Commit | Line | Data |
---|---|---|
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 | |
7865e817 FW |
33 | enum write_mode_t { |
34 | WRITE_FORCE, | |
35 | WRITE_APPEND | |
36 | }; | |
37 | ||
3de29cab SE |
38 | static u64 user_interval = ULLONG_MAX; |
39 | static u64 default_interval = 0; | |
a21ca2ca | 40 | |
de9ac07b | 41 | static unsigned int page_size; |
800cd25c | 42 | static unsigned int mmap_pages = UINT_MAX; |
f9212819 | 43 | static unsigned int user_freq = UINT_MAX; |
42e59d7d | 44 | static int freq = 1000; |
de9ac07b | 45 | static int output; |
529870e3 | 46 | static int pipe_output = 0; |
d7065adb | 47 | static const char *output_name = NULL; |
43bece79 | 48 | static bool group = false; |
1967936d | 49 | static int realtime_prio = 0; |
acac03fa | 50 | static bool nodelay = false; |
c0555642 | 51 | static bool raw_samples = false; |
9c90a61c | 52 | static bool sample_id_all_avail = true; |
c0555642 | 53 | static bool system_wide = false; |
42e59d7d | 54 | static pid_t target_pid = -1; |
d6d901c2 | 55 | static pid_t target_tid = -1; |
42e59d7d | 56 | static pid_t child_pid = -1; |
2e6cdf99 | 57 | static bool no_inherit = false; |
7865e817 | 58 | static enum write_mode_t write_mode = WRITE_FORCE; |
c0555642 IM |
59 | static bool call_graph = false; |
60 | static bool inherit_stat = false; | |
61 | static bool no_samples = false; | |
62 | static bool sample_address = false; | |
9c90a61c | 63 | static bool sample_time = false; |
a1ac1d3c | 64 | static bool no_buildid = false; |
baa2f6ce | 65 | static bool no_buildid_cache = false; |
361c99a6 | 66 | static struct perf_evlist *evsel_list; |
42e59d7d IM |
67 | |
68 | static long samples = 0; | |
42e59d7d | 69 | static u64 bytes_written = 0; |
a21ca2ca | 70 | |
42e59d7d | 71 | static int file_new = 1; |
6122e4e4 | 72 | static off_t post_processing_offset; |
7c6a1c65 | 73 | |
94c744b6 | 74 | static struct perf_session *session; |
c45c6ea2 | 75 | static const char *cpu_list; |
f5970550 | 76 | |
9215545e TZ |
77 | static void advance_output(size_t size) |
78 | { | |
79 | bytes_written += size; | |
80 | } | |
81 | ||
f5970550 PZ |
82 | static void write_output(void *buf, size_t size) |
83 | { | |
84 | while (size) { | |
85 | int ret = write(output, buf, size); | |
86 | ||
87 | if (ret < 0) | |
88 | die("failed to write"); | |
89 | ||
90 | size -= ret; | |
91 | buf += ret; | |
92 | ||
93 | bytes_written += ret; | |
94 | } | |
95 | } | |
96 | ||
8115d60c | 97 | static int process_synthesized_event(union perf_event *event, |
8d50e5b4 | 98 | struct perf_sample *sample __used, |
d8f66248 | 99 | struct perf_session *self __used) |
234fbbf5 | 100 | { |
6122e4e4 | 101 | write_output(event, event->header.size); |
234fbbf5 ACM |
102 | return 0; |
103 | } | |
104 | ||
744bd8aa | 105 | static void mmap_read(struct perf_mmap *md) |
de9ac07b | 106 | { |
744bd8aa | 107 | unsigned int head = perf_mmap__read_head(md); |
de9ac07b PZ |
108 | unsigned int old = md->prev; |
109 | unsigned char *data = md->base + page_size; | |
110 | unsigned long size; | |
111 | void *buf; | |
de9ac07b | 112 | |
dc82009a ACM |
113 | if (old == head) |
114 | return; | |
115 | ||
116 | samples++; | |
de9ac07b PZ |
117 | |
118 | size = head - old; | |
119 | ||
120 | if ((old & md->mask) + size != (head & md->mask)) { | |
121 | buf = &data[old & md->mask]; | |
122 | size = md->mask + 1 - (old & md->mask); | |
123 | old += size; | |
021e9f47 | 124 | |
6122e4e4 | 125 | write_output(buf, size); |
de9ac07b PZ |
126 | } |
127 | ||
128 | buf = &data[old & md->mask]; | |
129 | size = head - old; | |
130 | old += size; | |
021e9f47 | 131 | |
6122e4e4 | 132 | write_output(buf, size); |
de9ac07b PZ |
133 | |
134 | md->prev = old; | |
115d2d89 | 135 | perf_mmap__write_tail(md, old); |
de9ac07b PZ |
136 | } |
137 | ||
138 | static volatile int done = 0; | |
f7b7c26e | 139 | static volatile int signr = -1; |
de9ac07b | 140 | |
16c8a109 | 141 | static void sig_handler(int sig) |
