builtin-record.c 21.8 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/debug.h"
22
#include "util/session.h"
23
#include "util/symbol.h"
24
#include "util/cpumap.h"
25

26
#include <unistd.h>
27
#include <sched.h>
28
#include <sys/mman.h>
29

30 31 32 33 34
enum write_mode_t {
	WRITE_FORCE,
	WRITE_APPEND
};

35
static int			*fd[MAX_NR_CPUS][MAX_COUNTERS];
36

37 38
static u64			user_interval			= ULLONG_MAX;
static u64			default_interval		=      0;
39

40
static int			nr_cpus				=      0;
41
static unsigned int		page_size;
42
static unsigned int		mmap_pages			=    128;
43
static unsigned int		user_freq 			= UINT_MAX;
44
static int			freq				=   1000;
45
static int			output;
46
static int			pipe_output			=      0;
47
static const char		*output_name			= "perf.data";
48
static int			group				=      0;
49
static int			realtime_prio			=      0;
50 51
static bool			raw_samples			=  false;
static bool			system_wide			=  false;
52
static pid_t			target_pid			=     -1;
53 54 55
static pid_t			target_tid			=     -1;
static pid_t			*all_tids			=      NULL;
static int			thread_num			=      0;
56
static pid_t			child_pid			=     -1;
57
static bool			no_inherit			=  false;
58
static enum write_mode_t	write_mode			= WRITE_FORCE;
59 60 61 62
static bool			call_graph			=  false;
static bool			inherit_stat			=  false;
static bool			no_samples			=  false;
static bool			sample_address			=  false;
63
static bool			no_buildid			=  false;
64 65 66

static long			samples				=      0;
static u64			bytes_written			=      0;
67

68
static struct pollfd		*event_array;
69

70 71
static int			nr_poll				=      0;
static int			nr_cpu				=      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
struct mmap_data {
80 81 82 83
	int			counter;
	void			*base;
	unsigned int		mask;
	unsigned int		prev;
84 85
};

86
static struct mmap_data		mmap_array[MAX_NR_CPUS];
87

88
static unsigned long mmap_read_head(struct mmap_data *md)
89
{
90
	struct perf_event_mmap_page *pc = md->base;
91
	long head;
92 93 94 95 96 97 98

	head = pc->data_head;
	rmb();

	return head;
}

99 100
static void mmap_write_tail(struct mmap_data *md, unsigned long tail)
{
101
	struct perf_event_mmap_page *pc = md->base;
102 103 104 105 106 107 108 109

	/*
	 * ensure all reads are done before we write the tail out.
	 */
	/* mb(); */
	pc->data_tail = tail;
}

110 111 112 113 114
static void advance_output(size_t size)
{
	bytes_written += size;
}

115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
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;
	}
}

130 131
static int process_synthesized_event(event_t *event,
				     struct perf_session *self __used)
132
{
133
	write_output(event, event->header.size);
134 135 136
	return 0;
}

137 138 139 140 141 142 143 144 145 146 147
static void mmap_read(struct mmap_data *md)
{
	unsigned int head = mmap_read_head(md);
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	unsigned long size;
	void *buf;
	int diff;

	/*
	 * If we're further behind than half the buffer, there's a chance
148
	 * the writer will bite our tail and mess up the samples under us.
149 150 151 152 153 154
	 *
	 * If we somehow ended up ahead of the head, we got messed up.
	 *
	 * In either case, truncate and restart at head.
	 */
	diff = head - old;
155
	if (diff < 0) {
156
		fprintf(stderr, "WARNING: failed to keep up with mmap data\n");
157 158 159 160 161 162 163
		/*
		 * head points to a known good entry, start there.
		 */
		old = head;
	}

	if (old != head)
164
		samples++;
165 166 167 168 169 170 171

	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;
172

173
		write_output(buf, size);
174 175 176 177 178
	}

	buf = &data[old & md->mask];
	size = head - old;
	old += size;
179

180
	write_output(buf, size);
181 182

	md->prev = old;
183
	mmap_write_tail(md, old);
184 185 186
}

static volatile int done = 0;
187
static volatile int signr = -1;
188

189
static void sig_handler(int sig)
190
{
191
	done = 1;
192 193 194 195 196
	signr = sig;
}

static void sig_atexit(void)
{
197
	if (child_pid > 0)
198 199
		kill(child_pid, SIGTERM);

