1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * builtin-report.c
4 *
5 * Builtin report command: Analyze the perf.data input file,
6 * look up and read DSOs and symbol information and display
7 * a histogram of results, along various sorting keys.
8 */
9 #include "builtin.h"
10
11 #include "util/config.h"
12
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include "util/dso.h"
16 #include <linux/list.h>
17 #include <linux/rbtree.h>
18 #include <linux/err.h>
19 #include <linux/zalloc.h>
20 #include "util/map.h"
21 #include "util/symbol.h"
22 #include "util/map_symbol.h"
23 #include "util/mem-events.h"
24 #include "util/branch.h"
25 #include "util/callchain.h"
26 #include "util/values.h"
27
28 #include "perf.h"
29 #include "util/debug.h"
30 #include "util/evlist.h"
31 #include "util/evsel.h"
32 #include "util/evswitch.h"
33 #include "util/header.h"
34 #include "util/session.h"
35 #include "util/srcline.h"
36 #include "util/tool.h"
37
38 #include <subcmd/parse-options.h>
39 #include <subcmd/exec-cmd.h>
40 #include "util/parse-events.h"
41
42 #include "util/thread.h"
43 #include "util/sort.h"
44 #include "util/hist.h"
45 #include "util/data.h"
46 #include "arch/common.h"
47 #include "util/time-utils.h"
48 #include "util/auxtrace.h"
49 #include "util/units.h"
50 #include "util/util.h" // perf_tip()
51 #include "ui/ui.h"
52 #include "ui/progress.h"
53 #include "util/block-info.h"
54
55 #include <dlfcn.h>
56 #include <errno.h>
57 #include <inttypes.h>
58 #include <regex.h>
59 #include <linux/ctype.h>
60 #include <signal.h>
61 #include <linux/bitmap.h>
62 #include <linux/string.h>
63 #include <linux/stringify.h>
64 #include <linux/time64.h>
65 #include <sys/types.h>
66 #include <sys/stat.h>
67 #include <unistd.h>
68 #include <linux/mman.h>
69
70 #ifdef HAVE_LIBTRACEEVENT
71 #include <traceevent/event-parse.h>
72 #endif
73
74 struct report {
75 struct perf_tool tool;
76 struct perf_session *session;
77 struct evswitch evswitch;
78 #ifdef HAVE_SLANG_SUPPORT
79 bool use_tui;
80 #endif
81 #ifdef HAVE_GTK2_SUPPORT
82 bool use_gtk;
83 #endif
84 bool use_stdio;
85 bool show_full_info;
86 bool show_threads;
87 bool inverted_callchain;
88 bool mem_mode;
89 bool stats_mode;
90 bool tasks_mode;
91 bool mmaps_mode;
92 bool header;
93 bool header_only;
94 bool nonany_branch_mode;
95 bool group_set;
96 bool stitch_lbr;
97 bool disable_order;
98 bool skip_empty;
99 int max_stack;
100 struct perf_read_values show_threads_values;
101 struct annotation_options annotation_opts;
102 const char *pretty_printing_style;
103 const char *cpu_list;
104 const char *symbol_filter_str;
105 const char *time_str;
106 struct perf_time_interval *ptime_range;
107 int range_size;
108 int range_num;
109 float min_percent;
110 u64 nr_entries;
111 u64 queue_size;
112 u64 total_cycles;
113 int socket_filter;
114 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
115 struct branch_type_stat brtype_stat;
116 bool symbol_ipc;
117 bool total_cycles_mode;
118 struct block_report *block_reports;
119 int nr_block_reports;
120 };
121
report__config(const char * var,const char * value,void * cb)122 static int report__config(const char *var, const char *value, void *cb)
123 {
124 struct report *rep = cb;
125
126 if (!strcmp(var, "report.group")) {
127 symbol_conf.event_group = perf_config_bool(var, value);
128 return 0;
129 }
130 if (!strcmp(var, "report.percent-limit")) {
131 double pcnt = strtof(value, NULL);
132
133 rep->min_percent = pcnt;
134 callchain_param.min_percent = pcnt;
135 return 0;
136 }
137 if (!strcmp(var, "report.children")) {
138 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
139 return 0;
140 }
141 if (!strcmp(var, "report.queue-size"))
142 return perf_config_u64(&rep->queue_size, var, value);
143
144 if (!strcmp(var, "report.sort_order")) {
145 default_sort_order = strdup(value);
146 if (!default_sort_order) {
147 pr_err("Not enough memory for report.sort_order\n");
148 return -1;
149 }
150 return 0;
151 }
152
153 if (!strcmp(var, "report.skip-empty")) {
154 rep->skip_empty = perf_config_bool(var, value);
155 return 0;
156 }
157
158 pr_debug("%s variable unknown, ignoring...", var);
159 return 0;
160 }
161
hist_iter__report_callback(struct hist_entry_iter * iter,struct addr_location * al,bool single,void * arg)162 static int hist_iter__report_callback(struct hist_entry_iter *iter,
163 struct addr_location *al, bool single,
164 void *arg)
165 {
166 int err = 0;
167 struct report *rep = arg;
168 struct hist_entry *he = iter->he;
169 struct evsel *evsel = iter->evsel;
170 struct perf_sample *sample = iter->sample;
171 struct mem_info *mi;
172 struct branch_info *bi;
173
174 if (!ui__has_annotation() && !rep->symbol_ipc)
175 return 0;
176
177 if (sort__mode == SORT_MODE__BRANCH) {
178 bi = he->branch_info;
179 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
180 if (err)
181 goto out;
182
183 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
184
185 } else if (rep->mem_mode) {
186 mi = he->mem_info;
187 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
188 if (err)
189 goto out;
190
191 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
192
193 } else if (symbol_conf.cumulate_callchain) {
194 if (single)
195 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
196 } else {
197 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
198 }
199
200 out:
201 return err;
202 }
203
hist_iter__branch_callback(struct hist_entry_iter * iter,struct addr_location * al __maybe_unused,bool single __maybe_unused,void * arg)204 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
205 struct addr_location *al __maybe_unused,
206 bool single __maybe_unused,
207 void *arg)
208 {
209 struct hist_entry *he = iter->he;
210 struct report *rep = arg;
211 struct branch_info *bi = he->branch_info;
212 struct perf_sample *sample = iter->sample;
213 struct evsel *evsel = iter->evsel;
214 int err;
215
216 branch_type_count(&rep->brtype_stat, &bi->flags,
217 bi->from.addr, bi->to.addr);
218
219 if (!ui__has_annotation() && !rep->symbol_ipc)
220 return 0;
221
222 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
223 if (err)
224 goto out;
225
226 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
227
228 out:
229 return err;
230 }
231
setup_forced_leader(struct report * report,struct evlist * evlist)232 static void setup_forced_leader(struct report *report,
233 struct evlist *evlist)
234 {
235 if (report->group_set)
236 evlist__force_leader(evlist);
237 }
238
process_feature_event(struct perf_session * session,union perf_event * event)239 static int process_feature_event(struct perf_session *session,
240 union perf_event *event)
241 {
242 struct report *rep = container_of(session->tool, struct report, tool);
243
244 if (event->feat.feat_id < HEADER_LAST_FEATURE)
245 return perf_event__process_feature(session, event);
246
247 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
248 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
249 event->feat.feat_id);
250 return -1;
251 } else if (rep->header_only) {
252 session_done = 1;
253 }
254
255 /*
256 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
257 * means all features are received, now we can force the
258 * group if needed.
259 */
260 setup_forced_leader(rep, session->evlist);
261 return 0;
262 }
263
process_sample_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct evsel * evsel,struct machine * machine)264 static int process_sample_event(struct perf_tool *tool,
265 union perf_event *event,
266 struct perf_sample *sample,
267 struct evsel *evsel,
268 struct machine *machine)
269 {
270 struct report *rep = container_of(tool, struct report, tool);
271 struct addr_location al;
272 struct hist_entry_iter iter = {
273 .evsel = evsel,
274 .sample = sample,
275 .hide_unresolved = symbol_conf.hide_unresolved,
276 .add_entry_cb = hist_iter__report_callback,
277 };
278 int ret = 0;
279
280 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
281 sample->time)) {
282 return 0;
283 }
284
285 if (evswitch__discard(&rep->evswitch, evsel))
286 return 0;
287
288 addr_location__init(&al);
289 if (machine__resolve(machine, &al, sample) < 0) {
290 pr_debug("problem processing %d event, skipping it.\n",
291 event->header.type);
292 ret = -1;
293 goto out_put;
294 }
295
296 if (rep->stitch_lbr)
297 thread__set_lbr_stitch_enable(al.thread, true);
298
299 if (symbol_conf.hide_unresolved && al.sym == NULL)
300 goto out_put;
301
302 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
303 goto out_put;
304
305 if (sort__mode == SORT_MODE__BRANCH) {
306 /*
307 * A non-synthesized event might not have a branch stack if
308 * branch stacks have been synthesized (using itrace options).
309 */
310 if (!sample->branch_stack)
311 goto out_put;
312
313 iter.add_entry_cb = hist_iter__branch_callback;
314 iter.ops = &hist_iter_branch;
315 } else if (rep->mem_mode) {
316 iter.ops = &hist_iter_mem;
317 } else if (symbol_conf.cumulate_callchain) {
318 iter.ops = &hist_iter_cumulative;
319 } else {
320 iter.ops = &hist_iter_normal;
321 }
322
323 if (al.map != NULL)
324 map__dso(al.map)->hit = 1;
325
326 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
327 hist__account_cycles(sample->branch_stack, &al, sample,
328 rep->nonany_branch_mode,
329 &rep->total_cycles);
330 }
331
332 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
333 if (ret < 0)
334 pr_debug("problem adding hist entry, skipping event\n");
335 out_put:
336 addr_location__exit(&al);
337 return ret;
338 }
339
process_read_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample __maybe_unused,struct evsel * evsel,struct machine * machine __maybe_unused)340 static int process_read_event(struct perf_tool *tool,
341 union perf_event *event,
342 struct perf_sample *sample __maybe_unused,
343 struct evsel *evsel,
344 struct machine *machine __maybe_unused)
345 {
346 struct report *rep = container_of(tool, struct report, tool);
347
348 if (rep->show_threads) {
349 const char *name = evsel__name(evsel);
350 int err = perf_read_values_add_value(&rep->show_threads_values,
351 event->read.pid, event->read.tid,
352 evsel->core.idx,
353 name,
354 event->read.value);
355
356 if (err)
357 return err;
358 }
359
360 return 0;
361 }
362
363 /* For pipe mode, sample_type is not currently set */
report__setup_sample_type(struct report * rep)364 static int report__setup_sample_type(struct report *rep)
365 {
366 struct perf_session *session = rep->session;
367 u64 sample_type = evlist__combined_sample_type(session->evlist);
368 bool is_pipe = perf_data__is_pipe(session->data);
369 struct evsel *evsel;
370
371 if (session->itrace_synth_opts->callchain ||
372 session->itrace_synth_opts->add_callchain ||
373 (!is_pipe &&
374 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
375 !session->itrace_synth_opts->set))
376 sample_type |= PERF_SAMPLE_CALLCHAIN;
377
378 if (session->itrace_synth_opts->last_branch ||
379 session->itrace_synth_opts->add_last_branch)
380 sample_type |= PERF_SAMPLE_BRANCH_STACK;
381
382 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
383 if (perf_hpp_list.parent) {
384 ui__error("Selected --sort parent, but no "
385 "callchain data. Did you call "
386 "'perf record' without -g?\n");
387 return -EINVAL;
388 }
389 if (symbol_conf.use_callchain &&
390 !symbol_conf.show_branchflag_count) {
391 ui__error("Selected -g or --branch-history.\n"
392 "But no callchain or branch data.\n"
393 "Did you call 'perf record' without -g or -b?\n");
394 return -1;
395 }
396 } else if (!callchain_param.enabled &&
397 callchain_param.mode != CHAIN_NONE &&
398 !symbol_conf.use_callchain) {
399 symbol_conf.use_callchain = true;
400 if (callchain_register_param(&callchain_param) < 0) {
401 ui__error("Can't register callchain params.\n");
402 return -EINVAL;
403 }
404 }
405
406 if (symbol_conf.cumulate_callchain) {
407 /* Silently ignore if callchain is missing */
408 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
409 symbol_conf.cumulate_callchain = false;
410 perf_hpp__cancel_cumulate();
411 }
412 }
413
414 if (sort__mode == SORT_MODE__BRANCH) {
415 if (!is_pipe &&
416 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
417 ui__error("Selected -b but no branch data. "
418 "Did you call perf record without -b?\n");
419 return -1;
420 }
421 }
422
423 if (sort__mode == SORT_MODE__MEMORY) {
424 /*
425 * FIXUP: prior to kernel 5.18, Arm SPE missed to set
426 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward
427 * compatibility, set the bit if it's an old perf data file.
428 */
429 evlist__for_each_entry(session->evlist, evsel) {
430 if (strstr(evsel->name, "arm_spe") &&
431 !(sample_type & PERF_SAMPLE_DATA_SRC)) {
432 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC;
433 sample_type |= PERF_SAMPLE_DATA_SRC;
434 }
435 }
436
437 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
438 ui__error("Selected --mem-mode but no mem data. "
439 "Did you call perf record without -d?\n");
440 return -1;
441 }
442 }
443
444 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
445
446 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
447 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
448 "Please apply --call-graph lbr when recording.\n");
449 rep->stitch_lbr = false;
450 }
451
452 /* ??? handle more cases than just ANY? */
453 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
454 rep->nonany_branch_mode = true;
455
456 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
457 if (dwarf_callchain_users) {
458 ui__warning("Please install libunwind or libdw "
459 "development packages during the perf build.\n");
460 }
461 #endif
462
463 return 0;
464 }
465
sig_handler(int sig __maybe_unused)466 static void sig_handler(int sig __maybe_unused)
467 {
468 session_done = 1;
469 }
470
hists__fprintf_nr_sample_events(struct hists * hists,struct report * rep,const char * evname,FILE * fp)471 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
472 const char *evname, FILE *fp)
473 {
474 size_t ret;
475 char unit;
476 unsigned long nr_samples = hists->stats.nr_samples;
477 u64 nr_events = hists->stats.total_period;
478 struct evsel *evsel = hists_to_evsel(hists);
479 char buf[512];
480 size_t size = sizeof(buf);
481 int socked_id = hists->socket_filter;
482
483 if (quiet)
484 return 0;
485
486 if (symbol_conf.filter_relative) {
487 nr_samples = hists->stats.nr_non_filtered_samples;
488 nr_events = hists->stats.total_non_filtered_period;
489 }
490
491 if (evsel__is_group_event(evsel)) {
492 struct evsel *pos;
493
494 evsel__group_desc(evsel, buf, size);
495 evname = buf;
496
497 for_each_group_member(pos, evsel) {
498 const struct hists *pos_hists = evsel__hists(pos);
499
500 if (symbol_conf.filter_relative) {
501 nr_samples += pos_hists->stats.nr_non_filtered_samples;
502 nr_events += pos_hists->stats.total_non_filtered_period;
503 } else {
504 nr_samples += pos_hists->stats.nr_samples;
505 nr_events += pos_hists->stats.total_period;
506 }
507 }
508 }
509
510 nr_samples = convert_unit(nr_samples, &unit);
511 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
512 if (evname != NULL) {
513 ret += fprintf(fp, " of event%s '%s'",
514 evsel->core.nr_members > 1 ? "s" : "", evname);
515 }
516
517 if (rep->time_str)
518 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
519
520 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
521 ret += fprintf(fp, ", show reference callgraph");
522 }
523
524 if (rep->mem_mode) {
525 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
526 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
527 } else
528 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
529
530 if (socked_id > -1)
531 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
532
533 return ret + fprintf(fp, "\n#\n");
534 }
535
evlist__tui_block_hists_browse(struct evlist * evlist,struct report * rep)536 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep)
537 {
538 struct evsel *pos;
539 int i = 0, ret;
540
541 evlist__for_each_entry(evlist, pos) {
542 ret = report__browse_block_hists(&rep->block_reports[i++].hist,
543 rep->min_percent, pos,
544 &rep->session->header.env,
545 &rep->annotation_opts);
546 if (ret != 0)
547 return ret;
548 }
549
550 return 0;
551 }
552
evlist__tty_browse_hists(struct evlist * evlist,struct report * rep,const char * help)553 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
554 {
555 struct evsel *pos;
556 int i = 0;
557
558 if (!quiet) {
559 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
560 evlist->stats.total_lost_samples);
561 }
562
563 evlist__for_each_entry(evlist, pos) {
564 struct hists *hists = evsel__hists(pos);
565 const char *evname = evsel__name(pos);
566
567 if (symbol_conf.event_group && !evsel__is_group_leader(pos))
568 continue;
569
570 if (rep->skip_empty && !hists->stats.nr_samples)
571 continue;
572
573 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
574
575 if (rep->total_cycles_mode) {
576 report__browse_block_hists(&rep->block_reports[i++].hist,
577 rep->min_percent, pos,
578 NULL, NULL);
579 continue;
580 }
581
582 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
583 !(symbol_conf.use_callchain ||
584 symbol_conf.show_branchflag_count));
585 fprintf(stdout, "\n\n");
586 }
587
588 if (!quiet)
589 fprintf(stdout, "#\n# (%s)\n#\n", help);
590
591 if (rep->show_threads) {
592 bool style = !strcmp(rep->pretty_printing_style, "raw");
593 perf_read_values_display(stdout, &rep->show_threads_values,
594 style);
595 perf_read_values_destroy(&rep->show_threads_values);
596 }
597
598 if (sort__mode == SORT_MODE__BRANCH)
599 branch_type_stat_display(stdout, &rep->brtype_stat);
600
601 return 0;
602 }
603
report__warn_kptr_restrict(const struct report * rep)604 static void report__warn_kptr_restrict(const struct report *rep)
605 {
606 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
607 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
608
609 if (evlist__exclude_kernel(rep->session->evlist))
610 return;
611
612 if (kernel_map == NULL ||
613 (map__dso(kernel_map)->hit &&
614 (kernel_kmap->ref_reloc_sym == NULL ||
615 kernel_kmap->ref_reloc_sym->addr == 0))) {
616 const char *desc =
617 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
618 "can't be resolved.";
619
620 if (kernel_map && map__has_symbols(kernel_map)) {
621 desc = "If some relocation was applied (e.g. "
622 "kexec) symbols may be misresolved.";
623 }
624
625 ui__warning(
626 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
627 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
628 "Samples in kernel modules can't be resolved as well.\n\n",
629 desc);
630 }
631 }
632
report__gtk_browse_hists(struct report * rep,const char * help)633 static int report__gtk_browse_hists(struct report *rep, const char *help)
634 {
635 int (*hist_browser)(struct evlist *evlist, const char *help,
636 struct hist_browser_timer *timer, float min_pcnt);
637
638 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");
639
640 if (hist_browser == NULL) {
641 ui__error("GTK browser not found!\n");
642 return -1;
643 }
644
645 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
646 }
647
report__browse_hists(struct report * rep)648 static int report__browse_hists(struct report *rep)
649 {
650 int ret;
651 struct perf_session *session = rep->session;
652 struct evlist *evlist = session->evlist;
653 char *help = NULL, *path = NULL;
654
655 path = system_path(TIPDIR);
656 if (perf_tip(&help, path) || help == NULL) {
657 /* fallback for people who don't install perf ;-) */
658 free(path);
659 path = system_path(DOCDIR);
660 if (perf_tip(&help, path) || help == NULL)
661 help = strdup("Cannot load tips.txt file, please install perf!");
662 }
663 free(path);
664
665 switch (use_browser) {
666 case 1:
667 if (rep->total_cycles_mode) {
668 ret = evlist__tui_block_hists_browse(evlist, rep);
669 break;
670 }
671
672 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
673 &session->header.env, true, &rep->annotation_opts);
674 /*
675 * Usually "ret" is the last pressed key, and we only
676 * care if the key notifies us to switch data file.
677 */
678 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
679 ret = 0;
680 break;
681 case 2:
682 ret = report__gtk_browse_hists(rep, help);
683 break;
684 default:
685 ret = evlist__tty_browse_hists(evlist, rep, help);
686 break;
687 }
688 free(help);
689 return ret;
690 }
691
report__collapse_hists(struct report * rep)692 static int report__collapse_hists(struct report *rep)
693 {
694 struct ui_progress prog;
695 struct evsel *pos;
696 int ret = 0;
697
698 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
699
700 evlist__for_each_entry(rep->session->evlist, pos) {
701 struct hists *hists = evsel__hists(pos);
702
703 if (pos->core.idx == 0)
704 hists->symbol_filter_str = rep->symbol_filter_str;
705
706 hists->socket_filter = rep->socket_filter;
707
708 ret = hists__collapse_resort(hists, &prog);
709 if (ret < 0)
710 break;
711
712 /* Non-group events are considered as leader */
713 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
714 struct hists *leader_hists = evsel__hists(evsel__leader(pos));
715
716 hists__match(leader_hists, hists);
717 hists__link(leader_hists, hists);
718 }
719 }
720
721 ui_progress__finish();
722 return ret;
723 }
724
hists__resort_cb(struct hist_entry * he,void * arg)725 static int hists__resort_cb(struct hist_entry *he, void *arg)
726 {
727 struct report *rep = arg;
728 struct symbol *sym = he->ms.sym;
729
730 if (rep->symbol_ipc && sym && !sym->annotate2) {
731 struct evsel *evsel = hists_to_evsel(he->hists);
732
733 symbol__annotate2(&he->ms, evsel, &rep->annotation_opts, NULL);
734 }
735
736 return 0;
737 }
738
report__output_resort(struct report * rep)739 static void report__output_resort(struct report *rep)
740 {
741 struct ui_progress prog;
742 struct evsel *pos;
743
744 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
745
746 evlist__for_each_entry(rep->session->evlist, pos) {
747 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
748 }
749
750 ui_progress__finish();
751 }
752
count_sample_event(struct perf_tool * tool __maybe_unused,union perf_event * event __maybe_unused,struct perf_sample * sample __maybe_unused,struct evsel * evsel,struct machine * machine __maybe_unused)753 static int count_sample_event(struct perf_tool *tool __maybe_unused,
754 union perf_event *event __maybe_unused,
755 struct perf_sample *sample __maybe_unused,
756 struct evsel *evsel,
757 struct machine *machine __maybe_unused)
758 {
759 struct hists *hists = evsel__hists(evsel);
760
761 hists__inc_nr_events(hists);
762 return 0;
763 }
764
count_lost_samples_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine __maybe_unused)765 static int count_lost_samples_event(struct perf_tool *tool,
766 union perf_event *event,
767 struct perf_sample *sample,
768 struct machine *machine __maybe_unused)
769 {
770 struct report *rep = container_of(tool, struct report, tool);
771 struct evsel *evsel;
772
773 evsel = evlist__id2evsel(rep->session->evlist, sample->id);
774 if (evsel) {
775 hists__inc_nr_lost_samples(evsel__hists(evsel),
776 event->lost_samples.lost);
777 }
778 return 0;
779 }
780
781 static int process_attr(struct perf_tool *tool __maybe_unused,
782 union perf_event *event,
783 struct evlist **pevlist);
784
stats_setup(struct report * rep)785 static void stats_setup(struct report *rep)
786 {
787 memset(&rep->tool, 0, sizeof(rep->tool));
788 rep->tool.attr = process_attr;
789 rep->tool.sample = count_sample_event;
790 rep->tool.lost_samples = count_lost_samples_event;
791 rep->tool.no_warn = true;
792 }
793
stats_print(struct report * rep)794 static int stats_print(struct report *rep)
795 {
796 struct perf_session *session = rep->session;
797
798 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty);
799 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty);
800 return 0;
801 }
802
tasks_setup(struct report * rep)803 static void tasks_setup(struct report *rep)
804 {
805 memset(&rep->tool, 0, sizeof(rep->tool));
806 rep->tool.ordered_events = true;
807 if (rep->mmaps_mode) {
808 rep->tool.mmap = perf_event__process_mmap;
809 rep->tool.mmap2 = perf_event__process_mmap2;
810 }
811 rep->tool.attr = process_attr;
812 rep->tool.comm = perf_event__process_comm;
813 rep->tool.exit = perf_event__process_exit;
814 rep->tool.fork = perf_event__process_fork;
815 rep->tool.no_warn = true;
816 }
817
818 struct task {
819 struct thread *thread;
820 struct list_head list;
821 struct list_head children;
822 };
823
tasks_list(struct task * task,struct machine * machine)824 static struct task *tasks_list(struct task *task, struct machine *machine)
825 {
826 struct thread *parent_thread, *thread = task->thread;
827 struct task *parent_task;
828
829 /* Already listed. */
830 if (!list_empty(&task->list))
831 return NULL;
832
833 /* Last one in the chain. */
834 if (thread__ppid(thread) == -1)
835 return task;
836
837 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread));
838 if (!parent_thread)
839 return ERR_PTR(-ENOENT);
840
841 parent_task = thread__priv(parent_thread);
842 thread__put(parent_thread);
843 list_add_tail(&task->list, &parent_task->children);
844 return tasks_list(parent_task, machine);
845 }
846
maps__fprintf_task(struct maps * maps,int indent,FILE * fp)847 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
848 {
849 size_t printed = 0;
850 struct map_rb_node *rb_node;
851
852 maps__for_each_entry(maps, rb_node) {
853 struct map *map = rb_node->map;
854 const struct dso *dso = map__dso(map);
855 u32 prot = map__prot(map);
856
857 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
858 indent, "", map__start(map), map__end(map),
859 prot & PROT_READ ? 'r' : '-',
860 prot & PROT_WRITE ? 'w' : '-',
861 prot & PROT_EXEC ? 'x' : '-',
862 map__flags(map) ? 's' : 'p',
863 map__pgoff(map),
864 dso->id.ino, dso->name);
865 }
866
867 return printed;
868 }
869
task__print_level(struct task * task,FILE * fp,int level)870 static void task__print_level(struct task *task, FILE *fp, int level)
871 {
872 struct thread *thread = task->thread;
873 struct task *child;
874 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
875 thread__pid(thread), thread__tid(thread),
876 thread__ppid(thread), level, "");
877
878 fprintf(fp, "%s\n", thread__comm_str(thread));
879
880 maps__fprintf_task(thread__maps(thread), comm_indent, fp);
881
882 if (!list_empty(&task->children)) {
883 list_for_each_entry(child, &task->children, list)
884 task__print_level(child, fp, level + 1);
885 }
886 }
887
tasks_print(struct report * rep,FILE * fp)888 static int tasks_print(struct report *rep, FILE *fp)
889 {
890 struct perf_session *session = rep->session;
891 struct machine *machine = &session->machines.host;
892 struct task *tasks, *task;
893 unsigned int nr = 0, itask = 0, i;
894 struct rb_node *nd;
895 LIST_HEAD(list);
896
897 /*
898 * No locking needed while accessing machine->threads,
899 * because --tasks is single threaded command.
