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 if (ret != 0)
545 return ret;
546 }
547
548 return 0;
549 }
550
evlist__tty_browse_hists(struct evlist * evlist,struct report * rep,const char * help)551 static int evlist__tty_browse_hists(struct evlist *evlist, struct report *rep, const char *help)
552 {
553 struct evsel *pos;
554 int i = 0;
555
556 if (!quiet) {
557 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n",
558 evlist->stats.total_lost_samples);
559 }
560
561 evlist__for_each_entry(evlist, pos) {
562 struct hists *hists = evsel__hists(pos);
563 const char *evname = evsel__name(pos);
564
565 i++;
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 - 1].hist,
576 rep->min_percent, pos, NULL);
577 continue;
578 }
579
580 hists__fprintf(hists, !quiet, 0, 0, rep->min_percent, stdout,
581 !(symbol_conf.use_callchain ||
582 symbol_conf.show_branchflag_count));
583 fprintf(stdout, "\n\n");
584 }
585
586 if (!quiet)
587 fprintf(stdout, "#\n# (%s)\n#\n", help);
588
589 if (rep->show_threads) {
590 bool style = !strcmp(rep->pretty_printing_style, "raw");
591 perf_read_values_display(stdout, &rep->show_threads_values,
592 style);
593 perf_read_values_destroy(&rep->show_threads_values);
594 }
595
596 if (sort__mode == SORT_MODE__BRANCH)
597 branch_type_stat_display(stdout, &rep->brtype_stat);
598
599 return 0;
600 }
601
report__warn_kptr_restrict(const struct report * rep)602 static void report__warn_kptr_restrict(const struct report *rep)
603 {
604 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
605 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
606
607 if (evlist__exclude_kernel(rep->session->evlist))
608 return;
609
610 if (kernel_map == NULL ||
611 (map__dso(kernel_map)->hit &&
612 (kernel_kmap->ref_reloc_sym == NULL ||
613 kernel_kmap->ref_reloc_sym->addr == 0))) {
614 const char *desc =
615 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
616 "can't be resolved.";
617
618 if (kernel_map && map__has_symbols(kernel_map)) {
619 desc = "If some relocation was applied (e.g. "
620 "kexec) symbols may be misresolved.";
621 }
622
623 ui__warning(
624 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
625 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
626 "Samples in kernel modules can't be resolved as well.\n\n",
627 desc);
628 }
629 }
630
report__gtk_browse_hists(struct report * rep,const char * help)631 static int report__gtk_browse_hists(struct report *rep, const char *help)
632 {
633 int (*hist_browser)(struct evlist *evlist, const char *help,
634 struct hist_browser_timer *timer, float min_pcnt);
635
636 hist_browser = dlsym(perf_gtk_handle, "evlist__gtk_browse_hists");
637
638 if (hist_browser == NULL) {
639 ui__error("GTK browser not found!\n");
640 return -1;
641 }
642
643 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
644 }
645
report__browse_hists(struct report * rep)646 static int report__browse_hists(struct report *rep)
647 {
648 int ret;
649 struct perf_session *session = rep->session;
650 struct evlist *evlist = session->evlist;
651 char *help = NULL, *path = NULL;
652
653 path = system_path(TIPDIR);
654 if (perf_tip(&help, path) || help == NULL) {
655 /* fallback for people who don't install perf ;-) */
656 free(path);
657 path = system_path(DOCDIR);
658 if (perf_tip(&help, path) || help == NULL)
659 help = strdup("Cannot load tips.txt file, please install perf!");
660 }
661 free(path);
662
663 switch (use_browser) {
664 case 1:
665 if (rep->total_cycles_mode) {
666 ret = evlist__tui_block_hists_browse(evlist, rep);
667 break;
668 }
669
670 ret = evlist__tui_browse_hists(evlist, help, NULL, rep->min_percent,
671 &session->header.env, true);
672 /*
673 * Usually "ret" is the last pressed key, and we only
674 * care if the key notifies us to switch data file.
