Loading...
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#include <linux/stringify.h>
45
46struct report {
47 struct perf_tool tool;
48 struct perf_session *session;
49 bool use_tui, use_gtk, use_stdio;
50 bool dont_use_callchains;
51 bool show_full_info;
52 bool show_threads;
53 bool inverted_callchain;
54 bool mem_mode;
55 bool header;
56 bool header_only;
57 bool nonany_branch_mode;
58 int max_stack;
59 struct perf_read_values show_threads_values;
60 const char *pretty_printing_style;
61 const char *cpu_list;
62 const char *symbol_filter_str;
63 float min_percent;
64 u64 nr_entries;
65 u64 queue_size;
66 int socket_filter;
67 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
68};
69
70static int report__config(const char *var, const char *value, void *cb)
71{
72 struct report *rep = cb;
73
74 if (!strcmp(var, "report.group")) {
75 symbol_conf.event_group = perf_config_bool(var, value);
76 return 0;
77 }
78 if (!strcmp(var, "report.percent-limit")) {
79 double pcnt = strtof(value, NULL);
80
81 rep->min_percent = pcnt;
82 callchain_param.min_percent = pcnt;
83 return 0;
84 }
85 if (!strcmp(var, "report.children")) {
86 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
87 return 0;
88 }
89 if (!strcmp(var, "report.queue-size")) {
90 rep->queue_size = perf_config_u64(var, value);
91 return 0;
92 }
93
94 return 0;
95}
96
97static int hist_iter__report_callback(struct hist_entry_iter *iter,
98 struct addr_location *al, bool single,
99 void *arg)
100{
101 int err = 0;
102 struct report *rep = arg;
103 struct hist_entry *he = iter->he;
104 struct perf_evsel *evsel = iter->evsel;
105 struct mem_info *mi;
106 struct branch_info *bi;
107
108 if (!ui__has_annotation())
109 return 0;
110
111 hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
112 rep->nonany_branch_mode);
113
114 if (sort__mode == SORT_MODE__BRANCH) {
115 bi = he->branch_info;
116 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
117 if (err)
118 goto out;
119
120 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
121
122 } else if (rep->mem_mode) {
123 mi = he->mem_info;
124 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
125 if (err)
126 goto out;
127
128 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
129
130 } else if (symbol_conf.cumulate_callchain) {
131 if (single)
132 err = hist_entry__inc_addr_samples(he, evsel->idx,
133 al->addr);
134 } else {
135 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
136 }
137
138out:
139 return err;
140}
141
142static int process_sample_event(struct perf_tool *tool,
143 union perf_event *event,
144 struct perf_sample *sample,
145 struct perf_evsel *evsel,
146 struct machine *machine)
147{
148 struct report *rep = container_of(tool, struct report, tool);
149 struct addr_location al;
150 struct hist_entry_iter iter = {
151 .evsel = evsel,
152 .sample = sample,
153 .hide_unresolved = symbol_conf.hide_unresolved,
154 .add_entry_cb = hist_iter__report_callback,
155 };
156 int ret = 0;
157
158 if (machine__resolve(machine, &al, sample) < 0) {
159 pr_debug("problem processing %d event, skipping it.\n",
160 event->header.type);
161 return -1;
162 }
163
164 if (symbol_conf.hide_unresolved && al.sym == NULL)
165 goto out_put;
166
167 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
168 goto out_put;
169
170 if (sort__mode == SORT_MODE__BRANCH) {
171 /*
172 * A non-synthesized event might not have a branch stack if
173 * branch stacks have been synthesized (using itrace options).
174 */
175 if (!sample->branch_stack)
176 goto out_put;
177 iter.ops = &hist_iter_branch;
178 } else if (rep->mem_mode) {
179 iter.ops = &hist_iter_mem;
180 } else if (symbol_conf.cumulate_callchain) {
181 iter.ops = &hist_iter_cumulative;
182 } else {
183 iter.ops = &hist_iter_normal;
184 }
185
186 if (al.map != NULL)
187 al.map->dso->hit = 1;
188
189 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
190 if (ret < 0)
191 pr_debug("problem adding hist entry, skipping event\n");
192out_put:
193 addr_location__put(&al);
194 return ret;
195}
196
197static int process_read_event(struct perf_tool *tool,
198 union perf_event *event,
199 struct perf_sample *sample __maybe_unused,
200 struct perf_evsel *evsel,
201 struct machine *machine __maybe_unused)
202{
203 struct report *rep = container_of(tool, struct report, tool);
204
205 if (rep->show_threads) {
206 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
207 perf_read_values_add_value(&rep->show_threads_values,
208 event->read.pid, event->read.tid,
209 event->read.id,
210 name,
211 event->read.value);
212 }
213
214 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
215 evsel ? perf_evsel__name(evsel) : "FAIL",
216 event->read.value);
217
218 return 0;
219}
220
221/* For pipe mode, sample_type is not currently set */
222static int report__setup_sample_type(struct report *rep)
223{
224 struct perf_session *session = rep->session;
225 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
226 bool is_pipe = perf_data_file__is_pipe(session->file);
227
228 if (session->itrace_synth_opts->callchain ||
229 (!is_pipe &&
230 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
231 !session->itrace_synth_opts->set))
232 sample_type |= PERF_SAMPLE_CALLCHAIN;
233
234 if (session->itrace_synth_opts->last_branch)
235 sample_type |= PERF_SAMPLE_BRANCH_STACK;
236
237 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
238 if (sort__has_parent) {
239 ui__error("Selected --sort parent, but no "
240 "callchain data. Did you call "
241 "'perf record' without -g?\n");
242 return -EINVAL;
243 }
244 if (symbol_conf.use_callchain) {
245 ui__error("Selected -g or --branch-history but no "
246 "callchain data. Did\n"
247 "you call 'perf record' without -g?\n");
248 return -1;
249 }
250 } else if (!rep->dont_use_callchains &&
251 callchain_param.mode != CHAIN_NONE &&
252 !symbol_conf.use_callchain) {
253 symbol_conf.use_callchain = true;
254 if (callchain_register_param(&callchain_param) < 0) {
255 ui__error("Can't register callchain params.\n");
256 return -EINVAL;
257 }
258 }
259
260 if (symbol_conf.cumulate_callchain) {
261 /* Silently ignore if callchain is missing */
262 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
263 symbol_conf.cumulate_callchain = false;
264 perf_hpp__cancel_cumulate();
265 }
266 }
267
268 if (sort__mode == SORT_MODE__BRANCH) {
269 if (!is_pipe &&
270 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
271 ui__error("Selected -b but no branch data. "
272 "Did you call perf record without -b?\n");
273 return -1;
274 }
275 }
276
277 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
278 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
279 (sample_type & PERF_SAMPLE_STACK_USER))
280 callchain_param.record_mode = CALLCHAIN_DWARF;
281 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
282 callchain_param.record_mode = CALLCHAIN_LBR;
283 else
284 callchain_param.record_mode = CALLCHAIN_FP;
285 }
286
287 /* ??? handle more cases than just ANY? */
288 if (!(perf_evlist__combined_branch_type(session->evlist) &
289 PERF_SAMPLE_BRANCH_ANY))
290 rep->nonany_branch_mode = true;
291
292 return 0;
293}
294
295static void sig_handler(int sig __maybe_unused)
296{
297 session_done = 1;
298}
299
300static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
301 const char *evname, FILE *fp)
302{
303 size_t ret;
304 char unit;
305 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
306 u64 nr_events = hists->stats.total_period;
307 struct perf_evsel *evsel = hists_to_evsel(hists);
308 char buf[512];
309 size_t size = sizeof(buf);
310 int socked_id = hists->socket_filter;
311
312 if (symbol_conf.filter_relative) {
313 nr_samples = hists->stats.nr_non_filtered_samples;
314 nr_events = hists->stats.total_non_filtered_period;
315 }
316
317 if (perf_evsel__is_group_event(evsel)) {
318 struct perf_evsel *pos;
319
320 perf_evsel__group_desc(evsel, buf, size);
321 evname = buf;
322
323 for_each_group_member(pos, evsel) {
324 const struct hists *pos_hists = evsel__hists(pos);
325
326 if (symbol_conf.filter_relative) {
327 nr_samples += pos_hists->stats.nr_non_filtered_samples;
328 nr_events += pos_hists->stats.total_non_filtered_period;
329 } else {
330 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
331 nr_events += pos_hists->stats.total_period;
332 }
333 }
334 }
335
336 nr_samples = convert_unit(nr_samples, &unit);
337 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
338 if (evname != NULL)
339 ret += fprintf(fp, " of event '%s'", evname);
340
341 if (symbol_conf.show_ref_callgraph &&
342 strstr(evname, "call-graph=no")) {
343 ret += fprintf(fp, ", show reference callgraph");
344 }
345
346 if (rep->mem_mode) {
347 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
348 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
349 } else
350 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
351
352 if (socked_id > -1)
353 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
354
355 return ret + fprintf(fp, "\n#\n");
356}
357
358static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
359 struct report *rep,
360 const char *help)
361{
362 struct perf_evsel *pos;
363
364 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
365 evlist__for_each(evlist, pos) {
366 struct hists *hists = evsel__hists(pos);
367 const char *evname = perf_evsel__name(pos);
368
369 if (symbol_conf.event_group &&
370 !perf_evsel__is_group_leader(pos))
371 continue;
372
373 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
374 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
375 fprintf(stdout, "\n\n");
376 }
377
378 if (sort_order == NULL &&
379 parent_pattern == default_parent_pattern)
380 fprintf(stdout, "#\n# (%s)\n#\n", help);
381
382 if (rep->show_threads) {
383 bool style = !strcmp(rep->pretty_printing_style, "raw");
384 perf_read_values_display(stdout, &rep->show_threads_values,
385 style);
386 perf_read_values_destroy(&rep->show_threads_values);
387 }
388
389 return 0;
390}
391
392static void report__warn_kptr_restrict(const struct report *rep)
393{
394 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
395 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
396
397 if (kernel_map == NULL ||
398 (kernel_map->dso->hit &&
399 (kernel_kmap->ref_reloc_sym == NULL ||
400 kernel_kmap->ref_reloc_sym->addr == 0))) {
401 const char *desc =
402 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
403 "can't be resolved.";
404
405 if (kernel_map) {
406 const struct dso *kdso = kernel_map->dso;
407 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
408 desc = "If some relocation was applied (e.g. "
409 "kexec) symbols may be misresolved.";
410 }
411 }
412
413 ui__warning(
414"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
415"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
416"Samples in kernel modules can't be resolved as well.\n\n",
417 desc);
418 }
419}
420
421static int report__gtk_browse_hists(struct report *rep, const char *help)
422{
423 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
424 struct hist_browser_timer *timer, float min_pcnt);
425
426 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
427
428 if (hist_browser == NULL) {
429 ui__error("GTK browser not found!\n");
430 return -1;
431 }
432
433 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
434}
435
436static int report__browse_hists(struct report *rep)
437{
438 int ret;
439 struct perf_session *session = rep->session;
440 struct perf_evlist *evlist = session->evlist;
441 const char *help = perf_tip(system_path(TIPDIR));
442
443 if (help == NULL) {
444 /* fallback for people who don't install perf ;-) */
445 help = perf_tip(DOCDIR);
446 if (help == NULL)
447 help = "Cannot load tips.txt file, please install perf!";
448 }
449
450 switch (use_browser) {
451 case 1:
452 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
453 rep->min_percent,
454 &session->header.env);
455 /*
456 * Usually "ret" is the last pressed key, and we only
457 * care if the key notifies us to switch data file.
