[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] Re: [PATCH] CPUIDLE: revise tsc-save/restore to avoid big tsc skew between cpus

On 05/12/2008 11:30, "Wei, Gang" <gang.wei@xxxxxxxxx> wrote:

>> I tried extrapolating from t->stime_local_stamp, cpu_khz, and
>> t->local_tsc_stamp before I got into the current solution. It would still
>> bring accumulating skew, but in a slower increasing speed. I would like to
>> try it again with  t->tsc_scale instead of cpu_khz. If it is works, it would
>> really be simpler. Allow me some time.
> Below patch should be what you expected. It will still bring continuously
> increasing tsc skew. If I pin all domains' vcpus on 1 pcpu, the skew is
> increasing faster.

It looks about right to me, and better than using cpu_khz as in the current
code and the original patch. How much skew does it introduce? Is the skew a
problem? Bearing in mind that from the point of view of HVM guests, TSC
rates will be all over the place anyway if we are using P states (and the
host TSC is not invariant, which is actually the only time your code is
enabled anyway).

 -- Keir

Xen-devel mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.