async userspace gpio api [WAS: Re: PXA board chips control]

[Ben Nizette <squidgit@iinet.net.au>]

[bringing lkml in on this]

On Wed, 2008-05-28 at 00:37 -0700, David Brownell wrote:
> No, very different.  Those are more like device nodes, which
> get added and removed as part of routine operations (like
> physical connect/disconnect).
> 
> 
> > I would have thought -EINVAL was sufficient, no?
> 
> Surprising to see a read return a value at one moment, then
> fail with EINVAL the next, then return a value later ...
> no, there should be some distinguished value for "that mode
> would be meaningless right now".

Actually my thought (though I didn't come close to expressing it just
now) was to have an async behaviour control file (taking irq/poll/none
options) which wouldn't be writable while the gpio was an output.

The more I think about it though, the more I'm inclined to just allow
the user to select any of those options on any gpio regardless of
direction and trust them to know whether that makes sense for the
underlying hardware (sanity checking as needed).

	--Ben.