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" |
8f28827a | 21 | #include "util/debug.h" |
94c744b6 | 22 | #include "util/session.h" |
8d06367f | 23 | #include "util/symbol.h" |
a12b51c4 | 24 | #include "util/cpumap.h" |
7c6a1c65 | 25 | |
97124d5e | 26 | #include <unistd.h> |
de9ac07b | 27 | #include <sched.h> |
a41794cd | 28 | #include <sys/mman.h> |
de9ac07b | 29 | |
7865e817 FW |
30 | enum write_mode_t { |
31 | WRITE_FORCE, | |
32 | WRITE_APPEND | |
33 | }; | |
34 | ||
d6d901c2 | 35 | static int *fd[MAX_NR_CPUS][MAX_COUNTERS]; |
a21ca2ca | 36 | |
3de29cab SE |
37 | static u64 user_interval = ULLONG_MAX; |
38 | static u64 default_interval = 0; | |
a21ca2ca | 39 | |
42e59d7d | 40 | static int nr_cpus = 0; |
de9ac07b | 41 | static unsigned int page_size; |
42e59d7d | 42 | static unsigned int mmap_pages = 128; |
f9212819 | 43 | static unsigned int user_freq = UINT_MAX; |
42e59d7d | 44 | static int freq = 1000; |
de9ac07b | 45 | static int output; |
529870e3 | 46 | static int pipe_output = 0; |
23ac9cbe | 47 | static const char *output_name = "perf.data"; |
42e59d7d | 48 | static int group = 0; |
1967936d | 49 | static int realtime_prio = 0; |
c0555642 IM |
50 | static bool raw_samples = false; |
51 | static bool system_wide = false; | |
42e59d7d | 52 | static pid_t target_pid = -1; |
d6d901c2 ZY |
53 | static pid_t target_tid = -1; |
54 | static pid_t *all_tids = NULL; | |
55 | static int thread_num = 0; | |
42e59d7d | 56 | static pid_t child_pid = -1; |
2e6cdf99 | 57 | static bool no_inherit = false; |
7865e817 | 58 | static enum write_mode_t write_mode = WRITE_FORCE; |
c0555642 IM |
59 | static bool call_graph = false; |
60 | static bool inherit_stat = false; | |
61 | static bool no_samples = false; | |
62 | static bool sample_address = false; | |
a1ac1d3c | 63 | static bool no_buildid = false; |
42e59d7d IM |
64 | |
65 | static long samples = 0; | |
42e59d7d | 66 | static u64 bytes_written = 0; |
a21ca2ca | 67 | |
d6d901c2 | 68 | static struct pollfd *event_array; |
a21ca2ca | 69 | |
42e59d7d IM |
70 | static int nr_poll = 0; |
71 | static int nr_cpu = 0; | |
a21ca2ca | 72 | |
42e59d7d | 73 | static int file_new = 1; |
6122e4e4 | 74 | static off_t post_processing_offset; |
7c6a1c65 | 75 | |
94c744b6 | 76 | static struct perf_session *session; |
c45c6ea2 | 77 | static const char *cpu_list; |
f5970550 | 78 | |
de9ac07b | 79 | struct mmap_data { |
a21ca2ca IM |
80 | int counter; |
81 | void *base; | |
82 | unsigned int mask; | |
83 | unsigned int prev; | |
de9ac07b PZ |
84 | }; |
85 | ||
0e2e63dd | 86 | static struct mmap_data mmap_array[MAX_NR_CPUS]; |
a21ca2ca | 87 | |
9d91a6f7 | 88 | static unsigned long mmap_read_head(struct mmap_data *md) |
de9ac07b | 89 | { |
cdd6c482 | 90 | struct perf_event_mmap_page *pc = md->base; |
9d91a6f7 | 91 | long head; |
de9ac07b PZ |
92 | |
93 | head = pc->data_head; | |
94 | rmb(); | |
95 | ||
96 | return head; | |
97 | } | |
98 | ||
9d91a6f7 PZ |
99 | static void mmap_write_tail(struct mmap_data *md, unsigned long tail) |
100 | { | |
cdd6c482 | 101 | struct perf_event_mmap_page *pc = md->base; |
9d91a6f7 PZ |
102 | |
103 | /* | |
104 | * ensure all reads are done before we write the tail out. | |
105 | */ | |
106 | /* mb(); */ | |
107 | pc->data_tail = tail; | |
108 | } | |
109 | ||
9215545e TZ |
110 | static void advance_output(size_t size) |
111 | { | |
112 | bytes_written += size; | |
113 | } | |
114 | ||
f5970550 PZ |
115 | static void write_output(void *buf, size_t size) |
116 | { | |
117 | while (size) { | |
118 | int ret = write(output, buf, size); | |
119 | ||
120 | if (ret < 0) | |
121 | die("failed to write"); | |
122 | ||
123 | size -= ret; | |
124 | buf += ret; | |
125 | ||
126 | bytes_written += ret; | |
127 | } | |
128 | } | |
129 | ||
d8f66248 ACM |
130 | static int process_synthesized_event(event_t *event, |
131 | struct perf_session *self __used) | |
234fbbf5 | 132 | { |
6122e4e4 | 133 | write_output(event, event->header.size); |
234fbbf5 ACM |
134 | return 0; |
135 | } | |
136 | ||
de9ac07b PZ |
137 | static void mmap_read(struct mmap_data *md) |
138 | { | |
139 | unsigned int head = mmap_read_head(md); | |