200 201 202 203 204
	if (signr == -1)
		return;

	signal(signr, SIG_DFL);
	kill(getpid(), signr);
205 206
}

207 208
static int group_fd;

209
static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr)
210 211 212
{
	struct perf_header_attr *h_attr;

213 214
	if (nr < session->header.attrs) {
		h_attr = session->header.attr[nr];
215 216
	} else {
		h_attr = perf_header_attr__new(a);
217
		if (h_attr != NULL)
218
			if (perf_header__add_attr(&session->header, h_attr) < 0) {
219 220 221
				perf_header_attr__delete(h_attr);
				h_attr = NULL;
			}
222 223 224 225 226
	}

	return h_attr;
}

227
static void create_counter(int counter, int cpu)
228
{
L
Li Zefan 已提交
229
	char *filter = filters[counter];
230
	struct perf_event_attr *attr = attrs + counter;
231 232
	struct perf_header_attr *h_attr;
	int track = !counter; /* only the first counter needs these */
233
	int thread_index;
L
Li Zefan 已提交
234
	int ret;
235 236 237 238 239 240 241 242 243 244
	struct {
		u64 count;
		u64 time_enabled;
		u64 time_running;
		u64 id;
	} read_data;

	attr->read_format	= PERF_FORMAT_TOTAL_TIME_ENABLED |
				  PERF_FORMAT_TOTAL_TIME_RUNNING |
				  PERF_FORMAT_ID;
245

246
	attr->sample_type	|= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
247

248 249 250
	if (nr_counters > 1)
		attr->sample_type |= PERF_SAMPLE_ID;

251 252 253 254 255
	/*
	 * 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 &&
256
				     user_interval != ULLONG_MAX)) {
257 258 259 260 261 262 263
		if (freq) {
			attr->sample_type	|= PERF_SAMPLE_PERIOD;
			attr->freq		= 1;
			attr->sample_freq	= freq;
		} else {
			attr->sample_period = default_interval;
		}
264
	}
265

266 267 268 269 270 271
	if (no_samples)
		attr->sample_freq = 0;

	if (inherit_stat)
		attr->inherit_stat = 1;

272
	if (sample_address) {
273
		attr->sample_type	|= PERF_SAMPLE_ADDR;
274 275
		attr->mmap_data = track;
	}
276

277 278 279
	if (call_graph)
		attr->sample_type	|= PERF_SAMPLE_CALLCHAIN;

A
Arun Sharma 已提交
280 281 282
	if (system_wide)
		attr->sample_type	|= PERF_SAMPLE_CPU;

I
Ingo Molnar 已提交
283
	if (raw_samples) {
284
		attr->sample_type	|= PERF_SAMPLE_TIME;
285
		attr->sample_type	|= PERF_SAMPLE_RAW;
I
Ingo Molnar 已提交
286 287
		attr->sample_type	|= PERF_SAMPLE_CPU;
	}
288

289 290
	attr->mmap		= track;
	attr->comm		= track;
291 292
	attr->inherit		= !no_inherit;
	if (target_pid == -1 && target_tid == -1 && !system_wide) {
293
		attr->disabled = 1;
294
		attr->enable_on_exec = 1;
295
	}
296

297
	for (thread_index = 0; thread_index < thread_num; thread_index++) {
298
try_again:
299 300 301 302 303 304 305 306 307 308
		fd[nr_cpu][counter][thread_index] = sys_perf_event_open(attr,
				all_tids[thread_index], cpu, group_fd, 0);

		if (fd[nr_cpu][counter][thread_index] < 0) {
			int err = errno;

			if (err == EPERM || err == EACCES)
				die("Permission error - are you root?\n"
					"\t Consider tweaking"
					" /proc/sys/kernel/perf_event_paranoid.\n");
309
			else if (err ==  ENODEV && cpu_list) {
310 311 312
				die("No such device - did you specify"
					" an out-of-range profile CPU?\n");
			}
313

314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330
			/*
			 * 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)
					warning(" ... trying to fall back to cpu-clock-ticks\n");
				attr->type = PERF_TYPE_SOFTWARE;
				attr->config = PERF_COUNT_SW_CPU_CLOCK;
				goto try_again;
			}
			printf("\n");
			error("perfcounter syscall returned with %d (%s)\n",
					fd[nr_cpu][counter][thread_index], strerror(err));
331 332

