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" |
45694aa7 | 25 | #include "util/tool.h" |
8d06367f | 26 | #include "util/symbol.h" |
a12b51c4 | 27 | #include "util/cpumap.h" |
fd78260b | 28 | #include "util/thread_map.h" |
7c6a1c65 | 29 | |
97124d5e | 30 | #include <unistd.h> |
de9ac07b | 31 | #include <sched.h> |
a41794cd | 32 | #include <sys/mman.h> |
de9ac07b | 33 | |
7865e817 FW |
34 | enum write_mode_t { |
35 | WRITE_FORCE, | |
36 | WRITE_APPEND | |
37 | }; | |
38 | ||
d20deb64 | 39 | struct perf_record { |
45694aa7 | 40 | struct perf_tool tool; |
d20deb64 ACM |
41 | struct perf_record_opts opts; |
42 | u64 bytes_written; | |
43 | const char *output_name; | |
44 | struct perf_evlist *evlist; | |
45 | struct perf_session *session; | |
46 | const char *progname; | |
0d37aa34 | 47 | const char *uid_str; |
d20deb64 ACM |
48 | int output; |
49 | unsigned int page_size; | |
50 | int realtime_prio; | |
51 | enum write_mode_t write_mode; | |
52 | bool no_buildid; | |
53 | bool no_buildid_cache; | |
54 | bool force; | |
55 | bool file_new; | |
56 | bool append_file; | |
57 | long samples; | |
58 | off_t post_processing_offset; | |
0f82ebc4 | 59 | }; |
a21ca2ca | 60 | |
d20deb64 | 61 | static void advance_output(struct perf_record *rec, size_t size) |
9215545e | 62 | { |
d20deb64 | 63 | rec->bytes_written += size; |
9215545e TZ |
64 | } |
65 | ||
d20deb64 | 66 | static void write_output(struct perf_record *rec, void *buf, size_t size) |
f5970550 PZ |
67 | { |
68 | while (size) { | |
d20deb64 | 69 | int ret = write(rec->output, buf, size); |
f5970550 PZ |
70 | |
71 | if (ret < 0) | |
72 | die("failed to write"); | |
73 | ||
74 | size -= ret; | |
75 | buf += ret; | |
76 | ||
d20deb64 | 77 | rec->bytes_written += ret; |
f5970550 PZ |
78 | } |
79 | } | |
80 | ||
45694aa7 | 81 | static int process_synthesized_event(struct perf_tool *tool, |
d20deb64 | 82 | union perf_event *event, |
8d50e5b4 | 83 | struct perf_sample *sample __used, |
743eb868 | 84 | struct machine *machine __used) |
234fbbf5 | 85 | { |
45694aa7 | 86 | struct perf_record *rec = container_of(tool, struct perf_record, tool); |
d20deb64 | 87 | write_output(rec, event, event->header.size); |
234fbbf5 ACM |
88 | return 0; |
89 | } | |
90 | ||
d20deb64 ACM |
91 | static void perf_record__mmap_read(struct perf_record *rec, |
92 | struct perf_mmap *md) | |
de9ac07b | 93 | { |
744bd8aa | 94 | unsigned int head = perf_mmap__read_head(md); |
de9ac07b | 95 | unsigned int old = md->prev; |
d20deb64 | 96 | unsigned char *data = md->base + rec->page_size; |
de9ac07b PZ |
97 | unsigned long size; |
98 | void *buf; | |
de9ac07b | 99 | |
dc82009a ACM |
100 | if (old == head) |
101 | return; | |
102 | ||
d20deb64 | 103 | rec->samples++; |
de9ac07b PZ |
104 | |
105 | size = head - old; | |
106 | ||
107 | if ((old & md->mask) + size != (head & md->mask)) { | |
108 | buf = &data[old & md->mask]; | |
109 | size = md->mask + 1 - (old & md->mask); | |
110 | old += size; | |
021e9f47 | 111 | |
d20deb64 | 112 | write_output(rec, buf, size); |
de9ac07b PZ |
113 | } |
114 | ||
115 | buf = &data[old & md->mask]; | |
116 | size = head - old; | |
117 | old += size; | |
021e9f47 | 118 | |
d20deb64 | 119 | write_output(rec, buf, size); |
de9ac07b PZ |
120 | |
121 | md->prev = old; | |
115d2d89 | 122 | perf_mmap__write_tail(md, old); |
de9ac07b PZ |
123 | } |
124 | ||
125 | static volatile int done = 0; | |
f7b7c26e | 126 | static volatile int signr = -1; |
33e49ea7 | 127 | static volatile int child_finished = 0; |
de9ac07b | 128 | |
16c8a109 | 129 | static void sig_handler(int sig) |
de9ac07b | 130 | { |
33e49ea7 AK |
131 | if (sig == SIGCHLD) |
132 | child_finished = 1; | |
133 | ||
16c8a109 | 134 | done = 1; |
f7b7c26e PZ |
135 | signr = sig; |
136 | } | |
137 | ||
d20deb64 | 138 | static void perf_record__sig_exit(int exit_status __used, void *arg) |
f7b7c26e | 139 | { |
d20deb64 | 140 | struct perf_record *rec = arg; |
33e49ea7 AK |
141 | int status; |
142 | ||
d20deb64 | 143 | if (rec->evlist->workload.pid > 0) { |
33e49ea7 | 144 | if (!child_finished) |
d20deb64 | 145 | kill(rec->evlist->workload.pid, SIGTERM); |
33e49ea7 AK |
146 | |
