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; | |
47 | int output; | |
48 | unsigned int page_size; | |
49 | int realtime_prio; | |
50 | enum write_mode_t write_mode; | |
51 | bool no_buildid; | |
52 | bool no_buildid_cache; | |
53 | bool force; | |
54 | bool file_new; | |
55 | bool append_file; | |
56 | long samples; | |
57 | off_t post_processing_offset; | |
0f82ebc4 | 58 | }; |
a21ca2ca | 59 | |
d20deb64 | 60 | static void advance_output(struct perf_record *rec, size_t size) |
9215545e | 61 | { |
d20deb64 | 62 | rec->bytes_written += size; |
9215545e TZ |
63 | } |
64 | ||
d20deb64 | 65 | static void write_output(struct perf_record *rec, void *buf, size_t size) |
f5970550 PZ |
66 | { |
67 | while (size) { | |
d20deb64 | 68 | int ret = write(rec->output, buf, size); |
f5970550 PZ |
69 | |
70 | if (ret < 0) | |
71 | die("failed to write"); | |
72 | ||
73 | size -= ret; | |
74 | buf += ret; | |
75 | ||
d20deb64 | 76 | rec->bytes_written += ret; |
f5970550 PZ |
77 | } |
78 | } | |
79 | ||
45694aa7 | 80 | static int process_synthesized_event(struct perf_tool *tool, |
d20deb64 | 81 | union perf_event *event, |
8d50e5b4 | 82 | struct perf_sample *sample __used, |
743eb868 | 83 | struct machine *machine __used) |
234fbbf5 | 84 | { |
45694aa7 | 85 | struct perf_record *rec = container_of(tool, struct perf_record, tool); |
d20deb64 | 86 | write_output(rec, event, event->header.size); |
234fbbf5 ACM |
87 | return 0; |
88 | } | |
89 | ||
d20deb64 ACM |
90 | static void perf_record__mmap_read(struct perf_record *rec, |
91 | struct perf_mmap *md) | |
de9ac07b | 92 | { |
744bd8aa | 93 | unsigned int head = perf_mmap__read_head(md); |
de9ac07b | 94 | unsigned int old = md->prev; |
d20deb64 | 95 | unsigned char *data = md->base + rec->page_size; |
de9ac07b PZ |
96 | unsigned long size; |
97 | void *buf; | |
de9ac07b | 98 | |
dc82009a ACM |
99 | if (old == head) |
100 | return; | |
101 | ||
d20deb64 | 102 | rec->samples++; |
de9ac07b PZ |
103 | |
104 | size = head - old; | |
105 | ||
106 | if ((old & md->mask) + size != (head & md->mask)) { | |
107 | buf = &data[old & md->mask]; | |
108 | size = md->mask + 1 - (old & md->mask); | |
109 | old += size; | |
021e9f47 | 110 | |
d20deb64 | 111 | write_output(rec, buf, size); |
de9ac07b PZ |
112 | } |
113 | ||
114 | buf = &data[old & md->mask]; | |
115 | size = head - old; | |
116 | old += size; | |
021e9f47 | 117 | |
d20deb64 | 118 | write_output(rec, buf, size); |
de9ac07b PZ |
119 | |
120 | md->prev = old; | |
115d2d89 | 121 | perf_mmap__write_tail(md, old); |
de9ac07b PZ |
122 | } |
123 | ||
124 | static volatile int done = 0; | |
f7b7c26e | 125 | static volatile int signr = -1; |
33e49ea7 | 126 | static volatile int child_finished = 0; |
de9ac07b | 127 | |
16c8a109 | 128 | static void sig_handler(int sig) |
de9ac07b | 129 | { |
33e49ea7 AK |
130 | if (sig == SIGCHLD) |
131 | child_finished = 1; | |
132 | ||
16c8a109 | 133 | done = 1; |
f7b7c26e PZ |
134 | signr = sig; |
135 | } | |
136 | ||
d20deb64 | 137 | static void perf_record__sig_exit(int exit_status __used, void *arg) |
f7b7c26e | 138 | { |
d20deb64 | 139 | struct perf_record *rec = arg; |
33e49ea7 AK |
140 | int status; |
141 | ||
d20deb64 | 142 | if (rec->evlist->workload.pid > 0) { |
33e49ea7 | 143 | if (!child_finished) |
d20deb64 | 144 | kill(rec->evlist->workload.pid, SIGTERM); |
33e49ea7 AK |
145 | |
146 | wait(&status); | |
147 | if (WIFSIGNALED(status)) | |
d20deb64 | 148 | psignal(WTERMSIG(status), rec->progname); |
33e49ea7 | 149 | } |
933da83a | 150 | |
18483b81 | 151 | if (signr == -1 || signr == SIGUSR1) |
f7b7c26e PZ |
152 | return; |
153 | ||
154 | signal(signr, SIG_DFL); | |
155 | kill(getpid(), signr); | |
de9ac07b PZ |
156 | } |
157 | ||
a91e5431 ACM |
158 | static bool perf_evlist__equal(struct perf_evlist *evlist, |
159 | struct perf_evlist *other) | |
160 | { | |
161 | struct perf_evsel *pos, *pair; | |
162 | ||
163 | if (evlist->nr_entries != other->nr_entries) | |
164 | return false; | |
165 | ||
166 | pair = list_entry(other->entries.next, struct perf_evsel, node); | |
167 | ||
168 | list_for_each_entry(pos, &evlist->entries, node) { | |
169 | if (memcmp(&pos->attr, &pair->attr, sizeof(pos->attr) != 0)) | |
170 | return false; | |
171 | pair = list_entry(pair->node.next, struct perf_evsel, node); | |
172 | } | |
173 | ||
174 | return true; | |
175 | } | |
176 | ||
d20deb64 | 177 | static void perf_record__open(struct perf_record *rec) |
dd7927f4 | 178 | { |
727ab04e | 179 | struct perf_evsel *pos, *first; |
d20deb64 ACM |
