Re: [PATCH] arm64: dts: qcom: glymur: Fix missing SMMU stream IDs for ADSP

[Iranna Mundaganur]

On 31-08-2026 21:55, Bjorn Andersson wrote:
> On Fri, Aug 21, 2026 at 12:54:03PM +0530, Iranna Mundaganur wrote:
>> On 19-08-2026 17:16, Konrad Dybcio wrote:
>>> On 8/11/26 6:44 PM, Iranna Mundaganur wrote:
>>>> From: Sairamreddy Bojja <sairamreddy.bojja@xxxxxxxxxxxxxxxx>
>>>>
>>>> Add the missing SIDs for the LPASS eNPU masters to the ADSP
>>>> iommus list. These masters are exercised during audio record
>>>> use cases where Environmental Acoustic Intelligence (EAI)
>>>> models run on the eNPUs.
>>> Could you share a bit more details on how we can reproduce t his?
>> This use case is not enabled upstream yet, we validate it downstream with the
>> AudioReach application where EAI models run on eNPU0/eNPU1 and need memory
>> access from those masters.
>>
> Even more reason to provide a proper commit message, clearly describing
> the actual problem.

We are discussing this with the team and will get this fixed in the next re-spin.
Thanks for the review, Bjorn/Konard

Thanks,
Iranna

>
>>>> Currently all transactions from these eNPU masters will result
>>>> in global faults; fix them by adding the additional SIDs.
>>>>
>>>> Fixes: cd1d174c75f7 ("arm64: dts: qcom: glymur: Add ADSP and CDSP for Glymur SoC")
>>>> Signed-off-by: Sairamreddy Bojja <sairamreddy.bojja@xxxxxxxxxxxxxxxx>
>>>> Co-developed-by: Iranna Mundaganur <iranna.mundaganur@xxxxxxxxxxxxxxxx>
>>>> Signed-off-by: Iranna Mundaganur <iranna.mundaganur@xxxxxxxxxxxxxxxx>
>>>> ---
>>>>  arch/arm64/boot/dts/qcom/glymur.dtsi | 4 +++-
>>>>  1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/arch/arm64/boot/dts/qcom/glymur.dtsi b/arch/arm64/boot/dts/qcom/glymur.dtsi
>>>> index f04058d2e089..6b06e0acbcf0 100644
>>>> --- a/arch/arm64/boot/dts/qcom/glymur.dtsi
>>>> +++ b/arch/arm64/boot/dts/qcom/glymur.dtsi
>>>> @@ -4084,7 +4084,9 @@ remoteproc_adsp: remoteproc@6800000 {
>>>>  			compatible = "qcom,glymur-adsp-pas", "qcom,sm8550-adsp-pas";
>>>>  			reg = <0x0 0x06800000 0x0 0x10000>;
>>>>  
>>>> -			iommus = <&apps_smmu 0x1000 0x0>;
>>>> +			iommus = <&apps_smmu 0x1000 0x0>,
>>>> +				 <&apps_smmu 0x1040 0x20>,
>>> The correct SID base is 0x1060 (although the mask makes them equivalent)
>> Yes, will fix this up in the next re-spin.
>>
>>>> +				 <&apps_smmu 0x1080 0x0>;
>>> This will bind the eNPU streams together with the ADSP firmware stream
>>> into a single iommu domain - is that your intent?
>> Yes, we've been told that these SIDs would need access to all the memory
>> regions ADSP firmware stream is given access to, so we want them grouped
>> into a single domain.
>>
>>
>> Thanks,
>> Iranna
>>