147 | wait(&status); | |
148 | if (WIFSIGNALED(status)) | |
d20deb64 | 149 | psignal(WTERMSIG(status), rec->progname); |
33e49ea7 | 150 | } |
933da83a | 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 | ||
a91e5431 ACM |
159 | static bool perf_evlist__equal(struct perf_evlist *evlist, |
160 | struct perf_evlist *other) | |
161 | { | |
162 | struct perf_evsel *pos, *pair; | |
163 | ||
164 | if (evlist->nr_entries != other->nr_entries) | |
165 | return false; | |
166 | ||
167 | pair = list_entry(other->entries.next, struct perf_evsel, node); | |
168 | ||
169 | list_for_each_entry(pos, &evlist->entries, node) { | |
170 | if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0)) | |
171 | return false; | |
172 | pair = list_entry(pair->node.next, struct perf_evsel, node); | |
173 | } | |
174 | ||
175 | return true; | |
176 | } | |
177 | ||
d20deb64 | 178 | static void perf_record__open(struct perf_record *rec) |
dd7927f4 | 179 | { |
727ab04e | 180 | struct perf_evsel *pos, *first; |
d20deb64 ACM |
181 | struct perf_evlist *evlist = rec->evlist; |
182 | struct perf_session *session = rec->session; | |
183 | struct perf_record_opts *opts = &rec->opts; | |
dd7927f4 | 184 | |
727ab04e ACM |
185 | first = list_entry(evlist->entries.next, struct perf_evsel, node); |
186 | ||
d20deb64 | 187 | perf_evlist__config_attrs(evlist, opts); |
0f82ebc4 | 188 | |
dd7927f4 ACM |
189 | list_for_each_entry(pos, &evlist->entries, node) { |
190 | struct perf_event_attr *attr = &pos->attr; | |
727ab04e | 191 | struct xyarray *group_fd = NULL; |
dd7927f4 ACM |
192 | /* |
193 | * Check if parse_single_tracepoint_event has already asked for | |
194 | * PERF_SAMPLE_TIME. | |
195 | * | |
196 | * XXX this is kludgy but short term fix for problems introduced by | |
197 | * eac23d1c that broke 'perf script' by having different sample_types | |
198 | * when using multiple tracepoint events when we use a perf binary | |
199 | * that tries to use sample_id_all on an older kernel. | |
200 | * | |
201 | * We need to move counter creation to perf_session, support | |
202 | * different sample_types, etc. | |
203 | */ | |
204 | bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; | |
d6d901c2 | 205 | |
d20deb64 | 206 | if (opts->group && pos != first) |
727ab04e | 207 | group_fd = first->fd; |
0c978128 ACM |
208 | fallback_missing_features: |
209 | if (opts->exclude_guest_missing) | |
210 | attr->exclude_guest = attr->exclude_host = 0; | |
dd7927f4 | 211 | retry_sample_id: |
808e1226 | 212 | attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1; |
dd7927f4 | 213 | try_again: |
ed80f581 | 214 | if (perf_evsel__open(pos, evlist->cpus, evlist->threads, |
d20deb64 | 215 | opts->group, group_fd) < 0) { |
d6d901c2 ZY |
216 | int err = errno; |
217 | ||
c286c419 | 218 | if (err == EPERM || err == EACCES) { |
b8631e6e | 219 | ui__error_paranoid(); |
c286c419 | 220 | exit(EXIT_FAILURE); |
d20deb64 | 221 | } else if (err == ENODEV && opts->cpu_list) { |
d6d901c2 ZY |
222 | die("No such device - did you specify" |
223 | " an out-of-range profile CPU?\n"); | |
0c978128 ACM |
224 | } else if (err == EINVAL) { |
225 | if (!opts->exclude_guest_missing && | |
226 | (attr->exclude_guest || attr->exclude_host)) { | |
227 | pr_debug("Old kernel, cannot exclude " | |
228 | "guest or host samples.\n"); | |
229 | opts->exclude_guest_missing = true; | |
230 | goto fallback_missing_features; | |
808e1226 | 231 | } else if (!opts->sample_id_all_missing) { |
0c978128 ACM |
232 | /* |
233 | * Old kernel, no attr->sample_id_type_all field | |
234 | */ | |
808e1226 | 235 | opts->sample_id_all_missing = true; |
0c978128 ACM |
236 | if (!opts->sample_time && !opts->raw_samples && !time_needed) |
237 | attr->sample_type &= ~PERF_SAMPLE_TIME; | |
238 | ||
239 | goto retry_sample_id; | |
240 | } | |
d6d901c2 | 241 | } |
3da297a6 | 242 | |
d6d901c2 ZY |
243 | /* |
244 | * If it's cycles then fall back to hrtimer | |
245 | * based cpu-clock-tick sw counter, which | |
246 | * is always available even if no PMU support: | |
247 | */ | |
248 | if (attr->type == PERF_TYPE_HARDWARE | |
249 | && attr->config == PERF_COUNT_HW_CPU_CYCLES) { | |
250 | ||
