builtin-record.c 22.7 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/tool.h"
26
#include "util/symbol.h"
27
#include "util/cpumap.h"
28
#include "util/thread_map.h"
29

30
#include <unistd.h>
31
#include <sched.h>
32
#include <sys/mman.h>
33

34 35 36 37 38
enum write_mode_t {
	WRITE_FORCE,
	WRITE_APPEND
};

39
struct perf_record {
40
	struct perf_tool	tool;
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57
	struct perf_record_opts	opts;
	u64			bytes_written;
	const char		*output_name;
	struct perf_evlist	*evlist;
	struct perf_session	*session;
	const char		*progname;
	int			output;
	unsigned int		page_size;
	int			realtime_prio;
	enum write_mode_t	write_mode;
	bool			no_buildid;
	bool			no_buildid_cache;
	bool			force;
	bool			file_new;
	bool			append_file;
	long			samples;
	off_t			post_processing_offset;
58
};
59

60
static void advance_output(struct perf_record *rec, size_t size)
61
{
62
	rec->bytes_written += size;
63 64
}

65
static void write_output(struct perf_record *rec, void *buf, size_t size)
66 67
{
	while (size) {
68
		int ret = write(rec->output, buf, size);
69 70 71 72 73 74 75

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

		size -= ret;
		buf += ret;

76
		rec->bytes_written += ret;
77 78 79
	}
}

80
static int process_synthesized_event(struct perf_tool *tool,
81
				     union perf_event *event,
82
				     struct perf_sample *sample __used,
83
				     struct machine *machine __used)
84
{
85
	struct perf_record *rec = container_of(tool, struct perf_record, tool);
86
	write_output(rec, event, event->header.size);
87 88 89
	return 0;
}

90 91
static void perf_record__mmap_read(struct perf_record *rec,
				   struct perf_mmap *md)
92
{
93
	unsigned int head = perf_mmap__read_head(md);
94
	unsigned int old = md->prev;
95
	unsigned char *data = md->base + rec->page_size;
96 97 98
	unsigned long size;
	void *buf;

99 100 101
	if (old == head)
		return;

102
	rec->samples++;
103 104 105 106 107 108 109

	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;
110

111
		write_output(rec, buf, size);
112 113 114 115 116
	}

	buf = &data[old & md->mask];
	size = head - old;
	old += size;
117

118
	write_output(rec, buf, size);
119 120

	md->prev = old;
121
	perf_mmap__write_tail(md, old);
122 123 124
}

static volatile int done = 0;
125
static volatile int signr = -1;
126
static volatile int child_finished = 0;
127

128
static void sig_handler(int sig)
129
{
130 131 132
	if (sig == SIGCHLD)
		child_finished = 1;

133
	done = 1;
134 135 136
	signr = sig;
}

137
static void perf_record__sig_exit(int exit_status __used, void *arg)
138
{
139
	struct perf_record *rec = arg;
140 141
	int status;

142
	if (rec->evlist->workload.pid > 0) {
143
		if (!child_finished)
144
			kill(rec->evlist->workload.pid, SIGTERM);
145 146 147

		wait(&status);
		if (WIFSIGNALED(status))
148
			psignal(WTERMSIG(status), rec->progname);
149
	}
150

151
	if (signr == -1 || signr == SIGUSR1)
152 153 154 155
		return;

	signal(signr, SIG_DFL);
	kill(getpid(), signr);
156 157
}

158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176
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;
}

177
static void perf_record__open(struct perf_record *rec)
178
{
179
	struct perf_evsel *pos, *first;
180 181 182
	struct perf_evlist *evlist = rec->evlist;
	struct perf_session *session = rec->session;
	struct perf_record_opts *opts = &rec->opts;
183

184 185
	first = list_entry(evlist->entries.next, struct perf_evsel, node);

186
	perf_evlist__config_attrs(evlist, opts);
187

188 189
	list_for_each_entry(pos, &evlist->entries, node) {
		struct perf_event_attr *attr = &pos->attr;
190
		struct xyarray *group_fd = NULL;
191 192 193 194 195 196 197 198 199 200 201 202 203
		/*
		 * 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;
204

205
		if (opts->group && pos != first)
206
			group_fd = first->fd;
207
retry_sample_id:
208
		attr->sample_id_all = opts->sample_id_all_avail ? 1 : 0;
209
try_again:
210
		if (perf_evsel__open(pos, evlist->cpus, evlist->threads,
211
				     opts->group, group_fd) < 0) {
212 213
			int err = errno;