#if defined(__i386__) || defined(__x86_64__)
333 334 335 336 337
			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");
338 339
#endif

340 341 342
			die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
			exit(-1);
		}
343

344 345 346
		h_attr = get_header_attr(attr, counter);
		if (h_attr == NULL)
			die("nomem\n");
347

348 349 350 351 352
		if (!file_new) {
			if (memcmp(&h_attr->attr, attr, sizeof(*attr))) {
				fprintf(stderr, "incompatible append\n");
				exit(-1);
			}
353 354
		}

355 356 357 358
		if (read(fd[nr_cpu][counter][thread_index], &read_data, sizeof(read_data)) == -1) {
			perror("Unable to read perf file descriptor\n");
			exit(-1);
		}
359

360 361 362 363
		if (perf_header_attr__add_id(h_attr, read_data.id) < 0) {
			pr_warning("Not enough memory to add id\n");
			exit(-1);
		}
364

365 366
		assert(fd[nr_cpu][counter][thread_index] >= 0);
		fcntl(fd[nr_cpu][counter][thread_index], F_SETFL, O_NONBLOCK);
367

368 369 370 371 372
		/*
		 * First counter acts as the group leader:
		 */
		if (group && group_fd == -1)
			group_fd = fd[nr_cpu][counter][thread_index];
373

374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
		if (counter || thread_index) {
			ret = ioctl(fd[nr_cpu][counter][thread_index],
					PERF_EVENT_IOC_SET_OUTPUT,
					fd[nr_cpu][0][0]);
			if (ret) {
				error("failed to set output: %d (%s)\n", errno,
						strerror(errno));
				exit(-1);
			}
		} else {
			mmap_array[nr_cpu].counter = counter;
			mmap_array[nr_cpu].prev = 0;
			mmap_array[nr_cpu].mask = mmap_pages*page_size - 1;
			mmap_array[nr_cpu].base = mmap(NULL, (mmap_pages+1)*page_size,
				PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter][thread_index], 0);
			if (mmap_array[nr_cpu].base == MAP_FAILED) {
				error("failed to mmap with %d (%s)\n", errno, strerror(errno));
				exit(-1);
			}

			event_array[nr_poll].fd = fd[nr_cpu][counter][thread_index];
			event_array[nr_poll].events = POLLIN;
			nr_poll++;
397
		}
398

399 400 401 402 403 404 405 406
		if (filter != NULL) {
			ret = ioctl(fd[nr_cpu][counter][thread_index],
					PERF_EVENT_IOC_SET_FILTER, filter);
			if (ret) {
				error("failed to set filter with %d (%s)\n", errno,
						strerror(errno));
				exit(-1);
			}
L
Li Zefan 已提交
407 408
		}
	}
409
}
410

411
static void open_counters(int cpu)
412 413
{
	int counter;
414

415 416
	group_fd = -1;
	for (counter = 0; counter < nr_counters; counter++)
417
		create_counter(counter, cpu);
418

419 420 421
	nr_cpu++;
}

422 423 424 425
static int process_buildids(void)
{
	u64 size = lseek(output, 0, SEEK_CUR);

426 427 428
	if (size == 0)
		return 0;

429 430 431 432 433 434
	session->fd = output;
	return __perf_session__process_events(session, post_processing_offset,
					      size - post_processing_offset,
					      size, &build_id__mark_dso_hit_ops);
}

435 436
static void atexit_header(void)
{
437 438
	if (!pipe_output) {
		session->header.data_size += bytes_written;
439

440 441
		process_buildids();
		perf_header__write(&session->header, output, true);
442
		perf_session__delete(session);
443
		symbol__exit();
444
	}
445 446
}

447
static void event__synthesize_guest_os(struct machine *machine, void *data)
448 449
{
	int err;
450
	struct perf_session *psession = data;
451

452
	if (machine__is_host(machine))
453 454 455 456 457 458 459 460 461 462 463
		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.
	 */
	err = event__synthesize_modules(process_synthesized_event,
464
					psession, machine);
465 466
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
467
		       " relocation symbol.\n", machine->pid);
468 469 470 471 472 473

