builtin-record.c 12.3 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
#include "builtin.h"
9 10 11

#include "perf.h"

12
#include "util/util.h"
13
#include "util/parse-options.h"
14
#include "util/parse-events.h"
15
#include "util/string.h"
16

17
#include <unistd.h>
18 19
#include <sched.h>

20 21
#define ALIGN(x, a)		__ALIGN_MASK(x, (typeof(x))(a)-1)
#define __ALIGN_MASK(x, mask)	(((x)+(mask))&~(mask))
22

23
static int			fd[MAX_NR_CPUS][MAX_COUNTERS];
24 25 26

static long			default_interval		= 100000;

27
static int			nr_cpus				= 0;
28
static unsigned int		page_size;
29
static unsigned int		mmap_pages			= 128;
I
Ingo Molnar 已提交
30
static int			freq				= 0;
31
static int			output;
32
static const char		*output_name			= "perf.data";
33
static int			group				= 0;
34 35
static unsigned int		realtime_prio			= 0;
static int			system_wide			= 0;
36
static pid_t			target_pid			= -1;
37
static int			inherit				= 1;
38
static int			force				= 0;
I
Ingo Molnar 已提交
39
static int			append_file			= 0;
40
static int			verbose				= 0;
41

42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67
static long			samples;
static struct timeval		last_read;
static struct timeval		this_read;

static __u64			bytes_written;

static struct pollfd		event_array[MAX_NR_CPUS * MAX_COUNTERS];

static int			nr_poll;
static int			nr_cpu;

struct mmap_event {
	struct perf_event_header	header;
	__u32				pid;
	__u32				tid;
	__u64				start;
	__u64				len;
	__u64				pgoff;
	char				filename[PATH_MAX];
};

struct comm_event {
	struct perf_event_header	header;
	__u32				pid;
	__u32				tid;
	char				comm[16];
68 69
};

70

71
struct mmap_data {
72 73 74 75
	int			counter;
	void			*base;
	unsigned int		mask;
	unsigned int		prev;
76 77
};

78 79
static struct mmap_data		mmap_array[MAX_NR_CPUS][MAX_COUNTERS];

80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103
static unsigned int mmap_read_head(struct mmap_data *md)
{
	struct perf_counter_mmap_page *pc = md->base;
	int head;

	head = pc->data_head;
	rmb();

	return head;
}

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;

	gettimeofday(&this_read, NULL);

	/*
	 * If we're further behind than half the buffer, there's a chance
104
	 * the writer will bite our tail and mess up the samples under us.
105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
	 *
	 * If we somehow ended up ahead of the head, we got messed up.
	 *
	 * In either case, truncate and restart at head.
	 */
	diff = head - old;
	if (diff > md->mask / 2 || diff < 0) {
		struct timeval iv;
		unsigned long msecs;

		timersub(&this_read, &last_read, &iv);
		msecs = iv.tv_sec*1000 + iv.tv_usec/1000;

		fprintf(stderr, "WARNING: failed to keep up with mmap data."
				"  Last read %lu msecs ago.\n", msecs);

		/*
		 * head points to a known good entry, start there.
		 */
		old = head;
	}

	last_read = this_read;

	if (old != head)
130
		samples++;
131 132 133 134 135 136 137

	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;
138

139 140
		while (size) {
			int ret = write(output, buf, size);
141 142 143 144

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

145 146
			size -= ret;
			buf += ret;
147 148

			bytes_written += ret;
149 150 151 152 153 154
		}
	}

	buf = &data[old & md->mask];
	size = head - old;
	old += size;
155

156 157
	while (size) {
		int ret = write(output, buf, size);
158 159 160 161

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

162 163
		size -= ret;
		buf += ret;
164 165

		bytes_written += ret;
166 167 168 169 170 171 172
	}

	md->prev = old;
}

static volatile int done = 0;

173
static void sig_handler(int sig)
174
{
175
	done = 1;
176 177
}

178
static void pid_synthesize_comm_event(pid_t pid, int full)
179
{
180
	struct comm_event comm_ev;
181 182
	char filename[PATH_MAX];
	char bf[BUFSIZ];
183
	int fd, ret;
184
	size_t size;
185
	char *field, *sep;
186 187
	DIR *tasks;
	struct dirent dirent, *next;
188 189 190 191 192 193 194 195 196 197 198 199 200 201

	snprintf(filename, sizeof(filename), "/proc/%d/stat", pid);

	fd = open(filename, O_RDONLY);
	if (fd < 0) {
		fprintf(stderr, "couldn't open %s\n", filename);
		exit(EXIT_FAILURE);
	}
	if (read(fd, bf, sizeof(bf)) < 0) {
		fprintf(stderr, "couldn't read %s\n", filename);
		exit(EXIT_FAILURE);
	}
	close(fd);

