Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
abaff32a | 2 | /* |
bf9e1876 IM |
3 | * builtin-record.c |
4 | * | |
5 | * Builtin record command: Record the profile of a workload | |
6 | * (or a CPU, or a PID) into the perf.data output file - for | |
7 | * later analysis via perf report. | |
abaff32a | 8 | */ |
16f762a2 | 9 | #include "builtin.h" |
bf9e1876 | 10 | |
6122e4e4 | 11 | #include "util/build-id.h" |
4b6ab94e | 12 | #include <subcmd/parse-options.h> |
8ad8db37 | 13 | #include "util/parse-events.h" |
41840d21 | 14 | #include "util/config.h" |
6eda5838 | 15 | |
8f651eae | 16 | #include "util/callchain.h" |
f14d5707 | 17 | #include "util/cgroup.h" |
7c6a1c65 | 18 | #include "util/header.h" |
66e274f3 | 19 | #include "util/event.h" |
361c99a6 | 20 | #include "util/evlist.h" |
69aad6f1 | 21 | #include "util/evsel.h" |
8f28827a | 22 | #include "util/debug.h" |
e0fcfb08 | 23 | #include "util/mmap.h" |
aeb00b1a | 24 | #include "util/target.h" |
94c744b6 | 25 | #include "util/session.h" |
45694aa7 | 26 | #include "util/tool.h" |
8d06367f | 27 | #include "util/symbol.h" |
aeb00b1a | 28 | #include "util/record.h" |
a12b51c4 | 29 | #include "util/cpumap.h" |
fd78260b | 30 | #include "util/thread_map.h" |
f5fc1412 | 31 | #include "util/data.h" |
bcc84ec6 | 32 | #include "util/perf_regs.h" |
ef149c25 | 33 | #include "util/auxtrace.h" |
46bc29b9 | 34 | #include "util/tsc.h" |
f00898f4 | 35 | #include "util/parse-branch-options.h" |
bcc84ec6 | 36 | #include "util/parse-regs-options.h" |
40c7d246 | 37 | #include "util/perf_api_probe.h" |
71dc2326 | 38 | #include "util/llvm-utils.h" |
8690a2a7 | 39 | #include "util/bpf-loader.h" |
5f9cf599 | 40 | #include "util/trigger.h" |
a074865e | 41 | #include "util/perf-hooks.h" |
f13de660 | 42 | #include "util/cpu-set-sched.h" |
ea49e01c | 43 | #include "util/synthetic-events.h" |
c5e4027e | 44 | #include "util/time-utils.h" |
58db1d6e | 45 | #include "util/units.h" |
7b612e29 | 46 | #include "util/bpf-event.h" |
d99c22ea | 47 | #include "util/util.h" |
70943490 | 48 | #include "util/pfm.h" |
6953beb4 | 49 | #include "util/clockid.h" |
d8871ea7 | 50 | #include "asm/bug.h" |
c1a604df | 51 | #include "perf.h" |
7c6a1c65 | 52 | |
a43783ae | 53 | #include <errno.h> |
fd20e811 | 54 | #include <inttypes.h> |
67230479 | 55 | #include <locale.h> |
4208735d | 56 | #include <poll.h> |
d99c22ea | 57 | #include <pthread.h> |
97124d5e | 58 | #include <unistd.h> |
de9ac07b | 59 | #include <sched.h> |
9607ad3a | 60 | #include <signal.h> |
da231338 AM |
61 | #ifdef HAVE_EVENTFD_SUPPORT |
62 | #include <sys/eventfd.h> | |
63 | #endif | |
a41794cd | 64 | #include <sys/mman.h> |
4208735d | 65 | #include <sys/wait.h> |
eeb399b5 AH |
66 | #include <sys/types.h> |
67 | #include <sys/stat.h> | |
68 | #include <fcntl.h> | |
6ef81c55 | 69 | #include <linux/err.h> |
8520a98d | 70 | #include <linux/string.h> |
0693e680 | 71 | #include <linux/time64.h> |
d8f9da24 | 72 | #include <linux/zalloc.h> |
8384a260 | 73 | #include <linux/bitmap.h> |
d1e325cf | 74 | #include <sys/time.h> |
78da39fa | 75 | |
1b43b704 | 76 | struct switch_output { |
dc0c6127 | 77 | bool enabled; |
1b43b704 | 78 | bool signal; |
dc0c6127 | 79 | unsigned long size; |
bfacbe3b | 80 | unsigned long time; |
cb4e1ebb JO |
81 | const char *str; |
82 | bool set; | |
03724b2e AK |
83 | char **filenames; |
84 | int num_files; | |
85 | int cur_file; | |
1b43b704 JO |
86 | }; |
87 | ||
8c6f45a7 | 88 | struct record { |
45694aa7 | 89 | struct perf_tool tool; |
b4006796 | 90 | struct record_opts opts; |
d20deb64 | 91 | u64 bytes_written; |
8ceb41d7 | 92 | struct perf_data data; |
ef149c25 | 93 | struct auxtrace_record *itr; |
63503dba | 94 | struct evlist *evlist; |
d20deb64 | 95 | struct perf_session *session; |
bc477d79 | 96 | struct evlist *sb_evlist; |
899e5ffb | 97 | pthread_t thread_id; |
d20deb64 | 98 | int realtime_prio; |
899e5ffb | 99 | bool switch_output_event_set; |
d20deb64 | 100 | bool no_buildid; |
d2db9a98 | 101 | bool no_buildid_set; |
d20deb64 | 102 | bool no_buildid_cache; |
d2db9a98 | 103 | bool no_buildid_cache_set; |
6156681b | 104 | bool buildid_all; |
e29386c8 | 105 | bool buildid_mmap; |
ecfd7a9c | 106 | bool timestamp_filename; |
68588baf | 107 | bool timestamp_boundary; |
1b43b704 | 108 | struct switch_output switch_output; |
9f065194 | 109 | unsigned long long samples; |
8384a260 | 110 | struct mmap_cpu_mask affinity_mask; |
6d575816 | 111 | unsigned long output_max_size; /* = 0: unlimited */ |
0f82ebc4 | 112 | }; |
a21ca2ca | 113 | |
6d575816 JS |
114 | static volatile int done; |
115 | ||
dc0c6127 JO |
116 | static volatile int auxtrace_record__snapshot_started; |
117 | static DEFINE_TRIGGER(auxtrace_snapshot_trigger); | |
118 | static DEFINE_TRIGGER(switch_output_trigger); | |
119 | ||
9d2ed645 AB |
120 | static const char *affinity_tags[PERF_AFFINITY_MAX] = { |
121 | "SYS", "NODE", "CPU" | |
122 | }; | |
123 | ||
dc0c6127 JO |
124 | static bool switch_output_signal(struct record *rec) |
125 | { | |
126 | return rec->switch_output.signal && | |
127 | trigger_is_ready(&switch_output_trigger); | |
128 | } | |
129 | ||
130 | static bool switch_output_size(struct record *rec) | |
131 | { | |
132 | return rec->switch_output.size && | |
133 | trigger_is_ready(&switch_output_trigger) && | |
134 | (rec->bytes_written >= rec->switch_output.size); | |
135 | } | |
136 | ||
bfacbe3b JO |
137 | static bool switch_output_time(struct record *rec) |
138 | { | |
139 | return rec->switch_output.time && | |
140 | trigger_is_ready(&switch_output_trigger); | |
141 | } | |
142 | ||
6d575816 JS |
143 | static bool record__output_max_size_exceeded(struct record *rec) |
144 | { | |
145 | return rec->output_max_size && | |
146 | (rec->bytes_written >= rec->output_max_size); | |
147 | } | |
148 | ||
a5830532 | 149 | static int record__write(struct record *rec, struct mmap *map __maybe_unused, |
ded2b8fe | 150 | void *bf, size_t size) |
f5970550 | 151 | { |
ded2b8fe JO |
152 | struct perf_data_file *file = &rec->session->data->file; |
153 | ||
154 | if (perf_data_file__write(file, bf, size) < 0) { | |
50a9b868 JO |
155 | pr_err("failed to write perf data, error: %m\n"); |
156 | return -1; | |
f5970550 | 157 | } |
8d3eca20 | 158 | |
cf8b2e69 | 159 | rec->bytes_written += size; |
dc0c6127 | 160 | |
6d575816 JS |
161 | if (record__output_max_size_exceeded(rec) && !done) { |
162 | fprintf(stderr, "[ perf record: perf size limit reached (%" PRIu64 " KB)," | |
163 | " stopping session ]\n", | |
164 | rec->bytes_written >> 10); | |
165 | done = 1; | |
166 | } | |
167 | ||
dc0c6127 JO |
168 | if (switch_output_size(rec)) |
169 | trigger_hit(&switch_output_trigger); | |
170 | ||
8d3eca20 | 171 | return 0; |
f5970550 PZ |
172 | } |
173 | ||
ef781128 AB |
174 | static int record__aio_enabled(struct record *rec); |
175 | static int record__comp_enabled(struct record *rec); | |
5d7f4116 AB |
176 | static size_t zstd_compress(struct perf_session *session, void *dst, size_t dst_size, |
177 | void *src, size_t src_size); | |
178 | ||
d3d1af6f AB |
179 | #ifdef HAVE_AIO_SUPPORT |
180 | static int record__aio_write(struct aiocb *cblock, int trace_fd, | |
181 | void *buf, size_t size, off_t off) | |
182 | { | |
183 | int rc; | |
184 | ||
185 | cblock->aio_fildes = trace_fd; | |
186 | cblock->aio_buf = buf; | |
187 | cblock->aio_nbytes = size; | |
188 | cblock->aio_offset = off; | |
189 | cblock->aio_sigevent.sigev_notify = SIGEV_NONE; | |
190 | ||
191 | do { | |
192 | rc = aio_write(cblock); | |
193 | if (rc == 0) { | |
194 | break; | |
195 | } else if (errno != EAGAIN) { | |
196 | cblock->aio_fildes = -1; | |
197 | pr_err("failed to queue perf data, error: %m\n"); | |
198 | break; | |
199 | } | |
200 | } while (1); | |
201 | ||
202 | return rc; | |
203 | } | |
204 | ||
a5830532 | 205 | static int record__aio_complete(struct mmap *md, struct aiocb *cblock) |
d3d1af6f AB |
206 | { |
207 | void *rem_buf; | |
208 | off_t rem_off; | |
209 | size_t rem_size; | |
210 | int rc, aio_errno; | |
211 | ssize_t aio_ret, written; | |
212 | ||
213 | aio_errno = aio_error(cblock); | |
214 | if (aio_errno == EINPROGRESS) | |
215 | return 0; | |
216 | ||
217 | written = aio_ret = aio_return(cblock); | |
218 | if (aio_ret < 0) { | |
219 | if (aio_errno != EINTR) | |
220 | pr_err("failed to write perf data, error: %m\n"); | |
221 | written = 0; | |
222 | } | |
223 | ||
224 | rem_size = cblock->aio_nbytes - written; | |
225 | ||
226 | if (rem_size == 0) { | |
227 | cblock->aio_fildes = -1; | |
228 | /* | |
ef781128 AB |
229 | * md->refcount is incremented in record__aio_pushfn() for |
230 | * every aio write request started in record__aio_push() so | |
231 | * decrement it because the request is now complete. | |
d3d1af6f | 232 | */ |
80e53d11 | 233 | perf_mmap__put(&md->core); |
d3d1af6f AB |
234 | rc = 1; |
235 | } else { | |
236 | /* | |
237 | * aio write request may require restart with the | |
238 | * reminder if the kernel didn't write whole | |
239 | * chunk at once. | |
240 | */ | |
241 | rem_off = cblock->aio_offset + written; | |
242 | rem_buf = (void *)(cblock->aio_buf + written); | |
243 | record__aio_write(cblock, cblock->aio_fildes, | |
244 | rem_buf, rem_size, rem_off); | |
245 | rc = 0; | |
246 | } | |
247 | ||
248 | return rc; | |
249 | } | |
250 | ||
a5830532 | 251 | static int record__aio_sync(struct mmap *md, bool sync_all) |
d3d1af6f | 252 | { |
93f20c0f AB |
253 | struct aiocb **aiocb = md->aio.aiocb; |
254 | struct aiocb *cblocks = md->aio.cblocks; | |
d3d1af6f | 255 | struct timespec timeout = { 0, 1000 * 1000 * 1 }; /* 1ms */ |
93f20c0f | 256 | int i, do_suspend; |
d3d1af6f AB |
257 | |
258 | do { | |
93f20c0f AB |
259 | do_suspend = 0; |
260 | for (i = 0; i < md->aio.nr_cblocks; ++i) { | |
261 | if (cblocks[i].aio_fildes == -1 || record__aio_complete(md, &cblocks[i])) { | |
262 | if (sync_all) | |
263 | aiocb[i] = NULL; | |
264 | else | |
265 | return i; | |
266 | } else { | |
267 | /* | |
268 | * Started aio write is not complete yet | |
269 | * so it has to be waited before the | |
270 | * next allocation. | |
271 | */ | |
272 | aiocb[i] = &cblocks[i]; | |
273 | do_suspend = 1; | |
274 | } | |
275 | } | |
276 | if (!do_suspend) | |
277 | return -1; | |
d3d1af6f | 278 | |
93f20c0f | 279 | while (aio_suspend((const struct aiocb **)aiocb, md->aio.nr_cblocks, &timeout)) { |
d3d1af6f AB |
280 | if (!(errno == EAGAIN || errno == EINTR)) |
281 | pr_err("failed to sync perf data, error: %m\n"); | |
282 | } | |
283 | } while (1); | |
284 | } | |
285 | ||
ef781128 AB |
286 | struct record_aio { |
287 | struct record *rec; | |
288 | void *data; | |
289 | size_t size; | |
290 | }; | |
291 | ||
a5830532 | 292 | static int record__aio_pushfn(struct mmap *map, void *to, void *buf, size_t size) |
d3d1af6f | 293 | { |
ef781128 | 294 | struct record_aio *aio = to; |
d3d1af6f | 295 | |
ef781128 | 296 | /* |
547740f7 | 297 | * map->core.base data pointed by buf is copied into free map->aio.data[] buffer |
ef781128 AB |
298 | * to release space in the kernel buffer as fast as possible, calling |
299 | * perf_mmap__consume() from perf_mmap__push() function. | |
300 | * | |
301 | * That lets the kernel to proceed with storing more profiling data into | |
302 | * the kernel buffer earlier than other per-cpu kernel buffers are handled. | |
303 | * | |
304 | * Coping can be done in two steps in case the chunk of profiling data | |
305 | * crosses the upper bound of the kernel buffer. In this case we first move | |
306 | * part of data from map->start till the upper bound and then the reminder | |
307 | * from the beginning of the kernel buffer till the end of the data chunk. | |
308 | */ | |
309 | ||
310 | if (record__comp_enabled(aio->rec)) { | |
311 | size = zstd_compress(aio->rec->session, aio->data + aio->size, | |
bf59b305 | 312 | mmap__mmap_len(map) - aio->size, |
ef781128 AB |
313 | buf, size); |
314 | } else { | |
315 | memcpy(aio->data + aio->size, buf, size); | |
316 | } | |
317 | ||
318 | if (!aio->size) { | |
319 | /* | |
320 | * Increment map->refcount to guard map->aio.data[] buffer | |
321 | * from premature deallocation because map object can be | |
322 | * released earlier than aio write request started on | |
323 | * map->aio.data[] buffer is complete. | |
324 | * | |
325 | * perf_mmap__put() is done at record__aio_complete() | |
326 | * after started aio request completion or at record__aio_push() | |
327 | * if the request failed to start. | |
328 | */ | |
e75710f0 | 329 | perf_mmap__get(&map->core); |
ef781128 AB |
330 | } |
331 | ||
332 | aio->size += size; | |
333 | ||
334 | return size; | |
335 | } | |
336 | ||
a5830532 | 337 | static int record__aio_push(struct record *rec, struct mmap *map, off_t *off) |
ef781128 AB |
338 | { |
339 | int ret, idx; | |
340 | int trace_fd = rec->session->data->file.fd; | |
341 | struct record_aio aio = { .rec = rec, .size = 0 }; | |
d3d1af6f | 342 | |
ef781128 AB |
343 | /* |
344 | * Call record__aio_sync() to wait till map->aio.data[] buffer | |
345 | * becomes available after previous aio write operation. | |
346 | */ | |
347 | ||
348 | idx = record__aio_sync(map, false); | |
349 | aio.data = map->aio.data[idx]; | |
350 | ret = perf_mmap__push(map, &aio, record__aio_pushfn); | |
351 | if (ret != 0) /* ret > 0 - no data, ret < 0 - error */ | |
352 | return ret; | |
353 | ||
354 | rec->samples++; | |
355 | ret = record__aio_write(&(map->aio.cblocks[idx]), trace_fd, aio.data, aio.size, *off); | |
d3d1af6f | 356 | if (!ret) { |
ef781128 AB |
357 | *off += aio.size; |
358 | rec->bytes_written += aio.size; | |
d3d1af6f AB |
359 | if (switch_output_size(rec)) |
360 | trigger_hit(&switch_output_trigger); | |
ef781128 AB |
361 | } else { |
362 | /* | |
363 | * Decrement map->refcount incremented in record__aio_pushfn() | |
364 | * back if record__aio_write() operation failed to start, otherwise | |
365 | * map->refcount is decremented in record__aio_complete() after | |
366 | * aio write operation finishes successfully. | |
367 | */ | |
80e53d11 | 368 | perf_mmap__put(&map->core); |
d3d1af6f AB |
369 | } |
370 | ||
371 | return ret; | |
372 | } | |
373 | ||
374 | static off_t record__aio_get_pos(int trace_fd) | |
375 | { | |
376 | return lseek(trace_fd, 0, SEEK_CUR); | |
377 | } | |
378 | ||
379 | static void record__aio_set_pos(int trace_fd, off_t pos) | |
380 | { | |
381 | lseek(trace_fd, pos, SEEK_SET); | |