de9ac07b | 142 | { |
16c8a109 | 143 | done = 1; |
f7b7c26e PZ |
144 | signr = sig; |
145 | } | |
146 | ||
147 | static void sig_atexit(void) | |
148 | { | |
5ffc8881 | 149 | if (child_pid > 0) |
933da83a CW |
150 | kill(child_pid, SIGTERM); |
151 | ||
18483b81 | 152 | if (signr == -1 || signr == SIGUSR1) |
f7b7c26e PZ |
153 | return; |
154 | ||
155 | signal(signr, SIG_DFL); | |
156 | kill(getpid(), signr); | |
de9ac07b PZ |
157 | } |
158 | ||
dd7927f4 ACM |
159 | static void config_attr(struct perf_evsel *evsel, struct perf_evlist *evlist) |
160 | { | |
161 | struct perf_event_attr *attr = &evsel->attr; | |
162 | int track = !evsel->idx; /* only the first counter needs these */ | |
7c6a1c65 | 163 | |
5d2cd909 | 164 | attr->inherit = !no_inherit; |
7c6a1c65 PZ |
165 | attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | |
166 | PERF_FORMAT_TOTAL_TIME_RUNNING | | |
167 | PERF_FORMAT_ID; | |
16c8a109 | 168 | |
3a9f131f | 169 | attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID; |
3efa1cc9 | 170 | |
361c99a6 | 171 | if (evlist->nr_entries > 1) |
8907fd60 EM |
172 | attr->sample_type |= PERF_SAMPLE_ID; |
173 | ||
f9212819 FW |
174 | /* |
175 | * We default some events to a 1 default interval. But keep | |
176 | * it a weak assumption overridable by the user. | |
177 | */ | |
178 | if (!attr->sample_period || (user_freq != UINT_MAX && | |
3de29cab | 179 | user_interval != ULLONG_MAX)) { |
f9212819 FW |
180 | if (freq) { |
181 | attr->sample_type |= PERF_SAMPLE_PERIOD; | |
182 | attr->freq = 1; | |
183 | attr->sample_freq = freq; | |
184 | } else { | |
185 | attr->sample_period = default_interval; | |
186 | } | |
1dba15e7 | 187 | } |
3efa1cc9 | 188 | |
649c48a9 PZ |
189 | if (no_samples) |
190 | attr->sample_freq = 0; | |
191 | ||
192 | if (inherit_stat) | |
193 | attr->inherit_stat = 1; | |
194 | ||
3af9e859 | 195 | if (sample_address) { |
4bba828d | 196 | attr->sample_type |= PERF_SAMPLE_ADDR; |
3af9e859 EM |
197 | attr->mmap_data = track; |
198 | } | |
4bba828d | 199 | |
3efa1cc9 IM |
200 | if (call_graph) |
201 | attr->sample_type |= PERF_SAMPLE_CALLCHAIN; | |
202 | ||
f60f3593 AS |
203 | if (system_wide) |
204 | attr->sample_type |= PERF_SAMPLE_CPU; | |
205 | ||
a43d3f08 ACM |
206 | if (sample_id_all_avail && |
207 | (sample_time || system_wide || !no_inherit || cpu_list)) | |
9c90a61c ACM |
208 | attr->sample_type |= PERF_SAMPLE_TIME; |
209 | ||
cd6feeea | 210 | if (raw_samples) { |
6ddf259d | 211 | attr->sample_type |= PERF_SAMPLE_TIME; |
daac07b2 | 212 | attr->sample_type |= PERF_SAMPLE_RAW; |
cd6feeea IM |
213 | attr->sample_type |= PERF_SAMPLE_CPU; |
214 | } | |
f413cdb8 | 215 | |
acac03fa KS |
216 | if (nodelay) { |
217 | attr->watermark = 0; | |
218 | attr->wakeup_events = 1; | |
219 | } | |
220 | ||
a21ca2ca IM |
221 | attr->mmap = track; |
222 | attr->comm = track; | |
dd7927f4 | 223 | |
2e6cdf99 | 224 | if (target_pid == -1 && target_tid == -1 && !system_wide) { |
46be604b | 225 | attr->disabled = 1; |
bedbfdea | 226 | attr->enable_on_exec = 1; |
46be604b | 227 | } |
dd7927f4 | 228 | } |
bedbfdea | 229 | |
a91e5431 ACM |
230 | static bool perf_evlist__equal(struct perf_evlist *evlist, |
231 | struct perf_evlist *other) | |
232 | { | |
233 | struct perf_evsel *pos, *pair; | |
234 | ||
235 | if (evlist->nr_entries != other->nr_entries) | |
236 | return false; | |
237 | ||
238 | pair = list_entry(other->entries.next, struct perf_evsel, node); | |
239 | ||
240 | list_for_each_entry(pos, &evlist->entries, node) { | |
241 | if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0)) | |
242 | return false; | |
243 | pair = list_entry(pair->node.next, struct perf_evsel, node); | |
244 | } | |
245 | ||
246 | return true; | |
247 | } | |
248 | ||
dd7927f4 ACM |
249 | static void open_counters(struct perf_evlist *evlist) |
250 | { | |
251 | struct perf_evsel *pos; | |
dd7927f4 | 252 | |
5d2cd909 ACM |
253 | if (evlist->cpus->map[0] < 0) |
