Re: [PATCH 7/7] watchdog: s3c2410: Let kernel kick watchdog

[Guenter Roeck]

On 10/30/21 9:59 AM, Sam Protsenko wrote:
> On Sat, 30 Oct 2021 at 18:14, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > On 10/30/21 7:29 AM, Sam Protsenko wrote:
> > > On Fri, 29 Oct 2021 at 03:30, Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> > > > On 10/28/21 11:35 AM, Sam Protsenko wrote:
> > > > > When "tmr_atboot" module param is set, the watchdog is started in
> > > > > driver's probe. In that case, also set WDOG_HW_RUNNING bit to let
> > > > > watchdog core driver know it's running. This way wathcdog core can kick
> > > > > the watchdog for us (if CONFIG_WATCHDOG_HANDLE_BOOT_ENABLED option is
> > > > > enabled), until user space takes control.
> > > > >
> > > > > Signed-off-by: Sam Protsenko <semen.protsenko@xxxxxxxxxx>
> > > > > ---
> > > > >     drivers/watchdog/s3c2410_wdt.c | 26 +++++++++++++++-----------
> > > > >     1 file changed, 15 insertions(+), 11 deletions(-)
> > > > >
> > > > > diff --git a/drivers/watchdog/s3c2410_wdt.c b/drivers/watchdog/s3c2410_wdt.c
> > > > > index ca082b1226e3..9af014ff1468 100644
> > > > > --- a/drivers/watchdog/s3c2410_wdt.c
> > > > > +++ b/drivers/watchdog/s3c2410_wdt.c
> > > > > @@ -732,6 +732,21 @@ static int s3c2410wdt_probe(struct platform_device *pdev)
> > > > >         wdt->wdt_device.bootstatus = s3c2410wdt_get_bootstatus(wdt);
> > > > >         wdt->wdt_device.parent = dev;
> > > > >
> > > > > +     /*
> > > > > +      * If "tmr_atboot" param is non-zero, start the watchdog right now. Also
> > > > > +      * set WDOG_HW_RUNNING bit, so that watchdog core can kick the watchdog.
> > > > > +      *
> > > > > +      * If we're not enabling the watchdog, then ensure it is disabled if it
> > > > > +      * has been left running from the bootloader or other source.
> > > > > +      */
> > > > > +     if (tmr_atboot && started == 0) {
> > > > > +             dev_info(dev, "starting watchdog timer\n");
> > > > > +             s3c2410wdt_start(&wdt->wdt_device);
> > > > > +             set_bit(WDOG_HW_RUNNING, &wdt->wdt_device.status);
> > > > > +     } else if (!tmr_atboot) {
> > > > > +             s3c2410wdt_stop(&wdt->wdt_device);
> > > > > +     }
> > > > > +
> > > >
> > > > This doesn't cover the case where the watchdog is already enabled by the BIOS.
> > > > In that case, WDOG_HW_RUNNING won't be set, and the watchdog will time out
> > > > if the userspace handler is not loaded fast enough. The code should consistently
> > > > set WDOG_HW_RUNNING if the watchdog is running.
> > >
> > > As I understand, in the case when bootloader started the watchdog, the
> > > driver just stops it. You can see it in the code you replied to.
> > >
> > >       } else if (!tmr_atboot) {
> > >               s3c2410wdt_stop(&wdt->wdt_device);
> > >
> > > In other words, having "tmr_atboot" module param makes it irrelevant
> > > whether bootloader enabled WDT or no.
> >
> > Sure, but I am concerned about "if (tmr_atboot && started)", which doesn't
> > set WDOG_HW_RUNNING with your current code, and I was looking for something
> > like
> >
> >          if (tmr_atboot) {
> >                  if (!started) {
> >                          dev_info(dev, "starting watchdog timer\n");
> >                          s3c2410wdt_start(&wdt->wdt_device);
> >                  }
> >                  set_bit(WDOG_HW_RUNNING, &wdt->wdt_device.status);
> >          } else {
> >                  s3c2410wdt_stop(&wdt->wdt_device);
> >          }
>
> Wow, I really overlooked that case. Nice catch! Not having '} else {'
> section is vicious...
>
> Though if started != 0, it means s3c2410wdt_set_heartbeat() failed to
> set wdd->timeout, and without that the watchdog core won't be able to
> calculate correctly ping interval in watchdog_next_keepalive(), and
> WDOG_HW_RUNNING bit won't do much good, right? So I'll probably just
> call s3c2410wdt_stop() in that case, to be on the safe side.
>
> Also this 'started' variable name is misleading, I'll convert it to
> "bool timeout_ok" while at it.

This driver is a mess :-(. "started" true means that the driver doesn't
work as currently written because there is no known valid timeout. In
reality, s3c2410wdt_set_heartbeat() should in that case select a valid
timeout and set it. On top of that, a timeout value out of range should
never be passed to it in the first place. The check for "if (timeout < 1)"
is, in that context, pointless. The range check should happen in
s3c2410wdt_max_timeout(). If that range check is invalid, ie if
s3c2410wdt_set_heartbeat() fails even though the timeout is in the range
of 1 ..s3c2410wdt_max_timeout(), s3c2410wdt_max_timeout() is buggy.

The simplest fix (kludge/hack) might be to fail driver installation if
s3c2410wdt_set_heartbeat() fails.

Guenter