900 */
901
902 /* Count all the threads. */
903 for (i = 0; i < THREADS__TABLE_SIZE; i++)
904 nr += machine->threads[i].nr;
905
906 tasks = malloc(sizeof(*tasks) * nr);
907 if (!tasks)
908 return -ENOMEM;
909
910 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
911 struct threads *threads = &machine->threads[i];
912
913 for (nd = rb_first_cached(&threads->entries); nd;
914 nd = rb_next(nd)) {
915 task = tasks + itask++;
916
917 task->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
918 INIT_LIST_HEAD(&task->children);
919 INIT_LIST_HEAD(&task->list);
920 thread__set_priv(task->thread, task);
921 }
922 }
923
924 /*
925 * Iterate every task down to the unprocessed parent
926 * and link all in task children list. Task with no
927 * parent is added into 'list'.
928 */
929 for (itask = 0; itask < nr; itask++) {
930 task = tasks + itask;
931
932 if (!list_empty(&task->list))
933 continue;
934
935 task = tasks_list(task, machine);
936 if (IS_ERR(task)) {
937 pr_err("Error: failed to process tasks\n");
938 free(tasks);
939 return PTR_ERR(task);
940 }
941
942 if (task)
943 list_add_tail(&task->list, &list);
944 }
945
946 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
947
948 list_for_each_entry(task, &list, list)
949 task__print_level(task, fp, 0);
950
951 free(tasks);
952 return 0;
953 }
954
__cmd_report(struct report * rep)955 static int __cmd_report(struct report *rep)
956 {
957 int ret;
958 struct perf_session *session = rep->session;
959 struct evsel *pos;
960 struct perf_data *data = session->data;
961
962 signal(SIGINT, sig_handler);
963
964 if (rep->cpu_list) {
965 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
966 rep->cpu_bitmap);
967 if (ret) {
968 ui__error("failed to set cpu bitmap\n");
969 return ret;
970 }
971 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
972 }
973
974 if (rep->show_threads) {
975 ret = perf_read_values_init(&rep->show_threads_values);
976 if (ret)
977 return ret;
978 }
979
980 ret = report__setup_sample_type(rep);
981 if (ret) {
982 /* report__setup_sample_type() already showed error message */
983 return ret;
984 }
985
986 if (rep->stats_mode)
987 stats_setup(rep);
988
989 if (rep->tasks_mode)
990 tasks_setup(rep);
991
992 ret = perf_session__process_events(session);
993 if (ret) {
994 ui__error("failed to process sample\n");
995 return ret;
996 }
997
998 evlist__check_mem_load_aux(session->evlist);
999
1000 if (rep->stats_mode)
1001 return stats_print(rep);
1002
1003 if (rep->tasks_mode)
1004 return tasks_print(rep, stdout);
1005
1006 report__warn_kptr_restrict(rep);
1007
1008 evlist__for_each_entry(session->evlist, pos)
1009 rep->nr_entries += evsel__hists(pos)->nr_entries;
1010
1011 if (use_browser == 0) {
1012 if (verbose > 3)
1013 perf_session__fprintf(session, stdout);
1014
1015 if (verbose > 2)
1016 perf_session__fprintf_dsos(session, stdout);
1017
1018 if (dump_trace) {
1019 perf_session__fprintf_nr_events(session, stdout,
1020 rep->skip_empty);
1021 evlist__fprintf_nr_events(session->evlist, stdout,
1022 rep->skip_empty);
1023 return 0;
1024 }
1025 }
1026
1027 ret = report__collapse_hists(rep);
1028 if (ret) {
1029 ui__error("failed to process hist entry\n");
1030 return ret;
1031 }
1032
1033 if (session_done())
1034 return 0;
1035
1036 /*
1037 * recalculate number of entries after collapsing since it
1038 * might be changed during the collapse phase.
1039 */
1040 rep->nr_entries = 0;
1041 evlist__for_each_entry(session->evlist, pos)
1042 rep->nr_entries += evsel__hists(pos)->nr_entries;
1043
1044 if (rep->nr_entries == 0) {
1045 ui__error("The %s data has no samples!\n", data->path);
1046 return 0;
1047 }
1048
1049 report__output_resort(rep);
1050
1051 if (rep->total_cycles_mode) {
1052 int block_hpps[6] = {
1053 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
1054 PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
1055 PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
1056 PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
1057 PERF_HPP_REPORT__BLOCK_RANGE,
1058 PERF_HPP_REPORT__BLOCK_DSO,
1059 };
1060
1061 rep->block_reports = block_info__create_report(session->evlist,
1062 rep->total_cycles,
1063 block_hpps, 6,
1064 &rep->nr_block_reports);
1065 if (!rep->block_reports)
1066 return -1;
1067 }
1068
1069 return report__browse_hists(rep);
1070 }
1071
1072 static int
report_parse_callchain_opt(const struct option * opt,const char * arg,int unset)1073 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1074 {
1075 struct callchain_param *callchain = opt->value;
1076
1077 callchain->enabled = !unset;
1078 /*
1079 * --no-call-graph
1080 */
1081 if (unset) {
1082 symbol_conf.use_callchain = false;
1083 callchain->mode = CHAIN_NONE;
1084 return 0;
1085 }
1086
1087 return parse_callchain_report_opt(arg);
1088 }
1089
1090 static int
parse_time_quantum(const struct option * opt,const char * arg,int unset __maybe_unused)1091 parse_time_quantum(const struct option *opt, const char *arg,
1092 int unset __maybe_unused)
1093 {
1094 unsigned long *time_q = opt->value;
1095 char *end;
1096
1097 *time_q = strtoul(arg, &end, 0);
1098 if (end == arg)
1099 goto parse_err;
1100 if (*time_q == 0) {
1101 pr_err("time quantum cannot be 0");
1102 return -1;
1103 }
1104 end = skip_spaces(end);
1105 if (*end == 0)
1106 return 0;
1107 if (!strcmp(end, "s")) {
1108 *time_q *= NSEC_PER_SEC;
1109 return 0;
1110 }
1111 if (!strcmp(end, "ms")) {
1112 *time_q *= NSEC_PER_MSEC;
1113 return 0;
1114 }
1115 if (!strcmp(end, "us")) {
1116 *time_q *= NSEC_PER_USEC;
1117 return 0;
1118 }
1119 if (!strcmp(end, "ns"))
1120 return 0;
1121 parse_err:
1122 pr_err("Cannot parse time quantum `%s'\n", arg);
1123 return -1;
1124 }
1125
1126 int
report_parse_ignore_callees_opt(const struct option * opt __maybe_unused,const char * arg,int unset __maybe_unused)1127 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1128 const char *arg, int unset __maybe_unused)
1129 {
1130 if (arg) {
1131 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1132 if (err) {
1133 char buf[BUFSIZ];
1134 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1135 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1136 return -1;
1137 }
1138 have_ignore_callees = 1;
1139 }
1140
1141 return 0;
1142 }
1143
1144 static int
parse_branch_mode(const struct option * opt,const char * str __maybe_unused,int unset)1145 parse_branch_mode(const struct option *opt,
1146 const char *str __maybe_unused, int unset)
1147 {
1148 int *branch_mode = opt->value;
1149
1150 *branch_mode = !unset;
1151 return 0;
1152 }
1153
1154 static int
parse_percent_limit(const struct option * opt,const char * str,int unset __maybe_unused)1155 parse_percent_limit(const struct option *opt, const char *str,
1156 int unset __maybe_unused)
1157 {
1158 struct report *rep = opt->value;
1159 double pcnt = strtof(str, NULL);
1160
1161 rep->min_percent = pcnt;
1162 callchain_param.min_percent = pcnt;
1163 return 0;
1164 }
1165
process_attr(struct perf_tool * tool __maybe_unused,union perf_event * event,struct evlist ** pevlist)1166 static int process_attr(struct perf_tool *tool __maybe_unused,
1167 union perf_event *event,
1168 struct evlist **pevlist)
1169 {
1170 u64 sample_type;
1171 int err;
1172
1173 err = perf_event__process_attr(tool, event, pevlist);
