[dpdk-dev] Why the offloads of the guest's virtio-net network adapter are disabled when vhost-user is used?

[leo zhu]

Hi Tetsuya & Jijiang,
Great!  Thanks a lot for your explanations.

Thanks.
Leo

On Tue, Aug 25, 2015 at 11:27 AM, Liu, Jijiang <jijiang.liu at intel.com>
wrote:

>
>
> > -----Original Message-----
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of Tetsuya Mukawa
> > Sent: Tuesday, August 25, 2015 11:23 AM
> > To: leo zhu; dev at dpdk.org
> > Subject: Re: [dpdk-dev] Why the offloads of the guest's virtio-net
> network
> > adapter are disabled when vhost-user is used?
> >
> > On 2015/08/24 22:09, leo zhu wrote:
> > > Hi all,
> > >
> > > I am running the vhost sample application on my server.
> > >
> > > According to the dpdk-sample-applications-user-guide.pdf, I run the
> > > Virtual Machine with vhost-user enabled.
> > > Following is the command that is used to run the virtual machine.
> > >
> > >
> > >
> > >
> > >
> > >
> > > *qemu-system-x86_64 /root/leo/ubuntu-1.img -enable-kvm -m 1024 -
> > vnc :5
> > > -chardev
> > > \socket,id=char1,path=/root/leo/dpdk-2.0.0/examples/vhost/usvhost
> > > -netdev type=vhost-user, \id=mynet1,chardev=char1,vhostforce -device
> > > virtio-net-pci,mac=00:00:00:00:00:01,netdev=mynet1 \-object
> > > memory-backend-file,id=mem,size=1024M,mem-
> > path=/dev/hugepages,share=on
> > > -numa node,memdev=mem -mem-prealloc*
> > >
> > > After the Virtual Machine is started, I found the offloads of the
> > > Virtual Machine's virtio-net network adapter are all disabled*.* The
> > > offloads status is checked with command* ethtool -k eth0*. I try to
> > > enables the offloads with ethtool command, but it does not work.
> > >
> > > My questions are:
> > >
> > > 1. Can the offloads of the guest's virtio-net network adapter be
> > > enabled when vhost-user is used?
> >
> > Hi Leo,
> >
> > I guess we need additional implementations in librte_vhost to enable
> > offloads.
> >
> >
> > > 2. If the offloads can't be enabled when vhost-user is used, what is
> the
> > reason?
> >
> > Features are negotiated not olny between virtio-net driver on guest and
> > virtio-net device in QEMU, but also virtio-net device in QEMU and
> vhost-user
> > backend in librte_vhost.
> > As a result, if vhost-user backend doesn't support some features,
> virtio-net
> > driver on guest also cannot use them.
> >
> > Please see "lib/librte_vhost/virtio-net.c"
> Yes, you are correct.
>
> I'm working on the vhost TSO offload,  the offload set in struct
> 'virtio_net_hdr' need to be considered in both virtio-net and vhost side.
>
> > /* Features supported by this lib. */
> > #define VHOST_SUPPORTED_FEATURES ((1ULL <<
> > VIRTIO_NET_F_MRG_RXBUF) | \
> >                                 (1ULL << VIRTIO_NET_F_CTRL_VQ) | \
> >                                 (1ULL << VIRTIO_NET_F_CTRL_RX) | \
> >                                 (1ULL << VHOST_F_LOG_ALL))
> >
> > This is all current librte_vhost supports.
> >
> > Thanks,
> > Tetsuya
> >
> > > It will be great if someone from the forum could give the answers and
> clues.
> > >
> > > Thanks.
> > > Leo
>
>