cxgb4i,libcxgbi: add iSCSI DDP support
[linux/fpc-iii.git] / tools / perf / builtin-report.c
bloba87cb338bdf14b2d49c19b840fb99b3265e68997
1 /*
2 * builtin-report.c
4 * Builtin report command: Analyze the perf.data input file,
5 * look up and read DSOs and symbol information and display
6 * a histogram of results, along various sorting keys.
7 */
8 #include "builtin.h"
10 #include "util/util.h"
11 #include "util/cache.h"
13 #include "util/annotate.h"
14 #include "util/color.h"
15 #include <linux/list.h>
16 #include <linux/rbtree.h>
17 #include "util/symbol.h"
18 #include "util/callchain.h"
19 #include "util/strlist.h"
20 #include "util/values.h"
22 #include "perf.h"
23 #include "util/debug.h"
24 #include "util/evlist.h"
25 #include "util/evsel.h"
26 #include "util/header.h"
27 #include "util/session.h"
28 #include "util/tool.h"
30 #include <subcmd/parse-options.h>
31 #include <subcmd/exec-cmd.h>
32 #include "util/parse-events.h"
34 #include "util/thread.h"
35 #include "util/sort.h"
36 #include "util/hist.h"
37 #include "util/data.h"
38 #include "arch/common.h"
40 #include "util/auxtrace.h"
42 #include <dlfcn.h>
43 #include <linux/bitmap.h>
44 #include <linux/stringify.h>
46 struct report {
47 struct perf_tool tool;
48 struct perf_session *session;
49 bool use_tui, use_gtk, use_stdio;
50 bool show_full_info;
51 bool show_threads;
52 bool inverted_callchain;
53 bool mem_mode;
54 bool header;
55 bool header_only;
56 bool nonany_branch_mode;
57 int max_stack;
58 struct perf_read_values show_threads_values;
59 const char *pretty_printing_style;
60 const char *cpu_list;
61 const char *symbol_filter_str;
62 float min_percent;
63 u64 nr_entries;
64 u64 queue_size;
65 int socket_filter;
66 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
69 static int report__config(const char *var, const char *value, void *cb)
71 struct report *rep = cb;
73 if (!strcmp(var, "report.group")) {
74 symbol_conf.event_group = perf_config_bool(var, value);
75 return 0;
77 if (!strcmp(var, "report.percent-limit")) {
78 double pcnt = strtof(value, NULL);
80 rep->min_percent = pcnt;
81 callchain_param.min_percent = pcnt;
82 return 0;
84 if (!strcmp(var, "report.children")) {
85 symbol_conf.cumulate_callchain = perf_config_bool(var, value);
86 return 0;
88 if (!strcmp(var, "report.queue-size")) {
89 rep->queue_size = perf_config_u64(var, value);
90 return 0;
93 return 0;
96 static int hist_iter__report_callback(struct hist_entry_iter *iter,
97 struct addr_location *al, bool single,
98 void *arg)
100 int err = 0;
101 struct report *rep = arg;
102 struct hist_entry *he = iter->he;
103 struct perf_evsel *evsel = iter->evsel;
104 struct mem_info *mi;
105 struct branch_info *bi;
107 if (!ui__has_annotation())
108 return 0;
110 hist__account_cycles(iter->sample->branch_stack, al, iter->sample,
111 rep->nonany_branch_mode);
113 if (sort__mode == SORT_MODE__BRANCH) {
114 bi = he->branch_info;
115 err = addr_map_symbol__inc_samples(&bi->from, evsel->idx);
116 if (err)
117 goto out;
119 err = addr_map_symbol__inc_samples(&bi->to, evsel->idx);
121 } else if (rep->mem_mode) {
122 mi = he->mem_info;
123 err = addr_map_symbol__inc_samples(&mi->daddr, evsel->idx);
124 if (err)
125 goto out;
127 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
129 } else if (symbol_conf.cumulate_callchain) {
130 if (single)
131 err = hist_entry__inc_addr_samples(he, evsel->idx,
132 al->addr);
133 } else {
134 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
137 out:
138 return err;
141 static int process_sample_event(struct perf_tool *tool,
142 union perf_event *event,
143 struct perf_sample *sample,
144 struct perf_evsel *evsel,
145 struct machine *machine)
147 struct report *rep = container_of(tool, struct report, tool);
148 struct addr_location al;
149 struct hist_entry_iter iter = {
150 .evsel = evsel,
151 .sample = sample,
152 .hide_unresolved = symbol_conf.hide_unresolved,
153 .add_entry_cb = hist_iter__report_callback,
155 int ret = 0;
157 if (machine__resolve(machine, &al, sample) < 0) {
158 pr_debug("problem processing %d event, skipping it.\n",
159 event->header.type);
160 return -1;
163 if (symbol_conf.hide_unresolved && al.sym == NULL)
164 goto out_put;
166 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
167 goto out_put;
169 if (sort__mode == SORT_MODE__BRANCH) {
171 * A non-synthesized event might not have a branch stack if
172 * branch stacks have been synthesized (using itrace options).
