Open IoT & ELC 2017

The Linux kernel trace event subsystem provides a large array of tracepoints (and an infinite number of others can be added dynamically) which are used by various tracing utilities both inside and outside the kernel. Unless one-off external logic (code) is applied, there's no way to extract inter-event quantities such as latencies from the trace subsystem itself, even though the data to do that is already trivially there. This talk introduces a general-purpose mechanism for extracting inter-event values from arbitrary sets of events and shows how both one-off/one-line/programming-free experiments and higher-level tools such as latency histograms can be built on top of it.