Re: [PATCH] perf events: Add stalled cycles generic event - PERF_COUNT_HW_STALLED_CYCLES

[Ingo Molnar <mingo@elte.hu>]

* Ingo Molnar <mingo@elte.hu> wrote:

> > [...] why not pick cycles where no uops are retiring?
> > 
> > cycles_no_uops_retired = cycles - c["UOPS_RETIRED:ANY:c=1:t=1"]
> >
> > In the presence of C-states and some halted cycles, I found that I couldn't 
> > measure it via UOPS_RETIRED:ANY:c=1:i=1 because it counts halted cycles too 
> > and could be greater than (unhalted) cycles.
> 
> Agreed, good point.
> 
> You are right that it is more robust to pick 'the CPU was busy on our behalf' 
> metric instead of a 'CPU is idle' metric, because that way 'HLT' as a special 
> type of idling around does not have to be identified.

Sidenote, there's one advantage of the idle event: it's more meaningful to 
profile idle cycles - and it's easy to ignore the HLT loop in the profile 
output (we already do).

That way we get a 'hidden overhead' profile: a profile of frequently executed 
code which executes in the CPU in a suboptimal way.

So we should probably offer both events.

Thanks,

	Ingo