Re: [REGRESSION] x86, perf: counter freezing breaks rr

[Peter Zijlstra]

On Tue, Nov 20, 2018 at 12:11:44PM -0800, Andi Kleen wrote:
> > > > Given that we're already at rc3, and that this renders rr unusable,
> > > > we'd ask that counter freezing be disabled for the 4.20 release.
> > >
> > > The boot option should be good enough for the release?
> > 
> > I'm not entirely sure what you mean here. We want you to flip the
> > default boot option so this feature is off for this release. i.e. rr
> > should work by default on 4.20 and people should have to opt into the
> > inaccurate behavior if they want faster PMI servicing.
> 
> I don't think it's inaccurate, it's just different 
> than what you are used to.
> 
> For profiling including the kernel it's actually far more accurate
> because the count is stopped much earlier near the sampling
> point. Otherwise there is a considerable over count into
> the PMI handler.
> 
> In your case you limit the count to ring 3 so it's always cut off
> at the transition point into the kernel, while with freezing
> it's at the overflow point.

Ooh, so the thing does FREEZE_ON_OVERFLOW _not_ FREEZE_ON_PMI. Yes, that
can be a big difference.

See, FREEZE_ON_PMI, as advertised by the name, should have no observable
effect on counters limited to USR. But something like FREEZE_ON_OVERFLOW
will loose everything between the overflow and the eventual PMI, and by
freezing early we can't even compensate for it anymore either,
introducing drift in the period.

And I don't buy the over-count argument, the counter register shows how
far over you are; it triggers the overflow when we cross 0, it then
continues counting. So if you really care, you can throw away the
'over-count' at PMI time. That doesn't make it more reliable. We don't
magically get pt_regs from earlier on or any other state.

The only thing where it might make a difference is if you're running
multiple counters (groups in perf speak) and want to correlate the count
values. Then, and only then, does it matter.

Bah.