a recent xen kernel update causes dom0 to emit "Frag is bigger than frame." and then disconnect the network from the domu. this appears to be caused by tcp offload generating frames that are too big. there is a thread at:<br />
lists.xen.org/archives/html/xen-devel/2013-04/msg02590.html<br />
discussing it.<br />
<br />
dom0 emits logs like:<br />
vif14.0: Frag is bigger than frame.<br />
vif14.0: fatal error; disabling device<br />
<br />
i think i have seen some domu messages, but the last instance of this didn't cause any logs.<br />
<br />
a reboot of domu will hang during the ifdown phase (maybe trying to flush buffers?). doing a xm network-detach from dom0 will allow a clean shutdown, you can even attach a new vif via network-attach.<br />
<br />
supposedly, disabling offload functions will fix this. i have used:<br />
ethtool -K eth0 tx off tso off sg off<br />
<br />
which might be overkill, but seems to have fixed the issue.<br />
<br />
i have yet to be able to come up with a test case that will trigger this. i have only seen this on 3 out of 20 vms.<br />
<br />
dom0 and domu are running 2.6.18-348.6.1.el5xen, 64bit. everything up to date.<br />
<br />
i don't have an rhel system to test on, but i would guess this is an upstream issue.
↧