214
			if (err == EPERM || err == EACCES) {
215
				ui__error_paranoid();
216
				exit(EXIT_FAILURE);
217
			} else if (err ==  ENODEV && opts->cpu_list) {
218 219
				die("No such device - did you specify"
					" an out-of-range profile CPU?\n");
220
			} else if (err == EINVAL && opts->sample_id_all_avail) {
221 222 223
				/*
				 * Old kernel, no attr->sample_id_type_all field
				 */
224 225
				opts->sample_id_all_avail = false;
				if (!opts->sample_time && !opts->raw_samples && !time_needed)
226 227
					attr->sample_type &= ~PERF_SAMPLE_TIME;

228
				goto retry_sample_id;
229
			}
230

231 232 233 234 235 236 237 238 239
			/*
			 * 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)
240 241
					ui__warning("The cycles event is not supported, "
						    "trying to fall back to cpu-clock-ticks\n");
242 243 244 245
				attr->type = PERF_TYPE_SOFTWARE;
				attr->config = PERF_COUNT_SW_CPU_CLOCK;
				goto try_again;
			}
246 247 248 249 250 251 252

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

253
			printf("\n");
254
			error("sys_perf_event_open() syscall returned with %d (%s).  /bin/dmesg may provide additional information.\n",
255
			      err, strerror(err));
256 257

#if defined(__i386__) || defined(__x86_64__)
258 259 260 261 262
			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");
263 264
#endif

265
			die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
L
Li Zefan 已提交
266 267
		}
	}
268

269 270 271 272 273 274
	if (perf_evlist__set_filters(evlist)) {
		error("failed to set filter with %d (%s)\n", errno,
			strerror(errno));
		exit(-1);
	}

275 276 277 278 279 280 281
	if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) {
		if (errno == EPERM)
			die("Permission error mapping pages.\n"
			    "Consider increasing "
			    "/proc/sys/kernel/perf_event_mlock_kb,\n"
			    "or try again with a smaller value of -m/--mmap_pages.\n"
			    "(current value: %d)\n", opts->mmap_pages);
282
		die("failed to mmap with %d (%s)\n", errno, strerror(errno));
283
	}
284

285
	if (rec->file_new)
286 287 288 289 290 291 292 293 294
		session->evlist = evlist;
	else {
		if (!perf_evlist__equal(session->evlist, evlist)) {
			fprintf(stderr, "incompatible append\n");
			exit(-1);
		}
 	}

	perf_session__update_sample_type(session);
295 296
}

297
static int process_buildids(struct perf_record *rec)
298
{
299
	u64 size = lseek(rec->output, 0, SEEK_CUR);
300

301 302 303
	if (size == 0)
		return 0;

304 305 306
	rec->session->fd = rec->output;
	return __perf_session__process_events(rec->session, rec->post_processing_offset,
					      size - rec->post_processing_offset,
307 308 309
					      size, &build_id__mark_dso_hit_ops);
}

310
static void perf_record__exit(int status __used, void *arg)
311
{
312 313 314 315 316 317 318 319 320 321 322
	struct perf_record *rec = arg;

	if (!rec->opts.pipe_output) {
		rec->session->header.data_size += rec->bytes_written;

		if (!rec->no_buildid)
			process_buildids(rec);
		perf_session__write_header(rec->session, rec->evlist,
					   rec->output, true);
		perf_session__delete(rec->session);
		perf_evlist__delete(rec->evlist);
323
		symbol__exit();
324
	}
325 326
}

327
static void perf_event__synthesize_guest_os(struct machine *machine, void *data)
328 329
{
	int err;
330
	struct perf_tool *tool = data;
331

332
	if (machine__is_host(machine))
333 334 335 336 337 338 339 340 341 342
		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.
	 */
343
	err = perf_event__synthesize_modules(tool, process_synthesized_event,
344
					     machine);
345 346
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
347
		       " relocation symbol.\n", machine->pid);
348 349 350 351 352

	/*
	 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
	 * have no _text sometimes.
	 */
353
	err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event,
354
						 machine, "_text");
355
	if (err < 0)
356
		err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event,
357
							 machine, "_stext");
358 359
	if (err < 0)
		pr_err("Couldn't record guest kernel [%d]'s reference"
360
		       " relocation symbol.\n", machine->pid);
361 362
}

