Loading...
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/util.h"
12#include "util/config.h"
13
14#include "util/annotate.h"
15#include "util/color.h"
16#include <linux/list.h>
17#include <linux/rbtree.h>
18#include <linux/err.h>
19#include "util/symbol.h"
20#include "util/callchain.h"
21#include "util/values.h"
22
23#include "perf.h"
24#include "util/debug.h"
25#include "util/evlist.h"
26#include "util/evsel.h"
27#include "util/header.h"
28#include "util/session.h"
29#include "util/tool.h"
30
31#include <subcmd/parse-options.h>
32#include <subcmd/exec-cmd.h>
33#include "util/parse-events.h"
34
35#include "util/thread.h"
36#include "util/sort.h"
37#include "util/hist.h"
38#include "util/data.h"
39#include "arch/common.h"
40#include "util/time-utils.h"
41#include "util/auxtrace.h"
42#include "util/units.h"
43#include "util/branch.h"
44
45#include <dlfcn.h>
46#include <errno.h>
47#include <inttypes.h>
48#include <regex.h>
49#include <signal.h>
50#include <linux/bitmap.h>
51#include <linux/stringify.h>
52#include <sys/types.h>
53#include <sys/stat.h>
54#include <unistd.h>
55#include <linux/mman.h>
56
57struct report {
58 struct perf_tool tool;
59 struct perf_session *session;
60 bool use_tui, use_gtk, use_stdio;
61 bool show_full_info;
62 bool show_threads;
63 bool inverted_callchain;
64 bool mem_mode;
65 bool stats_mode;
66 bool tasks_mode;
67 bool mmaps_mode;
68 bool header;
69 bool header_only;
70 bool nonany_branch_mode;
71 bool group_set;
72 int max_stack;
73 struct perf_read_values show_threads_values;
74 const char *pretty_printing_style;
75 const char *cpu_list;
76 const char *symbol_filter_str;
77 const char *time_str;
78 struct perf_time_interval *ptime_range;
79 int range_size;
80 int range_num;
81 float min_percent;
82 u64 nr_entries;
83 u64 queue_size;
84 int socket_filter;
85 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
86 struct branch_type_stat brtype_stat;
87};
88
89static int report__config(const char *var, const char *value, void *cb)
90{
91 struct report *rep = cb;
92
93 if (!strcmp(var, "report.group")) {
94 symbol_conf.event_group = perf_config_bool(var, value);
95 return 0;
96 }
97 if (!strcmp(var, "report.percent-limit")) {
98 double pcnt = strtof(value, NULL);
99
100 rep->min_percent = pcnt;
101 callchain_param.min_percent = pcnt;
102 return 0;
103 }
104 if (!strcmp(var, "report.children")) {
105 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
106 return 0;
107 }
108 if (!strcmp(var, "report.queue-size"))
109 return perf_config_u64(&rep->queue_size, var, value);
110
111 if (!strcmp(var, "report.sort_order")) {
112 default_sort_order = strdup(value);
113 return 0;
114 }
115
116 return 0;
117}
118
119static int hist_iter__report_callback(struct hist_entry_iter *iter,
120 struct addr_location *al, bool single,
121 void *arg)
122{
123 int err = 0;
124 struct report *rep = arg;
125 struct hist_entry *he = iter->he;
126 struct perf_evsel *evsel = iter->evsel;
127 struct perf_sample *sample = iter->sample;
128 struct mem_info *mi;
129 struct branch_info *bi;
130
131 if (!ui__has_annotation())
132 return 0;
133
134 hist__account_cycles(sample->branch_stack, al, sample,
135 rep->nonany_branch_mode);
136
137 if (sort__mode == SORT_MODE__BRANCH) {
138 bi = he->branch_info;
139 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel->idx);
140 if (err)
141 goto out;
142
143 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel->idx);
144
145 } else if (rep->mem_mode) {
146 mi = he->mem_info;
147 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel->idx);
148 if (err)
149 goto out;
150
151 err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr);
152
153 } else if (symbol_conf.cumulate_callchain) {
154 if (single)
155 err = hist_entry__inc_addr_samples(he, sample, evsel->idx,
156 al->addr);
157 } else {
158 err = hist_entry__inc_addr_samples(he, sample, evsel->idx, al->addr);
159 }
160
161out:
162 return err;
163}
164
165static int hist_iter__branch_callback(struct hist_entry_iter *iter,
166 struct addr_location *al __maybe_unused,
167 bool single __maybe_unused,
168 void *arg)
169{
170 struct hist_entry *he = iter->he;
171 struct report *rep = arg;
172 struct branch_info *bi;
173 struct perf_sample *sample = iter->sample;
174 struct perf_evsel *evsel = iter->evsel;
175 int err;
176
177 if (!ui__has_annotation())
178 return 0;
179
180 hist__account_cycles(sample->branch_stack, al, sample,
181 rep->nonany_branch_mode);
182
183 bi = he->branch_info;
184 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel->idx);
185 if (err)
186 goto out;
187
188 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel->idx);
189
190 branch_type_count(&rep->brtype_stat, &bi->flags,
191 bi->from.addr, bi->to.addr);
192
193out:
194 return err;
195}
196
197/*
198 * Events in data file are not collect in groups, but we still want
199 * the group display. Set the artificial group and set the leader's
200 * forced_leader flag to notify the display code.
201 */
202static void setup_forced_leader(struct report *report,
203 struct perf_evlist *evlist)
204{
205 if (report->group_set && !evlist->nr_groups) {
206 struct perf_evsel *leader = perf_evlist__first(evlist);
207
208 perf_evlist__set_leader(evlist);
209 leader->forced_leader = true;
210 }
211}
212
213static int process_feature_event(struct perf_tool *tool,
214 union perf_event *event,
215 struct perf_session *session __maybe_unused)
216{
217 struct report *rep = container_of(tool, struct report, tool);
218
219 if (event->feat.feat_id < HEADER_LAST_FEATURE)
220 return perf_event__process_feature(tool, event, session);
221
222 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
223 pr_err("failed: wrong feature ID: %" PRIu64 "\n",
224 event->feat.feat_id);
225 return -1;
226 }
227
228 /*
229 * All features are received, we can force the
230 * group if needed.
231 */
232 setup_forced_leader(rep, session->evlist);
233 return 0;
234}
235
236static int process_sample_event(struct perf_tool *tool,
237 union perf_event *event,
238 struct perf_sample *sample,
239 struct perf_evsel *evsel,
240 struct machine *machine)
241{
242 struct report *rep = container_of(tool, struct report, tool);
243 struct addr_location al;
244 struct hist_entry_iter iter = {
245 .evsel = evsel,
246 .sample = sample,
247 .hide_unresolved = symbol_conf.hide_unresolved,
248 .add_entry_cb = hist_iter__report_callback,
249 };
250 int ret = 0;
251
252 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
253 sample->time)) {
254 return 0;
255 }
256
257 if (machine__resolve(machine, &al, sample) < 0) {
258 pr_debug("problem processing %d event, skipping it.\n",
259 event->header.type);
260 return -1;
261 }
262
263 if (symbol_conf.hide_unresolved && al.sym == NULL)
264 goto out_put;
265
266 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
267 goto out_put;
268
269 if (sort__mode == SORT_MODE__BRANCH) {
270 /*
271 * A non-synthesized event might not have a branch stack if
272 * branch stacks have been synthesized (using itrace options).
