[PATCH v2 1/3] dt-bindings: rtc: pcf8523: add DSM pm option for battery switch-over

[Alexandre Belloni]

On 27/07/2020 16:24:39+0100, Russell King - ARM Linux admin wrote:
> On Mon, Jul 27, 2020 at 04:49:38PM +0200, Alexandre Belloni wrote:
> > On 27/07/2020 10:45:53+0100, Russell King - ARM Linux admin wrote:
> > > > This is but this shouldn't be a DT property as it has to be changed
> > > > dynamically. I'm working on an ioctl interface to change this
> > > > configuration.
> > > 
> > > Why does it need to be changed dynamically?  If the hardware components
> > > are not fitted to allow the RTC to be safely used without DSM, then
> > > why should userspace be able to disable DSM?
> > 
> > For RTCs with a standby mode, you want to be able to return to standby
> > mode.
> > 
> > That would happen for example after factory flashing in that common use
> > case:
> >  - the board is manufactured
> >  - Vbackup is installed, the RTC switches to standby mode
> >  - the board is then booted to flash a system, Vprimary is now present,
> >    the RTC switches to DSM.
> > 
> > At this point, if the board is simply shut down, the RTC will start
> > draining Vbackup before leaving the factory. Instead, we want to be able
> > to return to standby mode until the final user switches the product on
> > for the first time.
> 
> I don't think you're understanding what's going on with this proposed
> patch.  The cubox-i does work today, and the RTC does survive most
> power-downs. There are situations where it doesn't.
> 
> So, let's take your process above.
> 
> - the board is manufactured
> - Vbackup is installed, the RTC switches to standby mode
> - the board is then booted to flash a system, Vprimary is now present
> - the board is powered down.  the RTC _might_ switch over to battery
>   if it notices the power failure in time, or it might not.  A random
>   sample of units leaving the factory have the RTC in standby mode.
>   Others are draining the battery.
> 
> I'm not saying what you propose isn't a good idea.  I'm questioning
> why we should expose this in the generic kernel on platforms where
> it's likely to end up with the RTC being corrupted.
> 

Note that I didn't say we should expose settings that are not working
but it is a different discussion. I was explaining why we need to be
able to change it dynamically.

> Now, I question your idea that units should leave the factory without
> the RTC being programmed.  We know that lovely systemd goes utterly
> bonkers if the system time is beyond INT_MAX.  If the RTC leaves
> standby mode containing a date which we translate beyond INT_MAX,
> systemd will refuse to boot the system, and the user will have no
> way to set the correct time.  The user returns the device to the
> supplier as faulty...

This is doesn't happen since v4.17.

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