382 | } | |
383 | ||
384 | static void record__aio_mmap_read_sync(struct record *rec) | |
385 | { | |
386 | int i; | |
63503dba | 387 | struct evlist *evlist = rec->evlist; |
a5830532 | 388 | struct mmap *maps = evlist->mmap; |
d3d1af6f | 389 | |
ef781128 | 390 | if (!record__aio_enabled(rec)) |
d3d1af6f AB |
391 | return; |
392 | ||
c976ee11 | 393 | for (i = 0; i < evlist->core.nr_mmaps; i++) { |
a5830532 | 394 | struct mmap *map = &maps[i]; |
d3d1af6f | 395 | |
547740f7 | 396 | if (map->core.base) |
93f20c0f | 397 | record__aio_sync(map, true); |
d3d1af6f AB |
398 | } |
399 | } | |
400 | ||
401 | static int nr_cblocks_default = 1; | |
93f20c0f | 402 | static int nr_cblocks_max = 4; |
d3d1af6f AB |
403 | |
404 | static int record__aio_parse(const struct option *opt, | |
93f20c0f | 405 | const char *str, |
d3d1af6f AB |
406 | int unset) |
407 | { | |
408 | struct record_opts *opts = (struct record_opts *)opt->value; | |
409 | ||
93f20c0f | 410 | if (unset) { |
d3d1af6f | 411 | opts->nr_cblocks = 0; |
93f20c0f AB |
412 | } else { |
413 | if (str) | |
414 | opts->nr_cblocks = strtol(str, NULL, 0); | |
415 | if (!opts->nr_cblocks) | |
416 | opts->nr_cblocks = nr_cblocks_default; | |
417 | } | |
d3d1af6f AB |
418 | |
419 | return 0; | |
420 | } | |
421 | #else /* HAVE_AIO_SUPPORT */ | |
93f20c0f AB |
422 | static int nr_cblocks_max = 0; |
423 | ||
a5830532 | 424 | static int record__aio_push(struct record *rec __maybe_unused, struct mmap *map __maybe_unused, |
ef781128 | 425 | off_t *off __maybe_unused) |
d3d1af6f AB |
426 | { |
427 | return -1; | |
428 | } | |
429 | ||
430 | static off_t record__aio_get_pos(int trace_fd __maybe_unused) | |
431 | { | |
432 | return -1; | |
433 | } | |
434 | ||
435 | static void record__aio_set_pos(int trace_fd __maybe_unused, off_t pos __maybe_unused) | |
436 | { | |
437 | } | |
438 | ||
439 | static void record__aio_mmap_read_sync(struct record *rec __maybe_unused) | |
440 | { | |
441 | } | |
442 | #endif | |
443 | ||
444 | static int record__aio_enabled(struct record *rec) | |
445 | { | |
446 | return rec->opts.nr_cblocks > 0; | |
447 | } | |
448 | ||
470530bb AB |
449 | #define MMAP_FLUSH_DEFAULT 1 |
450 | static int record__mmap_flush_parse(const struct option *opt, | |
451 | const char *str, | |
452 | int unset) | |
453 | { | |
454 | int flush_max; | |
455 | struct record_opts *opts = (struct record_opts *)opt->value; | |
456 | static struct parse_tag tags[] = { | |
457 | { .tag = 'B', .mult = 1 }, | |
458 | { .tag = 'K', .mult = 1 << 10 }, | |
459 | { .tag = 'M', .mult = 1 << 20 }, | |
460 | { .tag = 'G', .mult = 1 << 30 }, | |
461 | { .tag = 0 }, | |
462 | }; | |
463 | ||
464 | if (unset) | |
465 | return 0; | |
466 | ||
467 | if (str) { | |
468 | opts->mmap_flush = parse_tag_value(str, tags); | |
469 | if (opts->mmap_flush == (int)-1) | |
470 | opts->mmap_flush = strtol(str, NULL, 0); | |
471 | } | |
472 | ||
473 | if (!opts->mmap_flush) | |
474 | opts->mmap_flush = MMAP_FLUSH_DEFAULT; | |
475 | ||
9521b5f2 | 476 | flush_max = evlist__mmap_size(opts->mmap_pages); |
470530bb AB |
477 | flush_max /= 4; |
478 | if (opts->mmap_flush > flush_max) | |
479 | opts->mmap_flush = flush_max; | |
480 | ||
481 | return 0; | |
482 | } | |
483 | ||
504c1ad1 AB |
484 | #ifdef HAVE_ZSTD_SUPPORT |
485 | static unsigned int comp_level_default = 1; | |
486 | ||
487 | static int record__parse_comp_level(const struct option *opt, const char *str, int unset) | |
488 | { | |
489 | struct record_opts *opts = opt->value; | |
490 | ||
491 | if (unset) { | |
492 | opts->comp_level = 0; | |
493 | } else { | |
494 | if (str) | |
495 | opts->comp_level = strtol(str, NULL, 0); | |
496 | if (!opts->comp_level) | |
497 | opts->comp_level = comp_level_default; | |
498 | } | |
499 | ||
500 | return 0; | |
501 | } | |
502 | #endif | |
51255a8a AB |
503 | static unsigned int comp_level_max = 22; |
504 | ||
42e1fd80 AB |
505 | static int record__comp_enabled(struct record *rec) |
506 | { | |
507 | return rec->opts.comp_level > 0; | |
508 | } | |
509 | ||
45694aa7 | 510 | static int process_synthesized_event(struct perf_tool *tool, |
d20deb64 | 511 | union perf_event *event, |
1d037ca1 IT |
512 | struct perf_sample *sample __maybe_unused, |
513 | struct machine *machine __maybe_unused) | |
234fbbf5 | 514 | { |
8c6f45a7 | 515 | struct record *rec = container_of(tool, struct record, tool); |
ded2b8fe | 516 | return record__write(rec, NULL, event, event->header.size); |
234fbbf5 ACM |
517 | } |
518 | ||
d99c22ea SE |
519 | static int process_locked_synthesized_event(struct perf_tool *tool, |
520 | union perf_event *event, | |
521 | struct perf_sample *sample __maybe_unused, | |
522 | struct machine *machine __maybe_unused) | |
523 | { | |
524 | static pthread_mutex_t synth_lock = PTHREAD_MUTEX_INITIALIZER; | |
525 | int ret; | |
526 | ||
527 | pthread_mutex_lock(&synth_lock); | |
528 | ret = process_synthesized_event(tool, event, sample, machine); | |
529 | pthread_mutex_unlock(&synth_lock); | |
530 | return ret; | |
531 | } | |
532 | ||
a5830532 | 533 | static int record__pushfn(struct mmap *map, void *to, void *bf, size_t size) |
d37f1586 ACM |
534 | { |
535 | struct record *rec = to; | |
536 | ||
5d7f4116 | 537 | if (record__comp_enabled(rec)) { |
bf59b305 | 538 | size = zstd_compress(rec->session, map->data, mmap__mmap_len(map), bf, size); |
5d7f4116 AB |
539 | bf = map->data; |
540 | } | |
541 | ||
d37f1586 | 542 | rec->samples++; |
ded2b8fe | 543 | return record__write(rec, map, bf, size); |
d37f1586 ACM |
544 | } |
545 | ||
2dd6d8a1 AH |
546 | static volatile int signr = -1; |
547 | static volatile int child_finished; | |
da231338 AM |
548 | #ifdef HAVE_EVENTFD_SUPPORT |
549 | static int done_fd = -1; | |
550 | #endif | |
c0bdc1c4 | 551 | |
2dd6d8a1 AH |
552 | static void sig_handler(int sig) |
553 | { | |
554 | if (sig == SIGCHLD) | |
555 | child_finished = 1; | |
556 | else | |
557 | signr = sig; | |
558 | ||
559 | done = 1; | |
da231338 AM |
560 | #ifdef HAVE_EVENTFD_SUPPORT |
561 | { | |
562 | u64 tmp = 1; | |
563 | /* | |
564 | * It is possible for this signal handler to run after done is checked | |
565 | * in the main loop, but before the perf counter fds are polled. If this | |
566 | * happens, the poll() will continue to wait even though done is set, | |
567 | * and will only break out if either another signal is received, or the | |
568 | * counters are ready for read. To ensure the poll() doesn't sleep when | |
569 | * done is set, use an eventfd (done_fd) to wake up the poll(). | |
570 | */ | |
571 | if (write(done_fd, &tmp, sizeof(tmp)) < 0) | |
572 | pr_err("failed to signal wakeup fd, error: %m\n"); | |
573 | } | |
574 | #endif // HAVE_EVENTFD_SUPPORT | |
2dd6d8a1 AH |
575 | } |
576 | ||
a074865e WN |
577 | static void sigsegv_handler(int sig) |
578 | { | |
579 | perf_hooks__recover(); | |
580 | sighandler_dump_stack(sig); | |
581 | } | |
582 | ||
2dd6d8a1 AH |
583 | static void record__sig_exit(void) |
584 | { | |
585 | if (signr == -1) | |
586 | return; | |
587 | ||
588 | signal(signr, SIG_DFL); | |
589 | raise(signr); | |
590 | } | |
591 | ||
e31f0d01 AH |
592 | #ifdef HAVE_AUXTRACE_SUPPORT |
593 | ||
ef149c25 | 594 | static int record__process_auxtrace(struct perf_tool *tool, |
a5830532 | 595 | struct mmap *map, |
ef149c25 AH |
596 | union perf_event *event, void *data1, |
597 | size_t len1, void *data2, size_t len2) | |
598 | { | |
599 | struct record *rec = container_of(tool, struct record, tool); | |
8ceb41d7 | 600 | struct perf_data *data = &rec->data; |
ef149c25 AH |
601 | size_t padding; |
602 | u8 pad[8] = {0}; | |
603 | ||
46e201ef | 604 | if (!perf_data__is_pipe(data) && perf_data__is_single_file(data)) { |
99fa2984 | 605 | off_t file_offset; |
8ceb41d7 | 606 | int fd = perf_data__fd(data); |
99fa2984 AH |
607 | int err; |
608 | ||
609 | file_offset = lseek(fd, 0, SEEK_CUR); | |
610 | if (file_offset == -1) | |
611 | return -1; | |
612 | err = auxtrace_index__auxtrace_event(&rec->session->auxtrace_index, | |
613 | event, file_offset); | |
614 | if (err) | |
615 | return err; | |
616 | } | |
617 | ||
ef149c25 AH |
618 | /* event.auxtrace.size includes padding, see __auxtrace_mmap__read() */ |
619 | padding = (len1 + len2) & 7; | |
620 | if (padding) | |
621 | padding = 8 - padding; | |
622 | ||
ded2b8fe JO |
623 | record__write(rec, map, event, event->header.size); |
624 | record__write(rec, map, data1, len1); | |
ef149c25 | 625 | if (len2) |
ded2b8fe JO |
626 | record__write(rec, map, data2, len2); |
627 | record__write(rec, map, &pad, padding); | |
ef149c25 AH |
628 | |
629 | return 0; | |
630 | } | |
631 | ||
632 | static int record__auxtrace_mmap_read(struct record *rec, | |
a5830532 | 633 | struct mmap *map) |
ef149c25 AH |
634 | { |
635 | int ret; | |
636 | ||
e035f4ca | 637 | ret = auxtrace_mmap__read(map, rec->itr, &rec->tool, |
ef149c25 AH |
638 | record__process_auxtrace); |
639 | if (ret < 0) | |
640 | return ret; | |
641 | ||
642 | if (ret) | |
643 | rec->samples++; | |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
2dd6d8a1 | 648 | static int record__auxtrace_mmap_read_snapshot(struct record *rec, |
a5830532 | 649 | struct mmap *map) |
2dd6d8a1 AH |
650 | { |
651 | int ret; | |
652 | ||
e035f4ca | 653 | ret = auxtrace_mmap__read_snapshot(map, rec->itr, &rec->tool, |
2dd6d8a1 AH |
654 | record__process_auxtrace, |
655 | rec->opts.auxtrace_snapshot_size); | |
656 | if (ret < 0) | |
657 | return ret; | |
658 | ||
659 | if (ret) | |
660 | rec->samples++; | |
661 | ||
662 | return 0; | |
663 | } | |
664 | ||
665 | static int record__auxtrace_read_snapshot_all(struct record *rec) | |
666 | { | |
667 | int i; | |
668 | int rc = 0; | |
669 | ||
c976ee11 | 670 | for (i = 0; i < rec->evlist->core.nr_mmaps; i++) { |
a5830532 | 671 | struct mmap *map = &rec->evlist->mmap[i]; |
2dd6d8a1 | 672 | |
e035f4ca | 673 | if (!map->auxtrace_mmap.base) |
2dd6d8a1 AH |
674 | continue; |
675 | ||
e035f4ca | 676 | if (record__auxtrace_mmap_read_snapshot(rec, map) != 0) { |
2dd6d8a1 AH |
677 | rc = -1; |
678 | goto out; | |
679 | } | |
680 | } | |
681 | out: | |
682 | return rc; | |
683 | } | |
684 | ||
ce7b0e42 | 685 | static void record__read_auxtrace_snapshot(struct record *rec, bool on_exit) |
2dd6d8a1 AH |
686 | { |
687 | pr_debug("Recording AUX area tracing snapshot\n"); | |
688 | if (record__auxtrace_read_snapshot_all(rec) < 0) { | |
5f9cf599 | 689 | trigger_error(&auxtrace_snapshot_trigger); |
2dd6d8a1 | 690 | } else { |
ce7b0e42 | 691 | if (auxtrace_record__snapshot_finish(rec->itr, on_exit)) |
5f9cf599 WN |
692 | trigger_error(&auxtrace_snapshot_trigger); |
693 | else | |
694 | trigger_ready(&auxtrace_snapshot_trigger); | |
2dd6d8a1 AH |
695 | } |
696 | } | |
697 | ||
ce7b0e42 AS |
698 | static int record__auxtrace_snapshot_exit(struct record *rec) |
699 | { | |
700 | if (trigger_is_error(&auxtrace_snapshot_trigger)) | |
701 | return 0; | |
702 | ||
703 | if (!auxtrace_record__snapshot_started && | |
704 | auxtrace_record__snapshot_start(rec->itr)) | |
705 | return -1; | |
706 | ||
707 | record__read_auxtrace_snapshot(rec, true); | |
708 | if (trigger_is_error(&auxtrace_snapshot_trigger)) | |
709 | return -1; | |
710 | ||
711 | return 0; | |
712 | } | |
713 | ||
4b5ea3bd AH |
714 | static int record__auxtrace_init(struct record *rec) |
715 | { | |
716 | int err; | |
717 | ||
718 | if (!rec->itr) { | |
719 | rec->itr = auxtrace_record__init(rec->evlist, &err); | |
720 | if (err) | |
721 | return err; | |
722 | } | |
723 | ||
724 | err = auxtrace_parse_snapshot_options(rec->itr, &rec->opts, | |
725 | rec->opts.auxtrace_snapshot_opts); | |
726 | if (err) | |
727 | return err; | |
728 | ||
c0a6de06 AH |
729 | err = auxtrace_parse_sample_options(rec->itr, rec->evlist, &rec->opts, |
730 | rec->opts.auxtrace_sample_opts); | |
731 | if (err) | |
732 | return err; | |
733 | ||
d58b3f7e AH |
734 | auxtrace_regroup_aux_output(rec->evlist); |
735 | ||
4b5ea3bd AH |
736 | return auxtrace_parse_filters(rec->evlist); |
737 | } | |
738 | ||
e31f0d01 AH |
739 | #else |
740 | ||
741 | static inline | |
742 | int record__auxtrace_mmap_read(struct record *rec __maybe_unused, | |
a5830532 | 743 | struct mmap *map __maybe_unused) |
e31f0d01 AH |
744 | { |
745 | return 0; | |
746 | } | |
747 | ||
2dd6d8a1 | 748 | static inline |
ce7b0e42 AS |
749 | void record__read_auxtrace_snapshot(struct record *rec __maybe_unused, |
750 | bool on_exit __maybe_unused) | |
de9ac07b | 751 | { |
f7b7c26e PZ |
752 | } |
753 | ||
2dd6d8a1 AH |
754 | static inline |
755 | int auxtrace_record__snapshot_start(struct auxtrace_record *itr __maybe_unused) | |
f7b7c26e | 756 | { |
2dd6d8a1 | 757 | return 0; |
de9ac07b PZ |
758 | } |
759 | ||
ce7b0e42 AS |
760 | static inline |
761 | int record__auxtrace_snapshot_exit(struct record *rec __maybe_unused) | |
762 | { | |
763 | return 0; | |
764 | } | |
765 | ||
4b5ea3bd AH |
766 | static int record__auxtrace_init(struct record *rec __maybe_unused) |
767 | { | |
768 | return 0; | |
769 | } | |
770 | ||
2dd6d8a1 AH |
771 | #endif |
772 | ||
246eba8e AH |
773 | static int record__config_text_poke(struct evlist *evlist) |
774 | { | |
775 | struct evsel *evsel; | |
776 | int err; | |
777 | ||
778 | /* Nothing to do if text poke is already configured */ | |
779 | evlist__for_each_entry(evlist, evsel) { | |
780 | if (evsel->core.attr.text_poke) | |
781 | return 0; | |
782 | } | |
783 | ||
784 | err = parse_events(evlist, "dummy:u", NULL); | |
785 | if (err) | |
786 | return err; | |
787 | ||
788 | evsel = evlist__last(evlist); | |
789 | ||
790 | evsel->core.attr.freq = 0; | |
791 | evsel->core.attr.sample_period = 1; | |
792 | evsel->core.attr.text_poke = 1; | |
793 | evsel->core.attr.ksymbol = 1; | |
794 | ||
795 | evsel->core.system_wide = true; | |
796 | evsel->no_aux_samples = true; | |
797 | evsel->immediate = true; | |
798 | ||
799 | /* Text poke must be collected on all CPUs */ | |
800 | perf_cpu_map__put(evsel->core.own_cpus); | |
801 | evsel->core.own_cpus = perf_cpu_map__new(NULL); | |
802 | perf_cpu_map__put(evsel->core.cpus); | |
803 | evsel->core.cpus = perf_cpu_map__get(evsel->core.own_cpus); | |
804 | ||
805 | evsel__set_sample_bit(evsel, TIME); | |
806 | ||
807 | return 0; | |
808 | } | |
809 | ||
eeb399b5 AH |
810 | static bool record__kcore_readable(struct machine *machine) |
811 | { | |
812 | char kcore[PATH_MAX]; | |
