[Qemu-devel] QEMU RX offload issues with net device frontend that does not support virtio extensions

[Qemu-devel] QEMU RX offload issues with net device frontend that does not support virtio extensions

[Dmitry Fleytman]

Hello,

It looks like there is a problem in QEMU related to usage of frontends
without virtio extensions.

Normally on RX path QEMU receives packet metadata with offloads
information in virtio header.
In particular there is an information about checksumming state. For
example, when
packet is being transferred from guest to guest with offloads enabled
on transmitter side,
checksums won't be calculated, special flag will be set instead in
virtio header indicating that the checksum is ok.
Based on this flag device may notify guest that checksum verification
for this packet is not required.

Now, when virtio header is not supported on receiver side, there is no
way device may be notified about the checksum state.
Packet will be indicated to the guest as is and will be dropped by
guest's network stack.

I've tested this scenario with virtio devices and observed the issue described.

Do I miss something? Is there any configuration option or something
I'm not aware about?

Thanks in advance,
Dmitry Fleytman.

Re: [Qemu-devel] QEMU RX offload issues with net device frontend that does not support virtio extensions

[Stefan Hajnoczi]

On Wed, Mar 7, 2012 at 4:32 PM, Dmitry Fleytman
<dmitry.fleytman@ravellosystems.com> wrote:
> It looks like there is a problem in QEMU related to usage of frontends
> without virtio extensions.
>
> Normally on RX path QEMU receives packet metadata with offloads
> information in virtio header.

Have you tried -netdev tap,vnet_hdr=off ?

I guess the tricky thing is negotiating whether or not to enable the
host IFF_VNET_HDR flag.  (Especially with the QEMU "vlan" feature
where the tap isn't peered directly with another device.)

Stefan