ARM: clean up idle handlers

Let's factor out the need_resched() check instead of having it duplicated
in every pm_idle implementations to avoid inconsistencies (omap2_pm_idle
is missing it already).

The forceful re-enablement of IRQs after pm_idle has returned can go.
The warning certainly doesn't trigger for existing users.

To get rid of the pm_idle calling convention oddity, let's introduce
arm_pm_idle() allowing for the local_irq_enable() to be factored out
from SOC specific implementations. The default pm_idle function becomes
a wrapper for arm_pm_idle and it takes care of enabling IRQs closer to
where they are initially disabled.

And finally move the comment explaining the reason for that turning off
of IRQs to a more proper location.

Signed-off-by: Nicolas Pitre <nicolas.pitre@linaro.org>
Acked-and-tested-by: Jamie Iles <jamie@jamieiles.com>

arch/arm/include/asm/system.h

@@ -110,6 +110,7 @@ extern void cpu_init(void);
 
 void soft_restart(unsigned long);
 extern void (*arm_pm_restart)(char str, const char *cmd);
+extern void (*arm_pm_idle)(void);
 
 #define UDBG_UNDEFINED	(1 << 0)
 #define UDBG_SYSCALL	(1 << 1)