Re: [PATCH] Revert "watchdog: iTCO_wdt: Account for rebooting on second timeout"

[Guenter Roeck]

On 10/8/21 12:52 AM, Jan Kiszka wrote:
> On 08.10.21 02:33, Guenter Roeck wrote:
> > This reverts commit cb011044e34c ("watchdog: iTCO_wdt: Account for
> > rebooting on second timeout") and commit aec42642d91f ("watchdog: iTCO_wdt:
> > Fix detection of SMI-off case") since those patches cause a regression
> > on certain boards (https://bugzilla.kernel.org/show_bug.cgi?id=213809).
> >
> > While this revert may result in some boards to only reset after twice
> > the configured timeout value, that is still better than a watchdog reset
> > after half the configured value.
> >
> > Fixes: cb011044e34c ("watchdog: iTCO_wdt: Account for rebooting on second timeout")
> > Fixes: aec42642d91f ("watchdog: iTCO_wdt: Fix detection of SMI-off case")
> > Cc: Jan Kiszka <jan.kiszka@xxxxxxxxxxx>
> > Cc: Mantas Mikulėnas <grawity@xxxxxxxxx>
> > Reported-by: Javier S. Pedro <debbugs@xxxxxxxxxxxxxx>
> > Signed-off-by: Guenter Roeck <linux@xxxxxxxxxxxx>
> > ---
> >   drivers/watchdog/iTCO_wdt.c | 12 +++---------
> >   1 file changed, 3 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/watchdog/iTCO_wdt.c b/drivers/watchdog/iTCO_wdt.c
> > index 643c6c2d0b72..ced2fc0deb8c 100644
> > --- a/drivers/watchdog/iTCO_wdt.c
> > +++ b/drivers/watchdog/iTCO_wdt.c
> > @@ -71,8 +71,6 @@
> >   #define TCOBASE(p)	((p)->tco_res->start)
> >   /* SMI Control and Enable Register */
> >   #define SMI_EN(p)	((p)->smi_res->start)
> > -#define TCO_EN		(1 << 13)
> > -#define GBL_SMI_EN	(1 << 0)
> >   
> >   #define TCO_RLD(p)	(TCOBASE(p) + 0x00) /* TCO Timer Reload/Curr. Value */
> >   #define TCOv1_TMR(p)	(TCOBASE(p) + 0x01) /* TCOv1 Timer Initial Value*/
> > @@ -357,12 +355,8 @@ static int iTCO_wdt_set_timeout(struct watchdog_device *wd_dev, unsigned int t)
> >   
> >   	tmrval = seconds_to_ticks(p, t);
> >   
> > -	/*
> > -	 * If TCO SMIs are off, the timer counts down twice before rebooting.
> > -	 * Otherwise, the BIOS generally reboots when the SMI triggers.
> > -	 */
> > -	if (p->smi_res &&
> > -	    (inl(SMI_EN(p)) & (TCO_EN | GBL_SMI_EN)) != (TCO_EN | GBL_SMI_EN))
> > +	/* For TCO v1 the timer counts down twice before rebooting */
> > +	if (p->iTCO_version == 1)
> >   		tmrval /= 2;
> >   
> >   	/* from the specs: */
> > @@ -527,7 +521,7 @@ static int iTCO_wdt_probe(struct platform_device *pdev)
> >   		 * Disables TCO logic generating an SMI#
> >   		 */
> >   		val32 = inl(SMI_EN(p));
> > -		val32 &= ~TCO_EN;	/* Turn off SMI clearing watchdog */
> > +		val32 &= 0xffffdfff;	/* Turn off SMI clearing watchdog */
> >   		outl(val32, SMI_EN(p));
> >   	}
> >   
>
> Sigh, how broken is this architecture of the iTCO? Agreed, this leaves
> no option then.
>
> BTW, the fact that we saw an inconsistency in read-back timeout
> indicates that there is still an issue for the remaining /= 2 case
> (means v1), but I'm loosing interest in fixing those issues, given how
> hard it is to test broadly without breaking users first.

Agreed. This is because the /=2 handling is only implemented in
iTCO_wdt_set_timeout() without matching code in iTCO_wdt_get_timeleft().
I don't have hardware to test, so I am not going to touch that code
myself. We can address that if/when someone reports the actual problem
and has the ability to test a fix.

Thanks,
Guenter