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 const char *pretty_printing_style;
102 const char *cpu_list;
103 const char *symbol_filter_str;
104 const char *time_str;
105 struct perf_time_interval *ptime_range;
106 int range_size;
107 int range_num;
108 float min_percent;
109 u64 nr_entries;
110 u64 queue_size;
111 u64 total_cycles;
112 int socket_filter;
113 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
114 struct branch_type_stat brtype_stat;
115 bool symbol_ipc;
116 bool total_cycles_mode;
117 struct block_report *block_reports;
118 int nr_block_reports;
119 };
120
report__config(const char * var,const char * value,void * cb)121 static int report__config(const char *var, const char *value, void *cb)
122 {
123 struct report *rep = cb;
124
125 if (!strcmp(var, "report.group")) {
126 symbol_conf.event_group = perf_config_bool(var, value);
127 return 0;
128 }
129 if (!strcmp(var, "report.percent-limit")) {
130 double pcnt = strtof(value, NULL);
131
132 rep->min_percent = pcnt;
133 callchain_param.min_percent = pcnt;
134 return 0;
135 }
136 if (!strcmp(var, "report.children")) {
137 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
138 return 0;
139 }
140 if (!strcmp(var, "report.queue-size"))
141 return perf_config_u64(&rep->queue_size, var, value);
142
143 if (!strcmp(var, "report.sort_order")) {
144 default_sort_order = strdup(value);
145 if (!default_sort_order) {
146 pr_err("Not enough memory for report.sort_order\n");
147 return -1;
148 }
149 return 0;
150 }
151
152 if (!strcmp(var, "report.skip-empty")) {
153 rep->skip_empty = perf_config_bool(var, value);
154 return 0;
155 }
156
157 pr_debug("%s variable unknown, ignoring...", var);
158 return 0;
159 }
160
hist_iter__report_callback(struct hist_entry_iter * iter,struct addr_location * al,bool single,void * arg)161 static int hist_iter__report_callback(struct hist_entry_iter *iter,
162 struct addr_location *al, bool single,
163 void *arg)
164 {
165 int err = 0;
166 struct report *rep = arg;
167 struct hist_entry *he = iter->he;
168 struct evsel *evsel = iter->evsel;
169 struct perf_sample *sample = iter->sample;
170 struct mem_info *mi;
171 struct branch_info *bi;
172
173 if (!ui__has_annotation() && !rep->symbol_ipc)
174 return 0;
175
176 if (sort__mode == SORT_MODE__BRANCH) {
177 bi = he->branch_info;
178 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
179 if (err)
180 goto out;
181
182 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
183
184 } else if (rep->mem_mode) {
185 mi = he->mem_info;
186 err = addr_map_symbol__inc_samples(&mi->daddr, sample, evsel);
187 if (err)
188 goto out;
189
190 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
191
192 } else if (symbol_conf.cumulate_callchain) {
193 if (single)
194 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
195 } else {
196 err = hist_entry__inc_addr_samples(he, sample, evsel, al->addr);
197 }
198
199 out:
200 return err;
201 }
202
hist_iter__branch_callback(struct hist_entry_iter * iter,struct addr_location * al __maybe_unused,bool single __maybe_unused,void * arg)203 static int hist_iter__branch_callback(struct hist_entry_iter *iter,
204 struct addr_location *al __maybe_unused,
205 bool single __maybe_unused,
206 void *arg)
207 {
208 struct hist_entry *he = iter->he;
209 struct report *rep = arg;
210 struct branch_info *bi = he->branch_info;
211 struct perf_sample *sample = iter->sample;
212 struct evsel *evsel = iter->evsel;
213 int err;
214
215 branch_type_count(&rep->brtype_stat, &bi->flags,
216 bi->from.addr, bi->to.addr);
217
218 if (!ui__has_annotation() && !rep->symbol_ipc)
219 return 0;
220
221 err = addr_map_symbol__inc_samples(&bi->from, sample, evsel);
222 if (err)
223 goto out;
224
225 err = addr_map_symbol__inc_samples(&bi->to, sample, evsel);
226
227 out:
228 return err;
229 }
230
setup_forced_leader(struct report * report,struct evlist * evlist)231 static void setup_forced_leader(struct report *report,
232 struct evlist *evlist)
233 {
234 if (report->group_set)
235 evlist__force_leader(evlist);
236 }
237
process_feature_event(struct perf_session * session,union perf_event * event)238 static int process_feature_event(struct perf_session *session,
239 union perf_event *event)
240 {
241 struct report *rep = container_of(session->tool, struct report, tool);
242
243 if (event->feat.feat_id < HEADER_LAST_FEATURE)
244 return perf_event__process_feature(session, event);
245
246 if (event->feat.feat_id != HEADER_LAST_FEATURE) {
247 pr_err("failed: wrong feature ID: %" PRI_lu64 "\n",
248 event->feat.feat_id);
249 return -1;
250 } else if (rep->header_only) {
251 session_done = 1;
252 }
253
254 /*
255 * (feat_id = HEADER_LAST_FEATURE) is the end marker which
256 * means all features are received, now we can force the
257 * group if needed.
258 */
259 setup_forced_leader(rep, session->evlist);
260 return 0;
261 }
262
process_sample_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct evsel * evsel,struct machine * machine)263 static int process_sample_event(struct perf_tool *tool,
264 union perf_event *event,
265 struct perf_sample *sample,
266 struct evsel *evsel,
267 struct machine *machine)
268 {
269 struct report *rep = container_of(tool, struct report, tool);
270 struct addr_location al;
271 struct hist_entry_iter iter = {
272 .evsel = evsel,
273 .sample = sample,
274 .hide_unresolved = symbol_conf.hide_unresolved,
275 .add_entry_cb = hist_iter__report_callback,
276 };
277 int ret = 0;
278
279 if (perf_time__ranges_skip_sample(rep->ptime_range, rep->range_num,
280 sample->time)) {
281 return 0;
282 }
283
284 if (evswitch__discard(&rep->evswitch, evsel))
285 return 0;
286
287 addr_location__init(&al);
288 if (machine__resolve(machine, &al, sample) < 0) {
289 pr_debug("problem processing %d event, skipping it.\n",
290 event->header.type);
291 ret = -1;
292 goto out_put;
293 }
294
295 if (rep->stitch_lbr)
296 thread__set_lbr_stitch_enable(al.thread, true);
297
298 if (symbol_conf.hide_unresolved && al.sym == NULL)
299 goto out_put;
300
301 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
302 goto out_put;
303
304 if (sort__mode == SORT_MODE__BRANCH) {
305 /*
306 * A non-synthesized event might not have a branch stack if
307 * branch stacks have been synthesized (using itrace options).
