perf tools: Support callchain sorting based on addresses
[deliverable/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 "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 "arch/common.h"
37
38 #include <linux/bitmap.h>
39
40 struct perf_report {
41 struct perf_tool tool;
42 struct perf_session *session;
43 bool force, use_tui, use_gtk, use_stdio;
44 bool hide_unresolved;
45 bool dont_use_callchains;
46 bool show_full_info;
47 bool show_threads;
48 bool inverted_callchain;
49 bool mem_mode;
50 struct perf_read_values show_threads_values;
51 const char *pretty_printing_style;
52 symbol_filter_t annotate_init;
53 const char *cpu_list;
54 const char *symbol_filter_str;
55 float min_percent;
56 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
57 };
58
59 static int perf_report_config(const char *var, const char *value, void *cb)
60 {
61 if (!strcmp(var, "report.group")) {
62 symbol_conf.event_group = perf_config_bool(var, value);
63 return 0;
64 }
65 if (!strcmp(var, "report.percent-limit")) {
66 struct perf_report *rep = cb;
67 rep->min_percent = strtof(value, NULL);
68 return 0;
69 }
70
71 return perf_default_config(var, value, cb);
72 }
73
74 static int perf_report__add_mem_hist_entry(struct perf_tool *tool,
75 struct addr_location *al,
76 struct perf_sample *sample,
77 struct perf_evsel *evsel,
78 struct machine *machine,
79 union perf_event *event)
80 {
81 struct perf_report *rep = container_of(tool, struct perf_report, tool);
82 struct symbol *parent = NULL;
83 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
84 int err = 0;
85 struct hist_entry *he;
86 struct mem_info *mi, *mx;
87 uint64_t cost;
88
89 if ((sort__has_parent || symbol_conf.use_callchain) &&
90 sample->callchain) {
91 err = machine__resolve_callchain(machine, evsel, al->thread,
92 sample, &parent, al);
93 if (err)
94 return err;
95 }
96
97 mi = machine__resolve_mem(machine, al->thread, sample, cpumode);
98 if (!mi)
99 return -ENOMEM;
100
101 if (rep->hide_unresolved && !al->sym)
102 return 0;
103
104 cost = sample->weight;
105 if (!cost)
106 cost = 1;
107
108 /*
109 * must pass period=weight in order to get the correct
110 * sorting from hists__collapse_resort() which is solely
111 * based on periods. We want sorting be done on nr_events * weight
112 * and this is indirectly achieved by passing period=weight here
113 * and the he_stat__add_period() function.
114 */
115 he = __hists__add_mem_entry(&evsel->hists, al, parent, mi, cost, cost);
116 if (!he)
117 return -ENOMEM;
118
119 /*
120 * In the TUI browser, we are doing integrated annotation,
121 * so we don't allocate the extra space needed because the stdio
122 * code will not use it.
123 */
124 if (sort__has_sym && he->ms.sym && use_browser > 0) {
125 struct annotation *notes = symbol__annotation(he->ms.sym);
126
127 assert(evsel != NULL);
128
129 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
130 goto out;
131
132 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
133 if (err)
134 goto out;
135 }
136
137 if (sort__has_sym && he->mem_info->daddr.sym && use_browser > 0) {
138 struct annotation *notes;
139
140 mx = he->mem_info;
141
142 notes = symbol__annotation(mx->daddr.sym);
143 if (notes->src == NULL && symbol__alloc_hist(mx->daddr.sym) < 0)
144 goto out;
145
146 err = symbol__inc_addr_samples(mx->daddr.sym,
147 mx->daddr.map,
148 evsel->idx,
149 mx->daddr.al_addr);
150 if (err)
151 goto out;
152 }
153
154 evsel->hists.stats.total_period += cost;
155 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
156 err = 0;
157
158 if (symbol_conf.use_callchain) {
159 err = callchain_append(he->callchain,
160 &callchain_cursor,
161 sample->period);
162 }
163 out:
164 return err;
165 }
166
167 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
168 struct addr_location *al,
169 struct perf_sample *sample,
170 struct perf_evsel *evsel,
171 struct machine *machine)
172 {
173 struct perf_report *rep = container_of(tool, struct perf_report, tool);
174 struct symbol *parent = NULL;
175 int err = 0;
176 unsigned i;
177 struct hist_entry *he;
178 struct branch_info *bi, *bx;
179
180 if ((sort__has_parent || symbol_conf.use_callchain)
181 && sample->callchain) {
182 err = machine__resolve_callchain(machine, evsel, al->thread,
183 sample, &parent, al);
184 if (err)
185 return err;
186 }
187
188 bi = machine__resolve_bstack(machine, al->thread,
189 sample->branch_stack);
190 if (!bi)
191 return -ENOMEM;
192
193 for (i = 0; i < sample->branch_stack->nr; i++) {
194 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
195 continue;
196
197 err = -ENOMEM;
198
199 /*
200 * The report shows the percentage of total branches captured
201 * and not events sampled. Thus we use a pseudo period of 1.
