I've just fixed this for C1, in a fairly conservative way and C2 should do the same. Basically NativeCall::set_destination_mt_safe isn't MT-safe on Intel if the call spans a cache line. The most straightforward fix is to align the displacement word of CompiledIC and CompiledStaticCall to a word boundary. We could rely simply on having the whole displacement on a single cache line but you can't really rely on the atomicity of unaligned stores on intel either.