308 */
309 if (!sample->branch_stack)
310 goto out_put;
311
312 iter.add_entry_cb = hist_iter__branch_callback;
313 iter.ops = &hist_iter_branch;
314 } else if (rep->mem_mode) {
315 iter.ops = &hist_iter_mem;
316 } else if (symbol_conf.cumulate_callchain) {
317 iter.ops = &hist_iter_cumulative;
318 } else {
319 iter.ops = &hist_iter_normal;
320 }
321
322 if (al.map != NULL)
323 map__dso(al.map)->hit = 1;
324
325 if (ui__has_annotation() || rep->symbol_ipc || rep->total_cycles_mode) {
326 hist__account_cycles(sample->branch_stack, &al, sample,
327 rep->nonany_branch_mode,
328 &rep->total_cycles);
329 }
330
331 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
332 if (ret < 0)
333 pr_debug("problem adding hist entry, skipping event\n");
334 out_put:
335 addr_location__exit(&al);
336 return ret;
337 }
338
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)339 static int process_read_event(struct perf_tool *tool,
340 union perf_event *event,
341 struct perf_sample *sample __maybe_unused,
342 struct evsel *evsel,
343 struct machine *machine __maybe_unused)
344 {
345 struct report *rep = container_of(tool, struct report, tool);
346
347 if (rep->show_threads) {
348 const char *name = evsel__name(evsel);
349 int err = perf_read_values_add_value(&rep->show_threads_values,
350 event->read.pid, event->read.tid,
351 evsel->core.idx,
352 name,
353 event->read.value);
354
355 if (err)
356 return err;
357 }
358
359 return 0;
360 }
361
362 /* For pipe mode, sample_type is not currently set */
report__setup_sample_type(struct report * rep)363 static int report__setup_sample_type(struct report *rep)
364 {
365 struct perf_session *session = rep->session;
366 u64 sample_type = evlist__combined_sample_type(session->evlist);
367 bool is_pipe = perf_data__is_pipe(session->data);
368 struct evsel *evsel;
369
370 if (session->itrace_synth_opts->callchain ||
371 session->itrace_synth_opts->add_callchain ||
372 (!is_pipe &&
373 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
374 !session->itrace_synth_opts->set))
375 sample_type |= PERF_SAMPLE_CALLCHAIN;
376
377 if (session->itrace_synth_opts->last_branch ||
378 session->itrace_synth_opts->add_last_branch)
379 sample_type |= PERF_SAMPLE_BRANCH_STACK;
380
381 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
382 if (perf_hpp_list.parent) {
383 ui__error("Selected --sort parent, but no "
384 "callchain data. Did you call "
385 "'perf record' without -g?\n");
386 return -EINVAL;
387 }
388 if (symbol_conf.use_callchain &&
389 !symbol_conf.show_branchflag_count) {
390 ui__error("Selected -g or --branch-history.\n"
391 "But no callchain or branch data.\n"
392 "Did you call 'perf record' without -g or -b?\n");
393 return -1;
394 }
395 } else if (!callchain_param.enabled &&
396 callchain_param.mode != CHAIN_NONE &&
397 !symbol_conf.use_callchain) {
398 symbol_conf.use_callchain = true;
399 if (callchain_register_param(&callchain_param) < 0) {
400 ui__error("Can't register callchain params.\n");
401 return -EINVAL;
402 }
403 }
404
405 if (symbol_conf.cumulate_callchain) {
406 /* Silently ignore if callchain is missing */
407 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
408 symbol_conf.cumulate_callchain = false;
409 perf_hpp__cancel_cumulate();
410 }
411 }
412
413 if (sort__mode == SORT_MODE__BRANCH) {
414 if (!is_pipe &&
415 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
416 ui__error("Selected -b but no branch data. "
417 "Did you call perf record without -b?\n");
418 return -1;
419 }
420 }
421
422 if (sort__mode == SORT_MODE__MEMORY) {
423 /*
424 * FIXUP: prior to kernel 5.18, Arm SPE missed to set
425 * PERF_SAMPLE_DATA_SRC bit in sample type. For backward
426 * compatibility, set the bit if it's an old perf data file.
427 */
428 evlist__for_each_entry(session->evlist, evsel) {
429 if (strstr(evsel__name(evsel), "arm_spe") &&
430 !(sample_type & PERF_SAMPLE_DATA_SRC)) {
431 evsel->core.attr.sample_type |= PERF_SAMPLE_DATA_SRC;
432 sample_type |= PERF_SAMPLE_DATA_SRC;
433 }
434 }
435
436 if (!is_pipe && !(sample_type & PERF_SAMPLE_DATA_SRC)) {
437 ui__error("Selected --mem-mode but no mem data. "
438 "Did you call perf record without -d?\n");
439 return -1;
440 }
441 }
442
443 callchain_param_setup(sample_type, perf_env__arch(&rep->session->header.env));
444
445 if (rep->stitch_lbr && (callchain_param.record_mode != CALLCHAIN_LBR)) {
446 ui__warning("Can't find LBR callchain. Switch off --stitch-lbr.\n"
447 "Please apply --call-graph lbr when recording.\n");
448 rep->stitch_lbr = false;
449 }
450
451 /* ??? handle more cases than just ANY? */
452 if (!(evlist__combined_branch_type(session->evlist) & PERF_SAMPLE_BRANCH_ANY))
453 rep->nonany_branch_mode = true;
454
455 #if !defined(HAVE_LIBUNWIND_SUPPORT) && !defined(HAVE_DWARF_SUPPORT)
456 if (dwarf_callchain_users) {
457 ui__warning("Please install libunwind or libdw "
458 "development packages during the perf build.\n");
459 }
460 #endif
461
462 return 0;
463 }
464
sig_handler(int sig __maybe_unused)465 static void sig_handler(int sig __maybe_unused)
466 {
467 session_done = 1;
468 }
469
hists__fprintf_nr_sample_events(struct hists * hists,struct report * rep,const char * evname,FILE * fp)470 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
471 const char *evname, FILE *fp)
472 {
473 size_t ret;
474 char unit;
475 unsigned long nr_samples = hists->stats.nr_samples;
476 u64 nr_events = hists->stats.total_period;
477 struct evsel *evsel = hists_to_evsel(hists);
478 char buf[512];
479 size_t size = sizeof(buf);
480 int socked_id = hists->socket_filter;
481
482 if (quiet)
483 return 0;
484
485 if (symbol_conf.filter_relative) {
486 nr_samples = hists->stats.nr_non_filtered_samples;
