]>
Commit | Line | Data |
---|---|---|
f8a95309 ACM |
1 | /* |
2 | * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com> | |
3 | * | |
4 | * Parts came from builtin-{top,stat,record}.c, see those files for further | |
5 | * copyright notes. | |
6 | * | |
7 | * Released under the GPL v2. (and only v2, not any later version) | |
8 | */ | |
a8c9ae18 ACM |
9 | #include "util.h" |
10 | #include "debugfs.h" | |
5c581041 | 11 | #include <poll.h> |
f8a95309 ACM |
12 | #include "cpumap.h" |
13 | #include "thread_map.h" | |
361c99a6 ACM |
14 | #include "evlist.h" |
15 | #include "evsel.h" | |
35b9d88e | 16 | #include <unistd.h> |
361c99a6 | 17 | |
50d08e47 ACM |
18 | #include "parse-events.h" |
19 | ||
f8a95309 ACM |
20 | #include <sys/mman.h> |
21 | ||
70db7533 ACM |
22 | #include <linux/bitops.h> |
23 | #include <linux/hash.h> | |
24 | ||
f8a95309 | 25 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
a91e5431 | 26 | #define SID(e, x, y) xyarray__entry(e->sample_id, x, y) |
f8a95309 | 27 | |
7e2ed097 ACM |
28 | void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus, |
29 | struct thread_map *threads) | |
ef1d1af2 ACM |
30 | { |
31 | int i; | |
32 | ||
33 | for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) | |
34 | INIT_HLIST_HEAD(&evlist->heads[i]); | |
35 | INIT_LIST_HEAD(&evlist->entries); | |
7e2ed097 | 36 | perf_evlist__set_maps(evlist, cpus, threads); |
35b9d88e | 37 | evlist->workload.pid = -1; |
ef1d1af2 ACM |
38 | } |
39 | ||
7e2ed097 ACM |
40 | struct perf_evlist *perf_evlist__new(struct cpu_map *cpus, |
41 | struct thread_map *threads) | |
361c99a6 ACM |
42 | { |
43 | struct perf_evlist *evlist = zalloc(sizeof(*evlist)); | |
44 | ||
ef1d1af2 | 45 | if (evlist != NULL) |
7e2ed097 | 46 | perf_evlist__init(evlist, cpus, threads); |
361c99a6 ACM |
47 | |
48 | return evlist; | |
49 | } | |
50 | ||
0f82ebc4 ACM |
51 | void perf_evlist__config_attrs(struct perf_evlist *evlist, |
52 | struct perf_record_opts *opts) | |
53 | { | |
54 | struct perf_evsel *evsel; | |
55 | ||
56 | if (evlist->cpus->map[0] < 0) | |
57 | opts->no_inherit = true; | |
58 | ||
59 | list_for_each_entry(evsel, &evlist->entries, node) { | |
60 | perf_evsel__config(evsel, opts); | |
61 | ||
62 | if (evlist->nr_entries > 1) | |
63 | evsel->attr.sample_type |= PERF_SAMPLE_ID; | |
64 | } | |
65 | } | |
66 | ||
361c99a6 ACM |
67 | static void perf_evlist__purge(struct perf_evlist *evlist) |
68 | { | |
69 | struct perf_evsel *pos, *n; | |
70 | ||
71 | list_for_each_entry_safe(pos, n, &evlist->entries, node) { | |
72 | list_del_init(&pos->node); | |
73 | perf_evsel__delete(pos); | |
74 | } | |
75 | ||
76 | evlist->nr_entries = 0; | |
77 | } | |
78 | ||
ef1d1af2 | 79 | void perf_evlist__exit(struct perf_evlist *evlist) |
361c99a6 | 80 | { |
70db7533 | 81 | free(evlist->mmap); |
5c581041 | 82 | free(evlist->pollfd); |
ef1d1af2 ACM |
83 | evlist->mmap = NULL; |
84 | evlist->pollfd = NULL; | |
85 | } | |
86 | ||
87 | void perf_evlist__delete(struct perf_evlist *evlist) | |
88 | { | |
89 | perf_evlist__purge(evlist); | |
90 | perf_evlist__exit(evlist); | |
361c99a6 ACM |
91 | free(evlist); |
92 | } | |
93 | ||
94 | void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry) | |
95 | { | |
96 | list_add_tail(&entry->node, &evlist->entries); | |
97 | ++evlist->nr_entries; | |
98 | } | |
99 | ||
50d08e47 ACM |
100 | static void perf_evlist__splice_list_tail(struct perf_evlist *evlist, |
101 | struct list_head *list, | |
102 | int nr_entries) | |
103 | { | |
104 | list_splice_tail(list, &evlist->entries); | |
105 | evlist->nr_entries += nr_entries; | |
