Re: [PATCH 1/8] iommu: Decouple iommu_present() from bus ops

[Jason Gunthorpe]

On Mon, Jan 30, 2023 at 01:49:20PM +0000, Robin Murphy wrote:

> All in all, it's only actually the habanalabs ones that I'm slightly wary
> of, since they're effectively (mis)using device_iommu_mapped() to infer the
> DMA ops implementation, which could potentially go wrong (or at least *more*
> wrong) on POWER with this change. I guess the saving grace is that
> although

IMHO habana is not using the DMA API abstraction properly. If it
doesn't work on some archs is their bugs to deal with - we don't need
to complexify the core code to tiptoe around around such an abuse in
an obscure driver.

Jason