	/*
	 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
	 * have no _text sometimes.
	 */
	err = event__synthesize_kernel_mmap(process_synthesized_event,
474
					    psession, machine, "_text");
475 476
	if (err < 0)
		err = event__synthesize_kernel_mmap(process_synthesized_event,
477
						    psession, machine, "_stext");
478 479
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
480
		       " relocation symbol.\n", machine->pid);
481 482
}

483 484 485 486 487 488 489
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)
{
490
	int i;
491 492

	for (i = 0; i < nr_cpu; i++) {
493 494
		if (mmap_array[i].base)
			mmap_read(&mmap_array[i]);
495 496 497 498 499 500
	}

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

501
static int __cmd_record(int argc, const char **argv)
502 503
{
	int i, counter;
I
Ingo Molnar 已提交
504 505
	struct stat st;
	int flags;
506
	int err;
507
	unsigned long waking = 0;
508
	int child_ready_pipe[2], go_pipe[2];
509
	const bool forks = argc > 0;
510
	char buf;
511
	struct machine *machine;
512 513 514

	page_size = sysconf(_SC_PAGE_SIZE);

515 516 517 518
	atexit(sig_atexit);
	signal(SIGCHLD, sig_handler);
	signal(SIGINT, sig_handler);

519
	if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
520 521 522 523
		perror("failed to create pipes");
		exit(-1);
	}

524 525 526
	if (!strcmp(output_name, "-"))
		pipe_output = 1;
	else if (!stat(output_name, &st) && st.st_size) {
527
		if (write_mode == WRITE_FORCE) {
528 529 530 531 532
			char oldname[PATH_MAX];
			snprintf(oldname, sizeof(oldname), "%s.old",
				 output_name);
			unlink(oldname);
			rename(output_name, oldname);
533
		}
534 535
	} else if (write_mode == WRITE_APPEND) {
		write_mode = WRITE_FORCE;
536 537
	}

538
	flags = O_CREAT|O_RDWR;
539
	if (write_mode == WRITE_APPEND)
540
		file_new = 0;
I
Ingo Molnar 已提交
541 542 543
	else
		flags |= O_TRUNC;

544 545 546 547
	if (pipe_output)
		output = STDOUT_FILENO;
	else
		output = open(output_name, flags, S_IRUSR | S_IWUSR);
548 549 550 551 552
	if (output < 0) {
		perror("failed to create output file");
		exit(-1);
	}

553
	session = perf_session__new(output_name, O_WRONLY,
T
Tom Zanussi 已提交
554
				    write_mode == WRITE_FORCE, false);
555
	if (session == NULL) {
556 557 558 559
		pr_err("Not enough memory for reading perf file header\n");
		return -1;
	}

560
	if (!file_new) {
561
		err = perf_header__read(session, output);
562
		if (err < 0)
563
			goto out_delete_session;
564 565
	}

566
	if (have_tracepoints(attrs, nr_counters))
567
		perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
568

569 570 571
	/*
 	 * perf_session__delete(session) will be called at atexit_header()
	 */
572 573
	atexit(atexit_header);

574
	if (forks) {
575
		child_pid = fork();
576
		if (child_pid < 0) {
577 578 579
			perror("failed to fork");
			exit(-1);
		}
580

581
		if (!child_pid) {
582 583
			if (pipe_output)
				dup2(2, 1);
584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609
			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]);
			exit(-1);
610
		}
611

612 613 614
		if (!system_wide && target_tid == -1 && target_pid == -1)
			all_tids[0] = child_pid;

615 616 617 618 619 620 621 622 623 624 625 626
		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]);
	}

627 628 629 630 631 632 633 634
	nr_cpus = read_cpu_map(cpu_list);
	if (nr_cpus < 1) {
		perror("failed to collect number of CPUs\n");
		return -1;
	}

	if (!system_wide && no_inherit && !cpu_list) {
		open_counters(-1);
635 636
	} else {
		for (i = 0; i < nr_cpus; i++)
637
			open_counters(cpumap[i]);
638
	}
639

640 641 642 643 644
	if (pipe_output) {
		err = perf_header__write_pipe(output);
		if (err < 0)
			return err;
	} else if (file_new) {
645
		err = perf_header__write(&session->header, output, false);
646 647
		if (err < 0)
			return err;
648 649
	}

650 651
	post_processing_offset = lseek(output, 0, SEEK_CUR);