273 */
274 if (!sample->branch_stack)
275 goto out_put;
276
277 iter.add_entry_cb = hist_iter__branch_callback;
278 iter.ops = &hist_iter_branch;
279 } else if (rep->mem_mode) {
280 iter.ops = &hist_iter_mem;
281 } else if (symbol_conf.cumulate_callchain) {
282 iter.ops = &hist_iter_cumulative;
283 } else {
284 iter.ops = &hist_iter_normal;
285 }
286
287 if (al.map != NULL)
288 al.map->dso->hit = 1;
289
290 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
291 if (ret < 0)
292 pr_debug("problem adding hist entry, skipping event\n");
293out_put:
294 addr_location__put(&al);
295 return ret;
296}
297
298static int process_read_event(struct perf_tool *tool,
299 union perf_event *event,
300 struct perf_sample *sample __maybe_unused,
301 struct perf_evsel *evsel,
302 struct machine *machine __maybe_unused)
303{
304 struct report *rep = container_of(tool, struct report, tool);
305
306 if (rep->show_threads) {
307 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
308 int err = perf_read_values_add_value(&rep->show_threads_values,
309 event->read.pid, event->read.tid,
310 evsel->idx,
311 name,
312 event->read.value);
313
314 if (err)
315 return err;
316 }
317
318 return 0;
319}
320
321/* For pipe mode, sample_type is not currently set */
322static int report__setup_sample_type(struct report *rep)
323{
324 struct perf_session *session = rep->session;
325 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
326 bool is_pipe = perf_data__is_pipe(session->data);
327
328 if (session->itrace_synth_opts->callchain ||
329 (!is_pipe &&
330 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
331 !session->itrace_synth_opts->set))
332 sample_type |= PERF_SAMPLE_CALLCHAIN;
333
334 if (session->itrace_synth_opts->last_branch)
335 sample_type |= PERF_SAMPLE_BRANCH_STACK;
336
337 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
338 if (perf_hpp_list.parent) {
339 ui__error("Selected --sort parent, but no "
340 "callchain data. Did you call "
341 "'perf record' without -g?\n");
342 return -EINVAL;
343 }
344 if (symbol_conf.use_callchain &&
345 !symbol_conf.show_branchflag_count) {
346 ui__error("Selected -g or --branch-history.\n"
347 "But no callchain or branch data.\n"
348 "Did you call 'perf record' without -g or -b?\n");
349 return -1;
350 }
351 } else if (!callchain_param.enabled &&
352 callchain_param.mode != CHAIN_NONE &&
353 !symbol_conf.use_callchain) {
354 symbol_conf.use_callchain = true;
355 if (callchain_register_param(&callchain_param) < 0) {
356 ui__error("Can't register callchain params.\n");
357 return -EINVAL;
358 }
359 }
360
361 if (symbol_conf.cumulate_callchain) {
362 /* Silently ignore if callchain is missing */
363 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
364 symbol_conf.cumulate_callchain = false;
365 perf_hpp__cancel_cumulate();
366 }
367 }
368
369 if (sort__mode == SORT_MODE__BRANCH) {
370 if (!is_pipe &&
371 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
372 ui__error("Selected -b but no branch data. "
373 "Did you call perf record without -b?\n");
374 return -1;
375 }
376 }
377
378 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
379 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
380 (sample_type & PERF_SAMPLE_STACK_USER)) {
381 callchain_param.record_mode = CALLCHAIN_DWARF;
382 dwarf_callchain_users = true;
383 } else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
384 callchain_param.record_mode = CALLCHAIN_LBR;
385 else
386 callchain_param.record_mode = CALLCHAIN_FP;
387 }
388
389 /* ??? handle more cases than just ANY? */
390 if (!(perf_evlist__combined_branch_type(session->evlist) &
391 PERF_SAMPLE_BRANCH_ANY))
392 rep->nonany_branch_mode = true;
393
394 return 0;
395}
396
397static void sig_handler(int sig __maybe_unused)
398{
399 session_done = 1;
400}
401
402static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
403 const char *evname, FILE *fp)
404{
405 size_t ret;
406 char unit;
407 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
408 u64 nr_events = hists->stats.total_period;
409 struct perf_evsel *evsel = hists_to_evsel(hists);
410 char buf[512];
411 size_t size = sizeof(buf);
412 int socked_id = hists->socket_filter;
413
414 if (quiet)
415 return 0;
416
417 if (symbol_conf.filter_relative) {
418 nr_samples = hists->stats.nr_non_filtered_samples;
419 nr_events = hists->stats.total_non_filtered_period;
420 }
421
422 if (perf_evsel__is_group_event(evsel)) {
423 struct perf_evsel *pos;
424
425 perf_evsel__group_desc(evsel, buf, size);
426 evname = buf;
427
428 for_each_group_member(pos, evsel) {
429 const struct hists *pos_hists = evsel__hists(pos);
430
431 if (symbol_conf.filter_relative) {
432 nr_samples += pos_hists->stats.nr_non_filtered_samples;
433 nr_events += pos_hists->stats.total_non_filtered_period;
434 } else {
435 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
436 nr_events += pos_hists->stats.total_period;
437 }
438 }
439 }
440
441 nr_samples = convert_unit(nr_samples, &unit);
442 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
443 if (evname != NULL) {
444 ret += fprintf(fp, " of event%s '%s'",
445 evsel->nr_members > 1 ? "s" : "", evname);
446 }
447
448 if (rep->time_str)
449 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
450
451 if (symbol_conf.show_ref_callgraph &&
452 strstr(evname, "call-graph=no")) {
453 ret += fprintf(fp, ", show reference callgraph");
454 }
455
456 if (rep->mem_mode) {
457 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
458 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
459 } else
460 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
461
462 if (socked_id > -1)
463 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
464
465 return ret + fprintf(fp, "\n#\n");
466}
467
468static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
469 struct report *rep,
470 const char *help)
471{
472 struct perf_evsel *pos;
473
474 if (!quiet) {
475 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
476 evlist->stats.total_lost_samples);
477 }
478
479 evlist__for_each_entry(evlist, pos) {
480 struct hists *hists = evsel__hists(pos);
481 const char *evname = perf_evsel__name(pos);
482
483 if (symbol_conf.event_group &&
484 !perf_evsel__is_group_leader(pos))
485 continue;
486
487 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
488 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
489 symbol_conf.use_callchain ||
490 symbol_conf.show_branchflag_count);
491 fprintf(stdout, "\n\n");
492 }
493
494 if (!quiet)
495 fprintf(stdout, "#\n# (%s)\n#\n", help);
496
497 if (rep->show_threads) {
498 bool style = !strcmp(rep->pretty_printing_style, "raw");
499 perf_read_values_display(stdout, &rep->show_threads_values,
500 style);
501 perf_read_values_destroy(&rep->show_threads_values);
502 }
503
504 if (sort__mode == SORT_MODE__BRANCH)
505 branch_type_stat_display(stdout, &rep->brtype_stat);
506
507 return 0;
508}
509
510static void report__warn_kptr_restrict(const struct report *rep)
511{
512 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
513 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
514
515 if (perf_evlist__exclude_kernel(rep->session->evlist))
516 return;
517
518 if (kernel_map == NULL ||
519 (kernel_map->dso->hit &&
520 (kernel_kmap->ref_reloc_sym == NULL ||
521 kernel_kmap->ref_reloc_sym->addr == 0))) {
522 const char *desc =
523 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
524 "can't be resolved.";
525
526 if (kernel_map) {
527 const struct dso *kdso = kernel_map->dso;
528 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
529 desc = "If some relocation was applied (e.g. "
530 "kexec) symbols may be misresolved.";
531 }
532 }
533
534 ui__warning(
535"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
536"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
537"Samples in kernel modules can't be resolved as well.\n\n",
538 desc);
539 }
540}
541
542static int report__gtk_browse_hists(struct report *rep, const char *help)
543{
544 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
545 struct hist_browser_timer *timer, float min_pcnt);
546
547 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
548
549 if (hist_browser == NULL) {
550 ui__error("GTK browser not found!\n");
551 return -1;
552 }
553
554 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
555}
556
557static int report__browse_hists(struct report *rep)
558{
559 int ret;
560 struct perf_session *session = rep->session;
561 struct perf_evlist *evlist = session->evlist;
562 const char *help = perf_tip(system_path(TIPDIR));
563
564 if (help == NULL) {
565 /* fallback for people who don't install perf ;-) */
566 help = perf_tip(DOCDIR);
567 if (help == NULL)
568 help = "Cannot load tips.txt file, please install perf!";
569 }
570
571 switch (use_browser) {
572 case 1:
573 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
574 rep->min_percent,
575 &session->header.env,
576 true);
577 /*
578 * Usually "ret" is the last pressed key, and we only
579 * care if the key notifies us to switch data file.
580 */
581 if (ret != K_SWITCH_INPUT_DATA)
582 ret = 0;
583 break;
584 case 2:
585 ret = report__gtk_browse_hists(rep, help);
586 break;
587 default:
588 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
589 break;
590 }
591
592 return ret;
593}
594
595static int report__collapse_hists(struct report *rep)
596{
597 struct ui_progress prog;
598 struct perf_evsel *pos;
599 int ret = 0;
600
601 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
602
603 evlist__for_each_entry(rep->session->evlist, pos) {
604 struct hists *hists = evsel__hists(pos);
605
606 if (pos->idx == 0)
607 hists->symbol_filter_str = rep->symbol_filter_str;
608
609 hists->socket_filter = rep->socket_filter;
610
611 ret = hists__collapse_resort(hists, &prog);
612 if (ret < 0)
613 break;
614
615 /* Non-group events are considered as leader */
616 if (symbol_conf.event_group &&
617 !perf_evsel__is_group_leader(pos)) {
618 struct hists *leader_hists = evsel__hists(pos->leader);
619
620 hists__match(leader_hists, hists);
621 hists__link(leader_hists, hists);
622 }
623 }
624
625 ui_progress__finish();
626 return ret;
627}
628
629static void report__output_resort(struct report *rep)
630{
631 struct ui_progress prog;
632 struct perf_evsel *pos;
633
634 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
635
636 evlist__for_each_entry(rep->session->evlist, pos)
637 perf_evsel__output_resort(pos, &prog);
638
639 ui_progress__finish();
640}
641
642static void stats_setup(struct report *rep)
643{
644 memset(&rep->tool, 0, sizeof(rep->tool));
645 rep->tool.no_warn = true;
646}
647
648static int stats_print(struct report *rep)
649{
650 struct perf_session *session = rep->session;
651
652 perf_session__fprintf_nr_events(session, stdout);
653 return 0;
654}
655
656static void tasks_setup(struct report *rep)
657{
658 memset(&rep->tool, 0, sizeof(rep->tool));
659 rep->tool.ordered_events = true;
660 if (rep->mmaps_mode) {
661 rep->tool.mmap = perf_event__process_mmap;
662 rep->tool.mmap2 = perf_event__process_mmap2;
663 }
664 rep->tool.comm = perf_event__process_comm;
665 rep->tool.exit = perf_event__process_exit;
666 rep->tool.fork = perf_event__process_fork;
667 rep->tool.no_warn = true;
668}
669
670struct task {
671 struct thread *thread;
672 struct list_head list;
673 struct list_head children;
674};
675
676static struct task *tasks_list(struct task *task, struct machine *machine)
677{
678 struct thread *parent_thread, *thread = task->thread;
679 struct task *parent_task;
680
681 /* Already listed. */
682 if (!list_empty(&task->list))
683 return NULL;
684
685 /* Last one in the chain. */
686 if (thread->ppid == -1)
687 return task;
688
689 parent_thread = machine__find_thread(machine, -1, thread->ppid);
690 if (!parent_thread)
691 return ERR_PTR(-ENOENT);
692
693 parent_task = thread__priv(parent_thread);
694 list_add_tail(&task->list, &parent_task->children);
695 return tasks_list(parent_task, machine);
696}
697
698static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
699{
700 size_t printed = 0;
701 struct rb_node *nd;
702
703 for (nd = rb_first(&maps->entries); nd; nd = rb_next(nd)) {
704 struct map *map = rb_entry(nd, struct map, rb_node);
705
706 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
707 indent, "", map->start, map->end,
708 map->prot & PROT_READ ? 'r' : '-',
709 map->prot & PROT_WRITE ? 'w' : '-',
710 map->prot & PROT_EXEC ? 'x' : '-',
711 map->flags & MAP_SHARED ? 's' : 'p',
712 map->pgoff,
713 map->ino, map->dso->name);
714 }
715
716 return printed;
717}
718
719static int map_groups__fprintf_task(struct map_groups *mg, int indent, FILE *fp)
720{
721 int printed = 0, i;
722 for (i = 0; i < MAP__NR_TYPES; ++i)
723 printed += maps__fprintf_task(&mg->maps[i], indent, fp);
724 return printed;
725}
726
727static void task__print_level(struct task *task, FILE *fp, int level)
728{
729 struct thread *thread = task->thread;
730 struct task *child;
731 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
732 thread->pid_, thread->tid, thread->ppid,
733 level, "");
734
735 fprintf(fp, "%s\n", thread__comm_str(thread));
736
737 map_groups__fprintf_task(thread->mg, comm_indent, fp);
738
739 if (!list_empty(&task->children)) {
740 list_for_each_entry(child, &task->children, list)
741 task__print_level(child, fp, level + 1);
742 }
743}
744
745static int tasks_print(struct report *rep, FILE *fp)
746{
747 struct perf_session *session = rep->session;
748 struct machine *machine = &session->machines.host;
749 struct task *tasks, *task;
750 unsigned int nr = 0, itask = 0, i;
751 struct rb_node *nd;
752 LIST_HEAD(list);
753
754 /*
755 * No locking needed while accessing machine->threads,
756 * because --tasks is single threaded command.
757 */
758
759 /* Count all the threads. */
760 for (i = 0; i < THREADS__TABLE_SIZE; i++)
761 nr += machine->threads[i].nr;
762
763 tasks = malloc(sizeof(*tasks) * nr);
764 if (!tasks)
765 return -ENOMEM;
766
767 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
768 struct threads *threads = &machine->threads[i];
769
770 for (nd = rb_first(&threads->entries); nd; nd = rb_next(nd)) {
771 task = tasks + itask++;
772
773 task->thread = rb_entry(nd, struct thread, rb_node);
774 INIT_LIST_HEAD(&task->children);
775 INIT_LIST_HEAD(&task->list);
776 thread__set_priv(task->thread, task);
777 }
778 }
779
780 /*
781 * Iterate every task down to the unprocessed parent
782 * and link all in task children list. Task with no
783 * parent is added into 'list'.
784 */
785 for (itask = 0; itask < nr; itask++) {
786 task = tasks + itask;
787
788 if (!list_empty(&task->list))
789 continue;
790
791 task = tasks_list(task, machine);
792 if (IS_ERR(task)) {
793 pr_err("Error: failed to process tasks\n");
794 free(tasks);
795 return PTR_ERR(task);
796 }
797
798 if (task)
799 list_add_tail(&task->list, &list);
800 }
801
802 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
803
804 list_for_each_entry(task, &list, list)
805 task__print_level(task, fp, 0);
806
807 free(tasks);
808 return 0;
809}
810
811static int __cmd_report(struct report *rep)
812{
813 int ret;
814 struct perf_session *session = rep->session;
815 struct perf_evsel *pos;
816 struct perf_data *data = session->data;
817
818 signal(SIGINT, sig_handler);
819
820 if (rep->cpu_list) {
821 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
822 rep->cpu_bitmap);
823 if (ret) {
824 ui__error("failed to set cpu bitmap\n");
825 return ret;
826 }
827 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
828 }
829
830 if (rep->show_threads) {
831 ret = perf_read_values_init(&rep->show_threads_values);
832 if (ret)
833 return ret;
834 }
835
836 ret = report__setup_sample_type(rep);
837 if (ret) {
838 /* report__setup_sample_type() already showed error message */
839 return ret;
840 }
841
842 if (rep->stats_mode)
843 stats_setup(rep);
844
845 if (rep->tasks_mode)
846 tasks_setup(rep);
847
848 ret = perf_session__process_events(session);
849 if (ret) {
850 ui__error("failed to process sample\n");
851 return ret;
852 }
853
854 if (rep->stats_mode)
855 return stats_print(rep);
856
857 if (rep->tasks_mode)
858 return tasks_print(rep, stdout);
859
860 report__warn_kptr_restrict(rep);
861
862 evlist__for_each_entry(session->evlist, pos)
863 rep->nr_entries += evsel__hists(pos)->nr_entries;
864
865 if (use_browser == 0) {
866 if (verbose > 3)
867 perf_session__fprintf(session, stdout);
868
869 if (verbose > 2)
870 perf_session__fprintf_dsos(session, stdout);
871
872 if (dump_trace) {
873 perf_session__fprintf_nr_events(session, stdout);
874 perf_evlist__fprintf_nr_events(session->evlist, stdout);
875 return 0;
876 }
877 }
878
879 ret = report__collapse_hists(rep);
880 if (ret) {
881 ui__error("failed to process hist entry\n");
882 return ret;
883 }
884
885 if (session_done())
886 return 0;
887
888 /*
889 * recalculate number of entries after collapsing since it
890 * might be changed during the collapse phase.
891 */
892 rep->nr_entries = 0;
893 evlist__for_each_entry(session->evlist, pos)
894 rep->nr_entries += evsel__hists(pos)->nr_entries;
895
896 if (rep->nr_entries == 0) {
897 ui__error("The %s file has no samples!\n", data->file.path);
898 return 0;
899 }
900
901 report__output_resort(rep);
902
903 return report__browse_hists(rep);
904}
905
906static int
907report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
908{
909 struct callchain_param *callchain = opt->value;
910
911 callchain->enabled = !unset;
912 /*
913 * --no-call-graph
914 */
915 if (unset) {
916 symbol_conf.use_callchain = false;
917 callchain->mode = CHAIN_NONE;
918 return 0;
919 }
920
921 return parse_callchain_report_opt(arg);
922}
923
924int
925report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
926 const char *arg, int unset __maybe_unused)
927{
928 if (arg) {
929 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
930 if (err) {
931 char buf[BUFSIZ];
932 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
933 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
934 return -1;
935 }
936 have_ignore_callees = 1;
937 }
938
939 return 0;
940}
941
942static int
943parse_branch_mode(const struct option *opt,
944 const char *str __maybe_unused, int unset)
945{
946 int *branch_mode = opt->value;
947
948 *branch_mode = !unset;
949 return 0;
950}
951
952static int
953parse_percent_limit(const struct option *opt, const char *str,
954 int unset __maybe_unused)
955{
956 struct report *rep = opt->value;
957 double pcnt = strtof(str, NULL);
958
959 rep->min_percent = pcnt;
960 callchain_param.min_percent = pcnt;
961 return 0;
962}
963
964#define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
965
966const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
967 CALLCHAIN_REPORT_HELP
968 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
969
970int cmd_report(int argc, const char **argv)
971{
972 struct perf_session *session;
973 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
974 struct stat st;
975 bool has_br_stack = false;
976 int branch_mode = -1;
977 bool branch_call_mode = false;
978 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
979 const char * const report_usage[] = {
980 "perf report [<options>]",
981 NULL
982 };
983 struct report report = {
984 .tool = {
985 .sample = process_sample_event,
986 .mmap = perf_event__process_mmap,
987 .mmap2 = perf_event__process_mmap2,
988 .comm = perf_event__process_comm,
989 .namespaces = perf_event__process_namespaces,
990 .exit = perf_event__process_exit,
991 .fork = perf_event__process_fork,
992 .lost = perf_event__process_lost,
993 .read = process_read_event,
994 .attr = perf_event__process_attr,
995 .tracing_data = perf_event__process_tracing_data,
996 .build_id = perf_event__process_build_id,
997 .id_index = perf_event__process_id_index,
998 .auxtrace_info = perf_event__process_auxtrace_info,
999 .auxtrace = perf_event__process_auxtrace,
1000 .feature = process_feature_event,
1001 .ordered_events = true,
1002 .ordering_requires_timestamps = true,
1003 },
1004 .max_stack = PERF_MAX_STACK_DEPTH,
1005 .pretty_printing_style = "normal",
1006 .socket_filter = -1,
1007 };
1008 const struct option options[] = {
1009 OPT_STRING('i', "input", &input_name, "file",
1010 "input file name"),
1011 OPT_INCR('v', "verbose", &verbose,
1012 "be more verbose (show symbol address, etc)"),
1013 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any message"),
1014 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1015 "dump raw trace in ASCII"),
1016 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1017 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1018 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1019 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1020 "file", "vmlinux pathname"),
1021 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1022 "don't load vmlinux even if found"),
1023 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1024 "file", "kallsyms pathname"),
1025 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1026 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1027 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1028 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1029 "Show a column with the number of samples"),
1030 OPT_BOOLEAN('T', "threads", &report.show_threads,
1031 "Show per-thread event counters"),
1032 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1033 "pretty printing style key: normal raw"),
1034 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1035 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1036 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1037 "Use the stdio interface"),
1038 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1039 OPT_BOOLEAN(0, "header-only", &report.header_only,