202 */
203 he = __hists__add_branch_entry(&evsel->hists, al, parent,
204 &bi[i], 1, 1);
205 if (he) {
206 struct annotation *notes;
207 bx = he->branch_info;
208 if (bx->from.sym && use_browser == 1 && sort__has_sym) {
209 notes = symbol__annotation(bx->from.sym);
210 if (!notes->src
211 && symbol__alloc_hist(bx->from.sym) < 0)
212 goto out;
213
214 err = symbol__inc_addr_samples(bx->from.sym,
215 bx->from.map,
216 evsel->idx,
217 bx->from.al_addr);
218 if (err)
219 goto out;
220 }
221
222 if (bx->to.sym && use_browser == 1 && sort__has_sym) {
223 notes = symbol__annotation(bx->to.sym);
224 if (!notes->src
225 && symbol__alloc_hist(bx->to.sym) < 0)
226 goto out;
227
228 err = symbol__inc_addr_samples(bx->to.sym,
229 bx->to.map,
230 evsel->idx,
231 bx->to.al_addr);
232 if (err)
233 goto out;
234 }
235 evsel->hists.stats.total_period += 1;
236 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
237 } else
238 goto out;
239 }
240 err = 0;
241 out:
242 free(bi);
243 return err;
244 }
245
246 static int perf_evsel__add_hist_entry(struct perf_evsel *evsel,
247 struct addr_location *al,
248 struct perf_sample *sample,
249 struct machine *machine)
250 {
251 struct symbol *parent = NULL;
252 int err = 0;
253 struct hist_entry *he;
254
255 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
256 err = machine__resolve_callchain(machine, evsel, al->thread,
257 sample, &parent, al);
258 if (err)
259 return err;
260 }
261
262 he = __hists__add_entry(&evsel->hists, al, parent, sample->period,
263 sample->weight);
264 if (he == NULL)
265 return -ENOMEM;
266
267 if (symbol_conf.use_callchain) {
268 err = callchain_append(he->callchain,
269 &callchain_cursor,
270 sample->period);
271 if (err)
272 return err;
273 }
274 /*
275 * Only in the TUI browser we are doing integrated annotation,
276 * so we don't allocated the extra space needed because the stdio
277 * code will not use it.
