Re: condvar performance in .59 vs .60

Jamie Lokier wrote:

> There is a known bug in 2.6.0-test6 with FUTEX_REQUEUE, but I wonder
> if it's the same bug you're thinking of that causes v.60 to not use
> this operation.

No, has nothing to do with it.  I've seen the problem also on our 2.4
kernels with the nptl backport.  It might still be related or another
kernel problem or a logic problem.  In any case, I've disabled the
requeue use for pthread_cond_signal() for now.  It might be back.


> If there's a conceptual problem and NPTL is not going to use
> FUTEX_REQUEUE any more because of a condvar deadlock, then it may as
> well be removed from the kernel.

Not at all.  Requeue is essential for pthread_cond_broadcast().  Without
it countless of threads would be released at the same time, trying the
get the same resource.  This function, and not pthread_cond_signal, was
the main factor behind the creation of the requeue change.

-- 
--------------.                        ,-.            444 Castro Street
Ulrich Drepper \    ,-----------------'   \ Mountain View, CA 94041 USA
Red Hat         `--' drepper at redhat.com `---------------------------