builtin-report.c 15.5 KB
Newer Older
1 2 3 4 5 6 7
/*
 * builtin-report.c
 *
 * Builtin report command: Analyze the perf.data input file,
 * look up and read DSOs and symbol information and display
 * a histogram of results, along various sorting keys.
 */
8
#include "builtin.h"
9

10 11
#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"
21

22
#include "perf.h"
23
#include "util/debug.h"
24 25
#include "util/evlist.h"
#include "util/evsel.h"
26
#include "util/header.h"
27
#include "util/session.h"
28 29 30 31

#include "util/parse-options.h"
#include "util/parse-events.h"

32
#include "util/thread.h"
33
#include "util/sort.h"
34
#include "util/hist.h"
35

36 37
#include <linux/bitmap.h>

38
static char		const *input_name = "perf.data";
39

40
static bool		force, use_tui, use_stdio;
41
static bool		hide_unresolved;
42
static bool		dont_use_callchains;
43

44
static bool		show_threads;
45 46
static struct perf_read_values	show_threads_values;

47 48
static const char	default_pretty_printing_style[] = "normal";
static const char	*pretty_printing_style = default_pretty_printing_style;
49

50 51
static char		callchain_default_opt[] = "fractal,0.5,callee";
static bool		inverted_callchain;
52
static symbol_filter_t	annotate_init;
53

54 55 56
static const char	*cpu_list;
static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);

57
static int perf_session__add_hist_entry(struct perf_session *session,
58
					struct addr_location *al,
59 60
					struct perf_sample *sample,
					struct perf_evsel *evsel)
61
{
62
	struct symbol *parent = NULL;
63
	int err = 0;
64 65
	struct hist_entry *he;

66 67 68
	if ((sort__has_parent || symbol_conf.use_callchain) && sample->callchain) {
		err = perf_session__resolve_callchain(session, al->thread,
						      sample->callchain, &parent);
69 70
		if (err)
			return err;
71
	}
72

73
	he = __hists__add_entry(&evsel->hists, al, parent, sample->period);
74
	if (he == NULL)
75 76
		return -ENOMEM;

77
	if (symbol_conf.use_callchain) {
78 79
		err = callchain_append(he->callchain, &session->callchain_cursor,
				       sample->period);
80
		if (err)
81
			return err;
82 83 84 85 86 87
	}
	/*
	 * Only in the newt browser we are doing integrated annotation,
	 * so we don't allocated the extra space needed because the stdio
	 * code will not use it.
	 */
88 89
	if (al->sym != NULL && use_browser > 0) {
		struct annotation *notes = symbol__annotation(he->ms.sym);
90 91 92 93

		assert(evsel != NULL);

		err = -ENOMEM;
94
		if (notes->src == NULL &&
95 96 97 98
		    symbol__alloc_hist(he->ms.sym, session->evlist->nr_entries) < 0)
			goto out;

		err = hist_entry__inc_addr_samples(he, evsel->idx, al->addr);
99
	}
100

101 102 103
	evsel->hists.stats.total_period += sample->period;
	hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
out:
104
	return err;
105 106
}

107

108 109
static int process_sample_event(union perf_event *event,
				struct perf_sample *sample,
110
				struct perf_evsel *evsel,
111
				struct perf_session *session)
112
{
113
	struct addr_location al;
114

115
	if (perf_event__preprocess_sample(event, session, &al, sample,
116
					  annotate_init) < 0) {
117
		fprintf(stderr, "problem processing %d event, skipping it.\n",
118 119 120
			event->header.type);
		return -1;
	}
121

122
	if (al.filtered || (hide_unresolved && al.sym == NULL))
123
		return 0;
124

125 126 127
	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
		return 0;

128 129 130
	if (al.map != NULL)
		al.map->dso->hit = 1;

131
	if (perf_session__add_hist_entry(session, &al, sample, evsel)) {
132
		pr_debug("problem incrementing symbol period, skipping event\n");
133
		return -1;
134
	}
135

