On Wed, Jul 08, 2015 at 10:27:10PM +0300, Laurent Pinchart wrote:
> On Wednesday 08 July 2015 19:06:37 Mark Brown wrote:

> > I think this *might* be something that the efforts to reduce the amount
> > of out of tree embedded code can help with?  A part of that effort is
> > identifying areas that need fixing, there will be a large enough list I
> > imagine.  The hard part is always matching people up with things that
> > they are interested in though.

> There's also the issue of documentation. Datasheets for embedded SoCs are 
> often not publicly available. Even when out-of-tree code exists, the lack of 
> documentation makes it more difficult to understand it and port it to 
> mainline. I've seen many such cases in practice for camera sensor drivers that 
> are often provided by vendors with a small set (or even a single) hardcoded 
> configurations, using large register address/value tables. Porting such a 
> driver to the mainline APIs often require computing register values from 
> parameters received through the subsystem API, which requires a good 
> understanding of the hardware.

Yeah, if you're doing device support.  The generic subsystem bits are
less impacted by these issues, though you can't completely get away from
it.  Fortunately I don't think we've got a great shortage of problems
that need addressing so there should be some things that are tractable!