813 | int fd; | |
814 | ||
815 | scnprintf(kcore, sizeof(kcore), "%s/proc/kcore", machine->root_dir); | |
816 | ||
817 | fd = open(kcore, O_RDONLY); | |
818 | if (fd < 0) | |
819 | return false; | |
820 | ||
821 | close(fd); | |
822 | ||
823 | return true; | |
824 | } | |
825 | ||
826 | static int record__kcore_copy(struct machine *machine, struct perf_data *data) | |
827 | { | |
828 | char from_dir[PATH_MAX]; | |
829 | char kcore_dir[PATH_MAX]; | |
830 | int ret; | |
831 | ||
832 | snprintf(from_dir, sizeof(from_dir), "%s/proc", machine->root_dir); | |
833 | ||
834 | ret = perf_data__make_kcore_dir(data, kcore_dir, sizeof(kcore_dir)); | |
835 | if (ret) | |
836 | return ret; | |
837 | ||
838 | return kcore_copy(from_dir, kcore_dir); | |
839 | } | |
840 | ||
cda57a8c | 841 | static int record__mmap_evlist(struct record *rec, |
63503dba | 842 | struct evlist *evlist) |
cda57a8c WN |
843 | { |
844 | struct record_opts *opts = &rec->opts; | |
c0a6de06 AH |
845 | bool auxtrace_overwrite = opts->auxtrace_snapshot_mode || |
846 | opts->auxtrace_sample_mode; | |
cda57a8c WN |
847 | char msg[512]; |
848 | ||
f13de660 AB |
849 | if (opts->affinity != PERF_AFFINITY_SYS) |
850 | cpu__setup_cpunode_map(); | |
851 | ||
9521b5f2 | 852 | if (evlist__mmap_ex(evlist, opts->mmap_pages, |
cda57a8c | 853 | opts->auxtrace_mmap_pages, |
c0a6de06 | 854 | auxtrace_overwrite, |
470530bb | 855 | opts->nr_cblocks, opts->affinity, |
51255a8a | 856 | opts->mmap_flush, opts->comp_level) < 0) { |
cda57a8c WN |
857 | if (errno == EPERM) { |
858 | pr_err("Permission error mapping pages.\n" | |
859 | "Consider increasing " | |
860 | "/proc/sys/kernel/perf_event_mlock_kb,\n" | |
861 | "or try again with a smaller value of -m/--mmap_pages.\n" | |
862 | "(current value: %u,%u)\n", | |
863 | opts->mmap_pages, opts->auxtrace_mmap_pages); | |
864 | return -errno; | |
865 | } else { | |
866 | pr_err("failed to mmap with %d (%s)\n", errno, | |
c8b5f2c9 | 867 | str_error_r(errno, msg, sizeof(msg))); |
cda57a8c WN |
868 | if (errno) |
869 | return -errno; | |
870 | else | |
871 | return -EINVAL; | |
872 | } | |
873 | } | |
874 | return 0; | |
875 | } | |
876 | ||
877 | static int record__mmap(struct record *rec) | |
878 | { | |
879 | return record__mmap_evlist(rec, rec->evlist); | |
880 | } | |
881 | ||
8c6f45a7 | 882 | static int record__open(struct record *rec) |
dd7927f4 | 883 | { |
d6195a6a | 884 | char msg[BUFSIZ]; |
32dcd021 | 885 | struct evsel *pos; |
63503dba | 886 | struct evlist *evlist = rec->evlist; |
d20deb64 | 887 | struct perf_session *session = rec->session; |
b4006796 | 888 | struct record_opts *opts = &rec->opts; |
8d3eca20 | 889 | int rc = 0; |
dd7927f4 | 890 | |
d3dbf43c | 891 | /* |
0a892c1c IR |
892 | * For initial_delay or system wide, we need to add a dummy event so |
893 | * that we can track PERF_RECORD_MMAP to cover the delay of waiting or | |
894 | * event synthesis. | |
d3dbf43c | 895 | */ |
0a892c1c | 896 | if (opts->initial_delay || target__has_cpu(&opts->target)) { |
e80db255 | 897 | pos = evlist__get_tracking_event(evlist); |
442ad225 AH |
898 | if (!evsel__is_dummy_event(pos)) { |
899 | /* Set up dummy event. */ | |
facbf0b9 | 900 | if (evlist__add_dummy(evlist)) |
442ad225 AH |
901 | return -ENOMEM; |
902 | pos = evlist__last(evlist); | |
e80db255 | 903 | evlist__set_tracking_event(evlist, pos); |
442ad225 AH |
904 | } |
905 | ||
0a892c1c IR |
906 | /* |
907 | * Enable the dummy event when the process is forked for | |
908 | * initial_delay, immediately for system wide. | |
909 | */ | |
442ad225 | 910 | if (opts->initial_delay && !pos->immediate) |
0a892c1c IR |
911 | pos->core.attr.enable_on_exec = 1; |
912 | else | |
913 | pos->immediate = 1; | |
d3dbf43c ACM |
914 | } |
915 | ||
78e1bc25 | 916 | evlist__config(evlist, opts, &callchain_param); |
cac21425 | 917 | |
e5cadb93 | 918 | evlist__for_each_entry(evlist, pos) { |
dd7927f4 | 919 | try_again: |
af663bd0 | 920 | if (evsel__open(pos, pos->core.cpus, pos->core.threads) < 0) { |
ae430892 | 921 | if (evsel__fallback(pos, errno, msg, sizeof(msg))) { |
bb963e16 | 922 | if (verbose > 0) |
c0a54341 | 923 | ui__warning("%s\n", msg); |
d6d901c2 ZY |
924 | goto try_again; |
925 | } | |
cf99ad14 AK |
926 | if ((errno == EINVAL || errno == EBADF) && |
927 | pos->leader != pos && | |
928 | pos->weak_group) { | |
64b4778b | 929 | pos = evlist__reset_weak_group(evlist, pos, true); |
cf99ad14 AK |
930 | goto try_again; |
931 | } | |
56e52e85 | 932 | rc = -errno; |
2bb72dbb | 933 | evsel__open_strerror(pos, &opts->target, errno, msg, sizeof(msg)); |
56e52e85 | 934 | ui__error("%s\n", msg); |
8d3eca20 | 935 | goto out; |
c171b552 | 936 | } |
bfd8f72c AK |
937 | |
938 | pos->supported = true; | |
c171b552 | 939 | } |
a43d3f08 | 940 | |
78e1bc25 | 941 | if (symbol_conf.kptr_restrict && !evlist__exclude_kernel(evlist)) { |
c8b567c8 ACM |
942 | pr_warning( |
943 | "WARNING: Kernel address maps (/proc/{kallsyms,modules}) are restricted,\n" | |
944 | "check /proc/sys/kernel/kptr_restrict and /proc/sys/kernel/perf_event_paranoid.\n\n" | |
945 | "Samples in kernel functions may not be resolved if a suitable vmlinux\n" | |
946 | "file is not found in the buildid cache or in the vmlinux path.\n\n" | |
947 | "Samples in kernel modules won't be resolved at all.\n\n" | |
948 | "If some relocation was applied (e.g. kexec) symbols may be misresolved\n" | |
949 | "even with a suitable vmlinux or kallsyms file.\n\n"); | |
950 | } | |
951 | ||
24bf91a7 | 952 | if (evlist__apply_filters(evlist, &pos)) { |
62d94b00 | 953 | pr_err("failed to set filter \"%s\" on event %s with %d (%s)\n", |
8ab2e96d | 954 | pos->filter, evsel__name(pos), errno, |
c8b5f2c9 | 955 | str_error_r(errno, msg, sizeof(msg))); |
8d3eca20 | 956 | rc = -1; |
5d8bb1ec MP |
957 | goto out; |
958 | } | |
959 | ||
cda57a8c WN |
960 | rc = record__mmap(rec); |
961 | if (rc) | |
8d3eca20 | 962 | goto out; |
0a27d7f9 | 963 | |
563aecb2 | 964 | session->evlist = evlist; |
7b56cce2 | 965 | perf_session__set_id_hdr_size(session); |
8d3eca20 DA |
966 | out: |
967 | return rc; | |
16c8a109 PZ |
968 | } |
969 | ||
e3d59112 NK |
970 | static int process_sample_event(struct perf_tool *tool, |
971 | union perf_event *event, | |
972 | struct perf_sample *sample, | |
32dcd021 | 973 | struct evsel *evsel, |
e3d59112 NK |
974 | struct machine *machine) |
975 | { | |
976 | struct record *rec = container_of(tool, struct record, tool); | |
977 | ||
68588baf JY |
978 | if (rec->evlist->first_sample_time == 0) |
979 | rec->evlist->first_sample_time = sample->time; | |
980 | ||
981 | rec->evlist->last_sample_time = sample->time; | |
e3d59112 | 982 | |
68588baf JY |
983 | if (rec->buildid_all) |
984 | return 0; | |
985 | ||
986 | rec->samples++; | |
e3d59112 NK |
987 | return build_id__mark_dso_hit(tool, event, sample, evsel, machine); |
988 | } | |
989 | ||
8c6f45a7 | 990 | static int process_buildids(struct record *rec) |
6122e4e4 | 991 | { |
f5fc1412 | 992 | struct perf_session *session = rec->session; |
6122e4e4 | 993 | |
45112e89 | 994 | if (perf_data__size(&rec->data) == 0) |
9f591fd7 ACM |
995 | return 0; |
996 | ||
00dc8657 NK |
997 | /* |
998 | * During this process, it'll load kernel map and replace the | |
999 | * dso->long_name to a real pathname it found. In this case | |
1000 | * we prefer the vmlinux path like | |
1001 | * /lib/modules/3.16.4/build/vmlinux | |
1002 | * | |
1003 | * rather than build-id path (in debug directory). | |
1004 | * $HOME/.debug/.build-id/f0/6e17aa50adf4d00b88925e03775de107611551 | |
1005 | */ | |
1006 | symbol_conf.ignore_vmlinux_buildid = true; | |
1007 | ||
6156681b NK |
1008 | /* |
1009 | * If --buildid-all is given, it marks all DSO regardless of hits, | |
68588baf JY |
1010 | * so no need to process samples. But if timestamp_boundary is enabled, |
1011 | * it still needs to walk on all samples to get the timestamps of | |
1012 | * first/last samples. | |
6156681b | 1013 | */ |
68588baf | 1014 | if (rec->buildid_all && !rec->timestamp_boundary) |
6156681b NK |
1015 | rec->tool.sample = NULL; |
1016 | ||
b7b61cbe | 1017 | return perf_session__process_events(session); |
6122e4e4 ACM |
1018 | } |
1019 | ||
8115d60c | 1020 | static void perf_event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
1021 | { |
1022 | int err; | |
45694aa7 | 1023 | struct perf_tool *tool = data; |
a1645ce1 ZY |
1024 | /* |
1025 | *As for guest kernel when processing subcommand record&report, | |
1026 | *we arrange module mmap prior to guest kernel mmap and trigger | |
1027 | *a preload dso because default guest module symbols are loaded | |
1028 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
1029 | *method is used to avoid symbol missing when the first addr is | |
1030 | *in module instead of in guest kernel. | |
1031 | */ | |
45694aa7 | 1032 | err = perf_event__synthesize_modules(tool, process_synthesized_event, |
743eb868 | 1033 | machine); |
a1645ce1 ZY |
1034 | if (err < 0) |
1035 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 1036 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 1037 | |
a1645ce1 ZY |
1038 | /* |
1039 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
1040 | * have no _text sometimes. | |
1041 | */ | |
45694aa7 | 1042 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
0ae617be | 1043 | machine); |
a1645ce1 ZY |
1044 | if (err < 0) |
1045 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 1046 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
1047 | } |
1048 | ||
98402807 FW |
1049 | static struct perf_event_header finished_round_event = { |
1050 | .size = sizeof(struct perf_event_header), | |
1051 | .type = PERF_RECORD_FINISHED_ROUND, | |
1052 | }; | |
1053 | ||
a5830532 | 1054 | static void record__adjust_affinity(struct record *rec, struct mmap *map) |
f13de660 AB |
1055 | { |
1056 | if (rec->opts.affinity != PERF_AFFINITY_SYS && | |
8384a260 AB |
1057 | !bitmap_equal(rec->affinity_mask.bits, map->affinity_mask.bits, |
1058 | rec->affinity_mask.nbits)) { | |
1059 | bitmap_zero(rec->affinity_mask.bits, rec->affinity_mask.nbits); | |
1060 | bitmap_or(rec->affinity_mask.bits, rec->affinity_mask.bits, | |
1061 | map->affinity_mask.bits, rec->affinity_mask.nbits); | |
1062 | sched_setaffinity(0, MMAP_CPU_MASK_BYTES(&rec->affinity_mask), | |
1063 | (cpu_set_t *)rec->affinity_mask.bits); | |
1064 | if (verbose == 2) | |
1065 | mmap_cpu_mask__scnprintf(&rec->affinity_mask, "thread"); | |
f13de660 AB |
1066 | } |
1067 | } | |
1068 | ||
5d7f4116 AB |
1069 | static size_t process_comp_header(void *record, size_t increment) |
1070 | { | |
72932371 | 1071 | struct perf_record_compressed *event = record; |
5d7f4116 AB |
1072 | size_t size = sizeof(*event); |
1073 | ||
1074 | if (increment) { | |
1075 | event->header.size += increment; | |
1076 | return increment; | |
1077 | } | |
1078 | ||
1079 | event->header.type = PERF_RECORD_COMPRESSED; | |
1080 | event->header.size = size; | |
1081 | ||
1082 | return size; | |
1083 | } | |
1084 | ||
1085 | static size_t zstd_compress(struct perf_session *session, void *dst, size_t dst_size, | |
1086 | void *src, size_t src_size) | |
1087 | { | |
1088 | size_t compressed; | |
72932371 | 1089 | size_t max_record_size = PERF_SAMPLE_MAX_SIZE - sizeof(struct perf_record_compressed) - 1; |
5d7f4116 AB |
1090 | |
1091 | compressed = zstd_compress_stream_to_records(&session->zstd_data, dst, dst_size, src, src_size, | |
1092 | max_record_size, process_comp_header); | |
1093 | ||
1094 | session->bytes_transferred += src_size; | |
1095 | session->bytes_compressed += compressed; | |
1096 | ||
1097 | return compressed; | |
1098 | } | |
1099 | ||
63503dba | 1100 | static int record__mmap_read_evlist(struct record *rec, struct evlist *evlist, |
470530bb | 1101 | bool overwrite, bool synch) |
98402807 | 1102 | { |
dcabb507 | 1103 | u64 bytes_written = rec->bytes_written; |
0e2e63dd | 1104 | int i; |
8d3eca20 | 1105 | int rc = 0; |
a5830532 | 1106 | struct mmap *maps; |
d3d1af6f | 1107 | int trace_fd = rec->data.file.fd; |
ef781128 | 1108 | off_t off = 0; |
98402807 | 1109 | |
cb21686b WN |
1110 | if (!evlist) |
1111 | return 0; | |
ef149c25 | 1112 | |
0b72d69a | 1113 | maps = overwrite ? evlist->overwrite_mmap : evlist->mmap; |
a4ea0ec4 WN |
1114 | if (!maps) |
1115 | return 0; | |
1116 | ||
0b72d69a | 1117 | if (overwrite && evlist->bkw_mmap_state != BKW_MMAP_DATA_PENDING) |
54cc54de WN |
1118 | return 0; |
1119 | ||
d3d1af6f AB |
1120 | if (record__aio_enabled(rec)) |
1121 | off = record__aio_get_pos(trace_fd); | |
1122 | ||
c976ee11 | 1123 | for (i = 0; i < evlist->core.nr_mmaps; i++) { |
470530bb | 1124 | u64 flush = 0; |
a5830532 | 1125 | struct mmap *map = &maps[i]; |
cb21686b | 1126 | |
547740f7 | 1127 | if (map->core.base) { |
f13de660 | 1128 | record__adjust_affinity(rec, map); |
470530bb | 1129 | if (synch) { |
65aa2e6b JO |
1130 | flush = map->core.flush; |
1131 | map->core.flush = 1; | |
470530bb | 1132 | } |
d3d1af6f | 1133 | if (!record__aio_enabled(rec)) { |
ef781128 | 1134 | if (perf_mmap__push(map, rec, record__pushfn) < 0) { |
470530bb | 1135 | if (synch) |
65aa2e6b | 1136 | map->core.flush = flush; |
d3d1af6f AB |
1137 | rc = -1; |
1138 | goto out; | |
1139 | } | |
1140 | } else { | |
ef781128 | 1141 | if (record__aio_push(rec, map, &off) < 0) { |
d3d1af6f | 1142 | record__aio_set_pos(trace_fd, off); |
470530bb | 1143 | if (synch) |
65aa2e6b | 1144 | map->core.flush = flush; |
d3d1af6f AB |
1145 | rc = -1; |
1146 | goto out; | |
1147 | } | |
8d3eca20 | 1148 | } |
470530bb | 1149 | if (synch) |
65aa2e6b | 1150 | map->core.flush = flush; |
8d3eca20 | 1151 | } |
ef149c25 | 1152 | |
e035f4ca | 1153 | if (map->auxtrace_mmap.base && !rec->opts.auxtrace_snapshot_mode && |
c0a6de06 | 1154 | !rec->opts.auxtrace_sample_mode && |
e035f4ca | 1155 | record__auxtrace_mmap_read(rec, map) != 0) { |
ef149c25 AH |
1156 | rc = -1; |
1157 | goto out; | |
1158 | } | |
98402807 FW |
1159 | } |
1160 | ||
d3d1af6f AB |
1161 | if (record__aio_enabled(rec)) |
1162 | record__aio_set_pos(trace_fd, off); | |
1163 | ||
dcabb507 JO |
1164 | /* |
1165 | * Mark the round finished in case we wrote | |
1166 | * at least one event. | |
1167 | */ | |
1168 | if (bytes_written != rec->bytes_written) | |
ded2b8fe | 1169 | rc = record__write(rec, NULL, &finished_round_event, sizeof(finished_round_event)); |
8d3eca20 | 1170 | |
0b72d69a | 1171 | if (overwrite) |
