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"
29 #include "util/parse-options.h"
30 #include "util/parse-events.h"
32 #include "util/thread.h"
33 #include "util/sort.h"
34 #include "util/hist.h"
36 #include <linux/bitmap.h>
38 static struct perf_report {
39 char const *input_name;
40 bool force, use_tui, use_stdio;
42 bool dont_use_callchains;
45 bool inverted_callchain;
46 struct perf_read_values show_threads_values;
47 const char *pretty_printing_style;
48 symbol_filter_t annotate_init;
50 DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
52 .input_name = "perf.data",
53 .pretty_printing_style = "normal",
56 static char callchain_default_opt[] = "fractal,0.5,callee";
58 static int perf_session__add_hist_entry(struct perf_session *session,
59 struct addr_location *al,
60 struct perf_sample *sample,
61 struct perf_evsel *evsel)
63 struct symbol *parent = NULL;
65 struct hist_entry *he;
67 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
68 err = perf_session__resolve_callchain(session, evsel, al->thread,
69 sample->callchain, &parent);
74 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
78 if (symbol_conf.use_callchain) {
79 err = callchain_append(he->callchain,
80 &evsel->hists.callchain_cursor,
86 * Only in the newt browser we are doing integrated annotation,
87 * so we don't allocated the extra space needed because the stdio
88 * code will not use it.
90 if (al->sym != NULL && use_browser > 0) {
91 struct annotation *notes = symbol__annotation(he->ms.sym);
93 assert(evsel != NULL);
96 if (notes->src == NULL && symbol__alloc_hist(he->ms.sym) < 0)
99 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
102 evsel->hists.stats.total_period += sample->period;
103 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
109 static int process_sample_event(union perf_event *event,
110 struct perf_sample *sample,
111 struct perf_evsel *evsel,
112 struct perf_session *session)
114 struct addr_location al;
116 if (perf_event__preprocess_sample(event, session, &al, sample,
117 rep->annotate_init) < 0) {
118 fprintf(stderr, "problem processing %d event, skipping it.\n",
123 if (al.filtered || (rep->hide_unresolved && al.sym == NULL))
126 if (rep->cpu_list && !test_bit(sample->cpu, rep->cpu_bitmap))
130 al.map->dso->hit = 1;
132 if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
133 pr_debug("problem incrementing symbol period, skipping event\n");
140 static int process_read_event(union perf_event *event,
141 struct perf_sample *sample __used,
142 struct perf_session *session)
144 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
146 if (rep->show_threads) {
147 const char *name = evsel ? event_name(evsel) : "unknown";
148 perf_read_values_add_value(&rep->show_threads_values,
149 event->read.pid, event->read.tid,
155 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
156 evsel ? event_name(evsel) : "FAIL",
162 static int perf_session__setup_sample_type(struct perf_session *self)
164 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
165 if (sort__has_parent) {
166 ui__warning("Selected --sort parent, but no "
167 "callchain data. Did you call "
168 "'perf record' without -g?\n");
171 if (symbol_conf.use_callchain) {
172 ui__warning("Selected -g but no callchain data. Did "
173 "you call 'perf record' without -g?\n");
176 } else if (!rep->dont_use_callchains &&
177 callchain_param.mode != CHAIN_NONE &&
178 !symbol_conf.use_callchain) {
179 symbol_conf.use_callchain = true;
180 if (callchain_register_param(&callchain_param) < 0) {
181 ui__warning("Can't register callchain "
190 static struct perf_event_ops event_ops = {
191 .sample = process_sample_event,
192 .mmap = perf_event__process_mmap,
193 .comm = perf_event__process_comm,
194 .exit = perf_event__process_task,
195 .fork = perf_event__process_task,
196 .lost = perf_event__process_lost,
197 .read = process_read_event,
198 .attr = perf_event__process_attr,
199 .event_type = perf_event__process_event_type,
200 .tracing_data = perf_event__process_tracing_data,
201 .build_id = perf_event__process_build_id,
202 .ordered_samples = true,
203 .ordering_requires_timestamps = true,
206 extern volatile int session_done;
208 static void sig_handler(int sig __used)
213 static size_t hists__fprintf_nr_sample_events(struct hists *self,
214 const char *evname, FILE *fp)
218 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
220 nr_events = convert_unit(nr_events, &unit);
221 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
223 ret += fprintf(fp, " %s", evname);
224 return ret + fprintf(fp, "\n#\n");
227 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