363 364 365 366 367
static struct perf_event_header finished_round_event = {
	.size = sizeof(struct perf_event_header),
	.type = PERF_RECORD_FINISHED_ROUND,
};

368
static void perf_record__mmap_read_all(struct perf_record *rec)
369
{
370
	int i;
371

372 373 374
	for (i = 0; i < rec->evlist->nr_mmaps; i++) {
		if (rec->evlist->mmap[i].base)
			perf_record__mmap_read(rec, &rec->evlist->mmap[i]);
375 376
	}

377 378
	if (perf_header__has_feat(&rec->session->header, HEADER_TRACE_INFO))
		write_output(rec, &finished_round_event, sizeof(finished_round_event));
379 380
}

381
static int __cmd_record(struct perf_record *rec, int argc, const char **argv)
382
{
I
Ingo Molnar 已提交
383 384
	struct stat st;
	int flags;
385
	int err, output;
386
	unsigned long waking = 0;
387
	const bool forks = argc > 0;
388
	struct machine *machine;
389
	struct perf_tool *tool = &rec->tool;
390 391 392 393
	struct perf_record_opts *opts = &rec->opts;
	struct perf_evlist *evsel_list = rec->evlist;
	const char *output_name = rec->output_name;
	struct perf_session *session;
394

395
	rec->progname = argv[0];
396

397
	rec->page_size = sysconf(_SC_PAGE_SIZE);
398

399
	on_exit(perf_record__sig_exit, rec);
400 401
	signal(SIGCHLD, sig_handler);
	signal(SIGINT, sig_handler);
402
	signal(SIGUSR1, sig_handler);
403

404 405
	if (!output_name) {
		if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode))
406
			opts->pipe_output = true;
407
		else
408
			rec->output_name = output_name = "perf.data";
409 410 411
	}
	if (output_name) {
		if (!strcmp(output_name, "-"))
412
			opts->pipe_output = true;
413
		else if (!stat(output_name, &st) && st.st_size) {
414
			if (rec->write_mode == WRITE_FORCE) {
415 416 417 418 419 420
				char oldname[PATH_MAX];
				snprintf(oldname, sizeof(oldname), "%s.old",
					 output_name);
				unlink(oldname);
				rename(output_name, oldname);
			}
421 422
		} else if (rec->write_mode == WRITE_APPEND) {
			rec->write_mode = WRITE_FORCE;
423
		}
424 425
	}

426
	flags = O_CREAT|O_RDWR;
427 428
	if (rec->write_mode == WRITE_APPEND)
		rec->file_new = 0;
I
Ingo Molnar 已提交
429 430 431
	else
		flags |= O_TRUNC;

432
	if (opts->pipe_output)
433 434 435
		output = STDOUT_FILENO;
	else
		output = open(output_name, flags, S_IRUSR | S_IWUSR);
436 437 438 439 440
	if (output < 0) {
		perror("failed to create output file");
		exit(-1);
	}

441 442
	rec->output = output;

443
	session = perf_session__new(output_name, O_WRONLY,
444
				    rec->write_mode == WRITE_FORCE, false, NULL);
445
	if (session == NULL) {
446 447 448 449
		pr_err("Not enough memory for reading perf file header\n");
		return -1;
	}

450 451 452
	rec->session = session;

	if (!rec->no_buildid)
453 454
		perf_header__set_feat(&session->header, HEADER_BUILD_ID);

455
	if (!rec->file_new) {
456
		err = perf_session__read_header(session, output);
457
		if (err < 0)
458
			goto out_delete_session;
459 460
	}

461
	if (have_tracepoints(&evsel_list->entries))
462
		perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
463

464 465 466 467 468 469 470 471 472 473 474 475 476
	perf_header__set_feat(&session->header, HEADER_HOSTNAME);
	perf_header__set_feat(&session->header, HEADER_OSRELEASE);
	perf_header__set_feat(&session->header, HEADER_ARCH);
	perf_header__set_feat(&session->header, HEADER_CPUDESC);
	perf_header__set_feat(&session->header, HEADER_NRCPUS);
	perf_header__set_feat(&session->header, HEADER_EVENT_DESC);
	perf_header__set_feat(&session->header, HEADER_CMDLINE);
	perf_header__set_feat(&session->header, HEADER_VERSION);
	perf_header__set_feat(&session->header, HEADER_CPU_TOPOLOGY);
	perf_header__set_feat(&session->header, HEADER_TOTAL_MEM);
	perf_header__set_feat(&session->header, HEADER_NUMA_TOPOLOGY);
	perf_header__set_feat(&session->header, HEADER_CPUID);