106 | } | |
107 | ||
361c99a6 ACM |
108 | int perf_evlist__add_default(struct perf_evlist *evlist) |
109 | { | |
110 | struct perf_event_attr attr = { | |
111 | .type = PERF_TYPE_HARDWARE, | |
112 | .config = PERF_COUNT_HW_CPU_CYCLES, | |
113 | }; | |
1aed2671 JR |
114 | struct perf_evsel *evsel; |
115 | ||
116 | event_attr_init(&attr); | |
361c99a6 | 117 | |
1aed2671 | 118 | evsel = perf_evsel__new(&attr, 0); |
361c99a6 | 119 | if (evsel == NULL) |
cc2d86b0 SE |
120 | goto error; |
121 | ||
122 | /* use strdup() because free(evsel) assumes name is allocated */ | |
123 | evsel->name = strdup("cycles"); | |
124 | if (!evsel->name) | |
125 | goto error_free; | |
361c99a6 ACM |
126 | |
127 | perf_evlist__add(evlist, evsel); | |
128 | return 0; | |
cc2d86b0 SE |
129 | error_free: |
130 | perf_evsel__delete(evsel); | |
131 | error: | |
132 | return -ENOMEM; | |
361c99a6 | 133 | } |
5c581041 | 134 | |
50d08e47 ACM |
135 | int perf_evlist__add_attrs(struct perf_evlist *evlist, |
136 | struct perf_event_attr *attrs, size_t nr_attrs) | |
137 | { | |
138 | struct perf_evsel *evsel, *n; | |
139 | LIST_HEAD(head); | |
140 | size_t i; | |
141 | ||
142 | for (i = 0; i < nr_attrs; i++) { | |
143 | evsel = perf_evsel__new(attrs + i, evlist->nr_entries + i); | |
144 | if (evsel == NULL) | |
145 | goto out_delete_partial_list; | |
146 | list_add_tail(&evsel->node, &head); | |
147 | } | |
148 | ||
149 | perf_evlist__splice_list_tail(evlist, &head, nr_attrs); | |
150 | ||
151 | return 0; | |
152 | ||
153 | out_delete_partial_list: | |
154 | list_for_each_entry_safe(evsel, n, &head, node) | |
155 | perf_evsel__delete(evsel); | |
156 | return -1; | |
157 | } | |
158 | ||
a8c9ae18 ACM |
159 | static int trace_event__id(const char *evname) |
160 | { | |
161 | char *filename, *colon; | |
162 | int err = -1, fd; | |
163 | ||
164 | if (asprintf(&filename, "%s/%s/id", tracing_events_path, evname) < 0) | |
165 | return -1; | |
166 | ||
167 | colon = strrchr(filename, ':'); | |
168 | if (colon != NULL) | |
169 | *colon = '/'; | |
170 | ||
171 | fd = open(filename, O_RDONLY); | |
172 | if (fd >= 0) { | |
173 | char id[16]; | |
174 | if (read(fd, id, sizeof(id)) > 0) | |
175 | err = atoi(id); | |
176 | close(fd); | |
177 | } | |
178 | ||
179 | free(filename); | |
180 | return err; | |
181 | } | |
182 | ||
183 | int perf_evlist__add_tracepoints(struct perf_evlist *evlist, | |
184 | const char *tracepoints[], | |
185 | size_t nr_tracepoints) | |
186 | { | |
187 | int err; | |
188 | size_t i; | |
189 | struct perf_event_attr *attrs = zalloc(nr_tracepoints * sizeof(*attrs)); | |
190 | ||
191 | if (attrs == NULL) | |
192 | return -1; | |
193 | ||
194 | for (i = 0; i < nr_tracepoints; i++) { | |
195 | err = trace_event__id(tracepoints[i]); | |
196 | ||
197 | if (err < 0) | |
198 | goto out_free_attrs; | |
199 | ||
200 | attrs[i].type = PERF_TYPE_TRACEPOINT; | |
201 | attrs[i].config = err; | |
202 | attrs[i].sample_type = (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | | |
203 | PERF_SAMPLE_CPU); | |
204 | attrs[i].sample_period = 1; | |
205 | } | |
206 | ||
207 | err = perf_evlist__add_attrs(evlist, attrs, nr_tracepoints); | |
208 | out_free_attrs: | |
209 | free(attrs); | |
210 | return err; | |
211 | } | |
212 | ||
ee29be62 ACM |
213 | static struct perf_evsel * |
214 | perf_evlist__find_tracepoint_by_id(struct perf_evlist *evlist, int id) | |
215 | { | |
216 | struct perf_evsel *evsel; | |
217 | ||
218 | list_for_each_entry(evsel, &evlist->entries, node) { | |
219 | if (evsel->attr.type == PERF_TYPE_TRACEPOINT && | |
220 | (int)evsel->attr.config == id) | |
