Re: [PATCH bpf 5/6] libbpf: ringbuf: Prevent missed wakeups

From: Tamir Duberstein

Date: Fri Nov 20 2026 - 15:42:25 EST


> let's use regular kernel style

Fixed in v3.

> [...]

> The extra new variable feels overly complicated.

The final iterator uses one loop without got_new_data. It retains one
fenced BUSY retry to avoid missing a commit without spinning indefinitely.

> [...]

> We just add an smp_mb() here.

The tools generic mb() fallback can be only a compiler barrier.
__atomic_thread_fence() supplies the required StoreLoad ordering.

> [...]

> Why __ATOMIC_RELAXED here?

The stop flag publishes no other state. Relaxed atomics suffice;
volatile would leave a C data race.

> [...]

> Label is unnecessary

Removed in v3.