Root/
Source at commit ec7cab4cbb721bff91ec924ec691efd8daf36579 created 12 years 8 months ago. By Maarten ter Huurne, MIPS: JZ4740: A320: Updated quickstart documentation. | |
---|---|
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 | #include <poll.h> |
10 | #include "cpumap.h" |
11 | #include "thread_map.h" |
12 | #include "evlist.h" |
13 | #include "evsel.h" |
14 | #include "util.h" |
15 | #include "debug.h" |
16 | |
17 | #include <sys/mman.h> |
18 | |
19 | #include <linux/bitops.h> |
20 | #include <linux/hash.h> |
21 | |
22 | #define FD(e, x, y) (*(int *)xyarray__entry(e->fd, x, y)) |
23 | #define SID(e, x, y) xyarray__entry(e->sample_id, x, y) |
24 | |
25 | void perf_evlist__init(struct perf_evlist *evlist, struct cpu_map *cpus, |
26 | struct thread_map *threads) |
27 | { |
28 | int i; |
29 | |
30 | for (i = 0; i < PERF_EVLIST__HLIST_SIZE; ++i) |
31 | INIT_HLIST_HEAD(&evlist->heads[i]); |
32 | INIT_LIST_HEAD(&evlist->entries); |
33 | perf_evlist__set_maps(evlist, cpus, threads); |
34 | } |
35 | |
36 | struct perf_evlist *perf_evlist__new(struct cpu_map *cpus, |
37 | struct thread_map *threads) |
38 | { |
39 | struct perf_evlist *evlist = zalloc(sizeof(*evlist)); |
40 | |
41 | if (evlist != NULL) |
42 | perf_evlist__init(evlist, cpus, threads); |
43 | |
44 | return evlist; |
45 | } |
46 | |
47 | static void perf_evlist__purge(struct perf_evlist *evlist) |
48 | { |
49 | struct perf_evsel *pos, *n; |
50 | |
51 | list_for_each_entry_safe(pos, n, &evlist->entries, node) { |
52 | list_del_init(&pos->node); |
53 | perf_evsel__delete(pos); |
54 | } |
55 | |
56 | evlist->nr_entries = 0; |
57 | } |
58 | |
59 | void perf_evlist__exit(struct perf_evlist *evlist) |
60 | { |
61 | free(evlist->mmap); |
62 | free(evlist->pollfd); |
63 | evlist->mmap = NULL; |
64 | evlist->pollfd = NULL; |
65 | } |
66 | |
67 | void perf_evlist__delete(struct perf_evlist *evlist) |
68 | { |
69 | perf_evlist__purge(evlist); |
70 | perf_evlist__exit(evlist); |
71 | free(evlist); |
72 | } |
73 | |
74 | void perf_evlist__add(struct perf_evlist *evlist, struct perf_evsel *entry) |
75 | { |
76 | list_add_tail(&entry->node, &evlist->entries); |
77 | ++evlist->nr_entries; |
78 | } |
79 | |
80 | int perf_evlist__add_default(struct perf_evlist *evlist) |
81 | { |
82 | struct perf_event_attr attr = { |
83 | .type = PERF_TYPE_HARDWARE, |
84 | .config = PERF_COUNT_HW_CPU_CYCLES, |
85 | }; |
86 | struct perf_evsel *evsel = perf_evsel__new(&attr, 0); |
87 | |
88 | if (evsel == NULL) |
89 | return -ENOMEM; |
90 | |
91 | perf_evlist__add(evlist, evsel); |
92 | return 0; |
93 | } |
94 | |
95 | int perf_evlist__alloc_pollfd(struct perf_evlist *evlist) |
96 | { |
97 | int nfds = evlist->cpus->nr * evlist->threads->nr * evlist->nr_entries; |
98 | evlist->pollfd = malloc(sizeof(struct pollfd) * nfds); |
99 | return evlist->pollfd != NULL ? 0 : -ENOMEM; |
100 | } |
101 | |
102 | void perf_evlist__add_pollfd(struct perf_evlist *evlist, int fd) |
103 | { |
104 | fcntl(fd, F_SETFL, O_NONBLOCK); |
105 | evlist->pollfd[evlist->nr_fds].fd = fd; |
106 | evlist->pollfd[evlist->nr_fds].events = POLLIN; |
107 | evlist->nr_fds++; |
108 | } |
109 | |
110 | static void perf_evlist__id_hash(struct perf_evlist *evlist, |
111 | struct perf_evsel *evsel, |
112 | int cpu, int thread, u64 id) |
113 | { |
114 | int hash; |
115 | struct perf_sample_id *sid = SID(evsel, cpu, thread); |
116 | |
117 | sid->id = id; |
118 | sid->evsel = evsel; |
119 | hash = hash_64(sid->id, PERF_EVLIST__HLIST_BITS); |
120 | hlist_add_head(&sid->node, &evlist->heads[hash]); |
121 | } |
122 | |
