Re: [PATCH] x86/sev-es: Expose __sev_es_ghcb_hv_call() to call ghcb hv call out of sev code

[Tianyu Lan]

On 10/20/2021 10:39 PM, Borislav Petkov wrote:
> On Wed, Oct 20, 2021 at 10:23:06PM +0800, Tianyu Lan wrote:
> > This follows Joreg's previous comment and I implemented similar version in
> > the V! patchset([PATCH 05/13] HV: Add Write/Read MSR registers via ghcb page
> > https://lkml.org/lkml/2021/7/28/668).
> > "Instead, factor out a helper function which contains what Hyper-V needs and
> > use that in sev_es_ghcb_hv_call() and Hyper-V code."
> >
> > https://lkml.org/lkml/2021/8/2/375
>
> If you wanna point to mails on a mailing list, you simply do
>
> https://lore.kernel.org/r/<Message-id>
>
> No need to use some random, unreliable web pages.

OK. Thanks for suggestion.

> As to Joerg's suggestion, in the version I'm seeing, you're checking the
> *context* - and the one you sent today, avoids the __pa(ghcb) MSR write.
>
> So which is it?
>
> Because your current version will look at the context too, see
>
> 	return verify_exception_info(ghcb, ctxt);
>
> at the end of the function.
Both old and new patches are to avoid setting GHCB page address via MSR.
Paravisor is in charge of doing that and un-enlightened guest should not 
change it. The old one was in the patchset v1 "x86/Hyper-V: Add Hyper-V
Isolation VM support". The patch I sent today is based on your clean up 
patch and for review first. It should be in patchset "x86/Hyper-V: Add 
Hyper-V Isolation VM support."

> So is the issue what Tom said that "the paravisor uses the same GHCB MSR
> and GHCB protocol, it just can't use __pa() to get the address of the
> GHCB."?

Yes, hyper-V enables vTOM in the CVM and GHCB page PA reported by 
paravisor contains vTOM bit. We need to use memremap() to map ghcb page 
before accessing GHCB page. __pa() doesn't work for PA with vTOM bit.
Otherwise, guest should not set GHCB page address and avoid conflict 
with paravisor.

> If that is the case and the only thing you want is to avoid the GHCB PA
> write, then, in the future, we might drop that MSR write altogether on
> the enlightened Linux guests too and then the same function will be used
> by your paravisor and the Linux guest.

Yes, this is the target of the patch. Can we put the change in the 
Hyper-V patchset? Other patch has been fully reviewed.

Thanks.