477
	if (forks) {
478
		err = perf_evlist__prepare_workload(evsel_list, opts, argv);
479 480 481
		if (err < 0) {
			pr_err("Couldn't run the workload!\n");
			goto out_delete_session;
482 483 484
		}
	}

485
	perf_record__open(rec);
486

487
	/*
488
	 * perf_session__delete(session) will be called at perf_record__exit()
489
	 */
490
	on_exit(perf_record__exit, rec);
491

492
	if (opts->pipe_output) {
493 494 495
		err = perf_header__write_pipe(output);
		if (err < 0)
			return err;
496
	} else if (rec->file_new) {
497 498
		err = perf_session__write_header(session, evsel_list,
						 output, false);
499 500
		if (err < 0)
			return err;
501 502
	}

503
	rec->post_processing_offset = lseek(output, 0, SEEK_CUR);
504

505 506 507 508 509 510
	machine = perf_session__find_host_machine(session);
	if (!machine) {
		pr_err("Couldn't find native kernel information.\n");
		return -1;
	}

511
	if (opts->pipe_output) {
512
		err = perf_event__synthesize_attrs(tool, session,
513
						   process_synthesized_event);
514 515 516 517
		if (err < 0) {
			pr_err("Couldn't synthesize attrs.\n");
			return err;
		}
518

519
		err = perf_event__synthesize_event_types(tool, process_synthesized_event,
520
							 machine);
521 522 523 524
		if (err < 0) {
			pr_err("Couldn't synthesize event_types.\n");
			return err;
		}
525

526
		if (have_tracepoints(&evsel_list->entries)) {
527 528 529 530 531 532 533 534
			/*
			 * 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()
			 */
535
			err = perf_event__synthesize_tracing_data(tool, output, evsel_list,
536
								  process_synthesized_event);
537 538 539 540
			if (err <= 0) {
				pr_err("Couldn't record tracing data.\n");
				return err;
			}
541
			advance_output(rec, err);
542
		}
543 544
	}

545
	err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event,
546
						 machine, "_text");
547
	if (err < 0)
548
		err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event,
549
							 machine, "_stext");
550 551 552 553
	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");
554

555
	err = perf_event__synthesize_modules(tool, process_synthesized_event,
556
					     machine);
557 558 559 560 561
	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");

562
	if (perf_guest)
563
		perf_session__process_machines(session, tool,
564
					       perf_event__synthesize_guest_os);
565

566
	if (!opts->system_wide)
567
		perf_event__synthesize_thread_map(tool, evsel_list->threads,
568
						  process_synthesized_event,
569
						  machine);
570
	else
571
		perf_event__synthesize_threads(tool, process_synthesized_event,
572
					       machine);
573

574
	if (rec->realtime_prio) {
575 576
		struct sched_param param;

577
		param.sched_priority = rec->realtime_prio;
578
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
579
			pr_err("Could not set realtime priority.\n");
580 581 582 583
			exit(-1);
		}
	}

584 585
	perf_evlist__enable(evsel_list);

586 587 588
	/*
	 * Let the child rip
	 */
589
	if (forks)
590
		perf_evlist__start_workload(evsel_list);
591

592
	for (;;) {
593
		int hits = rec->samples;
594

595
		perf_record__mmap_read_all(rec);
596

597
		if (hits == rec->samples) {
598 599
			if (done)
				break;
600
			err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
601 602 603
			waking++;
		}

604 605
		if (done)
			perf_evlist__disable(evsel_list);
606 607
	}

608
	if (quiet || signr == SIGUSR1)
609 610
		return 0;

611 612
	fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);

613 614 615 616
	/*
	 * Approximate RIP event size: 24 bytes.
	 */
	fprintf(stderr,
617
		"[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n",
618
		(double)rec->bytes_written / 1024.0 / 1024.0,
619
		output_name,
620
		rec->bytes_written / 24);
621

622
	return 0;
623 624 625 626

out_delete_session:
	perf_session__delete(session);
	return err;
627
}
628 629

static const char * const record_usage[] = {
630 631
	"perf record [<options>] [<command>]",
	"perf record [<options>] -- <command> [<options>]",
632 633 634
	NULL
};

