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

From: Sean Christopherson
Date: Wed Apr 08 2020 - 12:03:02 EST


On Wed, Apr 08, 2020 at 08:46:02AM -0700, Christoph Hellwig wrote:
> On Wed, Apr 08, 2020 at 05:44:19PM +0200, Peter Zijlstra wrote:
> > On Wed, Apr 08, 2020 at 09:27:26AM -0400, Steven Rostedt wrote:
> > > On Tue, 07 Apr 2020 13:02:40 +0200
> > > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >
> > > > + if (insn_is_mov_CRn(&insn) || insn_is_mov_DRn(&insn)) {
> > > > + pr_err("Module writes to CRn or DRn, please use the proper accessors: %s\n", mod->name);
> > > > + return -ENOEXEC;
> > > > + }
> > > > +
> > >
> > > Something like this should be done for all modules, not just out of tree
> > > modules.
> >
> > I'm all for it; but people were worried scanning all modules was too
> > expensive (I don't really believe it is, module loading just can't be a
> > hot-path). Also, in-tree modules are audited a lot more than out of tree
> > magic voodoo crap.
>
> Scanning all modules seems safer. While we're at it - can be move the
> kvm bits using VMX to be always in the core kernel and just forbid
> modules from using those instructions entirely?

Practically speaking, no. Turning VMX on and off (literally VMXON/VMXOFF)
could be moved to helpers in the kernel, but KVM relies on inlining all
post-VMXON instructions (except for VMLAUNCH/VMRESUME) for performance.
VMLAUNCH/VMRESUME have their own caveats, moving them out of KVM would be
messy, to say the least.