136 137
	return 0;
}
I
Ingo Molnar 已提交
138

139 140
static int process_read_event(union perf_event *event,
			      struct perf_sample *sample __used,
141
			      struct perf_session *session)
142
{
143 144
	struct perf_evsel *evsel = perf_evlist__id2evsel(session->evlist,
							 event->read.id);
145
	if (show_threads) {
146
		const char *name = evsel ? event_name(evsel) : "unknown";
147 148 149 150 151 152 153
		perf_read_values_add_value(&show_threads_values,
					   event->read.pid, event->read.tid,
					   event->read.id,
					   name,
					   event->read.value);
	}

154
	dump_printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
155
		    evsel ? event_name(evsel) : "FAIL",
156
		    event->read.value);
157 158 159 160

	return 0;
}

161
static int perf_session__setup_sample_type(struct perf_session *self)
162
{
163
	if (!(self->sample_type & PERF_SAMPLE_CALLCHAIN)) {
164
		if (sort__has_parent) {
165 166 167
			ui__warning("Selected --sort parent, but no "
				    "callchain data. Did you call "
				    "'perf record' without -g?\n");
168
			return -EINVAL;
169
		}
170
		if (symbol_conf.use_callchain) {
171 172
			ui__warning("Selected -g but no callchain data. Did "
				    "you call 'perf record' without -g?\n");
173
			return -1;
174
		}
175 176
	} else if (!dont_use_callchains && callchain_param.mode != CHAIN_NONE &&
		   !symbol_conf.use_callchain) {
177
			symbol_conf.use_callchain = true;
178
			if (callchain_register_param(&callchain_param) < 0) {
179 180
				ui__warning("Can't register callchain "
					    "params.\n");
181
				return -EINVAL;
182
			}
183 184
	}

185 186
	return 0;
}
187

188
static struct perf_event_ops event_ops = {
189 190 191 192 193 194 195 196 197 198 199
	.sample		 = process_sample_event,
	.mmap		 = perf_event__process_mmap,
	.comm		 = perf_event__process_comm,
	.exit		 = perf_event__process_task,
	.fork		 = perf_event__process_task,
	.lost		 = perf_event__process_lost,
	.read		 = process_read_event,
	.attr		 = perf_event__process_attr,
	.event_type	 = perf_event__process_event_type,
	.tracing_data	 = perf_event__process_tracing_data,
	.build_id	 = perf_event__process_build_id,
200 201
	.ordered_samples = true,
	.ordering_requires_timestamps = true,
202
};
203

204 205
extern volatile int session_done;

206
static void sig_handler(int sig __used)
207 208 209 210
{
	session_done = 1;
}

211 212 213 214 215 216 217 218 219 220 221 222 223 224
static size_t hists__fprintf_nr_sample_events(struct hists *self,
					      const char *evname, FILE *fp)
{
	size_t ret;
	char unit;
	unsigned long nr_events = self->stats.nr_events[PERF_RECORD_SAMPLE];

	nr_events = convert_unit(nr_events, &unit);
	ret = fprintf(fp, "# Events: %lu%c", nr_events, unit);
	if (evname != NULL)
		ret += fprintf(fp, " %s", evname);
	return ret + fprintf(fp, "\n#\n");
}

225 226
static int perf_evlist__tty_browse_hists(struct perf_evlist *evlist,
					 const char *help)
227
{
228
	struct perf_evsel *pos;
229

230 231
	list_for_each_entry(pos, &evlist->entries, node) {
		struct hists *hists = &pos->hists;
232
		const char *evname = event_name(pos);
233 234

		hists__fprintf_nr_sample_events(hists, evname, stdout);
235
		hists__fprintf(hists, NULL, false, true, 0, 0, stdout);
236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253
		fprintf(stdout, "\n\n");
	}

	if (sort_order == default_sort_order &&
	    parent_pattern == default_parent_pattern) {
		fprintf(stdout, "#\n# (%s)\n#\n", help);

		if (show_threads) {
			bool style = !strcmp(pretty_printing_style, "raw");
			perf_read_values_display(stdout, &show_threads_values,
						 style);
			perf_read_values_destroy(&show_threads_values);
		}
	}

	return 0;
}

254 255
static int __cmd_report(void)
{
256
	int ret = -EINVAL;
257
	u64 nr_samples;
258
	struct perf_session *session;
259
	struct perf_evsel *pos;
260 261
	struct map *kernel_map;
	struct kmap *kernel_kmap;
262
	const char *help = "For a higher level overview, try: perf report --sort comm,dso";
263

