IRQ and FIQ disabling during sleep states.

Shilimkar, Santosh santosh.shilimkar at ti.com
Wed Jul 18 12:31:20 EDT 2012


On Wed, Jul 18, 2012 at 9:49 PM, Rob Lee <rob.lee at linaro.org> wrote:
> Hello Peter and Daniel,
>
> On Thu, Jul 12, 2012 at 11:05 AM, Daniel Lezcano
> <daniel.lezcano at linaro.org> wrote:
>> On 07/11/2012 02:30 PM, Peter De Schrijver wrote:
>>> Hi,
>>>
>>> I noticed several platforms exhibit different behaviour wrt disabling IRQ or
>>> FIQ when entering a CPU idle state. Tegra for example disables both FIQ and
>>> IRQ when entering CPU clockstop, but OMAP seems to only disable FIQs? Also
>>> the generic cpuidle_wrap_enter() function, enables IRQs after return from
>>> the idle state, but doesn't disable them.
>>>
>>> Some questions:
>>>
>>> 1) I assume we need to disable IRQ/FIQ to not confuse the cpuidle code due to
>>>    late interrupts?
>
> Part of the a cpuidle irq and time keeping consolidation and cleanup
> patch that went in a while back removed all the irq disable calls from
> arm platform code as they are they are already disabled at the
> arch/arm level in kernel/process.c before calling cpu_idle_call.  I
> meant to remove these redundant platform calls for all ARM platforms
> but it's possible I missed one for Tegra if its platform irq disable
> still exists.
>
>>> 2) What prevents us from unifying this?
> It may be good to also disable the FIQ's at the same time as IRQ's in
> arch/arm/kernel/process.c.  I had made some mention of also global
> disabling FIQs in my initial RFC submission and you can see the
> responses here:
> http://lists.infradead.org/pipermail/linux-arm-kernel/2011-December/076036.html
> Given the response, I didn't pursue making any changes to FIQ
> disabling.  But if would have suggested doing so at same location as
> irq's are being disabled for ARM, perhaps it would have been better
> received.  Now that many arm systems are using FIQs for handling
> interrupts in the secure trust zone world, that may change things as
> well.
>
Actually the FIQ disable on OMAP is bogus and can be cleaned up.
When FIQ is really used, it is used in secure world and it has banked
register so the kernel FIQ disable doesn't help anyways.

Regards
Santosh



More information about the linux-arm-kernel mailing list