Re: [PATCH/RFC 00/03] irqchip: renesas-irqc: Fine grained Runtime PM support

[Kevin Hilman <khilman@kernel.org>]

Geert Uytterhoeven <geert@linux-m68k.org> writes:

> On Thu, Apr 23, 2015 at 4:44 PM, Alan Stern <stern@rowland.harvard.edu> wrote:
>> On Thu, 23 Apr 2015, Geert Uytterhoeven wrote:
>>> >> I'm afraid you can't call pm_runtime_get_sync() from these methods, as
>>> >> they may be called from interrupt context.
>>> >
>>> > Ouch. I know the clock framework has prepare/enable separated with
>>> > context, but with the irqchip callbacks I suppose no such separation
>>>
>>> It's not the clock operations, but the pm_runtime operations that cannot be
>>> called from interrupt context.
>>
>> In fact the pm_runtime operations _can_ be called from interrupt
>> context, provided the driver has first invoked pm_runtime_irq_safe().
>> Of course, this requires that none of the runtime-PM callback routines
>> ever sleep or perform a blocking operation.
>>
>> This is all explained in Documentation/power/runtime_pm.txt (search for
>> "irq_safe").
>
> Perhaps that can help. We'll have to give it a try...
>
> I've always found this a bit strange when PM Domains are involved:
> pm_runtime_irq_safe(dev) applies to device dev, while the actual callbacks
> belong to the PM Domain code (the device's driver doesn't have any).

Currently PM domains don't support IRQ-safe devices very well.  Any
irq_safe devices will prevent the entire PM domain from ever powering
off.  There is some on-going work for this but it requires reworking the
genpd locking a bit.

Kevin