Re: [PATCH 3/3] net: smsc911x: add u16 workaround for pxa platforms

[Rob Herring]

On Mon, Oct 03, 2016 at 05:42:29PM +0100, Mark Rutland wrote:
> On Mon, Oct 03, 2016 at 05:09:13PM +0100, Russell King - ARM Linux wrote:
> > Please note that the binding doc for smsc,lan91c111.txt is slightly wrong
> > on two counts:
> > 
> > 1) compatible property:
> > 
> > compatible = "smsc,lan91c111";
> > 
> > vs the code:
> > 
> > static const struct of_device_id smc91x_match[] = {
> >         { .compatible = "smsc,lan91c94", },
> >         { .compatible = "smsc,lan91c111", },
> >         {},
> > };
> > MODULE_DEVICE_TABLE(of, smc91x_match);
> > 
> > So the binding document needs to mention that smsc,lan91c94 is a valid
> > compatible for this device.
> 
> Yes, it should.
> 
> > 2) reg-io-width property:
> > 
> > - reg-io-width : Mask of sizes (in bytes) of the IO accesses that
> >   are supported on the device.  Valid value for SMSC LAN91c111 are
> >   1, 2 or 4.  If it's omitted or invalid, the size would be 2 meaning
> >   16-bit access only.
>  
> > Moreover, look at the property name vs the binding description.  It's
> > property name says it's a width, but the description says it's a mask
> > of sizes - these really aren't the same thing.  Once you start
> > specifying these other legal masks, it makes a nonsense of the "width"
> > part of the name.  It's too late to try and fix this now though.
> 
> Indeed, as-is this is nonsense. :(
> 
> The best we can do here is to add a big fat notice regarding the
> misnaming; adding a new property is only giong to cause more confusion.

Just fix the text here removing the mask part. This is a common property 
and not a mask. The rest saying 1, 2, 4 being valid is correct. There 
are no occurences using this as a mask in kernel dts files either.

> 
> > The binding document really needs to get fixed - I'll try to cook up a
> > patch during this week to correct these points, but it probably needs
> > coordination if others are going to be changing this as well.
> 
> Thanks for handling both of these.
> 
> Given the historical rate of change of the binding document, I suspect
> the stuff for pxa platforms is going to be the only potential conflict.
> 
> Either all of that can go via the DT tree (independent of any new code),
> or we can ack the whole lot and it can all go via the net tree in one
> go.
> 
> Thanks,
> Mark.