487 nr_events = hists->stats.total_non_filtered_period;
488 }
489
490 if (evsel__is_group_event(evsel)) {
491 struct evsel *pos;
492
493 evsel__group_desc(evsel, buf, size);
494 evname = buf;
495
496 for_each_group_member(pos, evsel) {
497 const struct hists *pos_hists = evsel__hists(pos);
498
499 if (symbol_conf.filter_relative) {
500 nr_samples += pos_hists->stats.nr_non_filtered_samples;
501 nr_events += pos_hists->stats.total_non_filtered_period;
502 } else {
503 nr_samples += pos_hists->stats.nr_samples;
504 nr_events += pos_hists->stats.total_period;
505 }
506 }
507 }
508
509 nr_samples = convert_unit(nr_samples, &unit);
510 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
511 if (evname != NULL) {
512 ret += fprintf(fp, " of event%s '%s'",
513 evsel->core.nr_members > 1 ? "s" : "", evname);
514 }
515
516 if (rep->time_str)
517 ret += fprintf(fp, " (time slices: %s)", rep->time_str);
518
519 if (symbol_conf.show_ref_callgraph && evname && strstr(evname, "call-graph=no")) {
520 ret += fprintf(fp, ", show reference callgraph");
521 }
522
523 if (rep->mem_mode) {
524 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
525 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
526 } else
527 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
528
529 if (socked_id > -1)
530 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
531
532 return ret + fprintf(fp, "\n#\n");
533 }
534
evlist__tui_block_hists_browse(struct evlist * evlist,struct report * rep)535 static int evlist__tui_block_hists_browse(struct evlist *evlist, struct report *rep)
536 {
537 struct evsel *pos;
538 int i = 0, ret;
539
540 evlist__for_each_entry(evlist, pos) {
541 ret = report__browse_block_hists(&rep->block_reports[i++].hist,
542 rep->min_percent, pos,
543 &rep->session->header.env,
544 &annotate_opts);
545 if (ret != 0)
546 return ret;
547 }
548
549 return 0;
550 }
551
evlist__tty_browse_hists(struct evlist * evlist,struct report * rep,const char * help)552 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
553 {
554 struct evsel *pos;
555 int i = 0;
556
557 if (!quiet) {
558 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
559 evlist->stats.total_lost_samples);
560 }
561
562 evlist__for_each_entry(evlist, pos) {
563 struct hists *hists = evsel__hists(pos);
564 const char *evname = evsel__name(pos);
565
566 if (symbol_conf.event_group && !evsel__is_group_leader(pos))
567 continue;
568
569 if (rep->skip_empty && !hists->stats.nr_samples)
570 continue;
571
572 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
573
574 if (rep->total_cycles_mode) {
575 report__browse_block_hists(&rep->block_reports[i++].hist,
576 rep->min_percent, pos,
577 NULL, NULL);
578 continue;
579 }
580
581 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
582 !(symbol_conf.use_callchain ||
583 symbol_conf.show_branchflag_count));
584 fprintf(stdout, "\n\n");
585 }
586
587 if (!quiet)
588 fprintf(stdout, "#\n# (%s)\n#\n", help);
589
590 if (rep->show_threads) {
591 bool style = !strcmp(rep->pretty_printing_style, "raw");
592 perf_read_values_display(stdout, &rep->show_threads_values,
593 style);
594 perf_read_values_destroy(&rep->show_threads_values);
595 }
596
597 if (sort__mode == SORT_MODE__BRANCH)
598 branch_type_stat_display(stdout, &rep->brtype_stat);
599
600 return 0;
601 }
602
report__warn_kptr_restrict(const struct report * rep)603 static void report__warn_kptr_restrict(const struct report *rep)
604 {
605 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
606 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
607
608 if (evlist__exclude_kernel(rep->session->evlist))
609 return;
610
611 if (kernel_map == NULL ||
612 (map__dso(kernel_map)->hit &&
613 (kernel_kmap->ref_reloc_sym == NULL ||
614 kernel_kmap->ref_reloc_sym->addr == 0))) {
615 const char *desc =
616 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
617 "can't be resolved.";
618
619 if (kernel_map && map__has_symbols(kernel_map)) {
620 desc = "If some relocation was applied (e.g. "
621 "kexec) symbols may be misresolved.";
622 }
623
624 ui__warning(
625 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
626 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
627 "Samples in kernel modules can't be resolved as well.\n\n",
628 desc);
629 }
630 }
631
report__gtk_browse_hists(struct report * rep,const char * help)632 static int report__gtk_browse_hists(struct report *rep, const char *help)
633 {
634 int (*hist_browser)(struct evlist *evlist, const char *help,
635 struct hist_browser_timer *timer, float min_pcnt);
636
637 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");
638
639 if (hist_browser == NULL) {
640 ui__error("GTK browser not found!\n");
641 return -1;
642 }
643
644 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
645 }
646
report__browse_hists(struct report * rep)647 static int report__browse_hists(struct report *rep)
648 {
649 int ret;
650 struct perf_session *session = rep->session;
651 struct evlist *evlist = session->evlist;
652 char *help = NULL, *path = NULL;
653
654 path = system_path(TIPDIR);
655 if (perf_tip(&help, path) || help == NULL) {
656 /* fallback for people who don't install perf ;-) */
657 free(path);
658 path = system_path(DOCDIR);
659 if (perf_tip(&help, path) || help == NULL)
660 help = strdup("Cannot load tips.txt file, please install perf!");
661 }
662 free(path);
663
664 switch (use_browser) {
665 case 1:
666 if (rep->total_cycles_mode) {
667 ret = evlist__tui_block_hists_browse(evlist, rep);
668 break;
669 }
670
671 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
672 &session->header.env, true, &annotate_opts);
673 /*
674 * Usually "ret" is the last pressed key, and we only
675 * care if the key notifies us to switch data file.
