perf sort: Separate out memory-specific sort keys
[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 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
56 };
57
58 static int perf_report_config(const char *var, const char *value, void *cb)
59 {
60 if (!strcmp(var, "report.group")) {
61 symbol_conf.event_group = perf_config_bool(var, value);
62 return 0;
63 }
64
65 return perf_default_config(var, value, cb);
66 }
67
68 static int perf_report__add_mem_hist_entry(struct perf_tool *tool,
69 struct addr_location *al,
70 struct perf_sample *sample,
71 struct perf_evsel *evsel,
72 struct machine *machine,
73 union perf_event *event)
74 {
75 struct perf_report *rep = container_of(tool, struct perf_report, tool);
76 struct symbol *parent = NULL;
77 u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
78 int err = 0;
79 struct hist_entry *he;
80 struct mem_info *mi, *mx;
81 uint64_t cost;
82
83 if ((sort__has_parent || symbol_conf.use_callchain) &&
84 sample->callchain) {
85 err = machine__resolve_callchain(machine, evsel, al->thread,
86 sample, &parent);
87 if (err)
88 return err;
89 }
90
91 mi = machine__resolve_mem(machine, al->thread, sample, cpumode);
92 if (!mi)
93 return -ENOMEM;
94
95 if (rep->hide_unresolved && !al->sym)
96 return 0;
97
98 cost = sample->weight;
99 if (!cost)
100 cost = 1;
101
102 /*
103 * must pass period=weight in order to get the correct
104 * sorting from hists__collapse_resort() which is solely
105 * based on periods. We want sorting be done on nr_events * weight
106 * and this is indirectly achieved by passing period=weight here
107 * and the he_stat__add_period() function.
108 */
109 he = __hists__add_mem_entry(&evsel->hists, al, parent, mi, cost, cost);
110 if (!he)
111 return -ENOMEM;
112
113 /*
114 * In the TUI browser, we are doing integrated annotation,
115 * so we don't allocate the extra space needed because the stdio
116 * code will not use it.
117 */
118 if (sort__has_sym && he->ms.sym && use_browser > 0) {
119 struct annotation *notes = symbol__annotation(he->ms.sym);
120
121 assert(evsel != NULL);
122
123 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
124 goto out;
125
126 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
127 if (err)
128 goto out;
129 }
130
131 if (sort__has_sym && he->mem_info->daddr.sym && use_browser > 0) {
132 struct annotation *notes;
133
134 mx = he->mem_info;
135
136 notes = symbol__annotation(mx->daddr.sym);
137 if (notes->src == NULL && symbol__alloc_hist(mx->daddr.sym) < 0)
138 goto out;
139
140 err = symbol__inc_addr_samples(mx->daddr.sym,
141 mx->daddr.map,
142 evsel->idx,
143 mx->daddr.al_addr);
144 if (err)
145 goto out;
146 }
147
148 evsel->hists.stats.total_period += cost;
149 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
150 err = 0;
151
152 if (symbol_conf.use_callchain) {
153 err = callchain_append(he->callchain,
154 &callchain_cursor,
155 sample->period);
156 }
157 out:
158 return err;
159 }
160
161 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
162 struct addr_location *al,
163 struct perf_sample *sample,
164 struct perf_evsel *evsel,
165 struct machine *machine)
166 {
167 struct perf_report *rep = container_of(tool, struct perf_report, tool);
168 struct symbol *parent = NULL;
169 int err = 0;
170 unsigned i;
171 struct hist_entry *he;
172 struct branch_info *bi, *bx;
173
174 if ((sort__has_parent || symbol_conf.use_callchain)
175 && sample->callchain) {
176 err = machine__resolve_callchain(machine, evsel, al->thread,
177 sample, &parent);
178 if (err)
179 return err;
180 }
181
182 bi = machine__resolve_bstack(machine, al->thread,
183 sample->branch_stack);
184 if (!bi)
185 return -ENOMEM;
186
187 for (i = 0; i < sample->branch_stack->nr; i++) {
188 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
189 continue;
190
191 err = -ENOMEM;
192
193 /*
194 * The report shows the percentage of total branches captured
195 * and not events sampled. Thus we use a pseudo period of 1.
