]>
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" |
12864b31 | 17 | #include "target.h" |
69aad6f1 | 18 | |
c52b12ed | 19 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
727ab04e | 20 | #define GROUP_FD(group_fd, cpu) (*(int *)xyarray__entry(group_fd, cpu, 0)) |
c52b12ed | 21 | |
c2a70653 ACM |
22 | int __perf_evsel__sample_size(u64 sample_type) |
23 | { | |
24 | u64 mask = sample_type & PERF_SAMPLE_MASK; | |
25 | int size = 0; | |
26 | int i; | |
27 | ||
28 | for (i = 0; i < 64; i++) { | |
29 | if (mask & (1ULL << i)) | |
30 | size++; | |
31 | } | |
32 | ||
33 | size *= sizeof(u64); | |
34 | ||
35 | return size; | |
36 | } | |
37 | ||
4bf9ce1b | 38 | void hists__init(struct hists *hists) |
0e2a5f10 ACM |
39 | { |
40 | memset(hists, 0, sizeof(*hists)); | |
41 | hists->entries_in_array[0] = hists->entries_in_array[1] = RB_ROOT; | |
42 | hists->entries_in = &hists->entries_in_array[0]; | |
43 | hists->entries_collapsed = RB_ROOT; | |
44 | hists->entries = RB_ROOT; | |
45 | pthread_mutex_init(&hists->lock, NULL); | |
46 | } | |
47 | ||
ef1d1af2 ACM |
48 | void perf_evsel__init(struct perf_evsel *evsel, |
49 | struct perf_event_attr *attr, int idx) | |
50 | { | |
51 | evsel->idx = idx; | |
52 | evsel->attr = *attr; | |
53 | INIT_LIST_HEAD(&evsel->node); | |
1980c2eb | 54 | hists__init(&evsel->hists); |
ef1d1af2 ACM |
55 | } |
56 | ||
23a2f3ab | 57 | struct perf_evsel *perf_evsel__new(struct perf_event_attr *attr, int idx) |
69aad6f1 ACM |
58 | { |
59 | struct perf_evsel *evsel = zalloc(sizeof(*evsel)); | |
60 | ||
ef1d1af2 ACM |
61 | if (evsel != NULL) |
62 | perf_evsel__init(evsel, attr, idx); | |
69aad6f1 ACM |
63 | |
64 | return evsel; | |
65 | } | |
66 | ||
c410431c ACM |
67 | static const char *perf_evsel__hw_names[PERF_COUNT_HW_MAX] = { |
68 | "cycles", | |
69 | "instructions", | |
70 | "cache-references", | |
71 | "cache-misses", | |
72 | "branches", | |
73 | "branch-misses", | |
74 | "bus-cycles", | |
75 | "stalled-cycles-frontend", | |
76 | "stalled-cycles-backend", | |
77 | "ref-cycles", | |
78 | }; | |
79 | ||
dd4f5223 | 80 | static const char *__perf_evsel__hw_name(u64 config) |
c410431c ACM |
81 | { |
82 | if (config < PERF_COUNT_HW_MAX && perf_evsel__hw_names[config]) | |
83 | return perf_evsel__hw_names[config]; | |
84 | ||
85 | return "unknown-hardware"; | |
86 | } | |
87 | ||
27f18617 | 88 | static int perf_evsel__add_modifiers(struct perf_evsel *evsel, char *bf, size_t size) |
c410431c | 89 | { |
27f18617 | 90 | int colon = 0, r = 0; |
c410431c | 91 | struct perf_event_attr *attr = &evsel->attr; |
c410431c ACM |
92 | bool exclude_guest_default = false; |
93 | ||
94 | #define MOD_PRINT(context, mod) do { \ | |
95 | if (!attr->exclude_##context) { \ | |
27f18617 | 96 | if (!colon) colon = ++r; \ |
c410431c ACM |
97 | r += scnprintf(bf + r, size - r, "%c", mod); \ |
98 | } } while(0) | |
99 | ||
100 | if (attr->exclude_kernel || attr->exclude_user || attr->exclude_hv) { | |
101 | MOD_PRINT(kernel, 'k'); | |
102 | MOD_PRINT(user, 'u'); | |
103 | MOD_PRINT(hv, 'h'); | |
104 | exclude_guest_default = true; | |
105 | } | |
106 | ||
107 | if (attr->precise_ip) { | |
108 | if (!colon) | |
27f18617 | 109 | colon = ++r; |
c410431c ACM |
110 | r += scnprintf(bf + r, size - r, "%.*s", attr->precise_ip, "ppp"); |
111 | exclude_guest_default = true; | |
112 | } | |
113 | ||
114 | if (attr->exclude_host || attr->exclude_guest == exclude_guest_default) { | |
115 | MOD_PRINT(host, 'H'); | |
116 | MOD_PRINT(guest, 'G'); | |
117 | } | |
118 | #undef MOD_PRINT | |
119 | if (colon) | |
27f18617 | 120 | bf[colon - 1] = ':'; |
c410431c ACM |
121 | return r; |
122 | } | |
123 | ||
27f18617 ACM |
124 | static int perf_evsel__hw_name(struct perf_evsel *evsel, char *bf, size_t size) |
125 | { | |
