Commit | Line | Data |
---|---|---|
d58d1454 MD |
1 | /* |
2 | * lttng-context-perf-counters.c | |
3 | * | |
4 | * LTTng UST performance monitoring counters (perf-counters) integration. | |
5 | * | |
6 | * Copyright (C) 2009-2014 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> | |
7 | * | |
8 | * This library is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License as published by the Free Software Foundation; only | |
11 | * version 2.1 of the License. | |
12 | * | |
13 | * This library is distributed in the hope that it will be useful, | |
14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 | * Lesser General Public License for more details. | |
17 | * | |
18 | * You should have received a copy of the GNU Lesser General Public | |
19 | * License along with this library; if not, write to the Free Software | |
20 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
21 | */ | |
22 | ||
23 | #include <sys/types.h> | |
24 | #include <unistd.h> | |
25 | #include <string.h> | |
26 | #include <stdlib.h> | |
27 | #include <stdio.h> | |
28 | #include <sys/mman.h> | |
29 | #include <sys/syscall.h> | |
30 | #include <linux/perf_event.h> | |
31 | #include <lttng/ust-events.h> | |
32 | #include <lttng/ust-tracer.h> | |
33 | #include <lttng/ringbuffer-config.h> | |
34 | #include <urcu/system.h> | |
35 | #include <urcu/arch.h> | |
36 | #include <urcu/rculist.h> | |
37 | #include <helper.h> | |
38 | #include <urcu/ref.h> | |
39 | #include <usterr-signal-safe.h> | |
40 | #include <signal.h> | |
41 | #include "lttng-tracer-core.h" | |
42 | ||
43 | /* | |
44 | * We use a global perf counter key and iterate on per-thread RCU lists | |
45 | * of fields in the fast path, even though this is not strictly speaking | |
46 | * what would provide the best fast-path complexity, to ensure teardown | |
47 | * of sessions vs thread exit is handled racelessly. | |
48 | * | |
49 | * Updates and traversals of thread_list are protected by UST lock. | |
50 | * Updates to rcu_field_list are protected by UST lock. | |
51 | */ | |
52 | ||
53 | struct lttng_perf_counter_thread_field { | |
54 | struct lttng_perf_counter_field *field; /* Back reference */ | |
55 | struct perf_event_mmap_page *pc; | |
56 | struct cds_list_head thread_field_node; /* Per-field list of thread fields (node) */ | |
57 | struct cds_list_head rcu_field_node; /* RCU per-thread list of fields (node) */ | |
b9389e6e | 58 | int fd; /* Perf FD */ |
d58d1454 MD |
59 | }; |
60 | ||
61 | struct lttng_perf_counter_thread { | |
62 | struct cds_list_head rcu_field_list; /* RCU per-thread list of fields */ | |
63 | }; | |
64 | ||
65 | struct lttng_perf_counter_field { | |
66 | struct perf_event_attr attr; | |
67 | struct cds_list_head thread_field_list; /* Per-field list of thread fields */ | |
68 | }; | |
69 | ||
70 | static pthread_key_t perf_counter_key; | |
71 | ||
72 | static | |
53569322 | 73 | size_t perf_counter_get_size(struct lttng_ctx_field *field, size_t offset) |
d58d1454 MD |
74 | { |
75 | size_t size = 0; | |
76 | ||
77 | size += lib_ring_buffer_align(offset, lttng_alignof(uint64_t)); | |
78 | size += sizeof(uint64_t); | |
79 | return size; | |
80 | } | |
81 | ||
82 | #if defined(__x86_64__) || defined(__i386__) | |
83 | ||
84 | static | |
85 | uint64_t rdpmc(unsigned int counter) | |
86 | { | |
87 | unsigned int low, high; | |
88 | ||
89 | asm volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (counter)); | |
90 | ||