221 | return evsel; | |
222 | } | |
223 | ||
224 | return NULL; | |
225 | } | |
226 | ||
227 | int perf_evlist__set_tracepoints_handlers(struct perf_evlist *evlist, | |
228 | const struct perf_evsel_str_handler *assocs, | |
229 | size_t nr_assocs) | |
230 | { | |
231 | struct perf_evsel *evsel; | |
232 | int err; | |
233 | size_t i; | |
234 | ||
235 | for (i = 0; i < nr_assocs; i++) { | |
236 | err = trace_event__id(assocs[i].name); | |
237 | if (err < 0) | |
238 | goto out; | |
239 | ||
240 | evsel = perf_evlist__find_tracepoint_by_id(evlist, err); | |
241 | if (evsel == NULL) | |
242 | continue; | |
243 | ||
244 | err = -EEXIST; | |
245 | if (evsel->handler.func != NULL) | |
246 | goto out; | |
247 | evsel->handler.func = assocs[i].handler; | |
248 | } | |
249 | ||
250 | err = 0; | |
251 | out: | |
252 | return err; | |
253 | } | |
254 | ||
4152ab37 ACM |
255 | void perf_evlist__disable(struct perf_evlist *evlist) |
256 | { | |
257 | int cpu, thread; | |
258 | struct perf_evsel *pos; | |
259 | ||
260 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
261 | list_for_each_entry(pos, &evlist->entries, node) { | |
262 | for (thread = 0; thread < evlist->threads->nr; thread++) | |
263 | ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_DISABLE); | |
264 | } | |
265 | } | |
266 | } | |
267 | ||
764e16a3 DA |
268 | void perf_evlist__enable(struct perf_evlist *evlist) |
269 | { | |
270 | int cpu, thread; | |
271 | struct perf_evsel *pos; | |
272 | ||
273 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
274 | list_for_each_entry(pos, &evlist->entries, node) { | |
275 | for (thread = 0; thread < evlist->threads->nr; thread++) | |
276 | ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_ENABLE); | |
277 | } | |
278 | } | |
279 | } | |
280 | ||
806fb630 | 281 | static int perf_evlist__alloc_pollfd(struct perf_evlist *evlist) |
5c581041 | 282 | { |
7e2ed097 | 283 | int nfds = evlist->cpus->nr * evlist->threads->nr * evlist->nr_entries; |
5c581041 ACM |
284 | evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); |
285 | return evlist->pollfd != NULL ? 0 : -ENOMEM; | |
286 | } | |
70082dd9 ACM |
287 | |
288 | void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) | |
289 | { | |
290 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
291 | evlist->pollfd[evlist->nr_fds].fd = fd; | |
292 | evlist->pollfd[evlist->nr_fds].events = POLLIN; | |
293 | evlist->nr_fds++; | |
294 | } | |
70db7533 | 295 | |
a91e5431 ACM |
296 | static void perf_evlist__id_hash(struct perf_evlist *evlist, |
297 | struct perf_evsel *evsel, | |
298 | int cpu, int thread, u64 id) | |
3d3b5e95 ACM |
299 | { |
300 | int hash; | |
301 | struct perf_sample_id *sid = SID(evsel, cpu, thread); | |
302 | ||
303 | sid->id = id; | |
304 | sid->evsel = evsel; | |
305 | hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); | |
306 | hlist_add_head(&sid->node, &evlist->heads[hash]); | |
307 | } | |
308 | ||
a91e5431 ACM |
309 | void perf_evlist__id_add(struct perf_evlist *evlist, struct perf_evsel *evsel, |
310 | int cpu, int thread, u64 id) | |
311 | { | |
312 | perf_evlist__id_hash(evlist, evsel, cpu, thread, id); | |
313 | evsel->id[evsel->ids++] = id; | |
314 | } | |
315 | ||
316 | static int perf_evlist__id_add_fd(struct perf_evlist *evlist, | |
317 | struct perf_evsel *evsel, | |
318 | int cpu, int thread, int fd) | |
f8a95309 | 319 | { |
f8a95309 | 320 | u64 read_data[4] = { 0, }; |
3d3b5e95 | 321 | int id_idx = 1; /* The first entry is the counter value */ |
f8a95309 ACM |
322 | |
323 | if (!(evsel->attr.read_format & PERF_FORMAT_ID) || | |