1174 if (err)
1175 return err;
1176
1177 /*
1178 * Check if we need to enable callchains based
1179 * on events sample_type.
1180 */
1181 sample_type = evlist__combined_sample_type(*pevlist);
1182 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
1183 return 0;
1184 }
1185
cmd_report(int argc,const char ** argv)1186 int cmd_report(int argc, const char **argv)
1187 {
1188 struct perf_session *session;
1189 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1190 struct stat st;
1191 bool has_br_stack = false;
1192 int branch_mode = -1;
1193 int last_key = 0;
1194 bool branch_call_mode = false;
1195 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1196 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1197 CALLCHAIN_REPORT_HELP
1198 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1199 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1200 const char * const report_usage[] = {
1201 "perf report [<options>]",
1202 NULL
1203 };
1204 struct report report = {
1205 .tool = {
1206 .sample = process_sample_event,
1207 .mmap = perf_event__process_mmap,
1208 .mmap2 = perf_event__process_mmap2,
1209 .comm = perf_event__process_comm,
1210 .namespaces = perf_event__process_namespaces,
1211 .cgroup = perf_event__process_cgroup,
1212 .exit = perf_event__process_exit,
1213 .fork = perf_event__process_fork,
1214 .lost = perf_event__process_lost,
1215 .read = process_read_event,
1216 .attr = process_attr,
1217 #ifdef HAVE_LIBTRACEEVENT
1218 .tracing_data = perf_event__process_tracing_data,
1219 #endif
1220 .build_id = perf_event__process_build_id,
1221 .id_index = perf_event__process_id_index,
1222 .auxtrace_info = perf_event__process_auxtrace_info,
1223 .auxtrace = perf_event__process_auxtrace,
1224 .event_update = perf_event__process_event_update,
1225 .feature = process_feature_event,
1226 .ordered_events = true,
1227 .ordering_requires_timestamps = true,
1228 },
1229 .max_stack = PERF_MAX_STACK_DEPTH,
1230 .pretty_printing_style = "normal",
1231 .socket_filter = -1,
1232 .skip_empty = true,
1233 };
1234 char *sort_order_help = sort_help("sort by key(s):");
1235 char *field_order_help = sort_help("output field(s): overhead period sample ");
1236 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL;
1237 const struct option options[] = {
1238 OPT_STRING('i', "input", &input_name, "file",
1239 "input file name"),
1240 OPT_INCR('v', "verbose", &verbose,
1241 "be more verbose (show symbol address, etc)"),
1242 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
1243 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1244 "dump raw trace in ASCII"),
1245 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1246 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1247 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1248 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1249 "file", "vmlinux pathname"),
1250 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1251 "don't load vmlinux even if found"),
1252 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1253 "file", "kallsyms pathname"),
1254 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1255 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1256 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1257 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1258 "Show a column with the number of samples"),
1259 OPT_BOOLEAN('T', "threads", &report.show_threads,
1260 "Show per-thread event counters"),
1261 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1262 "pretty printing style key: normal raw"),
1263 #ifdef HAVE_SLANG_SUPPORT
1264 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1265 #endif
1266 #ifdef HAVE_GTK2_SUPPORT
1267 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1268 #endif
1269 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1270 "Use the stdio interface"),
1271 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1272 OPT_BOOLEAN(0, "header-only", &report.header_only,
1273 "Show only data header."),
1274 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1275 sort_order_help),
1276 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1277 field_order_help),
1278 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1279 "Show sample percentage for different cpu modes"),
1280 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1281 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1282 OPT_STRING('p', "parent", &parent_pattern, "regex",
1283 "regex filter to identify parent, see: '--sort parent'"),
1284 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1285 "Only display entries with parent-match"),
1286 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1287 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1288 report_callchain_help, &report_parse_callchain_opt,
1289 callchain_default_opt),
1290 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1291 "Accumulate callchains of children and show total overhead as well. "
1292 "Enabled by default, use --no-children to disable."),
1293 OPT_INTEGER(0, "max-stack", &report.max_stack,
1294 "Set the maximum stack depth when parsing the callchain, "
1295 "anything beyond the specified depth will be ignored. "
1296 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1297 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1298 "alias for inverted call graph"),
1299 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1300 "ignore callees of these functions in call graphs",
1301 report_parse_ignore_callees_opt),
1302 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1303 "only consider symbols in these dsos"),
1304 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1305 "only consider symbols in these comms"),
1306 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1307 "only consider symbols in these pids"),
1308 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1309 "only consider symbols in these tids"),
1310 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1311 "only consider these symbols"),
1312 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1313 "only show symbols that (partially) match with this filter"),
1314 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1315 "width[,width...]",
1316 "don't try to adjust column width, use these fixed values"),
1317 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1318 "separator for columns, no spaces will be added between "
1319 "columns '.' is reserved."),
1320 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1321 "Only display entries resolved to a symbol"),
1322 OPT_CALLBACK(0, "symfs", NULL, "directory",
1323 "Look for files with symbols relative to this directory",
1324 symbol__config_symfs),
1325 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1326 "list of cpus to profile"),
1327 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1328 "Display extended information about perf.data file"),
1329 OPT_BOOLEAN(0, "source", &report.annotation_opts.annotate_src,
1330 "Interleave source code with assembly code (default)"),
1331 OPT_BOOLEAN(0, "asm-raw", &report.annotation_opts.show_asm_raw,
1332 "Display raw encoding of assembly instructions (default)"),
1333 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1334 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1335 OPT_STRING(0, "prefix", &report.annotation_opts.prefix, "prefix",
