About mdio_bus for 82xx based board

[Scott Wood]

On Wed, Jul 25, 2007 at 05:22:40PM +0300, Alexandros Kostopoulos wrote:
> 1) When mdiobus_register() called from mii-bitbang.c (fs_enet_mdio_probe()  
> function) attemps to do a device_register for the mdio bus, it actually  
> registers the device with a bus_id in the form [0|1|...]:<phy_addr>, that  
> is the first part a simple integer. This, of course, happens because  
> fs_enet_of_init() (fsl_soc.c) does a  
> platform_device_register_simple("fsl-bb-mdio", i,  NULL, 0); with i being  
> the first part of the bus, starting from 0. Unfortunately, when  
> fs_init_phy() (fs_enet_main.c) calls phy_connect() and therefore  
> phy_attach() (phy_device.c), the latter attempts to find the device in the  
> mdio bus, but it searches using the bus_id registered in the net_device  
> struct, which is in the form of <resource address>:<phy_addr>, eg.  
> f0000000:0, and therefore it fails... I don't know if I am doing something  
> wrong here, so any hint would be greatly appreciated.

The code is broken.  Try applying the set of 61 patches I posted a week
or so ago (you'll need to apply them to Paul's tree from around the same
time, not the current tree).

> 2) Since there are two ethernet@<address> nodes in my device tree,  
> fs_of_enet_init() calls platform_device_register_simple("fsl-bb-mdio",...)  
> twice, therefore creating two mdio busses, 0 and 1, each having the same  
> two devices. For example, if I have two PHYs with addresses 1 and 5, I  
> will get two mdio busses and 4 devices, 0:1, 0:5, 1:1 and 1:5. Well, this  
> doesn't sound right to me, although I am not sure if this is a fatal  
> issue. Any comments?

It's not right -- my patchset gets rid of all of this mess.

> 3) Also, if I don't want to enter the phy interrupt in the device tree  
> (there is not one or I want to use PHY_POLL), what should I do? dtc seems  
> to not allow -1 as a value in the reg property.

Leave the property out altogether.

-Scott