Re: [PATCH v2 0/2] KVM: arm64: Stop stage-2 teardown scaling with the IPA size

[Fuad Tabba]

Hi Wei-Lin,

On Thu, 05 Nov 2026 18:59:55 +0000, Wei-Lin Chang <weilin.chang@xxxxxxx> wrote:
[...]
> I haven't looked very closely, but do functions that use
> stage2_apply_range() suffer a similar problem?
>
> If I'm not mistaken we previously realized unmapping a full IPA with
> kvm_vm_stage2_unmap_range() is slow partly for the same reason.

Yes, when the range is the whole IPA, as you found on your reverse map
series [1].

> I'm not sure whether optimizing stage2_destroy_range() and
> stage2_apply_range() together in a single series is preferrable though.

I think it should be a separate series. Returning where the walk
stopped was a small change for teardown, which runs on an unlinked
table with no lock held. The stage2_apply_range() callbacks are shared
(kvm_pgtable_stage2_unmap() has EL2 callers that treat a non-zero
return as an error), and the walks drop the mmu_lock between chunks,
so it's a different change. Only NV passes stage2_apply_range() the
whole IPA.

Cheers,
/fuad

[1] https://lore.kernel.org/all/eylfcgxs2xzu57czyuhkpmum45lpxyccyu2yh3mipr67aq3xo6@xinygxzavoja/