]>
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" | |
12864b31 | 14 | #include "target.h" |
361c99a6 ACM |
15 | #include "evlist.h" |
16 | #include "evsel.h" | |
35b9d88e | 17 | #include <unistd.h> |
361c99a6 | 18 | |
50d08e47 ACM |
19 | #include "parse-events.h" |
20 | ||
f8a95309 ACM |
21 | #include <sys/mman.h> |
22 | ||
70db7533 ACM |
23 | #include <linux/bitops.h> |
24 | #include <linux/hash.h> | |
25 | ||
f8a95309 | 26 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
a91e5431 | 27 | #define SID(e, x, y) xyarray__entry(e->sample_id, x, y) |
f8a95309 | 28 | |
7e2ed097 ACM |
29 | void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus, |
30 | struct thread_map *threads) | |
ef1d1af2 ACM |
31 | { |
32 | int i; | |
33 | ||
34 | for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) | |
35 | INIT_HLIST_HEAD(&evlist->heads[i]); | |
36 | INIT_LIST_HEAD(&evlist->entries); | |
7e2ed097 | 37 | perf_evlist__set_maps(evlist, cpus, threads); |
35b9d88e | 38 | evlist->workload.pid = -1; |
ef1d1af2 ACM |
39 | } |
40 | ||
7e2ed097 ACM |
41 | struct perf_evlist *perf_evlist__new(struct cpu_map *cpus, |
42 | struct thread_map *threads) | |
361c99a6 ACM |
43 | { |
44 | struct perf_evlist *evlist = zalloc(sizeof(*evlist)); | |
45 | ||
ef1d1af2 | 46 | if (evlist != NULL) |
7e2ed097 | 47 | perf_evlist__init(evlist, cpus, threads); |
361c99a6 ACM |
48 | |
49 | return evlist; | |
50 | } | |
51 | ||
0f82ebc4 ACM |
52 | void perf_evlist__config_attrs(struct perf_evlist *evlist, |
53 | struct perf_record_opts *opts) | |
54 | { | |
5090c6ae | 55 | struct perf_evsel *evsel, *first; |
0f82ebc4 ACM |
56 | |
57 | if (evlist->cpus->map[0] < 0) | |
58 | opts->no_inherit = true; | |
59 | ||
5090c6ae NK |
60 | first = list_entry(evlist->entries.next, struct perf_evsel, node); |
61 | ||
0f82ebc4 | 62 | list_for_each_entry(evsel, &evlist->entries, node) { |
5090c6ae | 63 | perf_evsel__config(evsel, opts, first); |
0f82ebc4 ACM |
64 | |
65 | if (evlist->nr_entries > 1) | |
66 | evsel->attr.sample_type |= PERF_SAMPLE_ID; | |
67 | } | |
68 | } | |
69 | ||
361c99a6 ACM |
70 | static void perf_evlist__purge(struct perf_evlist *evlist) |
71 | { | |
72 | struct perf_evsel *pos, *n; | |
73 | ||
74 | list_for_each_entry_safe(pos, n, &evlist->entries, node) { | |
75 | list_del_init(&pos->node); | |
76 | perf_evsel__delete(pos); | |
77 | } | |
78 | ||
79 | evlist->nr_entries = 0; | |
80 | } | |
81 | ||
ef1d1af2 | 82 | void perf_evlist__exit(struct perf_evlist *evlist) |
361c99a6 | 83 | { |
70db7533 | 84 | free(evlist->mmap); |
5c581041 | 85 | free(evlist->pollfd); |
ef1d1af2 ACM |
86 | evlist->mmap = NULL; |
87 | evlist->pollfd = NULL; | |
88 | } | |
89 | ||
90 | void perf_evlist__delete(struct perf_evlist *evlist) | |
91 | { | |
92 | perf_evlist__purge(evlist); | |
93 | perf_evlist__exit(evlist); | |
361c99a6 ACM |
94 | free(evlist); |
95 | } | |
96 | ||
97 | void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry) | |
98 | { | |
99 | list_add_tail(&entry->node, &evlist->entries); | |
100 | ++evlist->nr_entries; | |
101 | } | |
102 | ||
0529bc1f JO |
103 | void perf_evlist__splice_list_tail(struct perf_evlist *evlist, |
104 | struct list_head *list, | |
105 | int nr_entries) | |
50d08e47 ACM |
106 | { |
107 | list_splice_tail(list, &evlist->entries); | |
108 | evlist->nr_entries += nr_entries; | |
109 | } | |
110 | ||
361c99a6 ACM |
111 | int perf_evlist__add_default(struct perf_evlist *evlist) |
112 | { | |
113 | struct perf_event_attr attr = { | |
114 | .type = PERF_TYPE_HARDWARE, | |
115 | .config = PERF_COUNT_HW_CPU_CYCLES, | |
116 | }; | |
1aed2671 JR |
