On Sat, 31 Mar 2012, Andi Kleen wrote:

quoted

So if a guest exits due to an external event it's easy to inspect the
state of that guest and avoid to schedule away when it was interrupted
in a spinlock held section. That guest/host shared state needs to be

On a large system under high contention sleeping can perform surprisingly
well. pthread mutexes have a tendency to beat kernel spinlocks there.
So avoiding sleeping locks completely (that is what pv locks are
essentially) is not necessarily that good.

Care to back that up with numbers and proper trace evidence instead of
handwaving?

I've stared at RT traces and throughput problems on _LARGE_ machines
long enough to know what I'm talking about and I can provide evidence
in a split second.

Your proposal is probably only a good idea on low contention
and relatively small systems.

Sigh, you have really no fcking clue what you are talking about.

On RT we observed scalabilty problems way before hardware was
available to expose them. So what's your point?

Put up or shut up, really!

    tglx