Re: [BUG] futex_unlock_pi() hurts my brain and may cause application deadlock

[Ulrich Drepper]

On 5/30/07, Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> >       if (!(uval & FUTEX_OWNER_DIED)) {
> >               pagefault_disable();
> >               uval = futex_atomic_cmpxchg_inatomic(uaddr, current->pid, 0);
> >               pagefault_enable();
> >       }
> [...]
> This code is in futex_unlock_pi.  Can the owner of the mutex really die?
> Isn't the owner the one doing the unlock?

This is part of the implementation of robust PI futexes.  The
semantics is that if the owner of a robust futex dies the futex is
marked with FUTEX_OWNER_DIED.  The next locking thread then has to
clear that bit before calling pthread_mutex_unlock.  If the bit is
still set while unlocking the mutex is permanently marked unusable.  I
cannot say right now whether this is the semantics implemented above.
I'll have to check the glibc test suite whether we check for that
semantic.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/