ARM: a very small change.

[Andrew Haley]

atomic_copy64 is used by Object.clone().  It's used instead of
memcpy() because otherwise a copy might be interrupted partway
through copying a jlong.  We were using the generic code on ARM,
but we really should make sure we use an atomic load.  This only
really matters on multi-core ARMs; on others the 64-bit read is
restarted after an interrupt.

Andrew.


2012-05-28  Andrew Haley  <aph at redhat.com>

	* os_linux_zero.hpp (atomic_copy64): Use ldrexd for atomic reads
	on ARMv7.

diff -r 73a07d24174e src/os_cpu/linux_zero/vm/os_linux_zero.hpp
--- a/src/os_cpu/linux_zero/vm/os_linux_zero.hpp        Thu May 17 13:45:50 2012 -0400
+++ b/src/os_cpu/linux_zero/vm/os_linux_zero.hpp        Mon May 28 07:54:46 2012 -0400
@@ -48,6 +48,12 @@
                   "std %0, 0(%2)\n"
                   : "=r"(tmp)
                   : "a"(src), "a"(dst));
+#elif defined(__ARM_ARCH_7A__)
+    jlong tmp;
+    asm volatile ("ldrexd  %0, [%1]\n"
+                  : "=r"(tmp)
+                  : "r"(src), "m"(src));
+    *(jlong *) dst = tmp;
 #else
     *(jlong *) dst = *(jlong *) src;
 #endif