174 if (!sample->branch_stack)
175 goto out_put;
176 iter.ops = &hist_iter_branch;
177 } else if (rep->mem_mode) {
178 iter.ops = &hist_iter_mem;
179 } else if (symbol_conf.cumulate_callchain) {
180 iter.ops = &hist_iter_cumulative;
181 } else {
182 iter.ops = &hist_iter_normal;
185 if (al.map != NULL)
186 al.map->dso->hit = 1;
188 ret = hist_entry_iter__add(&iter, &al, rep->max_stack, rep);
189 if (ret < 0)
190 pr_debug("problem adding hist entry, skipping event\n");
191 out_put:
192 addr_location__put(&al);
193 return ret;
196 static int process_read_event(struct perf_tool *tool,
197 union perf_event *event,
198 struct perf_sample *sample __maybe_unused,
199 struct perf_evsel *evsel,
200 struct machine *machine __maybe_unused)
202 struct report *rep = container_of(tool, struct report, tool);
204 if (rep->show_threads) {
205 const char *name = evsel ? perf_evsel__name(evsel) : "unknown";
206 perf_read_values_add_value(&rep->show_threads_values,
207 event->read.pid, event->read.tid,
208 event->read.id,
209 name,
210 event->read.value);
213 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
214 evsel ? perf_evsel__name(evsel) : "FAIL",
215 event->read.value);
217 return 0;
220 /* For pipe mode, sample_type is not currently set */
221 static int report__setup_sample_type(struct report *rep)
223 struct perf_session *session = rep->session;
224 u64 sample_type = perf_evlist__combined_sample_type(session->evlist);
225 bool is_pipe = perf_data_file__is_pipe(session->file);
227 if (session->itrace_synth_opts->callchain ||
228 (!is_pipe &&
229 perf_header__has_feat(&session->header, HEADER_AUXTRACE) &&
230 !session->itrace_synth_opts->set))
231 sample_type |= PERF_SAMPLE_CALLCHAIN;
233 if (session->itrace_synth_opts->last_branch)
234 sample_type |= PERF_SAMPLE_BRANCH_STACK;
236 if (!is_pipe && !(sample_type & PERF_SAMPLE_CALLCHAIN)) {
237 if (perf_hpp_list.parent) {
238 ui__error("Selected --sort parent, but no "
239 "callchain data. Did you call "
240 "'perf record' without -g?\n");
241 return -EINVAL;
243 if (symbol_conf.use_callchain) {
244 ui__error("Selected -g or --branch-history but no "
245 "callchain data. Did\n"
246 "you call 'perf record' without -g?\n");
247 return -1;
249 } else if (!callchain_param.enabled &&
250 callchain_param.mode != CHAIN_NONE &&
251 !symbol_conf.use_callchain) {
252 symbol_conf.use_callchain = true;
253 if (callchain_register_param(&callchain_param) < 0) {
254 ui__error("Can't register callchain params.\n");
255 return -EINVAL;
259 if (symbol_conf.cumulate_callchain) {
260 /* Silently ignore if callchain is missing */
261 if (!(sample_type & PERF_SAMPLE_CALLCHAIN)) {
262 symbol_conf.cumulate_callchain = false;
263 perf_hpp__cancel_cumulate();
267 if (sort__mode == SORT_MODE__BRANCH) {
268 if (!is_pipe &&
269 !(sample_type & PERF_SAMPLE_BRANCH_STACK)) {
270 ui__error("Selected -b but no branch data. "
271 "Did you call perf record without -b?\n");
272 return -1;
276 if (symbol_conf.use_callchain || symbol_conf.cumulate_callchain) {
277 if ((sample_type & PERF_SAMPLE_REGS_USER) &&
278 (sample_type & PERF_SAMPLE_STACK_USER))
279 callchain_param.record_mode = CALLCHAIN_DWARF;
280 else if (sample_type & PERF_SAMPLE_BRANCH_STACK)
281 callchain_param.record_mode = CALLCHAIN_LBR;
282 else
283 callchain_param.record_mode = CALLCHAIN_FP;
286 /* ??? handle more cases than just ANY? */
287 if (!(perf_evlist__combined_branch_type(session->evlist) &
288 PERF_SAMPLE_BRANCH_ANY))
289 rep->nonany_branch_mode = true;
291 return 0;
294 static void sig_handler(int sig __maybe_unused)
296 session_done = 1;
299 static size_t hists__fprintf_nr_sample_events(struct hists *hists, struct report *rep,
300 const char *evname, FILE *fp)
302 size_t ret;
303 char unit;
