Simple question: we define __HAVE_RAS in <sys/types.h> if a $ARCH supports
RAS. Actually, besides a few regression tests where I recently added it,
nothing in userland tests __HAVE_RAS - AFAICT (and grep seems to agree).

So I suggest to not export __HAVE_RAS to userland at all, and have
everyone try rasctl() - which returns EOPNOTSUP if RAS is not available.
Fixing the regression tests is easy.

This would allow 32bit sparc64 kernels, which use sparc userland, to have
full RAS support.

Comments?

Martin