On Thu, May 16, 2013 at 09:24:05AM +0300, Michael S. Tsirkin wrote: > On Wed, May 15, 2013 at 12:00:03PM +0100, Nicholas Thomas wrote: > > Hi again, > > > > On Tue, 2013-05-14 at 15:49 +0100, Nicholas Thomas wrote: > > > /sys/devices/virtual/net/t100/tun_flags is 0x5002 - so it looks like > > > IFF_ONE_QUEUE was indeed unset by qemu (which is lacking the patch). It > > > surprises me, but that's probably my fault, rather than qemu's. > > > > > > I've rebuilt 1.4.1 with the IFF_ONE_QUEUE patch and tun_flags is now > > 0x7002; unfortunately, I'm still seeing this bug, twice in five trials. > > Symptoms in `ifconfig t100` now differ; overruns stays at 0, and > > "dropped" increases monotonically as I send packets. Those packets do > > appear if I tcpdump t100 on the host, but not if I tcpdump t100 on the > > guest. > > > > I've turned off gro in the guest, which makes no difference, and tried > > changing the queue sizes (post-hoc) in both guest and host, in the hope > > of causing them to be emptied out, clearing the condition; again to no > > effect. > > > > The VMs in question are bridged to a large (and busy) VLAN with no > > ingress filtering to speak of; I guess what's happening is that the > > transmit queue is filled up by that traffic while the guest is in ipxe, > > and it never gets out of that state when it happens... so maybe there is > > still an underlying problem? > > > > /Nick > > > Is this with or without vhost-net in host?
never mind, I see it's without. Try to enable vhost-net (you'll have to switch to -netdev syntax for that to work) and see if this help. If it does it's likely a qemu bug if not probably a guest bug. > -- > MST