On Fri, Jun 17, 2016 at 4:00 PM, Peter Maydell <peter.mayd...@linaro.org> wrote: > On 17 June 2016 at 11:20, Vijay Kilari <vijay.kil...@gmail.com> wrote: >> On Fri, Jun 17, 2016 at 1:12 AM, Richard Henderson <r...@twiddle.net> wrote: >>> On 06/14/2016 04:36 AM, Peter Maydell wrote: >>>> It would be better to delay the point at which we allocate >>>> the data structures which care about page size, rather than >>>> moving init of the CPU earlier. >>> >>> It would be *best* if we could re-initialize and re-allocate these data >>> structures so that we can follow the current page size as it changes. >>> >>> Yes, this might require flushing just about everything, but it's the kind of >>> thing that's likely to happen only at system startup. After that, >>> everything >>> benefits from having the correct (larger) page size. >> >> I tried shuffling the memory initialization code after cpu initialization. >> but it was full mess. So could not proceed further. >> >> However, I tried early creation cpu objects instead of doing it from >> machvirt_init. >> With this, initfn of the cpu model is called earlier and feature set is >> updated. >> With that I could fetch arm architecture info. Based on this we can >> choose page size. > > This won't work, because machvirt_init needs to be able to specify > properties of the CPU. I think we need to solve the issues with > dynamically reinitializing and reallocating the data structures, > as rth suggests.
In early init, only cpu object is created and does not specify properties of the CPU. machvirt_init will reuse already created cpu object and specify properties of the CPU. Regards Vijay
