On Fri, Nov 16, 2018 at 08:52:14AM -0800, Palmer Dabbelt wrote:

As far as I can tell, it looks like m68k, mips, and powerpc mention an 
IOMMU in their ports, don't set ARCH_HAS_SG_CHAIN, and with this patch set 
won't set ARCH_NO_SG_CHAIN.

m68k has no iommu, and not operations that operate on a scatterlist.

mips has a trivial iommu driver (jazzdma), but I wrote the current
instance of it, nad it is fine.

powerpc has various iommu, but actually enables ARCH_HAS_SG_CHAIN
unconditionally.

The issue is that I'm not sure how to 
determine what constitutes a horrible legacy IOMMU, at least with respect 
to not being able to use scatterlist chaining.

It basically means someone is iterating using manual pointer arithmetics
over a multi-element scatterlist instead of using the sg_next and
for_each_sg helpers.