652 653 654 655 656 657 658 659
	if (pipe_output) {
		err = event__synthesize_attrs(&session->header,
					      process_synthesized_event,
					      session);
		if (err < 0) {
			pr_err("Couldn't synthesize attrs.\n");
			return err;
		}
660 661 662 663 664 665 666

		err = event__synthesize_event_types(process_synthesized_event,
						    session);
		if (err < 0) {
			pr_err("Couldn't synthesize event_types.\n");
			return err;
		}
667

668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684
		if (have_tracepoints(attrs, nr_counters)) {
			/*
			 * 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()
			 */
			err = event__synthesize_tracing_data(output, attrs,
							     nr_counters,
							     process_synthesized_event,
							     session);
			if (err <= 0) {
				pr_err("Couldn't record tracing data.\n");
				return err;
			}
685
			advance_output(err);
686
		}
687 688
	}

689 690
	machine = perf_session__find_host_machine(session);
	if (!machine) {
691 692 693 694
		pr_err("Couldn't find native kernel information.\n");
		return -1;
	}

695
	err = event__synthesize_kernel_mmap(process_synthesized_event,
696
					    session, machine, "_text");
697 698
	if (err < 0)
		err = event__synthesize_kernel_mmap(process_synthesized_event,
699
						    session, machine, "_stext");
700 701 702
	if (err < 0) {
		pr_err("Couldn't record kernel reference relocation symbol.\n");
		return err;
703 704
	}

705
	err = event__synthesize_modules(process_synthesized_event,
706
					session, machine);
707 708 709
	if (err < 0) {
		pr_err("Couldn't record kernel reference relocation symbol.\n");
		return err;
710
	}
711
	if (perf_guest)
712
		perf_session__process_machines(session, event__synthesize_guest_os);
713

714
	if (!system_wide)
715
		event__synthesize_thread(target_tid, process_synthesized_event,
716
					 session);
717
	else
718
		event__synthesize_threads(process_synthesized_event, session);
719

720 721 722 723 724
	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
725
			pr_err("Could not set realtime priority.\n");
726 727 728 729
			exit(-1);
		}
	}

730 731 732
	/*
	 * Let the child rip
	 */
733 734
	if (forks)
		close(go_pipe[1]);
735

736
	for (;;) {
737
		int hits = samples;
738
		int thread;
739

740
		mmap_read_all();
741

742 743 744
		if (hits == samples) {
			if (done)
				break;
745
			err = poll(event_array, nr_poll, -1);
746 747 748 749 750
			waking++;
		}

		if (done) {
			for (i = 0; i < nr_cpu; i++) {
751 752 753 754 755 756 757 758 759
				for (counter = 0;
					counter < nr_counters;
					counter++) {
					for (thread = 0;
						thread < thread_num;
						thread++)
						ioctl(fd[i][counter][thread],
							PERF_EVENT_IOC_DISABLE);
				}
760
			}
761
		}
762 763
	}

764 765
	fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);

766 767 768 769
	/*
	 * Approximate RIP event size: 24 bytes.
	 */
	fprintf(stderr,
770
		"[ perf record: Captured and wrote %.3f MB %s (~%lld samples) ]\n",
771 772 773
		(double)bytes_written / 1024.0 / 1024.0,
		output_name,
		bytes_written / 24);
774

775
	return 0;
776 777 778 779

out_delete_session:
	perf_session__delete(session);
	return err;
780
}
781 782

static const char * const record_usage[] = {
783 784
	"perf record [<options>] [<command>]",
	"perf record [<options>] -- <command> [<options>]",
785 786 787
	NULL
};

788 789
static bool force, append_file;

