[rsyslog] FW: [Valgrind-users] helgrind points to race in pthread_mutex_lock?

[Rainer Gerhards]

Hi all,

An update on the abort case bug.

Below is one of the responses I got from the valgrind mailing list about
my inquiry. In short, the helgrind log I received seems not to be
absolutely correct and thus is unfortunately of very limited use for
troubleshooting.

There are new versions out and I will ask Lorenzo if it is possible to
run one of those. I'll do that via private mail as I need to forward
some more messages (unfortunately the valgrind mailing list archive
lacks behind several days, so I can not yet refer to the archive...).

Rainer

> ---------- Forwarded message ----------
> From: Julian Seward < >
> Date: Mon, Sep 22, 2008 at 8:54 PM
> Subject: Re: [Valgrind-users] helgrind points to race in
> pthread_mutex_lock?
> To: valgrind-users at lists.sourceforge.net
> Cc: Konstantin Serebryany < >, Rainer
> Gerhards < m>
> 
> 
> 
> On Monday 22 September 2008, Konstantin Serebryany wrote:
> 
> > helgrind is under active development now in two separate branches
> > (HGDEV & YARD).
> > The version in trunk is outdated. Please try YARD first:
> > svn co svn://svn.valgrind.org/valgrind/branches/YARD YARD
> 
> Yes, do try YARD; you might like it more than the 3.3.1 helgrind.
> 
> Nevertheless, the error you show:
> 
> > > ==31831== Possible data race during read of size 4 at 0x6622E8
> > > ==31831==    at 0x50464FE: pthread_mutex_lock
> (pthread_mutex_lock.c:87)
> > > ==31831==    by 0x4C2217B: pthread_mutex_lock
(hg_intercepts.c:400)
> > > ==31831==    by 0x420AFC: dbgGetThrdInfo (debug.c:604)
> > [...]
> 
> should have been suppressed, so you should not see it.  It is due to
an
> internal implementation limitation which is hard to avoid, so in some
> cases we suppress errors which we know cannot be from end-user code.
> 
> I think the relevant suppression in glibc-2.34567-NPTL-helgrind.supp
> is:
> 
>  {
>     helgrind-glibc25-101
>     Helgrind:Race
>     obj:/lib*/libpthread-2.5.*so
>     fun:pthread_*
>  }
> 
> (or corresponding other versions for glibc > 2.5, in the same file)
> 
> One possible reason it did not suppress the error is that you have
> debug information on/for your libpthread, which is an unusual setup
> that
> Helgrind has not been properly tested against.
> 
> But I am not 100% sure about that.
> 
> * if you uninstall the libc debuginfo rpm/deb, does that change
> anything?
> 
> * what glibc version are you using?
> 
> J