Re: [PATCH 1/6] arm64: dts: qcom: sm8250: drop incorrect domain idle states properties

[Dmitry Baryshkov]

On 29/03/2023 12:45, Ulf Hansson wrote:
> On Wed, 29 Mar 2023 at 00:51, Dmitry Baryshkov
> <dmitry.baryshkov@xxxxxxxxxx> wrote:
> > On 24/03/2023 09:38, Krzysztof Kozlowski wrote:
> > > Domain idle states do not use 'idle-state-name' and 'local-timer-stop':
> > >
> > >     sm8250-hdk.dtb: domain-idle-states: cluster-sleep-0: 'idle-state-name', 'local-timer-stop' do not match any of the regexes: 'pinctrl-[0-9]+'
> > >
> > > Reported-by: Neil Armstrong <neil.armstrong@xxxxxxxxxx>
> > > Link: https://lore.kernel.org/all/20230323-topic-sm8450-upstream-dt-bindings-fixes-v1-4-3ead1e418fe4@xxxxxxxxxx/
> > > Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@xxxxxxxxxx>
> > > ---
> > >    arch/arm64/boot/dts/qcom/sm8250.dtsi | 2 --
> > >    1 file changed, 2 deletions(-)
> > >
> > > diff --git a/arch/arm64/boot/dts/qcom/sm8250.dtsi b/arch/arm64/boot/dts/qcom/sm8250.dtsi
> > > index 79d67b466856..9cf2de87c632 100644
> > > --- a/arch/arm64/boot/dts/qcom/sm8250.dtsi
> > > +++ b/arch/arm64/boot/dts/qcom/sm8250.dtsi
> > > @@ -354,12 +354,10 @@ BIG_CPU_SLEEP_0: cpu-sleep-1-0 {
> > >                domain-idle-states {
> > >                        CLUSTER_SLEEP_0: cluster-sleep-0 {
> > >                                compatible = "domain-idle-state";
> > > -                             idle-state-name = "cluster-llcc-off";
> > >                                arm,psci-suspend-param = <0x4100c244>;
> > >                                entry-latency-us = <3264>;
> > >                                exit-latency-us = <6562>;
> > >                                min-residency-us = <9987>;
> > > -                             local-timer-stop;
> >
> > Hmm, so we support setting the broadcast timer when using plain PSCI
> > idle states, but not when using the domain-based idle states.
> >
> > Ulf, Rafael, Daniel, is that an omission for the domain-based idle
> > support? Or is it handled in some other way?
>
> I am not sure that we need a DT binding specifically for this, or do we?
>
> So far, the timer is managed from platform specific code. For some
> Qcom based platforms, the timer should be managed in
> rpmh_rsc_write_next_wakeup(), which makes use of
> dev_pm_genpd_get_next_hrtimer().

I'm not sure whether I fully follow this (I might lack some 
understanding here). I thought that with "local-timer-stop" (well, 
CPUIDLE_FLAG_TIMER_STOP), we were switching to the broadcast timer 
before a core/cluster goes into idle state with no local timer, while 
with  rpmh_rsc_write_next_wakeup() we write that only before shutting 
down the last core (see the comment at rpmh_flush()).

This might be an expected thing, I probably don't have enough 
understanding of the cpuidle internals to judge this.

> [...]
>
> Kind regards
> Uffe

--
With best wishes
Dmitry