251 | if (verbose) | |
ca6a4258 DA |
252 | ui__warning("The cycles event is not supported, " |
253 | "trying to fall back to cpu-clock-ticks\n"); | |
d6d901c2 ZY |
254 | attr->type = PERF_TYPE_SOFTWARE; |
255 | attr->config = PERF_COUNT_SW_CPU_CLOCK; | |
256 | goto try_again; | |
257 | } | |
ca6a4258 DA |
258 | |
259 | if (err == ENOENT) { | |
260 | ui__warning("The %s event is not supported.\n", | |
261 | event_name(pos)); | |
262 | exit(EXIT_FAILURE); | |
263 | } | |
264 | ||
d6d901c2 | 265 | printf("\n"); |
d9cf837e | 266 | error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", |
dd7927f4 | 267 | err, strerror(err)); |
bfd45118 SK |
268 | |
269 | #if defined(__i386__) || defined(__x86_64__) | |
d6d901c2 ZY |
270 | if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) |
271 | die("No hardware sampling interrupt available." | |
272 | " No APIC? If so then you can boot the kernel" | |
273 | " with the \"lapic\" boot parameter to" | |
274 | " force-enable it.\n"); | |
bfd45118 SK |
275 | #endif |
276 | ||
d6d901c2 | 277 | die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); |
c171b552 LZ |
278 | } |
279 | } | |
a43d3f08 | 280 | |
0a102479 FW |
281 | if (perf_evlist__set_filters(evlist)) { |
282 | error("failed to set filter with %d (%s)\n", errno, | |
283 | strerror(errno)); | |
284 | exit(-1); | |
285 | } | |
286 | ||
18e60939 NE |
287 | if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) { |
288 | if (errno == EPERM) | |
289 | die("Permission error mapping pages.\n" | |
290 | "Consider increasing " | |
291 | "/proc/sys/kernel/perf_event_mlock_kb,\n" | |
292 | "or try again with a smaller value of -m/--mmap_pages.\n" | |
293 | "(current value: %d)\n", opts->mmap_pages); | |
41d0d933 NE |
294 | else if (!is_power_of_2(opts->mmap_pages)) |
295 | die("--mmap_pages/-m value must be a power of two."); | |
296 | ||
0a27d7f9 | 297 | die("failed to mmap with %d (%s)\n", errno, strerror(errno)); |
18e60939 | 298 | } |
0a27d7f9 | 299 | |
d20deb64 | 300 | if (rec->file_new) |
a91e5431 ACM |
301 | session->evlist = evlist; |
302 | else { | |
303 | if (!perf_evlist__equal(session->evlist, evlist)) { | |
304 | fprintf(stderr, "incompatible append\n"); | |
305 | exit(-1); | |
306 | } | |
307 | } | |
308 | ||
309 | perf_session__update_sample_type(session); | |
16c8a109 PZ |
310 | } |
311 | ||
d20deb64 | 312 | static int process_buildids(struct perf_record *rec) |
6122e4e4 | 313 | { |
d20deb64 | 314 | u64 size = lseek(rec->output, 0, SEEK_CUR); |
6122e4e4 | 315 | |
9f591fd7 ACM |
316 | if (size == 0) |
317 | return 0; | |
318 | ||
d20deb64 ACM |
319 | rec->session->fd = rec->output; |
320 | return __perf_session__process_events(rec->session, rec->post_processing_offset, | |
321 | size - rec->post_processing_offset, | |
6122e4e4 ACM |
322 | size, &build_id__mark_dso_hit_ops); |
323 | } | |
324 | ||
d20deb64 | 325 | static void perf_record__exit(int status __used, void *arg) |
f5970550 | 326 | { |
d20deb64 ACM |
327 | struct perf_record *rec = arg; |
328 | ||
329 | if (!rec->opts.pipe_output) { | |
330 | rec->session->header.data_size += rec->bytes_written; | |
331 | ||
332 | if (!rec->no_buildid) | |
333 | process_buildids(rec); | |
334 | perf_session__write_header(rec->session, rec->evlist, | |
335 | rec->output, true); | |
336 | perf_session__delete(rec->session); | |
337 | perf_evlist__delete(rec->evlist); | |
d65a458b | 338 | symbol__exit(); |
c7929e47 | 339 | } |
f5970550 PZ |
340 | } |
341 | ||
8115d60c | 342 | static void perf_event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
343 | { |
344 | int err; | |
45694aa7 | 345 | struct perf_tool *tool = data; |
a1645ce1 | 346 | |
23346f21 | 347 | if (machine__is_host(machine)) |
a1645ce1 ZY |
348 | return; |
349 | ||
350 | /* | |
351 | *As for guest kernel when processing subcommand record&report, | |
352 | *we arrange module mmap prior to guest kernel mmap and trigger | |
353 | *a preload dso because default guest module symbols are loaded | |
354 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
355 | *method is used to avoid symbol missing when the first addr is | |
356 | *in module instead of in guest kernel. | |
