32-rc1 aka 32-rc2: warning at manage.c:361 (set_irq_wake), matrix-keypad related?

[Dmitry Torokhov]

On Wed, Oct 07, 2009 at 10:42:08PM +0800, Eric Miao wrote:
> On Wed, Oct 7, 2009 at 12:36 PM, Dmitry Torokhov
> <dmitry.torokhov at gmail.com> wrote:
> > On Tue, Oct 06, 2009 at 09:58:17AM +0200, Pavel Machek wrote:
> >> Hi!
> >>
> >> On Mon 2009-10-05 22:06:50, Dmitry Torokhov wrote:
> >> > On Wed, Sep 30, 2009 at 10:07:46PM +0200, Pavel Machek wrote:
> >> > >
> >> > > It complains about unbalanced irq 113 wake disable. That one belongs
> >> > > to matrix-keypad...
> >> >
> >> > I guess some of enable_irq_wake()s fail on your box. Do you see it if
> >> > you apply the patch below?
> >>
> >> I'll do short test. But... if you are right, your patch will just make
> >> your machine refuse to suspend... right?
> >>
> >> Ungood.
> >
> > It was just a debug patch.
> >
> >>
> >> Actually, it seems to break suspend (returns  -EINVAL while refusing
> >> to suspend), warnings are still there, and keyboard is dead after
> >> failed suspend... double plus ungood.
> >>
> >> Aha, so warning is solved: the one in the log is from gpio_buttons.
> >>
> >
> > OK, so it looks like your box refuses to set up one of the GPIOs as a
> > wakeup source... Hmm, either your box is wrong ;) or matrix_keypad
> > driver needs to maintain a separate list of wakeup GPIOs.
> >
> 
> This is due to the nature of PXA processor, where not every GPIO can
> be configured as a wakeup source. Mmm.... we can either return a
> pseudo value indicating setting wakeup on that GPIO is OK (which
> doesn't sound like a good idea), or we can just ignore the failure of
> enable_irq_wake() in matrix_keypad?

We ignore the failure right now in the mainline but that causes stack
traces on resume as we trying to disable not enabled wakeup GPIOs. That
was original Pavel's complaint.

-- 
Dmitry