I don't understand then, what is the problem?

You don’t actually need a static route on PC1. Since PC1 already has a default gateway (192.168.100.254), anything outside 192.168.100.0/24 is automatically sent there. So PC1 isn’t the source of the routing problem. The real issue is on the 10.10.10.x side: Device1 has no gateway. Without a gateway, it can receive packets coming from the Linux box, but it has no way to send replies back to a different subnet. That’s why the communication fails. The fix is simply to give Device1 a valid gateway pointing to the Linux box. Device1: IP: 10.10.10.2 Mask: 255.255.255.0 Gateway: 10.10.10.1 On the Linux box: sudo sysctl -w net.ipv4.ip_forward=1 This is enough to route between NIC1 and NIC2. If Device1 cannot be configured manually, the Linux box can offer a small DHCP service on NIC2. A minimal dnsmasq example: interface=enp4s0 dhcp-range=10.10.10.50,10.10.10.150,255.255.255.0,12h dhcp-option=3,10.10.10.1 This simply ensures Device1 learns “send everything through 10.10.10.1”. Once Device1 has a gateway, the Linux box will forward traffic correctly and PC1 will reach it without needing any static route at all. Useful resources: https://wiki.archlinux.org/title/Router https://askubuntu.com/questions/205017/how-to-define-the-range-of-ips-using-dhcp https://pingmynetwork.com/network/ccna-200-301/dhcp-dynamic-host-configuration-protocol

I don't think so, then my Virtual Machine run would fail the same way as you describe, but it doesn't, it works fine. Of course it is the meson64 kernel running on top of KVM/virt machine, the RTC for example is emulated, while for your real N2+ is a specific Amlogic SoC kernel driver I assume, so there something could be wrong. Or what about clocking tree, there are many failures possible. I know for rockchip64 there were 2 RTC modules, at some point in time 1 was chosen over the other, so a change, could also introduce issues. You could browse kernel changes/patches And also what about an unknown bad/corrupted diskblock or so. I normally use Btrfs for rootfs, so easy to check for and detect corruption. So maybe re-image again. I am still not sure if I used the same image as you, but up to you to track that. You can do the same thing as I did to see if it is maybe a specific Amlogic driver issue; Run that image as VM on the same N2+, you need a network bridge with main eth port or else use NAT. Or you make it first efi bootable, then it can run easier in virt-manager GUI. N2+ is fast enough to run VMs, I even do that on ROCK3A (2GB RAM, 4x Cortex-A55). Or use standard Debian Trixie kernel on real N2+, not sure if that works good enough. Or remove/disable RTC kernel module. Or also remove fake-hwclock stuff if you haven't already done that. I did not do it for this VM run, but when real RTC+battery, I would remove or disable it.

@Gavin Mundayhttps://github.com/NickAlilovic/build/blob/Radxa-A7A/config/boards/radxa-cubie-a7z.csc

found it wasnt using correct tree..;p