builtin-record.c 22.0 KB
Newer Older
I
Ingo Molnar 已提交
1
/*
2 3 4 5 6
 * builtin-record.c
 *
 * Builtin record command: Record the profile of a workload
 * (or a CPU, or a PID) into the perf.data output file - for
 * later analysis via perf report.
I
Ingo Molnar 已提交
7
 */
8 9
#define _FILE_OFFSET_BITS 64

10
#include "builtin.h"
11 12 13

#include "perf.h"

14
#include "util/build-id.h"
15
#include "util/util.h"
16
#include "util/parse-options.h"
17
#include "util/parse-events.h"
18

19
#include "util/header.h"
20
#include "util/event.h"
21
#include "util/evlist.h"
22
#include "util/evsel.h"
23
#include "util/debug.h"
24
#include "util/session.h"
25
#include "util/symbol.h"
26
#include "util/cpumap.h"
27
#include "util/thread_map.h"
28

29
#include <unistd.h>
30
#include <sched.h>
31
#include <sys/mman.h>
32

33 34
#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))

35 36 37 38 39
enum write_mode_t {
	WRITE_FORCE,
	WRITE_APPEND
};

40 41
static u64			user_interval			= ULLONG_MAX;
static u64			default_interval		=      0;
42

43
static unsigned int		page_size;
44
static unsigned int		mmap_pages			= UINT_MAX;
45
static unsigned int		user_freq 			= UINT_MAX;
46
static int			freq				=   1000;
47
static int			output;
48
static int			pipe_output			=      0;
49
static const char		*output_name			= NULL;
50
static int			group				=      0;
51
static int			realtime_prio			=      0;
52
static bool			nodelay				=  false;
53
static bool			raw_samples			=  false;
54
static bool			sample_id_all_avail		=   true;
55
static bool			system_wide			=  false;
56
static pid_t			target_pid			=     -1;
57
static pid_t			target_tid			=     -1;
58
static pid_t			child_pid			=     -1;
59
static bool			no_inherit			=  false;
60
static enum write_mode_t	write_mode			= WRITE_FORCE;
61 62 63 64
static bool			call_graph			=  false;
static bool			inherit_stat			=  false;
static bool			no_samples			=  false;
static bool			sample_address			=  false;
65
static bool			sample_time			=  false;
66
static bool			no_buildid			=  false;
67
static bool			no_buildid_cache		=  false;
68
static struct perf_evlist	*evsel_list;
69 70 71

static long			samples				=      0;
static u64			bytes_written			=      0;
72

73
static int			file_new			=      1;
74
static off_t			post_processing_offset;
75

76
static struct perf_session	*session;
77
static const char		*cpu_list;
78

79 80 81 82 83
static void advance_output(size_t size)
{
	bytes_written += size;
}

84 85 86 87 88 89 90 91 92 93 94 95 96 97 98
static void write_output(void *buf, size_t size)
{
	while (size) {
		int ret = write(output, buf, size);

		if (ret < 0)
			die("failed to write");

		size -= ret;
		buf += ret;

		bytes_written += ret;
	}
}

99
static int process_synthesized_event(union perf_event *event,
100
				     struct perf_sample *sample __used,
101
				     struct perf_session *self __used)
102
{
103
	write_output(event, event->header.size);
104 105 106
	return 0;
}

107
static void mmap_read(struct perf_mmap *md)
108
{
109
	unsigned int head = perf_mmap__read_head(md);
110 111 112 113 114
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	unsigned long size;
	void *buf;

115 116 117 118
	if (old == head)
		return;

	samples++;
119 120 121 122 123 124 125

	size = head - old;

	if ((old & md->mask) + size != (head & md->mask)) {
		buf = &data[old & md->mask];
		size = md->mask + 1 - (old & md->mask);
		old += size;
126

