On Wed, Mar 07, 2018 at 08:47:43AM +0100, Geert Uytterhoeven wrote:
> Hi Inami-san,
> 
> On Wed, Mar 7, 2018 at 2:48 AM, Gaku Inami <gaku.inami...@renesas.com> wrote:
> >> From: geert.uytterhoe...@gmail.com [mailto:geert.uytterhoe...@gmail.com] 
> >> On Behalf Of Geert Uytterhoeven
> >> Sent: Monday, March 5, 2018 7:24 PM
> > [snip]
> >>
> >> 411 for the A53 cores sounds a bit low to me, though.
> >> Documentation/devicetree/bindings/arm/cpu-capacity.txt uses 578.
> >>
> >> Perhaps you already took into account the maximum clock frequencies?
> >> According to the binding document, you should not do that (cfr.
> >> "final capacities are 1024 for cluster0 and 446 for cluster1" in the
> >> bindings doc).
> >
> > Thanks for your review.
> >
> > I set 411 for CA53 based on dhrystone measurement and current 
> > implementation.
> >
> > The average in 10 times of measurement as follows:
> >
> >   cpu   max-freq   dhrystone
> >   ---------------------------------
> >   A57   1500 MHz  15532585 lps/s
> >   A53   1200 MHz   6241541 lps/s
> >
> > With the value of CA57 is scaled at 1024, I end up with 411 for CA53.
> > However, since cpufreq is not available on renesas-devel-20180212-v4.16-rc1,
> > the final capacity is set by directly using capacity-dmips-mhz in dt as 
> > below.
> 
> Documentation/devicetree/bindings/arm/cpu-capacity.txt:
> 
> "capacity-dmips-mhz is an optional cpu node [1] property: u32 value
>  representing CPU capacity expressed in normalized DMIPS/MHz. At boot time, 
> the
>  maximum frequency available to the cpu is then used to calculate the capacity
>  value internally used by the kernel."
> 
> IIUIC, you should thus not use 15532585 and 6241541 directly, but
> scale them to the frequency.
> 
> So the formula for CA53 becomes:
> 
> 1024 / (15532585 / 1500) * (6241541 / 1200) = 514
> 
> > Considering cpufreq is available later, is it better to set the value(514)
> > for CA53 scaled by different maximum frequencies?
> 
> DT describes the hardware, not software limitations, so IMHO 514 is the 
> correct
> value.

Yes agreed. Please refresh this patchset accordingly.

Reply via email to