int pthread_barrier_wait(pthread_barrier_t *b)
{
int limit = b->_b_limit;
struct instance *inst;
/* Trivial case: count was set at 1 */
if (!limit) return PTHREAD_BARRIER_SERIAL_THREAD;
/* Process-shared barriers require a separate, inefficient wait */
if (limit < 0) return pshared_barrier_wait(b);
/* Otherwise we need a lock on the barrier object */
while (a_swap(&b->_b_lock, 1))
__wait(&b->_b_lock, &b->_b_waiters, 1, 1);
inst = b->_b_inst;
/* First thread to enter the barrier becomes the "instance owner" */
if (!inst) {
struct instance new_inst = { 0 };
int spins = 10000;
b->_b_inst = inst = &new_inst;
a_store(&b->_b_lock, 0);
if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
while (spins-- && !inst->finished)
a_spin();
a_inc(&inst->finished);
while (inst->finished == 1)
__syscall(SYS_futex, &inst->finished, FUTEX_WAIT,1,0);
return PTHREAD_BARRIER_SERIAL_THREAD;
}
/* Last thread to enter the barrier wakes all non-instance-owners */
if (++inst->count == limit) {
b->_b_inst = 0;
a_store(&b->_b_lock, 0);
if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
a_store(&inst->last, 1);
if (inst->waiters)
__wake(&inst->last, -1, 1);
} else {
a_store(&b->_b_lock, 0);
if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
__wait(&inst->last, &inst->waiters, 0, 1);
}
/* Last thread to exit the barrier wakes the instance owner */
if (a_fetch_add(&inst->count,-1)==1 && a_fetch_add(&inst->finished,1))
__wake(&inst->finished, 1, 1);
return 0;
}
char *tempnam(const char *dir, const char *pfx)
{
static int index;
char *s;
struct timespec ts;
int pid = getpid();
size_t l;
int n;
int try=0;
if (!dir) dir = P_tmpdir;
if (!pfx) pfx = "temp";
if (access(dir, R_OK|W_OK|X_OK) != 0)
return NULL;
l = strlen(dir) + 1 + strlen(pfx) + 3*(sizeof(int)*3+2) + 1;
s = malloc(l);
if (!s) return s;
do {
clock_gettime(CLOCK_REALTIME, &ts);
n = ts.tv_nsec ^ (uintptr_t)&s ^ (uintptr_t)s;
snprintf(s, l, "%s/%s-%d-%d-%x", dir, pfx, pid, a_fetch_add(&index, 1), n);
} while (!access(s, F_OK) && try++<MAXTRIES);
if (try>=MAXTRIES) {
free(s);
return 0;
}
return s;
}
int pthread_cond_broadcast(pthread_cond_t *c)
{
pthread_mutex_t *m;
if (!c->_c_waiters) return 0;
a_inc(&c->_c_seq);
#ifdef __EMSCRIPTEN__
// XXX Emscripten: TODO: This is suboptimal but works naively correctly for now. The Emscripten-specific code path below
// has a bug and does not work for some reason. Figure it out and remove this code block.
__wake(&c->_c_seq, -1, 0);
return 0;
#endif
/* If cond var is process-shared, simply wake all waiters. */
if (c->_c_mutex == (void *)-1) {
__wake(&c->_c_seq, -1, 0);
return 0;
}
/* Block waiters from returning so we can use the mutex. */
while (a_swap(&c->_c_lock, 1))
__wait(&c->_c_lock, &c->_c_lockwait, 1, 1);
if (!c->_c_waiters)
goto out;
m = c->_c_mutex;
/* Move waiter count to the mutex */
a_fetch_add(&m->_m_waiters, c->_c_waiters2);
c->_c_waiters2 = 0;
#ifdef __EMSCRIPTEN__
int futexResult;
do {
// XXX Emscripten: Bug, this does not work correctly.
