Make the semantic same as the x86 one. So far the ia64 p2m has the semantic similar to (the x86 p2m) + (page reference count). But the differece from the x86 p2m have caused the breakage and work around. This patch make the ia64 p2m semantic same as x86 to solve it. - get_page() only for grant table page mapping and foreign domain page mapping. - put_page() only for those pages. - guest_physmap_remove_page() doesn't touch PGC_allocated bit.

I have to admit this patch is ugly. There's much room for clean up, especially get_page() right before put_page(). But they aren't on performance critical path (memory exchange and decrease reservation). I'm rushing because I want to push this into 3.0.5.

Tsunehisa. I haven't tested arch_memory_op(). Could you verify it?