Commit | Line | Data |
---|---|---|
abaff32a | 1 | /* |
bf9e1876 IM |
2 | * builtin-record.c |
3 | * | |
4 | * Builtin record command: Record the profile of a workload | |
5 | * (or a CPU, or a PID) into the perf.data output file - for | |
6 | * later analysis via perf report. | |
abaff32a | 7 | */ |
b8f46c5a XG |
8 | #define _FILE_OFFSET_BITS 64 |
9 | ||
16f762a2 | 10 | #include "builtin.h" |
bf9e1876 IM |
11 | |
12 | #include "perf.h" | |
13 | ||
6122e4e4 | 14 | #include "util/build-id.h" |
6eda5838 | 15 | #include "util/util.h" |
0e9b20b8 | 16 | #include "util/parse-options.h" |
8ad8db37 | 17 | #include "util/parse-events.h" |
6eda5838 | 18 | |
7c6a1c65 | 19 | #include "util/header.h" |
66e274f3 | 20 | #include "util/event.h" |
69aad6f1 | 21 | #include "util/evsel.h" |
8f28827a | 22 | #include "util/debug.h" |
94c744b6 | 23 | #include "util/session.h" |
8d06367f | 24 | #include "util/symbol.h" |
a12b51c4 | 25 | #include "util/cpumap.h" |
7c6a1c65 | 26 | |
97124d5e | 27 | #include <unistd.h> |
de9ac07b | 28 | #include <sched.h> |
a41794cd | 29 | #include <sys/mman.h> |
de9ac07b | 30 | |
69aad6f1 ACM |
31 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
32 | ||
7865e817 FW |
33 | enum write_mode_t { |
34 | WRITE_FORCE, | |
35 | WRITE_APPEND | |
36 | }; | |
37 | ||
3de29cab SE |
38 | static u64 user_interval = ULLONG_MAX; |
39 | static u64 default_interval = 0; | |
640c03ce | 40 | static u64 sample_type; |
a21ca2ca | 41 | |
60d567e2 | 42 | static struct cpu_map *cpus; |
de9ac07b | 43 | static unsigned int page_size; |
42e59d7d | 44 | static unsigned int mmap_pages = 128; |
f9212819 | 45 | static unsigned int user_freq = UINT_MAX; |
42e59d7d | 46 | static int freq = 1000; |
de9ac07b | 47 | static int output; |
529870e3 | 48 | static int pipe_output = 0; |
23ac9cbe | 49 | static const char *output_name = "perf.data"; |
42e59d7d | 50 | static int group = 0; |
1967936d | 51 | static int realtime_prio = 0; |
acac03fa | 52 | static bool nodelay = false; |
c0555642 | 53 | static bool raw_samples = false; |
9c90a61c | 54 | static bool sample_id_all_avail = true; |
c0555642 | 55 | static bool system_wide = false; |
42e59d7d | 56 | static pid_t target_pid = -1; |
d6d901c2 | 57 | static pid_t target_tid = -1; |
5c98d466 | 58 | static struct thread_map *threads; |
42e59d7d | 59 | static pid_t child_pid = -1; |
2e6cdf99 | 60 | static bool no_inherit = false; |
7865e817 | 61 | static enum write_mode_t write_mode = WRITE_FORCE; |
c0555642 IM |
62 | static bool call_graph = false; |
63 | static bool inherit_stat = false; | |
64 | static bool no_samples = false; | |
65 | static bool sample_address = false; | |
9c90a61c | 66 | static bool sample_time = false; |
a1ac1d3c | 67 | static bool no_buildid = false; |
baa2f6ce | 68 | static bool no_buildid_cache = false; |
42e59d7d IM |
69 | |
70 | static long samples = 0; | |
42e59d7d | 71 | static u64 bytes_written = 0; |
a21ca2ca | 72 | |
d6d901c2 | 73 | static struct pollfd *event_array; |
a21ca2ca | 74 | |
42e59d7d IM |
75 | static int nr_poll = 0; |
76 | static int nr_cpu = 0; | |
a21ca2ca | 77 | |
42e59d7d | 78 | static int file_new = 1; |
6122e4e4 | 79 | static off_t post_processing_offset; |
7c6a1c65 | 80 | |
94c744b6 | 81 | static struct perf_session *session; |
c45c6ea2 | 82 | static const char *cpu_list; |
f5970550 | 83 | |
de9ac07b | 84 | struct mmap_data { |
a21ca2ca IM |
85 | void *base; |
86 | unsigned int mask; | |
87 | unsigned int prev; | |
de9ac07b PZ |
88 | }; |
89 | ||
0e2e63dd | 90 | static struct mmap_data mmap_array[MAX_NR_CPUS]; |
a21ca2ca | 91 | |
9d91a6f7 | 92 | static unsigned long mmap_read_head(struct mmap_data *md) |
de9ac07b | 93 | { |
cdd6c482 | 94 | struct perf_event_mmap_page *pc = md->base; |
9d91a6f7 | 95 | long head; |
de9ac07b PZ |
96 | |
97 | head = pc->data_head; | |
98 | rmb(); | |
99 | ||
100 | return head; | |
101 | } | |
102 | ||
9d91a6f7 PZ |
103 | static void mmap_write_tail(struct mmap_data *md, unsigned long tail) |
104 | { | |
cdd6c482 | 105 | struct perf_event_mmap_page *pc = md->base; |
9d91a6f7 PZ |
106 | |
107 | /* | |
108 | * ensure all reads are done before we write the tail out. | |
109 | */ | |
110 | /* mb(); */ | |
111 | pc->data_tail = tail; | |
112 | } | |
113 | ||
9215545e TZ |
114 | static void advance_output(size_t size) |
115 | { | |
116 | bytes_written += size; | |
117 | } | |
118 | ||
f5970550 PZ |
119 | static void write_output(void *buf, size_t size) |
120 | { | |
121 | while (size) { | |
122 | int ret = write(output, buf, size); | |
123 | ||
124 | if (ret < 0) | |
125 | die("failed to write"); | |
126 | ||