357 | */ | |
45694aa7 | 358 | err = perf_event__synthesize_modules(tool, process_synthesized_event, |
743eb868 | 359 | machine); |
a1645ce1 ZY |
360 | if (err < 0) |
361 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 362 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 363 | |
a1645ce1 ZY |
364 | /* |
365 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
366 | * have no _text sometimes. | |
367 | */ | |
45694aa7 | 368 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 369 | machine, "_text"); |
a1645ce1 | 370 | if (err < 0) |
45694aa7 | 371 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 372 | machine, "_stext"); |
a1645ce1 ZY |
373 | if (err < 0) |
374 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 375 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
376 | } |
377 | ||
98402807 FW |
378 | static struct perf_event_header finished_round_event = { |
379 | .size = sizeof(struct perf_event_header), | |
380 | .type = PERF_RECORD_FINISHED_ROUND, | |
381 | }; | |
382 | ||
d20deb64 | 383 | static void perf_record__mmap_read_all(struct perf_record *rec) |
98402807 | 384 | { |
0e2e63dd | 385 | int i; |
98402807 | 386 | |
d20deb64 ACM |
387 | for (i = 0; i < rec->evlist->nr_mmaps; i++) { |
388 | if (rec->evlist->mmap[i].base) | |
389 | perf_record__mmap_read(rec, &rec->evlist->mmap[i]); | |
98402807 FW |
390 | } |
391 | ||
d20deb64 ACM |
392 | if (perf_header__has_feat(&rec->session->header, HEADER_TRACE_INFO)) |
393 | write_output(rec, &finished_round_event, sizeof(finished_round_event)); | |
98402807 FW |
394 | } |
395 | ||
d20deb64 | 396 | static int __cmd_record(struct perf_record *rec, int argc, const char **argv) |
16c8a109 | 397 | { |
abaff32a | 398 | struct stat st; |
abaff32a | 399 | int flags; |
781ba9d2 | 400 | int err, output, feat; |
8b412664 | 401 | unsigned long waking = 0; |
46be604b | 402 | const bool forks = argc > 0; |
23346f21 | 403 | struct machine *machine; |
45694aa7 | 404 | struct perf_tool *tool = &rec->tool; |
d20deb64 ACM |
405 | struct perf_record_opts *opts = &rec->opts; |
406 | struct perf_evlist *evsel_list = rec->evlist; | |
407 | const char *output_name = rec->output_name; | |
408 | struct perf_session *session; | |
de9ac07b | 409 | |
d20deb64 | 410 | rec->progname = argv[0]; |
33e49ea7 | 411 | |
d20deb64 | 412 | rec->page_size = sysconf(_SC_PAGE_SIZE); |
de9ac07b | 413 | |
d20deb64 | 414 | on_exit(perf_record__sig_exit, rec); |
f5970550 PZ |
415 | signal(SIGCHLD, sig_handler); |
416 | signal(SIGINT, sig_handler); | |
18483b81 | 417 | signal(SIGUSR1, sig_handler); |
f5970550 | 418 | |
d7065adb FBH |
419 | if (!output_name) { |
420 | if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode)) | |
d20deb64 | 421 | opts->pipe_output = true; |
d7065adb | 422 | else |
d20deb64 | 423 | rec->output_name = output_name = "perf.data"; |
d7065adb FBH |
424 | } |
425 | if (output_name) { | |
426 | if (!strcmp(output_name, "-")) | |
d20deb64 | 427 | opts->pipe_output = true; |
d7065adb | 428 | else if (!stat(output_name, &st) && st.st_size) { |
d20deb64 | 429 | if (rec->write_mode == WRITE_FORCE) { |
d7065adb FBH |
430 | char oldname[PATH_MAX]; |
431 | snprintf(oldname, sizeof(oldname), "%s.old", | |
432 | output_name); | |
433 | unlink(oldname); | |
434 | rename(output_name, oldname); | |
435 | } | |
d20deb64 ACM |
436 | } else if (rec->write_mode == WRITE_APPEND) { |
437 | rec->write_mode = WRITE_FORCE; | |
266e0e21 | 438 | } |
97124d5e PZ |
439 | } |
440 | ||
f887f301 | 441 | flags = O_CREAT|O_RDWR; |
d20deb64 ACM |
442 | if (rec->write_mode == WRITE_APPEND) |
443 | rec->file_new = 0; | |
abaff32a IM |
444 | else |
445 | flags |= O_TRUNC; | |
446 | ||
d20deb64 | 447 | if (opts->pipe_output) |
529870e3 TZ |
448 | output = STDOUT_FILENO; |
449 | else | |
450 | output = open(output_name, flags, S_IRUSR | S_IWUSR); | |
de9ac07b PZ |
451 | if (output < 0) { |
452 | perror("failed to create output file"); | |
453 | exit(-1); | |
454 | } | |
455 | ||
d20deb64 ACM |
456 | rec->output = output; |
457 | ||
7865e817 | 458 | session = perf_session__new(output_name, O_WRONLY, |
d20deb64 | 459 | rec->write_mode == WRITE_FORCE, false, NULL); |