304 unsigned long nr_samples = hists->stats.nr_events[PERF_RECORD_SAMPLE];
305 u64 nr_events = hists->stats.total_period;
306 struct perf_evsel *evsel = hists_to_evsel(hists);
307 char buf[512];
308 size_t size = sizeof(buf);
309 int socked_id = hists->socket_filter;
311 if (symbol_conf.filter_relative) {
312 nr_samples = hists->stats.nr_non_filtered_samples;
313 nr_events = hists->stats.total_non_filtered_period;
316 if (perf_evsel__is_group_event(evsel)) {
317 struct perf_evsel *pos;
319 perf_evsel__group_desc(evsel, buf, size);
320 evname = buf;
322 for_each_group_member(pos, evsel) {
323 const struct hists *pos_hists = evsel__hists(pos);
325 if (symbol_conf.filter_relative) {
326 nr_samples += pos_hists->stats.nr_non_filtered_samples;
327 nr_events += pos_hists->stats.total_non_filtered_period;
328 } else {
329 nr_samples += pos_hists->stats.nr_events[PERF_RECORD_SAMPLE];
330 nr_events += pos_hists->stats.total_period;
335 nr_samples = convert_unit(nr_samples, &unit);
336 ret = fprintf(fp, "# Samples: %lu%c", nr_samples, unit);
337 if (evname != NULL)
338 ret += fprintf(fp, " of event '%s'", evname);
340 if (symbol_conf.show_ref_callgraph &&
341 strstr(evname, "call-graph=no")) {
342 ret += fprintf(fp, ", show reference callgraph");
345 if (rep->mem_mode) {
346 ret += fprintf(fp, "\n# Total weight : %" PRIu64, nr_events);
347 ret += fprintf(fp, "\n# Sort order : %s", sort_order ? : default_mem_sort_order);
348 } else
349 ret += fprintf(fp, "\n# Event count (approx.): %" PRIu64, nr_events);
351 if (socked_id > -1)
352 ret += fprintf(fp, "\n# Processor Socket: %d", socked_id);
354 return ret + fprintf(fp, "\n#\n");
357 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
358 struct report *rep,
359 const char *help)
361 struct perf_evsel *pos;
363 fprintf(stdout, "#\n# Total Lost Samples: %" PRIu64 "\n#\n", evlist->stats.total_lost_samples);
364 evlist__for_each(evlist, pos) {
365 struct hists *hists = evsel__hists(pos);
366 const char *evname = perf_evsel__name(pos);
368 if (symbol_conf.event_group &&
369 !perf_evsel__is_group_leader(pos))
370 continue;
372 hists__fprintf_nr_sample_events(hists, rep, evname, stdout);
373 hists__fprintf(hists, true, 0, 0, rep->min_percent, stdout);
374 fprintf(stdout, "\n\n");
377 if (sort_order == NULL &&
378 parent_pattern == default_parent_pattern)
379 fprintf(stdout, "#\n# (%s)\n#\n", help);
381 if (rep->show_threads) {
382 bool style = !strcmp(rep->pretty_printing_style, "raw");
383 perf_read_values_display(stdout, &rep->show_threads_values,
384 style);
385 perf_read_values_destroy(&rep->show_threads_values);
388 return 0;
391 static void report__warn_kptr_restrict(const struct report *rep)
393 struct map *kernel_map = machine__kernel_map(&rep->session->machines.host);
394 struct kmap *kernel_kmap = kernel_map ? map__kmap(kernel_map) : NULL;
396 if (kernel_map == NULL ||
397 (kernel_map->dso->hit &&
398 (kernel_kmap->ref_reloc_sym == NULL ||
399 kernel_kmap->ref_reloc_sym->addr == 0))) {
400 const char *desc =
401 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
402 "can't be resolved.";
404 if (kernel_map) {
405 const struct dso *kdso = kernel_map->dso;
406 if (!RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION])) {
407 desc = "If some relocation was applied (e.g. "
408 "kexec) symbols may be misresolved.";
412 ui__warning(
413 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
414 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
415 "Samples in kernel modules can't be resolved as well.\n\n",
416 desc);
420 static int report__gtk_browse_hists(struct report *rep, const char *help)
422 int (*hist_browser)(struct perf_evlist *evlist, const char *help,
423 struct hist_browser_timer *timer, float min_pcnt);