202
	/* 9027 (cat) R 6747 9027 6747 34816 9027 ... */
203
	memset(&comm_ev, 0, sizeof(comm_ev));
204 205 206 207 208 209 210 211
	field = strchr(bf, '(');
	if (field == NULL)
		goto out_failure;
	sep = strchr(++field, ')');
	if (sep == NULL)
		goto out_failure;
	size = sep - field;
	memcpy(comm_ev.comm, field, size++);
212 213

	comm_ev.pid = pid;
214 215 216
	comm_ev.header.type = PERF_EVENT_COMM;
	size = ALIGN(size, sizeof(uint64_t));
	comm_ev.header.size = sizeof(comm_ev) - (sizeof(comm_ev.comm) - size);
217

218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
	if (!full) {
		comm_ev.tid = pid;

		ret = write(output, &comm_ev, comm_ev.header.size);
		if (ret < 0) {
			perror("failed to write");
			exit(-1);
		}
		return;
	}

	snprintf(filename, sizeof(filename), "/proc/%d/task", pid);

	tasks = opendir(filename);
	while (!readdir_r(tasks, &dirent, &next) && next) {
		char *end;
		pid = strtol(dirent.d_name, &end, 10);
		if (*end)
			continue;

		comm_ev.tid = pid;

		ret = write(output, &comm_ev, comm_ev.header.size);
		if (ret < 0) {
			perror("failed to write");
			exit(-1);
		}
245
	}
246 247 248
	closedir(tasks);
	return;

249 250 251 252
out_failure:
	fprintf(stderr, "couldn't get COMM and pgid, malformed %s\n",
		filename);
	exit(EXIT_FAILURE);
253 254
}

255
static void pid_synthesize_mmap_samples(pid_t pid)
256 257 258 259 260 261 262 263 264 265 266 267
{
	char filename[PATH_MAX];
	FILE *fp;

	snprintf(filename, sizeof(filename), "/proc/%d/maps", pid);

	fp = fopen(filename, "r");
	if (fp == NULL) {
		fprintf(stderr, "couldn't open %s\n", filename);
		exit(EXIT_FAILURE);
	}
	while (1) {
268
		char bf[BUFSIZ], *pbf = bf;
269 270 271
		struct mmap_event mmap_ev = {
			.header.type = PERF_EVENT_MMAP,
		};
272
		int n;
273 274 275 276 277
		size_t size;
		if (fgets(bf, sizeof(bf), fp) == NULL)
			break;

		/* 00400000-0040c000 r-xp 00000000 fd:01 41038  /bin/cat */
278 279 280 281 282 283 284 285 286
		n = hex2u64(pbf, &mmap_ev.start);
		if (n < 0)
			continue;
		pbf += n + 1;
		n = hex2u64(pbf, &mmap_ev.len);
		if (n < 0)
			continue;
		pbf += n + 3;
		if (*pbf == 'x') { /* vm_exec */
287 288 289 290 291 292 293 294 295 296 297 298 299
			char *execname = strrchr(bf, ' ');

			if (execname == NULL || execname[1] != '/')
				continue;

			execname += 1;
			size = strlen(execname);
			execname[size - 1] = '\0'; /* Remove \n */
			memcpy(mmap_ev.filename, execname, size);
			size = ALIGN(size, sizeof(uint64_t));
			mmap_ev.len -= mmap_ev.start;
			mmap_ev.header.size = (sizeof(mmap_ev) -
					       (sizeof(mmap_ev.filename) - size));
300
			mmap_ev.pid = pid;
301 302 303 304 305 306 307 308 309 310 311 312
			mmap_ev.tid = pid;

			if (write(output, &mmap_ev, mmap_ev.header.size) < 0) {
				perror("failed to write");
				exit(-1);
			}
		}
	}

	fclose(fp);
}

313
static void synthesize_samples(void)
314 315 316 317 318 319 320 321 322 323 324 325 326 327 328
{
	DIR *proc;
	struct dirent dirent, *next;

