Re: [edk2-devel] [PATCH v2 2/2] ArmVirtPkg/ArmVirtQemu: Avoid early ID map on ThunderX

["Ard Biesheuvel" <ardb@kernel.org>]

On Thu, 19 Jan 2023 at 12:56, Marc Zyngier <maz@kernel.org> wrote:
>
> On Thu, 19 Jan 2023 11:11:34 +0000,
> Ard Biesheuvel <ardb@kernel.org> wrote:
> >
> >  (cc Marc)
> >
> > Context:
> > - on my TX2 (with the S1PTW r/o memslot fix applied), the new version
> > of ArmVirtQemu that uses an initial ID map in emulated NOR flash works
> > fine.
> > - in Oliver's case (which is a slightly different flavor of TX2), it
> > crashes extremely early, presumably at the point where this ID map is
> > activated.
>
> Oliver seems to have a vintage ThunderX (aka the worst arm64
> implementation in history!), so it is indeed a very different beat
> from TX2.
>

No, it's a TX2 alright - we disabled this ID map for TX1 already
because it tickles the I-cache corruption erratum (EDK2 runs at EL1
and the initial ID map is ASID scoped with non-global mappings, so
that we can switch to the actual ID map seamlessly)

> Without the kernel patch[1], I can trigger the issue pretty reliably,
> specially in the absence of THP. It all depends on the layout of the
> EDK2 object and the order in which pages get mapped.
>
> The first course of action would be to make sure that the patch is
> applied to the host kernel. If this still fails to boot, I'm happy to
> help investigating it.
>

indeed.