458 */
459 if (ret != K_SWITCH_INPUT_DATA)
460 ret = 0;
461 break;
462 case 2:
463 ret = report__gtk_browse_hists(rep, help);
464 break;
465 default:
466 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
467 break;
468 }
469
470 return ret;
471}
472
473static int report__collapse_hists(struct report *rep)
474{
475 struct ui_progress prog;
476 struct perf_evsel *pos;
477 int ret = 0;
478
479 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
480
481 evlist__for_each(rep->session->evlist, pos) {
482 struct hists *hists = evsel__hists(pos);
483
484 if (pos->idx == 0)
485 hists->symbol_filter_str = rep->symbol_filter_str;
486
487 hists->socket_filter = rep->socket_filter;
488
489 ret = hists__collapse_resort(hists, &prog);
490 if (ret < 0)
491 break;
492
493 /* Non-group events are considered as leader */
494 if (symbol_conf.event_group &&
495 !perf_evsel__is_group_leader(pos)) {
496 struct hists *leader_hists = evsel__hists(pos->leader);
497
498 hists__match(leader_hists, hists);
499 hists__link(leader_hists, hists);
500 }
501 }
502
503 ui_progress__finish();
504 return ret;
505}
506
507static void report__output_resort(struct report *rep)
508{
509 struct ui_progress prog;
510 struct perf_evsel *pos;
511
512 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
513
514 evlist__for_each(rep->session->evlist, pos)
515 perf_evsel__output_resort(pos, &prog);
516
517 ui_progress__finish();
518}
519
520static int __cmd_report(struct report *rep)
521{
522 int ret;
523 struct perf_session *session = rep->session;
524 struct perf_evsel *pos;
525 struct perf_data_file *file = session->file;
526
527 signal(SIGINT, sig_handler);
528
529 if (rep->cpu_list) {
530 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
531 rep->cpu_bitmap);
532 if (ret) {
533 ui__error("failed to set cpu bitmap\n");
534 return ret;
535 }
536 }
537
538 if (rep->show_threads)
539 perf_read_values_init(&rep->show_threads_values);
540
541 ret = report__setup_sample_type(rep);
542 if (ret) {
543 /* report__setup_sample_type() already showed error message */
544 return ret;
545 }
546
547 ret = perf_session__process_events(session);
548 if (ret) {
549 ui__error("failed to process sample\n");
550 return ret;
551 }
552
553 report__warn_kptr_restrict(rep);
554
555 evlist__for_each(session->evlist, pos)
556 rep->nr_entries += evsel__hists(pos)->nr_entries;
557
558 if (use_browser == 0) {
559 if (verbose > 3)
560 perf_session__fprintf(session, stdout);
561
562 if (verbose > 2)
563 perf_session__fprintf_dsos(session, stdout);
564
565 if (dump_trace) {
566 perf_session__fprintf_nr_events(session, stdout);
567 perf_evlist__fprintf_nr_events(session->evlist, stdout);
568 return 0;
569 }
570 }
571
572 ret = report__collapse_hists(rep);
573 if (ret) {
574 ui__error("failed to process hist entry\n");
575 return ret;
576 }
577
578 if (session_done())
579 return 0;
580
581 /*
582 * recalculate number of entries after collapsing since it
583 * might be changed during the collapse phase.
584 */
585 rep->nr_entries = 0;
586 evlist__for_each(session->evlist, pos)
587 rep->nr_entries += evsel__hists(pos)->nr_entries;
588
589 if (rep->nr_entries == 0) {
590 ui__error("The %s file has no samples!\n", file->path);
591 return 0;
592 }
593
594 report__output_resort(rep);
595
596 return report__browse_hists(rep);
597}
598
599static int
600report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
601{
602 struct report *rep = (struct report *)opt->value;
603
604 /*
605 * --no-call-graph
606 */
607 if (unset) {
608 rep->dont_use_callchains = true;
609 return 0;
610 }
611
612 return parse_callchain_report_opt(arg);
613}
614
615int
616report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
617 const char *arg, int unset __maybe_unused)
618{
619 if (arg) {
620 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
621 if (err) {
622 char buf[BUFSIZ];
623 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
624 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
625 return -1;
626 }
627 have_ignore_callees = 1;
628 }
629
630 return 0;
631}
632
633static int
634parse_branch_mode(const struct option *opt __maybe_unused,
635 const char *str __maybe_unused, int unset)
636{
637 int *branch_mode = opt->value;
638
639 *branch_mode = !unset;
640 return 0;
641}
642
643static int
644parse_percent_limit(const struct option *opt, const char *str,
645 int unset __maybe_unused)
646{
647 struct report *rep = opt->value;
648 double pcnt = strtof(str, NULL);
649
650 rep->min_percent = pcnt;
651 callchain_param.min_percent = pcnt;
652 return 0;
653}
654
655#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
656
657const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
658 CALLCHAIN_REPORT_HELP
659 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
660
661int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
662{
663 struct perf_session *session;
664 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
665 struct stat st;
666 bool has_br_stack = false;
667 int branch_mode = -1;
668 bool branch_call_mode = false;
669 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
670 const char * const report_usage[] = {
671 "perf report [<options>]",
672 NULL
673 };
674 struct report report = {
675 .tool = {
676 .sample = process_sample_event,
677 .mmap = perf_event__process_mmap,
678 .mmap2 = perf_event__process_mmap2,
679 .comm = perf_event__process_comm,
680 .exit = perf_event__process_exit,
681 .fork = perf_event__process_fork,
682 .lost = perf_event__process_lost,
683 .read = process_read_event,
684 .attr = perf_event__process_attr,
685 .tracing_data = perf_event__process_tracing_data,
686 .build_id = perf_event__process_build_id,
687 .id_index = perf_event__process_id_index,
688 .auxtrace_info = perf_event__process_auxtrace_info,
689 .auxtrace = perf_event__process_auxtrace,
690 .ordered_events = true,
691 .ordering_requires_timestamps = true,
692 },
693 .max_stack = PERF_MAX_STACK_DEPTH,
694 .pretty_printing_style = "normal",
695 .socket_filter = -1,
696 };
697 const struct option options[] = {
698 OPT_STRING('i', "input", &input_name, "file",
699 "input file name"),
700 OPT_INCR('v', "verbose", &verbose,
701 "be more verbose (show symbol address, etc)"),
702 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
703 "dump raw trace in ASCII"),
704 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
705 "file", "vmlinux pathname"),
706 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
707 "file", "kallsyms pathname"),
708 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
709 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
710 "load module symbols - WARNING: use only with -k and LIVE kernel"),
711 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
712 "Show a column with the number of samples"),
713 OPT_BOOLEAN('T', "threads", &report.show_threads,
714 "Show per-thread event counters"),
715 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
716 "pretty printing style key: normal raw"),
717 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
718 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
719 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
720 "Use the stdio interface"),
721 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
722 OPT_BOOLEAN(0, "header-only", &report.header_only,
723 "Show only data header."),
724 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
725 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
726 " Please refer the man page for the complete list."),
727 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
728 "output field(s): overhead, period, sample plus all of sort keys"),
729 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
730 "Show sample percentage for different cpu modes"),
731 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
732 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
733 OPT_STRING('p', "parent", &parent_pattern, "regex",
734 "regex filter to identify parent, see: '--sort parent'"),
735 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
736 "Only display entries with parent-match"),
737 OPT_CALLBACK_DEFAULT('g', "call-graph", &report,
738 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
739 report_callchain_help, &report_parse_callchain_opt,
740 callchain_default_opt),
741 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
742 "Accumulate callchains of children and show total overhead as well"),
743 OPT_INTEGER(0, "max-stack", &report.max_stack,
744 "Set the maximum stack depth when parsing the callchain, "
745 "anything beyond the specified depth will be ignored. "
746 "Default: " __stringify(PERF_MAX_STACK_DEPTH)),
747 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
748 "alias for inverted call graph"),
749 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
750 "ignore callees of these functions in call graphs",
751 report_parse_ignore_callees_opt),
752 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
