Re: tip/master falls off NOP cliff with KPTI under KVM

From: David Woodhouse
Date: Thu Jan 25 2018 - 02:53:57 EST


On Wed, 2018-01-24 at 16:53 -0800, Dexuan-Linux Cui wrote:
> On Wed, Jan 10, 2018 at 2:53 PM, Woodhouse, David <dwmw@xxxxxxxxxxxx> wrote:
> >
> > On Thu, 2018-01-11 at 01:34 +0300, Alexey Dobriyan wrote:
> > >
> > >
> > > Bisection points to
> > >
> > > ÂÂÂÂÂÂÂÂf3433c1010c6af61c9897f0f0447f81b991feac1 is the first bad commit
> > > ÂÂÂÂÂÂÂÂcommit f3433c1010c6af61c9897f0f0447f81b991feac1
> > > ÂÂÂÂÂÂÂÂAuthor: David Woodhouse <dwmw@xxxxxxxxxxxx>
> > > ÂÂÂÂÂÂÂÂDate:ÂÂÂTue Jan 9 14:43:11 2018 +0000
> > >
> > > ÂÂÂÂÂÂÂÂÂÂÂÂx86/retpoline/entry: Convert entry assembler indirect jumps
> > Thanks. We've fixed the underlying problem with the alternatives
> > mechanism, *and* changed the retpoline code not to actually rely on
> > said fix.
> Hi David and all,
> It looks the latest upstream tree did fix the issue.
> Can you please specify the related commit(s) that fixed the issue?
> I need to cherry-pick the fix. I suppose a quick reply from you would save
> Âme a lot of time. :-)

Hi Dexuan,

The above commit didn't ever make it into Linus' tree in that form; the
issues were fixed beforehand. So there isn't a subsequent commit that
fixes it. I think it might have just been removing some .align 16 from
nospec-branch.h?

The correct version has been backported to 4.9 and 4.4 releases
already; if you pulled in an early version directly from tip/x86/pti
then I'd recommend you drop it and pull in the real version instead.

Attachment: smime.p7s
Description: S/MIME cryptographic signature