1040 "Show only data header."),
1041 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1042 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
1043 " Please refer the man page for the complete list."),
1044 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1045 "output field(s): overhead, period, sample plus all of sort keys"),
1046 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1047 "Show sample percentage for different cpu modes"),
1048 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1049 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1050 OPT_STRING('p', "parent", &parent_pattern, "regex",
1051 "regex filter to identify parent, see: '--sort parent'"),
1052 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1053 "Only display entries with parent-match"),
1054 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1055 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1056 report_callchain_help, &report_parse_callchain_opt,
1057 callchain_default_opt),
1058 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1059 "Accumulate callchains of children and show total overhead as well"),
1060 OPT_INTEGER(0, "max-stack", &report.max_stack,
1061 "Set the maximum stack depth when parsing the callchain, "
1062 "anything beyond the specified depth will be ignored. "
1063 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1064 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1065 "alias for inverted call graph"),
1066 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1067 "ignore callees of these functions in call graphs",
1068 report_parse_ignore_callees_opt),
1069 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1070 "only consider symbols in these dsos"),
1071 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1072 "only consider symbols in these comms"),
1073 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1074 "only consider symbols in these pids"),
1075 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1076 "only consider symbols in these tids"),
1077 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1078 "only consider these symbols"),
1079 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1080 "only show symbols that (partially) match with this filter"),
1081 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1082 "width[,width...]",
1083 "don't try to adjust column width, use these fixed values"),
1084 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1085 "separator for columns, no spaces will be added between "
1086 "columns '.' is reserved."),
1087 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1088 "Only display entries resolved to a symbol"),
1089 OPT_CALLBACK(0, "symfs", NULL, "directory",
1090 "Look for files with symbols relative to this directory",
1091 symbol__config_symfs),
1092 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1093 "list of cpus to profile"),
1094 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1095 "Display extended information about perf.data file"),
1096 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
1097 "Interleave source code with assembly code (default)"),
1098 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
1099 "Display raw encoding of assembly instructions (default)"),
1100 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1101 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1102 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1103 "Show a column with the sum of periods"),
1104 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1105 "Show event group information together"),
1106 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1107 "use branch records for per branch histogram filling",
1108 parse_branch_mode),
1109 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1110 "add last branch records to call history"),
1111 OPT_STRING(0, "objdump", &objdump_path, "path",
1112 "objdump binary to use for disassembly and annotations"),
1113 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1114 "Disable symbol demangling"),
1115 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1116 "Enable kernel symbol demangling"),
1117 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1118 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1119 "Don't show entries under that percent", parse_percent_limit),
1120 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1121 "how to display percentage of filtered entries", parse_filter_percentage),
1122 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1123 "Instruction Tracing options",
1124 itrace_parse_synth_opts),
1125 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1126 "Show full source file name path for source lines"),
1127 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1128 "Show callgraph from reference event"),
1129 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1130 "only show processor socket that match with this filter"),
1131 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1132 "Show raw trace event output (do not use print fmt or plugins)"),
1133 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1134 "Show entries in a hierarchy"),
1135 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1136 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1137 stdio__config_color, "always"),
1138 OPT_STRING(0, "time", &report.time_str, "str",
1139 "Time span of interest (start,stop)"),
1140 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1141 "Show inline function"),
1142 OPT_END()
1143 };
1144 struct perf_data data = {
1145 .mode = PERF_DATA_MODE_READ,
1146 };
1147 int ret = hists__init();
1148
1149 if (ret < 0)
1150 return ret;
1151
1152 ret = perf_config(report__config, &report);
1153 if (ret)
1154 return ret;
1155
1156 argc = parse_options(argc, argv, options, report_usage, 0);
1157 if (argc) {
1158 /*
1159 * Special case: if there's an argument left then assume that
1160 * it's a symbol filter:
1161 */
1162 if (argc > 1)
1163 usage_with_options(report_usage, options);
1164
1165 report.symbol_filter_str = argv[0];
1166 }
1167
1168 if (report.mmaps_mode)
1169 report.tasks_mode = true;
1170
1171 if (quiet)
1172 perf_quiet_option();
1173
1174 if (symbol_conf.vmlinux_name &&
1175 access(symbol_conf.vmlinux_name, R_OK)) {
1176 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
1177 return -EINVAL;
1178 }
1179 if (symbol_conf.kallsyms_name &&
1180 access(symbol_conf.kallsyms_name, R_OK)) {
1181 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
1182 return -EINVAL;
1183 }
1184
1185 if (report.inverted_callchain)
1186 callchain_param.order = ORDER_CALLER;
1187 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1188 callchain_param.order = ORDER_CALLER;
1189
1190 if (itrace_synth_opts.callchain &&
1191 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1192 report.max_stack = itrace_synth_opts.callchain_sz;
1193
1194 if (!input_name || !strlen(input_name)) {
1195 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1196 input_name = "-";
1197 else
1198 input_name = "perf.data";
1199 }
1200
1201 data.file.path = input_name;
1202 data.force = symbol_conf.force;
1203
1204repeat:
1205 session = perf_session__new(&data, false, &report.tool);
1206 if (session == NULL)
1207 return -1;
1208
1209 if (report.queue_size) {
1210 ordered_events__set_alloc_size(&session->ordered_events,
1211 report.queue_size);
1212 }
1213
1214 session->itrace_synth_opts = &itrace_synth_opts;
1215
1216 report.session = session;
1217
1218 has_br_stack = perf_header__has_feat(&session->header,
1219 HEADER_BRANCH_STACK);
1220
1221 setup_forced_leader(&report, session->evlist);
1222
1223 if (itrace_synth_opts.last_branch)
1224 has_br_stack = true;
1225
1226 if (has_br_stack && branch_call_mode)
1227 symbol_conf.show_branchflag_count = true;
1228
1229 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1230
1231 /*
1232 * Branch mode is a tristate:
1233 * -1 means default, so decide based on the file having branch data.
