1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
|
#include <sys/membarrier.h>
#include <semaphore.h>
#include <signal.h>
#include <string.h>
#include "pthread_impl.h"
#include "syscall.h"
static void dummy_0(void)
{
}
static void dummy_1(pthread_t t)
{
}
weak_alias(dummy_0, __tl_lock);
weak_alias(dummy_0, __tl_unlock);
weak_alias(dummy_1, __tl_sync);
static sem_t barrier_sem;
static void bcast_barrier(int s)
{
sem_post(&barrier_sem);
}
int __membarrier(int cmd, int flags)
{
int r = __syscall(SYS_membarrier, cmd, flags);
/* Emulate the private expedited command, which is needed by the
* dynamic linker for installation of dynamic TLS, for older
* kernels that lack the syscall. Unlike the syscall, this only
* synchronizes with threads of the process, not other processes
* sharing the VM, but such sharing is not a supported usage
* anyway. */
if (r && cmd == MEMBARRIER_CMD_PRIVATE_EXPEDITED && !flags) {
pthread_t self=__pthread_self(), td;
sigset_t set;
__block_app_sigs(&set);
__tl_lock();
sem_init(&barrier_sem, 0, 0);
struct sigaction sa = {
.sa_flags = SA_RESTART,
.sa_handler = bcast_barrier
};
memset(&sa.sa_mask, -1, sizeof sa.sa_mask);
__libc_sigaction(SIGSYNCCALL, &sa, 0);
for (td=self->next; td!=self; td=td->next)
__syscall(SYS_tkill, td->tid, SIGSYNCCALL);
for (td=self->next; td!=self; td=td->next)
sem_wait(&barrier_sem);
sa.sa_handler = SIG_IGN;
__libc_sigaction(SIGSYNCCALL, &sa, 0);
sem_destroy(&barrier_sem);
__tl_unlock();
__restore_sigs(&set);
return 0;
}
return __syscall_ret(r);
}
void __membarrier_init(void)
{
/* If membarrier is linked, attempt to pre-register to be able to use
* the private expedited command before the process becomes multi-
* threaded, since registering later has bad, potentially unbounded
* latency. This syscall should be essentially free, and it's arguably
* a mistake in the API design that registration was even required.
* For other commands, registration may impose some cost, so it's left
* to the application to do so if desired. Unfortunately this means
* library code initialized after the process becomes multi-threaded
* cannot use these features without accepting registration latency. */
__syscall(SYS_membarrier, MEMBARRIER_CMD_REGISTER_PRIVATE_EXPEDITED, 0);
}
weak_alias(__membarrier, membarrier);
|
