builtin-record.c 22.6 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
#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
34
#define SID(e, x, y) xyarray__entry(e->id, x, y)
35

36 37 38 39 40
enum write_mode_t {
	WRITE_FORCE,
	WRITE_APPEND
};

41 42
static u64			user_interval			= ULLONG_MAX;
static u64			default_interval		=      0;
43
static u64			sample_type;
44

45
static unsigned int		page_size;
46
static unsigned int		mmap_pages			=    128;
47
static unsigned int		user_freq 			= UINT_MAX;
48
static int			freq				=   1000;
49
static int			output;
50
static int			pipe_output			=      0;
51
static const char		*output_name			= NULL;
52
static int			group				=      0;
53
static int			realtime_prio			=      0;
54
static bool			nodelay				=  false;
55
static bool			raw_samples			=  false;
56
static bool			sample_id_all_avail		=   true;
57
static bool			system_wide			=  false;
58
static pid_t			target_pid			=     -1;
59
static pid_t			target_tid			=     -1;
60
static pid_t			child_pid			=     -1;
61
static bool			no_inherit			=  false;
62
static enum write_mode_t	write_mode			= WRITE_FORCE;
63 64 65 66
static bool			call_graph			=  false;
static bool			inherit_stat			=  false;
static bool			no_samples			=  false;
static bool			sample_address			=  false;
67
static bool			sample_time			=  false;
68
static bool			no_buildid			=  false;
69
static bool			no_buildid_cache		=  false;
70
static struct perf_evlist	*evsel_list;
71 72 73

static long			samples				=      0;
static u64			bytes_written			=      0;
74

75
static int			file_new			=      1;
76
static off_t			post_processing_offset;
77

78
static struct perf_session	*session;
79
static const char		*cpu_list;
80

81 82 83 84 85
static void advance_output(size_t size)
{
	bytes_written += size;
}

86 87 88 89 90 91 92 93 94 95 96 97 98 99 100
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;
	}
}

101
static int process_synthesized_event(union perf_event *event,
102
				     struct perf_sample *sample __used,
103
				     struct perf_session *self __used)
104
{
105
	write_output(event, event->header.size);
106 107 108
	return 0;
}

109
static void mmap_read(struct perf_mmap *md)
110
{
111
	unsigned int head = perf_mmap__read_head(md);
112 113 114 115 116
	unsigned int old = md->prev;
	unsigned char *data = md->base + page_size;
	unsigned long size;
	void *buf;

117 118 119 120
	if (old == head)
		return;

	samples++;
121 122 123 124 125 126 127

	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;
128

129
		write_output(buf, size);
130 131 132 133 134
	}

	buf = &data[old & md->mask];
	size = head - old;
	old += size;
135

136
	write_output(buf, size);
137 138

	md->prev = old;
139
	perf_mmap__write_tail(md, old);
140 141 142
}

static volatile int done = 0;
143
static volatile int signr = -1;
144

145
static void sig_handler(int sig)
146
{
147
	done = 1;
148 149 150 151 152
	signr = sig;
}

static void sig_atexit(void)
{
153
	if (child_pid > 0)
154 155
		kill(child_pid, SIGTERM);

156
	if (signr == -1 || signr == SIGUSR1)
157 158 159 160
		return;

	signal(signr, SIG_DFL);
	kill(getpid(), signr);
161 162
}

163
static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr)
164 165 166
{
	struct perf_header_attr *h_attr;

167 168
	if (nr < session->header.attrs) {
		h_attr = session->header.attr[nr];
169 170
	} else {
		h_attr = perf_header_attr__new(a);
171
		if (h_attr != NULL)
172
			if (perf_header__add_attr(&session->header, h_attr) < 0) {
173 174 175
				perf_header_attr__delete(h_attr);
				h_attr = NULL;
			}
176 177 178 179 180
	}

	return h_attr;
}

181
static void create_counter(struct perf_evsel *evsel, int cpu)
182
{
183 184
	char *filter = evsel->filter;
	struct perf_event_attr *attr = &evsel->attr;
185
	struct perf_header_attr *h_attr;
186
	struct perf_sample_id *sid;
187
	int thread_index;
L
Li Zefan 已提交
188
	int ret;
189

