Re: [Xen-users] Potentially old bridging or xennet (netfront) bug, does this sound familiar?

[Ian Campbell <ian.campbell@xxxxxxxxxx>]

On Fri, 2016-01-08 at 10:47 -0700, Andrew Davidoff wrote:
> Hi,
> 
> I'm currently troubleshooting an issue with a CentOS 5.7 guest with
> kernel 2.6.18-308.4.1.el5xen running on a CentOS 6.6 dom0 with
> Xen4Centos 4.2.3-25 and kernel 3.10.20-11.el6.centos.alt.x86_64. After
> some amount of time (varies) the guest stops communicating on the
> network, and the symptoms are almost exactly as described here:
> 
> http://xen.1045712.n5.nabble.com/domU-loses-network-after-a-while-td32651
> 72.html
> 
> By "almost" I mean that the only behavior that hasn't been confirmed
> to be similar is seeing tx drops on the vif when things are broken. I
> hope to confirm this the next time this happens. I have not yet found
> a way to programmatically trigger the issue but unfortunately it
> happens somewhat frequently.
> 
> I realize the bug reported above doesn't apply here because the
> smartpoll feature that is responsible isn't part of the
> 2.6.18-308.4.1.el5xen kernel's xennet module, but because every other
> behavior reported in the bug details are exactly the same as what I'm
> experiencing, I'm lead to believe I might be hitting a different
> xennet domU bug.

It's not completely out of the question that it might be the same bug,
various things have been backported over the years, and in the time period
in question many of the things appearing in e.g. 2.6.32-*xen* kernels was
being forwarded ported from the old 2.6.18-*xen* fork to mainline.

So it isn't out of the question that, one way or another, this bug was in
both 2.6.18-308.blah and 2.6.32-5-xen-blah. But both are so long ago I
doubt anyone would be able to say for sure though.


> I have done a fairly exhaustive search for similar bugs and as part of
> that effort I wanted to see if this sounds familiar to anyone.
> 
> As an aside, I realize that running a newer guest is likely advisable,
> but I'm hoping to identify a root cause here before I make a change
> like that.

If upgrading the guest is too scary (or a problem for some other reason)
then maybe just updating the kernel to the latest 5.X (5.11 from the looks
of it) might be sufficient?

Having written all that -- isn't the smartpoll issue described at that link
a _backend_ one, i.e. it is the CenOS 6.6 3.10 kernel which would be at
fault (and that bug was surely fixed by then), so maybe that's all red-
herring.

Ian.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users