Re: [PATCH 1/2] rtc: da9063: set range

[Alexandre Belloni]

On 02/04/2019 10:33:37+0000, Steve Twiss wrote:
> Hi,
> 
> > > >  drivers/rtc/rtc-da9063.c | 3 +++
> > > >  1 file changed, 3 insertions(+)
> > > >
> > > > diff --git a/drivers/rtc/rtc-da9063.c b/drivers/rtc/rtc-da9063.c
> > > > index 1b792bcea3c7..53e690b0f3a2 100644
> > > > --- a/drivers/rtc/rtc-da9063.c
> > > > +++ b/drivers/rtc/rtc-da9063.c
> > > > @@ -475,6 +475,9 @@ static int da9063_rtc_probe(struct platform_device *pdev)
> > > >  	da9063_data_to_tm(data, &rtc->alarm_time, rtc);
> > > >  	rtc->rtc_sync = false;
> > > >
> > > > +	if (config->rtc_data_start != RTC_SEC)
> > > > +		rtc->rtc_dev->uie_unsupported = 1;
> > > > +
> > >
> > > I think we should have a comment here, like:
> > > /* FIXME: Make use of the TICK interrupt once the RTC core supports it */
> 
> Is this TICK interrupt suggestion to use the DA9063 TICK interrupt to simulate
> a second granularity in the AD alarm?
> 
> If I remember correctly, the original DA9063 patch set which was for AD silicon
> only, and which was sent to LKML before I took over looking at DA9063, used the
> DA9063 1-second TICK interrupt to count-down the seconds from the nearest
> minute in order to simulate second resolution on the RTC alarm for AD.
> 
> ... yes. Here it is. The original patch was from Krystian Garbaciak and tried to
> support RTC alarms on the AD silicon to a second resolution by counting down
> the DA9063 TICK interrupt:
> 
> https://marc.info/?l=lm-sensors&m=134613501230005&w=2
> 
> However, I dropped that patch completely and wrote a new RTC device driver
> because it didn't work in my tests.
> 
> The problem was: the TICK interrupt was indistinguishable from the ALARM
> interrupt for a wake event and when I tested AD silicon to wake up an Android
> device from suspend or power-off using the RTC IRQ, the device woke up on the
> ALARM minute (0 seconds), discovered it was not the correct time and immediately
> went back to sleep. Then it woke-up and returned back to sleep every TICK IRQ
> second until the correct alarm time was reached (up to 59 times!). At which point
> it woke up properly.
> 

No, the suggestion is to use the TICK interrupt to have a proper UIE
support even if the alarm has a minute granularity. As stated, this is
not yet supported by the core and need some work. Some RTCs have the
following in their set_alarm:

	if (tm->time.tm_sec) {
		time64_t alarm_time = rtc_tm_to_time64(&tm->time);

		alarm_time += 60 - tm->time.tm_sec;
		rtc_time64_to_tm(alarm_time, &tm->time);
	}

But my plan is to actually expose the capability to userspace and the
core so this doesn't have to be handled in the driver.


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