Extending sys/dev/mii

[Warner Losh]

On Feb 11, 2012, at 9:49 PM, Adrian Chadd wrote:

> On 11 February 2012 17:15, Warner Losh <imp at bsdimp.com> wrote:
>> 
>> On Feb 11, 2012, at 6:00 PM, Adrian Chadd wrote:
>>> I'd like to try and finally bring some sanity to the hardcoded PHY
>>> mask handling in if_arge (and make it actually work for AR71xx and
>>> AR724x - where AR71xx has one shared MDIO bus between both MACs, but
>>> AR724x has two independent MDIO busses..)
>> 
>> FDT would do that...
> 
> .. how would it bring sanity to the device driver?

FDT would encode the links between the MAC and the PHY.  FDT is a directed graph, not a tree, so you can have multiple paths to a device node that follow different domains.

> Right now the driver assumes that both arge0 and arge1 mdiobus are the
> same and uses the phymask setting to determine how/when to access
> registers (ie, trying to read/write from phy registers not in the
> phymask of argeX are denied.) It is one of the annoying issues with
> the AR7241 internal switch support as that switch hangs off of arge1's
> MII bus.

It would no longer need to do this, or make any assumptions at all.  FDT would tell it all.

> It may make it easier to specify the configuration but it doesn't fix
> the fundamentally wrong assumption.

Nope.  Bad assumptions would still need to be fixed.

> The trouble in this whole mess (where FDT may help) is that phy's for
> arge0 may actually sit on arge1. So you can't probe/attach the miibus
> on arge0 until you've probe/attached arge1, so the arge1 MII registers
> can be tickled.

The PHYs don't sit on arge1.  They sit on another device that the driver bogusly assumes is tightly coupled to arge1, when in fact it isn't.

> Ray/Stefan/others: if anything, I'd like to try and bring sanity to
> this particular thorny issue in -HEAD before we worry about switch PHY
> devices.

Warner