A tale about fixing eBPF spinlock issues in the Linux kernel
Summary
The article provides a deep technical analysis of a Linux kernel spinlock bug related to eBPF ring buffers. It traces a race between kernel context-switch code and non-maskable NMIs, explains the role of rqspinlock deadlock checks, and details patches that fixed the freezes and mitigations for NMIs. It also discusses why older kernels were susceptible and what this means for IT teams relying on eBPF-based observability.