425 hist_browser = dlsym(perf_gtk_handle, "perf_evlist__gtk_browse_hists");
427 if (hist_browser == NULL) {
428 ui__error("GTK browser not found!\n");
429 return -1;
432 return hist_browser(rep->session->evlist, help, NULL, rep->min_percent);
435 static int report__browse_hists(struct report *rep)
437 int ret;
438 struct perf_session *session = rep->session;
439 struct perf_evlist *evlist = session->evlist;
440 const char *help = perf_tip(system_path(TIPDIR));
442 if (help == NULL) {
443 /* fallback for people who don't install perf ;-) */
444 help = perf_tip(DOCDIR);
445 if (help == NULL)
446 help = "Cannot load tips.txt file, please install perf!";
449 switch (use_browser) {
450 case 1:
451 ret = perf_evlist__tui_browse_hists(evlist, help, NULL,
452 rep->min_percent,
453 &session->header.env);
455 * Usually "ret" is the last pressed key, and we only
456 * care if the key notifies us to switch data file.
458 if (ret != K_SWITCH_INPUT_DATA)
459 ret = 0;
460 break;
461 case 2:
462 ret = report__gtk_browse_hists(rep, help);
463 break;
464 default:
465 ret = perf_evlist__tty_browse_hists(evlist, rep, help);
466 break;
469 return ret;
472 static int report__collapse_hists(struct report *rep)
474 struct ui_progress prog;
475 struct perf_evsel *pos;
476 int ret = 0;
478 ui_progress__init(&prog, rep->nr_entries, "Merging related events...");
480 evlist__for_each(rep->session->evlist, pos) {
481 struct hists *hists = evsel__hists(pos);
483 if (pos->idx == 0)
484 hists->symbol_filter_str = rep->symbol_filter_str;
486 hists->socket_filter = rep->socket_filter;
488 ret = hists__collapse_resort(hists, &prog);
489 if (ret < 0)
490 break;
492 /* Non-group events are considered as leader */
493 if (symbol_conf.event_group &&
494 !perf_evsel__is_group_leader(pos)) {
495 struct hists *leader_hists = evsel__hists(pos->leader);
497 hists__match(leader_hists, hists);
498 hists__link(leader_hists, hists);
502 ui_progress__finish();
503 return ret;
506 static void report__output_resort(struct report *rep)
508 struct ui_progress prog;
509 struct perf_evsel *pos;
511 ui_progress__init(&prog, rep->nr_entries, "Sorting events for output...");
513 evlist__for_each(rep->session->evlist, pos)
514 perf_evsel__output_resort(pos, &prog);
516 ui_progress__finish();
519 static int __cmd_report(struct report *rep)
521 int ret;
522 struct perf_session *session = rep->session;
523 struct perf_evsel *pos;
524 struct perf_data_file *file = session->file;
526 signal(SIGINT, sig_handler);
528 if (rep->cpu_list) {
529 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
530 rep->cpu_bitmap);
531 if (ret) {
532 ui__error("failed to set cpu bitmap\n");
533 return ret;
537 if (rep->show_threads)
538 perf_read_values_init(&rep->show_threads_values);
540 ret = report__setup_sample_type(rep);
541 if (ret) {
542 /* report__setup_sample_type() already showed error message */
543 return ret;
546 ret = perf_session__process_events(session);
547 if (ret) {
548 ui__error("failed to process sample\n");
549 return ret;
552 report__warn_kptr_restrict(rep);
554 evlist__for_each(session->evlist, pos)
555 rep->nr_entries += evsel__hists(pos)->nr_entries;
557 if (use_browser == 0) {
558 if (verbose > 3)
559 perf_session__fprintf(session, stdout);
561 if (verbose > 2)
562 perf_session__fprintf_dsos(session, stdout);
564 if (dump_trace) {
565 perf_session__fprintf_nr_events(session, stdout);
566 perf_evlist__fprintf_nr_events(session->evlist, stdout);
567 return 0;
571 ret = report__collapse_hists(rep);
572 if (ret) {
573 ui__error("failed to process hist entry\n");
574 return ret;
577 if (session_done())
578 return 0;
581 * recalculate number of entries after collapsing since it
582 * might be changed during the collapse phase.