117 | struct perf_evsel *evsel; |
118 | ||
119 | event_attr_init(&attr); | |
361c99a6 | 120 | |
1aed2671 | 121 | evsel = perf_evsel__new(&attr, 0); |
361c99a6 | 122 | if (evsel == NULL) |
cc2d86b0 SE |
123 | goto error; |
124 | ||
125 | /* use strdup() because free(evsel) assumes name is allocated */ | |
126 | evsel->name = strdup("cycles"); | |
127 | if (!evsel->name) | |
128 | goto error_free; | |
361c99a6 ACM |
129 | |
130 | perf_evlist__add(evlist, evsel); | |
131 | return 0; | |
cc2d86b0 SE |
132 | error_free: |
133 | perf_evsel__delete(evsel); | |
134 | error: | |
135 | return -ENOMEM; | |
361c99a6 | 136 | } |
5c581041 | 137 | |
50d08e47 ACM |
138 | int perf_evlist__add_attrs(struct perf_evlist *evlist, |
139 | struct perf_event_attr *attrs, size_t nr_attrs) | |
140 | { | |
141 | struct perf_evsel *evsel, *n; | |
142 | LIST_HEAD(head); | |
143 | size_t i; | |
144 | ||
145 | for (i = 0; i < nr_attrs; i++) { | |
146 | evsel = perf_evsel__new(attrs + i, evlist->nr_entries + i); | |
147 | if (evsel == NULL) | |
148 | goto out_delete_partial_list; | |
149 | list_add_tail(&evsel->node, &head); | |
150 | } | |
151 | ||
152 | perf_evlist__splice_list_tail(evlist, &head, nr_attrs); | |
153 | ||
154 | return 0; | |
155 | ||
156 | out_delete_partial_list: | |
157 | list_for_each_entry_safe(evsel, n, &head, node) | |
158 | perf_evsel__delete(evsel); | |
159 | return -1; | |
160 | } | |
161 | ||
a8c9ae18 ACM |
162 | static int trace_event__id(const char *evname) |
163 | { | |
164 | char *filename, *colon; | |
165 | int err = -1, fd; | |
166 | ||
167 | if (asprintf(&filename, "%s/%s/id", tracing_events_path, evname) < 0) | |
168 | return -1; | |
169 | ||
170 | colon = strrchr(filename, ':'); | |
171 | if (colon != NULL) | |
172 | *colon = '/'; | |
173 | ||
174 | fd = open(filename, O_RDONLY); | |
175 | if (fd >= 0) { | |
176 | char id[16]; | |
177 | if (read(fd, id, sizeof(id)) > 0) | |
178 | err = atoi(id); | |
179 | close(fd); | |
180 | } | |
181 | ||
182 | free(filename); | |
183 | return err; | |
184 | } | |
185 | ||
186 | int perf_evlist__add_tracepoints(struct perf_evlist *evlist, | |
187 | const char *tracepoints[], | |
188 | size_t nr_tracepoints) | |
189 | { | |
190 | int err; | |
191 | size_t i; | |
192 | struct perf_event_attr *attrs = zalloc(nr_tracepoints * sizeof(*attrs)); | |
193 | ||
194 | if (attrs == NULL) | |
195 | return -1; | |
196 | ||
197 | for (i = 0; i < nr_tracepoints; i++) { | |
198 | err = trace_event__id(tracepoints[i]); | |
199 | ||
200 | if (err < 0) | |
201 | goto out_free_attrs; | |
202 | ||
203 | attrs[i].type = PERF_TYPE_TRACEPOINT; | |
204 | attrs[i].config = err; | |
205 | attrs[i].sample_type = (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | | |
206 | PERF_SAMPLE_CPU); | |
207 | attrs[i].sample_period = 1; | |
208 | } | |
209 | ||
210 | err = perf_evlist__add_attrs(evlist, attrs, nr_tracepoints); | |
211 | out_free_attrs: | |
212 | free(attrs); | |
213 | return err; | |
214 | } | |
215 | ||
ee29be62 ACM |
216 | static struct perf_evsel * |
217 | perf_evlist__find_tracepoint_by_id(struct perf_evlist *evlist, int id) | |
218 | { | |
219 | struct perf_evsel *evsel; | |
220 | ||
221 | list_for_each_entry(evsel, &evlist->entries, node) { | |
222 | if (evsel->attr.type == PERF_TYPE_TRACEPOINT && | |
223 | (int)evsel->attr.config == id) | |
224 | return evsel; | |
225 | } | |
226 | ||
227 | return NULL; | |
228 | } | |
229 | ||
230 | int perf_evlist__set_tracepoints_handlers(struct perf_evlist *evlist, | |
231 | const struct perf_evsel_str_handler *assocs, | |
232 | size_t nr_assocs) | |
233 | { | |