675 */
676 if (ret != K_SWITCH_INPUT_DATA && ret != K_RELOAD)
677 ret = 0;
678 break;
679 case 2:
680 ret = report__gtk_browse_hists(rep, help);
681 break;
682 default:
683 ret = evlist__tty_browse_hists(evlist, rep, help);
684 break;
685 }
686 free(help);
687 return ret;
688 }
689
report__collapse_hists(struct report * rep)690 static int report__collapse_hists(struct report *rep)
691 {
692 struct ui_progress prog;
693 struct evsel *pos;
694 int ret = 0;
695
696 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
697
698 evlist__for_each_entry(rep->session->evlist, pos) {
699 struct hists *hists = evsel__hists(pos);
700
701 if (pos->core.idx == 0)
702 hists->symbol_filter_str = rep->symbol_filter_str;
703
704 hists->socket_filter = rep->socket_filter;
705
706 ret = hists__collapse_resort(hists, &prog);
707 if (ret < 0)
708 break;
709
710 /* Non-group events are considered as leader */
711 if (symbol_conf.event_group && !evsel__is_group_leader(pos)) {
712 struct hists *leader_hists = evsel__hists(evsel__leader(pos));
713
714 hists__match(leader_hists, hists);
715 hists__link(leader_hists, hists);
716 }
717 }
718
719 ui_progress__finish();
720 return ret;
721 }
722
hists__resort_cb(struct hist_entry * he,void * arg)723 static int hists__resort_cb(struct hist_entry *he, void *arg)
724 {
725 struct report *rep = arg;
726 struct symbol *sym = he->ms.sym;
727
728 if (rep->symbol_ipc && sym && !sym->annotate2) {
729 struct evsel *evsel = hists_to_evsel(he->hists);
730
731 symbol__annotate2(&he->ms, evsel, NULL);
732 }
733
734 return 0;
735 }
736
report__output_resort(struct report * rep)737 static void report__output_resort(struct report *rep)
738 {
739 struct ui_progress prog;
740 struct evsel *pos;
741
742 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
743
744 evlist__for_each_entry(rep->session->evlist, pos) {
745 evsel__output_resort_cb(pos, &prog, hists__resort_cb, rep);
746 }
747
748 ui_progress__finish();
749 }
750
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)751 static int count_sample_event(struct perf_tool *tool __maybe_unused,
752 union perf_event *event __maybe_unused,
753 struct perf_sample *sample __maybe_unused,
754 struct evsel *evsel,
755 struct machine *machine __maybe_unused)
756 {
757 struct hists *hists = evsel__hists(evsel);
758
759 hists__inc_nr_events(hists);
760 return 0;
761 }
762
count_lost_samples_event(struct perf_tool * tool,union perf_event * event,struct perf_sample * sample,struct machine * machine __maybe_unused)763 static int count_lost_samples_event(struct perf_tool *tool,
764 union perf_event *event,
765 struct perf_sample *sample,
766 struct machine *machine __maybe_unused)
767 {
768 struct report *rep = container_of(tool, struct report, tool);
769 struct evsel *evsel;
770
771 evsel = evlist__id2evsel(rep->session->evlist, sample->id);
772 if (evsel) {
773 hists__inc_nr_lost_samples(evsel__hists(evsel),
774 event->lost_samples.lost);
775 }
776 return 0;
777 }
778
779 static int process_attr(struct perf_tool *tool __maybe_unused,
780 union perf_event *event,
781 struct evlist **pevlist);
782
stats_setup(struct report * rep)783 static void stats_setup(struct report *rep)
784 {
785 memset(&rep->tool, 0, sizeof(rep->tool));
786 rep->tool.attr = process_attr;
787 rep->tool.sample = count_sample_event;
788 rep->tool.lost_samples = count_lost_samples_event;
789 rep->tool.no_warn = true;
790 }
791
stats_print(struct report * rep)792 static int stats_print(struct report *rep)
793 {
794 struct perf_session *session = rep->session;
795
796 perf_session__fprintf_nr_events(session, stdout, rep->skip_empty);
797 evlist__fprintf_nr_events(session->evlist, stdout, rep->skip_empty);
798 return 0;
799 }
800
tasks_setup(struct report * rep)801 static void tasks_setup(struct report *rep)
802 {
803 memset(&rep->tool, 0, sizeof(rep->tool));
804 rep->tool.ordered_events = true;
805 if (rep->mmaps_mode) {
806 rep->tool.mmap = perf_event__process_mmap;
807 rep->tool.mmap2 = perf_event__process_mmap2;
808 }
809 rep->tool.attr = process_attr;
810 rep->tool.comm = perf_event__process_comm;
811 rep->tool.exit = perf_event__process_exit;
812 rep->tool.fork = perf_event__process_fork;
813 rep->tool.no_warn = true;
814 }
815
816 struct task {
817 struct thread *thread;
818 struct list_head list;
819 struct list_head children;
820 };
821
tasks_list(struct task * task,struct machine * machine)822 static struct task *tasks_list(struct task *task, struct machine *machine)
823 {
824 struct thread *parent_thread, *thread = task->thread;
825 struct task *parent_task;
826
827 /* Already listed. */
828 if (!list_empty(&task->list))
829 return NULL;
830
831 /* Last one in the chain. */
832 if (thread__ppid(thread) == -1)
833 return task;
834
835 parent_thread = machine__find_thread(machine, -1, thread__ppid(thread));
836 if (!parent_thread)
837 return ERR_PTR(-ENOENT);
838
839 parent_task = thread__priv(parent_thread);
840 thread__put(parent_thread);
841 list_add_tail(&task->list, &parent_task->children);
842 return tasks_list(parent_task, machine);
843 }
844
maps__fprintf_task(struct maps * maps,int indent,FILE * fp)845 static size_t maps__fprintf_task(struct maps *maps, int indent, FILE *fp)
846 {
847 size_t printed = 0;
848 struct map_rb_node *rb_node;
849
850 maps__for_each_entry(maps, rb_node) {
851 struct map *map = rb_node->map;
852 const struct dso *dso = map__dso(map);
853 u32 prot = map__prot(map);
854
855 printed += fprintf(fp, "%*s %" PRIx64 "-%" PRIx64 " %c%c%c%c %08" PRIx64 " %" PRIu64 " %s\n",
856 indent, "", map__start(map), map__end(map),
857 prot & PROT_READ ? 'r' : '-',
858 prot & PROT_WRITE ? 'w' : '-',
859 prot & PROT_EXEC ? 'x' : '-',
860 map__flags(map) ? 's' : 'p',
861 map__pgoff(map),
862 dso->id.ino, dso->name);
863 }
864
865 return printed;
866 }
867
task__print_level(struct task * task,FILE * fp,int level)868 static void task__print_level(struct task *task, FILE *fp, int level)
869 {
870 struct thread *thread = task->thread;
871 struct task *child;
872 int comm_indent = fprintf(fp, " %8d %8d %8d |%*s",
873 thread__pid(thread), thread__tid(thread),
874 thread__ppid(thread), level, "");
875
876 fprintf(fp, "%s\n", thread__comm_str(thread));
877
878 maps__fprintf_task(thread__maps(thread), comm_indent, fp);
879
880 if (!list_empty(&task->children)) {
881 list_for_each_entry(child, &task->children, list)
882 task__print_level(child, fp, level + 1);
883 }
884 }
885
tasks_print(struct report * rep,FILE * fp)886 static int tasks_print(struct report *rep, FILE *fp)
887 {
888 struct perf_session *session = rep->session;
889 struct machine *machine = &session->machines.host;
890 struct task *tasks, *task;
891 unsigned int nr = 0, itask = 0, i;
892 struct rb_node *nd;
893 LIST_HEAD(list);
894
895 /*
896 * No locking needed while accessing machine->threads,
897 * because --tasks is single threaded command.
