This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
Re: Unified tracing buffer
* Darren Hart (darren@dvhart.com) wrote:
> On Mon, Sep 22, 2008 at 7:02 PM, Mathieu Desnoyers <compudj@krystal.dyndns.org>
> > So only in the specific case of instrumentation of things like locking,
> > where it is possible to insure that instrumentation is synchronized with
> > the instrumented operation, does it make a difference to choose the TSC
> > (which implies a slight delta between the TSCs due to cache line delays
> > at synchronization and delay due to TSCs drifts caused by temperature)
> > over an atomic increment.
> >
>
> Hrm, i think that overlooks the other reason to use a time based counter over
> an atomic increment: you might care about time. Perhaps one might be less
> concerned with actual order tightly grouped events and more concerned with the
> actual time delta between more temporally distant events. In that case, using
> a clocksource would still be valuable. Although admitedtly the caller could
> embed that in their payload, but since we seem to agree we need some kind of
> counter, the time-based counter appears to be the most flexible.
>
> Thanks,
>
See my answer to Linus for a proposal on how to do both :)
Mathieu
> --
> Darren Hart
>
--
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F BA06 3F25 A8FE 3BAE 9A68