635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657
/*
 * XXX Ideally would be local to cmd_record() and passed to a perf_record__new
 * because we need to have access to it in perf_record__exit, that is called
 * after cmd_record() exits, but since record_options need to be accessible to
 * builtin-script, leave it here.
 *
 * At least we don't ouch it in all the other functions here directly.
 *
 * Just say no to tons of global variables, sigh.
 */
static struct perf_record record = {
	.opts = {
		.target_pid	     = -1,
		.target_tid	     = -1,
		.mmap_pages	     = UINT_MAX,
		.user_freq	     = UINT_MAX,
		.user_interval	     = ULLONG_MAX,
		.freq		     = 1000,
		.sample_id_all_avail = true,
	},
	.write_mode = WRITE_FORCE,
	.file_new   = true,
};
658

659 660 661 662 663 664 665
/*
 * XXX Will stay a global variable till we fix builtin-script.c to stop messing
 * with it and switch to use the library functions in perf_evlist that came
 * from builtin-record.c, i.e. use perf_record_opts,
 * perf_evlist__prepare_workload, etc instead of fork+exec'in 'perf record',
 * using pipes, etc.
 */
666
const struct option record_options[] = {
667
	OPT_CALLBACK('e', "event", &record.evlist, "event",
668
		     "event selector. use 'perf list' to list available events",
669
		     parse_events_option),
670
	OPT_CALLBACK(0, "filter", &record.evlist, "filter",
L
Li Zefan 已提交
671
		     "event filter", parse_filter),
672
	OPT_INTEGER('p', "pid", &record.opts.target_pid,
673
		    "record events on existing process id"),
674
	OPT_INTEGER('t', "tid", &record.opts.target_tid,
675
		    "record events on existing thread id"),
676
	OPT_INTEGER('r', "realtime", &record.realtime_prio,
677
		    "collect data with this RT SCHED_FIFO priority"),
678
	OPT_BOOLEAN('D', "no-delay", &record.opts.no_delay,
679
		    "collect data without buffering"),
680
	OPT_BOOLEAN('R', "raw-samples", &record.opts.raw_samples,
681
		    "collect raw sample records from all opened counters"),
682
	OPT_BOOLEAN('a', "all-cpus", &record.opts.system_wide,
683
			    "system-wide collection from all CPUs"),
684
	OPT_BOOLEAN('A', "append", &record.append_file,
I
Ingo Molnar 已提交
685
			    "append to the output file to do incremental profiling"),
686
	OPT_STRING('C', "cpu", &record.opts.cpu_list, "cpu",
687
		    "list of cpus to monitor"),
688
	OPT_BOOLEAN('f', "force", &record.force,
689
			"overwrite existing data file (deprecated)"),
690 691
	OPT_U64('c', "count", &record.opts.user_interval, "event period to sample"),
	OPT_STRING('o', "output", &record.output_name, "file",
I
Ingo Molnar 已提交
692
		    "output file name"),
693
	OPT_BOOLEAN('i', "no-inherit", &record.opts.no_inherit,
694
		    "child tasks do not inherit counters"),
695 696
	OPT_UINTEGER('F', "freq", &record.opts.user_freq, "profile at this frequency"),
	OPT_UINTEGER('m', "mmap-pages", &record.opts.mmap_pages,
697
		     "number of mmap data pages"),
698
	OPT_BOOLEAN(0, "group", &record.opts.group,
699
		    "put the counters into a counter group"),
700
	OPT_BOOLEAN('g', "call-graph", &record.opts.call_graph,
701
		    "do call-graph (stack chain/backtrace) recording"),
702
	OPT_INCR('v', "verbose", &verbose,
703
		    "be more verbose (show counter open errors, etc)"),
704
	OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
705
	OPT_BOOLEAN('s', "stat", &record.opts.inherit_stat,
706
		    "per thread counts"),
707
	OPT_BOOLEAN('d', "data", &record.opts.sample_address,
708
		    "Sample addresses"),
709
	OPT_BOOLEAN('T', "timestamp", &record.opts.sample_time, "Sample timestamps"),
710
	OPT_BOOLEAN('P', "period", &record.opts.period, "Sample period"),
711
	OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples,
712
		    "don't sample"),
713
	OPT_BOOLEAN('N', "no-buildid-cache", &record.no_buildid_cache,
714
		    "do not update the buildid cache"),
715
	OPT_BOOLEAN('B', "no-buildid", &record.no_buildid,
716
		    "do not collect buildids in perf.data"),
717
	OPT_CALLBACK('G', "cgroup", &record.evlist, "name",
S
Stephane Eranian 已提交
718 719
		     "monitor event in cgroup name only",
		     parse_cgroups),
720 721 722
	OPT_END()
};

