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.
10 #include "util/util.h"
12 #include "util/annotate.h"
13 #include "util/color.h"
14 #include <linux/list.h>
15 #include "util/cache.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"
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 "util/parse-options.h"
31 #include "util/parse-events.h"
33 #include "util/thread.h"
34 #include "util/sort.h"
35 #include "util/hist.h"
37 #include <linux/bitmap.h>
40 struct perf_tool tool;
41 struct perf_session *session;
42 char const *input_name;
43 bool force, use_tui, use_stdio;
45 bool dont_use_callchains;
48 bool inverted_callchain;
49 struct perf_read_values show_threads_values;
50 const char *pretty_printing_style;
51 symbol_filter_t annotate_init;
53 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
56 static int perf_report__add_branch_hist_entry(struct perf_tool *tool,
57 struct addr_location *al,
58 struct perf_sample *sample,
59 struct perf_evsel *evsel,
60 struct machine *machine)
62 struct perf_report *rep = container_of(tool, struct perf_report, tool);
63 struct symbol *parent = NULL;
66 struct hist_entry *he;
67 struct branch_info *bi;
69 if ((sort__has_parent || symbol_conf.use_callchain)
70 && sample->callchain) {
71 err = machine__resolve_callchain(machine, evsel, al->thread,
72 sample->callchain, &parent);
77 bi = machine__resolve_bstack(machine, al->thread,
78 sample->branch_stack);
82 for (i = 0; i < sample->branch_stack->nr; i++) {
83 if (rep->hide_unresolved && !(bi[i].from.sym && bi[i].to.sym))
86 * The report shows the percentage of total branches captured
87 * and not events sampled. Thus we use a pseudo period of 1.
89 he = __hists__add_branch_entry(&evsel->hists, al, parent,
92 evsel->hists.stats.total_period += 1;
93 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
100 static int perf_evsel__add_hist_entry(struct perf_evsel *evsel,
101 struct addr_location *al,
102 struct perf_sample *sample,
103 struct machine *machine)
105 struct symbol *parent = NULL;
107 struct hist_entry *he;
109 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
110 err = machine__resolve_callchain(machine, evsel, al->thread,
111 sample->callchain, &parent);
116 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
120 if (symbol_conf.use_callchain) {
121 err = callchain_append(he->callchain,
122 &evsel->hists.callchain_cursor,
128 * Only in the newt browser we are doing integrated annotation,
129 * so we don't allocated the extra space needed because the stdio
130 * code will not use it.
132 if (al->sym != NULL && use_browser > 0) {
133 struct annotation *notes = symbol__annotation(he->ms.sym);
135 assert(evsel != NULL);
138 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
141 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
144 evsel->hists.stats.total_period += sample->period;
145 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
151 static int process_sample_event(struct perf_tool *tool,
152 union perf_event *event,
153 struct perf_sample *sample,
154 struct perf_evsel *evsel,
155 struct machine *machine)
157 struct perf_report *rep = container_of(tool, struct perf_report, tool);
158 struct addr_location al;
160 if (perf_event__preprocess_sample(event, machine, &al, sample,
161 rep->annotate_init) < 0) {
162 fprintf(stderr, "problem processing %d event, skipping it.\n",
167 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
170 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
173 if (sort__branch_mode == 1) {
174 if (perf_report__add_branch_hist_entry(tool, &al, sample,
176 pr_debug("problem adding lbr entry, skipping event\n");
181 al.map->dso->hit = 1;
183 if (perf_evsel__add_hist_entry(evsel, &al, sample, machine)) {
184 pr_debug("problem incrementing symbol period, skipping event\n");
191 static int process_read_event(struct perf_tool *tool,
192 union perf_event *event,
193 struct perf_sample *sample __used,
194 struct perf_evsel *evsel,
195 struct machine *machine __used)
197 struct perf_report *rep = container_of(tool, struct perf_report, tool);
199 if (rep->show_threads) {
200 const char *name = evsel ? event_name(evsel) : "unknown";
201 perf_read_values_add_value(&rep->show_threads_values,
202 event->read.pid, event->read.tid,
208 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
209 evsel ? event_name(evsel) : "FAIL",
215 static int perf_report__setup_sample_type(struct perf_report *rep)
217 struct perf_session *self = rep->session;
219 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
220 if (sort__has_parent) {
221 ui__warning("Selected --sort parent, but no "
222 "callchain data. Did you call "
223 "'perf record' without -g?\n");
226 if (symbol_conf.use_callchain) {
227 ui__warning("Selected -g but no callchain data. Did "
228 "you call 'perf record' without -g?\n");
231 } else if (!rep->dont_use_callchains &&
232 callchain_param.mode != CHAIN_NONE &&
233 !symbol_conf.use_callchain) {
234 symbol_conf.use_callchain = true;
235 if (callchain_register_param(&callchain_param) < 0) {
236 ui__warning("Can't register callchain "
242 if (sort__branch_mode == 1) {
243 if (!(self->sample_type & PERF_SAMPLE_BRANCH_STACK)) {
244 fprintf(stderr, "selected -b but no branch data."