898 */
899
900 /* Count all the threads. */
901 for (i = 0; i < THREADS__TABLE_SIZE; i++)
902 nr += machine->threads[i].nr;
903
904 tasks = malloc(sizeof(*tasks) * nr);
905 if (!tasks)
906 return -ENOMEM;
907
908 for (i = 0; i < THREADS__TABLE_SIZE; i++) {
909 struct threads *threads = &machine->threads[i];
910
911 for (nd = rb_first_cached(&threads->entries); nd;
912 nd = rb_next(nd)) {
913 task = tasks + itask++;
914
915 task->thread = rb_entry(nd, struct thread_rb_node, rb_node)->thread;
916 INIT_LIST_HEAD(&task->children);
917 INIT_LIST_HEAD(&task->list);
918 thread__set_priv(task->thread, task);
919 }
920 }
921
922 /*
923 * Iterate every task down to the unprocessed parent
924 * and link all in task children list. Task with no
925 * parent is added into 'list'.
926 */
927 for (itask = 0; itask < nr; itask++) {
928 task = tasks + itask;
929
930 if (!list_empty(&task->list))
931 continue;
932
933 task = tasks_list(task, machine);
934 if (IS_ERR(task)) {
935 pr_err("Error: failed to process tasks\n");
936 free(tasks);
937 return PTR_ERR(task);
938 }
939
940 if (task)
941 list_add_tail(&task->list, &list);
942 }
943
944 fprintf(fp, "# %8s %8s %8s %s\n", "pid", "tid", "ppid", "comm");
945
946 list_for_each_entry(task, &list, list)
947 task__print_level(task, fp, 0);
948
949 free(tasks);
950 return 0;
951 }
952
__cmd_report(struct report * rep)953 static int __cmd_report(struct report *rep)
954 {
955 int ret;
956 struct perf_session *session = rep->session;
957 struct evsel *pos;
958 struct perf_data *data = session->data;
959
960 signal(SIGINT, sig_handler);
961
962 if (rep->cpu_list) {
963 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
964 rep->cpu_bitmap);
965 if (ret) {
966 ui__error("failed to set cpu bitmap\n");
967 return ret;
968 }
969 session->itrace_synth_opts->cpu_bitmap = rep->cpu_bitmap;
970 }
971
972 if (rep->show_threads) {
973 ret = perf_read_values_init(&rep->show_threads_values);
974 if (ret)
975 return ret;
976 }
977
978 ret = report__setup_sample_type(rep);
979 if (ret) {
980 /* report__setup_sample_type() already showed error message */
981 return ret;
982 }
983
984 if (rep->stats_mode)
985 stats_setup(rep);
986
987 if (rep->tasks_mode)
988 tasks_setup(rep);
989
990 ret = perf_session__process_events(session);
991 if (ret) {
992 ui__error("failed to process sample\n");
993 return ret;
994 }
995
996 evlist__check_mem_load_aux(session->evlist);
997
998 if (rep->stats_mode)
999 return stats_print(rep);
1000
1001 if (rep->tasks_mode)
1002 return tasks_print(rep, stdout);
1003
1004 report__warn_kptr_restrict(rep);
1005
1006 evlist__for_each_entry(session->evlist, pos)
1007 rep->nr_entries += evsel__hists(pos)->nr_entries;
1008
1009 if (use_browser == 0) {
1010 if (verbose > 3)
1011 perf_session__fprintf(session, stdout);
1012
1013 if (verbose > 2)
1014 perf_session__fprintf_dsos(session, stdout);
1015
1016 if (dump_trace) {
1017 perf_session__fprintf_nr_events(session, stdout,
1018 rep->skip_empty);
1019 evlist__fprintf_nr_events(session->evlist, stdout,
1020 rep->skip_empty);
1021 return 0;
1022 }
1023 }
1024
1025 ret = report__collapse_hists(rep);
1026 if (ret) {
1027 ui__error("failed to process hist entry\n");
1028 return ret;
1029 }
1030
1031 if (session_done())
1032 return 0;
1033
1034 /*
1035 * recalculate number of entries after collapsing since it
1036 * might be changed during the collapse phase.
