[xdp-hints] Re: [PATCH bpf-next v4 3/4] net: stmmac: Add launch time support to XDP ZC

[Stanislav Fomichev <stfomichev@gmail.com>]

On 01/09, Song, Yoong Siang wrote:
> On Wednesday, January 8, 2025 1:08 AM, Stanislav Fomichev <stfomichev@gmail.com> wrote:
> >On 01/06, Song Yoong Siang wrote:
> >> Enable launch time (Time-Based Scheduling) support to XDP zero copy via XDP
> >> Tx metadata framework.
> >>
> >> This patch is tested with tools/testing/selftests/bpf/xdp_hw_metadata on
> >> Intel Tiger Lake platform. Below are the test steps and result.
> >>
> >> Test Steps:
> >> 1. Add mqprio qdisc:
> >>    $ sudo tc qdisc add dev enp0s30f4 handle 8001: parent root mqprio num_tc
> >>      4 map 0 1 2 3 3 3 3 3 3 3 3 3 3 3 3 3 queues 1@0 1@1 1@2 1@3 hw 0
> >>
> >> 2. Enable launch time hardware offload on hardware queue 1:
> >>    $ sudo tc qdisc replace dev enp0s30f4 parent 8001:2 etf offload clockid
> >>      CLOCK_TAI delta 500000
> >>
> >> 3. Add an ingress qdisc:
> >>    $ sudo tc qdisc add dev enp0s30f4 ingress
> >>
> >> 4. Add a flower filter to route incoming packet with VLAN priority 1 into
> >>    hardware queue 1:
> >>    $ sudo tc filter add dev enp0s30f4 parent ffff: protocol 802.1Q flower
> >>      vlan_prio 1 hw_tc 1
> >>
> >> 5. Enable VLAN tag stripping:
> >>    $ sudo ethtool -K enp0s30f4 rxvlan on
> >>
> >> 6. Start xdp_hw_metadata selftest application:
> >>    $ sudo ./xdp_hw_metadata enp0s30f4 -l 1000000000
> >>
> >> 7. Send an UDP packet with VLAN priority 1 to port 9091 of DUT.
> >
> >Tangential: I wonder whether we can add the setup steps to the
> >xdp_hw_metadata tool? It is useful to have one command to run that
> >takes care of all the details. Same way it already enables HW
> >tstamping..
> >
> >Or, if not the full setup, some kind of detection we can signal to the
> >user that some things might be missing?
> 
> Sure. I can try to add the setup steps into xdp_hw_metadata
> by using ioctl() function. However, there are some device specific
> command, like the number of queue, their priority,
> how they route the incoming packet, etc. I will try to find out
> a common way that can work for most of the devices,
> at least work for both igc and stmmac.

We can query the number of queues (and everything else you need) in the
tool, similar to what we do in
testing/selftests/drivers/net/hw/ncdevmem.c, take a look. But if it's
too complicated, maybe at least print these commands on startup and tell
the user to run them.

The reason I'm asking is that I hope that we eventually can run this
tool from (automatic) testing/selftests/drivers/net/hw testsuite to
make sure the metadata stuff keeps working.