91 | return low | ((uint64_t) high) << 32; | |
92 | } | |
93 | ||
b9389e6e JD |
94 | static bool arch_perf_use_read(void) |
95 | { | |
96 | return false; | |
97 | } | |
98 | ||
d58d1454 | 99 | static |
d286ad50 JD |
100 | uint64_t read_perf_counter( |
101 | struct lttng_perf_counter_thread_field *thread_field) | |
d58d1454 MD |
102 | { |
103 | uint32_t seq, idx; | |
104 | uint64_t count; | |
d286ad50 | 105 | struct perf_event_mmap_page *pc = thread_field->pc; |
d58d1454 MD |
106 | |
107 | if (caa_unlikely(!pc)) | |
108 | return 0; | |
109 | ||
110 | do { | |
111 | seq = CMM_LOAD_SHARED(pc->lock); | |
112 | cmm_barrier(); | |
113 | ||
114 | idx = pc->index; | |
115 | if (idx) | |
116 | count = pc->offset + rdpmc(idx - 1); | |
117 | else | |
118 | count = 0; | |
119 | ||
120 | cmm_barrier(); | |
121 | } while (CMM_LOAD_SHARED(pc->lock) != seq); | |
122 | ||
123 | return count; | |
124 | } | |
125 | ||
d286ad50 JD |
126 | #elif defined (__ARM_ARCH_7A__) |
127 | ||
128 | static bool arch_perf_use_read(void) | |
129 | { | |
130 | return true; | |
131 | } | |
132 | ||
133 | static | |
134 | uint64_t read_perf_counter( | |
135 | struct lttng_perf_counter_thread_field *thread_field) | |
136 | { | |
137 | uint64_t count; | |
138 | ||
139 | if (caa_unlikely(thread_field->fd < 0)) | |
140 | return 0; | |
141 | ||
142 | if (caa_unlikely(read(thread_field->fd, &count, sizeof(count)) | |
143 | < sizeof(count))) | |
144 | return 0; | |
145 | ||
146 | return count; | |
147 | } | |
148 | ||
149 | #else /* defined(__x86_64__) || defined(__i386__) || defined(__ARM_ARCH_7A__) */ | |
150 | ||
151 | #error "Perf event counters are only supported on x86 and ARMv7 so far." | |
152 | ||
153 | #endif /* #else defined(__x86_64__) || defined(__i386__) || defined(__ARM_ARCH_7A__) */ | |
154 | ||
d58d1454 MD |
155 | static |
156 | int sys_perf_event_open(struct perf_event_attr *attr, | |
157 | pid_t pid, int cpu, int group_fd, | |
158 | unsigned long flags) | |
159 | { | |
160 | return syscall(SYS_perf_event_open, attr, pid, cpu, | |
161 | group_fd, flags); | |
162 | } | |
163 | ||
164 | static | |
b9389e6e | 165 | int open_perf_fd(struct perf_event_attr *attr) |
d58d1454 | 166 | { |
b9389e6e | 167 | int fd; |
d58d1454 MD |
168 | |
169 | fd = sys_perf_event_open(attr, 0, -1, -1, 0); | |
170 | if (fd < 0) | |
b9389e6e JD |
171 | return -1; |
172 | ||
173 | return fd; | |
174 | } | |
175 | ||
d286ad50 JD |
176 | static |
177 | void close_perf_fd(int fd) | |
178 | { | |
179 | int ret; | |
180 | ||
181 | if (fd < 0) | |
182 | return; | |
183 | ||
184 | ret = close(fd); | |
185 | if (ret) { | |
186 | perror("Error closing LTTng-UST perf memory mapping FD"); | |
187 | } | |
188 | } | |
189 | ||
b9389e6e JD |
190 | static |
191 | struct perf_event_mmap_page *setup_perf( | |
192 | struct lttng_perf_counter_thread_field *thread_field) | |
193 | { | |
194 | void *perf_addr; | |
d58d1454 MD |
195 | |
196 | perf_addr = mmap(NULL, sizeof(struct perf_event_mmap_page), | |
b9389e6e | 197 | PROT_READ, MAP_SHARED, thread_field->fd, 0); |
d58d1454 | 198 | if (perf_addr == MAP_FAILED) |
b9389e6e JD |
199 | perf_addr = NULL; |
200 | ||
201 | if (!arch_perf_use_read()) { | |
202 | close_perf_fd(thread_field->fd); | |
203 | thread_field->fd = -1; | |
6c2125af | 204 | } |
b9389e6e | 205 | |
d58d1454 MD |
206 | return perf_addr; |
207 | } | |
208 | ||
209 | static | |
210 | void unmap_perf_page(struct perf_event_mmap_page *pc) | |
211 | { | |
212 | int ret; | |