ade9d208 | 1172 | evlist__toggle_bkw_mmap(evlist, BKW_MMAP_EMPTY); |
8d3eca20 DA |
1173 | out: |
1174 | return rc; | |
98402807 FW |
1175 | } |
1176 | ||
470530bb | 1177 | static int record__mmap_read_all(struct record *rec, bool synch) |
cb21686b WN |
1178 | { |
1179 | int err; | |
1180 | ||
470530bb | 1181 | err = record__mmap_read_evlist(rec, rec->evlist, false, synch); |
cb21686b WN |
1182 | if (err) |
1183 | return err; | |
1184 | ||
470530bb | 1185 | return record__mmap_read_evlist(rec, rec->evlist, true, synch); |
cb21686b WN |
1186 | } |
1187 | ||
8c6f45a7 | 1188 | static void record__init_features(struct record *rec) |
57706abc | 1189 | { |
57706abc DA |
1190 | struct perf_session *session = rec->session; |
1191 | int feat; | |
1192 | ||
1193 | for (feat = HEADER_FIRST_FEATURE; feat < HEADER_LAST_FEATURE; feat++) | |
1194 | perf_header__set_feat(&session->header, feat); | |
1195 | ||
1196 | if (rec->no_buildid) | |
1197 | perf_header__clear_feat(&session->header, HEADER_BUILD_ID); | |
1198 | ||
ce9036a6 | 1199 | if (!have_tracepoints(&rec->evlist->core.entries)) |
57706abc DA |
1200 | perf_header__clear_feat(&session->header, HEADER_TRACING_DATA); |
1201 | ||
1202 | if (!rec->opts.branch_stack) | |
1203 | perf_header__clear_feat(&session->header, HEADER_BRANCH_STACK); | |
ef149c25 AH |
1204 | |
1205 | if (!rec->opts.full_auxtrace) | |
1206 | perf_header__clear_feat(&session->header, HEADER_AUXTRACE); | |
ffa517ad | 1207 | |
cf790516 AB |
1208 | if (!(rec->opts.use_clockid && rec->opts.clockid_res_ns)) |
1209 | perf_header__clear_feat(&session->header, HEADER_CLOCKID); | |
1210 | ||
d1e325cf JO |
1211 | if (!rec->opts.use_clockid) |
1212 | perf_header__clear_feat(&session->header, HEADER_CLOCK_DATA); | |
1213 | ||
258031c0 | 1214 | perf_header__clear_feat(&session->header, HEADER_DIR_FORMAT); |
42e1fd80 AB |
1215 | if (!record__comp_enabled(rec)) |
1216 | perf_header__clear_feat(&session->header, HEADER_COMPRESSED); | |
258031c0 | 1217 | |
ffa517ad | 1218 | perf_header__clear_feat(&session->header, HEADER_STAT); |
57706abc DA |
1219 | } |
1220 | ||
e1ab48ba WN |
1221 | static void |
1222 | record__finish_output(struct record *rec) | |
1223 | { | |
8ceb41d7 JO |
1224 | struct perf_data *data = &rec->data; |
1225 | int fd = perf_data__fd(data); | |
e1ab48ba | 1226 | |
8ceb41d7 | 1227 | if (data->is_pipe) |
e1ab48ba WN |
1228 | return; |
1229 | ||
1230 | rec->session->header.data_size += rec->bytes_written; | |
45112e89 | 1231 | data->file.size = lseek(perf_data__fd(data), 0, SEEK_CUR); |
e1ab48ba WN |
1232 | |
1233 | if (!rec->no_buildid) { | |
1234 | process_buildids(rec); | |
1235 | ||
1236 | if (rec->buildid_all) | |
1237 | dsos__hit_all(rec->session); | |
1238 | } | |
1239 | perf_session__write_header(rec->session, rec->evlist, fd, true); | |
1240 | ||
1241 | return; | |
1242 | } | |
1243 | ||
4ea648ae | 1244 | static int record__synthesize_workload(struct record *rec, bool tail) |
be7b0c9e | 1245 | { |
9d6aae72 | 1246 | int err; |
9749b90e | 1247 | struct perf_thread_map *thread_map; |
be7b0c9e | 1248 | |
4ea648ae WN |
1249 | if (rec->opts.tail_synthesize != tail) |
1250 | return 0; | |
1251 | ||
9d6aae72 ACM |
1252 | thread_map = thread_map__new_by_tid(rec->evlist->workload.pid); |
1253 | if (thread_map == NULL) | |
1254 | return -1; | |
1255 | ||
1256 | err = perf_event__synthesize_thread_map(&rec->tool, thread_map, | |
be7b0c9e WN |
1257 | process_synthesized_event, |
1258 | &rec->session->machines.host, | |
3fcb10e4 | 1259 | rec->opts.sample_address); |
7836e52e | 1260 | perf_thread_map__put(thread_map); |
9d6aae72 | 1261 | return err; |
be7b0c9e WN |
1262 | } |
1263 | ||
4ea648ae | 1264 | static int record__synthesize(struct record *rec, bool tail); |
3c1cb7e3 | 1265 | |
ecfd7a9c WN |
1266 | static int |
1267 | record__switch_output(struct record *rec, bool at_exit) | |
1268 | { | |
8ceb41d7 | 1269 | struct perf_data *data = &rec->data; |
ecfd7a9c | 1270 | int fd, err; |
03724b2e | 1271 | char *new_filename; |
ecfd7a9c WN |
1272 | |
1273 | /* Same Size: "2015122520103046"*/ | |
1274 | char timestamp[] = "InvalidTimestamp"; | |
1275 | ||
d3d1af6f AB |
1276 | record__aio_mmap_read_sync(rec); |
1277 | ||
4ea648ae WN |
1278 | record__synthesize(rec, true); |
1279 | if (target__none(&rec->opts.target)) | |
1280 | record__synthesize_workload(rec, true); | |
1281 | ||
ecfd7a9c WN |
1282 | rec->samples = 0; |
1283 | record__finish_output(rec); | |
1284 | err = fetch_current_timestamp(timestamp, sizeof(timestamp)); | |
1285 | if (err) { | |
1286 | pr_err("Failed to get current timestamp\n"); | |
1287 | return -EINVAL; | |
1288 | } | |
1289 | ||
8ceb41d7 | 1290 | fd = perf_data__switch(data, timestamp, |
ecfd7a9c | 1291 | rec->session->header.data_offset, |
03724b2e | 1292 | at_exit, &new_filename); |
ecfd7a9c WN |
1293 | if (fd >= 0 && !at_exit) { |
1294 | rec->bytes_written = 0; | |
1295 | rec->session->header.data_size = 0; | |
1296 | } | |
1297 | ||
1298 | if (!quiet) | |
1299 | fprintf(stderr, "[ perf record: Dump %s.%s ]\n", | |
2d4f2799 | 1300 | data->path, timestamp); |
3c1cb7e3 | 1301 | |
03724b2e AK |
1302 | if (rec->switch_output.num_files) { |
1303 | int n = rec->switch_output.cur_file + 1; | |
1304 | ||
1305 | if (n >= rec->switch_output.num_files) | |
1306 | n = 0; | |
1307 | rec->switch_output.cur_file = n; | |
1308 | if (rec->switch_output.filenames[n]) { | |
1309 | remove(rec->switch_output.filenames[n]); | |
d8f9da24 | 1310 | zfree(&rec->switch_output.filenames[n]); |
03724b2e AK |
1311 | } |
1312 | rec->switch_output.filenames[n] = new_filename; | |
1313 | } else { | |
1314 | free(new_filename); | |
1315 | } | |
1316 | ||
3c1cb7e3 | 1317 | /* Output tracking events */ |
be7b0c9e | 1318 | if (!at_exit) { |
4ea648ae | 1319 | record__synthesize(rec, false); |
3c1cb7e3 | 1320 | |
be7b0c9e WN |
1321 | /* |
1322 | * In 'perf record --switch-output' without -a, | |
1323 | * record__synthesize() in record__switch_output() won't | |
1324 | * generate tracking events because there's no thread_map | |
1325 | * in evlist. Which causes newly created perf.data doesn't | |
1326 | * contain map and comm information. | |
1327 | * Create a fake thread_map and directly call | |
1328 | * perf_event__synthesize_thread_map() for those events. | |
1329 | */ | |
1330 | if (target__none(&rec->opts.target)) | |
4ea648ae | 1331 | record__synthesize_workload(rec, false); |
be7b0c9e | 1332 | } |
ecfd7a9c WN |
1333 | return fd; |
1334 | } | |
1335 | ||
f33cbe72 ACM |
1336 | static volatile int workload_exec_errno; |
1337 | ||
1338 | /* | |
7b392ef0 | 1339 | * evlist__prepare_workload will send a SIGUSR1 |
f33cbe72 ACM |
1340 | * if the fork fails, since we asked by setting its |
1341 | * want_signal to true. | |
1342 | */ | |
45604710 NK |
1343 | static void workload_exec_failed_signal(int signo __maybe_unused, |
1344 | siginfo_t *info, | |
f33cbe72 ACM |
1345 | void *ucontext __maybe_unused) |
1346 | { | |
1347 | workload_exec_errno = info->si_value.sival_int; | |
1348 | done = 1; | |
f33cbe72 ACM |
1349 | child_finished = 1; |
1350 | } | |
1351 | ||
2dd6d8a1 | 1352 | static void snapshot_sig_handler(int sig); |
bfacbe3b | 1353 | static void alarm_sig_handler(int sig); |
2dd6d8a1 | 1354 | |
db0ea13c | 1355 | static const struct perf_event_mmap_page *evlist__pick_pc(struct evlist *evlist) |
ee667f94 | 1356 | { |
b2cb615d | 1357 | if (evlist) { |
547740f7 JO |
1358 | if (evlist->mmap && evlist->mmap[0].core.base) |
1359 | return evlist->mmap[0].core.base; | |
1360 | if (evlist->overwrite_mmap && evlist->overwrite_mmap[0].core.base) | |
1361 | return evlist->overwrite_mmap[0].core.base; | |
b2cb615d | 1362 | } |
ee667f94 WN |
1363 | return NULL; |
1364 | } | |
1365 | ||
c45628b0 WN |
1366 | static const struct perf_event_mmap_page *record__pick_pc(struct record *rec) |
1367 | { | |
db0ea13c | 1368 | const struct perf_event_mmap_page *pc = evlist__pick_pc(rec->evlist); |
ee667f94 WN |
1369 | if (pc) |
1370 | return pc; | |
c45628b0 WN |
1371 | return NULL; |
1372 | } | |
1373 | ||
4ea648ae | 1374 | static int record__synthesize(struct record *rec, bool tail) |
c45c86eb WN |
1375 | { |
1376 | struct perf_session *session = rec->session; | |
1377 | struct machine *machine = &session->machines.host; | |
8ceb41d7 | 1378 | struct perf_data *data = &rec->data; |
c45c86eb WN |
1379 | struct record_opts *opts = &rec->opts; |
1380 | struct perf_tool *tool = &rec->tool; | |
8ceb41d7 | 1381 | int fd = perf_data__fd(data); |
c45c86eb | 1382 | int err = 0; |
d99c22ea | 1383 | event_op f = process_synthesized_event; |
c45c86eb | 1384 | |
4ea648ae WN |
1385 | if (rec->opts.tail_synthesize != tail) |
1386 | return 0; | |
1387 | ||
8ceb41d7 | 1388 | if (data->is_pipe) { |
a2015516 JO |
1389 | /* |
1390 | * We need to synthesize events first, because some | |
1391 | * features works on top of them (on report side). | |
1392 | */ | |
318ec184 | 1393 | err = perf_event__synthesize_attrs(tool, rec->evlist, |
c45c86eb WN |
1394 | process_synthesized_event); |
1395 | if (err < 0) { | |
1396 | pr_err("Couldn't synthesize attrs.\n"); | |
1397 | goto out; | |
1398 | } | |
1399 | ||
a2015516 JO |
1400 | err = perf_event__synthesize_features(tool, session, rec->evlist, |
1401 | process_synthesized_event); | |
1402 | if (err < 0) { | |
1403 | pr_err("Couldn't synthesize features.\n"); | |
1404 | return err; | |
1405 | } | |
1406 | ||
ce9036a6 | 1407 | if (have_tracepoints(&rec->evlist->core.entries)) { |
c45c86eb WN |
1408 | /* |
1409 | * FIXME err <= 0 here actually means that | |
1410 | * there were no tracepoints so its not really | |
1411 | * an error, just that we don't need to | |
1412 | * synthesize anything. We really have to | |
1413 | * return this more properly and also | |
1414 | * propagate errors that now are calling die() | |
1415 | */ | |
1416 | err = perf_event__synthesize_tracing_data(tool, fd, rec->evlist, | |
1417 | process_synthesized_event); | |
1418 | if (err <= 0) { | |
1419 | pr_err("Couldn't record tracing data.\n"); | |
1420 | goto out; | |
1421 | } | |
1422 | rec->bytes_written += err; | |
1423 | } | |
1424 | } | |
1425 | ||
c45628b0 | 1426 | err = perf_event__synth_time_conv(record__pick_pc(rec), tool, |
46bc29b9 AH |
1427 | process_synthesized_event, machine); |
1428 | if (err) | |
1429 | goto out; | |
1430 | ||
c0a6de06 AH |
1431 | /* Synthesize id_index before auxtrace_info */ |
1432 | if (rec->opts.auxtrace_sample_mode) { | |
1433 | err = perf_event__synthesize_id_index(tool, | |
1434 | process_synthesized_event, | |
1435 | session->evlist, machine); | |
1436 | if (err) | |
1437 | goto out; | |
1438 | } | |
1439 | ||
c45c86eb WN |
1440 | if (rec->opts.full_auxtrace) { |
1441 | err = perf_event__synthesize_auxtrace_info(rec->itr, tool, | |
1442 | session, process_synthesized_event); | |
1443 | if (err) | |
1444 | goto out; | |
1445 | } | |
1446 | ||
78e1bc25 | 1447 | if (!evlist__exclude_kernel(rec->evlist)) { |
6c443954 ACM |
1448 | err = perf_event__synthesize_kernel_mmap(tool, process_synthesized_event, |
1449 | machine); | |
1450 | WARN_ONCE(err < 0, "Couldn't record kernel reference relocation symbol\n" | |
1451 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
1452 | "Check /proc/kallsyms permission or run as root.\n"); | |
1453 | ||
1454 | err = perf_event__synthesize_modules(tool, process_synthesized_event, | |
1455 | machine); | |
1456 | WARN_ONCE(err < 0, "Couldn't record kernel module information.\n" | |
1457 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
1458 | "Check /proc/modules permission or run as root.\n"); | |
1459 | } | |
c45c86eb WN |
1460 | |
1461 | if (perf_guest) { | |
1462 | machines__process_guests(&session->machines, | |
1463 | perf_event__synthesize_guest_os, tool); | |
1464 | } | |
1465 | ||
bfd8f72c AK |
1466 | err = perf_event__synthesize_extra_attr(&rec->tool, |
1467 | rec->evlist, | |
1468 | process_synthesized_event, | |
1469 | data->is_pipe); | |
1470 | if (err) | |
1471 | goto out; | |
1472 | ||
03617c22 | 1473 | err = perf_event__synthesize_thread_map2(&rec->tool, rec->evlist->core.threads, |
373565d2 AK |
1474 | process_synthesized_event, |
1475 | NULL); | |
1476 | if (err < 0) { | |
1477 | pr_err("Couldn't synthesize thread map.\n"); | |
1478 | return err; | |
1479 | } | |
1480 | ||
f72f901d | 1481 | err = perf_event__synthesize_cpu_map(&rec->tool, rec->evlist->core.cpus, |
373565d2 AK |
1482 | process_synthesized_event, NULL); |
1483 | if (err < 0) { | |
1484 | pr_err("Couldn't synthesize cpu map.\n"); | |
1485 | return err; | |
1486 | } | |
1487 | ||
e5416950 | 1488 | err = perf_event__synthesize_bpf_events(session, process_synthesized_event, |
7b612e29 SL |
1489 | machine, opts); |
1490 | if (err < 0) | |
1491 | pr_warning("Couldn't synthesize bpf events.\n"); | |
1492 | ||
ab64069f NK |
1493 | err = perf_event__synthesize_cgroups(tool, process_synthesized_event, |
1494 | machine); | |
1495 | if (err < 0) | |
1496 | pr_warning("Couldn't synthesize cgroup events.\n"); | |
1497 | ||
d99c22ea SE |
1498 | if (rec->opts.nr_threads_synthesize > 1) { |
1499 | perf_set_multithreaded(); | |
1500 | f = process_locked_synthesized_event; | |
1501 | } | |
1502 | ||
03617c22 | 1503 | err = __machine__synthesize_threads(machine, tool, &opts->target, rec->evlist->core.threads, |
d99c22ea SE |
1504 | f, opts->sample_address, |
1505 | rec->opts.nr_threads_synthesize); | |
1506 | ||
1507 | if (rec->opts.nr_threads_synthesize > 1) | |
1508 | perf_set_singlethreaded(); | |
1509 | ||
c45c86eb WN |
1510 | out: |
1511 | return err; | |
1512 | } | |
1513 | ||
899e5ffb ACM |
1514 | static int record__process_signal_event(union perf_event *event __maybe_unused, void *data) |
1515 | { | |
1516 | struct record *rec = data; | |
1517 | pthread_kill(rec->thread_id, SIGUSR2); | |
1518 | return 0; | |
1519 | } | |
1520 | ||
23cbb41c ACM |
1521 | static int record__setup_sb_evlist(struct record *rec) |
1522 | { | |
1523 | struct record_opts *opts = &rec->opts; | |
1524 | ||
1525 | if (rec->sb_evlist != NULL) { | |
1526 | /* | |
1527 | * We get here if --switch-output-event populated the | |
1528 | * sb_evlist, so associate a callback that will send a SIGUSR2 | |
1529 | * to the main thread. | |
1530 | */ | |
1531 | evlist__set_cb(rec->sb_evlist, record__process_signal_event, rec); | |
1532 | rec->thread_id = pthread_self(); | |
1533 | } | |
1101c872 | 1534 | #ifdef HAVE_LIBBPF_SUPPORT |