234 | struct perf_evsel *evsel; | |
235 | int err; | |
236 | size_t i; | |
237 | ||
238 | for (i = 0; i < nr_assocs; i++) { | |
239 | err = trace_event__id(assocs[i].name); | |
240 | if (err < 0) | |
241 | goto out; | |
242 | ||
243 | evsel = perf_evlist__find_tracepoint_by_id(evlist, err); | |
244 | if (evsel == NULL) | |
245 | continue; | |
246 | ||
247 | err = -EEXIST; | |
248 | if (evsel->handler.func != NULL) | |
249 | goto out; | |
250 | evsel->handler.func = assocs[i].handler; | |
251 | } | |
252 | ||
253 | err = 0; | |
254 | out: | |
255 | return err; | |
256 | } | |
257 | ||
4152ab37 ACM |
258 | void perf_evlist__disable(struct perf_evlist *evlist) |
259 | { | |
260 | int cpu, thread; | |
261 | struct perf_evsel *pos; | |
262 | ||
263 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
264 | list_for_each_entry(pos, &evlist->entries, node) { | |
265 | for (thread = 0; thread < evlist->threads->nr; thread++) | |
266 | ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_DISABLE); | |
267 | } | |
268 | } | |
269 | } | |
270 | ||
764e16a3 DA |
271 | void perf_evlist__enable(struct perf_evlist *evlist) |
272 | { | |
273 | int cpu, thread; | |
274 | struct perf_evsel *pos; | |
275 | ||
276 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
277 | list_for_each_entry(pos, &evlist->entries, node) { | |
278 | for (thread = 0; thread < evlist->threads->nr; thread++) | |
279 | ioctl(FD(pos, cpu, thread), PERF_EVENT_IOC_ENABLE); | |
280 | } | |
281 | } | |
282 | } | |
283 | ||
806fb630 | 284 | static int perf_evlist__alloc_pollfd(struct perf_evlist *evlist) |
5c581041 | 285 | { |
7e2ed097 | 286 | int nfds = evlist->cpus->nr * evlist->threads->nr * evlist->nr_entries; |
5c581041 ACM |
287 | evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); |
288 | return evlist->pollfd != NULL ? 0 : -ENOMEM; | |
289 | } | |
70082dd9 ACM |
290 | |
291 | void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) | |
292 | { | |
293 | fcntl(fd, F_SETFL, O_NONBLOCK); | |
294 | evlist->pollfd[evlist->nr_fds].fd = fd; | |
295 | evlist->pollfd[evlist->nr_fds].events = POLLIN; | |
296 | evlist->nr_fds++; | |
297 | } | |
70db7533 | 298 | |
a91e5431 ACM |
299 | static void perf_evlist__id_hash(struct perf_evlist *evlist, |
300 | struct perf_evsel *evsel, | |
301 | int cpu, int thread, u64 id) | |
3d3b5e95 ACM |
302 | { |
303 | int hash; | |
304 | struct perf_sample_id *sid = SID(evsel, cpu, thread); | |
305 | ||
306 | sid->id = id; | |
307 | sid->evsel = evsel; | |
308 | hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); | |
309 | hlist_add_head(&sid->node, &evlist->heads[hash]); | |
310 | } | |
311 | ||
a91e5431 ACM |
312 | void perf_evlist__id_add(struct perf_evlist *evlist, struct perf_evsel *evsel, |
313 | int cpu, int thread, u64 id) | |
314 | { | |
315 | perf_evlist__id_hash(evlist, evsel, cpu, thread, id); | |
316 | evsel->id[evsel->ids++] = id; | |
317 | } | |
318 | ||
319 | static int perf_evlist__id_add_fd(struct perf_evlist *evlist, | |
320 | struct perf_evsel *evsel, | |
321 | int cpu, int thread, int fd) | |
f8a95309 | 322 | { |
f8a95309 | 323 | u64 read_data[4] = { 0, }; |
3d3b5e95 | 324 | int id_idx = 1; /* The first entry is the counter value */ |
f8a95309 ACM |
325 | |
326 | if (!(evsel->attr.read_format & PERF_FORMAT_ID) || | |
327 | read(fd, &read_data, sizeof(read_data)) == -1) | |
328 | return -1; | |
329 | ||
330 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) | |
331 | ++id_idx; | |
332 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) | |
333 | ++id_idx; | |
334 | ||
