Re: [PATCH 4/4] x86,module: Detect CRn and DRn manipulation

[Kees Cook]

On Tue, Apr 07, 2020 at 08:13:25PM +0200, Peter Zijlstra wrote:
> On Tue, Apr 07, 2020 at 10:01:04AM -0700, Kees Cook wrote:
> > On Tue, Apr 07, 2020 at 01:02:40PM +0200, Peter Zijlstra wrote:
> > > Since we now have infrastructure to analyze module text, disallow
> > > modules that write to CRn and DRn registers.
> > > 
> > > Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
> > > ---
> > >  arch/x86/kernel/module.c |   21 +++++++++++++++++++++
> > >  1 file changed, 21 insertions(+)
> > > 
> > > --- a/arch/x86/kernel/module.c
> > > +++ b/arch/x86/kernel/module.c
> > > @@ -266,6 +266,22 @@ static bool insn_is_vmx(struct insn *ins
> > >  	return false;
> > >  }
> > >  
> > > +static bool insn_is_mov_CRn(struct insn *insn)
> > > +{
> > > +	if (insn->opcode.bytes[0] == 0x0f && insn->opcode.bytes[1] == 0x22)
> > > +		return true;
> > 
> > I always cringe at numeric literals. Would it be overkill to add defines
> > for these (and the others that have comments next to them in 3/4)? It
> > makes stuff easier to grep, etc. (e.g. we have register names in
> > arch/x86/include/asm/asm.h, do we need instruction names somewhere else?
> > I assume objtool has a bunch of this too...)
> 
> objtool does not, have a peek at tools/objtool/arch/x86/decode.c

Eek.

> I'm not sure what the best way is here, the x86 opcode map is a
> disaster. Even the mnemonic doesn't help us here, as that's just MOV :/

Yeah, I'm not sure either. I guess leave this as-is.

-- 
Kees Cook