Re: Question about ktime_get_mono_fast_ns() non-monotonic behavior

[Yosry Ahmed]

On Thu, Oct 13, 2022 at 7:23 PM John Stultz <jstultz@xxxxxxxxxx> wrote:
>
> On Thu, Oct 13, 2022 at 7:15 PM John Stultz <jstultz@xxxxxxxxxx> wrote:
> >
> > On Thu, Oct 13, 2022 at 6:29 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Oct 13, 2022 at 5:03 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
> > > >
> > > > On Wed, Oct 12, 2022 at 8:07 PM John Stultz <jstultz@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Wed, Oct 12, 2022 at 8:02 PM John Stultz <jstultz@xxxxxxxxxx> wrote:
> > > > > > On Mon, Sep 26, 2022 at 2:18 PM Yosry Ahmed <yosryahmed@xxxxxxxxxx> wrote:
> > > > > > So I think it reasonable to say its bounded by approximately  2 *
> > > > > > NSEC_PER_SEC/HZ +/- 11%.
> > > > >
> > > > > Sorry, this should be 2*NSEC_PER_SEC/HZ * 0.11
> > > >
> > > > Thanks so much for the detailed response :)
> > > >
> > > > IIUC this error bound is in ns. So on a 2 GHz cpu the bound is 0.11 ns
> > > > (essentially 0)? I feel like I miscalculated, this error bound is too
> > > > good to be true.
> > >
> > > Never mind, I thought HZ is the cpu speed for some reason. It's the
> > > number of jiffies per second, right?
> >
> > Correct.
> >
> > > So if HZ is 1000, the error bound is actually ~2 ms, which is very
> > > large considering that the unit is ns.
> >
> > Uh, for HZ=1000, I think it's closer to 220us, but yes, for HZ=100 2.2ms.
>
> And again, it has been awhile since I've been deep in this code, so
> I'd not be surprised if I'm missing something and the worst case may
> be larger (things like SMIs or virtualization stalling the timekeeping
> update for longer than a tick).  So no promises, but this feels pretty
> close to the expected bound. If you can't handle time inconsistencies,
> you need to use the normal locked accessors.
>

Thanks a lot for the clarification. The inconsistencies may be too
large, depending on what we are trying to measure, so it is something
that we need to keep in mind.

> thanks
> -john