On 04/05/2021 12:08:58+0200, Martin Kaiser wrote:

Thus wrote Alexandre Belloni (alexandre.belloni@bootlin.com):

quoted

quoted

 	platform_set_drvdata(pdev, imxdi);

quoted

quoted

+	device_set_wakeup_capable(&pdev->dev, true);

quoted

Maybe it makes sense to simply use device_init_wakeup here.

the explanation for device_init_wakeup

"By default, most devices should leave wakeup disabled. The exceptions
are devices that everyone expects to be wakeup sources: keyboards, power
buttons, ..."

made me think that device_set_wakeup_capable is more appropriate here. I
can change this if you want.

Doesn't everyone expect the RTC to be a wakeup source? :)

However, if I compile rtc-imxdi as a module and use device_init_wakeup,
the module can't be unloaded any more. The reason is that alarmtimer
(kernel/time/alarmtimer.c) starts using rtc-imxdi as its backing rtc
device and holds a reference to it. It seems that alarmtimer has no way
to relinquish its backing rtc device, regardless of any pending alarms.

What is the right approach here? Are there any rtc drivers that act as a
wakeup source and can still be unloaded if compiled as a module?

Yes, when you don't have alarmtimer ;)
I honestly think the RTC selection needs to be a bit more dynamic but at
the same time, it would not be great to change it at suspend time. I
guess the best way would be to allow module unloading and tracking when
the RTC disappears.


-- 
Alexandre Belloni, co-owner and COO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com