Re: [Xen-devel] [PATCH] x86/time: use correct (local) time stamp in constant-TSC calibration fast path

[Joao Martins <joao.m.martins@xxxxxxxxxx>]

On 06/10/2016 10:29 AM, Jan Beulich wrote:
>>>> On 10.06.16 at 08:59, <JBeulich@xxxxxxxx> wrote:
>> Actually, since local time gets seeded from platform time in
>> init_percpu_time(), I don't think we can do away with
>> maintaining platform time.
>
Yeah, I agree. But the case of my previous message was towards
improvement potential of the rendezvous after this patch (for
the fast path). Platform time overflow could be another example
but probably a bit borderline as certain clocksources have very
short intervals.

> And it looks like this seeding is where much of the remaining backwards
> deltas are coming from: While on local_time_calibration()'s slow path
> we use the platform time as reference (and hence the seeding is fine),
> on the fast path we don't, and hence using the platform time in
> init_percpu_time() to initialize local stime introduces a discontinuity.

I'll do some testing too on your patch that's addressing this on Monday (today
is an holiday in my country).

Joao

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel