Re: [PATCH v5 1/6] KVM: arm64: Introduce KVM_PGTABLE_S2_GUEST stage-2 flag

[wangyanan (Y)]

Hi Quentin,

On 2021/6/2 18:43, Quentin Perret wrote:
> Hi Yanan,
>
> On Thursday 15 Apr 2021 at 19:50:27 (+0800), Yanan Wang wrote:
> > diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
> > index c3674c47d48c..a43cbe697b37 100644
> > --- a/arch/arm64/include/asm/kvm_pgtable.h
> > +++ b/arch/arm64/include/asm/kvm_pgtable.h
> > @@ -61,10 +61,12 @@ struct kvm_pgtable_mm_ops {
> >    * @KVM_PGTABLE_S2_NOFWB:	Don't enforce Normal-WB even if the CPUs have
> >    *				ARM64_HAS_STAGE2_FWB.
> >    * @KVM_PGTABLE_S2_IDMAP:	Only use identity mappings.
> > + * @KVM_PGTABLE_S2_GUEST:	Whether the page-tables are guest stage-2.
> >    */
> >   enum kvm_pgtable_stage2_flags {
> >   	KVM_PGTABLE_S2_NOFWB			= BIT(0),
> >   	KVM_PGTABLE_S2_IDMAP			= BIT(1),
> > +	KVM_PGTABLE_S2_GUEST			= BIT(2),
> Assuming that we need this flag (maybe not given Marc's suggestion on
> another patch), I'd recommend re-naming it to explain _what_ it does,
> rather than _who_ is using it.
I agree with this.
> That's the principle we followed for e.g. KVM_PGTABLE_S2_IDMAP, so we
> should be consistent here as well.
But I think maybe we don't need the new flag anymore with Marc's suggestion.
Currently the CMOs right before installation or update of a PTE are 
guest-specific.
So if we also take the new optional callbacks as guest specific, then a 
new flag is
not necessary because we can check whether the callbacks have been 
initialized
to determine if we are managing a guest S2 PTE.

Thanks,
Yanan
> Thanks,
> Quentin
> .