264 265
	signal(SIGINT, sig_handler);

266
	session = perf_session__new(input_name, O_RDONLY, force, false, &event_ops);
267 268 269
	if (session == NULL)
		return -ENOMEM;

270 271 272 273 274 275
	if (cpu_list) {
		ret = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap);
		if (ret)
			goto out_delete;
	}

276 277
	if (show_threads)
		perf_read_values_init(&show_threads_values);
278

279 280 281 282
	ret = perf_session__setup_sample_type(session);
	if (ret)
		goto out_delete;

283
	ret = perf_session__process_events(session, &event_ops);
284
	if (ret)
285
		goto out_delete;
286

287 288 289 290 291 292 293 294
	kernel_map = session->host_machine.vmlinux_maps[MAP__FUNCTION];
	kernel_kmap = map__kmap(kernel_map);
	if (kernel_map == NULL ||
	    (kernel_map->dso->hit &&
	     (kernel_kmap->ref_reloc_sym == NULL ||
	      kernel_kmap->ref_reloc_sym->addr == 0))) {
		const struct dso *kdso = kernel_map->dso;

295 296 297 298
		ui__warning(
"Kernel address maps (/proc/{kallsyms,modules}) were restricted.\n\n"
"Check /proc/sys/kernel/kptr_restrict before running 'perf record'.\n\n%s\n\n"
"Samples in kernel modules can't be resolved as well.\n\n",
299
			    RB_EMPTY_ROOT(&kdso->symbols[MAP__FUNCTION]) ?
300 301 302
"As no suitable kallsyms nor vmlinux was found, kernel samples\n"
"can't be resolved." :
"If some relocation was applied (e.g. kexec) symbols may be misresolved.");
303 304
	}

305
	if (dump_trace) {
306
		perf_session__fprintf_nr_events(session, stdout);
307
		goto out_delete;
308
	}
309

310
	if (verbose > 3)
311
		perf_session__fprintf(session, stdout);
312

313
	if (verbose > 2)
314
		perf_session__fprintf_dsos(session, stdout);
315

316 317 318
	nr_samples = 0;
	list_for_each_entry(pos, &session->evlist->entries, node) {
		struct hists *hists = &pos->hists;
319

320
		hists__collapse_resort(hists);
321
		hists__output_resort(hists);
322 323 324 325 326 327
		nr_samples += hists->stats.nr_events[PERF_RECORD_SAMPLE];
	}

	if (nr_samples == 0) {
		ui__warning("The %s file has no samples!\n", input_name);
		goto out_delete;
328 329
	}

330
	if (use_browser > 0)
331
		perf_evlist__tui_browse_hists(session->evlist, help);
332
	else
333
		perf_evlist__tty_browse_hists(session->evlist, help);
334

335
out_delete:
336 337 338 339 340 341 342 343 344 345 346 347
	/*
	 * Speed up the exit process, for large files this can
	 * take quite a while.
	 *
	 * XXX Enable this when using valgrind or if we ever
	 * librarize this command.
	 *
	 * Also experiment with obstacks to see how much speed
	 * up we'll get here.
	 *
 	 * perf_session__delete(session);
 	 */
348
	return ret;
349 350
}

351 352
static int
parse_callchain_opt(const struct option *opt __used, const char *arg,
353
		    int unset)
354
{
355
	char *tok, *tok2;
356 357
	char *endptr;

358 359 360 361 362 363 364 365
	/*
	 * --no-call-graph
	 */
	if (unset) {
		dont_use_callchains = true;
		return 0;
	}

366
	symbol_conf.use_callchain = true;
367 368 369 370

	if (!arg)
		return 0;

371 372 373 374 375 376
	tok = strtok((char *)arg, ",");
	if (!tok)
		return -1;