a91e5431 | 335 | perf_evlist__id_add(evlist, evsel, cpu, thread, read_data[id_idx]); |
f8a95309 ACM |
336 | return 0; |
337 | } | |
338 | ||
70db7533 ACM |
339 | struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) |
340 | { | |
341 | struct hlist_head *head; | |
342 | struct hlist_node *pos; | |
343 | struct perf_sample_id *sid; | |
344 | int hash; | |
345 | ||
346 | if (evlist->nr_entries == 1) | |
347 | return list_entry(evlist->entries.next, struct perf_evsel, node); | |
348 | ||
349 | hash = hash_64(id, PERF_EVLIST__HLIST_BITS); | |
350 | head = &evlist->heads[hash]; | |
351 | ||
352 | hlist_for_each_entry(sid, pos, head, node) | |
353 | if (sid->id == id) | |
354 | return sid->evsel; | |
30e68bcc NK |
355 | |
356 | if (!perf_evlist__sample_id_all(evlist)) | |
357 | return list_entry(evlist->entries.next, struct perf_evsel, node); | |
358 | ||
70db7533 ACM |
359 | return NULL; |
360 | } | |
04391deb | 361 | |
aece948f | 362 | union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx) |
04391deb ACM |
363 | { |
364 | /* XXX Move this to perf.c, making it generally available */ | |
365 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); | |
aece948f | 366 | struct perf_mmap *md = &evlist->mmap[idx]; |
04391deb ACM |
367 | unsigned int head = perf_mmap__read_head(md); |
368 | unsigned int old = md->prev; | |
369 | unsigned char *data = md->base + page_size; | |
8115d60c | 370 | union perf_event *event = NULL; |
04391deb | 371 | |
7bb41152 | 372 | if (evlist->overwrite) { |
04391deb | 373 | /* |
7bb41152 ACM |
374 | * If we're further behind than half the buffer, there's a chance |
375 | * the writer will bite our tail and mess up the samples under us. | |
376 | * | |
377 | * If we somehow ended up ahead of the head, we got messed up. | |
378 | * | |
379 | * In either case, truncate and restart at head. | |
04391deb | 380 | */ |
7bb41152 ACM |
381 | int diff = head - old; |
382 | if (diff > md->mask / 2 || diff < 0) { | |
383 | fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); | |
384 | ||
385 | /* | |
386 | * head points to a known good entry, start there. | |
387 | */ | |
388 | old = head; | |
389 | } | |
04391deb ACM |
390 | } |
391 | ||
392 | if (old != head) { | |
393 | size_t size; | |
394 | ||
8115d60c | 395 | event = (union perf_event *)&data[old & md->mask]; |
04391deb ACM |
396 | size = event->header.size; |
397 | ||
398 | /* | |
399 | * Event straddles the mmap boundary -- header should always | |
400 | * be inside due to u64 alignment of output. | |
401 | */ | |
402 | if ((old & md->mask) + size != ((old + size) & md->mask)) { | |
403 | unsigned int offset = old; | |
404 | unsigned int len = min(sizeof(*event), size), cpy; | |
405 | void *dst = &evlist->event_copy; | |
406 | ||
407 | do { | |
408 | cpy = min(md->mask + 1 - (offset & md->mask), len); | |
409 | memcpy(dst, &data[offset & md->mask], cpy); | |
410 | offset += cpy; | |
411 | dst += cpy; | |
412 | len -= cpy; | |
413 | } while (len); | |
414 | ||
415 | event = &evlist->event_copy; | |
416 | } | |
417 | ||
418 | old += size; | |
419 | } | |
420 | ||
421 | md->prev = old; | |
7bb41152 ACM |
422 | |
423 | if (!evlist->overwrite) | |
424 | perf_mmap__write_tail(md, old); | |
425 | ||
04391deb ACM |
426 | return event; |
427 | } | |
f8a95309 | 428 | |
7e2ed097 | 429 | void perf_evlist__munmap(struct perf_evlist *evlist) |
f8a95309 | 430 | { |
aece948f | 431 | int i; |
f8a95309 | 432 | |
aece948f ACM |
433 | for (i = 0; i < evlist->nr_mmaps; i++) { |
434 | if (evlist->mmap[i].base != NULL) { | |
435 | munmap(evlist->mmap[i].base, evlist->mmap_len); | |
436 | evlist->mmap[i].base = NULL; | |