324 | read(fd, &read_data, sizeof(read_data)) == -1) | |
325 | return -1; | |
326 | ||
327 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) | |
328 | ++id_idx; | |
329 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) | |
330 | ++id_idx; | |
331 | ||
a91e5431 | 332 | perf_evlist__id_add(evlist, evsel, cpu, thread, read_data[id_idx]); |
f8a95309 ACM |
333 | return 0; |
334 | } | |
335 | ||
70db7533 ACM |
336 | struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) |
337 | { | |
338 | struct hlist_head *head; | |
339 | struct hlist_node *pos; | |
340 | struct perf_sample_id *sid; | |
341 | int hash; | |
342 | ||
343 | if (evlist->nr_entries == 1) | |
344 | return list_entry(evlist->entries.next, struct perf_evsel, node); | |
345 | ||
346 | hash = hash_64(id, PERF_EVLIST__HLIST_BITS); | |
347 | head = &evlist->heads[hash]; | |
348 | ||
349 | hlist_for_each_entry(sid, pos, head, node) | |
350 | if (sid->id == id) | |
351 | return sid->evsel; | |
352 | return NULL; | |
353 | } | |
04391deb | 354 | |
aece948f | 355 | union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx) |
04391deb ACM |
356 | { |
357 | /* XXX Move this to perf.c, making it generally available */ | |
358 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); | |
aece948f | 359 | struct perf_mmap *md = &evlist->mmap[idx]; |
04391deb ACM |
360 | unsigned int head = perf_mmap__read_head(md); |
361 | unsigned int old = md->prev; | |
362 | unsigned char *data = md->base + page_size; | |
8115d60c | 363 | union perf_event *event = NULL; |
04391deb | 364 | |
7bb41152 | 365 | if (evlist->overwrite) { |
04391deb | 366 | /* |
7bb41152 ACM |
367 | * If we're further behind than half the buffer, there's a chance |
368 | * the writer will bite our tail and mess up the samples under us. | |
369 | * | |
370 | * If we somehow ended up ahead of the head, we got messed up. | |
371 | * | |
372 | * In either case, truncate and restart at head. | |
04391deb | 373 | */ |
7bb41152 ACM |
374 | int diff = head - old; |
375 | if (diff > md->mask / 2 || diff < 0) { | |
376 | fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); | |
377 | ||
378 | /* | |
379 | * head points to a known good entry, start there. | |
380 | */ | |
381 | old = head; | |
382 | } | |
04391deb ACM |
383 | } |
384 | ||
385 | if (old != head) { | |
386 | size_t size; | |
387 | ||
8115d60c | 388 | event = (union perf_event *)&data[old & md->mask]; |
04391deb ACM |
389 | size = event->header.size; |
390 | ||
391 | /* | |
392 | * Event straddles the mmap boundary -- header should always | |
393 | * be inside due to u64 alignment of output. | |
394 | */ | |
395 | if ((old & md->mask) + size != ((old + size) & md->mask)) { | |
396 | unsigned int offset = old; | |
397 | unsigned int len = min(sizeof(*event), size), cpy; | |
398 | void *dst = &evlist->event_copy; | |
399 | ||
400 | do { | |
401 | cpy = min(md->mask + 1 - (offset & md->mask), len); | |
402 | memcpy(dst, &data[offset & md->mask], cpy); | |
403 | offset += cpy; | |
404 | dst += cpy; | |
405 | len -= cpy; | |
406 | } while (len); | |
407 | ||
408 | event = &evlist->event_copy; | |
409 | } | |
410 | ||
411 | old += size; | |
412 | } | |
413 | ||
414 | md->prev = old; | |
7bb41152 ACM |
415 | |
416 | if (!evlist->overwrite) | |
417 | perf_mmap__write_tail(md, old); | |
418 | ||
04391deb ACM |
419 | return event; |
420 | } | |
f8a95309 | 421 | |
7e2ed097 | 422 | void perf_evlist__munmap(struct perf_evlist *evlist) |
f8a95309 | 423 | { |
aece948f | 424 | int i; |
f8a95309 | 425 | |
aece948f ACM |
426 | for (i = 0; i < evlist->nr_mmaps; i++) { |