1037 */
1038 rep->nr_entries = 0;
1039 evlist__for_each_entry(session->evlist, pos)
1040 rep->nr_entries += evsel__hists(pos)->nr_entries;
1041
1042 if (rep->nr_entries == 0) {
1043 ui__error("The %s data has no samples!\n", data->path);
1044 return 0;
1045 }
1046
1047 report__output_resort(rep);
1048
1049 if (rep->total_cycles_mode) {
1050 int block_hpps[6] = {
1051 PERF_HPP_REPORT__BLOCK_TOTAL_CYCLES_PCT,
1052 PERF_HPP_REPORT__BLOCK_LBR_CYCLES,
1053 PERF_HPP_REPORT__BLOCK_CYCLES_PCT,
1054 PERF_HPP_REPORT__BLOCK_AVG_CYCLES,
1055 PERF_HPP_REPORT__BLOCK_RANGE,
1056 PERF_HPP_REPORT__BLOCK_DSO,
1057 };
1058
1059 rep->block_reports = block_info__create_report(session->evlist,
1060 rep->total_cycles,
1061 block_hpps, 6,
1062 &rep->nr_block_reports);
1063 if (!rep->block_reports)
1064 return -1;
1065 }
1066
1067 return report__browse_hists(rep);
1068 }
1069
1070 static int
report_parse_callchain_opt(const struct option * opt,const char * arg,int unset)1071 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
1072 {
1073 struct callchain_param *callchain = opt->value;
1074
1075 callchain->enabled = !unset;
1076 /*
1077 * --no-call-graph
1078 */
1079 if (unset) {
1080 symbol_conf.use_callchain = false;
1081 callchain->mode = CHAIN_NONE;
1082 return 0;
1083 }
1084
1085 return parse_callchain_report_opt(arg);
1086 }
1087
1088 static int
parse_time_quantum(const struct option * opt,const char * arg,int unset __maybe_unused)1089 parse_time_quantum(const struct option *opt, const char *arg,
1090 int unset __maybe_unused)
1091 {
1092 unsigned long *time_q = opt->value;
1093 char *end;
1094
1095 *time_q = strtoul(arg, &end, 0);
1096 if (end == arg)
1097 goto parse_err;
1098 if (*time_q == 0) {
1099 pr_err("time quantum cannot be 0");
1100 return -1;
1101 }
1102 end = skip_spaces(end);
1103 if (*end == 0)
1104 return 0;
1105 if (!strcmp(end, "s")) {
1106 *time_q *= NSEC_PER_SEC;
1107 return 0;
1108 }
1109 if (!strcmp(end, "ms")) {
1110 *time_q *= NSEC_PER_MSEC;
1111 return 0;
1112 }
1113 if (!strcmp(end, "us")) {
1114 *time_q *= NSEC_PER_USEC;
1115 return 0;
1116 }
1117 if (!strcmp(end, "ns"))
1118 return 0;
1119 parse_err:
1120 pr_err("Cannot parse time quantum `%s'\n", arg);
1121 return -1;
1122 }
1123
1124 int
report_parse_ignore_callees_opt(const struct option * opt __maybe_unused,const char * arg,int unset __maybe_unused)1125 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
1126 const char *arg, int unset __maybe_unused)
1127 {
1128 if (arg) {
1129 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
1130 if (err) {
1131 char buf[BUFSIZ];
1132 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
1133 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
1134 return -1;
1135 }
1136 have_ignore_callees = 1;
1137 }
1138
1139 return 0;
1140 }
1141
1142 static int
parse_branch_mode(const struct option * opt,const char * str __maybe_unused,int unset)1143 parse_branch_mode(const struct option *opt,
1144 const char *str __maybe_unused, int unset)
1145 {
1146 int *branch_mode = opt->value;
1147
1148 *branch_mode = !unset;
1149 return 0;
1150 }
1151
1152 static int
parse_percent_limit(const struct option * opt,const char * str,int unset __maybe_unused)1153 parse_percent_limit(const struct option *opt, const char *str,
1154 int unset __maybe_unused)
1155 {
1156 struct report *rep = opt->value;
1157 double pcnt = strtof(str, NULL);
1158
1159 rep->min_percent = pcnt;
1160 callchain_param.min_percent = pcnt;
1161 return 0;
1162 }
1163
process_attr(struct perf_tool * tool __maybe_unused,union perf_event * event,struct evlist ** pevlist)1164 static int process_attr(struct perf_tool *tool __maybe_unused,
1165 union perf_event *event,
1166 struct evlist **pevlist)
1167 {
1168 u64 sample_type;
1169 int err;
1170
1171 err = perf_event__process_attr(tool, event, pevlist);
1172 if (err)
1173 return err;
1174
1175 /*
1176 * Check if we need to enable callchains based
1177 * on events sample_type.