753 "only consider symbols in these dsos"),
754 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
755 "only consider symbols in these comms"),
756 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
757 "only consider symbols in these pids"),
758 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
759 "only consider symbols in these tids"),
760 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
761 "only consider these symbols"),
762 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
763 "only show symbols that (partially) match with this filter"),
764 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
765 "width[,width...]",
766 "don't try to adjust column width, use these fixed values"),
767 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
768 "separator for columns, no spaces will be added between "
769 "columns '.' is reserved."),
770 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
771 "Only display entries resolved to a symbol"),
772 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
773 "Look for files with symbols relative to this directory"),
774 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
775 "list of cpus to profile"),
776 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
777 "Display extended information about perf.data file"),
778 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
779 "Interleave source code with assembly code (default)"),
780 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
781 "Display raw encoding of assembly instructions (default)"),
782 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
783 "Specify disassembler style (e.g. -M intel for intel syntax)"),
784 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
785 "Show a column with the sum of periods"),
786 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
787 "Show event group information together"),
788 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
789 "use branch records for per branch histogram filling",
790 parse_branch_mode),
791 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
792 "add last branch records to call history"),
793 OPT_STRING(0, "objdump", &objdump_path, "path",
794 "objdump binary to use for disassembly and annotations"),
795 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
796 "Disable symbol demangling"),
797 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
798 "Enable kernel symbol demangling"),
799 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
800 OPT_CALLBACK(0, "percent-limit", &report, "percent",
801 "Don't show entries under that percent", parse_percent_limit),
802 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
803 "how to display percentage of filtered entries", parse_filter_percentage),
804 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
805 "Instruction Tracing options",
806 itrace_parse_synth_opts),
807 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
808 "Show full source file name path for source lines"),
809 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
810 "Show callgraph from reference event"),
811 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
812 "only show processor socket that match with this filter"),
813 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
814 "Show raw trace event output (do not use print fmt or plugins)"),
815 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
816 "Show entries in a hierarchy"),
817 OPT_END()
818 };
819 struct perf_data_file file = {
820 .mode = PERF_DATA_MODE_READ,
821 };
822 int ret = hists__init();
823
824 if (ret < 0)
825 return ret;
826
827 perf_config(report__config, &report);
828
829 argc = parse_options(argc, argv, options, report_usage, 0);
830 if (argc) {
831 /*
832 * Special case: if there's an argument left then assume that
833 * it's a symbol filter:
834 */
835 if (argc > 1)
836 usage_with_options(report_usage, options);
837
838 report.symbol_filter_str = argv[0];
839 }
840
841 if (symbol_conf.vmlinux_name &&
842 access(symbol_conf.vmlinux_name, R_OK)) {
843 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
844 return -EINVAL;
845 }
846 if (symbol_conf.kallsyms_name &&
847 access(symbol_conf.kallsyms_name, R_OK)) {
848 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
849 return -EINVAL;
850 }
851
852 if (report.use_stdio)
853 use_browser = 0;
854 else if (report.use_tui)
855 use_browser = 1;
856 else if (report.use_gtk)
857 use_browser = 2;
858
859 if (report.inverted_callchain)
860 callchain_param.order = ORDER_CALLER;
861 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
862 callchain_param.order = ORDER_CALLER;
863
864 if (itrace_synth_opts.callchain &&
865 (int)itrace_synth_opts.callchain_sz > report.max_stack)
866 report.max_stack = itrace_synth_opts.callchain_sz;
867
868 if (!input_name || !strlen(input_name)) {
869 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
870 input_name = "-";
871 else
872 input_name = "perf.data";
873 }
874
875 file.path = input_name;
876 file.force = symbol_conf.force;
877
878repeat:
879 session = perf_session__new(&file, false, &report.tool);
880 if (session == NULL)
881 return -1;
882
883 if (report.queue_size) {
884 ordered_events__set_alloc_size(&session->ordered_events,
885 report.queue_size);
886 }
887
888 session->itrace_synth_opts = &itrace_synth_opts;
889
890 report.session = session;
891
892 has_br_stack = perf_header__has_feat(&session->header,
893 HEADER_BRANCH_STACK);
894
895 if (itrace_synth_opts.last_branch)
896 has_br_stack = true;
897
898 /*
899 * Branch mode is a tristate:
900 * -1 means default, so decide based on the file having branch data.
901 * 0/1 means the user chose a mode.
902 */
903 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
904 !branch_call_mode) {
905 sort__mode = SORT_MODE__BRANCH;
906 symbol_conf.cumulate_callchain = false;
907 }
908 if (branch_call_mode) {
909 callchain_param.key = CCKEY_ADDRESS;
910 callchain_param.branch_callstack = 1;
911 symbol_conf.use_callchain = true;
912 callchain_register_param(&callchain_param);
913 if (sort_order == NULL)
914 sort_order = "srcline,symbol,dso";
915 }
916
917 if (report.mem_mode) {
918 if (sort__mode == SORT_MODE__BRANCH) {
919 pr_err("branch and mem mode incompatible\n");
920 goto error;
921 }
922 sort__mode = SORT_MODE__MEMORY;
923 symbol_conf.cumulate_callchain = false;
924 }
925
926 if (symbol_conf.report_hierarchy) {
927 /* disable incompatible options */
928 symbol_conf.event_group = false;
929 symbol_conf.cumulate_callchain = false;
930
931 if (field_order) {
932 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
933 parse_options_usage(report_usage, options, "F", 1);
934 parse_options_usage(NULL, options, "hierarchy", 0);
935 goto error;
936 }
937
938 sort__need_collapse = true;
939 }
940
941 /* Force tty output for header output and per-thread stat. */
942 if (report.header || report.header_only || report.show_threads)
943 use_browser = 0;
944
945 if (strcmp(input_name, "-") != 0)
946 setup_browser(true);
947 else
948 use_browser = 0;
949
950 if (setup_sorting(session->evlist) < 0) {
951 if (sort_order)
952 parse_options_usage(report_usage, options, "s", 1);
953 if (field_order)
954 parse_options_usage(sort_order ? NULL : report_usage,
955 options, "F", 1);
956 goto error;
957 }
958
959 if (report.header || report.header_only) {
960 perf_session__fprintf_info(session, stdout,
961 report.show_full_info);
962 if (report.header_only) {
963 ret = 0;
964 goto error;
965 }
966 } else if (use_browser == 0) {
967 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
968 stdout);
969 }
970
971 /*
972 * Only in the TUI browser we are doing integrated annotation,
973 * so don't allocate extra space that won't be used in the stdio
974 * implementation.
975 */
976 if (ui__has_annotation()) {
977 symbol_conf.priv_size = sizeof(struct annotation);
978 machines__set_symbol_filter(&session->machines,
979 symbol__annotate_init);
980 /*
981 * For searching by name on the "Browse map details".
982 * providing it only in verbose mode not to bloat too
983 * much struct symbol.
984 */
985 if (verbose) {
986 /*
987 * XXX: Need to provide a less kludgy way to ask for
988 * more space per symbol, the u32 is for the index on
989 * the ui browser.
990 * See symbol__browser_index.
991 */
992 symbol_conf.priv_size += sizeof(u32);
993 symbol_conf.sort_by_name = true;
994 }
995 }
996
997 if (symbol__init(&session->header.env) < 0)
998 goto error;
999
1000 sort__setup_elide(stdout);
1001
1002 ret = __cmd_report(&report);
1003 if (ret == K_SWITCH_INPUT_DATA) {
1004 perf_session__delete(session);
1005 goto repeat;
1006 } else
1007 ret = 0;
1008
1009error:
1010 perf_session__delete(session);
1011 return ret;
1012}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * builtin-report.c
4 *
5 * Builtin report command: Analyze the perf.data input file,
6 * look up and read DSOs and symbol information and display
7 * a histogram of results, along various sorting keys.