126 | int r = scnprintf(bf, size, "%s", __perf_evsel__hw_name(evsel->attr.config)); | |
127 | return r + perf_evsel__add_modifiers(evsel, bf + r, size - r); | |
128 | } | |
129 | ||
335c2f5d ACM |
130 | static const char *perf_evsel__sw_names[PERF_COUNT_SW_MAX] = { |
131 | "cpu-clock", | |
132 | "task-clock", | |
133 | "page-faults", | |
134 | "context-switches", | |
135 | "CPU-migrations", | |
136 | "minor-faults", | |
137 | "major-faults", | |
138 | "alignment-faults", | |
139 | "emulation-faults", | |
140 | }; | |
141 | ||
dd4f5223 | 142 | static const char *__perf_evsel__sw_name(u64 config) |
335c2f5d ACM |
143 | { |
144 | if (config < PERF_COUNT_SW_MAX && perf_evsel__sw_names[config]) | |
145 | return perf_evsel__sw_names[config]; | |
146 | return "unknown-software"; | |
147 | } | |
148 | ||
149 | static int perf_evsel__sw_name(struct perf_evsel *evsel, char *bf, size_t size) | |
150 | { | |
151 | int r = scnprintf(bf, size, "%s", __perf_evsel__sw_name(evsel->attr.config)); | |
152 | return r + perf_evsel__add_modifiers(evsel, bf + r, size - r); | |
153 | } | |
154 | ||
0b668bc9 ACM |
155 | const char *perf_evsel__hw_cache[PERF_COUNT_HW_CACHE_MAX] |
156 | [PERF_EVSEL__MAX_ALIASES] = { | |
157 | { "L1-dcache", "l1-d", "l1d", "L1-data", }, | |
158 | { "L1-icache", "l1-i", "l1i", "L1-instruction", }, | |
159 | { "LLC", "L2", }, | |
160 | { "dTLB", "d-tlb", "Data-TLB", }, | |
161 | { "iTLB", "i-tlb", "Instruction-TLB", }, | |
162 | { "branch", "branches", "bpu", "btb", "bpc", }, | |
163 | { "node", }, | |
164 | }; | |
165 | ||
166 | const char *perf_evsel__hw_cache_op[PERF_COUNT_HW_CACHE_OP_MAX] | |
167 | [PERF_EVSEL__MAX_ALIASES] = { | |
168 | { "load", "loads", "read", }, | |
169 | { "store", "stores", "write", }, | |
170 | { "prefetch", "prefetches", "speculative-read", "speculative-load", }, | |
171 | }; | |
172 | ||
173 | const char *perf_evsel__hw_cache_result[PERF_COUNT_HW_CACHE_RESULT_MAX] | |
174 | [PERF_EVSEL__MAX_ALIASES] = { | |
175 | { "refs", "Reference", "ops", "access", }, | |
176 | { "misses", "miss", }, | |
177 | }; | |
178 | ||
179 | #define C(x) PERF_COUNT_HW_CACHE_##x | |
180 | #define CACHE_READ (1 << C(OP_READ)) | |
181 | #define CACHE_WRITE (1 << C(OP_WRITE)) | |
182 | #define CACHE_PREFETCH (1 << C(OP_PREFETCH)) | |
183 | #define COP(x) (1 << x) | |
184 | ||
185 | /* | |
186 | * cache operartion stat | |
187 | * L1I : Read and prefetch only | |
188 | * ITLB and BPU : Read-only | |
189 | */ | |
190 | static unsigned long perf_evsel__hw_cache_stat[C(MAX)] = { | |
191 | [C(L1D)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), | |
192 | [C(L1I)] = (CACHE_READ | CACHE_PREFETCH), | |
193 | [C(LL)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), | |
194 | [C(DTLB)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), | |
195 | [C(ITLB)] = (CACHE_READ), | |
196 | [C(BPU)] = (CACHE_READ), | |
197 | [C(NODE)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), | |
198 | }; | |
199 | ||
200 | bool perf_evsel__is_cache_op_valid(u8 type, u8 op) | |
201 | { | |
202 | if (perf_evsel__hw_cache_stat[type] & COP(op)) | |
203 | return true; /* valid */ | |
204 | else | |
205 | return false; /* invalid */ | |
206 | } | |
207 | ||
208 | int __perf_evsel__hw_cache_type_op_res_name(u8 type, u8 op, u8 result, | |
209 | char *bf, size_t size) | |
210 | { | |
211 | if (result) { | |
212 | return scnprintf(bf, size, "%s-%s-%s", perf_evsel__hw_cache[type][0], | |
213 | perf_evsel__hw_cache_op[op][0], | |
214 | perf_evsel__hw_cache_result[result][0]); | |
215 | } | |
216 | ||
217 | return scnprintf(bf, size, "%s-%s", perf_evsel__hw_cache[type][0], | |
218 | perf_evsel__hw_cache_op[op][1]); | |
219 | } | |
220 | ||
dd4f5223 | 221 | static int __perf_evsel__hw_cache_name(u64 config, char *bf, size_t size) |
0b668bc9 ACM |
222 | { |
