tl;dr: apps are auto-linked to LTTng, which is a tracing framework. It contains calls to the membarrier syscall, which in some cases waits until every thread has context-switched at least once, which is kinda slow. New kernel option solves this.

In order to debug the slow-down issue they use the command `perf trace --duration 1 ./a.out' on where a.out is

  int main()
  {
  	return 0;
  }

and is built using `gcc -O2 -g -ldl -llttng-ust lttng-test.c'

Why is their tracing framework not able to trace itself? Is that a churlish question?

Lttng-ust is used primarily for static tracepoints. They needed a syscall level visibility, so strace / perf trace is a better tool.

I see. Thank you!