190
	for (thread_index = 0; thread_index < evsel_list->threads->nr; thread_index++) {
191 192 193 194 195 196 197 198 199 200 201
		h_attr = get_header_attr(attr, evsel->idx);
		if (h_attr == NULL)
			die("nomem\n");

		if (!file_new) {
			if (memcmp(&h_attr->attr, attr, sizeof(*attr))) {
				fprintf(stderr, "incompatible append\n");
				exit(-1);
			}
		}

202 203
		sid = SID(evsel, cpu, thread_index);
		if (perf_header_attr__add_id(h_attr, sid->id) < 0) {
204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226
			pr_warning("Not enough memory to add id\n");
			exit(-1);
		}

		if (filter != NULL) {
			ret = ioctl(FD(evsel, cpu, thread_index),
				    PERF_EVENT_IOC_SET_FILTER, filter);
			if (ret) {
				error("failed to set filter with %d (%s)\n", errno,
						strerror(errno));
				exit(-1);
			}
		}
	}

	if (!sample_type)
		sample_type = attr->sample_type;
}

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 */
227 228 229 230

	attr->read_format	= PERF_FORMAT_TOTAL_TIME_ENABLED |
				  PERF_FORMAT_TOTAL_TIME_RUNNING |
				  PERF_FORMAT_ID;
231

232
	attr->sample_type	|= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
233

234
	if (evlist->nr_entries > 1)
235 236
		attr->sample_type |= PERF_SAMPLE_ID;

237 238 239 240 241
	/*
	 * 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 &&
242
				     user_interval != ULLONG_MAX)) {
243 244 245 246 247 248 249
		if (freq) {
			attr->sample_type	|= PERF_SAMPLE_PERIOD;
			attr->freq		= 1;
			attr->sample_freq	= freq;
		} else {
			attr->sample_period = default_interval;
		}
250
	}
251

252 253 254 255 256 257
	if (no_samples)
		attr->sample_freq = 0;

	if (inherit_stat)
		attr->inherit_stat = 1;

258
	if (sample_address) {
259
		attr->sample_type	|= PERF_SAMPLE_ADDR;
260 261
		attr->mmap_data = track;
	}
262

263 264 265
	if (call_graph)
		attr->sample_type	|= PERF_SAMPLE_CALLCHAIN;

A
Arun Sharma 已提交
266 267 268
	if (system_wide)
		attr->sample_type	|= PERF_SAMPLE_CPU;

269 270
	if (sample_id_all_avail &&
	    (sample_time || system_wide || !no_inherit || cpu_list))
271 272
		attr->sample_type	|= PERF_SAMPLE_TIME;

I
Ingo Molnar 已提交
273
	if (raw_samples) {
274
		attr->sample_type	|= PERF_SAMPLE_TIME;
275
		attr->sample_type	|= PERF_SAMPLE_RAW;
I
Ingo Molnar 已提交
276 277
		attr->sample_type	|= PERF_SAMPLE_CPU;
	}
278

279 280 281 282 283
	if (nodelay) {
		attr->watermark = 0;
		attr->wakeup_events = 1;
	}

284 285
	attr->mmap		= track;
	attr->comm		= track;
286

287
	if (target_pid == -1 && target_tid == -1 && !system_wide) {
288
		attr->disabled = 1;
289
		attr->enable_on_exec = 1;
290
	}
291
}
292

293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
static void open_counters(struct perf_evlist *evlist)
{
	struct perf_evsel *pos;
	int cpu;