223 | u8 op, result, type = (config >> 0) & 0xff; | |
224 | const char *err = "unknown-ext-hardware-cache-type"; | |
225 | ||
226 | if (type > PERF_COUNT_HW_CACHE_MAX) | |
227 | goto out_err; | |
228 | ||
229 | op = (config >> 8) & 0xff; | |
230 | err = "unknown-ext-hardware-cache-op"; | |
231 | if (op > PERF_COUNT_HW_CACHE_OP_MAX) | |
232 | goto out_err; | |
233 | ||
234 | result = (config >> 16) & 0xff; | |
235 | err = "unknown-ext-hardware-cache-result"; | |
236 | if (result > PERF_COUNT_HW_CACHE_RESULT_MAX) | |
237 | goto out_err; | |
238 | ||
239 | err = "invalid-cache"; | |
240 | if (!perf_evsel__is_cache_op_valid(type, op)) | |
241 | goto out_err; | |
242 | ||
243 | return __perf_evsel__hw_cache_type_op_res_name(type, op, result, bf, size); | |
244 | out_err: | |
245 | return scnprintf(bf, size, "%s", err); | |
246 | } | |
247 | ||
248 | static int perf_evsel__hw_cache_name(struct perf_evsel *evsel, char *bf, size_t size) | |
249 | { | |
250 | int ret = __perf_evsel__hw_cache_name(evsel->attr.config, bf, size); | |
251 | return ret + perf_evsel__add_modifiers(evsel, bf + ret, size - ret); | |
252 | } | |
253 | ||
6eef3d9c ACM |
254 | static int perf_evsel__raw_name(struct perf_evsel *evsel, char *bf, size_t size) |
255 | { | |
256 | int ret = scnprintf(bf, size, "raw 0x%" PRIx64, evsel->attr.config); | |
257 | return ret + perf_evsel__add_modifiers(evsel, bf + ret, size - ret); | |
258 | } | |
259 | ||
7289f83c | 260 | const char *perf_evsel__name(struct perf_evsel *evsel) |
a4460836 | 261 | { |
7289f83c | 262 | char bf[128]; |
a4460836 | 263 | |
7289f83c ACM |
264 | if (evsel->name) |
265 | return evsel->name; | |
c410431c ACM |
266 | |
267 | switch (evsel->attr.type) { | |
268 | case PERF_TYPE_RAW: | |
6eef3d9c | 269 | perf_evsel__raw_name(evsel, bf, sizeof(bf)); |
c410431c ACM |
270 | break; |
271 | ||
272 | case PERF_TYPE_HARDWARE: | |
7289f83c | 273 | perf_evsel__hw_name(evsel, bf, sizeof(bf)); |
c410431c | 274 | break; |
0b668bc9 ACM |
275 | |
276 | case PERF_TYPE_HW_CACHE: | |
7289f83c | 277 | perf_evsel__hw_cache_name(evsel, bf, sizeof(bf)); |
0b668bc9 ACM |
278 | break; |
279 | ||
335c2f5d | 280 | case PERF_TYPE_SOFTWARE: |
7289f83c | 281 | perf_evsel__sw_name(evsel, bf, sizeof(bf)); |
335c2f5d ACM |
282 | break; |
283 | ||
a4460836 | 284 | case PERF_TYPE_TRACEPOINT: |
7289f83c | 285 | scnprintf(bf, sizeof(bf), "%s", "unknown tracepoint"); |
a4460836 ACM |
286 | break; |
287 | ||
c410431c | 288 | default: |
7289f83c | 289 | scnprintf(bf, sizeof(bf), "%s", "unknown attr type"); |
a4460836 | 290 | break; |
c410431c ACM |
291 | } |
292 | ||
7289f83c ACM |
293 | evsel->name = strdup(bf); |
294 | ||
295 | return evsel->name ?: "unknown"; | |
c410431c ACM |
296 | } |
297 | ||
5090c6ae NK |
298 | void perf_evsel__config(struct perf_evsel *evsel, struct perf_record_opts *opts, |
299 | struct perf_evsel *first) | |
0f82ebc4 ACM |
300 | { |
301 | struct perf_event_attr *attr = &evsel->attr; | |
302 | int track = !evsel->idx; /* only the first counter needs these */ | |
303 | ||
5e1c81d9 | 304 | attr->disabled = 1; |
808e1226 | 305 | attr->sample_id_all = opts->sample_id_all_missing ? 0 : 1; |
0f82ebc4 ACM |
306 | attr->inherit = !opts->no_inherit; |
307 | attr->read_format = PERF_FORMAT_TOTAL_TIME_ENABLED | | |
308 | PERF_FORMAT_TOTAL_TIME_RUNNING | | |
309 | PERF_FORMAT_ID; | |
310 | ||
311 | attr->sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID; | |
312 | ||
313 | /* | |
314 | * We default some events to a 1 default interval. But keep | |
315 | * it a weak assumption overridable by the user. | |
316 | */ | |
317 | if (!attr->sample_period || (opts->user_freq != UINT_MAX && | |
318 | opts->user_interval != ULLONG_MAX)) { | |
319 | if (opts->freq) { | |
320 | attr->sample_type |= PERF_SAMPLE_PERIOD; | |
321 | attr->freq = 1; | |