230 struct perf_evsel *pos;
232 list_for_each_entry(pos, &evlist->entries, node) {
233 struct hists *hists = &pos->hists;
234 const char *evname = event_name(pos);
236 hists__fprintf_nr_sample_events(hists, evname, stdout);
237 hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
238 fprintf(stdout, "\n\n");
241 if (sort_order == default_sort_order &&
242 parent_pattern == default_parent_pattern) {
243 fprintf(stdout, "#\n# (%s)\n#\n", help);
245 if (rep->show_threads) {
246 bool style = !strcmp(rep->pretty_printing_style, "raw");
247 perf_read_values_display(stdout, &rep->show_threads_values,
249 perf_read_values_destroy(&rep->show_threads_values);
256 static int __cmd_report(void)
260 struct perf_session *session;
261 struct perf_evsel *pos;
262 struct map *kernel_map;
263 struct kmap *kernel_kmap;
264 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
266 signal(SIGINT, sig_handler);
268 session = perf_session__new(rep->input_name, O_RDONLY,
269 rep->force, false, &event_ops);
274 ret = perf_session__cpu_bitmap(session, rep->cpu_list,
280 if (use_browser <= 0)
281 perf_session__fprintf_info(session, stdout, rep->show_full_info);
283 if (rep->show_threads)
284 perf_read_values_init(&rep->show_threads_values);
286 ret = perf_session__setup_sample_type(session);
290 ret = perf_session__process_events(session, &event_ops);
294 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
295 kernel_kmap = map__kmap(kernel_map);
296 if (kernel_map == NULL ||
297 (kernel_map->dso->hit &&
298 (kernel_kmap->ref_reloc_sym == NULL ||
299 kernel_kmap->ref_reloc_sym->addr == 0))) {
300 const struct dso *kdso = kernel_map->dso;
303 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
304 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
305 "Samples in kernel modules can't be resolved as well.\n\n",
306 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
307 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
308 "can't be resolved." :
309 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
313 perf_session__fprintf_nr_events(session, stdout);
318 perf_session__fprintf(session, stdout);
321 perf_session__fprintf_dsos(session, stdout);
324 list_for_each_entry(pos, &session->evlist->entries, node) {
325 struct hists *hists = &pos->hists;
327 hists__collapse_resort(hists);
328 hists__output_resort(hists);
329 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
332 if (nr_samples == 0) {
333 ui__warning("The %s file has no samples!\n",
338 if (use_browser > 0) {
339 perf_evlist__tui_browse_hists(session->evlist, help,
342 perf_evlist__tty_browse_hists(session->evlist, help);
346 * Speed up the exit process, for large files this can
347 * take quite a while.
349 * XXX Enable this when using valgrind or if we ever
350 * librarize this command.
352 * Also experiment with obstacks to see how much speed
355 * perf_session__delete(session);
361 parse_callchain_opt(const struct option *opt __used, const char *arg,
371 rep->dont_use_callchains = true;
375 symbol_conf.use_callchain = true;
380 tok = strtok((char *)arg, ",");
384 /* get the output mode */
385 if (!strncmp(tok, "graph", strlen(arg)))
386 callchain_param.mode = CHAIN_GRAPH_ABS;
388 else if (!strncmp(tok, "flat", strlen(arg)))
389 callchain_param.mode = CHAIN_FLAT;
391 else if (!strncmp(tok, "fractal", strlen(arg)))
392 callchain_param.mode = CHAIN_GRAPH_REL;
394 else if (!strncmp(tok, "none", strlen(arg))) {
395 callchain_param.mode = CHAIN_NONE;
396 symbol_conf.use_callchain = false;
404 /* get the min percentage */
405 tok = strtok(NULL, ",");
409 callchain_param.min_percent = strtod(tok, &endptr);
413 /* get the print limit */
414 tok2 = strtok(NULL, ",");
418 if (tok2[0] != 'c') {
419 callchain_param.print_limit = strtod(tok2, &endptr);
420 tok2 = strtok(NULL, ",");
425 /* get the call chain order */
426 if (!strcmp(tok2, "caller"))
427 callchain_param.order = ORDER_CALLER;
428 else if (!strcmp(tok2, "callee"))
429 callchain_param.order = ORDER_CALLEE;
433 if (callchain_register_param(&callchain_param) < 0) {
434 fprintf(stderr, "Can't register callchain params\n");
440 static const char * const report_usage[] = {
441 "perf report [<options>] <command>",
445 static const struct option options[] = {
446 OPT_STRING('i', "input", &report.input_name, "file",
448 OPT_INCR('v', "verbose", &verbose,
449 "be more verbose (show symbol address, etc)"),
450 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
451 "dump raw trace in ASCII"),
452 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
453 "file", "vmlinux pathname"),