	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;
313

314 315 316 317
		config_attr(pos, evlist);
retry_sample_id:
		attr->sample_id_all = sample_id_all_avail ? 1 : 0;
try_again:
318 319
		if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group,
				     !no_inherit) < 0) {
320 321 322 323 324 325
			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");
326
			else if (err ==  ENODEV && cpu_list) {
327 328
				die("No such device - did you specify"
					" an out-of-range profile CPU?\n");
329 330 331 332 333
			} else if (err == EINVAL && sample_id_all_avail) {
				/*
				 * Old kernel, no attr->sample_id_type_all field
				 */
				sample_id_all_avail = false;
334
				if (!sample_time && !raw_samples && !time_needed)
335 336
					attr->sample_type &= ~PERF_SAMPLE_TIME;

337
				goto retry_sample_id;
338
			}
339

340 341 342 343 344 345 346 347 348 349 350 351 352 353 354
			/*
			 * 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");
355
			error("sys_perf_event_open() syscall returned with %d (%s).  /bin/dmesg may provide additional information.\n",
356
			      err, strerror(err));
357 358

#if defined(__i386__) || defined(__x86_64__)
359 360 361 362 363
			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");
364 365
#endif

366
			die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
L
Li Zefan 已提交
367 368
		}
	}
369

370
	if (perf_evlist__mmap(evlist, mmap_pages, false) < 0)
371 372
		die("failed to mmap with %d (%s)\n", errno, strerror(errno));

373
	for (cpu = 0; cpu < evsel_list->cpus->nr; ++cpu) {
374
		list_for_each_entry(pos, &evlist->entries, node)
375
			create_counter(pos, cpu);
376
	}
377 378
}

379 380 381 382
static int process_buildids(void)
{
	u64 size = lseek(output, 0, SEEK_CUR);

383 384 385
	if (size == 0)
		return 0;

386 387 388 389 390 391
	session->fd = output;
	return __perf_session__process_events(session, post_processing_offset,
					      size - post_processing_offset,
					      size, &build_id__mark_dso_hit_ops);
}

392 393
static void atexit_header(void)
{
394 395
	if (!pipe_output) {
		session->header.data_size += bytes_written;
396

397 398
		if (!no_buildid)
			process_buildids();
399
		perf_header__write(&session->header, evsel_list, output, true);
400
		perf_session__delete(session);
401
		perf_evlist__delete(evsel_list);
402
		symbol__exit();
403
	}
404 405
}

406
static void perf_event__synthesize_guest_os(struct machine *machine, void *data)
407 408
{
	int err;
409
	struct perf_session *psession = data;
410

411
	if (machine__is_host(machine))
412 413 414 415 416 417 418 419 420 421
		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.
	 */
422 423
	err = perf_event__synthesize_modules(process_synthesized_event,
					     psession, machine);
424 425
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
426
		       " relocation symbol.\n", machine->pid);
427 428 429 430 431

	/*
	 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
	 * have no _text sometimes.
	 */
432 433
	err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
						 psession, machine, "_text");
434
	if (err < 0)
435 436 437
		err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
							 psession, machine,
							 "_stext");
438 439
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
440
		       " relocation symbol.\n", machine->pid);
441 442
}

443 444 445 446 447 448 449
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)
{
450
	int i;
451

452
	for (i = 0; i < evsel_list->cpus->nr; i++) {
453 454
		if (evsel_list->mmap[i].base)
			mmap_read(&evsel_list->mmap[i]);
455 456 457 458 459 460
	}

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

461
static int __cmd_record(int argc, const char **argv)
462
{
463
	int i;
I
Ingo Molnar 已提交
464 465
	struct stat st;
	int flags;
466
	int err;
467
	unsigned long waking = 0;
468
	int child_ready_pipe[2], go_pipe[2];
469
	const bool forks = argc > 0;
470
	char buf;
471
	struct machine *machine;
472 473 474

	page_size = sysconf(_SC_PAGE_SIZE);