254 | no_inherit = true; | |
255 | ||
dd7927f4 ACM |
256 | list_for_each_entry(pos, &evlist->entries, node) { |
257 | struct perf_event_attr *attr = &pos->attr; | |
258 | /* | |
259 | * Check if parse_single_tracepoint_event has already asked for | |
260 | * PERF_SAMPLE_TIME. | |
261 | * | |
262 | * XXX this is kludgy but short term fix for problems introduced by | |
263 | * eac23d1c that broke 'perf script' by having different sample_types | |
264 | * when using multiple tracepoint events when we use a perf binary | |
265 | * that tries to use sample_id_all on an older kernel. | |
266 | * | |
267 | * We need to move counter creation to perf_session, support | |
268 | * different sample_types, etc. | |
269 | */ | |
270 | bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; | |
d6d901c2 | 271 | |
dd7927f4 ACM |
272 | config_attr(pos, evlist); |
273 | retry_sample_id: | |
274 | attr->sample_id_all = sample_id_all_avail ? 1 : 0; | |
275 | try_again: | |
5d2cd909 | 276 | if (perf_evsel__open(pos, evlist->cpus, evlist->threads, group) < 0) { |
d6d901c2 ZY |
277 | int err = errno; |
278 | ||
c286c419 ACM |
279 | if (err == EPERM || err == EACCES) { |
280 | ui__warning_paranoid(); | |
281 | exit(EXIT_FAILURE); | |
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) | |
ca6a4258 DA |
305 | ui__warning("The cycles event is not supported, " |
306 | "trying to fall back to cpu-clock-ticks\n"); | |
d6d901c2 ZY |
307 | attr->type = PERF_TYPE_SOFTWARE; |
308 | attr->config = PERF_COUNT_SW_CPU_CLOCK; | |
309 | goto try_again; | |
310 | } | |
ca6a4258 DA |
311 | |
312 | if (err == ENOENT) { | |
313 | ui__warning("The %s event is not supported.\n", | |
314 | event_name(pos)); | |
315 | exit(EXIT_FAILURE); | |
316 | } | |
317 | ||
d6d901c2 | 318 | printf("\n"); |
d9cf837e | 319 | error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", |
dd7927f4 | 320 | err, strerror(err)); |
bfd45118 SK |
321 | |
322 | #if defined(__i386__) || defined(__x86_64__) | |
d6d901c2 ZY |
323 | if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) |
324 | die("No hardware sampling interrupt available." | |
325 | " No APIC? If so then you can boot the kernel" | |
326 | " with the \"lapic\" boot parameter to" | |
327 | " force-enable it.\n"); | |
bfd45118 SK |
328 | #endif |
329 | ||
d6d901c2 | 330 | die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); |
c171b552 LZ |
331 | } |
332 | } | |
a43d3f08 | 333 | |
0a102479 FW |
334 | if (perf_evlist__set_filters(evlist)) { |
335 | error("failed to set filter with %d (%s)\n", errno, | |
336 | strerror(errno)); | |
337 | exit(-1); | |
338 | } | |
339 | ||
7e2ed097 | 340 | if (perf_evlist__mmap(evlist, mmap_pages, false) < 0) |
0a27d7f9 ACM |
341 | die("failed to mmap with %d (%s)\n", errno, strerror(errno)); |
342 | ||
a91e5431 ACM |
343 | if (file_new) |
344 | session->evlist = evlist; | |
345 | else { | |
346 | if (!perf_evlist__equal(session->evlist, evlist)) { | |
347 | fprintf(stderr, "incompatible append\n"); | |
348 | exit(-1); | |
349 | } | |
350 | } | |
351 | ||
352 | perf_session__update_sample_type(session); | |
16c8a109 PZ |
353 | } |
354 | ||
6122e4e4 ACM |
355 | static int process_buildids(void) |
356 | { | |
357 | u64 size = lseek(output, 0, SEEK_CUR); | |
358 | ||
9f591fd7 ACM |
359 | if (size == 0) |
360 | return 0; | |
361 | ||
6122e4e4 ACM |
362 | session->fd = output; |
363 | return __perf_session__process_events(session, post_processing_offset, | |
364 | size - post_processing_offset, | |
365 | size, &build_id__mark_dso_hit_ops); | |
366 | } | |
367 | ||
f5970550 PZ |
368 | static void atexit_header(void) |
369 | { | |
c7929e47 TZ |
370 | if (!pipe_output) { |
371 | session->header.data_size += bytes_written; | |
f5970550 | 372 | |
baa2f6ce ACM |
373 | if (!no_buildid) |
374 | process_buildids(); | |
a91e5431 | 375 | perf_session__write_header(session, evsel_list, output, true); |
39d17dac | 376 | perf_session__delete(session); |
