x86 WBINVD prefix allocations

[Christian Ludloff]

> Prefixes not acting as such, but as insns is bogus.

For completeness: the opposite direction exists though.

D5 – AAD vs REX2
D6 – SALC vs L1OM

> While with segment overrides that's not as obvious, try using
> data16 or one of the REP prefixes with e.g. MOVAPS.

While EVEX with 66/F2/F3/F0 or LL=11b results in #UD
on legacy x86 processors, those encodings may have a
different meaning on evolved x86 processors. (At which
point code like [1] simply evolves too.)

> It's still not really "normal", though.

ABIs tend to evolve.

x86 is no exception.

--
C.

[1] https://inbox.sourceware.org/binutils/e1c1d593-4588-08e0-d3f7-0cd42b0dc6e4@suse.com/T/#u