127 | size -= ret; | |
128 | buf += ret; | |
129 | ||
130 | bytes_written += ret; | |
131 | } | |
132 | } | |
133 | ||
d8f66248 | 134 | static int process_synthesized_event(event_t *event, |
640c03ce | 135 | struct sample_data *sample __used, |
d8f66248 | 136 | struct perf_session *self __used) |
234fbbf5 | 137 | { |
6122e4e4 | 138 | write_output(event, event->header.size); |
234fbbf5 ACM |
139 | return 0; |
140 | } | |
141 | ||
de9ac07b PZ |
142 | static void mmap_read(struct mmap_data *md) |
143 | { | |
144 | unsigned int head = mmap_read_head(md); | |
145 | unsigned int old = md->prev; | |
146 | unsigned char *data = md->base + page_size; | |
147 | unsigned long size; | |
148 | void *buf; | |
149 | int diff; | |
150 | ||
de9ac07b PZ |
151 | /* |
152 | * If we're further behind than half the buffer, there's a chance | |
2debbc83 | 153 | * the writer will bite our tail and mess up the samples under us. |
de9ac07b PZ |
154 | * |
155 | * If we somehow ended up ahead of the head, we got messed up. | |
156 | * | |
157 | * In either case, truncate and restart at head. | |
158 | */ | |
159 | diff = head - old; | |
9d91a6f7 | 160 | if (diff < 0) { |
ef365cef | 161 | fprintf(stderr, "WARNING: failed to keep up with mmap data\n"); |
de9ac07b PZ |
162 | /* |
163 | * head points to a known good entry, start there. | |
164 | */ | |
165 | old = head; | |
166 | } | |
167 | ||
de9ac07b | 168 | if (old != head) |
2debbc83 | 169 | samples++; |
de9ac07b PZ |
170 | |
171 | size = head - old; | |
172 | ||
173 | if ((old & md->mask) + size != (head & md->mask)) { | |
174 | buf = &data[old & md->mask]; | |
175 | size = md->mask + 1 - (old & md->mask); | |
176 | old += size; | |
021e9f47 | 177 | |
6122e4e4 | 178 | write_output(buf, size); |
de9ac07b PZ |
179 | } |
180 | ||
181 | buf = &data[old & md->mask]; | |
182 | size = head - old; | |
183 | old += size; | |
021e9f47 | 184 | |
6122e4e4 | 185 | write_output(buf, size); |
de9ac07b PZ |
186 | |
187 | md->prev = old; | |
9d91a6f7 | 188 | mmap_write_tail(md, old); |
de9ac07b PZ |
189 | } |
190 | ||
191 | static volatile int done = 0; | |
f7b7c26e | 192 | static volatile int signr = -1; |
de9ac07b | 193 | |
16c8a109 | 194 | static void sig_handler(int sig) |
de9ac07b | 195 | { |
16c8a109 | 196 | done = 1; |
f7b7c26e PZ |
197 | signr = sig; |
198 | } | |
199 | ||
200 | static void sig_atexit(void) | |
201 | { | |
5ffc8881 | 202 | if (child_pid > 0) |
933da83a CW |
203 | kill(child_pid, SIGTERM); |
204 | ||
18483b81 | 205 | if (signr == -1 || signr == SIGUSR1) |
f7b7c26e PZ |
206 | return; |
207 | ||
208 | signal(signr, SIG_DFL); | |
209 | kill(getpid(), signr); | |
de9ac07b PZ |
210 | } |
211 | ||
f250c030 IM |
212 | static int group_fd; |
213 | ||
cdd6c482 | 214 | static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr) |
7c6a1c65 PZ |
215 | { |
216 | struct perf_header_attr *h_attr; | |
217 | ||
94c744b6 ACM |
218 | if (nr < session->header.attrs) { |
219 | h_attr = session->header.attr[nr]; | |
7c6a1c65 PZ |
220 | } else { |
221 | h_attr = perf_header_attr__new(a); | |
dc79c0fc | 222 | if (h_attr != NULL) |
94c744b6 | 223 | if (perf_header__add_attr(&session->header, h_attr) < 0) { |
11deb1f9 ACM |
224 | perf_header_attr__delete(h_attr); |
225 | h_attr = NULL; | |
226 | } | |
7c6a1c65 PZ |
227 | } |
228 | ||
229 | return h_attr; | |
230 | } | |
231 | ||
69aad6f1 | 232 | static void create_counter(struct perf_evsel *evsel, int cpu) |
de9ac07b | 233 | { |
69aad6f1 ACM |
234 | char *filter = evsel->filter; |
235 | struct perf_event_attr *attr = &evsel->attr; | |
7c6a1c65 | 236 | struct perf_header_attr *h_attr; |
69aad6f1 | 237 | int track = !evsel->idx; /* only the first counter needs these */ |
d6d901c2 | 238 | int thread_index; |
c171b552 | 239 | int ret; |
7c6a1c65 PZ |
240 | struct { |
241 | u64 count; | |
242 | u64 time_enabled; | |
243 | u64 time_running; | |
244 | u64 id; | |
245 | } read_data; | |
a43d3f08 ACM |
246 | /* |
247 | * Check if parse_single_tracepoint_event has already asked for | |
248 | * PERF_SAMPLE_TIME. | |
249 | * | |
250 | * XXX this is kludgy but short term fix for problems introduced by | |
251 | * eac23d1c that broke 'perf script' by having different sample_types | |
252 | * when using multiple tracepoint events when we use a perf binary | |
253 | * that tries to use sample_id_all on an older kernel. | |
254 | * | |
255 | * We need to move counter creation to perf_session, support | |
256 | * different sample_types, etc. | |
257 | */ | |
258 | bool time_needed = attr->sample_type & PERF_SAMPLE_TIME; | |
7c6a1c65 PZ |
259 | |