f8a95309 ACM |
437 | } |
438 | } | |
aece948f ACM |
439 | |
440 | free(evlist->mmap); | |
441 | evlist->mmap = NULL; | |
f8a95309 ACM |
442 | } |
443 | ||
806fb630 | 444 | static int perf_evlist__alloc_mmap(struct perf_evlist *evlist) |
f8a95309 | 445 | { |
aece948f ACM |
446 | evlist->nr_mmaps = evlist->cpus->nr; |
447 | if (evlist->cpus->map[0] == -1) | |
448 | evlist->nr_mmaps = evlist->threads->nr; | |
449 | evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap)); | |
f8a95309 ACM |
450 | return evlist->mmap != NULL ? 0 : -ENOMEM; |
451 | } | |
452 | ||
bccdaba0 | 453 | static int __perf_evlist__mmap(struct perf_evlist *evlist, |
aece948f | 454 | int idx, int prot, int mask, int fd) |
f8a95309 | 455 | { |
aece948f ACM |
456 | evlist->mmap[idx].prev = 0; |
457 | evlist->mmap[idx].mask = mask; | |
458 | evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot, | |
f8a95309 | 459 | MAP_SHARED, fd, 0); |
301b195d NE |
460 | if (evlist->mmap[idx].base == MAP_FAILED) { |
461 | evlist->mmap[idx].base = NULL; | |
f8a95309 | 462 | return -1; |
301b195d | 463 | } |
f8a95309 ACM |
464 | |
465 | perf_evlist__add_pollfd(evlist, fd); | |
466 | return 0; | |
467 | } | |
468 | ||
aece948f ACM |
469 | static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mask) |
470 | { | |
471 | struct perf_evsel *evsel; | |
472 | int cpu, thread; | |
473 | ||
474 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
475 | int output = -1; | |
476 | ||
477 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
478 | list_for_each_entry(evsel, &evlist->entries, node) { | |
479 | int fd = FD(evsel, cpu, thread); | |
480 | ||
481 | if (output == -1) { | |
482 | output = fd; | |
bccdaba0 | 483 | if (__perf_evlist__mmap(evlist, cpu, |
aece948f ACM |
484 | prot, mask, output) < 0) |
485 | goto out_unmap; | |
486 | } else { | |
487 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) | |
488 | goto out_unmap; | |
489 | } | |
490 | ||
491 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
492 | perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) | |
493 | goto out_unmap; | |
494 | } | |
495 | } | |
496 | } | |
497 | ||
498 | return 0; | |
499 | ||
500 | out_unmap: | |
501 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { | |
502 | if (evlist->mmap[cpu].base != NULL) { | |
503 | munmap(evlist->mmap[cpu].base, evlist->mmap_len); | |
504 | evlist->mmap[cpu].base = NULL; | |
505 | } | |
506 | } | |
507 | return -1; | |
508 | } | |
509 | ||
510 | static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, int mask) | |
511 | { | |
512 | struct perf_evsel *evsel; | |
513 | int thread; | |
514 | ||
515 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
516 | int output = -1; | |
517 | ||
518 | list_for_each_entry(evsel, &evlist->entries, node) { | |
519 | int fd = FD(evsel, 0, thread); | |
520 | ||
521 | if (output == -1) { | |
522 | output = fd; | |
bccdaba0 | 523 | if (__perf_evlist__mmap(evlist, thread, |
aece948f ACM |
524 | prot, mask, output) < 0) |
525 | goto out_unmap; | |
526 | } else { | |
527 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) | |
528 | goto out_unmap; | |
529 | } | |
530 | ||
531 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
532 | perf_evlist__id_add_fd(evlist, evsel, 0, thread, fd) < 0) | |
533 | goto out_unmap; | |
534 | } | |
535 | } | |
536 | ||
537 | return 0; | |
538 | ||
539 | out_unmap: | |
540 | for (thread = 0; thread < evlist->threads->nr; thread++) { | |
541 | if (evlist->mmap[thread].base != NULL) { | |
542 | munmap(evlist->mmap[thread].base, evlist->mmap_len); | |
543 | evlist->mmap[thread].base = NULL; | |
