[PATCH 1/5] PowerPC 74xx: Katana Qp device tree

[Dale Farnsworth]

On Sun, Dec 16, 2007 at 05:40:56PM +1100, David Gibson wrote:
> On Mon, Dec 10, 2007 at 02:18:16PM -0700, Dale Farnsworth wrote:
> > David Gibson wrote:
> > > On Thu, Nov 29, 2007 at 06:28:36PM +0300, Andrei Dolnikov wrote:
> > > > Device tree source file for the Emerson Katana Qp board
> > 
> > [snip]
> > 
> > > > +	mv64x60 at f8100000 { /* Marvell Discovery */
> > > > +		#address-cells = <1>;
> > > > +		#size-cells = <1>;
> > > > +		model = "mv64460";			/* Default */
> > > > +		compatible = "marvell,mv64x60";
> > 
> > [snip]
> > 
> > > > +		mdio {
> > > 
> > > There must be some way of actuall accessing the mdio bus, so this node
> > > ought to have a 'reg' property and unit address.
> > 
> > There is no way for the cpu to directly access the mdio bus.  The
> > mdio bus is internally accessed by the ethernet MAC.  That being the
> > case, maybe it makes more sense to move the mdio node inside of the
> > multiethernet node, as follows, but I don't see how we can give it
> > a reg property or a unit address.
> 
> Ah, I see.  If the mdio interface isn't distinct from the other
> pieces, then it probably shouldn't get a device node at all.  Having
> an explicit mdio bus with the phys hanging off it is convenient for
> hardware which actually works that way, but it doesn't have to be done
> like that.
> 
> Of course, then the question is where to hang the phy nodes, which
> look like they have information you need.  Since you already have a
> local addressing scheme for the MACs under the multiethernet, what
> probably makes sense would be to hang the phys directly under the
> multiethernet, using an encoding scheme for the reg properties so that
> the MACs and PHYs aren't confused (say, MACs are 0x0, 0x1, 0x2, PHYs
> are 0x80000000, 0x80000001, 0x80000002).

Ugh.  They really are two separate address spaces.  One is an
intra-register index, and the other really is an mdio address,
even though it's only indirectly addressable.  Combining them would
obfuscate.  I'm proceding with an mdio node under the multiethernet
node as I posted below.  Let me know if you feel strongly that that
should be changed.

> Incidentally, although I suggested it, I'm not all that fond of the
> "multiethernet" name, it was just the first thing that occurred to me.

I'm not fond of it either.  Sometimes, naming is hard. :)
I'll see if we can come up with something better.

Thanks,
-Dale

> > 
> > 		multiethernet at 2000 {
> > 			reg = <0x2000 0x2000>;
> > 			ethernet at 0 {
> > 				device_type = "network";
> > 				compatible = "marvell,mv64360-eth";
> > 				reg = <0>;
> > 				interrupts = <32>;
> > 				interrupt-parent = <&PIC>;
> > 				phy = <&PHY0>;
> > 				local-mac-address = [ 00 00 00 00 00 00 ];
> > 			};
> > 			ethernet at 1 {
> > 				device_type = "network";
> > 				compatible = "marvell,mv64360-eth";
> > 				reg = <1>;
> > 				interrupts = <33>;
> > 				interrupt-parent = <&PIC>;
> > 				phy = <&PHY1>;
> > 				local-mac-address = [ 00 00 00 00 00 00 ];
> > 			};
> > 			mdio {
> > 				#address-cells = <1>;
> > 				#size-cells = <0>;
> > 				device_type = "mdio";
> > 				compatible = "marvell,mv64360-mdio";
> > 				PHY0: ethernet-phy at 1 {
> > 					device_type = "ethernet-phy";
> > 					compatible = "broadcom,bcm5421";
> > 					interrupts = <76>;	/* GPP 12 */
> > 					interrupt-parent = <&PIC>;
> > 					reg = <1>;
> > 				};
> > 				PHY1: ethernet-phy at 3 {
> > 					device_type = "ethernet-phy";
> > 					compatible = "broadcom,bcm5421";
> > 					interrupts = <76>;	/* GPP 12 */
> > 					interrupt-parent = <&PIC>;
> > 					reg = <3>;
> > 				};
> > 			};
> > 		};