Re: [PATCH 2/2] watchdog: gpio: add configurable minimum interval

[Rob Herring <robh@kernel.org>]

On Fri, Jul 08, 2022 at 04:44:25AM -0700, Guenter Roeck wrote:
> On Thu, Jul 07, 2022 at 11:00:45AM +0200, Isaac True wrote:
> > On Tue, 5 Jul 2022 at 21:56, Rob Herring <robh@kernel.org> wrote:
> > >
> > > On Mon, Jul 04, 2022 at 01:05:04PM +0200, Isaac True wrote:
> > > > On Fri, 1 Jul 2022 at 19:48, Guenter Roeck <linux@roeck-us.net> wrote:
> > > > >
> > > > > On 7/1/22 10:18, Rob Herring wrote:
> > > > > > On Wed, Jun 29, 2022 at 01:06:26PM +0200, Isaac True wrote:
> > > > > >> Add the "min_hw_margin_ms" parameter to gpio_wdt devices, allowing a
> > > > > >> minimum interval to be specified, stopping watchdog devices from being
> > > > > >> fed too quickly if they require a certain interval between feeds.
> > > > > >
> > > > > > I assume there is some real platform with a real problem you are trying
> > > > > > to solve? Details please.
> > > > > >
> > > > >
> > > > > Agreed, this should be explained in more detail.
> > > >
> > > > Yes this is a real platform using a TI TPS3850 watchdog chip. With
> > > > this chip you can configure a "window" which can detect early faults
> > > > (i.e. too frequent) in addition to the standard watchdog features. I
> > > > needed to add this minimum timeout to avoid watchdog resets in
> > > > situations such as where first U-Boot and then the Linux kernel feed
> > > > the watchdog with too short of an interval between them, or when
> > > > systemd was configured to use the watchdog device and was feeding it
> > > > too soon after taking over from the kernel.
> > > >
> > > > > > Can you just hardcode some min? Maybe 10% of the max or something. Is
> > > > > > there a downside to a larger than necessary min?
> > > > > >
> > > > >
> > > > > That would result in extra overhead in the watchdog core which would not
> > > > > be required for all other hardware using this driver. I'd rather avoid that.
> > > > >
> > > >
> > > > In the case of the TI TPS3850, the minimum timeout is configurable, so
> > > > I didn't want to add a hard-coded value to the driver.
> > > >
> > > > > > Wouldn't be better to fix this without requiring a DT change and that
> > > > > > could work on stable kernels if needed.
> > > > > >
> > > > >
> > > > > Presumably that is some new hardware. Most of the watchdog drivers
> > > > > needing this value can derive it from the compatible property. The
> > > > > gpio watchdog driver is a bit different since it is supposed to work
> > > > > on a variety of hardware using gpio pins for watchdog control.
> > > > >
> > > >
> > > > Yes this is new hardware. This use case is also not very common as
> > > > most watchdog chips don't have this window function or a minimum
> > > > interval, at least in my experience, so I did not want to make it the
> > > > default for everything.
> > >
> > > Okay. However the existing property you copied has 2 problems. It uses
> > > underscores rather than hypens and doesn't use a standard unit suffix.
> > > So 'min-hw-margin-ms'.
> > >
> > > Though maybe a new property instead:
> > >
> > > timeout-range-ms = <min max>;
> > >
> > > That's somewhat aligned to 'timeout-sec', and IMO, clearer meaning than
> > > 'hw margin'.
> > >
> > > Rob
> > 
> > I agree that both the original property name and the new one aren't
> > great, but I didn't want to go changing the existing property for
> > everyone.  I could definitely add a new "timeout-range-ms" property -
> > should that be added in parallel to the original hw_margin_ms (i.e.
> > you can use one or the other), or completely replace the original?
> 
> I wonder how that made it in in the first place. Embarrassed ...
> of course it wasn't reviewed by a DT maintainer. I'd suggest to mark
> the old property as deprecated (if that is possible) and define
> the new one.

Seems doable. Only 1 user upstream.

Also, note that there is also 'rohm,hw-timeout-ms' which is a range.

Rob