196 */
197 he = __hists__add_branch_entry(&evsel->hists, al, parent,
198 &bi[i], 1, 1);
199 if (he) {
200 struct annotation *notes;
201 bx = he->branch_info;
202 if (bx->from.sym && use_browser == 1 && sort__has_sym) {
203 notes = symbol__annotation(bx->from.sym);
204 if (!notes->src
205 && symbol__alloc_hist(bx->from.sym) < 0)
206 goto out;
207
208 err = symbol__inc_addr_samples(bx->from.sym,
209 bx->from.map,
210 evsel->idx,
211 bx->from.al_addr);
212 if (err)
213 goto out;
214 }
215
216 if (bx->to.sym && use_browser == 1 && sort__has_sym) {
217 notes = symbol__annotation(bx->to.sym);
218 if (!notes->src
219 && symbol__alloc_hist(bx->to.sym) < 0)
220 goto out;
221
222 err = symbol__inc_addr_samples(bx->to.sym,
223 bx->to.map,
224 evsel->idx,
225 bx->to.al_addr);
226 if (err)
227 goto out;
228 }
229 evsel->hists.stats.total_period += 1;
230 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
231 } else
232 goto out;
233 }
234 err = 0;
235 out:
236 free(bi);
237 return err;
238 }
239
240 static int perf_evsel__add_hist_entry(struct perf_evsel *evsel,
241 struct addr_location *al,
242 struct perf_sample *sample,
243 struct machine *machine)
244 {
245 struct symbol *parent = NULL;
246 int err = 0;
247 struct hist_entry *he;
248
249 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
250 err = machine__resolve_callchain(machine, evsel, al->thread,
251 sample, &parent);
252 if (err)
253 return err;
254 }
255
256 he = __hists__add_entry(&evsel->hists, al, parent, sample->period,
257 sample->weight);
258 if (he == NULL)
259 return -ENOMEM;
260
261 if (symbol_conf.use_callchain) {
262 err = callchain_append(he->callchain,
263 &callchain_cursor,
264 sample->period);
265 if (err)
266 return err;
267 }
268 /*
269 * Only in the TUI browser we are doing integrated annotation,
270 * so we don't allocated the extra space needed because the stdio
271 * code will not use it.
272 */
273 if (he->ms.sym != NULL && use_browser == 1 && sort__has_sym) {
274 struct annotation *notes = symbol__annotation(he->ms.sym);
275
276 assert(evsel != NULL);
277
278 err = -ENOMEM;
279 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
280 goto out;
281
282 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
283 }
284
285 evsel->hists.stats.total_period += sample->period;
286 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
287 out:
288 return err;
289 }
290
291
292 static int process_sample_event(struct perf_tool *tool,
293 union perf_event *event,
294 struct perf_sample *sample,
295 struct perf_evsel *evsel,
296 struct machine *machine)
297 {
298 struct perf_report *rep = container_of(tool, struct perf_report, tool);
299 struct addr_location al;
300
301 if (perf_event__preprocess_sample(event, machine, &al, sample,
302 rep->annotate_init) < 0) {
303 fprintf(stderr, "problem processing %d event, skipping it.\n",
304 event->header.type);
305 return -1;
306 }
307
308 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
309 return 0;
310
311 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
312 return 0;
313
314 if (sort__mode == SORT_MODE__BRANCH) {
315 if (perf_report__add_branch_hist_entry(tool, &al, sample,
316 evsel, machine)) {
317 pr_debug("problem adding lbr entry, skipping event\n");
318 return -1;
319 }
320 } else if (rep->mem_mode == 1) {
321 if (perf_report__add_mem_hist_entry(tool, &al, sample,
322 evsel, machine, event)) {
323 pr_debug("problem adding mem entry, skipping event\n");
324 return -1;
325 }
326 } else {
327 if (al.map != NULL)
328 al.map->dso->hit = 1;
329
330 if (perf_evsel__add_hist_entry(evsel, &al, sample, machine)) {
