On Tue, May 5, 2020 at 12:52 PM Robin Murphy <robin.murphy@xxxxxxx> wrote: > > On 2020-05-05 5:51 pm, Andre Przywara wrote: > > Date: Mon, 4 May 2020 12:43:18 +0100 > > Subject: [PATCH 03/16] dt-bindings: ehci/ohci: Allow iommus property > > > > A OHCI/EHCI controller could be behind an IOMMU, in which case an iommus > > property assigns the stream ID for this device. > > Hmm, pretty much any DMA master device could be behind an IOMMU, and in > a way that's basically entirely irrelevant to the device itself (and > thus the consumer of its binding). Is there a better way we can handle > this than repeating such "semi-standard" properties in individual > bindings for ever more? We did that for pinctrl-* properties which I was somewhat lukewarm on. I think a lot more nodes have pins than DMA, so I prefer to keep this explicit. We do allow more than 1 iommus entry which might be device specific, so somewhere we need to capture how many are allowed/valid. We could define a 'dma-device.yaml' schema and reference that on all DMA capable device. Not sure that saves much though. Rob