676 */
677 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
678 ret = 0;
679 break;
680 case 2:
681 ret = report__gtk_browse_hists(rep, help);
682 break;
683 default:
684 ret = evlist__tty_browse_hists(evlist, rep, help);
685 break;
686 }
687 free(help);
688 return ret;
689 }
690
report__collapse_hists(struct report * rep)691 static int report__collapse_hists(struct report *rep)
692 {
693 struct ui_progress prog;
694 struct evsel *pos;
695 int ret = 0;
696
697 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
698
699 evlist__for_each_entry(rep->session->evlist, pos) {
700 struct hists *hists = evsel__hists(pos);
701
702 if (pos->core.idx == 0)
703 hists->symbol_filter_str = rep->symbol_filter_str;
704
705 hists->socket_filter = rep->socket_filter;
706
707 ret = hists__collapse_resort(hists, &prog);
708 if (ret < 0)
709 break;
710
711 /* Non-group events are considered as leader */
712 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
713 struct hists *leader_hists = evsel__hists(evsel__leader(pos));
714
715 hists__match(leader_hists, hists);
716 hists__link(leader_hists, hists);
717 }
718 }
719
720 ui_progress__finish();
721 return ret;
722 }
723
hists__resort_cb(struct hist_entry * he,void * arg)724 static int hists__resort_cb(struct hist_entry *he, void *arg)
725 {
726 struct report *rep = arg;
727 struct symbol *sym = he->ms.sym;
728
729 if (rep->symbol_ipc && sym && !sym->annotate2) {
730 struct evsel *evsel = hists_to_evsel(he->hists);
731
732 symbol__annotate2(&he->ms, evsel, NULL);
733 }
734
735 return 0;
736 }
737
report__output_resort(struct report * rep)738 static void report__output_resort(struct report *rep)
739 {
740 struct ui_progress prog;
741 struct evsel *pos;
742
743 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
744
745 evlist__for_each_entry(rep->session->evlist, pos) {
746 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
747 }
748
749 ui_progress__finish();
750 }
751
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)752 static int count_sample_event(struct perf_tool *tool __maybe_unused,
753 union perf_event *event __maybe_unused,
754 struct perf_sample *sample __maybe_unused,
755 struct evsel *evsel,
756 struct machine *machine __maybe_unused)
757 {
758 struct hists *hists = evsel__hists(evsel);
759
760 hists__inc_nr_events(hists);
761 return 0;
762 }
763
count_lost_samples_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine __maybe_unused)764 static int count_lost_samples_event(struct perf_tool *tool,
765 union perf_event *event,
766 struct perf_sample *sample,
767 struct machine *machine __maybe_unused)
768 {
769 struct report *rep = container_of(tool, struct report, tool);
770 struct evsel *evsel;
771
772 evsel = evlist__id2evsel(rep->session->evlist, sample->id);
773 if (evsel) {
774 hists__inc_nr_lost_samples(evsel__hists(evsel),
775 event->lost_samples.lost);
776 }
777 return 0;
778 }
779
780 static int process_attr(struct perf_tool *tool __maybe_unused,
781 union perf_event *event,
782 struct evlist **pevlist);
783
stats_setup(struct report * rep)784 static void stats_setup(struct report *rep)
785 {
786 memset(&rep->tool, 0, sizeof(rep->tool));
787 rep->tool.attr = process_attr;
788 rep->tool.sample = count_sample_event;
789 rep->tool.lost_samples = count_lost_samples_event;
790 rep->tool.no_warn = true;
791 }
792
stats_print(struct report * rep)793 static int stats_print(struct report *rep)
794 {
795 struct perf_session *session = rep->session;
796
797 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty);
798 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty);
799 return 0;
800 }
801
tasks_setup(struct report * rep)802 static void tasks_setup(struct report *rep)
803 {
804 memset(&rep->tool, 0, sizeof(rep->tool));
805 rep->tool.ordered_events = true;
806 if (rep->mmaps_mode) {
807 rep->tool.mmap = perf_event__process_mmap;
808 rep->tool.mmap2 = perf_event__process_mmap2;
809 }
810 rep->tool.attr = process_attr;
811 rep->tool.comm = perf_event__process_comm;
812 rep->tool.exit = perf_event__process_exit;
813 rep->tool.fork = perf_event__process_fork;
814 rep->tool.no_warn = true;
815 }
816
817 struct task {
818 struct thread *thread;
819 struct list_head list;
820 struct list_head children;
821 };
822
tasks_list(struct task * task,struct machine * machine)823 static struct task *tasks_list(struct task *task, struct machine *machine)
824 {
825 struct thread *parent_thread, *thread = task->thread;
826 struct task *parent_task;
827
828 /* Already listed. */
829 if (!list_empty(&task->list))
830 return NULL;
831
832 /* Last one in the chain. */
833 if (thread__ppid(thread) == -1)
834 return task;
835
836 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread));
837 if (!parent_thread)
838 return ERR_PTR(-ENOENT);
839
840 parent_task = thread__priv(parent_thread);
841 thread__put(parent_thread);
842 list_add_tail(&task->list, &parent_task->children);
843 return tasks_list(parent_task, machine);
844 }
845
maps__fprintf_task(struct maps * maps,int indent,FILE * fp)846 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
847 {
848 size_t printed = 0;
849 struct map_rb_node *rb_node;
850
851 maps__for_each_entry(maps, rb_node) {
852 struct map *map = rb_node->map;
853 const struct dso *dso = map__dso(map);
854 u32 prot = map__prot(map);
855
856 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
857 indent, "", map__start(map), map__end(map),
858 prot & PROT_READ ? 'r' : '-',
859 prot & PROT_WRITE ? 'w' : '-',
860 prot & PROT_EXEC ? 'x' : '-',
861 map__flags(map) ? 's' : 'p',
862 map__pgoff(map),
863 dso->id.ino, dso->name);
864 }
865
866 return printed;
867 }
868
task__print_level(struct task * task,FILE * fp,int level)869 static void task__print_level(struct task *task, FILE *fp, int level)
870 {
871 struct thread *thread = task->thread;
872 struct task *child;
873 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
874 thread__pid(thread), thread__tid(thread),
875 thread__ppid(thread), level, "");
876
877 fprintf(fp, "%s\n", thread__comm_str(thread));
878
879 maps__fprintf_task(thread__maps(thread), comm_indent, fp);
880
881 if (!list_empty(&task->children)) {
882 list_for_each_entry(child, &task->children, list)
883 task__print_level(child, fp, level + 1);
884 }
885 }
886
tasks_print(struct report * rep,FILE * fp)887 static int tasks_print(struct report *rep, FILE *fp)
888 {
889 struct perf_session *session = rep->session;
890 struct machine *machine = &session->machines.host;
891 struct task *tasks, *task;
892 unsigned int nr = 0, itask = 0, i;
893 struct rb_node *nd;
894 LIST_HEAD(list);
895
896 /*
897 * No locking needed while accessing machine->threads,
898 * because --tasks is single threaded command.
