Re: [clocksource] 8901ecc231: stress-ng.lockbus.ops_per_sec -9.5% regression

[Paul E. McKenney]

On Fri, Aug 06, 2021 at 10:10:00AM +0800, Chao Gao wrote:
> On Thu, Aug 05, 2021 at 08:37:27AM -0700, Paul E. McKenney wrote:
> >On Thu, Aug 05, 2021 at 01:39:40PM +0800, Chao Gao wrote:
> >> [snip]
> >> >> This patch works well; no false-positive (marking TSC unstable) in a
> >> >> 10hr stress test.
> >> >
> >> >Very good, thank you!  May I add your Tested-by?
> >> 
> >> sure.
> >> Tested-by: Chao Gao <chao.gao@xxxxxxxxx>
> >
> >Very good, thank you!  I will apply this on the next rebase.
> >
> >> >I expect that I will need to modify the patch a bit more to check for
> >> >a system where it is -never- able to get a good fine-grained read from
> >> >the clock.
> >> 
> >> Agreed.
> >> 
> >> >And it might be that your test run ended up in that state.
> >> 
> >> Not that case judging from kernel logs. Coarse-grained check happened 6475
> >> times in 43k seconds (by grep "coarse-grained skew check" in kernel logs).
> >> So, still many checks were fine-grained.
> >
> >Whew!  ;-)
> >
> >So about once per 13 clocksource watchdog checks.
> >
> >To Andi's point, do you have enough information in your console log to
> >work out the longest run of course-grained clocksource checks?
> 
> Yes. 5 consecutive course-grained clocksource checks. Note that
> considering the reinitialization after course-grained check, in my
> calculation, two course-grained checks are considered consecutive if
> they happens in 1s(+/- 0.3s).

Very good, thank you!

So it seems eminently reasonable to have the clocksource watchdog complain
bitterly for more than (say) 100 consecutive course-grained checks.

I am thinking in terms of a separate patch for this purpose.

Thoughts?

							Thanx, Paul