On 2022-11-15 13:06, Konrad Dybcio wrote:
On 15/11/2022 14:00, Krzysztof Kozlowski wrote:
On 15/11/2022 13:54, Konrad Dybcio wrote:
On 14/11/2022 17:58, Krzysztof Kozlowski wrote:
On 14/11/2022 16:53, Konrad Dybcio wrote:
On 14/11/2022 14:00, Krzysztof Kozlowski wrote:
On 14/11/2022 12:17, Konrad Dybcio wrote:
On 14/11/2022 12:01, Krzysztof Kozlowski wrote:
On 14/11/2022 11:42, Konrad Dybcio wrote:
Some SMMUs require that a vote is held on as much as 3 separate
PDs
(hello Qualcomm). Allow it in bindings.
Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxx>
---
Changes since v1:
- Add minItems
Documentation/devicetree/bindings/iommu/arm,smmu.yaml | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git
a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
index 9066e6df1ba1..82bc696de662 100644
--- a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
+++ b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
@@ -159,7 +159,8 @@ properties:
through the TCU's programming interface.
power-domains:
- maxItems: 1
+ minItems: 0
It cannot be 0.
minItems: 1
Anyway you still need to restrict it per variant, as I said in
previous
version.
Hm.. I'm not entirely sure what you mean.. Should I add a list of
compatibles
Yes and limit it to maxItems: 1 for "else".
I tried adding:
- if:
properties:
compatible:
contains:
enum:
- qcom,sm6375-smmu-500
then:
properties:
power-domains:
minItems: 3
maxItems: 3
else:
properties:
power-domains:
maxItems: 1
Right under the nvidia reg if-else in the allOf, but dtbs_check throws
errors like:
/home/konrad/linux/arch/arm64/boot/dts/qcom/msm8998-sony-xperia-yoshino-poplar.dtb:
iommu@5040000: 'power-domains' does not match any of the regexes:
'pinctrl-[0-9]+'
Any clues as to why?
I don't know what code do you have there, but generic pattern is:
https://elixir.bootlin.com/linux/v5.19-rc6/source/Documentation/devicetree/bindings/clock/samsung,exynos7-clock.yaml#L38
I tried many things, but I still don't seem to get a hang of it.. Here's
my current diff rebased on top of Dmitry's recent cleanups (available at
[1])
diff --git a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
index 28f5720824cd..55759aebc4a0 100644
--- a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
+++ b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
@@ -200,7 +200,7 @@ properties:
maxItems: 7
power-domains:
As I mentioned before - minItems: 1.
But not all SMMUs require a power domain :/
Right, so it's not a required property. However if it *is* present, then
it needs to reference at least one power domain, because having an empty
property, i.e.:
power-domains = <>;
or
power-domains;
makes no sense whatsoever.
Thanks,
Robin.
Just like the link I gave you.
- maxItems: 1
+ maxItems: 3
nvidia,memory-controller:
description: |
@@ -364,6 +364,26 @@ allOf:
- description: interface clock required to access smmu's
registers
through the TCU's programming interface.
+ - if:
+ properties:
+ compatible:
+ contains:
+ const: qcom,sm6375-smmu-500
+ then:
+ properties:
+ power-domains:
+ items:
+ - description: SNoC MMU TBU RT GDSC
+ - description: SNoC MMU TBU NRT GDSC
+ - description: SNoC TURING MMU TBU0 GDSC
+
+ required:
+ - power-domains
+ else:
+ properties:
+ power-domains:
+ maxItems: 1
+
examples:
- |+
/* SMMU with stream matching or stream indexing */
In my eyes, this should work, but I still get errors like:
/home/konrad/linux/arch/arm64/boot/dts/qcom/sm8250-hdk.dtb:
iommu@3da0000: power-domains: [[108, 0]] is too short
as if the else: path was never taken..
It was, but the top-level property said that minItems=3 (implicitly), so
it is too short.
So the top-level properties take precedence over the ones that come from
the if-then-else?? Ugh.
Konrad
Best regards,
Krzysztof
