Merge branch 'tip/perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/roste...
[linux-2.6-block.git] / tools / perf / builtin-record.c
CommitLineData
abaff32a 1/*
bf9e1876
IM
2 * builtin-record.c
3 *
4 * Builtin record command: Record the profile of a workload
5 * (or a CPU, or a PID) into the perf.data output file - for
6 * later analysis via perf report.
abaff32a 7 */
b8f46c5a
XG
8#define _FILE_OFFSET_BITS 64
9
16f762a2 10#include "builtin.h"
bf9e1876
IM
11
12#include "perf.h"
13
6122e4e4 14#include "util/build-id.h"
6eda5838 15#include "util/util.h"
0e9b20b8 16#include "util/parse-options.h"
8ad8db37 17#include "util/parse-events.h"
6eda5838 18
7c6a1c65 19#include "util/header.h"
66e274f3 20#include "util/event.h"
361c99a6 21#include "util/evlist.h"
69aad6f1 22#include "util/evsel.h"
8f28827a 23#include "util/debug.h"
94c744b6 24#include "util/session.h"
8d06367f 25#include "util/symbol.h"
a12b51c4 26#include "util/cpumap.h"
fd78260b 27#include "util/thread_map.h"
7c6a1c65 28
97124d5e 29#include <unistd.h>
de9ac07b 30#include <sched.h>
a41794cd 31#include <sys/mman.h>
de9ac07b 32
69aad6f1
ACM
33#define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y))
34
7865e817
FW
35enum write_mode_t {
36 WRITE_FORCE,
37 WRITE_APPEND
38};
39
3de29cab
SE
40static u64 user_interval = ULLONG_MAX;
41static u64 default_interval = 0;
a21ca2ca 42
de9ac07b 43static unsigned int page_size;
800cd25c 44static unsigned int mmap_pages = UINT_MAX;
f9212819 45static unsigned int user_freq = UINT_MAX;
42e59d7d 46static int freq = 1000;
de9ac07b 47static int output;
529870e3 48static int pipe_output = 0;
d7065adb 49static const char *output_name = NULL;
42e59d7d 50static int group = 0;
1967936d 51static int realtime_prio = 0;
acac03fa 52static bool nodelay = false;
c0555642 53static bool raw_samples = false;
9c90a61c 54static bool sample_id_all_avail = true;
c0555642 55static bool system_wide = false;
42e59d7d 56static pid_t target_pid = -1;
d6d901c2 57static pid_t target_tid = -1;
42e59d7d 58static pid_t child_pid = -1;
2e6cdf99 59static bool no_inherit = false;
7865e817 60static enum write_mode_t write_mode = WRITE_FORCE;
c0555642
IM
61static bool call_graph = false;
62static bool inherit_stat = false;
63static bool no_samples = false;
64static bool sample_address = false;
9c90a61c 65static bool sample_time = false;
a1ac1d3c 66static bool no_buildid = false;
baa2f6ce 67static bool no_buildid_cache = false;
361c99a6 68static struct perf_evlist *evsel_list;
42e59d7d
IM
69
70static long samples = 0;
42e59d7d 71static u64 bytes_written = 0;
a21ca2ca 72
42e59d7d 73static int file_new = 1;
6122e4e4 74static off_t post_processing_offset;
7c6a1c65 75
94c744b6 76static struct perf_session *session;
c45c6ea2 77static const char *cpu_list;
f5970550 78
9215545e
TZ
79static void advance_output(size_t size)
80{
81 bytes_written += size;
82}
83
f5970550
PZ
84static void write_output(void *buf, size_t size)
85{
86 while (size) {
87 int ret = write(output, buf, size);
88
89 if (ret < 0)
90 die("failed to write");
91
92 size -= ret;
93 buf += ret;
94
95 bytes_written += ret;
96 }
97}
98
8115d60c 99static int process_synthesized_event(union perf_event *event,
8d50e5b4 100 struct perf_sample *sample __used,
d8f66248 101 struct perf_session *self __used)
234fbbf5 102{
6122e4e4 103 write_output(event, event->header.size);
234fbbf5
ACM
104 return 0;
105}
106
744bd8aa 107static void mmap_read(struct perf_mmap *md)
de9ac07b 108{
744bd8aa 109 unsigned int head = perf_mmap__read_head(md);
de9ac07b
PZ
110 unsigned int old = md->prev;
111 unsigned char *data = md->base + page_size;
112 unsigned long size;
113 void *buf;
de9ac07b 114
dc82009a
ACM
115 if (old == head)
116 return;
117
118 samples++;
de9ac07b
PZ
119
120 size = head - old;
121
122 if ((old & md->mask) + size != (head & md->mask)) {
123 buf = &data[old & md->mask];
124 size = md->mask + 1 - (old & md->mask);
125 old += size;