278 */
279 if (he->ms.sym != NULL && use_browser == 1 && sort__has_sym) {
280 struct annotation *notes = symbol__annotation(he->ms.sym);
281
282 assert(evsel != NULL);
283
284 err = -ENOMEM;
285 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
286 goto out;
287
288 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
289 }
290
291 evsel->hists.stats.total_period += sample->period;
292 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
293 out:
294 return err;
295 }
296
297
298 static int process_sample_event(struct perf_tool *tool,
299 union perf_event *event,
300 struct perf_sample *sample,
301 struct perf_evsel *evsel,
302 struct machine *machine)
303 {
304 struct perf_report *rep = container_of(tool, struct perf_report, tool);
305 struct addr_location al;
306 int ret;
307
308 if (perf_event__preprocess_sample(event, machine, &al, sample,
309 rep->annotate_init) < 0) {
310 fprintf(stderr, "problem processing %d event, skipping it.\n",
311 event->header.type);
312 return -1;
313 }
314
315 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
316 return 0;
317
318 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
319 return 0;
320
321 if (sort__mode == SORT_MODE__BRANCH) {
322 ret = perf_report__add_branch_hist_entry(tool, &al, sample,
323 evsel, machine);
324 if (ret < 0)
325 pr_debug("problem adding lbr entry, skipping event\n");
326 } else if (rep->mem_mode == 1) {
327 ret = perf_report__add_mem_hist_entry(tool, &al, sample,
328 evsel, machine, event);
329 if (ret < 0)
330 pr_debug("problem adding mem entry, skipping event\n");
331 } else {
332 if (al.map != NULL)
333 al.map->dso->hit = 1;
334
335 ret = perf_evsel__add_hist_entry(evsel, &al, sample, machine);
336 if (ret < 0)
337 pr_debug("problem incrementing symbol period, skipping event\n");
338 }
339 return ret;
340 }
341
342 static int process_read_event(struct perf_tool *tool,
343 union perf_event *event,
344 struct perf_sample *sample __maybe_unused,
345 struct perf_evsel *evsel,
346 struct machine *machine __maybe_unused)
347 {
348 struct perf_report *rep = container_of(tool, struct perf_report, tool);
349
350 if (rep->show_threads) {
351 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
352 perf_read_values_add_value(&rep->show_threads_values,
353 event->read.pid, event->read.tid,
354 event->read.id,
355 name,
356 event->read.value);
357 }
358
359 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
360 evsel ? perf_evsel__name(evsel) : "FAIL",
361 event->read.value);
362
363 return 0;
364 }
365
366 /* For pipe mode, sample_type is not currently set */
367 static int perf_report__setup_sample_type(struct perf_report *rep)
368 {
369 struct perf_session *self = rep->session;
370 u64 sample_type = perf_evlist__sample_type(self->evlist);
371
372 if (!self->fd_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
373 if (sort__has_parent) {
374 ui__error("Selected --sort parent, but no "
375 "callchain data. Did you call "
376 "'perf record' without -g?\n");
377 return -EINVAL;
378 }
379 if (symbol_conf.use_callchain) {
380 ui__error("Selected -g but no callchain data. Did "
381 "you call 'perf record' without -g?\n");
382 return -1;
383 }
384 } else if (!rep->dont_use_callchains &&
385 callchain_param.mode != CHAIN_NONE &&
386 !symbol_conf.use_callchain) {
387 symbol_conf.use_callchain = true;
388 if (callchain_register_param(&callchain_param) < 0) {
389 ui__error("Can't register callchain params.\n");
390 return -EINVAL;
391 }
392 }
393
394 if (sort__mode == SORT_MODE__BRANCH) {
395 if (!self->fd_pipe &&
396 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
397 ui__error("Selected -b but no branch data. "
398 "Did you call perf record without -b?\n");
399 return -1;
400 }
401 }
402
403 return 0;
404 }
405
406 extern volatile int session_done;
407
408 static void sig_handler(int sig __maybe_unused)
409 {
410 session_done = 1;
411 }
412
413 static size_t hists__fprintf_nr_sample_events(struct perf_report *rep,
414 struct hists *self,
415 const char *evname, FILE *fp)
416 {
417 size_t ret;
418 char unit;
419 unsigned long nr_samples = self->stats.nr_events[PERF_RECORD_SAMPLE];
420 u64 nr_events = self->stats.total_period;
421 struct perf_evsel *evsel = hists_to_evsel(self);
422 char buf[512];
423 size_t size = sizeof(buf);
424
425 if (perf_evsel__is_group_event(evsel)) {
426 struct perf_evsel *pos;
427
428 perf_evsel__group_desc(evsel, buf, size);
429 evname = buf;
430
431 for_each_group_member(pos, evsel) {
432 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
