this implementation is rather heavy-weight, but it's the first
solution i've found that's actually correct. all waiters actually wait
twice at the barrier so that they can synchronize exit, and they hold
a "vm lock" that prevents changes to virtual memory mappings (and
blocks pthread_barrier_destroy) until all waiters are finished
inspecting the barrier.

thus, it is safe for any thread to destroy and/or unmap the barrier's
memory as soon as pthread_barrier_wait returns, without further
synchronization.

per POSIX: The mprotect() function shall change the access protections
to be that specified by prot for those whole pages containing any part
of the address space of the process starting at address addr and
continuing for len bytes.

on the other hand, linux mprotect fails with EINVAL if the base
address and/or length is not page-aligned, so we have to align them
before making the syscall.

the check against MADV_DONTNEED to because linux MADV_DONTNEED
semantics conflict dangerously with the POSIX semantics

- hide all the legacy xxxxxx32 name cruft in syscall.h so the actual
source files can be clean and uniform across all archs.

- cleanup llseek/lseek and mmap2/mmap handling for 32/64 bit systems

- alternate implementation for nice if the target lacks nice syscall