123 | void perf_evlist__id_add(struct perf_evlist *evlist, struct perf_evsel *evsel, |
124 | int cpu, int thread, u64 id) |
125 | { |
126 | perf_evlist__id_hash(evlist, evsel, cpu, thread, id); |
127 | evsel->id[evsel->ids++] = id; |
128 | } |
129 | |
130 | static int perf_evlist__id_add_fd(struct perf_evlist *evlist, |
131 | struct perf_evsel *evsel, |
132 | int cpu, int thread, int fd) |
133 | { |
134 | u64 read_data[4] = { 0, }; |
135 | int id_idx = 1; /* The first entry is the counter value */ |
136 | |
137 | if (!(evsel->attr.read_format & PERF_FORMAT_ID) || |
138 | read(fd, &read_data, sizeof(read_data)) == -1) |
139 | return -1; |
140 | |
141 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) |
142 | ++id_idx; |
143 | if (evsel->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) |
144 | ++id_idx; |
145 | |
146 | perf_evlist__id_add(evlist, evsel, cpu, thread, read_data[id_idx]); |
147 | return 0; |
148 | } |
149 | |
150 | struct perf_evsel *perf_evlist__id2evsel(struct perf_evlist *evlist, u64 id) |
151 | { |
152 | struct hlist_head *head; |
153 | struct hlist_node *pos; |
154 | struct perf_sample_id *sid; |
155 | int hash; |
156 | |
157 | if (evlist->nr_entries == 1) |
158 | return list_entry(evlist->entries.next, struct perf_evsel, node); |
159 | |
160 | hash = hash_64(id, PERF_EVLIST__HLIST_BITS); |
161 | head = &evlist->heads[hash]; |
162 | |
163 | hlist_for_each_entry(sid, pos, head, node) |
164 | if (sid->id == id) |
165 | return sid->evsel; |
166 | return NULL; |
167 | } |
168 | |
169 | union perf_event *perf_evlist__mmap_read(struct perf_evlist *evlist, int idx) |
170 | { |
171 | /* XXX Move this to perf.c, making it generally available */ |
172 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); |
173 | struct perf_mmap *md = &evlist->mmap[idx]; |
174 | unsigned int head = perf_mmap__read_head(md); |
175 | unsigned int old = md->prev; |
176 | unsigned char *data = md->base + page_size; |
177 | union perf_event *event = NULL; |
178 | |
179 | if (evlist->overwrite) { |
180 | /* |
181 | * If we're further behind than half the buffer, there's a chance |
182 | * the writer will bite our tail and mess up the samples under us. |
183 | * |
184 | * If we somehow ended up ahead of the head, we got messed up. |
185 | * |
186 | * In either case, truncate and restart at head. |
187 | */ |
188 | int diff = head - old; |
189 | if (diff > md->mask / 2 || diff < 0) { |
190 | fprintf(stderr, "WARNING: failed to keep up with mmap data.\n"); |
191 | |
192 | /* |
193 | * head points to a known good entry, start there. |
194 | */ |
195 | old = head; |
196 | } |
197 | } |
198 | |
199 | if (old != head) { |
200 | size_t size; |
201 | |
202 | event = (union perf_event *)&data[old & md->mask]; |
203 | size = event->header.size; |
204 | |
205 | /* |
206 | * Event straddles the mmap boundary -- header should always |
207 | * be inside due to u64 alignment of output. |
208 | */ |
209 | if ((old & md->mask) + size != ((old + size) & md->mask)) { |
210 | unsigned int offset = old; |
211 | unsigned int len = min(sizeof(*event), size), cpy; |
212 | void *dst = &evlist->event_copy; |
213 | |
214 | do { |
215 | cpy = min(md->mask + 1 - (offset & md->mask), len); |
216 | memcpy(dst, &data[offset & md->mask], cpy); |
217 | offset += cpy; |
218 | dst += cpy; |
219 | len -= cpy; |
220 | } while (len); |
221 | |
222 | event = &evlist->event_copy; |
223 | } |
224 | |
225 | old += size; |
226 | } |
227 | |
228 | md->prev = old; |
229 | |
230 | if (!evlist->overwrite) |
231 | perf_mmap__write_tail(md, old); |
232 | |
233 | return event; |
234 | } |
235 | |
236 | void perf_evlist__munmap(struct perf_evlist *evlist) |
237 | { |
238 | int i; |
239 | |
240 | for (i = 0; i < evlist->nr_mmaps; i++) { |
241 | if (evlist->mmap[i].base != NULL) { |
242 | munmap(evlist->mmap[i].base, evlist->mmap_len); |
243 | evlist->mmap[i].base = NULL; |
244 | } |
245 | } |
246 | |
247 | free(evlist->mmap); |
248 | evlist->mmap = NULL; |
249 | } |
250 | |
251 | int perf_evlist__alloc_mmap(struct perf_evlist *evlist) |
252 | { |
253 | evlist->nr_mmaps = evlist->cpus->nr; |
254 | if (evlist->cpus->map[0] == -1) |
255 | evlist->nr_mmaps = evlist->threads->nr; |
256 | evlist->mmap = zalloc(evlist->nr_mmaps * sizeof(struct perf_mmap)); |
257 | return evlist->mmap != NULL ? 0 : -ENOMEM; |
258 | } |
259 | |
260 | static int __perf_evlist__mmap(struct perf_evlist *evlist, struct perf_evsel *evsel, |
261 | int idx, int prot, int mask, int fd) |
262 | { |
263 | evlist->mmap[idx].prev = 0; |
264 | evlist->mmap[idx].mask = mask; |
265 | evlist->mmap[idx].base = mmap(NULL, evlist->mmap_len, prot, |
266 | MAP_SHARED, fd, 0); |
267 | if (evlist->mmap[idx].base == MAP_FAILED) { |
268 | if (evlist->cpus->map[idx] == -1 && evsel->attr.inherit) |
269 | ui__warning("Inherit is not allowed on per-task " |
270 | "events using mmap.\n"); |
271 | return -1; |
272 | } |
273 | |
274 | perf_evlist__add_pollfd(evlist, fd); |
275 | return 0; |
276 | } |
277 | |
278 | static int perf_evlist__mmap_per_cpu(struct perf_evlist *evlist, int prot, int mask) |
279 | { |
280 | struct perf_evsel *evsel; |
281 | int cpu, thread; |
282 | |
283 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { |
284 | int output = -1; |
285 | |
286 | for (thread = 0; thread < evlist->threads->nr; thread++) { |
287 | list_for_each_entry(evsel, &evlist->entries, node) { |
288 | int fd = FD(evsel, cpu, thread); |
289 | |
290 | if (output == -1) { |
291 | output = fd; |
292 | if (__perf_evlist__mmap(evlist, evsel, cpu, |
293 | prot, mask, output) < 0) |
294 | goto out_unmap; |
295 | } else { |
296 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) |
297 | goto out_unmap; |
298 | } |
299 | |
300 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && |
301 | perf_evlist__id_add_fd(evlist, evsel, cpu, thread, fd) < 0) |
302 | goto out_unmap; |
303 | } |
304 | } |
305 | } |
306 | |
307 | return 0; |
308 | |
309 | out_unmap: |
310 | for (cpu = 0; cpu < evlist->cpus->nr; cpu++) { |
311 | if (evlist->mmap[cpu].base != NULL) { |
312 | munmap(evlist->mmap[cpu].base, evlist->mmap_len); |
313 | evlist->mmap[cpu].base = NULL; |
314 | } |
315 | } |
316 | return -1; |
317 | } |
318 | |
319 | static int perf_evlist__mmap_per_thread(struct perf_evlist *evlist, int prot, int mask) |
320 | { |
321 | struct perf_evsel *evsel; |
322 | int thread; |
323 | |
324 | for (thread = 0; thread < evlist->threads->nr; thread++) { |
325 | int output = -1; |
326 | |
327 | list_for_each_entry(evsel, &evlist->entries, node) { |
328 | int fd = FD(evsel, 0, thread); |
329 | |
330 | if (output == -1) { |
331 | output = fd; |
332 | if (__perf_evlist__mmap(evlist, evsel, thread, |
333 | prot, mask, output) < 0) |
334 | goto out_unmap; |
335 | } else { |
336 | if (ioctl(fd, PERF_EVENT_IOC_SET_OUTPUT, output) != 0) |
337 | goto out_unmap; |
338 | } |
339 | |
340 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && |
341 | perf_evlist__id_add_fd(evlist, evsel, 0, thread, fd) < 0) |
342 | goto out_unmap; |
343 | } |
344 | } |
345 | |
346 | return 0; |
347 | |
348 | out_unmap: |
349 | for (thread = 0; thread < evlist->threads->nr; thread++) { |
350 | if (evlist->mmap[thread].base != NULL) { |
