Hi,

These are most of the patches I had in x86/lazy and are the preparatory patches
for x86 to eventually run witn MMU_LAZY_TLB_REFCOUNT=n.

The rebase and submission prompted by Rik's recent context switch optimization
which touched unuse_temporary_mm() in a way incompatible with these patches.

I'll see if I can find time this merge window to also finish that last patch
so that x86 can indeed switch to MMU_LAZY_TLB_REFCOUNT=n, that too should help
with context switch times.