Hi Peter, On 7/10/20 11:47 AM, Peter Maydell wrote: > On Wed, 8 Jul 2020 at 15:20, Eric Auger <eric.au...@redhat.com> wrote: >> >> Expose the RIL bit so that the guest driver uses range >> invalidation. Range invalidation being an SMMU3.2 feature, >> let AIDR advertise SMMUv3.2 support. >> >> Signed-off-by: Eric Auger <eric.au...@redhat.com> > > I think that to advertise SMMUv3.2 we would also need to > set the IDR3.BBML field to something non-zero. That means > we need to analyze our implementation of the caching of page > table structures to see if we need to do anything different > (per the behaviours and guarantees described in section 3.21.1 > of the spec). you're right. I need to further study this feature.
I felt difficult to find out which features are mandatory for a given revision number. > > Alternatively, we could take advantage of the language > in section 2.5 that says that a v3.x implementation is > allowed to implement features from v3.(x+1), and just > set the RIL bit while leaving AIDR advertising us as v3.1. Indeed :-) Thank you for the review! Eric > > thanks > -- PMM >