As I've said, there's not much other choice here. ?We _have_ to have an
identity mapping in place to bring the MMU back online, or we'll crash.
That's non-optional. ?So if this is a problem for cpuidle, cpuidle needs
to ensure that a page table switch has happened prior to the CPU going
down which _can_ be safely modified.
I fully agree with you, but what I was suggesting is to create
temporary identity page table at boot time, the resume code will load
that page table (Context ID will be loaded with unique ASID), enable
mmu, then the original page table and Context ID will be restored.
saeed