On Thu, Aug 16, 2012 at 14:00, Mark Kettenis wrote:
> A whitelist-based approach is probably best. What you could do is
> attach this timecounter on platforms that are not whitelisted as well,
> but give it a fairly low tc_quality if it isn't whitelisted. That way
> people can still select the TS
> Date: Wed, 15 Aug 2012 13:25:55 -0400
> From: Ted Unangst
>
> So for more than a decade, the rdtsc instruction has in theory been
> the fastest most accurate way to measure elapsed time on x86. Except
> when it doesn't quite work.
It's not obvious though that it is the most stable clock in th
On Aug 15, 2012, at 8:17 PM, Ted Unangst wrote:
> On Wed, Aug 15, 2012 at 13:25, Ted Unangst wrote:
>
> I will probably rename this just "tsc" after some prodding from mikeb,
> that's a better name. I tend to focus on the instruction used, but we
> should name it after the counter.
>
>> +r
On Wed, Aug 15, 2012 at 13:25, Ted Unangst wrote:
I will probably rename this just "tsc" after some prodding from mikeb,
that's a better name. I tend to focus on the instruction used, but we
should name it after the counter.
> + rdtsc_timecounter.tc_frequency = cpuspeed;
> + /* cpuspeed
So for more than a decade, the rdtsc instruction has in theory been
the fastest most accurate way to measure elapsed time on x86. Except
when it doesn't quite work.
The main issues are 1) frequency changing as a result of power
management, and 2) different values on different CPUs in SMP systems.
