Re: [PATCH -v8][RFC] mutex: implement adaptive spinning

[Nick Piggin <npiggin@suse.de>]

On Wed, Jan 14, 2009 at 05:46:39PM +0100, Peter Zijlstra wrote:
> On Mon, 2009-01-12 at 19:32 +0200, Avi Kivity wrote:
> > Peter Zijlstra wrote:
> > > Spinlocks can use 'pure' MCS locks.
> > >   
> > 
> > How about this, then.  In mutex_lock(), keep wait_lock locked and only 
> > release it when scheduling out.  Waiter spinning naturally follows.  If 
> > spinlocks are cache friendly (are they today?) 
> 
> (no they're not, Nick's ticket locks still spin on a shared cacheline
> IIRC -- the MCS locks mentioned could fix this)

It reminds me. I wrote a basic variation of MCS spinlocks a while back. And
converted dcache lock to use it, which showed large dbench improvements on
a big machine (of course for different reasons than the dbench improvements
in this threaed).

http://lkml.org/lkml/2008/8/28/24

Each "lock" object is sane in size because given set of spin-local queues may
only be used once per lock stack. But any spinlocks within a mutex acquisition
will always be at the bottom of such a stack anyway, by definition.

If you can use any code or concept for your code, that would be great.

 
> > we inherit that.  If 
> > there is no contention on the mutex, then we don't need to reacquire the 
> > wait_lock on mutex_unlock() (not that the atomic op is that expensive 
> > these days).
> 
> That might actually work, although we'd have to move the
> __mutex_slowpath_needs_to_unlock() branch outside wait_lock otherwise
> we'll deadlock :-)
> 
> It might be worth trying this if we get serious fairness issues with the
> current construct.