Merge branch 'sched-fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[GitHub/mt8127/android_kernel_alcatel_ttab.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;
42e59d7d 44static unsigned int mmap_pages = 128;
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
PZ
165
166 attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
167 PERF_FORMAT_TOTAL_TIME_RUNNING |
168 PERF_FORMAT_ID;
16c8a109 169
3a9f131f 170 attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID;
3efa1cc9 171
361c99a6 172 if (evlist->nr_entries > 1)
8907fd60
EM
173 attr->sample_type |= PERF_SAMPLE_ID;
174
f9212819
FW
175 /*
176 * We default some events to a 1 default interval. But keep
177 * it a weak assumption overridable by the user.
178 */
179 if (!attr->sample_period || (user_freq != UINT_MAX &&
3de29cab 180 user_interval != ULLONG_MAX)) {
f9212819
FW
181 if (freq) {
182 attr->sample_type |= PERF_SAMPLE_PERIOD;
183 attr->freq = 1;
184 attr->sample_freq = freq;
185 } else {
186 attr->sample_period = default_interval;
187 }
1dba15e7 188 }
3efa1cc9 189
649c48a9
PZ
190 if (no_samples)
191 attr->sample_freq = 0;
192
193 if (inherit_stat)
194 attr->inherit_stat = 1;
195
3af9e859 196 if (sample_address) {
4bba828d 197 attr->sample_type |= PERF_SAMPLE_ADDR;
3af9e859
EM
198 attr->mmap_data = track;
199 }
4bba828d 200
3efa1cc9
IM
201 if (call_graph)
202 attr->sample_type |= PERF_SAMPLE_CALLCHAIN;
203
f60f3593
AS
204 if (system_wide)
205 attr->sample_type |= PERF_SAMPLE_CPU;
206
a43d3f08
ACM
207 if (sample_id_all_avail &&
208 (sample_time || system_wide || !no_inherit || cpu_list))
9c90a61c
ACM
209 attr->sample_type |= PERF_SAMPLE_TIME;
210
cd6feeea 211 if (raw_samples) {
6ddf259d 212 attr->sample_type |= PERF_SAMPLE_TIME;
daac07b2 213 attr->sample_type |= PERF_SAMPLE_RAW;
cd6feeea
IM
214 attr->sample_type |= PERF_SAMPLE_CPU;
215 }
f413cdb8 216
acac03fa
KS
217 if (nodelay) {
218 attr->watermark = 0;
219 attr->wakeup_events = 1;
220 }
221
a21ca2ca
IM
222 attr->mmap = track;
223 attr->comm = track;
dd7927f4 224
2e6cdf99 225 if (target_pid == -1 && target_tid == -1 && !system_wide) {
46be604b 226 attr->disabled = 1;
bedbfdea 227 attr->enable_on_exec = 1;
46be604b 228 }
dd7927f4 229}
bedbfdea 230
a91e5431
ACM
231static bool perf_evlist__equal(struct perf_evlist *evlist,
232 struct perf_evlist *other)
233{
234 struct perf_evsel *pos, *pair;
235
236 if (evlist->nr_entries != other->nr_entries)
237 return false;
238
239 pair = list_entry(other->entries.next, struct perf_evsel, node);
240
241 list_for_each_entry(pos, &evlist->entries, node) {
242 if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0))
243 return false;
244 pair = list_entry(pair->node.next, struct perf_evsel, node);
245 }
246
247 return true;
248}
249
dd7927f4
ACM
250static void open_counters(struct perf_evlist *evlist)
251{
252 struct perf_evsel *pos;
dd7927f4
ACM
253
254 list_for_each_entry(pos, &evlist->entries, node) {
255 struct perf_event_attr *attr = &pos->attr;
256 /*
257 * Check if parse_single_tracepoint_event has already asked for
258 * PERF_SAMPLE_TIME.
259 *
260 * XXX this is kludgy but short term fix for problems introduced by
261 * eac23d1c that broke 'perf script' by having different sample_types
262 * when using multiple tracepoint events when we use a perf binary
263 * that tries to use sample_id_all on an older kernel.
264 *
265 * We need to move counter creation to perf_session, support
266 * different sample_types, etc.
267 */
268 bool time_needed = attr->sample_type & PERF_SAMPLE_TIME;
d6d901c2 269
dd7927f4
ACM
270 config_attr(pos, evlist);
271retry_sample_id:
272 attr->sample_id_all = sample_id_all_avail ? 1 : 0;
273try_again:
7e2ed097
ACM
274 if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group,
275 !no_inherit) < 0) {
d6d901c2
ZY
276 int err = errno;
277
278 if (err == EPERM || err == EACCES)
279 die("Permission error - are you root?\n"
280 "\t Consider tweaking"
281 " /proc/sys/kernel/perf_event_paranoid.\n");
c45c6ea2 282 else if (err == ENODEV && cpu_list) {
d6d901c2
ZY
283 die("No such device - did you specify"
284 " an out-of-range profile CPU?\n");
9c90a61c
ACM
285 } else if (err == EINVAL && sample_id_all_avail) {
286 /*
287 * Old kernel, no attr->sample_id_type_all field
288 */
289 sample_id_all_avail = false;
a43d3f08 290 if (!sample_time && !raw_samples && !time_needed)
eac23d1c
IM
291 attr->sample_type &= ~PERF_SAMPLE_TIME;
292
9c90a61c 293 goto retry_sample_id;
d6d901c2 294 }
3da297a6 295
d6d901c2
ZY
296 /*
297 * If it's cycles then fall back to hrtimer
298 * based cpu-clock-tick sw counter, which
299 * is always available even if no PMU support:
300 */
301 if (attr->type == PERF_TYPE_HARDWARE
302 && attr->config == PERF_COUNT_HW_CPU_CYCLES) {
303
304 if (verbose)
305 warning(" ... trying to fall back to cpu-clock-ticks\n");
306 attr->type = PERF_TYPE_SOFTWARE;
307 attr->config = PERF_COUNT_SW_CPU_CLOCK;
308 goto try_again;
309 }
310 printf("\n");
d9cf837e 311 error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n",
dd7927f4 312 err, strerror(err));
bfd45118
SK
313
314#if defined(__i386__) || defined(__x86_64__)
d6d901c2
ZY
315 if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP)
316 die("No hardware sampling interrupt available."
317 " No APIC? If so then you can boot the kernel"
318 " with the \"lapic\" boot parameter to"
319 " force-enable it.\n");
bfd45118
SK
320#endif
321
d6d901c2 322 die("No CONFIG_PERF_EVENTS=y kernel support configured?\n");
c171b552
LZ
323 }
324 }
a43d3f08 325
0a102479
FW
326 if (perf_evlist__set_filters(evlist)) {
327 error("failed to set filter with %d (%s)\n", errno,
328 strerror(errno));
329 exit(-1);
330 }
331
7e2ed097 332 if (perf_evlist__mmap(evlist, mmap_pages, false) < 0)
0a27d7f9
ACM
333 die("failed to mmap with %d (%s)\n", errno, strerror(errno));
334
a91e5431
ACM
335 if (file_new)
336 session->evlist = evlist;
337 else {
338 if (!perf_evlist__equal(session->evlist, evlist)) {
339 fprintf(stderr, "incompatible append\n");
340 exit(-1);
341 }
342 }
343
344 perf_session__update_sample_type(session);
16c8a109
PZ
345}
346
6122e4e4
ACM
347static int process_buildids(void)
348{
349 u64 size = lseek(output, 0, SEEK_CUR);
350
9f591fd7
ACM
351 if (size == 0)
352 return 0;
353
6122e4e4
ACM
354 session->fd = output;
355 return __perf_session__process_events(session, post_processing_offset,
356 size - post_processing_offset,
357 size, &build_id__mark_dso_hit_ops);
358}
359
f5970550
PZ
360static void atexit_header(void)
361{
c7929e47
TZ
362 if (!pipe_output) {
363 session->header.data_size += bytes_written;
f5970550 364
baa2f6ce
ACM
365 if (!no_buildid)
366 process_buildids();
a91e5431 367 perf_session__write_header(session, evsel_list, output, true);
39d17dac 368 perf_session__delete(session);
361c99a6 369 perf_evlist__delete(evsel_list);
d65a458b 370 symbol__exit();
c7929e47 371 }
f5970550
PZ
372}
373
8115d60c 374static void perf_event__synthesize_guest_os(struct machine *machine, void *data)
a1645ce1
ZY
375{
376 int err;
23346f21 377 struct perf_session *psession = data;
a1645ce1 378
23346f21 379 if (machine__is_host(machine))
a1645ce1
ZY
380 return;
381
382 /*
383 *As for guest kernel when processing subcommand record&report,
384 *we arrange module mmap prior to guest kernel mmap and trigger
385 *a preload dso because default guest module symbols are loaded
386 *from guest kallsyms instead of /lib/modules/XXX/XXX. This
387 *method is used to avoid symbol missing when the first addr is
388 *in module instead of in guest kernel.
