Re: [generalized cache events] Re: [PATCH 1/1] perf tools: Add missing user space support for config1/config2

[Ingo Molnar <mingo@elte.hu>]

* Andi Kleen <ak@linux.intel.com> wrote:

> One example of this we had recently in the kernel: 
> 
> function accesses three global objects. Scalability tanks when the test is 
> run with more CPUs.  Now the hit is near the three memory accesses. Which one
> is the one that is actually bouncing cache lines?

that's not an example - you are still only giving vague, untestable, 
unverifiable references. You need to give us something specific and 
reproducible - preferably a testcase.

Peter and me are doing lots of scalability work in the core kernel and for most 
problems i've met it was enough if we knew the function name - the scalability 
problem is typically very obvious from that point on - and an annotated profile 
makes it even more obvious.

I've never met a situation what you describe, that it was not possible to 
disambiguate a real SMP bounce - and i've been fixing SMP bounces in the kernel 
for over ten years.

So you really will have to back up your point with an accurate, reproducible 
testcase - vague statements like the ones you are making i do not accept at 
face value, sorry.

Thanks,

	Ingo