433 nr_events += pos->hists.stats.total_period;
434 }
435 }
436
437 nr_samples = convert_unit(nr_samples, &unit);
438 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
439 if (evname != NULL)
440 ret += fprintf(fp, " of event '%s'", evname);
441
442 if (rep->mem_mode) {
443 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
444 ret += fprintf(fp, "\n# Sort order : %s", sort_order);
445 } else
446 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
447 return ret + fprintf(fp, "\n#\n");
448 }
449
450 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
451 struct perf_report *rep,
452 const char *help)
453 {
454 struct perf_evsel *pos;
455
456 list_for_each_entry(pos, &evlist->entries, node) {
457 struct hists *hists = &pos->hists;
458 const char *evname = perf_evsel__name(pos);
459
460 if (symbol_conf.event_group &&
461 !perf_evsel__is_group_leader(pos))
462 continue;
463
464 hists__fprintf_nr_sample_events(rep, hists, evname, stdout);
465 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
466 fprintf(stdout, "\n\n");
467 }
468
469 if (sort_order == default_sort_order &&
470 parent_pattern == default_parent_pattern) {
471 fprintf(stdout, "#\n# (%s)\n#\n", help);
472
473 if (rep->show_threads) {
474 bool style = !strcmp(rep->pretty_printing_style, "raw");
475 perf_read_values_display(stdout, &rep->show_threads_values,
476 style);
477 perf_read_values_destroy(&rep->show_threads_values);
478 }
479 }
480
481 return 0;
482 }
483
484 static int __cmd_report(struct perf_report *rep)
485 {
486 int ret = -EINVAL;
487 u64 nr_samples;
488 struct perf_session *session = rep->session;
489 struct perf_evsel *pos;
490 struct map *kernel_map;
491 struct kmap *kernel_kmap;
492 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
493
494 signal(SIGINT, sig_handler);
495
496 if (rep->cpu_list) {
497 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
498 rep->cpu_bitmap);
499 if (ret)
500 return ret;
501 }
502
503 if (use_browser <= 0)
504 perf_session__fprintf_info(session, stdout, rep->show_full_info);
505
506 if (rep->show_threads)
507 perf_read_values_init(&rep->show_threads_values);
508
509 ret = perf_report__setup_sample_type(rep);
510 if (ret)
511 return ret;
512
513 ret = perf_session__process_events(session, &rep->tool);
514 if (ret)
515 return ret;
516
517 kernel_map = session->machines.host.vmlinux_maps[MAP__FUNCTION];
518 kernel_kmap = map__kmap(kernel_map);
519 if (kernel_map == NULL ||
520 (kernel_map->dso->hit &&
521 (kernel_kmap->ref_reloc_sym == NULL ||
522 kernel_kmap->ref_reloc_sym->addr == 0))) {
523 const char *desc =
524 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
525 "can't be resolved.";
526
527 if (kernel_map) {
528 const struct dso *kdso = kernel_map->dso;
529 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
530 desc = "If some relocation was applied (e.g. "
531 "kexec) symbols may be misresolved.";
532 }
533 }
534
535 ui__warning(
536 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
537 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
538 "Samples in kernel modules can't be resolved as well.\n\n",
539 desc);
540 }
541
542 if (verbose > 3)
543 perf_session__fprintf(session, stdout);
544
545 if (verbose > 2)
546 perf_session__fprintf_dsos(session, stdout);
547
548 if (dump_trace) {
549 perf_session__fprintf_nr_events(session, stdout);
550 return 0;
551 }
552
553 nr_samples = 0;
554 list_for_each_entry(pos, &session->evlist->entries, node) {
555 struct hists *hists = &pos->hists;
556
557 if (pos->idx == 0)
558 hists->symbol_filter_str = rep->symbol_filter_str;
559
560 hists__collapse_resort(hists);
561 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
562
563 /* Non-group events are considered as leader */
564 if (symbol_conf.event_group &&
565 !perf_evsel__is_group_leader(pos)) {
566 struct hists *leader_hists = &pos->leader->hists;
567
568 hists__match(leader_hists, hists);
569 hists__link(leader_hists, hists);
570 }
571 }
572
573 if (nr_samples == 0) {
574 ui__error("The %s file has no samples!\n", session->filename);
575 return 0;
576 }
577
578 list_for_each_entry(pos, &session->evlist->entries, node)
579 hists__output_resort(&pos->hists);
580
581 if (use_browser > 0) {
582 if (use_browser == 1) {
583 ret = perf_evlist__tui_browse_hists(session->evlist,
584 help, NULL,
585 rep->min_percent,
586 &session->header.env);
587 /*
588 * Usually "ret" is the last pressed key, and we only
589 * care if the key notifies us to switch data file.
