diff options
author | Rich Felker <dalias@aerifal.cx> | 2011-09-28 18:57:18 -0400 |
---|---|---|
committer | Rich Felker <dalias@aerifal.cx> | 2011-09-28 18:57:18 -0400 |
commit | 9cee9307e3b8ed31e772d564d1dc53e6d888acba (patch) | |
tree | 90355be0615b94ba2cd5ddf7581173862ae7a063 /src/thread/pthread_barrier_wait.c | |
parent | 95b147967216a057d51a1b970da777486be716e3 (diff) | |
download | musl-9cee9307e3b8ed31e772d564d1dc53e6d888acba.tar.gz musl-9cee9307e3b8ed31e772d564d1dc53e6d888acba.tar.bz2 musl-9cee9307e3b8ed31e772d564d1dc53e6d888acba.tar.xz musl-9cee9307e3b8ed31e772d564d1dc53e6d888acba.zip |
improve pshared barriers
eliminate the sequence number field and instead use the counter as the
futex because of the way the lock is held, sequence numbers are
completely useless, and this frees up a field in the barrier structure
to be used as a waiter count for the count futex, which lets us avoid
some syscalls in the best case.
as of now, self-synchronized destruction and unmapping should be fully
safe. before any thread can return from the barrier, all threads in
the barrier have obtained the vm lock, and each holds a shared lock on
the barrier. the barrier memory is not inspected after the shared lock
count reaches 0, nor after the vm lock is released.
Diffstat (limited to 'src/thread/pthread_barrier_wait.c')
-rw-r--r-- | src/thread/pthread_barrier_wait.c | 24 |
1 files changed, 13 insertions, 11 deletions
diff --git a/src/thread/pthread_barrier_wait.c b/src/thread/pthread_barrier_wait.c index 9b1edbf9..71f7b5fe 100644 --- a/src/thread/pthread_barrier_wait.c +++ b/src/thread/pthread_barrier_wait.c @@ -21,7 +21,6 @@ void __vm_unlock(void) static int pshared_barrier_wait(pthread_barrier_t *b) { int limit = (b->_b_limit & INT_MAX) + 1; - int seq; int ret = 0; int v, w; @@ -30,33 +29,36 @@ static int pshared_barrier_wait(pthread_barrier_t *b) while ((v=a_cas(&b->_b_lock, 0, limit))) __wait(&b->_b_lock, &b->_b_waiters, v, 0); - seq = b->_b_seq; - + /* 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; - b->_b_seq++; - __wake(&b->_b_seq, -1, 0); + 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); - __wait(&b->_b_seq, 0, seq, 0); + while ((v=b->_b_count)>0) + __wait(&b->_b_count, &b->_b_waiters2, v, 0); } __vm_lock(+1); - if (a_fetch_add(&b->_b_count, -1)==1) { - b->_b_seq++; - __wake(&b->_b_seq, -1, 0); + /* 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 { - __wait(&b->_b_seq, 0, seq+1, 0); + 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-1 & INT_MAX) != v); + } 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); |