94c744b6 | 460 | if (session == NULL) { |
a9a70bbc ACM |
461 | pr_err("Not enough memory for reading perf file header\n"); |
462 | return -1; | |
463 | } | |
464 | ||
d20deb64 ACM |
465 | rec->session = session; |
466 | ||
781ba9d2 RR |
467 | for (feat = HEADER_FIRST_FEATURE; feat < HEADER_LAST_FEATURE; feat++) |
468 | perf_header__set_feat(&session->header, feat); | |
469 | ||
470 | if (rec->no_buildid) | |
471 | perf_header__clear_feat(&session->header, HEADER_BUILD_ID); | |
472 | ||
473 | if (!have_tracepoints(&evsel_list->entries)) | |
474 | perf_header__clear_feat(&session->header, HEADER_TRACE_INFO); | |
baa2f6ce | 475 | |
d20deb64 | 476 | if (!rec->file_new) { |
a91e5431 | 477 | err = perf_session__read_header(session, output); |
4dc0a04b | 478 | if (err < 0) |
39d17dac | 479 | goto out_delete_session; |
4dc0a04b ACM |
480 | } |
481 | ||
d4db3f16 | 482 | if (forks) { |
d20deb64 | 483 | err = perf_evlist__prepare_workload(evsel_list, opts, argv); |
35b9d88e ACM |
484 | if (err < 0) { |
485 | pr_err("Couldn't run the workload!\n"); | |
486 | goto out_delete_session; | |
856e9660 | 487 | } |
856e9660 PZ |
488 | } |
489 | ||
d20deb64 | 490 | perf_record__open(rec); |
de9ac07b | 491 | |
712a4b60 | 492 | /* |
d20deb64 | 493 | * perf_session__delete(session) will be called at perf_record__exit() |
712a4b60 | 494 | */ |
d20deb64 | 495 | on_exit(perf_record__exit, rec); |
712a4b60 | 496 | |
d20deb64 | 497 | if (opts->pipe_output) { |
529870e3 TZ |
498 | err = perf_header__write_pipe(output); |
499 | if (err < 0) | |
500 | return err; | |
d20deb64 | 501 | } else if (rec->file_new) { |
a91e5431 ACM |
502 | err = perf_session__write_header(session, evsel_list, |
503 | output, false); | |
d5eed904 ACM |
504 | if (err < 0) |
505 | return err; | |
56b03f3c ACM |
506 | } |
507 | ||
d3665498 | 508 | if (!rec->no_buildid |
e20960c0 | 509 | && !perf_header__has_feat(&session->header, HEADER_BUILD_ID)) { |
d3665498 | 510 | pr_err("Couldn't generate buildids. " |
e20960c0 RR |
511 | "Use --no-buildid to profile anyway.\n"); |
512 | return -1; | |
513 | } | |
514 | ||
d20deb64 | 515 | rec->post_processing_offset = lseek(output, 0, SEEK_CUR); |
6122e4e4 | 516 | |
743eb868 ACM |
517 | machine = perf_session__find_host_machine(session); |
518 | if (!machine) { | |
519 | pr_err("Couldn't find native kernel information.\n"); | |
520 | return -1; | |
521 | } | |
522 | ||
d20deb64 | 523 | if (opts->pipe_output) { |
45694aa7 | 524 | err = perf_event__synthesize_attrs(tool, session, |
d20deb64 | 525 | process_synthesized_event); |
2c46dbb5 TZ |
526 | if (err < 0) { |
527 | pr_err("Couldn't synthesize attrs.\n"); | |
528 | return err; | |
529 | } | |
cd19a035 | 530 | |
45694aa7 | 531 | err = perf_event__synthesize_event_types(tool, process_synthesized_event, |
743eb868 | 532 | machine); |
cd19a035 TZ |
533 | if (err < 0) { |
534 | pr_err("Couldn't synthesize event_types.\n"); | |
535 | return err; | |
536 | } | |
9215545e | 537 | |
361c99a6 | 538 | if (have_tracepoints(&evsel_list->entries)) { |
63e0c771 TZ |
539 | /* |
540 | * FIXME err <= 0 here actually means that | |
541 | * there were no tracepoints so its not really | |
542 | * an error, just that we don't need to | |
543 | * synthesize anything. We really have to | |
544 | * return this more properly and also | |
545 | * propagate errors that now are calling die() | |
546 | */ | |
45694aa7 | 547 | err = perf_event__synthesize_tracing_data(tool, output, evsel_list, |
743eb868 | 548 | process_synthesized_event); |
63e0c771 TZ |
549 | if (err <= 0) { |
550 | pr_err("Couldn't record tracing data.\n"); | |
551 | return err; | |
552 | } | |
d20deb64 | 553 | advance_output(rec, err); |
63e0c771 | 554 | } |
2c46dbb5 TZ |
555 | } |
556 | ||
45694aa7 | 557 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 558 | machine, "_text"); |
70162138 | 559 | if (err < 0) |
45694aa7 | 560 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 561 | machine, "_stext"); |
c1a3a4b9 ACM |
562 | if (err < 0) |
563 | pr_err("Couldn't record kernel reference relocation symbol\n" | |
564 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
