| Message ID | 20220512071858.10805-2-qianfanguijin@163.com (mailing list archive) |
|---|---|
| State | New, archived |
| Delegated to: | viresh kumar |
| Headers | show |
| Series | [v3,1/2] ARM: dts: sun8i-r40: add opp table for cpu | expand |
Hi, On Thu, May 12, 2022 at 03:18:58PM +0800, qianfanguijin@163.com wrote: > From: qianfan Zhao <qianfanguijin@163.com> > > sun8i-r40 actived cpufreq feature now, let's add "cpu-supply" node on > board. > > Signed-off-by: qianfan Zhao <qianfanguijin@163.com> > --- > arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts | 4 ++++ > arch/arm/boot/dts/sun8i-r40-feta40i.dtsi | 4 ++++ > arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts | 4 ++++ > arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts | 4 ++++ > 4 files changed, 16 insertions(+) > > diff --git a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > index a6a1087a0c9b..4f30018ec4a2 100644 > --- a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > +++ b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > @@ -113,6 +113,10 @@ &ahci { > status = "okay"; > }; > > +&cpu0 { > + cpu-supply = <®_dcdc2>; > +}; > + This will break bisection on those boards. Indeed, you added the OPPs on the first patch, and if you only apply that patch, the boards in the second patch will be missing their CPU regulator. The kernel will then ramp up the frequency to the highest OPP, but will not change the voltage, resulting in a crash. There's a similar issue for all the boards that don't have a regulator in the first place. The way we worked around this for the other SoCs is to have a DTSI with the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), and only include that DTSI on boards that have a CPU regulator hooked in. Maxime
在 2022/5/13 15:38, Maxime Ripard 写道: > Hi, > > On Thu, May 12, 2022 at 03:18:58PM +0800, qianfanguijin@163.com wrote: >> From: qianfan Zhao <qianfanguijin@163.com> >> >> sun8i-r40 actived cpufreq feature now, let's add "cpu-supply" node on >> board. >> >> Signed-off-by: qianfan Zhao <qianfanguijin@163.com> >> --- >> arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts | 4 ++++ >> arch/arm/boot/dts/sun8i-r40-feta40i.dtsi | 4 ++++ >> arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts | 4 ++++ >> arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts | 4 ++++ >> 4 files changed, 16 insertions(+) >> >> diff --git a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >> index a6a1087a0c9b..4f30018ec4a2 100644 >> --- a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >> +++ b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >> @@ -113,6 +113,10 @@ &ahci { >> status = "okay"; >> }; >> >> +&cpu0 { >> + cpu-supply = <®_dcdc2>; >> +}; >> + > This will break bisection on those boards. Indeed, you added the OPPs on > the first patch, and if you only apply that patch, the boards in the > second patch will be missing their CPU regulator. The kernel will then > ramp up the frequency to the highest OPP, but will not change the > voltage, resulting in a crash. This is a good point and I will merge those two patch. > > There's a similar issue for all the boards that don't have a regulator > in the first place. > > The way we worked around this for the other SoCs is to have a DTSI with > the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), > and only include that DTSI on boards that have a CPU regulator hooked in. Is this really necessary? It seems like every board based on sun8i-r40 have a cpu regulator. > > Maxime
On Fri, May 13, 2022 at 03:48:50PM +0800, qianfan wrote: > > > 在 2022/5/13 15:38, Maxime Ripard 写道: > > Hi, > > > > On Thu, May 12, 2022 at 03:18:58PM +0800, qianfanguijin@163.com wrote: > > > From: qianfan Zhao <qianfanguijin@163.com> > > > > > > sun8i-r40 actived cpufreq feature now, let's add "cpu-supply" node on > > > board. > > > > > > Signed-off-by: qianfan Zhao <qianfanguijin@163.com> > > > --- > > > arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts | 4 ++++ > > > arch/arm/boot/dts/sun8i-r40-feta40i.dtsi | 4 ++++ > > > arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts | 4 ++++ > > > arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts | 4 ++++ > > > 4 files changed, 16 insertions(+) > > > > > > diff --git a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > > > index a6a1087a0c9b..4f30018ec4a2 100644 > > > --- a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > > > +++ b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts > > > @@ -113,6 +113,10 @@ &ahci { > > > status = "okay"; > > > }; > > > +&cpu0 { > > > + cpu-supply = <®_dcdc2>; > > > +}; > > > + > > This will break bisection on those boards. Indeed, you added the OPPs on > > the first patch, and if you only apply that patch, the boards in the > > second patch will be missing their CPU regulator. The kernel will then > > ramp up the frequency to the highest OPP, but will not change the > > voltage, resulting in a crash. > > This is a good point and I will merge those two patch. That's not what I meant to say. Those two patches are great separated. You can invert them though. > > There's a similar issue for all the boards that don't have a regulator > > in the first place. > > > > The way we worked around this for the other SoCs is to have a DTSI with > > the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), > > and only include that DTSI on boards that have a CPU regulator hooked in. > > Is this really necessary? It seems like every board based on sun8i-r40 > have a cpu regulator. This probably won't be the case whenever someone starts a new design, and then they'll face random crashes for no apparent reason, and waste a lot of time in the process. Whereas the alternative is that you would be missing some OPPs, something that is fairly easy to figure out. Maxime
在 2022/5/13 16:15, Maxime Ripard 写道: > On Fri, May 13, 2022 at 03:48:50PM +0800, qianfan wrote: >> >> 在 2022/5/13 15:38, Maxime Ripard 写道: >>> Hi, >>> >>> On Thu, May 12, 2022 at 03:18:58PM +0800, qianfanguijin@163.com wrote: >>>> From: qianfan Zhao <qianfanguijin@163.com> >>>> >>>> sun8i-r40 actived cpufreq feature now, let's add "cpu-supply" node on >>>> board. >>>> >>>> Signed-off-by: qianfan Zhao <qianfanguijin@163.com> >>>> --- >>>> arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts | 4 ++++ >>>> arch/arm/boot/dts/sun8i-r40-feta40i.dtsi | 4 ++++ >>>> arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts | 4 ++++ >>>> arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts | 4 ++++ >>>> 4 files changed, 16 insertions(+) >>>> >>>> diff --git a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >>>> index a6a1087a0c9b..4f30018ec4a2 100644 >>>> --- a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >>>> +++ b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts >>>> @@ -113,6 +113,10 @@ &ahci { >>>> status = "okay"; >>>> }; >>>> +&cpu0 { >>>> + cpu-supply = <®_dcdc2>; >>>> +}; >>>> + >>> This will break bisection on those boards. Indeed, you added the OPPs on >>> the first patch, and if you only apply that patch, the boards in the >>> second patch will be missing their CPU regulator. The kernel will then >>> ramp up the frequency to the highest OPP, but will not change the >>> voltage, resulting in a crash. >> This is a good point and I will merge those two patch. > That's not what I meant to say. Those two patches are great separated. > You can invert them though. haha, invert them, so interesting. good idea. > >>> There's a similar issue for all the boards that don't have a regulator >>> in the first place. >>> >>> The way we worked around this for the other SoCs is to have a DTSI with >>> the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), >>> and only include that DTSI on boards that have a CPU regulator hooked in. >> Is this really necessary? It seems like every board based on sun8i-r40 >> have a cpu regulator. > This probably won't be the case whenever someone starts a new design, > and then they'll face random crashes for no apparent reason, and waste a > lot of time in the process. > > Whereas the alternative is that you would be missing some OPPs, > something that is fairly easy to figure out. How about remove the OPPs which greate that 1.08G in sun8i-r40.dtsi, If some boards want to run at a higher frequency, can add them byself in the board's file. > > Maxime
