[dpdk-dev] DPDK Virtio Checksum Errors

Maxime Coquelin maxime.coquelin at redhat.com
Tue Aug 6 09:55:36 CEST 2019


Hi,

On 8/5/19 8:09 PM, kumaraparameshwaran rathinavel wrote:
> Hi All,
> 
> I had recently tried using TSO on the DPDK virtio guest driver and I get
> the TCP checksum errors at the client side. I tried running iperf and
> Apache bench
> 
> We have been using the same TCP/IP stack for other DPDK pmd drivers and
> they are working fine. I did notice that incase of virtio alone we do
> calculate the IP payload in pseudo check sum, please correct me if I am
> wrong. This is the guest side implementation. Is this applicable for all
> versions of qemu virtio
> 
> In the host the OVS is run in a Kolla inside container and it is in
> non-dpdk mode. The tcpdump from the physical interface shows that the
> packet is not modified from the way it was sent by the guest.
> 
> After the checksum errors the TCP stack of the guest  sends the MTU sized
> packet for which the TSO failed and this time the checksum at the client
> side did not fail. In the client I had disabled GRO and LRO, so that we
> compute checksum on each frame.
> 
> Also the checksum errors are seen only when the VMs communicating are
> present in different physical host, when VMs in same host I do not see any
> checksum errors.
> 
> Please let me know if such issues were seen earlier and any pointers on
> debugging further would be helpful.

Which version of DPDK are you using?

There was a fix recently that may fix your issue [0]:

commit 00a5ea02e12fc1409901c0149ce87c28a286ea60
Author: Dilshod Urazov <dilshod.urazov at oktetlabs.ru>
Date:   Mon Jun 17 12:31:37 2019 +0100

    net/virtio: add Tx preparation

    Virtio requires pseudo-header checksum in TCP/UDP checksum to do
    offload, but it was lost when Tx prepare is introduced. Also
    rte_validate_tx_offload() should be used to validate Tx offloads.

    Also it is incorrect to do virtio_tso_fix_cksum() after prepend
    to mbuf without taking prepended size into account, since layer 2/3/4
    lengths provide incorrect offsets after prepend.

    Fixes: 4fb7e803eb1a ("ethdev: add Tx preparation")
    Cc: stable at dpdk.org

    Signed-off-by: Dilshod Urazov <dilshod.urazov at oktetlabs.ru>
    Signed-off-by: Andrew Rybchenko <arybchenko at solarflare.com>
    Reviewed-by: Tiwei Bie <tiwei.bie at intel.com>


> Thanks,
> Param
>

https://www.mail-archive.com/dev@dpdk.org/msg134448.html


More information about the dev mailing list