790
static const struct option options[] = {
791
	OPT_CALLBACK('e', "event", NULL, "event",
792 793
		     "event selector. use 'perf list' to list available events",
		     parse_events),
L
Li Zefan 已提交
794 795
	OPT_CALLBACK(0, "filter", NULL, "filter",
		     "event filter", parse_filter),
796
	OPT_INTEGER('p', "pid", &target_pid,
797 798 799
		    "record events on existing process id"),
	OPT_INTEGER('t', "tid", &target_tid,
		    "record events on existing thread id"),
800 801
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
802 803
	OPT_BOOLEAN('R', "raw-samples", &raw_samples,
		    "collect raw sample records from all opened counters"),
804 805
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
I
Ingo Molnar 已提交
806 807
	OPT_BOOLEAN('A', "append", &append_file,
			    "append to the output file to do incremental profiling"),
808 809
	OPT_STRING('C', "cpu", &cpu_list, "cpu",
		    "list of cpus to monitor"),
810
	OPT_BOOLEAN('f', "force", &force,
811
			"overwrite existing data file (deprecated)"),
812
	OPT_U64('c', "count", &user_interval, "event period to sample"),
I
Ingo Molnar 已提交
813 814
	OPT_STRING('o', "output", &output_name, "file",
		    "output file name"),
815 816
	OPT_BOOLEAN('i', "no-inherit", &no_inherit,
		    "child tasks do not inherit counters"),
817 818
	OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"),
	OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
819 820
	OPT_BOOLEAN('g', "call-graph", &call_graph,
		    "do call-graph (stack chain/backtrace) recording"),
821
	OPT_INCR('v', "verbose", &verbose,
822
		    "be more verbose (show counter open errors, etc)"),
823 824
	OPT_BOOLEAN('s', "stat", &inherit_stat,
		    "per thread counts"),
825 826
	OPT_BOOLEAN('d', "data", &sample_address,
		    "Sample addresses"),
827 828
	OPT_BOOLEAN('n', "no-samples", &no_samples,
		    "don't sample"),
829 830
	OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid,
		    "do not update the buildid cache"),
831 832 833
	OPT_END()
};

834
int cmd_record(int argc, const char **argv, const char *prefix __used)
835
{
836
	int i, j, err = -ENOMEM;
837

838
	argc = parse_options(argc, argv, options, record_usage,
839
			    PARSE_OPT_STOP_AT_NON_OPTION);
840
	if (!argc && target_pid == -1 && target_tid == -1 &&
841
		!system_wide && !cpu_list)
842 843
		usage_with_options(record_usage, options);

844 845 846 847 848 849 850 851 852 853
	if (force && append_file) {
		fprintf(stderr, "Can't overwrite and append at the same time."
				" You need to choose between -f and -A");
		usage_with_options(record_usage, options);
	} else if (append_file) {
		write_mode = WRITE_APPEND;
	} else {
		write_mode = WRITE_FORCE;
	}

854
	symbol__init();
855 856
	if (no_buildid)
		disable_buildid_cache();
857

858 859 860 861 862
	if (!nr_counters) {
		nr_counters	= 1;
		attrs[0].type	= PERF_TYPE_HARDWARE;
		attrs[0].config = PERF_COUNT_HW_CPU_CYCLES;
	}
863

864 865 866 867 868 869 870 871 872 873 874
	if (target_pid != -1) {
		target_tid = target_pid;
		thread_num = find_all_tid(target_pid, &all_tids);
		if (thread_num <= 0) {
			fprintf(stderr, "Can't find all threads of pid %d\n",
					target_pid);
			usage_with_options(record_usage, options);
		}
	} else {
		all_tids=malloc(sizeof(pid_t));
		if (!all_tids)
875
			goto out_symbol_exit;
876 877 878 879 880 881 882 883

		all_tids[0] = target_tid;
		thread_num = 1;
	}

	for (i = 0; i < MAX_NR_CPUS; i++) {
		for (j = 0; j < MAX_COUNTERS; j++) {
			fd[i][j] = malloc(sizeof(int)*thread_num);
884
			if (!fd[i][j])
885
				goto out_free_fd;
886 887 888 889 890
		}
	}
	event_array = malloc(
		sizeof(struct pollfd)*MAX_NR_CPUS*MAX_COUNTERS*thread_num);
	if (!event_array)
891
		goto out_free_fd;
892

893
	if (user_interval != ULLONG_MAX)
894 895 896 897
		default_interval = user_interval;
	if (user_freq != UINT_MAX)
		freq = user_freq;

898 899 900 901 902 903 904 905 906
	/*
	 * 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");
907 908
		err = -EINVAL;
		goto out_free_event_array;
909 910
	}

911 912 913 914 915 916 917 918 919 920 921
	err = __cmd_record(argc, argv);

out_free_event_array:
	free(event_array);
out_free_fd:
	for (i = 0; i < MAX_NR_CPUS; i++) {
		for (j = 0; j < MAX_COUNTERS; j++)
			free(fd[i][j]);
	}
	free(all_tids);
	all_tids = NULL;
922 923
out_symbol_exit:
	symbol__exit();
924
	return err;
925
}