[PATCH v3 07/10] watchdog: wdt-uclass.c: handle all DM watchdogs in watchdog_reset()

[Simon Glass]

Hi Rasmus,

On Tue, 3 Aug 2021 at 00:48, Rasmus Villemoes
<rasmus.villemoes at prevas.dk> wrote:
>
> On 02/08/2021 21.20, Simon Glass wrote:
> > Hi Rasmus,
> >
> >> But let me one last time repeat why I think the post_probe approach is
> >> the cleanest and a natural fit for DM: post_probe is (AIUI) a place
> >> where a uclass can do some action it wants done for every device
> >> belonging to that uclass. When CONFIG_WATCHDOG_AUTOSTART is set,
> >> wdt_uclass does have such an action. When it's not set, the post_probe
> >> method is a no-op.
> >
> > I can certainly see you point. It definitely looks very natural for DM.
> >
> > But I still think the problem of a device possibly resetting the board
> > when probed (despite the various watchdog-reset things sprinkled
> > throughout the code)
>
> But that is the _very point_ of having the developer be able to opt in
> [1] to starting the watchdog device ASAP (i.e. as soon as U-Boot is
> capable of handling it). If some part of the boot process hangs or
> U-Boot goes into an inf-loop without hitting a WATCHDOG_RESET in there,
> one wants (or may want) the board to reset and hopefully follow some
> other path the next time around. If the developer/project doesn't want
> or need that, WATCHDOG_AUTOSTART can be disabled. And if the developer

That requires a rebuild though.

This is more of a driver-model thing than anything else and I know it
seems like a minor point. But the way it is designed at present,
probing a device should get it ready for use and not actually start it
doing something. That is the way all(?) devices work and I think it
makes sense to keep watchdogs consistent with that.

> wants to make use of this, it makes sense to have the watchdog device
> monitor as much of the boot process as possible.

This is not an issue, since you enable all watchdogs immediately at present.

>
> [1] ok, for historical reasons it's default-y, so actually an opt-out
> knob, but the developer does have a choice regardless.
>

Regards,
Simon