213 | ||
214 | if (!pc) | |
215 | return; | |
216 | ret = munmap(pc, sizeof(struct perf_event_mmap_page)); | |
217 | if (ret < 0) { | |
218 | PERROR("Error in munmap"); | |
219 | abort(); | |
220 | } | |
221 | } | |
222 | ||
223 | static | |
224 | struct lttng_perf_counter_thread *alloc_perf_counter_thread(void) | |
225 | { | |
226 | struct lttng_perf_counter_thread *perf_thread; | |
227 | sigset_t newmask, oldmask; | |
228 | int ret; | |
229 | ||
230 | ret = sigfillset(&newmask); | |
231 | if (ret) | |
232 | abort(); | |
233 | ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask); | |
234 | if (ret) | |
235 | abort(); | |
236 | /* Check again with signals disabled */ | |
237 | perf_thread = pthread_getspecific(perf_counter_key); | |
238 | if (perf_thread) | |
239 | goto skip; | |
240 | perf_thread = zmalloc(sizeof(*perf_thread)); | |
241 | if (!perf_thread) | |
242 | abort(); | |
243 | CDS_INIT_LIST_HEAD(&perf_thread->rcu_field_list); | |
244 | ret = pthread_setspecific(perf_counter_key, perf_thread); | |
245 | if (ret) | |
246 | abort(); | |
247 | skip: | |
248 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
249 | if (ret) | |
250 | abort(); | |
251 | return perf_thread; | |
252 | } | |
253 | ||
254 | static | |
255 | struct lttng_perf_counter_thread_field * | |
256 | add_thread_field(struct lttng_perf_counter_field *perf_field, | |
257 | struct lttng_perf_counter_thread *perf_thread) | |
258 | { | |
259 | struct lttng_perf_counter_thread_field *thread_field; | |
260 | sigset_t newmask, oldmask; | |
261 | int ret; | |
262 | ||
263 | ret = sigfillset(&newmask); | |
264 | if (ret) | |
265 | abort(); | |
266 | ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask); | |
267 | if (ret) | |
268 | abort(); | |
269 | /* Check again with signals disabled */ | |
270 | cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list, | |
271 | rcu_field_node) { | |
272 | if (thread_field->field == perf_field) | |
273 | goto skip; | |
274 | } | |
275 | thread_field = zmalloc(sizeof(*thread_field)); | |
276 | if (!thread_field) | |
277 | abort(); | |
278 | thread_field->field = perf_field; | |
b9389e6e JD |
279 | thread_field->fd = open_perf_fd(&perf_field->attr); |
280 | if (thread_field->fd >= 0) | |
281 | thread_field->pc = setup_perf(thread_field); | |
282 | /* | |
283 | * Note: thread_field->pc can be NULL if setup_perf() fails. | |
284 | * Also, thread_field->fd can be -1 if open_perf_fd() fails. | |
285 | */ | |
d58d1454 MD |
286 | ust_lock_nocheck(); |
287 | cds_list_add_rcu(&thread_field->rcu_field_node, | |
288 | &perf_thread->rcu_field_list); | |
289 | cds_list_add(&thread_field->thread_field_node, | |
290 | &perf_field->thread_field_list); | |
291 | ust_unlock(); | |
292 | skip: | |
293 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
294 | if (ret) | |
295 | abort(); | |
296 | return thread_field; | |
297 | } | |
298 | ||
299 | static | |
300 | struct lttng_perf_counter_thread_field * | |
301 | get_thread_field(struct lttng_perf_counter_field *field) | |
302 | { | |
303 | struct lttng_perf_counter_thread *perf_thread; | |
304 | struct lttng_perf_counter_thread_field *thread_field; | |
305 | ||
306 | perf_thread = pthread_getspecific(perf_counter_key); | |
307 | if (!perf_thread) | |
308 | perf_thread = alloc_perf_counter_thread(); | |
309 | cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list, | |
310 | rcu_field_node) { | |
311 | if (thread_field->field == field) | |