584 rep->nr_entries = 0;
585 evlist__for_each(session->evlist, pos)
586 rep->nr_entries += evsel__hists(pos)->nr_entries;
588 if (rep->nr_entries == 0) {
589 ui__error("The %s file has no samples!\n", file->path);
590 return 0;
593 report__output_resort(rep);
595 return report__browse_hists(rep);
598 static int
599 report_parse_callchain_opt(const struct option *opt, const char *arg, int unset)
601 struct callchain_param *callchain = opt->value;
603 callchain->enabled = !unset;
605 * --no-call-graph
607 if (unset) {
608 symbol_conf.use_callchain = false;
609 callchain->mode = CHAIN_NONE;
610 return 0;
613 return parse_callchain_report_opt(arg);
617 report_parse_ignore_callees_opt(const struct option *opt __maybe_unused,
618 const char *arg, int unset __maybe_unused)
620 if (arg) {
621 int err = regcomp(&ignore_callees_regex, arg, REG_EXTENDED);
622 if (err) {
623 char buf[BUFSIZ];
624 regerror(err, &ignore_callees_regex, buf, sizeof(buf));
625 pr_err("Invalid --ignore-callees regex: %s\n%s", arg, buf);
626 return -1;
628 have_ignore_callees = 1;
631 return 0;
634 static int
635 parse_branch_mode(const struct option *opt __maybe_unused,
636 const char *str __maybe_unused, int unset)
638 int *branch_mode = opt->value;
640 *branch_mode = !unset;
641 return 0;
644 static int
645 parse_percent_limit(const struct option *opt, const char *str,
646 int unset __maybe_unused)
648 struct report *rep = opt->value;
649 double pcnt = strtof(str, NULL);
651 rep->min_percent = pcnt;
652 callchain_param.min_percent = pcnt;
653 return 0;
656 #define CALLCHAIN_DEFAULT_OPT "graph,0.5,caller,function,percent"
658 const char report_callchain_help[] = "Display call graph (stack chain/backtrace):\n\n"
659 CALLCHAIN_REPORT_HELP
660 "\n\t\t\t\tDefault: " CALLCHAIN_DEFAULT_OPT;
662 int cmd_report(int argc, const char **argv, const char *prefix __maybe_unused)
664 struct perf_session *session;
665 struct itrace_synth_opts itrace_synth_opts = { .set = 0, };
666 struct stat st;
667 bool has_br_stack = false;
668 int branch_mode = -1;
669 bool branch_call_mode = false;
670 char callchain_default_opt[] = CALLCHAIN_DEFAULT_OPT;
671 const char * const report_usage[] = {
672 "perf report [<options>]",
673 NULL
675 struct report report = {
676 .tool = {
677 .sample = process_sample_event,
678 .mmap = perf_event__process_mmap,
679 .mmap2 = perf_event__process_mmap2,
680 .comm = perf_event__process_comm,
681 .exit = perf_event__process_exit,
682 .fork = perf_event__process_fork,
683 .lost = perf_event__process_lost,
684 .read = process_read_event,
685 .attr = perf_event__process_attr,
686 .tracing_data = perf_event__process_tracing_data,
687 .build_id = perf_event__process_build_id,
688 .id_index = perf_event__process_id_index,
689 .auxtrace_info = perf_event__process_auxtrace_info,
690 .auxtrace = perf_event__process_auxtrace,
691 .ordered_events = true,
692 .ordering_requires_timestamps = true,
694 .max_stack = PERF_MAX_STACK_DEPTH,
695 .pretty_printing_style = "normal",
696 .socket_filter = -1,
698 const struct option options[] = {
699 OPT_STRING('i', "input", &input_name, "file",
700 "input file name"),
701 OPT_INCR('v', "verbose", &verbose,
702 "be more verbose (show symbol address, etc)"),
703 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
704 "dump raw trace in ASCII"),
705 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
706 "file", "vmlinux pathname"),
707 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
708 "file", "kallsyms pathname"),
709 OPT_BOOLEAN('f', "force", &symbol_conf.force, "don't complain, do it"),
710 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
711 "load module symbols - WARNING: use only with -k and LIVE kernel"),
712 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
713 "Show a column with the number of samples"),
714 OPT_BOOLEAN('T', "threads", &report.show_threads,
715 "Show per-thread event counters"),