260 | attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | | |
261 | PERF_FORMAT_TOTAL_TIME_RUNNING | | |
262 | PERF_FORMAT_ID; | |
16c8a109 | 263 | |
3a9f131f | 264 | attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID; |
3efa1cc9 | 265 | |
8907fd60 EM |
266 | if (nr_counters > 1) |
267 | attr->sample_type |= PERF_SAMPLE_ID; | |
268 | ||
f9212819 FW |
269 | /* |
270 | * We default some events to a 1 default interval. But keep | |
271 | * it a weak assumption overridable by the user. | |
272 | */ | |
273 | if (!attr->sample_period || (user_freq != UINT_MAX && | |
3de29cab | 274 | user_interval != ULLONG_MAX)) { |
f9212819 FW |
275 | if (freq) { |
276 | attr->sample_type |= PERF_SAMPLE_PERIOD; | |
277 | attr->freq = 1; | |
278 | attr->sample_freq = freq; | |
279 | } else { | |
280 | attr->sample_period = default_interval; | |
281 | } | |
1dba15e7 | 282 | } |
3efa1cc9 | 283 | |
649c48a9 PZ |
284 | if (no_samples) |
285 | attr->sample_freq = 0; | |
286 | ||
287 | if (inherit_stat) | |
288 | attr->inherit_stat = 1; | |
289 | ||
3af9e859 | 290 | if (sample_address) { |
4bba828d | 291 | attr->sample_type |= PERF_SAMPLE_ADDR; |
3af9e859 EM |
292 | attr->mmap_data = track; |
293 | } | |
4bba828d | 294 | |
3efa1cc9 IM |
295 | if (call_graph) |
296 | attr->sample_type |= PERF_SAMPLE_CALLCHAIN; | |
297 | ||
f60f3593 AS |
298 | if (system_wide) |
299 | attr->sample_type |= PERF_SAMPLE_CPU; | |
300 | ||
a43d3f08 ACM |
301 | if (sample_id_all_avail && |
302 | (sample_time || system_wide || !no_inherit || cpu_list)) | |
9c90a61c ACM |
303 | attr->sample_type |= PERF_SAMPLE_TIME; |
304 | ||
cd6feeea | 305 | if (raw_samples) { |
6ddf259d | 306 | attr->sample_type |= PERF_SAMPLE_TIME; |
daac07b2 | 307 | attr->sample_type |= PERF_SAMPLE_RAW; |
cd6feeea IM |
308 | attr->sample_type |= PERF_SAMPLE_CPU; |
309 | } | |
f413cdb8 | 310 | |
acac03fa KS |
311 | if (nodelay) { |
312 | attr->watermark = 0; | |
313 | attr->wakeup_events = 1; | |
314 | } | |
315 | ||
a21ca2ca IM |
316 | attr->mmap = track; |
317 | attr->comm = track; | |
2e6cdf99 SE |
318 | attr->inherit = !no_inherit; |
319 | if (target_pid == -1 && target_tid == -1 && !system_wide) { | |
46be604b | 320 | attr->disabled = 1; |
bedbfdea | 321 | attr->enable_on_exec = 1; |
46be604b | 322 | } |
9c90a61c ACM |
323 | retry_sample_id: |
324 | attr->sample_id_all = sample_id_all_avail ? 1 : 0; | |
bedbfdea | 325 | |
5c98d466 | 326 | for (thread_index = 0; thread_index < threads->nr; thread_index++) { |
3da297a6 | 327 | try_again: |
5c98d466 | 328 | FD(evsel, nr_cpu, thread_index) = sys_perf_event_open(attr, threads->map[thread_index], cpu, group_fd, 0); |
d6d901c2 | 329 | |
69aad6f1 | 330 | if (FD(evsel, nr_cpu, thread_index) < 0) { |
d6d901c2 ZY |
331 | int err = errno; |
332 | ||
333 | if (err == EPERM || err == EACCES) | |
334 | die("Permission error - are you root?\n" | |
335 | "\t Consider tweaking" | |
336 | " /proc/sys/kernel/perf_event_paranoid.\n"); | |
c45c6ea2 | 337 | else if (err == ENODEV && cpu_list) { |
d6d901c2 ZY |
338 | die("No such device - did you specify" |
339 | " an out-of-range profile CPU?\n"); | |
9c90a61c ACM |
340 | } else if (err == EINVAL && sample_id_all_avail) { |
341 | /* | |
342 | * Old kernel, no attr->sample_id_type_all field | |
343 | */ | |
344 | sample_id_all_avail = false; | |
a43d3f08 | 345 | if (!sample_time && !raw_samples && !time_needed) |
eac23d1c IM |
346 | attr->sample_type &= ~PERF_SAMPLE_TIME; |
347 | ||
9c90a61c | 348 | goto retry_sample_id; |
d6d901c2 | 349 | } |
3da297a6 | 350 | |
d6d901c2 ZY |
351 | /* |
352 | * If it's cycles then fall back to hrtimer | |
353 | * based cpu-clock-tick sw counter, which | |
354 | * is always available even if no PMU support: | |
355 | */ | |
356 | if (attr->type == PERF_TYPE_HARDWARE | |
357 | && attr->config == PERF_COUNT_HW_CPU_CYCLES) { | |
358 | ||
359 | if (verbose) | |
360 | warning(" ... trying to fall back to cpu-clock-ticks\n"); | |
361 | attr->type = PERF_TYPE_SOFTWARE; | |
362 | attr->config = PERF_COUNT_SW_CPU_CLOCK; | |
363 | goto try_again; | |
364 | } | |
365 | printf("\n"); | |
d9cf837e | 366 | error("sys_perf_event_open() syscall returned with %d (%s). /bin/dmesg may provide additional information.\n", |
69aad6f1 | 367 | FD(evsel, nr_cpu, thread_index), strerror(err)); |
bfd45118 SK |
368 | |
369 | #if defined(__i386__) || defined(__x86_64__) | |
d6d901c2 ZY |
370 | if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) |
371 | die("No hardware sampling interrupt available." | |
372 | " No APIC? If so then you can boot the kernel" | |
373 | " with the \"lapic\" boot parameter to" | |