389 */
8115d60c
ACM
390 err = perf_event__synthesize_modules(process_synthesized_event,
391 psession, machine);
a1645ce1
ZY
392 if (err < 0)
393 pr_err("Couldn't record guest kernel [%d]'s reference"
23346f21 394 " relocation symbol.\n", machine->pid);
a1645ce1 395
a1645ce1
ZY
396 /*
397 * We use _stext for guest kernel because guest kernel's /proc/kallsyms
398 * have no _text sometimes.
399 */
8115d60c
ACM
400 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
401 psession, machine, "_text");
a1645ce1 402 if (err < 0)
8115d60c
ACM
403 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
404 psession, machine,
405 "_stext");
a1645ce1
ZY
406 if (err < 0)
407 pr_err("Couldn't record guest kernel [%d]'s reference"
23346f21 408 " relocation symbol.\n", machine->pid);
a1645ce1
ZY
409}
410
98402807
FW
411static struct perf_event_header finished_round_event = {
412 .size = sizeof(struct perf_event_header),
413 .type = PERF_RECORD_FINISHED_ROUND,
414};
415
416static void mmap_read_all(void)
417{
0e2e63dd 418 int i;
98402807 419
7e2ed097 420 for (i = 0; i < evsel_list->cpus->nr; i++) {
0a27d7f9
ACM
421 if (evsel_list->mmap[i].base)
422 mmap_read(&evsel_list->mmap[i]);
98402807
FW
423 }
424
425 if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO))
426 write_output(&finished_round_event, sizeof(finished_round_event));
427}
428
d4db3f16 429static int __cmd_record(int argc, const char **argv)
16c8a109 430{
69aad6f1 431 int i;
abaff32a 432 struct stat st;
abaff32a 433 int flags;
4dc0a04b 434 int err;
8b412664 435 unsigned long waking = 0;
856e9660 436 int child_ready_pipe[2], go_pipe[2];
46be604b 437 const bool forks = argc > 0;
856e9660 438 char buf;
23346f21 439 struct machine *machine;
de9ac07b
PZ
440
441 page_size = sysconf(_SC_PAGE_SIZE);
de9ac07b 442
f5970550
PZ
443 atexit(sig_atexit);
444 signal(SIGCHLD, sig_handler);
445 signal(SIGINT, sig_handler);
18483b81 446 signal(SIGUSR1, sig_handler);
f5970550 447
d4db3f16 448 if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) {
856e9660
PZ
449 perror("failed to create pipes");
450 exit(-1);
451 }
452
d7065adb
FBH
453 if (!output_name) {
454 if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode))
455 pipe_output = 1;
456 else
457 output_name = "perf.data";
458 }
459 if (output_name) {
460 if (!strcmp(output_name, "-"))
461 pipe_output = 1;
462 else if (!stat(output_name, &st) && st.st_size) {
463 if (write_mode == WRITE_FORCE) {
464 char oldname[PATH_MAX];
465 snprintf(oldname, sizeof(oldname), "%s.old",
466 output_name);
467 unlink(oldname);
468 rename(output_name, oldname);
469 }
470 } else if (write_mode == WRITE_APPEND) {
471 write_mode = WRITE_FORCE;
266e0e21 472 }
97124d5e
PZ
473 }
474
f887f301 475 flags = O_CREAT|O_RDWR;
7865e817 476 if (write_mode == WRITE_APPEND)
f5970550 477 file_new = 0;
abaff32a
IM
478 else
479 flags |= O_TRUNC;
480
529870e3
TZ
481 if (pipe_output)
482 output = STDOUT_FILENO;
483 else
484 output = open(output_name, flags, S_IRUSR | S_IWUSR);
de9ac07b
PZ
485 if (output < 0) {
486 perror("failed to create output file");
487 exit(-1);
488 }
489
7865e817 490 session = perf_session__new(output_name, O_WRONLY,
