[svsm-devel] GUIDs in the GHCB and SVSM specifications

[Dionna Amalie Glaze]

On Tue, Mar 25, 2025 at 11:50 AM Dionna Amalie Glaze
<dionnaglaze at google.com> wrote:
>
> On Wed, Feb 26, 2025 at 2:40 PM Tom Lendacky <thomas.lendacky at amd.com> wrote:
> >
> > All,
> >
> > There has been some talk about how the GUIDs in the GHCB and SVSM
> > specifications should be encoded [1]. The term GUID brings encoding
> > ambiguity with it.
> >
> > - RFC 4122 specifies the format of a GUID/UUID and the encodingas
> > network byte order.
> >
> > - The UEFI specification says that their GUID format is that of RFC
> > 4122, but that the encoding is different. UEFI has time_low, time_mid,
> > and time_hi_and_version as little-endian encoded.
> >
> > Given this, the GHCB and SVSM specifications need to be updated to
> > specify the expected encoding of the GUIDs.
> >
> > The GHCB specification will add that the certificate GUIDs follow RFC
> > 4122 in format and encoding (network byte order). This is based on
> > existing users and tools that have already (likely) been using network
> > byte order.
> >
> > The SVSM specification, however, because support has already been added
> > to the Linux kernel using guid library APIs, will follow RFC 4122 in
> > format with an encoding of time_low, time_mid, and time_hi_and_version
> > as little-endian. In Linux, the guid library APIs are guid_parse(),
> > guid_gen(), etc.
> >
> > If there are no objections, I'll make those changes.
>
> What we know is that the service GUID is in little endian order, but
> what we don't know is if the GUID table of the attest all services
> protocol should be the same as is provided in GHCB (big endian).
> I haven't heard back on the other thread about a more general
> principle that maybe we should follow little-endian in and big-endian
> out at least for GUIDs, since reports still use little endian numbers.
> I'm writing up the implementation of the GUID table serializer right
> now, and it'd be nice to know it can have the same test vectors as the
> GUID table marshaling/unmarshaling code I have in go-sev-guest.
>

I suppose nevermind since the headers are slightly different, the
parsers are incompatible. I'll use little endian.

> >
> > Thanks,
> > Tom
> >
> > [1] https://github.com/coconut-svsm/svsm/pull/541#discussion_r1963201009
> >
>
>
> --
> -Dionna Glaze, PhD, CISSP, CCSP (she/her)



-- 
-Dionna Glaze, PhD, CISSP, CCSP (she/her)