312 | return thread_field; | |
313 | } | |
314 | /* perf_counter_thread_field not found, need to add one */ | |
315 | return add_thread_field(field, perf_thread); | |
316 | } | |
317 | ||
318 | static | |
319 | uint64_t wrapper_perf_counter_read(struct lttng_ctx_field *field) | |
320 | { | |
321 | struct lttng_perf_counter_field *perf_field; | |
322 | struct lttng_perf_counter_thread_field *perf_thread_field; | |
323 | ||
324 | perf_field = field->u.perf_counter; | |
325 | perf_thread_field = get_thread_field(perf_field); | |
d286ad50 | 326 | return read_perf_counter(perf_thread_field); |
d58d1454 MD |
327 | } |
328 | ||
329 | static | |
330 | void perf_counter_record(struct lttng_ctx_field *field, | |
331 | struct lttng_ust_lib_ring_buffer_ctx *ctx, | |
332 | struct lttng_channel *chan) | |
333 | { | |
334 | uint64_t value; | |
335 | ||
336 | value = wrapper_perf_counter_read(field); | |
337 | lib_ring_buffer_align_ctx(ctx, lttng_alignof(value)); | |
338 | chan->ops->event_write(ctx, &value, sizeof(value)); | |
339 | } | |
340 | ||
341 | static | |
342 | void perf_counter_get_value(struct lttng_ctx_field *field, | |
53569322 | 343 | struct lttng_ctx_value *value) |
d58d1454 MD |
344 | { |
345 | uint64_t v; | |
346 | ||
347 | v = wrapper_perf_counter_read(field); | |
53569322 | 348 | value->u.s64 = v; |
d58d1454 MD |
349 | } |
350 | ||
351 | /* Called with UST lock held */ | |
352 | static | |
353 | void lttng_destroy_perf_thread_field( | |
354 | struct lttng_perf_counter_thread_field *thread_field) | |
355 | { | |
b9389e6e | 356 | close_perf_fd(thread_field->fd); |
d58d1454 MD |
357 | unmap_perf_page(thread_field->pc); |
358 | cds_list_del_rcu(&thread_field->rcu_field_node); | |
359 | cds_list_del(&thread_field->thread_field_node); | |
360 | free(thread_field); | |
361 | } | |
362 | ||
363 | static | |
364 | void lttng_destroy_perf_thread_key(void *_key) | |
365 | { | |
366 | struct lttng_perf_counter_thread *perf_thread = _key; | |
367 | struct lttng_perf_counter_thread_field *pos, *p; | |
368 | ||
369 | ust_lock_nocheck(); | |
370 | cds_list_for_each_entry_safe(pos, p, &perf_thread->rcu_field_list, | |
371 | rcu_field_node) | |
372 | lttng_destroy_perf_thread_field(pos); | |
373 | ust_unlock(); | |
374 | free(perf_thread); | |
375 | } | |
376 | ||
377 | /* Called with UST lock held */ | |
378 | static | |
379 | void lttng_destroy_perf_counter_field(struct lttng_ctx_field *field) | |
380 | { | |
381 | struct lttng_perf_counter_field *perf_field; | |
382 | struct lttng_perf_counter_thread_field *pos, *p; | |
383 | ||
384 | free((char *) field->event_field.name); | |
385 | perf_field = field->u.perf_counter; | |
386 | /* | |
387 | * This put is performed when no threads can concurrently | |
388 | * perform a "get" concurrently, thanks to urcu-bp grace | |
389 | * period. | |
390 | */ | |
391 | cds_list_for_each_entry_safe(pos, p, &perf_field->thread_field_list, | |
392 | thread_field_node) | |
393 | lttng_destroy_perf_thread_field(pos); | |
394 | free(perf_field); | |
395 | } | |
396 | ||
d286ad50 JD |
397 | #ifdef __ARM_ARCH_7A__ |
398 | ||
399 | static | |
400 | int perf_get_exclude_kernel(void) | |
401 | { | |
402 | return 0; | |
403 | } | |
404 | ||
405 | #else /* __ARM_ARCH_7A__ */ | |
406 | ||
407 | static | |
408 | int perf_get_exclude_kernel(void) | |
409 | { | |
410 | return 1; | |
411 | } | |
412 | ||
413 | #endif /* __ARM_ARCH_7A__ */ | |
414 | ||
d58d1454 MD |
415 | /* Called with UST lock held */ |
