[PATCH v2 5/6] watchdog: at91sam9: request the irq with IRQF_NO_SUSPEND

[Boris Brezillon]

On Thu, 5 Mar 2015 12:17:23 +0100
Boris Brezillon <boris.brezillon at free-electrons.com> wrote:

> Hi Boris,

     ^ Mark,

I'm suffering from a dual personality disorder :-)

> 
> On Thu, 5 Mar 2015 10:53:08 +0000
> Mark Rutland <mark.rutland at arm.com> wrote:
> 
> > Hi Boris,
> > 
> > I'd missed the fact that this was for SW watchdog as opposed to HW
> > watchdog, which may explain my confusion.
> > 
> > [...]
> > 
> > > > >  		err = request_irq(wdt->irq, wdt_interrupt,
> > > > > -				  IRQF_SHARED | IRQF_IRQPOLL,
> > > > > +				  IRQF_SHARED | IRQF_IRQPOLL |
> > > > > +				  IRQF_NO_SUSPEND,
> > > > 
> > > > I'm a little confused by this. What happens if the watchdog fires when
> > > > we're actually in the suspended state (when IRQF_NO_SUSPEND interrupts
> > > > aren't guaranteed to be delivered).
> > > 
> > > It reboot the system.
> > 
> > Is the timer we use to ping the watchdog guaranted to result in a wakeup
> > before an interrupt will be triggered? If so, then I think we're ok.
> 
> It should be (I don't recall exactly what the logic is, but it's at
> least half the watchdog time limit).
> 
> > 
> > If not, then don't we need to clear a potentially pending watchdog irq
> > at resume time so at to not immediately reboot the machine? I couldn't
> > see any logic to that effect in the driver.
> 
> That depends on what we want.
> If we want the watchdog to be inactive when entering suspend, then we
> shouldn't reboot the machine when receiving a watchdog irq while the
> system is suspended.
> ITOH, with the hardware mode (reset handled by the watchdog IP) you
> can't disable the watchdog when entering suspend, so I would expect the
> same behavior for the SW mode.
> 
> > 
> > Regardless, if the only reason we care about taking the interrupt during
> > the suspend/resume phases is due to the timer sharing the IRQ, then
> > shouldn't we be using IRQF_COND_SUSPEND?
> 
> I'm not sure, but IMO this interrupt should be flagged as NO_SUSPEND,
> because it's here to reset the system (even if it is suspended).
> If you flag the irq line as COND_SUSPEND, and atmel decide to give this
> peripheral its own IRQ line (on new SoCs), then your watchdog will not
> reboot the system when it is suspended.
> Another solution would be to support wakeup for this peripheral and
> delay the system reboot until it has resumed.
> 
> Anyway, if we decide to go for the wakeup approach, I'd prefer to post
> another patch on top of this one.
> 
> Best Regards,
> 
> Boris
> 
> 



-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com