361c99a6 | 377 | perf_evlist__delete(evsel_list); |
d65a458b | 378 | symbol__exit(); |
c7929e47 | 379 | } |
f5970550 PZ |
380 | } |
381 | ||
8115d60c | 382 | static void perf_event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
383 | { |
384 | int err; | |
23346f21 | 385 | struct perf_session *psession = data; |
a1645ce1 | 386 | |
23346f21 | 387 | if (machine__is_host(machine)) |
a1645ce1 ZY |
388 | return; |
389 | ||
390 | /* | |
391 | *As for guest kernel when processing subcommand record&report, | |
392 | *we arrange module mmap prior to guest kernel mmap and trigger | |
393 | *a preload dso because default guest module symbols are loaded | |
394 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
395 | *method is used to avoid symbol missing when the first addr is | |
396 | *in module instead of in guest kernel. | |
397 | */ | |
8115d60c ACM |
398 | err = perf_event__synthesize_modules(process_synthesized_event, |
399 | psession, machine); | |
a1645ce1 ZY |
400 | if (err < 0) |
401 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 402 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 403 | |
a1645ce1 ZY |
404 | /* |
405 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
406 | * have no _text sometimes. | |
407 | */ | |
8115d60c ACM |
408 | err = perf_event__synthesize_kernel_mmap(process_synthesized_event, |
409 | psession, machine, "_text"); | |
a1645ce1 | 410 | if (err < 0) |
8115d60c ACM |
411 | err = perf_event__synthesize_kernel_mmap(process_synthesized_event, |
412 | psession, machine, | |
413 | "_stext"); | |
a1645ce1 ZY |
414 | if (err < 0) |
415 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 416 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
417 | } |
418 | ||
98402807 FW |
419 | static struct perf_event_header finished_round_event = { |
420 | .size = sizeof(struct perf_event_header), | |
421 | .type = PERF_RECORD_FINISHED_ROUND, | |
422 | }; | |
423 | ||
424 | static void mmap_read_all(void) | |
425 | { | |
0e2e63dd | 426 | int i; |
98402807 | 427 | |
aece948f | 428 | for (i = 0; i < evsel_list->nr_mmaps; i++) { |
0a27d7f9 ACM |
429 | if (evsel_list->mmap[i].base) |
430 | mmap_read(&evsel_list->mmap[i]); | |
98402807 FW |
431 | } |
432 | ||
433 | if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO)) | |
434 | write_output(&finished_round_event, sizeof(finished_round_event)); | |
435 | } | |
436 | ||
d4db3f16 | 437 | static int __cmd_record(int argc, const char **argv) |
16c8a109 | 438 | { |
abaff32a | 439 | struct stat st; |
abaff32a | 440 | int flags; |
4dc0a04b | 441 | int err; |
8b412664 | 442 | unsigned long waking = 0; |
856e9660 | 443 | int child_ready_pipe[2], go_pipe[2]; |
46be604b | 444 | const bool forks = argc > 0; |
856e9660 | 445 | char buf; |
23346f21 | 446 | struct machine *machine; |
de9ac07b PZ |
447 | |
448 | page_size = sysconf(_SC_PAGE_SIZE); | |
de9ac07b | 449 | |
f5970550 PZ |
450 | atexit(sig_atexit); |
451 | signal(SIGCHLD, sig_handler); | |
452 | signal(SIGINT, sig_handler); | |
18483b81 | 453 | signal(SIGUSR1, sig_handler); |
f5970550 | 454 | |
d4db3f16 | 455 | if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { |
856e9660 PZ |
456 | perror("failed to create pipes"); |
457 | exit(-1); | |
458 | } | |
459 | ||
d7065adb FBH |
460 | if (!output_name) { |
461 | if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode)) | |
462 | pipe_output = 1; | |
463 | else | |
464 | output_name = "perf.data"; | |
465 | } | |
466 | if (output_name) { | |
467 | if (!strcmp(output_name, "-")) | |
468 | pipe_output = 1; | |
469 | else if (!stat(output_name, &st) && st.st_size) { | |
470 | if (write_mode == WRITE_FORCE) { | |
471 | char oldname[PATH_MAX]; | |
472 | snprintf(oldname, sizeof(oldname), "%s.old", | |
473 | output_name); | |
474 | unlink(oldname); | |
475 | rename(output_name, oldname); | |
476 | } | |
477 | } else if (write_mode == WRITE_APPEND) { | |
478 | write_mode = WRITE_FORCE; | |
