On ARM there is a problem where the interrupt handler stalls when they
are coming faster than the kernel can handle. The problem seems to
occur on RT primarily, but the problem is also valid for non-RT
kernels.

The problem is twofold:
* the handle_simple_irq() mechanism is used for GPIO, but because the GPIO
interrupt source is actually an edge triggered interrupt source, the
handle_edge_irq() mechanism must be used. While using the simple_irq()
mechanisms edges can be missed for either mainline as RT kernels.
The simple_irq mechanism is *never* meant to be used for these types
of interrupts. See the thread at: http://lkml.org/lkml/2007/11/26/73
* The RT kernels has a problem that the interrupt get masked forever while
the interrupt thread is running and a new interrupt arrives.
In the interrupt threads there is masking done in the handle_simple_irq()
path, while a simple_irq typically cannot be masked.

This patch only solves the first bullet, which is enough for AT91, by
moving the GPIO interrupt handler towards the handle_edge_irq().
To solve the problem in the simple_irq() path a seperate fix has to be done,
but as it is no longer used by AT91, that fix will not affect AT91.

Tested on:
* AT91rm9200-ek, and proprietary board
* AT91SAM9261-ek. (This patches also solves the problem that the DM9000 does
                   not work on this board while using PREEMPT-RT)
Signed-off-by: Remy Bohmer <linux@bohmer.net>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>