140 | unsigned int old = md->prev; | |
141 | unsigned char *data = md->base + page_size; | |
142 | unsigned long size; | |
143 | void *buf; | |
144 | int diff; | |
145 | ||
de9ac07b PZ |
146 | /* |
147 | * If we're further behind than half the buffer, there's a chance | |
2debbc83 | 148 | * the writer will bite our tail and mess up the samples under us. |
de9ac07b PZ |
149 | * |
150 | * If we somehow ended up ahead of the head, we got messed up. | |
151 | * | |
152 | * In either case, truncate and restart at head. | |
153 | */ | |
154 | diff = head - old; | |
9d91a6f7 | 155 | if (diff < 0) { |
ef365cef | 156 | fprintf(stderr, "WARNING: failed to keep up with mmap data\n"); |
de9ac07b PZ |
157 | /* |
158 | * head points to a known good entry, start there. | |
159 | */ | |
160 | old = head; | |
161 | } | |
162 | ||
de9ac07b | 163 | if (old != head) |
2debbc83 | 164 | samples++; |
de9ac07b PZ |
165 | |
166 | size = head - old; | |
167 | ||
168 | if ((old & md->mask) + size != (head & md->mask)) { | |
169 | buf = &data[old & md->mask]; | |
170 | size = md->mask + 1 - (old & md->mask); | |
171 | old += size; | |
021e9f47 | 172 | |
6122e4e4 | 173 | write_output(buf, size); |
de9ac07b PZ |
174 | } |
175 | ||
176 | buf = &data[old & md->mask]; | |
177 | size = head - old; | |
178 | old += size; | |
021e9f47 | 179 | |
6122e4e4 | 180 | write_output(buf, size); |
de9ac07b PZ |
181 | |
182 | md->prev = old; | |
9d91a6f7 | 183 | mmap_write_tail(md, old); |
de9ac07b PZ |
184 | } |
185 | ||
186 | static volatile int done = 0; | |
f7b7c26e | 187 | static volatile int signr = -1; |
de9ac07b | 188 | |
16c8a109 | 189 | static void sig_handler(int sig) |
de9ac07b | 190 | { |
16c8a109 | 191 | done = 1; |
f7b7c26e PZ |
192 | signr = sig; |
193 | } | |
194 | ||
195 | static void sig_atexit(void) | |
196 | { | |
5ffc8881 | 197 | if (child_pid > 0) |
933da83a CW |
198 | kill(child_pid, SIGTERM); |
199 | ||
f7b7c26e PZ |
200 | if (signr == -1) |
201 | return; | |
202 | ||
203 | signal(signr, SIG_DFL); | |
204 | kill(getpid(), signr); | |
de9ac07b PZ |
205 | } |
206 | ||
f250c030 IM |
207 | static int group_fd; |
208 | ||
cdd6c482 | 209 | static struct perf_header_attr *get_header_attr(struct perf_event_attr *a, int nr) |
7c6a1c65 PZ |
210 | { |
211 | struct perf_header_attr *h_attr; | |
212 | ||
94c744b6 ACM |
213 | if (nr < session->header.attrs) { |
214 | h_attr = session->header.attr[nr]; | |
7c6a1c65 PZ |
215 | } else { |
216 | h_attr = perf_header_attr__new(a); | |
dc79c0fc | 217 | if (h_attr != NULL) |
94c744b6 | 218 | if (perf_header__add_attr(&session->header, h_attr) < 0) { |
11deb1f9 ACM |
219 | perf_header_attr__delete(h_attr); |
220 | h_attr = NULL; | |
221 | } | |
7c6a1c65 PZ |
222 | } |
223 | ||
224 | return h_attr; | |
225 | } | |
226 | ||
d6d901c2 | 227 | static void create_counter(int counter, int cpu) |
de9ac07b | 228 | { |
c171b552 | 229 | char *filter = filters[counter]; |
cdd6c482 | 230 | struct perf_event_attr *attr = attrs + counter; |
7c6a1c65 PZ |
231 | struct perf_header_attr *h_attr; |
232 | int track = !counter; /* only the first counter needs these */ | |
d6d901c2 | 233 | int thread_index; |
c171b552 | 234 | int ret; |
7c6a1c65 PZ |
235 | struct { |
236 | u64 count; | |
237 | u64 time_enabled; | |
238 | u64 time_running; | |
239 | u64 id; | |
240 | } read_data; | |
241 | ||
242 | attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | | |
243 | PERF_FORMAT_TOTAL_TIME_RUNNING | | |
244 | PERF_FORMAT_ID; | |
16c8a109 | 245 | |
3a9f131f | 246 | attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID; |
3efa1cc9 | 247 | |
8907fd60 EM |
248 | if (nr_counters > 1) |
249 | attr->sample_type |= PERF_SAMPLE_ID; | |
250 | ||
f9212819 FW |
251 | /* |
252 | * We default some events to a 1 default interval. But keep | |
253 | * it a weak assumption overridable by the user. | |
254 | */ | |
255 | if (!attr->sample_period || (user_freq != UINT_MAX && | |
3de29cab | 256 | user_interval != ULLONG_MAX)) { |
f9212819 FW |
257 | if (freq) { |
258 | attr->sample_type |= PERF_SAMPLE_PERIOD; | |
259 | attr->freq = 1; | |
260 | attr->sample_freq = freq; | |
261 | } else { | |