021e9f47 126
6122e4e4 127 write_output(buf, size);
de9ac07b
PZ
128 }
129
130 buf = &data[old & md->mask];
131 size = head - old;
132 old += size;
021e9f47 133
6122e4e4 134 write_output(buf, size);
de9ac07b
PZ
135
136 md->prev = old;
115d2d89 137 perf_mmap__write_tail(md, old);
de9ac07b
PZ
138}
139
140static volatile int done = 0;
f7b7c26e 141static volatile int signr = -1;
de9ac07b 142
16c8a109 143static void sig_handler(int sig)
de9ac07b 144{
16c8a109 145 done = 1;
f7b7c26e
PZ
146 signr = sig;
147}
148
149static void sig_atexit(void)
150{
5ffc8881 151 if (child_pid > 0)
933da83a
CW
152 kill(child_pid, SIGTERM);
153
18483b81 154 if (signr == -1 || signr == SIGUSR1)
f7b7c26e
PZ
155 return;
156
157 signal(signr, SIG_DFL);
158 kill(getpid(), signr);
de9ac07b
PZ
159}
160
dd7927f4
ACM
161static void config_attr(struct perf_evsel *evsel, struct perf_evlist *evlist)
162{
163 struct perf_event_attr *attr = &evsel->attr;
164 int track = !evsel->idx; /* only the first counter needs these */
7c6a1c65 165
5d2cd909 166 attr->inherit = !no_inherit;
7c6a1c65
PZ
167 attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
168 PERF_FORMAT_TOTAL_TIME_RUNNING |
169 PERF_FORMAT_ID;
16c8a109 170
3a9f131f 171 attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
3efa1cc9 172
361c99a6 173 if (evlist->nr_entries > 1)
8907fd60
EM
174 attr->sample_type |= PERF_SAMPLE_ID;
175
f9212819
FW
176 /*
177 * We default some events to a 1 default interval. But keep
178 * it a weak assumption overridable by the user.
179 */
180 if (!attr->sample_period || (user_freq != UINT_MAX &&
3de29cab 181 user_interval != ULLONG_MAX)) {
f9212819
FW
182 if (freq) {
183 attr->sample_type |= PERF_SAMPLE_PERIOD;
184 attr->freq = 1;
185 attr->sample_freq = freq;
186 } else {
187 attr->sample_period = default_interval;
188 }
1dba15e7 189 }
3efa1cc9 190
649c48a9
PZ
191 if (no_samples)
192 attr->sample_freq = 0;
193
194 if (inherit_stat)
195 attr->inherit_stat = 1;
196
3af9e859 197 if (sample_address) {
4bba828d 198 attr->sample_type |= PERF_SAMPLE_ADDR;
3af9e859
EM
199 attr->mmap_data = track;
200 }
4bba828d 201
3efa1cc9
IM
202 if (call_graph)
203 attr->sample_type |= PERF_SAMPLE_CALLCHAIN;
204
f60f3593
AS
205 if (system_wide)
206 attr->sample_type |= PERF_SAMPLE_CPU;
207
a43d3f08
ACM
208 if (sample_id_all_avail &&
209 (sample_time || system_wide || !no_inherit || cpu_list))
9c90a61c
ACM
210 attr->sample_type |= PERF_SAMPLE_TIME;
211
cd6feeea 212 if (raw_samples) {
6ddf259d 213 attr->sample_type |= PERF_SAMPLE_TIME;
daac07b2 214 attr->sample_type |= PERF_SAMPLE_RAW;
cd6feeea
IM
215 attr->sample_type |= PERF_SAMPLE_CPU;
216 }
f413cdb8 217
acac03fa
KS
218 if (nodelay) {
219 attr->watermark = 0;
220 attr->wakeup_events = 1;
221 }
222
a21ca2ca
IM
223 attr->mmap = track;
224 attr->comm = track;
dd7927f4 225
2e6cdf99 226 if (target_pid == -1 && target_tid == -1 && !system_wide) {
46be604b 227 attr->disabled = 1;
bedbfdea 228 attr->enable_on_exec = 1;
46be604b 229 }
dd7927f4 230}
bedbfdea 231
a91e5431
ACM
232static bool perf_evlist__equal(struct perf_evlist *evlist,
233 struct perf_evlist *other)
234{
235 struct perf_evsel *pos, *pair;
236
237 if (evlist->nr_entries != other->nr_entries)
238 return false;
239
240 pair = list_entry(other->entries.next, struct perf_evsel, node);
241
242 list_for_each_entry(pos, &evlist->entries, node) {
243 if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0))
244 return false;
245 pair = list_entry(pair->node.next, struct perf_evsel, node);
246 }
247
248 return true;
249}
250
dd7927f4
ACM
251static void open_counters(struct perf_evlist *evlist)
252{
253 struct perf_evsel *pos;
dd7927f4 254
5d2cd909
ACM
255 if (evlist->cpus->map[0] < 0)
256 no_inherit = true;
257
dd7927f4
ACM
258 list_for_each_entry(pos, &evlist->entries, node) {
259 struct perf_event_attr *attr = &pos->attr;
260 /*
261 * Check if parse_single_tracepoint_event has already asked for
262 * PERF_SAMPLE_TIME.
