Re: [PATCHv6] mtd: gpio-nand: add device tree bindings

[Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>]

On Fri, Oct 14, 2011 at 03:21:49PM +0100, Jamie Iles wrote:
> Hi Artem,
> 
> On Fri, Oct 14, 2011 at 04:48:20PM +0300, Artem Bityutskiy wrote:
> > On Wed, 2011-10-12 at 00:10 +0100, Jamie Iles wrote:
> > > +#ifdef CONFIG_OF
> > > +static const struct of_device_id gpio_nand_id_table[] = {
> > > +	{ .compatible = "gpio-control-nand" },
> > > +	{}
> > > +};
> > > +MODULE_DEVICE_TABLE(of, gpio_nand_id_table);
> > ...
> > > +#else /* CONFIG_OF */
> > ...
> > > +#endif /* CONFIG_OF */
> > 
> > I wonder, why it is either OF of platform data? What if I want my kernel
> > to fall-back to platform data if device tree data is absent? What is the
> > general policy? Sorry, I am not very well aware of the DT stuff. But off
> 
> I think the general policy is that for device tree everything should be 
> in the device tree.  There is a mechanism for device tree platforms to 
> pass platform data too, but I believe this is more as a tool for 
> migrating existing platforms to device tree.
> 
> Also, the device tree binding should be well documented - if the device 
> is present in the tree it should have all of the required properties.  
> If the device isn't there at all then it won't get registered.
> 
> > the top of my head, it is logical when things go like this: I have a
> > kernel with working platform data, but I can change that dynamically by
> > feeding it a device tree configuration. Hmm?
> 
> I think in general platform data and device tree should be mutually 
> exclusive.

No, Artem is correct.  It is *not* okay to break platform_data support
simply by turning on CONFIG_OF.  The driver must be able to support
both, and to choose its data source at runtime.

Essentially, turning on CONFIG_OF adds the ability to boot with a
device tree while still being able to boot on legacy platforms.

g.