716 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
717 "pretty printing style key: normal raw"),
718 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
719 OPT_BOOLEAN(0, "gtk", &report.use_gtk, "Use the GTK2 interface"),
720 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
721 "Use the stdio interface"),
722 OPT_BOOLEAN(0, "header", &report.header, "Show data header."),
723 OPT_BOOLEAN(0, "header-only", &report.header_only,
724 "Show only data header."),
725 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
726 "sort by key(s): pid, comm, dso, symbol, parent, cpu, srcline, ..."
727 " Please refer the man page for the complete list."),
728 OPT_STRING('F', "fields", &field_order, "key[,keys...]",
729 "output field(s): overhead, period, sample plus all of sort keys"),
730 OPT_BOOLEAN(0, "show-cpu-utilization", &symbol_conf.show_cpu_utilization,
731 "Show sample percentage for different cpu modes"),
732 OPT_BOOLEAN_FLAG(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
733 "Show sample percentage for different cpu modes", PARSE_OPT_HIDDEN),
734 OPT_STRING('p', "parent", &parent_pattern, "regex",
735 "regex filter to identify parent, see: '--sort parent'"),
736 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
737 "Only display entries with parent-match"),
738 OPT_CALLBACK_DEFAULT('g', "call-graph", &callchain_param,
739 "print_type,threshold[,print_limit],order,sort_key[,branch],value",
740 report_callchain_help, &report_parse_callchain_opt,
741 callchain_default_opt),
742 OPT_BOOLEAN(0, "children", &symbol_conf.cumulate_callchain,
743 "Accumulate callchains of children and show total overhead as well"),
744 OPT_INTEGER(0, "max-stack", &report.max_stack,
745 "Set the maximum stack depth when parsing the callchain, "
746 "anything beyond the specified depth will be ignored. "
747 "Default: kernel.perf_event_max_stack or " __stringify(PERF_MAX_STACK_DEPTH)),
748 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
749 "alias for inverted call graph"),
750 OPT_CALLBACK(0, "ignore-callees", NULL, "regex",
751 "ignore callees of these functions in call graphs",
752 report_parse_ignore_callees_opt),
753 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
754 "only consider symbols in these dsos"),
755 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
756 "only consider symbols in these comms"),
757 OPT_STRING(0, "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
758 "only consider symbols in these pids"),
759 OPT_STRING(0, "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
760 "only consider symbols in these tids"),
761 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
762 "only consider these symbols"),
763 OPT_STRING(0, "symbol-filter", &report.symbol_filter_str, "filter",
764 "only show symbols that (partially) match with this filter"),
765 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
766 "width[,width...]",
767 "don't try to adjust column width, use these fixed values"),
768 OPT_STRING_NOEMPTY('t', "field-separator", &symbol_conf.field_sep, "separator",
769 "separator for columns, no spaces will be added between "
770 "columns '.' is reserved."),
771 OPT_BOOLEAN('U', "hide-unresolved", &symbol_conf.hide_unresolved,
772 "Only display entries resolved to a symbol"),
773 OPT_CALLBACK(0, "symfs", NULL, "directory",
774 "Look for files with symbols relative to this directory",
775 symbol__config_symfs),
776 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
777 "list of cpus to profile"),
778 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
779 "Display extended information about perf.data file"),
780 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
781 "Interleave source code with assembly code (default)"),
782 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
783 "Display raw encoding of assembly instructions (default)"),
784 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
785 "Specify disassembler style (e.g. -M intel for intel syntax)"),