266e0e21 | 479 | } |
97124d5e PZ |
480 | } |
481 | ||
f887f301 | 482 | flags = O_CREAT|O_RDWR; |
7865e817 | 483 | if (write_mode == WRITE_APPEND) |
f5970550 | 484 | file_new = 0; |
abaff32a IM |
485 | else |
486 | flags |= O_TRUNC; | |
487 | ||
529870e3 TZ |
488 | if (pipe_output) |
489 | output = STDOUT_FILENO; | |
490 | else | |
491 | output = open(output_name, flags, S_IRUSR | S_IWUSR); | |
de9ac07b PZ |
492 | if (output < 0) { |
493 | perror("failed to create output file"); | |
494 | exit(-1); | |
495 | } | |
496 | ||
7865e817 | 497 | session = perf_session__new(output_name, O_WRONLY, |
21ef97f0 | 498 | write_mode == WRITE_FORCE, false, NULL); |
94c744b6 | 499 | if (session == NULL) { |
a9a70bbc ACM |
500 | pr_err("Not enough memory for reading perf file header\n"); |
501 | return -1; | |
502 | } | |
503 | ||
baa2f6ce ACM |
504 | if (!no_buildid) |
505 | perf_header__set_feat(&session->header, HEADER_BUILD_ID); | |
506 | ||
4dc0a04b | 507 | if (!file_new) { |
a91e5431 | 508 | err = perf_session__read_header(session, output); |
4dc0a04b | 509 | if (err < 0) |
39d17dac | 510 | goto out_delete_session; |
4dc0a04b ACM |
511 | } |
512 | ||
361c99a6 | 513 | if (have_tracepoints(&evsel_list->entries)) |
94c744b6 | 514 | perf_header__set_feat(&session->header, HEADER_TRACE_INFO); |
03456a15 | 515 | |
800cd25c FW |
516 | /* 512 kiB: default amount of unprivileged mlocked memory */ |
517 | if (mmap_pages == UINT_MAX) | |
518 | mmap_pages = (512 * 1024) / page_size; | |
519 | ||
d4db3f16 | 520 | if (forks) { |
46be604b | 521 | child_pid = fork(); |
2fb750e8 | 522 | if (child_pid < 0) { |
856e9660 PZ |
523 | perror("failed to fork"); |
524 | exit(-1); | |
525 | } | |
7c6a1c65 | 526 | |
46be604b | 527 | if (!child_pid) { |
529870e3 TZ |
528 | if (pipe_output) |
529 | dup2(2, 1); | |
856e9660 PZ |
530 | close(child_ready_pipe[0]); |
531 | close(go_pipe[1]); | |
532 | fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC); | |
533 | ||
534 | /* | |
535 | * Do a dummy execvp to get the PLT entry resolved, | |
536 | * so we avoid the resolver overhead on the real | |
537 | * execvp call. | |
538 | */ | |
539 | execvp("", (char **)argv); | |
540 | ||
541 | /* | |
542 | * Tell the parent we're ready to go | |
543 | */ | |
544 | close(child_ready_pipe[1]); | |
545 | ||
546 | /* | |
547 | * Wait until the parent tells us to go. | |
548 | */ | |
549 | if (read(go_pipe[0], &buf, 1) == -1) | |
550 | perror("unable to read pipe"); | |
551 | ||
552 | execvp(argv[0], (char **)argv); | |
553 | ||
554 | perror(argv[0]); | |
18483b81 | 555 | kill(getppid(), SIGUSR1); |
856e9660 | 556 | exit(-1); |
0a5ac846 | 557 | } |
856e9660 | 558 | |
d6d901c2 | 559 | if (!system_wide && target_tid == -1 && target_pid == -1) |
7e2ed097 | 560 | evsel_list->threads->map[0] = child_pid; |
d6d901c2 | 561 | |
856e9660 PZ |
562 | close(child_ready_pipe[1]); |
563 | close(go_pipe[0]); | |
564 | /* | |
565 | * wait for child to settle | |
566 | */ | |
567 | if (read(child_ready_pipe[0], &buf, 1) == -1) { | |
568 | perror("unable to read pipe"); | |
569 | exit(-1); | |
570 | } | |
571 | close(child_ready_pipe[0]); | |
572 | } | |
573 | ||
dd7927f4 | 574 | open_counters(evsel_list); |
de9ac07b | 575 | |
712a4b60 ACM |
576 | /* |
577 | * perf_session__delete(session) will be called at atexit_header() | |
578 | */ | |
579 | atexit(atexit_header); | |
580 | ||
529870e3 TZ |
581 | if (pipe_output) { |
582 | err = perf_header__write_pipe(output); | |
583 | if (err < 0) | |
584 | return err; | |
585 | } else if (file_new) { | |
a91e5431 ACM |
586 | err = perf_session__write_header(session, evsel_list, |
587 | output, false); | |
d5eed904 ACM |
588 | if (err < 0) |
589 | return err; | |
56b03f3c ACM |
590 | } |
591 | ||
6122e4e4 ACM |
592 | post_processing_offset = lseek(output, 0, SEEK_CUR); |
593 | ||
2c46dbb5 | 594 | if (pipe_output) { |
a91e5431 ACM |
595 | err = perf_session__synthesize_attrs(session, |
