Xen devel - Oct 2010 - New CPUs, now get: NETDEV WATCHDOG: eth0: transmit timed out

At Linode we recently altered our server build spec from Intel L5520 to 
L5630.  Nothing else has changed as far as we can tell, however with 
this new build we''re experiencing a new problem -- permanent loss of 
networking after some time (measured in days) on these new machines, with:

NETDEV WATCHDOG: eth0: transmit timed out <-- kiss of network death

The link is remains active on switch, yet the NIC stops receiving any 
interrupts.  No amount of prodding wakes it back up...

Some data points:

2.6.18.8 @ 931 contains an older igb driver
2.6.18.8 @ 1038 contains newest igb driver (as of last week)

2.6.18.8 @ 931 works perfectly on all our equipment prior to L5630
2.6.18.8 @ 1038 times out on everything

Motherboard BIOS version is the same.
Upgrading BIOS on affected boxes has no effect.

A year or two back (after 931), I had to build a newer 2.6.18.8 for 
whatever reason and decided to include the newest igb drivers at that 
time.  I eventually had to roll this back because the NICs started 
timing out.  However, even our "good" build is timing out on the new 
spec machine.

These machines don''t appear to present the problem when on bare metal.

dmesg:

http://theshore.net/~caker/xen/BUGS/nic-timeout/

What we''re trying:

1) On an affected machine, we''re swapping out the L5630 back to the
L5520.
2) Moving from Xen 3.4.1 to Xen 3.4.4-rc1-pre
3) Xen 3.4.4-rc1-pre along with 2.6.32.23-g41a85de5 dom0

This certainly appears as some strange incompatibility with Xen, dom0, 
and/or the NIC driver.  No more interrupts being delivered is suspicious.

I''d be grateful for any insight!

Thanks,
-Chris


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel