1 /*
2 * builtin-report.c
3 *
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"
9
10 #include "util/util.h"
11
12 #include "util/annotate.h"
13 #include "util/color.h"
14 /* ANDROID_CHANGE_BEGIN */
15 #if 0
16 #include <linux/list.h>
17 #include "util/cache.h"
18 #include <linux/rbtree.h>
19 #else
20 #include "util/include/linux/list.h"
21 #include "util/cache.h"
22 #include "util/include/linux/rbtree.h"
23 #endif
24 /* ANDROID_CHANGE_END */
25 #include "util/symbol.h"
26 #include "util/callchain.h"
27 #include "util/strlist.h"
28 #include "util/values.h"
29
30 #include "perf.h"
31 #include "util/debug.h"
32 #include "util/evlist.h"
33 #include "util/evsel.h"
34 #include "util/header.h"
35 #include "util/session.h"
36
37 #include "util/parse-options.h"
38 #include "util/parse-events.h"
39
40 #include "util/thread.h"
41 #include "util/sort.h"
42 #include "util/hist.h"
43
44 static char const *input_name = "perf.data";
45
46 static bool force, use_tui, use_stdio;
47 static bool hide_unresolved;
48 static bool dont_use_callchains;
49
50 static bool show_threads;
51 static struct perf_read_values show_threads_values;
52
53 static const char default_pretty_printing_style[] = "normal";
54 static const char *pretty_printing_style = default_pretty_printing_style;
55
56 static char callchain_default_opt[] = "fractal,0.5";
57 static symbol_filter_t annotate_init;
58
perf_session__add_hist_entry(struct perf_session * session,struct addr_location * al,struct perf_sample * sample,struct perf_evsel * evsel)59 static int perf_session__add_hist_entry(struct perf_session *session,
60 struct addr_location *al,
61 struct perf_sample *sample,
62 struct perf_evsel *evsel)
63 {
64 struct symbol *parent = NULL;
65 int err = 0;
66 struct hist_entry *he;
67
68 if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
69 err = perf_session__resolve_callchain(session, al->thread,
70 sample->callchain, &parent);
71 if (err)
72 return err;
73 }
74
75 he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
76 if (he == NULL)
77 return -ENOMEM;
78
79 if (symbol_conf.use_callchain) {
80 err = callchain_append(he->callchain, &session->callchain_cursor,
81 sample->period);
82 if (err)
83 return err;
84 }
85 /*
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.
89 */
90 if (al->sym != NULL && use_browser > 0) {
91 struct annotation *notes = symbol__annotation(he->ms.sym);
92
93 assert(evsel != NULL);
94
95 err = -ENOMEM;
96 if (notes->src == NULL &&
97 symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0)
98 goto out;
99
100 err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
101 }
102
103 evsel->hists.stats.total_period += sample->period;
104 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
105 out:
106 return err;
107 }
108
109
process_sample_event(union perf_event * event,struct perf_sample * sample,struct perf_evsel * evsel,struct perf_session * session)110 static int process_sample_event(union perf_event *event,
111 struct perf_sample *sample,
112 struct perf_evsel *evsel,
113 struct perf_session *session)
114 {
115 struct addr_location al;
116
117 if (perf_event__preprocess_sample(event, session, &al, sample,
118 annotate_init) < 0) {
119 fprintf(stderr, "problem processing %d event, skipping it.\n",
120 event->header.type);
121 return -1;
122 }
123
124 if (al.filtered || (hide_unresolved && al.sym == NULL))
125 return 0;
126
127 if (al.map != NULL)
128 al.map->dso->hit = 1;
129
130 if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
131 pr_debug("problem incrementing symbol period, skipping event\n");
132 return -1;
133 }
134
135 return 0;
136 }
137
process_read_event(union perf_event * event,struct perf_sample * sample __used,struct perf_session * session)138 static int process_read_event(union perf_event *event,
139 struct perf_sample *sample __used,
140 struct perf_session *session)
141 {
142 struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
143 event->read.id);
144 if (show_threads) {
145 const char *name = evsel ? event_name(evsel) : "unknown";
146 perf_read_values_add_value(&show_threads_values,
147 event->read.pid, event->read.tid,
148 event->read.id,
149 name,
150 event->read.value);
151 }
152
153 dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
154 evsel ? event_name(evsel) : "FAIL",
155 event->read.value);
156
157 return 0;
158 }
159
perf_session__setup_sample_type(struct perf_session * self)160 static int perf_session__setup_sample_type(struct perf_session *self)
161 {
162 if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
163 if (sort__has_parent) {
164 fprintf(stderr, "selected --sort parent, but no"
165 " callchain data. Did you call"
166 " perf record without -g?\n");
167 return -EINVAL;
168 }
169 if (symbol_conf.use_callchain) {
170 fprintf(stderr, "selected -g but no callchain data."
171 " Did you call perf record without"
172 " -g?\n");
173 return -1;
174 }
175 } else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
176 !symbol_conf.use_callchain) {
177 symbol_conf.use_callchain = true;
178 if (callchain_register_param(&callchain_param) < 0) {
179 fprintf(stderr, "Can't register callchain"
180 " params\n");
181 return -EINVAL;
182 }
183 }
184
185 return 0;
186 }
187
188 static struct perf_event_ops event_ops = {
189 .sample = process_sample_event,
190 .mmap = perf_event__process_mmap,
191 .comm = perf_event__process_comm,
192 .exit = perf_event__process_task,
193 .fork = perf_event__process_task,
194 .lost = perf_event__process_lost,
195 .read = process_read_event,
196 .attr = perf_event__process_attr,
197 .event_type = perf_event__process_event_type,
198 .tracing_data = perf_event__process_tracing_data,
199 .build_id = perf_event__process_build_id,
200 .ordered_samples = true,
201 .ordering_requires_timestamps = true,
202 };
203
204 extern volatile int session_done;
205
sig_handler(int sig __used)206 static void sig_handler(int sig __used)
207 {
208 session_done = 1;
209 }
210
hists__fprintf_nr_sample_events(struct hists * self,const char * evname,FILE * fp)211 static size_t hists__fprintf_nr_sample_events(struct hists *self,
212 const char *evname, FILE *fp)
213 {
214 size_t ret;
215 char unit;
216 unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];
217
218 nr_events = convert_unit(nr_events, &unit);
219 ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
220 if (evname != NULL)
221 ret += fprintf(fp, " %s", evname);
222 return ret + fprintf(fp, "\n#\n");
223 }
224
perf_evlist__tty_browse_hists(struct perf_evlist * evlist,const char * help)225 static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
226 const char *help)
227 {
228 struct perf_evsel *pos;
229
230 list_for_each_entry(pos, &evlist->entries, node) {
231 struct hists *hists = &pos->hists;
232 const char *evname = NULL;
233
234 if (rb_first(&hists->entries) != rb_last(&hists->entries))
235 evname = event_name(pos);
236
237 hists__fprintf_nr_sample_events(hists, evname, stdout);
238 hists__fprintf(hists, NULL, false, stdout);
239 fprintf(stdout, "\n\n");
240 }
241
242 if (sort_order == default_sort_order &&
243 parent_pattern == default_parent_pattern) {
244 fprintf(stdout, "#\n# (%s)\n#\n", help);
245
246 if (show_threads) {
247 bool style = !strcmp(pretty_printing_style, "raw");
248 perf_read_values_display(stdout, &show_threads_values,
249 style);
250 perf_read_values_destroy(&show_threads_values);
251 }
252 }
253
254 return 0;
255 }
256
__cmd_report(void)257 static int __cmd_report(void)
258 {
259 int ret = -EINVAL;
260 u64 nr_samples;
261 struct perf_session *session;
262 struct perf_evsel *pos;
263 struct map *kernel_map;
264 struct kmap *kernel_kmap;
265 const char *help = "For a higher level overview, try: perf report --sort comm,dso";
266
267 signal(SIGINT, sig_handler);
268
269 session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops);
270 if (session == NULL)
271 return -ENOMEM;
272
273 if (show_threads)
274 perf_read_values_init(&show_threads_values);
275
276 ret = perf_session__setup_sample_type(session);
277 if (ret)
278 goto out_delete;
279
280 ret = perf_session__process_events(session, &event_ops);
281 if (ret)
282 goto out_delete;
283
284 kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
285 kernel_kmap = map__kmap(kernel_map);
286 if (kernel_map == NULL ||
287 (kernel_map->dso->hit &&
288 (kernel_kmap->ref_reloc_sym == NULL ||
289 kernel_kmap->ref_reloc_sym->addr == 0))) {
290 const struct dso *kdso = kernel_map->dso;
291
292 ui__warning(
293 "Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
294 "Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
295 "Samples in kernel modules can't be resolved as well.\n\n",
296 RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
297 "As no suitable kallsyms nor vmlinux was found, kernel samples\n"
298 "can't be resolved." :
299 "If some relocation was applied (e.g. kexec) symbols may be misresolved.");
300 }
301
302 if (dump_trace) {
303 perf_session__fprintf_nr_events(session, stdout);
304 goto out_delete;
305 }
306
307 if (verbose > 3)
308 perf_session__fprintf(session, stdout);
309
310 if (verbose > 2)
311 perf_session__fprintf_dsos(session, stdout);
312
313 nr_samples = 0;
314 list_for_each_entry(pos, &session->evlist->entries, node) {
315 struct hists *hists = &pos->hists;
316
317 hists__collapse_resort(hists);
318 hists__output_resort(hists);
319 nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
320 }
321
322 if (nr_samples == 0) {
323 ui__warning("The %s file has no samples!\n", input_name);
324 goto out_delete;
325 }
326
327 if (use_browser > 0)
328 perf_evlist__tui_browse_hists(session->evlist, help);
329 else
330 perf_evlist__tty_browse_hists(session->evlist, help);
331
332 out_delete:
333 /*
334 * Speed up the exit process, for large files this can
335 * take quite a while.
336 *
337 * XXX Enable this when using valgrind or if we ever
338 * librarize this command.
339 *
340 * Also experiment with obstacks to see how much speed
341 * up we'll get here.
342 *
343 * perf_session__delete(session);
344 */
345 return ret;
346 }
347
348 static int
parse_callchain_opt(const struct option * opt __used,const char * arg,int unset)349 parse_callchain_opt(const struct option *opt __used, const char *arg,
350 int unset)
351 {
352 char *tok, *tok2;
353 char *endptr;
354
355 /*
356 * --no-call-graph
357 */
358 if (unset) {
359 dont_use_callchains = true;
360 return 0;
361 }
362
363 symbol_conf.use_callchain = true;
364
365 if (!arg)
366 return 0;
367
368 tok = strtok((char *)arg, ",");
369 if (!tok)
370 return -1;
371
372 /* get the output mode */
373 if (!strncmp(tok, "graph", strlen(arg)))
374 callchain_param.mode = CHAIN_GRAPH_ABS;
375
376 else if (!strncmp(tok, "flat", strlen(arg)))
377 callchain_param.mode = CHAIN_FLAT;
378
379 else if (!strncmp(tok, "fractal", strlen(arg)))
380 callchain_param.mode = CHAIN_GRAPH_REL;
381
382 else if (!strncmp(tok, "none", strlen(arg))) {
383 callchain_param.mode = CHAIN_NONE;
384 symbol_conf.use_callchain = false;
385
386 return 0;
387 }
388
389 else
390 return -1;
391
392 /* get the min percentage */
393 tok = strtok(NULL, ",");
394 if (!tok)
395 goto setup;
396
397 tok2 = strtok(NULL, ",");
398 callchain_param.min_percent = strtod(tok, &endptr);
399 if (tok == endptr)
400 return -1;
401
402 if (tok2)
403 callchain_param.print_limit = strtod(tok2, &endptr);
404 setup:
405 if (callchain_register_param(&callchain_param) < 0) {
406 fprintf(stderr, "Can't register callchain params\n");
407 return -1;
408 }
409 return 0;
410 }
411
412 static const char * const report_usage[] = {
413 "perf report [<options>] <command>",
414 NULL
415 };
416
417 static const struct option options[] = {
418 OPT_STRING('i', "input", &input_name, "file",
419 "input file name"),
420 OPT_INCR('v', "verbose", &verbose,
421 "be more verbose (show symbol address, etc)"),
422 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
423 "dump raw trace in ASCII"),
424 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
425 "file", "vmlinux pathname"),
426 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
427 "file", "kallsyms pathname"),
428 OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
429 OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
430 "load module symbols - WARNING: use only with -k and LIVE kernel"),