322 | attr->sample_freq = opts->freq; | |
323 | } else { | |
324 | attr->sample_period = opts->default_interval; | |
325 | } | |
326 | } | |
327 | ||
328 | if (opts->no_samples) | |
329 | attr->sample_freq = 0; | |
330 | ||
331 | if (opts->inherit_stat) | |
332 | attr->inherit_stat = 1; | |
333 | ||
334 | if (opts->sample_address) { | |
335 | attr->sample_type |= PERF_SAMPLE_ADDR; | |
336 | attr->mmap_data = track; | |
337 | } | |
338 | ||
339 | if (opts->call_graph) | |
340 | attr->sample_type |= PERF_SAMPLE_CALLCHAIN; | |
341 | ||
e40ee742 | 342 | if (perf_target__has_cpu(&opts->target)) |
0f82ebc4 ACM |
343 | attr->sample_type |= PERF_SAMPLE_CPU; |
344 | ||
3e76ac78 AV |
345 | if (opts->period) |
346 | attr->sample_type |= PERF_SAMPLE_PERIOD; | |
347 | ||
808e1226 | 348 | if (!opts->sample_id_all_missing && |
d67356e7 | 349 | (opts->sample_time || !opts->no_inherit || |
aa22dd49 | 350 | perf_target__has_cpu(&opts->target))) |
0f82ebc4 ACM |
351 | attr->sample_type |= PERF_SAMPLE_TIME; |
352 | ||
353 | if (opts->raw_samples) { | |
354 | attr->sample_type |= PERF_SAMPLE_TIME; | |
355 | attr->sample_type |= PERF_SAMPLE_RAW; | |
356 | attr->sample_type |= PERF_SAMPLE_CPU; | |
357 | } | |
358 | ||
359 | if (opts->no_delay) { | |
360 | attr->watermark = 0; | |
361 | attr->wakeup_events = 1; | |
362 | } | |
bdfebd84 RAV |
363 | if (opts->branch_stack) { |
364 | attr->sample_type |= PERF_SAMPLE_BRANCH_STACK; | |
365 | attr->branch_sample_type = opts->branch_stack; | |
366 | } | |
0f82ebc4 ACM |
367 | |
368 | attr->mmap = track; | |
369 | attr->comm = track; | |
370 | ||
d67356e7 NK |
371 | if (perf_target__none(&opts->target) && |
372 | (!opts->group || evsel == first)) { | |
0f82ebc4 ACM |
373 | attr->enable_on_exec = 1; |
374 | } | |
375 | } | |
376 | ||
69aad6f1 ACM |
377 | int perf_evsel__alloc_fd(struct perf_evsel *evsel, int ncpus, int nthreads) |
378 | { | |
4af4c955 | 379 | int cpu, thread; |
69aad6f1 | 380 | evsel->fd = xyarray__new(ncpus, nthreads, sizeof(int)); |
4af4c955 DA |
381 | |
382 | if (evsel->fd) { | |
383 | for (cpu = 0; cpu < ncpus; cpu++) { | |
384 | for (thread = 0; thread < nthreads; thread++) { | |
385 | FD(evsel, cpu, thread) = -1; | |
386 | } | |
387 | } | |
388 | } | |
389 | ||
69aad6f1 ACM |
390 | return evsel->fd != NULL ? 0 : -ENOMEM; |
391 | } | |
392 | ||
70db7533 ACM |
393 | int perf_evsel__alloc_id(struct perf_evsel *evsel, int ncpus, int nthreads) |
394 | { | |
a91e5431 ACM |
395 | evsel->sample_id = xyarray__new(ncpus, nthreads, sizeof(struct perf_sample_id)); |
396 | if (evsel->sample_id == NULL) | |
397 | return -ENOMEM; | |
398 | ||
399 | evsel->id = zalloc(ncpus * nthreads * sizeof(u64)); | |
400 | if (evsel->id == NULL) { | |
401 | xyarray__delete(evsel->sample_id); | |
402 | evsel->sample_id = NULL; | |
403 | return -ENOMEM; | |
404 | } | |
405 | ||
406 | return 0; | |
70db7533 ACM |
407 | } |
408 | ||
c52b12ed ACM |
409 | int perf_evsel__alloc_counts(struct perf_evsel *evsel, int ncpus) |
410 | { | |
411 | evsel->counts = zalloc((sizeof(*evsel->counts) + | |
412 | (ncpus * sizeof(struct perf_counts_values)))); | |
413 | return evsel->counts != NULL ? 0 : -ENOMEM; | |
414 | } | |
415 | ||
69aad6f1 ACM |
416 | void perf_evsel__free_fd(struct perf_evsel *evsel) |
417 | { | |
418 | xyarray__delete(evsel->fd); | |
419 | evsel->fd = NULL; | |
420 | } | |
421 | ||
70db7533 ACM |
422 | void perf_evsel__free_id(struct perf_evsel *evsel) |
423 | { | |
a91e5431 ACM |
424 | xyarray__delete(evsel->sample_id); |
425 | evsel->sample_id = NULL; | |
426 | free(evsel->id); | |
70db7533 ACM |
427 | evsel->id = NULL; |
428 | } | |
429 | ||
c52b12ed ACM |
430 | void perf_evsel__close_fd(struct perf_evsel *evsel, int ncpus, int nthreads) |
431 | { | |
432 | int cpu, thread; | |
433 | ||