475 476 477
	atexit(sig_atexit);
	signal(SIGCHLD, sig_handler);
	signal(SIGINT, sig_handler);
478
	signal(SIGUSR1, sig_handler);
479

480
	if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
481 482 483 484
		perror("failed to create pipes");
		exit(-1);
	}

485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503
	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;
504
		}
505 506
	}

507
	flags = O_CREAT|O_RDWR;
508
	if (write_mode == WRITE_APPEND)
509
		file_new = 0;
I
Ingo Molnar 已提交
510 511 512
	else
		flags |= O_TRUNC;

513 514 515 516
	if (pipe_output)
		output = STDOUT_FILENO;
	else
		output = open(output_name, flags, S_IRUSR | S_IWUSR);
517 518 519 520 521
	if (output < 0) {
		perror("failed to create output file");
		exit(-1);
	}

522
	session = perf_session__new(output_name, O_WRONLY,
523
				    write_mode == WRITE_FORCE, false, NULL);
524
	if (session == NULL) {
525 526 527 528
		pr_err("Not enough memory for reading perf file header\n");
		return -1;
	}

529 530 531
	if (!no_buildid)
		perf_header__set_feat(&session->header, HEADER_BUILD_ID);

532
	if (!file_new) {
533
		err = perf_header__read(session, output);
534
		if (err < 0)
535
			goto out_delete_session;
536 537
	}

538
	if (have_tracepoints(&evsel_list->entries))
539
		perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
540

541 542 543
	/*
 	 * perf_session__delete(session) will be called at atexit_header()
	 */
544 545
	atexit(atexit_header);

546
	if (forks) {
547
		child_pid = fork();
548
		if (child_pid < 0) {
549 550 551
			perror("failed to fork");
			exit(-1);
		}
552

553
		if (!child_pid) {
554 555
			if (pipe_output)
				dup2(2, 1);
556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
			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]);
581
			kill(getppid(), SIGUSR1);
582
			exit(-1);
583
		}
584

585
		if (!system_wide && target_tid == -1 && target_pid == -1)
586
			evsel_list->threads->map[0] = child_pid;
587

588 589 590 591 592 593 594 595 596 597 598 599
		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]);
	}

600
	open_counters(evsel_list);
601

602 603
	perf_session__set_sample_type(session, sample_type);

604 605 606 607 608
	if (pipe_output) {
		err = perf_header__write_pipe(output);
		if (err < 0)
			return err;
	} else if (file_new) {
609 610
		err = perf_header__write(&session->header, evsel_list,
					 output, false);
611 612
		if (err < 0)
			return err;
613 614
	}

615 616
	post_processing_offset = lseek(output, 0, SEEK_CUR);

617 618
	perf_session__set_sample_id_all(session, sample_id_all_avail);

619
	if (pipe_output) {
620 621 622
		err = perf_event__synthesize_attrs(&session->header,
						   process_synthesized_event,
						   session);
623 624 625 626
		if (err < 0) {
			pr_err("Couldn't synthesize attrs.\n");
			return err;
		}
627

628 629
		err = perf_event__synthesize_event_types(process_synthesized_event,
							 session);
630 631 632 633
		if (err < 0) {
			pr_err("Couldn't synthesize event_types.\n");
			return err;
		}
634

635
		if (have_tracepoints(&evsel_list->entries)) {
636 637 638 639 640 641 642 643
			/*
			 * 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()
			 */
644 645 646
			err = perf_event__synthesize_tracing_data(output, evsel_list,
								  process_synthesized_event,
								  session);
647 648 649 650
			if (err <= 0) {
				pr_err("Couldn't record tracing data.\n");
				return err;
			}
651
			advance_output(err);
652
		}
653 654
	}

655 656
	machine = perf_session__find_host_machine(session);
	if (!machine) {
657 658 659 660
		pr_err("Couldn't find native kernel information.\n");
		return -1;
	}

661 662
	err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
						 session, machine, "_text");
663
	if (err < 0)
664 665
		err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
							 session, machine, "_stext");
666 667 668 669
	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");
670

