Re: [PATCH v8 13/20] KVM: x86: Reject EVEX-prefixed instructions

[Jim Mattson]

On Thu, Nov 5, 2026 at 8:56 PM Jim Mattson <jmattson@xxxxxxxxxx> wrote:
>
> On Wed, Sep 30, 2026 at 4:54 PM Chang S. Bae <chang.seok.bae@xxxxxxxxx> wrote:
> >
> > Explicitly mark EVEX-prefixed opcodes (0x62) as unsupported.
> >
> > Signed-off-by: Chang S. Bae <chang.seok.bae@xxxxxxxxx>
> > ---
> >  arch/x86/kvm/emulate.c | 5 +++++
> >  1 file changed, 5 insertions(+)
> >
> > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> > index eba4e9b7cd5c..f0d5a40eba27 100644
> > --- a/arch/x86/kvm/emulate.c
> > +++ b/arch/x86/kvm/emulate.c
> > @@ -5052,6 +5052,11 @@ int x86_decode_insn(struct x86_emulate_ctxt *ctxt, void *insn, int insn_len, int
> >         if (opcode.flags & ModRM)
> >                 ctxt->modrm = insn_fetch(u8, ctxt);
> >
> > +       /* EVEX-prefixed instructions are not implemented */
> > +       if (ctxt->opcode_len == 1 && ctxt->b == 0x62 &&
> > +           (mode == X86EMUL_MODE_PROT64 || (ctxt->modrm & 0xc0) == 0xc0))
> > +               opcode.flags = NotImpl;
>
> What is this patch actually doing? opcode_table[0x62] is already N
> (NotImpl). Moreover, because opcode_table[0x62] does not have the ModRM
> flag and init_decode_cache() does not clear ctxt->modrm, "(ctxt->modrm &
> 0xc0) == 0xc0" in non-64-bit mode inspects a stale ctxt->modrm from
> whatever instruction was emulated previously.
>
> If the intent is to mark opcode_table[0x62] with No64 (or NoRex2) so
> that REX2 (with M0=0) followed by 0x62 raises #UD, then:
> 1. opcode_table[0x62] actually needs that flag, and
> 2. this EVEX check must be gated on ctxt->rex_prefix != REX2_PREFIX,
>    or else it will clobber opcode.flags with NotImpl on "D5 00 62" and
>    return EMULATION_FAILED instead of injecting #UD.

Again, on EMULATION_FAILED,  x86_emulate_instruction() will raise #UD,
so this second paragraph is nonsense. However, the key point from the
first paragraph--that opcode_table[0x62] is already N
(NotImpl)--stands. As a result, this patch is a NOP.