544 | } | |
545 | } | |
546 | return -1; | |
547 | } | |
548 | ||
f8a95309 ACM |
549 | /** perf_evlist__mmap - Create per cpu maps to receive events |
550 | * | |
551 | * @evlist - list of events | |
f8a95309 ACM |
552 | * @pages - map length in pages |
553 | * @overwrite - overwrite older events? | |
554 | * | |
555 | * If overwrite is false the user needs to signal event consuption using: | |
556 | * | |
557 | * struct perf_mmap *m = &evlist->mmap[cpu]; | |
558 | * unsigned int head = perf_mmap__read_head(m); | |
559 | * | |
560 | * perf_mmap__write_tail(m, head) | |
7e2ed097 ACM |
561 | * |
562 | * Using perf_evlist__read_on_cpu does this automatically. | |
f8a95309 | 563 | */ |
50a682ce ACM |
564 | int perf_evlist__mmap(struct perf_evlist *evlist, unsigned int pages, |
565 | bool overwrite) | |
f8a95309 ACM |
566 | { |
567 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); | |
aece948f | 568 | struct perf_evsel *evsel; |
7e2ed097 ACM |
569 | const struct cpu_map *cpus = evlist->cpus; |
570 | const struct thread_map *threads = evlist->threads; | |
50a682ce ACM |
571 | int prot = PROT_READ | (overwrite ? 0 : PROT_WRITE), mask; |
572 | ||
573 | /* 512 kiB: default amount of unprivileged mlocked memory */ | |
574 | if (pages == UINT_MAX) | |
575 | pages = (512 * 1024) / page_size; | |
41d0d933 NE |
576 | else if (!is_power_of_2(pages)) |
577 | return -EINVAL; | |
50a682ce ACM |
578 | |
579 | mask = pages * page_size - 1; | |
f8a95309 | 580 | |
7e2ed097 | 581 | if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0) |
f8a95309 ACM |
582 | return -ENOMEM; |
583 | ||
7e2ed097 | 584 | if (evlist->pollfd == NULL && perf_evlist__alloc_pollfd(evlist) < 0) |
f8a95309 ACM |
585 | return -ENOMEM; |
586 | ||
587 | evlist->overwrite = overwrite; | |
588 | evlist->mmap_len = (pages + 1) * page_size; | |
f8a95309 ACM |
589 | |
590 | list_for_each_entry(evsel, &evlist->entries, node) { | |
591 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && | |
a91e5431 | 592 | evsel->sample_id == NULL && |
f8a95309 ACM |
593 | perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) |
594 | return -ENOMEM; | |
f8a95309 ACM |
595 | } |
596 | ||
aece948f ACM |
597 | if (evlist->cpus->map[0] == -1) |
598 | return perf_evlist__mmap_per_thread(evlist, prot, mask); | |
f8a95309 | 599 | |
aece948f | 600 | return perf_evlist__mmap_per_cpu(evlist, prot, mask); |
f8a95309 | 601 | } |
7e2ed097 | 602 | |
b809ac10 NK |
603 | int perf_evlist__create_maps(struct perf_evlist *evlist, |
604 | struct perf_target *target) | |
7e2ed097 | 605 | { |
b809ac10 NK |
606 | evlist->threads = thread_map__new_str(target->pid, target->tid, |
607 | target->uid); | |
7e2ed097 ACM |
608 | |
609 | if (evlist->threads == NULL) | |
610 | return -1; | |
611 | ||
879d77d0 | 612 | if (perf_target__has_task(target)) |
d67356e7 | 613 | evlist->cpus = cpu_map__dummy_new(); |
d1cb9fce NK |
614 | else if (!perf_target__has_cpu(target) && !target->uses_mmap) |
615 | evlist->cpus = cpu_map__dummy_new(); | |
879d77d0 NK |
616 | else |
617 | evlist->cpus = cpu_map__new(target->cpu_list); | |
7e2ed097 ACM |
618 | |
619 | if (evlist->cpus == NULL) | |
620 | goto out_delete_threads; | |
621 | ||
622 | return 0; | |
623 | ||
624 | out_delete_threads: | |
625 | thread_map__delete(evlist->threads); | |
626 | return -1; | |
627 | } | |
628 | ||
629 | void perf_evlist__delete_maps(struct perf_evlist *evlist) | |
630 | { | |
631 | cpu_map__delete(evlist->cpus); | |
632 | thread_map__delete(evlist->threads); | |
633 | evlist->cpus = NULL; | |
634 | evlist->threads = NULL; | |
635 | } | |
0a102479 FW |
636 | |
637 | int perf_evlist__set_filters(struct perf_evlist *evlist) | |
638 | { | |
639 | const struct thread_map *threads = evlist->threads; | |
640 | const struct cpu_map *cpus = evlist->cpus; | |
641 | struct perf_evsel *evsel; | |
642 | char *filter; | |
643 | int thread; | |
644 | int cpu; | |
645 | int err; | |
646 | int fd; | |
647 | ||
648 | list_for_each_entry(evsel, &evlist->entries, node) { | |
649 | filter = evsel->filter; | |
650 | if (!filter) | |
651 | continue; | |
652 | for (cpu = 0; cpu < cpus->nr; cpu++) { | |
653 | for (thread = 0; thread < threads->nr; thread++) { | |
654 | fd = FD(evsel, cpu, thread); | |
655 | err = ioctl(fd, PERF_EVENT_IOC_SET_FILTER, filter); | |
656 | if (err) | |
657 | return err; | |
658 | } | |
659 | } | |
660 | } | |
661 | ||
662 | return 0; | |
663 | } | |
74429964 | 664 | |
c2a70653 | 665 | bool perf_evlist__valid_sample_type(const struct perf_evlist *evlist) |
74429964 | 666 | { |
c2a70653 ACM |
667 | struct perf_evsel *pos, *first; |
668 | ||
669 | pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
670 | ||
671 | list_for_each_entry_continue(pos, &evlist->entries, node) { | |
672 | if (first->attr.sample_type != pos->attr.sample_type) | |
673 | return false; | |
74429964 FW |
674 | } |
675 | ||
c2a70653 | 676 | return true; |
74429964 FW |
677 | } |
678 | ||
c2a70653 ACM |
679 | u64 perf_evlist__sample_type(const struct perf_evlist *evlist) |
680 | { | |
681 | struct perf_evsel *first; | |
682 | ||
683 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
684 | return first->attr.sample_type; | |
685 | } | |
686 | ||
81e36bff ACM |
687 | u16 perf_evlist__id_hdr_size(const struct perf_evlist *evlist) |
688 | { | |
689 | struct perf_evsel *first; | |
690 | struct perf_sample *data; | |
691 | u64 sample_type; | |
692 | u16 size = 0; | |
693 | ||
694 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
695 | ||
696 | if (!first->attr.sample_id_all) | |
697 | goto out; | |
698 | ||
699 | sample_type = first->attr.sample_type; | |
700 | ||
701 | if (sample_type & PERF_SAMPLE_TID) | |
702 | size += sizeof(data->tid) * 2; | |
703 | ||
704 | if (sample_type & PERF_SAMPLE_TIME) | |
705 | size += sizeof(data->time); | |
706 | ||
707 | if (sample_type & PERF_SAMPLE_ID) | |
708 | size += sizeof(data->id); | |
709 | ||
710 | if (sample_type & PERF_SAMPLE_STREAM_ID) | |
711 | size += sizeof(data->stream_id); | |
712 | ||
713 | if (sample_type & PERF_SAMPLE_CPU) | |
714 | size += sizeof(data->cpu) * 2; | |
715 | out: | |
716 | return size; | |
717 | } | |
718 | ||
c2a70653 | 719 | bool perf_evlist__valid_sample_id_all(const struct perf_evlist *evlist) |
74429964 | 720 | { |
c2a70653 ACM |
721 | struct perf_evsel *pos, *first; |
722 | ||
723 | pos = first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
724 | ||
725 | list_for_each_entry_continue(pos, &evlist->entries, node) { | |
726 | if (first->attr.sample_id_all != pos->attr.sample_id_all) | |
727 | return false; | |
74429964 FW |
728 | } |
729 | ||
c2a70653 ACM |
730 | return true; |
731 | } | |
732 | ||
733 | bool perf_evlist__sample_id_all(const struct perf_evlist *evlist) | |
734 | { | |
735 | struct perf_evsel *first; | |
736 | ||
737 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
738 | return first->attr.sample_id_all; | |
74429964 | 739 | } |
81cce8de ACM |
740 | |
741 | void perf_evlist__set_selected(struct perf_evlist *evlist, | |
742 | struct perf_evsel *evsel) | |
743 | { | |
744 | evlist->selected = evsel; | |
745 | } | |
727ab04e ACM |
746 | |
747 | int perf_evlist__open(struct perf_evlist *evlist, bool group) | |