127
		write_output(buf, size);
128 129 130 131 132
	}

	buf = &data[old & md->mask];
	size = head - old;
	old += size;
133

134
	write_output(buf, size);
135 136

	md->prev = old;
137
	perf_mmap__write_tail(md, old);
138 139 140
}

static volatile int done = 0;
141
static volatile int signr = -1;
142

143
static void sig_handler(int sig)
144
{
145
	done = 1;
146 147 148 149 150
	signr = sig;
}

static void sig_atexit(void)
{
151
	if (child_pid > 0)
152 153
		kill(child_pid, SIGTERM);

154
	if (signr == -1 || signr == SIGUSR1)
155 156 157 158
		return;

	signal(signr, SIG_DFL);
	kill(getpid(), signr);
159 160
}

161 162 163 164
static void config_attr(struct perf_evsel *evsel, struct perf_evlist *evlist)
{
	struct perf_event_attr *attr = &evsel->attr;
	int track = !evsel->idx; /* only the first counter needs these */
165

166
	attr->inherit		= !no_inherit;
167 168 169
	attr->read_format	= PERF_FORMAT_TOTAL_TIME_ENABLED |
				  PERF_FORMAT_TOTAL_TIME_RUNNING |
				  PERF_FORMAT_ID;
170

171
	attr->sample_type	|= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
172

173
	if (evlist->nr_entries > 1)
174 175
		attr->sample_type |= PERF_SAMPLE_ID;

176 177 178 179 180
	/*
	 * We default some events to a 1 default interval. But keep
	 * it a weak assumption overridable by the user.
	 */
	if (!attr->sample_period || (user_freq != UINT_MAX &&
181
				     user_interval != ULLONG_MAX)) {
182 183 184 185 186 187 188
		if (freq) {
			attr->sample_type	|= PERF_SAMPLE_PERIOD;
			attr->freq		= 1;
			attr->sample_freq	= freq;
		} else {
			attr->sample_period = default_interval;
		}
189
	}
190

191 192 193 194 195 196
	if (no_samples)
		attr->sample_freq = 0;

	if (inherit_stat)
		attr->inherit_stat = 1;

197
	if (sample_address) {
198
		attr->sample_type	|= PERF_SAMPLE_ADDR;
199 200
		attr->mmap_data = track;
	}
201

202 203 204
	if (call_graph)
		attr->sample_type	|= PERF_SAMPLE_CALLCHAIN;

A
Arun Sharma 已提交
205 206 207
	if (system_wide)
		attr->sample_type	|= PERF_SAMPLE_CPU;

208 209
	if (sample_id_all_avail &&
	    (sample_time || system_wide || !no_inherit || cpu_list))
210 211
		attr->sample_type	|= PERF_SAMPLE_TIME;

I
Ingo Molnar 已提交
212
	if (raw_samples) {
213
		attr->sample_type	|= PERF_SAMPLE_TIME;
214
		attr->sample_type	|= PERF_SAMPLE_RAW;
I
Ingo Molnar 已提交
215 216
		attr->sample_type	|= PERF_SAMPLE_CPU;
	}
217

218 219 220 221 222
	if (nodelay) {
		attr->watermark = 0;
		attr->wakeup_events = 1;
	}

223 224
	attr->mmap		= track;
	attr->comm		= track;
225

226
	if (target_pid == -1 && target_tid == -1 && !system_wide) {
227
		attr->disabled = 1;
228
		attr->enable_on_exec = 1;
229
	}
230
}
231

232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
static bool perf_evlist__equal(struct perf_evlist *evlist,
			       struct perf_evlist *other)
{
	struct perf_evsel *pos, *pair;

	if (evlist->nr_entries != other->nr_entries)
		return false;

	pair = list_entry(other->entries.next, struct perf_evsel, node);

	list_for_each_entry(pos, &evlist->entries, node) {
		if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0))
			return false;
		pair = list_entry(pair->node.next, struct perf_evsel, node);
	}

	return true;
}

251 252 253 254
static void open_counters(struct perf_evlist *evlist)
{
	struct perf_evsel *pos;

255 256 257
	if (evlist->cpus->map[0] < 0)
		no_inherit = true;

