Re: [PATCH 4/8] arm64: dts: qcom: shikra: Add CAMSS node

[Bryan O'Donoghue]

On 01/06/2026 15:27, Loic Poulain wrote:
> > Yes I think CDM should be a standalone block and BTW use standard DMA
> > APIs with its own IOMMU entry.
> Interesting, so CDM should be a dedicated subnode as well, and not
> part of the OPE?
> Then, If we want to use it from OPE at some point we would need some
> 'cdm' phandle in OPE node.

Your question makes me think again.

CAMSS:
- IFE/CDM
- IFE Lite/CDM

HFI/BPS/IPE
- BPS/CDM
- IPE/CDM

OPE:
- OPE/CDM

I believe those are the mappings of the various SIDs too.

Come to think of it there is every reason to model the CDM block as

cdm-ife0@ {
    iommus = <IFE0_SID>;
};

cdm-ope0@ {
    iommus = <etc>;
}

And then do use the dma engine API to shift data. I don't see why the 
IFE, OPE or BPS needs to map the IOMMUs and if you are using the BPS via 
HFI you'd simply leave the cdm-bps@ {} disabled in your dtsi anyway.

We should excise CDM IOMMUs unless/until we come to implement CDMs as 
standalone nodes.

I'm open to being corrected of course :)
---
bod