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