Atomic/SMP update, part 3.

Update ARM atomic ops to use LDREX/STREX.  Stripped out #if 0 chunk.

Insert explicit memory barriers in pthread and semaphore code.

For bug 2721865.

Change-Id: I0f153b797753a655702d8be41679273d1d5d6ae7
diff --git a/libc/bionic/pthread.c b/libc/bionic/pthread.c
index ae44b06..709e612 100644
--- a/libc/bionic/pthread.c
+++ b/libc/bionic/pthread.c
@@ -44,6 +44,7 @@
 #include <assert.h>
 #include <malloc.h>
 #include <linux/futex.h>
+#include <cutils/atomic-inline.h>
 
 extern int  __pthread_clone(int (*fn)(void*), void *child_stack, int flags, void *arg);
 extern void _exit_with_stack_teardown(void * stackBase, int stackSize, int retCode);
@@ -936,6 +937,7 @@
         while (__atomic_swap(shared|2, &mutex->value ) != (shared|0))
             __futex_syscall4(&mutex->value, wait_op, shared|2, 0);
     }
+    ANDROID_MEMBAR_FULL();
 }
 
 /*
@@ -945,6 +947,8 @@
 static __inline__ void
 _normal_unlock(pthread_mutex_t*  mutex)
 {
+    ANDROID_MEMBAR_FULL();
+
     /* We need to preserve the shared flag during operations */
     int  shared = mutex->value & MUTEX_SHARED_MASK;
 
@@ -1144,8 +1148,10 @@
     /* Handle common case first */
     if ( __likely(mtype == MUTEX_TYPE_NORMAL) )
     {
-        if (__atomic_cmpxchg(shared|0, shared|1, &mutex->value) == 0)
+        if (__atomic_cmpxchg(shared|0, shared|1, &mutex->value) == 0) {
+            ANDROID_MEMBAR_FULL();
             return 0;
+        }
 
         return EBUSY;
     }
@@ -1241,9 +1247,11 @@
     {
         int  wait_op = shared ? FUTEX_WAIT : FUTEX_WAIT_PRIVATE;
 
-        /* fast path for unconteded lock */
-        if (__atomic_cmpxchg(shared|0, shared|1, &mutex->value) == 0)
+        /* fast path for uncontended lock */
+        if (__atomic_cmpxchg(shared|0, shared|1, &mutex->value) == 0) {
+            ANDROID_MEMBAR_FULL();
             return 0;
+        }
 
         /* loop while needed */
         while (__atomic_swap(shared|2, &mutex->value) != (shared|0)) {
@@ -1252,6 +1260,7 @@
 
             __futex_syscall4(&mutex->value, wait_op, shared|2, &ts);
         }
+        ANDROID_MEMBAR_FULL();
         return 0;
     }
 
diff --git a/libc/bionic/semaphore.c b/libc/bionic/semaphore.c
index 84b9314..b624943 100644
--- a/libc/bionic/semaphore.c
+++ b/libc/bionic/semaphore.c
@@ -30,6 +30,7 @@
 #include <sys/time.h>
 #include <sys/atomics.h>
 #include <time.h>
+#include <cutils/atomic-inline.h>
 
 int sem_init(sem_t *sem, int pshared, unsigned int value)
 {
@@ -103,6 +104,7 @@
     return old;
 }
 
+/* lock a semaphore */
 int sem_wait(sem_t *sem)
 {
     if (sem == NULL) {
@@ -116,6 +118,7 @@
 
         __futex_wait(&sem->count, 0, 0);
     }
+    ANDROID_MEMBAR_FULL();
     return 0;
 }
 
@@ -130,8 +133,10 @@
 
     /* POSIX says we need to try to decrement the semaphore
      * before checking the timeout value */
-    if (__atomic_dec_if_positive(&sem->count))
+    if (__atomic_dec_if_positive(&sem->count)) {
+        ANDROID_MEMBAR_FULL();
         return 0;
+    }
 
     /* check it as per Posix */
     if (abs_timeout == NULL    ||
@@ -169,17 +174,21 @@
             return -1;
         }
 
-        if (__atomic_dec_if_positive(&sem->count))
+        if (__atomic_dec_if_positive(&sem->count)) {
+            ANDROID_MEMBAR_FULL();
             break;
+        }
     }
     return 0;
 }
 
+/* unlock a semaphore */
 int sem_post(sem_t *sem)
 {
     if (sem == NULL)
         return EINVAL;
 
+    ANDROID_MEMBAR_FULL();
     if (__atomic_inc((volatile int*)&sem->count) >= 0)
         __futex_wake(&sem->count, 1);
 
@@ -194,6 +203,7 @@
     }
 
     if (__atomic_dec_if_positive(&sem->count) > 0) {
+        ANDROID_MEMBAR_FULL();
         return 0;
     } else {
         errno = EAGAIN;