180 | struct perf_evlist *evlist = rec->evlist; |
181 | struct perf_session *session = rec->session; | |
182 | struct perf_record_opts *opts = &rec->opts; | |
dd7927f4 | 183 | |
727ab04e ACM |
184 | first = list_entry(evlist->entries.next, struct perf_evsel, node); |
185 | ||
d20deb64 | 186 | perf_evlist__config_attrs(evlist, opts); |
0f82ebc4 | 187 | |
dd7927f4 ACM |
188 | list_for_each_entry(pos, &evlist->entries, node) { |
189 | struct perf_event_attr *attr = &pos->attr; | |
727ab04e | 190 | struct xyarray *group_fd = NULL; |
dd7927f4 ACM |
191 | /* |
192 | * Check if parse_single_tracepoint_event has already asked for | |
193 | * PERF_SAMPLE_TIME. | |
194 | * | |
195 | * XXX this is kludgy but short term fix for problems introduced by | |
196 | * eac23d1c that broke 'perf script' by having different sample_types | |
197 | * when using multiple tracepoint events when we use a perf binary | |
198 | * that tries to use sample_id_all on an older kernel. | |
199 | * | |
200 | * We need to move counter creation to perf_session, support | |
201 | * different sample_types, etc. | |
202 | */ | |
203 | bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; | |
d6d901c2 | 204 | |
d20deb64 | 205 | if (opts->group && pos != first) |
727ab04e | 206 | group_fd = first->fd; |
0c978128 ACM |
207 | fallback_missing_features: |
208 | if (opts->exclude_guest_missing) | |
209 | attr->exclude_guest = attr->exclude_host = 0; | |
dd7927f4 | 210 | retry_sample_id: |
808e1226 | 211 | attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1; |
dd7927f4 | 212 | try_again: |
ed80f581 | 213 | if (perf_evsel__open(pos, evlist->cpus, evlist->threads, |
d20deb64 | 214 | opts->group, group_fd) < 0) { |
d6d901c2 ZY |
215 | int err = errno; |
216 | ||
c286c419 | 217 | if (err == EPERM || err == EACCES) { |
b8631e6e | 218 | ui__error_paranoid(); |
c286c419 | 219 | exit(EXIT_FAILURE); |
bea03405 | 220 | } else if (err == ENODEV && opts->target.cpu_list) { |
d6d901c2 ZY |
221 | die("No such device - did you specify" |
222 | " an out-of-range profile CPU?\n"); | |
0c978128 ACM |
223 | } else if (err == EINVAL) { |
224 | if (!opts->exclude_guest_missing && | |
225 | (attr->exclude_guest || attr->exclude_host)) { | |
226 | pr_debug("Old kernel, cannot exclude " | |
227 | "guest or host samples.\n"); | |
228 | opts->exclude_guest_missing = true; | |
229 | goto fallback_missing_features; | |
808e1226 | 230 | } else if (!opts->sample_id_all_missing) { |
0c978128 ACM |
231 | /* |
232 | * Old kernel, no attr->sample_id_type_all field | |
233 | */ | |
808e1226 | 234 | opts->sample_id_all_missing = true; |
0c978128 ACM |
235 | if (!opts->sample_time && !opts->raw_samples && !time_needed) |
236 | attr->sample_type &= ~PERF_SAMPLE_TIME; | |
237 | ||
238 | goto retry_sample_id; | |
239 | } | |
d6d901c2 | 240 | } |
3da297a6 | 241 | |
d6d901c2 ZY |
242 | /* |
243 | * If it's cycles then fall back to hrtimer | |
244 | * based cpu-clock-tick sw counter, which | |
028d455b DA |
245 | * is always available even if no PMU support. |
246 | * | |
247 | * PPC returns ENXIO until 2.6.37 (behavior changed | |
248 | * with commit b0a873e). | |
d6d901c2 | 249 | */ |
028d455b DA |
250 | if ((err == ENOENT || err == ENXIO) |
251 | && attr->type == PERF_TYPE_HARDWARE | |
d6d901c2 ZY |
252 | && attr->config == PERF_COUNT_HW_CPU_CYCLES) { |
253 | ||
254 | if (verbose) | |
ca6a4258 DA |
255 | ui__warning("The cycles event is not supported, " |
256 | "trying to fall back to cpu-clock-ticks\n"); | |
d6d901c2 ZY |
257 | attr->type = PERF_TYPE_SOFTWARE; |
258 | attr->config = PERF_COUNT_SW_CPU_CLOCK; | |
d1cae34d DA |
259 | if (pos->name) { |
260 | free(pos->name); | |
261 | pos->name = NULL; | |
262 | } | |
d6d901c2 ZY |
263 | goto try_again; |
264 | } | |
ca6a4258 DA |
265 | |
266 | if (err == ENOENT) { | |
267 | ui__warning("The %s event is not supported.\n", | |
268 | event_name(pos)); | |
269 | exit(EXIT_FAILURE); | |
270 | } | |
271 | ||
d6d901c2 | 272 | printf("\n"); |
d9cf837e | 273 | error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", |
dd7927f4 | 274 | err, strerror(err)); |
bfd45118 SK |
275 | |
276 | #if defined(__i386__) || defined(__x86_64__) | |
d6d901c2 ZY |
277 | if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) |
278 | die("No hardware sampling interrupt available." | |
279 | " No APIC? If so then you can boot the kernel" | |
280 | " with the \"lapic\" boot parameter to" | |
281 | " force-enable it.\n"); | |
bfd45118 SK |
282 | #endif |
283 | ||