899 */
900
901 /* Count all the threads. */
902 for (i = 0; i < THREADS__TABLE_SIZE; i++)
903 nr += machine->threads[i].nr;
904
905 tasks = malloc(sizeof(*tasks) * nr);
906 if (!tasks)
907 return -ENOMEM;
908
909 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
910 struct threads *threads = &machine->threads[i];
911
912 for (nd = rb_first_cached(&threads->entries); nd;
913 nd = rb_next(nd)) {
914 task = tasks + itask++;
915
916 task->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
917 INIT_LIST_HEAD(&task->children);
918 INIT_LIST_HEAD(&task->list);
919 thread__set_priv(task->thread, task);
920 }
921 }
922
923 /*
924 * Iterate every task down to the unprocessed parent
925 * and link all in task children list. Task with no
926 * parent is added into 'list'.
927 */
928 for (itask = 0; itask < nr; itask++) {
929 task = tasks + itask;
930
931 if (!list_empty(&task->list))
932 continue;
933
934 task = tasks_list(task, machine);
935 if (IS_ERR(task)) {
936 pr_err("Error: failed to process tasks\n");
937 free(tasks);
938 return PTR_ERR(task);
939 }
940
941 if (task)
942 list_add_tail(&task->list, &list);
943 }
944
945 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
946
947 list_for_each_entry(task, &list, list)
948 task__print_level(task, fp, 0);
949
950 free(tasks);
951 return 0;
952 }
953
__cmd_report(struct report * rep)954 static int __cmd_report(struct report *rep)
955 {
956 int ret;
957 struct perf_session *session = rep->session;
958 struct evsel *pos;
959 struct perf_data *data = session->data;
960
961 signal(SIGINT, sig_handler);
962
963 if (rep->cpu_list) {
964 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
965 rep->cpu_bitmap);
966 if (ret) {
967 ui__error("failed to set cpu bitmap\n");
968 return ret;
969 }
970 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
971 }
972
973 if (rep->show_threads) {
974 ret = perf_read_values_init(&rep->show_threads_values);
975 if (ret)
976 return ret;
977 }
978
979 ret = report__setup_sample_type(rep);
980 if (ret) {
981 /* report__setup_sample_type() already showed error message */
982 return ret;
983 }
984
985 if (rep->stats_mode)
986 stats_setup(rep);
987
988 if (rep->tasks_mode)
989 tasks_setup(rep);
990
991 ret = perf_session__process_events(session);
992 if (ret) {
993 ui__error("failed to process sample\n");
994 return ret;
995 }
996
997 evlist__check_mem_load_aux(session->evlist);
998
999 if (rep->stats_mode)
1000 return stats_print(rep);
1001
1002 if (rep->tasks_mode)
1003 return tasks_print(rep, stdout);
1004
1005 report__warn_kptr_restrict(rep);
1006
1007 evlist__for_each_entry(session->evlist, pos)
1008 rep->nr_entries += evsel__hists(pos)->nr_entries;
1009
1010 if (use_browser == 0) {
1011 if (verbose > 3)
1012 perf_session__fprintf(session, stdout);
1013
1014 if (verbose > 2)
1015 perf_session__fprintf_dsos(session, stdout);
1016
1017 if (dump_trace) {
1018 perf_session__fprintf_nr_events(session, stdout,
1019 rep->skip_empty);
1020 evlist__fprintf_nr_events(session->evlist, stdout,
1021 rep->skip_empty);
1022 return 0;
1023 }
1024 }
1025
1026 ret = report__collapse_hists(rep);
1027 if (ret) {
1028 ui__error("failed to process hist entry\n");
1029 return ret;
1030 }
1031
1032 if (session_done())
1033 return 0;
1034
1035 /*
1036 * recalculate number of entries after collapsing since it
1037 * might be changed during the collapse phase.
1038 */
1039 rep->nr_entries = 0;
1040 evlist__for_each_entry(session->evlist, pos)
1041 rep->nr_entries += evsel__hists(pos)->nr_entries;
1042
1043 if (rep->nr_entries == 0) {
1044 ui__error("The %s data has no samples!\n", data->path);
1045 return 0;
1046 }
1047
1048 report__output_resort(rep);
1049
1050 if (rep->total_cycles_mode) {
1051 int block_hpps[6] = {
1052 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
1053 PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
1054 PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
1055 PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
1056 PERF_HPP_REPORT__BLOCK_RANGE,
1057 PERF_HPP_REPORT__BLOCK_DSO,
1058 };
1059
1060 rep->block_reports = block_info__create_report(session->evlist,
1061 rep->total_cycles,
1062 block_hpps, 6,
1063 &rep->nr_block_reports);
1064 if (!rep->block_reports)
1065 return -1;
1066 }
1067
1068 return report__browse_hists(rep);
1069 }
1070
1071 static int
report_parse_callchain_opt(const struct option * opt,const char * arg,int unset)1072 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1073 {
1074 struct callchain_param *callchain = opt->value;
1075
1076 callchain->enabled = !unset;
1077 /*
1078 * --no-call-graph
1079 */
1080 if (unset) {
1081 symbol_conf.use_callchain = false;
1082 callchain->mode = CHAIN_NONE;
1083 return 0;
1084 }
1085
1086 return parse_callchain_report_opt(arg);
1087 }
1088
1089 static int
parse_time_quantum(const struct option * opt,const char * arg,int unset __maybe_unused)1090 parse_time_quantum(const struct option *opt, const char *arg,
1091 int unset __maybe_unused)
1092 {
1093 unsigned long *time_q = opt->value;
1094 char *end;
1095
1096 *time_q = strtoul(arg, &end, 0);
1097 if (end == arg)
1098 goto parse_err;
1099 if (*time_q == 0) {
1100 pr_err("time quantum cannot be 0");
1101 return -1;
1102 }
1103 end = skip_spaces(end);
1104 if (*end == 0)
1105 return 0;
1106 if (!strcmp(end, "s")) {
1107 *time_q *= NSEC_PER_SEC;
1108 return 0;
1109 }
1110 if (!strcmp(end, "ms")) {
1111 *time_q *= NSEC_PER_MSEC;
1112 return 0;
1113 }
1114 if (!strcmp(end, "us")) {
1115 *time_q *= NSEC_PER_USEC;
1116 return 0;
1117 }
1118 if (!strcmp(end, "ns"))
1119 return 0;
1120 parse_err:
1121 pr_err("Cannot parse time quantum `%s'\n", arg);
1122 return -1;
1123 }
1124
1125 int
report_parse_ignore_callees_opt(const struct option * opt __maybe_unused,const char * arg,int unset __maybe_unused)1126 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1127 const char *arg, int unset __maybe_unused)
1128 {
1129 if (arg) {
1130 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1131 if (err) {
1132 char buf[BUFSIZ];
1133 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1134 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1135 return -1;
1136 }
1137 have_ignore_callees = 1;
1138 }
1139
1140 return 0;
1141 }
1142
1143 static int
parse_branch_mode(const struct option * opt,const char * str __maybe_unused,int unset)1144 parse_branch_mode(const struct option *opt,
1145 const char *str __maybe_unused, int unset)
1146 {
1147 int *branch_mode = opt->value;
1148
1149 *branch_mode = !unset;
1150 return 0;
1151 }
1152
1153 static int
parse_percent_limit(const struct option * opt,const char * str,int unset __maybe_unused)1154 parse_percent_limit(const struct option *opt, const char *str,
1155 int unset __maybe_unused)
1156 {
1157 struct report *rep = opt->value;
1158 double pcnt = strtof(str, NULL);
1159
1160 rep->min_percent = pcnt;
1161 callchain_param.min_percent = pcnt;
1162 return 0;
1163 }
1164
process_attr(struct perf_tool * tool __maybe_unused,union perf_event * event,struct evlist ** pevlist)1165 static int process_attr(struct perf_tool *tool __maybe_unused,
1166 union perf_event *event,
1167 struct evlist **pevlist)
1168 {
1169 u64 sample_type;
1170 int err;
1171
1172 err = perf_event__process_attr(tool, event, pevlist);
1173 if (err)
1174 return err;
1175
1176 /*
1177 * Check if we need to enable callchains based
1178 * on events sample_type.
