Date:   Tue, 14 Mar 2023 22:35:25 -0400
From:   Etienne Champetier <champetier.etienne@...il.com>
To:     Vladimir Oltean <vladimir.oltean@....com>
Cc:     Tobias Waldekranz <tobias@...dekranz.com>,
        Linux Netdev List <netdev@...r.kernel.org>
Subject: Re: mv88e6xxx / MV88E6176 + VLAN-aware unusable in 5.15.98 (ok in
 5.10.168) (resend)


Le 13/03/2023 à 18:30, Vladimir Oltean a écrit :
> Hi Etienne,
>
> On Sun, Mar 12, 2023 at 12:41:32AM -0500, Etienne Champetier wrote:
>> I get tagged frame with VID 3 on lan4 (at least some multicast & broadcast), but lan4 is not a member of VLAN 3
>> Also unicast frames from wifi to lan4 exit tagged with VID 2, broadcast frames are fine (verifed with scapy)
>> Reverting
>> 5bded8259ee3 "net: dsa: mv88e6xxx: isolate the ATU databases of standalone and bridged ports" from Vladimir
>> and
>> b80dc51b72e2 "net: dsa: mv88e6xxx: Only allow LAG offload on supported hardware"
>> 57e661aae6a8 "net: dsa: mv88e6xxx: Link aggregation support"
>> from Tobias allow me to get back to 5.10 behavior / working system.
>>
>> On the OpenWrt side, 5.15 is the latest supported kernel, so I was not able to try more recent for now.
> I don't know and I am not able to reproduce this on Turris MOX with a linux-5.15.y
> kernel fromhttps://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git.
>
> Could we approach this from the other end? I would like to try and
> reproduce the issue with the kernel you are using. But I have no idea
> how to use OpenWRT or to navigate through its build system. Could you
> help me figure out which source code is built for the Omnia board (plus
> additional OpenWRT patches, if any)?

OpenWrt doesn't support Turris Mox, but here is what is built for Omnia 
as far as I understand

- Linux 5.15.98: 
https://github.com/openwrt/openwrt/blob/0aedf916df364771be47ffda8ff3465250ecee77/include/kernel-5.15

- some generic patches (backport-5.15 / pending-5.15 / hack-5.15): 
https://github.com/openwrt/openwrt/tree/0aedf916df364771be47ffda8ff3465250ecee77/target/linux/generic

- some arch specific patches: 
https://github.com/openwrt/openwrt/tree/0aedf916df364771be47ffda8ff3465250ecee77/target/linux/mvebu/patches-5.15

(not 100% sure in what order they are applied)

- config is generated by taking config-5.15 in generic, mvebu and 
mvebu/cortexa9 and somehow merging them

The wifi code (mac80211 / ath10k) uses kernel backports, so it's 
actually 6.1-rc8 based 
https://github.com/openwrt/openwrt/blob/0aedf916df364771be47ffda8ff3465250ecee77/package/kernel/mac80211/Makefile


While writting this I stumbled upon 
https://github.com/openwrt/openwrt/blob/master/target/linux/generic/hack-5.15/600-bridge_offload.patch,

reverting it fixes half of the problem (frame tagged that should be 
untagged), but I still see frames with VID 3 on a port that is not a 
member of VLAN 3.

I'll continue to look at 'hack' and old 'pending' patches that were 
never accepted,

and loop back with the author of the bridge_offload patch.


> I might also ask you to provide a reproducer for the issue using regular
> iproute2 tools starting from an unconfigured system (bridge, ip, etc, as
> opposed to the network manager from OpenWRT and its /etc/config/network
> configuration file), if this wouldn't be too much effort.

I don't have a serial cable right now, but if reverting suspicious patches

is not enough I'll likely go down that route.


> Also, not clear which interface exactly you mean by "wifi" ("unicast
> frames from wifi to lan4 exit tagged with VID 2").

wifi interface here would be wlan0 / bridged to vlan2

> Thanks.

Powered by blists - more mailing lists