On Wed, Aug 31, 2022 at 08:12:09AM +0200, Mattias Forsblad wrote:

Please elaborate why this may pose a problem, I might have missed
some information.

Because as I said, you need to enable it over the CPU port connected to
the master which goes up and is operational, and you don't have this
information at probe time.

quoted

If you decide to rework this using the master netdev, you can use
dsa_tag_protocol_overhead(master->dsa_ptr->tag_ops). Or even reserve
enough headroom for the larger header (EDSA) and be done with it.
But then you need to construct a different header depending on whether
DSA or EDSA is used.

So in the new version a la qca8k we need the 'extra' parameter to
see if we need space for EDSA header, thus we need run through the tagger.
We can discuss that in the next version.

"thus we need run through the tagger" -> is this a justification that
you're going to keep the tag_ops->inband_xmit?

You don't _have_ to, you already have access to the tagging protocol in
use via chip->tag_protocol, you can derive from that if you need the E
in EDSA or not, and still keep everything within the switch driver.

quoted

Could you please explain for me what will setting skb->pkt_type to
PACKET_OUTGOING achieve?

I though it was prudent, will remove if it's not needed.

I honestly don't know what it does.