565 | "Check /proc/kallsyms permission or run as root.\n"); | |
b7cece76 | 566 | |
45694aa7 | 567 | err = perf_event__synthesize_modules(tool, process_synthesized_event, |
743eb868 | 568 | machine); |
c1a3a4b9 ACM |
569 | if (err < 0) |
570 | pr_err("Couldn't record kernel module information.\n" | |
571 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
572 | "Check /proc/modules permission or run as root.\n"); | |
573 | ||
a1645ce1 | 574 | if (perf_guest) |
45694aa7 | 575 | perf_session__process_machines(session, tool, |
8115d60c | 576 | perf_event__synthesize_guest_os); |
7c6a1c65 | 577 | |
d20deb64 | 578 | if (!opts->system_wide) |
45694aa7 | 579 | perf_event__synthesize_thread_map(tool, evsel_list->threads, |
7c940c18 | 580 | process_synthesized_event, |
743eb868 | 581 | machine); |
234fbbf5 | 582 | else |
45694aa7 | 583 | perf_event__synthesize_threads(tool, process_synthesized_event, |
743eb868 | 584 | machine); |
7c6a1c65 | 585 | |
d20deb64 | 586 | if (rec->realtime_prio) { |
de9ac07b PZ |
587 | struct sched_param param; |
588 | ||
d20deb64 | 589 | param.sched_priority = rec->realtime_prio; |
de9ac07b | 590 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { |
6beba7ad | 591 | pr_err("Could not set realtime priority.\n"); |
de9ac07b PZ |
592 | exit(-1); |
593 | } | |
594 | } | |
595 | ||
764e16a3 DA |
596 | perf_evlist__enable(evsel_list); |
597 | ||
856e9660 PZ |
598 | /* |
599 | * Let the child rip | |
600 | */ | |
d4db3f16 | 601 | if (forks) |
35b9d88e | 602 | perf_evlist__start_workload(evsel_list); |
856e9660 | 603 | |
649c48a9 | 604 | for (;;) { |
d20deb64 | 605 | int hits = rec->samples; |
de9ac07b | 606 | |
d20deb64 | 607 | perf_record__mmap_read_all(rec); |
de9ac07b | 608 | |
d20deb64 | 609 | if (hits == rec->samples) { |
649c48a9 PZ |
610 | if (done) |
611 | break; | |
5c581041 | 612 | err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1); |
8b412664 PZ |
613 | waking++; |
614 | } | |
615 | ||
4152ab37 ACM |
616 | if (done) |
617 | perf_evlist__disable(evsel_list); | |
de9ac07b PZ |
618 | } |
619 | ||
18483b81 | 620 | if (quiet || signr == SIGUSR1) |
b44308f5 ACM |
621 | return 0; |
622 | ||
8b412664 PZ |
623 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
624 | ||
021e9f47 IM |
625 | /* |
626 | * Approximate RIP event size: 24 bytes. | |
627 | */ | |
628 | fprintf(stderr, | |
9486aa38 | 629 | "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n", |
d20deb64 | 630 | (double)rec->bytes_written / 1024.0 / 1024.0, |
021e9f47 | 631 | output_name, |
d20deb64 | 632 | rec->bytes_written / 24); |
addc2785 | 633 | |
de9ac07b | 634 | return 0; |
39d17dac ACM |
635 | |
636 | out_delete_session: | |
637 | perf_session__delete(session); | |
638 | return err; | |
de9ac07b | 639 | } |
0e9b20b8 | 640 | |
0e9b20b8 | 641 | static const char * const record_usage[] = { |
9e096753 MG |
642 | "perf record [<options>] [<command>]", |
643 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
644 | NULL |
645 | }; | |
646 | ||
d20deb64 ACM |
647 | /* |
648 | * XXX Ideally would be local to cmd_record() and passed to a perf_record__new | |
649 | * because we need to have access to it in perf_record__exit, that is called | |
650 | * after cmd_record() exits, but since record_options need to be accessible to | |
651 | * builtin-script, leave it here. | |
652 | * | |
653 | * At least we don't ouch it in all the other functions here directly. | |
654 | * | |
655 | * Just say no to tons of global variables, sigh. | |
656 | */ | |
657 | static struct perf_record record = { | |
658 | .opts = { | |
d20deb64 ACM |
659 | .mmap_pages = UINT_MAX, |
660 | .user_freq = UINT_MAX, | |
661 | .user_interval = ULLONG_MAX, | |
662 | .freq = 1000, | |
d20deb64 ACM |
663 | }, |
664 | .write_mode = WRITE_FORCE, | |
665 | .file_new = true, | |
666 | }; | |
7865e817 | 667 | |
d20deb64 ACM |
668 | /* |
669 | * XXX Will stay a global variable till we fix builtin-script.c to stop messing | |
670 | * with it and switch to use the library functions in perf_evlist that came | |
671 | * from builtin-record.c, i.e. use perf_record_opts, | |
672 | * perf_evlist__prepare_workload, etc instead of fork+exec'in 'perf record', | |
