----- Original Message ----- From: "Ben Greear" <greearb@xxxxxxxxxxxxxxx> >> I'm getting desperate to find a fix. I've been trying to work through >> this >> for a few weeks, and I've made very little progress. I can provide >> traffic >> dumps or any other information. Any help would be appreciated. I must be >> missing something simple. > > What 'driver' does Xen use for ethernet? It's likely that this driver > has some hard-coded constraints on packet size that does not take the > extra 4 byte VLAN header into account. The specific card in question is a sk-9e22 with the non-GPL sk98lin driver (v8.24.1.3 (01)) [as I understand it, the driver won't go into the kernel as there are two drivers in one]. I recently noticed a version v8.28 and will try that, but I couldn't find a changelog, so don't know if it will help. The NIC and the switch are all capable of an MTU of 9000. I'm using an MTU of 1500 to try and overcome any potential issue (as well as a potential problem that the Xen driver may only support a 4k frame). In the past I have tried an e1000 driver (As the machine also has a dual pro/1000 mt card in it), and not had any success. I'll try it again. > Or, it could be a checksum support issue or similar. I used ethtool to turn off checksum support, as well as TSO and scatter gatter. I didn't observe any change. > You probably should cross-post this with the bridging and/or xen mailing > lists...or just send to netdev@xxxxxxxxxxxxxxx I stuck a copy of the message in netdev over the weekend. One comment about possible MTU issues. > (I think this is probably more a Xen/Bridging problem than VLANs, though > it's possible there is a nasty interaction between all three.) I have made several posts in the xen group without finding a resolution. It seems that either people aren't using VLAN much with xen (perhaps because it's mainly developers running it), or perhaps it just works for them. I agree that it is likely to be an interaction between the drivers (or something silly I'm doing or not doing). What I have found is that if I remove the VLAN interfaces, and use physicals NICs I can easily get things working. It's strange that I have got ICMP pings up to 1460 through, but I can't get UDP 100byte DNS packets through. Greg :-)
