[ovs-dev] [PATCH V5 1/2] netdev-dpdk: DPDK v17.11 upgrade

[Jan Scheurich]

> From: Mark Kavanagh [mailto:mark.b.kavanagh at intel.com]
> Sent: Thursday, 07 December, 2017 18:46
> 
>  #. (Optional) Configure DPDK as a shared library
> @@ -583,7 +583,21 @@ Limitations
>    The latest list of validated firmware versions can be found in the `DPDK
>    release notes`_.
> 
> -.. _DPDK release notes: http://dpdk.org/doc/guides/rel_notes/release_17_05.html
> +.. _DPDK release notes: http://dpdk.org/doc/guides/rel_notes/release_17_11.html
> +
> +- The DPDK v17.11 virtio net driver contains a bug that prevents guest DPDK
> +  applications from receiving packets. This only occurs when guest-bound traffic
> +  is live before a DPDK application is started within the guest, and where two
> +  or more forwarding cores are used. As such, it is not recommended for guests
> +  which execute DPDK applications to upgrade to DPDK v17.11.
> +
> +  Note that this issue does not occur when guest network devices are controlled
> +  by the guest kernel.

I don't think this aspect should be documented as a limitation of OVS. It is a bug that can affect DPDK 17.11 guest applications under certain conditions. If at all we should have the such a statement in the dpdk/vhost_user.rst document (see below).

I am missing a statement here that OVS-DPDK itself, when running in a VM with virtio interfaces, can be affected by this bug of the virtio PMD. A work-around until the bug is fixed could be to make sure mergeable rx buffers are enabled for the virtio ports in Qemu and vhost backend on the host, so that the virtio PMD does not use the faulty vector Rx function.

> +
> +  Details regarding the virtio driver bug are available on the `DPDK mailing
> +  list`_.
> +
> +.. _DPDK mailing list: http://dpdk.org/ml/archives/dev/2017-December/082801.html
> 
>  Reporting Bugs
>  --------------
> diff --git a/Documentation/topics/dpdk/ring.rst b/Documentation/topics/dpdk/ring.rst
> index ad9d7a5..8d0ede8 100644
> --- a/Documentation/topics/dpdk/ring.rst
> +++ b/Documentation/topics/dpdk/ring.rst
> @@ -77,4 +77,4 @@ DPDK. However, this functionality was removed because:
>  - :doc:`vhost-user interfaces <vhost-user>` are the defacto DPDK-based path to
>    guests
> 
> -.. _DPDK documentation: https://dpdk.readthedocs.io/en/v17.05/prog_guide/ring_lib.html
> +.. _DPDK documentation: https://dpdk.readthedocs.io/en/v17.11/prog_guide/ring_lib.html
> diff --git a/Documentation/topics/dpdk/vhost-user.rst b/Documentation/topics/dpdk/vhost-user.rst
> index 74ac06e..1a5de24 100644
> --- a/Documentation/topics/dpdk/vhost-user.rst
> +++ b/Documentation/topics/dpdk/vhost-user.rst
> @@ -287,6 +287,11 @@ application in the VM.
> 
>    Support for DPDK in the guest requires QEMU >= 2.2
> 
> +.. important::
> +
> +  DPDK v17.11 virtio PMD contains a bug that affects testpmd/DPDK guest
> +  applications. As such, guests should not upgrade to 17.11.

If such a temporary bug warning is really wanted here, it needs to be more qualified.