core: Add generic MMIO access dispatching

core: Add generic MMIO access dispatching

This introduces the infrastructure for generically dispatching MMIO
accesses. Handlers can now be registered for MMIO regions on a pre-cell
basis, removing the need for probing multiple handlers that decide
themselves if they are in charge.

The backing data structures consist of two sorted tables: one is
describing the region locations and sizes, the other one is holding, in
identical order, the handler and a corresponding opaque parameter.

Dispatching works lock-free and can even run in parallel with region
registration or removal. That latter two steps are protected against
concurrent invocation via a per-cell spinlock.

In order to preallocate sufficient space during cell setup, arch
architecture has to implement arch_mmio_count_regions that calculates
the maximum number of MMIO regions a cell may register during its
lifetime, typically based on static values and the cell configuration.
So far these functions are implemented as dummies because the MMIO
infrastructure is not yet used.

Signed-off-by: Jan Kiszka <jan.kiszka@siemens.com>