1178 */
1179 sample_type = evlist__combined_sample_type(*pevlist);
1180 callchain_param_setup(sample_type, perf_env__arch((*pevlist)->env));
1181 return 0;
1182 }
1183
cmd_report(int argc,const char ** argv)1184 int cmd_report(int argc, const char **argv)
1185 {
1186 struct perf_session *session;
1187 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
1188 struct stat st;
1189 bool has_br_stack = false;
1190 int branch_mode = -1;
1191 int last_key = 0;
1192 bool branch_call_mode = false;
1193 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
1194 static const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
1195 CALLCHAIN_REPORT_HELP
1196 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
1197 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
1198 const char * const report_usage[] = {
1199 "perf report [<options>]",
1200 NULL
1201 };
1202 struct report report = {
1203 .tool = {
1204 .sample = process_sample_event,
1205 .mmap = perf_event__process_mmap,
1206 .mmap2 = perf_event__process_mmap2,
1207 .comm = perf_event__process_comm,
1208 .namespaces = perf_event__process_namespaces,
1209 .cgroup = perf_event__process_cgroup,
1210 .exit = perf_event__process_exit,
1211 .fork = perf_event__process_fork,
1212 .lost = perf_event__process_lost,
1213 .read = process_read_event,
1214 .attr = process_attr,
1215 #ifdef HAVE_LIBTRACEEVENT
1216 .tracing_data = perf_event__process_tracing_data,
1217 #endif
1218 .build_id = perf_event__process_build_id,
1219 .id_index = perf_event__process_id_index,
1220 .auxtrace_info = perf_event__process_auxtrace_info,
1221 .auxtrace = perf_event__process_auxtrace,
1222 .event_update = perf_event__process_event_update,
1223 .feature = process_feature_event,
1224 .ordered_events = true,
1225 .ordering_requires_timestamps = true,
1226 },
1227 .max_stack = PERF_MAX_STACK_DEPTH,
1228 .pretty_printing_style = "normal",
1229 .socket_filter = -1,
1230 .skip_empty = true,
1231 };
1232 char *sort_order_help = sort_help("sort by key(s):");
1233 char *field_order_help = sort_help("output field(s): overhead period sample ");
1234 const char *disassembler_style = NULL, *objdump_path = NULL, *addr2line_path = NULL;
1235 const struct option options[] = {
1236 OPT_STRING('i', "input", &input_name, "file",
1237 "input file name"),
1238 OPT_INCR('v', "verbose", &verbose,
1239 "be more verbose (show symbol address, etc)"),
1240 OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
1241 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1242 "dump raw trace in ASCII"),
1243 OPT_BOOLEAN(0, "stats", &report.stats_mode, "Display event stats"),
1244 OPT_BOOLEAN(0, "tasks", &report.tasks_mode, "Display recorded tasks"),
1245 OPT_BOOLEAN(0, "mmaps", &report.mmaps_mode, "Display recorded tasks memory maps"),
1246 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
1247 "file", "vmlinux pathname"),
1248 OPT_BOOLEAN(0, "ignore-vmlinux", &symbol_conf.ignore_vmlinux,
1249 "don't load vmlinux even if found"),
1250 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
1251 "file", "kallsyms pathname"),
1252 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
1253 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
1254 "load module symbols - WARNING: use only with -k and LIVE kernel"),
1255 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
1256 "Show a column with the number of samples"),
1257 OPT_BOOLEAN('T', "threads", &report.show_threads,
1258 "Show per-thread event counters"),
1259 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
1260 "pretty printing style key: normal raw"),
1261 #ifdef HAVE_SLANG_SUPPORT
1262 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
1263 #endif
1264 #ifdef HAVE_GTK2_SUPPORT
1265 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
1266 #endif
1267 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
1268 "Use the stdio interface"),
1269 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
1270 OPT_BOOLEAN(0, "header-only", &report.header_only,
1271 "Show only data header."),
1272 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1273 sort_order_help),
1274 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
1275 field_order_help),
1276 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
1277 "Show sample percentage for different cpu modes"),
1278 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
1279 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
1280 OPT_STRING('p', "parent", &parent_pattern, "regex",
1281 "regex filter to identify parent, see: '--sort parent'"),
1282 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
1283 "Only display entries with parent-match"),
1284 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
1285 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
1286 report_callchain_help, &report_parse_callchain_opt,
1287 callchain_default_opt),
1288 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
1289 "Accumulate callchains of children and show total overhead as well. "
1290 "Enabled by default, use --no-children to disable."),
1291 OPT_INTEGER(0, "max-stack", &report.max_stack,
1292 "Set the maximum stack depth when parsing the callchain, "
1293 "anything beyond the specified depth will be ignored. "
1294 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
1295 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
1296 "alias for inverted call graph"),
1297 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
1298 "ignore callees of these functions in call graphs",
1299 report_parse_ignore_callees_opt),
1300 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
1301 "only consider symbols in these dsos"),
1302 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
1303 "only consider symbols in these comms"),
