();
510 }
511
512 static int __cmd_report(struct report *rep)
513 {
514         int ret;
515         struct perf_session *session = rep->session;
516         struct perf_evsel *pos;
517         struct perf_data_file *file = session->file;
518
519         signal(SIGINT, sig_handler);
520
521         if (rep->cpu_list) {
522                 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
523                                                rep->cpu_bitmap);
524                 if (ret) {
525                         ui__error("failed to set cpu bitmap\n");
526                         return ret;
527                 }
528         }
529
530         if (rep->show_threads)
531                 perf_read_values_init(&rep->show_threads_values);
532
533         ret = report__setup_sample_type(rep);
534         if (ret) {
535                 /* report__setup_sample_type() already showed error message */
536                 return ret;
537         }
538
539         ret = perf_session__process_events(session);
540         if (ret) {
541                 ui__error("failed to process sample\n");
542                 return ret;
543         }
544
545         report__warn_kptr_restrict(rep);
546
547         evlist__for_each(session->evlist, pos)
548                 rep->nr_entries += evsel__hists(pos)->nr_entries;
549
550         if (use_browser == 0) {
551                 if (verbose > 3)
552                         perf_session__fprintf(session, stdout);
553
554                 if (verbose > 2)
555                         perf_session__fprintf_dsos(session, stdout);
556
557                 if (dump_trace) {
558                         perf_session__fprintf_nr_events(session, stdout);
559                         perf_evlist__fprintf_nr_events(session->evlist, stdout);
560                         return 0;
561                 }
562         }
563
564         report__collapse_hists(rep);
565
566         if (session_done())
567                 return 0;
568
569         /*
570          * recalculate number of entries after collapsing since it
571          * might be changed during the collapse phase.
572          */
573         rep->nr_entries = 0;
574         evlist__for_each(session->evlist, pos)
575                 rep->nr_entries += evsel__hists(pos)->nr_entries;
576
577         if (rep->nr_entries == 0) {
578                 ui__error("The %s file has no samples!\n", file->path);
579                 return 0;
580         }
581
582         report__output_resort(rep);
583
584         return report__browse_hists(rep);
585 }
586
587 static int
588 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
589 {
590         struct report *rep = (struct report *)opt->value;
591
592         /*
593          * --no-call-graph
594          */
595         if (unset) {
596                 rep->dont_use_callchains = true;
597                 return 0;
598         }
599
600         return parse_callchain_report_opt(arg);
601 }
602
603 int
604 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
605                                 const char *arg, int unset __maybe_unused)
606 {
607         if (arg) {
608                 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
609                 if (err) {
610                         char buf[BUFSIZ];
611                         regerror(err, &ignore_callees_regex, buf, sizeof(buf));
612                         pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
613                         return -1;
614                 }
615                 have_ignore_callees = 1;
616         }
617
618         return 0;
619 }
620
621 static int
622 parse_branch_mode(const struct option *opt __maybe_unused,
623                   const char *str __maybe_unused, int unset)
624 {
625         int *branch_mode = opt->value;
626
627         *branch_mode = !unset;
628         return 0;
629 }
630
631 static int
632 parse_percent_limit(const struct option *opt, const char *str,
633                     int unset __maybe_unused)
634 {
635         struct report *rep = opt->value;
636
637         rep->min_percent = strtof(str, NULL);
638         return 0;
639 }
640
641 #define CALLCHAIN_DEFAULT_OPT  "graph,0.5,caller,function,percent"
642
643 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
644                                      CALLCHAIN_REPORT_HELP
645                                      "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
646
647 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
648 {
649         struct perf_session *session;
650         struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
651         struct stat st;
652         bool has_br_stack = false;
653         int branch_mode = -1;
654         bool branch_call_mode = false;
655         char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
656         const char * const report_usage[] = {
657                 "perf report [<options>]",
658                 NULL
659         };
660         struct report report = {
661                 .tool = {
662                         .sample          = process_sample_event,
663                         .mmap            = perf_event__process_mmap,
664                         .mmap2           = perf_event__process_mmap2,
665                         .comm            = perf_event__process_comm,
666                         .exit            = perf_event__process_exit,
667                         .fork            = perf_event__process_fork,
668                         .lost            = perf_event__process_lost,
669                         .read            = process_read_event,
670                         .attr            = perf_event__process_attr,
671                         .tracing_data    = perf_event__process_tracing_data,
672                         .build_id        = perf_event__process_build_id,
673                         .id_index        = perf_event__process_id_index,
674                         .auxtrace_info   = perf_event__process_auxtrace_info,
675                         .auxtrace        = perf_event__process_auxtrace,
676                         .ordered_events  = true,
677                         .ordering_requires_timestamps = true,
678                 },
679                 .max_stack               = PERF_MAX_STACK_DEPTH,
680                 .pretty_printing_style   = "normal",
681                 .socket_filter           = -1,
682         };
683         const struct option options[] = {
684         OPT_STRING('i', "input", &input_name, "file",
685                     "input file name"),
686         OPT_INCR('v', "verbose", &verbose,
687                     "be more verbose (show symbol address, etc)"),
688         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
689                     "dump raw trace in ASCII"),
690         OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
691                    "file", "vmlinux pathname"),
692         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
693                    "file", "kallsyms pathname"),
694         OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
695         OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