263 *
264 * XXX this is kludgy but short term fix for problems introduced by
265 * eac23d1c that broke 'perf script' by having different sample_types
266 * when using multiple tracepoint events when we use a perf binary
267 * that tries to use sample_id_all on an older kernel.
268 *
269 * We need to move counter creation to perf_session, support
270 * different sample_types, etc.
271 */
272 bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
d6d901c2 273
dd7927f4
ACM
274 config_attr(pos, evlist);
275retry_sample_id:
276 attr->sample_id_all = sample_id_all_avail ? 1 : 0;
277try_again:
5d2cd909 278 if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group) < 0) {
d6d901c2
ZY
279 int err = errno;
280
c286c419
ACM
281 if (err == EPERM || err == EACCES) {
282 ui__warning_paranoid();
283 exit(EXIT_FAILURE);
284 } else if (err == ENODEV && cpu_list) {
d6d901c2
ZY
285 die("No such device - did you specify"
286 " an out-of-range profile CPU?\n");
9c90a61c
ACM
287 } else if (err == EINVAL && sample_id_all_avail) {
288 /*
289 * Old kernel, no attr->sample_id_type_all field
290 */
291 sample_id_all_avail = false;
a43d3f08 292 if (!sample_time && !raw_samples && !time_needed)
eac23d1c
IM
293 attr->sample_type &= ~PERF_SAMPLE_TIME;
294
9c90a61c 295 goto retry_sample_id;
d6d901c2 296 }
3da297a6 297
d6d901c2
ZY
298 /*
299 * If it's cycles then fall back to hrtimer
300 * based cpu-clock-tick sw counter, which
301 * is always available even if no PMU support:
302 */
303 if (attr->type == PERF_TYPE_HARDWARE
304 && attr->config == PERF_COUNT_HW_CPU_CYCLES) {
305
306 if (verbose)
ca6a4258
DA
307 ui__warning("The cycles event is not supported, "
308 "trying to fall back to cpu-clock-ticks\n");
d6d901c2
ZY
309 attr->type = PERF_TYPE_SOFTWARE;
310 attr->config = PERF_COUNT_SW_CPU_CLOCK;
311 goto try_again;
312 }
ca6a4258
DA
313
314 if (err == ENOENT) {
315 ui__warning("The %s event is not supported.\n",
316 event_name(pos));
317 exit(EXIT_FAILURE);
318 }
319
d6d901c2 320 printf("\n");
d9cf837e 321 error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n",
dd7927f4 322 err, strerror(err));
bfd45118
SK
323
324#if defined(__i386__) || defined(__x86_64__)
d6d901c2
ZY
325 if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
326 die("No hardware sampling interrupt available."
