Avoiding the page cache?

Hello,

I'm trying to start using kvm to do some research here at MIT. I would like to try out some interesting page mapping ideas we have that would be easier to try out in a hypervisor than inside the linux kernel. However, I would like to start hacking a simple version of kvm that doesn't do any shadow page table caching as this would make things much easier. I noticed that some earlier versions of kvm in the mainline kernel repository (eg 399badf315bd...) around January 2007 do this, but I had trouble getting them to work with qemu on an x86_64 processor (seems like x86_64 was not supported that well in the kvm branch of qemu back then). How hard do you guys think in would be to change the current code to avoid using the cache?

I'm guessing I would have to change mmu_free_roots() to free all the pages in the shadow page table, but beyond that I'm a bit lost. Can anybody help me out with some small pointers to get me started?

Thanks!

Marek