[dpdk-dev] [PATCH 1/3] MAINTAINERS remove experimental tag from vdev_netvsc

[Stephen Hemminger]

On Tue, 06 Oct 2020 00:41:55 +0200
Thomas Monjalon <thomas at monjalon.net> wrote:

> 15/09/2020 16:47, Stephen Hemminger:
> > Ferruh Yigit <ferruh.yigit at intel.com> wrote:  
> > > On 9/15/2020 3:03 AM, Stephen Hemminger wrote:  
> > > > Vdev_netvsc has been around for several years. It no longer needs
> > > > to be marked experimental.
> > > > 
> > > > Signed-off-by: Stephen Hemminger <stephen at networkplumber.org>
> > > > ---
> > > > -Microsoft vdev_netvsc - EXPERIMENTAL
> > > > +Microsoft vdev_netvsc
> > > >  M: Matan Azrad <matan at mellanox.com>
> > > >  F: drivers/net/vdev_netvsc/
> > > >  F: doc/guides/nics/vdev_netvsc.rst  
> > > 
> > > As far as I remember 'vdev_netvsc' was interim solution until 'netvsc'
> > > was ready. In this patchset 'netvsc' is also becoming mature.
> > > 
> > > Wouldn't be easier to keep 'vdev_netvsc' experimental to be able to
> > > remove it soon?  
> > 
> > Let me discuss with Long Li and management.
> > Maybe replace EXPERIMENTAL with DEPRECATED in 20.11.  
> 
> Would be strange to switch from experimental to deprecated :)
> 
> +Cc Matan
> 
> I think you still need this platform driver (with failsafe and tap)
> in case you need rte_flow. Or is it well supported with netvsc PMD?

This needs more discussion. Netvsc PMD does not support rte_flow because
there is not a good/complete implementation of rte_flow library in pure
software form. It might be possible using the BPF stuff that tap supports.

I do not know of anybody who is using vdev_netvsc/failsafe with rte_flow.
Part of the problem is that the TAP implementation of rte_flow supports
a much smaller subset of features than the VF device (MLX).