Fix stack alignment of child threads on ARM.

src/nonsfi/linux/linux_pthread_private.c

Index: src/nonsfi/linux/linux_pthread_private.c
diff --git a/src/nonsfi/linux/linux_pthread_private.c b/src/nonsfi/linux/linux_pthread_private.c
index cb9316871374b53bc772317a064405c27fa4fb7d..def51d4f2aa015c46be0ddbc36747e93b8f4fdc1 100644
--- a/src/nonsfi/linux/linux_pthread_private.c
+++ b/src/nonsfi/linux/linux_pthread_private.c
@@ -95,10 +95,10 @@ static int nacl_irt_thread_create(void (*start_func)(void), void *stack,
                        /*
                         * In child thread. Clear the frame pointer to
                         * prevent debuggers from unwinding beyond this,
-                        * pop the stack to get start_func and call it.
+                        * load start_func from the stack and call it.
                         */
                        "mov fp, #0\n"
-                       "pop {r0}\n"
+                       "ldr r0, [sp]\n"
                        "blx r0\n"
                        /* start_func never finishes. */
                        "bkpt #0\n"