Hi Alexandre,

On Thu, Jul 16, 2020 at 08:18:16PM +0200, Alexandre Belloni wrote:
> On 16/07/2020 16:47:05+0200, Uwe Kleine-König wrote:
> > On Thu, Aug 22, 2019 at 03:19:35PM +0200, Bruno Thomsen wrote:
> > > Add partial support for the watchdog functionality of
> > > both PCF2127 and PCF2129 chips.
> > 
> > I have a board here with a pcf2127 that has the #RST pin
> > not connected.
> > 
> > The problem this creates is: The bootloader arms the SoC's watchdog and
> > jumps into Linux. The pcf2127 driver happens to load first, so watchdog0
> > is provided by the RTC (but non-functional). Systemd is configured to
> > feed the watchdog, but happens to feed the wrong one, so the machine
> > resets shortly after it is up :-|
> > 
> > So I wonder if we need a dt property that tells the driver if the RST
> > line is connected or not.
> 
> I guess the current solution is to set WatchdogDevice to point to a link
> that is updated by udev thus ensuring it points to the correct watchdog
> device regardless of the probe order.
> 
> This would be similar to the /dev/rtc symlink, pointing to the systohc
> RTC by default (even if I don't really like that heuristic).
> 
> What you suggest is somewhat okay but doesn't really solve the issue if
> both watchdogs are functional and systemd still doesn't pick the one
> that is armed by the bootloader.

Yes, my suggestion doesn't solve the problem "Oh, there are two
watchdogs, which should I feed?". But IMHO in this case there shouldn't
be a watchdog device provided at all by an RTC that can be a watchdog
but isn't wired up correctly.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |