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 | */ | |
9 | ||
936be503 DA |
10 | #include <byteswap.h> |
11 | #include "asm/bug.h" | |
69aad6f1 | 12 | #include "evsel.h" |
70082dd9 | 13 | #include "evlist.h" |
69aad6f1 | 14 | #include "util.h" |
86bd5e86 | 15 | #include "cpumap.h" |
fd78260b | 16 | #include "thread_map.h" |
69aad6f1 | 17 | |
c52b12ed | 18 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
727ab04e | 19 | #define GROUP_FD(group_fd, cpu) (*(int *)xyarray__entry(group_fd, cpu, 0)) |
c52b12ed | 20 | |
c2a70653 ACM |
21 | int __perf_evsel__sample_size(u64 sample_type) |
22 | { | |
23 | u64 mask = sample_type & PERF_SAMPLE_MASK; | |
24 | int size = 0; | |
25 | int i; | |
26 | ||
27 | for (i = 0; i < 64; i++) { | |
28 | if (mask & (1ULL << i)) | |
29 | size++; | |
30 | } | |
31 | ||
32 | size *= sizeof(u64); | |
33 | ||
34 | return size; | |
35 | } | |
36 | ||
0e2a5f10 ACM |
37 | static void hists__init(struct hists *hists) |
38 | { | |
39 | memset(hists, 0, sizeof(*hists)); | |
40 | hists->entries_in_array[0] = hists->entries_in_array[1] = RB_ROOT; | |
41 | hists->entries_in = &hists->entries_in_array[0]; | |
42 | hists->entries_collapsed = RB_ROOT; | |
43 | hists->entries = RB_ROOT; | |
44 | pthread_mutex_init(&hists->lock, NULL); | |
45 | } | |
46 | ||
ef1d1af2 ACM |
47 | void perf_evsel__init(struct perf_evsel *evsel, |
48 | struct perf_event_attr *attr, int idx) | |
49 | { | |
50 | evsel->idx = idx; | |
51 | evsel->attr = *attr; | |
52 | INIT_LIST_HEAD(&evsel->node); | |
1980c2eb | 53 | hists__init(&evsel->hists); |
ef1d1af2 ACM |
54 | } |
55 | ||
23a2f3ab | 56 | struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) |
69aad6f1 ACM |
57 | { |
58 | struct perf_evsel *evsel = zalloc(sizeof(*evsel)); | |
59 | ||
ef1d1af2 ACM |
60 | if (evsel != NULL) |
61 | perf_evsel__init(evsel, attr, idx); | |
69aad6f1 ACM |
62 | |
63 | return evsel; | |
64 | } | |
65 | ||
66 | int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads) | |
67 | { | |
4af4c955 | 68 | int cpu, thread; |
69aad6f1 | 69 | evsel->fd = xyarray__new(ncpus, nthreads, sizeof(int)); |
4af4c955 DA |
70 | |
71 | if (evsel->fd) { | |
72 | for (cpu = 0; cpu < ncpus; cpu++) { | |
73 | for (thread = 0; thread < nthreads; thread++) { | |
74 | FD(evsel, cpu, thread) = -1; | |
75 | } | |
76 | } | |
77 | } | |
78 | ||
69aad6f1 ACM |
79 | return evsel->fd != NULL ? 0 : -ENOMEM; |
80 | } | |
81 | ||
70db7533 ACM |
82 | int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads) |
83 | { | |
a91e5431 ACM |
84 | evsel->sample_id = xyarray__new(ncpus, nthreads, sizeof(struct perf_sample_id)); |
85 | if (evsel->sample_id == NULL) | |
86 | return -ENOMEM; | |
87 | ||
88 | evsel->id = zalloc(ncpus * nthreads * sizeof(u64)); | |
89 | if (evsel->id == NULL) { | |
90 | xyarray__delete(evsel->sample_id); | |
91 | evsel->sample_id = NULL; | |
92 | return -ENOMEM; | |
93 | } | |
94 | ||
95 | return 0; | |
70db7533 ACM |
96 | } |
97 | ||
c52b12ed ACM |
98 | int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus) |
99 | { | |
100 | evsel->counts = zalloc((sizeof(*evsel->counts) + | |
101 | (ncpus * sizeof(struct perf_counts_values)))); | |
102 | return evsel->counts != NULL ? 0 : -ENOMEM; | |
103 | } | |
104 | ||
69aad6f1 ACM |
105 | void perf_evsel__free_fd(struct perf_evsel *evsel) |
106 | { | |
107 | xyarray__delete(evsel->fd); | |
108 | evsel->fd = NULL; | |
109 | } | |
110 | ||
70db7533 ACM |
111 | void perf_evsel__free_id(struct perf_evsel *evsel) |
112 | { | |
a91e5431 ACM |
113 | xyarray__delete(evsel->sample_id); |
114 | evsel->sample_id = NULL; | |
115 | free(evsel->id); | |
70db7533 ACM |
116 | evsel->id = NULL; |
117 | } | |
118 | ||
c52b12ed ACM |
119 | void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) |
120 | { | |
121 | int cpu, thread; | |
122 | ||
123 | for (cpu = 0; cpu < ncpus; cpu++) | |
124 | for (thread = 0; thread < nthreads; ++thread) { | |
125 | close(FD(evsel, cpu, thread)); | |
126 | FD(evsel, cpu, thread) = -1; | |
127 | } | |
128 | } | |
129 | ||
ef1d1af2 | 130 | void perf_evsel__exit(struct perf_evsel *evsel) |
69aad6f1 ACM |
131 | { |
132 | assert(list_empty(&evsel->node)); | |
133 | xyarray__delete(evsel->fd); | |
a91e5431 ACM |
134 | xyarray__delete(evsel->sample_id); |
135 | free(evsel->id); | |
ef1d1af2 ACM |
136 | } |
137 | ||
138 | void perf_evsel__delete(struct perf_evsel *evsel) | |
139 | { | |
140 | perf_evsel__exit(evsel); | |
023695d9 | 141 | close_cgroup(evsel->cgrp); |
f0c55bcf | 142 | free(evsel->name); |
69aad6f1 ACM |
143 | free(evsel); |
144 | } | |
c52b12ed ACM |
145 | |
146 | int __perf_evsel__read_on_cpu(struct perf_evsel *evsel, | |
147 | int cpu, int thread, bool scale) | |
148 | { | |
149 | struct perf_counts_values count; | |
150 | size_t nv = scale ? 3 : 1; | |
151 | ||
152 | if (FD(evsel, cpu, thread) < 0) | |
153 | return -EINVAL; | |
154 | ||
4eed11d5 ACM |
155 | if (evsel->counts == NULL && perf_evsel__alloc_counts(evsel, cpu + 1) < 0) |
156 | return -ENOMEM; | |
157 | ||
c52b12ed ACM |
158 | if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0) |
159 | return -errno; | |
160 | ||
161 | if (scale) { | |
162 | if (count.run == 0) | |
163 | count.val = 0; | |
164 | else if (count.run < count.ena) | |
165 | count.val = (u64)((double)count.val * count.ena / count.run + 0.5); | |
166 | } else | |
167 | count.ena = count.run = 0; | |
168 | ||
169 | evsel->counts->cpu[cpu] = count; | |
170 | return 0; | |
171 | } | |
172 | ||
173 | int __perf_evsel__read(struct perf_evsel *evsel, | |
174 | int ncpus, int nthreads, bool scale) | |
175 | { | |
176 | size_t nv = scale ? 3 : 1; | |
177 | int cpu, thread; | |
178 | struct perf_counts_values *aggr = &evsel->counts->aggr, count; | |
179 | ||
52bcd994 | 180 | aggr->val = aggr->ena = aggr->run = 0; |
c52b12ed ACM |
181 | |
182 | for (cpu = 0; cpu < ncpus; cpu++) { | |
183 | for (thread = 0; thread < nthreads; thread++) { | |
184 | if (FD(evsel, cpu, thread) < 0) | |
185 | continue; | |
186 | ||
187 | if (readn(FD(evsel, cpu, thread), | |
188 | &count, nv * sizeof(u64)) < 0) | |
189 | return -errno; | |
190 | ||
191 | aggr->val += count.val; | |
192 | if (scale) { | |
193 | aggr->ena += count.ena; | |
194 | aggr->run += count.run; | |
195 | } | |
196 | } | |
197 | } | |
198 | ||
199 | evsel->counts->scaled = 0; | |
200 | if (scale) { | |
201 | if (aggr->run == 0) { | |