431 OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
432 "Show a column with the number of samples"),
433 OPT_BOOLEAN('T', "threads", &show_threads,
434 "Show per-thread event counters"),
435 OPT_STRING(0, "pretty", &pretty_printing_style, "key",
436 "pretty printing style key: normal raw"),
437 OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
438 OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
439 OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
440 "sort by key(s): pid, comm, dso, symbol, parent"),
441 OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
442 "Show sample percentage for different cpu modes"),
443 OPT_STRING('p', "parent", &parent_pattern, "regex",
444 "regex filter to identify parent, see: '--sort parent'"),
445 OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
446 "Only display entries with parent-match"),
447 OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent",
448 "Display callchains using output_type (graph, flat, fractal, or none) and min percent threshold. "
449 "Default: fractal,0.5", &parse_callchain_opt, callchain_default_opt),
450 OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
451 "only consider symbols in these dsos"),
452 OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
453 "only consider symbols in these comms"),
454 OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
455 "only consider these symbols"),
456 OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
457 "width[,width...]",
458 "don't try to adjust column width, use these fixed values"),
459 OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
460 "separator for columns, no spaces will be added between "
461 "columns '.' is reserved."),
462 OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
463 "Only display entries resolved to a symbol"),
464 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
465 "Look for files with symbols relative to this directory"),
466 OPT_END()
467 };
468
cmd_report(int argc,const char ** argv,const char * prefix __used)469 int cmd_report(int argc, const char **argv, const char *prefix __used)
470 {
471 argc = parse_options(argc, argv, options, report_usage, 0);
472
473 if (use_stdio)
474 use_browser = 0;
475 else if (use_tui)
476 use_browser = 1;
477
478 if (strcmp(input_name, "-") != 0)
479 setup_browser(true);
480 else
481 use_browser = 0;
482 /*
483 * Only in the newt browser we are doing integrated annotation,
484 * so don't allocate extra space that won't be used in the stdio
485 * implementation.
486 */
487 if (use_browser > 0) {
488 symbol_conf.priv_size = sizeof(struct annotation);
489 annotate_init = symbol__annotate_init;
490 /*
491 * For searching by name on the "Browse map details".
492 * providing it only in verbose mode not to bloat too
493 * much struct symbol.
494 */
495 if (verbose) {
496 /*
497 * XXX: Need to provide a less kludgy way to ask for
498 * more space per symbol, the u32 is for the index on
499 * the ui browser.
500 * See symbol__browser_index.
501 */
502 symbol_conf.priv_size += sizeof(u32);
503 symbol_conf.sort_by_name = true;
504 }
505 }
506
507 if (symbol__init() < 0)
508 return -1;
509
510 setup_sorting(report_usage, options);
511
512 if (parent_pattern != default_parent_pattern) {
513 if (sort_dimension__add("parent") < 0)
514 return -1;
515 sort_parent.elide = 1;
516 } else
517 symbol_conf.exclude_other = false;
518
519 /*
520 * Any (unrecognized) arguments left?
521 */
522 if (argc)
523 usage_with_options(report_usage, options);
524
525 sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
526 sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
527 sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
528
529 return __cmd_report();
530 }
531