Re: [PATCH v4 2/5] dt-bindings: iommu: Add Broadcom BCM2712 IOMMU

[Daniel Drake <dan@reactivated.net>]

Hi!,

On 03/09/2026 18:36, Jim Quinlan wrote:
> What are your thoughts on allowing the iommu work without the cache
> property, ie having its presence be optional?  I work on the Broadcom
> version of the Rpi4/5 SoCs,  and  (a) on certain workloads the cache
> slightly degrades performance or (b) on some chips there is no cache
> associated with a specific iommu.  This is probably on me to implement
> these possibilities, but I just wanted to give you a heads-up.


For (a) I think the DT must encode the relationship between iommu & 
iommuc in any case, because the DT's job is to describe the hardware.

If the user wishes to disable use of the iommu cache then that would be 
a software configuration choice. This control could perhaps be exposed 
via a module parameter, or by iommu_device_sysfs_add() to expose a 
writable attribute. Ultimately which approach is suitable would be a 
decision of the iommu subsystem maintainers.

For (b) we could just make the property optional when support for such a 
chip arrives, or alternatively we could use a different compatible 
string for the cache-less iommus and have the cache property only 
required for certain specific compatible strings (see 
mediatek,iommu.yaml for an example where some properties are only 
required for certain compatible values).

I'd be happy to review patches for these cases once the driver support 
is added.

Thanks!
Daniel