Re: [PATCH v22 11/23] KVM: arm64: Add VM specific callback for S2 MMU operations

[Suzuki K Poulose]

On 06/10/2026 11:36, Suzuki K Poulose wrote:
> On 06/10/2026 10:24, Marc Zyngier wrote:
> > On Mon, 05 Oct 2026 10:07:42 +0100,
> > Suzuki K Poulose <suzuki.poulose@xxxxxxx> wrote:
> > > Add VM type specific S2 MMU operation backends which can be 
> > > initialized per
> > > VM flavor, to keep the handling cleaner.
> > >
> > > Signed-off-by: Suzuki K Poulose <suzuki.poulose@xxxxxxx>
> > > ---
> > > Change since v21:
> > >   - Define all vm_s2_ops call back. All calls are mandatory.
> > >   - Define callback for each flavor, disjointing the non-protetcted 
> > > pKVM and
> > >     normal KVM (VHE & nVHE) and remove the KVM_PGT_FN() hacks.
> >
> > It is a bit annoying that we still have part of the operations being
> > indirected by kvm_vm_s2_ops, and others by KVM_PGT_FN(), which is
> > still there. I was hoping that we'd have only one indirection. after
> > this patch.
>
> I agree and I did think about it. The issue is, some of these calls are
> deep burried from the higher leve dispatcher callbacks (e.g.,
> user_mem_abort->kvm_pgtable_stage2_map). We could go all in and remove
> all of them if you are happy with that change. Also, some of the call
> paths aren't valid for certain VM types, that adds quite a lot of dummy
> callbacks (now that they all are mandatory. e.g., pVM or Realms).

For the record, as discussed, there are callbacks that don't have a kvm
instance available, e.g. kvm_pgtable_stage2_free_unlinked() where we
only have a page and level. Further abstractions will be explored in a
future series.

Suzuki

> Cheers
> Suzuki
>
> >     M.