671 672
	err = perf_event__synthesize_modules(process_synthesized_event,
					     session, machine);
673 674 675 676 677
	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");

678
	if (perf_guest)
679 680
		perf_session__process_machines(session,
					       perf_event__synthesize_guest_os);
681

682
	if (!system_wide)
683 684 685
		perf_event__synthesize_thread_map(evsel_list->threads,
						  process_synthesized_event,
						  session);
686
	else
687 688
		perf_event__synthesize_threads(process_synthesized_event,
					       session);
689

690 691 692 693 694
	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
695
			pr_err("Could not set realtime priority.\n");
696 697 698 699
			exit(-1);
		}
	}

700 701 702
	/*
	 * Let the child rip
	 */
703 704
	if (forks)
		close(go_pipe[1]);
705

706
	for (;;) {
707
		int hits = samples;
708
		int thread;
709

710
		mmap_read_all();
711

712 713 714
		if (hits == samples) {
			if (done)
				break;
715
			err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
716 717 718 719
			waking++;
		}

		if (done) {
720
			for (i = 0; i < evsel_list->cpus->nr; i++) {
721 722
				struct perf_evsel *pos;

723
				list_for_each_entry(pos, &evsel_list->entries, node) {
724
					for (thread = 0;
725
						thread < evsel_list->threads->nr;
726
						thread++)
727
						ioctl(FD(pos, i, thread),
728 729
							PERF_EVENT_IOC_DISABLE);
				}
730
			}
731
		}
732 733
	}

734
	if (quiet || signr == SIGUSR1)
735 736
		return 0;

737 738
	fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);

739 740 741 742
	/*
	 * Approximate RIP event size: 24 bytes.
	 */
	fprintf(stderr,
743
		"[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n",
744 745 746
		(double)bytes_written / 1024.0 / 1024.0,
		output_name,
		bytes_written / 24);
747

748
	return 0;
749 750 751 752

out_delete_session:
	perf_session__delete(session);
	return err;
753
}
754 755

static const char * const record_usage[] = {
756 757
	"perf record [<options>] [<command>]",
	"perf record [<options>] -- <command> [<options>]",
758 759 760
	NULL
};

761 762
static bool force, append_file;

763
const struct option record_options[] = {
764
	OPT_CALLBACK('e', "event", &evsel_list, "event",
765 766
		     "event selector. use 'perf list' to list available events",
		     parse_events),
767
	OPT_CALLBACK(0, "filter", &evsel_list, "filter",
L
Li Zefan 已提交
768
		     "event filter", parse_filter),
769
	OPT_INTEGER('p', "pid", &target_pid,
770 771 772
		    "record events on existing process id"),
	OPT_INTEGER('t', "tid", &target_tid,
		    "record events on existing thread id"),
773 774
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
775 776
	OPT_BOOLEAN('D', "no-delay", &nodelay,
		    "collect data without buffering"),
777 778
	OPT_BOOLEAN('R', "raw-samples", &raw_samples,
		    "collect raw sample records from all opened counters"),
779 780
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
I
Ingo Molnar 已提交
781 782
	OPT_BOOLEAN('A', "append", &append_file,
			    "append to the output file to do incremental profiling"),
783 784
	OPT_STRING('C', "cpu", &cpu_list, "cpu",
		    "list of cpus to monitor"),
785
	OPT_BOOLEAN('f', "force", &force,
786
			"overwrite existing data file (deprecated)"),
787
	OPT_U64('c', "count", &user_interval, "event period to sample"),
I
Ingo Molnar 已提交
788 789
	OPT_STRING('o', "output", &output_name, "file",
		    "output file name"),
790 791
	OPT_BOOLEAN('i', "no-inherit", &no_inherit,
		    "child tasks do not inherit counters"),
792 793
	OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"),
	OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
794 795
	OPT_BOOLEAN('g', "call-graph", &call_graph,
		    "do call-graph (stack chain/backtrace) recording"),
796
	OPT_INCR('v', "verbose", &verbose,
797
		    "be more verbose (show counter open errors, etc)"),
798
	OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
799 800
	OPT_BOOLEAN('s', "stat", &inherit_stat,
		    "per thread counts"),
801 802
	OPT_BOOLEAN('d', "data", &sample_address,
		    "Sample addresses"),
803
	OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"),
804 805
	OPT_BOOLEAN('n', "no-samples", &no_samples,
		    "don't sample"),
806
	OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache,
807
		    "do not update the buildid cache"),
808 809
	OPT_BOOLEAN('B', "no-buildid", &no_buildid,
		    "do not collect buildids in perf.data"),
S
Stephane Eranian 已提交
810 811 812
	OPT_CALLBACK('G', "cgroup", &evsel_list, "name",
		     "monitor event in cgroup name only",
		     parse_cgroups),
813 814 815
	OPT_END()
};