202 | evsel->counts->scaled = -1; | |
203 | aggr->val = 0; | |
204 | return 0; | |
205 | } | |
206 | ||
207 | if (aggr->run < aggr->ena) { | |
208 | evsel->counts->scaled = 1; | |
209 | aggr->val = (u64)((double)aggr->val * aggr->ena / aggr->run + 0.5); | |
210 | } | |
211 | } else | |
212 | aggr->ena = aggr->run = 0; | |
213 | ||
214 | return 0; | |
215 | } | |
48290609 | 216 | |
0252208e | 217 | static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, |
727ab04e ACM |
218 | struct thread_map *threads, bool group, |
219 | struct xyarray *group_fds) | |
48290609 | 220 | { |
0252208e | 221 | int cpu, thread; |
023695d9 | 222 | unsigned long flags = 0; |
727ab04e | 223 | int pid = -1, err; |
48290609 | 224 | |
0252208e ACM |
225 | if (evsel->fd == NULL && |
226 | perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0) | |
727ab04e | 227 | return -ENOMEM; |
4eed11d5 | 228 | |
023695d9 SE |
229 | if (evsel->cgrp) { |
230 | flags = PERF_FLAG_PID_CGROUP; | |
231 | pid = evsel->cgrp->fd; | |
232 | } | |
233 | ||
86bd5e86 | 234 | for (cpu = 0; cpu < cpus->nr; cpu++) { |
727ab04e | 235 | int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1; |
9d04f178 | 236 | |
0252208e | 237 | for (thread = 0; thread < threads->nr; thread++) { |
023695d9 SE |
238 | |
239 | if (!evsel->cgrp) | |
240 | pid = threads->map[thread]; | |
241 | ||
0252208e | 242 | FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, |
023695d9 | 243 | pid, |
f08199d3 | 244 | cpus->map[cpu], |
023695d9 | 245 | group_fd, flags); |
727ab04e ACM |
246 | if (FD(evsel, cpu, thread) < 0) { |
247 | err = -errno; | |
0252208e | 248 | goto out_close; |
727ab04e | 249 | } |
f08199d3 ACM |
250 | |
251 | if (group && group_fd == -1) | |
252 | group_fd = FD(evsel, cpu, thread); | |
0252208e | 253 | } |
48290609 ACM |
254 | } |
255 | ||
256 | return 0; | |
257 | ||
258 | out_close: | |
0252208e ACM |
259 | do { |
260 | while (--thread >= 0) { | |
261 | close(FD(evsel, cpu, thread)); | |
262 | FD(evsel, cpu, thread) = -1; | |
263 | } | |
264 | thread = threads->nr; | |
265 | } while (--cpu >= 0); | |
727ab04e ACM |
266 | return err; |
267 | } | |
268 | ||
269 | void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads) | |
270 | { | |
271 | if (evsel->fd == NULL) | |
272 | return; | |
273 | ||
274 | perf_evsel__close_fd(evsel, ncpus, nthreads); | |
275 | perf_evsel__free_fd(evsel); | |
276 | evsel->fd = NULL; | |
48290609 ACM |
277 | } |
278 | ||
0252208e ACM |
279 | static struct { |
280 | struct cpu_map map; | |
281 | int cpus[1]; | |
282 | } empty_cpu_map = { | |
283 | .map.nr = 1, | |
284 | .cpus = { -1, }, | |
285 | }; | |
286 | ||
287 | static struct { | |
288 | struct thread_map map; | |
289 | int threads[1]; | |
290 | } empty_thread_map = { | |
291 | .map.nr = 1, | |
292 | .threads = { -1, }, | |
293 | }; | |
294 | ||
f08199d3 | 295 | int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, |
727ab04e ACM |
296 | struct thread_map *threads, bool group, |
297 | struct xyarray *group_fd) | |
48290609 | 298 | { |
0252208e ACM |
299 | if (cpus == NULL) { |
300 | /* Work around old compiler warnings about strict aliasing */ | |
301 | cpus = &empty_cpu_map.map; | |