673 | * using pipes, etc. | |
674 | */ | |
bca647aa | 675 | const struct option record_options[] = { |
d20deb64 | 676 | OPT_CALLBACK('e', "event", &record.evlist, "event", |
86847b62 | 677 | "event selector. use 'perf list' to list available events", |
f120f9d5 | 678 | parse_events_option), |
d20deb64 | 679 | OPT_CALLBACK(0, "filter", &record.evlist, "filter", |
c171b552 | 680 | "event filter", parse_filter), |
b52956c9 | 681 | OPT_STRING('p', "pid", &record.opts.target_pid, "pid", |
d6d901c2 | 682 | "record events on existing process id"), |
b52956c9 | 683 | OPT_STRING('t', "tid", &record.opts.target_tid, "tid", |
d6d901c2 | 684 | "record events on existing thread id"), |
d20deb64 | 685 | OPT_INTEGER('r', "realtime", &record.realtime_prio, |
0e9b20b8 | 686 | "collect data with this RT SCHED_FIFO priority"), |
d20deb64 | 687 | OPT_BOOLEAN('D', "no-delay", &record.opts.no_delay, |
acac03fa | 688 | "collect data without buffering"), |
d20deb64 | 689 | OPT_BOOLEAN('R', "raw-samples", &record.opts.raw_samples, |
daac07b2 | 690 | "collect raw sample records from all opened counters"), |
d20deb64 | 691 | OPT_BOOLEAN('a', "all-cpus", &record.opts.system_wide, |
0e9b20b8 | 692 | "system-wide collection from all CPUs"), |
d20deb64 | 693 | OPT_BOOLEAN('A', "append", &record.append_file, |
abaff32a | 694 | "append to the output file to do incremental profiling"), |
d20deb64 | 695 | OPT_STRING('C', "cpu", &record.opts.cpu_list, "cpu", |
c45c6ea2 | 696 | "list of cpus to monitor"), |
d20deb64 | 697 | OPT_BOOLEAN('f', "force", &record.force, |
7865e817 | 698 | "overwrite existing data file (deprecated)"), |
d20deb64 ACM |
699 | OPT_U64('c', "count", &record.opts.user_interval, "event period to sample"), |
700 | OPT_STRING('o', "output", &record.output_name, "file", | |
abaff32a | 701 | "output file name"), |
d20deb64 | 702 | OPT_BOOLEAN('i', "no-inherit", &record.opts.no_inherit, |
2e6cdf99 | 703 | "child tasks do not inherit counters"), |
d20deb64 ACM |
704 | OPT_UINTEGER('F', "freq", &record.opts.user_freq, "profile at this frequency"), |
705 | OPT_UINTEGER('m', "mmap-pages", &record.opts.mmap_pages, | |
01c2d99b | 706 | "number of mmap data pages"), |
d20deb64 | 707 | OPT_BOOLEAN(0, "group", &record.opts.group, |
43bece79 | 708 | "put the counters into a counter group"), |
d20deb64 | 709 | OPT_BOOLEAN('g', "call-graph", &record.opts.call_graph, |
3efa1cc9 | 710 | "do call-graph (stack chain/backtrace) recording"), |
c0555642 | 711 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 712 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 713 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
d20deb64 | 714 | OPT_BOOLEAN('s', "stat", &record.opts.inherit_stat, |
649c48a9 | 715 | "per thread counts"), |
d20deb64 | 716 | OPT_BOOLEAN('d', "data", &record.opts.sample_address, |
4bba828d | 717 | "Sample addresses"), |
d20deb64 | 718 | OPT_BOOLEAN('T', "timestamp", &record.opts.sample_time, "Sample timestamps"), |
3e76ac78 | 719 | OPT_BOOLEAN('P', "period", &record.opts.period, "Sample period"), |
d20deb64 | 720 | OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples, |
649c48a9 | 721 | "don't sample"), |
d20deb64 | 722 | OPT_BOOLEAN('N', "no-buildid-cache", &record.no_buildid_cache, |
a1ac1d3c | 723 | "do not update the buildid cache"), |
d20deb64 | 724 | OPT_BOOLEAN('B', "no-buildid", &record.no_buildid, |
baa2f6ce | 725 | "do not collect buildids in perf.data"), |
d20deb64 | 726 | OPT_CALLBACK('G', "cgroup", &record.evlist, "name", |
023695d9 SE |
727 | "monitor event in cgroup name only", |
728 | parse_cgroups), | |
0d37aa34 | 729 | OPT_STRING('u', "uid", &record.uid_str, "user", "user to profile"), |
0e9b20b8 IM |
730 | OPT_END() |
731 | }; | |
732 | ||
f37a291c | 733 | int cmd_record(int argc, const char **argv, const char *prefix __used) |
0e9b20b8 | 734 | { |
69aad6f1 ACM |
735 | int err = -ENOMEM; |
736 | struct perf_evsel *pos; | |
d20deb64 ACM |
737 | struct perf_evlist *evsel_list; |
738 | struct perf_record *rec = &record; | |
0e9b20b8 | 739 | |
fbe96f29 SE |
740 | perf_header__set_cmdline(argc, argv); |
741 | ||
7e2ed097 | 742 | evsel_list = perf_evlist__new(NULL, NULL); |