262 | attr->sample_period = default_interval; | |
263 | } | |
1dba15e7 | 264 | } |
3efa1cc9 | 265 | |
649c48a9 PZ |
266 | if (no_samples) |
267 | attr->sample_freq = 0; | |
268 | ||
269 | if (inherit_stat) | |
270 | attr->inherit_stat = 1; | |
271 | ||
3af9e859 | 272 | if (sample_address) { |
4bba828d | 273 | attr->sample_type |= PERF_SAMPLE_ADDR; |
3af9e859 EM |
274 | attr->mmap_data = track; |
275 | } | |
4bba828d | 276 | |
3efa1cc9 IM |
277 | if (call_graph) |
278 | attr->sample_type |= PERF_SAMPLE_CALLCHAIN; | |
279 | ||
f60f3593 AS |
280 | if (system_wide) |
281 | attr->sample_type |= PERF_SAMPLE_CPU; | |
282 | ||
cd6feeea | 283 | if (raw_samples) { |
6ddf259d | 284 | attr->sample_type |= PERF_SAMPLE_TIME; |
daac07b2 | 285 | attr->sample_type |= PERF_SAMPLE_RAW; |
cd6feeea IM |
286 | attr->sample_type |= PERF_SAMPLE_CPU; |
287 | } | |
f413cdb8 | 288 | |
a21ca2ca IM |
289 | attr->mmap = track; |
290 | attr->comm = track; | |
2e6cdf99 SE |
291 | attr->inherit = !no_inherit; |
292 | if (target_pid == -1 && target_tid == -1 && !system_wide) { | |
46be604b | 293 | attr->disabled = 1; |
bedbfdea | 294 | attr->enable_on_exec = 1; |
46be604b | 295 | } |
bedbfdea | 296 | |
d6d901c2 | 297 | for (thread_index = 0; thread_index < thread_num; thread_index++) { |
3da297a6 | 298 | try_again: |
d6d901c2 ZY |
299 | fd[nr_cpu][counter][thread_index] = sys_perf_event_open(attr, |
300 | all_tids[thread_index], cpu, group_fd, 0); | |
301 | ||
302 | if (fd[nr_cpu][counter][thread_index] < 0) { | |
303 | int err = errno; | |
304 | ||
305 | if (err == EPERM || err == EACCES) | |
306 | die("Permission error - are you root?\n" | |
307 | "\t Consider tweaking" | |
308 | " /proc/sys/kernel/perf_event_paranoid.\n"); | |
c45c6ea2 | 309 | else if (err == ENODEV && cpu_list) { |
d6d901c2 ZY |
310 | die("No such device - did you specify" |
311 | " an out-of-range profile CPU?\n"); | |
312 | } | |
3da297a6 | 313 | |
d6d901c2 ZY |
314 | /* |
315 | * If it's cycles then fall back to hrtimer | |
316 | * based cpu-clock-tick sw counter, which | |
317 | * is always available even if no PMU support: | |
318 | */ | |
319 | if (attr->type == PERF_TYPE_HARDWARE | |
320 | && attr->config == PERF_COUNT_HW_CPU_CYCLES) { | |
321 | ||
322 | if (verbose) | |
323 | warning(" ... trying to fall back to cpu-clock-ticks\n"); | |
324 | attr->type = PERF_TYPE_SOFTWARE; | |
325 | attr->config = PERF_COUNT_SW_CPU_CLOCK; | |
326 | goto try_again; | |
327 | } | |
328 | printf("\n"); | |
329 | error("perfcounter syscall returned with %d (%s)\n", | |
330 | fd[nr_cpu][counter][thread_index], strerror(err)); | |
bfd45118 SK |
331 | |
332 | #if defined(__i386__) || defined(__x86_64__) | |
d6d901c2 ZY |
333 | if (attr->type == PERF_TYPE_HARDWARE && err == EOPNOTSUPP) |
334 | die("No hardware sampling interrupt available." | |
335 | " No APIC? If so then you can boot the kernel" | |
336 | " with the \"lapic\" boot parameter to" | |
337 | " force-enable it.\n"); | |
bfd45118 SK |
338 | #endif |
339 | ||
d6d901c2 ZY |
340 | die("No CONFIG_PERF_EVENTS=y kernel support configured?\n"); |
341 | exit(-1); | |
342 | } | |
3da297a6 | 343 | |
d6d901c2 ZY |
344 | h_attr = get_header_attr(attr, counter); |
345 | if (h_attr == NULL) | |
346 | die("nomem\n"); | |
7c6a1c65 | 347 | |
d6d901c2 ZY |
348 | if (!file_new) { |
349 | if (memcmp(&h_attr->attr, attr, sizeof(*attr))) { | |
350 | fprintf(stderr, "incompatible append\n"); | |
351 | exit(-1); | |
352 | } | |
7c6a1c65 | 353 | } |
7c6a1c65 | 354 | |
d6d901c2 | 355 | if (read(fd[nr_cpu][counter][thread_index], &read_data, sizeof(read_data)) == -1) { |
0ab7368f | 356 | perror("Unable to read perf file descriptor"); |
d6d901c2 ZY |
357 | exit(-1); |
358 | } | |
7c6a1c65 | 359 | |
d6d901c2 ZY |
360 | if (perf_header_attr__add_id(h_attr, read_data.id) < 0) { |
361 | pr_warning("Not enough memory to add id\n"); | |
362 | exit(-1); | |
363 | } | |
7c6a1c65 | 364 | |
d6d901c2 ZY |
365 | assert(fd[nr_cpu][counter][thread_index] >= 0); |
366 | fcntl(fd[nr_cpu][counter][thread_index], F_SETFL, O_NONBLOCK); | |
16c8a109 | 367 | |
d6d901c2 ZY |
368 | /* |
369 | * First counter acts as the group leader: | |
