When investigating some performance issues, I finally ended up in gthread-posix.c. There I found code such as:
static void __attribute__((noinline))
g_mutex_lock_slowpath (GMutex *mutex)
{
/* Set to 2 to indicate contention. If it was zero before then we
* just acquired the lock.
*
* Otherwise, sleep for as long as the 2 remains...
*/
while (exchange_acquire (&mutex->i[0], 2) != 0)
syscall (__NR_futex, &mutex->i[0], (gsize) FUTEX_WAIT, (gsize) 2, NULL);
}
I am curious as to why it doesn't use FUTEX_WAIT_PRIVATE here and it other places. At least on the ARM the non-private futexes are significantly slower, and I was under the impression that glib is for multithreading rather than interprocess communication in shared memory.
