Re: [PATCH v3] arm64: dts: mediatek: Add mt8192 power domains controller

[Enric Balletbo Serra]

Hi Weiyi,

Missatge de Weiyi Lu <weiyi.lu@xxxxxxxxxxxx> del dia dj., 18 de febr.
2021 a les 11:31:
>
> On Mon, 2020-11-30 at 19:16 +0800, Weiyi Lu wrote:
> > On Fri, 2020-11-27 at 13:42 +0100, Matthias Brugger wrote:
> > >
> > > On 19/11/2020 15:13, Enric Balletbo Serra wrote:
> > > > Hi Weiyi,
> > > >
> > > > Missatge de Weiyi Lu <weiyi.lu@xxxxxxxxxxxx> del dia dj., 19 de nov.
> > > > 2020 a les 14:10:
> > > >>
> > > >> On Thu, 2020-11-19 at 13:13 +0100, Enric Balletbo Serra wrote:
> > > >>> Hi Weiyi,
> > > >>>
> > > >>> Thank you for the patch
> > > >>>
> > > >>> Missatge de Weiyi Lu <weiyi.lu@xxxxxxxxxxxx> del dia dj., 19 de nov.
> > > >>> 2020 a les 11:48:
> > > >>>>
> > > >>>> Add power domains controller node for SoC mt8192
> > > >>>>
> > > >>>> Signed-off-by: Weiyi Lu <weiyi.lu@xxxxxxxxxxxx>
> > > >>>> ---
> > > [...]
> > > >>>> +                       /* System Power Manager */
> > > >>>> +                       spm: power-controller {
> > > >>>> +                               compatible = "mediatek,mt8192-power-controller";
> > > >>>> +                               #address-cells = <1>;
> > > >>>> +                               #size-cells = <0>;
> > > >>>> +                               #power-domain-cells = <1>;
> > > >>>> +
> > > >>>> +                               /* power domain of the SoC */
> > > >>>> +                               audio@MT8192_POWER_DOMAIN_AUDIO {
> > > >>>
> > > >>> If you run the dt_bindings_check it should return some errors, as all
> > > >>> these node names should be 'power-domain@'. Which is a bit annoying
> > > >>> because then you will get a bunch of errors like this:
> > > >>>
> > > >>> [    1.969110] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    1.976997] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    1.984828] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    1.992657] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.000685] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.008566] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.016395] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.024221] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.032049] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.039874] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.047699] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.055524] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.063352] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>> [    2.071176] debugfs: Directory 'power-domain' with parent
> > > >>> 'pm_genpd' already present!
> > > >>>
> > > >>> But that's another problem that should be handled in debugfs system.
> > > >>>
> > > >>
> > > >> Indeed...so I chose to use different name in dts to avoid problems in
> > > >> debugfs. It does violate the naming rules.
> > > >>
> > > >
> > > > But your binding will not pass (or trigger warnings) the dtb check
> > > > then. Rob was clear that names should be generic. Proper fix is fix
> > > > debugfs not the binding.
> > > >
> > >
> > > By the way, is anybody working on this debugfs issue?
> > >
> >
> > I think we can solve this problem by adding "name" to the struct
> > scpsys_domain_data and use this domain_data->name as the genpd.name.
> > This is very simple. But I want to know if you both like it?
> >
>
> Hi Enric and Matthias,
>
> May I have your opinions on how you might to fix this issue?
> I'll try to give another name to each power domain in the
> scpsys_domain_data and register power domain with this name like below
>
> struct scpsys_domain_data {
>         ...
> +       char *name;
>  };
>
>
> -       pd->genpd.name = node->name;
> +       pd->genpd.name = pd->data->name ?: node->name;
>
>
> Does it violate the naming rules to some extent? or it's acceptable?
>

I think it could be acceptable, I think doesn't violate any rule. My
doubt here is we should name in a generic way in the form
'power-domain@addr' getting the last part of the full node name to
avoid conflicts or if the name should be driver-specific. I think it
makes sense to be driver-specific as is more helpful for debugging
purposes. If we do driver-specific (with data->name) I'd fail if is
not supplied.

Thanks,
  Enric

> > > Regards,
> > > Matthias
> >
> >
> > _______________________________________________
> > Linux-mediatek mailing list
> > Linux-mediatek@xxxxxxxxxxxxxxxxxxx
> > http://lists.infradead.org/mailman/listinfo/linux-mediatek
>