Kernel panic with AF_XDP + zero copy

I'm benchmarking a new AF_XDP code path and my tester generates around 80MPPS. The application on the receiving end is an AF_XDP zero copy app, which just counts the incoming packets. On the receiver I see two different panics:

First type of panic:

[panic-1.txt](https://github.com/user-attachments/files/27582366/panic-1.txt)

Second type of panic:

[panic-2.txt](https://github.com/user-attachments/files/27582390/panic-2.txt)

Both machines have two Intel E810 NICs that are configured as bonded, but the AF_XDP program of course operates at a layer below that, so I don't expect it to make a difference. RDMA is enabled.

The device being used here is also the main NIC of the machine, so concurrently to receiving traffic that is redirected in an XDP program it also receives traffic that goes to the kernel. As the first panic is related to skb handling it seems like the interaction between the kernel TCP/IP stack and zero copy AF_XDP might have an issue.

**Distro**: Ubuntu Server 26.04 LTS
**Kernel**: linux-image-7.0.0-15-generic

[dmesg.txt](https://github.com/user-attachments/files/27583228/dmesg.txt)