21ef97f0 491 write_mode == WRITE_FORCE, false, NULL);
94c744b6 492 if (session == NULL) {
a9a70bbc
ACM
493 pr_err("Not enough memory for reading perf file header\n");
494 return -1;
495 }
496
baa2f6ce
ACM
497 if (!no_buildid)
498 perf_header__set_feat(&session->header, HEADER_BUILD_ID);
499
4dc0a04b 500 if (!file_new) {
a91e5431 501 err = perf_session__read_header(session, output);
4dc0a04b 502 if (err < 0)
39d17dac 503 goto out_delete_session;
4dc0a04b
ACM
504 }
505
361c99a6 506 if (have_tracepoints(&evsel_list->entries))
94c744b6 507 perf_header__set_feat(&session->header, HEADER_TRACE_INFO);
03456a15 508
d4db3f16 509 if (forks) {
46be604b 510 child_pid = fork();
2fb750e8 511 if (child_pid < 0) {
856e9660
PZ
512 perror("failed to fork");
513 exit(-1);
514 }
7c6a1c65 515
46be604b 516 if (!child_pid) {
529870e3
TZ
517 if (pipe_output)
518 dup2(2, 1);
856e9660
PZ
519 close(child_ready_pipe[0]);
520 close(go_pipe[1]);
521 fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC);
522
523 /*
524 * Do a dummy execvp to get the PLT entry resolved,
525 * so we avoid the resolver overhead on the real
526 * execvp call.
527 */
528 execvp("", (char **)argv);
529
530 /*
531 * Tell the parent we're ready to go
532 */
533 close(child_ready_pipe[1]);
534
535 /*
536 * Wait until the parent tells us to go.
537 */
538 if (read(go_pipe[0], &buf, 1) == -1)
539 perror("unable to read pipe");
540
541 execvp(argv[0], (char **)argv);
542
543 perror(argv[0]);
18483b81 544 kill(getppid(), SIGUSR1);
856e9660 545 exit(-1);
0a5ac846 546 }
856e9660 547
d6d901c2 548 if (!system_wide && target_tid == -1 && target_pid == -1)
7e2ed097 549 evsel_list->threads->map[0] = child_pid;
d6d901c2 550
856e9660
PZ
551 close(child_ready_pipe[1]);
552 close(go_pipe[0]);
553 /*
554 * wait for child to settle
555 */
556 if (read(child_ready_pipe[0], &buf, 1) == -1) {
557 perror("unable to read pipe");
558 exit(-1);
559 }
560 close(child_ready_pipe[0]);
561 }
562
dd7927f4 563 open_counters(evsel_list);
de9ac07b 564
712a4b60
ACM
565 /*
566 * perf_session__delete(session) will be called at atexit_header()
567 */
568 atexit(atexit_header);
569
529870e3
TZ
570 if (pipe_output) {
571 err = perf_header__write_pipe(output);
572 if (err < 0)
573 return err;
574 } else if (file_new) {
a91e5431
ACM
575 err = perf_session__write_header(session, evsel_list,
576 output, false);
d5eed904
ACM
577 if (err < 0)
578 return err;
56b03f3c
ACM
579 }
580
6122e4e4
ACM
581 post_processing_offset = lseek(output, 0, SEEK_CUR);
582
2c46dbb5 583 if (pipe_output) {
a91e5431
ACM
584 err = perf_session__synthesize_attrs(session,
585 process_synthesized_event);
2c46dbb5
TZ
586 if (err < 0) {
587 pr_err("Couldn't synthesize attrs.\n");
588 return err;
589 }
cd19a035 590
8115d60c
ACM
591 err = perf_event__synthesize_event_types(process_synthesized_event,
592 session);
cd19a035
TZ
593 if (err < 0) {
594 pr_err("Couldn't synthesize event_types.\n");
595 return err;
596 }
9215545e 597
361c99a6 598 if (have_tracepoints(&evsel_list->entries)) {
63e0c771
TZ
599 /*
600 * FIXME err <= 0 here actually means that
601 * there were no tracepoints so its not really
602 * an error, just that we don't need to