370 | */ | |
371 | if (group && group_fd == -1) | |
372 | group_fd = fd[nr_cpu][counter][thread_index]; | |
f250c030 | 373 | |
0e2e63dd PZ |
374 | if (counter || thread_index) { |
375 | ret = ioctl(fd[nr_cpu][counter][thread_index], | |
376 | PERF_EVENT_IOC_SET_OUTPUT, | |
377 | fd[nr_cpu][0][0]); | |
378 | if (ret) { | |
379 | error("failed to set output: %d (%s)\n", errno, | |
380 | strerror(errno)); | |
381 | exit(-1); | |
382 | } | |
383 | } else { | |
384 | mmap_array[nr_cpu].counter = counter; | |
385 | mmap_array[nr_cpu].prev = 0; | |
386 | mmap_array[nr_cpu].mask = mmap_pages*page_size - 1; | |
387 | mmap_array[nr_cpu].base = mmap(NULL, (mmap_pages+1)*page_size, | |
388 | PROT_READ|PROT_WRITE, MAP_SHARED, fd[nr_cpu][counter][thread_index], 0); | |
389 | if (mmap_array[nr_cpu].base == MAP_FAILED) { | |
390 | error("failed to mmap with %d (%s)\n", errno, strerror(errno)); | |
391 | exit(-1); | |
392 | } | |
393 | ||
394 | event_array[nr_poll].fd = fd[nr_cpu][counter][thread_index]; | |
395 | event_array[nr_poll].events = POLLIN; | |
396 | nr_poll++; | |
ea57c4f5 | 397 | } |
d1302522 | 398 | |
d6d901c2 ZY |
399 | if (filter != NULL) { |
400 | ret = ioctl(fd[nr_cpu][counter][thread_index], | |
401 | PERF_EVENT_IOC_SET_FILTER, filter); | |
402 | if (ret) { | |
403 | error("failed to set filter with %d (%s)\n", errno, | |
404 | strerror(errno)); | |
405 | exit(-1); | |
406 | } | |
c171b552 LZ |
407 | } |
408 | } | |
f250c030 | 409 | } |
f2521b6e | 410 | |
d6d901c2 | 411 | static void open_counters(int cpu) |
f250c030 IM |
412 | { |
413 | int counter; | |
16c8a109 | 414 | |
f250c030 IM |
415 | group_fd = -1; |
416 | for (counter = 0; counter < nr_counters; counter++) | |
d6d901c2 | 417 | create_counter(counter, cpu); |
f250c030 | 418 | |
16c8a109 PZ |
419 | nr_cpu++; |
420 | } | |
421 | ||
6122e4e4 ACM |
422 | static int process_buildids(void) |
423 | { | |
424 | u64 size = lseek(output, 0, SEEK_CUR); | |
425 | ||
9f591fd7 ACM |
426 | if (size == 0) |
427 | return 0; | |
428 | ||
6122e4e4 ACM |
429 | session->fd = output; |
430 | return __perf_session__process_events(session, post_processing_offset, | |
431 | size - post_processing_offset, | |
432 | size, &build_id__mark_dso_hit_ops); | |
433 | } | |
434 | ||
f5970550 PZ |
435 | static void atexit_header(void) |
436 | { | |
c7929e47 TZ |
437 | if (!pipe_output) { |
438 | session->header.data_size += bytes_written; | |
f5970550 | 439 | |
c7929e47 TZ |
440 | process_buildids(); |
441 | perf_header__write(&session->header, output, true); | |
39d17dac | 442 | perf_session__delete(session); |
d65a458b | 443 | symbol__exit(); |
c7929e47 | 444 | } |
f5970550 PZ |
445 | } |
446 | ||
23346f21 | 447 | static void event__synthesize_guest_os(struct machine *machine, void *data) |
a1645ce1 ZY |
448 | { |
449 | int err; | |
23346f21 | 450 | struct perf_session *psession = data; |
a1645ce1 | 451 | |
23346f21 | 452 | if (machine__is_host(machine)) |
a1645ce1 ZY |
453 | return; |
454 | ||
455 | /* | |
456 | *As for guest kernel when processing subcommand record&report, | |
457 | *we arrange module mmap prior to guest kernel mmap and trigger | |
458 | *a preload dso because default guest module symbols are loaded | |
459 | *from guest kallsyms instead of /lib/modules/XXX/XXX. This | |
460 | *method is used to avoid symbol missing when the first addr is | |
461 | *in module instead of in guest kernel. | |
462 | */ | |
463 | err = event__synthesize_modules(process_synthesized_event, | |
23346f21 | 464 | psession, machine); |
a1645ce1 ZY |
465 | if (err < 0) |
466 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 467 | " relocation symbol.\n", machine->pid); |
a1645ce1 | 468 | |
a1645ce1 ZY |
469 | /* |
470 | * We use _stext for guest kernel because guest kernel's /proc/kallsyms | |
471 | * have no _text sometimes. | |
472 | */ | |
473 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 474 | psession, machine, "_text"); |
a1645ce1 ZY |
475 | if (err < 0) |
476 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 477 | psession, machine, "_stext"); |
a1645ce1 ZY |
478 | if (err < 0) |
479 | pr_err("Couldn't record guest kernel [%d]'s reference" | |