1234 * 0/1 means the user chose a mode.
1235 */
1236 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1237 !branch_call_mode) {
1238 sort__mode = SORT_MODE__BRANCH;
1239 symbol_conf.cumulate_callchain = false;
1240 }
1241 if (branch_call_mode) {
1242 callchain_param.key = CCKEY_ADDRESS;
1243 callchain_param.branch_callstack = 1;
1244 symbol_conf.use_callchain = true;
1245 callchain_register_param(&callchain_param);
1246 if (sort_order == NULL)
1247 sort_order = "srcline,symbol,dso";
1248 }
1249
1250 if (report.mem_mode) {
1251 if (sort__mode == SORT_MODE__BRANCH) {
1252 pr_err("branch and mem mode incompatible\n");
1253 goto error;
1254 }
1255 sort__mode = SORT_MODE__MEMORY;
1256 symbol_conf.cumulate_callchain = false;
1257 }
1258
1259 if (symbol_conf.report_hierarchy) {
1260 /* disable incompatible options */
1261 symbol_conf.cumulate_callchain = false;
1262
1263 if (field_order) {
1264 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1265 parse_options_usage(report_usage, options, "F", 1);
1266 parse_options_usage(NULL, options, "hierarchy", 0);
1267 goto error;
1268 }
1269
1270 perf_hpp_list.need_collapse = true;
1271 }
1272
1273 if (report.use_stdio)
1274 use_browser = 0;
1275 else if (report.use_tui)
1276 use_browser = 1;
1277 else if (report.use_gtk)
1278 use_browser = 2;
1279
1280 /* Force tty output for header output and per-thread stat. */
1281 if (report.header || report.header_only || report.show_threads)
1282 use_browser = 0;
1283 if (report.header || report.header_only)
1284 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1285 if (report.show_full_info)
1286 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1287 if (report.stats_mode || report.tasks_mode)
1288 use_browser = 0;
1289 if (report.stats_mode && report.tasks_mode) {
1290 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1291 goto error;
1292 }
1293
1294 if (strcmp(input_name, "-") != 0)
1295 setup_browser(true);
1296 else
1297 use_browser = 0;
1298
1299 if (setup_sorting(session->evlist) < 0) {
1300 if (sort_order)
1301 parse_options_usage(report_usage, options, "s", 1);
1302 if (field_order)
1303 parse_options_usage(sort_order ? NULL : report_usage,
1304 options, "F", 1);
1305 goto error;
1306 }
1307
1308 if ((report.header || report.header_only) && !quiet) {
1309 perf_session__fprintf_info(session, stdout,
1310 report.show_full_info);
1311 if (report.header_only) {
1312 ret = 0;
1313 goto error;
1314 }
1315 } else if (use_browser == 0 && !quiet &&
1316 !report.stats_mode && !report.tasks_mode) {
1317 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1318 stdout);
1319 }
1320
1321 /*
1322 * Only in the TUI browser we are doing integrated annotation,
1323 * so don't allocate extra space that won't be used in the stdio
1324 * implementation.
1325 */
1326 if (ui__has_annotation()) {
1327 ret = symbol__annotation_init();
1328 if (ret < 0)
1329 goto error;
1330 /*
1331 * For searching by name on the "Browse map details".
1332 * providing it only in verbose mode not to bloat too
1333 * much struct symbol.
1334 */
1335 if (verbose > 0) {
1336 /*
1337 * XXX: Need to provide a less kludgy way to ask for
1338 * more space per symbol, the u32 is for the index on
1339 * the ui browser.
1340 * See symbol__browser_index.
1341 */
1342 symbol_conf.priv_size += sizeof(u32);
1343 symbol_conf.sort_by_name = true;
1344 }
1345 annotation_config__init();
1346 }
1347
1348 if (symbol__init(&session->header.env) < 0)
1349 goto error;
1350
1351 report.ptime_range = perf_time__range_alloc(report.time_str,
1352 &report.range_size);
1353 if (!report.ptime_range) {
1354 ret = -ENOMEM;
1355 goto error;
1356 }
1357
1358 if (perf_time__parse_str(report.ptime_range, report.time_str) != 0) {
1359 if (session->evlist->first_sample_time == 0 &&
1360 session->evlist->last_sample_time == 0) {
1361 pr_err("HINT: no first/last sample time found in perf data.\n"
1362 "Please use latest perf binary to execute 'perf record'\n"
1363 "(if '--buildid-all' is enabled, please set '--timestamp-boundary').\n");
1364 ret = -EINVAL;
1365 goto error;
1366 }
1367
1368 report.range_num = perf_time__percent_parse_str(
1369 report.ptime_range, report.range_size,
1370 report.time_str,
1371 session->evlist->first_sample_time,
1372 session->evlist->last_sample_time);
1373
1374 if (report.range_num < 0) {
1375 pr_err("Invalid time string\n");
1376 ret = -EINVAL;
1377 goto error;
1378 }
1379 } else {
1380 report.range_num = 1;
1381 }
1382
1383 if (session->tevent.pevent &&
1384 pevent_set_function_resolver(session->tevent.pevent,
1385 machine__resolve_kernel_addr,
1386 &session->machines.host) < 0) {
1387 pr_err("%s: failed to set libtraceevent function resolver\n",
1388 __func__);
1389 return -1;
1390 }
1391
1392 sort__setup_elide(stdout);
1393
1394 ret = __cmd_report(&report);
1395 if (ret == K_SWITCH_INPUT_DATA) {
1396 perf_session__delete(session);
1397 goto repeat;
1398 } else
1399 ret = 0;
1400
1401error:
1402 zfree(&report.ptime_range);
1403
1404 perf_session__delete(session);
1405 return ret;
1406}
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
12#include "util/annotate.h"
13#include "util/color.h"
14#include <linux/list.h>
15#include "util/cache.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
29#include "util/parse-options.h"
30#include "util/parse-events.h"
31
32#include "util/thread.h"
33#include "util/sort.h"
34#include "util/hist.h"
35
36#include <linux/bitmap.h>
37
38static char const *input_name = "perf.data";
39
40static bool force, use_tui, use_stdio;
41static bool hide_unresolved;
42static bool dont_use_callchains;
43
44static bool show_threads;
45static struct perf_read_values show_threads_values;
46
47static const char default_pretty_printing_style[] = "normal";
48static const char *pretty_printing_style = default_pretty_printing_style;
49
50static char callchain_default_opt[] = "fractal,0.5,callee";
51static bool inverted_callchain;
52static symbol_filter_t annotate_init;
53
54static const char *cpu_list;
55static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
56
57static int perf_session__add_hist_entry(struct perf_session *session,
58 struct addr_location *al,
59 struct perf_sample *sample,
60 struct perf_evsel *evsel)
61{
62 struct symbol *parent = NULL;
63 int err = 0;
64 struct hist_entry *he;
65
66 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
67 err = perf_session__resolve_callchain(session, al->thread,
68 sample->callchain, &parent);
69 if (err)
70 return err;
71 }
72
73 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
74 if (he == NULL)
75 return -ENOMEM;
76
77 if (symbol_conf.use_callchain) {
78 err = callchain_append(he->callchain, &session->callchain_cursor,
79 sample->period);
80 if (err)
81 return err;
82 }
83 /*
84 * Only in the newt browser we are doing integrated annotation,
85 * so we don't allocated the extra space needed because the stdio
86 * code will not use it.