1304 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
1305 "only consider symbols in these pids"),
1306 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
1307 "only consider symbols in these tids"),
1308 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
1309 "only consider these symbols"),
1310 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
1311 "only show symbols that (partially) match with this filter"),
1312 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
1313 "width[,width...]",
1314 "don't try to adjust column width, use these fixed values"),
1315 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
1316 "separator for columns, no spaces will be added between "
1317 "columns '.' is reserved."),
1318 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
1319 "Only display entries resolved to a symbol"),
1320 OPT_CALLBACK(0, "symfs", NULL, "directory",
1321 "Look for files with symbols relative to this directory",
1322 symbol__config_symfs),
1323 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
1324 "list of cpus to profile"),
1325 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
1326 "Display extended information about perf.data file"),
1327 OPT_BOOLEAN(0, "source", &annotate_opts.annotate_src,
1328 "Interleave source code with assembly code (default)"),
1329 OPT_BOOLEAN(0, "asm-raw", &annotate_opts.show_asm_raw,
1330 "Display raw encoding of assembly instructions (default)"),
1331 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
1332 "Specify disassembler style (e.g. -M intel for intel syntax)"),
1333 OPT_STRING(0, "prefix", &annotate_opts.prefix, "prefix",
1334 "Add prefix to source file path names in programs (with --prefix-strip)"),
1335 OPT_STRING(0, "prefix-strip", &annotate_opts.prefix_strip, "N",
1336 "Strip first N entries of source file path name in programs (with --prefix)"),
1337 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
1338 "Show a column with the sum of periods"),
1339 OPT_BOOLEAN_SET(0, "group", &symbol_conf.event_group, &report.group_set,
1340 "Show event group information together"),
1341 OPT_INTEGER(0, "group-sort-idx", &symbol_conf.group_sort_idx,
1342 "Sort the output by the event at the index n in group. "
1343 "If n is invalid, sort by the first event. "
1344 "WARNING: should be used on grouped events."),
1345 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
1346 "use branch records for per branch histogram filling",
1347 parse_branch_mode),
1348 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
1349 "add last branch records to call history"),
1350 OPT_STRING(0, "objdump", &objdump_path, "path",
1351 "objdump binary to use for disassembly and annotations"),
1352 OPT_STRING(0, "addr2line", &addr2line_path, "path",
1353 "addr2line binary to use for line numbers"),
1354 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
1355 "Disable symbol demangling"),
1356 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
1357 "Enable kernel symbol demangling"),
1358 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
1359 OPT_INTEGER(0, "samples", &symbol_conf.res_sample,
1360 "Number of samples to save per histogram entry for individual browsing"),
1361 OPT_CALLBACK(0, "percent-limit", &report, "percent",
1362 "Don't show entries under that percent", parse_percent_limit),
1363 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
1364 "how to display percentage of filtered entries", parse_filter_percentage),
1365 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
1366 "Instruction Tracing options\n" ITRACE_HELP,
1367 itrace_parse_synth_opts),
1368 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
1369 "Show full source file name path for source lines"),
1370 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
1371 "Show callgraph from reference event"),
1372 OPT_BOOLEAN(0, "stitch-lbr", &report.stitch_lbr,
1373 "Enable LBR callgraph stitching approach"),
1374 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
1375 "only show processor socket that match with this filter"),
1376 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
1377 "Show raw trace event output (do not use print fmt or plugins)"),
1378 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
1379 "Show entries in a hierarchy"),
1380 OPT_CALLBACK_DEFAULT(0, "stdio-color", NULL, "mode",
1381 "'always' (default), 'never' or 'auto' only applicable to --stdio mode",
1382 stdio__config_color, "always"),
1383 OPT_STRING(0, "time", &report.time_str, "str",
1384 "Time span of interest (start,stop)"),
1385 OPT_BOOLEAN(0, "inline", &symbol_conf.inline_name,
1386 "Show inline function"),
1387 OPT_CALLBACK(0, "percent-type", &annotate_opts, "local-period",
1388 "Set percent type local/global-period/hits",
1389 annotate_parse_percent_type),
1390 OPT_BOOLEAN(0, "ns", &symbol_conf.nanosecs, "Show times in nanosecs"),
1391 OPT_CALLBACK(0, "time-quantum", &symbol_conf.time_quantum, "time (ms|us|ns|s)",
1392 "Set time quantum for time sort key (default 100ms)",
1393 parse_time_quantum),
1394 OPTS_EVSWITCH(&report.evswitch),
1395 OPT_BOOLEAN(0, "total-cycles", &report.total_cycles_mode,
1396 "Sort all blocks by 'Sampled Cycles%'"),
1397 OPT_BOOLEAN(0, "disable-order", &report.disable_order,
1398 "Disable raw trace ordering"),
1399 OPT_BOOLEAN(0, "skip-empty", &report.skip_empty,
1400 "Do not display empty (or dummy) events in the output"),
1401 OPT_END()
1402 };
1403 struct perf_data data = {
1404 .mode = PERF_DATA_MODE_READ,
1405 };
1406 int ret = hists__init();
1407 char sort_tmp[128];
1408
1409 if (ret < 0)
1410 goto exit;
1411
1412 /*
1413 * tasks_mode require access to exited threads to list those that are in
1414 * the data file. Off-cpu events are synthesized after other events and
1415 * reference exited threads.
