This is the mail archive of the
systemtap@sourceware.org
mailing list for the systemtap project.
RE: Summary of nightly testing 20061016
- From: "Stone, Joshua I" <joshua dot i dot stone at intel dot com>
- To: "William Cohen" <wcohen at redhat dot com>
- Cc: <systemtap at sourceware dot org>
- Date: Mon, 16 Oct 2006 10:53:51 -0700
- Subject: RE: Summary of nightly testing 20061016
On Monday, October 16, 2006 8:05 AM, William Cohen wrote:
> The timers.stp probe didn't register on FC5 i686 machine It failed
> with rc=-16, "Device or resource busy" because oprofile was already
> using the timer. So this means that only one probe could be using the
> timer mechanism, and no concurrent use?
For the plain timers, concurrent use is fine -- it's only a problem for
timer.profile. This is because of an API change that happened in
2.6.10. Previously, the profile timer was accessed with
register_profile_notifier, which used a notifier_block (allowing
unlimited registrants). In 2.6.10 it switched to register_timer_hook,
which uses a single pointer to store the callback (allowing only one
registrant).
Probably someone figured that oprofile was the only one that would ever
use the timer profiler, so this change was an "optimization". It does
reduce the work for making the callback, but the functionality is more
limited.
Josh