327 " No APIC? If so then you can boot the kernel"
328 " with the \"lapic\" boot parameter to"
329 " force-enable it.\n");
bfd45118
SK
330#endif
331
d6d901c2 332 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
c171b552
LZ
333 }
334 }
a43d3f08 335
0a102479
FW
336 if (perf_evlist__set_filters(evlist)) {
337 error("failed to set filter with %d (%s)\n", errno,
338 strerror(errno));
339 exit(-1);
340 }
341
7e2ed097 342 if (perf_evlist__mmap(evlist, mmap_pages, false) < 0)
0a27d7f9
ACM
343 die("failed to mmap with %d (%s)\n", errno, strerror(errno));
344
a91e5431
ACM
345 if (file_new)
346 session->evlist = evlist;
347 else {
348 if (!perf_evlist__equal(session->evlist, evlist)) {
349 fprintf(stderr, "incompatible append\n");
350 exit(-1);
351 }
352 }
353
354 perf_session__update_sample_type(session);
16c8a109
PZ
355}
356
6122e4e4
ACM
357static int process_buildids(void)
358{
359 u64 size = lseek(output, 0, SEEK_CUR);
360
9f591fd7
ACM
361 if (size == 0)
362 return 0;
363
6122e4e4
ACM
364 session->fd = output;
365 return __perf_session__process_events(session, post_processing_offset,
366 size - post_processing_offset,
367 size, &build_id__mark_dso_hit_ops);
368}
369
f5970550
PZ
370static void atexit_header(void)
371{
c7929e47
TZ
372 if (!pipe_output) {
373 session->header.data_size += bytes_written;
f5970550 374
baa2f6ce
ACM
375 if (!no_buildid)
376 process_buildids();
a91e5431 377 perf_session__write_header(session, evsel_list, output, true);
39d17dac 378 perf_session__delete(session);
361c99a6 379 perf_evlist__delete(evsel_list);
d65a458b 380 symbol__exit();
c7929e47 381 }
f5970550
PZ
382}
383
8115d60c 384static void perf_event__synthesize_guest_os(struct machine *machine, void *data)
a1645ce1
ZY
385{
386 int err;
23346f21 387 struct perf_session *psession = data;
a1645ce1 388
23346f21 389 if (machine__is_host(machine))
a1645ce1
ZY
390 return;
391
392 /*
393 *As for guest kernel when processing subcommand record&report,
394 *we arrange module mmap prior to guest kernel mmap and trigger
395 *a preload dso because default guest module symbols are loaded
396 *from guest kallsyms instead of /lib/modules/XXX/XXX. This
397 *method is used to avoid symbol missing when the first addr is
398 *in module instead of in guest kernel.
399 */
8115d60c
ACM
400 err = perf_event__synthesize_modules(process_synthesized_event,
401 psession, machine);
a1645ce1
ZY
402 if (err < 0)
403 pr_err("Couldn't record guest kernel [%d]'s reference"
23346f21 404 " relocation symbol.\n", machine->pid);
a1645ce1 405
a1645ce1
ZY
406 /*
407 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
408 * have no _text sometimes.
409 */
8115d60c
ACM
410 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
411 psession, machine, "_text");
a1645ce1 412 if (err < 0)
8115d60c
ACM
413 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
414 psession, machine,
415 "_stext");
a1645ce1
ZY
416 if (err < 0)
417 pr_err("Couldn't record guest kernel [%d]'s reference"
23346f21 418 " relocation symbol.\n", machine->pid);
a1645ce1
ZY
419}
420
98402807
FW
421static struct perf_event_header finished_round_event = {
422 .size = sizeof(struct perf_event_header),
423 .type = PERF_RECORD_FINISHED_ROUND,
424};
425
426static void mmap_read_all(void)
427{
0e2e63dd 428 int i;
98402807 429
aece948f 430 for (i = 0; i < evsel_list->nr_mmaps; i++) {
0a27d7f9
ACM
431 if (evsel_list->mmap[i].base)
432 mmap_read(&evsel_list->mmap[i]);