8 */
9#include "builtin.h"
10
11#include "util/config.h"
12
13#include "util/annotate.h"
14#include "util/color.h"
15#include "util/dso.h"
16#include <linux/list.h>
17#include <linux/rbtree.h>
18#include <linux/err.h>
19#include <linux/zalloc.h>
20#include "util/map.h"
21#include "util/symbol.h"
22#include "util/map_symbol.h"
23#include "util/mem-events.h"
24#include "util/branch.h"
25#include "util/callchain.h"
26#include "util/values.h"
27
28#include "perf.h"
29#include "util/debug.h"
30#include "util/evlist.h"
31#include "util/evsel.h"
32#include "util/evswitch.h"
33#include "util/header.h"
34#include "util/session.h"
35#include "util/srcline.h"
36#include "util/tool.h"
37
38#include <subcmd/parse-options.h>
39#include <subcmd/exec-cmd.h>
40#include "util/parse-events.h"
41
42#include "util/thread.h"
43#include "util/sort.h"
44#include "util/hist.h"
45#include "util/data.h"
46#include "arch/common.h"
47#include "util/time-utils.h"
48#include "util/auxtrace.h"
49#include "util/units.h"
50#include "util/util.h" // perf_tip()
51#include "ui/ui.h"
52#include "ui/progress.h"
53#include "util/block-info.h"
54
55#include <dlfcn.h>
56#include <errno.h>
57#include <inttypes.h>
58#include <regex.h>
59#include <linux/ctype.h>
60#include <signal.h>
61#include <linux/bitmap.h>
62#include <linux/string.h>
63#include <linux/stringify.h>
64#include <linux/time64.h>
65#include <sys/types.h>
66#include <sys/stat.h>
67#include <unistd.h>
68#include <linux/mman.h>
69
70struct report {
71 struct perf_tool tool;
72 struct perf_session *session;
73 struct evswitch evswitch;
74 bool use_tui, use_gtk, use_stdio;
75 bool show_full_info;
76 bool show_threads;
77 bool inverted_callchain;
78 bool mem_mode;
79 bool stats_mode;
80 bool tasks_mode;
81 bool mmaps_mode;
82 bool header;
83 bool header_only;
84 bool nonany_branch_mode;
85 bool group_set;
86 bool stitch_lbr;
87 int max_stack;
88 struct perf_read_values show_threads_values;
89 struct annotation_options annotation_opts;
90 const char *pretty_printing_style;
91 const char *cpu_list;
92 const char *symbol_filter_str;
93 const char *time_str;
94 struct perf_time_interval *ptime_range;
95 int range_size;
96 int range_num;
97 float min_percent;
98 u64 nr_entries;
99 u64 queue_size;
100 u64 total_cycles;
101 int socket_filter;
102 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
103 struct branch_type_stat brtype_stat;
104 bool symbol_ipc;
105 bool total_cycles_mode;
106 struct block_report *block_reports;
107 int nr_block_reports;
108};
109
110static int report__config(const char *var, const char *value, void *cb)
111{
112 struct report *rep = cb;
113
114 if (!strcmp(var, "report.group")) {
115 symbol_conf.event_group = perf_config_bool(var, value);
116 return 0;
117 }
118 if (!strcmp(var, "report.percent-limit")) {
119 double pcnt = strtof(value, NULL);
120
121 rep->min_percent = pcnt;
122 callchain_param.min_percent = pcnt;
123 return 0;
124 }
125 if (!strcmp(var, "report.children")) {
126 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
127 return 0;
128 }
129 if (!strcmp(var, "report.queue-size"))
130 return perf_config_u64(&rep->queue_size, var, value);
131
132 if (!strcmp(var, "report.sort_order")) {
133 default_sort_order = strdup(value);
134 return 0;
135 }
136
137 return 0;
138}
139
140static int hist_iter__report_callback(struct hist_entry_iter *iter,
141 struct addr_location *al, bool single,
142 void *arg)
143{
144 int err = 0;
145 struct report *rep = arg;
146 struct hist_entry *he = iter->he;
147 struct evsel *evsel = iter->evsel;
148 struct perf_sample *sample = iter->sample;
149 struct mem_info *mi;
150 struct branch_info *bi;
151
152 if (!ui__has_annotation() && !rep->symbol_ipc)
153 return 0;
154
155 if (sort__mode == SORT_MODE__BRANCH) {
156 bi = he->branch_info;
157 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
158 if (err)
159 goto out;
160
161 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
162
163 } else if (rep->mem_mode) {
164 mi = he->mem_info;
165 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
166 if (err)
167 goto out;
168
169 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
170
171 } else if (symbol_conf.cumulate_callchain) {
172 if (single)
173 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
174 } else {
175 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
176 }
177
178out:
179 return err;
180}
181
182static int hist_iter__branch_callback(struct hist_entry_iter *iter,
183 struct addr_location *al __maybe_unused,
184 bool single __maybe_unused,
185 void *arg)
186{
187 struct hist_entry *he = iter->he;
188 struct report *rep = arg;
189 struct branch_info *bi = he->branch_info;
190 struct perf_sample *sample = iter->sample;
191 struct evsel *evsel = iter->evsel;
192 int err;
193
194 branch_type_count(&rep->brtype_stat, &bi->flags,
195 bi->from.addr, bi->to.addr);
196
197 if (!ui__has_annotation() && !rep->symbol_ipc)
198 return 0;
199
200 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
201 if (err)
202 goto out;
203
204 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
205
206out:
207 return err;
208}
209
210static void setup_forced_leader(struct report *report,
211 struct evlist *evlist)
212{
213 if (report->group_set)
214 perf_evlist__force_leader(evlist);
215}
216
217static int process_feature_event(struct perf_session *session,
218 union perf_event *event)
219{
220 struct report *rep = container_of(session->tool, struct report, tool);
221
222 if (event->feat.feat_id < HEADER_LAST_FEATURE)
223 return perf_event__process_feature(session, event);
224
225 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
226 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
227 event->feat.feat_id);
228 return -1;
229 }
230
231 /*
232 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
233 * means all features are received, now we can force the
234 * group if needed.
235 */
236 setup_forced_leader(rep, session->evlist);
237 return 0;
238}
239
240static int process_sample_event(struct perf_tool *tool,
241 union perf_event *event,
242 struct perf_sample *sample,
243 struct evsel *evsel,
244 struct machine *machine)
245{
246 struct report *rep = container_of(tool, struct report, tool);
247 struct addr_location al;
248 struct hist_entry_iter iter = {
249 .evsel = evsel,
250 .sample = sample,
251 .hide_unresolved = symbol_conf.hide_unresolved,
252 .add_entry_cb = hist_iter__report_callback,
253 };
254 int ret = 0;
255
256 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
257 sample->time)) {
258 return 0;
259 }
260
261 if (evswitch__discard(&rep->evswitch, evsel))
262 return 0;
263
264 if (machine__resolve(machine, &al, sample) < 0) {
265 pr_debug("problem processing %d event, skipping it.\n",
266 event->header.type);
267 return -1;
268 }
269
270 if (rep->stitch_lbr)
271 al.thread->lbr_stitch_enable = true;
272
273 if (symbol_conf.hide_unresolved && al.sym == NULL)
274 goto out_put;
275
276 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
277 goto out_put;
278
279 if (sort__mode == SORT_MODE__BRANCH) {
280 /*
281 * A non-synthesized event might not have a branch stack if
282 * branch stacks have been synthesized (using itrace options).
