Re: [PATCH] x86: Disallow APX instruction with length > 15 bytes

["H.J. Lu" <hjl.tools@gmail.com>]

On Thu, Feb 1, 2024 at 11:23 PM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 01.02.2024 23:47, H.J. Lu wrote:
> > --- a/gas/config/tc-i386.c
> > +++ b/gas/config/tc-i386.c
> > @@ -11772,8 +11772,14 @@ output_insn (const struct last_insn *last_insn)
> >       {
> >         j = encoding_length (insn_start_frag, insn_start_off, frag_more (0));
> >         if (j > 15)
> > -         as_warn (_("instruction length of %u bytes exceeds the limit of 15"),
> > -                  j);
> > +         {
> > +           if (i.tm.cpu.bitfield.cpuapx_f)
> > +             as_bad (_("instruction length of %u bytes exceeds the limit of 15"),
> > +                     j);
> > +           else
> > +             as_warn (_("instruction length of %u bytes exceeds the limit of 15"),
> > +                      j);
> > +         }
>
> Why would APX insns be different from others? IOW I continue to think that

No.  It is just very easy to generate invalid instructions with APX.

> having a warning here is good enough, uniformly. And it's quite sad that

We ran into this with real codes and triggered run-time errors.

> with introducing APX the limit isn't raised, to accommodate all valid insn
> forms (not considering ones with redundant prefixes, of course). _That_
> would then permit special casing APX here, in _not_ warning anymore.
>

Here is the v2 patch:

https://sourceware.org/pipermail/binutils/2024-February/132285.html

to change warning to error for all instructions.

-- 
H.J.