Re: [PATCH 7/7] ARM: dts: add OPPs for i.MX23/i.MX28

[Stefan Wahren <stefan.wahren@i2se.com>]

Hi Juergen,

> Juergen Borleis <juergen@kreuzholzen.de> hat am 29. März 2015 um 16:40
> geschrieben:
>
>
> Hi Stefan,
>
> Stefan Wahren wrote:
> > > Juergen Borleis <juergen@kreuzholzen.de> hat am 24. März 2015 um 21:45
> > > geschrieben:
> > >
> > > Stefan Wahren wrote:
> > > > [...]
> > > > diff --git a/arch/arm/boot/dts/imx28.dtsi
> > > > b/arch/arm/boot/dts/imx28.dtsi index 98c1be6..21c1921 100644
> > > > --- a/arch/arm/boot/dts/imx28.dtsi
> > > > +++ b/arch/arm/boot/dts/imx28.dtsi
> > > > @@ -38,12 +38,23 @@
> > > > };
> > > >
> > > > cpus {
> > > > - #address-cells = <0>;
> > > > + #address-cells = <1>;
> > > > #size-cells = <0>;
> > > >
> > > > - cpu {
> > > > + cpu@0 {
> > > > compatible = "arm,arm926ej-s";
> > > > device_type = "cpu";
> > > > + reg = <0x0>;
> > > > + operating-points = <
> > > > + /* kHz uV */
> > > > + 261819 1350000
> > > > + 360000 1350000
> > > > + 392728 1450000
> > > > + 454737 1550000
> > > > + >;
> > > > + clocks = <&clks 4>;
> > > > + clock-latency = <61036>; /* two CLK32 periods */
> > > > + cpu-supply = <&reg_vddd>;
> > > > };
> > > > };
> > >
> > > Maybe you should take into account not to reduce VDD below 1.55 V if the
> > > SDRAM controller runs above 196 MHz. The i.MX28 datasheet[1] lists these
> > > restrictions. VDD powers the SDRAM controller as well. From the datasheet
> > > the table "Frequency versus Voltage for EMICLK" shows:
> > >
> > > EMICLK Fmax (MHz)
> > > VDDD (V) DDR2 mDDR
> > > --------------------------------
> > > 1.550 205.71 205.71
> > > 1.450 196.36 196.36
> > > 1.350 196.36 196.36
> > >
> > > jbe
> > >
> > > [1]
> > > i.MX28 Applications
> > > Processors for Consumer
> > > Products
> > > Silicon Version 1.2
> > >
> > > Document Number: IMX28CEC
> > > Rev. 3, 07/2012
> >
> > the only chance that i see to meet this constraint without introducing a
> > new cpufreq driver is to change clock ref_cpu
> > and ref_emi.
> >
> > Do you think this solution of a virtual clock [1] can be applied here, too?
>
> This would imply we can change the SDRAM timing at run-time on the fly. IMHO
> that is not possible (at least not easy).

it's possible since the cpufreq driver in the FSL can do it. But the code isn't
nice and contains some workarounds.

> I think when the driver detects the SDRAM controller runs above 196 MHz the
> clock for the CPU core can still be
> changed, but VDD must be kept at 1.55 V.

The cpufreq-dt doesn't know anything about SDRAM controller and i think it isn't
necessary if we use 1.55 V for all
operating points.

>
> Juergen

Stefan