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> | |
649fb6b3 | 28 | #include <stdbool.h> |
d58d1454 MD |
29 | #include <sys/mman.h> |
30 | #include <sys/syscall.h> | |
d58d1454 MD |
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> | |
77d7fa98 | 41 | #include "perf_event.h" |
d58d1454 MD |
42 | #include "lttng-tracer-core.h" |
43 | ||
44 | /* | |
45 | * We use a global perf counter key and iterate on per-thread RCU lists | |
46 | * of fields in the fast path, even though this is not strictly speaking | |
47 | * what would provide the best fast-path complexity, to ensure teardown | |
48 | * of sessions vs thread exit is handled racelessly. | |
49 | * | |
50 | * Updates and traversals of thread_list are protected by UST lock. | |
51 | * Updates to rcu_field_list are protected by UST lock. | |
52 | */ | |
53 | ||
54 | struct lttng_perf_counter_thread_field { | |
55 | struct lttng_perf_counter_field *field; /* Back reference */ | |
56 | struct perf_event_mmap_page *pc; | |
57 | struct cds_list_head thread_field_node; /* Per-field list of thread fields (node) */ | |
58 | struct cds_list_head rcu_field_node; /* RCU per-thread list of fields (node) */ | |
b9389e6e | 59 | int fd; /* Perf FD */ |
d58d1454 MD |
60 | }; |
61 | ||
62 | struct lttng_perf_counter_thread { | |
63 | struct cds_list_head rcu_field_list; /* RCU per-thread list of fields */ | |
64 | }; | |
65 | ||
66 | struct lttng_perf_counter_field { | |
67 | struct perf_event_attr attr; | |
68 | struct cds_list_head thread_field_list; /* Per-field list of thread fields */ | |
69 | }; | |
70 | ||
71 | static pthread_key_t perf_counter_key; | |
72 | ||
73 | static | |
53569322 | 74 | size_t perf_counter_get_size(struct lttng_ctx_field *field, size_t offset) |
d58d1454 MD |
75 | { |
76 | size_t size = 0; | |
77 | ||
78 | size += lib_ring_buffer_align(offset, lttng_alignof(uint64_t)); | |
79 | size += sizeof(uint64_t); | |
80 | return size; | |
81 | } | |
82 | ||
a3a8d943 MD |
83 | static |
84 | uint64_t read_perf_counter_syscall( | |
85 | struct lttng_perf_counter_thread_field *thread_field) | |
86 | { | |
87 | uint64_t count; | |
88 | ||
89 | if (caa_unlikely(thread_field->fd < 0)) | |
90 | return 0; | |
91 | ||
92 | if (caa_unlikely(read(thread_field->fd, &count, sizeof(count)) | |
93 | < sizeof(count))) | |
94 | return 0; | |
95 | ||
96 | return count; | |
97 | } | |
98 | ||
d58d1454 MD |
99 | #if defined(__x86_64__) || defined(__i386__) |
100 | ||
101 | static | |
102 | uint64_t rdpmc(unsigned int counter) | |
103 | { | |
104 | unsigned int low, high; | |
105 | ||
106 | asm volatile("rdpmc" : "=a" (low), "=d" (high) : "c" (counter)); | |
107 | ||
108 | return low | ((uint64_t) high) << 32; | |
109 | } | |
110 | ||
77d7fa98 MD |
111 | static |
112 | bool has_rdpmc(struct perf_event_mmap_page *pc) | |
113 | { | |
114 | if (caa_unlikely(!pc->cap_bit0_is_deprecated)) | |
115 | return false; | |
116 | /* Since Linux kernel 3.12. */ | |
117 | return pc->cap_user_rdpmc; | |
118 | } | |
119 | ||
d58d1454 | 120 | static |
a3a8d943 | 121 | uint64_t arch_read_perf_counter( |
d286ad50 | 122 | struct lttng_perf_counter_thread_field *thread_field) |
d58d1454 MD |
123 | { |
124 | uint32_t seq, idx; | |
125 | uint64_t count; | |
d286ad50 | 126 | struct perf_event_mmap_page *pc = thread_field->pc; |
d58d1454 MD |
127 | |
128 | if (caa_unlikely(!pc)) | |
129 | return 0; | |
130 | ||
131 | do { | |
132 | seq = CMM_LOAD_SHARED(pc->lock); | |
