RFR: 8336256: memcpy short value to int local is incorrect in VtableStubs::unsafe_hash

[Thomas Stuefe]

On Fri, 12 Jul 2024 08:16:18 GMT, Richard Reingruber <rrich at openjdk.org> wrote:

> > Was that a big endian problem?
> 
> We got the crashes given in the JBS issue (see above) when building on PPC64 big endian (AIX and Linux). I do think this is also incorrect with little endian order since excessive bytes after `VtableStub::_index` are copied into the local variable `vtable_index`. On LE the excessive bytes are in the high word of `vtable_index`. Maybe they are masked out by the hash function. On BE the excessive bytes end up in the low word and the short value from `_index` ends up in the high word of `vtable_index` and the hash function would mask away the bytes from `_index`.
> 
> @offamitkumar have you observed the issues too on S390?

@reinrich thanks for clarifying :)

in any case, patch makes sense

-------------

PR Comment: https://git.openjdk.org/jdk/pull/20137#issuecomment-2225080751