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

[Guenter Roeck <linux@roeck-us.net>]

On Tue, Oct 24, 2017 at 10:13:22AM -0600, Jerry Hoemann wrote:
> On Sun, Oct 22, 2017 at 07:55:19AM -0700, Guenter Roeck wrote:
> > On 10/21/2017 06:41 PM, Jerry Hoemann wrote:
> > > 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@hpe.com>
> > > > > ---
> > > > >    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.
> > > 
> > 
> > I don't see this patch as a fix. It adds functionality. the other two patches
> > are fixes and should come first.
> > 
> > > 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.
> > > 
> > Yes, but I don't think that is fixed with this patch. That would be patch 3/3, no ?
> > 
> > Guenter
> 
> Oh, the test program had problems too.
> 
> While a pretimeout doesn't have to crash the system they can. hpwdt most
> certainly does.
> 
> So, my read of how to safely handle a wdt is this:
> Let Pretimeout == 0 for wdt that don't support a pre timeout.
> A user space applications needs to ensure that:
> 
>    1. Timeout > Pretimeout
>    2. Ping < Timeout - Pretimeout
> 
> The test application did neither.  But since hpwdt wasn't supplying
> the pretimeout, even the corrected test application would have still
> failed.
> 
> Am I missing something?  How should user space daemons take into
> account pretimeouts?
> 

The watchdog core should reject bad timout (and pretimeout) values.
Of course, the hpwdt driver may not do this, since it doesn't use
the watchdog core. My take is that the driver should be converted
to use the watchdog core instead of trying to fix it.

Now, if you insist to improve the driver's pretimeout handling
instead of converting it to use the watchdog core, that is fine
with me. Please note though that I'll defer the review of such
patches to Wim since I strongly believe that it is the wrong approach.

Thanks,
Guenter