	/* get the output mode */
	if (!strncmp(tok, "graph", strlen(arg)))
377
		callchain_param.mode = CHAIN_GRAPH_ABS;
378

379
	else if (!strncmp(tok, "flat", strlen(arg)))
380 381 382 383 384
		callchain_param.mode = CHAIN_FLAT;

	else if (!strncmp(tok, "fractal", strlen(arg)))
		callchain_param.mode = CHAIN_GRAPH_REL;

385 386
	else if (!strncmp(tok, "none", strlen(arg))) {
		callchain_param.mode = CHAIN_NONE;
387
		symbol_conf.use_callchain = false;
388 389 390 391

		return 0;
	}

392 393 394
	else
		return -1;

395 396 397
	/* get the min percentage */
	tok = strtok(NULL, ",");
	if (!tok)
398
		goto setup;
399

400
	callchain_param.min_percent = strtod(tok, &endptr);
401 402 403
	if (tok == endptr)
		return -1;

404 405 406 407 408 409
	/* get the print limit */
	tok2 = strtok(NULL, ",");
	if (!tok2)
		goto setup;

	if (tok2[0] != 'c') {
410
		callchain_param.print_limit = strtod(tok2, &endptr);
411 412 413 414 415 416 417 418 419 420 421 422
		tok2 = strtok(NULL, ",");
		if (!tok2)
			goto setup;
	}

	/* get the call chain order */
	if (!strcmp(tok2, "caller"))
		callchain_param.order = ORDER_CALLER;
	else if (!strcmp(tok2, "callee"))
		callchain_param.order = ORDER_CALLEE;
	else
		return -1;
423
setup:
424
	if (callchain_register_param(&callchain_param) < 0) {
425 426 427
		fprintf(stderr, "Can't register callchain params\n");
		return -1;
	}
428 429 430
	return 0;
}

431
static const char * const report_usage[] = {
432 433 434 435 436 437 438
	"perf report [<options>] <command>",
	NULL
};

static const struct option options[] = {
	OPT_STRING('i', "input", &input_name, "file",
		    "input file name"),
439
	OPT_INCR('v', "verbose", &verbose,
440
		    "be more verbose (show symbol address, etc)"),
441 442
	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
		    "dump raw trace in ASCII"),
443 444
	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
		   "file", "vmlinux pathname"),
445 446
	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
		   "file", "kallsyms pathname"),
447
	OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
448
	OPT_BOOLEAN('m', "modules", &symbol_conf.use_modules,
449
		    "load module symbols - WARNING: use only with -k and LIVE kernel"),
450
	OPT_BOOLEAN('n', "show-nr-samples", &symbol_conf.show_nr_samples,
451
		    "Show a column with the number of samples"),
452 453
	OPT_BOOLEAN('T', "threads", &show_threads,
		    "Show per-thread event counters"),
454 455
	OPT_STRING(0, "pretty", &pretty_printing_style, "key",
		   "pretty printing style key: normal raw"),
456 457
	OPT_BOOLEAN(0, "tui", &use_tui, "Use the TUI interface"),
	OPT_BOOLEAN(0, "stdio", &use_stdio, "Use the stdio interface"),
458
	OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
459
		   "sort by key(s): pid, comm, dso, symbol, parent"),
460 461
	OPT_BOOLEAN(0, "showcpuutilization", &symbol_conf.show_cpu_utilization,
		    "Show sample percentage for different cpu modes"),
462 463
	OPT_STRING('p', "parent", &parent_pattern, "regex",
		   "regex filter to identify parent, see: '--sort parent'"),
464
	OPT_BOOLEAN('x', "exclude-other", &symbol_conf.exclude_other,
465
		    "Only display entries with parent-match"),
466 467 468 469
	OPT_CALLBACK_DEFAULT('g', "call-graph", NULL, "output_type,min_percent, call_order",
		     "Display callchains using output_type (graph, flat, fractal, or none) , min percent threshold and callchain order. "
		     "Default: fractal,0.5,callee", &parse_callchain_opt, callchain_default_opt),
	OPT_BOOLEAN('G', "inverted", &inverted_callchain, "alias for inverted call graph"),
470
	OPT_STRING('d', "dsos", &symbol_conf.dso_list_str, "dso[,dso...]",
471
		   "only consider symbols in these dsos"),
472
	OPT_STRING('C', "comms", &symbol_conf.comm_list_str, "comm[,comm...]",
473
		   "only consider symbols in these comms"),
474
	OPT_STRING('S', "symbols", &symbol_conf.sym_list_str, "symbol[,symbol...]",
475
		   "only consider these symbols"),
476
	OPT_STRING('w', "column-widths", &symbol_conf.col_width_list_str,
477 478
		   "width[,width...]",
		   "don't try to adjust column width, use these fixed values"),
479
	OPT_STRING('t', "field-separator", &symbol_conf.field_sep, "separator",
480 481
		   "separator for columns, no spaces will be added between "
		   "columns '.' is reserved."),
482 483
	OPT_BOOLEAN('U', "hide-unresolved", &hide_unresolved,
		    "Only display entries resolved to a symbol"),
484 485
	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
		    "Look for files with symbols relative to this directory"),
486
	OPT_STRING('c', "cpu", &cpu_list, "cpu", "list of cpus to profile"),
487 488
	OPT_STRING('M', "disassembler-style", &disassembler_style, "disassembler style",
		   "Specify disassembler style (e.g. -M intel for intel syntax)"),
489 490
	OPT_BOOLEAN(0, "show-total-period", &symbol_conf.show_total_period,
		    "Show a column with the sum of periods"),
491 492 493
	OPT_END()
};