23cbb41c ACM |
1535 | if (!opts->no_bpf_event) { |
1536 | if (rec->sb_evlist == NULL) { | |
1537 | rec->sb_evlist = evlist__new(); | |
1538 | ||
1539 | if (rec->sb_evlist == NULL) { | |
1540 | pr_err("Couldn't create side band evlist.\n."); | |
1541 | return -1; | |
1542 | } | |
1543 | } | |
1544 | ||
1545 | if (evlist__add_bpf_sb_event(rec->sb_evlist, &rec->session->header.env)) { | |
1546 | pr_err("Couldn't ask for PERF_RECORD_BPF_EVENT side band events.\n."); | |
1547 | return -1; | |
1548 | } | |
1549 | } | |
1101c872 | 1550 | #endif |
08c83997 | 1551 | if (evlist__start_sb_thread(rec->sb_evlist, &rec->opts.target)) { |
23cbb41c ACM |
1552 | pr_debug("Couldn't start the BPF side band thread:\nBPF programs starting from now on won't be annotatable\n"); |
1553 | opts->no_bpf_event = true; | |
1554 | } | |
1555 | ||
1556 | return 0; | |
1557 | } | |
1558 | ||
d1e325cf JO |
1559 | static int record__init_clock(struct record *rec) |
1560 | { | |
1561 | struct perf_session *session = rec->session; | |
1562 | struct timespec ref_clockid; | |
1563 | struct timeval ref_tod; | |
1564 | u64 ref; | |
1565 | ||
1566 | if (!rec->opts.use_clockid) | |
1567 | return 0; | |
1568 | ||
9d88a1a1 JO |
1569 | if (rec->opts.use_clockid && rec->opts.clockid_res_ns) |
1570 | session->header.env.clock.clockid_res_ns = rec->opts.clockid_res_ns; | |
1571 | ||
d1e325cf JO |
1572 | session->header.env.clock.clockid = rec->opts.clockid; |
1573 | ||
1574 | if (gettimeofday(&ref_tod, NULL) != 0) { | |
1575 | pr_err("gettimeofday failed, cannot set reference time.\n"); | |
1576 | return -1; | |
1577 | } | |
1578 | ||
1579 | if (clock_gettime(rec->opts.clockid, &ref_clockid)) { | |
1580 | pr_err("clock_gettime failed, cannot set reference time.\n"); | |
1581 | return -1; | |
1582 | } | |
1583 | ||
1584 | ref = (u64) ref_tod.tv_sec * NSEC_PER_SEC + | |
1585 | (u64) ref_tod.tv_usec * NSEC_PER_USEC; | |
1586 | ||
1587 | session->header.env.clock.tod_ns = ref; | |
1588 | ||
1589 | ref = (u64) ref_clockid.tv_sec * NSEC_PER_SEC + | |
1590 | (u64) ref_clockid.tv_nsec; | |
1591 | ||
1592 | session->header.env.clock.clockid_ns = ref; | |
1593 | return 0; | |
1594 | } | |
1595 | ||
d20aff15 AH |
1596 | static void hit_auxtrace_snapshot_trigger(struct record *rec) |
1597 | { | |
1598 | if (trigger_is_ready(&auxtrace_snapshot_trigger)) { | |
1599 | trigger_hit(&auxtrace_snapshot_trigger); | |
1600 | auxtrace_record__snapshot_started = 1; | |
1601 | if (auxtrace_record__snapshot_start(rec->itr)) | |
1602 | trigger_error(&auxtrace_snapshot_trigger); | |
1603 | } | |
1604 | } | |
1605 | ||
8c6f45a7 | 1606 | static int __cmd_record(struct record *rec, int argc, const char **argv) |
16c8a109 | 1607 | { |
57706abc | 1608 | int err; |
45604710 | 1609 | int status = 0; |
8b412664 | 1610 | unsigned long waking = 0; |
46be604b | 1611 | const bool forks = argc > 0; |
45694aa7 | 1612 | struct perf_tool *tool = &rec->tool; |
b4006796 | 1613 | struct record_opts *opts = &rec->opts; |
8ceb41d7 | 1614 | struct perf_data *data = &rec->data; |
d20deb64 | 1615 | struct perf_session *session; |
6dcf45ef | 1616 | bool disabled = false, draining = false; |
42aa276f | 1617 | int fd; |
d3c8c08e | 1618 | float ratio = 0; |
acce0223 | 1619 | enum evlist_ctl_cmd cmd = EVLIST_CTL_CMD_UNSUPPORTED; |
de9ac07b | 1620 | |
45604710 | 1621 | atexit(record__sig_exit); |
f5970550 PZ |
1622 | signal(SIGCHLD, sig_handler); |
1623 | signal(SIGINT, sig_handler); | |
804f7ac7 | 1624 | signal(SIGTERM, sig_handler); |
a074865e | 1625 | signal(SIGSEGV, sigsegv_handler); |
c0bdc1c4 | 1626 | |
f3b3614a HB |
1627 | if (rec->opts.record_namespaces) |
1628 | tool->namespace_events = true; | |
1629 | ||
8fb4b679 NK |
1630 | if (rec->opts.record_cgroup) { |
1631 | #ifdef HAVE_FILE_HANDLE | |
1632 | tool->cgroup_events = true; | |
1633 | #else | |
1634 | pr_err("cgroup tracking is not supported\n"); | |
1635 | return -1; | |
1636 | #endif | |
1637 | } | |
1638 | ||
dc0c6127 | 1639 | if (rec->opts.auxtrace_snapshot_mode || rec->switch_output.enabled) { |
2dd6d8a1 | 1640 | signal(SIGUSR2, snapshot_sig_handler); |
3c1cb7e3 WN |
1641 | if (rec->opts.auxtrace_snapshot_mode) |
1642 | trigger_on(&auxtrace_snapshot_trigger); | |
dc0c6127 | 1643 | if (rec->switch_output.enabled) |
3c1cb7e3 | 1644 | trigger_on(&switch_output_trigger); |
c0bdc1c4 | 1645 | } else { |
2dd6d8a1 | 1646 | signal(SIGUSR2, SIG_IGN); |
c0bdc1c4 | 1647 | } |
f5970550 | 1648 | |
8ceb41d7 | 1649 | session = perf_session__new(data, false, tool); |
6ef81c55 | 1650 | if (IS_ERR(session)) { |
ffa91880 | 1651 | pr_err("Perf session creation failed.\n"); |
6ef81c55 | 1652 | return PTR_ERR(session); |
a9a70bbc ACM |
1653 | } |
1654 | ||
8ceb41d7 | 1655 | fd = perf_data__fd(data); |
d20deb64 ACM |
1656 | rec->session = session; |
1657 | ||
5d7f4116 AB |
1658 | if (zstd_init(&session->zstd_data, rec->opts.comp_level) < 0) { |
1659 | pr_err("Compression initialization failed.\n"); | |
1660 | return -1; | |
1661 | } | |
da231338 AM |
1662 | #ifdef HAVE_EVENTFD_SUPPORT |
1663 | done_fd = eventfd(0, EFD_NONBLOCK); | |
1664 | if (done_fd < 0) { | |
1665 | pr_err("Failed to create wakeup eventfd, error: %m\n"); | |
1666 | status = -1; | |
1667 | goto out_delete_session; | |
1668 | } | |
e16c2ce7 | 1669 | err = evlist__add_wakeup_eventfd(rec->evlist, done_fd); |
da231338 AM |
1670 | if (err < 0) { |
1671 | pr_err("Failed to add wakeup eventfd to poll list\n"); | |
1672 | status = err; | |
1673 | goto out_delete_session; | |
1674 | } | |
1675 | #endif // HAVE_EVENTFD_SUPPORT | |
5d7f4116 AB |
1676 | |
1677 | session->header.env.comp_type = PERF_COMP_ZSTD; | |
1678 | session->header.env.comp_level = rec->opts.comp_level; | |
1679 | ||
eeb399b5 AH |
1680 | if (rec->opts.kcore && |
1681 | !record__kcore_readable(&session->machines.host)) { | |
1682 | pr_err("ERROR: kcore is not readable.\n"); | |
1683 | return -1; | |
1684 | } | |
1685 | ||
d1e325cf JO |
1686 | if (record__init_clock(rec)) |
1687 | return -1; | |
1688 | ||
8c6f45a7 | 1689 | record__init_features(rec); |
330aa675 | 1690 | |
d4db3f16 | 1691 | if (forks) { |
7b392ef0 ACM |
1692 | err = evlist__prepare_workload(rec->evlist, &opts->target, argv, data->is_pipe, |
1693 | workload_exec_failed_signal); | |
35b9d88e ACM |
1694 | if (err < 0) { |
1695 | pr_err("Couldn't run the workload!\n"); | |
45604710 | 1696 | status = err; |
35b9d88e | 1697 | goto out_delete_session; |
856e9660 | 1698 | } |
856e9660 PZ |
1699 | } |
1700 | ||
ad46e48c JO |
1701 | /* |
1702 | * If we have just single event and are sending data | |
1703 | * through pipe, we need to force the ids allocation, | |
1704 | * because we synthesize event name through the pipe | |
1705 | * and need the id for that. | |
1706 | */ | |
6484d2f9 | 1707 | if (data->is_pipe && rec->evlist->core.nr_entries == 1) |
ad46e48c JO |
1708 | rec->opts.sample_id = true; |
1709 | ||
8c6f45a7 | 1710 | if (record__open(rec) != 0) { |
8d3eca20 | 1711 | err = -1; |
45604710 | 1712 | goto out_child; |
8d3eca20 | 1713 | } |
f6fa4375 | 1714 | session->header.env.comp_mmap_len = session->evlist->core.mmap_len; |
de9ac07b | 1715 | |
eeb399b5 AH |
1716 | if (rec->opts.kcore) { |
1717 | err = record__kcore_copy(&session->machines.host, data); | |
1718 | if (err) { | |
1719 | pr_err("ERROR: Failed to copy kcore\n"); | |
1720 | goto out_child; | |
1721 | } | |
1722 | } | |
1723 | ||
8690a2a7 WN |
1724 | err = bpf__apply_obj_config(); |
1725 | if (err) { | |
1726 | char errbuf[BUFSIZ]; | |
1727 | ||
1728 | bpf__strerror_apply_obj_config(err, errbuf, sizeof(errbuf)); | |
1729 | pr_err("ERROR: Apply config to BPF failed: %s\n", | |
1730 | errbuf); | |
1731 | goto out_child; | |
1732 | } | |
1733 | ||
cca8482c AH |
1734 | /* |
1735 | * Normally perf_session__new would do this, but it doesn't have the | |
1736 | * evlist. | |
1737 | */ | |
8cedf3a5 | 1738 | if (rec->tool.ordered_events && !evlist__sample_id_all(rec->evlist)) { |
cca8482c AH |
1739 | pr_warning("WARNING: No sample_id_all support, falling back to unordered processing\n"); |
1740 | rec->tool.ordered_events = false; | |
1741 | } | |
1742 | ||
3e2be2da | 1743 | if (!rec->evlist->nr_groups) |
a8bb559b NK |
1744 | perf_header__clear_feat(&session->header, HEADER_GROUP_DESC); |
1745 | ||
8ceb41d7 | 1746 | if (data->is_pipe) { |
42aa276f | 1747 | err = perf_header__write_pipe(fd); |
529870e3 | 1748 | if (err < 0) |
45604710 | 1749 | goto out_child; |
563aecb2 | 1750 | } else { |
42aa276f | 1751 | err = perf_session__write_header(session, rec->evlist, fd, false); |
d5eed904 | 1752 | if (err < 0) |
45604710 | 1753 | goto out_child; |
56b03f3c ACM |
1754 | } |
1755 | ||
b38d85ef | 1756 | err = -1; |
d3665498 | 1757 | if (!rec->no_buildid |
e20960c0 | 1758 | && !perf_header__has_feat(&session->header, HEADER_BUILD_ID)) { |
d3665498 | 1759 | pr_err("Couldn't generate buildids. " |
e20960c0 | 1760 | "Use --no-buildid to profile anyway.\n"); |
45604710 | 1761 | goto out_child; |
e20960c0 RR |
1762 | } |
1763 | ||
23cbb41c ACM |
1764 | err = record__setup_sb_evlist(rec); |
1765 | if (err) | |
1766 | goto out_child; | |
657ee553 | 1767 | |
4ea648ae | 1768 | err = record__synthesize(rec, false); |
c45c86eb | 1769 | if (err < 0) |
45604710 | 1770 | goto out_child; |
8d3eca20 | 1771 | |
d20deb64 | 1772 | if (rec->realtime_prio) { |
de9ac07b PZ |
1773 | struct sched_param param; |
1774 | ||
d20deb64 | 1775 | param.sched_priority = rec->realtime_prio; |
de9ac07b | 1776 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { |
6beba7ad | 1777 | pr_err("Could not set realtime priority.\n"); |
8d3eca20 | 1778 | err = -1; |
45604710 | 1779 | goto out_child; |
de9ac07b PZ |
1780 | } |
1781 | } | |
1782 | ||
774cb499 JO |
1783 | /* |
1784 | * When perf is starting the traced process, all the events | |
1785 | * (apart from group members) have enable_on_exec=1 set, | |
1786 | * so don't spoil it by prematurely enabling them. | |
1787 | */ | |
6619a53e | 1788 | if (!target__none(&opts->target) && !opts->initial_delay) |
1c87f165 | 1789 | evlist__enable(rec->evlist); |
764e16a3 | 1790 | |
856e9660 PZ |
1791 | /* |
1792 | * Let the child rip | |
1793 | */ | |
e803cf97 | 1794 | if (forks) { |
20a8a3cf | 1795 | struct machine *machine = &session->machines.host; |
e5bed564 | 1796 | union perf_event *event; |
e907caf3 | 1797 | pid_t tgid; |
e5bed564 NK |
1798 | |
1799 | event = malloc(sizeof(event->comm) + machine->id_hdr_size); | |
1800 | if (event == NULL) { | |
1801 | err = -ENOMEM; | |
1802 | goto out_child; | |
1803 | } | |
1804 | ||
e803cf97 NK |
1805 | /* |
1806 | * Some H/W events are generated before COMM event | |
1807 | * which is emitted during exec(), so perf script | |
1808 | * cannot see a correct process name for those events. | |
1809 | * Synthesize COMM event to prevent it. | |
1810 | */ | |
e907caf3 HB |
1811 | tgid = perf_event__synthesize_comm(tool, event, |
1812 | rec->evlist->workload.pid, | |
1813 | process_synthesized_event, | |
1814 | machine); | |
1815 | free(event); | |
1816 | ||
1817 | if (tgid == -1) | |
1818 | goto out_child; | |
1819 | ||
1820 | event = malloc(sizeof(event->namespaces) + | |
1821 | (NR_NAMESPACES * sizeof(struct perf_ns_link_info)) + | |
1822 | machine->id_hdr_size); | |
1823 | if (event == NULL) { | |
1824 | err = -ENOMEM; | |
1825 | goto out_child; | |
1826 | } | |
1827 | ||
1828 | /* | |
1829 | * Synthesize NAMESPACES event for the command specified. | |
1830 | */ | |
1831 | perf_event__synthesize_namespaces(tool, event, | |
1832 | rec->evlist->workload.pid, | |
1833 | tgid, process_synthesized_event, | |
1834 | machine); | |
e5bed564 | 1835 | free(event); |
e803cf97 | 1836 | |
7b392ef0 | 1837 | evlist__start_workload(rec->evlist); |
e803cf97 | 1838 | } |
856e9660 | 1839 | |
1d078ccb AB |
1840 | if (evlist__initialize_ctlfd(rec->evlist, opts->ctl_fd, opts->ctl_fd_ack)) |
1841 | goto out_child; | |
1842 | ||
6619a53e | 1843 | if (opts->initial_delay) { |
68cd3b45 AB |
1844 | pr_info(EVLIST_DISABLED_MSG); |
1845 | if (opts->initial_delay > 0) { | |
1846 | usleep(opts->initial_delay * USEC_PER_MSEC); | |
1847 | evlist__enable(rec->evlist); | |
1848 | pr_info(EVLIST_ENABLED_MSG); | |
1849 | } | |
6619a53e AK |
1850 | } |
1851 | ||
5f9cf599 | 1852 | trigger_ready(&auxtrace_snapshot_trigger); |
3c1cb7e3 | 1853 | trigger_ready(&switch_output_trigger); |
a074865e | 1854 | perf_hooks__invoke_record_start(); |
649c48a9 | 1855 | for (;;) { |
9f065194 | 1856 | unsigned long long hits = rec->samples; |
de9ac07b | 1857 | |
05737464 WN |
1858 | /* |
1859 | * rec->evlist->bkw_mmap_state is possible to be | |
1860 | * BKW_MMAP_EMPTY here: when done == true and | |
1861 | * hits != rec->samples in previous round. | |
1862 | * | |
ade9d208 | 1863 | * evlist__toggle_bkw_mmap ensure we never |
05737464 WN |
1864 | * convert BKW_MMAP_EMPTY to BKW_MMAP_DATA_PENDING. |
1865 | */ | |
1866 | if (trigger_is_hit(&switch_output_trigger) || done || draining) | |
ade9d208 | 1867 | evlist__toggle_bkw_mmap(rec->evlist, BKW_MMAP_DATA_PENDING); |
05737464 | 1868 | |
470530bb | 1869 | if (record__mmap_read_all(rec, false) < 0) { |
5f9cf599 | 1870 | trigger_error(&auxtrace_snapshot_trigger); |
3c1cb7e3 | 1871 | trigger_error(&switch_output_trigger); |
8d3eca20 | 1872 | err = -1; |
45604710 | 1873 | goto out_child; |
8d3eca20 | 1874 | } |
de9ac07b | 1875 | |
2dd6d8a1 AH |
1876 | if (auxtrace_record__snapshot_started) { |
1877 | auxtrace_record__snapshot_started = 0; | |
5f9cf599 | 1878 | if (!trigger_is_error(&auxtrace_snapshot_trigger)) |
ce7b0e42 | 1879 | record__read_auxtrace_snapshot(rec, false); |
5f9cf599 | 1880 | if (trigger_is_error(&auxtrace_snapshot_trigger)) { |
2dd6d8a1 AH |
1881 | pr_err("AUX area tracing snapshot failed\n"); |
1882 | err = -1; | |
1883 | goto out_child; | |
1884 | } | |
1885 | } | |
1886 | ||
3c1cb7e3 | 1887 | if (trigger_is_hit(&switch_output_trigger)) { |
05737464 WN |
1888 | /* |
1889 | * If switch_output_trigger is hit, the data in | |
1890 | * overwritable ring buffer should have been collected, | |
1891 | * so bkw_mmap_state should be set to BKW_MMAP_EMPTY. | |
1892 | * | |
1893 | * If SIGUSR2 raise after or during record__mmap_read_all(), | |
1894 | * record__mmap_read_all() didn't collect data from | |
1895 | * overwritable ring buffer. Read again. | |
1896 | */ | |
1897 | if (rec->evlist->bkw_mmap_state == BKW_MMAP_RUNNING) | |
1898 | continue; | |