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
	list_for_each_entry(pos, &evlist->entries, node) {
		struct perf_event_attr *attr = &pos->attr;
		/*
		 * Check if parse_single_tracepoint_event has already asked for
		 * PERF_SAMPLE_TIME.
		 *
		 * XXX this is kludgy but short term fix for problems introduced by
		 * eac23d1c that broke 'perf script' by having different sample_types
		 * when using multiple tracepoint events when we use a perf binary
		 * that tries to use sample_id_all on an older kernel.
		 *
		 * We need to move counter creation to perf_session, support
		 * different sample_types, etc.
		 */
		bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
273

274 275 276 277
		config_attr(pos, evlist);
retry_sample_id:
		attr->sample_id_all = sample_id_all_avail ? 1 : 0;
try_again:
278
		if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group) < 0) {
279 280
			int err = errno;

281 282 283 284
			if (err == EPERM || err == EACCES) {
				ui__warning_paranoid();
				exit(EXIT_FAILURE);
			} else if (err ==  ENODEV && cpu_list) {
285 286
				die("No such device - did you specify"
					" an out-of-range profile CPU?\n");
287 288 289 290 291
			} else if (err == EINVAL && sample_id_all_avail) {
				/*
				 * Old kernel, no attr->sample_id_type_all field
				 */
				sample_id_all_avail = false;
292
				if (!sample_time && !raw_samples && !time_needed)
293 294
					attr->sample_type &= ~PERF_SAMPLE_TIME;

295
				goto retry_sample_id;
296
			}
297

298 299 300 301 302 303 304 305 306
			/*
			 * If it's cycles then fall back to hrtimer
			 * based cpu-clock-tick sw counter, which
			 * is always available even if no PMU support:
			 */
			if (attr->type == PERF_TYPE_HARDWARE
					&& attr->config == PERF_COUNT_HW_CPU_CYCLES) {

				if (verbose)
307 308
					ui__warning("The cycles event is not supported, "
						    "trying to fall back to cpu-clock-ticks\n");
309 310 311 312
				attr->type = PERF_TYPE_SOFTWARE;
				attr->config = PERF_COUNT_SW_CPU_CLOCK;
				goto try_again;
			}
313 314 315 316 317 318 319

			if (err == ENOENT) {
				ui__warning("The %s event is not supported.\n",
					    event_name(pos));
				exit(EXIT_FAILURE);
			}

320
			printf("\n");
321
			error("sys_perf_event_open() syscall returned with %d (%s).  /bin/dmesg may provide additional information.\n",
322
			      err, strerror(err));
323 324

#if defined(__i386__) || defined(__x86_64__)
325 326 327 328 329
			if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
				die("No hardware sampling interrupt available."
				    " No APIC? If so then you can boot the kernel"
				    " with the \"lapic\" boot parameter to"
				    " force-enable it.\n");
330 331
#endif

332
			die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
L
Li Zefan 已提交
333 334
		}
	}
335

336 337 338 339 340 341
	if (perf_evlist__set_filters(evlist)) {
		error("failed to set filter with %d (%s)\n", errno,
			strerror(errno));
		exit(-1);
	}

342
	if (perf_evlist__mmap(evlist, mmap_pages, false) < 0)
343 344
		die("failed to mmap with %d (%s)\n", errno, strerror(errno));

345 346 347 348 349 350 351 352 353 354
	if (file_new)
		session->evlist = evlist;
	else {
		if (!perf_evlist__equal(session->evlist, evlist)) {
			fprintf(stderr, "incompatible append\n");
			exit(-1);
		}
 	}

	perf_session__update_sample_type(session);
355 356
}

357 358 359 360
static int process_buildids(void)
{
	u64 size = lseek(output, 0, SEEK_CUR);

361 362 363
	if (size == 0)
		return 0;

364 365 366 367 368 369
	session->fd = output;
	return __perf_session__process_events(session, post_processing_offset,
					      size - post_processing_offset,
					      size, &build_id__mark_dso_hit_ops);
}

370 371
static void atexit_header(void)
{
372 373
	if (!pipe_output) {
		session->header.data_size += bytes_written;
374

