Re: [GIT PULL] x86/urgent for v5.11-rc7

[Peter Zijlstra <peterz@infradead.org>]

On Sun, Feb 07, 2021 at 10:31:32AM -0800, Andy Lutomirski wrote:
> 
> > On Feb 7, 2021, at 10:19 AM, Dave Hansen <dave.hansen@intel.com> wrote:
> > 
> > On 2/7/21 9:58 AM, Borislav Petkov wrote:
> >>> On Sun, Feb 07, 2021 at 09:49:18AM -0800, Linus Torvalds wrote:
> >>> On Sun, Feb 7, 2021 at 2:40 AM Borislav Petkov <bp@suse.de> wrote:
> >>>> - Disable CET instrumentation in the kernel so that gcc doesn't add
> >>>> ENDBR64 to kernel code and thus confuse tracing.
> >>> So this is clearly the right thing to do for now, but I wonder if
> >>> people have a plan for actually enabling CET and endbr at cpl0 at some
> >>> point?
> >> It probably is an item on some Intel manager's to-enable list. So far,
> >> the CET enablement concentrates only on userspace but dhansen might know
> >> more about future plans. CCed.
> > 
> > It's definitely on our radar to look at after CET userspace.
> > 
> > The only question for me is whether it will be worth doing with the
> > exiting kernel entry/exit architecture.
> 
> I assume you mean: is anyone sufficiently inspired to try to handle
> NMI correctly?  I have a whole pile of nacks saved up for incorrect
> implementations, although I will try to wrap them in polite
> explanations of precisely what is wrong :)

Yeah, the IST stack recursion possibilities are 'fun'. But IIRC CET-SS
has far more problems than just NMI. It will also run into all the ROP
tricks we pull for return tracing, CALL emulation and other lovely
things.