It looks like TLS implementation may rely on extensive work done
during dynamic link stage (search for "linker" in the document
http://people.redhat.com/drepper/tls.pdf). Is there some way to force
gcc or the linker to generate code that uses static TLS allocation
(TLS being "thread local storage")? A quick glance at the gcc man page
shows the option

        -ftls-model=<model>

which might give some ability to control this? But it looks like the
most promising option might already be the default? I've not done
anything with multiple threads in a C/C++ program so I'm out of my
depth here.

Dave W