374 | " force-enable it.\n"); | |
bfd45118 SK |
375 | #endif |
376 | ||
d6d901c2 ZY |
377 | die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); |
378 | exit(-1); | |
379 | } | |
3da297a6 | 380 | |
69aad6f1 | 381 | h_attr = get_header_attr(attr, evsel->idx); |
d6d901c2 ZY |
382 | if (h_attr == NULL) |
383 | die("nomem\n"); | |
7c6a1c65 | 384 | |
d6d901c2 ZY |
385 | if (!file_new) { |
386 | if (memcmp(&h_attr->attr, attr, sizeof(*attr))) { | |
387 | fprintf(stderr, "incompatible append\n"); | |
388 | exit(-1); | |
389 | } | |
7c6a1c65 | 390 | } |
7c6a1c65 | 391 | |
69aad6f1 | 392 | if (read(FD(evsel, nr_cpu, thread_index), &read_data, sizeof(read_data)) == -1) { |
0ab7368f | 393 | perror("Unable to read perf file descriptor"); |
d6d901c2 ZY |
394 | exit(-1); |
395 | } | |
7c6a1c65 | 396 | |
d6d901c2 ZY |
397 | if (perf_header_attr__add_id(h_attr, read_data.id) < 0) { |
398 | pr_warning("Not enough memory to add id\n"); | |
399 | exit(-1); | |
400 | } | |
7c6a1c65 | 401 | |
69aad6f1 ACM |
402 | assert(FD(evsel, nr_cpu, thread_index) >= 0); |
403 | fcntl(FD(evsel, nr_cpu, thread_index), F_SETFL, O_NONBLOCK); | |
16c8a109 | 404 | |
d6d901c2 ZY |
405 | /* |
406 | * First counter acts as the group leader: | |
407 | */ | |
408 | if (group && group_fd == -1) | |
69aad6f1 ACM |
409 | group_fd = FD(evsel, nr_cpu, thread_index); |
410 | ||
411 | if (evsel->idx || thread_index) { | |
412 | struct perf_evsel *first; | |
413 | first = list_entry(evsel_list.next, struct perf_evsel, node); | |
414 | ret = ioctl(FD(evsel, nr_cpu, thread_index), | |
415 | PERF_EVENT_IOC_SET_OUTPUT, | |
416 | FD(first, nr_cpu, 0)); | |
0e2e63dd PZ |
417 | if (ret) { |
418 | error("failed to set output: %d (%s)\n", errno, | |
419 | strerror(errno)); | |
420 | exit(-1); | |
421 | } | |
422 | } else { | |
0e2e63dd PZ |
423 | mmap_array[nr_cpu].prev = 0; |
424 | mmap_array[nr_cpu].mask = mmap_pages*page_size - 1; | |
425 | mmap_array[nr_cpu].base = mmap(NULL, (mmap_pages+1)*page_size, | |
69aad6f1 | 426 | PROT_READ | PROT_WRITE, MAP_SHARED, FD(evsel, nr_cpu, thread_index), 0); |
0e2e63dd PZ |
427 | if (mmap_array[nr_cpu].base == MAP_FAILED) { |
428 | error("failed to mmap with %d (%s)\n", errno, strerror(errno)); | |
429 | exit(-1); | |
430 | } | |
431 | ||
69aad6f1 | 432 | event_array[nr_poll].fd = FD(evsel, nr_cpu, thread_index); |
0e2e63dd PZ |
433 | event_array[nr_poll].events = POLLIN; |
434 | nr_poll++; | |
ea57c4f5 | 435 | } |
d1302522 | 436 | |
d6d901c2 | 437 | if (filter != NULL) { |
69aad6f1 ACM |
438 | ret = ioctl(FD(evsel, nr_cpu, thread_index), |
439 | PERF_EVENT_IOC_SET_FILTER, filter); | |
d6d901c2 ZY |
440 | if (ret) { |
441 | error("failed to set filter with %d (%s)\n", errno, | |
442 | strerror(errno)); | |
443 | exit(-1); | |
444 | } | |
c171b552 LZ |
445 | } |
446 | } | |
a43d3f08 ACM |
447 | |
448 | if (!sample_type) | |
449 | sample_type = attr->sample_type; | |
f250c030 | 450 | } |
f2521b6e | 451 | |
d6d901c2 | 452 | static void open_counters(int cpu) |
f250c030 | 453 | { |
69aad6f1 | 454 | struct perf_evsel *pos; |
16c8a109 | 455 | |
f250c030 | 456 | group_fd = -1; |
69aad6f1 ACM |
457 | |
458 | list_for_each_entry(pos, &evsel_list, node) | |
459 | create_counter(pos, cpu); | |
f250c030 | 460 | |
16c8a109 PZ |
461 | nr_cpu++; |
462 | } | |
463 | ||
6122e4e4 ACM |
464 | static int process_buildids(void) |
465 | { | |
466 | u64 size = lseek(output, 0, SEEK_CUR); | |
467 | ||
9f591fd7 ACM |
468 | if (size == 0) |
469 | return 0; | |
470 | ||
6122e4e4 ACM |
471 | session->fd = output; |
472 | return __perf_session__process_events(session, post_processing_offset, | |
473 | size - post_processing_offset, | |
474 | size, &build_id__mark_dso_hit_ops); | |
475 | } | |
476 | ||
f5970550 PZ |
477 | static void atexit_header(void) |
478 | { | |
c7929e47 TZ |
479 | if (!pipe_output) { |
480 | session->header.data_size += bytes_written; | |
f5970550 | 481 | |
baa2f6ce ACM |
482 | if (!no_buildid) |
483 | process_buildids(); | |
c7929e47 | 484 | perf_header__write(&session->header, output, true); |
39d17dac | 485 | perf_session__delete(session); |
bd3bfe9e | 486 | perf_evsel_list__delete(); |
d65a458b | 487 | symbol__exit(); |
c7929e47 | 488 | } |
f5970550 PZ |
489 | } |
490 | ||
23346f21 | 491 | static void event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
492 | { |
493 | int err; | |
23346f21 | 494 | struct perf_session *psession = data; |
a1645ce1 | 495 | |
23346f21 | 496 | if (machine__is_host(machine)) |
a1645ce1 ZY |
497 | return; |
498 | ||
499 | /* | |
500 | *As for guest kernel when processing subcommand record&report, | |
