Closed xfxian closed 8 years ago
Hi, are you using multiple queues on ixgbe?
There is some problem with it. (I'll try to solve it on next week) Can you set single queue through ethtool?
Cheers, Stefano
Hi Stefano,
it seems it was defaulting to 8 queues, so I tried to set it to one:
root@host:~# ethtool -L net_d combined 1
root@host:~# ip link set dev eth0 down
root@host:~# ip link set dev eth0 up
root@host:~# ./vale-ctl -g eth0
bdg_ctl [165] eth0: 1 queues.
root@host:~# ./vale-ctl -a vale0:eth0
Here it stacktraces again. Could it somehow be related to IOMMU settings? Right now I am booting with "intel_iommu=on", do I also need to set "iommu=pt" or some other value? Thank you in advance.
Best regards, Jacob
Hi,
Are you using the version d4bf89c of netmap? (as pointed by the ptnetmap repo) I tested it and indeed it causes a kernel panic on the host when you try to attach a physical NIC to VALE.
The problem seems to be already fixed in most recent versions of netmap. I tried with 579b43 and it worked fine.
After that, in the guest, you can use either an updated version of netmap or the version pointed by the ptnetmap repo. At least for me, both versions worked fine when used in the guest with both e1000 and virtio drivers, the issue is only when using VALE + a physical NIC.
Cheers, Jose
Hi, I'm trying to connect my VM's netmap-aware service through the physical 10GB interface on the host using ptnetmap.
I'm running Ubuntu 14.04.3 with 3.13.0-66-generic. For higher Kernel versions of this release, the ptnetmap compilation fails (3.16 and 3.19).
I can reproduce the attached stacktrace by loading ptnetmap and attaching the interface like this:
This is where the system crashes. It also happens if I start qemu with a -netdev netmap,ifname=netmap:eth0, thus directly connecting to the netmap interface, when I
ip link set dev eth0 up
in the guest then.When I am using vanilla netmap, I can do all the previous steps successfully and send packages to the network with pkt-gen.
Stacktrace: