Re: [PATCH v5] watchdog: add driver for StreamLabs USB watchdog device

[Alexey Klimov]

On Tue, Jul 26, 2022 at 4:24 AM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
>

[...]

> > > > +
> > > > +     streamlabs_wdt->intf = intf;
> > > > +     usb_set_intfdata(intf, &streamlabs_wdt->wdt_dev);
> > > > +     watchdog_set_drvdata(&streamlabs_wdt->wdt_dev, streamlabs_wdt);
> > > > +     watchdog_set_nowayout(&streamlabs_wdt->wdt_dev, nowayout);
> > > > +
> > > > +     retval = usb_streamlabs_wdt_stop(&streamlabs_wdt->wdt_dev);
> > > > +     if (retval)
> > > > +             return -ENODEV;
> > > > +
> > >
> > > A comment explaining why the watchdog is explicitly stopped when running
> > > might be useful.
> >
> > What do you mean by saying "when running"?
> > Everytime during my testing the initial state is "stopped" on
> > boot/power on/after reset, so not sure what you mean by saying "when
> > running".
>
> Should I have used the term "active" ? "started" ?
>
> > There is a comment above that explains the stop command but I will
> > add/change comments that explain things better.
> > The point of executing "stop" command here is to check that device
> > being probed behaves like we expect it to. This is a bit paranoid
> > check since I am a not 100% sure that all devices with such USB ids
> > are watchdogs -- that's why additional checks for usbdev->product and
> > usbdev->manufacturer and this stop command that doesn't change the
> > initial state. In theory, I can remove this stop command at all.
> >
>
> Normally one does not want a watchdog to stop if it is running (started ?
> active ? pick your term) when the device is instantiated to avoid gaps
> in watchdog coverage. The watchdog core provides a flag, WDOG_HW_RUNNING,
> for this purpose (sorry, there is the 'running' term again). It is used
> by many drivers, and ensures that the time from loading the Linux kernel
> to opening the watchdog device is protected by the watchdog.
>
> Calling the stop function during probe suggests that you at least
> considered the possibility that the watchdog may be running/started/active.
> Per your explanation, you still want it to stop explicitly if/when that
> happens.  All I am asking you is to add a comment explaining why this is
> not needed/wanted/relevant/supported for this driver. One explanation
> might, for example, be that the state can not be detected reliably.

Thanks for the explanation. I was confused initially by the phrase
"explicitly stopped when running" since in reality it is "explicitly
stopped when stopped" (or not active, not running).
On the second thought, I can issue a start command and indicate to the
watchdog core via set_bit(WDOG_HW_RUNNING, &wdd->status) that it is
running or return -ENODEV if the start fails instead of stopping the
watchdog. I just would like to have a command sent to the device in
->probe() that checks that the device behaves like expected. If there
are no objects I'll change it like this.

Best regards,
Alexey