Re: CLONE_DETACHED and exit notification (was user-vm-unlock-2.5.31-A2)

[Jamie Lokier <lk@tantalophile.demon.co.uk>]

Ingo Molnar wrote:
> there are some practical problems with making the notification a futex,
> not a simple flag. Eg. futexes right now do not force any lock-counter
> format upon userspace. Futexes can be used as mutexes, conditional
> variables, read-write locks, all of which have different atomic counter
> formats and uses.

Agreed; futexes are lovely because they are so generic.

> By doing the TID-release notification via a futex the actual format of
> the lock is forced, which is a cleanliness problem. Just writing $0 to
> the TID pointer is a robust thing on the other hand.

Quite.  There is no lock; the the futex is used in its purest form, as
a wait queue.

      TID  = thread is alive
      zero = thread is gone

There's no "lock-counter format", because this isn't a lock -- it's a
wakeup.  There no need for atomicity either, because the listener only
reads, it doesn't write.  I'm not sure if a PROT_SEM word is required
for cache coherency, but if it is, your current implementation requires
one too.

Here's a synchronous thread_join-style waiter; it is architecture-neutral:

	while (tid = *tid_address) != 0)
		retval = sys_futex (tid_address, FUTEX_WAIT, tid, 0);

-- Jamie