87 */
88 if (al->sym != NULL && use_browser > 0) {
89 struct annotation *notes = symbol__annotation(he->ms.sym);
90
91 assert(evsel != NULL);
92
93 err = -ENOMEM;
94 if (notes->src == NULL &&
95 symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0)
96 goto out;
97
98 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
99 }
100
101 evsel->hists.stats.total_period += sample->period;
102 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
103out:
104 return err;
105}
106
107
108static int process_sample_event(union perf_event *event,
109 struct perf_sample *sample,
110 struct perf_evsel *evsel,
111 struct perf_session *session)
112{
113 struct addr_location al;
114
115 if (perf_event__preprocess_sample(event, session, &al, sample,
116 annotate_init) < 0) {
117 fprintf(stderr, "problem processing %d event, skipping it.\n",
118 event->header.type);
119 return -1;
120 }
121
122 if (al.filtered || (hide_unresolved && al.sym == NULL))
123 return 0;
124
125 if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
126 return 0;
127
128 if (al.map != NULL)
129 al.map->dso->hit = 1;
130
131 if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
132 pr_debug("problem incrementing symbol period, skipping event\n");
133 return -1;
134 }
135
136 return 0;
137}
138
139static int process_read_event(union perf_event *event,
140 struct perf_sample *sample __used,
141 struct perf_session *session)
142{
143 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
144 event->read.id);
145 if (show_threads) {
146 const char *name = evsel ? event_name(evsel) : "unknown";
147 perf_read_values_add_value(&show_threads_values,
148 event->read.pid, event->read.tid,
149 event->read.id,
150 name,
151 event->read.value);
152 }
153
154 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
155 evsel ? event_name(evsel) : "FAIL",
156 event->read.value);
157
158 return 0;
159}
160
161static int perf_session__setup_sample_type(struct perf_session *self)
162{
163 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
164 if (sort__has_parent) {
165 ui__warning("Selected --sort parent, but no "
166 "callchain data. Did you call "
167 "'perf record' without -g?\n");
168 return -EINVAL;
169 }
170 if (symbol_conf.use_callchain) {
171 ui__warning("Selected -g but no callchain data. Did "
172 "you call 'perf record' without -g?\n");
173 return -1;
174 }
175 } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
176 !symbol_conf.use_callchain) {
177 symbol_conf.use_callchain = true;
178 if (callchain_register_param(&callchain_param) < 0) {
179 ui__warning("Can't register callchain "
180 "params.\n");
181 return -EINVAL;
182 }
183 }
184
185 return 0;
186}
187
188static struct perf_event_ops event_ops = {
189 .sample = process_sample_event,
190 .mmap = perf_event__process_mmap,
191 .comm = perf_event__process_comm,
192 .exit = perf_event__process_task,
193 .fork = perf_event__process_task,
194 .lost = perf_event__process_lost,
195 .read = process_read_event,
196 .attr = perf_event__process_attr,
197 .event_type = perf_event__process_event_type,
198 .tracing_data = perf_event__process_tracing_data,
199 .build_id = perf_event__process_build_id,
200 .ordered_samples = true,
201 .ordering_requires_timestamps = true,
202};
203
204extern volatile int session_done;
205
206static void sig_handler(int sig __used)
207{
208 session_done = 1;
209}
210
211static size_t hists__fprintf_nr_sample_events(struct hists *self,
212 const char *evname, FILE *fp)
213{
214 size_t ret;
215 char unit;
216 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
217
218 nr_events = convert_unit(nr_events, &unit);
219 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
220 if (evname != NULL)
221 ret += fprintf(fp, " %s", evname);
222 return ret + fprintf(fp, "\n#\n");
223}
224
225static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
226 const char *help)
227{
228 struct perf_evsel *pos;
229
230 list_for_each_entry(pos, &evlist->entries, node) {
231 struct hists *hists = &pos->hists;
232 const char *evname = NULL;
233
234 if (rb_first(&hists->entries) != rb_last(&hists->entries))
235 evname = event_name(pos);
236
237 hists__fprintf_nr_sample_events(hists, evname, stdout);
238 hists__fprintf(hists, NULL, false, stdout);
239 fprintf(stdout, "\n\n");
240 }
241
242 if (sort_order == default_sort_order &&
243 parent_pattern == default_parent_pattern) {
244 fprintf(stdout, "#\n# (%s)\n#\n", help);
245
246 if (show_threads) {
247 bool style = !strcmp(pretty_printing_style, "raw");
248 perf_read_values_display(stdout, &show_threads_values,
249 style);
250 perf_read_values_destroy(&show_threads_values);
251 }
252 }
253
254 return 0;
255}
256
257static int __cmd_report(void)
258{
259 int ret = -EINVAL;
260 u64 nr_samples;
261 struct perf_session *session;
262 struct perf_evsel *pos;
263 struct map *kernel_map;
264 struct kmap *kernel_kmap;
265 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
266
267 signal(SIGINT, sig_handler);
268
269 session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops);
270 if (session == NULL)
271 return -ENOMEM;
272
273 if (cpu_list) {
274 ret = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap);
275 if (ret)
276 goto out_delete;
277 }
278
279 if (show_threads)
280 perf_read_values_init(&show_threads_values);
281
282 ret = perf_session__setup_sample_type(session);
283 if (ret)
284 goto out_delete;
285
286 ret = perf_session__process_events(session, &event_ops);
287 if (ret)
288 goto out_delete;
289
290 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
291 kernel_kmap = map__kmap(kernel_map);
292 if (kernel_map == NULL ||
293 (kernel_map->dso->hit &&
294 (kernel_kmap->ref_reloc_sym == NULL ||
295 kernel_kmap->ref_reloc_sym->addr == 0))) {
296 const struct dso *kdso = kernel_map->dso;
297
298 ui__warning(
299"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
300"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
301"Samples in kernel modules can't be resolved as well.\n\n",
302 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
303"As no suitable kallsyms nor vmlinux was found, kernel samples\n"
304"can't be resolved." :
305"If some relocation was applied (e.g. kexec) symbols may be misresolved.");
306 }
307
308 if (dump_trace) {
309 perf_session__fprintf_nr_events(session, stdout);
310 goto out_delete;
311 }
312
313 if (verbose > 3)
314 perf_session__fprintf(session, stdout);
315
316 if (verbose > 2)
317 perf_session__fprintf_dsos(session, stdout);
318
319 nr_samples = 0;
320 list_for_each_entry(pos, &session->evlist->entries, node) {
321 struct hists *hists = &pos->hists;
322
323 hists__collapse_resort(hists);
324 hists__output_resort(hists);
325 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
326 }
327
328 if (nr_samples == 0) {
329 ui__warning("The %s file has no samples!\n", input_name);
330 goto out_delete;
331 }
332
333 if (use_browser > 0)
334 perf_evlist__tui_browse_hists(session->evlist, help);
335 else
336 perf_evlist__tty_browse_hists(session->evlist, help);
337
338out_delete:
339 /*
340 * Speed up the exit process, for large files this can
341 * take quite a while.