98402807
FW
433 }
434
435 if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO))
436 write_output(&finished_round_event, sizeof(finished_round_event));
437}
438
d4db3f16 439static int __cmd_record(int argc, const char **argv)
16c8a109 440{
69aad6f1 441 int i;
abaff32a 442 struct stat st;
abaff32a 443 int flags;
4dc0a04b 444 int err;
8b412664 445 unsigned long waking = 0;
856e9660 446 int child_ready_pipe[2], go_pipe[2];
46be604b 447 const bool forks = argc > 0;
856e9660 448 char buf;
23346f21 449 struct machine *machine;
de9ac07b
PZ
450
451 page_size = sysconf(_SC_PAGE_SIZE);
de9ac07b 452
f5970550
PZ
453 atexit(sig_atexit);
454 signal(SIGCHLD, sig_handler);
455 signal(SIGINT, sig_handler);
18483b81 456 signal(SIGUSR1, sig_handler);
f5970550 457
d4db3f16 458 if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
856e9660
PZ
459 perror("failed to create pipes");
460 exit(-1);
461 }
462
d7065adb
FBH
463 if (!output_name) {
464 if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode))
465 pipe_output = 1;
466 else
467 output_name = "perf.data";
468 }
469 if (output_name) {
470 if (!strcmp(output_name, "-"))
471 pipe_output = 1;
472 else if (!stat(output_name, &st) && st.st_size) {
473 if (write_mode == WRITE_FORCE) {
474 char oldname[PATH_MAX];
475 snprintf(oldname, sizeof(oldname), "%s.old",
476 output_name);
477 unlink(oldname);
478 rename(output_name, oldname);
479 }
480 } else if (write_mode == WRITE_APPEND) {
481 write_mode = WRITE_FORCE;
266e0e21 482 }
97124d5e
PZ
483 }
484
f887f301 485 flags = O_CREAT|O_RDWR;
7865e817 486 if (write_mode == WRITE_APPEND)
f5970550 487 file_new = 0;
abaff32a
IM
488 else
489 flags |= O_TRUNC;
490
529870e3
TZ
491 if (pipe_output)
492 output = STDOUT_FILENO;
493 else
494 output = open(output_name, flags, S_IRUSR | S_IWUSR);
de9ac07b
PZ
495 if (output < 0) {
496 perror("failed to create output file");
497 exit(-1);
498 }
499
7865e817 500 session = perf_session__new(output_name, O_WRONLY,
21ef97f0 501 write_mode == WRITE_FORCE, false, NULL);
94c744b6 502 if (session == NULL) {
a9a70bbc
ACM
503 pr_err("Not enough memory for reading perf file header\n");
504 return -1;
505 }
506
baa2f6ce
ACM
507 if (!no_buildid)
508 perf_header__set_feat(&session->header, HEADER_BUILD_ID);
509
4dc0a04b 510 if (!file_new) {
a91e5431 511 err = perf_session__read_header(session, output);
4dc0a04b 512 if (err < 0)
39d17dac 513 goto out_delete_session;
4dc0a04b
ACM
514 }
515
361c99a6 516 if (have_tracepoints(&evsel_list->entries))
94c744b6 517 perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
03456a15 518
800cd25c
FW
519 /* 512 kiB: default amount of unprivileged mlocked memory */
520 if (mmap_pages == UINT_MAX)
521 mmap_pages = (512 * 1024) / page_size;
522
d4db3f16 523 if (forks) {
46be604b 524 child_pid = fork();
2fb750e8 525 if (child_pid < 0) {
856e9660
PZ
526 perror("failed to fork");
527 exit(-1);
528 }
7c6a1c65 529
46be604b 530 if (!child_pid) {
529870e3
TZ
531 if (pipe_output)
532 dup2(2, 1);
856e9660
PZ
533 close(child_ready_pipe[0]);
534 close(go_pipe[1]);
535 fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);
536
537 /*
538 * Do a dummy execvp to get the PLT entry resolved,
539 * so we avoid the resolver overhead on the real
540 * execvp call.
541 */
542 execvp("", (char **)argv);
543
544 /*
545 * Tell the parent we're ready to go
546 */
547 close(child_ready_pipe[1]);
548
549 /*
550 * Wait until the parent tells us to go.
