>> 在 2016年2月22日，19:03，Peter Maydell <peter.mayd...@linaro.org> 写道： >> >>> On 22 February 2016 at 10:28, hitmoon <zxq_yx_...@163.com> wrote: >>> 在 2016年02月22日 17:24, Peter Maydell 写道: >>> Just noticed this won't work as it is -- the vmstate >>> struct is for the M48t59State*, but the ISA and >>> Sysbus wrappers have their own structs which are >>> what the dc->vmsd will be wanting to operate on. >>> You'd need extra VMState structs I think and >>> somebody who knows migration better than me to say >>> whether that is a migration compat break. > >> It seems that the old code also use the same vmstate structure. Maybe it's a >> common structure which will not be used at the same time. > > The old code passes vmstate_register() a pointer to the > M48t59State, and so the offsets in the vmstate line up correctly > with the fields it wants to access. If you use dc->vmsd then the > pointer that is (implicitly) used is the pointer to the device > structure itself, which is not the same address as the > M48t59State embedded inside that struct. > > You'd basically need to have extra VMState structures for the > devices themselves which just said "inside this M48txxISAState > is an M48t59State", and "inside this M48txxSysBusState is a > M48t59State" and referred to the existing vmstate for the > M48t59State. > > thanks > -- PMM
I will have a closer look.
