Re: [PATCH v2 net 2/3] net: dsa: don't advertise 'rx-vlan-filter' if VLAN filtering not global

[Vladimir Oltean <olteanv@gmail.com>]

On Tue, Mar 23, 2021 at 09:16:03AM -0700, Florian Fainelli wrote:
> > Would you be so kind to try this fixup patch on top?
>
> That works for me, thank you! So for the whole patch when you resend,
> you can add:
>
> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com>
> Tested-by: Florian Fainelli <f.fainelli@gmail.com>

Thanks again for testing!

> > Although I am much less confident now about submitting this as a bugfix
> > patch to go to stable trees. But I also kind of dislike the idea that
> > Tobias' patch (which returns -EOPNOTSUPP in dsa_slave_vlan_rx_add_vid)
> > only masks the problem and makes issues harder to reproduce.
> >
> > Tobias, how bad is your problem? Do you mind if we tackle it in net-next?
> > Also, again, any chance you could make mv88e6xxx not refuse the 8021q
> > VLAN IDs?
>
> I was thinking the same last night while sending my results, as far as I
> can tell the switches that have global VLAN filtering or hellcreek are
> not broken currently right?

Yes.

> If only mv88e6xxx seems to be requiring special treatment, how do we
> feel about adding an argument to port_vlan_add() and port_vlan_del()
> that tell us the context in which they are called, that is via 802.1q
> upper, or via bridge and have mv88e6xxx ignore the former but not the
> latter?

How would you then describe to .port_vlan_add() those VLANs that don't
come either from the bridge nor from 8021q uppers, but from direct calls
to vlan_vid_add? A VLAN is a VLAN, and a driver with
configure_vlan_while_not_filtering should accept it.

If mv88e6xxx refuses this right away:

ip link add link lan0 name lan0.100 type vlan id 100

Then traffic through lan0.100 will be broken as soon as we do:

ip link add br0 type bridge vlan_filtering 1
ip link set lan0 master br0

So I believe we should be looking at how to make the Marvell driver
accept the VLAN, not how to help it refuse it in other ways.