Chris Wright wrote: >* Zachary Amsden (zach@xxxxxxxxxx) wrote: > > >>Doesn't that require 16 pages per CPU? That seems excessive to impose >>on a native build. Perhaps we could get away with 1 page per CPU for >>the GDT on native boots and bump that up to 16 if compiling for a >>virtualized sub-architecture - i.e. move GDT to a page aligned struct >>for native (doesn't cost too much), and give it MACH_GDT_PAGES of space >>which is defined by the sub-architecture. >> >> > >For Xen, the gdt is one page per cpu (so it's not one page per table >entry). > > I thought Xen itself used segments like 0xe020 ... but I haven't looked in a while. In any case, this is good news. One page per GDT per cpu is good for everybody, and hopefully an overhead that can be lived with on native builds. Zach