3c1cb7e3 WN |
1899 | trigger_ready(&switch_output_trigger); |
1900 | ||
05737464 WN |
1901 | /* |
1902 | * Reenable events in overwrite ring buffer after | |
1903 | * record__mmap_read_all(): we should have collected | |
1904 | * data from it. | |
1905 | */ | |
ade9d208 | 1906 | evlist__toggle_bkw_mmap(rec->evlist, BKW_MMAP_RUNNING); |
05737464 | 1907 | |
3c1cb7e3 WN |
1908 | if (!quiet) |
1909 | fprintf(stderr, "[ perf record: dump data: Woken up %ld times ]\n", | |
1910 | waking); | |
1911 | waking = 0; | |
1912 | fd = record__switch_output(rec, false); | |
1913 | if (fd < 0) { | |
1914 | pr_err("Failed to switch to new file\n"); | |
1915 | trigger_error(&switch_output_trigger); | |
1916 | err = fd; | |
1917 | goto out_child; | |
1918 | } | |
bfacbe3b JO |
1919 | |
1920 | /* re-arm the alarm */ | |
1921 | if (rec->switch_output.time) | |
1922 | alarm(rec->switch_output.time); | |
3c1cb7e3 WN |
1923 | } |
1924 | ||
d20deb64 | 1925 | if (hits == rec->samples) { |
6dcf45ef | 1926 | if (done || draining) |
649c48a9 | 1927 | break; |
80ab2987 | 1928 | err = evlist__poll(rec->evlist, -1); |
a515114f JO |
1929 | /* |
1930 | * Propagate error, only if there's any. Ignore positive | |
1931 | * number of returned events and interrupt error. | |
1932 | */ | |
1933 | if (err > 0 || (err < 0 && errno == EINTR)) | |
45604710 | 1934 | err = 0; |
8b412664 | 1935 | waking++; |
6dcf45ef | 1936 | |
f4009e7b | 1937 | if (evlist__filter_pollfd(rec->evlist, POLLERR | POLLHUP) == 0) |
6dcf45ef | 1938 | draining = true; |
8b412664 PZ |
1939 | } |
1940 | ||
acce0223 AB |
1941 | if (evlist__ctlfd_process(rec->evlist, &cmd) > 0) { |
1942 | switch (cmd) { | |
d20aff15 AH |
1943 | case EVLIST_CTL_CMD_SNAPSHOT: |
1944 | hit_auxtrace_snapshot_trigger(rec); | |
1945 | evlist__ctlfd_ack(rec->evlist); | |
1946 | break; | |
f186cd61 JO |
1947 | case EVLIST_CTL_CMD_STOP: |
1948 | done = 1; | |
1949 | break; | |
acce0223 AB |
1950 | case EVLIST_CTL_CMD_ACK: |
1951 | case EVLIST_CTL_CMD_UNSUPPORTED: | |
991ae4eb JO |
1952 | case EVLIST_CTL_CMD_ENABLE: |
1953 | case EVLIST_CTL_CMD_DISABLE: | |
142544a9 | 1954 | case EVLIST_CTL_CMD_EVLIST: |
47fddcb4 | 1955 | case EVLIST_CTL_CMD_PING: |
acce0223 AB |
1956 | default: |
1957 | break; | |
1958 | } | |
1959 | } | |
1960 | ||
774cb499 JO |
1961 | /* |
1962 | * When perf is starting the traced process, at the end events | |
1963 | * die with the process and we wait for that. Thus no need to | |
1964 | * disable events in this case. | |
1965 | */ | |
602ad878 | 1966 | if (done && !disabled && !target__none(&opts->target)) { |
5f9cf599 | 1967 | trigger_off(&auxtrace_snapshot_trigger); |
e74676de | 1968 | evlist__disable(rec->evlist); |
2711926a JO |
1969 | disabled = true; |
1970 | } | |
de9ac07b | 1971 | } |
ce7b0e42 | 1972 | |
5f9cf599 | 1973 | trigger_off(&auxtrace_snapshot_trigger); |
3c1cb7e3 | 1974 | trigger_off(&switch_output_trigger); |
de9ac07b | 1975 | |
ce7b0e42 AS |
1976 | if (opts->auxtrace_snapshot_on_exit) |
1977 | record__auxtrace_snapshot_exit(rec); | |
1978 | ||
f33cbe72 | 1979 | if (forks && workload_exec_errno) { |
3535a696 | 1980 | char msg[STRERR_BUFSIZE], strevsels[2048]; |
c8b5f2c9 | 1981 | const char *emsg = str_error_r(workload_exec_errno, msg, sizeof(msg)); |
3535a696 ACM |
1982 | |
1983 | evlist__scnprintf_evsels(rec->evlist, sizeof(strevsels), strevsels); | |
1984 | ||
1985 | pr_err("Failed to collect '%s' for the '%s' workload: %s\n", | |
1986 | strevsels, argv[0], emsg); | |
f33cbe72 | 1987 | err = -1; |
45604710 | 1988 | goto out_child; |
f33cbe72 ACM |
1989 | } |
1990 | ||
e3d59112 | 1991 | if (!quiet) |
45604710 | 1992 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
b44308f5 | 1993 | |
4ea648ae WN |
1994 | if (target__none(&rec->opts.target)) |
1995 | record__synthesize_workload(rec, true); | |
1996 | ||
45604710 | 1997 | out_child: |
1d078ccb | 1998 | evlist__finalize_ctlfd(rec->evlist); |
470530bb | 1999 | record__mmap_read_all(rec, true); |
d3d1af6f AB |
2000 | record__aio_mmap_read_sync(rec); |
2001 | ||
d3c8c08e AB |
2002 | if (rec->session->bytes_transferred && rec->session->bytes_compressed) { |
2003 | ratio = (float)rec->session->bytes_transferred/(float)rec->session->bytes_compressed; | |
2004 | session->header.env.comp_ratio = ratio + 0.5; | |
2005 | } | |
2006 | ||
45604710 NK |
2007 | if (forks) { |
2008 | int exit_status; | |
addc2785 | 2009 | |
45604710 NK |
2010 | if (!child_finished) |
2011 | kill(rec->evlist->workload.pid, SIGTERM); | |
2012 | ||
2013 | wait(&exit_status); | |
2014 | ||
2015 | if (err < 0) | |
2016 | status = err; | |
2017 | else if (WIFEXITED(exit_status)) | |
2018 | status = WEXITSTATUS(exit_status); | |
2019 | else if (WIFSIGNALED(exit_status)) | |
2020 | signr = WTERMSIG(exit_status); | |
2021 | } else | |
2022 | status = err; | |
2023 | ||
4ea648ae | 2024 | record__synthesize(rec, true); |
e3d59112 NK |
2025 | /* this will be recalculated during process_buildids() */ |
2026 | rec->samples = 0; | |
2027 | ||
ecfd7a9c WN |
2028 | if (!err) { |
2029 | if (!rec->timestamp_filename) { | |
2030 | record__finish_output(rec); | |
2031 | } else { | |
2032 | fd = record__switch_output(rec, true); | |
2033 | if (fd < 0) { | |
2034 | status = fd; | |
2035 | goto out_delete_session; | |
2036 | } | |
2037 | } | |
2038 | } | |
39d17dac | 2039 | |
a074865e WN |
2040 | perf_hooks__invoke_record_end(); |
2041 | ||
e3d59112 NK |
2042 | if (!err && !quiet) { |
2043 | char samples[128]; | |
ecfd7a9c WN |
2044 | const char *postfix = rec->timestamp_filename ? |
2045 | ".<timestamp>" : ""; | |
e3d59112 | 2046 | |
ef149c25 | 2047 | if (rec->samples && !rec->opts.full_auxtrace) |
e3d59112 NK |
2048 | scnprintf(samples, sizeof(samples), |
2049 | " (%" PRIu64 " samples)", rec->samples); | |
2050 | else | |
2051 | samples[0] = '\0'; | |
2052 | ||
d3c8c08e | 2053 | fprintf(stderr, "[ perf record: Captured and wrote %.3f MB %s%s%s", |
8ceb41d7 | 2054 | perf_data__size(data) / 1024.0 / 1024.0, |
2d4f2799 | 2055 | data->path, postfix, samples); |
d3c8c08e AB |
2056 | if (ratio) { |
2057 | fprintf(stderr, ", compressed (original %.3f MB, ratio is %.3f)", | |
2058 | rec->session->bytes_transferred / 1024.0 / 1024.0, | |
2059 | ratio); | |
2060 | } | |
2061 | fprintf(stderr, " ]\n"); | |
e3d59112 NK |
2062 | } |
2063 | ||
39d17dac | 2064 | out_delete_session: |
da231338 AM |
2065 | #ifdef HAVE_EVENTFD_SUPPORT |
2066 | if (done_fd >= 0) | |
2067 | close(done_fd); | |
2068 | #endif | |
5d7f4116 | 2069 | zstd_fini(&session->zstd_data); |
39d17dac | 2070 | perf_session__delete(session); |
657ee553 SL |
2071 | |
2072 | if (!opts->no_bpf_event) | |
08c83997 | 2073 | evlist__stop_sb_thread(rec->sb_evlist); |
45604710 | 2074 | return status; |
de9ac07b | 2075 | } |
0e9b20b8 | 2076 | |
0883e820 | 2077 | static void callchain_debug(struct callchain_param *callchain) |
09b0fd45 | 2078 | { |
aad2b21c | 2079 | static const char *str[CALLCHAIN_MAX] = { "NONE", "FP", "DWARF", "LBR" }; |
a601fdff | 2080 | |
0883e820 | 2081 | pr_debug("callchain: type %s\n", str[callchain->record_mode]); |
26d33022 | 2082 | |
0883e820 | 2083 | if (callchain->record_mode == CALLCHAIN_DWARF) |
09b0fd45 | 2084 | pr_debug("callchain: stack dump size %d\n", |
0883e820 | 2085 | callchain->dump_size); |
09b0fd45 JO |
2086 | } |
2087 | ||
0883e820 ACM |
2088 | int record_opts__parse_callchain(struct record_opts *record, |
2089 | struct callchain_param *callchain, | |
2090 | const char *arg, bool unset) | |
09b0fd45 | 2091 | { |
09b0fd45 | 2092 | int ret; |
0883e820 | 2093 | callchain->enabled = !unset; |
eb853e80 | 2094 | |
09b0fd45 JO |
2095 | /* --no-call-graph */ |
2096 | if (unset) { | |
0883e820 | 2097 | callchain->record_mode = CALLCHAIN_NONE; |
09b0fd45 JO |
2098 | pr_debug("callchain: disabled\n"); |
2099 | return 0; | |
2100 | } | |
2101 | ||
0883e820 | 2102 | ret = parse_callchain_record_opt(arg, callchain); |
5c0cf224 JO |
2103 | if (!ret) { |
2104 | /* Enable data address sampling for DWARF unwind. */ | |
0883e820 | 2105 | if (callchain->record_mode == CALLCHAIN_DWARF) |
5c0cf224 | 2106 | record->sample_address = true; |
0883e820 | 2107 | callchain_debug(callchain); |
5c0cf224 | 2108 | } |
26d33022 JO |
2109 | |
2110 | return ret; | |
2111 | } | |
2112 | ||
0883e820 ACM |
2113 | int record_parse_callchain_opt(const struct option *opt, |
2114 | const char *arg, | |
2115 | int unset) | |
2116 | { | |
2117 | return record_opts__parse_callchain(opt->value, &callchain_param, arg, unset); | |
2118 | } | |
2119 | ||
c421e80b | 2120 | int record_callchain_opt(const struct option *opt, |
09b0fd45 JO |
2121 | const char *arg __maybe_unused, |
2122 | int unset __maybe_unused) | |
2123 | { | |
2ddd5c04 | 2124 | struct callchain_param *callchain = opt->value; |
c421e80b | 2125 | |
2ddd5c04 | 2126 | callchain->enabled = true; |
09b0fd45 | 2127 | |
2ddd5c04 ACM |
2128 | if (callchain->record_mode == CALLCHAIN_NONE) |
2129 | callchain->record_mode = CALLCHAIN_FP; | |
eb853e80 | 2130 | |
2ddd5c04 | 2131 | callchain_debug(callchain); |
09b0fd45 JO |
2132 | return 0; |
2133 | } | |
2134 | ||
eb853e80 JO |
2135 | static int perf_record_config(const char *var, const char *value, void *cb) |
2136 | { | |
7a29c087 NK |
2137 | struct record *rec = cb; |
2138 | ||
2139 | if (!strcmp(var, "record.build-id")) { | |
2140 | if (!strcmp(value, "cache")) | |
2141 | rec->no_buildid_cache = false; | |
2142 | else if (!strcmp(value, "no-cache")) | |
2143 | rec->no_buildid_cache = true; | |
2144 | else if (!strcmp(value, "skip")) | |
2145 | rec->no_buildid = true; | |
e29386c8 JO |
2146 | else if (!strcmp(value, "mmap")) |
2147 | rec->buildid_mmap = true; | |
7a29c087 NK |
2148 | else |
2149 | return -1; | |
2150 | return 0; | |
2151 | } | |
cff17205 YX |
2152 | if (!strcmp(var, "record.call-graph")) { |
2153 | var = "call-graph.record-mode"; | |
2154 | return perf_default_config(var, value, cb); | |
2155 | } | |
93f20c0f AB |
2156 | #ifdef HAVE_AIO_SUPPORT |
2157 | if (!strcmp(var, "record.aio")) { | |
2158 | rec->opts.nr_cblocks = strtol(value, NULL, 0); | |
2159 | if (!rec->opts.nr_cblocks) | |
2160 | rec->opts.nr_cblocks = nr_cblocks_default; | |
2161 | } | |
2162 | #endif | |
eb853e80 | 2163 | |
cff17205 | 2164 | return 0; |
eb853e80 JO |
2165 | } |
2166 | ||
814c8c38 | 2167 | |
f4fe11b7 AB |
2168 | static int record__parse_affinity(const struct option *opt, const char *str, int unset) |
2169 | { | |
2170 | struct record_opts *opts = (struct record_opts *)opt->value; | |
2171 | ||
2172 | if (unset || !str) | |
2173 | return 0; | |
2174 | ||
2175 | if (!strcasecmp(str, "node")) | |
2176 | opts->affinity = PERF_AFFINITY_NODE; | |
2177 | else if (!strcasecmp(str, "cpu")) | |
2178 | opts->affinity = PERF_AFFINITY_CPU; | |
2179 | ||
2180 | return 0; | |
2181 | } | |
2182 | ||
6d575816 JS |
2183 | static int parse_output_max_size(const struct option *opt, |
2184 | const char *str, int unset) | |
2185 | { | |
2186 | unsigned long *s = (unsigned long *)opt->value; | |
2187 | static struct parse_tag tags_size[] = { | |
2188 | { .tag = 'B', .mult = 1 }, | |
2189 | { .tag = 'K', .mult = 1 << 10 }, | |
2190 | { .tag = 'M', .mult = 1 << 20 }, | |
2191 | { .tag = 'G', .mult = 1 << 30 }, | |
2192 | { .tag = 0 }, | |
2193 | }; | |
2194 | unsigned long val; | |
2195 | ||
2196 | if (unset) { | |
2197 | *s = 0; | |
2198 | return 0; | |
2199 | } | |
2200 | ||
2201 | val = parse_tag_value(str, tags_size); | |
2202 | if (val != (unsigned long) -1) { | |
2203 | *s = val; | |
2204 | return 0; | |
2205 | } | |
2206 | ||
2207 | return -1; | |
2208 | } | |
2209 | ||
e9db1310 AH |
2210 | static int record__parse_mmap_pages(const struct option *opt, |
2211 | const char *str, | |
2212 | int unset __maybe_unused) | |
2213 | { | |
2214 | struct record_opts *opts = opt->value; | |
2215 | char *s, *p; | |
2216 | unsigned int mmap_pages; | |
2217 | int ret; | |
2218 | ||
2219 | if (!str) | |
2220 | return -EINVAL; | |
2221 | ||
2222 | s = strdup(str); | |
2223 | if (!s) | |
2224 | return -ENOMEM; | |
2225 | ||
2226 | p = strchr(s, ','); | |
2227 | if (p) | |
2228 | *p = '\0'; | |
2229 | ||
2230 | if (*s) { | |
25f84702 | 2231 | ret = __evlist__parse_mmap_pages(&mmap_pages, s); |
e9db1310 AH |
2232 | if (ret) |
2233 | goto out_free; | |
2234 | opts->mmap_pages = mmap_pages; | |
2235 | } | |
2236 | ||
2237 | if (!p) { | |
2238 | ret = 0; | |
2239 | goto out_free; | |
2240 | } | |
2241 | ||
25f84702 | 2242 | ret = __evlist__parse_mmap_pages(&mmap_pages, p + 1); |
e9db1310 AH |
2243 | if (ret) |
2244 | goto out_free; | |
2245 | ||
2246 | opts->auxtrace_mmap_pages = mmap_pages; | |
2247 | ||
2248 | out_free: | |
2249 | free(s); | |
2250 | return ret; | |
2251 | } | |
2252 | ||
1d078ccb AB |
2253 | static int parse_control_option(const struct option *opt, |
2254 | const char *str, | |
2255 | int unset __maybe_unused) | |
2256 | { | |
9864a66d | 2257 | struct record_opts *opts = opt->value; |
1d078ccb | 2258 | |
a8fcbd26 AH |
2259 | return evlist__parse_control(str, &opts->ctl_fd, &opts->ctl_fd_ack, &opts->ctl_fd_close); |
2260 | } | |
2261 | ||
0c582449 JO |
2262 | static void switch_output_size_warn(struct record *rec) |
2263 | { | |
9521b5f2 | 2264 | u64 wakeup_size = evlist__mmap_size(rec->opts.mmap_pages); |
0c582449 JO |
2265 | struct switch_output *s = &rec->switch_output; |
2266 | ||
2267 | wakeup_size /= 2; | |
2268 | ||
2269 | if (s->size < wakeup_size) { | |
2270 | char buf[100]; | |
2271 | ||
2272 | unit_number__scnprintf(buf, sizeof(buf), wakeup_size); | |
2273 | pr_warning("WARNING: switch-output data size lower than " | |
2274 | "wakeup kernel buffer size (%s) " | |
2275 | "expect bigger perf.data sizes\n", buf); | |
2276 | } | |
2277 | } | |
2278 | ||
cb4e1ebb JO |
2279 | static int switch_output_setup(struct record *rec) |
2280 | { | |
2281 | struct switch_output *s = &rec->switch_output; | |
dc0c6127 JO |
2282 | static struct parse_tag tags_size[] = { |
2283 | { .tag = 'B', .mult = 1 }, | |
2284 | { .tag = 'K', .mult = 1 << 10 }, | |
2285 | { .tag = 'M', .mult = 1 << 20 }, | |
2286 | { .tag = 'G', .mult = 1 << 30 }, | |
2287 | { .tag = 0 }, | |
2288 | }; | |
bfacbe3b JO |
2289 | static struct parse_tag tags_time[] = { |
2290 | { .tag = 's', .mult = 1 }, | |
2291 | { .tag = 'm', .mult = 60 }, | |
2292 | { .tag = 'h', .mult = 60*60 }, | |
2293 | { .tag = 'd', .mult = 60*60*24 }, | |
2294 | { .tag = 0 }, | |