245 " Did you call perf record without"
254 extern volatile int session_done;
256 static void sig_handler(int sig __used)
261 static size_t hists__fprintf_nr_sample_events(struct hists *self,
262 const char *evname, FILE *fp)
266 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
268 nr_events = convert_unit(nr_events, &unit);
269 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
271 ret += fprintf(fp, " %s", evname);
272 return ret + fprintf(fp, "\n#\n");
275 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
276 struct perf_report *rep,
279 struct perf_evsel *pos;
281 list_for_each_entry(pos, &evlist->entries, node) {
282 struct hists *hists = &pos->hists;
283 const char *evname = event_name(pos);
285 hists__fprintf_nr_sample_events(hists, evname, stdout);
286 hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
287 fprintf(stdout, "\n\n");
290 if (sort_order == default_sort_order &&
291 parent_pattern == default_parent_pattern) {
292 fprintf(stdout, "#\n# (%s)\n#\n", help);
294 if (rep->show_threads) {
295 bool style = !strcmp(rep->pretty_printing_style, "raw");
296 perf_read_values_display(stdout, &rep->show_threads_values,
298 perf_read_values_destroy(&rep->show_threads_values);
305 static int __cmd_report(struct perf_report *rep)
309 struct perf_session *session = rep->session;
310 struct perf_evsel *pos;
311 struct map *kernel_map;
312 struct kmap *kernel_kmap;
313 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
315 signal(SIGINT, sig_handler);
318 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
324 if (use_browser <= 0)
325 perf_session__fprintf_info(session, stdout, rep->show_full_info);
327 if (rep->show_threads)
328 perf_read_values_init(&rep->show_threads_values);
330 ret = perf_report__setup_sample_type(rep);
334 ret = perf_session__process_events(session, &rep->tool);
338 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
339 kernel_kmap = map__kmap(kernel_map);
340 if (kernel_map == NULL ||
341 (kernel_map->dso->hit &&
342 (kernel_kmap->ref_reloc_sym == NULL ||
343 kernel_kmap->ref_reloc_sym->addr == 0))) {
344 const struct dso *kdso = kernel_map->dso;
347 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
348 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
349 "Samples in kernel modules can't be resolved as well.\n\n",
350 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
351 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
352 "can't be resolved." :
353 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
357 perf_session__fprintf_nr_events(session, stdout);
362 perf_session__fprintf(session, stdout);
365 perf_session__fprintf_dsos(session, stdout);
368 list_for_each_entry(pos, &session->evlist->entries, node) {
369 struct hists *hists = &pos->hists;
371 hists__collapse_resort(hists);
372 hists__output_resort(hists);
373 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
376 if (nr_samples == 0) {
377 ui__warning("The %s file has no samples!\n", session->filename);
381 if (use_browser > 0) {
382 perf_evlist__tui_browse_hists(session->evlist, help,
385 perf_evlist__tty_browse_hists(session->evlist, rep, help);
389 * Speed up the exit process, for large files this can
390 * take quite a while.