603 * synthesize anything. We really have to
604 * return this more properly and also
605 * propagate errors that now are calling die()
606 */
8115d60c
ACM
607 err = perf_event__synthesize_tracing_data(output, evsel_list,
608 process_synthesized_event,
609 session);
63e0c771
TZ
610 if (err <= 0) {
611 pr_err("Couldn't record tracing data.\n");
612 return err;
613 }
2c9faa06 614 advance_output(err);
63e0c771 615 }
2c46dbb5
TZ
616 }
617
23346f21
ACM
618 machine = perf_session__find_host_machine(session);
619 if (!machine) {
a1645ce1
ZY
620 pr_err("Couldn't find native kernel information.\n");
621 return -1;
622 }
623
8115d60c
ACM
624 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
625 session, machine, "_text");
70162138 626 if (err < 0)
8115d60c
ACM
627 err = perf_event__synthesize_kernel_mmap(process_synthesized_event,
628 session, machine, "_stext");
c1a3a4b9
ACM
629 if (err < 0)
630 pr_err("Couldn't record kernel reference relocation symbol\n"
631 "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
632 "Check /proc/kallsyms permission or run as root.\n");
b7cece76 633
8115d60c
ACM
634 err = perf_event__synthesize_modules(process_synthesized_event,
635 session, machine);
c1a3a4b9
ACM
636 if (err < 0)
637 pr_err("Couldn't record kernel module information.\n"
638 "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n"
639 "Check /proc/modules permission or run as root.\n");
640
a1645ce1 641 if (perf_guest)
8115d60c
ACM
642 perf_session__process_machines(session,
643 perf_event__synthesize_guest_os);
7c6a1c65 644
cf103a14 645 if (!system_wide)
7c940c18
ACM
646 perf_event__synthesize_thread_map(evsel_list->threads,
647 process_synthesized_event,
648 session);
234fbbf5 649 else
8115d60c
ACM
650 perf_event__synthesize_threads(process_synthesized_event,
651 session);
7c6a1c65 652
de9ac07b
PZ
653 if (realtime_prio) {
654 struct sched_param param;
655
656 param.sched_priority = realtime_prio;
657 if (sched_setscheduler(0, SCHED_FIFO, &param)) {
6beba7ad 658 pr_err("Could not set realtime priority.\n");
de9ac07b
PZ
659 exit(-1);
660 }
661 }
662
856e9660
PZ
663 /*
664 * Let the child rip
665 */
d4db3f16
ACM
666 if (forks)
667 close(go_pipe[1]);
856e9660 668
649c48a9 669 for (;;) {
2debbc83 670 int hits = samples;
d6d901c2 671 int thread;
de9ac07b 672
98402807 673 mmap_read_all();
de9ac07b 674
649c48a9
PZ
675 if (hits == samples) {
676 if (done)
677 break;
5c581041 678 err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1);
8b412664
PZ
679 waking++;
680 }
681
682 if (done) {
7e2ed097 683 for (i = 0; i < evsel_list->cpus->nr; i++) {
69aad6f1
ACM
684 struct perf_evsel *pos;
685
361c99a6 686 list_for_each_entry(pos, &evsel_list->entries, node) {
d6d901c2 687 for (thread = 0;
7e2ed097 688 thread < evsel_list->threads->nr;
d6d901c2 689 thread++)
69aad6f1 690 ioctl(FD(pos, i, thread),
d6d901c2
ZY
691 PERF_EVENT_IOC_DISABLE);
692 }
8b412664 693 }
649c48a9 694 }
de9ac07b
PZ
695 }
696
18483b81 697 if (quiet || signr == SIGUSR1)
b44308f5
ACM
698 return 0;
699
8b412664
PZ
700 fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking);
701
021e9f47
IM
702 /*
703 * Approximate RIP event size: 24 bytes.
