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 static char const *input_name = "perf.data";
38 static bool force, use_tui, use_stdio;
39 static bool hide_unresolved;
40 static bool dont_use_callchains;
42 static bool show_threads;
43 static struct perf_read_values show_threads_values;
45 static const char default_pretty_printing_style[] = "normal";
46 static const char *pretty_printing_style = default_pretty_printing_style;
48 static char callchain_default_opt[] = "fractal,0.5";
49 static symbol_filter_t annotate_init;
51 static int perf_session__add_hist_entry(struct perf_session *session,
52 struct addr_location *al,
53 struct perf_sample *sample,
54 struct perf_evsel *evsel)
56 struct symbol *parent = NULL;
58 struct hist_entry *he;
60 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
61 err = perf_session__resolve_callchain(session, al->thread,
62 sample->callchain, &parent);
67 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
71 if (symbol_conf.use_callchain) {
72 err = callchain_append(he->callchain, &session->callchain_cursor,
78 * Only in the newt browser we are doing integrated annotation,
79 * so we don't allocated the extra space needed because the stdio
80 * code will not use it.
82 if (al->sym != NULL && use_browser > 0) {
83 struct annotation *notes = symbol__annotation(he->ms.sym);
85 assert(evsel != NULL);
88 if (notes->src == NULL &&
89 symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0)
92 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
95 evsel->hists.stats.total_period += sample->period;
96 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
102 static int process_sample_event(union perf_event *event,
103 struct perf_sample *sample,
104 struct perf_evsel *evsel,
105 struct perf_session *session)
107 struct addr_location al;
109 if (perf_event__preprocess_sample(event, session, &al, sample,
110 annotate_init) < 0) {
111 fprintf(stderr, "problem processing %d event, skipping it.\n",
116 if (al.filtered || (hide_unresolved && al.sym == NULL))
120 al.map->dso->hit = 1;
122 if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
123 pr_debug("problem incrementing symbol period, skipping event\n");
130 static int process_read_event(union perf_event *event,
131 struct perf_sample *sample __used,
132 struct perf_session *session)
134 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
137 const char *name = evsel ? event_name(evsel) : "unknown";
138 perf_read_values_add_value(&show_threads_values,
139 event->read.pid, event->read.tid,
145 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
146 evsel ? event_name(evsel) : "FAIL",
152 static int perf_session__setup_sample_type(struct perf_session *self)
154 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
155 if (sort__has_parent) {
156 fprintf(stderr, "selected --sort parent, but no"
157 " callchain data. Did you call"
158 " perf record without -g?\n");
161 if (symbol_conf.use_callchain) {
162 fprintf(stderr, "selected -g but no callchain data."
163 " Did you call perf record without"
167 } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
168 !symbol_conf.use_callchain) {
169 symbol_conf.use_callchain = true;
170 if (callchain_register_param(&callchain_param) < 0) {
171 fprintf(stderr, "Can't register callchain"
180 static struct perf_event_ops event_ops = {
181 .sample = process_sample_event,
182 .mmap = perf_event__process_mmap,
183 .comm = perf_event__process_comm,
184 .exit = perf_event__process_task,
185 .fork = perf_event__process_task,
186 .lost = perf_event__process_lost,
187 .read = process_read_event,
188 .attr = perf_event__process_attr,
189 .event_type = perf_event__process_event_type,
190 .tracing_data = perf_event__process_tracing_data,
191 .build_id = perf_event__process_build_id,
192 .ordered_samples = true,
193 .ordering_requires_timestamps = true,
196 extern volatile int session_done;
198 static void sig_handler(int sig __used)
203 static size_t hists__fprintf_nr_sample_events(struct hists *self,
204 const char *evname, FILE *fp)
208 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
210 nr_events = convert_unit(nr_events, &unit);
211 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
213 ret += fprintf(fp, " %s", evname);
214 return ret + fprintf(fp, "\n#\n");
217 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
220 struct perf_evsel *pos;
222 list_for_each_entry(pos, &evlist->entries, node) {
223 struct hists *hists = &pos->hists;
224 const char *evname = NULL;
226 if (rb_first(&hists->entries) != rb_last(&hists->entries))
227 evname = event_name(pos);
229 hists__fprintf_nr_sample_events(hists, evname, stdout);
230 hists__fprintf(hists, NULL, false, stdout);
231 fprintf(stdout, "\n\n");
234 if (sort_order == default_sort_order &&
235 parent_pattern == default_parent_pattern) {
236 fprintf(stdout, "#\n# (%s)\n#\n", help);
239 bool style = !strcmp(pretty_printing_style, "raw");
240 perf_read_values_display(stdout, &show_threads_values,
242 perf_read_values_destroy(&show_threads_values);
249 static int __cmd_report(void)
253 struct perf_session *session;
254 struct perf_evsel *pos;
255 struct map *kernel_map;
256 struct kmap *kernel_kmap;
257 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
259 signal(SIGINT, sig_handler);
261 session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops);
266 perf_read_values_init(&show_threads_values);
268 ret = perf_session__setup_sample_type(session);
272 ret = perf_session__process_events(session, &event_ops);
276 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
277 kernel_kmap = map__kmap(kernel_map);
278 if (kernel_map == NULL ||
279 (kernel_map->dso->hit &&
280 (kernel_kmap->ref_reloc_sym == NULL ||
281 kernel_kmap->ref_reloc_sym->addr == 0))) {
282 const struct dso *kdso = kernel_map->dso;
284 ui__warning("Kernel address maps "
285 "(/proc/{kallsyms,modules}) were restricted, "
286 "check /proc/sys/kernel/kptr_restrict before "
287 "running 'perf record'.\n\n%s\n\n"
288 "Samples in kernel modules can't be resolved "
290 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
291 "As no suitable kallsyms nor vmlinux was found, "
292 "kernel samples can't be resolved." :
293 "If some relocation was applied (e.g. kexec) "
294 "symbols may be misresolved.");
298 perf_session__fprintf_nr_events(session, stdout);
303 perf_session__fprintf(session, stdout);
306 perf_session__fprintf_dsos(session, stdout);
309 list_for_each_entry(pos, &session->evlist->entries, node) {
310 struct hists *hists = &pos->hists;
312 hists__collapse_resort(hists);
313 hists__output_resort(hists);
314 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
317 if (nr_samples == 0) {
318 ui__warning("The %s file has no samples!\n", input_name);
323 perf_evlist__tui_browse_hists(session->evlist, help);
325 perf_evlist__tty_browse_hists(session->evlist, help);
329 * Speed up the exit process, for large files this can
330 * take quite a while.