501 | *we arrange module mmap prior to guest kernel mmap and trigger | |
502 | *a preload dso because default guest module symbols are loaded | |
503 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
504 | *method is used to avoid symbol missing when the first addr is | |
505 | *in module instead of in guest kernel. | |
506 | */ | |
507 | err = event__synthesize_modules(process_synthesized_event, | |
23346f21 | 508 | psession, machine); |
a1645ce1 ZY |
509 | if (err < 0) |
510 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 511 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 512 | |
a1645ce1 ZY |
513 | /* |
514 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
515 | * have no _text sometimes. | |
516 | */ | |
517 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 518 | psession, machine, "_text"); |
a1645ce1 ZY |
519 | if (err < 0) |
520 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 521 | psession, machine, "_stext"); |
a1645ce1 ZY |
522 | if (err < 0) |
523 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 524 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
525 | } |
526 | ||
98402807 FW |
527 | static struct perf_event_header finished_round_event = { |
528 | .size = sizeof(struct perf_event_header), | |
529 | .type = PERF_RECORD_FINISHED_ROUND, | |
530 | }; | |
531 | ||
532 | static void mmap_read_all(void) | |
533 | { | |
0e2e63dd | 534 | int i; |
98402807 FW |
535 | |
536 | for (i = 0; i < nr_cpu; i++) { | |
0e2e63dd PZ |
537 | if (mmap_array[i].base) |
538 | mmap_read(&mmap_array[i]); | |
98402807 FW |
539 | } |
540 | ||
541 | if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO)) | |
542 | write_output(&finished_round_event, sizeof(finished_round_event)); | |
543 | } | |
544 | ||
d4db3f16 | 545 | static int __cmd_record(int argc, const char **argv) |
16c8a109 | 546 | { |
69aad6f1 | 547 | int i; |
abaff32a | 548 | struct stat st; |
abaff32a | 549 | int flags; |
4dc0a04b | 550 | int err; |
8b412664 | 551 | unsigned long waking = 0; |
856e9660 | 552 | int child_ready_pipe[2], go_pipe[2]; |
46be604b | 553 | const bool forks = argc > 0; |
856e9660 | 554 | char buf; |
23346f21 | 555 | struct machine *machine; |
de9ac07b PZ |
556 | |
557 | page_size = sysconf(_SC_PAGE_SIZE); | |
de9ac07b | 558 | |
f5970550 PZ |
559 | atexit(sig_atexit); |
560 | signal(SIGCHLD, sig_handler); | |
561 | signal(SIGINT, sig_handler); | |
18483b81 | 562 | signal(SIGUSR1, sig_handler); |
f5970550 | 563 | |
d4db3f16 | 564 | if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { |
856e9660 PZ |
565 | perror("failed to create pipes"); |
566 | exit(-1); | |
567 | } | |
568 | ||
529870e3 TZ |
569 | if (!strcmp(output_name, "-")) |
570 | pipe_output = 1; | |
571 | else if (!stat(output_name, &st) && st.st_size) { | |
7865e817 | 572 | if (write_mode == WRITE_FORCE) { |
b38d3464 ACM |
573 | char oldname[PATH_MAX]; |
574 | snprintf(oldname, sizeof(oldname), "%s.old", | |
575 | output_name); | |
576 | unlink(oldname); | |
577 | rename(output_name, oldname); | |
266e0e21 | 578 | } |
7865e817 FW |
579 | } else if (write_mode == WRITE_APPEND) { |
580 | write_mode = WRITE_FORCE; | |
97124d5e PZ |
581 | } |
582 | ||
f887f301 | 583 | flags = O_CREAT|O_RDWR; |
7865e817 | 584 | if (write_mode == WRITE_APPEND) |
f5970550 | 585 | file_new = 0; |
abaff32a IM |
586 | else |
587 | flags |= O_TRUNC; | |
588 | ||
529870e3 TZ |
589 | if (pipe_output) |
590 | output = STDOUT_FILENO; | |
591 | else | |
592 | output = open(output_name, flags, S_IRUSR | S_IWUSR); | |
de9ac07b PZ |
593 | if (output < 0) { |
594 | perror("failed to create output file"); | |
595 | exit(-1); | |
596 | } | |
597 | ||
7865e817 | 598 | session = perf_session__new(output_name, O_WRONLY, |
21ef97f0 | 599 | write_mode == WRITE_FORCE, false, NULL); |
94c744b6 | 600 | if (session == NULL) { |
a9a70bbc ACM |
601 | pr_err("Not enough memory for reading perf file header\n"); |
602 | return -1; | |
603 | } | |
604 | ||
baa2f6ce ACM |
605 | if (!no_buildid) |
606 | perf_header__set_feat(&session->header, HEADER_BUILD_ID); | |
607 | ||
4dc0a04b | 608 | if (!file_new) { |
8dc58101 | 609 | err = perf_header__read(session, output); |
4dc0a04b | 610 | if (err < 0) |
39d17dac | 611 | goto out_delete_session; |
4dc0a04b ACM |
612 | } |
613 | ||
69aad6f1 | 614 | if (have_tracepoints(&evsel_list)) |
94c744b6 | 615 | perf_header__set_feat(&session->header, HEADER_TRACE_INFO); |
03456a15 | 616 | |
39d17dac ACM |
617 | /* |
618 | * perf_session__delete(session) will be called at atexit_header() | |
619 | */ | |
f5970550 PZ |
