[PATCH v2] mtd: dataflash: add device tree probe support

[Shawn Guo]

On Wed, Jul 20, 2011 at 08:32:45AM +0300, Artem Bityutskiy wrote:
> On Wed, 2011-07-20 at 13:28 +0800, Shawn Guo wrote:
> > On Wed, Jul 20, 2011 at 08:17:45AM +0300, Artem Bityutskiy wrote:
> > > On Wed, 2011-07-20 at 12:55 +0800, Shawn Guo wrote:
> > > > On Wed, Jul 20, 2011 at 07:40:38AM +0300, Artem Bityutskiy wrote:
> > > > > On Fri, 2011-07-15 at 12:49 +0800, Shawn Guo wrote:
> > > > > > > > +static const struct of_device_id dataflash_dt_ids[] = {
> > > > > > > > +	{ .compatible = "atmel,at45xxx", },
> > > > > > > > +	{ .compatible = "atmel,dataflash", },
> > > > > > > > +	{ /* sentinel */ }
> > > > > > > > +};
> > > > > > > > +
> > > > > > > 
> > > > > > > This should be protected with a #ifdef CONFIG_OF/#else/#endif, and
> > > > > > > there should be a MODULE_DEVICE_TABLE().
> > > > > > > 
> > > > > > I personally hate #ifdef stuff.  But okay, I can do it since there
> > > > > > are people being concerned by this little waste of space.
> > > > > 
> > > > > I guess the question is - will it compile and work if CONFIG_OF is
> > > > > unset?
> > > > > 
> > > > Yes, it will compile, as 'struct of_device_id' is defined in
> > > > include/linux/mod_devicetable.h unconditionally.
> > > 
> > > And it will work correctly even though dataflash_dt_ids is not NULL, it
> > > will not confuse MTD layer?
> > > 
> > I think for non-dt case, dataflash_dt_ids is not used anyway.  So yes,
> > it will not confuse MTD layer, at least from my testing.
> 
> And it is not error-prone? I mean, it is not very likely that someone
> changing MTD could make wrong assumptions and break dataflash driver? If
> so, then we probably do not need ifdefs indeed.
> 
No, it's not error-prone.  The dataflash_dt_ids assigned to
.of_match_table is used by OF device core to match driver and device.
I do not think MTD change could impact this match.

The argument about this #ifdef thing was the driver could be used by
some platforms that will never migrate to device tree.  This table
will be just a waste of bytes for those platforms, if we do not
protect the table by '#ifdef CONFIG_OF'.

-- 
Regards,
Shawn