427 | if (evlist->mmap[i].base != NULL) { | |
428 | munmap(evlist->mmap[i].base, evlist->mmap_len); | |
429 | evlist->mmap[i].base = NULL; | |
f8a95309 ACM |
430 | } |
431 | } | |
aece948f ACM |
432 | |
433 | free(evlist->mmap); | |
434 | evlist->mmap = NULL; | |
f8a95309 ACM |
435 | } |
436 | ||
806fb630 | 437 | static int perf_evlist__alloc_mmap(struct perf_evlist *evlist) |
f8a95309 | 438 | { |
aece948f ACM |
439 | evlist->nr_mmaps = evlist->cpus->nr; |
440 | if (evlist->cpus->map[0] == -1) | |
441 | evlist->nr_mmaps = evlist->threads->nr; | |
442 | evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap)); | |
f8a95309 ACM |
443 | return evlist->mmap != NULL ? 0 : -ENOMEM; |
444 | } | |
445 | ||
bccdaba0 | 446 | static int __perf_evlist__mmap(struct perf_evlist *evlist, |
aece948f | 447 | int idx, int prot, int mask, int fd) |
f8a95309 | 448 | { |
aece948f ACM |
449 | evlist->mmap[idx].prev = 0; |
450 | evlist->mmap[idx].mask = mask; | |
451 | evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot, | |
f8a95309 | 452 | MAP_SHARED, fd, 0); |
301b195d NE |
453 | if (evlist->mmap[idx].base == MAP_FAILED) { |
454 | evlist->mmap[idx].base = NULL; | |
f8a95309 | 455 | return -1; |
301b195d | 456 | } |
f8a95309 ACM |
457 | |
458 | perf_evlist__add_pollfd(evlist, fd); | |
459 | return 0; | |
460 | } | |
461 | ||
aece948f ACM |
462 | static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mask) |
463 | { | |
464 | struct perf_evsel *evsel; | |
465 | int cpu, thread; | |
466 | ||
467 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
468 | int output = -1; | |
469 | ||
470 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
471 | list_for_each_entry(evsel, &evlist->entries, node) { | |
472 | int fd = FD(evsel, cpu, thread); | |
473 | ||
474 | if (output == -1) { | |
475 | output = fd; | |
bccdaba0 | 476 | if (__perf_evlist__mmap(evlist, cpu, |
aece948f ACM |
477 | prot, mask, output) < 0) |
478 | goto out_unmap; | |
479 | } else { | |
480 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) | |
481 | goto out_unmap; | |
482 | } | |
483 | ||
484 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
485 | perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) | |
486 | goto out_unmap; | |
487 | } | |
488 | } | |
489 | } | |
490 | ||
491 | return 0; | |
492 | ||
493 | out_unmap: | |
494 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
495 | if (evlist->mmap[cpu].base != NULL) { | |
496 | munmap(evlist->mmap[cpu].base, evlist->mmap_len); | |
497 | evlist->mmap[cpu].base = NULL; | |
498 | } | |
499 | } | |
500 | return -1; | |
501 | } | |
502 | ||
503 | static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, int mask) | |
504 | { | |
505 | struct perf_evsel *evsel; | |
506 | int thread; | |
507 | ||
508 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
509 | int output = -1; | |
510 | ||
511 | list_for_each_entry(evsel, &evlist->entries, node) { | |
512 | int fd = FD(evsel, 0, thread); | |
513 | ||
514 | if (output == -1) { | |
515 | output = fd; | |
bccdaba0 | 516 | if (__perf_evlist__mmap(evlist, thread, |
aece948f ACM |
517 | prot, mask, output) < 0) |
518 | goto out_unmap; | |
519 | } else { | |
520 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) | |
521 | goto out_unmap; | |
522 | } | |
523 | ||
524 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
525 | perf_evlist__id_add_fd(evlist, evsel, 0, thread, fd) < 0) | |
526 | goto out_unmap; | |
527 | } | |
528 | } | |
529 | ||
530 | return 0; | |
531 | ||
532 | out_unmap: | |
533 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