283 */
284 if (!sample->branch_stack)
285 goto out_put;
286
287 iter.add_entry_cb = hist_iter__branch_callback;
288 iter.ops = &hist_iter_branch;
289 } else if (rep->mem_mode) {
290 iter.ops = &hist_iter_mem;
291 } else if (symbol_conf.cumulate_callchain) {
292 iter.ops = &hist_iter_cumulative;
293 } else {
294 iter.ops = &hist_iter_normal;
295 }
296
297 if (al.map != NULL)
298 al.map->dso->hit = 1;
299
300 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
301 hist__account_cycles(sample->branch_stack, &al, sample,
302 rep->nonany_branch_mode,
303 &rep->total_cycles);
304 }
305
306 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
307 if (ret < 0)
308 pr_debug("problem adding hist entry, skipping event\n");
309out_put:
310 addr_location__put(&al);
311 return ret;
312}
313
314static int process_read_event(struct perf_tool *tool,
315 union perf_event *event,
316 struct perf_sample *sample __maybe_unused,
317 struct evsel *evsel,
318 struct machine *machine __maybe_unused)
319{
320 struct report *rep = container_of(tool, struct report, tool);
321
322 if (rep->show_threads) {
323 const char *name = evsel__name(evsel);
324 int err = perf_read_values_add_value(&rep->show_threads_values,
325 event->read.pid, event->read.tid,
326 evsel->idx,
327 name,
328 event->read.value);
329
330 if (err)
331 return err;
332 }
333
334 return 0;
335}
336
337/* For pipe mode, sample_type is not currently set */
338static int report__setup_sample_type(struct report *rep)
339{
340 struct perf_session *session = rep->session;
341 u64 sample_type = evlist__combined_sample_type(session->evlist);
342 bool is_pipe = perf_data__is_pipe(session->data);
343
344 if (session->itrace_synth_opts->callchain ||
345 session->itrace_synth_opts->add_callchain ||
346 (!is_pipe &&
347 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
348 !session->itrace_synth_opts->set))
349 sample_type |= PERF_SAMPLE_CALLCHAIN;
350
351 if (session->itrace_synth_opts->last_branch ||
352 session->itrace_synth_opts->add_last_branch)
353 sample_type |= PERF_SAMPLE_BRANCH_STACK;
354
355 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
356 if (perf_hpp_list.parent) {
357 ui__error("Selected --sort parent, but no "
358 "callchain data. Did you call "
359 "'perf record' without -g?\n");
360 return -EINVAL;
361 }
362 if (symbol_conf.use_callchain &&
363 !symbol_conf.show_branchflag_count) {
364 ui__error("Selected -g or --branch-history.\n"
365 "But no callchain or branch data.\n"
366 "Did you call 'perf record' without -g or -b?\n");
367 return -1;
368 }
369 } else if (!callchain_param.enabled &&
370 callchain_param.mode != CHAIN_NONE &&
371 !symbol_conf.use_callchain) {
372 symbol_conf.use_callchain = true;
373 if (callchain_register_param(&callchain_param) < 0) {
374 ui__error("Can't register callchain params.\n");
375 return -EINVAL;
376 }
377 }
378
379 if (symbol_conf.cumulate_callchain) {
380 /* Silently ignore if callchain is missing */
381 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
382 symbol_conf.cumulate_callchain = false;
383 perf_hpp__cancel_cumulate();
384 }
385 }
386
387 if (sort__mode == SORT_MODE__BRANCH) {
388 if (!is_pipe &&
389 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
390 ui__error("Selected -b but no branch data. "
391 "Did you call perf record without -b?\n");
392 return -1;
393 }
394 }
395
396 if (sort__mode == SORT_MODE__MEMORY) {
397 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
398 ui__error("Selected --mem-mode but no mem data. "
399 "Did you call perf record without -d?\n");
400 return -1;
401 }
402 }
403
404 callchain_param_setup(sample_type);
405
406 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
407 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
408 "Please apply --call-graph lbr when recording.\n");
409 rep->stitch_lbr = false;
410 }
411
412 /* ??? handle more cases than just ANY? */
413 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
414 rep->nonany_branch_mode = true;
415
416#if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
417 if (dwarf_callchain_users) {
418 ui__warning("Please install libunwind or libdw "
419 "development packages during the perf build.\n");
420 }
421#endif
422
423 return 0;
424}
425
426static void sig_handler(int sig __maybe_unused)
427{
428 session_done = 1;
429}
430
431static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
432 const char *evname, FILE *fp)
433{
434 size_t ret;
435 char unit;
436 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
437 u64 nr_events = hists->stats.total_period;
438 struct evsel *evsel = hists_to_evsel(hists);
439 char buf[512];
440 size_t size = sizeof(buf);
441 int socked_id = hists->socket_filter;
442
443 if (quiet)
444 return 0;
445
446 if (symbol_conf.filter_relative) {
447 nr_samples = hists->stats.nr_non_filtered_samples;
448 nr_events = hists->stats.total_non_filtered_period;
449 }
450
451 if (evsel__is_group_event(evsel)) {
452 struct evsel *pos;
453
454 evsel__group_desc(evsel, buf, size);
455 evname = buf;
456
457 for_each_group_member(pos, evsel) {
458 const struct hists *pos_hists = evsel__hists(pos);
459
460 if (symbol_conf.filter_relative) {
461 nr_samples += pos_hists->stats.nr_non_filtered_samples;
462 nr_events += pos_hists->stats.total_non_filtered_period;
463 } else {
464 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
465 nr_events += pos_hists->stats.total_period;
466 }
467 }
468 }
469
470 nr_samples = convert_unit(nr_samples, &unit);
471 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
472 if (evname != NULL) {
473 ret += fprintf(fp, " of event%s '%s'",
474 evsel->core.nr_members > 1 ? "s" : "", evname);
475 }
476
477 if (rep->time_str)
478 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
479
480 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
481 ret += fprintf(fp, ", show reference callgraph");
482 }
483
484 if (rep->mem_mode) {
485 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
486 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
487 } else
488 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
489
490 if (socked_id > -1)
491 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
492
493 return ret + fprintf(fp, "\n#\n");
494}
495
496static int perf_evlist__tui_block_hists_browse(struct evlist *evlist,
497 struct report *rep)
498{
499 struct evsel *pos;
500 int i = 0, ret;
501
502 evlist__for_each_entry(evlist, pos) {
503 ret = report__browse_block_hists(&rep->block_reports[i++].hist,
504 rep->min_percent, pos,
505 &rep->session->header.env,
506 &rep->annotation_opts);
507 if (ret != 0)
508 return ret;
509 }
510
511 return 0;
512}
513
514static int perf_evlist__tty_browse_hists(struct evlist *evlist,
515 struct report *rep,
516 const char *help)
517{
518 struct evsel *pos;
519 int i = 0;
520
521 if (!quiet) {
522 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
523 evlist->stats.total_lost_samples);
524 }
525
526 evlist__for_each_entry(evlist, pos) {
527 struct hists *hists = evsel__hists(pos);
528 const char *evname = evsel__name(pos);
529
530 if (symbol_conf.event_group && !evsel__is_group_leader(pos))
531 continue;
532
533 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
534
535 if (rep->total_cycles_mode) {
536 report__browse_block_hists(&rep->block_reports[i++].hist,
537 rep->min_percent, pos,
538 NULL, NULL);
539 continue;
540 }
541
542 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
543 !(symbol_conf.use_callchain ||
544 symbol_conf.show_branchflag_count));
545 fprintf(stdout, "\n\n");
546 }
547
548 if (!quiet)
549 fprintf(stdout, "#\n# (%s)\n#\n", help);
550
551 if (rep->show_threads) {
552 bool style = !strcmp(rep->pretty_printing_style, "raw");
553 perf_read_values_display(stdout, &rep->show_threads_values,
554 style);
555 perf_read_values_destroy(&rep->show_threads_values);
556 }
557
558 if (sort__mode == SORT_MODE__BRANCH)
559 branch_type_stat_display(stdout, &rep->brtype_stat);
560
561 return 0;
562}
563
564static void report__warn_kptr_restrict(const struct report *rep)
565{
566 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
567 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
568
569 if (perf_evlist__exclude_kernel(rep->session->evlist))
570 return;
571
572 if (kernel_map == NULL ||
573 (kernel_map->dso->hit &&
574 (kernel_kmap->ref_reloc_sym == NULL ||
575 kernel_kmap->ref_reloc_sym->addr == 0))) {
576 const char *desc =
577 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
578 "can't be resolved.";
579
580 if (kernel_map && map__has_symbols(kernel_map)) {
581 desc = "If some relocation was applied (e.g. "
582 "kexec) symbols may be misresolved.";
583 }
584
585 ui__warning(
586"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
587"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
588"Samples in kernel modules can't be resolved as well.\n\n",
589 desc);
590 }
591}
592
593static int report__gtk_browse_hists(struct report *rep, const char *help)
594{
595 int (*hist_browser)(struct evlist *evlist, const char *help,
596 struct hist_browser_timer *timer, float min_pcnt);
597
598 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
599
600 if (hist_browser == NULL) {
601 ui__error("GTK browser not found!\n");
602 return -1;
603 }
604
605 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
606}
607
608static int report__browse_hists(struct report *rep)
609{
610 int ret;
611 struct perf_session *session = rep->session;
612 struct evlist *evlist = session->evlist;
613 const char *help = perf_tip(system_path(TIPDIR));
614
615 if (help == NULL) {
616 /* fallback for people who don't install perf ;-) */
617 help = perf_tip(DOCDIR);
618 if (help == NULL)
619 help = "Cannot load tips.txt file, please install perf!";
620 }
621
622 switch (use_browser) {
623 case 1:
624 if (rep->total_cycles_mode) {
625 ret = perf_evlist__tui_block_hists_browse(evlist, rep);
626 break;
627 }
628
629 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
630 rep->min_percent,
631 &session->header.env,
632 true, &rep->annotation_opts);
633 /*
634 * Usually "ret" is the last pressed key, and we only
635 * care if the key notifies us to switch data file.
