Re: [RFC] disabling RTC irq during release

[David Brownell <david-b@pacbell.net>]

On Wednesday 04 July 2007, Rodolfo Giometti wrote:
> Hello,
> 
> Looking at other rtc drivers I noticed that during the release()
> method we should disable IRQs as follow:
> 
> static void ds1307_release(struct device *dev)
> {
>         struct ds1307   *ds1307 = dev_get_drvdata(dev);
> 
>         if (ds1307->irq >= 0) {
>                 ds1307->irqen = 0;
>                 ds1307_update_alarm(ds1307);
>         }
> }

On the RTC list, the outcome of that discussion was that
the release() method -- not generally implemented at this
time, and only supporting the /dev/rtcN access mode -- should
only disable the IRQs enabled through that /dev/rtcN file.

Now, it turns out that of the three kinds of RTC IRQ (alarm,
once-per-second, and 2^N-per-second), two of those modes can
be accessed through in-kernel APIs outside /dev/rtcN calls:

 - 2^N-per-second "periodic" irqs, using rtc_irq_set_freq()
   and rtc_irq_set_state()

 - alarm irqs, using rtc_set_alarm()

So it would be incorrect to disable those IRQs, when some
other kernel activity had enabled them.  But for now it
would be correct to always disable once-per-second "update"
IRQs in a release() method.

There's a separate issue -- longstanding, not new to the
RTC framework -- of how several consumers of RTC IRQs ought
to share the different kinds of IRQ.


> But if I wish using RTC IRQ line to turn also the board ON, not just
> to put to sleep it, I should leave the IRQ line on!

That can be done with /sys/class/rtcN/device/power/wakealarm,
writing it to the time (in seconds) the RTC alarm should kick
the board's power-on circuitry.

- Dave