1416 */
1417 symbol_conf.keep_exited_threads = true;
1418
1419 annotation_options__init(&annotate_opts);
1420
1421 ret = perf_config(report__config, &report);
1422 if (ret)
1423 goto exit;
1424
1425 argc = parse_options(argc, argv, options, report_usage, 0);
1426 if (argc) {
1427 /*
1428 * Special case: if there's an argument left then assume that
1429 * it's a symbol filter:
1430 */
1431 if (argc > 1)
1432 usage_with_options(report_usage, options);
1433
1434 report.symbol_filter_str = argv[0];
1435 }
1436
1437 if (disassembler_style) {
1438 annotate_opts.disassembler_style = strdup(disassembler_style);
1439 if (!annotate_opts.disassembler_style)
1440 return -ENOMEM;
1441 }
1442 if (objdump_path) {
1443 annotate_opts.objdump_path = strdup(objdump_path);
1444 if (!annotate_opts.objdump_path)
1445 return -ENOMEM;
1446 }
1447 if (addr2line_path) {
1448 symbol_conf.addr2line_path = strdup(addr2line_path);
1449 if (!symbol_conf.addr2line_path)
1450 return -ENOMEM;
1451 }
1452
1453 if (annotate_check_args(&annotate_opts) < 0) {
1454 ret = -EINVAL;
1455 goto exit;
1456 }
1457
1458 if (report.mmaps_mode)
1459 report.tasks_mode = true;
1460
1461 if (dump_trace && report.disable_order)
1462 report.tool.ordered_events = false;
1463
1464 if (quiet)
1465 perf_quiet_option();
1466
1467 ret = symbol__validate_sym_arguments();
1468 if (ret)
1469 goto exit;
1470
1471 if (report.inverted_callchain)
1472 callchain_param.order = ORDER_CALLER;
1473 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
1474 callchain_param.order = ORDER_CALLER;
1475
1476 if ((itrace_synth_opts.callchain || itrace_synth_opts.add_callchain) &&
1477 (int)itrace_synth_opts.callchain_sz > report.max_stack)
1478 report.max_stack = itrace_synth_opts.callchain_sz;
1479
1480 if (!input_name || !strlen(input_name)) {
1481 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
1482 input_name = "-";
1483 else
1484 input_name = "perf.data";
1485 }
1486
1487 data.path = input_name;
1488 data.force = symbol_conf.force;
1489
1490 repeat:
1491 session = perf_session__new(&data, &report.tool);
1492 if (IS_ERR(session)) {
1493 ret = PTR_ERR(session);
1494 goto exit;
1495 }
1496
1497 ret = evswitch__init(&report.evswitch, session->evlist, stderr);
1498 if (ret)
1499 goto exit;
1500
1501 if (zstd_init(&(session->zstd_data), 0) < 0)
1502 pr_warning("Decompression initialization failed. Reported data may be incomplete.\n");
1503
1504 if (report.queue_size) {
1505 ordered_events__set_alloc_size(&session->ordered_events,
1506 report.queue_size);
1507 }
1508
1509 session->itrace_synth_opts = &itrace_synth_opts;
1510
1511 report.session = session;
1512
1513 has_br_stack = perf_header__has_feat(&session->header,
1514 HEADER_BRANCH_STACK);
1515 if (evlist__combined_sample_type(session->evlist) & PERF_SAMPLE_STACK_USER)
1516 has_br_stack = false;
1517
1518 setup_forced_leader(&report, session->evlist);
1519
1520 if (symbol_conf.group_sort_idx && evlist__nr_groups(session->evlist) == 0) {
1521 parse_options_usage(NULL, options, "group-sort-idx", 0);
1522 ret = -EINVAL;
1523 goto error;
1524 }
1525
1526 if (itrace_synth_opts.last_branch || itrace_synth_opts.add_last_branch)
1527 has_br_stack = true;
1528
1529 if (has_br_stack && branch_call_mode)
1530 symbol_conf.show_branchflag_count = true;
1531
1532 memset(&report.brtype_stat, 0, sizeof(struct branch_type_stat));
1533
1534 /*
1535 * Branch mode is a tristate:
1536 * -1 means default, so decide based on the file having branch data.
1537 * 0/1 means the user chose a mode.
1538 */
1539 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
1540 !branch_call_mode) {
1541 sort__mode = SORT_MODE__BRANCH;
1542 symbol_conf.cumulate_callchain = false;
1543 }
1544 if (branch_call_mode) {
1545 callchain_param.key = CCKEY_ADDRESS;
1546 callchain_param.branch_callstack = true;
1547 symbol_conf.use_callchain = true;
1548 callchain_register_param(&callchain_param);
1549 if (sort_order == NULL)
1550 sort_order = "srcline,symbol,dso";
1551 }
1552
1553 if (report.mem_mode) {
1554 if (sort__mode == SORT_MODE__BRANCH) {
1555 pr_err("branch and mem mode incompatible\n");
1556 goto error;
1557 }
1558 sort__mode = SORT_MODE__MEMORY;
1559 symbol_conf.cumulate_callchain = false;
1560 }
1561
1562 if (symbol_conf.report_hierarchy) {
1563 /* disable incompatible options */
1564 symbol_conf.cumulate_callchain = false;
1565
1566 if (field_order) {
1567 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
1568 parse_options_usage(report_usage, options, "F", 1);
1569 parse_options_usage(NULL, options, "hierarchy", 0);
1570 goto error;
1571 }
1572
1573 perf_hpp_list.need_collapse = true;
1574 }
1575
1576 if (report.use_stdio)
1577 use_browser = 0;
1578 #ifdef HAVE_SLANG_SUPPORT