392 * XXX Enable this when using valgrind or if we ever
393 * librarize this command.
395 * Also experiment with obstacks to see how much speed
398 * perf_session__delete(session);
404 parse_callchain_opt(const struct option *opt, const char *arg, int unset)
406 struct perf_report *rep = (struct perf_report *)opt->value;
414 rep->dont_use_callchains = true;
418 symbol_conf.use_callchain = true;
423 tok = strtok((char *)arg, ",");
427 /* get the output mode */
428 if (!strncmp(tok, "graph", strlen(arg)))
429 callchain_param.mode = CHAIN_GRAPH_ABS;
431 else if (!strncmp(tok, "flat", strlen(arg)))
432 callchain_param.mode = CHAIN_FLAT;
434 else if (!strncmp(tok, "fractal", strlen(arg)))
435 callchain_param.mode = CHAIN_GRAPH_REL;
437 else if (!strncmp(tok, "none", strlen(arg))) {
438 callchain_param.mode = CHAIN_NONE;
439 symbol_conf.use_callchain = false;
447 /* get the min percentage */
448 tok = strtok(NULL, ",");
452 callchain_param.min_percent = strtod(tok, &endptr);
456 /* get the print limit */
457 tok2 = strtok(NULL, ",");
461 if (tok2[0] != 'c') {
462 callchain_param.print_limit = strtoul(tok2, &endptr, 0);
463 tok2 = strtok(NULL, ",");
468 /* get the call chain order */
469 if (!strcmp(tok2, "caller"))
470 callchain_param.order = ORDER_CALLER;
471 else if (!strcmp(tok2, "callee"))
472 callchain_param.order = ORDER_CALLEE;
476 if (callchain_register_param(&callchain_param) < 0) {
477 fprintf(stderr, "Can't register callchain params\n");
484 parse_branch_mode(const struct option *opt __used, const char *str __used, int unset)
486 sort__branch_mode = !unset;
490 int cmd_report(int argc, const char **argv, const char *prefix __used)
492 struct perf_session *session;
494 bool has_br_stack = false;
496 char callchain_default_opt[] = "fractal,0.5,callee";
497 const char * const report_usage[] = {
498 "perf report [<options>]",
501 struct perf_report report = {
503 .sample = process_sample_event,
504 .mmap = perf_event__process_mmap,
505 .comm = perf_event__process_comm,
506 .exit = perf_event__process_task,
507 .fork = perf_event__process_task,
508 .lost = perf_event__process_lost,
509 .read = process_read_event,
510 .attr = perf_event__process_attr,
511 .event_type = perf_event__process_event_type,
512 .tracing_data = perf_event__process_tracing_data,
513 .build_id = perf_event__process_build_id,
514 .ordered_samples = true,
515 .ordering_requires_timestamps = true,
517 .pretty_printing_style = "normal",
519 const struct option options[] = {
520 OPT_STRING('i', "input", &report.input_name, "file",
522 OPT_INCR('v', "verbose", &verbose,
523 "be more verbose (show symbol address, etc)"),
524 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
525 "dump raw trace in ASCII"),
526 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
527 "file", "vmlinux pathname"),
528 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
529 "file", "kallsyms pathname"),
530 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
531 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
532 "load module symbols - WARNING: use only with -k and LIVE kernel"),
533 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
534 "Show a column with the number of samples"),
535 OPT_BOOLEAN('T', "threads", &report.show_threads,
536 "Show per-thread event counters"),
537 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
538 "pretty printing style key: normal raw"),
539 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
540 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
541 "Use the stdio interface"),
542 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
543 "sort by key(s): pid, comm, dso, symbol, parent, dso_to,"
544 " dso_from, symbol_to, symbol_from, mispredict"),
545 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
546 "Show sample percentage for different cpu modes"),
547 OPT_STRING('p', "parent", &parent_pattern, "regex",
548 "regex filter to identify parent, see: '--sort parent'"),
549 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
550 "Only display entries with parent-match"),
551 OPT_CALLBACK_DEFAULT('g', "call-graph", &report, "output_type,min_percent[,print_limit],call_order",
552 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold, optional print limit and callchain order. "
553 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
554 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