434 | for (cpu = 0; cpu < ncpus; cpu++) | |
435 | for (thread = 0; thread < nthreads; ++thread) { | |
436 | close(FD(evsel, cpu, thread)); | |
437 | FD(evsel, cpu, thread) = -1; | |
438 | } | |
439 | } | |
440 | ||
ef1d1af2 | 441 | void perf_evsel__exit(struct perf_evsel *evsel) |
69aad6f1 ACM |
442 | { |
443 | assert(list_empty(&evsel->node)); | |
444 | xyarray__delete(evsel->fd); | |
a91e5431 ACM |
445 | xyarray__delete(evsel->sample_id); |
446 | free(evsel->id); | |
ef1d1af2 ACM |
447 | } |
448 | ||
449 | void perf_evsel__delete(struct perf_evsel *evsel) | |
450 | { | |
451 | perf_evsel__exit(evsel); | |
023695d9 | 452 | close_cgroup(evsel->cgrp); |
f0c55bcf | 453 | free(evsel->name); |
69aad6f1 ACM |
454 | free(evsel); |
455 | } | |
c52b12ed ACM |
456 | |
457 | int __perf_evsel__read_on_cpu(struct perf_evsel *evsel, | |
458 | int cpu, int thread, bool scale) | |
459 | { | |
460 | struct perf_counts_values count; | |
461 | size_t nv = scale ? 3 : 1; | |
462 | ||
463 | if (FD(evsel, cpu, thread) < 0) | |
464 | return -EINVAL; | |
465 | ||
4eed11d5 ACM |
466 | if (evsel->counts == NULL && perf_evsel__alloc_counts(evsel, cpu + 1) < 0) |
467 | return -ENOMEM; | |
468 | ||
c52b12ed ACM |
469 | if (readn(FD(evsel, cpu, thread), &count, nv * sizeof(u64)) < 0) |
470 | return -errno; | |
471 | ||
472 | if (scale) { | |
473 | if (count.run == 0) | |
474 | count.val = 0; | |
475 | else if (count.run < count.ena) | |
476 | count.val = (u64)((double)count.val * count.ena / count.run + 0.5); | |
477 | } else | |
478 | count.ena = count.run = 0; | |
479 | ||
480 | evsel->counts->cpu[cpu] = count; | |
481 | return 0; | |
482 | } | |
483 | ||
484 | int __perf_evsel__read(struct perf_evsel *evsel, | |
485 | int ncpus, int nthreads, bool scale) | |
486 | { | |
487 | size_t nv = scale ? 3 : 1; | |
488 | int cpu, thread; | |
489 | struct perf_counts_values *aggr = &evsel->counts->aggr, count; | |
490 | ||
52bcd994 | 491 | aggr->val = aggr->ena = aggr->run = 0; |
c52b12ed ACM |
492 | |
493 | for (cpu = 0; cpu < ncpus; cpu++) { | |
494 | for (thread = 0; thread < nthreads; thread++) { | |
495 | if (FD(evsel, cpu, thread) < 0) | |
496 | continue; | |
497 | ||
498 | if (readn(FD(evsel, cpu, thread), | |
499 | &count, nv * sizeof(u64)) < 0) | |
500 | return -errno; | |
501 | ||
502 | aggr->val += count.val; | |
503 | if (scale) { | |
504 | aggr->ena += count.ena; | |
505 | aggr->run += count.run; | |
506 | } | |
507 | } | |
508 | } | |
509 | ||
510 | evsel->counts->scaled = 0; | |
511 | if (scale) { | |
512 | if (aggr->run == 0) { | |
513 | evsel->counts->scaled = -1; | |
514 | aggr->val = 0; | |
515 | return 0; | |
516 | } | |
517 | ||
518 | if (aggr->run < aggr->ena) { | |
519 | evsel->counts->scaled = 1; | |
520 | aggr->val = (u64)((double)aggr->val * aggr->ena / aggr->run + 0.5); | |
521 | } | |
522 | } else | |
523 | aggr->ena = aggr->run = 0; | |
524 | ||
525 | return 0; | |
526 | } | |
48290609 | 527 | |
0252208e | 528 | static int __perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, |
727ab04e ACM |
529 | struct thread_map *threads, bool group, |
530 | struct xyarray *group_fds) | |
48290609 | 531 | { |
0252208e | 532 | int cpu, thread; |
023695d9 | 533 | unsigned long flags = 0; |
727ab04e | 534 | int pid = -1, err; |
48290609 | 535 | |
0252208e ACM |
536 | if (evsel->fd == NULL && |
537 | perf_evsel__alloc_fd(evsel, cpus->nr, threads->nr) < 0) | |
727ab04e | 538 | return -ENOMEM; |
4eed11d5 | 539 | |
023695d9 SE |
540 | if (evsel->cgrp) { |
541 | flags = PERF_FLAG_PID_CGROUP; | |
542 | pid = evsel->cgrp->fd; | |
543 | } | |
544 | ||
86bd5e86 | 545 | for (cpu = 0; cpu < cpus->nr; cpu++) { |
727ab04e | 546 | int group_fd = group_fds ? GROUP_FD(group_fds, cpu) : -1; |
9d04f178 | 547 | |
0252208e | 548 | for (thread = 0; thread < threads->nr; thread++) { |