786 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
787 "Show a column with the sum of periods"),
788 OPT_BOOLEAN(0, "group", &symbol_conf.event_group,
789 "Show event group information together"),
790 OPT_CALLBACK_NOOPT('b', "branch-stack", &branch_mode, "",
791 "use branch records for per branch histogram filling",
792 parse_branch_mode),
793 OPT_BOOLEAN(0, "branch-history", &branch_call_mode,
794 "add last branch records to call history"),
795 OPT_STRING(0, "objdump", &objdump_path, "path",
796 "objdump binary to use for disassembly and annotations"),
797 OPT_BOOLEAN(0, "demangle", &symbol_conf.demangle,
798 "Disable symbol demangling"),
799 OPT_BOOLEAN(0, "demangle-kernel", &symbol_conf.demangle_kernel,
800 "Enable kernel symbol demangling"),
801 OPT_BOOLEAN(0, "mem-mode", &report.mem_mode, "mem access profile"),
802 OPT_CALLBACK(0, "percent-limit", &report, "percent",
803 "Don't show entries under that percent", parse_percent_limit),
804 OPT_CALLBACK(0, "percentage", NULL, "relative|absolute",
805 "how to display percentage of filtered entries", parse_filter_percentage),
806 OPT_CALLBACK_OPTARG(0, "itrace", &itrace_synth_opts, NULL, "opts",
807 "Instruction Tracing options",
808 itrace_parse_synth_opts),
809 OPT_BOOLEAN(0, "full-source-path", &srcline_full_filename,
810 "Show full source file name path for source lines"),
811 OPT_BOOLEAN(0, "show-ref-call-graph", &symbol_conf.show_ref_callgraph,
812 "Show callgraph from reference event"),
813 OPT_INTEGER(0, "socket-filter", &report.socket_filter,
814 "only show processor socket that match with this filter"),
815 OPT_BOOLEAN(0, "raw-trace", &symbol_conf.raw_trace,
816 "Show raw trace event output (do not use print fmt or plugins)"),
817 OPT_BOOLEAN(0, "hierarchy", &symbol_conf.report_hierarchy,
818 "Show entries in a hierarchy"),
819 OPT_END()
821 struct perf_data_file file = {
822 .mode = PERF_DATA_MODE_READ,
824 int ret = hists__init();
826 if (ret < 0)
827 return ret;
829 perf_config(report__config, &report);
831 argc = parse_options(argc, argv, options, report_usage, 0);
832 if (argc) {
834 * Special case: if there's an argument left then assume that
835 * it's a symbol filter:
837 if (argc > 1)
838 usage_with_options(report_usage, options);
840 report.symbol_filter_str = argv[0];
843 if (symbol_conf.vmlinux_name &&
844 access(symbol_conf.vmlinux_name, R_OK)) {
845 pr_err("Invalid file: %s\n", symbol_conf.vmlinux_name);
846 return -EINVAL;
848 if (symbol_conf.kallsyms_name &&
849 access(symbol_conf.kallsyms_name, R_OK)) {
850 pr_err("Invalid file: %s\n", symbol_conf.kallsyms_name);
851 return -EINVAL;
854 if (report.use_stdio)
855 use_browser = 0;
856 else if (report.use_tui)
857 use_browser = 1;
858 else if (report.use_gtk)
859 use_browser = 2;
861 if (report.inverted_callchain)
862 callchain_param.order = ORDER_CALLER;
863 if (symbol_conf.cumulate_callchain && !callchain_param.order_set)
864 callchain_param.order = ORDER_CALLER;
866 if (itrace_synth_opts.callchain &&
867 (int)itrace_synth_opts.callchain_sz > report.max_stack)
868 report.max_stack = itrace_synth_opts.callchain_sz;
870 if (!input_name || !strlen(input_name)) {
871 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
872 input_name = "-";
873 else
874 input_name = "perf.data";
877 file.path = input_name;
878 file.force = symbol_conf.force;
880 repeat:
881 session = perf_session__new(&file, false, &report.tool);
882 if (session == NULL)
883 return -1;
885 if (report.queue_size) {
886 ordered_events__set_alloc_size(&session->ordered_events,
887 report.queue_size);
890 session->itrace_synth_opts = &itrace_synth_opts;
892 report.session = session;
894 has_br_stack = perf_header__has_feat(&session->header,
895 HEADER_BRANCH_STACK);
897 if (itrace_synth_opts.last_branch)
898 has_br_stack = true;
901 * Branch mode is a tristate:
902 * -1 means default, so decide based on the file having branch data.