1579 else if (report.use_tui)
1580 use_browser = 1;
1581 #endif
1582 #ifdef HAVE_GTK2_SUPPORT
1583 else if (report.use_gtk)
1584 use_browser = 2;
1585 #endif
1586
1587 /* Force tty output for header output and per-thread stat. */
1588 if (report.header || report.header_only || report.show_threads)
1589 use_browser = 0;
1590 if (report.header || report.header_only)
1591 report.tool.show_feat_hdr = SHOW_FEAT_HEADER;
1592 if (report.show_full_info)
1593 report.tool.show_feat_hdr = SHOW_FEAT_HEADER_FULL_INFO;
1594 if (report.stats_mode || report.tasks_mode)
1595 use_browser = 0;
1596 if (report.stats_mode && report.tasks_mode) {
1597 pr_err("Error: --tasks and --mmaps can't be used together with --stats\n");
1598 goto error;
1599 }
1600
1601 if (report.total_cycles_mode) {
1602 if (sort__mode != SORT_MODE__BRANCH)
1603 report.total_cycles_mode = false;
1604 else
1605 sort_order = NULL;
1606 }
1607
1608 if (strcmp(input_name, "-") != 0)
1609 setup_browser(true);
1610 else
1611 use_browser = 0;
1612
1613 if (sort_order && strstr(sort_order, "ipc")) {
1614 parse_options_usage(report_usage, options, "s", 1);
1615 goto error;
1616 }
1617
1618 if (sort_order && strstr(sort_order, "symbol")) {
1619 if (sort__mode == SORT_MODE__BRANCH) {
1620 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1621 sort_order, "ipc_lbr");
1622 report.symbol_ipc = true;
1623 } else {
1624 snprintf(sort_tmp, sizeof(sort_tmp), "%s,%s",
1625 sort_order, "ipc_null");
1626 }
1627
1628 sort_order = sort_tmp;
1629 }
1630
1631 if ((last_key != K_SWITCH_INPUT_DATA && last_key != K_RELOAD) &&
1632 (setup_sorting(session->evlist) < 0)) {
1633 if (sort_order)
1634 parse_options_usage(report_usage, options, "s", 1);
1635 if (field_order)
1636 parse_options_usage(sort_order ? NULL : report_usage,
1637 options, "F", 1);
1638 goto error;
1639 }
1640
1641 if ((report.header || report.header_only) && !quiet) {
1642 perf_session__fprintf_info(session, stdout,
1643 report.show_full_info);
1644 if (report.header_only) {
1645 if (data.is_pipe) {
1646 /*
1647 * we need to process first few records
1648 * which contains PERF_RECORD_HEADER_FEATURE.
1649 */
1650 perf_session__process_events(session);
1651 }
1652 ret = 0;
1653 goto error;
1654 }
1655 } else if (use_browser == 0 && !quiet &&
1656 !report.stats_mode && !report.tasks_mode) {
1657 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
1658 stdout);
1659 }
1660
1661 /*
1662 * Only in the TUI browser we are doing integrated annotation,
1663 * so don't allocate extra space that won't be used in the stdio
1664 * implementation.
1665 */
1666 if (ui__has_annotation() || report.symbol_ipc ||
1667 report.total_cycles_mode) {
1668 ret = symbol__annotation_init();
1669 if (ret < 0)
1670 goto error;
1671 /*
1672 * For searching by name on the "Browse map details".
1673 * providing it only in verbose mode not to bloat too
1674 * much struct symbol.
1675 */
1676 if (verbose > 0) {
1677 /*
1678 * XXX: Need to provide a less kludgy way to ask for
1679 * more space per symbol, the u32 is for the index on
1680 * the ui browser.
1681 * See symbol__browser_index.
1682 */
1683 symbol_conf.priv_size += sizeof(u32);
1684 }
1685 annotation_config__init(&annotate_opts);
1686 }
1687
1688 if (symbol__init(&session->header.env) < 0)
1689 goto error;
1690
1691 if (report.time_str) {
1692 ret = perf_time__parse_for_ranges(report.time_str, session,
1693 &report.ptime_range,
1694 &report.range_size,
1695 &report.range_num);
1696 if (ret < 0)
1697 goto error;
1698
1699 itrace_synth_opts__set_time_range(&itrace_synth_opts,
1700 report.ptime_range,
1701 report.range_num);
1702 }
1703
1704 #ifdef HAVE_LIBTRACEEVENT
1705 if (session->tevent.pevent &&
1706 tep_set_function_resolver(session->tevent.pevent,
1707 machine__resolve_kernel_addr,
1708 &session->machines.host) < 0) {
1709 pr_err("%s: failed to set libtraceevent function resolver\n",
1710 __func__);
1711 return -1;
1712 }
1713 #endif
1714 sort__setup_elide(stdout);
1715
1716 ret = __cmd_report(&report);
1717 if (ret == K_SWITCH_INPUT_DATA || ret == K_RELOAD) {
1718 perf_session__delete(session);
1719 last_key = K_SWITCH_INPUT_DATA;
1720 goto repeat;
1721 } else
1722 ret = 0;
1723
1724 error:
1725 if (report.ptime_range) {
1726 itrace_synth_opts__clear_time_range(&itrace_synth_opts);
1727 zfree(&report.ptime_range);
1728 }
1729
1730 if (report.block_reports) {
1731 block_info__free_report(report.block_reports,
1732 report.nr_block_reports);
1733 report.block_reports = NULL;
1734 }
1735
1736 zstd_fini(&(session->zstd_data));
1737 perf_session__delete(session);
1738 exit:
1739 annotation_options__exit(&annotate_opts);
1740 free(sort_order_help);
1741 free(field_order_help);
1742 return ret;
1743 }
1744