The current GIC per-cpu interrupts (aka PPIs) suffer from a number of
problems:

- They use a completely separate scheme to handle the interrupts,
  mostly because the PPI concept doesn't really match the kernel view
  of an interrupt.
- PPIs can only be used by the timer code, unless we add more low-level
  assembly code.
- The local timer code can only be used by devices generating PPIs,
  and not SPIs.
- At least one platform (msm) has started implementing its own
  alternative scheme.
- Some low-level code gets duplicated, as usual...

As the previous solution which tried to map PPIs to normal interrupts
has been proved to be buggy, I've opted to a much simpler scheme
(based on Russell's input).

The proposed solution is to handle the interrupt using the same path
as SPIs, with a common handler for all PPIs. Each PPI can be requested
using gic_request_ppi(), similar to request_irq(). The local timer
code is updated to reflect these changes.

Patches against next-20110801, tested on PB11MP. As this patch series
is quite different from the previous one, I've dropped all previous
acks from platform maintainers.