596 | process_synthesized_event); | |
2c46dbb5 TZ |
597 | if (err < 0) { |
598 | pr_err("Couldn't synthesize attrs.\n"); | |
599 | return err; | |
600 | } | |
cd19a035 | 601 | |
8115d60c ACM |
602 | err = perf_event__synthesize_event_types(process_synthesized_event, |
603 | session); | |
cd19a035 TZ |
604 | if (err < 0) { |
605 | pr_err("Couldn't synthesize event_types.\n"); | |
606 | return err; | |
607 | } | |
9215545e | 608 | |
361c99a6 | 609 | if (have_tracepoints(&evsel_list->entries)) { |
63e0c771 TZ |
610 | /* |
611 | * FIXME err <= 0 here actually means that | |
612 | * there were no tracepoints so its not really | |
613 | * an error, just that we don't need to | |
614 | * synthesize anything. We really have to | |
615 | * return this more properly and also | |
616 | * propagate errors that now are calling die() | |
617 | */ | |
8115d60c ACM |
618 | err = perf_event__synthesize_tracing_data(output, evsel_list, |
619 | process_synthesized_event, | |
620 | session); | |
63e0c771 TZ |
621 | if (err <= 0) { |
622 | pr_err("Couldn't record tracing data.\n"); | |
623 | return err; | |
624 | } | |
2c9faa06 | 625 | advance_output(err); |
63e0c771 | 626 | } |
2c46dbb5 TZ |
627 | } |
628 | ||
23346f21 ACM |
629 | machine = perf_session__find_host_machine(session); |
630 | if (!machine) { | |
a1645ce1 ZY |
631 | pr_err("Couldn't find native kernel information.\n"); |
632 | return -1; | |
633 | } | |
634 | ||
8115d60c ACM |
635 | err = perf_event__synthesize_kernel_mmap(process_synthesized_event, |
636 | session, machine, "_text"); | |
70162138 | 637 | if (err < 0) |
8115d60c ACM |
638 | err = perf_event__synthesize_kernel_mmap(process_synthesized_event, |
639 | session, machine, "_stext"); | |
c1a3a4b9 ACM |
640 | if (err < 0) |
641 | pr_err("Couldn't record kernel reference relocation symbol\n" | |
642 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
643 | "Check /proc/kallsyms permission or run as root.\n"); | |
b7cece76 | 644 | |
8115d60c ACM |
645 | err = perf_event__synthesize_modules(process_synthesized_event, |
646 | session, machine); | |
c1a3a4b9 ACM |
647 | if (err < 0) |
648 | pr_err("Couldn't record kernel module information.\n" | |
649 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
650 | "Check /proc/modules permission or run as root.\n"); | |
651 | ||
a1645ce1 | 652 | if (perf_guest) |
8115d60c ACM |
653 | perf_session__process_machines(session, |
654 | perf_event__synthesize_guest_os); | |
7c6a1c65 | 655 | |
cf103a14 | 656 | if (!system_wide) |
7c940c18 ACM |
657 | perf_event__synthesize_thread_map(evsel_list->threads, |
658 | process_synthesized_event, | |
659 | session); | |
234fbbf5 | 660 | else |
8115d60c ACM |
661 | perf_event__synthesize_threads(process_synthesized_event, |
662 | session); | |
7c6a1c65 | 663 | |
de9ac07b PZ |
664 | if (realtime_prio) { |
665 | struct sched_param param; | |
666 | ||
667 | param.sched_priority = realtime_prio; | |
668 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { | |
6beba7ad | 669 | pr_err("Could not set realtime priority.\n"); |
de9ac07b PZ |
670 | exit(-1); |
671 | } | |
672 | } | |
673 | ||
856e9660 PZ |
674 | /* |
675 | * Let the child rip | |
676 | */ | |
d4db3f16 ACM |
677 | if (forks) |
678 | close(go_pipe[1]); | |
856e9660 | 679 | |
649c48a9 | 680 | for (;;) { |
2debbc83 | 681 | int hits = samples; |
de9ac07b | 682 | |
98402807 | 683 | mmap_read_all(); |
de9ac07b | 684 | |
649c48a9 PZ |
685 | if (hits == samples) { |
686 | if (done) | |
687 | break; | |
5c581041 | 688 | err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1); |
8b412664 PZ |
689 | waking++; |
690 | } | |
691 | ||
4152ab37 ACM |
692 | if (done) |
693 | perf_evlist__disable(evsel_list); | |
de9ac07b PZ |
694 | } |
695 | ||
18483b81 | 696 | if (quiet || signr == SIGUSR1) |
b44308f5 ACM |
697 | return 0; |
698 | ||
8b412664 PZ |
699 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
700 | ||
021e9f47 IM |
701 | /* |