d6d901c2 | 284 | die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); |
c171b552 LZ |
285 | } |
286 | } | |
a43d3f08 | 287 | |
0a102479 FW |
288 | if (perf_evlist__set_filters(evlist)) { |
289 | error("failed to set filter with %d (%s)\n", errno, | |
290 | strerror(errno)); | |
291 | exit(-1); | |
292 | } | |
293 | ||
18e60939 NE |
294 | if (perf_evlist__mmap(evlist, opts->mmap_pages, false) < 0) { |
295 | if (errno == EPERM) | |
296 | die("Permission error mapping pages.\n" | |
297 | "Consider increasing " | |
298 | "/proc/sys/kernel/perf_event_mlock_kb,\n" | |
299 | "or try again with a smaller value of -m/--mmap_pages.\n" | |
300 | "(current value: %d)\n", opts->mmap_pages); | |
41d0d933 NE |
301 | else if (!is_power_of_2(opts->mmap_pages)) |
302 | die("--mmap_pages/-m value must be a power of two."); | |
303 | ||
0a27d7f9 | 304 | die("failed to mmap with %d (%s)\n", errno, strerror(errno)); |
18e60939 | 305 | } |
0a27d7f9 | 306 | |
d20deb64 | 307 | if (rec->file_new) |
a91e5431 ACM |
308 | session->evlist = evlist; |
309 | else { | |
310 | if (!perf_evlist__equal(session->evlist, evlist)) { | |
311 | fprintf(stderr, "incompatible append\n"); | |
312 | exit(-1); | |
313 | } | |
314 | } | |
315 | ||
316 | perf_session__update_sample_type(session); | |
16c8a109 PZ |
317 | } |
318 | ||
d20deb64 | 319 | static int process_buildids(struct perf_record *rec) |
6122e4e4 | 320 | { |
d20deb64 | 321 | u64 size = lseek(rec->output, 0, SEEK_CUR); |
6122e4e4 | 322 | |
9f591fd7 ACM |
323 | if (size == 0) |
324 | return 0; | |
325 | ||
d20deb64 ACM |
326 | rec->session->fd = rec->output; |
327 | return __perf_session__process_events(rec->session, rec->post_processing_offset, | |
328 | size - rec->post_processing_offset, | |
6122e4e4 ACM |
329 | size, &build_id__mark_dso_hit_ops); |
330 | } | |
331 | ||
d20deb64 | 332 | static void perf_record__exit(int status __used, void *arg) |
f5970550 | 333 | { |
d20deb64 ACM |
334 | struct perf_record *rec = arg; |
335 | ||
336 | if (!rec->opts.pipe_output) { | |
337 | rec->session->header.data_size += rec->bytes_written; | |
338 | ||
339 | if (!rec->no_buildid) | |
340 | process_buildids(rec); | |
341 | perf_session__write_header(rec->session, rec->evlist, | |
342 | rec->output, true); | |
343 | perf_session__delete(rec->session); | |
344 | perf_evlist__delete(rec->evlist); | |
d65a458b | 345 | symbol__exit(); |
c7929e47 | 346 | } |
f5970550 PZ |
347 | } |
348 | ||
8115d60c | 349 | static void perf_event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
350 | { |
351 | int err; | |
45694aa7 | 352 | struct perf_tool *tool = data; |
a1645ce1 | 353 | |
23346f21 | 354 | if (machine__is_host(machine)) |
a1645ce1 ZY |
355 | return; |
356 | ||
357 | /* | |
358 | *As for guest kernel when processing subcommand record&report, | |
359 | *we arrange module mmap prior to guest kernel mmap and trigger | |
360 | *a preload dso because default guest module symbols are loaded | |
361 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
362 | *method is used to avoid symbol missing when the first addr is | |
363 | *in module instead of in guest kernel. | |
364 | */ | |
45694aa7 | 365 | err = perf_event__synthesize_modules(tool, process_synthesized_event, |
743eb868 | 366 | machine); |
a1645ce1 ZY |
367 | if (err < 0) |
368 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 369 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 370 | |
a1645ce1 ZY |
371 | /* |
372 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
373 | * have no _text sometimes. | |
374 | */ | |
45694aa7 | 375 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 376 | machine, "_text"); |
a1645ce1 | 377 | if (err < 0) |
45694aa7 | 378 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 379 | machine, "_stext"); |
a1645ce1 ZY |
380 | if (err < 0) |
381 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 382 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
383 | } |
384 | ||
98402807 FW |
385 | static struct perf_event_header finished_round_event = { |
386 | .size = sizeof(struct perf_event_header), | |
387 | .type = PERF_RECORD_FINISHED_ROUND, | |
388 | }; | |
389 | ||
d20deb64 | 390 | static void perf_record__mmap_read_all(struct perf_record *rec) |
98402807 | 391 | { |
0e2e63dd | 392 | int i; |
98402807 | 393 | |
d20deb64 ACM |
394 | for (i = 0; i < rec->evlist->nr_mmaps; i++) { |
395 | if (rec->evlist->mmap[i].base) | |
396 | perf_record__mmap_read(rec, &rec->evlist->mmap[i]); | |
98402807 FW |
397 | } |
398 | ||
2eeaaa09 | 399 | if (perf_header__has_feat(&rec->session->header, HEADER_TRACING_DATA)) |