023695d9 SE |
549 | |
550 | if (!evsel->cgrp) | |
551 | pid = threads->map[thread]; | |
552 | ||
0252208e | 553 | FD(evsel, cpu, thread) = sys_perf_event_open(&evsel->attr, |
023695d9 | 554 | pid, |
f08199d3 | 555 | cpus->map[cpu], |
023695d9 | 556 | group_fd, flags); |
727ab04e ACM |
557 | if (FD(evsel, cpu, thread) < 0) { |
558 | err = -errno; | |
0252208e | 559 | goto out_close; |
727ab04e | 560 | } |
f08199d3 ACM |
561 | |
562 | if (group && group_fd == -1) | |
563 | group_fd = FD(evsel, cpu, thread); | |
0252208e | 564 | } |
48290609 ACM |
565 | } |
566 | ||
567 | return 0; | |
568 | ||
569 | out_close: | |
0252208e ACM |
570 | do { |
571 | while (--thread >= 0) { | |
572 | close(FD(evsel, cpu, thread)); | |
573 | FD(evsel, cpu, thread) = -1; | |
574 | } | |
575 | thread = threads->nr; | |
576 | } while (--cpu >= 0); | |
727ab04e ACM |
577 | return err; |
578 | } | |
579 | ||
580 | void perf_evsel__close(struct perf_evsel *evsel, int ncpus, int nthreads) | |
581 | { | |
582 | if (evsel->fd == NULL) | |
583 | return; | |
584 | ||
585 | perf_evsel__close_fd(evsel, ncpus, nthreads); | |
586 | perf_evsel__free_fd(evsel); | |
587 | evsel->fd = NULL; | |
48290609 ACM |
588 | } |
589 | ||
0252208e ACM |
590 | static struct { |
591 | struct cpu_map map; | |
592 | int cpus[1]; | |
593 | } empty_cpu_map = { | |
594 | .map.nr = 1, | |
595 | .cpus = { -1, }, | |
596 | }; | |
597 | ||
598 | static struct { | |
599 | struct thread_map map; | |
600 | int threads[1]; | |
601 | } empty_thread_map = { | |
602 | .map.nr = 1, | |
603 | .threads = { -1, }, | |
604 | }; | |
605 | ||
f08199d3 | 606 | int perf_evsel__open(struct perf_evsel *evsel, struct cpu_map *cpus, |
727ab04e ACM |
607 | struct thread_map *threads, bool group, |
608 | struct xyarray *group_fd) | |
48290609 | 609 | { |
0252208e ACM |
610 | if (cpus == NULL) { |
611 | /* Work around old compiler warnings about strict aliasing */ | |
612 | cpus = &empty_cpu_map.map; | |
48290609 ACM |
613 | } |
614 | ||
0252208e ACM |
615 | if (threads == NULL) |
616 | threads = &empty_thread_map.map; | |
48290609 | 617 | |
727ab04e | 618 | return __perf_evsel__open(evsel, cpus, threads, group, group_fd); |
48290609 ACM |
619 | } |
620 | ||
f08199d3 | 621 | int perf_evsel__open_per_cpu(struct perf_evsel *evsel, |
727ab04e ACM |
622 | struct cpu_map *cpus, bool group, |
623 | struct xyarray *group_fd) | |
48290609 | 624 | { |
727ab04e ACM |
625 | return __perf_evsel__open(evsel, cpus, &empty_thread_map.map, group, |
626 | group_fd); | |
0252208e | 627 | } |
48290609 | 628 | |
f08199d3 | 629 | int perf_evsel__open_per_thread(struct perf_evsel *evsel, |
727ab04e ACM |
630 | struct thread_map *threads, bool group, |
631 | struct xyarray *group_fd) | |
0252208e | 632 | { |
727ab04e ACM |
633 | return __perf_evsel__open(evsel, &empty_cpu_map.map, threads, group, |
634 | group_fd); | |
48290609 | 635 | } |
70082dd9 | 636 | |
8115d60c | 637 | static int perf_event__parse_id_sample(const union perf_event *event, u64 type, |
37073f9e JO |
638 | struct perf_sample *sample, |
639 | bool swapped) | |
d0dd74e8 ACM |
640 | { |
641 | const u64 *array = event->sample.array; | |
37073f9e | 642 | union u64_swap u; |
d0dd74e8 ACM |
643 | |
644 | array += ((event->header.size - | |
645 | sizeof(event->header)) / sizeof(u64)) - 1; | |
646 | ||
647 | if (type & PERF_SAMPLE_CPU) { | |
37073f9e JO |
648 | u.val64 = *array; |
649 | if (swapped) { | |
650 | /* undo swap of u64, then swap on individual u32s */ | |
651 | u.val64 = bswap_64(u.val64); | |
652 | u.val32[0] = bswap_32(u.val32[0]); | |
653 | } | |
654 | ||
655 | sample->cpu = u.val32[0]; | |
d0dd74e8 ACM |
656 | array--; |
657 | } | |
658 | ||
659 | if (type & PERF_SAMPLE_STREAM_ID) { | |
660 | sample->stream_id = *array; | |
661 | array--; | |
662 | } | |