On Fri, May 13, 2022 at 04:23:20PM +0800, qianfan wrote: > > > > There's a similar issue for all the boards that don't have a regulator > > > > in the first place. > > > > > > > > The way we worked around this for the other SoCs is to have a DTSI with > > > > the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), > > > > and only include that DTSI on boards that have a CPU regulator hooked in. > > > Is this really necessary? It seems like every board based on sun8i-r40 > > > have a cpu regulator. > > This probably won't be the case whenever someone starts a new design, > > and then they'll face random crashes for no apparent reason, and waste a > > lot of time in the process. > > > > Whereas the alternative is that you would be missing some OPPs, > > something that is fairly easy to figure out. > > How about remove the OPPs which greate that 1.08G in sun8i-r40.dtsi, > If some boards want to run at a higher frequency, can add them byself > in the board's file. You did all the work to support and test them already. It's a bit of a waste to do that and not include it. Just do a DTSI like we did for the A64 for example. Maxime
On Mon, May 16, 2022 at 3:53 PM Maxime Ripard <maxime@cerno.tech> wrote: > > On Fri, May 13, 2022 at 04:23:20PM +0800, qianfan wrote: > > > > > There's a similar issue for all the boards that don't have a regulator > > > > > in the first place. > > > > > > > > > > The way we worked around this for the other SoCs is to have a DTSI with > > > > > the OPPs with a frequency higher than what U-Boot boots with (1008MHz?), > > > > > and only include that DTSI on boards that have a CPU regulator hooked in. > > > > Is this really necessary? It seems like every board based on sun8i-r40 > > > > have a cpu regulator. > > > This probably won't be the case whenever someone starts a new design, > > > and then they'll face random crashes for no apparent reason, and waste a > > > lot of time in the process. > > > > > > Whereas the alternative is that you would be missing some OPPs, > > > something that is fairly easy to figure out. > > > > How about remove the OPPs which greate that 1.08G in sun8i-r40.dtsi, > > If some boards want to run at a higher frequency, can add them byself > > in the board's file. > > You did all the work to support and test them already. It's a bit of a > waste to do that and not include it. > > Just do a DTSI like we did for the A64 for example. There's also no guarantee that the board boots up at 1.08G. The board may be set to boot up at a slightly lower frequency / voltage combination. Or maybe the board's supply voltage simply isn't stable enough for sustained high CPU usage at 1.08G. Letting the kernel assume that it is OK to run at some OPP is not a good idea. The boards should explicitly include the default OPP table, or define their own, while adding a proper CPU supply at the same time. Regards ChenYu
diff --git a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts index a6a1087a0c9b..4f30018ec4a2 100644 --- a/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts +++ b/arch/arm/boot/dts/sun8i-r40-bananapi-m2-ultra.dts @@ -113,6 +113,10 @@ &ahci { status = "okay"; }; +&cpu0 { + cpu-supply = <®_dcdc2>; +}; + &de { status = "okay"; }; diff --git a/arch/arm/boot/dts/sun8i-r40-feta40i.dtsi b/arch/arm/boot/dts/sun8i-r40-feta40i.dtsi index 265e0fa57a32..b872b51a346d 100644 --- a/arch/arm/boot/dts/sun8i-r40-feta40i.dtsi +++ b/arch/arm/boot/dts/sun8i-r40-feta40i.dtsi @@ -6,6 +6,10 @@ #include "sun8i-r40.dtsi" +&cpu0 { + cpu-supply = <®_dcdc2>; +}; + &i2c0 { status = "okay"; diff --git a/arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts b/arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts index 6931aaab2382..0eb1990742ff 100644 --- a/arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts +++ b/arch/arm/boot/dts/sun8i-t3-cqa3t-bv3.dts @@ -88,6 +88,10 @@ &ahci { status = "okay"; }; +&cpu0 { + cpu-supply = <®_dcdc2>; +}; + &de { status = "okay"; }; diff --git a/arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts b/arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts index 47954551f573..fdf8bd12faaa 100644 --- a/arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts +++ b/arch/arm/boot/dts/sun8i-v40-bananapi-m2-berry.dts @@ -107,6 +107,10 @@ &ahci { status = "okay"; }; +&cpu0 { + cpu-supply = <®_dcdc2>; +}; + &de { status = "okay"; };