1 files changed, 37 insertions, 0 deletions
diff --git a/lib/librthread/arch/powerpc/_atomic_lock.c b/lib/librthread/arch/powerpc/_atomic_lock.c
new file mode 100644
index 00000000000..aa07fc3a97b
--- /dev/null
+++ b/lib/librthread/arch/powerpc/_atomic_lock.c
@@ -0,0 +1,37 @@
+/*	$OpenBSD: _atomic_lock.c,v 1.1 2005/12/04 04:08:07 brad Exp $	*/
+/*
+ * Atomic lock for powerpc
+ */
+
+#include "spinlock.h"
+
+int
+_atomic_lock(volatile _spinlock_lock_t *lock)
+{
+	_spinlock_lock_t old;
+
+	__asm__("1: lwarx 0,0,%1   \n"
+		"   stwcx. %2,0,%1 \n"
+		"   bne- 1b        \n"
+		"   mr %0, 0	   \n"
+		: "=r" (old), "=r" (lock)
+		: "r" (_SPINLOCK_LOCKED), "1" (lock) : "0"
+	);
+
+	return (old != _SPINLOCK_UNLOCKED);
+
+	/*
+	 * Dale <rahnds@openbsd.org> says:
+	 *   Side note. to prevent two processes from accessing
+	 *   the same address with the lwarx in one instrution
+	 *   and the stwcx in another process, the current powerpc
+	 *   kernel uses a stwcx instruction without the corresponding
+	 *   lwarx which causes any reservation of a process
+	 *   to be removed.  if a context switch occurs
+	 *   between the two accesses the store will not occur
+	 *   and the condition code will cause it to loop. If on
+	 *   a dual processor machine, the reserve will cause
+	 *   appropriate bus cycle accesses to notify other
+	 *   processors.
+	 */
+}