361c99a6 ACM |
743 | if (evsel_list == NULL) |
744 | return -ENOMEM; | |
745 | ||
d20deb64 ACM |
746 | rec->evlist = evsel_list; |
747 | ||
bca647aa | 748 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 749 | PARSE_OPT_STOP_AT_NON_OPTION); |
b52956c9 | 750 | if (!argc && !rec->opts.target_pid && !rec->opts.target_tid && |
0d37aa34 | 751 | !rec->opts.system_wide && !rec->opts.cpu_list && !rec->uid_str) |
bca647aa | 752 | usage_with_options(record_usage, record_options); |
0e9b20b8 | 753 | |
d20deb64 | 754 | if (rec->force && rec->append_file) { |
7865e817 FW |
755 | fprintf(stderr, "Can't overwrite and append at the same time." |
756 | " You need to choose between -f and -A"); | |
bca647aa | 757 | usage_with_options(record_usage, record_options); |
d20deb64 ACM |
758 | } else if (rec->append_file) { |
759 | rec->write_mode = WRITE_APPEND; | |
7865e817 | 760 | } else { |
d20deb64 | 761 | rec->write_mode = WRITE_FORCE; |
7865e817 FW |
762 | } |
763 | ||
d20deb64 | 764 | if (nr_cgroups && !rec->opts.system_wide) { |
023695d9 SE |
765 | fprintf(stderr, "cgroup monitoring only available in" |
766 | " system-wide mode\n"); | |
767 | usage_with_options(record_usage, record_options); | |
768 | } | |
769 | ||
655000e7 | 770 | symbol__init(); |
baa2f6ce | 771 | |
ec80fde7 | 772 | if (symbol_conf.kptr_restrict) |
646aaea6 ACM |
773 | pr_warning( |
774 | "WARNING: Kernel address maps (/proc/{kallsyms,modules}) are restricted,\n" | |
775 | "check /proc/sys/kernel/kptr_restrict.\n\n" | |
776 | "Samples in kernel functions may not be resolved if a suitable vmlinux\n" | |
777 | "file is not found in the buildid cache or in the vmlinux path.\n\n" | |
778 | "Samples in kernel modules won't be resolved at all.\n\n" | |
779 | "If some relocation was applied (e.g. kexec) symbols may be misresolved\n" | |
780 | "even with a suitable vmlinux or kallsyms file.\n\n"); | |
ec80fde7 | 781 | |
d20deb64 | 782 | if (rec->no_buildid_cache || rec->no_buildid) |
a1ac1d3c | 783 | disable_buildid_cache(); |
655000e7 | 784 | |
361c99a6 ACM |
785 | if (evsel_list->nr_entries == 0 && |
786 | perf_evlist__add_default(evsel_list) < 0) { | |
69aad6f1 ACM |
787 | pr_err("Not enough memory for event selector list\n"); |
788 | goto out_symbol_exit; | |
bbd36e5e | 789 | } |
0e9b20b8 | 790 | |
0d37aa34 ACM |
791 | rec->opts.uid = parse_target_uid(rec->uid_str, rec->opts.target_tid, |
792 | rec->opts.target_pid); | |
793 | if (rec->uid_str != NULL && rec->opts.uid == UINT_MAX - 1) | |
794 | goto out_free_fd; | |
795 | ||
b52956c9 | 796 | if (rec->opts.target_pid) |
d20deb64 | 797 | rec->opts.target_tid = rec->opts.target_pid; |
d6d901c2 | 798 | |
d20deb64 | 799 | if (perf_evlist__create_maps(evsel_list, rec->opts.target_pid, |
0d37aa34 ACM |
800 | rec->opts.target_tid, rec->opts.uid, |
801 | rec->opts.cpu_list) < 0) | |
dd7927f4 | 802 | usage_with_options(record_usage, record_options); |
69aad6f1 | 803 | |
361c99a6 | 804 | list_for_each_entry(pos, &evsel_list->entries, node) { |
ad7f4e3f ACM |
805 | if (perf_header__push_event(pos->attr.config, event_name(pos))) |
806 | goto out_free_fd; | |
d6d901c2 | 807 | } |
5c581041 | 808 | |
d20deb64 ACM |
809 | if (rec->opts.user_interval != ULLONG_MAX) |
810 | rec->opts.default_interval = rec->opts.user_interval; | |
811 | if (rec->opts.user_freq != UINT_MAX) | |
812 | rec->opts.freq = rec->opts.user_freq; | |
f9212819 | 813 | |
7e4ff9e3 MG |
814 | /* |
815 | * User specified count overrides default frequency. | |
816 | */ | |
d20deb64 ACM |
817 | if (rec->opts.default_interval) |
818 | rec->opts.freq = 0; | |
819 | else if (rec->opts.freq) { | |
820 | rec->opts.default_interval = rec->opts.freq; | |
7e4ff9e3 MG |
821 | } else { |
822 | fprintf(stderr, "frequency and count are zero, aborting\n"); | |
39d17dac | 823 | err = -EINVAL; |
5c581041 | 824 | goto out_free_fd; |
7e4ff9e3 MG |
825 | } |
826 | ||
d20deb64 | 827 | err = __cmd_record(&record, argc, argv); |
39d17dac | 828 | out_free_fd: |
7e2ed097 | 829 | perf_evlist__delete_maps(evsel_list); |
d65a458b ACM |
830 | out_symbol_exit: |
831 | symbol__exit(); | |
39d17dac | 832 | return err; |
0e9b20b8 | 833 | } |