133 | cmm_barrier(); | |
134 | ||
135 | idx = pc->index; | |
77d7fa98 | 136 | if (caa_likely(has_rdpmc(pc) && idx)) { |
4f58f54f MD |
137 | int64_t pmcval; |
138 | ||
139 | pmcval = rdpmc(idx - 1); | |
140 | /* Sign-extend the pmc register result. */ | |
141 | pmcval <<= 64 - pc->pmc_width; | |
142 | pmcval >>= 64 - pc->pmc_width; | |
143 | count = pc->offset + pmcval; | |
144 | } else { | |
a3a8d943 MD |
145 | /* Fall-back on system call if rdpmc cannot be used. */ |
146 | return read_perf_counter_syscall(thread_field); | |
4f58f54f | 147 | } |
d58d1454 MD |
148 | cmm_barrier(); |
149 | } while (CMM_LOAD_SHARED(pc->lock) != seq); | |
150 | ||
151 | return count; | |
152 | } | |
153 | ||
d286ad50 | 154 | static |
a3a8d943 | 155 | int arch_perf_keep_fd(struct lttng_perf_counter_thread_field *thread_field) |
d286ad50 | 156 | { |
a3a8d943 | 157 | struct perf_event_mmap_page *pc = thread_field->pc; |
d286ad50 | 158 | |
a3a8d943 | 159 | if (!pc) |
d286ad50 | 160 | return 0; |
77d7fa98 | 161 | return !has_rdpmc(pc); |
a3a8d943 | 162 | } |
d286ad50 | 163 | |
a3a8d943 | 164 | #else |
d286ad50 | 165 | |
a3a8d943 MD |
166 | /* Generic (slow) implementation using a read system call. */ |
167 | static | |
168 | uint64_t arch_read_perf_counter( | |
169 | struct lttng_perf_counter_thread_field *thread_field) | |
170 | { | |
171 | return read_perf_counter_syscall(thread_field); | |
d286ad50 JD |
172 | } |
173 | ||
a3a8d943 MD |
174 | static |
175 | int arch_perf_keep_fd(struct lttng_perf_counter_thread_field *thread_field) | |
176 | { | |
177 | return 1; | |
178 | } | |
d286ad50 | 179 | |
a3a8d943 | 180 | #endif |
d286ad50 | 181 | |
d58d1454 MD |
182 | static |
183 | int sys_perf_event_open(struct perf_event_attr *attr, | |
184 | pid_t pid, int cpu, int group_fd, | |
185 | unsigned long flags) | |
186 | { | |
187 | return syscall(SYS_perf_event_open, attr, pid, cpu, | |
188 | group_fd, flags); | |
189 | } | |
190 | ||
191 | static | |
b9389e6e | 192 | int open_perf_fd(struct perf_event_attr *attr) |
d58d1454 | 193 | { |
b9389e6e | 194 | int fd; |
d58d1454 MD |
195 | |
196 | fd = sys_perf_event_open(attr, 0, -1, -1, 0); | |
197 | if (fd < 0) | |
b9389e6e JD |
198 | return -1; |
199 | ||
200 | return fd; | |
201 | } | |
202 | ||
d286ad50 JD |
203 | static |
204 | void close_perf_fd(int fd) | |
205 | { | |
206 | int ret; | |
207 | ||
208 | if (fd < 0) | |
209 | return; | |
210 | ||
211 | ret = close(fd); | |
212 | if (ret) { | |
213 | perror("Error closing LTTng-UST perf memory mapping FD"); | |
214 | } | |
215 | } | |
216 | ||
77d7fa98 | 217 | static void setup_perf(struct lttng_perf_counter_thread_field *thread_field) |
b9389e6e JD |
218 | { |
219 | void *perf_addr; | |
d58d1454 MD |
220 | |
221 | perf_addr = mmap(NULL, sizeof(struct perf_event_mmap_page), | |
b9389e6e | 222 | PROT_READ, MAP_SHARED, thread_field->fd, 0); |
d58d1454 | 223 | if (perf_addr == MAP_FAILED) |
b9389e6e | 224 | perf_addr = NULL; |
77d7fa98 | 225 | thread_field->pc = perf_addr; |
b9389e6e | 226 | |
a3a8d943 | 227 | if (!arch_perf_keep_fd(thread_field)) { |
b9389e6e JD |
228 | close_perf_fd(thread_field->fd); |
229 | thread_field->fd = -1; | |
6c2125af | 230 | } |
d58d1454 MD |
231 | } |
232 | ||
233 | static | |
234 | void unmap_perf_page(struct perf_event_mmap_page *pc) | |
235 | { | |
236 | int ret; | |
237 | ||
238 | if (!pc) | |
239 | return; | |
240 | ret = munmap(pc, sizeof(struct perf_event_mmap_page)); | |
