Memory barriers vs lock/unlock

[Peter Dimov]

Paolo Carlini wrote:
> Paolo Carlini wrote:
>
>> Don't pay too much attention to those, for now. Likely are due to a
>> MT bug affecting old glibcs
>> (http://sources.redhat.com/ml/libc-hacker/2004-07/msg00034.html)
>>
>>
> Actually, there is a real issue, on machines using up to date glibcs.
>
> I filed other/24757, about this.

Have you tried calling _S_initialize before spawning the threads? If the 
problem disappears, then _S_initialize isn't thread safe. If it persists... 
I'm out of ideas :-)