On Fri, Apr 08, 2016 at 09:38:25PM +0200, Mark Kettenis wrote:
> > Date: Fri, 8 Apr 2016 20:26:14 +0200
> > From: Patrick Wildt <patr...@blueri.se>
> >
> > +void
> > +mainbus_iterate(struct device *self, struct device *match, int node)
> > +{
> > + for (;
> > + node != 0;
> > + node = OF_peer(node))
> > + {
> > + /* skip nodes that are already handled by some driver */
> > + if (!mainbus_is_attached(node))
> > + mainbus_attach_node(self, match, node);
> > +
> > + mainbus_iterate(self, match, OF_child(node));
> > + }
> > +}
>
> Sorry, but this really is wrong. You're completely ignoring the tree
> aspect if the FDT. That is not going to work for more complex designs
> with i2c busses or pci busses. And it is going to be essential if you
> want to support suspend/resume.
Fully agreed. I am aware of the limitations of the approach taken in
this diff and also agree that this does not scale well.
>
> The autoconf framework is really quite clever and flexible. It is
> possible to re-use glue code quite easily. As files.conf(5)
> documents, device don't directly attach to parents, but to attributes.
> You can re-use those attributes, and parents can define multiple
> attributes.
>
> You could for example define a generic fdt attribute. Simple devices
> can then attach directly to that attribute. And you could also have
> some soc-specific attributes, that you then use to attach devices that
> need some additional glue.
>
> It'll take a bit of time to figure out the best way to define your
> attributes. But the end result will be worth it. And don't be afraid
> to make mistakes. As long as you drivers use the same "attach_args"
> structure, we can move around attributes without the need to change
> the code. So it will be easy to adjust the design as we learn more.
Thank you for the input. I will try to gather more information about
the autoconf framework with those attributes in mind. I will come up
with something better once I have understood that and came up with a
better design and implementation.
Thanks,
Patrick
>
> Cheers,
>
> Mark
>
