Re: [RFC PATCH 02/12] KVM: arm64: Allow visiting block PTEs in post-order

[Ricardo Koller <ricarkol@google.com>]

On Mon, Nov 14, 2022 at 06:48:04PM +0000, Oliver Upton wrote:
> On Sat, Nov 12, 2022 at 08:17:04AM +0000, Ricardo Koller wrote:
> > The page table walker does not visit block PTEs in post-order. But there
> > are some cases where doing so would be beneficial, for example: breaking a
> > 1G block PTE into a full tree in post-order avoids visiting the new tree.
> > 
> > Allow post order visits of block PTEs. This will be used in a subsequent
> > commit for eagerly breaking huge pages.
> > 
> > Signed-off-by: Ricardo Koller <ricarkol@google.com>
> > ---
> >  arch/arm64/include/asm/kvm_pgtable.h |  4 ++--
> >  arch/arm64/kvm/hyp/nvhe/setup.c      |  2 +-
> >  arch/arm64/kvm/hyp/pgtable.c         | 25 ++++++++++++-------------
> >  3 files changed, 15 insertions(+), 16 deletions(-)
> > 
> > diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
> > index e2edeed462e8..d2e4a5032146 100644
> > --- a/arch/arm64/include/asm/kvm_pgtable.h
> > +++ b/arch/arm64/include/asm/kvm_pgtable.h
> > @@ -255,7 +255,7 @@ struct kvm_pgtable {
> >   *					entries.
> >   * @KVM_PGTABLE_WALK_TABLE_PRE:		Visit table entries before their
> >   *					children.
> > - * @KVM_PGTABLE_WALK_TABLE_POST:	Visit table entries after their
> > + * @KVM_PGTABLE_WALK_POST:		Visit leaf or table entries after their
> >   *					children.
> 
> It is not immediately obvious from this change alone that promoting the
> post-order traversal of every walker to cover leaf + table PTEs is safe.
> 
> Have you considered using a flag for just leaf post-order visits?
> 

Not using this commit in v1. There's no (noticeable) perf benefit from
avoiding visiting the new split tree.

Thanks,
Ricardo

> --
> Thanks,
> Oliver