48290609 ACM |
302 | } |
303 | ||
0252208e ACM |
304 | if (threads == NULL) |
305 | threads = &empty_thread_map.map; | |
48290609 | 306 | |
727ab04e | 307 | return __perf_evsel__open(evsel, cpus, threads, group, group_fd); |
48290609 ACM |
308 | } |
309 | ||
f08199d3 | 310 | int perf_evsel__open_per_cpu(struct perf_evsel *evsel, |
727ab04e ACM |
311 | struct cpu_map *cpus, bool group, |
312 | struct xyarray *group_fd) | |
48290609 | 313 | { |
727ab04e ACM |
314 | return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group, |
315 | group_fd); | |
0252208e | 316 | } |
48290609 | 317 | |
f08199d3 | 318 | int perf_evsel__open_per_thread(struct perf_evsel *evsel, |
727ab04e ACM |
319 | struct thread_map *threads, bool group, |
320 | struct xyarray *group_fd) | |
0252208e | 321 | { |
727ab04e ACM |
322 | return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, |
323 | group_fd); | |
48290609 | 324 | } |
70082dd9 | 325 | |
8115d60c ACM |
326 | static int perf_event__parse_id_sample(const union perf_event *event, u64 type, |
327 | struct perf_sample *sample) | |
d0dd74e8 ACM |
328 | { |
329 | const u64 *array = event->sample.array; | |
330 | ||
331 | array += ((event->header.size - | |
332 | sizeof(event->header)) / sizeof(u64)) - 1; | |
333 | ||
334 | if (type & PERF_SAMPLE_CPU) { | |
335 | u32 *p = (u32 *)array; | |
336 | sample->cpu = *p; | |
337 | array--; | |
338 | } | |
339 | ||
340 | if (type & PERF_SAMPLE_STREAM_ID) { | |
341 | sample->stream_id = *array; | |
342 | array--; | |
343 | } | |
344 | ||
345 | if (type & PERF_SAMPLE_ID) { | |
346 | sample->id = *array; | |
347 | array--; | |
348 | } | |
349 | ||
350 | if (type & PERF_SAMPLE_TIME) { | |
351 | sample->time = *array; | |
352 | array--; | |
353 | } | |
354 | ||
355 | if (type & PERF_SAMPLE_TID) { | |
356 | u32 *p = (u32 *)array; | |
357 | sample->pid = p[0]; | |
358 | sample->tid = p[1]; | |
359 | } | |
360 | ||
361 | return 0; | |
362 | } | |
363 | ||
98e1da90 FW |
364 | static bool sample_overlap(const union perf_event *event, |
365 | const void *offset, u64 size) | |
366 | { | |
367 | const void *base = event; | |
368 | ||
369 | if (offset + size > base + event->header.size) | |
370 | return true; | |
371 | ||
372 | return false; | |
373 | } | |
374 | ||
8115d60c | 375 | int perf_event__parse_sample(const union perf_event *event, u64 type, |
a2854124 | 376 | int sample_size, bool sample_id_all, |
936be503 | 377 | struct perf_sample *data, bool swapped) |
d0dd74e8 ACM |
378 | { |
379 | const u64 *array; | |
380 | ||
936be503 DA |
381 | /* |
382 | * used for cross-endian analysis. See git commit 65014ab3 | |
383 | * for why this goofiness is needed. | |
384 | */ | |
385 | union { | |
386 | u64 val64; | |
387 | u32 val32[2]; | |
388 | } u; | |
389 | ||
390 | ||
d0dd74e8 ACM |
391 | data->cpu = data->pid = data->tid = -1; |
392 | data->stream_id = data->id = data->time = -1ULL; | |
393 | ||
394 | if (event->header.type != PERF_RECORD_SAMPLE) { | |
395 | if (!sample_id_all) | |
396 | return 0; | |
8115d60c | 397 | return perf_event__parse_id_sample(event, type, data); |
d0dd74e8 ACM |
398 | } |
399 | ||
400 | array = event->sample.array; | |
401 | ||