375 376
		if (!no_buildid)
			process_buildids();
377
		perf_session__write_header(session, evsel_list, output, true);
378
		perf_session__delete(session);
379
		perf_evlist__delete(evsel_list);
380
		symbol__exit();
381
	}
382 383
}

384
static void perf_event__synthesize_guest_os(struct machine *machine, void *data)
385 386
{
	int err;
387
	struct perf_session *psession = data;
388

389
	if (machine__is_host(machine))
390 391 392 393 394 395 396 397 398 399
		return;

	/*
	 *As for guest kernel when processing subcommand record&report,
	 *we arrange module mmap prior to guest kernel mmap and trigger
	 *a preload dso because default guest module symbols are loaded
	 *from guest kallsyms instead of /lib/modules/XXX/XXX. This
	 *method is used to avoid symbol missing when the first addr is
	 *in module instead of in guest kernel.
	 */
400 401
	err = perf_event__synthesize_modules(process_synthesized_event,
					     psession, machine);
402 403
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
404
		       " relocation symbol.\n", machine->pid);
405 406 407 408 409

	/*
	 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
	 * have no _text sometimes.
	 */
410 411
	err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
						 psession, machine, "_text");
412
	if (err < 0)
413 414 415
		err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
							 psession, machine,
							 "_stext");
416 417
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
418
		       " relocation symbol.\n", machine->pid);
419 420
}

421 422 423 424 425 426 427
static struct perf_event_header finished_round_event = {
	.size = sizeof(struct perf_event_header),
	.type = PERF_RECORD_FINISHED_ROUND,
};

static void mmap_read_all(void)
{
428
	int i;
429

430
	for (i = 0; i < evsel_list->nr_mmaps; i++) {
431 432
		if (evsel_list->mmap[i].base)
			mmap_read(&evsel_list->mmap[i]);
433 434 435 436 437 438
	}

	if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO))
		write_output(&finished_round_event, sizeof(finished_round_event));
}

439
static int __cmd_record(int argc, const char **argv)
440
{
441
	int i;
I
Ingo Molnar 已提交
442 443
	struct stat st;
	int flags;
444
	int err;
445
	unsigned long waking = 0;
446
	int child_ready_pipe[2], go_pipe[2];
447
	const bool forks = argc > 0;
448
	char buf;
449
	struct machine *machine;
450 451 452

	page_size = sysconf(_SC_PAGE_SIZE);

453 454 455
	atexit(sig_atexit);
	signal(SIGCHLD, sig_handler);
	signal(SIGINT, sig_handler);
456
	signal(SIGUSR1, sig_handler);
457

458
	if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
459 460 461 462
		perror("failed to create pipes");
		exit(-1);
	}

463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481
	if (!output_name) {
		if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode))
			pipe_output = 1;
		else
			output_name = "perf.data";
	}
	if (output_name) {
		if (!strcmp(output_name, "-"))
			pipe_output = 1;
		else if (!stat(output_name, &st) && st.st_size) {
			if (write_mode == WRITE_FORCE) {
				char oldname[PATH_MAX];
				snprintf(oldname, sizeof(oldname), "%s.old",
					 output_name);
				unlink(oldname);
				rename(output_name, oldname);
			}
		} else if (write_mode == WRITE_APPEND) {
			write_mode = WRITE_FORCE;
482
		}
483 484
	}

485
	flags = O_CREAT|O_RDWR;
486
	if (write_mode == WRITE_APPEND)
487
		file_new = 0;
I
Ingo Molnar 已提交
488 489 490
	else
		flags |= O_TRUNC;

491 492 493 494
	if (pipe_output)
		output = STDOUT_FILENO;
	else
		output = open(output_name, flags, S_IRUSR | S_IWUSR);
495 496 497 498 499
	if (output < 0) {
		perror("failed to create output file");
		exit(-1);
	}

500
	session = perf_session__new(output_name, O_WRONLY,
501
				    write_mode == WRITE_FORCE, false, NULL);
502
	if (session == NULL) {
503 504 505 506
		pr_err("Not enough memory for reading perf file header\n");
		return -1;
	}

507 508 509
	if (!no_buildid)
		perf_header__set_feat(&session->header, HEADER_BUILD_ID);