23346f21 | 480 | " relocation symbol.\n", machine->pid); |
a1645ce1 ZY |
481 | } |
482 | ||
98402807 FW |
483 | static struct perf_event_header finished_round_event = { |
484 | .size = sizeof(struct perf_event_header), | |
485 | .type = PERF_RECORD_FINISHED_ROUND, | |
486 | }; | |
487 | ||
488 | static void mmap_read_all(void) | |
489 | { | |
0e2e63dd | 490 | int i; |
98402807 FW |
491 | |
492 | for (i = 0; i < nr_cpu; i++) { | |
0e2e63dd PZ |
493 | if (mmap_array[i].base) |
494 | mmap_read(&mmap_array[i]); | |
98402807 FW |
495 | } |
496 | ||
497 | if (perf_header__has_feat(&session->header, HEADER_TRACE_INFO)) | |
498 | write_output(&finished_round_event, sizeof(finished_round_event)); | |
499 | } | |
500 | ||
d4db3f16 | 501 | static int __cmd_record(int argc, const char **argv) |
16c8a109 PZ |
502 | { |
503 | int i, counter; | |
abaff32a | 504 | struct stat st; |
abaff32a | 505 | int flags; |
4dc0a04b | 506 | int err; |
8b412664 | 507 | unsigned long waking = 0; |
856e9660 | 508 | int child_ready_pipe[2], go_pipe[2]; |
46be604b | 509 | const bool forks = argc > 0; |
856e9660 | 510 | char buf; |
23346f21 | 511 | struct machine *machine; |
de9ac07b PZ |
512 | |
513 | page_size = sysconf(_SC_PAGE_SIZE); | |
de9ac07b | 514 | |
f5970550 PZ |
515 | atexit(sig_atexit); |
516 | signal(SIGCHLD, sig_handler); | |
517 | signal(SIGINT, sig_handler); | |
518 | ||
d4db3f16 | 519 | if (forks && (pipe(child_ready_pipe) < 0 || pipe(go_pipe) < 0)) { |
856e9660 PZ |
520 | perror("failed to create pipes"); |
521 | exit(-1); | |
522 | } | |
523 | ||
529870e3 TZ |
524 | if (!strcmp(output_name, "-")) |
525 | pipe_output = 1; | |
526 | else if (!stat(output_name, &st) && st.st_size) { | |
7865e817 | 527 | if (write_mode == WRITE_FORCE) { |
b38d3464 ACM |
528 | char oldname[PATH_MAX]; |
529 | snprintf(oldname, sizeof(oldname), "%s.old", | |
530 | output_name); | |
531 | unlink(oldname); | |
532 | rename(output_name, oldname); | |
266e0e21 | 533 | } |
7865e817 FW |
534 | } else if (write_mode == WRITE_APPEND) { |
535 | write_mode = WRITE_FORCE; | |
97124d5e PZ |
536 | } |
537 | ||
f887f301 | 538 | flags = O_CREAT|O_RDWR; |
7865e817 | 539 | if (write_mode == WRITE_APPEND) |
f5970550 | 540 | file_new = 0; |
abaff32a IM |
541 | else |
542 | flags |= O_TRUNC; | |
543 | ||
529870e3 TZ |
544 | if (pipe_output) |
545 | output = STDOUT_FILENO; | |
546 | else | |
547 | output = open(output_name, flags, S_IRUSR | S_IWUSR); | |
de9ac07b PZ |
548 | if (output < 0) { |
549 | perror("failed to create output file"); | |
550 | exit(-1); | |
551 | } | |
552 | ||
7865e817 | 553 | session = perf_session__new(output_name, O_WRONLY, |
454c407e | 554 | write_mode == WRITE_FORCE, false); |
94c744b6 | 555 | if (session == NULL) { |
a9a70bbc ACM |
556 | pr_err("Not enough memory for reading perf file header\n"); |
557 | return -1; | |
558 | } | |
559 | ||
4dc0a04b | 560 | if (!file_new) { |
8dc58101 | 561 | err = perf_header__read(session, output); |
4dc0a04b | 562 | if (err < 0) |
39d17dac | 563 | goto out_delete_session; |
4dc0a04b ACM |
564 | } |
565 | ||
db620b1c | 566 | if (have_tracepoints(attrs, nr_counters)) |
94c744b6 | 567 | perf_header__set_feat(&session->header, HEADER_TRACE_INFO); |
03456a15 | 568 | |
39d17dac ACM |
569 | /* |
570 | * perf_session__delete(session) will be called at atexit_header() | |
571 | */ | |
f5970550 PZ |
572 | atexit(atexit_header); |
573 | ||
d4db3f16 | 574 | if (forks) { |
46be604b | 575 | child_pid = fork(); |
2fb750e8 | 576 | if (child_pid < 0) { |
856e9660 PZ |
577 | perror("failed to fork"); |
578 | exit(-1); | |
579 | } | |
7c6a1c65 | 580 | |
46be604b | 581 | if (!child_pid) { |
529870e3 TZ |
582 | if (pipe_output) |
583 | dup2(2, 1); | |
856e9660 PZ |
584 | close(child_ready_pipe[0]); |
585 | close(go_pipe[1]); | |
586 | fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC); | |
587 | ||
588 | /* | |
589 | * Do a dummy execvp to get the PLT entry resolved, | |
590 | * so we avoid the resolver overhead on the real | |
591 | * execvp call. | |
592 | */ | |
593 | execvp("", (char **)argv); | |
594 | ||
595 | /* | |
596 | * Tell the parent we're ready to go | |