663 | ||
664 | if (type & PERF_SAMPLE_ID) { | |
665 | sample->id = *array; | |
666 | array--; | |
667 | } | |
668 | ||
669 | if (type & PERF_SAMPLE_TIME) { | |
670 | sample->time = *array; | |
671 | array--; | |
672 | } | |
673 | ||
674 | if (type & PERF_SAMPLE_TID) { | |
37073f9e JO |
675 | u.val64 = *array; |
676 | if (swapped) { | |
677 | /* undo swap of u64, then swap on individual u32s */ | |
678 | u.val64 = bswap_64(u.val64); | |
679 | u.val32[0] = bswap_32(u.val32[0]); | |
680 | u.val32[1] = bswap_32(u.val32[1]); | |
681 | } | |
682 | ||
683 | sample->pid = u.val32[0]; | |
684 | sample->tid = u.val32[1]; | |
d0dd74e8 ACM |
685 | } |
686 | ||
687 | return 0; | |
688 | } | |
689 | ||
98e1da90 FW |
690 | static bool sample_overlap(const union perf_event *event, |
691 | const void *offset, u64 size) | |
692 | { | |
693 | const void *base = event; | |
694 | ||
695 | if (offset + size > base + event->header.size) | |
696 | return true; | |
697 | ||
698 | return false; | |
699 | } | |
700 | ||
8115d60c | 701 | int perf_event__parse_sample(const union perf_event *event, u64 type, |
a2854124 | 702 | int sample_size, bool sample_id_all, |
936be503 | 703 | struct perf_sample *data, bool swapped) |
d0dd74e8 ACM |
704 | { |
705 | const u64 *array; | |
706 | ||
936be503 DA |
707 | /* |
708 | * used for cross-endian analysis. See git commit 65014ab3 | |
709 | * for why this goofiness is needed. | |
710 | */ | |
6a11f92e | 711 | union u64_swap u; |
936be503 | 712 | |
f3bda2c9 | 713 | memset(data, 0, sizeof(*data)); |
d0dd74e8 ACM |
714 | data->cpu = data->pid = data->tid = -1; |
715 | data->stream_id = data->id = data->time = -1ULL; | |
a4a03fc7 | 716 | data->period = 1; |
d0dd74e8 ACM |
717 | |
718 | if (event->header.type != PERF_RECORD_SAMPLE) { | |
719 | if (!sample_id_all) | |
720 | return 0; | |
37073f9e | 721 | return perf_event__parse_id_sample(event, type, data, swapped); |
d0dd74e8 ACM |
722 | } |
723 | ||
724 | array = event->sample.array; | |
725 | ||
a2854124 FW |
726 | if (sample_size + sizeof(event->header) > event->header.size) |
727 | return -EFAULT; | |
728 | ||
d0dd74e8 ACM |
729 | if (type & PERF_SAMPLE_IP) { |
730 | data->ip = event->ip.ip; | |
731 | array++; | |
732 | } | |
733 | ||
734 | if (type & PERF_SAMPLE_TID) { | |
936be503 DA |
735 | u.val64 = *array; |
736 | if (swapped) { | |
737 | /* undo swap of u64, then swap on individual u32s */ | |
738 | u.val64 = bswap_64(u.val64); | |
739 | u.val32[0] = bswap_32(u.val32[0]); | |
740 | u.val32[1] = bswap_32(u.val32[1]); | |
741 | } | |
742 | ||
743 | data->pid = u.val32[0]; | |
744 | data->tid = u.val32[1]; | |
d0dd74e8 ACM |
745 | array++; |
746 | } | |
747 | ||
748 | if (type & PERF_SAMPLE_TIME) { | |
749 | data->time = *array; | |
750 | array++; | |
751 | } | |
752 | ||
7cec0922 | 753 | data->addr = 0; |
d0dd74e8 ACM |
754 | if (type & PERF_SAMPLE_ADDR) { |
755 | data->addr = *array; | |
756 | array++; | |
757 | } | |
758 | ||
759 | data->id = -1ULL; | |
760 | if (type & PERF_SAMPLE_ID) { | |
761 | data->id = *array; | |
762 | array++; | |
763 | } | |
764 | ||
765 | if (type & PERF_SAMPLE_STREAM_ID) { | |
766 | data->stream_id = *array; | |
767 | array++; | |
768 | } | |
769 | ||
770 | if (type & PERF_SAMPLE_CPU) { | |
936be503 DA |
771 | |
772 | u.val64 = *array; | |
773 | if (swapped) { | |
774 | /* undo swap of u64, then swap on individual u32s */ | |
775 | u.val64 = bswap_64(u.val64); | |
776 | u.val32[0] = bswap_32(u.val32[0]); | |
777 | } | |
778 | ||
779 | data->cpu = u.val32[0]; | |
d0dd74e8 ACM |
780 | array++; |
781 | } | |
782 | ||
783 | if (type & PERF_SAMPLE_PERIOD) { | |
784 | data->period = *array; | |
785 | array++; | |
786 | } | |
787 | ||
788 | if (type & PERF_SAMPLE_READ) { | |
f9d36996 | 789 | fprintf(stderr, "PERF_SAMPLE_READ is unsupported for now\n"); |