	proc = opendir("/proc");

	while (!readdir_r(proc, &dirent, &next) && next) {
		char *end;
		pid_t pid;

		pid = strtol(dirent.d_name, &end, 10);
		if (*end) /* only interested in proper numerical dirents */
			continue;

		pid_synthesize_comm_event(pid, 1);
329
		pid_synthesize_mmap_samples(pid);
330 331 332 333 334
	}

	closedir(proc);
}

335 336 337
static int group_fd;

static void create_counter(int counter, int cpu, pid_t pid)
338
{
339
	struct perf_counter_attr *attr = attrs + counter;
340 341
	int track = 1;

342
	attr->sample_type	= PERF_SAMPLE_IP | PERF_SAMPLE_TID | PERF_SAMPLE_PERIOD;
343
	if (freq) {
344 345
		attr->freq		= 1;
		attr->sample_freq	= freq;
346
	}
347 348 349
	attr->mmap		= track;
	attr->comm		= track;
	attr->inherit		= (cpu < 0) && inherit;
350

351
	track = 0; /* only the first counter needs these */
352

353
try_again:
354
	fd[nr_cpu][counter] = sys_perf_counter_open(attr, pid, cpu, group_fd, 0);
355

356 357
	if (fd[nr_cpu][counter] < 0) {
		int err = errno;
358

359 360
		if (verbose)
			error("sys_perf_counter_open() syscall returned with %d (%s)\n",
361 362
				fd[nr_cpu][counter], strerror(err));
		if (err == EPERM)
363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378
			die("Permission error - are you root?\n");

		/*
		 * 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_CPU_CYCLES) {

			if (verbose)
				warning(" ... trying to fall back to cpu-clock-ticks\n");
			attr->type = PERF_TYPE_SOFTWARE;
			attr->config = PERF_COUNT_CPU_CLOCK;
			goto try_again;
		}
379 380
		exit(-1);
	}
381

382 383
	assert(fd[nr_cpu][counter] >= 0);
	fcntl(fd[nr_cpu][counter], F_SETFL, O_NONBLOCK);
384

385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404
	/*
	 * First counter acts as the group leader:
	 */
	if (group && group_fd == -1)
		group_fd = fd[nr_cpu][counter];

	event_array[nr_poll].fd = fd[nr_cpu][counter];
	event_array[nr_poll].events = POLLIN;
	nr_poll++;

	mmap_array[nr_cpu][counter].counter = counter;
	mmap_array[nr_cpu][counter].prev = 0;
	mmap_array[nr_cpu][counter].mask = mmap_pages*page_size - 1;
	mmap_array[nr_cpu][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
			PROT_READ, MAP_SHARED, fd[nr_cpu][counter], 0);
	if (mmap_array[nr_cpu][counter].base == MAP_FAILED) {
		error("failed to mmap with %d (%s)\n", errno, strerror(errno));
		exit(-1);
	}
}
405

406 407 408
static void open_counters(int cpu, pid_t pid)
{
	int counter;
409

410 411
	if (pid > 0) {
		pid_synthesize_comm_event(pid, 0);
412
		pid_synthesize_mmap_samples(pid);
413
	}
414 415 416 417 418

	group_fd = -1;
	for (counter = 0; counter < nr_counters; counter++)
		create_counter(counter, cpu, pid);

419 420 421
	nr_cpu++;
}

422
static int __cmd_record(int argc, const char **argv)
423 424
{
	int i, counter;
I
Ingo Molnar 已提交
425
	struct stat st;
426
	pid_t pid;
I
Ingo Molnar 已提交
427
	int flags;
428 429 430 431 432 433 434
	int ret;

	page_size = sysconf(_SC_PAGE_SIZE);
	nr_cpus = sysconf(_SC_NPROCESSORS_ONLN);
	assert(nr_cpus <= MAX_NR_CPUS);
	assert(nr_cpus >= 0);

I
Ingo Molnar 已提交
435 436
	if (!stat(output_name, &st) && !force && !append_file) {
		fprintf(stderr, "Error, output file %s exists, use -A to append or -f to overwrite.\n",
437 438 439 440
				output_name);
		exit(-1);
	}

