Andrew Doran wrote:

> There's a global flag that's tested (ktrace_on) before any function calls
> are made. In the common case (tracing off) it should be marginally cheaper.

Have you measured this with say lat_syscall from lmbench?

Cheers,
Simon.