620 | atexit(atexit_header); |
621 | ||
d4db3f16 | 622 | if (forks) { |
46be604b | 623 | child_pid = fork(); |
2fb750e8 | 624 | if (child_pid < 0) { |
856e9660 PZ |
625 | perror("failed to fork"); |
626 | exit(-1); | |
627 | } | |
7c6a1c65 | 628 | |
46be604b | 629 | if (!child_pid) { |
529870e3 TZ |
630 | if (pipe_output) |
631 | dup2(2, 1); | |
856e9660 PZ |
632 | close(child_ready_pipe[0]); |
633 | close(go_pipe[1]); | |
634 | fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC); | |
635 | ||
636 | /* | |
637 | * Do a dummy execvp to get the PLT entry resolved, | |
638 | * so we avoid the resolver overhead on the real | |
639 | * execvp call. | |
640 | */ | |
641 | execvp("", (char **)argv); | |
642 | ||
643 | /* | |
644 | * Tell the parent we're ready to go | |
645 | */ | |
646 | close(child_ready_pipe[1]); | |
647 | ||
648 | /* | |
649 | * Wait until the parent tells us to go. | |
650 | */ | |
651 | if (read(go_pipe[0], &buf, 1) == -1) | |
652 | perror("unable to read pipe"); | |
653 | ||
654 | execvp(argv[0], (char **)argv); | |
655 | ||
656 | perror(argv[0]); | |
18483b81 | 657 | kill(getppid(), SIGUSR1); |
856e9660 | 658 | exit(-1); |
0a5ac846 | 659 | } |
856e9660 | 660 | |
d6d901c2 | 661 | if (!system_wide && target_tid == -1 && target_pid == -1) |
5c98d466 | 662 | threads->map[0] = child_pid; |
d6d901c2 | 663 | |
856e9660 PZ |
664 | close(child_ready_pipe[1]); |
665 | close(go_pipe[0]); | |
666 | /* | |
667 | * wait for child to settle | |
668 | */ | |
669 | if (read(child_ready_pipe[0], &buf, 1) == -1) { | |
670 | perror("unable to read pipe"); | |
671 | exit(-1); | |
672 | } | |
673 | close(child_ready_pipe[0]); | |
674 | } | |
675 | ||
c45c6ea2 SE |
676 | if (!system_wide && no_inherit && !cpu_list) { |
677 | open_counters(-1); | |
856e9660 | 678 | } else { |
60d567e2 ACM |
679 | for (i = 0; i < cpus->nr; i++) |
680 | open_counters(cpus->map[i]); | |
0a5ac846 | 681 | } |
de9ac07b | 682 | |
640c03ce ACM |
683 | perf_session__set_sample_type(session, sample_type); |
684 | ||
529870e3 TZ |
685 | if (pipe_output) { |
686 | err = perf_header__write_pipe(output); | |
687 | if (err < 0) | |
688 | return err; | |
689 | } else if (file_new) { | |
94c744b6 | 690 | err = perf_header__write(&session->header, output, false); |
d5eed904 ACM |
691 | if (err < 0) |
692 | return err; | |
56b03f3c ACM |
693 | } |
694 | ||
6122e4e4 ACM |
695 | post_processing_offset = lseek(output, 0, SEEK_CUR); |
696 | ||
9c90a61c ACM |
697 | perf_session__set_sample_id_all(session, sample_id_all_avail); |
698 | ||
2c46dbb5 TZ |
699 | if (pipe_output) { |
700 | err = event__synthesize_attrs(&session->header, | |
701 | process_synthesized_event, | |
702 | session); | |
703 | if (err < 0) { | |
704 | pr_err("Couldn't synthesize attrs.\n"); | |
705 | return err; | |
706 | } | |
cd19a035 TZ |
707 | |
708 | err = event__synthesize_event_types(process_synthesized_event, | |
709 | session); | |
710 | if (err < 0) { | |
711 | pr_err("Couldn't synthesize event_types.\n"); | |
712 | return err; | |
713 | } | |
9215545e | 714 | |
69aad6f1 | 715 | if (have_tracepoints(&evsel_list)) { |
63e0c771 TZ |
716 | /* |
717 | * FIXME err <= 0 here actually means that | |
718 | * there were no tracepoints so its not really | |
719 | * an error, just that we don't need to | |
720 | * synthesize anything. We really have to | |
721 | * return this more properly and also | |
722 | * propagate errors that now are calling die() | |
723 | */ | |
69aad6f1 | 724 | err = event__synthesize_tracing_data(output, &evsel_list, |
63e0c771 TZ |
725 | process_synthesized_event, |
726 | session); | |
727 | if (err <= 0) { | |
728 | pr_err("Couldn't record tracing data.\n"); | |
729 | return err; | |
730 | } | |
2c9faa06 | 731 | advance_output(err); |
63e0c771 | 732 | } |
2c46dbb5 TZ |
733 | } |
734 | ||
23346f21 ACM |
735 | machine = perf_session__find_host_machine(session); |
736 | if (!machine) { | |
a1645ce1 ZY |
737 | pr_err("Couldn't find native kernel information.\n"); |
738 | return -1; | |
739 | } | |
740 | ||
56b03f3c | 741 | err = event__synthesize_kernel_mmap(process_synthesized_event, |
23346f21 | 742 | session, machine, "_text"); |
70162138 ACM |
743 | if (err < 0) |
744 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 745 | session, machine, "_stext"); |
c1a3a4b9 ACM |
746 | if (err < 0) |
747 | pr_err("Couldn't record kernel reference relocation symbol\n" | |
748 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
749 | "Check /proc/kallsyms permission or run as root.\n"); | |
b7cece76 | 750 | |
a1645ce1 | 751 | err = event__synthesize_modules(process_synthesized_event, |