534 | if (evlist->mmap[thread].base != NULL) { | |
535 | munmap(evlist->mmap[thread].base, evlist->mmap_len); | |
536 | evlist->mmap[thread].base = NULL; | |
537 | } | |
538 | } | |
539 | return -1; | |
540 | } | |
541 | ||
f8a95309 ACM |
542 | /** perf_evlist__mmap - Create per cpu maps to receive events |
543 | * | |
544 | * @evlist - list of events | |
f8a95309 ACM |
545 | * @pages - map length in pages |
546 | * @overwrite - overwrite older events? | |
547 | * | |
548 | * If overwrite is false the user needs to signal event consuption using: | |
549 | * | |
550 | * struct perf_mmap *m = &evlist->mmap[cpu]; | |
551 | * unsigned int head = perf_mmap__read_head(m); | |
552 | * | |
553 | * perf_mmap__write_tail(m, head) | |
7e2ed097 ACM |
554 | * |
555 | * Using perf_evlist__read_on_cpu does this automatically. | |
f8a95309 | 556 | */ |
50a682ce ACM |
557 | int perf_evlist__mmap(struct perf_evlist *evlist, unsigned int pages, |
558 | bool overwrite) | |
f8a95309 ACM |
559 | { |
560 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); | |
aece948f | 561 | struct perf_evsel *evsel; |
7e2ed097 ACM |
562 | const struct cpu_map *cpus = evlist->cpus; |
563 | const struct thread_map *threads = evlist->threads; | |
50a682ce ACM |
564 | int prot = PROT_READ | (overwrite ? 0 : PROT_WRITE), mask; |
565 | ||
566 | /* 512 kiB: default amount of unprivileged mlocked memory */ | |
567 | if (pages == UINT_MAX) | |
568 | pages = (512 * 1024) / page_size; | |
41d0d933 NE |
569 | else if (!is_power_of_2(pages)) |
570 | return -EINVAL; | |
50a682ce ACM |
571 | |
572 | mask = pages * page_size - 1; | |
f8a95309 | 573 | |
7e2ed097 | 574 | if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0) |
f8a95309 ACM |
575 | return -ENOMEM; |
576 | ||
7e2ed097 | 577 | if (evlist->pollfd == NULL && perf_evlist__alloc_pollfd(evlist) < 0) |
f8a95309 ACM |
578 | return -ENOMEM; |
579 | ||
580 | evlist->overwrite = overwrite; | |
581 | evlist->mmap_len = (pages + 1) * page_size; | |
f8a95309 ACM |
582 | |
583 | list_for_each_entry(evsel, &evlist->entries, node) { | |
584 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
a91e5431 | 585 | evsel->sample_id == NULL && |
f8a95309 ACM |
586 | perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) |
587 | return -ENOMEM; | |
f8a95309 ACM |
588 | } |
589 | ||
aece948f ACM |
590 | if (evlist->cpus->map[0] == -1) |
591 | return perf_evlist__mmap_per_thread(evlist, prot, mask); | |
f8a95309 | 592 | |
aece948f | 593 | return perf_evlist__mmap_per_cpu(evlist, prot, mask); |
f8a95309 | 594 | } |
7e2ed097 ACM |
595 | |
596 | int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, | |
0d37aa34 | 597 | pid_t target_tid, uid_t uid, const char *cpu_list) |
7e2ed097 | 598 | { |
0d37aa34 | 599 | evlist->threads = thread_map__new(target_pid, target_tid, uid); |
7e2ed097 ACM |
600 | |
601 | if (evlist->threads == NULL) | |
602 | return -1; | |
603 | ||
0d37aa34 | 604 | if (uid != UINT_MAX || (cpu_list == NULL && target_tid != -1)) |
7e2ed097 ACM |
605 | evlist->cpus = cpu_map__dummy_new(); |
606 | else | |
607 | evlist->cpus = cpu_map__new(cpu_list); | |
608 | ||
609 | if (evlist->cpus == NULL) | |
610 | goto out_delete_threads; | |
611 | ||
612 | return 0; | |
613 | ||
614 | out_delete_threads: | |
615 | thread_map__delete(evlist->threads); | |
616 | return -1; | |
617 | } | |
618 | ||
619 | void perf_evlist__delete_maps(struct perf_evlist *evlist) | |
620 | { | |
621 | cpu_map__delete(evlist->cpus); | |
622 | thread_map__delete(evlist->threads); | |
623 | evlist->cpus = NULL; | |
624 | evlist->threads = NULL; | |