555 "alias for inverted call graph"),
556 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
557 "only consider symbols in these dsos"),
558 OPT_STRING('c', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
559 "only consider symbols in these comms"),
560 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
561 "only consider these symbols"),
562 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
564 "don't try to adjust column width, use these fixed values"),
565 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
566 "separator for columns, no spaces will be added between "
567 "columns '.' is reserved."),
568 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
569 "Only display entries resolved to a symbol"),
570 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
571 "Look for files with symbols relative to this directory"),
572 OPT_STRING('C', "cpu", &report.cpu_list, "cpu",
573 "list of cpus to profile"),
574 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
575 "Display extended information about perf.data file"),
576 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
577 "Interleave source code with assembly code (default)"),
578 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
579 "Display raw encoding of assembly instructions (default)"),
580 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
581 "Specify disassembler style (e.g. -M intel for intel syntax)"),
582 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
583 "Show a column with the sum of periods"),
584 OPT_CALLBACK_NOOPT('b', "branch-stack", &sort__branch_mode, "",
585 "use branch records for histogram filling", parse_branch_mode),
589 argc = parse_options(argc, argv, options, report_usage, 0);
591 if (report.use_stdio)
593 else if (report.use_tui)
596 if (report.inverted_callchain)
597 callchain_param.order = ORDER_CALLER;
599 if (!report.input_name || !strlen(report.input_name)) {
600 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
601 report.input_name = "-";
603 report.input_name = "perf.data";
605 session = perf_session__new(report.input_name, O_RDONLY,
606 report.force, false, &report.tool);
610 report.session = session;
612 has_br_stack = perf_header__has_feat(&session->header,
613 HEADER_BRANCH_STACK);
615 if (sort__branch_mode == -1 && has_br_stack)
616 sort__branch_mode = 1;
618 if (sort__branch_mode == 1) {
620 fprintf(stderr, "Warning: TUI interface not supported"
621 " in branch mode\n");
622 if (symbol_conf.dso_list_str != NULL)
623 fprintf(stderr, "Warning: dso filtering not supported"
624 " in branch mode\n");
625 if (symbol_conf.sym_list_str != NULL)
626 fprintf(stderr, "Warning: symbol filtering not"
627 " supported in branch mode\n");
629 report.use_stdio = true;
632 symbol_conf.dso_list_str = NULL;
633 symbol_conf.sym_list_str = NULL;
636 * if no sort_order is provided, then specify branch-mode
639 if (sort_order == default_sort_order)
640 sort_order = "comm,dso_from,symbol_from,"
643 } else if (strcmp(report.input_name, "-") != 0) {
650 * Only in the newt browser we are doing integrated annotation,
651 * so don't allocate extra space that won't be used in the stdio
654 if (use_browser > 0) {
655 symbol_conf.priv_size = sizeof(struct annotation);
656 report.annotate_init = symbol__annotate_init;
658 * For searching by name on the "Browse map details".
659 * providing it only in verbose mode not to bloat too
660 * much struct symbol.
664 * XXX: Need to provide a less kludgy way to ask for
665 * more space per symbol, the u32 is for the index on
667 * See symbol__browser_index.
669 symbol_conf.priv_size += sizeof(u32);
670 symbol_conf.sort_by_name = true;
674 if (symbol__init() < 0)
677 setup_sorting(report_usage, options);
679 if (parent_pattern != default_parent_pattern) {
680 if (sort_dimension__add("parent") < 0)
684 * Only show the parent fields if we explicitly
685 * sort that way. If we only use parent machinery
686 * for filtering, we don't want it.
688 if (!strstr(sort_order, "parent"))
689 sort_parent.elide = 1;
691 symbol_conf.exclude_other = false;
694 * Any (unrecognized) arguments left?
697 usage_with_options(report_usage, options);
699 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
700 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
701 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
703 ret = __cmd_report(&report);
705 perf_session__delete(session);