Re: [PATCH v2 net 4/4] net: dsa: mv88e6xxx: isolate the ATU databases of standalone and bridged ports

On Tue, Oct 05, 2021 at 08:32:06PM +0200, Tobias Waldekranz wrote:

I believe this patch gets aaaalmost everything right. But I think we
need to set the port's PVID to 4095 when it is attached to a
VLAN-unaware bridge. Unfortunately I won't be able to prove it until I
get back to the office tomorrow (need physical access to move some
cables around), but I will go out on a limb and present the theory now
anyway :)

Basically, it has to do with the same issue that you have identified in
the CPU tx path. On user-port ingress, setting the port's default FID is
enough to make it select the correct database -- on a single chip
system. Once you're going over a DSA port though, you need to carry that
information in the tag, just like when sending from the CPU. So in this
scenario:

  CPU
   |    .----.
.--0--. | .--0--.
| sw0 | | | sw1 |
'-1-2-' | '-1-2-'
    '---'

Say that sw0p1 and sw1p1 are members of a VLAN-unaware bridge and sw1p2
is in standalone mode. Packets from both sw1p1 and sw1p2 will ingress on
sw0p2 with VID 0 - the port can only have one default FID - yet it has
to map one flow to FID 1 and one to FID 0.

Setting the PVID of sw1p1 should provide that missing piece of
information.

Hey, Spiderman!

Yes, that makes sense, and is something I can actually test, no need to
wait until tomorrow. Give me a few hours, I need to finish something else first.