590 */
591 if (ret != K_SWITCH_INPUT_DATA)
592 ret = 0;
593
594 } else if (use_browser == 2) {
595 perf_evlist__gtk_browse_hists(session->evlist, help,
596 NULL, rep->min_percent);
597 }
598 } else
599 perf_evlist__tty_browse_hists(session->evlist, rep, help);
600
601 return ret;
602 }
603
604 static int
605 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
606 {
607 struct perf_report *rep = (struct perf_report *)opt->value;
608 char *tok, *tok2;
609 char *endptr;
610
611 /*
612 * --no-call-graph
613 */
614 if (unset) {
615 rep->dont_use_callchains = true;
616 return 0;
617 }
618
619 symbol_conf.use_callchain = true;
620
621 if (!arg)
622 return 0;
623
624 tok = strtok((char *)arg, ",");
625 if (!tok)
626 return -1;
627
628 /* get the output mode */
629 if (!strncmp(tok, "graph", strlen(arg)))
630 callchain_param.mode = CHAIN_GRAPH_ABS;
631
632 else if (!strncmp(tok, "flat", strlen(arg)))
633 callchain_param.mode = CHAIN_FLAT;
634
635 else if (!strncmp(tok, "fractal", strlen(arg)))
636 callchain_param.mode = CHAIN_GRAPH_REL;
637
638 else if (!strncmp(tok, "none", strlen(arg))) {
639 callchain_param.mode = CHAIN_NONE;
640 symbol_conf.use_callchain = false;
641
642 return 0;
643 }
644
645 else
646 return -1;
647
648 /* get the min percentage */
649 tok = strtok(NULL, ",");
650 if (!tok)
651 goto setup;
652
653 callchain_param.min_percent = strtod(tok, &endptr);
654 if (tok == endptr)
655 return -1;
656
657 /* get the print limit */
658 tok2 = strtok(NULL, ",");
659 if (!tok2)
660 goto setup;
661
662 if (tok2[0] != 'c') {
663 callchain_param.print_limit = strtoul(tok2, &endptr, 0);
664 tok2 = strtok(NULL, ",");
665 if (!tok2)
666 goto setup;
667 }
668
669 /* get the call chain order */
670 if (!strncmp(tok2, "caller", strlen("caller")))
671 callchain_param.order = ORDER_CALLER;
672 else if (!strncmp(tok2, "callee", strlen("callee")))
673 callchain_param.order = ORDER_CALLEE;
674 else
675 return -1;
676
677 /* Get the sort key */
678 tok2 = strtok(NULL, ",");
679 if (!tok2)
680 goto setup;
681 if (!strncmp(tok2, "function", strlen("function")))
682 callchain_param.key = CCKEY_FUNCTION;
683 else if (!strncmp(tok2, "address", strlen("address")))
684 callchain_param.key = CCKEY_ADDRESS;
685 else
686 return -1;
687 setup:
688 if (callchain_register_param(&callchain_param) < 0) {
689 fprintf(stderr, "Can't register callchain params\n");
690 return -1;
691 }
692 return 0;
693 }
694
695 int
696 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
697 const char *arg, int unset __maybe_unused)
698 {
699 if (arg) {
700 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
701 if (err) {
702 char buf[BUFSIZ];
703 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
704 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
705 return -1;
706 }
707 have_ignore_callees = 1;
708 }
709
710 return 0;
711 }
712
713 static int
714 parse_branch_mode(const struct option *opt __maybe_unused,
715 const char *str __maybe_unused, int unset)
716 {
717 int *branch_mode = opt->value;
718
719 *branch_mode = !unset;
720 return 0;
721 }
722
723 static int
724 parse_percent_limit(const struct option *opt, const char *str,
725 int unset __maybe_unused)
726 {
727 struct perf_report *rep = opt->value;
728
729 rep->min_percent = strtof(str, NULL);
730 return 0;
731 }
732
733 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
734 {
735 struct perf_session *session;
736 struct stat st;
737 bool has_br_stack = false;
738 int branch_mode = -1;
739 int ret = -1;
740 char callchain_default_opt[] = "fractal,0.5,callee";
741 const char * const report_usage[] = {
742 "perf report [<options>]",
743 NULL
744 };
745 struct perf_report report = {
746 .tool = {
747 .sample = process_sample_event,
748 .mmap = perf_event__process_mmap,
749 .comm = perf_event__process_comm,
750 .exit = perf_event__process_exit,
751 .fork = perf_event__process_fork,
752 .lost = perf_event__process_lost,
753 .read = process_read_event,
754 .attr = perf_event__process_attr,
755 .tracing_data = perf_event__process_tracing_data,
