On 11 October 2016 at 12:20, Richard Henderson <r...@twiddle.net> wrote: > On 10/11/2016 12:08 PM, Peter Maydell wrote: >> I would ideally have liked to finalize things much later, but >> this is in practice hugely difficult because so many things >> (in particular all the address space/memory system code) >> assume the target page size is known.
> Unfortunate. I suppose that 4k is still better than 1k, but > I was hoping to get 16k or 64k (or higher) when the OS is > configured to use such. I.e. totally dynamically configurable > upon write to the appropriate cpu register. I think that would run into problems with migration: the migration stream all works in guest-pages of ram and a mismatch means migration doesn't work. > Given how the memory subsystem already dynamically reconfigures > itself for changes in address_space topology, I assumed page size > changes would be trivial and fall out naturally. The trouble is that all the data structures work in terms of page sizes (even though we support sub-page allocations those are still done by carving up a page-size chunk). It could probably be done but it looked like a gargantuan task so I decided this was a better compromise. thanks -- PMM