23346f21 | 752 | session, machine); |
c1a3a4b9 ACM |
753 | if (err < 0) |
754 | pr_err("Couldn't record kernel module information.\n" | |
755 | "Symbol resolution may be skewed if relocation was used (e.g. kexec).\n" | |
756 | "Check /proc/modules permission or run as root.\n"); | |
757 | ||
a1645ce1 | 758 | if (perf_guest) |
23346f21 | 759 | perf_session__process_machines(session, event__synthesize_guest_os); |
7c6a1c65 | 760 | |
cf103a14 | 761 | if (!system_wide) |
d6d901c2 | 762 | event__synthesize_thread(target_tid, process_synthesized_event, |
d8f66248 | 763 | session); |
234fbbf5 | 764 | else |
d8f66248 | 765 | event__synthesize_threads(process_synthesized_event, session); |
7c6a1c65 | 766 | |
de9ac07b PZ |
767 | if (realtime_prio) { |
768 | struct sched_param param; | |
769 | ||
770 | param.sched_priority = realtime_prio; | |
771 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { | |
6beba7ad | 772 | pr_err("Could not set realtime priority.\n"); |
de9ac07b PZ |
773 | exit(-1); |
774 | } | |
775 | } | |
776 | ||
856e9660 PZ |
777 | /* |
778 | * Let the child rip | |
779 | */ | |
d4db3f16 ACM |
780 | if (forks) |
781 | close(go_pipe[1]); | |
856e9660 | 782 | |
649c48a9 | 783 | for (;;) { |
2debbc83 | 784 | int hits = samples; |
d6d901c2 | 785 | int thread; |
de9ac07b | 786 | |
98402807 | 787 | mmap_read_all(); |
de9ac07b | 788 | |
649c48a9 PZ |
789 | if (hits == samples) { |
790 | if (done) | |
791 | break; | |
4dc0a04b | 792 | err = poll(event_array, nr_poll, -1); |
8b412664 PZ |
793 | waking++; |
794 | } | |
795 | ||
796 | if (done) { | |
797 | for (i = 0; i < nr_cpu; i++) { | |
69aad6f1 ACM |
798 | struct perf_evsel *pos; |
799 | ||
800 | list_for_each_entry(pos, &evsel_list, node) { | |
d6d901c2 | 801 | for (thread = 0; |
5c98d466 | 802 | thread < threads->nr; |
d6d901c2 | 803 | thread++) |
69aad6f1 | 804 | ioctl(FD(pos, i, thread), |
d6d901c2 ZY |
805 | PERF_EVENT_IOC_DISABLE); |
806 | } | |
8b412664 | 807 | } |
649c48a9 | 808 | } |
de9ac07b PZ |
809 | } |
810 | ||
18483b81 | 811 | if (quiet || signr == SIGUSR1) |
b44308f5 ACM |
812 | return 0; |
813 | ||
8b412664 PZ |
814 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
815 | ||
021e9f47 IM |
816 | /* |
817 | * Approximate RIP event size: 24 bytes. | |
818 | */ | |
819 | fprintf(stderr, | |
9486aa38 | 820 | "[ perf record: Captured and wrote %.3f MB %s (~%" PRIu64 " samples) ]\n", |
021e9f47 IM |
821 | (double)bytes_written / 1024.0 / 1024.0, |
822 | output_name, | |
823 | bytes_written / 24); | |
addc2785 | 824 | |
de9ac07b | 825 | return 0; |
39d17dac ACM |
826 | |
827 | out_delete_session: | |
828 | perf_session__delete(session); | |
829 | return err; | |
de9ac07b | 830 | } |
0e9b20b8 | 831 | |
0e9b20b8 | 832 | static const char * const record_usage[] = { |
9e096753 MG |
833 | "perf record [<options>] [<command>]", |
834 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
835 | NULL |
836 | }; | |
837 | ||
7865e817 FW |
838 | static bool force, append_file; |
839 | ||
bca647aa | 840 | const struct option record_options[] = { |
0e9b20b8 | 841 | OPT_CALLBACK('e', "event", NULL, "event", |
86847b62 TG |
842 | "event selector. use 'perf list' to list available events", |
843 | parse_events), | |
c171b552 LZ |
844 | OPT_CALLBACK(0, "filter", NULL, "filter", |
845 | "event filter", parse_filter), | |
0e9b20b8 | 846 | OPT_INTEGER('p', "pid", &target_pid, |
d6d901c2 ZY |
847 | "record events on existing process id"), |
848 | OPT_INTEGER('t', "tid", &target_tid, | |
849 | "record events on existing thread id"), | |
0e9b20b8 IM |
850 | OPT_INTEGER('r', "realtime", &realtime_prio, |
851 | "collect data with this RT SCHED_FIFO priority"), | |
acac03fa KS |
852 | OPT_BOOLEAN('D', "no-delay", &nodelay, |
853 | "collect data without buffering"), | |
daac07b2 FW |
854 | OPT_BOOLEAN('R', "raw-samples", &raw_samples, |
855 | "collect raw sample records from all opened counters"), | |
0e9b20b8 IM |
856 | OPT_BOOLEAN('a', "all-cpus", &system_wide, |
857 | "system-wide collection from all CPUs"), | |
abaff32a IM |
858 | OPT_BOOLEAN('A', "append", &append_file, |
859 | "append to the output file to do incremental profiling"), | |
c45c6ea2 SE |
860 | OPT_STRING('C', "cpu", &cpu_list, "cpu", |
861 | "list of cpus to monitor"), | |
97124d5e | 862 | OPT_BOOLEAN('f', "force", &force, |
7865e817 | 863 | "overwrite existing data file (deprecated)"), |
3de29cab | 864 | OPT_U64('c', "count", &user_interval, "event period to sample"), |
abaff32a IM |
865 | OPT_STRING('o', "output", &output_name, "file", |
866 | "output file name"), | |