332 * XXX Enable this when using valgrind or if we ever
333 * librarize this command.
335 * Also experiment with obstacks to see how much speed
338 * perf_session__delete(session);
344 parse_callchain_opt(const struct option *opt __used, const char *arg,
354 dont_use_callchains = true;
358 symbol_conf.use_callchain = true;
363 tok = strtok((char *)arg, ",");
367 /* get the output mode */
368 if (!strncmp(tok, "graph", strlen(arg)))
369 callchain_param.mode = CHAIN_GRAPH_ABS;
371 else if (!strncmp(tok, "flat", strlen(arg)))
372 callchain_param.mode = CHAIN_FLAT;
374 else if (!strncmp(tok, "fractal", strlen(arg)))
375 callchain_param.mode = CHAIN_GRAPH_REL;
377 else if (!strncmp(tok, "none", strlen(arg))) {
378 callchain_param.mode = CHAIN_NONE;
379 symbol_conf.use_callchain = false;
387 /* get the min percentage */
388 tok = strtok(NULL, ",");
392 tok2 = strtok(NULL, ",");
393 callchain_param.min_percent = strtod(tok, &endptr);
398 callchain_param.print_limit = strtod(tok2, &endptr);
400 if (callchain_register_param(&callchain_param) < 0) {
401 fprintf(stderr, "Can't register callchain params\n");
407 static const char * const report_usage[] = {
408 "perf report [<options>] <command>",
412 static const struct option options[] = {
413 OPT_STRING('i', "input", &input_name, "file",
415 OPT_INCR('v', "verbose", &verbose,
416 "be more verbose (show symbol address, etc)"),
417 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
418 "dump raw trace in ASCII"),
419 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
420 "file", "vmlinux pathname"),
421 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
422 "file", "kallsyms pathname"),
423 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
424 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
425 "load module symbols - WARNING: use only with -k and LIVE kernel"),
426 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
427 "Show a column with the number of samples"),
428 OPT_BOOLEAN('T', "threads", &show_threads,
429 "Show per-thread event counters"),
430 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
431 "pretty printing style key: normal raw"),
432 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
433 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
434 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
435 "sort by key(s): pid, comm, dso, symbol, parent"),
436 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
437 "Show sample percentage for different cpu modes"),
438 OPT_STRING('p', "parent", &parent_pattern, "regex",
439 "regex filter to identify parent, see: '--sort parent'"),
440 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
441 "Only display entries with parent-match"),
442 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
443 "Display callchains using output_type (graph, flat, fractal, or none) and min percent threshold. "
444 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
445 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
446 "only consider symbols in these dsos"),
447 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
448 "only consider symbols in these comms"),
449 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
450 "only consider these symbols"),
451 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
453 "don't try to adjust column width, use these fixed values"),
454 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
455 "separator for columns, no spaces will be added between "
456 "columns '.' is reserved."),
457 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
458 "Only display entries resolved to a symbol"),
459 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
460 "Look for files with symbols relative to this directory"),
464 int cmd_report(int argc, const char **argv, const char *prefix __used)
466 argc = parse_options(argc, argv, options, report_usage, 0);
473 if (strcmp(input_name, "-") != 0)
478 * Only in the newt browser we are doing integrated annotation,
479 * so don't allocate extra space that won't be used in the stdio
482 if (use_browser > 0) {
483 symbol_conf.priv_size = sizeof(struct annotation);
484 annotate_init = symbol__annotate_init;
486 * For searching by name on the "Browse map details".
487 * providing it only in verbose mode not to bloat too
488 * much struct symbol.
492 * XXX: Need to provide a less kludgy way to ask for
493 * more space per symbol, the u32 is for the index on
495 * See symbol__browser_index.
497 symbol_conf.priv_size += sizeof(u32);
498 symbol_conf.sort_by_name = true;
502 if (symbol__init() < 0)
505 setup_sorting(report_usage, options);
507 if (parent_pattern != default_parent_pattern) {
508 if (sort_dimension__add("parent") < 0)
510 sort_parent.elide = 1;
512 symbol_conf.exclude_other = false;
515 * Any (unrecognized) arguments left?
518 usage_with_options(report_usage, options);
520 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
521 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
522 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
524 return __cmd_report();