704 */
705 fprintf(stderr,
9486aa38 706 "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n",
021e9f47
IM
707 (double)bytes_written / 1024.0 / 1024.0,
708 output_name,
709 bytes_written / 24);
addc2785 710
de9ac07b 711 return 0;
39d17dac
ACM
712
713out_delete_session:
714 perf_session__delete(session);
715 return err;
de9ac07b 716}
0e9b20b8 717
0e9b20b8 718static const char * const record_usage[] = {
9e096753
MG
719 "perf record [<options>] [<command>]",
720 "perf record [<options>] -- <command> [<options>]",
0e9b20b8
IM
721 NULL
722};
723
7865e817
FW
724static bool force, append_file;
725
bca647aa 726const struct option record_options[] = {
361c99a6 727 OPT_CALLBACK('e', "event", &evsel_list, "event",
86847b62
TG
728 "event selector. use 'perf list' to list available events",
729 parse_events),
361c99a6 730 OPT_CALLBACK(0, "filter", &evsel_list, "filter",
c171b552 731 "event filter", parse_filter),
0e9b20b8 732 OPT_INTEGER('p', "pid", &target_pid,
d6d901c2
ZY
733 "record events on existing process id"),
734 OPT_INTEGER('t', "tid", &target_tid,
735 "record events on existing thread id"),
0e9b20b8
IM
736 OPT_INTEGER('r', "realtime", &realtime_prio,
737 "collect data with this RT SCHED_FIFO priority"),
acac03fa
KS
738 OPT_BOOLEAN('D', "no-delay", &nodelay,
739 "collect data without buffering"),
daac07b2
FW
740 OPT_BOOLEAN('R', "raw-samples", &raw_samples,
741 "collect raw sample records from all opened counters"),
0e9b20b8
IM
742 OPT_BOOLEAN('a', "all-cpus", &system_wide,
743 "system-wide collection from all CPUs"),
abaff32a
IM
744 OPT_BOOLEAN('A', "append", &append_file,
745 "append to the output file to do incremental profiling"),
c45c6ea2
SE
746 OPT_STRING('C', "cpu", &cpu_list, "cpu",
747 "list of cpus to monitor"),
97124d5e 748 OPT_BOOLEAN('f', "force", &force,
7865e817 749 "overwrite existing data file (deprecated)"),
3de29cab 750 OPT_U64('c', "count", &user_interval, "event period to sample"),
abaff32a
IM
751 OPT_STRING('o', "output", &output_name, "file",
752 "output file name"),
2e6cdf99
SE
753 OPT_BOOLEAN('i', "no-inherit", &no_inherit,
754 "child tasks do not inherit counters"),
1967936d
ACM
755 OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"),
756 OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"),
3efa1cc9
IM
757 OPT_BOOLEAN('g', "call-graph", &call_graph,
758 "do call-graph (stack chain/backtrace) recording"),
c0555642 759 OPT_INCR('v', "verbose", &verbose,
3da297a6 760 "be more verbose (show counter open errors, etc)"),
b44308f5 761 OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"),
649c48a9
PZ
762 OPT_BOOLEAN('s', "stat", &inherit_stat,
763 "per thread counts"),
4bba828d
AB
764 OPT_BOOLEAN('d', "data", &sample_address,
765 "Sample addresses"),
9c90a61c 766 OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"),
649c48a9
PZ
767 OPT_BOOLEAN('n', "no-samples", &no_samples,
768 "don't sample"),
baa2f6ce 769 OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache,
a1ac1d3c 770 "do not update the buildid cache"),
baa2f6ce
ACM
771 OPT_BOOLEAN('B', "no-buildid", &no_buildid,
772 "do not collect buildids in perf.data"),
023695d9
SE
773 OPT_CALLBACK('G', "cgroup", &evsel_list, "name",
774 "monitor event in cgroup name only",
775 parse_cgroups),
0e9b20b8
IM
776 OPT_END()
777};
778
f37a291c 779int cmd_record(int argc, const char **argv, const char *prefix __used)
0e9b20b8 780{
69aad6f1
ACM
781 int err = -ENOMEM;
782 struct perf_evsel *pos;
0e9b20b8 783
7e2ed097 784 evsel_list = perf_evlist__new(NULL, NULL);
361c99a6
ACM
785 if (evsel_list == NULL)
786 return -ENOMEM;
787
bca647aa 788 argc = parse_options(argc, argv, record_options, record_usage,
655000e7 789 PARSE_OPT_STOP_AT_NON_OPTION);
d6d901c2 790 if (!argc && target_pid == -1 && target_tid == -1 &&
c45c6ea2 791 !system_wide && !cpu_list)
bca647aa 792 usage_with_options(record_usage, record_options);
0e9b20b8 793
7865e817
FW
794 if (force && append_file) {
795 fprintf(stderr, "Can't overwrite and append at the same time."