723
int cmd_record(int argc, const char **argv, const char *prefix __used)
724
{
725 726
	int err = -ENOMEM;
	struct perf_evsel *pos;
727 728
	struct perf_evlist *evsel_list;
	struct perf_record *rec = &record;
729

730 731
	perf_header__set_cmdline(argc, argv);

732
	evsel_list = perf_evlist__new(NULL, NULL);
733 734 735
	if (evsel_list == NULL)
		return -ENOMEM;

736 737
	rec->evlist = evsel_list;

738
	argc = parse_options(argc, argv, record_options, record_usage,
739
			    PARSE_OPT_STOP_AT_NON_OPTION);
740 741
	if (!argc && rec->opts.target_pid == -1 && rec->opts.target_tid == -1 &&
		!rec->opts.system_wide && !rec->opts.cpu_list)
742
		usage_with_options(record_usage, record_options);
743

744
	if (rec->force && rec->append_file) {
745 746
		fprintf(stderr, "Can't overwrite and append at the same time."
				" You need to choose between -f and -A");
747
		usage_with_options(record_usage, record_options);
748 749
	} else if (rec->append_file) {
		rec->write_mode = WRITE_APPEND;
750
	} else {
751
		rec->write_mode = WRITE_FORCE;
752 753
	}

754
	if (nr_cgroups && !rec->opts.system_wide) {
S
Stephane Eranian 已提交
755 756 757 758 759
		fprintf(stderr, "cgroup monitoring only available in"
			" system-wide mode\n");
		usage_with_options(record_usage, record_options);
	}

760
	symbol__init();
761

762
	if (symbol_conf.kptr_restrict)
763 764 765 766 767 768 769 770
		pr_warning(
"WARNING: Kernel address maps (/proc/{kallsyms,modules}) are restricted,\n"
"check /proc/sys/kernel/kptr_restrict.\n\n"
"Samples in kernel functions may not be resolved if a suitable vmlinux\n"
"file is not found in the buildid cache or in the vmlinux path.\n\n"
"Samples in kernel modules won't be resolved at all.\n\n"
"If some relocation was applied (e.g. kexec) symbols may be misresolved\n"
"even with a suitable vmlinux or kallsyms file.\n\n");
771

772
	if (rec->no_buildid_cache || rec->no_buildid)
773
		disable_buildid_cache();
774

775 776
	if (evsel_list->nr_entries == 0 &&
	    perf_evlist__add_default(evsel_list) < 0) {
777 778
		pr_err("Not enough memory for event selector list\n");
		goto out_symbol_exit;
779
	}
780

781 782
	if (rec->opts.target_pid != -1)
		rec->opts.target_tid = rec->opts.target_pid;
783

784 785
	if (perf_evlist__create_maps(evsel_list, rec->opts.target_pid,
				     rec->opts.target_tid, rec->opts.cpu_list) < 0)
786
		usage_with_options(record_usage, record_options);
787

788
	list_for_each_entry(pos, &evsel_list->entries, node) {
789 790
		if (perf_header__push_event(pos->attr.config, event_name(pos)))
			goto out_free_fd;
791
	}
792

793 794 795 796
	if (rec->opts.user_interval != ULLONG_MAX)
		rec->opts.default_interval = rec->opts.user_interval;
	if (rec->opts.user_freq != UINT_MAX)
		rec->opts.freq = rec->opts.user_freq;
797

798 799 800
	/*
	 * User specified count overrides default frequency.
	 */
801 802 803 804
	if (rec->opts.default_interval)
		rec->opts.freq = 0;
	else if (rec->opts.freq) {
		rec->opts.default_interval = rec->opts.freq;
805 806
	} else {
		fprintf(stderr, "frequency and count are zero, aborting\n");
807
		err = -EINVAL;
808
		goto out_free_fd;
809 810
	}

811
	err = __cmd_record(&record, argc, argv);
812
out_free_fd:
813
	perf_evlist__delete_maps(evsel_list);
814 815
out_symbol_exit:
	symbol__exit();
816
	return err;
817
}