On Tue, Oct 29, 2019 at 02:45:59AM +0100, Ondřej Jirman wrote:
> On Mon, Oct 28, 2019 at 05:12:50PM -0700, Dmitry Torokhov wrote:
> > On Tue, Oct 29, 2019 at 12:56:26AM +0100, Ondřej Jirman wrote:
> > > On Mon, Oct 28, 2019 at 04:38:28PM -0700, Dmitry Torokhov wrote:
> > > > > +
> > > > > + error = dev_pm_set_wake_irq(dev, irq);
> > > > > + if (error) {
> > > > > + dev_err(dev, "Could not set wake IRQ\n");
> > > > > + return error;
> > > > > + }
> > > > > +
> > > >
> > > > I wonder if we could teach platform driver core to handle this for us.
> > >
> > > Not sure, some drivers do enable/disable wake_irq by hand in suspend/resume
> > > callbacks, so it would probably need to be opt-in somehow. I guess calling the
> > > function like this is one way to make it opt-in.
> > >
> > > The other way may be by passing a flag somewhere, like to
> > > request_threaded_irq. Did you have something more concrete in mind?
> >
> > I think it is perfectly fine to continue using enable_irq_wake and
> > disable_irq_wake from the driver while marking irq as being wake irq
> > form the core.
>
> I see, it looks like irq_set_irq_wake will track the calls via wake_depth
>
> https://elixir.bootlin.com/linux/latest/source/kernel/irq/manage.c#L714
>
> But all irqs are not necessarily wake irqs, no? So it still may need to be
> opt-in somehow.
I thought we'd do that for IRQ named "wakeirq" or the very first IRQ if
there is no named IRQ, and when we have the "wakeup-source" property,
similarly to what we do in I2C bus.
Thanks.
--
Dmitry