241 | if (ret < 0) { | |
242 | PERROR("Error in munmap"); | |
243 | abort(); | |
244 | } | |
245 | } | |
246 | ||
247 | static | |
248 | struct lttng_perf_counter_thread *alloc_perf_counter_thread(void) | |
249 | { | |
250 | struct lttng_perf_counter_thread *perf_thread; | |
251 | sigset_t newmask, oldmask; | |
252 | int ret; | |
253 | ||
254 | ret = sigfillset(&newmask); | |
255 | if (ret) | |
256 | abort(); | |
257 | ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask); | |
258 | if (ret) | |
259 | abort(); | |
260 | /* Check again with signals disabled */ | |
261 | perf_thread = pthread_getspecific(perf_counter_key); | |
262 | if (perf_thread) | |
263 | goto skip; | |
264 | perf_thread = zmalloc(sizeof(*perf_thread)); | |
265 | if (!perf_thread) | |
266 | abort(); | |
267 | CDS_INIT_LIST_HEAD(&perf_thread->rcu_field_list); | |
268 | ret = pthread_setspecific(perf_counter_key, perf_thread); | |
269 | if (ret) | |
270 | abort(); | |
271 | skip: | |
272 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
273 | if (ret) | |
274 | abort(); | |
275 | return perf_thread; | |
276 | } | |
277 | ||
278 | static | |
279 | struct lttng_perf_counter_thread_field * | |
280 | add_thread_field(struct lttng_perf_counter_field *perf_field, | |
281 | struct lttng_perf_counter_thread *perf_thread) | |
282 | { | |
283 | struct lttng_perf_counter_thread_field *thread_field; | |
284 | sigset_t newmask, oldmask; | |
285 | int ret; | |
286 | ||
287 | ret = sigfillset(&newmask); | |
288 | if (ret) | |
289 | abort(); | |
290 | ret = pthread_sigmask(SIG_BLOCK, &newmask, &oldmask); | |
291 | if (ret) | |
292 | abort(); | |
293 | /* Check again with signals disabled */ | |
294 | cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list, | |
295 | rcu_field_node) { | |
296 | if (thread_field->field == perf_field) | |
297 | goto skip; | |
298 | } | |
299 | thread_field = zmalloc(sizeof(*thread_field)); | |
300 | if (!thread_field) | |
301 | abort(); | |
302 | thread_field->field = perf_field; | |
b9389e6e JD |
303 | thread_field->fd = open_perf_fd(&perf_field->attr); |
304 | if (thread_field->fd >= 0) | |
77d7fa98 | 305 | setup_perf(thread_field); |
b9389e6e JD |
306 | /* |
307 | * Note: thread_field->pc can be NULL if setup_perf() fails. | |
308 | * Also, thread_field->fd can be -1 if open_perf_fd() fails. | |
309 | */ | |
d58d1454 MD |
310 | ust_lock_nocheck(); |
311 | cds_list_add_rcu(&thread_field->rcu_field_node, | |
312 | &perf_thread->rcu_field_list); | |
313 | cds_list_add(&thread_field->thread_field_node, | |
314 | &perf_field->thread_field_list); | |
315 | ust_unlock(); | |
316 | skip: | |
317 | ret = pthread_sigmask(SIG_SETMASK, &oldmask, NULL); | |
318 | if (ret) | |
319 | abort(); | |
320 | return thread_field; | |
321 | } | |
322 | ||
323 | static | |
324 | struct lttng_perf_counter_thread_field * | |
325 | get_thread_field(struct lttng_perf_counter_field *field) | |
326 | { | |
327 | struct lttng_perf_counter_thread *perf_thread; | |
328 | struct lttng_perf_counter_thread_field *thread_field; | |
329 | ||
330 | perf_thread = pthread_getspecific(perf_counter_key); | |
331 | if (!perf_thread) | |
332 | perf_thread = alloc_perf_counter_thread(); | |
333 | cds_list_for_each_entry_rcu(thread_field, &perf_thread->rcu_field_list, | |
334 | rcu_field_node) { | |
335 | if (thread_field->field == field) | |
336 | return thread_field; | |
337 | } | |