756 .build_id = perf_event__process_build_id,
757 .ordered_samples = true,
758 .ordering_requires_timestamps = true,
759 },
760 .pretty_printing_style = "normal",
761 };
762 const struct option options[] = {
763 OPT_STRING('i', "input", &input_name, "file",
764 "input file name"),
765 OPT_INCR('v', "verbose", &verbose,
766 "be more verbose (show symbol address, etc)"),
767 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
768 "dump raw trace in ASCII"),
769 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
770 "file", "vmlinux pathname"),
771 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
772 "file", "kallsyms pathname"),
773 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
774 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
775 "load module symbols - WARNING: use only with -k and LIVE kernel"),
776 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
777 "Show a column with the number of samples"),
778 OPT_BOOLEAN('T', "threads", &report.show_threads,
779 "Show per-thread event counters"),
780 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
781 "pretty printing style key: normal raw"),
782 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
783 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
784 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
785 "Use the stdio interface"),
786 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
787 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
788 " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
789 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
790 "snoop, locked"),
791 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
792 "Show sample percentage for different cpu modes"),
793 OPT_STRING('p', "parent", &parent_pattern, "regex",
794 "regex filter to identify parent, see: '--sort parent'"),
795 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
796 "Only display entries with parent-match"),
797 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
798 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit, callchain order, key (function or address). "
799 "Default: fractal,0.5,callee,function", &parse_callchain_opt, callchain_default_opt),
800 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
801 "alias for inverted call graph"),
802 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
803 "ignore callees of these functions in call graphs",
804 report_parse_ignore_callees_opt),
805 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
806 "only consider symbols in these dsos"),
807 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
808 "only consider symbols in these comms"),
809 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
810 "only consider these symbols"),
811 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
812 "only show symbols that (partially) match with this filter"),
813 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
814 "width[,width...]",
815 "don't try to adjust column width, use these fixed values"),
816 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
817 "separator for columns, no spaces will be added between "
818 "columns '.' is reserved."),
819 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
820 "Only display entries resolved to a symbol"),
821 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
822 "Look for files with symbols relative to this directory"),
823 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
824 "list of cpus to profile"),
825 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
826 "Display extended information about perf.data file"),
827 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
828 "Interleave source code with assembly code (default)"),
829 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
830 "Display raw encoding of assembly instructions (default)"),
831 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
832 "Specify disassembler style (e.g. -M intel for intel syntax)"),
833 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