510
	if (!file_new) {
511
		err = perf_session__read_header(session, output);
512
		if (err < 0)
513
			goto out_delete_session;
514 515
	}

516
	if (have_tracepoints(&evsel_list->entries))
517
		perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
518

519 520 521 522
	/* 512 kiB: default amount of unprivileged mlocked memory */
	if (mmap_pages == UINT_MAX)
		mmap_pages = (512 * 1024) / page_size;

523
	if (forks) {
524
		child_pid = fork();
525
		if (child_pid < 0) {
526 527 528
			perror("failed to fork");
			exit(-1);
		}
529

530
		if (!child_pid) {
531 532
			if (pipe_output)
				dup2(2, 1);
533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557
			close(child_ready_pipe[0]);
			close(go_pipe[1]);
			fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);

			/*
			 * Do a dummy execvp to get the PLT entry resolved,
			 * so we avoid the resolver overhead on the real
			 * execvp call.
			 */
			execvp("", (char **)argv);

			/*
			 * Tell the parent we're ready to go
			 */
			close(child_ready_pipe[1]);

			/*
			 * Wait until the parent tells us to go.
			 */
			if (read(go_pipe[0], &buf, 1) == -1)
				perror("unable to read pipe");

			execvp(argv[0], (char **)argv);

			perror(argv[0]);
558
			kill(getppid(), SIGUSR1);
559
			exit(-1);
560
		}
561

562
		if (!system_wide && target_tid == -1 && target_pid == -1)
563
			evsel_list->threads->map[0] = child_pid;
564

565 566 567 568 569 570 571 572 573 574 575 576
		close(child_ready_pipe[1]);
		close(go_pipe[0]);
		/*
		 * wait for child to settle
		 */
		if (read(child_ready_pipe[0], &buf, 1) == -1) {
			perror("unable to read pipe");
			exit(-1);
		}
		close(child_ready_pipe[0]);
	}

577
	open_counters(evsel_list);
578

579 580 581 582 583
	/*
	 * perf_session__delete(session) will be called at atexit_header()
	 */
	atexit(atexit_header);

584 585 586 587 588
	if (pipe_output) {
		err = perf_header__write_pipe(output);
		if (err < 0)
			return err;
	} else if (file_new) {
589 590
		err = perf_session__write_header(session, evsel_list,
						 output, false);
591 592
		if (err < 0)
			return err;
593 594
	}

595 596
	post_processing_offset = lseek(output, 0, SEEK_CUR);

597
	if (pipe_output) {
598 599
		err = perf_session__synthesize_attrs(session,
						     process_synthesized_event);
600 601 602 603
		if (err < 0) {
			pr_err("Couldn't synthesize attrs.\n");
			return err;
		}
604

605 606
		err = perf_event__synthesize_event_types(process_synthesized_event,
							 session);
607 608 609 610
		if (err < 0) {
			pr_err("Couldn't synthesize event_types.\n");
			return err;
		}
611

612
		if (have_tracepoints(&evsel_list->entries)) {
613 614 615 616 617 618 619 620
			/*
			 * FIXME err <= 0 here actually means that
			 * there were no tracepoints so its not really
			 * an error, just that we don't need to
			 * synthesize anything.  We really have to
			 * return this more properly and also
			 * propagate errors that now are calling die()
			 */
621 622 623
			err = perf_event__synthesize_tracing_data(output, evsel_list,
								  process_synthesized_event,
								  session);
624 625 626 627
			if (err <= 0) {
				pr_err("Couldn't record tracing data.\n");
				return err;
			}
628
			advance_output(err);
629
		}
630 631
	}

632 633
	machine = perf_session__find_host_machine(session);
	if (!machine) {
634 635 636 637
		pr_err("Couldn't find native kernel information.\n");
		return -1;
	}

638 639
	err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
						 session, machine, "_text");
640
	if (err < 0)
641 642
		err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
							 session, machine, "_stext");
643 644 645 646
	if (err < 0)
		pr_err("Couldn't record kernel reference relocation symbol\n"
		       "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
		       "Check /proc/kallsyms permission or run as root.\n");
647

