DragonFlyBSD

I've got an skeleton prototype roughly similar to the fbsd implementation.

Semaphore names are regular files. Each semaphore has a system-wide state, basically its value, stored in a shared memory region. There is also a per-process state for dealing with multiple sem_open() calls.

The semaphore value is manipulated via atomic instructions. sem_wait() calls into the kernel (umtx_sleep()) if and only if the calling thread must sleep (semaphore value < 0). Similarly sem_post() calls into the kernel (umtx_wakeup()) if and only if there are waiters (value 0 >1). umtx_sleep() could be replaced in a subsequent step with a semaphore-optimized syscall for avoiding user<>kernel transitions when several threads concurrently call sem_wait() with value <= 0.

I'll post the code when fleshed out and cleaned up.

Implemented by commit 1a09a5c7d3dc38e61fc28441d066b071ad659dc0