597 | */ | |
598 | close(child_ready_pipe[1]); | |
599 | ||
600 | /* | |
601 | * Wait until the parent tells us to go. | |
602 | */ | |
603 | if (read(go_pipe[0], &buf, 1) == -1) | |
604 | perror("unable to read pipe"); | |
605 | ||
606 | execvp(argv[0], (char **)argv); | |
607 | ||
608 | perror(argv[0]); | |
609 | exit(-1); | |
0a5ac846 | 610 | } |
856e9660 | 611 | |
d6d901c2 ZY |
612 | if (!system_wide && target_tid == -1 && target_pid == -1) |
613 | all_tids[0] = child_pid; | |
614 | ||
856e9660 PZ |
615 | close(child_ready_pipe[1]); |
616 | close(go_pipe[0]); | |
617 | /* | |
618 | * wait for child to settle | |
619 | */ | |
620 | if (read(child_ready_pipe[0], &buf, 1) == -1) { | |
621 | perror("unable to read pipe"); | |
622 | exit(-1); | |
623 | } | |
624 | close(child_ready_pipe[0]); | |
625 | } | |
626 | ||
c45c6ea2 SE |
627 | nr_cpus = read_cpu_map(cpu_list); |
628 | if (nr_cpus < 1) { | |
0ab7368f | 629 | perror("failed to collect number of CPUs"); |
c45c6ea2 SE |
630 | return -1; |
631 | } | |
632 | ||
633 | if (!system_wide && no_inherit && !cpu_list) { | |
634 | open_counters(-1); | |
856e9660 PZ |
635 | } else { |
636 | for (i = 0; i < nr_cpus; i++) | |
d6d901c2 | 637 | open_counters(cpumap[i]); |
0a5ac846 | 638 | } |
de9ac07b | 639 | |
529870e3 TZ |
640 | if (pipe_output) { |
641 | err = perf_header__write_pipe(output); | |
642 | if (err < 0) | |
643 | return err; | |
644 | } else if (file_new) { | |
94c744b6 | 645 | err = perf_header__write(&session->header, output, false); |
d5eed904 ACM |
646 | if (err < 0) |
647 | return err; | |
56b03f3c ACM |
648 | } |
649 | ||
6122e4e4 ACM |
650 | post_processing_offset = lseek(output, 0, SEEK_CUR); |
651 | ||
2c46dbb5 TZ |
652 | if (pipe_output) { |
653 | err = event__synthesize_attrs(&session->header, | |
654 | process_synthesized_event, | |
655 | session); | |
656 | if (err < 0) { | |
657 | pr_err("Couldn't synthesize attrs.\n"); | |
658 | return err; | |
659 | } | |
cd19a035 TZ |
660 | |
661 | err = event__synthesize_event_types(process_synthesized_event, | |
662 | session); | |
663 | if (err < 0) { | |
664 | pr_err("Couldn't synthesize event_types.\n"); | |
665 | return err; | |
666 | } | |
9215545e | 667 | |
63e0c771 TZ |
668 | if (have_tracepoints(attrs, nr_counters)) { |
669 | /* | |
670 | * FIXME err <= 0 here actually means that | |
671 | * there were no tracepoints so its not really | |
672 | * an error, just that we don't need to | |
673 | * synthesize anything. We really have to | |
674 | * return this more properly and also | |
675 | * propagate errors that now are calling die() | |
676 | */ | |
677 | err = event__synthesize_tracing_data(output, attrs, | |
678 | nr_counters, | |
679 | process_synthesized_event, | |
680 | session); | |
681 | if (err <= 0) { | |
682 | pr_err("Couldn't record tracing data.\n"); | |
683 | return err; | |
684 | } | |
2c9faa06 | 685 | advance_output(err); |
63e0c771 | 686 | } |
2c46dbb5 TZ |
687 | } |
688 | ||
23346f21 ACM |
689 | machine = perf_session__find_host_machine(session); |
690 | if (!machine) { | |
a1645ce1 ZY |
691 | pr_err("Couldn't find native kernel information.\n"); |
692 | return -1; | |
693 | } | |
694 | ||
56b03f3c | 695 | err = event__synthesize_kernel_mmap(process_synthesized_event, |
23346f21 | 696 | session, machine, "_text"); |
70162138 ACM |
697 | if (err < 0) |
698 | err = event__synthesize_kernel_mmap(process_synthesized_event, | |
23346f21 | 699 | session, machine, "_stext"); |
56b03f3c ACM |
700 | if (err < 0) { |
701 | pr_err("Couldn't record kernel reference relocation symbol.\n"); | |
702 | return err; | |
b7cece76 ACM |
703 | } |
704 | ||
a1645ce1 | 705 | err = event__synthesize_modules(process_synthesized_event, |
23346f21 | 706 | session, machine); |
b7cece76 ACM |
707 | if (err < 0) { |
708 | pr_err("Couldn't record kernel reference relocation symbol.\n"); | |
709 | return err; | |
d5eed904 | 710 | } |
a1645ce1 | 711 | if (perf_guest) |
23346f21 | 712 | perf_session__process_machines(session, event__synthesize_guest_os); |
7c6a1c65 | 713 | |
cf103a14 | 714 | if (!system_wide) |