454 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
455 "file", "kallsyms pathname"),
456 OPT_BOOLEAN('f', "force", &report.force, "don't complain, do it"),
457 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
458 "load module symbols - WARNING: use only with -k and LIVE kernel"),
459 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
460 "Show a column with the number of samples"),
461 OPT_BOOLEAN('T', "threads", &report.show_threads,
462 "Show per-thread event counters"),
463 OPT_STRING(0, "pretty", &report.pretty_printing_style, "key",
464 "pretty printing style key: normal raw"),
465 OPT_BOOLEAN(0, "tui", &report.use_tui, "Use the TUI interface"),
466 OPT_BOOLEAN(0, "stdio", &report.use_stdio,
467 "Use the stdio interface"),
468 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
469 "sort by key(s): pid, comm, dso, symbol, parent"),
470 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
471 "Show sample percentage for different cpu modes"),
472 OPT_STRING('p', "parent", &parent_pattern, "regex",
473 "regex filter to identify parent, see: '--sort parent'"),
474 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
475 "Only display entries with parent-match"),
476 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent, call_order",
477 "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. "
478 "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
479 OPT_BOOLEAN('G', "inverted", &report.inverted_callchain,
480 "alias for inverted call graph"),
481 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
482 "only consider symbols in these dsos"),
483 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
484 "only consider symbols in these comms"),
485 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
486 "only consider these symbols"),
487 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
489 "don't try to adjust column width, use these fixed values"),
490 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
491 "separator for columns, no spaces will be added between "
492 "columns '.' is reserved."),
493 OPT_BOOLEAN('U', "hide-unresolved", &report.hide_unresolved,
494 "Only display entries resolved to a symbol"),
495 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
496 "Look for files with symbols relative to this directory"),
497 OPT_STRING('c', "cpu", &report.cpu_list, "cpu",
498 "list of cpus to profile"),
499 OPT_BOOLEAN('I', "show-info", &report.show_full_info,
500 "Display extended information about perf.data file"),
501 OPT_BOOLEAN(0, "source", &symbol_conf.annotate_src,
502 "Interleave source code with assembly code (default)"),
503 OPT_BOOLEAN(0, "asm-raw", &symbol_conf.annotate_asm_raw,
504 "Display raw encoding of assembly instructions (default)"),
505 OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
506 "Specify disassembler style (e.g. -M intel for intel syntax)"),
507 OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
508 "Show a column with the sum of periods"),
512 int cmd_report(int argc, const char **argv, const char *prefix __used)
514 argc = parse_options(argc, argv, options, report_usage, 0);
516 if (report.use_stdio)
518 else if (report.use_tui)
521 if (report.inverted_callchain)
522 callchain_param.order = ORDER_CALLER;
524 if (strcmp(report.input_name, "-") != 0)
529 * Only in the newt browser we are doing integrated annotation,
530 * so don't allocate extra space that won't be used in the stdio
533 if (use_browser > 0) {
534 symbol_conf.priv_size = sizeof(struct annotation);
535 report.annotate_init = symbol__annotate_init;
537 * For searching by name on the "Browse map details".
538 * providing it only in verbose mode not to bloat too
539 * much struct symbol.
543 * XXX: Need to provide a less kludgy way to ask for
544 * more space per symbol, the u32 is for the index on
546 * See symbol__browser_index.
548 symbol_conf.priv_size += sizeof(u32);
549 symbol_conf.sort_by_name = true;
553 if (symbol__init() < 0)
556 setup_sorting(report_usage, options);
558 if (parent_pattern != default_parent_pattern) {
559 if (sort_dimension__add("parent") < 0)
563 * Only show the parent fields if we explicitly
564 * sort that way. If we only use parent machinery
565 * for filtering, we don't want it.
567 if (!strstr(sort_order, "parent"))
568 sort_parent.elide = 1;
570 symbol_conf.exclude_other = false;
573 * Any (unrecognized) arguments left?
576 usage_with_options(report_usage, options);
578 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
579 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
580 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
582 return __cmd_report();