[CentOS] tg3 BCM5755 intermittantly stops working after upgrade to 5.3.

Wed Apr 22 09:18:32 UTC 2009
Mangesh S. Umbarje <mangesh at gmrt.ncra.tifr.res.in>

            The machine has a static IP. If I try to change speed or 
autoneg or both the network completely stops. It starts working only after 
I issue service network restart.

Regards,
Mangesh

On Tue, 21 Apr 2009, Michael Iatrou wrote:

> When the date was Tuesday 21 April 2009, Mangesh S. Umbarje wrote:
>
>>            I have checked the physical connectivity which is perfectly
>> fine. This machine is very critical which we need to keep running as much
>> as possible. So I had gone to older kernel 2.6.18-92.1.17.el5. But along
>> with this kernel, I had added one more Dlink ethernet card which shows to
>> be
>>
>> 05:09.0 Ethernet controller: D-Link System Inc DGE-530T Gigabit Ethernet
>> Adapter (rev 11) (rev 11)
>>
>> and it uses the skge driver.
>>
>>            So I could test the machine with this new card and
>> kernel-2.6.18-92.1.17.el5 togather. Still I do get the network breaks but
>> the frequency reduced to factor of Ten. The corresponding messages seen
>> in the /var/log/messages are as follow.
>>
>> Apr 21 10:18:32 kernel: skge eth1: Link is down.
>> Apr 21 10:18:36 kernel: skge eth1: Link is up at 1000 Mbps, full duplex,
>> flow control none Apr 21 10:18:55 kernel: skge eth1: Link is down.
>> Apr 21 10:18:57 kernel: skge eth1: Link is up at 1000 Mbps, full duplex,
>> flow control none Apr 21 10:18:58 kernel: skge eth1: Link is down.
>> Apr 21 10:19:01 kernel: skge eth1: Link is up at 1000 Mbps, full duplex,
>> flow control none
>
> Try to disable auto-negotiation for the NIC, using ethtool. Set manually the
> speed to 100 and then 1000 Mbps and see what happens.
>
> The machine has static IP or it uses DHCP? It's a good idea to do some
> traffic capture and check if there is any correlation between the traffic
> and the link state.
>
>