816
int cmd_record(int argc, const char **argv, const char *prefix __used)
817
{
818 819
	int err = -ENOMEM;
	struct perf_evsel *pos;
820

821
	evsel_list = perf_evlist__new(NULL, NULL);
822 823 824
	if (evsel_list == NULL)
		return -ENOMEM;

825
	argc = parse_options(argc, argv, record_options, record_usage,
826
			    PARSE_OPT_STOP_AT_NON_OPTION);
827
	if (!argc && target_pid == -1 && target_tid == -1 &&
828
		!system_wide && !cpu_list)
829
		usage_with_options(record_usage, record_options);
830

831 832 833
	if (force && append_file) {
		fprintf(stderr, "Can't overwrite and append at the same time."
				" You need to choose between -f and -A");
834
		usage_with_options(record_usage, record_options);
835 836 837 838 839 840
	} else if (append_file) {
		write_mode = WRITE_APPEND;
	} else {
		write_mode = WRITE_FORCE;
	}

S
Stephane Eranian 已提交
841 842 843 844 845 846
	if (nr_cgroups && !system_wide) {
		fprintf(stderr, "cgroup monitoring only available in"
			" system-wide mode\n");
		usage_with_options(record_usage, record_options);
	}

847
	symbol__init();
848 849

	if (no_buildid_cache || no_buildid)
850
		disable_buildid_cache();
851

852 853
	if (evsel_list->nr_entries == 0 &&
	    perf_evlist__add_default(evsel_list) < 0) {
854 855
		pr_err("Not enough memory for event selector list\n");
		goto out_symbol_exit;
856
	}
857

858
	if (target_pid != -1)
859 860
		target_tid = target_pid;

861 862
	if (perf_evlist__create_maps(evsel_list, target_pid,
				     target_tid, cpu_list) < 0)
863
		usage_with_options(record_usage, record_options);
864

865
	list_for_each_entry(pos, &evsel_list->entries, node) {
866 867
		if (perf_evsel__alloc_fd(pos, evsel_list->cpus->nr,
					 evsel_list->threads->nr) < 0)
868
			goto out_free_fd;
869 870
		if (perf_header__push_event(pos->attr.config, event_name(pos)))
			goto out_free_fd;
871
	}
872

873
	if (perf_evlist__alloc_pollfd(evsel_list) < 0)
874
		goto out_free_fd;
875

876
	if (user_interval != ULLONG_MAX)
877 878 879 880
		default_interval = user_interval;
	if (user_freq != UINT_MAX)
		freq = user_freq;

881 882 883 884 885 886 887 888 889
	/*
	 * 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");
890
		err = -EINVAL;
891
		goto out_free_fd;
892 893
	}

894 895
	err = __cmd_record(argc, argv);
out_free_fd:
896
	perf_evlist__delete_maps(evsel_list);
897 898
out_symbol_exit:
	symbol__exit();
899
	return err;
900
}