Re: [PATCH 1/3] watchdog: hpwdt: add ioctl WDIOC_GETPRETIMEOUT

[Jerry Hoemann]

On Fri, Oct 20, 2017 at 07:25:20PM -0700, Guenter Roeck wrote:
> On 10/20/2017 03:54 PM, Jerry Hoemann wrote:
> > Add support for WDIOC_GETPRETIMEOUT ioctl so that user applications
> > can determine when the NMI should arrive.
> > 
> > Signed-off-by: Jerry Hoemann <jerry.hoemann@xxxxxxx>
> > ---
> >   drivers/watchdog/hpwdt.c | 7 +++++++
> >   1 file changed, 7 insertions(+)
> > 
> > diff --git a/drivers/watchdog/hpwdt.c b/drivers/watchdog/hpwdt.c
> > index 67fbe35..ef54b03 100644
> > --- a/drivers/watchdog/hpwdt.c
> > +++ b/drivers/watchdog/hpwdt.c
> > @@ -50,6 +50,7 @@
> >   static bool nowayout = WATCHDOG_NOWAYOUT;
> >   static char expect_release;
> >   static unsigned long hpwdt_is_open;
> > +static const int pretimeout = 9;
> >   static void __iomem *pci_mem_addr;		/* the PCI-memory address */
> >   static unsigned long __iomem *hpwdt_timer_reg;
> > @@ -622,6 +623,12 @@ static long hpwdt_ioctl(struct file *file, unsigned int cmd,
> >   		}
> >   		break;
> > +	case WDIOC_GETPRETIMEOUT:
> > +		ret = copy_to_user(argp, &pretimeout, sizeof(pretimeout));
> > +		if (ret)
> > +			ret = -EFAULT;
> > +		break;
> > +
> >   	case WDIOC_SETTIMEOUT:
> >   		ret = get_user(new_margin, p);
> >   		if (ret)
> > 
> 
> Can you please convert the driver to use the watchdog subsystem instead ?
> If there are still improvements needed afterwards, they can still be
> implemented, but we really should not make improvements which are
> already supported by the watchdog core.

I will look into converting the driver, but would like to get this
fix in independently.

SuSE brought https://bugzilla.novell.com/show_bug.cgi?id=1042933
to my attention earlier this summer.  The submitter was trying to
develop a watchdog test where the ping rate was set to be the
Timeout/2.

The test worked fine until (Timeout/2) < PreTimeout.  At this point
an NMI would be delivered to the system before the test could refresh
the timer.

I came to the view that a watchdog that implements a pre-timeout NMI
where the value of the pre-timeout is not known programmatically as having
a defect.

This problem has been around a long time and we could live with it, but
figured while I was in fixing other problems, I'd fix this one as well.

Thanks

-- 

-----------------------------------------------------------------------------
Jerry Hoemann                  Software Engineer   Hewlett Packard Enterprise
-----------------------------------------------------------------------------