> Am 28.10.2024 um 16:58 schrieb Robin Murphy <robin.murphy@xxxxxxx>:
>
>
> Oh fiddle... this is getting complex...
>
> I was testing the of_xlate patch rather than the diff below, which should avoid the deferred probe timeout but otherwise achieve much the same effect. Trouble is, that turns out to only "fix" things back to another breakage due to 4720287c7bf7 ("iommu: Remove struct iommu_ops *iommu from arch_setup_dma_ops()"), wherein we lost the fragile combination of setting up the IOMMU while still causing arm_setup_iommu_dma_ops() to ignore it. Thus what now happens here is the domain from the ARM DMA ops gets in the way of the driver attaching its own. Looks like I missed that with the remoteproc driver since it only starts trying to do IOMMU API stuff upon successfully loading firmware, which I didn't have.
>
> I think the most practical answer is going to be to give the ISP and remoteproc drivers the same workaround for that as the various DRM drivers use, lemme write those up too...
Sounds good. Please let me know if you have something I can test.
Best regards and thanks,
Nikolaus
