Re: [RFC] micrel KSZ8041 disable auto negotiation with fiber

["Aníbal Almeida Pinto" <anibal.pinto@efacec.com>]

Em Ter 19 Jun 2012 18:08:29 WEST, Ben Hutchings escreveu:
> On Tue, 2012-06-19 at 15:45 +0100, Aníbal Almeida Pinto wrote:
>> Hi,
>>
>> I am working on a custom board based on a OMAP L138 with a KSZ8041TL-FTL
>> phy that have fibber and copper support.
>>
>> When using with fibber the board can't connect to a switch, only after
>> exec :
>>
>> 	ethtool -s eth1 speed 100 duplex full autoneg off
>>
>> the phy get link and start working.
>>
>> On some switches it appears that autoneg don't work well with fibber,
>> ethtool reports that its at 10MB/s after exec
>>
>> 	ethtool -s eth1 autoneg on
>
> I'm not aware of any specification for autoneg over fibre, so I imagine
> the PHY is being programmed with an invalid configuration and sending
> weird signals.
>
>> The auto negotiation bit on phy register is disable on start when using
>> the fibber but Linux don't appear to look at it.
>>
>> Found a thread [1] that solve the problem but don't appear to be fully
>> accepted.
>>
>> The problem is modifying the phy code without interfere with ethtool use.
>>
>> Any official/accepted solution to this problem ?
>
> The net driver or PHY driver needs to disable autoneg by default when
> connected to fibre, and reject attempts to turn it on.
>
> As for *how* the driver is supposed to know whether that's the case, I
> have no idea...

On this chip you have a pin where you can specify if is fibber or 
copper, at power up it cleat/set the bit of auto negotiation.

The driver can save the bit state after phy power up or reset, and with 
that validate if is copper or fibber.

In the case of fibber shouldn't allow change the auto negotiation.

>
>> [1] - http://marc.info/?l=linuxppc-embedded&m=131107263711714
>
> ...which seems to be where the previous discussion ended.
>
> Ben.
>