648 649
	err = perf_event__synthesize_modules(process_synthesized_event,
					     session, machine);
650 651 652 653 654
	if (err < 0)
		pr_err("Couldn't record kernel module information.\n"
		       "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
		       "Check /proc/modules permission or run as root.\n");

655
	if (perf_guest)
656 657
		perf_session__process_machines(session,
					       perf_event__synthesize_guest_os);
658

659
	if (!system_wide)
660 661 662
		perf_event__synthesize_thread_map(evsel_list->threads,
						  process_synthesized_event,
						  session);
663
	else
664 665
		perf_event__synthesize_threads(process_synthesized_event,
					       session);
666

667 668 669 670 671
	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
672
			pr_err("Could not set realtime priority.\n");
673 674 675 676
			exit(-1);
		}
	}

677 678 679
	/*
	 * Let the child rip
	 */
680 681
	if (forks)
		close(go_pipe[1]);
682

683
	for (;;) {
684
		int hits = samples;
685
		int thread;
686

687
		mmap_read_all();
688

689 690 691
		if (hits == samples) {
			if (done)
				break;
692
			err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
693 694 695 696
			waking++;
		}

		if (done) {
697
			for (i = 0; i < evsel_list->cpus->nr; i++) {
698 699
				struct perf_evsel *pos;

700
				list_for_each_entry(pos, &evsel_list->entries, node) {
701
					for (thread = 0;
702
						thread < evsel_list->threads->nr;
703
						thread++)
704
						ioctl(FD(pos, i, thread),
705 706
							PERF_EVENT_IOC_DISABLE);
				}
707
			}
708
		}
709 710
	}

711
	if (quiet || signr == SIGUSR1)
712 713
		return 0;

714 715
	fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);

716 717 718 719
	/*
	 * Approximate RIP event size: 24 bytes.
	 */
	fprintf(stderr,
720
		"[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n",
721 722 723
		(double)bytes_written / 1024.0 / 1024.0,
		output_name,
		bytes_written / 24);
724

725
	return 0;
726 727 728 729

out_delete_session:
	perf_session__delete(session);
	return err;
730
}
731 732

static const char * const record_usage[] = {
733 734
	"perf record [<options>] [<command>]",
	"perf record [<options>] -- <command> [<options>]",
735 736 737
	NULL
};

738 739
static bool force, append_file;

740
const struct option record_options[] = {
741
	OPT_CALLBACK('e', "event", &evsel_list, "event",
742 743
		     "event selector. use 'perf list' to list available events",
		     parse_events),
744
	OPT_CALLBACK(0, "filter", &evsel_list, "filter",
L
Li Zefan 已提交
745
		     "event filter", parse_filter),
746
	OPT_INTEGER('p', "pid", &target_pid,
747 748 749
		    "record events on existing process id"),
	OPT_INTEGER('t', "tid", &target_tid,
		    "record events on existing thread id"),
750 751
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
752 753
	OPT_BOOLEAN('D', "no-delay", &nodelay,
		    "collect data without buffering"),
754 755
	OPT_BOOLEAN('R', "raw-samples", &raw_samples,
		    "collect raw sample records from all opened counters"),
756 757
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
I
Ingo Molnar 已提交
758 759
	OPT_BOOLEAN('A', "append", &append_file,
			    "append to the output file to do incremental profiling"),
760 761
	OPT_STRING('C', "cpu", &cpu_list, "cpu",
		    "list of cpus to monitor"),
762
	OPT_BOOLEAN('f', "force", &force,
763
			"overwrite existing data file (deprecated)"),
764
	OPT_U64('c', "count", &user_interval, "event period to sample"),
I
Ingo Molnar 已提交
765 766
	OPT_STRING('o', "output", &output_name, "file",
		    "output file name"),
767 768
	OPT_BOOLEAN('i', "no-inherit", &no_inherit,
		    "child tasks do not inherit counters"),
769 770
	OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"),
	OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
771 772
	OPT_BOOLEAN('g', "call-graph", &call_graph,
		    "do call-graph (stack chain/backtrace) recording"),
773
	OPT_INCR('v', "verbose", &verbose,
774
		    "be more verbose (show counter open errors, etc)"),
775
	OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
776 777
	OPT_BOOLEAN('s', "stat", &inherit_stat,
		    "per thread counts"),
778 779
	OPT_BOOLEAN('d', "data", &sample_address,
		    "Sample addresses"),
780
	OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"),
781 782
	OPT_BOOLEAN('n', "no-samples", &no_samples,
		    "don't sample"),
783
	OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache,
784
		    "do not update the buildid cache"),
785 786
	OPT_BOOLEAN('B', "no-buildid", &no_buildid,
		    "do not collect buildids in perf.data"),
S
Stephane Eranian 已提交
787 788 789
	OPT_CALLBACK('G', "cgroup", &evsel_list, "name",
		     "monitor event in cgroup name only",
		     parse_cgroups),
790 791 792
	OPT_END()
};