551 */
552 if (read(go_pipe[0], &buf, 1) == -1)
553 perror("unable to read pipe");
554
555 execvp(argv[0], (char **)argv);
556
557 perror(argv[0]);
18483b81 558 kill(getppid(), SIGUSR1);
856e9660 559 exit(-1);
0a5ac846 560 }
856e9660 561
d6d901c2 562 if (!system_wide && target_tid == -1 && target_pid == -1)
7e2ed097 563 evsel_list->threads->map[0] = child_pid;
d6d901c2 564
856e9660
PZ
565 close(child_ready_pipe[1]);
566 close(go_pipe[0]);
567 /*
568 * wait for child to settle
569 */
570 if (read(child_ready_pipe[0], &buf, 1) == -1) {
571 perror("unable to read pipe");
572 exit(-1);
573 }
574 close(child_ready_pipe[0]);
575 }
576
dd7927f4 577 open_counters(evsel_list);
de9ac07b 578
712a4b60
ACM
579 /*
580 * perf_session__delete(session) will be called at atexit_header()
581 */
582 atexit(atexit_header);
583
529870e3
TZ
584 if (pipe_output) {
585 err = perf_header__write_pipe(output);
586 if (err < 0)
587 return err;
588 } else if (file_new) {
a91e5431
ACM
589 err = perf_session__write_header(session, evsel_list,
590 output, false);
d5eed904
ACM
591 if (err < 0)
592 return err;
56b03f3c
ACM
593 }
594
6122e4e4
ACM
595 post_processing_offset = lseek(output, 0, SEEK_CUR);
596
2c46dbb5 597 if (pipe_output) {
a91e5431
ACM
598 err = perf_session__synthesize_attrs(session,
599 process_synthesized_event);
2c46dbb5
TZ
600 if (err < 0) {
601 pr_err("Couldn't synthesize attrs.\n");
602 return err;
603 }
cd19a035 604
8115d60c
ACM
605 err = perf_event__synthesize_event_types(process_synthesized_event,
606 session);
cd19a035
TZ
607 if (err < 0) {
608 pr_err("Couldn't synthesize event_types.\n");
609 return err;
610 }
9215545e 611
361c99a6 612 if (have_tracepoints(&evsel_list->entries)) {
63e0c771
TZ
613 /*
614 * FIXME err <= 0 here actually means that
615 * there were no tracepoints so its not really
616 * an error, just that we don't need to
617 * synthesize anything. We really have to
618 * return this more properly and also
619 * propagate errors that now are calling die()
620 */
8115d60c
ACM
621 err = perf_event__synthesize_tracing_data(output, evsel_list,
622 process_synthesized_event,
623 session);
63e0c771
TZ
624 if (err <= 0) {
625 pr_err("Couldn't record tracing data.\n");
626 return err;
627 }
2c9faa06 628 advance_output(err);
63e0c771 629 }
2c46dbb5
TZ
630 }
631
23346f21
ACM
632 machine = perf_session__find_host_machine(session);
633 if (!machine) {
a1645ce1
ZY
634 pr_err("Couldn't find native kernel information.\n");
635 return -1;
636 }
637
8115d60c
ACM
638 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
639 session, machine, "_text");
70162138 640 if (err < 0)
8115d60c
ACM
641 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
642 session, machine, "_stext");
c1a3a4b9
ACM
643 if (err < 0)
644 pr_err("Couldn't record kernel reference relocation symbol\n"
645 "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
646 "Check /proc/kallsyms permission or run as root.\n");
b7cece76 647
8115d60c
ACM
648 err = perf_event__synthesize_modules(process_synthesized_event,
649 session, machine);
c1a3a4b9
ACM
650 if (err < 0)
651 pr_err("Couldn't record kernel module information.\n"
652 "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
653 "Check /proc/modules permission or run as root.\n");
654
a1645ce1 655 if (perf_guest)
8115d60c
ACM
656 perf_session__process_machines(session,
657 perf_event__synthesize_guest_os);
7c6a1c65 658
cf103a14 659 if (!system_wide)
7c940c18
ACM
660 perf_event__synthesize_thread_map(evsel_list->threads,
661 process_synthesized_event,
662 session);
234fbbf5 663 else
8115d60c
ACM
664 perf_event__synthesize_threads(process_synthesized_event,
665 session);
7c6a1c65 666
de9ac07b
PZ
667 if (realtime_prio) {
668 struct sched_param param;
669
670 param.sched_priority = realtime_prio;
671 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
6beba7ad 672 pr_err("Could not set realtime priority.\n");
de9ac07b
PZ
673 exit(-1);
674 }
675 }
676
856e9660
PZ
677 /*
678 * Let the child rip
679 */
d4db3f16
ACM
680 if (forks)
681 close(go_pipe[1]);
856e9660 682
649c48a9 683 for (;;) {
2debbc83 684 int hits = samples;
d6d901c2 685 int thread;
de9ac07b 686
98402807 687 mmap_read_all();
de9ac07b 688
649c48a9
PZ
689 if (hits == samples) {
690 if (done)
691 break;
5c581041 692 err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
8b412664
PZ
693 waking++;
694 }
695
696 if (done) {
7e2ed097 697 for (i = 0; i < evsel_list->cpus->nr; i++) {
69aad6f1
ACM
698 struct perf_evsel *pos;
699
361c99a6 700 list_for_each_entry(pos, &evsel_list->entries, node) {
d6d901c2 701 for (thread = 0;
7e2ed097 702 thread < evsel_list->threads->nr;
d6d901c2 703 thread++)
69aad6f1 704 ioctl(FD(pos, i, thread),
d6d901c2
ZY
705 PERF_EVENT_IOC_DISABLE);
706 }
8b412664 707 }
649c48a9 708 }
de9ac07b
PZ
709 }
710
18483b81 711 if (quiet || signr == SIGUSR1)
b44308f5
ACM
712 return 0;
713
8b412664
PZ
714 fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);
715
021e9f47
IM
716 /*
717 * Approximate RIP event size: 24 bytes.