d20deb64 | 400 | write_output(rec, &finished_round_event, sizeof(finished_round_event)); |
98402807 FW |
401 | } |
402 | ||
d20deb64 | 403 | static int __cmd_record(struct perf_record *rec, int argc, const char **argv) |
16c8a109 | 404 | { |
abaff32a | 405 | struct stat st; |
abaff32a | 406 | int flags; |
781ba9d2 | 407 | int err, output, feat; |
8b412664 | 408 | unsigned long waking = 0; |
46be604b | 409 | const bool forks = argc > 0; |
23346f21 | 410 | struct machine *machine; |
45694aa7 | 411 | struct perf_tool *tool = &rec->tool; |
d20deb64 ACM |
412 | struct perf_record_opts *opts = &rec->opts; |
413 | struct perf_evlist *evsel_list = rec->evlist; | |
414 | const char *output_name = rec->output_name; | |
415 | struct perf_session *session; | |
de9ac07b | 416 | |
d20deb64 | 417 | rec->progname = argv[0]; |
33e49ea7 | 418 | |
d20deb64 | 419 | rec->page_size = sysconf(_SC_PAGE_SIZE); |
de9ac07b | 420 | |
d20deb64 | 421 | on_exit(perf_record__sig_exit, rec); |
f5970550 PZ |
422 | signal(SIGCHLD, sig_handler); |
423 | signal(SIGINT, sig_handler); | |
18483b81 | 424 | signal(SIGUSR1, sig_handler); |
f5970550 | 425 | |
d7065adb FBH |
426 | if (!output_name) { |
427 | if (!fstat(STDOUT_FILENO, &st) && S_ISFIFO(st.st_mode)) | |
d20deb64 | 428 | opts->pipe_output = true; |
d7065adb | 429 | else |
d20deb64 | 430 | rec->output_name = output_name = "perf.data"; |
d7065adb FBH |
431 | } |
432 | if (output_name) { | |
433 | if (!strcmp(output_name, "-")) | |
d20deb64 | 434 | opts->pipe_output = true; |
d7065adb | 435 | else if (!stat(output_name, &st) && st.st_size) { |
d20deb64 | 436 | if (rec->write_mode == WRITE_FORCE) { |
d7065adb FBH |
437 | char oldname[PATH_MAX]; |
438 | snprintf(oldname, sizeof(oldname), "%s.old", | |
439 | output_name); | |
440 | unlink(oldname); | |
441 | rename(output_name, oldname); | |
442 | } | |
d20deb64 ACM |
443 | } else if (rec->write_mode == WRITE_APPEND) { |
444 | rec->write_mode = WRITE_FORCE; | |
266e0e21 | 445 | } |
97124d5e PZ |
446 | } |
447 | ||
f887f301 | 448 | flags = O_CREAT|O_RDWR; |
d20deb64 ACM |
449 | if (rec->write_mode == WRITE_APPEND) |
450 | rec->file_new = 0; | |
abaff32a IM |
451 | else |
452 | flags |= O_TRUNC; | |
453 | ||
d20deb64 | 454 | if (opts->pipe_output) |
529870e3 TZ |
455 | output = STDOUT_FILENO; |
456 | else | |
457 | output = open(output_name, flags, S_IRUSR | S_IWUSR); | |
de9ac07b PZ |
458 | if (output < 0) { |
459 | perror("failed to create output file"); | |
460 | exit(-1); | |
461 | } | |
462 | ||
d20deb64 ACM |
463 | rec->output = output; |
464 | ||
7865e817 | 465 | session = perf_session__new(output_name, O_WRONLY, |
d20deb64 | 466 | rec->write_mode == WRITE_FORCE, false, NULL); |
94c744b6 | 467 | if (session == NULL) { |
a9a70bbc ACM |
468 | pr_err("Not enough memory for reading perf file header\n"); |
469 | return -1; | |
470 | } | |
471 | ||
d20deb64 ACM |
472 | rec->session = session; |
473 | ||
781ba9d2 RR |
474 | for (feat = HEADER_FIRST_FEATURE; feat < HEADER_LAST_FEATURE; feat++) |
475 | perf_header__set_feat(&session->header, feat); | |
476 | ||
477 | if (rec->no_buildid) | |
478 | perf_header__clear_feat(&session->header, HEADER_BUILD_ID); | |
479 | ||
480 | if (!have_tracepoints(&evsel_list->entries)) | |
2eeaaa09 | 481 | perf_header__clear_feat(&session->header, HEADER_TRACING_DATA); |
baa2f6ce | 482 | |
330aa675 SE |
483 | if (!rec->opts.branch_stack) |
484 | perf_header__clear_feat(&session->header, HEADER_BRANCH_STACK); | |
485 | ||
d20deb64 | 486 | if (!rec->file_new) { |
a91e5431 | 487 | err = perf_session__read_header(session, output); |
4dc0a04b | 488 | if (err < 0) |
39d17dac | 489 | goto out_delete_session; |
4dc0a04b ACM |
490 | } |
491 | ||
d4db3f16 | 492 | if (forks) { |
d20deb64 | 493 | err = perf_evlist__prepare_workload(evsel_list, opts, argv); |
35b9d88e ACM |
494 | if (err < 0) { |
495 | pr_err("Couldn't run the workload!\n"); | |
496 | goto out_delete_session; | |
856e9660 | 497 | } |
856e9660 PZ |
498 | } |
499 | ||
d20deb64 | 500 | perf_record__open(rec); |
de9ac07b | 501 | |
712a4b60 | 502 | /* |
d20deb64 | 503 | * perf_session__delete(session) will be called at perf_record__exit() |
712a4b60 | 504 | */ |
d20deb64 | 505 | on_exit(perf_record__exit, rec); |
712a4b60 | 506 | |
d20deb64 | 507 | if (opts->pipe_output) { |
529870e3 TZ |
508 | err = perf_header__write_pipe(output); |
509 | if (err < 0) | |
510 | return err; | |
d20deb64 | 511 | } else if (rec->file_new) { |
a91e5431 ACM |