636 */
637 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
638 ret = 0;
639 break;
640 case 2:
641 ret = report__gtk_browse_hists(rep, help);
642 break;
643 default:
644 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
645 break;
646 }
647
648 return ret;
649}
650
651static int report__collapse_hists(struct report *rep)
652{
653 struct ui_progress prog;
654 struct evsel *pos;
655 int ret = 0;
656
657 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
658
659 evlist__for_each_entry(rep->session->evlist, pos) {
660 struct hists *hists = evsel__hists(pos);
661
662 if (pos->idx == 0)
663 hists->symbol_filter_str = rep->symbol_filter_str;
664
665 hists->socket_filter = rep->socket_filter;
666
667 ret = hists__collapse_resort(hists, &prog);
668 if (ret < 0)
669 break;
670
671 /* Non-group events are considered as leader */
672 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
673 struct hists *leader_hists = evsel__hists(pos->leader);
674
675 hists__match(leader_hists, hists);
676 hists__link(leader_hists, hists);
677 }
678 }
679
680 ui_progress__finish();
681 return ret;
682}
683
684static int hists__resort_cb(struct hist_entry *he, void *arg)
685{
686 struct report *rep = arg;
687 struct symbol *sym = he->ms.sym;
688
689 if (rep->symbol_ipc && sym && !sym->annotate2) {
690 struct evsel *evsel = hists_to_evsel(he->hists);
691
692 symbol__annotate2(&he->ms, evsel,
693 &annotation__default_options, NULL);
694 }
695
696 return 0;
697}
698
699static void report__output_resort(struct report *rep)
700{
701 struct ui_progress prog;
702 struct evsel *pos;
703
704 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
705
706 evlist__for_each_entry(rep->session->evlist, pos) {
707 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
708 }
709
710 ui_progress__finish();
711}
712
713static void stats_setup(struct report *rep)
714{
715 memset(&rep->tool, 0, sizeof(rep->tool));
716 rep->tool.no_warn = true;
717}
718
719static int stats_print(struct report *rep)
720{
721 struct perf_session *session = rep->session;
722
723 perf_session__fprintf_nr_events(session, stdout);
724 return 0;
725}
726
727static void tasks_setup(struct report *rep)
728{
729 memset(&rep->tool, 0, sizeof(rep->tool));
730 rep->tool.ordered_events = true;
731 if (rep->mmaps_mode) {
732 rep->tool.mmap = perf_event__process_mmap;
733 rep->tool.mmap2 = perf_event__process_mmap2;
734 }
735 rep->tool.comm = perf_event__process_comm;
736 rep->tool.exit = perf_event__process_exit;
737 rep->tool.fork = perf_event__process_fork;
738 rep->tool.no_warn = true;
739}
740
741struct task {
742 struct thread *thread;
743 struct list_head list;
744 struct list_head children;
745};
746
747static struct task *tasks_list(struct task *task, struct machine *machine)
748{
749 struct thread *parent_thread, *thread = task->thread;
750 struct task *parent_task;
751
752 /* Already listed. */
753 if (!list_empty(&task->list))
754 return NULL;
755
756 /* Last one in the chain. */
757 if (thread->ppid == -1)
758 return task;
759
760 parent_thread = machine__find_thread(machine, -1, thread->ppid);
761 if (!parent_thread)
762 return ERR_PTR(-ENOENT);
763
764 parent_task = thread__priv(parent_thread);
765 list_add_tail(&task->list, &parent_task->children);
766 return tasks_list(parent_task, machine);
767}
768
769static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
770{
771 size_t printed = 0;
772 struct map *map;
773
774 maps__for_each_entry(maps, map) {
775 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
776 indent, "", map->start, map->end,
777 map->prot & PROT_READ ? 'r' : '-',
778 map->prot & PROT_WRITE ? 'w' : '-',
779 map->prot & PROT_EXEC ? 'x' : '-',
780 map->flags & MAP_SHARED ? 's' : 'p',
781 map->pgoff,
782 map->dso->id.ino, map->dso->name);
783 }
784
785 return printed;
786}
787
788static void task__print_level(struct task *task, FILE *fp, int level)
789{
790 struct thread *thread = task->thread;
791 struct task *child;
792 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
793 thread->pid_, thread->tid, thread->ppid,
794 level, "");
795
796 fprintf(fp, "%s\n", thread__comm_str(thread));
797
798 maps__fprintf_task(thread->maps, comm_indent, fp);
799
800 if (!list_empty(&task->children)) {
801 list_for_each_entry(child, &task->children, list)
802 task__print_level(child, fp, level + 1);
803 }
804}
805
806static int tasks_print(struct report *rep, FILE *fp)
807{
808 struct perf_session *session = rep->session;
809 struct machine *machine = &session->machines.host;
810 struct task *tasks, *task;
811 unsigned int nr = 0, itask = 0, i;
812 struct rb_node *nd;
813 LIST_HEAD(list);
814
815 /*
816 * No locking needed while accessing machine->threads,
817 * because --tasks is single threaded command.
818 */
819
820 /* Count all the threads. */
821 for (i = 0; i < THREADS__TABLE_SIZE; i++)
822 nr += machine->threads[i].nr;
823
824 tasks = malloc(sizeof(*tasks) * nr);
825 if (!tasks)
826 return -ENOMEM;
827
828 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
829 struct threads *threads = &machine->threads[i];
830
831 for (nd = rb_first_cached(&threads->entries); nd;
832 nd = rb_next(nd)) {
833 task = tasks + itask++;
834
835 task->thread = rb_entry(nd, struct thread, rb_node);
836 INIT_LIST_HEAD(&task->children);
837 INIT_LIST_HEAD(&task->list);
838 thread__set_priv(task->thread, task);
839 }
840 }
841
842 /*
843 * Iterate every task down to the unprocessed parent
844 * and link all in task children list. Task with no
845 * parent is added into 'list'.
846 */
847 for (itask = 0; itask < nr; itask++) {
848 task = tasks + itask;
849
850 if (!list_empty(&task->list))
851 continue;
852
853 task = tasks_list(task, machine);
854 if (IS_ERR(task)) {
855 pr_err("Error: failed to process tasks\n");
856 free(tasks);
857 return PTR_ERR(task);
858 }
859
860 if (task)
861 list_add_tail(&task->list, &list);
862 }
863
864 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
865
866 list_for_each_entry(task, &list, list)
867 task__print_level(task, fp, 0);
868
869 free(tasks);
870 return 0;
871}
872
873static int __cmd_report(struct report *rep)
874{
875 int ret;
876 struct perf_session *session = rep->session;
877 struct evsel *pos;
878 struct perf_data *data = session->data;
879
880 signal(SIGINT, sig_handler);
881
882 if (rep->cpu_list) {
883 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
884 rep->cpu_bitmap);
885 if (ret) {
886 ui__error("failed to set cpu bitmap\n");
887 return ret;
888 }
889 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
890 }
891
892 if (rep->show_threads) {
893 ret = perf_read_values_init(&rep->show_threads_values);
894 if (ret)
895 return ret;
896 }
897
898 ret = report__setup_sample_type(rep);
899 if (ret) {
900 /* report__setup_sample_type() already showed error message */
901 return ret;
902 }
903
904 if (rep->stats_mode)
905 stats_setup(rep);
906
907 if (rep->tasks_mode)
908 tasks_setup(rep);
909
910 ret = perf_session__process_events(session);
911 if (ret) {
912 ui__error("failed to process sample\n");
913 return ret;
914 }
915
916 if (rep->stats_mode)
917 return stats_print(rep);
918
919 if (rep->tasks_mode)
920 return tasks_print(rep, stdout);
921
922 report__warn_kptr_restrict(rep);
923
924 evlist__for_each_entry(session->evlist, pos)
925 rep->nr_entries += evsel__hists(pos)->nr_entries;
926
927 if (use_browser == 0) {
928 if (verbose > 3)
929 perf_session__fprintf(session, stdout);
930
931 if (verbose > 2)
932 perf_session__fprintf_dsos(session, stdout);
933
934 if (dump_trace) {
935 perf_session__fprintf_nr_events(session, stdout);
936 perf_evlist__fprintf_nr_events(session->evlist, stdout);
937 return 0;
938 }
939 }
940
941 ret = report__collapse_hists(rep);
942 if (ret) {
943 ui__error("failed to process hist entry\n");
944 return ret;
945 }
946
947 if (session_done())
948 return 0;
949
950 /*
951 * recalculate number of entries after collapsing since it
952 * might be changed during the collapse phase.
