[dpdk-dev] [RFC v2 0/5] rte_flow extension for vSwitch acceleration

Alex Rosenbaum rosenbaumalex at gmail.com
Thu Dec 21 14:01:36 CET 2017


On Thu, Dec 21, 2017 at 4:35 AM, Qi Zhang <qi.z.zhang at intel.com> wrote:
> This patch extend rte_flow API.
> The purpose is to provide comfortable programming interface for virtual switch
> software (such as OVS) to take advantage of incoming device's vSwitch acceleration
> capability when using DPDK as data plane.
>
> Below is summary of changes:
>
> 1. Support to specify flow's destination as an ethdev interface.
>
> Add action RTE_FLOW_ACTION_TYPE_ETHDEV_PORT, use port_id as the identification
> of the destitation. A typical use case is, with a smart NIC used for vSwitch
> acceleration, flow is defined to redirect packet between switch port that is
> managed by a Port Representor.
> See patch for Port Representor: http://dpdk.org/dev/patchwork/patch/31458/

If we are going to have a representation of a switch in DPDK then I we
need real PMD's for the switch ports, and not virtual PMD's as in the
Port Representor model.
These real ports will probably require specific switch_ops, which are
different than the NIC (eth_dev_ops). At least few additional on top
of NIC ops.
We will need to support Tx and Rx burst on these switch ports and
configure ACL's, flooding mode and more.

Alex


More information about the dev mailing list