512 | err = perf_session__write_header(session, evsel_list, |
513 | output, false); | |
d5eed904 ACM |
514 | if (err < 0) |
515 | return err; | |
56b03f3c ACM |
516 | } |
517 | ||
d3665498 | 518 | if (!rec->no_buildid |
e20960c0 | 519 | && !perf_header__has_feat(&session->header, HEADER_BUILD_ID)) { |
d3665498 | 520 | pr_err("Couldn't generate buildids. " |
e20960c0 RR |
521 | "Use --no-buildid to profile anyway.\n"); |
522 | return -1; | |
523 | } | |
524 | ||
d20deb64 | 525 | rec->post_processing_offset = lseek(output, 0, SEEK_CUR); |
6122e4e4 | 526 | |
743eb868 ACM |
527 | machine = perf_session__find_host_machine(session); |
528 | if (!machine) { | |
529 | pr_err("Couldn't find native kernel information.\n"); | |
530 | return -1; | |
531 | } | |
532 | ||
d20deb64 | 533 | if (opts->pipe_output) { |
45694aa7 | 534 | err = perf_event__synthesize_attrs(tool, session, |
d20deb64 | 535 | process_synthesized_event); |
2c46dbb5 TZ |
536 | if (err < 0) { |
537 | pr_err("Couldn't synthesize attrs.\n"); | |
538 | return err; | |
539 | } | |
cd19a035 | 540 | |
45694aa7 | 541 | err = perf_event__synthesize_event_types(tool, process_synthesized_event, |
743eb868 | 542 | machine); |
cd19a035 TZ |
543 | if (err < 0) { |
544 | pr_err("Couldn't synthesize event_types.\n"); | |
545 | return err; | |
546 | } | |
9215545e | 547 | |
361c99a6 | 548 | if (have_tracepoints(&evsel_list->entries)) { |
63e0c771 TZ |
549 | /* |
550 | * FIXME err <= 0 here actually means that | |
551 | * there were no tracepoints so its not really | |
552 | * an error, just that we don't need to | |
553 | * synthesize anything. We really have to | |
554 | * return this more properly and also | |
555 | * propagate errors that now are calling die() | |
556 | */ | |
45694aa7 | 557 | err = perf_event__synthesize_tracing_data(tool, output, evsel_list, |
743eb868 | 558 | process_synthesized_event); |
63e0c771 TZ |
559 | if (err <= 0) { |
560 | pr_err("Couldn't record tracing data.\n"); | |
561 | return err; | |
562 | } | |
d20deb64 | 563 | advance_output(rec, err); |
63e0c771 | 564 | } |
2c46dbb5 TZ |
565 | } |
566 | ||
45694aa7 | 567 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 568 | machine, "_text"); |
70162138 | 569 | if (err < 0) |
45694aa7 | 570 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
743eb868 | 571 | machine, "_stext"); |
c1a3a4b9 ACM |
572 | if (err < 0) |
573 | pr_err("Couldn't record kernel reference relocation symbol\n" | |
574 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
575 | "Check /proc/kallsyms permission or run as root.\n"); | |
b7cece76 | 576 | |
45694aa7 | 577 | err = perf_event__synthesize_modules(tool, process_synthesized_event, |
743eb868 | 578 | machine); |
c1a3a4b9 ACM |
579 | if (err < 0) |
580 | pr_err("Couldn't record kernel module information.\n" | |
581 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
582 | "Check /proc/modules permission or run as root.\n"); | |
583 | ||
a1645ce1 | 584 | if (perf_guest) |
45694aa7 | 585 | perf_session__process_machines(session, tool, |
8115d60c | 586 | perf_event__synthesize_guest_os); |
7c6a1c65 | 587 | |
bea03405 | 588 | if (!opts->target.system_wide) |
45694aa7 | 589 | perf_event__synthesize_thread_map(tool, evsel_list->threads, |
7c940c18 | 590 | process_synthesized_event, |
743eb868 | 591 | machine); |
234fbbf5 | 592 | else |
45694aa7 | 593 | perf_event__synthesize_threads(tool, process_synthesized_event, |
743eb868 | 594 | machine); |
7c6a1c65 | 595 | |
d20deb64 | 596 | if (rec->realtime_prio) { |
de9ac07b PZ |
597 | struct sched_param param; |
598 | ||
d20deb64 | 599 | param.sched_priority = rec->realtime_prio; |
de9ac07b | 600 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { |
6beba7ad | 601 | pr_err("Could not set realtime priority.\n"); |
de9ac07b PZ |
602 | exit(-1); |
603 | } | |
604 | } | |
605 | ||
764e16a3 DA |
606 | perf_evlist__enable(evsel_list); |
607 | ||
856e9660 PZ |
608 | /* |
609 | * Let the child rip | |
610 | */ | |
d4db3f16 | 611 | if (forks) |
35b9d88e | 612 | perf_evlist__start_workload(evsel_list); |
856e9660 | 613 | |
649c48a9 | 614 | for (;;) { |
d20deb64 | 615 | int hits = rec->samples; |
de9ac07b | 616 | |
d20deb64 | 617 | perf_record__mmap_read_all(rec); |
de9ac07b | 618 | |
d20deb64 | 619 | if (hits == rec->samples) { |
649c48a9 PZ |
620 | if (done) |
621 | break; | |
5c581041 | 622 | err = poll(evsel_list->pollfd, evsel_list->nr_fds, -1); |
8b412664 PZ |
623 | waking++; |
624 | } | |
625 | ||
4152ab37 ACM |
626 | if (done) |
627 | perf_evlist__disable(evsel_list); | |
de9ac07b PZ |
