On 2021-12-20 10:53, Andrew Lunn wrote: >> # ip link set eth0 up >> # ip link show eth0 >> 2: eth0: mtu 1500 qdisc fq_codel >> state UP >> mode DEFAULT group default qlen 1000 >> link/ether ee:f1:a8:e7:0c:8f brd ff:ff:ff:ff:ff:ff >> >> ... now link is up with the same mac address >> >> # ifconfig eth0 >> eth0: flags=4163 mtu 1500 >> inet6 2a02:2f09:3d09:1b00:ecf1:a8ff:fee7:c8f prefixlen 64 >> scopeid >> 0x0 >> inet6 fe80::ecf1:a8ff:fee7:c8f prefixlen 64 scopeid >> 0x20 >> inet6 fd00:cafe::c772:d58f:1061:4df3 prefixlen 64 scopeid >> 0x0 >> inet6 fd00:cafe::ecf1:a8ff:fee7:c8f prefixlen 64 scopeid >> 0x0 >> inet6 2a02:2f09:3d09:1b00:eef4:bb1d:f9d5:febc prefixlen 64 >> scopeid >> 0x0 > > These make use of multicast. Router Advertisements are multicast, both > at L2 and L3, by the router to the Internet. > >> # ping -6 -c 10 ipv6.l.google.com >> PING ipv6.l.google.com(ipv6.l.google.com (2a00:1450:400d:806::200e)) >> 56 data >> bytes >> From odroid (2a02:2f09:3d09:1b00:eef4:bb1d:f9d5:febc) icmp_seq=1 >> Destination >> unreachable: Address unreachable >> ... >> --- ipv6.l.google.com ping statistics --- >> 10 packets transmitted, 0 received, +7 errors, 100% packet loss, time >> 9311ms > > ping uses unicast. > >> ... neither does it work for my local router, so network layer 3 does >> not >> seem to work > > I think it is still pointing towards L2 unicast. > >> next, when i have time and if there's still no progress, i think i >> should >> try to insert: >> >> ret = smsc95xx_reset(dev); >> if (ret) >> goto free_pdata; >> >> before >> >> ret = phy_connect_direct(dev->net, pdata->phydev, >> &smsc95xx_handle_link_change, >> PHY_INTERFACE_MODE_MII); >> >> in smsc95xx_bind() to try to emulate the old behavior for the first >> call to >> start_phy(). > > Yes, that will be in interesting experiment. Something in > smsc95xx_reset() is required. > > Andrew well, i recompiled 5.15.10 with that change and network is still not working. then i disabled NetworkManager and rebooted. after boot i issued following commands: # ethtool -d eth0 # ip link set dev eth0 up # ethtool -d eth0 for 5.15.5 (working kernel), 5.15.8 (unmodified, not working kernel), 5.15.10 (modified as above, not working kernel). obviously, in these register dumps there are 3 different mac addresses (8 bytes at offset 0x0104) ... but the rest could provide a clue (for someone who can make sense of those values). Gabriel