351 | munmap(evlist->mmap[thread].base, evlist->mmap_len); |
352 | evlist->mmap[thread].base = NULL; |
353 | } |
354 | } |
355 | return -1; |
356 | } |
357 | |
358 | /** perf_evlist__mmap - Create per cpu maps to receive events |
359 | * |
360 | * @evlist - list of events |
361 | * @pages - map length in pages |
362 | * @overwrite - overwrite older events? |
363 | * |
364 | * If overwrite is false the user needs to signal event consuption using: |
365 | * |
366 | * struct perf_mmap *m = &evlist->mmap[cpu]; |
367 | * unsigned int head = perf_mmap__read_head(m); |
368 | * |
369 | * perf_mmap__write_tail(m, head) |
370 | * |
371 | * Using perf_evlist__read_on_cpu does this automatically. |
372 | */ |
373 | int perf_evlist__mmap(struct perf_evlist *evlist, int pages, bool overwrite) |
374 | { |
375 | unsigned int page_size = sysconf(_SC_PAGE_SIZE); |
376 | int mask = pages * page_size - 1; |
377 | struct perf_evsel *evsel; |
378 | const struct cpu_map *cpus = evlist->cpus; |
379 | const struct thread_map *threads = evlist->threads; |
380 | int prot = PROT_READ | (overwrite ? 0 : PROT_WRITE); |
381 | |
382 | if (evlist->mmap == NULL && perf_evlist__alloc_mmap(evlist) < 0) |
383 | return -ENOMEM; |
384 | |
385 | if (evlist->pollfd == NULL && perf_evlist__alloc_pollfd(evlist) < 0) |
386 | return -ENOMEM; |
387 | |
388 | evlist->overwrite = overwrite; |
389 | evlist->mmap_len = (pages + 1) * page_size; |
390 | |
391 | list_for_each_entry(evsel, &evlist->entries, node) { |
392 | if ((evsel->attr.read_format & PERF_FORMAT_ID) && |
393 | evsel->sample_id == NULL && |
394 | perf_evsel__alloc_id(evsel, cpus->nr, threads->nr) < 0) |
395 | return -ENOMEM; |
396 | } |
397 | |
398 | if (evlist->cpus->map[0] == -1) |
399 | return perf_evlist__mmap_per_thread(evlist, prot, mask); |
400 | |
401 | return perf_evlist__mmap_per_cpu(evlist, prot, mask); |
402 | } |
403 | |
404 | int perf_evlist__create_maps(struct perf_evlist *evlist, pid_t target_pid, |
405 | pid_t target_tid, const char *cpu_list) |
406 | { |
407 | evlist->threads = thread_map__new(target_pid, target_tid); |
408 | |
409 | if (evlist->threads == NULL) |
410 | return -1; |
411 | |
412 | if (cpu_list == NULL && target_tid != -1) |
413 | evlist->cpus = cpu_map__dummy_new(); |
414 | else |
415 | evlist->cpus = cpu_map__new(cpu_list); |
416 | |
417 | if (evlist->cpus == NULL) |
418 | goto out_delete_threads; |
419 | |
420 | return 0; |
421 | |
422 | out_delete_threads: |
423 | thread_map__delete(evlist->threads); |
424 | return -1; |
425 | } |
426 | |
427 | void perf_evlist__delete_maps(struct perf_evlist *evlist) |
428 | { |
429 | cpu_map__delete(evlist->cpus); |
430 | thread_map__delete(evlist->threads); |
431 | evlist->cpus = NULL; |
432 | evlist->threads = NULL; |
433 | } |
434 | |
435 | int perf_evlist__set_filters(struct perf_evlist *evlist) |
436 | { |
437 | const struct thread_map *threads = evlist->threads; |
438 | const struct cpu_map *cpus = evlist->cpus; |
439 | struct perf_evsel *evsel; |
440 | char *filter; |
441 | int thread; |
442 | int cpu; |
443 | int err; |
444 | int fd; |
445 | |
446 | list_for_each_entry(evsel, &evlist->entries, node) { |
447 | filter = evsel->filter; |
448 | if (!filter) |
449 | continue; |
450 | for (cpu = 0; cpu < cpus->nr; cpu++) { |
451 | for (thread = 0; thread < threads->nr; thread++) { |
452 | fd = FD(evsel, cpu, thread); |
453 | err = ioctl(fd, PERF_EVENT_IOC_SET_FILTER, filter); |
454 | if (err) |
455 | return err; |
456 | } |
457 | } |
458 | } |
459 | |
460 | return 0; |
461 | } |
462 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9