I
Ingo Molnar 已提交
441 442 443 444 445 446 447
	flags = O_CREAT|O_RDWR;
	if (append_file)
		flags |= O_APPEND;
	else
		flags |= O_TRUNC;

	output = open(output_name, flags, S_IRUSR|S_IWUSR);
448 449 450 451 452
	if (output < 0) {
		perror("failed to create output file");
		exit(-1);
	}

453
	if (!system_wide) {
454
		open_counters(-1, target_pid != -1 ? target_pid : getpid());
455 456
	} else for (i = 0; i < nr_cpus; i++)
		open_counters(i, target_pid);
457

458 459
	signal(SIGCHLD, sig_handler);
	signal(SIGINT, sig_handler);
460

461
	if (target_pid == -1 && argc) {
462 463 464
		pid = fork();
		if (pid < 0)
			perror("failed to fork");
465

466
		if (!pid) {
467
			if (execvp(argv[0], (char **)argv)) {
468 469 470
				perror(argv[0]);
				exit(-1);
			}
471 472 473 474 475 476 477 478 479 480 481 482 483
		}
	}

	if (realtime_prio) {
		struct sched_param param;

		param.sched_priority = realtime_prio;
		if (sched_setscheduler(0, SCHED_FIFO, &param)) {
			printf("Could not set realtime priority.\n");
			exit(-1);
		}
	}

484
	if (system_wide)
485
		synthesize_samples();
486 487

	while (!done) {
488
		int hits = samples;
489

490
		for (i = 0; i < nr_cpu; i++) {
491 492 493 494
			for (counter = 0; counter < nr_counters; counter++)
				mmap_read(&mmap_array[i][counter]);
		}

495
		if (hits == samples)
496 497 498
			ret = poll(event_array, nr_poll, 100);
	}

499 500 501 502
	/*
	 * Approximate RIP event size: 24 bytes.
	 */
	fprintf(stderr,
503
		"[ perf record: Captured and wrote %.3f MB %s (~%lld samples) ]\n",
504 505 506
		(double)bytes_written / 1024.0 / 1024.0,
		output_name,
		bytes_written / 24);
507

508 509
	return 0;
}
510 511

static const char * const record_usage[] = {
512 513
	"perf record [<options>] [<command>]",
	"perf record [<options>] -- <command> [<options>]",
514 515 516
	NULL
};

517
static const struct option options[] = {
518
	OPT_CALLBACK('e', "event", NULL, "event",
519 520
		     "event selector. use 'perf list' to list available events",
		     parse_events),
521 522 523 524 525 526
	OPT_INTEGER('p', "pid", &target_pid,
		    "record events on existing pid"),
	OPT_INTEGER('r', "realtime", &realtime_prio,
		    "collect data with this RT SCHED_FIFO priority"),
	OPT_BOOLEAN('a', "all-cpus", &system_wide,
			    "system-wide collection from all CPUs"),
I
Ingo Molnar 已提交
527 528
	OPT_BOOLEAN('A', "append", &append_file,
			    "append to the output file to do incremental profiling"),
529 530
	OPT_BOOLEAN('f', "force", &force,
			"overwrite existing data file"),
531
	OPT_LONG('c', "count", &default_interval,
I
Ingo Molnar 已提交
532 533 534 535 536
		    "event period to sample"),
	OPT_STRING('o', "output", &output_name, "file",
		    "output file name"),
	OPT_BOOLEAN('i', "inherit", &inherit,
		    "child tasks inherit counters"),
I
Ingo Molnar 已提交
537 538
	OPT_INTEGER('F', "freq", &freq,
		    "profile at this frequency"),
I
Ingo Molnar 已提交
539 540
	OPT_INTEGER('m', "mmap-pages", &mmap_pages,
		    "number of mmap data pages"),
541 542
	OPT_BOOLEAN('v', "verbose", &verbose,
		    "be more verbose (show counter open errors, etc)"),
543 544 545 546 547 548 549 550
	OPT_END()
};

int cmd_record(int argc, const char **argv, const char *prefix)
{
	int counter;

	argc = parse_options(argc, argv, options, record_usage, 0);
551
	if (!argc && target_pid == -1 && !system_wide)
552 553
		usage_with_options(record_usage, options);

554
	if (!nr_counters)
555 556 557
		nr_counters = 1;

	for (counter = 0; counter < nr_counters; counter++) {
558
		if (attrs[counter].sample_period)
559 560
			continue;

561
		attrs[counter].sample_period = default_interval;
562 563 564 565
	}

	return __cmd_record(argc, argv);
}