On 25 December 2016 at 03:39, Paul Blakey <pa...@mellanox.com> wrote:
> flow offloading by tc needs ingress qdisc on the device.
> Deleting the ingress qdisc was done in order to flush
> policing filters, so instead we just flush the filter and
> leave the ingress added (and add it if there wasn't any).
>
> Signed-off-by: Paul Blakey <pa...@mellanox.com>
> Reviewed-by: Roi Dayan <r...@mellanox.com>

I'm concerned that this patch will break a bunch of things.

I don't think we generally want an ingress qdisc unless there is
specific policing configuration, or hardware offloads is enabled. This
looks like it sets it up regardless.

I suspect that if you changed policing configuration, it will not take
effect with this patch.

Hardware offloads is orthogonal to policing; I suggest that there
should be a new netdev-provider API to enable hardware offloads, which
will ensure that the qdiscs are set up appropriately.

If policing is configured with a rate/burst, I would expect this to
take effect. If this policing is not possible to be offloaded to
hardware, then I'd expect either OVS complains in the logs that
hardware offloads are configured with policing, and describe how it
works - either A), policing is ignored and flows are offloaded, or B)
policing works, but flows are not offloaded.
_______________________________________________
dev mailing list
d...@openvswitch.org
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to