1336 "Add prefix to source file path names in programs (with --prefix-strip)"),
1337 OPT_STRING(0, "prefix-strip", &report.annotation_opts.prefix_strip, "N",
1338 "Strip first N entries of source file path name in programs (with --prefix)"),
1339 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1340 "Show a column with the sum of periods"),
1341 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1342 "Show event group information together"),
1343 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1344 "Sort the output by the event at the index n in group. "
1345 "If n is invalid, sort by the first event. "
1346 "WARNING: should be used on grouped events."),
1347 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1348 "use branch records for per branch histogram filling",
1349 parse_branch_mode),
1350 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1351 "add last branch records to call history"),
1352 OPT_STRING(0, "objdump", &objdump_path, "path",
1353 "objdump binary to use for disassembly and annotations"),
1354 OPT_STRING(0, "addr2line", &addr2line_path, "path",
1355 "addr2line binary to use for line numbers"),
1356 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1357 "Disable symbol demangling"),
1358 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1359 "Enable kernel symbol demangling"),
1360 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1361 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1362 "Number of samples to save per histogram entry for individual browsing"),
1363 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1364 "Don't show entries under that percent", parse_percent_limit),
1365 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1366 "how to display percentage of filtered entries", parse_filter_percentage),
1367 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1368 "Instruction Tracing options\n" ITRACE_HELP,
1369 itrace_parse_synth_opts),
1370 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1371 "Show full source file name path for source lines"),
1372 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1373 "Show callgraph from reference event"),
1374 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1375 "Enable LBR callgraph stitching approach"),
1376 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1377 "only show processor socket that match with this filter"),
1378 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1379 "Show raw trace event output (do not use print fmt or plugins)"),
1380 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1381 "Show entries in a hierarchy"),
1382 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1383 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1384 stdio__config_color, "always"),
1385 OPT_STRING(0, "time", &report.time_str, "str",
1386 "Time span of interest (start,stop)"),
1387 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1388 "Show inline function"),
1389 OPT_CALLBACK(0, "percent-type", &report.annotation_opts, "local-period",
1390 "Set percent type local/global-period/hits",
1391 annotate_parse_percent_type),
1392 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1393 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1394 "Set time quantum for time sort key (default 100ms)",
1395 parse_time_quantum),
1396 OPTS_EVSWITCH(&report.evswitch),
1397 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1398 "Sort all blocks by 'Sampled Cycles%'"),
1399 OPT_BOOLEAN(0, "disable-order", &report.disable_order,
1400 "Disable raw trace ordering"),
1401 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
1402 "Do not display empty (or dummy) events in the output"),
1403 OPT_END()
1404 };
1405 struct perf_data data = {
1406 .mode = PERF_DATA_MODE_READ,
1407 };
1408 int ret = hists__init();
1409 char sort_tmp[128];
1410
1411 if (ret < 0)
1412 goto exit;
1413
1414 annotation_options__init(&report.annotation_opts);
1415
1416 ret = perf_config(report__config, &report);
1417 if (ret)
1418 goto exit;
1419
1420 argc = parse_options(argc, argv, options, report_usage, 0);
1421 if (argc) {
1422 /*
1423 * Special case: if there's an argument left then assume that
1424 * it's a symbol filter:
1425 */
1426 if (argc > 1)
1427 usage_with_options(report_usage, options);
1428
1429 report.symbol_filter_str = argv[0];
1430 }
1431
1432 if (disassembler_style) {
1433 report.annotation_opts.disassembler_style = strdup(disassembler_style);
1434 if (!report.annotation_opts.disassembler_style)
1435 return -ENOMEM;
1436 }
1437 if (objdump_path) {
1438 report.annotation_opts.objdump_path = strdup(objdump_path);
1439 if (!report.annotation_opts.objdump_path)
1440 return -ENOMEM;
1441 }
1442 if (addr2line_path) {
1443 symbol_conf.addr2line_path = strdup(addr2line_path);
1444 if (!symbol_conf.addr2line_path)
1445 return -ENOMEM;
1446 }
1447
1448 if (annotate_check_args(&report.annotation_opts) < 0) {
1449 ret = -EINVAL;
1450 goto exit;
1451 }
1452
1453 if (report.mmaps_mode)
1454 report.tasks_mode = true;
1455
1456 if (dump_trace && report.disable_order)
1457 report.tool.ordered_events = false;
1458
1459 if (quiet)
1460 perf_quiet_option();
1461
1462 ret = symbol__validate_sym_arguments();
1463 if (ret)
1464 goto exit;
1465
1466 if (report.inverted_callchain)
1467 callchain_param.order = ORDER_CALLER;
1468 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1469 callchain_param.order = ORDER_CALLER;
1470
1471 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1472 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1473 report.max_stack = itrace_synth_opts.callchain_sz;
1474
1475 if (!input_name || !strlen(input_name)) {
1476 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1477 input_name = "-";
1478 else
1479 input_name = "perf.data";
1480 }
1481
1482 data.path = input_name;
1483 data.force = symbol_conf.force;
1484
1485 repeat:
1486 session = perf_session__new(&data, &report.tool);
1487 if (IS_ERR(session)) {
1488 ret = PTR_ERR(session);
1489 goto exit;
1490 }
1491
1492 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1493 if (ret)
1494 goto exit;
1495
1496 if (zstd_init(&(session->zstd_data), 0) < 0)
1497 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1498
1499 if (report.queue_size) {
1500 ordered_events__set_alloc_size(&session->ordered_events,
1501 report.queue_size);
1502 }
1503
1504 session->itrace_synth_opts = &itrace_synth_opts;
1505
1506 report.session = session;
1507
1508 has_br_stack = perf_header__has_feat(&session->header,
1509 HEADER_BRANCH_STACK);
1510 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1511 has_br_stack = false;
1512
1513 setup_forced_leader(&report, session->evlist);
1514
1515 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) {
1516 parse_options_usage(NULL, options, "group-sort-idx", 0);
1517 ret = -EINVAL;
1518 goto error;
1519 }
1520
1521 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1522 has_br_stack = true;
1523
1524 if (has_br_stack && branch_call_mode)
1525 symbol_conf.show_branchflag_count = true;
1526
1527 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1528
1529 /*
1530 * Branch mode is a tristate:
1531 * -1 means default, so decide based on the file having branch data.