d6d901c2 | 715 | event__synthesize_thread(target_tid, process_synthesized_event, |
d8f66248 | 716 | session); |
234fbbf5 | 717 | else |
d8f66248 | 718 | event__synthesize_threads(process_synthesized_event, session); |
7c6a1c65 | 719 | |
de9ac07b PZ |
720 | if (realtime_prio) { |
721 | struct sched_param param; | |
722 | ||
723 | param.sched_priority = realtime_prio; | |
724 | if (sched_setscheduler(0, SCHED_FIFO, ¶m)) { | |
6beba7ad | 725 | pr_err("Could not set realtime priority.\n"); |
de9ac07b PZ |
726 | exit(-1); |
727 | } | |
728 | } | |
729 | ||
856e9660 PZ |
730 | /* |
731 | * Let the child rip | |
732 | */ | |
d4db3f16 ACM |
733 | if (forks) |
734 | close(go_pipe[1]); | |
856e9660 | 735 | |
649c48a9 | 736 | for (;;) { |
2debbc83 | 737 | int hits = samples; |
d6d901c2 | 738 | int thread; |
de9ac07b | 739 | |
98402807 | 740 | mmap_read_all(); |
de9ac07b | 741 | |
649c48a9 PZ |
742 | if (hits == samples) { |
743 | if (done) | |
744 | break; | |
4dc0a04b | 745 | err = poll(event_array, nr_poll, -1); |
8b412664 PZ |
746 | waking++; |
747 | } | |
748 | ||
749 | if (done) { | |
750 | for (i = 0; i < nr_cpu; i++) { | |
d6d901c2 ZY |
751 | for (counter = 0; |
752 | counter < nr_counters; | |
753 | counter++) { | |
754 | for (thread = 0; | |
755 | thread < thread_num; | |
756 | thread++) | |
757 | ioctl(fd[i][counter][thread], | |
758 | PERF_EVENT_IOC_DISABLE); | |
759 | } | |
8b412664 | 760 | } |
649c48a9 | 761 | } |
de9ac07b PZ |
762 | } |
763 | ||
b44308f5 ACM |
764 | if (quiet) |
765 | return 0; | |
766 | ||
8b412664 PZ |
767 | fprintf(stderr, "[ perf record: Woken up %ld times to write data ]\n", waking); |
768 | ||
021e9f47 IM |
769 | /* |
770 | * Approximate RIP event size: 24 bytes. | |
771 | */ | |
772 | fprintf(stderr, | |
2debbc83 | 773 | "[ perf record: Captured and wrote %.3f MB %s (~%lld samples) ]\n", |
021e9f47 IM |
774 | (double)bytes_written / 1024.0 / 1024.0, |
775 | output_name, | |
776 | bytes_written / 24); | |
addc2785 | 777 | |
de9ac07b | 778 | return 0; |
39d17dac ACM |
779 | |
780 | out_delete_session: | |
781 | perf_session__delete(session); | |
782 | return err; | |
de9ac07b | 783 | } |
0e9b20b8 | 784 | |
0e9b20b8 | 785 | static const char * const record_usage[] = { |
9e096753 MG |
786 | "perf record [<options>] [<command>]", |
787 | "perf record [<options>] -- <command> [<options>]", | |
0e9b20b8 IM |
788 | NULL |
789 | }; | |
790 | ||
7865e817 FW |
791 | static bool force, append_file; |
792 | ||
bca647aa | 793 | const struct option record_options[] = { |
0e9b20b8 | 794 | OPT_CALLBACK('e', "event", NULL, "event", |
86847b62 TG |
795 | "event selector. use 'perf list' to list available events", |
796 | parse_events), | |
c171b552 LZ |
797 | OPT_CALLBACK(0, "filter", NULL, "filter", |
798 | "event filter", parse_filter), | |
0e9b20b8 | 799 | OPT_INTEGER('p', "pid", &target_pid, |
d6d901c2 ZY |
800 | "record events on existing process id"), |
801 | OPT_INTEGER('t', "tid", &target_tid, | |
802 | "record events on existing thread id"), | |
0e9b20b8 IM |
803 | OPT_INTEGER('r', "realtime", &realtime_prio, |
804 | "collect data with this RT SCHED_FIFO priority"), | |
daac07b2 FW |
805 | OPT_BOOLEAN('R', "raw-samples", &raw_samples, |
806 | "collect raw sample records from all opened counters"), | |
0e9b20b8 IM |
807 | OPT_BOOLEAN('a', "all-cpus", &system_wide, |
808 | "system-wide collection from all CPUs"), | |
abaff32a IM |
809 | OPT_BOOLEAN('A', "append", &append_file, |
810 | "append to the output file to do incremental profiling"), | |
c45c6ea2 SE |
811 | OPT_STRING('C', "cpu", &cpu_list, "cpu", |
812 | "list of cpus to monitor"), | |
97124d5e | 813 | OPT_BOOLEAN('f', "force", &force, |
7865e817 | 814 | "overwrite existing data file (deprecated)"), |
3de29cab | 815 | OPT_U64('c', "count", &user_interval, "event period to sample"), |
abaff32a IM |
816 | OPT_STRING('o', "output", &output_name, "file", |
817 | "output file name"), | |
2e6cdf99 SE |
818 | OPT_BOOLEAN('i', "no-inherit", &no_inherit, |
819 | "child tasks do not inherit counters"), | |
1967936d ACM |
820 | OPT_UINTEGER('F', "freq", &user_freq, "profile at this frequency"), |