342 *
343 * XXX Enable this when using valgrind or if we ever
344 * librarize this command.
345 *
346 * Also experiment with obstacks to see how much speed
347 * up we'll get here.
348 *
349 * perf_session__delete(session);
350 */
351 return ret;
352}
353
354static int
355parse_callchain_opt(const struct option *opt __used, const char *arg,
356 int unset)
357{
358 char *tok, *tok2;
359 char *endptr;
360
361 /*
362 * --no-call-graph
363 */
364 if (unset) {
365 dont_use_callchains = true;
366 return 0;
367 }
368
369 symbol_conf.use_callchain = true;
370
371 if (!arg)
372 return 0;
373
374 tok = strtok((char *)arg, ",");
375 if (!tok)
376 return -1;
377
378 /* get the output mode */
379 if (!strncmp(tok, "graph", strlen(arg)))
380 callchain_param.mode = CHAIN_GRAPH_ABS;
381
382 else if (!strncmp(tok, "flat", strlen(arg)))
383 callchain_param.mode = CHAIN_FLAT;
384
385 else if (!strncmp(tok, "fractal", strlen(arg)))
386 callchain_param.mode = CHAIN_GRAPH_REL;
387
388 else if (!strncmp(tok, "none", strlen(arg))) {
389 callchain_param.mode = CHAIN_NONE;
390 symbol_conf.use_callchain = false;
391
392 return 0;
393 }
394
395 else
396 return -1;
397
398 /* get the min percentage */
399 tok = strtok(NULL, ",");
400 if (!tok)
401 goto setup;
402
403 callchain_param.min_percent = strtod(tok, &endptr);
404 if (tok == endptr)
405 return -1;
406
407 /* get the print limit */
408 tok2 = strtok(NULL, ",");
409 if (!tok2)
410 goto setup;
411
412 if (tok2[0] != 'c') {
413 callchain_param.print_limit = strtod(tok2, &endptr);
414 tok2 = strtok(NULL, ",");
415 if (!tok2)
416 goto setup;
417 }
418
419 /* get the call chain order */
420 if (!strcmp(tok2, "caller"))
421 callchain_param.order = ORDER_CALLER;
422 else if (!strcmp(tok2, "callee"))
423 callchain_param.order = ORDER_CALLEE;
424 else
425 return -1;
426setup:
427 if (callchain_register_param(&callchain_param) < 0) {
428 fprintf(stderr, "Can't register callchain params\n");
429 return -1;
430 }
431 return 0;
432}
433
434static const char * const report_usage[] = {
435 "perf report [<options>] <command>",
436 NULL
437};
438
439static const struct option options[] = {
440 OPT_STRING('i', "input", &input_name, "file",
441 "input file name"),
442 OPT_INCR('v', "verbose", &verbose,
443 "be more verbose (show symbol address, etc)"),
444 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
445 "dump raw trace in ASCII"),
446 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
447 "file", "vmlinux pathname"),
448 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
449 "file", "kallsyms pathname"),
450 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
451 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
452 "load module symbols - WARNING: use only with -k and LIVE kernel"),
453 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
454 "Show a column with the number of samples"),
455 OPT_BOOLEAN('T', "threads", &show_threads,
456 "Show per-thread event counters"),
457 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
458 "pretty printing style key: normal raw"),
459 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
460 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
461 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
462 "sort by key(s): pid, comm, dso, symbol, parent"),
463 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
464 "Show sample percentage for different cpu modes"),
465 OPT_STRING('p', "parent", &parent_pattern, "regex",
466 "regex filter to identify parent, see: '--sort parent'"),
467 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
468 "Only display entries with parent-match"),
469 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent, call_order",
470 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. "
471 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
472 OPT_BOOLEAN('G', "inverted", &inverted_callchain, "alias for inverted call graph"),
473 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
474 "only consider symbols in these dsos"),
475 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
476 "only consider symbols in these comms"),
477 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
478 "only consider these symbols"),
479 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
480 "width[,width...]",
481 "don't try to adjust column width, use these fixed values"),
482 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
483 "separator for columns, no spaces will be added between "
484 "columns '.' is reserved."),
485 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
486 "Only display entries resolved to a symbol"),
487 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
488 "Look for files with symbols relative to this directory"),
489 OPT_STRING('c', "cpu", &cpu_list, "cpu", "list of cpus to profile"),
490 OPT_END()
491};
492
493int cmd_report(int argc, const char **argv, const char *prefix __used)
494{
495 argc = parse_options(argc, argv, options, report_usage, 0);
496
497 if (use_stdio)
498 use_browser = 0;
499 else if (use_tui)
500 use_browser = 1;
501
502 if (inverted_callchain)
503 callchain_param.order = ORDER_CALLER;
504
505 if (strcmp(input_name, "-") != 0)
506 setup_browser(true);
507 else
508 use_browser = 0;
509 /*
510 * Only in the newt browser we are doing integrated annotation,
511 * so don't allocate extra space that won't be used in the stdio
512 * implementation.
513 */
514 if (use_browser > 0) {
515 symbol_conf.priv_size = sizeof(struct annotation);
516 annotate_init = symbol__annotate_init;
517 /*
518 * For searching by name on the "Browse map details".
519 * providing it only in verbose mode not to bloat too
520 * much struct symbol.
521 */
522 if (verbose) {
523 /*
524 * XXX: Need to provide a less kludgy way to ask for
525 * more space per symbol, the u32 is for the index on
526 * the ui browser.
527 * See symbol__browser_index.
528 */
529 symbol_conf.priv_size += sizeof(u32);
530 symbol_conf.sort_by_name = true;
531 }
532 }
533
534 if (symbol__init() < 0)
535 return -1;
536
537 setup_sorting(report_usage, options);
538
539 if (parent_pattern != default_parent_pattern) {
540 if (sort_dimension__add("parent") < 0)
541 return -1;
542
543 /*
544 * Only show the parent fields if we explicitly
545 * sort that way. If we only use parent machinery
546 * for filtering, we don't want it.
547 */
548 if (!strstr(sort_order, "parent"))
549 sort_parent.elide = 1;
550 } else
551 symbol_conf.exclude_other = false;
552
553 /*
554 * Any (unrecognized) arguments left?
555 */
556 if (argc)
557 usage_with_options(report_usage, options);
558
559 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
560 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
561 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
562
563 return __cmd_report();
564}