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