696                     "load module symbols - WARNING: use only with -k and LIVE kernel"),
697         OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
698                     "Show a column with the number of samples"),
699         OPT_BOOLEAN('T', "threads", &report.show_threads,
700                     "Show per-thread event counters"),
701         OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
702                    "pretty printing style key: normal raw"),
703         OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
704         OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
705         OPT_BOOLEAN(0, "stdio", &report.use_stdio,
706                     "Use the stdio interface"),
707         OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
708         OPT_BOOLEAN(0, "header-only", &report.header_only,
709                     "Show only data header."),
710         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
711                    "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
712                    " Please refer the man page for the complete list."),
713         OPT_STRING('F', "fields", &field_order, "key[,keys...]",
714                    "output field(s): overhead, period, sample plus all of sort keys"),
715         OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
716                     "Show sample percentage for different cpu modes"),
717         OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
718                     "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
719         OPT_STRING('p', "parent", &parent_pattern, "regex",
720                    "regex filter to identify parent, see: '--sort parent'"),
721         OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
722                     "Only display entries with parent-match"),
723         OPT_CALLBACK_DEFAULT('g', "call-graph", &report,
724                              "print_type,threshold[,print_limit],order,sort_key[,branch],value",
725                              report_callchain_help, &report_parse_callchain_opt,
726                              callchain_default_opt),
727         OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
728                     "Accumulate callchains of children and show total overhead as well"),
729         OPT_INTEGER(0, "max-stack", &report.max_stack,
730                     "Set the maximum stack depth when parsing the callchain, "
731                     "anything beyond the specified depth will be ignored. "
732                     "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
733         OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
734                     "alias for inverted call graph"),
735         OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
736                    "ignore callees of these functions in call graphs",
737                    report_parse_ignore_callees_opt),
738         OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
739                    "only consider symbols in these dsos"),
740         OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
741                    "only consider symbols in these comms"),
742         OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
743                    "only consider symbols in these pids"),
744         OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
745                    "only consider symbols in these tids"),
746         OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
747                    "only consider these symbols"),
748         OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
749                    "only show symbols that (partially) match with this filter"),
750         OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
751                    "width[,width...]",
752                    "don't try to adjust column width, use these fixed values"),
753         OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
754                    "separator for columns, no spaces will be added between "
755                    "columns '.' is reserved."),
756         OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
757                     "Only display entries resolved to a symbol"),
758         OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
759                     "Look for files with symbols relative to this directory"),
760         OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
761                    "list of cpus to profile"),
762         OPT_BOOLEAN('I', "show-info", &report.show_full_info,
763                     "Display extended information about perf.data file"),
764         OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
765                     "Interleave source code with assembly code (default)"),
766         OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
767                     "Display raw encoding of assembly instructions (default)"),
768         OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
769                    "Specify disassembler style (e.g. -M intel for intel syntax)"),
770         OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
771                     "Show a column with the sum of periods"),
772         OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
773                     "Show event group information together"),
774         OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
775                     "use branch records for per branch histogram filling",
776                     parse_branch_mode),
777         OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
778                     "add last branch records to call history"),
779         OPT_STRING(0, "objdump", &objdump_path, "path",
780                    "objdump binary to use for disassembly and annotations"),
781         OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
782                     "Disable symbol demangling"),
783         OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
784                     "Enable kernel symbol demangling"),
785         OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
786         OPT_CALLBACK(0, "percent-limit", &report, "percent",
787                      "Don't show entries under that percent", parse_percent_limit),
788         OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
789                      "how to display percentage of filtered entries", parse_filter_percentage),
790         OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
791                             "Instruction Tracing options",
792                             itrace_parse_synth_opts),
793         OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
794                         "Show full source file name path for source lines"),
795         OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
796                     "Show callgraph from reference event"),
797         OPT_INTEGER(0, "socket-filter", &report.socket_filter,
798                     "only show processor socket that match with this filter"),
