syscall tracing overheads: utrace vs. kprobes

[Frank Ch. Eigler]

Hi -

On Tue, Apr 28, 2009 at 11:10:10AM -0700, Roland McGrath wrote:
> [...]
> The microbenchmark makes that seem like more of a sensical comparison than
> it really is.  They are really apples and oranges.  The TIF_SYSCALL_TRACE
> types (process.syscall) add some overhead to every system call.  The probe
> types (kprobe/tracepoint/marker) add overhead only to the probed call.

Certainly, in general.  But in this specific test, only the under-test
system calls occurred in essnetially the whole system, so the overhead
measurements were in a way the bare minimums imposed by the kprobes
vs. utrace callback infrastructure itself.

> In real situations [...]  the distribution of overheads will be
> quite different.

Or rather, the basic overhead quanta measured above may be multiplied
along several different axes.


- FChE