>-----Original Message-----
>From: [email protected]
>[mailto:[email protected]]
>Sent: Friday, September 05, 2014 3:44 AM
>To: [email protected]
>Subject: [E1000-devel] ixgbe and pausing NICs
>
>
> Hello,
>
> I have been facing a problem with NICs going down
>temporarily and intermittently, and there's a possibilty the ixgbe driver
>plays a central role. The behaviour is depending on the type of switch at
>the other end. Looking for some expert advice here.
>
> In brief:
> - Intel 82599EB connected to Nexus 4900 switch works fine.
> - Intel X540-AT2 connected to Nexus 3064 switch works fine.
> - 82599EB connected to Nexus 3064, or X540-AT2 to Nexus
>4900, have intermittent interface down events lasting a few seconds.
>There may be two such events one day, none the next, and 50 the day
>after, just as an example. Traffic levels do not seem to play a role,
>but then, with mrtg, I may not have the required resolution.
>
>Sep  5 10:56:16 host kernel: ixgbe 0000:04:00.0: eth0: NIC Link is Down
>Sep  5 10:56:20 host kernel: ixgbe 0000:04:00.0: eth0: NIC Link is Up 10 Gbps, 
>Flow Control: RX/TX

Are these the only messages from the driver in dmesg?

> What could be the reason, and can I get debug info from the driver to help
> iwth understanding ? All servers run CentOS5/6 and the driver versions
> involved are 3.9.15-k, 3.17.3, and 3.18.7. While there are different server
> makes and models affected, this behavior definitely follows the NIC+switch
> combination. During the event, there is zero traffic through the interface
> (according to tcpdump), as you would expect with link down.

A link down event can be caused by removing the cable or a fluctuation in the 
signal between the link partners.
The driver does not have much control over it other than setting the PHY.

> My current working theory is that flow control settings may be the culprit.
> All servers have RX/TX on. The Nexus 4900 has RX/TX on, the 3064 off. I found
> that ethtool doesn't allow me to switch RX/TX off unless I also switch autoneg
> off, so I'm reduced to playing with the resp. switch settings. I take it that
> flow control (layer 2?) is not something that is auto-negotiated between
> NIC/driver and switch?

It is very unlikely that flow control has anything to do with it, but if you 
suspect it you can disable it from the driver's side:

#ethtool -A ethX autoneg off tx off rx off

You can see the result in the link up message as the one from your email.

Thanks,
Emil
 
>
> Any ideas?
>
> I only came across this list a few days ago. Neither Cisco nor Intel community
> forums have been any help so far.
>
>
>------------------------------------------------------------------------------
>Slashdot TV.
>Video for Nerds.  Stuff that matters.
>http://tv.slashdot.org/
>_______________________________________________
>E1000-devel mailing list
>[email protected]
>https://lists.sourceforge.net/lists/listinfo/e1000-devel
>To learn more about Intel® Ethernet, visit
>http://communities.intel.com/community/wired

------------------------------------------------------------------------------
Slashdot TV.  
Video for Nerds.  Stuff that matters.
http://tv.slashdot.org/
_______________________________________________
E1000-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel® Ethernet, visit 
http://communities.intel.com/community/wired

Reply via email to