628 | } |
629 | ||
18483b81 | 630 | if (quiet || signr == SIGUSR1) |
b44308f5 ACM |
631 | return 0; |
632 | ||
8b412664 PZ |
633 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
634 | ||
021e9f47 IM |
635 | /* |
636 | * Approximate RIP event size: 24 bytes. | |
637 | */ | |
638 | fprintf(stderr, | |
9486aa38 | 639 | "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n", |
d20deb64 | 640 | (double)rec->bytes_written / 1024.0 / 1024.0, |
021e9f47 | 641 | output_name, |
d20deb64 | 642 | rec->bytes_written / 24); |
addc2785 | 643 | |
de9ac07b | 644 | return 0; |
39d17dac ACM |
645 | |
646 | out_delete_session: | |
647 | perf_session__delete(session); | |
648 | return err; | |
de9ac07b | 649 | } |
0e9b20b8 | 650 | |
bdfebd84 RAV |
651 | #define BRANCH_OPT(n, m) \ |
652 | { .name = n, .mode = (m) } | |
653 | ||
654 | #define BRANCH_END { .name = NULL } | |
655 | ||
656 | struct branch_mode { | |
657 | const char *name; | |
658 | int mode; | |
659 | }; | |
660 | ||
661 | static const struct branch_mode branch_modes[] = { | |
662 | BRANCH_OPT("u", PERF_SAMPLE_BRANCH_USER), | |
663 | BRANCH_OPT("k", PERF_SAMPLE_BRANCH_KERNEL), | |
664 | BRANCH_OPT("hv", PERF_SAMPLE_BRANCH_HV), | |
665 | BRANCH_OPT("any", PERF_SAMPLE_BRANCH_ANY), | |
666 | BRANCH_OPT("any_call", PERF_SAMPLE_BRANCH_ANY_CALL), | |
667 | BRANCH_OPT("any_ret", PERF_SAMPLE_BRANCH_ANY_RETURN), | |
668 | BRANCH_OPT("ind_call", PERF_SAMPLE_BRANCH_IND_CALL), | |
669 | BRANCH_END | |
670 | }; | |
671 | ||
672 | static int | |
a5aabdac | 673 | parse_branch_stack(const struct option *opt, const char *str, int unset) |
bdfebd84 RAV |
674 | { |
675 | #define ONLY_PLM \ | |
676 | (PERF_SAMPLE_BRANCH_USER |\ | |
677 | PERF_SAMPLE_BRANCH_KERNEL |\ | |
678 | PERF_SAMPLE_BRANCH_HV) | |
679 | ||
680 | uint64_t *mode = (uint64_t *)opt->value; | |
681 | const struct branch_mode *br; | |
a5aabdac | 682 | char *s, *os = NULL, *p; |
bdfebd84 RAV |
683 | int ret = -1; |
684 | ||
a5aabdac SE |
685 | if (unset) |
686 | return 0; | |
bdfebd84 | 687 | |
a5aabdac SE |
688 | /* |
689 | * cannot set it twice, -b + --branch-filter for instance | |
690 | */ | |
691 | if (*mode) | |
bdfebd84 RAV |
692 | return -1; |
693 | ||
a5aabdac SE |
694 | /* str may be NULL in case no arg is passed to -b */ |
695 | if (str) { | |
696 | /* because str is read-only */ | |
697 | s = os = strdup(str); | |
698 | if (!s) | |
699 | return -1; | |
700 | ||
701 | for (;;) { | |
702 | p = strchr(s, ','); | |
703 | if (p) | |
704 | *p = '\0'; | |
705 | ||
706 | for (br = branch_modes; br->name; br++) { | |
707 | if (!strcasecmp(s, br->name)) | |
708 | break; | |
709 | } | |
710 | if (!br->name) { | |
711 | ui__warning("unknown branch filter %s," | |
712 | " check man page\n", s); | |
713 | goto error; | |
714 | } | |
bdfebd84 | 715 | |
a5aabdac | 716 | *mode |= br->mode; |
bdfebd84 | 717 | |
a5aabdac SE |
718 | if (!p) |
719 | break; | |
bdfebd84 | 720 | |
a5aabdac SE |
721 | s = p + 1; |
722 | } | |
bdfebd84 RAV |
723 | } |
724 | ret = 0; | |
725 | ||
a5aabdac | 726 | /* default to any branch */ |
bdfebd84 | 727 | if ((*mode & ~ONLY_PLM) == 0) { |
a5aabdac | 728 | *mode = PERF_SAMPLE_BRANCH_ANY; |
bdfebd84 RAV |
729 | } |
730 | error: | |
731 | free(os); | |
732 | return ret; | |
733 | } | |
734 | ||
0e9b20b8 | 735 | static const char * const record_usage[] = { |
9e096753 MG |
736 | "perf record [<options>] [<command>]", |
737 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
738 | NULL |
739 | }; | |
740 | ||
d20deb64 ACM |
741 | /* |
742 | * XXX Ideally would be local to cmd_record() and passed to a perf_record__new | |
743 | * because we need to have access to it in perf_record__exit, that is called | |
744 | * after cmd_record() exits, but since record_options need to be accessible to | |
745 | * builtin-script, leave it here. | |
746 | * | |
747 | * At least we don't ouch it in all the other functions here directly. | |
748 | * | |
749 | * Just say no to tons of global variables, sigh. | |
750 | */ | |
751 | static struct perf_record record = { | |
752 | .opts = { | |
d20deb64 ACM |
753 | .mmap_pages = UINT_MAX, |
754 | .user_freq = UINT_MAX, | |
755 | .user_interval = ULLONG_MAX, | |
756 | .freq = 1000, | |
d1cb9fce NK |
757 | .target = { |
758 | .uses_mmap = true, | |
759 | }, | |
d20deb64 ACM |
760 | }, |
761 | .write_mode = WRITE_FORCE, | |
762 | .file_new = true, | |
763 | }; | |
7865e817 | 764 | |
d20deb64 ACM |
765 | /* |
766 | * XXX Will stay a global variable till we fix builtin-script.c to stop messing | |
