Re: Spinlocks, yet again: analysis and proposed patches

On Tue, 13 Sep 2005 Tom Lane wrote :
> "Min Xu (Hsu)" <xu(at)cs(dot)wisc(dot)edu> writes:
> > ...If this were the case, perhaps first fetch the spin lock with read-only
> > permission should have helped.
>
> But the cmpb instruction in the 8.0 version of TAS would have done that,
> and I think we've already established that the cmpb is a loss on most
> machines (except maybe single-physical-CPU Xeons). I suggested in my
> other message that it might help to grab write permission on the cache
> line before actually trying to acquire the spin lock --- but I don't see
> how getting read-only permission first will help.

Yes, I agree. What I was trying to say was that if the second scenario
you hypothesize were true, i.e. fetching write-able line before actually
trying to acquire the spin lock would cause another processor to slow
down its execution inside the critical section, then fetching read-only
lines would have helped. As you said, however, experimental results
shows fetching read-only lines didn't help, which led me wonder whether the
second scenario your described was really happening.