2e6cdf99 SE |
867 | OPT_BOOLEAN('i', "no-inherit", &no_inherit, |
868 | "child tasks do not inherit counters"), | |
1967936d ACM |
869 | OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"), |
870 | OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), | |
3efa1cc9 IM |
871 | OPT_BOOLEAN('g', "call-graph", &call_graph, |
872 | "do call-graph (stack chain/backtrace) recording"), | |
c0555642 | 873 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 874 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 875 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
649c48a9 PZ |
876 | OPT_BOOLEAN('s', "stat", &inherit_stat, |
877 | "per thread counts"), | |
4bba828d AB |
878 | OPT_BOOLEAN('d', "data", &sample_address, |
879 | "Sample addresses"), | |
9c90a61c | 880 | OPT_BOOLEAN('T', "timestamp", &sample_time, "Sample timestamps"), |
649c48a9 PZ |
881 | OPT_BOOLEAN('n', "no-samples", &no_samples, |
882 | "don't sample"), | |
baa2f6ce | 883 | OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid_cache, |
a1ac1d3c | 884 | "do not update the buildid cache"), |
baa2f6ce ACM |
885 | OPT_BOOLEAN('B', "no-buildid", &no_buildid, |
886 | "do not collect buildids in perf.data"), | |
0e9b20b8 IM |
887 | OPT_END() |
888 | }; | |
889 | ||
f37a291c | 890 | int cmd_record(int argc, const char **argv, const char *prefix __used) |
0e9b20b8 | 891 | { |
69aad6f1 ACM |
892 | int err = -ENOMEM; |
893 | struct perf_evsel *pos; | |
0e9b20b8 | 894 | |
bca647aa | 895 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 896 | PARSE_OPT_STOP_AT_NON_OPTION); |
d6d901c2 | 897 | if (!argc && target_pid == -1 && target_tid == -1 && |
c45c6ea2 | 898 | !system_wide && !cpu_list) |
bca647aa | 899 | usage_with_options(record_usage, record_options); |
0e9b20b8 | 900 | |
7865e817 FW |
901 | if (force && append_file) { |
902 | fprintf(stderr, "Can't overwrite and append at the same time." | |
903 | " You need to choose between -f and -A"); | |
bca647aa | 904 | usage_with_options(record_usage, record_options); |
7865e817 FW |
905 | } else if (append_file) { |
906 | write_mode = WRITE_APPEND; | |
907 | } else { | |
908 | write_mode = WRITE_FORCE; | |
909 | } | |
910 | ||
655000e7 | 911 | symbol__init(); |
baa2f6ce ACM |
912 | |
913 | if (no_buildid_cache || no_buildid) | |
a1ac1d3c | 914 | disable_buildid_cache(); |
655000e7 | 915 | |
69aad6f1 ACM |
916 | if (list_empty(&evsel_list) && perf_evsel_list__create_default() < 0) { |
917 | pr_err("Not enough memory for event selector list\n"); | |
918 | goto out_symbol_exit; | |
bbd36e5e | 919 | } |
0e9b20b8 | 920 | |
5c98d466 | 921 | if (target_pid != -1) |
d6d901c2 | 922 | target_tid = target_pid; |
d6d901c2 | 923 | |
5c98d466 ACM |
924 | threads = thread_map__new(target_pid, target_tid); |
925 | if (threads == NULL) { | |
926 | pr_err("Problems finding threads of monitor\n"); | |
927 | usage_with_options(record_usage, record_options); | |
d6d901c2 ZY |
928 | } |
929 | ||
60d567e2 ACM |
930 | cpus = cpu_map__new(cpu_list); |
931 | if (cpus == NULL) { | |
932 | perror("failed to parse CPUs map"); | |
69aad6f1 ACM |
933 | return -1; |
934 | } | |
935 | ||
936 | list_for_each_entry(pos, &evsel_list, node) { | |
5c98d466 | 937 | if (perf_evsel__alloc_fd(pos, cpus->nr, threads->nr) < 0) |
69aad6f1 | 938 | goto out_free_fd; |
ad7f4e3f ACM |
939 | if (perf_header__push_event(pos->attr.config, event_name(pos))) |
940 | goto out_free_fd; | |
d6d901c2 | 941 | } |
5c98d466 ACM |
942 | event_array = malloc((sizeof(struct pollfd) * MAX_NR_CPUS * |
943 | MAX_COUNTERS * threads->nr)); | |
d6d901c2 | 944 | if (!event_array) |
39d17dac | 945 | goto out_free_fd; |
d6d901c2 | 946 | |
3de29cab | 947 | if (user_interval != ULLONG_MAX) |
f9212819 FW |
948 | default_interval = user_interval; |
949 | if (user_freq != UINT_MAX) | |
950 | freq = user_freq; | |
951 | ||
7e4ff9e3 MG |
952 | /* |
953 | * User specified count overrides default frequency. | |
954 | */ | |
955 | if (default_interval) | |
956 | freq = 0; | |
957 | else if (freq) { | |
958 | default_interval = freq; | |
959 | } else { | |
960 | fprintf(stderr, "frequency and count are zero, aborting\n"); | |
39d17dac ACM |
961 | err = -EINVAL; |
962 | goto out_free_event_array; | |
7e4ff9e3 MG |
963 | } |
964 | ||
39d17dac ACM |
965 | err = __cmd_record(argc, argv); |
966 | ||
967 | out_free_event_array: | |
968 | free(event_array); | |
969 | out_free_fd: | |
5c98d466 ACM |
970 | thread_map__delete(threads); |
971 | threads = NULL; | |
d65a458b ACM |
972 | out_symbol_exit: |
973 | symbol__exit(); | |
39d17dac | 974 | return err; |
0e9b20b8 | 975 | } |