On 4 December 2013 12:33, Fedorov Sergey <s.fedo...@samsung.com> wrote: > > On 12/04/2013 03:18 PM, Peter Maydell wrote: >> >> On 4 December 2013 10:58, Peter Crosthwaite >> <peter.crosthwa...@xilinx.com> wrote: >>> >>> So what im proposing is just a slightly more general patch. Is it >>> really any more complicated than just applying your change pattern for >>> the hyp mode? >> >> I think it would be, because of the wrinkle that hyp mode doesn't >> have a banked LR, so the existing "assume we can just translate >> the mode into a single index good for both LR and SP" logic >> would break. >> >> The minimal change if we wanted to keep VMSD bumps to a >> minimum would be to increase the size of the banked_spsr[] >> and banked_r13[] arrays to allow for Hyp mode but do nothing >> else (except add a comment about it I guess). > > > If we want to bump VMSD just once for monitor + hypervisor mode then we need > to add ELR_hyp register definition too. I think then it would be better to > implement hypervisor mode and its special banking scheme, too. But I doubt > it worth to combine these two things into one patch.
It's possible to add single new fields to the VMState without requiring a compatibility break, by marking the new field as "only present in version X or greater"; new elements on the end of arrays are a little fiddlier. But yes, I think we should just not worry about possible future Hyp mode now. Let's stick with your current patch. thanks -- PMM