903 * 0/1 means the user chose a mode.
905 if (((branch_mode == -1 && has_br_stack) || branch_mode == 1) &&
906 !branch_call_mode) {
907 sort__mode = SORT_MODE__BRANCH;
908 symbol_conf.cumulate_callchain = false;
910 if (branch_call_mode) {
911 callchain_param.key = CCKEY_ADDRESS;
912 callchain_param.branch_callstack = 1;
913 symbol_conf.use_callchain = true;
914 callchain_register_param(&callchain_param);
915 if (sort_order == NULL)
916 sort_order = "srcline,symbol,dso";
919 if (report.mem_mode) {
920 if (sort__mode == SORT_MODE__BRANCH) {
921 pr_err("branch and mem mode incompatible\n");
922 goto error;
924 sort__mode = SORT_MODE__MEMORY;
925 symbol_conf.cumulate_callchain = false;
928 if (symbol_conf.report_hierarchy) {
929 /* disable incompatible options */
930 symbol_conf.event_group = false;
931 symbol_conf.cumulate_callchain = false;
933 if (field_order) {
934 pr_err("Error: --hierarchy and --fields options cannot be used together\n");
935 parse_options_usage(report_usage, options, "F", 1);
936 parse_options_usage(NULL, options, "hierarchy", 0);
937 goto error;
940 perf_hpp_list.need_collapse = true;
943 /* Force tty output for header output and per-thread stat. */
944 if (report.header || report.header_only || report.show_threads)
945 use_browser = 0;
947 if (strcmp(input_name, "-") != 0)
948 setup_browser(true);
949 else
950 use_browser = 0;
952 if (setup_sorting(session->evlist) < 0) {
953 if (sort_order)
954 parse_options_usage(report_usage, options, "s", 1);
955 if (field_order)
956 parse_options_usage(sort_order ? NULL : report_usage,
957 options, "F", 1);
958 goto error;
961 if (report.header || report.header_only) {
962 perf_session__fprintf_info(session, stdout,
963 report.show_full_info);
964 if (report.header_only) {
965 ret = 0;
966 goto error;
968 } else if (use_browser == 0) {
969 fputs("# To display the perf.data header info, please use --header/--header-only options.\n#\n",
970 stdout);
974 * Only in the TUI browser we are doing integrated annotation,
975 * so don't allocate extra space that won't be used in the stdio
976 * implementation.
978 if (ui__has_annotation()) {
979 symbol_conf.priv_size = sizeof(struct annotation);
980 machines__set_symbol_filter(&session->machines,
981 symbol__annotate_init);
983 * For searching by name on the "Browse map details".
984 * providing it only in verbose mode not to bloat too
985 * much struct symbol.
987 if (verbose) {
989 * XXX: Need to provide a less kludgy way to ask for
990 * more space per symbol, the u32 is for the index on
991 * the ui browser.
992 * See symbol__browser_index.
994 symbol_conf.priv_size += sizeof(u32);
995 symbol_conf.sort_by_name = true;
999 if (symbol__init(&session->header.env) < 0)
1000 goto error;
1002 sort__setup_elide(stdout);
1004 ret = __cmd_report(&report);
1005 if (ret == K_SWITCH_INPUT_DATA) {
1006 perf_session__delete(session);
1007 goto repeat;
1008 } else
1009 ret = 0;
1011 error:
1012 perf_session__delete(session);
1013 return ret;