2295 | }; | |
dc0c6127 | 2296 | unsigned long val; |
cb4e1ebb | 2297 | |
899e5ffb ACM |
2298 | /* |
2299 | * If we're using --switch-output-events, then we imply its | |
2300 | * --switch-output=signal, as we'll send a SIGUSR2 from the side band | |
2301 | * thread to its parent. | |
2302 | */ | |
2303 | if (rec->switch_output_event_set) | |
2304 | goto do_signal; | |
2305 | ||
cb4e1ebb JO |
2306 | if (!s->set) |
2307 | return 0; | |
2308 | ||
2309 | if (!strcmp(s->str, "signal")) { | |
899e5ffb | 2310 | do_signal: |
cb4e1ebb JO |
2311 | s->signal = true; |
2312 | pr_debug("switch-output with SIGUSR2 signal\n"); | |
dc0c6127 JO |
2313 | goto enabled; |
2314 | } | |
2315 | ||
2316 | val = parse_tag_value(s->str, tags_size); | |
2317 | if (val != (unsigned long) -1) { | |
2318 | s->size = val; | |
2319 | pr_debug("switch-output with %s size threshold\n", s->str); | |
2320 | goto enabled; | |
cb4e1ebb JO |
2321 | } |
2322 | ||
bfacbe3b JO |
2323 | val = parse_tag_value(s->str, tags_time); |
2324 | if (val != (unsigned long) -1) { | |
2325 | s->time = val; | |
2326 | pr_debug("switch-output with %s time threshold (%lu seconds)\n", | |
2327 | s->str, s->time); | |
2328 | goto enabled; | |
2329 | } | |
2330 | ||
cb4e1ebb | 2331 | return -1; |
dc0c6127 JO |
2332 | |
2333 | enabled: | |
2334 | rec->timestamp_filename = true; | |
2335 | s->enabled = true; | |
0c582449 JO |
2336 | |
2337 | if (s->size && !rec->opts.no_buffering) | |
2338 | switch_output_size_warn(rec); | |
2339 | ||
dc0c6127 | 2340 | return 0; |
cb4e1ebb JO |
2341 | } |
2342 | ||
e5b2c207 | 2343 | static const char * const __record_usage[] = { |
9e096753 MG |
2344 | "perf record [<options>] [<command>]", |
2345 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
2346 | NULL |
2347 | }; | |
e5b2c207 | 2348 | const char * const *record_usage = __record_usage; |
0e9b20b8 | 2349 | |
6e0a9b3d ACM |
2350 | static int build_id__process_mmap(struct perf_tool *tool, union perf_event *event, |
2351 | struct perf_sample *sample, struct machine *machine) | |
2352 | { | |
2353 | /* | |
2354 | * We already have the kernel maps, put in place via perf_session__create_kernel_maps() | |
2355 | * no need to add them twice. | |
2356 | */ | |
2357 | if (!(event->header.misc & PERF_RECORD_MISC_USER)) | |
2358 | return 0; | |
2359 | return perf_event__process_mmap(tool, event, sample, machine); | |
2360 | } | |
2361 | ||
2362 | static int build_id__process_mmap2(struct perf_tool *tool, union perf_event *event, | |
2363 | struct perf_sample *sample, struct machine *machine) | |
2364 | { | |
2365 | /* | |
2366 | * We already have the kernel maps, put in place via perf_session__create_kernel_maps() | |
2367 | * no need to add them twice. | |
2368 | */ | |
2369 | if (!(event->header.misc & PERF_RECORD_MISC_USER)) | |
2370 | return 0; | |
2371 | ||
2372 | return perf_event__process_mmap2(tool, event, sample, machine); | |
2373 | } | |
2374 | ||
d20deb64 | 2375 | /* |
8c6f45a7 ACM |
2376 | * XXX Ideally would be local to cmd_record() and passed to a record__new |
2377 | * because we need to have access to it in record__exit, that is called | |
d20deb64 ACM |
2378 | * after cmd_record() exits, but since record_options need to be accessible to |
2379 | * builtin-script, leave it here. | |
2380 | * | |
2381 | * At least we don't ouch it in all the other functions here directly. | |
2382 | * | |
2383 | * Just say no to tons of global variables, sigh. | |
2384 | */ | |
8c6f45a7 | 2385 | static struct record record = { |
d20deb64 | 2386 | .opts = { |
8affc2b8 | 2387 | .sample_time = true, |
d20deb64 ACM |
2388 | .mmap_pages = UINT_MAX, |
2389 | .user_freq = UINT_MAX, | |
2390 | .user_interval = ULLONG_MAX, | |
447a6013 | 2391 | .freq = 4000, |
d1cb9fce NK |
2392 | .target = { |
2393 | .uses_mmap = true, | |
3aa5939d | 2394 | .default_per_cpu = true, |
d1cb9fce | 2395 | }, |
470530bb | 2396 | .mmap_flush = MMAP_FLUSH_DEFAULT, |
d99c22ea | 2397 | .nr_threads_synthesize = 1, |
1d078ccb AB |
2398 | .ctl_fd = -1, |
2399 | .ctl_fd_ack = -1, | |
d20deb64 | 2400 | }, |
e3d59112 NK |
2401 | .tool = { |
2402 | .sample = process_sample_event, | |
2403 | .fork = perf_event__process_fork, | |
cca8482c | 2404 | .exit = perf_event__process_exit, |
e3d59112 | 2405 | .comm = perf_event__process_comm, |
f3b3614a | 2406 | .namespaces = perf_event__process_namespaces, |
6e0a9b3d ACM |
2407 | .mmap = build_id__process_mmap, |
2408 | .mmap2 = build_id__process_mmap2, | |
cca8482c | 2409 | .ordered_events = true, |
e3d59112 | 2410 | }, |
d20deb64 | 2411 | }; |
7865e817 | 2412 | |
76a26549 NK |
2413 | const char record_callchain_help[] = CALLCHAIN_RECORD_HELP |
2414 | "\n\t\t\t\tDefault: fp"; | |
61eaa3be | 2415 | |
0aab2136 WN |
2416 | static bool dry_run; |
2417 | ||
d20deb64 ACM |
2418 | /* |
2419 | * XXX Will stay a global variable till we fix builtin-script.c to stop messing | |
2420 | * with it and switch to use the library functions in perf_evlist that came | |
b4006796 | 2421 | * from builtin-record.c, i.e. use record_opts, |
7b392ef0 | 2422 | * evlist__prepare_workload, etc instead of fork+exec'in 'perf record', |
d20deb64 ACM |
2423 | * using pipes, etc. |
2424 | */ | |
efd21307 | 2425 | static struct option __record_options[] = { |
d20deb64 | 2426 | OPT_CALLBACK('e', "event", &record.evlist, "event", |
86847b62 | 2427 | "event selector. use 'perf list' to list available events", |
f120f9d5 | 2428 | parse_events_option), |
d20deb64 | 2429 | OPT_CALLBACK(0, "filter", &record.evlist, "filter", |
c171b552 | 2430 | "event filter", parse_filter), |
4ba1faa1 WN |
2431 | OPT_CALLBACK_NOOPT(0, "exclude-perf", &record.evlist, |
2432 | NULL, "don't record events from perf itself", | |
2433 | exclude_perf), | |
bea03405 | 2434 | OPT_STRING('p', "pid", &record.opts.target.pid, "pid", |
d6d901c2 | 2435 | "record events on existing process id"), |
bea03405 | 2436 | OPT_STRING('t', "tid", &record.opts.target.tid, "tid", |
d6d901c2 | 2437 | "record events on existing thread id"), |
d20deb64 | 2438 | OPT_INTEGER('r', "realtime", &record.realtime_prio, |
0e9b20b8 | 2439 | "collect data with this RT SCHED_FIFO priority"), |
509051ea | 2440 | OPT_BOOLEAN(0, "no-buffering", &record.opts.no_buffering, |
acac03fa | 2441 | "collect data without buffering"), |
d20deb64 | 2442 | OPT_BOOLEAN('R', "raw-samples", &record.opts.raw_samples, |
daac07b2 | 2443 | "collect raw sample records from all opened counters"), |
bea03405 | 2444 | OPT_BOOLEAN('a', "all-cpus", &record.opts.target.system_wide, |
0e9b20b8 | 2445 | "system-wide collection from all CPUs"), |
bea03405 | 2446 | OPT_STRING('C', "cpu", &record.opts.target.cpu_list, "cpu", |
c45c6ea2 | 2447 | "list of cpus to monitor"), |
d20deb64 | 2448 | OPT_U64('c', "count", &record.opts.user_interval, "event period to sample"), |
2d4f2799 | 2449 | OPT_STRING('o', "output", &record.data.path, "file", |
abaff32a | 2450 | "output file name"), |
69e7e5b0 AH |
2451 | OPT_BOOLEAN_SET('i', "no-inherit", &record.opts.no_inherit, |
2452 | &record.opts.no_inherit_set, | |
2453 | "child tasks do not inherit counters"), | |
4ea648ae WN |
2454 | OPT_BOOLEAN(0, "tail-synthesize", &record.opts.tail_synthesize, |
2455 | "synthesize non-sample events at the end of output"), | |
626a6b78 | 2456 | OPT_BOOLEAN(0, "overwrite", &record.opts.overwrite, "use overwrite mode"), |
a060c1f1 | 2457 | OPT_BOOLEAN(0, "no-bpf-event", &record.opts.no_bpf_event, "do not record bpf events"), |
b09c2364 ACM |
2458 | OPT_BOOLEAN(0, "strict-freq", &record.opts.strict_freq, |
2459 | "Fail if the specified frequency can't be used"), | |
67230479 ACM |
2460 | OPT_CALLBACK('F', "freq", &record.opts, "freq or 'max'", |
2461 | "profile at this frequency", | |
2462 | record__parse_freq), | |
e9db1310 AH |
2463 | OPT_CALLBACK('m', "mmap-pages", &record.opts, "pages[,pages]", |
2464 | "number of mmap data pages and AUX area tracing mmap pages", | |
2465 | record__parse_mmap_pages), | |
470530bb AB |
2466 | OPT_CALLBACK(0, "mmap-flush", &record.opts, "number", |
2467 | "Minimal number of bytes that is extracted from mmap data pages (default: 1)", | |
2468 | record__mmap_flush_parse), | |
d20deb64 | 2469 | OPT_BOOLEAN(0, "group", &record.opts.group, |
43bece79 | 2470 | "put the counters into a counter group"), |
2ddd5c04 | 2471 | OPT_CALLBACK_NOOPT('g', NULL, &callchain_param, |
09b0fd45 JO |
2472 | NULL, "enables call-graph recording" , |
2473 | &record_callchain_opt), | |
2474 | OPT_CALLBACK(0, "call-graph", &record.opts, | |
76a26549 | 2475 | "record_mode[,record_size]", record_callchain_help, |
09b0fd45 | 2476 | &record_parse_callchain_opt), |
c0555642 | 2477 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 2478 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 2479 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
d20deb64 | 2480 | OPT_BOOLEAN('s', "stat", &record.opts.inherit_stat, |
649c48a9 | 2481 | "per thread counts"), |
56100321 | 2482 | OPT_BOOLEAN('d', "data", &record.opts.sample_address, "Record the sample addresses"), |
3b0a5daa KL |
2483 | OPT_BOOLEAN(0, "phys-data", &record.opts.sample_phys_addr, |
2484 | "Record the sample physical addresses"), | |
542b88fd KL |
2485 | OPT_BOOLEAN(0, "data-page-size", &record.opts.sample_data_page_size, |
2486 | "Record the sampled data address data page size"), | |
c1de7f3d KL |
2487 | OPT_BOOLEAN(0, "code-page-size", &record.opts.sample_code_page_size, |
2488 | "Record the sampled code address (ip) page size"), | |
b6f35ed7 | 2489 | OPT_BOOLEAN(0, "sample-cpu", &record.opts.sample_cpu, "Record the sample cpu"), |
3abebc55 AH |
2490 | OPT_BOOLEAN_SET('T', "timestamp", &record.opts.sample_time, |
2491 | &record.opts.sample_time_set, | |
2492 | "Record the sample timestamps"), | |
f290aa1f JO |
2493 | OPT_BOOLEAN_SET('P', "period", &record.opts.period, &record.opts.period_set, |
2494 | "Record the sample period"), | |
d20deb64 | 2495 | OPT_BOOLEAN('n', "no-samples", &record.opts.no_samples, |
649c48a9 | 2496 | "don't sample"), |
d2db9a98 WN |
2497 | OPT_BOOLEAN_SET('N', "no-buildid-cache", &record.no_buildid_cache, |
2498 | &record.no_buildid_cache_set, | |
2499 | "do not update the buildid cache"), | |
2500 | OPT_BOOLEAN_SET('B', "no-buildid", &record.no_buildid, | |
2501 | &record.no_buildid_set, | |
2502 | "do not collect buildids in perf.data"), | |
d20deb64 | 2503 | OPT_CALLBACK('G', "cgroup", &record.evlist, "name", |
023695d9 SE |
2504 | "monitor event in cgroup name only", |
2505 | parse_cgroups), | |
68cd3b45 AB |
2506 | OPT_INTEGER('D', "delay", &record.opts.initial_delay, |
2507 | "ms to wait before starting measurement after program start (-1: start with events disabled)"), | |
eeb399b5 | 2508 | OPT_BOOLEAN(0, "kcore", &record.opts.kcore, "copy /proc/kcore"), |
bea03405 NK |
2509 | OPT_STRING('u', "uid", &record.opts.target.uid_str, "user", |
2510 | "user to profile"), | |
a5aabdac SE |
2511 | |
2512 | OPT_CALLBACK_NOOPT('b', "branch-any", &record.opts.branch_stack, | |
2513 | "branch any", "sample any taken branches", | |
2514 | parse_branch_stack), | |
2515 | ||
2516 | OPT_CALLBACK('j', "branch-filter", &record.opts.branch_stack, | |
2517 | "branch filter mask", "branch stack filter modes", | |
bdfebd84 | 2518 | parse_branch_stack), |
05484298 AK |
2519 | OPT_BOOLEAN('W', "weight", &record.opts.sample_weight, |
2520 | "sample by weight (on special events only)"), | |
475eeab9 AK |
2521 | OPT_BOOLEAN(0, "transaction", &record.opts.sample_transaction, |
2522 | "sample transaction flags (special events only)"), | |
3aa5939d AH |
2523 | OPT_BOOLEAN(0, "per-thread", &record.opts.target.per_thread, |
2524 | "use per-thread mmaps"), | |
bcc84ec6 SE |
2525 | OPT_CALLBACK_OPTARG('I', "intr-regs", &record.opts.sample_intr_regs, NULL, "any register", |
2526 | "sample selected machine registers on interrupt," | |
aeea9062 | 2527 | " use '-I?' to list register names", parse_intr_regs), |
84c41742 AK |
2528 | OPT_CALLBACK_OPTARG(0, "user-regs", &record.opts.sample_user_regs, NULL, "any register", |
2529 | "sample selected machine registers on interrupt," | |
aeea9062 | 2530 | " use '--user-regs=?' to list register names", parse_user_regs), |
85c273d2 AK |
2531 | OPT_BOOLEAN(0, "running-time", &record.opts.running_time, |
2532 | "Record running/enabled time of read (:S) events"), | |
814c8c38 PZ |
2533 | OPT_CALLBACK('k', "clockid", &record.opts, |
2534 | "clockid", "clockid to use for events, see clock_gettime()", | |
2535 | parse_clockid), | |
2dd6d8a1 AH |
2536 | OPT_STRING_OPTARG('S', "snapshot", &record.opts.auxtrace_snapshot_opts, |
2537 | "opts", "AUX area tracing Snapshot Mode", ""), | |
c0a6de06 AH |
2538 | OPT_STRING_OPTARG(0, "aux-sample", &record.opts.auxtrace_sample_opts, |
2539 | "opts", "sample AUX area", ""), | |
3fcb10e4 | 2540 | OPT_UINTEGER(0, "proc-map-timeout", &proc_map_timeout, |
9d9cad76 | 2541 | "per thread proc mmap processing timeout in ms"), |
f3b3614a HB |
2542 | OPT_BOOLEAN(0, "namespaces", &record.opts.record_namespaces, |
2543 | "Record namespaces events"), | |
8fb4b679 NK |
2544 | OPT_BOOLEAN(0, "all-cgroups", &record.opts.record_cgroup, |
2545 | "Record cgroup events"), | |
16b4b4e1 AH |
2546 | OPT_BOOLEAN_SET(0, "switch-events", &record.opts.record_switch_events, |
2547 | &record.opts.record_switch_events_set, | |
2548 | "Record context switch events"), | |
85723885 JO |
2549 | OPT_BOOLEAN_FLAG(0, "all-kernel", &record.opts.all_kernel, |
2550 | "Configure all used events to run in kernel space.", | |
2551 | PARSE_OPT_EXCLUSIVE), | |
2552 | OPT_BOOLEAN_FLAG(0, "all-user", &record.opts.all_user, | |
2553 | "Configure all used events to run in user space.", | |
2554 | PARSE_OPT_EXCLUSIVE), | |
53651b28 | 2555 | OPT_BOOLEAN(0, "kernel-callchains", &record.opts.kernel_callchains, |
2556 | "collect kernel callchains"), | |
2557 | OPT_BOOLEAN(0, "user-callchains", &record.opts.user_callchains, | |
2558 | "collect user callchains"), | |
71dc2326 WN |
2559 | OPT_STRING(0, "clang-path", &llvm_param.clang_path, "clang path", |
2560 | "clang binary to use for compiling BPF scriptlets"), | |
2561 | OPT_STRING(0, "clang-opt", &llvm_param.clang_opt, "clang options", | |
2562 | "options passed to clang when compiling BPF scriptlets"), | |
7efe0e03 HK |
2563 | OPT_STRING(0, "vmlinux", &symbol_conf.vmlinux_name, |
2564 | "file", "vmlinux pathname"), | |
6156681b NK |
2565 | OPT_BOOLEAN(0, "buildid-all", &record.buildid_all, |
