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

[Boris Brezillon <boris.brezillon@free-electrons.com>]

Rafael, Alexandre,

On Wed, 11 Mar 2015 02:03:08 +0100
"Rafael J. Wysocki" <rjw@rjwysocki.net> wrote:

> On Tuesday, March 10, 2015 11:33:05 PM Alexandre Belloni wrote:
> > On 10/03/2015 at 23:31:52 +0100, Rafael J. Wysocki wrote :
> > > On Tuesday, March 10, 2015 10:33:17 PM Alexandre Belloni wrote:
> > > > Hi,
> > > > 
> > > > On 09/03/2015 at 15:30:01 +0100, Rafael J. Wysocki wrote :
> > > > > > > > Actaully, your platform should just refuse to enter suspend-to-RAM
> > > > > > > > when hw watchdog is enabled.
> > > > > > > 
> > > > > > > Quite likely, depending on how exactly the suspend is implemented.
> > > > > > >
> > > > > > 
> > > > > > We've had absolutely zero complain on that. It is quite clear in the
> > > > > > datasheet that failing to refresh the watchdog once started will lead to
> > > > > > a reset and that it is impossible to stop.
> > > > > > It is actually quite convenient to also ensure that you can actually
> > > > > > wake up from suspend because that can obviously go wrong.
> > > > > 
> > > > > I gather then that the suspend implementation is such that touching the
> > > > > watchdog periodically while suspended is not a problem.
> > > > > 
> > > > > Again, can you please tell me how suspend is implemented on at91?
> > > > > 
> > > > 
> > > > It actually depends on the architecture (at91rm9200, at91sam9 or sama5)
> > > > but basically, the clocks are switched off in almost all the peripheral
> > > > drivers then the ram self refresh activated, the master clock is
> > > > switched off using code running from SRAM and the core is then waiting
> > > > for interrupt.
> > > 
> > > OK, so it looks like enable_irq_wake() doesn't actually affect the hardware
> > > on those platforms, is that correct?
> > > 
> > 
> > I didn't exactly look in details but apart from the wakeup from gpio
> > handling (keeping the pio controller clocked in the case one of its gpio
> > has wakeup enabled), I don't think it does much more. It uses
> > irq_gc_set_wake().
> 
> Well, that only modifies gc->wake_active, so no hardware interactions.

I'm not sure I understand the whole discussion, but calling
enable_irq_wake() does affect suspend behavior on at91 platforms.
Take a look at the suspend() implementation [1], it's making use of the
wake_active field (modified by irq_gc_set_wake) when entering suspend
in order to keep wakeup IRQ sources enabled. 

Best Regards,

Boris


[1]http://lxr.free-electrons.com/source/drivers/irqchip/irq-atmel-aic.c#L106

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