The Linux trace toolkit's next generation

The Linux trace toolkit's next generation

Posted Jan 10, 2008 4:32 UTC (Thu) by jd (guest, #26381)
Parent article: The Linux trace toolkit's next generation

There are several projects that provide architecture-independent access to the hardware clocks, PAPI is one. Although this is a relatively minor part of LTT-ng, it would presumably make sense if there is an agreed way to do this - or, perhaps if one project is already installed, any subsequent project should be configurable to use that as the access point. The problem with multiple methods of accessing the same data is that if you use multiple profilers, they may give inconsistent and incompatible results.

---

(Log in to post comments)

Yes, I would be happy to do that. However, reading timestamps in a tracer is a bit trickier
than the average scenario, especially because of NMI context tracing. This is why I developed
algorithms that keeps track of 64 bits counters that can be read atomically, even if the
underlying hardware only provides a 32 bits counter.

So I guess it would make sense to use the LTTng timestamping infrastructure for other uses,
but the opposite is not necessarily true because of the reentrancy constraints.

Mathieu