mutex-bug in recent releng_[67]?

[Tim Liddelow]

Hi,

I saw this post after trawling the archives trying to find a solution to a
problem I'm having - and I can say that I'm seeing the same issue, in this
case with CruiseControl.

We've been running CruiseControl on 6.2-STABLE for a while without issues,
and about a week ago (after the box was rebooted into a new kernel) started
to see java processes consuming 100% CPU.  I have confirmed that the problem
doesn't occur when I use the Linux JDK.  So something has changed; what I
cannot tell (I've tried to find any kernel commits that may be related) is
exactly where.

I've seen the followup posts regarding reader/writer processes - I'm just
not sure this bug is limited to the Java exec(), but perhaps it is tickled
by it.  To reproduce, I've compiled CruiseControl (from sourceforge) and it
hangs during the running of unit tests, whereas with the Linux JVM it
doesn't.

I think we may have a real bug here.  Sorry I can't give more precise
information - I'm still investigating.

Cheers
Tim.