futexResult = emscripten_futex_wake_or_requeue(&c->_c_seq, !m->_m_type || (m->_m_lock&INT_MAX)!=pthread_self()->tid,
c->_c_seq, &m->_m_lock);
} while(futexResult == -EAGAIN);
#else
/* Perform the futex requeue, waking one waiter unless we know
* that the calling thread holds the mutex. */
__syscall(SYS_futex, &c->_c_seq, FUTEX_REQUEUE,
!m->_m_type || (m->_m_lock&INT_MAX)!=pthread_self()->tid,
INT_MAX, &m->_m_lock);
#endif
out:
a_store(&c->_c_lock, 0);
if (c->_c_lockwait) __wake(&c->_c_lock, 1, 0);
return 0;
}
static int pshared_barrier_wait(pthread_barrier_t *b)
{
int limit = (b->_b_limit & INT_MAX) + 1;
int ret = 0;
int v, w;
if (limit==1) return PTHREAD_BARRIER_SERIAL_THREAD;
while ((v=a_cas(&b->_b_lock, 0, limit)))
__wait(&b->_b_lock, &b->_b_waiters, v, 0);
/* Wait for <limit> threads to get to the barrier */
if (++b->_b_count == limit) {
a_store(&b->_b_count, 0);
ret = PTHREAD_BARRIER_SERIAL_THREAD;
if (b->_b_waiters2) __wake(&b->_b_count, -1, 0);
} else {
a_store(&b->_b_lock, 0);
if (b->_b_waiters) __wake(&b->_b_lock, 1, 0);
while ((v=b->_b_count)>0)
__wait(&b->_b_count, &b->_b_waiters2, v, 0);
}
__vm_lock_impl(+1);
/* Ensure all threads have a vm lock before proceeding */
if (a_fetch_add(&b->_b_count, -1)==1-limit) {
a_store(&b->_b_count, 0);
if (b->_b_waiters2) __wake(&b->_b_count, -1, 0);
} else {
while ((v=b->_b_count))
__wait(&b->_b_count, &b->_b_waiters2, v, 0);
}
/* Perform a recursive unlock suitable for self-sync'd destruction */
do {
v = b->_b_lock;
w = b->_b_waiters;
} while (a_cas(&b->_b_lock, v, v==INT_MIN+1 ? 0 : v-1) != v);
/* Wake a thread waiting to reuse or destroy the barrier */
if (v==INT_MIN+1 || (v==1 && w))
__wake(&b->_b_lock, 1, 0);
__vm_unlock_impl();
return ret;
}
int pthread_cond_broadcast(pthread_cond_t *c)
{
pthread_mutex_t *m;
if (!c->_c_waiters) return 0;
a_inc(&c->_c_seq);
/* If cond var is process-shared, simply wake all waiters. */
if (c->_c_mutex == (void *)-1) {
__wake(&c->_c_seq, -1, 0);
return 0;
}
/* Block waiters from returning so we can use the mutex. */
while (a_swap(&c->_c_lock, 1))
__wait(&c->_c_lock, &c->_c_lockwait, 1, 1);
if (!c->_c_waiters)
goto out;
m = c->_c_mutex;
/* Move waiter count to the mutex */
a_fetch_add(&m->_m_waiters, c->_c_waiters2);
c->_c_waiters2 = 0;
/* Perform the futex requeue, waking one waiter unless we know
* that the calling thread holds the mutex. */
__syscall(SYS_futex, &c->_c_seq, FUTEX_REQUEUE,
!m->_m_type || (m->_m_lock&INT_MAX)!=pthread_self()->tid,
INT_MAX, &m->_m_lock);
out:
a_store(&c->_c_lock, 0);
if (c->_c_lockwait) __wake(&c->_c_lock, 1, 0);
return 0;
}
void __vm_unlock(void)
{
int inc = vmlock[0]>0 ? -1 : 1;
if (a_fetch_add(vmlock, inc)==-inc && vmlock[1])
__wake(vmlock, -1, 1);
}
void __vm_unlock() {
if (a_fetch_add(vmlock, -1) == 1 && vmlock[1])
__wake(vmlock, -1, 1);
}
_Noreturn void pthread_exit(void *result)
{
pthread_t self = pthread_self();
sigset_t set;
self->result = result;
while (self->cancelbuf) {
void (*f)(void *) = self->cancelbuf->__f;
void *x = self->cancelbuf->__x;
self->cancelbuf = self->cancelbuf->__next;
f(x);
}
__pthread_tsd_run_dtors();
__lock(self->exitlock);
/* Mark this thread dead before decrementing count */
__lock(self->killlock);
self->dead = 1;
/* Block all signals before decrementing the live thread count.
* This is important to ensure that dynamically allocated TLS
* is not under-allocated/over-committed, and possibly for other
* reasons as well. */
__block_all_sigs(&set);
/* Wait to unlock the kill lock, which governs functions like
* pthread_kill which target a thread id, until signals have
* been blocked. This precludes observation of the thread id
* as a live thread (with application code running in it) after
* the thread was reported dead by ESRCH being returned. */
__unlock(self->killlock);
/* It's impossible to determine whether this is "the last thread"
* until performing the atomic decrement, since multiple threads
* could exit at the same time. For the last thread, revert the
* decrement and unblock signals to give the atexit handlers and
* stdio cleanup code a consistent state. */
if (a_fetch_add(&libc.threads_minus_1, -1)==0) {
libc.threads_minus_1 = 0;
__restore_sigs(&set);
exit(0);
}
if (self->detached && self->map_base) {
/* Detached threads must avoid the kernel clear_child_tid
* feature, since the virtual address will have been
* unmapped and possibly already reused by a new mapping
* at the time the kernel would perform the write. In
* the case of threads that started out detached, the
* initial clone flags are correct, but if the thread was
* detached later (== 2), we need to clear it here. */
if (self->detached == 2) __syscall(SYS_set_tid_address, 0);
/* The following call unmaps the thread's stack mapping
* and then exits without touching the stack. */
__unmapself(self->map_base, self->map_size);
}
for (;;) __syscall(SYS_exit, 0);
}