1179 */
1180 sample_type = evlist__combined_sample_type(*pevlist);
1181 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
1182 return 0;
1183 }
1184
cmd_report(int argc,const char ** argv)1185 int cmd_report(int argc, const char **argv)
1186 {
1187 struct perf_session *session;
1188 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1189 struct stat st;
1190 bool has_br_stack = false;
1191 int branch_mode = -1;
1192 int last_key = 0;
1193 bool branch_call_mode = false;
1194 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1195 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1196 CALLCHAIN_REPORT_HELP
1197 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1198 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1199 const char * const report_usage[] = {
1200 "perf report [<options>]",
1201 NULL
1202 };
1203 struct report report = {
1204 .tool = {
1205 .sample = process_sample_event,
1206 .mmap = perf_event__process_mmap,
1207 .mmap2 = perf_event__process_mmap2,
1208 .comm = perf_event__process_comm,
1209 .namespaces = perf_event__process_namespaces,
1210 .cgroup = perf_event__process_cgroup,
1211 .exit = perf_event__process_exit,
1212 .fork = perf_event__process_fork,
1213 .lost = perf_event__process_lost,
1214 .read = process_read_event,
1215 .attr = process_attr,
1216 #ifdef HAVE_LIBTRACEEVENT
1217 .tracing_data = perf_event__process_tracing_data,
1218 #endif
1219 .build_id = perf_event__process_build_id,
1220 .id_index = perf_event__process_id_index,
1221 .auxtrace_info = perf_event__process_auxtrace_info,
1222 .auxtrace = perf_event__process_auxtrace,
1223 .event_update = perf_event__process_event_update,
1224 .feature = process_feature_event,
1225 .ordered_events = true,
1226 .ordering_requires_timestamps = true,
1227 },
1228 .max_stack = PERF_MAX_STACK_DEPTH,
1229 .pretty_printing_style = "normal",
1230 .socket_filter = -1,
1231 .skip_empty = true,
1232 };
1233 char *sort_order_help = sort_help("sort by key(s):");
1234 char *field_order_help = sort_help("output field(s): overhead period sample ");
1235 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL;
1236 const struct option options[] = {
1237 OPT_STRING('i', "input", &input_name, "file",
1238 "input file name"),
1239 OPT_INCR('v', "verbose", &verbose,
1240 "be more verbose (show symbol address, etc)"),
1241 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
1242 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1243 "dump raw trace in ASCII"),
1244 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1245 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1246 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1247 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1248 "file", "vmlinux pathname"),
1249 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1250 "don't load vmlinux even if found"),
1251 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1252 "file", "kallsyms pathname"),
1253 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1254 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1255 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1256 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1257 "Show a column with the number of samples"),
1258 OPT_BOOLEAN('T', "threads", &report.show_threads,
1259 "Show per-thread event counters"),
1260 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1261 "pretty printing style key: normal raw"),
1262 #ifdef HAVE_SLANG_SUPPORT
1263 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1264 #endif
1265 #ifdef HAVE_GTK2_SUPPORT
1266 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1267 #endif
1268 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1269 "Use the stdio interface"),
1270 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1271 OPT_BOOLEAN(0, "header-only", &report.header_only,
1272 "Show only data header."),
1273 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1274 sort_order_help),
1275 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1276 field_order_help),
1277 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1278 "Show sample percentage for different cpu modes"),
1279 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1280 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1281 OPT_STRING('p', "parent", &parent_pattern, "regex",
1282 "regex filter to identify parent, see: '--sort parent'"),
1283 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1284 "Only display entries with parent-match"),
1285 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1286 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1287 report_callchain_help, &report_parse_callchain_opt,
1288 callchain_default_opt),
1289 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1290 "Accumulate callchains of children and show total overhead as well. "
1291 "Enabled by default, use --no-children to disable."),
1292 OPT_INTEGER(0, "max-stack", &report.max_stack,
1293 "Set the maximum stack depth when parsing the callchain, "
1294 "anything beyond the specified depth will be ignored. "
1295 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1296 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1297 "alias for inverted call graph"),
1298 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1299 "ignore callees of these functions in call graphs",
1300 report_parse_ignore_callees_opt),
1301 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1302 "only consider symbols in these dsos"),
1303 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1304 "only consider symbols in these comms"),
1305 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1306 "only consider symbols in these pids"),
1307 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1308 "only consider symbols in these tids"),
1309 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1310 "only consider these symbols"),
1311 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1312 "only show symbols that (partially) match with this filter"),
1313 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1314 "width[,width...]",
1315 "don't try to adjust column width, use these fixed values"),
1316 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1317 "separator for columns, no spaces will be added between "
1318 "columns '.' is reserved."),
1319 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1320 "Only display entries resolved to a symbol"),
1321 OPT_CALLBACK(0, "symfs", NULL, "directory",
1322 "Look for files with symbols relative to this directory",
1323 symbol__config_symfs),
1324 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1325 "list of cpus to profile"),
1326 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1327 "Display extended information about perf.data file"),
1328 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src,
1329 "Interleave source code with assembly code (default)"),
1330 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw,
1331 "Display raw encoding of assembly instructions (default)"),
1332 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1333 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1334 OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix",
1335 "Add prefix to source file path names in programs (with --prefix-strip)"),
