[ovs-discuss] could not create netdev enp2s0 of unknown type dpdk

Christian Ehrhardt christian.ehrhardt at canonical.com
Wed Nov 23 07:04:46 UTC 2016

On Tue, Nov 22, 2016 at 8:31 PM, Shivaram Mysore <shivaram.mysore at gmail.com>
> # more /etc/dpdk/interfaces
> pci 0000:04:00.1 uio_pci_generic

Hrm - it seems you cards need igb_uio - at least from your initial snipped
of dpdk_devbind --status.
There you had:
0000:02:00.0 'I211 Gigabit Network Connection' drv=igb_uio unused=igb

I used several ixgbe cards via uio_pci_generic and never needed igb_uio,
but for whatever reason yours don't list that as potential driver.
Chime in if anybody has ideas on that please.
So assuming your card is special and you need igb_uio, then you'd also want
to list igb_uio in /etc/dpdk/interfaces so they are bound to that driver

> Should this really be commented out and just use the following commands?
> # modprobe uio
> # insmod /usr/local/src/dpdk-stable-16.07.1/build/kmod/igb_uio.ko

You have a self build igb_uio kernel module here.
Out-of-tree modules are bad anyway, but I don't want to go down that route
with the discussion.
But you see just the tip of the iceberg with the module signing.
Of course you could sign it following some guides, but then what on kernel
updates, ...

> what is the correct procedure?

The "correct" - if in the scope of out-of-tree modules you want to call
anytihng "correct" is DKMS.
Starting with Ubuntu 16.10 we deliver DPDK 16.07 including DKMS based
modules for igb_uio and kni
You have the benefit of:
- those versions are matching your other parts of dpdk brought by the
- they are updated when DPDK gets updated
- the automatically (try to) build for your installed kernels and are
refreshed on updates
=> https://help.ubuntu.com/community/DKMS
=> http://packages.ubuntu.com/yakkety/dpdk-igb-uio-dkms

Christian Ehrhardt
Software Engineer, Ubuntu Server
Canonical Ltd
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20161123/8be7555f/attachment.html>

More information about the discuss mailing list