[PATCH 0/2] Remove syscon compatible from google,gs101-pmu node

[PATCH 0/2] Remove syscon compatible from google,gs101-pmu node

[Peter Griffin]

Hi folks,

This series removes the syscon compatible from the pmu_system_controller
node, by updating both yaml bindings and device tree.

Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
"syscon" compatible") it is possible to register a regmap without the
syscon compatible in the node.

As noted in that commit, it isn't really correct to claim we are compatible
with syscon, as a MMIO regmap created by syscon driver won't work on gs101.
Removing the syscon compatible means the syscon driver won't ever attempt
to create a mmio regmap.

Currently we rely on exynos-pmu running and registering its regmap at a
very early initcall level, so no mmio created regmap is returned. For
pinctrl driver that runs at the same initcall level as exynos-pmu today we
have a custom exynos_get_pmu_regmap_by_phandle() API that supports
-EPROBE_DEFER.

However with the changes proposed in [1] -EPROBE_DEFER will become
supported in the syscon driver directly making this whole approach more
robust especially in a highly modularized system with other drivers at the
same initcall level. We should also be able to remove the custom API
referenced above.

Technically this is a ABI break but no other platforms are affected. With
an old DT we will have the behaviour of today (rely on early initcall
levels). But once [1] is merged, with a new DT we will benefit from
-EPROBE_DEFER.

regards,

Peter

Link: https://lore.kernel.org/lkml/aQdHmrchkmOr34r3@stanley.mountain/ [1]

Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
---
Peter Griffin (2):
      dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu
      arm64: dts: exynos: gs101: remove syscon compatible from pmu node

 Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml | 3 ++-
 arch/arm64/boot/dts/exynos/google/gs101.dtsi                  | 2 +-
 2 files changed, 3 insertions(+), 2 deletions(-)
---
base-commit: 72fb0170ef1f45addf726319c52a0562b6913707
change-id: 20251102-remove-pmu-syscon-compat-dbbc492a4eea

Best regards,
-- 
Peter Griffin <peter.griffin@linaro.org>

[PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Peter Griffin]

Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
"syscon" compatible") it is possible to register a regmap without the
syscon compatible in the node.

Update the bindings for google,gs101-pmu so that the syscon compatible is
no longer required. As it isn't really correct to claim we are compatible with
syscon (as a mmio regmap created by syscon will not work on gs101).

Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
---
 Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml b/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
index be1441193feeea1c12d0d5370168e9316c8dcf76..c5dd132a2b6948f8eda93e4a32ddd1b57a83fcc0 100644
--- a/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
+++ b/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
@@ -34,9 +34,10 @@ select:
 properties:
   compatible:
     oneOf:
+      - items:
+          - const: google,gs101-pmu
       - items:
           - enum:
-              - google,gs101-pmu
               - samsung,exynos3250-pmu
               - samsung,exynos4210-pmu
               - samsung,exynos4212-pmu

-- 
2.51.1.930.gacf6e81ea2-goog

[PATCH 2/2] arm64: dts: exynos: gs101: remove syscon compatible from pmu node

[Peter Griffin]

Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
"syscon" compatible") it is possible to register a regmap without the
syscon compatible in the node.

As mentioned in that commit, it's not correct to claim we are compatible
with syscon, as a MMIO regmap created by syscon won't work. Removing the
syscon compatible means syscon driver won't ever create a mmio regmap.

Note this isn't usually an issue today as exynos-pmu runs at an early
initcall so the custom regmap will have been registered first. However
changes proposed in [1] will bring -EPROBE_DEFER support to syscon allowing
this mechanism to be more robust, especially in highly modularized systems.

Technically this is a ABI break but no other platforms are affected.

Link: https://lore.kernel.org/lkml/aQdHmrchkmOr34r3@stanley.mountain/ [1]
Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
---
 arch/arm64/boot/dts/exynos/google/gs101.dtsi | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm64/boot/dts/exynos/google/gs101.dtsi b/arch/arm64/boot/dts/exynos/google/gs101.dtsi
index d06d1d05f36408137a8acd98e43d48ea7d4f4292..e1a7d33fd4a369f7b352b81d2070beb62a0ced16 100644
--- a/arch/arm64/boot/dts/exynos/google/gs101.dtsi
+++ b/arch/arm64/boot/dts/exynos/google/gs101.dtsi
@@ -1705,7 +1705,7 @@ sysreg_apm: syscon@17420000 {
 		};
 
 		pmu_system_controller: system-controller@17460000 {
-			compatible = "google,gs101-pmu", "syscon";
+			compatible = "google,gs101-pmu";
 			reg = <0x17460000 0x10000>;
 			google,pmu-intr-gen-syscon = <&pmu_intr_gen>;
 

-- 
2.51.1.930.gacf6e81ea2-goog

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Krzysztof Kozlowski]

On Mon, Nov 03, 2025 at 08:03:10AM +0000, Peter Griffin wrote:
> Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
> "syscon" compatible") it is possible to register a regmap without the
> syscon compatible in the node.
> 
> Update the bindings for google,gs101-pmu so that the syscon compatible is

That's an ABI break...

> no longer required. As it isn't really correct to claim we are compatible with
> syscon (as a mmio regmap created by syscon will not work on gs101).

... with kind of a reason, but then the question I have: was the
standard MMIO regmap exposed via syscon ever working for any part of
this PMU?

Original posting here:
https://lore.kernel.org/all/20231209233106.147416-2-peter.griffin@linaro.org/
did not change PMU driver and did not express that the syscon is broken.

Best regards,
Krzysztof

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Peter Griffin]

Hi Krzysztof,

On Wed, 5 Nov 2025 at 08:55, Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On Mon, Nov 03, 2025 at 08:03:10AM +0000, Peter Griffin wrote:
> > Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
> > "syscon" compatible") it is possible to register a regmap without the
> > syscon compatible in the node.
> >
> > Update the bindings for google,gs101-pmu so that the syscon compatible is
>
> That's an ABI break...
>
> > no longer required. As it isn't really correct to claim we are compatible with
> > syscon (as a mmio regmap created by syscon will not work on gs101).
>
> ... with kind of a reason, but then the question I have: was the
> standard MMIO regmap exposed via syscon ever working for any part of
> this PMU?
>
> Original posting here:
> https://lore.kernel.org/all/20231209233106.147416-2-peter.griffin@linaro.org/
> did not change PMU driver and did not express that the syscon is broken.

With the benefit of hindsight, no, PMU write register access was never
working for any part of the PMU.

On the initial posting I didn't hit this as we were only disabling the
Watchdog on boot (which doesn't write any PMU registers). Once the
issue with accessing PMU regs became clear, I first suspected some
missing clock. Then later it became apparent we needed a SMC call and
after some initial proposals the syscon driver was enhanced with the
of_syscon_register_regmap() API in 769cb63166d9 ("mfd: syscon: Add
of_syscon_register_regmap() API"). At the time that commit was merged,
syscon driver required a "syscon" compatible, but since ba5095ebbc7a
("mfd: syscon: Allow syscon nodes without a "syscon" compatible") that
is no longer the case.

Peter.

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Krzysztof Kozlowski]

On Tue, Nov 11, 2025 at 10:18:47AM +0000, Peter Griffin wrote:
> Hi Krzysztof,
> 
> On Wed, 5 Nov 2025 at 08:55, Krzysztof Kozlowski <krzk@kernel.org> wrote:
> >
> > On Mon, Nov 03, 2025 at 08:03:10AM +0000, Peter Griffin wrote:
> > > Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
> > > "syscon" compatible") it is possible to register a regmap without the
> > > syscon compatible in the node.
> > >
> > > Update the bindings for google,gs101-pmu so that the syscon compatible is
> >
> > That's an ABI break...
> >
> > > no longer required. As it isn't really correct to claim we are compatible with
> > > syscon (as a mmio regmap created by syscon will not work on gs101).
> >
> > ... with kind of a reason, but then the question I have: was the
> > standard MMIO regmap exposed via syscon ever working for any part of
> > this PMU?
> >
> > Original posting here:
> > https://lore.kernel.org/all/20231209233106.147416-2-peter.griffin@linaro.org/
> > did not change PMU driver and did not express that the syscon is broken.
> 
> With the benefit of hindsight, no, PMU write register access was never
> working for any part of the PMU.

Please add it to the last paragraph of your commit msg, that this was
never working in practice thus changing ABI is justified.

Best regards,
Krzysztof

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Krzysztof Kozlowski]

On Mon, Nov 03, 2025 at 08:03:10AM +0000, Peter Griffin wrote:
> Since commit ba5095ebbc7a ("mfd: syscon: Allow syscon nodes without a
> "syscon" compatible") it is possible to register a regmap without the
> syscon compatible in the node.
> 
> Update the bindings for google,gs101-pmu so that the syscon compatible is
> no longer required. As it isn't really correct to claim we are compatible with
> syscon (as a mmio regmap created by syscon will not work on gs101).
> 
> Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
> ---
>  Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml b/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
> index be1441193feeea1c12d0d5370168e9316c8dcf76..c5dd132a2b6948f8eda93e4a32ddd1b57a83fcc0 100644
> --- a/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
> +++ b/Documentation/devicetree/bindings/soc/samsung/exynos-pmu.yaml
> @@ -34,9 +34,10 @@ select:
>  properties:
>    compatible:
>      oneOf:
> +      - items:

I expect new version (see other comment), so also please change items
into enum and drop const from below. items is redundant for one entry in
compatible (by convention) and having it as enum already makes it ready
for growing the enumeration for future devices.

> +          - const: google,gs101-pmu

Best regards,
Krzysztof

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Krzysztof Kozlowski]

On 12/11/2025 11:22, Krzysztof Kozlowski wrote:
>>      oneOf:
>> +      - items:
> 
> I expect new version (see other comment), so also please change items
> into enum and drop const from below. items is redundant for one entry in
> compatible (by convention) and having it as enum already makes it ready
> for growing the enumeration for future devices.
> 
>> +          - const: google,gs101-pmu
> 


Ah, and there is also checkpatch warning about length of commit msg.

Best regards,
Krzysztof

Re: [PATCH 1/2] dt-bindings: soc: samsung: exynos-pmu: remove syscon for google,gs101-pmu

[Peter Griffin]

Hi Krzysztof,

Thanks for the review feedback!

On Wed, 12 Nov 2025 at 10:23, Krzysztof Kozlowski <krzk@kernel.org> wrote:
>
> On 12/11/2025 11:22, Krzysztof Kozlowski wrote:
> >>      oneOf:
> >> +      - items:
> >
> > I expect new version (see other comment), so also please change items
> > into enum and drop const from below. items is redundant for one entry in
> > compatible (by convention) and having it as enum already makes it ready
> > for growing the enumeration for future devices.
> >
> >> +          - const: google,gs101-pmu
> >
>
>
> Ah, and there is also checkpatch warning about length of commit msg.

I just sent a new version addressing this and the other comments.

Thanks,

Peter