331 pr_debug("problem incrementing symbol period, skipping event\n");
332 return -1;
333 }
334 }
335 return 0;
336 }
337
338 static int process_read_event(struct perf_tool *tool,
339 union perf_event *event,
340 struct perf_sample *sample __maybe_unused,
341 struct perf_evsel *evsel,
342 struct machine *machine __maybe_unused)
343 {
344 struct perf_report *rep = container_of(tool, struct perf_report, tool);
345
346 if (rep->show_threads) {
347 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
348 perf_read_values_add_value(&rep->show_threads_values,
349 event->read.pid, event->read.tid,
350 event->read.id,
351 name,
352 event->read.value);
353 }
354
355 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
356 evsel ? perf_evsel__name(evsel) : "FAIL",
357 event->read.value);
358
359 return 0;
360 }
361
362 /* For pipe mode, sample_type is not currently set */
363 static int perf_report__setup_sample_type(struct perf_report *rep)
364 {
365 struct perf_session *self = rep->session;
366 u64 sample_type = perf_evlist__sample_type(self->evlist);
367
368 if (!self->fd_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
369 if (sort__has_parent) {
370 ui__error("Selected --sort parent, but no "
371 "callchain data. Did you call "
372 "'perf record' without -g?\n");
373 return -EINVAL;
374 }
375 if (symbol_conf.use_callchain) {
376 ui__error("Selected -g but no callchain data. Did "
377 "you call 'perf record' without -g?\n");
378 return -1;
379 }
380 } else if (!rep->dont_use_callchains &&
381 callchain_param.mode != CHAIN_NONE &&
382 !symbol_conf.use_callchain) {
383 symbol_conf.use_callchain = true;
384 if (callchain_register_param(&callchain_param) < 0) {
385 ui__error("Can't register callchain params.\n");
386 return -EINVAL;
387 }
388 }
389
390 if (sort__mode == SORT_MODE__BRANCH) {
391 if (!self->fd_pipe &&
392 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
393 ui__error("Selected -b but no branch data. "
394 "Did you call perf record without -b?\n");
395 return -1;
396 }
397 }
398
399 return 0;
400 }
401
402 extern volatile int session_done;
403
404 static void sig_handler(int sig __maybe_unused)
405 {
406 session_done = 1;
407 }
408
409 static size_t hists__fprintf_nr_sample_events(struct perf_report *rep,
410 struct hists *self,
411 const char *evname, FILE *fp)
412 {
413 size_t ret;
414 char unit;
415 unsigned long nr_samples = self->stats.nr_events[PERF_RECORD_SAMPLE];
416 u64 nr_events = self->stats.total_period;
417 struct perf_evsel *evsel = hists_to_evsel(self);
418 char buf[512];
419 size_t size = sizeof(buf);
420
421 if (perf_evsel__is_group_event(evsel)) {
422 struct perf_evsel *pos;
423
424 perf_evsel__group_desc(evsel, buf, size);
425 evname = buf;
426
427 for_each_group_member(pos, evsel) {
428 nr_samples += pos->hists.stats.nr_events[PERF_RECORD_SAMPLE];
429 nr_events += pos->hists.stats.total_period;
430 }
431 }
432
433 nr_samples = convert_unit(nr_samples, &unit);
434 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
435 if (evname != NULL)
436 ret += fprintf(fp, " of event '%s'", evname);
437
438 if (rep->mem_mode) {
439 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
440 ret += fprintf(fp, "\n# Sort order : %s", sort_order);
441 } else
442 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
443 return ret + fprintf(fp, "\n#\n");
444 }
445
446 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
447 struct perf_report *rep,
448 const char *help)
449 {
450 struct perf_evsel *pos;
451
452 list_for_each_entry(pos, &evlist->entries, node) {
