Fix some race condition on arm64

* move barrier() to architecture depended region
* add barrier() in issue_ipi, kprintf, map_virtual
* enable the workaround for cavium thunderx

kernel/mem.c

@@ -1617,6 +1617,7 @@ void *ihk_mc_map_virtual(unsigned long phys, int npages,
 			return NULL;
 		}
 	}
+	barrier();
 	return (char *)p + offset;
 }