f20 netfront failures

[Bill McGonigle]

I've hit this bug a few times on f20 with a fairly boring config:

http://lists.xen.org/archives/html/xen-devel/2014-12/msg01045.html

currently:

xen-4.3.3-5.fc20.x86_64
  3.17.3-200.fc20.x86_64

One f20 DomU (16GB), one el6 DomU (2GB), i7 desktop w/ 32GB RAM (12GB
Dom0).  The only thing not completely bog-standard is that my 'physical'
interface is on a vlan tag on each bridge, but Dom0 network continues to
hum along at the time, so I don't suppose that's a factor.

Symptom is the DomU network just goes away.  It appears to be load
related (I was doing video encoding over NFS).  dmesg says:

[76819.472975] vif vif-2-0 vif2.0: txreq.offset: 8ee, size: 3858, end: 6144
[76819.473012] vif vif-2-0 vif2.0: fatal error; disabling device
[76819.482474] brbfc: port 2(vif2.0) entered disabled state

A workaround is to xl save the domU to a checkpoint file (have to use -c
and destroy it), then restore it, and things continue happily.  I wasn't
able to figure out a way to tell Xen to just restart the network device
(it appears to be attached and up after Xen decides it's failed).

I'll be applying the 3-line kernel patch here; do we stand any chance of
getting something like this cherry picked into the Fedora kernel?  It's
not upstream as of 3.18:

http://lxr.free-electrons.com/source/drivers/net/xen-netfront.c#L628

I can advocate on xen-devel if needed.

-Bill

f20 netfront failures [SOLVED]

[Bill McGonigle]

On 12/11/14 13:28, Bill McGonigle wrote:
> [76819.472975] vif vif-2-0 vif2.0: txreq.offset: 8ee, size: 3858, end: 6144
> [76819.473012] vif vif-2-0 vif2.0: fatal error; disabling device
> [76819.482474] brbfc: port 2(vif2.0) entered disabled state

Followup: a few days later, this patch hit lkml:

https://lkml.org/lkml/2014/12/14/147

The Fedora 3.17.8 kernel definitely has it (in patch-3.17.8.xz).

The problem is in the DomU kernel, so that's where the updates need to
happen.  3.17.3 was really giving me headaches lately as a DomU - to the
point that I'd have to reboot and hope I could 'xl console' and yum
could download the update before the port got disabled.  I don't know
why the problem go so much more severe this week - the domU and dom0
kernels haven't been updated for a few weeks.  I rebooted to deal with
some bum memory, so something apparently is different other than the
kernels.

Anyway, after updating the DomU kernel to 3.17.8 the networking has been
stable so far.

-Bill