702 | * Approximate RIP event size: 24 bytes. | |
703 | */ | |
704 | fprintf(stderr, | |
9486aa38 | 705 | "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n", |
021e9f47 IM |
706 | (double)bytes_written / 1024.0 / 1024.0, |
707 | output_name, | |
708 | bytes_written / 24); | |
addc2785 | 709 | |
de9ac07b | 710 | return 0; |
39d17dac ACM |
711 | |
712 | out_delete_session: | |
713 | perf_session__delete(session); | |
714 | return err; | |
de9ac07b | 715 | } |
0e9b20b8 | 716 | |
0e9b20b8 | 717 | static const char * const record_usage[] = { |
9e096753 MG |
718 | "perf record [<options>] [<command>]", |
719 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
720 | NULL |
721 | }; | |
722 | ||
7865e817 FW |
723 | static bool force, append_file; |
724 | ||
bca647aa | 725 | const struct option record_options[] = { |
361c99a6 | 726 | OPT_CALLBACK('e', "event", &evsel_list, "event", |
86847b62 | 727 | "event selector. use 'perf list' to list available events", |
f120f9d5 | 728 | parse_events_option), |
361c99a6 | 729 | OPT_CALLBACK(0, "filter", &evsel_list, "filter", |
c171b552 | 730 | "event filter", parse_filter), |
0e9b20b8 | 731 | OPT_INTEGER('p', "pid", &target_pid, |
d6d901c2 ZY |
732 | "record events on existing process id"), |
733 | OPT_INTEGER('t', "tid", &target_tid, | |
734 | "record events on existing thread id"), | |
0e9b20b8 IM |
735 | OPT_INTEGER('r', "realtime", &realtime_prio, |
736 | "collect data with this RT SCHED_FIFO priority"), | |
acac03fa KS |
737 | OPT_BOOLEAN('D', "no-delay", &nodelay, |
738 | "collect data without buffering"), | |
daac07b2 FW |
739 | OPT_BOOLEAN('R', "raw-samples", &raw_samples, |
740 | "collect raw sample records from all opened counters"), | |
0e9b20b8 IM |
741 | OPT_BOOLEAN('a', "all-cpus", &system_wide, |
742 | "system-wide collection from all CPUs"), | |
abaff32a IM |
743 | OPT_BOOLEAN('A', "append", &append_file, |
744 | "append to the output file to do incremental profiling"), | |
c45c6ea2 SE |
745 | OPT_STRING('C', "cpu", &cpu_list, "cpu", |
746 | "list of cpus to monitor"), | |
97124d5e | 747 | OPT_BOOLEAN('f', "force", &force, |
7865e817 | 748 | "overwrite existing data file (deprecated)"), |
3de29cab | 749 | OPT_U64('c', "count", &user_interval, "event period to sample"), |
abaff32a IM |
750 | OPT_STRING('o', "output", &output_name, "file", |
751 | "output file name"), | |
2e6cdf99 SE |
752 | OPT_BOOLEAN('i', "no-inherit", &no_inherit, |
753 | "child tasks do not inherit counters"), | |
1967936d ACM |
754 | OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"), |
755 | OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), | |
43bece79 LM |
756 | OPT_BOOLEAN(0, "group", &group, |
757 | "put the counters into a counter group"), | |
3efa1cc9 IM |
758 | OPT_BOOLEAN('g', "call-graph", &call_graph, |
759 | "do call-graph (stack chain/backtrace) recording"), | |
c0555642 | 760 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 761 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 762 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
649c48a9 PZ |
763 | OPT_BOOLEAN('s', "stat", &inherit_stat, |
764 | "per thread counts"), | |
4bba828d AB |
765 | OPT_BOOLEAN('d', "data", &sample_address, |
766 | "Sample addresses"), | |
9c90a61c | 767 | OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"), |
649c48a9 PZ |
768 | OPT_BOOLEAN('n', "no-samples", &no_samples, |
769 | "don't sample"), | |
baa2f6ce | 770 | OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache, |
a1ac1d3c | 771 | "do not update the buildid cache"), |
baa2f6ce ACM |
772 | OPT_BOOLEAN('B', "no-buildid", &no_buildid, |
773 | "do not collect buildids in perf.data"), | |
023695d9 SE |
774 | OPT_CALLBACK('G', "cgroup", &evsel_list, "name", |
775 | "monitor event in cgroup name only", | |
776 | parse_cgroups), | |
0e9b20b8 IM |
777 | OPT_END() |
778 | }; | |
779 | ||
f37a291c | 780 | int cmd_record(int argc, const char **argv, const char *prefix __used) |
0e9b20b8 | 781 | { |
69aad6f1 ACM |
782 | int err = -ENOMEM; |
783 | struct perf_evsel *pos; | |
0e9b20b8 | 784 | |
7e2ed097 | 785 | evsel_list = perf_evlist__new(NULL, NULL); |
361c99a6 ACM |
786 | if (evsel_list == NULL) |
787 | return -ENOMEM; | |
788 | ||
bca647aa | 789 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 790 | PARSE_OPT_STOP_AT_NON_OPTION); |
d6d901c2 | 791 | if (!argc && target_pid == -1 && target_tid == -1 && |
c45c6ea2 | 792 | !system_wide && !cpu_list) |
bca647aa | 793 | usage_with_options(record_usage, record_options); |
0e9b20b8 | 794 | |
7865e817 FW |
795 | if (force && append_file) { |
796 | fprintf(stderr, "Can't overwrite and append at the same time." | |
797 | " You need to choose between -f and -A"); | |
bca647aa | 798 | usage_with_options(record_usage, record_options); |
7865e817 FW |
799 | } else if (append_file) { |
800 | write_mode = WRITE_APPEND; | |
801 | } else { | |
802 | write_mode = WRITE_FORCE; | |
803 | } | |
804 | ||
023695d9 SE |
805 | if (nr_cgroups && !system_wide) { |
806 | fprintf(stderr, "cgroup monitoring only available in" | |
807 | " system-wide mode\n"); | |
808 | usage_with_options(record_usage, record_options); | |
809 | } | |
810 | ||
655000e7 | 811 | symbol__init(); |
baa2f6ce | 812 | |
ec80fde7 | 813 | if (symbol_conf.kptr_restrict) |
646aaea6 ACM |
814 | pr_warning( |
815 | "WARNING: Kernel address maps (/proc/{kallsyms,modules}) are restricted,\n" | |
816 | "check /proc/sys/kernel/kptr_restrict.\n\n" | |
817 | "Samples in kernel functions may not be resolved if a suitable vmlinux\n" | |
818 | "file is not found in the buildid cache or in the vmlinux path.\n\n" | |
819 | "Samples in kernel modules won't be resolved at all.\n\n" | |
820 | "If some relocation was applied (e.g. kexec) symbols may be misresolved\n" | |
821 | "even with a suitable vmlinux or kallsyms file.\n\n"); | |
ec80fde7 | 822 | |
baa2f6ce | 823 | if (no_buildid_cache || no_buildid) |
a1ac1d3c | 824 | disable_buildid_cache(); |
655000e7 | 825 | |
361c99a6 ACM |
826 | if (evsel_list->nr_entries == 0 && |
827 | perf_evlist__add_default(evsel_list) < 0) { | |
69aad6f1 ACM |
828 | pr_err("Not enough memory for event selector list\n"); |
829 | goto out_symbol_exit; | |
bbd36e5e | 830 | } |
0e9b20b8 | 831 | |
5c98d466 | 832 | if (target_pid != -1) |
d6d901c2 | 833 | target_tid = target_pid; |
d6d901c2 | 834 | |
7e2ed097 ACM |
835 | if (perf_evlist__create_maps(evsel_list, target_pid, |
836 | target_tid, cpu_list) < 0) | |
dd7927f4 | 837 | usage_with_options(record_usage, record_options); |
69aad6f1 | 838 | |
361c99a6 | 839 | list_for_each_entry(pos, &evsel_list->entries, node) { |
7e2ed097 ACM |
840 | if (perf_evsel__alloc_fd(pos, evsel_list->cpus->nr, |
841 | evsel_list->threads->nr) < 0) | |
69aad6f1 | 842 | goto out_free_fd; |
ad7f4e3f ACM |
843 | if (perf_header__push_event(pos->attr.config, event_name(pos))) |
844 | goto out_free_fd; | |
d6d901c2 | 845 | } |
5c581041 | 846 | |
7e2ed097 | 847 | if (perf_evlist__alloc_pollfd(evsel_list) < 0) |
39d17dac | 848 | goto out_free_fd; |
d6d901c2 | 849 | |
3de29cab | 850 | if (user_interval != ULLONG_MAX) |
f9212819 FW |
851 | default_interval = user_interval; |
852 | if (user_freq != UINT_MAX) | |
853 | freq = user_freq; | |
854 | ||
7e4ff9e3 MG |
855 | /* |
856 | * User specified count overrides default frequency. | |
857 | */ | |
858 | if (default_interval) | |
859 | freq = 0; | |
860 | else if (freq) { | |
861 | default_interval = freq; | |
862 | } else { | |
863 | fprintf(stderr, "frequency and count are zero, aborting\n"); | |
39d17dac | 864 | err = -EINVAL; |
5c581041 | 865 | goto out_free_fd; |
7e4ff9e3 MG |
866 | } |
867 | ||
39d17dac | 868 | err = __cmd_record(argc, argv); |
39d17dac | 869 | out_free_fd: |
7e2ed097 | 870 | perf_evlist__delete_maps(evsel_list); |
d65a458b ACM |
871 | out_symbol_exit: |
872 | symbol__exit(); | |
39d17dac | 873 | return err; |
0e9b20b8 | 874 | } |