821 | OPT_UINTEGER('m', "mmap-pages", &mmap_pages, "number of mmap data pages"), | |
3efa1cc9 IM |
822 | OPT_BOOLEAN('g', "call-graph", &call_graph, |
823 | "do call-graph (stack chain/backtrace) recording"), | |
c0555642 | 824 | OPT_INCR('v', "verbose", &verbose, |
3da297a6 | 825 | "be more verbose (show counter open errors, etc)"), |
b44308f5 | 826 | OPT_BOOLEAN('q', "quiet", &quiet, "don't print any message"), |
649c48a9 PZ |
827 | OPT_BOOLEAN('s', "stat", &inherit_stat, |
828 | "per thread counts"), | |
4bba828d AB |
829 | OPT_BOOLEAN('d', "data", &sample_address, |
830 | "Sample addresses"), | |
649c48a9 PZ |
831 | OPT_BOOLEAN('n', "no-samples", &no_samples, |
832 | "don't sample"), | |
a1ac1d3c SE |
833 | OPT_BOOLEAN('N', "no-buildid-cache", &no_buildid, |
834 | "do not update the buildid cache"), | |
0e9b20b8 IM |
835 | OPT_END() |
836 | }; | |
837 | ||
f37a291c | 838 | int cmd_record(int argc, const char **argv, const char *prefix __used) |
0e9b20b8 | 839 | { |
39d17dac | 840 | int i, j, err = -ENOMEM; |
0e9b20b8 | 841 | |
bca647aa | 842 | argc = parse_options(argc, argv, record_options, record_usage, |
655000e7 | 843 | PARSE_OPT_STOP_AT_NON_OPTION); |
d6d901c2 | 844 | if (!argc && target_pid == -1 && target_tid == -1 && |
c45c6ea2 | 845 | !system_wide && !cpu_list) |
bca647aa | 846 | usage_with_options(record_usage, record_options); |
0e9b20b8 | 847 | |
7865e817 FW |
848 | if (force && append_file) { |
849 | fprintf(stderr, "Can't overwrite and append at the same time." | |
850 | " You need to choose between -f and -A"); | |
bca647aa | 851 | usage_with_options(record_usage, record_options); |
7865e817 FW |
852 | } else if (append_file) { |
853 | write_mode = WRITE_APPEND; | |
854 | } else { | |
855 | write_mode = WRITE_FORCE; | |
856 | } | |
857 | ||
655000e7 | 858 | symbol__init(); |
a1ac1d3c SE |
859 | if (no_buildid) |
860 | disable_buildid_cache(); | |
655000e7 | 861 | |
bbd36e5e PZ |
862 | if (!nr_counters) { |
863 | nr_counters = 1; | |
864 | attrs[0].type = PERF_TYPE_HARDWARE; | |
865 | attrs[0].config = PERF_COUNT_HW_CPU_CYCLES; | |
866 | } | |
0e9b20b8 | 867 | |
d6d901c2 ZY |
868 | if (target_pid != -1) { |
869 | target_tid = target_pid; | |
870 | thread_num = find_all_tid(target_pid, &all_tids); | |
871 | if (thread_num <= 0) { | |
872 | fprintf(stderr, "Can't find all threads of pid %d\n", | |
873 | target_pid); | |
bca647aa | 874 | usage_with_options(record_usage, record_options); |
d6d901c2 ZY |
875 | } |
876 | } else { | |
877 | all_tids=malloc(sizeof(pid_t)); | |
878 | if (!all_tids) | |
d65a458b | 879 | goto out_symbol_exit; |
d6d901c2 ZY |
880 | |
881 | all_tids[0] = target_tid; | |
882 | thread_num = 1; | |
883 | } | |
884 | ||
885 | for (i = 0; i < MAX_NR_CPUS; i++) { | |
886 | for (j = 0; j < MAX_COUNTERS; j++) { | |
887 | fd[i][j] = malloc(sizeof(int)*thread_num); | |
0e2e63dd | 888 | if (!fd[i][j]) |
39d17dac | 889 | goto out_free_fd; |
d6d901c2 ZY |
890 | } |
891 | } | |
892 | event_array = malloc( | |
893 | sizeof(struct pollfd)*MAX_NR_CPUS*MAX_COUNTERS*thread_num); | |
894 | if (!event_array) | |
39d17dac | 895 | goto out_free_fd; |
d6d901c2 | 896 | |
3de29cab | 897 | if (user_interval != ULLONG_MAX) |
f9212819 FW |
898 | default_interval = user_interval; |
899 | if (user_freq != UINT_MAX) | |
900 | freq = user_freq; | |
901 | ||
7e4ff9e3 MG |
902 | /* |
903 | * User specified count overrides default frequency. | |
904 | */ | |
905 | if (default_interval) | |
906 | freq = 0; | |
907 | else if (freq) { | |
908 | default_interval = freq; | |
909 | } else { | |
910 | fprintf(stderr, "frequency and count are zero, aborting\n"); | |
39d17dac ACM |
911 | err = -EINVAL; |
912 | goto out_free_event_array; | |
7e4ff9e3 MG |
913 | } |
914 | ||
39d17dac ACM |
915 | err = __cmd_record(argc, argv); |
916 | ||
917 | out_free_event_array: | |
918 | free(event_array); | |
919 | out_free_fd: | |
920 | for (i = 0; i < MAX_NR_CPUS; i++) { | |
921 | for (j = 0; j < MAX_COUNTERS; j++) | |
922 | free(fd[i][j]); | |
923 | } | |
924 | free(all_tids); | |
925 | all_tids = NULL; | |
d65a458b ACM |
926 | out_symbol_exit: |
927 | symbol__exit(); | |
39d17dac | 928 | return err; |
0e9b20b8 | 929 | } |