494
int cmd_report(int argc, const char **argv, const char *prefix __used)
495
{
496 497
	argc = parse_options(argc, argv, options, report_usage, 0);

498 499 500 501 502
	if (use_stdio)
		use_browser = 0;
	else if (use_tui)
		use_browser = 1;

503 504 505
	if (inverted_callchain)
		callchain_param.order = ORDER_CALLER;

506
	if (strcmp(input_name, "-") != 0)
507
		setup_browser(true);
508 509
	else
		use_browser = 0;
510 511 512 513 514
	/*
	 * Only in the newt browser we are doing integrated annotation,
	 * so don't allocate extra space that won't be used in the stdio
	 * implementation.
	 */
515
	if (use_browser > 0) {
516
		symbol_conf.priv_size = sizeof(struct annotation);
517
		annotate_init	      = symbol__annotate_init;
518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533
		/*
 		 * For searching by name on the "Browse map details".
 		 * providing it only in verbose mode not to bloat too
 		 * much struct symbol.
 		 */
		if (verbose) {
			/*
			 * XXX: Need to provide a less kludgy way to ask for
			 * more space per symbol, the u32 is for the index on
			 * the ui browser.
			 * See symbol__browser_index.
			 */
			symbol_conf.priv_size += sizeof(u32);
			symbol_conf.sort_by_name = true;
		}
	}
534

535
	if (symbol__init() < 0)
536
		return -1;
537

538
	setup_sorting(report_usage, options);
539

540
	if (parent_pattern != default_parent_pattern) {
541 542
		if (sort_dimension__add("parent") < 0)
			return -1;
543 544 545 546 547 548 549 550

		/*
		 * Only show the parent fields if we explicitly
		 * sort that way. If we only use parent machinery
		 * for filtering, we don't want it.
		 */
		if (!strstr(sort_order, "parent"))
			sort_parent.elide = 1;
551
	} else
552
		symbol_conf.exclude_other = false;
553

554 555 556 557 558 559
	/*
	 * Any (unrecognized) arguments left?
	 */
	if (argc)
		usage_with_options(report_usage, options);

560 561 562
	sort_entry__setup_elide(&sort_dso, symbol_conf.dso_list, "dso", stdout);
	sort_entry__setup_elide(&sort_comm, symbol_conf.comm_list, "comm", stdout);
	sort_entry__setup_elide(&sort_sym, symbol_conf.sym_list, "symbol", stdout);
563

564 565
	return __cmd_report();
}