On Tue, Dec 01, 2020 at 05:41:54PM +0100, H. Nikolaus Schaller wrote:
> > Am 01.12.2020 um 17:20 schrieb Mark Brown <broonie@kernel.org>:

> > I think the problem here is more to do with where we started than where
> > we're going or how we got there - things have been glued together or
> > happened to work in ways that mean I'm not sure we reasonably understand
> > the situation we started from or all the requirements it has.  As you
> > say I'm not sure anything beyond throwing the API away and starting
> > afresh would really help here, but that's not really how we tend to do
> > things for a bunch of very good reasons.

> I think the key problem is GPIO_ACTIVE_HIGH 0 and GPIO_ACTIVE_LOW 1
> in device tree blobs. But that is so fundamental that we have to live with it.
> So I guess that even a new API from scratch wouldn't improve that.

Yeah, that's definitely part of it - more generally there's multiple
places trying to determine if the signal is inverted with different
interactions/expectations.  Having it in an ABI definitely contributes a
lot to causing trouble though.