Re: [PATCH] x86_64,xen,espfix: Initialize espfix on secondary CPUs

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Tue, Jul 15, 2014 at 08:44:39AM -0700, Andy Lutomirski wrote:
> On Tue, Jul 15, 2014 at 8:38 AM, Konrad Rzeszutek Wilk
> <konrad.wilk@oracle.com> wrote:
> > On Tue, Jul 15, 2014 at 08:26:41AM -0700, Andy Lutomirski wrote:
> >> Xen doesn't call start_secondary.
> >
> > Duh!
> >>
> >> Cc: stable@vger.kernel.org
> >> Signed-off-by: Andy Lutomirski <luto@amacapital.net>
> >> ---
> >>
> >> espfix still doesn't seem to work on Xen (it goes boom in some way that
> >> I don't understand right now), but initializing all CPUs instead of just
> >> one of them seems like a good start.
> >>
> >> ISTM the right fix is probably to shove the espfix logic into
> >> native_iret and to tweak the paravirt logic so that native_iret always
> >> gets invoked.  I suspect that Xen will need its own implementation of
> >> espfix64 in the hypervisor and that, ultimately, someone may want to
> >> stop initializing espfix64 at all on Xen guests.
> >
> > I think just disallowing would be preferrable.
> 
> Disabling what?
> 
> Sorry, my flu-addled brain needs more clarity.  I'm currently working
> on a patch on top of this one to move all of the espfix64 invocation
> logic into native_iret, which will have the effect of preventing it
> from being used on Xen.
> 
> Is that what you mean?

Yes. I presume the logic to deal with the bits losing information
has to be dealt in the Xen case somehow. Peter asked whether the 
Xen IRET handles a 16-bit stack segment - and if it restores all of the
RSP then we are OK.

I don't have yet that information and my brain is a in low-power right
now (-ENOSLEEP).