Kernel lockup via incompatible module-function BPF trampoline

Incompatible BPF module-function trampolines in Linux 6.17 on LoongArch allow local users to lock the kernel or disrupt module traffic. The LoongArch __arch_prepare_bpf_trampoline path treated module text as an ordinary original-call target instead of rejecting it, producing an unsafe fentry trampoline. Triggering the fault requires permission to load and attach a BPF tracing program to a kernel-module function; effects include a system lockup or loss of service from a traced module such as WireGuard.