767 | * with it and switch to use the library functions in perf_evlist that came | |
768 | * from builtin-record.c, i.e. use perf_record_opts, | |
769 | * perf_evlist__prepare_workload, etc instead of fork+exec'in 'perf record', | |
770 | * using pipes, etc. | |
771 | */ | |
bca647aa | 772 | const struct option record_options[] = { |
d20deb64 | 773 | OPT_CALLBACK('e', "event", &record.evlist, "event", |
86847b62 | 774 | "event selector. use 'perf list' to list available events", |
f120f9d5 | 775 | parse_events_option), |
d20deb64 | 776 | OPT_CALLBACK(0, "filter", &record.evlist, "filter", |
c171b552 | 777 | "event filter", parse_filter), |
bea03405 | 778 | OPT_STRING('p', "pid", &record.opts.target.pid, "pid", |
d6d901c2 | 779 | "record events on existing process id"), |
bea03405 | 780 | OPT_STRING('t', "tid", &record.opts.target.tid, "tid", |
d6d901c2 | 781 | "record events on existing thread id"), |
d20deb64 | 782 | OPT_INTEGER('r', "realtime", &record.realtime_prio, |
0e9b20b8 | 783 | "collect data with this RT SCHED_FIFO priority"), |
d20deb64 | 784 | OPT_BOOLEAN('D', "no-delay", &record.opts.no_delay, |
acac03fa | 785 | "collect data without buffering"), |
d20deb64 | 786 | OPT_BOOLEAN('R', "raw-samples", &record.opts.raw_samples, |
daac07b2 | 787 | "collect raw sample records from all opened counters"), |
bea03405 | 788 | OPT_BOOLEAN('a', "all-cpus", &record.opts.target.system_wide, |
0e9b20b8 | 789 | "system-wide collection from all CPUs"), |
d20deb64 | 790 | OPT_BOOLEAN('A', "append", &record.append_file, |
abaff32a | 791 | "append to the output file to do incremental profiling"), |
bea03405 | 792 | OPT_STRING('C', "cpu", &record.opts.target.cpu_list, "cpu", |
c45c6ea2 | 793 | "list of cpus to monitor"), |
d20deb64 | 794 | OPT_BOOLEAN('f', "force", &record.force, |
7865e817 | 795 | "overwrite existing data file (deprecated)"), |
d20deb64 ACM |
796 | OPT_U64('c', "count", &record.opts.user_interval, "event period to sample"), |
797 | OPT_STRING('o', "output", &record.output_name, "file", | |
abaff32a | 798 | "output file name"), |
d20deb64 | 799 | OPT_BOOLEAN('i', "no-inherit", &record.opts.no_inherit, |
2e6cdf99 | 800 | "child tasks do not inherit counters"), |
d20deb64 ACM |
801 | OPT_UINTEGER('F', "freq", &record.opts.user_freq, "profile at this frequency"), |
802 | OPT_UINTEGER('m', "mmap-pages", &record.opts.mmap_pages, | |
01c2d99b | 803 | "number of mmap data pages"), |
d20deb64 | 804 | OPT_BOOLEAN(0, "group", &record.opts.group, |
43bece79 | 805 | "put the counters into a counter group"), |
d20deb64 | 806 | OPT_BOOLEAN('g', "call-graph", &record.opts.call_graph, |
3efa1cc9 | 807 | "do call-graph (stack chain/backtrace) recording"), |
c0555642 | 808 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 809 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 810 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
d20deb64 | 811 | OPT_BOOLEAN('s', "stat", &record.opts.inherit_stat, |
649c48a9 | 812 | "per thread counts"), |
d20deb64 | 813 | OPT_BOOLEAN('d', "data", &record.opts.sample_address, |
4bba828d | 814 | "Sample addresses"), |
d20deb64 | 815 | OPT_BOOLEAN('T', "timestamp", &record.opts.sample_time, "Sample timestamps"), |
3e76ac78 | 816 | OPT_BOOLEAN('P', "period", &record.opts.period, "Sample period"), |
d20deb64 | 817 | OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples, |
649c48a9 | 818 | "don't sample"), |
d20deb64 | 819 | OPT_BOOLEAN('N', "no-buildid-cache", &record.no_buildid_cache, |
a1ac1d3c | 820 | "do not update the buildid cache"), |
d20deb64 | 821 | OPT_BOOLEAN('B', "no-buildid", &record.no_buildid, |
baa2f6ce | 822 | "do not collect buildids in perf.data"), |
d20deb64 | 823 | OPT_CALLBACK('G', "cgroup", &record.evlist, "name", |
023695d9 SE |
824 | "monitor event in cgroup name only", |
825 | parse_cgroups), | |
bea03405 NK |
826 | OPT_STRING('u', "uid", &record.opts.target.uid_str, "user", |
827 | "user to profile"), | |
a5aabdac SE |
828 | |
829 | OPT_CALLBACK_NOOPT('b', "branch-any", &record.opts.branch_stack, | |
830 | "branch any", "sample any taken branches", | |
831 | parse_branch_stack), | |
832 | ||
833 | OPT_CALLBACK('j', "branch-filter", &record.opts.branch_stack, | |
834 | "branch filter mask", "branch stack filter modes", | |
bdfebd84 | 835 | parse_branch_stack), |
0e9b20b8 IM |
836 | OPT_END() |
837 | }; | |
838 | ||
f37a291c | 839 | int cmd_record(int argc, const char **argv, const char *prefix __used) |
0e9b20b8 | 840 | { |
69aad6f1 ACM |
841 | int err = -ENOMEM; |
842 | struct perf_evsel *pos; | |
