After much tracing through the code, I found out why the Apache DSO
support doesn't work on NetBSD/pmax.  The master httpd process loads all the
shared objects just fine, but the server process it forks performs an internal
re-start which does a dlclose() on all the modules, then tries to dlopen()
them again.  Apparently, ld.elf_so doesn't allow a process to dlopen() a
shared object again after a dlclose() on it.
    I've written a simple test program which shows the same behavior.  All it
does is dlopen() a library, dlclose() it, and tries to dlopen() it again.  It
locks up on the second dlopen().  Is this a known behavior, or should I send a
pr on it?