[PATCH 3/4] cputime/powerpc/s390: make scaled cputime arch specific

Martin Schwidefsky schwidefsky at de.ibm.com
Thu Nov 3 08:22:00 AEDT 2016


On Wed, 2 Nov 2016 10:38:20 +0100
Stanislaw Gruszka <sgruszka at redhat.com> wrote:

> On Wed, Nov 02, 2016 at 10:11:22AM +0100, Christian Borntraeger wrote:
> > On 10/31/2016 01:36 PM, Stanislaw Gruszka wrote:
> > > Only s390 and powerpc have hardware facilities allowing to measure
> > > cputimes scaled by frequency. On all other architectures
> > > utimescaled/stimescaled are equal to utime/stime (however they are
> > > accounted separately).
> > > 
> > > Patch remove {u,s}timescaled accounting on all architectures except
> > > powerpc and s390, where those values are explicitly accounted on proper
> > > places.
> > 
> > If we remove it everywhere else (and assuming that there are no users then)
> > I aks myself if we should remove this as well from s390.
> 
> There is one user of scaled cputimes values, it is taskstats (to users
> space are exported ac_utimescaled, ac_stimescaled and
> cpu_scaled_run_real_total which is calculated based on scaled times).
> However on other than powerpc and s390 architectures scaled times are
> equal to normal times (this is also true for older powerpc's without
> SPURR/PURR registers).

The taskstats interface is the only user of the scaled cputime that
I am aware of. It is hard to say how valuable the information is.
Without the scaled number in the taskstats inferface you have no means
to detect if your workload has been affected by a another SMT thread
running on the same core.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.



More information about the Linuxppc-dev mailing list