799         OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
800                     "Show raw trace event output (do not use print fmt or plugins)"),
801         OPT_END()
802         };
803         struct perf_data_file file = {
804                 .mode  = PERF_DATA_MODE_READ,
805         };
806         int ret = hists__init();
807
808         if (ret < 0)
809                 return ret;
810
811         perf_config(report__config, &report);
812
813         argc = parse_options(argc, argv, options, report_usage, 0);
814         if (argc) {
815                 /*
816                  * Special case: if there's an argument left then assume that
817                  * it's a symbol filter:
818                  */
819                 if (argc > 1)
820                         usage_with_options(report_usage, options);
821
822                 report.symbol_filter_str = argv[0];
823         }
824
825         if (symbol_conf.vmlinux_name &&
826             access(symbol_conf.vmlinux_name, R_OK)) {
827                 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
828                 return -EINVAL;
829         }
830         if (symbol_conf.kallsyms_name &&
831             access(symbol_conf.kallsyms_name, R_OK)) {
832                 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
833                 return -EINVAL;
834         }
835
836         if (report.use_stdio)
837                 use_browser = 0;
838         else if (report.use_tui)
839                 use_browser = 1;
840         else if (report.use_gtk)
841                 use_browser = 2;
842
843         if (report.inverted_callchain)
844                 callchain_param.order = ORDER_CALLER;
845         if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
846                 callchain_param.order = ORDER_CALLER;
847
848         if (itrace_synth_opts.callchain &&
849             (int)itrace_synth_opts.callchain_sz > report.max_stack)
850                 report.max_stack = itrace_synth_opts.callchain_sz;
851
852         if (!input_name || !strlen(input_name)) {
853                 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
854                         input_name = "-";
855                 else
856                         input_name = "perf.data";
857         }
858
859         file.path  = input_name;
860         file.force = symbol_conf.force;
861
862 repeat:
863         session = perf_session__new(&file, false, &report.tool);
864         if (session == NULL)
865                 return -1;
866
867         if (report.queue_size) {
868                 ordered_events__set_alloc_size(&session->ordered_events,
869                                                report.queue_size);
870         }
871
872         session->itrace_synth_opts = &itrace_synth_opts;
873
874         report.session = session;
875
876         has_br_stack = perf_header__has_feat(&session->header,
877                                              HEADER_BRANCH_STACK);
878
879         if (itrace_synth_opts.last_branch)
880                 has_br_stack = true;
881
882         /*
883          * Branch mode is a tristate:
884          * -1 means default, so decide based on the file having branch data.
885          * 0/1 means the user chose a mode.
886          */
887         if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
888             !branch_call_mode) {
889                 sort__mode = SORT_MODE__BRANCH;
890                 symbol_conf.cumulate_callchain = false;
891         }
892         if (branch_call_mode) {
893                 callchain_param.key = CCKEY_ADDRESS;
894                 callchain_param.branch_callstack = 1;
895                 symbol_conf.use_callchain = true;
896                 callchain_register_param(&callchain_param);
897                 if (sort_order == NULL)
898                         sort_order = "srcline,symbol,dso";
899         }
900
901         if (report.mem_mode) {
902                 if (sort__mode == SORT_MODE__BRANCH) {
903                         pr_err("branch and mem mode incompatible\n");
904                         goto error;
905                 }
906                 sort__mode = SORT_MODE__MEMORY;
907                 symbol_conf.cumulate_callchain = false;
908         }
909
910         if (setup_sorting(session->evlist) < 0) {
911                 if (sort_order)
912                         parse_options_usage(report_usage, options, "s", 1);
913                 if (field_order)
914                         parse_options_usage(sort_order ? NULL : report_usage,
915                                             options, "F", 1);
916                 goto error;
917         }
918
919         /* Force tty output for header output and per-thread stat. */
920         if (report.header || report.header_only || report.show_threads)
921                 use_browser = 0;
922
923         if (strcmp(input_name, "-") != 0)
924                 setup_browser(true);
925         else
926                 use_browser = 0;
927
928         if (report.header || report.header_only) {
929                 perf_session__fprintf_info(session, stdout,
930                                            report.show_full_info);
931                 if (report.header_only) {
932                         ret = 0;
933                         goto error;
934                 }
935         } else if (use_browser == 0) {
936                 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
937                       stdout);
938         }
939
940         /*
941          * Only in the TUI browser we are doing integrated annotation,
942          * so don't allocate extra space that won't be used in the stdio
943          * implementation.
944          */
945         if (ui__has_annotation()) {
946                 symbol_conf.priv_size = sizeof(struct annotation);
947                 machines__set_symbol_filter(&session->machines,
948                                             symbol__annotate_init);
949                 /*
950                  * For searching by name on the "Browse map details".
951                  * providing it only in verbose mode not to bloat too
952                  * much struct symbol.
953                  */
954                 if (verbose) {
955                         /*
956                          * XXX: Need to provide a less kludgy way to ask for
957                          * more space per symbol, the u32 is for the index on
958                          * the ui browser.
959                          * See symbol__browser_index.
960                          */
961                         symbol_conf.priv_size += sizeof(u32);
962                         symbol_conf.sort_by_name = true;
963                 }
964         }
965
966         if (symbol__init(&session->header.env) < 0)
967                 goto error;
968
969         sort__setup_elide(stdout);
970
971         ret = __cmd_report(&report);
972         if (ret == K_SWITCH_INPUT_DATA) {
973                 perf_session__delete(session);
974                 goto repeat;
975         } else
976                 ret = 0;
977
978 error:
979         perf_session__delete(session);
980         return ret;
981 }