748 | { | |
749 | struct perf_evsel *evsel, *first; | |
750 | int err, ncpus, nthreads; | |
751 | ||
752 | first = list_entry(evlist->entries.next, struct perf_evsel, node); | |
753 | ||
754 | list_for_each_entry(evsel, &evlist->entries, node) { | |
755 | struct xyarray *group_fd = NULL; | |
756 | ||
757 | if (group && evsel != first) | |
758 | group_fd = first->fd; | |
759 | ||
760 | err = perf_evsel__open(evsel, evlist->cpus, evlist->threads, | |
761 | group, group_fd); | |
762 | if (err < 0) | |
763 | goto out_err; | |
764 | } | |
765 | ||
766 | return 0; | |
767 | out_err: | |
768 | ncpus = evlist->cpus ? evlist->cpus->nr : 1; | |
769 | nthreads = evlist->threads ? evlist->threads->nr : 1; | |
770 | ||
771 | list_for_each_entry_reverse(evsel, &evlist->entries, node) | |
772 | perf_evsel__close(evsel, ncpus, nthreads); | |
773 | ||
41c21a68 | 774 | errno = -err; |
727ab04e ACM |
775 | return err; |
776 | } | |
35b9d88e ACM |
777 | |
778 | int perf_evlist__prepare_workload(struct perf_evlist *evlist, | |
779 | struct perf_record_opts *opts, | |
780 | const char *argv[]) | |
781 | { | |
782 | int child_ready_pipe[2], go_pipe[2]; | |
783 | char bf; | |
784 | ||
785 | if (pipe(child_ready_pipe) < 0) { | |
786 | perror("failed to create 'ready' pipe"); | |
787 | return -1; | |
788 | } | |
789 | ||
790 | if (pipe(go_pipe) < 0) { | |
791 | perror("failed to create 'go' pipe"); | |
792 | goto out_close_ready_pipe; | |
793 | } | |
794 | ||
795 | evlist->workload.pid = fork(); | |
796 | if (evlist->workload.pid < 0) { | |
797 | perror("failed to fork"); | |
798 | goto out_close_pipes; | |
799 | } | |
800 | ||
801 | if (!evlist->workload.pid) { | |
802 | if (opts->pipe_output) | |
803 | dup2(2, 1); | |
804 | ||
805 | close(child_ready_pipe[0]); | |
806 | close(go_pipe[1]); | |
807 | fcntl(go_pipe[0], F_SETFD, FD_CLOEXEC); | |
808 | ||
809 | /* | |
810 | * Do a dummy execvp to get the PLT entry resolved, | |
811 | * so we avoid the resolver overhead on the real | |
812 | * execvp call. | |
813 | */ | |
814 | execvp("", (char **)argv); | |
815 | ||
816 | /* | |
817 | * Tell the parent we're ready to go | |
818 | */ | |
819 | close(child_ready_pipe[1]); | |
820 | ||
821 | /* | |
822 | * Wait until the parent tells us to go. | |
823 | */ | |
824 | if (read(go_pipe[0], &bf, 1) == -1) | |
825 | perror("unable to read pipe"); | |
826 | ||
827 | execvp(argv[0], (char **)argv); | |
828 | ||
829 | perror(argv[0]); | |
830 | kill(getppid(), SIGUSR1); | |
831 | exit(-1); | |
832 | } | |
833 | ||
d67356e7 | 834 | if (perf_target__none(&opts->target)) |
35b9d88e ACM |
835 | evlist->threads->map[0] = evlist->workload.pid; |
836 | ||
837 | close(child_ready_pipe[1]); | |
838 | close(go_pipe[0]); | |
839 | /* | |
840 | * wait for child to settle | |
841 | */ | |
842 | if (read(child_ready_pipe[0], &bf, 1) == -1) { | |
843 | perror("unable to read pipe"); | |
844 | goto out_close_pipes; | |
845 | } | |
846 | ||
847 | evlist->workload.cork_fd = go_pipe[1]; | |
848 | close(child_ready_pipe[0]); | |
849 | return 0; | |
850 | ||
851 | out_close_pipes: | |
852 | close(go_pipe[0]); | |
853 | close(go_pipe[1]); | |
854 | out_close_ready_pipe: | |
855 | close(child_ready_pipe[0]); | |
856 | close(child_ready_pipe[1]); | |
857 | return -1; | |
858 | } | |
859 | ||
860 | int perf_evlist__start_workload(struct perf_evlist *evlist) | |
861 | { | |
862 | if (evlist->workload.cork_fd > 0) { | |
863 | /* | |
864 | * Remove the cork, let it rip! | |
865 | */ | |
866 | return close(evlist->workload.cork_fd); | |
867 | } | |
868 | ||
869 | return 0; | |
870 | } |