1532 * 0/1 means the user chose a mode.
1533 */
1534 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1535 !branch_call_mode) {
1536 sort__mode = SORT_MODE__BRANCH;
1537 symbol_conf.cumulate_callchain = false;
1538 }
1539 if (branch_call_mode) {
1540 callchain_param.key = CCKEY_ADDRESS;
1541 callchain_param.branch_callstack = true;
1542 symbol_conf.use_callchain = true;
1543 callchain_register_param(&callchain_param);
1544 if (sort_order == NULL)
1545 sort_order = "srcline,symbol,dso";
1546 }
1547
1548 if (report.mem_mode) {
1549 if (sort__mode == SORT_MODE__BRANCH) {
1550 pr_err("branch and mem mode incompatible\n");
1551 goto error;
1552 }
1553 sort__mode = SORT_MODE__MEMORY;
1554 symbol_conf.cumulate_callchain = false;
1555 }
1556
1557 if (symbol_conf.report_hierarchy) {
1558 /* disable incompatible options */
1559 symbol_conf.cumulate_callchain = false;
1560
1561 if (field_order) {
1562 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1563 parse_options_usage(report_usage, options, "F", 1);
1564 parse_options_usage(NULL, options, "hierarchy", 0);
1565 goto error;
1566 }
1567
1568 perf_hpp_list.need_collapse = true;
1569 }
1570
1571 if (report.use_stdio)
1572 use_browser = 0;
1573 #ifdef HAVE_SLANG_SUPPORT
1574 else if (report.use_tui)
1575 use_browser = 1;
1576 #endif
1577 #ifdef HAVE_GTK2_SUPPORT
1578 else if (report.use_gtk)
1579 use_browser = 2;
1580 #endif
1581
1582 /* Force tty output for header output and per-thread stat. */
1583 if (report.header || report.header_only || report.show_threads)
1584 use_browser = 0;
1585 if (report.header || report.header_only)
1586 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1587 if (report.show_full_info)
1588 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1589 if (report.stats_mode || report.tasks_mode)
1590 use_browser = 0;
1591 if (report.stats_mode && report.tasks_mode) {
1592 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1593 goto error;
1594 }
1595
1596 if (report.total_cycles_mode) {
1597 if (sort__mode != SORT_MODE__BRANCH)
1598 report.total_cycles_mode = false;
1599 else
1600 sort_order = NULL;
1601 }
1602
1603 if (strcmp(input_name, "-") != 0)
1604 setup_browser(true);
1605 else
1606 use_browser = 0;
1607
1608 if (sort_order && strstr(sort_order, "ipc")) {
1609 parse_options_usage(report_usage, options, "s", 1);
1610 goto error;
1611 }
1612
1613 if (sort_order && strstr(sort_order, "symbol")) {
1614 if (sort__mode == SORT_MODE__BRANCH) {
1615 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1616 sort_order, "ipc_lbr");
1617 report.symbol_ipc = true;
1618 } else {
1619 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1620 sort_order, "ipc_null");
1621 }
1622
1623 sort_order = sort_tmp;
1624 }
1625
1626 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1627 (setup_sorting(session->evlist) < 0)) {
1628 if (sort_order)
1629 parse_options_usage(report_usage, options, "s", 1);
1630 if (field_order)
1631 parse_options_usage(sort_order ? NULL : report_usage,
1632 options, "F", 1);
1633 goto error;
1634 }
1635
1636 if ((report.header || report.header_only) && !quiet) {
1637 perf_session__fprintf_info(session, stdout,
1638 report.show_full_info);
1639 if (report.header_only) {
1640 if (data.is_pipe) {
1641 /*
1642 * we need to process first few records
1643 * which contains PERF_RECORD_HEADER_FEATURE.
1644 */
1645 perf_session__process_events(session);
1646 }
1647 ret = 0;
1648 goto error;
1649 }
1650 } else if (use_browser == 0 && !quiet &&
1651 !report.stats_mode && !report.tasks_mode) {
1652 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1653 stdout);
1654 }
1655
1656 /*
1657 * Only in the TUI browser we are doing integrated annotation,
1658 * so don't allocate extra space that won't be used in the stdio
1659 * implementation.
1660 */
1661 if (ui__has_annotation() || report.symbol_ipc ||
1662 report.total_cycles_mode) {
1663 ret = symbol__annotation_init();
1664 if (ret < 0)
1665 goto error;
1666 /*
1667 * For searching by name on the "Browse map details".
1668 * providing it only in verbose mode not to bloat too
1669 * much struct symbol.
1670 */
1671 if (verbose > 0) {
1672 /*
1673 * XXX: Need to provide a less kludgy way to ask for
1674 * more space per symbol, the u32 is for the index on
1675 * the ui browser.
1676 * See symbol__browser_index.
1677 */
1678 symbol_conf.priv_size += sizeof(u32);
1679 }
1680 annotation_config__init(&report.annotation_opts);
1681 }
1682
1683 if (symbol__init(&session->header.env) < 0)
1684 goto error;
1685
1686 if (report.time_str) {
1687 ret = perf_time__parse_for_ranges(report.time_str, session,
1688 &report.ptime_range,
1689 &report.range_size,
1690 &report.range_num);
1691 if (ret < 0)
1692 goto error;
1693
1694 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1695 report.ptime_range,
1696 report.range_num);
1697 }
1698
1699 #ifdef HAVE_LIBTRACEEVENT
1700 if (session->tevent.pevent &&
1701 tep_set_function_resolver(session->tevent.pevent,
1702 machine__resolve_kernel_addr,
1703 &session->machines.host) < 0) {
1704 pr_err("%s: failed to set libtraceevent function resolver\n",
1705 __func__);
1706 return -1;
1707 }
1708 #endif
1709 sort__setup_elide(stdout);
1710
1711 ret = __cmd_report(&report);
1712 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1713 perf_session__delete(session);
1714 last_key = K_SWITCH_INPUT_DATA;
1715 goto repeat;
1716 } else
1717 ret = 0;
1718
1719 error:
1720 if (report.ptime_range) {
1721 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1722 zfree(&report.ptime_range);
1723 }
1724
1725 if (report.block_reports) {
1726 block_info__free_report(report.block_reports,
1727 report.nr_block_reports);
1728 report.block_reports = NULL;
1729 }
1730
1731 zstd_fini(&(session->zstd_data));
1732 perf_session__delete(session);
1733 exit:
1734 annotation_options__exit(&report.annotation_opts);
1735 free(sort_order_help);
1736 free(field_order_help);
1737 return ret;
1738 }
1739