Odd behaviour with 860P and LXT970

Odd behaviour with 860P and LXT970

[Gessner, Matt]

Greetings,

We're using an 860P's FEC with an LXT970 PHY.
We're seeing something odd and would like to know
if anyone has either seen the same behaviour or
(hopefully) knows how to fix it.

We're using 2.2, but when I compare it to the 2.4
version, there are really no significant differences
(outside of things like 'struct device' vs. 'struct net_device').
In fact, what we're running is the same thing, just modified
to fit into 2.2.

The behaviour is that when we are connected to a hub
that can auto-negotiate speed and duplex and we change
the link speed (unplug off 100Mb/half and plug into 10Mb/half)
the link does not work.  We get the interrupts and the status
reads correctly, but we can't use the interface any longer.

If we unplug it from 100Mb/half and plug it back into the same
port it works just fine; likewise for 10Mb.  It's the switching
rates that causes the problem.

We appreciate any ideas.

Regards,

Matt Gessner


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: Odd behaviour with 860P and LXT970

[Dan Malek]

"Gessner, Matt" wrote:

> ..... It's the switching
> rates that causes the problem.

Take an o'scope and look at the clocks from the PHY to the FEC.  The
PHY is responsible for providing the data clocks, and there is nothing
from "our" side that is clock speed related (except the MDIO clock, which
isn't the concern here).  The only thing we have to know in the driver is
full/half duplex and reconfigure the FEC if necessary.

It almost sounds like the PHY didn't negotiate or determine the clock speed
correctly.  I tend to take the simple approach and always let the PHY
autonegotiate.  That is, I don't tell it with commands to use a particular
link speed, just read what it found.  The PHY command interface in the FEC
driver seems to be growing more complex and forcing modes of operation, when
I never did that in the original driver.

That's about all I can think of at this hour.


	-- Dan

** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/