338 | /* perf_counter_thread_field not found, need to add one */ | |
339 | return add_thread_field(field, perf_thread); | |
340 | } | |
341 | ||
342 | static | |
343 | uint64_t wrapper_perf_counter_read(struct lttng_ctx_field *field) | |
344 | { | |
345 | struct lttng_perf_counter_field *perf_field; | |
346 | struct lttng_perf_counter_thread_field *perf_thread_field; | |
347 | ||
348 | perf_field = field->u.perf_counter; | |
349 | perf_thread_field = get_thread_field(perf_field); | |
a3a8d943 | 350 | return arch_read_perf_counter(perf_thread_field); |
d58d1454 MD |
351 | } |
352 | ||
353 | static | |
354 | void perf_counter_record(struct lttng_ctx_field *field, | |
355 | struct lttng_ust_lib_ring_buffer_ctx *ctx, | |
356 | struct lttng_channel *chan) | |
357 | { | |
358 | uint64_t value; | |
359 | ||
360 | value = wrapper_perf_counter_read(field); | |
361 | lib_ring_buffer_align_ctx(ctx, lttng_alignof(value)); | |
362 | chan->ops->event_write(ctx, &value, sizeof(value)); | |
363 | } | |
364 | ||
365 | static | |
366 | void perf_counter_get_value(struct lttng_ctx_field *field, | |
53569322 | 367 | struct lttng_ctx_value *value) |
d58d1454 MD |
368 | { |
369 | uint64_t v; | |
370 | ||
371 | v = wrapper_perf_counter_read(field); | |
53569322 | 372 | value->u.s64 = v; |
d58d1454 MD |
373 | } |
374 | ||
375 | /* Called with UST lock held */ | |
376 | static | |
377 | void lttng_destroy_perf_thread_field( | |
378 | struct lttng_perf_counter_thread_field *thread_field) | |
379 | { | |
b9389e6e | 380 | close_perf_fd(thread_field->fd); |
d58d1454 MD |
381 | unmap_perf_page(thread_field->pc); |
382 | cds_list_del_rcu(&thread_field->rcu_field_node); | |
383 | cds_list_del(&thread_field->thread_field_node); | |
384 | free(thread_field); | |
385 | } | |
386 | ||
387 | static | |
388 | void lttng_destroy_perf_thread_key(void *_key) | |
389 | { | |
390 | struct lttng_perf_counter_thread *perf_thread = _key; | |
391 | struct lttng_perf_counter_thread_field *pos, *p; | |
392 | ||
393 | ust_lock_nocheck(); | |
394 | cds_list_for_each_entry_safe(pos, p, &perf_thread->rcu_field_list, | |
395 | rcu_field_node) | |
396 | lttng_destroy_perf_thread_field(pos); | |
397 | ust_unlock(); | |
398 | free(perf_thread); | |
399 | } | |
400 | ||
401 | /* Called with UST lock held */ | |
402 | static | |
403 | void lttng_destroy_perf_counter_field(struct lttng_ctx_field *field) | |
404 | { | |
405 | struct lttng_perf_counter_field *perf_field; | |
406 | struct lttng_perf_counter_thread_field *pos, *p; | |
407 | ||
408 | free((char *) field->event_field.name); | |
409 | perf_field = field->u.perf_counter; | |
410 | /* | |
411 | * This put is performed when no threads can concurrently | |
412 | * perform a "get" concurrently, thanks to urcu-bp grace | |
413 | * period. | |
414 | */ | |
415 | cds_list_for_each_entry_safe(pos, p, &perf_field->thread_field_list, | |
416 | thread_field_node) | |
417 | lttng_destroy_perf_thread_field(pos); | |
418 | free(perf_field); | |
419 | } | |
420 | ||
d286ad50 JD |
421 | #ifdef __ARM_ARCH_7A__ |
422 | ||
423 | static | |
424 | int perf_get_exclude_kernel(void) | |
425 | { | |
426 | return 0; | |
427 | } | |
428 | ||
429 | #else /* __ARM_ARCH_7A__ */ | |
430 | ||
431 | static | |
432 | int perf_get_exclude_kernel(void) | |
433 | { | |
434 | return 1; | |
435 | } | |
436 | ||
437 | #endif /* __ARM_ARCH_7A__ */ | |
438 | ||
d58d1454 MD |
439 | /* Called with UST lock held */ |
