Hi Peter, On Mon, 28 Oct 2019 10:12:29 +0100 you wrote: > But now that I wrote it, I'm a little scared of what I had to do for > __perf_sw_event(). Let me ponder that a little bit more. Do you have an idea how to resolve this? For context, I maintain a Go library that interacts with perf_events for the purpose of attaching eBPF kprobes. Users keep asking for an explicit tracefs fallback since the miss counter is so valuable to them. It'd be great to find a solution! Best Lorenz
