RE: SNP: qemu upstream vs AMD fork on kernel 6.11

[Roth, Michael]

[AMD Official Use Only - AMD Internal Distribution Only]

> -----Original Message-----
> From: Paolo Bonzini <pbonzini@redhat.com>
> Sent: Monday, August 5, 2024 8:36 AM
> To: Arvid Picciani <arvid@kraud.cloud>
> Cc: Roth, Michael <Michael.Roth@amd.com>; qemu-discuss@nongnu.org
> Subject: Re: SNP: qemu upstream vs AMD fork on kernel 6.11
>
> On Mon, Aug 5, 2024 at 3:26 PM Arvid Picciani <arvid@kraud.cloud> wrote:
> >
> > Hi,
> >
> > with linux 6.11 it looks like kvm SNP host api is finally there.
> >
> > However, current qemu upstream appears to be tested against a much
> > older kernel by redhat (according to the libvirt IRC its via coconut
> > svsm),
>
> It's tested against both 6.11 (actually a slightly older branch in
> kvm.git) and the CentOS Stream 9 kernel. The CentOS Stream 9 kernel
> has the same SNP code as Linux 6.11.
>
> > while the current head of the amd fork on github works just
> > fine with 6.11.
>
> The current head of the AMD fork (https://github.com/AMDESE/qemu
> snp-latest) is not in line with the upstream kernel, for example:
>
> 66e7fbadfc8 i386/sev: Add KVM_EXIT_VMGEXIT handling for Page State Changes
> e9898385037 i386/sev: Add KVM_EXIT_VMGEXIT handling for Page State
> Changes (MSR-based)

It actually adds the KVM_HC_MAP_GPA_RANGE handling on top, and then drops all 
KVM_EXIT_VMGEXIT handling completely as part of:

  commit 62434c7dd57fcf2e02f6765b9d0d2588b5e032d3
  Author: Michael Roth <michael.roth@amd.com>
  Date:   Wed May 1 01:14:39 2024 -0500

      *i386/sev: Rework GHCB extended guest request handling

      TODO: drop the PSC patches completely so this rework can become a clean
      standalone patch.

The v4 patches that actually went into QEMU 9.1 have this changeset 
appropriately split/squashed into their proper place, but I think the upstream 
behavior should be about the same as what is in snp-latest QEMU tree currently.

If snp-latest works, but upstream QEMU does not, then there is potentially a 
regression upstream. However I just re-tested latest QEMU commit (f9851d2ffef5) 
against upstream KVM and didn't see any issue.

QEMU 9.1 is already is hard-freeze, so if there is a confirmed breakage please 
share the full details and steps to reproduce so it can hopefully be sorted 
before the 9.1 release.

-Mike

>
> What problems are you seeing with Linux 6.11?
>
> Paolo
>
> > there's also this old patch missing to enable virtiofsd
> > https://lists.gnu.org/archive/html/qemu-devel/2022-01/msg03456.html
> >
> > i would rebase that and send it again, but current main is broken on
> > 6.11, so wondering where qemu is heading.
> >
> > thanks
> >
> >
> >
> > --
> > https://kraudcloud.com/
> > devguard GmbH, Berlin, Geschäftsführer: Arvid E Picciani
> > Handelsregister: Amtsgericht Charlottenburg (Berlin) HRB 195184 B
> >