416 | int lttng_add_perf_counter_to_ctx(uint32_t type, | |
417 | uint64_t config, | |
418 | const char *name, | |
419 | struct lttng_ctx **ctx) | |
420 | { | |
421 | struct lttng_ctx_field *field; | |
422 | struct lttng_perf_counter_field *perf_field; | |
d58d1454 MD |
423 | char *name_alloc; |
424 | int ret; | |
425 | ||
426 | name_alloc = strdup(name); | |
427 | if (!name_alloc) { | |
428 | ret = -ENOMEM; | |
429 | goto name_alloc_error; | |
430 | } | |
431 | perf_field = zmalloc(sizeof(*perf_field)); | |
432 | if (!perf_field) { | |
433 | ret = -ENOMEM; | |
434 | goto perf_field_alloc_error; | |
435 | } | |
436 | field = lttng_append_context(ctx); | |
437 | if (!field) { | |
438 | ret = -ENOMEM; | |
439 | goto append_context_error; | |
440 | } | |
441 | if (lttng_find_context(*ctx, name_alloc)) { | |
442 | ret = -EEXIST; | |
443 | goto find_error; | |
444 | } | |
445 | ||
446 | field->destroy = lttng_destroy_perf_counter_field; | |
447 | ||
448 | field->event_field.name = name_alloc; | |
449 | field->event_field.type.atype = atype_integer; | |
450 | field->event_field.type.u.basic.integer.size = | |
451 | sizeof(uint64_t) * CHAR_BIT; | |
452 | field->event_field.type.u.basic.integer.alignment = | |
453 | lttng_alignof(uint64_t) * CHAR_BIT; | |
454 | field->event_field.type.u.basic.integer.signedness = | |
455 | lttng_is_signed_type(uint64_t); | |
456 | field->event_field.type.u.basic.integer.reverse_byte_order = 0; | |
457 | field->event_field.type.u.basic.integer.base = 10; | |
458 | field->event_field.type.u.basic.integer.encoding = lttng_encode_none; | |
459 | field->get_size = perf_counter_get_size; | |
460 | field->record = perf_counter_record; | |
461 | field->get_value = perf_counter_get_value; | |
462 | ||
463 | perf_field->attr.type = type; | |
464 | perf_field->attr.config = config; | |
d286ad50 | 465 | perf_field->attr.exclude_kernel = perf_get_exclude_kernel(); |
d58d1454 MD |
466 | CDS_INIT_LIST_HEAD(&perf_field->thread_field_list); |
467 | field->u.perf_counter = perf_field; | |
468 | ||
469 | /* Ensure that this perf counter can be used in this process. */ | |
b9389e6e JD |
470 | ret = open_perf_fd(&perf_field->attr); |
471 | if (ret < 0) { | |
d58d1454 MD |
472 | ret = -ENODEV; |
473 | goto setup_error; | |
474 | } | |
b9389e6e | 475 | close_perf_fd(ret); |
d58d1454 MD |
476 | |
477 | /* | |
478 | * Contexts can only be added before tracing is started, so we | |
479 | * don't have to synchronize against concurrent threads using | |
480 | * the field here. | |
481 | */ | |
482 | ||
b2cc986a | 483 | lttng_context_update(*ctx); |
d58d1454 MD |
484 | return 0; |
485 | ||
486 | setup_error: | |
487 | find_error: | |
488 | lttng_remove_context_field(ctx, field); | |
489 | append_context_error: | |
490 | free(perf_field); | |
491 | perf_field_alloc_error: | |
492 | free(name_alloc); | |
493 | name_alloc_error: | |
494 | return ret; | |
495 | } | |
496 | ||
497 | int lttng_perf_counter_init(void) | |
498 | { | |
499 | int ret; | |
500 | ||
501 | ret = pthread_key_create(&perf_counter_key, | |
502 | lttng_destroy_perf_thread_key); | |
503 | if (ret) | |
504 | ret = -ret; | |
505 | return ret; | |
506 | } | |
507 | ||
508 | void lttng_perf_counter_exit(void) | |
509 | { | |
510 | int ret; | |
511 | ||
512 | ret = pthread_key_delete(perf_counter_key); | |
513 | if (ret) { | |
514 | errno = ret; | |
515 | PERROR("Error in pthread_key_delete"); | |
516 | } | |
517 | } |