453 struct hists *hists = &pos->hists;
454 const char *evname = perf_evsel__name(pos);
455
456 if (symbol_conf.event_group &&
457 !perf_evsel__is_group_leader(pos))
458 continue;
459
460 hists__fprintf_nr_sample_events(rep, hists, evname, stdout);
461 hists__fprintf(hists, true, 0, 0, stdout);
462 fprintf(stdout, "\n\n");
463 }
464
465 if (sort_order == default_sort_order &&
466 parent_pattern == default_parent_pattern) {
467 fprintf(stdout, "#\n# (%s)\n#\n", help);
468
469 if (rep->show_threads) {
470 bool style = !strcmp(rep->pretty_printing_style, "raw");
471 perf_read_values_display(stdout, &rep->show_threads_values,
472 style);
473 perf_read_values_destroy(&rep->show_threads_values);
474 }
475 }
476
477 return 0;
478 }
479
480 static int __cmd_report(struct perf_report *rep)
481 {
482 int ret = -EINVAL;
483 u64 nr_samples;
484 struct perf_session *session = rep->session;
485 struct perf_evsel *pos;
486 struct map *kernel_map;
487 struct kmap *kernel_kmap;
488 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
489
490 signal(SIGINT, sig_handler);
491
492 if (rep->cpu_list) {
493 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
494 rep->cpu_bitmap);
495 if (ret)
496 goto out_delete;
497 }
498
499 if (use_browser <= 0)
500 perf_session__fprintf_info(session, stdout, rep->show_full_info);
501
502 if (rep->show_threads)
503 perf_read_values_init(&rep->show_threads_values);
504
505 ret = perf_report__setup_sample_type(rep);
506 if (ret)
507 goto out_delete;
508
509 ret = perf_session__process_events(session, &rep->tool);
510 if (ret)
511 goto out_delete;
512
513 kernel_map = session->machines.host.vmlinux_maps[MAP__FUNCTION];
514 kernel_kmap = map__kmap(kernel_map);
515 if (kernel_map == NULL ||
516 (kernel_map->dso->hit &&
517 (kernel_kmap->ref_reloc_sym == NULL ||
518 kernel_kmap->ref_reloc_sym->addr == 0))) {
519 const char *desc =
520 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
521 "can't be resolved.";
522
523 if (kernel_map) {
524 const struct dso *kdso = kernel_map->dso;
525 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
526 desc = "If some relocation was applied (e.g. "
527 "kexec) symbols may be misresolved.";
528 }
529 }
530
531 ui__warning(
532 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
533 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
534 "Samples in kernel modules can't be resolved as well.\n\n",
535 desc);
536 }
537
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 goto out_delete;
547 }
548
549 nr_samples = 0;
550 list_for_each_entry(pos, &session->evlist->entries, node) {
551 struct hists *hists = &pos->hists;
552
553 if (pos->idx == 0)
554 hists->symbol_filter_str = rep->symbol_filter_str;
555
556 hists__collapse_resort(hists);
557 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
558
559 /* Non-group events are considered as leader */
560 if (symbol_conf.event_group &&
561 !perf_evsel__is_group_leader(pos)) {
562 struct hists *leader_hists = &pos->leader->hists;
563
564 hists__match(leader_hists, hists);
565 hists__link(leader_hists, hists);
566 }
567 }
568
569 if (nr_samples == 0) {
570 ui__error("The %s file has no samples!\n", session->filename);
571 goto out_delete;
572 }
573
574 list_for_each_entry(pos, &session->evlist->entries, node)
575 hists__output_resort(&pos->hists);
576
577 if (use_browser > 0) {
578 if (use_browser == 1) {
579 ret = perf_evlist__tui_browse_hists(session->evlist,
580 help,
581 NULL,
582 &session->header.env);
583 /*
584 * Usually "ret" is the last pressed key, and we only
585 * care if the key notifies us to switch data file.