440 | int lttng_add_perf_counter_to_ctx(uint32_t type, | |
441 | uint64_t config, | |
442 | const char *name, | |
443 | struct lttng_ctx **ctx) | |
444 | { | |
445 | struct lttng_ctx_field *field; | |
446 | struct lttng_perf_counter_field *perf_field; | |
d58d1454 MD |
447 | char *name_alloc; |
448 | int ret; | |
449 | ||
450 | name_alloc = strdup(name); | |
451 | if (!name_alloc) { | |
452 | ret = -ENOMEM; | |
453 | goto name_alloc_error; | |
454 | } | |
455 | perf_field = zmalloc(sizeof(*perf_field)); | |
456 | if (!perf_field) { | |
457 | ret = -ENOMEM; | |
458 | goto perf_field_alloc_error; | |
459 | } | |
460 | field = lttng_append_context(ctx); | |
461 | if (!field) { | |
462 | ret = -ENOMEM; | |
463 | goto append_context_error; | |
464 | } | |
465 | if (lttng_find_context(*ctx, name_alloc)) { | |
466 | ret = -EEXIST; | |
467 | goto find_error; | |
468 | } | |
469 | ||
470 | field->destroy = lttng_destroy_perf_counter_field; | |
471 | ||
472 | field->event_field.name = name_alloc; | |
473 | field->event_field.type.atype = atype_integer; | |
474 | field->event_field.type.u.basic.integer.size = | |
475 | sizeof(uint64_t) * CHAR_BIT; | |
476 | field->event_field.type.u.basic.integer.alignment = | |
477 | lttng_alignof(uint64_t) * CHAR_BIT; | |
478 | field->event_field.type.u.basic.integer.signedness = | |
479 | lttng_is_signed_type(uint64_t); | |
480 | field->event_field.type.u.basic.integer.reverse_byte_order = 0; | |
481 | field->event_field.type.u.basic.integer.base = 10; | |
482 | field->event_field.type.u.basic.integer.encoding = lttng_encode_none; | |
483 | field->get_size = perf_counter_get_size; | |
484 | field->record = perf_counter_record; | |
485 | field->get_value = perf_counter_get_value; | |
486 | ||
487 | perf_field->attr.type = type; | |
488 | perf_field->attr.config = config; | |
d286ad50 | 489 | perf_field->attr.exclude_kernel = perf_get_exclude_kernel(); |
d58d1454 MD |
490 | CDS_INIT_LIST_HEAD(&perf_field->thread_field_list); |
491 | field->u.perf_counter = perf_field; | |
492 | ||
493 | /* Ensure that this perf counter can be used in this process. */ | |
b9389e6e JD |
494 | ret = open_perf_fd(&perf_field->attr); |
495 | if (ret < 0) { | |
d58d1454 MD |
496 | ret = -ENODEV; |
497 | goto setup_error; | |
498 | } | |
b9389e6e | 499 | close_perf_fd(ret); |
d58d1454 MD |
500 | |
501 | /* | |
502 | * Contexts can only be added before tracing is started, so we | |
503 | * don't have to synchronize against concurrent threads using | |
504 | * the field here. | |
505 | */ | |
506 | ||
b2cc986a | 507 | lttng_context_update(*ctx); |
d58d1454 MD |
508 | return 0; |
509 | ||
510 | setup_error: | |
511 | find_error: | |
512 | lttng_remove_context_field(ctx, field); | |
513 | append_context_error: | |
514 | free(perf_field); | |
515 | perf_field_alloc_error: | |
516 | free(name_alloc); | |
517 | name_alloc_error: | |
518 | return ret; | |
519 | } | |
520 | ||
521 | int lttng_perf_counter_init(void) | |
522 | { | |
523 | int ret; | |
524 | ||
525 | ret = pthread_key_create(&perf_counter_key, | |
526 | lttng_destroy_perf_thread_key); | |
527 | if (ret) | |
528 | ret = -ret; | |
529 | return ret; | |
530 | } | |
531 | ||
532 | void lttng_perf_counter_exit(void) | |
533 | { | |
534 | int ret; | |
535 | ||
536 | ret = pthread_key_delete(perf_counter_key); | |
537 | if (ret) { | |
538 | errno = ret; | |
539 | PERROR("Error in pthread_key_delete"); | |
540 | } | |
541 | } |