Re: [gentoo-amd64] Dual gigabit Intel NICs problem.

[Juan Fco. Giordana]

Thank you both for you answers.

Bob Sanders wrote:
> It doesn't matter what the LED is saying there is no link detected.  It
> seems the phy on eth1 is good but the path from the phy to the
> ethernet chip is bad.

Bob, what is strange is that regardless the LED status, ethtool says 
that on server1 the link has not been detected and on server2 it says 
the contrary.

eth1 on server1 -> Link detected: no
eth1 on server2 -> Link detected: yes

Also other noticeable differences:

eth1 on server1 ->
	Supported ports: [ TP ]
	Port: Twisted Pair

eth1 on server2 ->
	Supported ports: [ TP MII ]
	Port: MII

I'm not sure what you mean with the PHY on the ethernet device so I'm 
taking a look to the Ethernet howto to get a clue.

Also, just to try something on server2:

# ethtool -s eth1 speed 100 duplex full port tp
Cannot set new settings: Invalid argument
   not setting speed
   not setting duplex
   not setting port

> Server two is working, only that should be running the EEPro100 driver,
> not the e1000e driver.

Sorry for not mentioning this. I was taking this server out of the 
equation since it was reporting that everything was working.

The driver this server was using was e100. I've just recompiled the 
kernel and switched to the eepro100 driver. I didn't know that two 
drivers for the same adapter could be found on the same kernel.

If I understood well the eepro100 is the open source version of this 
driver and e100 is the Intel's version.

> btw - have you had the colo center AC power cycle these systems?
> Especially, the first one?  And I mean completely remove the AC.
>
> In some cases, there are areas in the ethernet chip's nvram that
> only get initialized when all power is removed.  A system reset,
> and even a reset/setup by the bios won't clear some locations.
>
> The updated driver prevents almost all of these issues from
> occuring.  But they can still, on a few systems, occur.

Not sure about this. We switched some memory modules recently but I'm 
not sure if they completely removed the AC cord.

Also, this could be related since in our first attempt to add more 
memory to this server the company that sold it to us sent incompatible 
modules and we have a very weird situation. They added the new modules, 
found out they were not compatible, plugged back the old ones and when 
they turned the server on, it hanged on the "Loading initrd ..." stage.

It was a nightmare. I was using the livecd kernel and that night I had 
to compile a new one to try "something" to make it work again.

Finally I got it working and a couple weeks later we upǵraded the memory 
smoothly.

I'll see if we can turn both server off tonight.

Thank you again for your help.