mirror of
https://github.com/openwrt/openwrt.git
synced 2026-10-14 06:38:29 +04:00
Aleksei Sviridkin
244d2f67a5
On the Keenetic MT7981 boards the LAN port takes its MAC from rf-eeprom
0x4, and mt76 reads the same slot for phy0. The 2.4 GHz radio and the
LAN port therefore share one address. Found on a Keenetic KN-1012 while
writing down its MAC map: eth0, br-lan and phy0 all read
d4:9c:53:xx:xx:a1.
This is the phy0 side of the collision fixed for phy1 in commit
cc96ed1e6d ("mediatek: filogic: fix Keenetic phy1 MAC collision"). Give
phy0 the slot with the LA bit set. The userspace VIF allocator keeps bit
7 of the first octet, so the extra phy0 VIFs stay apart from phy1 and
its VIFs, which have bit 7 flipped, and from the ethernet addresses,
which have the LA bit clear. With num_global_macaddr set, the allocator
adds to the base instead, and phy0 + 1 was the WAN address wherever the
0xa slot holds the base + 1, as on the KN-1012; that goes away too. The
2.4 GHz BSSID of these boards changes accordingly. On KAP-630 and
NAP-630 the ethernet port uses the 0xa slot instead, so there it only
changes the BSSID, and the family keeps one 2.4 GHz addressing
convention.
The Buddy 6 and Buddy 6 SE extenders (KN-3411, KN-4410, NC-3411,
NC-4410) have the same collision on their single port: on an NC-4410,
phy0-ap0 came up with d4:9c:53:xx:xx:23, the address its LAN has. They
only need the phy0 part. Their 0xa slot is empty (zeros on that unit),
so mt76 already derives phy1 as the family fixup does, 56:9c:53:xx:xx:23
there.
Tested on the KN-1012: phy0 comes up as d6:9c:53:xx:xx:a1 and phy1 stays
56:9c:53:xx:xx:a1. The VIF allocator hands phy0 d2/de/da and phy1
52/5e/5a in the first octet, none of them equal to the LAN or WAN
address. With two APs on phy0 and one on phy1, all bridged into br-lan,
every bridge port has its own local address.
Assisted-by: LLM
Signed-off-by: Aleksei Sviridkin <f@lex.la>
Link: https://github.com/openwrt/openwrt/pull/25450
Signed-off-by: Jonas Jelonek <jonas@jonasjelonek.de>