Re: [PATCH v2] input: cros_ec_keyb: Report wakeup events

[Dmitry Torokhov]

On Mon, Apr 03, 2017 at 01:53:53PM -0700, Brian Norris wrote:
> + others
> 
> On Mon, Apr 03, 2017 at 11:43:36AM -0700, Dmitry Torokhov wrote:
> > On Sun, Apr 02, 2017 at 08:07:39AM +0800, Jeffy Chen wrote:
> > > Report wakeup events when process events.
> > > 
> > > Signed-off-by: Jeffy Chen <jeffy.chen@xxxxxxxxxxxxxx>
> > > ---
> > > 
> > > Changes in v2:
> > > Remove unneeded dts changes.
> > > 
> > >  drivers/input/keyboard/cros_ec_keyb.c | 9 +++++++++
> > >  1 file changed, 9 insertions(+)
> > > 
> > > diff --git a/drivers/input/keyboard/cros_ec_keyb.c b/drivers/input/keyboard/cros_ec_keyb.c
> > > index 6a250d6..a93d55f 100644
> > > --- a/drivers/input/keyboard/cros_ec_keyb.c
> > > +++ b/drivers/input/keyboard/cros_ec_keyb.c
> > > @@ -286,6 +286,9 @@ static int cros_ec_keyb_work(struct notifier_block *nb,
> > >  		return NOTIFY_DONE;
> > >  	}
> > >  
> > > +	if (device_may_wakeup(ckdev->dev))
> > > +		pm_wakeup_event(ckdev->dev, 0);
> > > +
> > >  	return NOTIFY_OK;
> > >  }
> > >  
> > > @@ -632,6 +635,12 @@ static int cros_ec_keyb_probe(struct platform_device *pdev)
> > >  		return err;
> > >  	}
> > >  
> > > +	err = device_init_wakeup(dev, 1);
> > 
> > I would prefer if we did not mark cros_ec devices as wakeup sources
> > unconditionally. Your original patch series was better (except it failed
> > to parse the "wakeup-source" property that you introduced.
> 
> I'm curious, why is this keyboard device different than any other keyboard
> device? I see several other drivers in drivers/input/keyboard/ that do an
> unconditional 'device_init_wakeup(..., 1)'. Keyboards tend to be wakeup
> devices...

If we did something before it does not mean we should continue doing
this forever. I think providing an option to mark device as wakeup
capable should be left to the platform.

> 
> Also, what's the idea behind sub-devices vs. the main cros-ec device reporting
> wakeups? Right now, we have this in drivers/mfd/cros_ec.c:
> 
> static irqreturn_t ec_irq_thread(int irq, void *data)
> {
>         struct cros_ec_device *ec_dev = data;
>         int ret;
> 
>         if (device_may_wakeup(ec_dev->dev))
>                 pm_wakeup_event(ec_dev->dev, 0);
> 
>         ret = cros_ec_get_next_event(ec_dev);
>         if (ret > 0)
>                 blocking_notifier_call_chain(&ec_dev->event_notifier,
>                                              0, ec_dev);
>         return IRQ_HANDLED;
> }
> 
> But now, we're going to start double-reporting wakeups? Is that
> expected?

No, and not always (below).

> 
> I think we have a similar overlap with the RTC driver (which is being
> upstreamed now?):
> 
> https://lkml.org/lkml/2017/2/14/658
> [PATCH v3 3/4] rtc: cros-ec: add cros-ec-rtc driver.
> 
> except that also goes through the trouble of enabling/disabling wakeup for the
> EC IRQ. It seems to me (though I haven't dug in thoroughly) like the
> main MFD shouldn't really be doing the wakeup reporting at all, and we
> should depend on the sub-devices to do this. (i.e., the current patchset
> is a step in the right direction, but it's not 100%.)
> 
> Anyway, I could be wrong about the above, but I think we should make
> sure there's a consistent answer across the drivers tree.

Hm, it appears we have quite a mess. SPI-based EC declares entire EC as
wakeup source (unconditionally I must add; we do mention "wakeup-source"
in binding document at least). I2C-based EC does not call
device_init_wakeup() at all, presumably that is what caused the calls to
be added into sub-drivers.

We need to resolve this one way or another. You probably do not want to
wake up any time you move your device (accelerometer or other sensors),
so I would try to move this property into individual devices, and try to
come up with a reasonable binding.

Thanks.

-- 
Dmitry