953 */
954 rep->nr_entries = 0;
955 evlist__for_each_entry(session->evlist, pos)
956 rep->nr_entries += evsel__hists(pos)->nr_entries;
957
958 if (rep->nr_entries == 0) {
959 ui__error("The %s data has no samples!\n", data->path);
960 return 0;
961 }
962
963 report__output_resort(rep);
964
965 if (rep->total_cycles_mode) {
966 int block_hpps[6] = {
967 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
968 PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
969 PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
970 PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
971 PERF_HPP_REPORT__BLOCK_RANGE,
972 PERF_HPP_REPORT__BLOCK_DSO,
973 };
974
975 rep->block_reports = block_info__create_report(session->evlist,
976 rep->total_cycles,
977 block_hpps, 6,
978 &rep->nr_block_reports);
979 if (!rep->block_reports)
980 return -1;
981 }
982
983 return report__browse_hists(rep);
984}
985
986static int
987report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
988{
989 struct callchain_param *callchain = opt->value;
990
991 callchain->enabled = !unset;
992 /*
993 * --no-call-graph
994 */
995 if (unset) {
996 symbol_conf.use_callchain = false;
997 callchain->mode = CHAIN_NONE;
998 return 0;
999 }
1000
1001 return parse_callchain_report_opt(arg);
1002}
1003
1004static int
1005parse_time_quantum(const struct option *opt, const char *arg,
1006 int unset __maybe_unused)
1007{
1008 unsigned long *time_q = opt->value;
1009 char *end;
1010
1011 *time_q = strtoul(arg, &end, 0);
1012 if (end == arg)
1013 goto parse_err;
1014 if (*time_q == 0) {
1015 pr_err("time quantum cannot be 0");
1016 return -1;
1017 }
1018 end = skip_spaces(end);
1019 if (*end == 0)
1020 return 0;
1021 if (!strcmp(end, "s")) {
1022 *time_q *= NSEC_PER_SEC;
1023 return 0;
1024 }
1025 if (!strcmp(end, "ms")) {
1026 *time_q *= NSEC_PER_MSEC;
1027 return 0;
1028 }
1029 if (!strcmp(end, "us")) {
1030 *time_q *= NSEC_PER_USEC;
1031 return 0;
1032 }
1033 if (!strcmp(end, "ns"))
1034 return 0;
1035parse_err:
1036 pr_err("Cannot parse time quantum `%s'\n", arg);
1037 return -1;
1038}
1039
1040int
1041report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1042 const char *arg, int unset __maybe_unused)
1043{
1044 if (arg) {
1045 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1046 if (err) {
1047 char buf[BUFSIZ];
1048 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1049 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1050 return -1;
1051 }
1052 have_ignore_callees = 1;
1053 }
1054
1055 return 0;
1056}
1057
1058static int
1059parse_branch_mode(const struct option *opt,
1060 const char *str __maybe_unused, int unset)
1061{
1062 int *branch_mode = opt->value;
1063
1064 *branch_mode = !unset;
1065 return 0;
1066}
1067
1068static int
1069parse_percent_limit(const struct option *opt, const char *str,
1070 int unset __maybe_unused)
1071{
1072 struct report *rep = opt->value;
1073 double pcnt = strtof(str, NULL);
1074
1075 rep->min_percent = pcnt;
1076 callchain_param.min_percent = pcnt;
1077 return 0;
1078}
1079
1080static int process_attr(struct perf_tool *tool __maybe_unused,
1081 union perf_event *event,
1082 struct evlist **pevlist)
1083{
1084 u64 sample_type;
1085 int err;
1086
1087 err = perf_event__process_attr(tool, event, pevlist);
1088 if (err)
1089 return err;
1090
1091 /*
1092 * Check if we need to enable callchains based
1093 * on events sample_type.
1094 */
1095 sample_type = evlist__combined_sample_type(*pevlist);
1096 callchain_param_setup(sample_type);
1097 return 0;
1098}
1099
1100int cmd_report(int argc, const char **argv)
1101{
1102 struct perf_session *session;
1103 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1104 struct stat st;
1105 bool has_br_stack = false;
1106 int branch_mode = -1;
1107 int last_key = 0;
1108 bool branch_call_mode = false;
1109#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1110 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1111 CALLCHAIN_REPORT_HELP
1112 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1113 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1114 const char * const report_usage[] = {
1115 "perf report [<options>]",
1116 NULL
1117 };
1118 struct report report = {
1119 .tool = {
1120 .sample = process_sample_event,
1121 .mmap = perf_event__process_mmap,
1122 .mmap2 = perf_event__process_mmap2,
1123 .comm = perf_event__process_comm,
1124 .namespaces = perf_event__process_namespaces,
1125 .cgroup = perf_event__process_cgroup,
1126 .exit = perf_event__process_exit,
1127 .fork = perf_event__process_fork,
1128 .lost = perf_event__process_lost,
1129 .read = process_read_event,
1130 .attr = process_attr,
1131 .tracing_data = perf_event__process_tracing_data,
1132 .build_id = perf_event__process_build_id,
1133 .id_index = perf_event__process_id_index,
1134 .auxtrace_info = perf_event__process_auxtrace_info,
1135 .auxtrace = perf_event__process_auxtrace,
1136 .event_update = perf_event__process_event_update,
1137 .feature = process_feature_event,
1138 .ordered_events = true,
1139 .ordering_requires_timestamps = true,
1140 },
1141 .max_stack = PERF_MAX_STACK_DEPTH,
1142 .pretty_printing_style = "normal",
1143 .socket_filter = -1,
1144 .annotation_opts = annotation__default_options,
1145 };
1146 const struct option options[] = {
1147 OPT_STRING('i', "input", &input_name, "file",
1148 "input file name"),
1149 OPT_INCR('v', "verbose", &verbose,
1150 "be more verbose (show symbol address, etc)"),
1151 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1152 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1153 "dump raw trace in ASCII"),
1154 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1155 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1156 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1157 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1158 "file", "vmlinux pathname"),
1159 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1160 "don't load vmlinux even if found"),
1161 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1162 "file", "kallsyms pathname"),
1163 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1164 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1165 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1166 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1167 "Show a column with the number of samples"),
1168 OPT_BOOLEAN('T', "threads", &report.show_threads,
1169 "Show per-thread event counters"),
1170 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1171 "pretty printing style key: normal raw"),
1172 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1173 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1174 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1175 "Use the stdio interface"),
1176 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1177 OPT_BOOLEAN(0, "header-only", &report.header_only,
1178 "Show only data header."),
1179 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1180 sort_help("sort by key(s):")),
1181 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1182 sort_help("output field(s): overhead period sample ")),
1183 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1184 "Show sample percentage for different cpu modes"),
1185 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1186 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1187 OPT_STRING('p', "parent", &parent_pattern, "regex",
1188 "regex filter to identify parent, see: '--sort parent'"),
1189 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1190 "Only display entries with parent-match"),
1191 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1192 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1193 report_callchain_help, &report_parse_callchain_opt,
1194 callchain_default_opt),
1195 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1196 "Accumulate callchains of children and show total overhead as well. "
1197 "Enabled by default, use --no-children to disable."),
1198 OPT_INTEGER(0, "max-stack", &report.max_stack,
1199 "Set the maximum stack depth when parsing the callchain, "
1200 "anything beyond the specified depth will be ignored. "
1201 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1202 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1203 "alias for inverted call graph"),
1204 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1205 "ignore callees of these functions in call graphs",
1206 report_parse_ignore_callees_opt),
1207 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1208 "only consider symbols in these dsos"),
1209 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1210 "only consider symbols in these comms"),
1211 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1212 "only consider symbols in these pids"),
1213 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1214 "only consider symbols in these tids"),
1215 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1216 "only consider these symbols"),
1217 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1218 "only show symbols that (partially) match with this filter"),
1219 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1220 "width[,width...]",
1221 "don't try to adjust column width, use these fixed values"),
1222 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1223 "separator for columns, no spaces will be added between "
1224 "columns '.' is reserved."),
1225 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1226 "Only display entries resolved to a symbol"),
1227 OPT_CALLBACK(0, "symfs", NULL, "directory",
1228 "Look for files with symbols relative to this directory",
1229 symbol__config_symfs),
1230 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1231 "list of cpus to profile"),
1232 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1233 "Display extended information about perf.data file"),
1234 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1235 "Interleave source code with assembly code (default)"),
1236 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1237 "Display raw encoding of assembly instructions (default)"),
1238 OPT_STRING('M', "disassembler-style", &report.annotation_opts.disassembler_style, "disassembler style",
1239 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1240 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix",
1241 "Add prefix to source file path names in programs (with --prefix-strip)"),
1242 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N",
1243 "Strip first N entries of source file path name in programs (with --prefix)"),
1244 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1245 "Show a column with the sum of periods"),
1246 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1247 "Show event group information together"),
1248 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1249 "Sort the output by the event at the index n in group. "
1250 "If n is invalid, sort by the first event. "
1251 "WARNING: should be used on grouped events."),
1252 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1253 "use branch records for per branch histogram filling",
1254 parse_branch_mode),
1255 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1256 "add last branch records to call history"),
1257 OPT_STRING(0, "objdump", &report.annotation_opts.objdump_path, "path",
1258 "objdump binary to use for disassembly and annotations"),
1259 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1260 "Disable symbol demangling"),
1261 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1262 "Enable kernel symbol demangling"),
1263 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1264 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1265 "Number of samples to save per histogram entry for individual browsing"),
1266 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1267 "Don't show entries under that percent", parse_percent_limit),
1268 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1269 "how to display percentage of filtered entries", parse_filter_percentage),
1270 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1271 "Instruction Tracing options\n" ITRACE_HELP,
1272 itrace_parse_synth_opts),
1273 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1274 "Show full source file name path for source lines"),
1275 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1276 "Show callgraph from reference event"),
1277 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1278 "Enable LBR callgraph stitching approach"),
1279 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1280 "only show processor socket that match with this filter"),
1281 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1282 "Show raw trace event output (do not use print fmt or plugins)"),
1283 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1284 "Show entries in a hierarchy"),
1285 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1286 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1287 stdio__config_color, "always"),
1288 OPT_STRING(0, "time", &report.time_str, "str",
1289 "Time span of interest (start,stop)"),
1290 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1291 "Show inline function"),
1292 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1293 "Set percent type local/global-period/hits",
1294 annotate_parse_percent_type),
1295 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1296 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1297 "Set time quantum for time sort key (default 100ms)",
1298 parse_time_quantum),
1299 OPTS_EVSWITCH(&report.evswitch),
1300 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1301 "Sort all blocks by 'Sampled Cycles%'"),
1302 OPT_END()
1303 };
1304 struct perf_data data = {
1305 .mode = PERF_DATA_MODE_READ,
1306 };
1307 int ret = hists__init();
1308 char sort_tmp[128];
1309
1310 if (ret < 0)
1311 return ret;
1312
1313 ret = perf_config(report__config, &report);
1314 if (ret)
1315 return ret;
1316
1317 argc = parse_options(argc, argv, options, report_usage, 0);
1318 if (argc) {
1319 /*
1320 * Special case: if there's an argument left then assume that
1321 * it's a symbol filter:
1322 */
1323 if (argc > 1)
1324 usage_with_options(report_usage, options);
1325
1326 report.symbol_filter_str = argv[0];
1327 }
1328
1329 if (annotate_check_args(&report.annotation_opts) < 0)
1330 return -EINVAL;
1331
1332 if (report.mmaps_mode)
1333 report.tasks_mode = true;
1334
1335 if (dump_trace)
1336 report.tool.ordered_events = false;
1337
1338 if (quiet)
1339 perf_quiet_option();
1340
1341 if (symbol_conf.vmlinux_name &&
1342 access(symbol_conf.vmlinux_name, R_OK)) {
1343 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1344 return -EINVAL;
1345 }
1346 if (symbol_conf.kallsyms_name &&
1347 access(symbol_conf.kallsyms_name, R_OK)) {
1348 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1349 return -EINVAL;
1350 }
1351
1352 if (report.inverted_callchain)
1353 callchain_param.order = ORDER_CALLER;
1354 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1355 callchain_param.order = ORDER_CALLER;
1356
1357 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1358 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1359 report.max_stack = itrace_synth_opts.callchain_sz;
1360
1361 if (!input_name || !strlen(input_name)) {
1362 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1363 input_name = "-";
1364 else
1365 input_name = "perf.data";
1366 }
1367
1368 data.path = input_name;
1369 data.force = symbol_conf.force;
1370
1371repeat:
1372 session = perf_session__new(&data, false, &report.tool);
1373 if (IS_ERR(session))
1374 return PTR_ERR(session);
1375
1376 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1377 if (ret)
1378 return ret;
1379
1380 if (zstd_init(&(session->zstd_data), 0) < 0)
1381 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1382
1383 if (report.queue_size) {
1384 ordered_events__set_alloc_size(&session->ordered_events,
1385 report.queue_size);
1386 }
1387
1388 session->itrace_synth_opts = &itrace_synth_opts;
1389
1390 report.session = session;
1391
1392 has_br_stack = perf_header__has_feat(&session->header,
1393 HEADER_BRANCH_STACK);
1394 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1395 has_br_stack = false;
1396
1397 setup_forced_leader(&report, session->evlist);
1398
1399 if (symbol_conf.group_sort_idx && !session->evlist->nr_groups) {
1400 parse_options_usage(NULL, options, "group-sort-idx", 0);
1401 ret = -EINVAL;
1402 goto error;
1403 }
1404
1405 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1406 has_br_stack = true;
1407
1408 if (has_br_stack && branch_call_mode)
1409 symbol_conf.show_branchflag_count = true;
1410
1411 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1412
1413 /*
1414 * Branch mode is a tristate:
1415 * -1 means default, so decide based on the file having branch data.
1416 * 0/1 means the user chose a mode.
1417 */
1418 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1419 !branch_call_mode) {
1420 sort__mode = SORT_MODE__BRANCH;
1421 symbol_conf.cumulate_callchain = false;
1422 }
1423 if (branch_call_mode) {
1424 callchain_param.key = CCKEY_ADDRESS;
1425 callchain_param.branch_callstack = true;
1426 symbol_conf.use_callchain = true;
1427 callchain_register_param(&callchain_param);
1428 if (sort_order == NULL)
1429 sort_order = "srcline,symbol,dso";
1430 }
1431
1432 if (report.mem_mode) {
1433 if (sort__mode == SORT_MODE__BRANCH) {
1434 pr_err("branch and mem mode incompatible\n");
1435 goto error;
1436 }
1437 sort__mode = SORT_MODE__MEMORY;
1438 symbol_conf.cumulate_callchain = false;
1439 }
1440
1441 if (symbol_conf.report_hierarchy) {
1442 /* disable incompatible options */
1443 symbol_conf.cumulate_callchain = false;
1444
1445 if (field_order) {
1446 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1447 parse_options_usage(report_usage, options, "F", 1);
1448 parse_options_usage(NULL, options, "hierarchy", 0);
1449 goto error;
1450 }
1451
1452 perf_hpp_list.need_collapse = true;
1453 }
1454
1455 if (report.use_stdio)
1456 use_browser = 0;
1457 else if (report.use_tui)
1458 use_browser = 1;
1459 else if (report.use_gtk)
1460 use_browser = 2;
1461
1462 /* Force tty output for header output and per-thread stat. */
1463 if (report.header || report.header_only || report.show_threads)
1464 use_browser = 0;
1465 if (report.header || report.header_only)
1466 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1467 if (report.show_full_info)
1468 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1469 if (report.stats_mode || report.tasks_mode)
1470 use_browser = 0;
1471 if (report.stats_mode && report.tasks_mode) {
1472 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1473 goto error;
1474 }
1475
1476 if (report.total_cycles_mode) {
1477 if (sort__mode != SORT_MODE__BRANCH)
1478 report.total_cycles_mode = false;
1479 else
1480 sort_order = NULL;
1481 }
1482
1483 if (strcmp(input_name, "-") != 0)
1484 setup_browser(true);
1485 else
1486 use_browser = 0;
1487
1488 if (sort_order && strstr(sort_order, "ipc")) {
1489 parse_options_usage(report_usage, options, "s", 1);
1490 goto error;
1491 }
1492
1493 if (sort_order && strstr(sort_order, "symbol")) {
1494 if (sort__mode == SORT_MODE__BRANCH) {
1495 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1496 sort_order, "ipc_lbr");
1497 report.symbol_ipc = true;
1498 } else {
1499 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1500 sort_order, "ipc_null");
1501 }
1502
1503 sort_order = sort_tmp;
1504 }
1505
1506 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1507 (setup_sorting(session->evlist) < 0)) {
1508 if (sort_order)
1509 parse_options_usage(report_usage, options, "s", 1);
1510 if (field_order)
1511 parse_options_usage(sort_order ? NULL : report_usage,
1512 options, "F", 1);
1513 goto error;
1514 }
1515
1516 if ((report.header || report.header_only) && !quiet) {
1517 perf_session__fprintf_info(session, stdout,
1518 report.show_full_info);
1519 if (report.header_only) {
1520 ret = 0;
1521 goto error;
1522 }
1523 } else if (use_browser == 0 && !quiet &&
1524 !report.stats_mode && !report.tasks_mode) {
1525 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1526 stdout);
1527 }
1528
1529 /*
1530 * Only in the TUI browser we are doing integrated annotation,
1531 * so don't allocate extra space that won't be used in the stdio
1532 * implementation.
1533 */
1534 if (ui__has_annotation() || report.symbol_ipc ||
1535 report.total_cycles_mode) {
1536 ret = symbol__annotation_init();
1537 if (ret < 0)
1538 goto error;
1539 /*
1540 * For searching by name on the "Browse map details".
1541 * providing it only in verbose mode not to bloat too
1542 * much struct symbol.
1543 */
1544 if (verbose > 0) {
1545 /*
1546 * XXX: Need to provide a less kludgy way to ask for
1547 * more space per symbol, the u32 is for the index on
1548 * the ui browser.
1549 * See symbol__browser_index.
1550 */
1551 symbol_conf.priv_size += sizeof(u32);
1552 symbol_conf.sort_by_name = true;
1553 }
1554 annotation_config__init(&report.annotation_opts);
1555 }
1556
1557 if (symbol__init(&session->header.env) < 0)
1558 goto error;
1559
1560 if (report.time_str) {
1561 ret = perf_time__parse_for_ranges(report.time_str, session,
1562 &report.ptime_range,
1563 &report.range_size,
1564 &report.range_num);
1565 if (ret < 0)
1566 goto error;
1567
1568 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1569 report.ptime_range,
1570 report.range_num);
1571 }
1572
1573 if (session->tevent.pevent &&
1574 tep_set_function_resolver(session->tevent.pevent,
1575 machine__resolve_kernel_addr,
1576 &session->machines.host) < 0) {
1577 pr_err("%s: failed to set libtraceevent function resolver\n",
1578 __func__);
1579 return -1;
1580 }
1581
1582 sort__setup_elide(stdout);
1583
1584 ret = __cmd_report(&report);
1585 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1586 perf_session__delete(session);
1587 last_key = K_SWITCH_INPUT_DATA;
1588 goto repeat;
1589 } else
1590 ret = 0;
1591
1592error:
1593 if (report.ptime_range) {
1594 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1595 zfree(&report.ptime_range);
1596 }
1597
1598 if (report.block_reports) {
1599 block_info__free_report(report.block_reports,
1600 report.nr_block_reports);
1601 report.block_reports = NULL;
1602 }
1603
1604 zstd_fini(&(session->zstd_data));
1605 perf_session__delete(session);
1606 return ret;
1607}