793
int cmd_record(int argc, const char **argv, const char *prefix __used)
794
{
795 796
	int err = -ENOMEM;
	struct perf_evsel *pos;
797

798
	evsel_list = perf_evlist__new(NULL, NULL);
799 800 801
	if (evsel_list == NULL)
		return -ENOMEM;

802
	argc = parse_options(argc, argv, record_options, record_usage,
803
			    PARSE_OPT_STOP_AT_NON_OPTION);
804
	if (!argc && target_pid == -1 && target_tid == -1 &&
805
		!system_wide && !cpu_list)
806
		usage_with_options(record_usage, record_options);
807

808 809 810
	if (force && append_file) {
		fprintf(stderr, "Can't overwrite and append at the same time."
				" You need to choose between -f and -A");
811
		usage_with_options(record_usage, record_options);
812 813 814 815 816 817
	} else if (append_file) {
		write_mode = WRITE_APPEND;
	} else {
		write_mode = WRITE_FORCE;
	}

S
Stephane Eranian 已提交
818 819 820 821 822 823
	if (nr_cgroups && !system_wide) {
		fprintf(stderr, "cgroup monitoring only available in"
			" system-wide mode\n");
		usage_with_options(record_usage, record_options);
	}

824
	symbol__init();
825 826

	if (no_buildid_cache || no_buildid)
827
		disable_buildid_cache();
828

829 830
	if (evsel_list->nr_entries == 0 &&
	    perf_evlist__add_default(evsel_list) < 0) {
831 832
		pr_err("Not enough memory for event selector list\n");
		goto out_symbol_exit;
833
	}
834

835
	if (target_pid != -1)
836 837
		target_tid = target_pid;

838 839
	if (perf_evlist__create_maps(evsel_list, target_pid,
				     target_tid, cpu_list) < 0)
840
		usage_with_options(record_usage, record_options);
841

842
	list_for_each_entry(pos, &evsel_list->entries, node) {
843 844
		if (perf_evsel__alloc_fd(pos, evsel_list->cpus->nr,
					 evsel_list->threads->nr) < 0)
845
			goto out_free_fd;
846 847
		if (perf_header__push_event(pos->attr.config, event_name(pos)))
			goto out_free_fd;
848
	}
849

850
	if (perf_evlist__alloc_pollfd(evsel_list) < 0)
851
		goto out_free_fd;
852

853
	if (user_interval != ULLONG_MAX)
854 855 856 857
		default_interval = user_interval;
	if (user_freq != UINT_MAX)
		freq = user_freq;

858 859 860 861 862 863 864 865 866
	/*
	 * User specified count overrides default frequency.
	 */
	if (default_interval)
		freq = 0;
	else if (freq) {
		default_interval = freq;
	} else {
		fprintf(stderr, "frequency and count are zero, aborting\n");
867
		err = -EINVAL;
868
		goto out_free_fd;
869 870
	}

871 872
	err = __cmd_record(argc, argv);
out_free_fd:
873
	perf_evlist__delete_maps(evsel_list);
874 875
out_symbol_exit:
	symbol__exit();
876
	return err;
877
}