Re: [PATCH 6/8] arm64: dts: qcom: eliza: Add GPU SMMU node

[Konrad Dybcio]

On 7/13/26 9:47 PM, Akhil P Oommen wrote:
> On 7/6/2026 4:38 PM, Konrad Dybcio wrote:
>> On 7/6/26 1:07 PM, Konrad Dybcio wrote:
>>> On 7/5/26 10:14 AM, Akhil P Oommen wrote:
>>>> From: Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx>
>>>>
>>>> Add the nodes to describe the GPU SMMU.
>>>>
>>>> Signed-off-by: Abel Vesa <abel.vesa@xxxxxxxxxxxxxxxx>
>>>> Signed-off-by: Akhil P Oommen <akhilpo@xxxxxxxxxxxxxxxx>
>>>> ---
>>>>  arch/arm64/boot/dts/qcom/eliza.dtsi | 38 +++++++++++++++++++++++++++++++++++++
>>>>  1 file changed, 38 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/boot/dts/qcom/eliza.dtsi b/arch/arm64/boot/dts/qcom/eliza.dtsi
>>>> index c5635f22e2a7..e5b8377e6c3a 100644
>>>> --- a/arch/arm64/boot/dts/qcom/eliza.dtsi
>>>> +++ b/arch/arm64/boot/dts/qcom/eliza.dtsi
>>>> @@ -2674,6 +2674,44 @@ mdss_dp0_out: endpoint {
>>>>  			};
>>>>  		};
>>>>  
>>>> +		adreno_smmu: iommu@3da0000 {
>>>> +			compatible = "qcom,eliza-smmu-500", "qcom,adreno-smmu",
>>>> +				     "qcom,smmu-500", "arm,mmu-500";
>>>> +			reg = <0x0 0x03da0000 0x0 0x40000>;
>>>> +			#iommu-cells = <2>;
>>>> +			#global-interrupts = <1>;
>>>> +			interrupts = <GIC_SPI 673 IRQ_TYPE_LEVEL_HIGH>,
>>>
>>> 674 is the correct globla
>>
>> (modulo the issue i explained below)
>>
>> [...]
>>
>>> This list is not quite correct. It must be sorted by the context index,
>>> to which a given interrupt corresponds to - the driver relies on that
>>> to give you information about where a context fault happens
>>>
>>> Moreover, I see that the interrupt sheet has the bug where some
>>> interrupts are offset by 32 from the base vector, and others are offset
>>> by 31 (i.e. the SPI number doesn't equal irq vector - 32) - please find
>>> out which mapping is correct with the right folks
> 
> I reviewed this and looks correct to me. Also, this matches the
> downstream configuration. Did you confuse with Dest Port vs Vector ID in
> the documentation?

What I'm trying to say is that the documentation is confusing, we've had
that issue in the past on other platforms. If you see that the correct
irqs are firing, it's fine

Konrad