d20deb64 ACM |
843 | struct perf_evlist *evsel_list; |
844 | struct perf_record *rec = &record; | |
16ad2ffb | 845 | char errbuf[BUFSIZ]; |
0e9b20b8 | 846 | |
fbe96f29 SE |
847 | perf_header__set_cmdline(argc, argv); |
848 | ||
7e2ed097 | 849 | evsel_list = perf_evlist__new(NULL, NULL); |
361c99a6 ACM |
850 | if (evsel_list == NULL) |
851 | return -ENOMEM; | |
852 | ||
d20deb64 ACM |
853 | rec->evlist = evsel_list; |
854 | ||
bca647aa | 855 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 856 | PARSE_OPT_STOP_AT_NON_OPTION); |
d67356e7 | 857 | if (!argc && perf_target__none(&rec->opts.target)) |
bca647aa | 858 | usage_with_options(record_usage, record_options); |
0e9b20b8 | 859 | |
d20deb64 | 860 | if (rec->force && rec->append_file) { |
7865e817 FW |
861 | fprintf(stderr, "Can't overwrite and append at the same time." |
862 | " You need to choose between -f and -A"); | |
bca647aa | 863 | usage_with_options(record_usage, record_options); |
d20deb64 ACM |
864 | } else if (rec->append_file) { |
865 | rec->write_mode = WRITE_APPEND; | |
7865e817 | 866 | } else { |
d20deb64 | 867 | rec->write_mode = WRITE_FORCE; |
7865e817 FW |
868 | } |
869 | ||
bea03405 | 870 | if (nr_cgroups && !rec->opts.target.system_wide) { |
023695d9 SE |
871 | fprintf(stderr, "cgroup monitoring only available in" |
872 | " system-wide mode\n"); | |
873 | usage_with_options(record_usage, record_options); | |
874 | } | |
875 | ||
655000e7 | 876 | symbol__init(); |
baa2f6ce | 877 | |
ec80fde7 | 878 | if (symbol_conf.kptr_restrict) |
646aaea6 ACM |
879 | pr_warning( |
880 | "WARNING: Kernel address maps (/proc/{kallsyms,modules}) are restricted,\n" | |
881 | "check /proc/sys/kernel/kptr_restrict.\n\n" | |
882 | "Samples in kernel functions may not be resolved if a suitable vmlinux\n" | |
883 | "file is not found in the buildid cache or in the vmlinux path.\n\n" | |
884 | "Samples in kernel modules won't be resolved at all.\n\n" | |
885 | "If some relocation was applied (e.g. kexec) symbols may be misresolved\n" | |
886 | "even with a suitable vmlinux or kallsyms file.\n\n"); | |
ec80fde7 | 887 | |
d20deb64 | 888 | if (rec->no_buildid_cache || rec->no_buildid) |
a1ac1d3c | 889 | disable_buildid_cache(); |
655000e7 | 890 | |
361c99a6 ACM |
891 | if (evsel_list->nr_entries == 0 && |
892 | perf_evlist__add_default(evsel_list) < 0) { | |
69aad6f1 ACM |
893 | pr_err("Not enough memory for event selector list\n"); |
894 | goto out_symbol_exit; | |
bbd36e5e | 895 | } |
0e9b20b8 | 896 | |
16ad2ffb NK |
897 | err = perf_target__validate(&rec->opts.target); |
898 | if (err) { | |
899 | perf_target__strerror(&rec->opts.target, err, errbuf, BUFSIZ); | |
900 | ui__warning("%s", errbuf); | |
901 | } | |
902 | ||
903 | err = perf_target__parse_uid(&rec->opts.target); | |
904 | if (err) { | |
905 | int saved_errno = errno; | |
4bd0f2d2 | 906 | |
16ad2ffb NK |
907 | perf_target__strerror(&rec->opts.target, err, errbuf, BUFSIZ); |
908 | ui__warning("%s", errbuf); | |
909 | ||
910 | err = -saved_errno; | |
0d37aa34 | 911 | goto out_free_fd; |
16ad2ffb | 912 | } |
0d37aa34 | 913 | |
16ad2ffb | 914 | err = -ENOMEM; |
b809ac10 | 915 | if (perf_evlist__create_maps(evsel_list, &rec->opts.target) < 0) |
dd7927f4 | 916 | usage_with_options(record_usage, record_options); |
69aad6f1 | 917 | |
361c99a6 | 918 | list_for_each_entry(pos, &evsel_list->entries, node) { |
ad7f4e3f ACM |
919 | if (perf_header__push_event(pos->attr.config, event_name(pos))) |
920 | goto out_free_fd; | |
d6d901c2 | 921 | } |
5c581041 | 922 | |
d20deb64 ACM |
923 | if (rec->opts.user_interval != ULLONG_MAX) |
924 | rec->opts.default_interval = rec->opts.user_interval; | |
925 | if (rec->opts.user_freq != UINT_MAX) | |
926 | rec->opts.freq = rec->opts.user_freq; | |
f9212819 | 927 | |
7e4ff9e3 MG |
928 | /* |
929 | * User specified count overrides default frequency. | |
930 | */ | |
d20deb64 ACM |
931 | if (rec->opts.default_interval) |
932 | rec->opts.freq = 0; | |
933 | else if (rec->opts.freq) { | |
934 | rec->opts.default_interval = rec->opts.freq; | |
7e4ff9e3 MG |
935 | } else { |
936 | fprintf(stderr, "frequency and count are zero, aborting\n"); | |
39d17dac | 937 | err = -EINVAL; |
5c581041 | 938 | goto out_free_fd; |
7e4ff9e3 MG |
939 | } |
940 | ||
d20deb64 | 941 | err = __cmd_record(&record, argc, argv); |
39d17dac | 942 | out_free_fd: |
7e2ed097 | 943 | perf_evlist__delete_maps(evsel_list); |
d65a458b ACM |
944 | out_symbol_exit: |
945 | symbol__exit(); | |
39d17dac | 946 | return err; |
0e9b20b8 | 947 | } |