On Sat, Oct 07, 2017 at 10:27 +0000, Adam Steen wrote:
> On Sat, Oct 7, 2017 at 5:52 PM, Adam Steen <a...@adamsteen.com.au> wrote:
>
> > On Fri, Oct 06, 2017 at 03:58:18PM +0200, Mike Belopuhov wrote:
> > > Hi,
> > >
> > > An experimental change to use TSC as a timecounter source on a variety
> > > of modern Intel and AMD CPUs has been just committed and enabled on
> > > OpenBSD/amd64 thanks to the work done by Adam Steen.
> > >
> > > The rationale is, quoting the commit message:
> > >
> > > If frequency of an invariant (non-stop) time stamp counter is measured
> > > using an independent working timecounter that has a known frequency, we
> > > can assume that the measured TSC frequency is as good as the resolution
> > > of the timecounter that we use to perform the measurement. This lets us
> > > switch from this high quality but expensive source to the cheaper TSC
> > > without sacrificing precision on a wide range of modern CPUs.
> > >
> > > You can query and change the current timecounter source in the runtime
> > > via sysctl:
> > >
> > > % sysctl kern.timecounter.{choice,hardware}
> > > kern.timecounter.choice=i8254(0) tsc(2000) acpihpet0(1000)
> > acpitimer0(1000) dummy(-1000000)
> > > kern.timecounter.hardware=tsc
> > >
> > > Please make sure your NTP drift (/var/db/ntpd.drift) stays within
> > -20..+20
> > > or at least is not worse than it is right now.
> > >
> > > And finally, please make sure to run a "make config" when building the
> > > kernel to update offset tables because of the cpu_info structure changes.
> > >
> > > Regards,
> > > Mike
> > >
> >
> > Hi,
> >
> > Now that we have an accurate tsc frequency, I would like to expose this
> > information to userland via a sysctl.
> >
> > The diff below exposes the tsc frequency and if it is invariant.
> >
> > Cheers,
> > Adam
> >
> >
> Please ignore that diff, looks like i had some dregs from a older diff i
> had been testing
Yeah, I've renamed a few things and deliberately skipped the sysctl part
so that we can test for regressions and don't mess with exposed interfaces
if we need to back out. So I'd wait a few weeks for people to get it
running on a variety of systems and report the fallout if any.