2566 | "Record build-id of all DSOs regardless of hits"), | |
e29386c8 JO |
2567 | OPT_BOOLEAN(0, "buildid-mmap", &record.buildid_mmap, |
2568 | "Record build-id in map events"), | |
ecfd7a9c WN |
2569 | OPT_BOOLEAN(0, "timestamp-filename", &record.timestamp_filename, |
2570 | "append timestamp to output filename"), | |
68588baf JY |
2571 | OPT_BOOLEAN(0, "timestamp-boundary", &record.timestamp_boundary, |
2572 | "Record timestamp boundary (time of first/last samples)"), | |
cb4e1ebb | 2573 | OPT_STRING_OPTARG_SET(0, "switch-output", &record.switch_output.str, |
c38dab7d AK |
2574 | &record.switch_output.set, "signal or size[BKMG] or time[smhd]", |
2575 | "Switch output when receiving SIGUSR2 (signal) or cross a size or time threshold", | |
dc0c6127 | 2576 | "signal"), |
899e5ffb ACM |
2577 | OPT_CALLBACK_SET(0, "switch-output-event", &record.sb_evlist, &record.switch_output_event_set, "switch output event", |
2578 | "switch output event selector. use 'perf list' to list available events", | |
2579 | parse_events_option_new_evlist), | |
03724b2e AK |
2580 | OPT_INTEGER(0, "switch-max-files", &record.switch_output.num_files, |
2581 | "Limit number of switch output generated files"), | |
0aab2136 WN |
2582 | OPT_BOOLEAN(0, "dry-run", &dry_run, |
2583 | "Parse options then exit"), | |
d3d1af6f | 2584 | #ifdef HAVE_AIO_SUPPORT |
93f20c0f AB |
2585 | OPT_CALLBACK_OPTARG(0, "aio", &record.opts, |
2586 | &nr_cblocks_default, "n", "Use <n> control blocks in asynchronous trace writing mode (default: 1, max: 4)", | |
d3d1af6f AB |
2587 | record__aio_parse), |
2588 | #endif | |
f4fe11b7 AB |
2589 | OPT_CALLBACK(0, "affinity", &record.opts, "node|cpu", |
2590 | "Set affinity mask of trace reading thread to NUMA node cpu mask or cpu of processed mmap buffer", | |
2591 | record__parse_affinity), | |
504c1ad1 AB |
2592 | #ifdef HAVE_ZSTD_SUPPORT |
2593 | OPT_CALLBACK_OPTARG('z', "compression-level", &record.opts, &comp_level_default, | |
2594 | "n", "Compressed records using specified level (default: 1 - fastest compression, 22 - greatest compression)", | |
2595 | record__parse_comp_level), | |
2596 | #endif | |
6d575816 JS |
2597 | OPT_CALLBACK(0, "max-size", &record.output_max_size, |
2598 | "size", "Limit the maximum size of the output file", parse_output_max_size), | |
d99c22ea SE |
2599 | OPT_UINTEGER(0, "num-thread-synthesize", |
2600 | &record.opts.nr_threads_synthesize, | |
2601 | "number of threads to run for event synthesis"), | |
70943490 SE |
2602 | #ifdef HAVE_LIBPFM |
2603 | OPT_CALLBACK(0, "pfm-events", &record.evlist, "event", | |
2604 | "libpfm4 event selector. use 'perf list' to list available events", | |
2605 | parse_libpfm_events_option), | |
2606 | #endif | |
a8fcbd26 | 2607 | OPT_CALLBACK(0, "control", &record.opts, "fd:ctl-fd[,ack-fd] or fifo:ctl-fifo[,ack-fifo]", |
d20aff15 AH |
2608 | "Listen on ctl-fd descriptor for command to control measurement ('enable': enable events, 'disable': disable events,\n" |
2609 | "\t\t\t 'snapshot': AUX area tracing snapshot).\n" | |
a8fcbd26 AH |
2610 | "\t\t\t Optionally send control command completion ('ack\\n') to ack-fd descriptor.\n" |
2611 | "\t\t\t Alternatively, ctl-fifo / ack-fifo will be opened and used as ctl-fd / ack-fd.", | |
1d078ccb | 2612 | parse_control_option), |
0e9b20b8 IM |
2613 | OPT_END() |
2614 | }; | |
2615 | ||
e5b2c207 NK |
2616 | struct option *record_options = __record_options; |
2617 | ||
b0ad8ea6 | 2618 | int cmd_record(int argc, const char **argv) |
0e9b20b8 | 2619 | { |
ef149c25 | 2620 | int err; |
8c6f45a7 | 2621 | struct record *rec = &record; |
16ad2ffb | 2622 | char errbuf[BUFSIZ]; |
0e9b20b8 | 2623 | |
67230479 ACM |
2624 | setlocale(LC_ALL, ""); |
2625 | ||
48e1cab1 WN |
2626 | #ifndef HAVE_LIBBPF_SUPPORT |
2627 | # define set_nobuild(s, l, c) set_option_nobuild(record_options, s, l, "NO_LIBBPF=1", c) | |
2628 | set_nobuild('\0', "clang-path", true); | |
2629 | set_nobuild('\0', "clang-opt", true); | |
2630 | # undef set_nobuild | |
7efe0e03 HK |
2631 | #endif |
2632 | ||
2633 | #ifndef HAVE_BPF_PROLOGUE | |
2634 | # if !defined (HAVE_DWARF_SUPPORT) | |
2635 | # define REASON "NO_DWARF=1" | |
2636 | # elif !defined (HAVE_LIBBPF_SUPPORT) | |
2637 | # define REASON "NO_LIBBPF=1" | |
2638 | # else | |
2639 | # define REASON "this architecture doesn't support BPF prologue" | |
2640 | # endif | |
2641 | # define set_nobuild(s, l, c) set_option_nobuild(record_options, s, l, REASON, c) | |
2642 | set_nobuild('\0', "vmlinux", true); | |
2643 | # undef set_nobuild | |
2644 | # undef REASON | |
48e1cab1 WN |
2645 | #endif |
2646 | ||
9d2ed645 AB |
2647 | rec->opts.affinity = PERF_AFFINITY_SYS; |
2648 | ||
0f98b11c | 2649 | rec->evlist = evlist__new(); |
3e2be2da | 2650 | if (rec->evlist == NULL) |
361c99a6 ACM |
2651 | return -ENOMEM; |
2652 | ||
ecc4c561 ACM |
2653 | err = perf_config(perf_record_config, rec); |
2654 | if (err) | |
2655 | return err; | |
eb853e80 | 2656 | |
bca647aa | 2657 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 2658 | PARSE_OPT_STOP_AT_NON_OPTION); |
68ba3235 NK |
2659 | if (quiet) |
2660 | perf_quiet_option(); | |
483635a9 JO |
2661 | |
2662 | /* Make system wide (-a) the default target. */ | |
602ad878 | 2663 | if (!argc && target__none(&rec->opts.target)) |
483635a9 | 2664 | rec->opts.target.system_wide = true; |
0e9b20b8 | 2665 | |
bea03405 | 2666 | if (nr_cgroups && !rec->opts.target.system_wide) { |
c7118369 NK |
2667 | usage_with_options_msg(record_usage, record_options, |
2668 | "cgroup monitoring only available in system-wide mode"); | |
2669 | ||
023695d9 | 2670 | } |
504c1ad1 | 2671 | |
e29386c8 JO |
2672 | if (rec->buildid_mmap) { |
2673 | if (!perf_can_record_build_id()) { | |
2674 | pr_err("Failed: no support to record build id in mmap events, update your kernel.\n"); | |
2675 | err = -EINVAL; | |
2676 | goto out_opts; | |
2677 | } | |
2678 | pr_debug("Enabling build id in mmap2 events.\n"); | |
2679 | /* Enable mmap build id synthesizing. */ | |
2680 | symbol_conf.buildid_mmap2 = true; | |
2681 | /* Enable perf_event_attr::build_id bit. */ | |
2682 | rec->opts.build_id = true; | |
2683 | /* Disable build id cache. */ | |
2684 | rec->no_buildid = true; | |
2685 | } | |
2686 | ||
eeb399b5 AH |
2687 | if (rec->opts.kcore) |
2688 | rec->data.is_dir = true; | |
2689 | ||
504c1ad1 AB |
2690 | if (rec->opts.comp_level != 0) { |
2691 | pr_debug("Compression enabled, disabling build id collection at the end of the session.\n"); | |
2692 | rec->no_buildid = true; | |
2693 | } | |
2694 | ||
b757bb09 AH |
2695 | if (rec->opts.record_switch_events && |
2696 | !perf_can_record_switch_events()) { | |
c7118369 NK |
2697 | ui__error("kernel does not support recording context switch events\n"); |
2698 | parse_options_usage(record_usage, record_options, "switch-events", 0); | |
a8fcbd26 AH |
2699 | err = -EINVAL; |
2700 | goto out_opts; | |
b757bb09 | 2701 | } |
023695d9 | 2702 | |
cb4e1ebb JO |
2703 | if (switch_output_setup(rec)) { |
2704 | parse_options_usage(record_usage, record_options, "switch-output", 0); | |
a8fcbd26 AH |
2705 | err = -EINVAL; |
2706 | goto out_opts; | |
cb4e1ebb JO |
2707 | } |
2708 | ||
bfacbe3b JO |
2709 | if (rec->switch_output.time) { |
2710 | signal(SIGALRM, alarm_sig_handler); | |
2711 | alarm(rec->switch_output.time); | |
2712 | } | |
2713 | ||
03724b2e AK |
2714 | if (rec->switch_output.num_files) { |
2715 | rec->switch_output.filenames = calloc(sizeof(char *), | |
2716 | rec->switch_output.num_files); | |
a8fcbd26 AH |
2717 | if (!rec->switch_output.filenames) { |
2718 | err = -EINVAL; | |
2719 | goto out_opts; | |
2720 | } | |
03724b2e AK |
2721 | } |
2722 | ||
1b36c03e AH |
2723 | /* |
2724 | * Allow aliases to facilitate the lookup of symbols for address | |
2725 | * filters. Refer to auxtrace_parse_filters(). | |
2726 | */ | |
2727 | symbol_conf.allow_aliases = true; | |
2728 | ||
2729 | symbol__init(NULL); | |
2730 | ||
8384a260 AB |
2731 | if (rec->opts.affinity != PERF_AFFINITY_SYS) { |
2732 | rec->affinity_mask.nbits = cpu__max_cpu(); | |
2733 | rec->affinity_mask.bits = bitmap_alloc(rec->affinity_mask.nbits); | |
2734 | if (!rec->affinity_mask.bits) { | |
2735 | pr_err("Failed to allocate thread mask for %zd cpus\n", rec->affinity_mask.nbits); | |
a8fcbd26 AH |
2736 | err = -ENOMEM; |
2737 | goto out_opts; | |
8384a260 AB |
2738 | } |
2739 | pr_debug2("thread mask[%zd]: empty\n", rec->affinity_mask.nbits); | |
2740 | } | |
2741 | ||
4b5ea3bd | 2742 | err = record__auxtrace_init(rec); |
1b36c03e AH |
2743 | if (err) |
2744 | goto out; | |
2745 | ||
0aab2136 | 2746 | if (dry_run) |
5c01ad60 | 2747 | goto out; |
0aab2136 | 2748 | |
d7888573 WN |
2749 | err = bpf__setup_stdout(rec->evlist); |
2750 | if (err) { | |
2751 | bpf__strerror_setup_stdout(rec->evlist, err, errbuf, sizeof(errbuf)); | |
2752 | pr_err("ERROR: Setup BPF stdout failed: %s\n", | |
2753 | errbuf); | |
5c01ad60 | 2754 | goto out; |
d7888573 WN |
2755 | } |
2756 | ||
ef149c25 AH |
2757 | err = -ENOMEM; |
2758 | ||
0c1d46a8 | 2759 | if (rec->no_buildid_cache || rec->no_buildid) { |
a1ac1d3c | 2760 | disable_buildid_cache(); |
dc0c6127 | 2761 | } else if (rec->switch_output.enabled) { |
0c1d46a8 WN |
2762 | /* |
2763 | * In 'perf record --switch-output', disable buildid | |
2764 | * generation by default to reduce data file switching | |
2765 | * overhead. Still generate buildid if they are required | |
2766 | * explicitly using | |
2767 | * | |
60437ac0 | 2768 | * perf record --switch-output --no-no-buildid \ |
0c1d46a8 WN |
2769 | * --no-no-buildid-cache |
2770 | * | |
2771 | * Following code equals to: | |
2772 | * | |
2773 | * if ((rec->no_buildid || !rec->no_buildid_set) && | |
2774 | * (rec->no_buildid_cache || !rec->no_buildid_cache_set)) | |
2775 | * disable_buildid_cache(); | |
2776 | */ | |
2777 | bool disable = true; | |
2778 | ||
2779 | if (rec->no_buildid_set && !rec->no_buildid) | |
2780 | disable = false; | |
2781 | if (rec->no_buildid_cache_set && !rec->no_buildid_cache) | |
2782 | disable = false; | |
2783 | if (disable) { | |
2784 | rec->no_buildid = true; | |
2785 | rec->no_buildid_cache = true; | |
2786 | disable_buildid_cache(); | |
2787 | } | |
2788 | } | |
655000e7 | 2789 | |
4ea648ae WN |
2790 | if (record.opts.overwrite) |
2791 | record.opts.tail_synthesize = true; | |
2792 | ||
6484d2f9 | 2793 | if (rec->evlist->core.nr_entries == 0 && |
e251abee | 2794 | __evlist__add_default(rec->evlist, !record.opts.no_samples) < 0) { |
69aad6f1 | 2795 | pr_err("Not enough memory for event selector list\n"); |
394c01ed | 2796 | goto out; |
bbd36e5e | 2797 | } |
0e9b20b8 | 2798 | |
69e7e5b0 AH |
2799 | if (rec->opts.target.tid && !rec->opts.no_inherit_set) |
2800 | rec->opts.no_inherit = true; | |
2801 | ||
602ad878 | 2802 | err = target__validate(&rec->opts.target); |
16ad2ffb | 2803 | if (err) { |
602ad878 | 2804 | target__strerror(&rec->opts.target, err, errbuf, BUFSIZ); |
c3dec27b | 2805 | ui__warning("%s\n", errbuf); |
16ad2ffb NK |
2806 | } |
2807 | ||
602ad878 | 2808 | err = target__parse_uid(&rec->opts.target); |
16ad2ffb NK |
2809 | if (err) { |
2810 | int saved_errno = errno; | |
4bd0f2d2 | 2811 | |
602ad878 | 2812 | target__strerror(&rec->opts.target, err, errbuf, BUFSIZ); |
3780f488 | 2813 | ui__error("%s", errbuf); |
16ad2ffb NK |
2814 | |
2815 | err = -saved_errno; | |
394c01ed | 2816 | goto out; |
16ad2ffb | 2817 | } |
0d37aa34 | 2818 | |
ca800068 MZ |
2819 | /* Enable ignoring missing threads when -u/-p option is defined. */ |
2820 | rec->opts.ignore_missing_thread = rec->opts.target.uid != UINT_MAX || rec->opts.target.pid; | |
23dc4f15 | 2821 | |
16ad2ffb | 2822 | err = -ENOMEM; |
7748bb71 | 2823 | if (evlist__create_maps(rec->evlist, &rec->opts.target) < 0) |
dd7927f4 | 2824 | usage_with_options(record_usage, record_options); |
69aad6f1 | 2825 | |
ef149c25 AH |
2826 | err = auxtrace_record__options(rec->itr, rec->evlist, &rec->opts); |
2827 | if (err) | |
394c01ed | 2828 | goto out; |
ef149c25 | 2829 | |
6156681b NK |
2830 | /* |
2831 | * We take all buildids when the file contains | |
2832 | * AUX area tracing data because we do not decode the | |
2833 | * trace because it would take too long. | |
2834 | */ | |
2835 | if (rec->opts.full_auxtrace) | |
2836 | rec->buildid_all = true; | |
2837 | ||
246eba8e AH |
2838 | if (rec->opts.text_poke) { |
2839 | err = record__config_text_poke(rec->evlist); | |
2840 | if (err) { | |
2841 | pr_err("record__config_text_poke failed, error %d\n", err); | |
2842 | goto out; | |
2843 | } | |
2844 | } | |
2845 | ||
b4006796 | 2846 | if (record_opts__config(&rec->opts)) { |
39d17dac | 2847 | err = -EINVAL; |
394c01ed | 2848 | goto out; |
7e4ff9e3 MG |
2849 | } |
2850 | ||
93f20c0f AB |
2851 | if (rec->opts.nr_cblocks > nr_cblocks_max) |
2852 | rec->opts.nr_cblocks = nr_cblocks_max; | |
5d7f4116 | 2853 | pr_debug("nr_cblocks: %d\n", rec->opts.nr_cblocks); |
d3d1af6f | 2854 | |
9d2ed645 | 2855 | pr_debug("affinity: %s\n", affinity_tags[rec->opts.affinity]); |
470530bb | 2856 | pr_debug("mmap flush: %d\n", rec->opts.mmap_flush); |
9d2ed645 | 2857 | |
51255a8a AB |
2858 | if (rec->opts.comp_level > comp_level_max) |
2859 | rec->opts.comp_level = comp_level_max; | |
2860 | pr_debug("comp level: %d\n", rec->opts.comp_level); | |
2861 | ||
d20deb64 | 2862 | err = __cmd_record(&record, argc, argv); |
394c01ed | 2863 | out: |
8384a260 | 2864 | bitmap_free(rec->affinity_mask.bits); |
c12995a5 | 2865 | evlist__delete(rec->evlist); |
d65a458b | 2866 | symbol__exit(); |
ef149c25 | 2867 | auxtrace_record__free(rec->itr); |
a8fcbd26 | 2868 | out_opts: |
ee7fe31e | 2869 | evlist__close_control(rec->opts.ctl_fd, rec->opts.ctl_fd_ack, &rec->opts.ctl_fd_close); |
39d17dac | 2870 | return err; |
0e9b20b8 | 2871 | } |
2dd6d8a1 AH |
2872 | |
2873 | static void snapshot_sig_handler(int sig __maybe_unused) | |
2874 | { | |
dc0c6127 JO |
2875 | struct record *rec = &record; |
2876 | ||
d20aff15 | 2877 | hit_auxtrace_snapshot_trigger(rec); |
3c1cb7e3 | 2878 | |
dc0c6127 | 2879 | if (switch_output_signal(rec)) |
3c1cb7e3 | 2880 | trigger_hit(&switch_output_trigger); |
2dd6d8a1 | 2881 | } |
bfacbe3b JO |
2882 | |
2883 | static void alarm_sig_handler(int sig __maybe_unused) | |
2884 | { | |
2885 | struct record *rec = &record; | |
2886 | ||
2887 | if (switch_output_time(rec)) | |
2888 | trigger_hit(&switch_output_trigger); | |
2889 | } |