1336 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N",
1337 "Strip first N entries of source file path name in programs (with --prefix)"),
1338 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1339 "Show a column with the sum of periods"),
1340 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1341 "Show event group information together"),
1342 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1343 "Sort the output by the event at the index n in group. "
1344 "If n is invalid, sort by the first event. "
1345 "WARNING: should be used on grouped events."),
1346 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1347 "use branch records for per branch histogram filling",
1348 parse_branch_mode),
1349 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1350 "add last branch records to call history"),
1351 OPT_STRING(0, "objdump", &objdump_path, "path",
1352 "objdump binary to use for disassembly and annotations"),
1353 OPT_STRING(0, "addr2line", &addr2line_path, "path",
1354 "addr2line binary to use for line numbers"),
1355 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1356 "Disable symbol demangling"),
1357 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1358 "Enable kernel symbol demangling"),
1359 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1360 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1361 "Number of samples to save per histogram entry for individual browsing"),
1362 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1363 "Don't show entries under that percent", parse_percent_limit),
1364 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1365 "how to display percentage of filtered entries", parse_filter_percentage),
1366 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1367 "Instruction Tracing options\n" ITRACE_HELP,
1368 itrace_parse_synth_opts),
1369 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1370 "Show full source file name path for source lines"),
1371 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1372 "Show callgraph from reference event"),
1373 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1374 "Enable LBR callgraph stitching approach"),
1375 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1376 "only show processor socket that match with this filter"),
1377 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1378 "Show raw trace event output (do not use print fmt or plugins)"),
1379 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1380 "Show entries in a hierarchy"),
1381 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1382 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1383 stdio__config_color, "always"),
1384 OPT_STRING(0, "time", &report.time_str, "str",
1385 "Time span of interest (start,stop)"),
1386 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1387 "Show inline function"),
1388 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period",
1389 "Set percent type local/global-period/hits",
1390 annotate_parse_percent_type),
1391 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1392 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1393 "Set time quantum for time sort key (default 100ms)",
1394 parse_time_quantum),
1395 OPTS_EVSWITCH(&report.evswitch),
1396 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1397 "Sort all blocks by 'Sampled Cycles%'"),
1398 OPT_BOOLEAN(0, "disable-order", &report.disable_order,
1399 "Disable raw trace ordering"),
1400 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
1401 "Do not display empty (or dummy) events in the output"),
1402 OPT_END()
1403 };
1404 struct perf_data data = {
1405 .mode = PERF_DATA_MODE_READ,
1406 };
1407 int ret = hists__init();
1408 char sort_tmp[128];
1409
1410 if (ret < 0)
1411 goto exit;
1412
1413 /*
1414 * tasks_mode require access to exited threads to list those that are in
1415 * the data file. Off-cpu events are synthesized after other events and
1416 * reference exited threads.
1417 */
1418 symbol_conf.keep_exited_threads = true;
1419
1420 annotation_options__init(&annotate_opts);
1421
1422 ret = perf_config(report__config, &report);
1423 if (ret)
1424 goto exit;
1425
1426 argc = parse_options(argc, argv, options, report_usage, 0);
1427 if (argc) {
1428 /*
1429 * Special case: if there's an argument left then assume that
1430 * it's a symbol filter:
1431 */
1432 if (argc > 1)
1433 usage_with_options(report_usage, options);
1434
1435 report.symbol_filter_str = argv[0];
1436 }
1437
1438 if (disassembler_style) {
1439 annotate_opts.disassembler_style = strdup(disassembler_style);
1440 if (!annotate_opts.disassembler_style)
1441 return -ENOMEM;
1442 }
1443 if (objdump_path) {
1444 annotate_opts.objdump_path = strdup(objdump_path);
1445 if (!annotate_opts.objdump_path)
1446 return -ENOMEM;
1447 }
1448 if (addr2line_path) {
1449 symbol_conf.addr2line_path = strdup(addr2line_path);
1450 if (!symbol_conf.addr2line_path)
1451 return -ENOMEM;
1452 }
1453
1454 if (annotate_check_args(&annotate_opts) < 0) {
1455 ret = -EINVAL;
1456 goto exit;
1457 }
1458
1459 if (report.mmaps_mode)
1460 report.tasks_mode = true;
1461
1462 if (dump_trace && report.disable_order)
1463 report.tool.ordered_events = false;
1464
1465 if (quiet)
1466 perf_quiet_option();
1467
1468 ret = symbol__validate_sym_arguments();
1469 if (ret)
1470 goto exit;
1471
1472 if (report.inverted_callchain)
1473 callchain_param.order = ORDER_CALLER;
1474 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1475 callchain_param.order = ORDER_CALLER;
1476
1477 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1478 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1479 report.max_stack = itrace_synth_opts.callchain_sz;
1480
1481 if (!input_name || !strlen(input_name)) {
1482 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1483 input_name = "-";
1484 else
1485 input_name = "perf.data";
1486 }
1487
1488 data.path = input_name;
1489 data.force = symbol_conf.force;
1490
1491 repeat:
1492 session = perf_session__new(&data, &report.tool);
1493 if (IS_ERR(session)) {
1494 ret = PTR_ERR(session);
1495 goto exit;
1496 }
1497
1498 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1499 if (ret)
1500 goto exit;
1501
1502 if (zstd_init(&(session->zstd_data), 0) < 0)
1503 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1504
1505 if (report.queue_size) {
1506 ordered_events__set_alloc_size(&session->ordered_events,
1507 report.queue_size);
1508 }
1509
1510 session->itrace_synth_opts = &itrace_synth_opts;
1511
1512 report.session = session;
1513
1514 has_br_stack = perf_header__has_feat(&session->header,
1515 HEADER_BRANCH_STACK);
1516 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1517 has_br_stack = false;
1518
1519 setup_forced_leader(&report, session->evlist);
1520
1521 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) {
1522 parse_options_usage(NULL, options, "group-sort-idx", 0);
1523 ret = -EINVAL;
1524 goto error;
1525 }
1526
1527 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1528 has_br_stack = true;
1529
1530 if (has_br_stack && branch_call_mode)
1531 symbol_conf.show_branchflag_count = true;
1532
1533 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1534
1535 /*
1536 * Branch mode is a tristate:
1537 * -1 means default, so decide based on the file having branch data.
