]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - tools/perf/builtin-report.c
Merge branch 'perf-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[mirror_ubuntu-zesty-kernel.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 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 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 = rep->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 (rep->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 = "For a higher level overview, try: perf report --sort comm,dso";
438
439 switch (use_browser) {
440 case 1:
441 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
442 rep->min_percent,
443 &session->header.env);
444 /*
445 * Usually "ret" is the last pressed key, and we only
446 * care if the key notifies us to switch data file.
447 */
448 if (ret != K_SWITCH_INPUT_DATA)
449 ret = 0;
450 break;
451 case 2:
452 ret = report__gtk_browse_hists(rep, help);
453 break;
454 default:
455 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
456 break;
457 }
458
459 return ret;
460 }
461
462 static void report__collapse_hists(struct report *rep)
463 {
464 struct ui_progress prog;
465 struct perf_evsel *pos;
466
467 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
468
469 evlist__for_each(rep->session->evlist, pos) {
470 struct hists *hists = evsel__hists(pos);
471
472 if (pos->idx == 0)
473 hists->symbol_filter_str = rep->symbol_filter_str;
474
475 hists->socket_filter = rep->socket_filter;
476
477 hists__collapse_resort(hists, &prog);
478
479 /* Non-group events are considered as leader */
480 if (symbol_conf.event_group &&
481 !perf_evsel__is_group_leader(pos)) {
482 struct hists *leader_hists = evsel__hists(pos->leader);
483
484 hists__match(leader_hists, hists);
485 hists__link(leader_hists, hists);
486 }
487 }
488
489 ui_progress__finish();
490 }
491
492 static void report__output_resort(struct report *rep)
493 {
494 struct ui_progress prog;
495 struct perf_evsel *pos;
496
497 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
498
499 evlist__for_each(rep->session->evlist, pos)
500 hists__output_resort(evsel__hists(pos), &prog);
501
502 ui_progress__finish();
503 }
504
505 static int __cmd_report(struct report *rep)
506 {
507 int ret;
508 struct perf_session *session = rep->session;
509 struct perf_evsel *pos;
510 struct perf_data_file *file = session->file;
511
512 signal(SIGINT, sig_handler);
513
514 if (rep->cpu_list) {
515 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
516 rep->cpu_bitmap);
517 if (ret)
518 return ret;
519 }
520
521 if (rep->show_threads)
522 perf_read_values_init(&rep->show_threads_values);
523
524 ret = report__setup_sample_type(rep);
525 if (ret)
526 return ret;
527
528 ret = perf_session__process_events(session);
529 if (ret)
530 return ret;
531
532 report__warn_kptr_restrict(rep);
533
534 evlist__for_each(session->evlist, pos)
535 rep->nr_entries += evsel__hists(pos)->nr_entries;
536
537 if (use_browser == 0) {
538 if (verbose > 3)
539 perf_session__fprintf(session, stdout);
540
541 if (verbose > 2)
542 perf_session__fprintf_dsos(session, stdout);
543
544 if (dump_trace) {
545 perf_session__fprintf_nr_events(session, stdout);
546 perf_evlist__fprintf_nr_events(session->evlist, stdout);
547 return 0;
548 }
549 }
550
551 report__collapse_hists(rep);
552
553 if (session_done())
554 return 0;
555
556 /*
557 * recalculate number of entries after collapsing since it
558 * might be changed during the collapse phase.
559 */
560 rep->nr_entries = 0;
561 evlist__for_each(session->evlist, pos)
562 rep->nr_entries += evsel__hists(pos)->nr_entries;
563
564 if (rep->nr_entries == 0) {
565 ui__error("The %s file has no samples!\n", file->path);
566 return 0;
567 }
568
569 report__output_resort(rep);
570
571 return report__browse_hists(rep);
572 }
573
574 static int
575 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
576 {
577 struct report *rep = (struct report *)opt->value;
578
579 /*
580 * --no-call-graph
581 */
582 if (unset) {
583 rep->dont_use_callchains = true;
584 return 0;
585 }
586
587 return parse_callchain_report_opt(arg);
588 }
589
590 int
591 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
592 const char *arg, int unset __maybe_unused)
593 {
594 if (arg) {
595 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
596 if (err) {
597 char buf[BUFSIZ];
598 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
599 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
600 return -1;
601 }
602 have_ignore_callees = 1;
603 }
604
605 return 0;
606 }
607
608 static int
609 parse_branch_mode(const struct option *opt __maybe_unused,
610 const char *str __maybe_unused, int unset)
611 {
612 int *branch_mode = opt->value;
613
614 *branch_mode = !unset;
615 return 0;
616 }
617
618 static int
619 parse_percent_limit(const struct option *opt, const char *str,
620 int unset __maybe_unused)
621 {
622 struct report *rep = opt->value;
623
624 rep->min_percent = strtof(str, NULL);
625 return 0;
626 }
627
628 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function"
629
630 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
631 CALLCHAIN_REPORT_HELP
632 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
633
634 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
635 {
636 struct perf_session *session;
637 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
638 struct stat st;
639 bool has_br_stack = false;
640 int branch_mode = -1;
641 bool branch_call_mode = false;
642 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
643 const char * const report_usage[] = {
644 "perf report [<options>]",
645 NULL
646 };
647 struct report report = {
648 .tool = {
649 .sample = process_sample_event,
650 .mmap = perf_event__process_mmap,
651 .mmap2 = perf_event__process_mmap2,
652 .comm = perf_event__process_comm,
653 .exit = perf_event__process_exit,
654 .fork = perf_event__process_fork,
655 .lost = perf_event__process_lost,
656 .read = process_read_event,
657 .attr = perf_event__process_attr,
658 .tracing_data = perf_event__process_tracing_data,
659 .build_id = perf_event__process_build_id,
660 .id_index = perf_event__process_id_index,
661 .auxtrace_info = perf_event__process_auxtrace_info,
662 .auxtrace = perf_event__process_auxtrace,
663 .ordered_events = true,
664 .ordering_requires_timestamps = true,
665 },
666 .max_stack = PERF_MAX_STACK_DEPTH,
667 .pretty_printing_style = "normal",
668 .socket_filter = -1,
669 };
670 const struct option options[] = {
671 OPT_STRING('i', "input", &input_name, "file",
672 "input file name"),
673 OPT_INCR('v', "verbose", &verbose,
674 "be more verbose (show symbol address, etc)"),
675 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
676 "dump raw trace in ASCII"),
677 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
678 "file", "vmlinux pathname"),
679 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
680 "file", "kallsyms pathname"),
681 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
682 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
683 "load module symbols - WARNING: use only with -k and LIVE kernel"),
684 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
685 "Show a column with the number of samples"),
686 OPT_BOOLEAN('T', "threads", &report.show_threads,
687 "Show per-thread event counters"),
688 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
689 "pretty printing style key: normal raw"),
690 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
691 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
692 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
693 "Use the stdio interface"),
694 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
695 OPT_BOOLEAN(0, "header-only", &report.header_only,
696 "Show only data header."),
697 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
698 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
699 " Please refer the man page for the complete list."),
700 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
701 "output field(s): overhead, period, sample plus all of sort keys"),
702 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
703 "Show sample percentage for different cpu modes"),
704 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
705 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
706 OPT_STRING('p', "parent", &parent_pattern, "regex",
707 "regex filter to identify parent, see: '--sort parent'"),
708 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
709 "Only display entries with parent-match"),
710 OPT_CALLBACK_DEFAULT('g', "call-graph", &report,
711 "print_type,threshold[,print_limit],order,sort_key[,branch]",
712 report_callchain_help, &report_parse_callchain_opt,
713 callchain_default_opt),
714 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
715 "Accumulate callchains of children and show total overhead as well"),
716 OPT_INTEGER(0, "max-stack", &report.max_stack,
717 "Set the maximum stack depth when parsing the callchain, "
718 "anything beyond the specified depth will be ignored. "
719 "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
720 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
721 "alias for inverted call graph"),
722 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
723 "ignore callees of these functions in call graphs",
724 report_parse_ignore_callees_opt),
725 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
726 "only consider symbols in these dsos"),
727 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
728 "only consider symbols in these comms"),
729 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
730 "only consider symbols in these pids"),
731 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
732 "only consider symbols in these tids"),
733 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
734 "only consider these symbols"),
735 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
736 "only show symbols that (partially) match with this filter"),
737 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
738 "width[,width...]",
739 "don't try to adjust column width, use these fixed values"),
740 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
741 "separator for columns, no spaces will be added between "
742 "columns '.' is reserved."),
743 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
744 "Only display entries resolved to a symbol"),
745 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
746 "Look for files with symbols relative to this directory"),
747 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
748 "list of cpus to profile"),
749 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
750 "Display extended information about perf.data file"),
751 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
752 "Interleave source code with assembly code (default)"),
753 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
754 "Display raw encoding of assembly instructions (default)"),
755 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
756 "Specify disassembler style (e.g. -M intel for intel syntax)"),
757 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
758 "Show a column with the sum of periods"),
759 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
760 "Show event group information together"),
761 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
762 "use branch records for per branch histogram filling",
763 parse_branch_mode),
764 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
765 "add last branch records to call history"),
766 OPT_STRING(0, "objdump", &objdump_path, "path",
767 "objdump binary to use for disassembly and annotations"),
