RE: [virtio-dev] RE: [virtio-comment] RE: [PATCH v9] virtio-net: support inner header hash

[Parav Pandit <parav@nvidia.com>]

> From: Michael S. Tsirkin <mst@redhat.com>
> Sent: Tuesday, February 21, 2023 12:44 PM
> 
> On Tue, Feb 21, 2023 at 05:40:51PM +0000, Parav Pandit wrote:
> >
> > > From: Michael S. Tsirkin <mst@redhat.com>
> > > Sent: Tuesday, February 21, 2023 12:14 PM
> > > > The part that I am missing is, how do to reuse
> > > > virtio_net_hash_config and say
> > > ignore all the existing fields related to rss, but only consider
> > > hash_tunnel_types?
> > >
> > > Like a union?  The answer is, don't. Just lay out fields one after another.
> > >
> > In that case driver needs to fill up all the fields which are not
> > related to hash_tunnel_types and the device also needs to compare with
> > the previous config and ignore it.  Doesnât look like a good use of
> > existing commands and sw/fw usage for it.  Shouldnât we have the
> > explicit command for setting tunnel types?
> 
> I don't know what's proposed at this point, this is too vague.
Proposal is to have new command like how Heng drafted in latest email.

> I feel which tunnels to hash for inner header is not different from which
> transports to hash. If device wants to know what changes it can compare. I
> expect generally devices will just apply the new config without caring what
> changed exactly.
In a device when things are programmed, it often requires removing the previous configuration and re-apply the new one.
This results in wrong steering for several tens of micro to milli seconds for no apparent reason.
Hence, comparison is often needed for the best experience.
And those are just overheads on the device and driver side without any apparent gain other than reusing some structure of rss.
Hence, a separate command is more efficient choice.