586 */
587 if (ret != K_SWITCH_INPUT_DATA)
588 ret = 0;
589
590 } else if (use_browser == 2) {
591 perf_evlist__gtk_browse_hists(session->evlist, help,
592 NULL);
593 }
594 } else
595 perf_evlist__tty_browse_hists(session->evlist, rep, help);
596
597 out_delete:
598 /*
599 * Speed up the exit process, for large files this can
600 * take quite a while.
601 *
602 * XXX Enable this when using valgrind or if we ever
603 * librarize this command.
604 *
605 * Also experiment with obstacks to see how much speed
606 * up we'll get here.
607 *
608 * perf_session__delete(session);
609 */
610 return ret;
611 }
612
613 static int
614 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
615 {
616 struct perf_report *rep = (struct perf_report *)opt->value;
617 char *tok, *tok2;
618 char *endptr;
619
620 /*
621 * --no-call-graph
622 */
623 if (unset) {
624 rep->dont_use_callchains = true;
625 return 0;
626 }
627
628 symbol_conf.use_callchain = true;
629
630 if (!arg)
631 return 0;
632
633 tok = strtok((char *)arg, ",");
634 if (!tok)
635 return -1;
636
637 /* get the output mode */
638 if (!strncmp(tok, "graph", strlen(arg)))
639 callchain_param.mode = CHAIN_GRAPH_ABS;
640
641 else if (!strncmp(tok, "flat", strlen(arg)))
642 callchain_param.mode = CHAIN_FLAT;
643
644 else if (!strncmp(tok, "fractal", strlen(arg)))
645 callchain_param.mode = CHAIN_GRAPH_REL;
646
647 else if (!strncmp(tok, "none", strlen(arg))) {
648 callchain_param.mode = CHAIN_NONE;
649 symbol_conf.use_callchain = false;
650
651 return 0;
652 }
653
654 else
655 return -1;
656
657 /* get the min percentage */
658 tok = strtok(NULL, ",");
659 if (!tok)
660 goto setup;
661
662 callchain_param.min_percent = strtod(tok, &endptr);
663 if (tok == endptr)
664 return -1;
665
666 /* get the print limit */
667 tok2 = strtok(NULL, ",");
668 if (!tok2)
669 goto setup;
670
671 if (tok2[0] != 'c') {
672 callchain_param.print_limit = strtoul(tok2, &endptr, 0);
673 tok2 = strtok(NULL, ",");
674 if (!tok2)
675 goto setup;
676 }
677
678 /* get the call chain order */
679 if (!strcmp(tok2, "caller"))
680 callchain_param.order = ORDER_CALLER;
681 else if (!strcmp(tok2, "callee"))
682 callchain_param.order = ORDER_CALLEE;
683 else
684 return -1;
685 setup:
686 if (callchain_register_param(&callchain_param) < 0) {
687 fprintf(stderr, "Can't register callchain params\n");
688 return -1;
689 }
690 return 0;
691 }
692
693 static int
694 parse_branch_mode(const struct option *opt __maybe_unused,
695 const char *str __maybe_unused, int unset)
696 {
697 int *branch_mode = opt->value;
698
699 *branch_mode = !unset;
700 return 0;
701 }
702
703 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
704 {
705 struct perf_session *session;
706 struct stat st;
707 bool has_br_stack = false;
708 int branch_mode = -1;
709 int ret = -1;
710 char callchain_default_opt[] = "fractal,0.5,callee";
711 const char * const report_usage[] = {
712 "perf report [<options>]",
713 NULL
714 };
715 struct perf_report report = {
716 .tool = {
717 .sample = process_sample_event,
718 .mmap = perf_event__process_mmap,
719 .comm = perf_event__process_comm,
720 .exit = perf_event__process_exit,
721 .fork = perf_event__process_fork,
722 .lost = perf_event__process_lost,
723 .read = process_read_event,
724 .attr = perf_event__process_attr,
725 .event_type = perf_event__process_event_type,
726 .tracing_data = perf_event__process_tracing_data,
727 .build_id = perf_event__process_build_id,
728 .ordered_samples = true,
729 .ordering_requires_timestamps = true,
730 },
731 .pretty_printing_style = "normal",
732 };
733 const struct option options[] = {
734 OPT_STRING('i', "input", &input_name, "file",
735 "input file name"),
736 OPT_INCR('v', "verbose", &verbose,
737 "be more verbose (show symbol address, etc)"),
738 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
739 "dump raw trace in ASCII"),
740 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
741 "file", "vmlinux pathname"),
742 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
743 "file", "kallsyms pathname"),
744 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
745 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
746 "load module symbols - WARNING: use only with -k and LIVE kernel"),
747 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
748 "Show a column with the number of samples"),
749 OPT_BOOLEAN('T', "threads", &report.show_threads,
750 "Show per-thread event counters"),
751 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
752 "pretty printing style key: normal raw"),
753 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
754 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
755 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
756 "Use the stdio interface"),
757 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
758 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline,"
759 " dso_to, dso_from, symbol_to, symbol_from, mispredict,"
760 " weight, local_weight, mem, symbol_daddr, dso_daddr, tlb, "
761 "snoop, locked"),
762 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
763 "Show sample percentage for different cpu modes"),
764 OPT_STRING('p', "parent", &parent_pattern, "regex",
765 "regex filter to identify parent, see: '--sort parent'"),
766 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
767 "Only display entries with parent-match"),
768 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
769 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit and callchain order. "
770 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
771 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
772 "alias for inverted call graph"),
773 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
774 "only consider symbols in these dsos"),
775 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
776 "only consider symbols in these comms"),
777 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
778 "only consider these symbols"),
779 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
780 "only show symbols that (partially) match with this filter"),
781 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
782 "width[,width...]",
783 "don't try to adjust column width, use these fixed values"),
784 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
785 "separator for columns, no spaces will be added between "
786 "columns '.' is reserved."),
787 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
788 "Only display entries resolved to a symbol"),
789 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
790 "Look for files with symbols relative to this directory"),
791 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
792 "list of cpus to profile"),
793 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
794 "Display extended information about perf.data file"),
795 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
796 "Interleave source code with assembly code (default)"),
797 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
798 "Display raw encoding of assembly instructions (default)"),
799 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
800 "Specify disassembler style (e.g. -M intel for intel syntax)"),
801 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
802 "Show a column with the sum of periods"),
803 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
804 "Show event group information together"),
805 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
806 "use branch records for histogram filling", parse_branch_mode),
807 OPT_STRING(0, "objdump", &objdump_path, "path",
808 "objdump binary to use for disassembly and annotations"),
809 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
810 "Disable symbol demangling"),
811 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
812 OPT_END()
813 };
814
815 perf_config(perf_report_config, NULL);
816
817 argc = parse_options(argc, argv, options, report_usage, 0);
818
819 if (report.use_stdio)
820 use_browser = 0;
821 else if (report.use_tui)
822 use_browser = 1;
823 else if (report.use_gtk)
824 use_browser = 2;
825
826 if (report.inverted_callchain)
827 callchain_param.order = ORDER_CALLER;
828
829 if (!input_name || !strlen(input_name)) {
830 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
831 input_name = "-";
832 else
833 input_name = "perf.data";
834 }
835
