czw., 26 lut 2026 o 22:26 Florian Fainelli
[off-list ref] napisał(a):

On 2/26/26 12:24, Rafał Miłecki wrote:

quoted

W dniu 8.05.2025 o 11:14, Jonas Gorski pisze:

quoted

When bridged ports and standalone ports share a VLAN, e.g. via VLAN
uppers, or untagged traffic with a vlan unaware bridge, the ASIC will
still try to forward traffic to known FDB entries on standalone ports.
But since the port VLAN masks prevent forwarding to bridged ports, this
traffic will be dropped.

This e.g. can be observed in the bridge_vlan_unaware ping tests, where
this breaks pinging with learning on.

Work around this by enabling the simplified EAP mode on switches
supporting it for standalone ports, which causes the ASIC to redirect
traffic of unknown source MAC addresses to the CPU port.

Since standalone ports do not learn, there are no known source MAC
addresses, so effectively this redirects all incoming traffic to the CPU
port.

This change broke standalone ports (those not being part of bridge) on
Northstar (BCM5301X) devices. We got reports from users about WAN ports
on routers not working anymore:
https://github.com/openwrt/openwrt/issues/21187
https://github.com/openwrt/openwrt/issues/21349

Can you take another look at those changes, see if something me be
wrong / missing?

Checking the BCM5301X datasheet, the register definitions look correct.

The first report appears to be for a Netgear R7000 which AFAICT would
utilize port 8 of the switch as the CPU port, which would be correct.

The other report is for a Netgear R6250 where port 5 would be acting as
the CPU port, that could be an issue unless the switch is also
programmed to use port 5 as the IMP port.

Could you try to program GMNGCFG (page 0x02, offset 0) bits 7:6 to have
the value 0b11 which is dual IMP port (IMPO is port 8 and IMP1 is port
5) and see if that helps?

On my Luxul XWR-3150 (it uses gmac0 and port 5) the value of
B53_MGMT_PAGE (0x02) / B53_GLOBAL_CONFIG (0x00) is 0xc2. Nothing to
set there I suppose?

-- 
Rafał