796 " You need to choose between -f and -A");
bca647aa 797 usage_with_options(record_usage, record_options);
7865e817
FW
798 } else if (append_file) {
799 write_mode = WRITE_APPEND;
800 } else {
801 write_mode = WRITE_FORCE;
802 }
803
023695d9
SE
804 if (nr_cgroups && !system_wide) {
805 fprintf(stderr, "cgroup monitoring only available in"
806 " system-wide mode\n");
807 usage_with_options(record_usage, record_options);
808 }
809
655000e7 810 symbol__init();
baa2f6ce
ACM
811
812 if (no_buildid_cache || no_buildid)
a1ac1d3c 813 disable_buildid_cache();
655000e7 814
361c99a6
ACM
815 if (evsel_list->nr_entries == 0 &&
816 perf_evlist__add_default(evsel_list) < 0) {
69aad6f1
ACM
817 pr_err("Not enough memory for event selector list\n");
818 goto out_symbol_exit;
bbd36e5e 819 }
0e9b20b8 820
5c98d466 821 if (target_pid != -1)
d6d901c2 822 target_tid = target_pid;
d6d901c2 823
7e2ed097
ACM
824 if (perf_evlist__create_maps(evsel_list, target_pid,
825 target_tid, cpu_list) < 0)
dd7927f4 826 usage_with_options(record_usage, record_options);
69aad6f1 827
361c99a6 828 list_for_each_entry(pos, &evsel_list->entries, node) {
7e2ed097
ACM
829 if (perf_evsel__alloc_fd(pos, evsel_list->cpus->nr,
830 evsel_list->threads->nr) < 0)
69aad6f1 831 goto out_free_fd;
ad7f4e3f
ACM
832 if (perf_header__push_event(pos->attr.config, event_name(pos)))
833 goto out_free_fd;
d6d901c2 834 }
5c581041 835
7e2ed097 836 if (perf_evlist__alloc_pollfd(evsel_list) < 0)
39d17dac 837 goto out_free_fd;
d6d901c2 838
3de29cab 839 if (user_interval != ULLONG_MAX)
f9212819
FW
840 default_interval = user_interval;
841 if (user_freq != UINT_MAX)
842 freq = user_freq;
843
7e4ff9e3
MG
844 /*
845 * User specified count overrides default frequency.
846 */
847 if (default_interval)
848 freq = 0;
849 else if (freq) {
850 default_interval = freq;
851 } else {
852 fprintf(stderr, "frequency and count are zero, aborting\n");
39d17dac 853 err = -EINVAL;
5c581041 854 goto out_free_fd;
7e4ff9e3
MG
855 }
856
39d17dac 857 err = __cmd_record(argc, argv);
39d17dac 858out_free_fd:
7e2ed097 859 perf_evlist__delete_maps(evsel_list);
d65a458b
ACM
860out_symbol_exit:
861 symbol__exit();
39d17dac 862 return err;
0e9b20b8 863}