625 | } | |
0a102479 FW |
626 | |
627 | int perf_evlist__set_filters(struct perf_evlist *evlist) | |
628 | { | |
629 | const struct thread_map *threads = evlist->threads; | |
630 | const struct cpu_map *cpus = evlist->cpus; | |
631 | struct perf_evsel *evsel; | |
632 | char *filter; | |
633 | int thread; | |
634 | int cpu; | |
635 | int err; | |
636 | int fd; | |
637 | ||
638 | list_for_each_entry(evsel, &evlist->entries, node) { | |
639 | filter = evsel->filter; | |
640 | if (!filter) | |
641 | continue; | |
642 | for (cpu = 0; cpu < cpus->nr; cpu++) { | |
643 | for (thread = 0; thread < threads->nr; thread++) { | |
644 | fd = FD(evsel, cpu, thread); | |
645 | err = ioctl(fd, PERF_EVENT_IOC_SET_FILTER, filter); | |
646 | if (err) | |
647 | return err; | |
648 | } | |
649 | } | |
650 | } | |
651 | ||
652 | return 0; | |
653 | } | |
74429964 | 654 | |
c2a70653 | 655 | bool perf_evlist__valid_sample_type(const struct perf_evlist *evlist) |
74429964 | 656 | { |
c2a70653 ACM |
657 | struct perf_evsel *pos, *first; |
658 | ||
659 | pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
660 | ||
661 | list_for_each_entry_continue(pos, &evlist->entries, node) { | |
662 | if (first->attr.sample_type != pos->attr.sample_type) | |
663 | return false; | |
74429964 FW |
664 | } |
665 | ||
c2a70653 | 666 | return true; |
74429964 FW |
667 | } |
668 | ||
c2a70653 ACM |
669 | u64 perf_evlist__sample_type(const struct perf_evlist *evlist) |
670 | { | |
671 | struct perf_evsel *first; | |
672 | ||
673 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
674 | return first->attr.sample_type; | |
675 | } | |
676 | ||
81e36bff ACM |
677 | u16 perf_evlist__id_hdr_size(const struct perf_evlist *evlist) |
678 | { | |
679 | struct perf_evsel *first; | |
680 | struct perf_sample *data; | |
681 | u64 sample_type; | |
682 | u16 size = 0; | |
683 | ||
684 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
685 | ||
686 | if (!first->attr.sample_id_all) | |
687 | goto out; | |
688 | ||
689 | sample_type = first->attr.sample_type; | |
690 | ||
691 | if (sample_type & PERF_SAMPLE_TID) | |
692 | size += sizeof(data->tid) * 2; | |
693 | ||
694 | if (sample_type & PERF_SAMPLE_TIME) | |
695 | size += sizeof(data->time); | |
696 | ||
697 | if (sample_type & PERF_SAMPLE_ID) | |
698 | size += sizeof(data->id); | |
699 | ||
700 | if (sample_type & PERF_SAMPLE_STREAM_ID) | |
701 | size += sizeof(data->stream_id); | |
702 | ||
703 | if (sample_type & PERF_SAMPLE_CPU) | |
704 | size += sizeof(data->cpu) * 2; | |
705 | out: | |
706 | return size; | |
707 | } | |
708 | ||
c2a70653 | 709 | bool perf_evlist__valid_sample_id_all(const struct perf_evlist *evlist) |
74429964 | 710 | { |
c2a70653 ACM |
711 | struct perf_evsel *pos, *first; |
712 | ||
713 | pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
714 | ||
715 | list_for_each_entry_continue(pos, &evlist->entries, node) { | |
716 | if (first->attr.sample_id_all != pos->attr.sample_id_all) | |
717 | return false; | |
74429964 FW |
718 | } |
719 | ||
c2a70653 ACM |
720 | return true; |
721 | } | |
722 | ||
723 | bool perf_evlist__sample_id_all(const struct perf_evlist *evlist) | |
724 | { | |
725 | struct perf_evsel *first; | |
726 | ||
727 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
728 | return first->attr.sample_id_all; | |
74429964 | 729 | } |
81cce8de ACM |
730 | |
731 | void perf_evlist__set_selected(struct perf_evlist *evlist, | |
732 | struct perf_evsel *evsel) | |
733 | { | |
734 | evlist->selected = evsel; | |
735 | } | |
727ab04e ACM |
736 | |
737 | int perf_evlist__open(struct perf_evlist *evlist, bool group) | |