d0dd74e8 ACM |
790 | return -1; |
791 | } | |
792 | ||
793 | if (type & PERF_SAMPLE_CALLCHAIN) { | |
98e1da90 FW |
794 | if (sample_overlap(event, array, sizeof(data->callchain->nr))) |
795 | return -EFAULT; | |
796 | ||
d0dd74e8 | 797 | data->callchain = (struct ip_callchain *)array; |
98e1da90 FW |
798 | |
799 | if (sample_overlap(event, array, data->callchain->nr)) | |
800 | return -EFAULT; | |
801 | ||
d0dd74e8 ACM |
802 | array += 1 + data->callchain->nr; |
803 | } | |
804 | ||
805 | if (type & PERF_SAMPLE_RAW) { | |
8e303f20 JO |
806 | const u64 *pdata; |
807 | ||
936be503 DA |
808 | u.val64 = *array; |
809 | if (WARN_ONCE(swapped, | |
810 | "Endianness of raw data not corrected!\n")) { | |
811 | /* undo swap of u64, then swap on individual u32s */ | |
812 | u.val64 = bswap_64(u.val64); | |
813 | u.val32[0] = bswap_32(u.val32[0]); | |
814 | u.val32[1] = bswap_32(u.val32[1]); | |
815 | } | |
98e1da90 FW |
816 | |
817 | if (sample_overlap(event, array, sizeof(u32))) | |
818 | return -EFAULT; | |
819 | ||
936be503 | 820 | data->raw_size = u.val32[0]; |
8e303f20 | 821 | pdata = (void *) array + sizeof(u32); |
98e1da90 | 822 | |
8e303f20 | 823 | if (sample_overlap(event, pdata, data->raw_size)) |
98e1da90 FW |
824 | return -EFAULT; |
825 | ||
8e303f20 | 826 | data->raw_data = (void *) pdata; |
fa30c964 SE |
827 | |
828 | array = (void *)array + data->raw_size + sizeof(u32); | |
d0dd74e8 ACM |
829 | } |
830 | ||
b5387528 RAV |
831 | if (type & PERF_SAMPLE_BRANCH_STACK) { |
832 | u64 sz; | |
833 | ||
834 | data->branch_stack = (struct branch_stack *)array; | |
835 | array++; /* nr */ | |
836 | ||
837 | sz = data->branch_stack->nr * sizeof(struct branch_entry); | |
838 | sz /= sizeof(u64); | |
839 | array += sz; | |
840 | } | |
d0dd74e8 ACM |
841 | return 0; |
842 | } | |
74eec26f AV |
843 | |
844 | int perf_event__synthesize_sample(union perf_event *event, u64 type, | |
845 | const struct perf_sample *sample, | |
846 | bool swapped) | |
847 | { | |
848 | u64 *array; | |
849 | ||
850 | /* | |
851 | * used for cross-endian analysis. See git commit 65014ab3 | |
852 | * for why this goofiness is needed. | |
853 | */ | |
6a11f92e | 854 | union u64_swap u; |
74eec26f AV |
855 | |
856 | array = event->sample.array; | |
857 | ||
858 | if (type & PERF_SAMPLE_IP) { | |
859 | event->ip.ip = sample->ip; | |
860 | array++; | |
861 | } | |
862 | ||
863 | if (type & PERF_SAMPLE_TID) { | |
864 | u.val32[0] = sample->pid; | |
865 | u.val32[1] = sample->tid; | |
866 | if (swapped) { | |
867 | /* | |
868 | * Inverse of what is done in perf_event__parse_sample | |
869 | */ | |
870 | u.val32[0] = bswap_32(u.val32[0]); | |
871 | u.val32[1] = bswap_32(u.val32[1]); | |
872 | u.val64 = bswap_64(u.val64); | |
873 | } | |
874 | ||
875 | *array = u.val64; | |
876 | array++; | |
877 | } | |
878 | ||
879 | if (type & PERF_SAMPLE_TIME) { | |
880 | *array = sample->time; | |
881 | array++; | |
882 | } | |
883 | ||
884 | if (type & PERF_SAMPLE_ADDR) { | |
885 | *array = sample->addr; | |
886 | array++; | |
887 | } | |
888 | ||
889 | if (type & PERF_SAMPLE_ID) { | |
890 | *array = sample->id; | |
891 | array++; | |
892 | } | |
893 | ||
894 | if (type & PERF_SAMPLE_STREAM_ID) { | |
895 | *array = sample->stream_id; | |
896 | array++; | |
897 | } | |
898 | ||
899 | if (type & PERF_SAMPLE_CPU) { | |
900 | u.val32[0] = sample->cpu; | |
901 | if (swapped) { | |
902 | /* | |
903 | * Inverse of what is done in perf_event__parse_sample | |
904 | */ | |
905 | u.val32[0] = bswap_32(u.val32[0]); | |
906 | u.val64 = bswap_64(u.val64); | |
907 | } | |
908 | *array = u.val64; | |
909 | array++; | |
910 | } | |
911 | ||
912 | if (type & PERF_SAMPLE_PERIOD) { | |
913 | *array = sample->period; | |
914 | array++; | |
915 | } | |
916 | ||
917 | return 0; | |
918 | } |