834 "Show a column with the sum of periods"),
835 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
836 "Show event group information together"),
837 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
838 "use branch records for histogram filling", parse_branch_mode),
839 OPT_STRING(0, "objdump", &objdump_path, "path",
840 "objdump binary to use for disassembly and annotations"),
841 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
842 "Disable symbol demangling"),
843 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
844 OPT_CALLBACK(0, "percent-limit", &report, "percent",
845 "Don't show entries under that percent", parse_percent_limit),
846 OPT_END()
847 };
848
849 perf_config(perf_report_config, &report);
850
851 argc = parse_options(argc, argv, options, report_usage, 0);
852
853 if (report.use_stdio)
854 use_browser = 0;
855 else if (report.use_tui)
856 use_browser = 1;
857 else if (report.use_gtk)
858 use_browser = 2;
859
860 if (report.inverted_callchain)
861 callchain_param.order = ORDER_CALLER;
862
863 if (!input_name || !strlen(input_name)) {
864 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
865 input_name = "-";
866 else
867 input_name = "perf.data";
868 }
869
870 if (strcmp(input_name, "-") != 0)
871 setup_browser(true);
872 else {
873 use_browser = 0;
874 perf_hpp__init();
875 }
876
877 repeat:
878 session = perf_session__new(input_name, O_RDONLY,
879 report.force, false, &report.tool);
880 if (session == NULL)
881 return -ENOMEM;
882
883 report.session = session;
884
885 has_br_stack = perf_header__has_feat(&session->header,
886 HEADER_BRANCH_STACK);
887
888 if (branch_mode == -1 && has_br_stack)
889 sort__mode = SORT_MODE__BRANCH;
890
891 /* sort__mode could be NORMAL if --no-branch-stack */
892 if (sort__mode == SORT_MODE__BRANCH) {
893 /*
894 * if no sort_order is provided, then specify
895 * branch-mode specific order
896 */
897 if (sort_order == default_sort_order)
898 sort_order = "comm,dso_from,symbol_from,"
899 "dso_to,symbol_to";
900
901 }
902 if (report.mem_mode) {
903 if (sort__mode == SORT_MODE__BRANCH) {
904 fprintf(stderr, "branch and mem mode incompatible\n");
905 goto error;
906 }
907 sort__mode = SORT_MODE__MEMORY;
908
909 /*
910 * if no sort_order is provided, then specify
911 * branch-mode specific order
912 */
913 if (sort_order == default_sort_order)
914 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
915 }
916
917 if (setup_sorting() < 0)
918 usage_with_options(report_usage, options);
919
920 /*
921 * Only in the TUI browser we are doing integrated annotation,
922 * so don't allocate extra space that won't be used in the stdio
923 * implementation.
924 */
925 if (use_browser == 1 && sort__has_sym) {
926 symbol_conf.priv_size = sizeof(struct annotation);
927 report.annotate_init = symbol__annotate_init;
928 /*
929 * For searching by name on the "Browse map details".
930 * providing it only in verbose mode not to bloat too
931 * much struct symbol.
932 */
933 if (verbose) {
934 /*
935 * XXX: Need to provide a less kludgy way to ask for
936 * more space per symbol, the u32 is for the index on
937 * the ui browser.
938 * See symbol__browser_index.
939 */
940 symbol_conf.priv_size += sizeof(u32);
941 symbol_conf.sort_by_name = true;
942 }
943 }
944
945 if (symbol__init() < 0)
946 goto error;
947
948 if (parent_pattern != default_parent_pattern) {
949 if (sort_dimension__add("parent") < 0)
950 goto error;
951 }
952
953 if (argc) {
954 /*
955 * Special case: if there's an argument left then assume that
956 * it's a symbol filter:
957 */
958 if (argc > 1)
959 usage_with_options(report_usage, options);
960
961 report.symbol_filter_str = argv[0];
962 }
963
964 sort__setup_elide(stdout);
965
966 ret = __cmd_report(&report);
967 if (ret == K_SWITCH_INPUT_DATA) {
968 perf_session__delete(session);
969 goto repeat;
970 } else
971 ret = 0;
972
973 error:
974 perf_session__delete(session);
975 return ret;
976 }
This page took 0.105647 seconds and 6 git commands to generate.