738 | { | |
739 | struct perf_evsel *evsel, *first; | |
740 | int err, ncpus, nthreads; | |
741 | ||
742 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
743 | ||
744 | list_for_each_entry(evsel, &evlist->entries, node) { | |
745 | struct xyarray *group_fd = NULL; | |
746 | ||
747 | if (group && evsel != first) | |
748 | group_fd = first->fd; | |
749 | ||
750 | err = perf_evsel__open(evsel, evlist->cpus, evlist->threads, | |
751 | group, group_fd); | |
752 | if (err < 0) | |
753 | goto out_err; | |
754 | } | |
755 | ||
756 | return 0; | |
757 | out_err: | |
758 | ncpus = evlist->cpus ? evlist->cpus->nr : 1; | |
759 | nthreads = evlist->threads ? evlist->threads->nr : 1; | |
760 | ||
761 | list_for_each_entry_reverse(evsel, &evlist->entries, node) | |
762 | perf_evsel__close(evsel, ncpus, nthreads); | |
763 | ||
764 | return err; | |
765 | } | |
35b9d88e ACM |
766 | |
767 | int perf_evlist__prepare_workload(struct perf_evlist *evlist, | |
768 | struct perf_record_opts *opts, | |
769 | const char *argv[]) | |
770 | { | |
771 | int child_ready_pipe[2], go_pipe[2]; | |
772 | char bf; | |
773 | ||
774 | if (pipe(child_ready_pipe) < 0) { | |
775 | perror("failed to create 'ready' pipe"); | |
776 | return -1; | |
777 | } | |
778 | ||
779 | if (pipe(go_pipe) < 0) { | |
780 | perror("failed to create 'go' pipe"); | |
781 | goto out_close_ready_pipe; | |
782 | } | |
783 | ||
784 | evlist->workload.pid = fork(); | |
785 | if (evlist->workload.pid < 0) { | |
786 | perror("failed to fork"); | |
787 | goto out_close_pipes; | |
788 | } | |
789 | ||
790 | if (!evlist->workload.pid) { | |
791 | if (opts->pipe_output) | |
792 | dup2(2, 1); | |
793 | ||
794 | close(child_ready_pipe[0]); | |
795 | close(go_pipe[1]); | |
796 | fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC); | |
797 | ||
798 | /* | |
799 | * Do a dummy execvp to get the PLT entry resolved, | |
800 | * so we avoid the resolver overhead on the real | |
801 | * execvp call. | |
802 | */ | |
803 | execvp("", (char **)argv); | |
804 | ||
805 | /* | |
806 | * Tell the parent we're ready to go | |
807 | */ | |
808 | close(child_ready_pipe[1]); | |
809 | ||
810 | /* | |
811 | * Wait until the parent tells us to go. | |
812 | */ | |
813 | if (read(go_pipe[0], &bf, 1) == -1) | |
814 | perror("unable to read pipe"); | |
815 | ||
816 | execvp(argv[0], (char **)argv); | |
817 | ||
818 | perror(argv[0]); | |
819 | kill(getppid(), SIGUSR1); | |
820 | exit(-1); | |
821 | } | |
822 | ||
823 | if (!opts->system_wide && opts->target_tid == -1 && opts->target_pid == -1) | |
824 | evlist->threads->map[0] = evlist->workload.pid; | |
825 | ||
826 | close(child_ready_pipe[1]); | |
827 | close(go_pipe[0]); | |
828 | /* | |
829 | * wait for child to settle | |
830 | */ | |
831 | if (read(child_ready_pipe[0], &bf, 1) == -1) { | |
832 | perror("unable to read pipe"); | |
833 | goto out_close_pipes; | |
834 | } | |
835 | ||
836 | evlist->workload.cork_fd = go_pipe[1]; | |
837 | close(child_ready_pipe[0]); | |
838 | return 0; | |
839 | ||
840 | out_close_pipes: | |
841 | close(go_pipe[0]); | |
842 | close(go_pipe[1]); | |
843 | out_close_ready_pipe: | |
844 | close(child_ready_pipe[0]); | |
845 | close(child_ready_pipe[1]); | |
846 | return -1; | |
847 | } | |
848 | ||
849 | int perf_evlist__start_workload(struct perf_evlist *evlist) | |
850 | { | |
851 | if (evlist->workload.cork_fd > 0) { | |
852 | /* | |
853 | * Remove the cork, let it rip! | |
854 | */ | |
855 | return close(evlist->workload.cork_fd); | |
856 | } | |
857 | ||
858 | return 0; | |
859 | } |