1538 * 0/1 means the user chose a mode.
1539 */
1540 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1541 !branch_call_mode) {
1542 sort__mode = SORT_MODE__BRANCH;
1543 symbol_conf.cumulate_callchain = false;
1544 }
1545 if (branch_call_mode) {
1546 callchain_param.key = CCKEY_ADDRESS;
1547 callchain_param.branch_callstack = true;
1548 symbol_conf.use_callchain = true;
1549 callchain_register_param(&callchain_param);
1550 if (sort_order == NULL)
1551 sort_order = "srcline,symbol,dso";
1552 }
1553
1554 if (report.mem_mode) {
1555 if (sort__mode == SORT_MODE__BRANCH) {
1556 pr_err("branch and mem mode incompatible\n");
1557 goto error;
1558 }
1559 sort__mode = SORT_MODE__MEMORY;
1560 symbol_conf.cumulate_callchain = false;
1561 }
1562
1563 if (symbol_conf.report_hierarchy) {
1564 /* disable incompatible options */
1565 symbol_conf.cumulate_callchain = false;
1566
1567 if (field_order) {
1568 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1569 parse_options_usage(report_usage, options, "F", 1);
1570 parse_options_usage(NULL, options, "hierarchy", 0);
1571 goto error;
1572 }
1573
1574 perf_hpp_list.need_collapse = true;
1575 }
1576
1577 if (report.use_stdio)
1578 use_browser = 0;
1579 #ifdef HAVE_SLANG_SUPPORT
1580 else if (report.use_tui)
1581 use_browser = 1;
1582 #endif
1583 #ifdef HAVE_GTK2_SUPPORT
1584 else if (report.use_gtk)
1585 use_browser = 2;
1586 #endif
1587
1588 /* Force tty output for header output and per-thread stat. */
1589 if (report.header || report.header_only || report.show_threads)
1590 use_browser = 0;
1591 if (report.header || report.header_only)
1592 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1593 if (report.show_full_info)
1594 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1595 if (report.stats_mode || report.tasks_mode)
1596 use_browser = 0;
1597 if (report.stats_mode && report.tasks_mode) {
1598 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1599 goto error;
1600 }
1601
1602 if (report.total_cycles_mode) {
1603 if (sort__mode != SORT_MODE__BRANCH)
1604 report.total_cycles_mode = false;
1605 else
1606 sort_order = NULL;
1607 }
1608
1609 if (strcmp(input_name, "-") != 0)
1610 setup_browser(true);
1611 else
1612 use_browser = 0;
1613
1614 if (sort_order && strstr(sort_order, "ipc")) {
1615 parse_options_usage(report_usage, options, "s", 1);
1616 goto error;
1617 }
1618
1619 if (sort_order && strstr(sort_order, "symbol")) {
1620 if (sort__mode == SORT_MODE__BRANCH) {
1621 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1622 sort_order, "ipc_lbr");
1623 report.symbol_ipc = true;
1624 } else {
1625 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1626 sort_order, "ipc_null");
1627 }
1628
1629 sort_order = sort_tmp;
1630 }
1631
1632 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1633 (setup_sorting(session->evlist) < 0)) {
1634 if (sort_order)
1635 parse_options_usage(report_usage, options, "s", 1);
1636 if (field_order)
1637 parse_options_usage(sort_order ? NULL : report_usage,
1638 options, "F", 1);
1639 goto error;
1640 }
1641
1642 if ((report.header || report.header_only) && !quiet) {
1643 perf_session__fprintf_info(session, stdout,
1644 report.show_full_info);
1645 if (report.header_only) {
1646 if (data.is_pipe) {
1647 /*
1648 * we need to process first few records
1649 * which contains PERF_RECORD_HEADER_FEATURE.
1650 */
1651 perf_session__process_events(session);
1652 }
1653 ret = 0;
1654 goto error;
1655 }
1656 } else if (use_browser == 0 && !quiet &&
1657 !report.stats_mode && !report.tasks_mode) {
1658 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1659 stdout);
1660 }
1661
1662 /*
1663 * Only in the TUI browser we are doing integrated annotation,
1664 * so don't allocate extra space that won't be used in the stdio
1665 * implementation.
1666 */
1667 if (ui__has_annotation() || report.symbol_ipc ||
1668 report.total_cycles_mode) {
1669 ret = symbol__annotation_init();
1670 if (ret < 0)
1671 goto error;
1672 /*
1673 * For searching by name on the "Browse map details".
1674 * providing it only in verbose mode not to bloat too
1675 * much struct symbol.
1676 */
1677 if (verbose > 0) {
1678 /*
1679 * XXX: Need to provide a less kludgy way to ask for
1680 * more space per symbol, the u32 is for the index on
1681 * the ui browser.
1682 * See symbol__browser_index.
1683 */
1684 symbol_conf.priv_size += sizeof(u32);
1685 }
1686 annotation_config__init(&annotate_opts);
1687 }
1688
1689 if (symbol__init(&session->header.env) < 0)
1690 goto error;
1691
1692 if (report.time_str) {
1693 ret = perf_time__parse_for_ranges(report.time_str, session,
1694 &report.ptime_range,
1695 &report.range_size,
1696 &report.range_num);
1697 if (ret < 0)
1698 goto error;
1699
1700 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1701 report.ptime_range,
1702 report.range_num);
1703 }
1704
1705 #ifdef HAVE_LIBTRACEEVENT
1706 if (session->tevent.pevent &&
1707 tep_set_function_resolver(session->tevent.pevent,
1708 machine__resolve_kernel_addr,
1709 &session->machines.host) < 0) {
1710 pr_err("%s: failed to set libtraceevent function resolver\n",
1711 __func__);
1712 return -1;
1713 }
1714 #endif
1715 sort__setup_elide(stdout);
1716
1717 ret = __cmd_report(&report);
1718 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1719 perf_session__delete(session);
1720 last_key = K_SWITCH_INPUT_DATA;
1721 goto repeat;
1722 } else
1723 ret = 0;
1724
1725 error:
1726 if (report.ptime_range) {
1727 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1728 zfree(&report.ptime_range);
1729 }
1730
1731 if (report.block_reports) {
1732 block_info__free_report(report.block_reports,
1733 report.nr_block_reports);
1734 report.block_reports = NULL;
1735 }
1736
1737 zstd_fini(&(session->zstd_data));
1738 perf_session__delete(session);
1739 exit:
1740 annotation_options__exit(&annotate_opts);
1741 free(sort_order_help);
1742 free(field_order_help);
1743 return ret;
1744 }
1745