718 */
719 fprintf(stderr,
9486aa38 720 "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n",
021e9f47
IM
721 (double)bytes_written / 1024.0 / 1024.0,
722 output_name,
723 bytes_written / 24);
addc2785 724
de9ac07b 725 return 0;
39d17dac
ACM
726
727out_delete_session:
728 perf_session__delete(session);
729 return err;
de9ac07b 730}
0e9b20b8 731
0e9b20b8 732static const char * const record_usage[] = {
9e096753
MG
733 "perf record [<options>] [<command>]",
734 "perf record [<options>] -- <command> [<options>]",
0e9b20b8
IM
735 NULL
736};
737
7865e817
FW
738static bool force, append_file;
739
bca647aa 740const struct option record_options[] = {
361c99a6 741 OPT_CALLBACK('e', "event", &evsel_list, "event",
86847b62
TG
742 "event selector. use 'perf list' to list available events",
743 parse_events),
361c99a6 744 OPT_CALLBACK(0, "filter", &evsel_list, "filter",
c171b552 745 "event filter", parse_filter),
0e9b20b8 746 OPT_INTEGER('p', "pid", &target_pid,
d6d901c2
ZY
747 "record events on existing process id"),
748 OPT_INTEGER('t', "tid", &target_tid,
749 "record events on existing thread id"),
0e9b20b8
IM
750 OPT_INTEGER('r', "realtime", &realtime_prio,
751 "collect data with this RT SCHED_FIFO priority"),
acac03fa
KS
752 OPT_BOOLEAN('D', "no-delay", &nodelay,
753 "collect data without buffering"),
daac07b2
FW
754 OPT_BOOLEAN('R', "raw-samples", &raw_samples,
755 "collect raw sample records from all opened counters"),
0e9b20b8
IM
756 OPT_BOOLEAN('a', "all-cpus", &system_wide,
757 "system-wide collection from all CPUs"),
abaff32a
IM
758 OPT_BOOLEAN('A', "append", &append_file,
759 "append to the output file to do incremental profiling"),
c45c6ea2
SE
760 OPT_STRING('C', "cpu", &cpu_list, "cpu",
761 "list of cpus to monitor"),
97124d5e 762 OPT_BOOLEAN('f', "force", &force,
7865e817 763 "overwrite existing data file (deprecated)"),
3de29cab 764 OPT_U64('c', "count", &user_interval, "event period to sample"),
abaff32a
IM
765 OPT_STRING('o', "output", &output_name, "file",
766 "output file name"),
2e6cdf99
SE
767 OPT_BOOLEAN('i', "no-inherit", &no_inherit,
768 "child tasks do not inherit counters"),
1967936d
ACM
769 OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"),
770 OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
3efa1cc9
IM
771 OPT_BOOLEAN('g', "call-graph", &call_graph,
772 "do call-graph (stack chain/backtrace) recording"),
c0555642 773 OPT_INCR('v', "verbose", &verbose,
3da297a6 774 "be more verbose (show counter open errors, etc)"),
b44308f5 775 OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
649c48a9
PZ
776 OPT_BOOLEAN('s', "stat", &inherit_stat,
777 "per thread counts"),
4bba828d
AB
778 OPT_BOOLEAN('d', "data", &sample_address,
779 "Sample addresses"),
9c90a61c 780 OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"),
649c48a9
PZ
781 OPT_BOOLEAN('n', "no-samples", &no_samples,
782 "don't sample"),
baa2f6ce 783 OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache,
a1ac1d3c 784 "do not update the buildid cache"),
baa2f6ce
ACM
785 OPT_BOOLEAN('B', "no-buildid", &no_buildid,
786 "do not collect buildids in perf.data"),
023695d9
SE
787 OPT_CALLBACK('G', "cgroup", &evsel_list, "name",
788 "monitor event in cgroup name only",
789 parse_cgroups),
0e9b20b8
IM
790 OPT_END()
791};
792
f37a291c 793int cmd_record(int argc, const char **argv, const char *prefix __used)
0e9b20b8 794{
69aad6f1
ACM
795 int err = -ENOMEM;
796 struct perf_evsel *pos;
0e9b20b8 797
7e2ed097 798 evsel_list = perf_evlist__new(NULL, NULL);
361c99a6
ACM
799 if (evsel_list == NULL)
800 return -ENOMEM;
801
bca647aa 802 argc = parse_options(argc, argv, record_options, record_usage,
655000e7 803 PARSE_OPT_STOP_AT_NON_OPTION);
d6d901c2 804 if (!argc && target_pid == -1 && target_tid == -1 &&
c45c6ea2 805 !system_wide && !cpu_list)
bca647aa 806 usage_with_options(record_usage, record_options);
0e9b20b8 807
7865e817
FW
808 if (force && append_file) {
809 fprintf(stderr, "Can't overwrite and append at the same time."
810 " You need to choose between -f and -A");
bca647aa 811 usage_with_options(record_usage, record_options);
7865e817
FW
812 } else if (append_file) {
813 write_mode = WRITE_APPEND;
814 } else {
815 write_mode = WRITE_FORCE;
816 }
817
023695d9
SE
818 if (nr_cgroups && !system_wide) {
819 fprintf(stderr, "cgroup monitoring only available in"
820 " system-wide mode\n");
821 usage_with_options(record_usage, record_options);
822 }
823
655000e7 824 symbol__init();
baa2f6ce 825
ec80fde7
ACM
826 if (symbol_conf.kptr_restrict)
827 pr_warning("WARNING: Kernel address maps "
828 "(/proc/{kallsyms,modules}) are restricted, "
829 "check /proc/sys/kernel/kptr_restrict.\n\n"
830 "Samples in kernel functions may not be resolved "
831 "if a suitable vmlinux file is not found in the "
832 "buildid cache or in the vmlinux path.\n\n"
833 "Samples in kernel modules won't be resolved "
834 "at all.\n\n"
835 "If some relocation was applied (e.g. kexec) "
836 "symbols may be misresolved even with a suitable "
837 "vmlinux or kallsyms file.\n\n");
838
baa2f6ce 839 if (no_buildid_cache || no_buildid)
a1ac1d3c 840 disable_buildid_cache();
655000e7 841
361c99a6
ACM
842 if (evsel_list->nr_entries == 0 &&
843 perf_evlist__add_default(evsel_list) < 0) {
69aad6f1
ACM
844 pr_err("Not enough memory for event selector list\n");
845 goto out_symbol_exit;
bbd36e5e 846 }
0e9b20b8 847
5c98d466 848 if (target_pid != -1)
d6d901c2 849 target_tid = target_pid;
d6d901c2 850
7e2ed097
ACM
851 if (perf_evlist__create_maps(evsel_list, target_pid,
852 target_tid, cpu_list) < 0)
dd7927f4 853 usage_with_options(record_usage, record_options);
69aad6f1 854
361c99a6 855 list_for_each_entry(pos, &evsel_list->entries, node) {
7e2ed097
ACM
856 if (perf_evsel__alloc_fd(pos, evsel_list->cpus->nr,
857 evsel_list->threads->nr) < 0)
69aad6f1 858 goto out_free_fd;
ad7f4e3f
ACM
859 if (perf_header__push_event(pos->attr.config, event_name(pos)))
860 goto out_free_fd;
d6d901c2 861 }
5c581041 862
7e2ed097 863 if (perf_evlist__alloc_pollfd(evsel_list) < 0)
39d17dac 864 goto out_free_fd;
d6d901c2 865
3de29cab 866 if (user_interval != ULLONG_MAX)
f9212819
FW
867 default_interval = user_interval;
868 if (user_freq != UINT_MAX)
869 freq = user_freq;
870
7e4ff9e3
MG
871 /*
872 * User specified count overrides default frequency.
873 */
874 if (default_interval)
875 freq = 0;
876 else if (freq) {
877 default_interval = freq;
878 } else {
879 fprintf(stderr, "frequency and count are zero, aborting\n");
39d17dac 880 err = -EINVAL;
5c581041 881 goto out_free_fd;
7e4ff9e3
MG
882 }
883
39d17dac 884 err = __cmd_record(argc, argv);
39d17dac 885out_free_fd:
7e2ed097 886 perf_evlist__delete_maps(evsel_list);
d65a458b
ACM
887out_symbol_exit:
888 symbol__exit();
39d17dac 889 return err;
0e9b20b8 890}