Re: [RFC][PATCH RT 0/2] futex priority based wakeup

["Sébastien Dugué" <sebastien.dugue@bull.net>]

On Tue, 2006-05-16 at 06:36 -0400, Jakub Jelinek wrote:
> On Thu, May 11, 2006 at 10:56:57AM +0200, S?bastien Dugu? wrote:
> >   Hm, I wonder what would be the effect of having the external mutex
> > and __data.__lock use a PI futex and __data.__futex use a regular
> > futex when the waiters on __data.__futex are requeued on the external
> > mutex during a broadcast.
> 
> Well, either glibc can stop using requeue if the mutex is PI mutex (and use
> the slower fallback), or kernel would need to handle requeueing from normal to
> PI futex.

  Yes, or make the condvar use a PI futex and implement a 
futex_requeue_pi() and have glibc use that in the broadcast case. I 
think the requeue thing is a good idea and should not be dropped.

> 
> > > > But, there is a problem here - pthread_cond_{signal,broadcast} don't
> > > > have any associated mutexes, so you often don't know which type
> > > > of protocol you want to use for the internal condvar lock.
> > 
> >   Just a wild guess here, but in the broadcast or signal path, couldn't
> > __data.__mutex be looked up to determine what protocol to use for the
> > __data.__futex?
> 
> Not always.
> Say if you do:
> thread1 (low prio)	thread2 (very high prio)		thread3 (mid prio)
> pthread_cond_signal (&cv)
> # first use of cv in the program, no mutex has been ever used with this
> # condvar
> lll_mutex_lock (&cv->__data.__lock)
> 			pthread_cond_wait (&cv, &pi_mutex)
> 			lll_mutex_lock (&cv->__data.__lock)
> 								use_all_CPU
> 			# Then thread2 is stuck, waiting on thread1 which waits on thread3
> 
> At pthread_cond_signal enter time you don't know the type of associated
> mutex, so you don't know which kind of internal lock to use.
> It doesn't have to be the first use of cv in the program, similarly it can
> be any pthread_cond_{signal,broadcast} called while there are no threads
> in pthread_cond_*wait on that cv.

  OK, it makes sense now, thanks.

  Sébastien.