Re: [PATCH 1/2] ARM: dts: rockchip: increase vdd_arm voltage for rk3066a-rayeager board

[Heiko Stuebner <heiko-4mtYJXux2i+zQB+pC5nmwQ@public.gmane.org>]

Hi Andy, Eddie,

Am Freitag, 15. Januar 2016, 09:11:16 schrieb Andy Yan:
> On 2016年01月14日 21:50, Eddie Cai wrote:
> > Hi Heiko
> > 
> > I don't want open source version kernel support too high frequency.
> > Many people use RK3066/PX2 on their board. We don't know if their
> > layout is good enough to support higher frequency. ie. bad layout will
> > cause more voltage drop on the VDD_CPU power rail. So it may need
> > higher voltage. I think stable is more important here. People want
> > higher frequency can add more opp depend on their board's layout.
> > 
> > 2016-01-14 21:20 GMT+08:00 Andy Yan <andy.yan@rock-chips.com>:
> >> Hi Heiko:
> >> 
> >> On 2016年01月14日 20:45, Heiko Stuebner wrote:
> >>> Hi Andy,
> >>> 
> >>> Am Donnerstag, 14. Januar 2016, 20:34:20 schrieb Andy Yan:
> >>>> The current vdd_arm voltage is too low, increase it will make
> >>>> the system more stable.
> >>>> 
> >>>> Signed-off-by: Andy Yan <andy.yan@rock-chips.com>
> >>>> ---
> >>>> 
> >>>>    arch/arm/boot/dts/rk3066a-rayeager.dts | 8 ++++++++
> >>>>    1 file changed, 8 insertions(+)
> >>>> 
> >>>> diff --git a/arch/arm/boot/dts/rk3066a-rayeager.dts
> >>>> b/arch/arm/boot/dts/rk3066a-rayeager.dts index 0bf6876..3d4c104
> >>>> 100644
> >>>> --- a/arch/arm/boot/dts/rk3066a-rayeager.dts
> >>>> +++ b/arch/arm/boot/dts/rk3066a-rayeager.dts
> >>>> @@ -159,6 +159,14 @@
> >>>> 
> >>>>    &cpu0 {
> >>>>    
> >>>>          cpu0-supply = <&vdd_arm>;
> >>>> 
> >>>> +       operating-points = <
> >>>> +               /* kHz    uV */
> >>>> +               1008000 1125000
> >>>> +               816000  1125000
> >>>> +               600000  1100000
> >>>> +               504000  1100000
> >>>> +               312000  1075000
> >>>> +       >;
> >>>> 
> >>>>    };
> >>>>    
> >>>>    &emac {
> >>> 
> >>> Does this only apply to the rayeager-board / px2 soc or for all
> >>> rk3066a/px2
> >>> boards? I.e. the original values come from some sdk kernel and I
> >>> might've
> >>> transcribed that wrong at the time and I remember having problems
> >>> trying
> >>> to
> >>> reach higher frequencies as well.
> >>> 
> >>> So could you check what the full operating points looks like for both
> >>> and
> >>> if
> >>> this can move to the rk3066a.dtsi, as both should be able to reach at
> >>> least
> >>> 1.4GHz / 1.6GHz.
> >>> 
> >>> 
> >>> Thanks
> >>> Heiko
> >>> 
> >>> ______
> >>> 
> >>      I have checked the sdk kernel, the opp for the sdk board is like 
bellow:
> >>   /**
> >>   * dvfs_cpu_logic_table: table for arm and logic dvfs
> >>   * @frequency   : arm frequency
> >>   * @cpu_volt    : arm voltage depend on frequency
> >>   * @logic_volt  : logic voltage arm requests depend on frequency
> >>   * comments     : min arm/logic voltage
> >>   */
> >> 
> >> static struct dvfs_arm_table dvfs_cpu_logic_table[] = {
> >> 
> >>          {.frequency = 252 * 1000,       .cpu_volt = 1075 * 1000,
> >> 
> >> .logic_volt = 1125 * 1000},//0.975V/1.000V
> >> 
> >>          {.frequency = 504 * 1000,       .cpu_volt = 1100 * 1000,
> >> 
> >> .logic_volt = 1125 * 1000},//0.975V/1.000V
> >> 
> >>          {.frequency = 816 * 1000,       .cpu_volt = 1125 * 1000,
> >> 
> >> .logic_volt = 1150 * 1000},//1.000V/1.025V
> >> 
> >>          {.frequency = 1008 * 1000,      .cpu_volt = 1125 * 1000,
> >> 
> >> .logic_volt = 1150 * 1000},//1.025V/1.050V
> >> 
> >>          {.frequency = 1200 * 1000,      .cpu_volt = 1175 * 1000,
> >> 
> >> .logic_volt = 1200 * 1000},//1.100V/1.050V
> >> 
> >>          {.frequency = 1272 * 1000,      .cpu_volt = 1225 * 1000,
> >> 
> >> .logic_volt = 1200 * 1000},//1.150V/1.100V
> >> 
> >>          {.frequency = 1416 * 1000,      .cpu_volt = 1300 * 1000,
> >> 
> >> .logic_volt = 1200 * 1000},//1.225V/1.100V
> >> 
> >>          {.frequency = 1512 * 1000,      .cpu_volt = 1350 * 1000,
> >> 
> >> .logic_volt = 1250 * 1000},//1.300V/1.150V
> >> 
> >>          {.frequency = 1608 * 1000,      .cpu_volt = 1425 * 1000,
> >> 
> >> .logic_volt = 1300 * 1000},//1.325V/1.175V
> >> 
> >>          {.frequency = CPUFREQ_TABLE_END},
> >> 
> >> };
> >> 
> >>   This table is made base on mass stress test. So is a safe opp table
> >>   for
> >> 
> >> most rk3066a base boards. But it some
> >> 
> >>   board is not so well in power supply circuit design, the may also
> >>   need to
> >> 
> >> adjust it.
> >> 
> >>   From the table we see that the logic_voltage need to increase  as the
> >> 
> >> cpufreq increase. But it seems that the current mainline kernel does
> >> not
> >> support adjust the logic voltage when cpufreq change. So we can only
> >> scale the cpufreq to 1.4GHZ by fixed the logic_volt at 1.2V.
> >> 
> >>   If you like, I can move it to rk3066a.dtsi with the max freq at
> >>   1008MHZ, as>> 
> >> I haven't  test to 1.4GHZ  with the mainline kernel. Or give me some
> >> time to test to see if the rayeager board can run to 1.4G stable with
> >> the mainline .
>    I have run Rayeager  board with cpufreq freely change from
> 312MHZ~1416MHZ for 11 hours now, it's stable. Maybe we can set the
> default opp table for rk3066a/PX2 from 312MHZ~1416HZ in current mainline.

Thanks for doing that stress test and making that the default setting would 
be a nice way to go :-) .

I just checked, the other 2 rk3066a boards we currently support (Hayou 
Marsboard and BQ Curie2) use the exact same power-layout (tps65210 + pwm-
regulator for vdd_log), so should be quite save to do.


Heiko
Heiko

_______________________________________________
Linux-rockchip mailing list
Linux-rockchip@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-rockchip