On 26 July 2015 at 14:52, Pavel Fedin <p.fe...@samsung.com> wrote: > And could you be not so strict on this file?
OK, so having thought about this this week, this is what I suggest. For this patch series we should retain this class hierarchy, but remove all the fields corresponding to state information which isn't used by the KVM subclass (and all the accessor macro definitions, and that GICV3_NCPU definition, and so on). We should have a migration information struct which says "not migratable". That gives us the right structure into which we can add the correct state fields when we get the design right (and for whichever of emulation and KVM state save/load gets there first). We do need to make sure we get right: * the class hierarchy * the API to the rest of QEMU (meaning of GPIO pins, memory regions, any qdev/qom properties) (and in the other parts of the patchset, also that we get the right memory map for the virt board with the GICv3 in it and that the mechanism for selecting GICv2 or GICv3 or "whatever you've got" is good.) thanks -- PMM