768 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
769 "Disable symbol demangling"),
770 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
771 "Enable kernel symbol demangling"),
772 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
773 OPT_CALLBACK(0, "percent-limit", &report, "percent",
774 "Don't show entries under that percent", parse_percent_limit),
775 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
776 "how to display percentage of filtered entries", parse_filter_percentage),
777 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
778 "Instruction Tracing options",
779 itrace_parse_synth_opts),
780 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
781 "Show full source file name path for source lines"),
782 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
783 "Show callgraph from reference event"),
784 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
785 "only show processor socket that match with this filter"),
786 OPT_END()
787 };
788 struct perf_data_file file = {
789 .mode = PERF_DATA_MODE_READ,
790 };
791 int ret = hists__init();
792
793 if (ret < 0)
794 return ret;
795
796 perf_config(report__config, &report);
797
798 argc = parse_options(argc, argv, options, report_usage, 0);
799
800 if (symbol_conf.vmlinux_name &&
801 access(symbol_conf.vmlinux_name, R_OK)) {
802 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
803 return -EINVAL;
804 }
805 if (symbol_conf.kallsyms_name &&
806 access(symbol_conf.kallsyms_name, R_OK)) {
807 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
808 return -EINVAL;
809 }
810
811 if (report.use_stdio)
812 use_browser = 0;
813 else if (report.use_tui)
814 use_browser = 1;
815 else if (report.use_gtk)
816 use_browser = 2;
817
818 if (report.inverted_callchain)
819 callchain_param.order = ORDER_CALLER;
820 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
821 callchain_param.order = ORDER_CALLER;
822
823 if (itrace_synth_opts.callchain &&
824 (int)itrace_synth_opts.callchain_sz > report.max_stack)
825 report.max_stack = itrace_synth_opts.callchain_sz;
826
827 if (!input_name || !strlen(input_name)) {
828 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
829 input_name = "-";
830 else
831 input_name = "perf.data";
832 }
833
834 file.path = input_name;
835 file.force = symbol_conf.force;
836
837 repeat:
838 session = perf_session__new(&file, false, &report.tool);
839 if (session == NULL)
840 return -1;
841
842 if (report.queue_size) {
843 ordered_events__set_alloc_size(&session->ordered_events,
844 report.queue_size);
845 }
846
847 session->itrace_synth_opts = &itrace_synth_opts;
848
849 report.session = session;
850
851 has_br_stack = perf_header__has_feat(&session->header,
852 HEADER_BRANCH_STACK);
853
854 if (itrace_synth_opts.last_branch)
855 has_br_stack = true;
856
857 /*
858 * Branch mode is a tristate:
859 * -1 means default, so decide based on the file having branch data.
860 * 0/1 means the user chose a mode.
861 */
862 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
863 !branch_call_mode) {
864 sort__mode = SORT_MODE__BRANCH;
865 symbol_conf.cumulate_callchain = false;
866 }
867 if (branch_call_mode) {
868 callchain_param.key = CCKEY_ADDRESS;
869 callchain_param.branch_callstack = 1;
870 symbol_conf.use_callchain = true;
871 callchain_register_param(&callchain_param);
872 if (sort_order == NULL)
873 sort_order = "srcline,symbol,dso";
874 }
875
876 if (report.mem_mode) {
877 if (sort__mode == SORT_MODE__BRANCH) {
878 pr_err("branch and mem mode incompatible\n");
879 goto error;
880 }
881 sort__mode = SORT_MODE__MEMORY;
882 symbol_conf.cumulate_callchain = false;
883 }
884
885 if (setup_sorting() < 0) {
886 if (sort_order)
887 parse_options_usage(report_usage, options, "s", 1);
888 if (field_order)
889 parse_options_usage(sort_order ? NULL : report_usage,
890 options, "F", 1);
891 goto error;
892 }
893
894 /* Force tty output for header output and per-thread stat. */
895 if (report.header || report.header_only || report.show_threads)
896 use_browser = 0;
897
898 if (strcmp(input_name, "-") != 0)
899 setup_browser(true);
900 else
901 use_browser = 0;
902
903 if (report.header || report.header_only) {
904 perf_session__fprintf_info(session, stdout,
905 report.show_full_info);
906 if (report.header_only) {
907 ret = 0;
908 goto error;
909 }
910 } else if (use_browser == 0) {
911 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
912 stdout);
913 }
914
915 /*
916 * Only in the TUI browser we are doing integrated annotation,
917 * so don't allocate extra space that won't be used in the stdio
918 * implementation.
919 */
920 if (ui__has_annotation()) {
921 symbol_conf.priv_size = sizeof(struct annotation);
922 machines__set_symbol_filter(&session->machines,
923 symbol__annotate_init);
924 /*
925 * For searching by name on the "Browse map details".
926 * providing it only in verbose mode not to bloat too
927 * much struct symbol.
928 */
929 if (verbose) {
930 /*
931 * XXX: Need to provide a less kludgy way to ask for
932 * more space per symbol, the u32 is for the index on
933 * the ui browser.
934 * See symbol__browser_index.
935 */
936 symbol_conf.priv_size += sizeof(u32);
937 symbol_conf.sort_by_name = true;
938 }
939 }
940
941 if (symbol__init(&session->header.env) < 0)
942 goto error;
943
944 if (argc) {
945 /*
946 * Special case: if there's an argument left then assume that
947 * it's a symbol filter:
948 */
949 if (argc > 1)
950 usage_with_options(report_usage, options);
951
952 report.symbol_filter_str = argv[0];
953 }
954
955 sort__setup_elide(stdout);
956
957 ret = __cmd_report(&report);
958 if (ret == K_SWITCH_INPUT_DATA) {
959 perf_session__delete(session);
960 goto repeat;
961 } else
962 ret = 0;
963
964 error:
965 perf_session__delete(session);
966 return ret;
967 }