Re: [PATCH v5 1/1] x86/cpufeatures: Implement Predictive Store Forwarding control.

[Kees Cook <keescook@chromium.org>]

On Mon, May 10, 2021 at 11:44:03PM +0200, Thomas Gleixner wrote:
> On Mon, May 10 2021 at 06:10, RK Saripalli wrote:
> > On 5/7/2021 10:13 AM, Thomas Gleixner wrote:
> >> What's wrong with just treating this in the same way in which we treat
> >> all other speculative vulnerabilities and provide a consistent picture
> >> to the user?
> >> 
> >> Something like the below. You get the idea.
> >
> > Thomas, thank you very much for the comments.
> >
> > I provided the links to the original patches which treat PSF similar to other
> > speculative vulnerabilities.
> >
> > Could you review them please?. The first patch is the cover letter.
> >
> > https://lore.kernel.org/lkml/20210406155004.230790-1-rsaripal@amd.com/
> > https://lore.kernel.org/lkml/20210406155004.230790-2-rsaripal@amd.com/
> > https://lore.kernel.org/lkml/20210406155004.230790-3-rsaripal@amd.com/
> > https://lore.kernel.org/lkml/20210406155004.230790-4-rsaripal@amd.com/
> > https://lore.kernel.org/lkml/20210406155004.230790-5-rsaripal@amd.com/
> > https://lore.kernel.org/lkml/20210406155004.230790-6-rsaripal@amd.com/
> 
> They are going into the right direction, i.e. detection and reporting.
> 
> Vs. mitigation control the question is whether we need the full
> machinery of prctl/seccomp and so forth especially under the aspect that
> the SSBD mitigation already covers the PSF issue.
> 
> So for the start a simple on/off might be good enough.
> 
> Kees, any opinions?

I agree: if PSF is a subset of SSBD, there's no need for the additional
machinery.

On a related topic, what happened to Andi's patch to switch the seccomp
defaults? I can't find it now...

-- 
Kees Cook