836 if (strcmp(input_name, "-") != 0)
837 setup_browser(true);
838 else {
839 use_browser = 0;
840 perf_hpp__column_enable(PERF_HPP__OVERHEAD);
841 perf_hpp__init();
842 }
843
844 repeat:
845 session = perf_session__new(input_name, O_RDONLY,
846 report.force, false, &report.tool);
847 if (session == NULL)
848 return -ENOMEM;
849
850 report.session = session;
851
852 has_br_stack = perf_header__has_feat(&session->header,
853 HEADER_BRANCH_STACK);
854
855 if (branch_mode == -1 && has_br_stack)
856 sort__mode = SORT_MODE__BRANCH;
857
858 /* sort__mode could be NORMAL if --no-branch-stack */
859 if (sort__mode == SORT_MODE__BRANCH) {
860 /*
861 * if no sort_order is provided, then specify
862 * branch-mode specific order
863 */
864 if (sort_order == default_sort_order)
865 sort_order = "comm,dso_from,symbol_from,"
866 "dso_to,symbol_to";
867
868 }
869 if (report.mem_mode) {
870 if (sort__mode == SORT_MODE__BRANCH) {
871 fprintf(stderr, "branch and mem mode incompatible\n");
872 goto error;
873 }
874 sort__mode = SORT_MODE__MEMORY;
875
876 /*
877 * if no sort_order is provided, then specify
878 * branch-mode specific order
879 */
880 if (sort_order == default_sort_order)
881 sort_order = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked";
882 }
883
884 if (setup_sorting() < 0)
885 usage_with_options(report_usage, options);
886
887 /*
888 * Only in the TUI browser we are doing integrated annotation,
889 * so don't allocate extra space that won't be used in the stdio
890 * implementation.
891 */
892 if (use_browser == 1 && sort__has_sym) {
893 symbol_conf.priv_size = sizeof(struct annotation);
894 report.annotate_init = symbol__annotate_init;
895 /*
896 * For searching by name on the "Browse map details".
897 * providing it only in verbose mode not to bloat too
898 * much struct symbol.
899 */
900 if (verbose) {
901 /*
902 * XXX: Need to provide a less kludgy way to ask for
903 * more space per symbol, the u32 is for the index on
904 * the ui browser.
905 * See symbol__browser_index.
906 */
907 symbol_conf.priv_size += sizeof(u32);
908 symbol_conf.sort_by_name = true;
909 }
910 }
911
912 if (symbol__init() < 0)
913 goto error;
914
915 if (parent_pattern != default_parent_pattern) {
916 if (sort_dimension__add("parent") < 0)
917 goto error;
918
919 /*
920 * Only show the parent fields if we explicitly
921 * sort that way. If we only use parent machinery
922 * for filtering, we don't want it.
923 */
924 if (!strstr(sort_order, "parent"))
925 sort_parent.elide = 1;
926 } else
927 symbol_conf.exclude_other = false;
928
929 if (argc) {
930 /*
931 * Special case: if there's an argument left then assume that
932 * it's a symbol filter:
933 */
934 if (argc > 1)
935 usage_with_options(report_usage, options);
936
937 report.symbol_filter_str = argv[0];
938 }
939
940 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
941
942 if (sort__mode == SORT_MODE__BRANCH) {
943 sort_entry__setup_elide(&sort_dso_from, symbol_conf.dso_from_list, "dso_from", stdout);
944 sort_entry__setup_elide(&sort_dso_to, symbol_conf.dso_to_list, "dso_to", stdout);
945 sort_entry__setup_elide(&sort_sym_from, symbol_conf.sym_from_list, "sym_from", stdout);
946 sort_entry__setup_elide(&sort_sym_to, symbol_conf.sym_to_list, "sym_to", stdout);
947 } else {
948 if (report.mem_mode) {
949 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "symbol_daddr", stdout);
950 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso_daddr", stdout);
951 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "mem", stdout);
952 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "local_weight", stdout);
953 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "tlb", stdout);
954 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "snoop", stdout);
955 }
956 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
957 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
958 }
959
960 ret = __cmd_report(&report);
961 if (ret == K_SWITCH_INPUT_DATA) {
962 perf_session__delete(session);
963 goto repeat;
964 } else
965 ret = 0;
966
967 error:
968 perf_session__delete(session);
969 return ret;
970 }
This page took 0.062275 seconds and 6 git commands to generate.