a2854124 FW |
402 | if (sample_size + sizeof(event->header) > event->header.size) |
403 | return -EFAULT; | |
404 | ||
d0dd74e8 ACM |
405 | if (type & PERF_SAMPLE_IP) { |
406 | data->ip = event->ip.ip; | |
407 | array++; | |
408 | } | |
409 | ||
410 | if (type & PERF_SAMPLE_TID) { | |
936be503 DA |
411 | u.val64 = *array; |
412 | if (swapped) { | |
413 | /* undo swap of u64, then swap on individual u32s */ | |
414 | u.val64 = bswap_64(u.val64); | |
415 | u.val32[0] = bswap_32(u.val32[0]); | |
416 | u.val32[1] = bswap_32(u.val32[1]); | |
417 | } | |
418 | ||
419 | data->pid = u.val32[0]; | |
420 | data->tid = u.val32[1]; | |
d0dd74e8 ACM |
421 | array++; |
422 | } | |
423 | ||
424 | if (type & PERF_SAMPLE_TIME) { | |
425 | data->time = *array; | |
426 | array++; | |
427 | } | |
428 | ||
7cec0922 | 429 | data->addr = 0; |
d0dd74e8 ACM |
430 | if (type & PERF_SAMPLE_ADDR) { |
431 | data->addr = *array; | |
432 | array++; | |
433 | } | |
434 | ||
435 | data->id = -1ULL; | |
436 | if (type & PERF_SAMPLE_ID) { | |
437 | data->id = *array; | |
438 | array++; | |
439 | } | |
440 | ||
441 | if (type & PERF_SAMPLE_STREAM_ID) { | |
442 | data->stream_id = *array; | |
443 | array++; | |
444 | } | |
445 | ||
446 | if (type & PERF_SAMPLE_CPU) { | |
936be503 DA |
447 | |
448 | u.val64 = *array; | |
449 | if (swapped) { | |
450 | /* undo swap of u64, then swap on individual u32s */ | |
451 | u.val64 = bswap_64(u.val64); | |
452 | u.val32[0] = bswap_32(u.val32[0]); | |
453 | } | |
454 | ||
455 | data->cpu = u.val32[0]; | |
d0dd74e8 ACM |
456 | array++; |
457 | } | |
458 | ||
459 | if (type & PERF_SAMPLE_PERIOD) { | |
460 | data->period = *array; | |
461 | array++; | |
462 | } | |
463 | ||
464 | if (type & PERF_SAMPLE_READ) { | |
465 | fprintf(stderr, "PERF_SAMPLE_READ is unsuported for now\n"); | |
466 | return -1; | |
467 | } | |
468 | ||
469 | if (type & PERF_SAMPLE_CALLCHAIN) { | |
98e1da90 FW |
470 | if (sample_overlap(event, array, sizeof(data->callchain->nr))) |
471 | return -EFAULT; | |
472 | ||
d0dd74e8 | 473 | data->callchain = (struct ip_callchain *)array; |
98e1da90 FW |
474 | |
475 | if (sample_overlap(event, array, data->callchain->nr)) | |
476 | return -EFAULT; | |
477 | ||
d0dd74e8 ACM |
478 | array += 1 + data->callchain->nr; |
479 | } | |
480 | ||
481 | if (type & PERF_SAMPLE_RAW) { | |
8e303f20 JO |
482 | const u64 *pdata; |
483 | ||
936be503 DA |
484 | u.val64 = *array; |
485 | if (WARN_ONCE(swapped, | |
486 | "Endianness of raw data not corrected!\n")) { | |
487 | /* undo swap of u64, then swap on individual u32s */ | |
488 | u.val64 = bswap_64(u.val64); | |
489 | u.val32[0] = bswap_32(u.val32[0]); | |
490 | u.val32[1] = bswap_32(u.val32[1]); | |
491 | } | |
98e1da90 FW |
492 | |
493 | if (sample_overlap(event, array, sizeof(u32))) | |
494 | return -EFAULT; | |
495 | ||
936be503 | 496 | data->raw_size = u.val32[0]; |
8e303f20 | 497 | pdata = (void *) array + sizeof(u32); |
98e1da90 | 498 | |
8e303f20 | 499 | if (sample_overlap(event, pdata, data->raw_size)) |
98e1da90 FW |
500 | return -EFAULT; |
501 | ||
8e303f20 | 502 | data->raw_data = (void *) pdata; |
d0dd74e8 ACM |
503 | } |
504 | ||
505 | return 0; | |
506 | } |