Can we get back to the basic premise of the proposed patch?
The current code in of_dma_configure() contains a hack that allows the
dma-ranges property to specify a mask instead of a size. The binding
in the specification allows a size and does not allow a mask.
The hack was added to account for one or more dts files that did not
follow the specification. In the mail list discussion of the hack
you said "Also, we need a WARN here so DTs get fixed."
The hack was first present in Linux 4.1. The only in-tree dts that
incorrectly contained a mask instead of a size in dma-ranges was
arch/arm64/boot/dts/amd/amd-seattle-soc.dtsi
That .dtsi was fixed by
commit c91cb9123cdd ("dtb: amd: Fix DMA ranges in device tree")
The fix was present in Linux 4.6, May 15, 2016.
I would like to remove the hack. I think that enough time has
elapsed to allow this change.
If we have no cases of what I'm concerned about, then removing it is
fine. Is this a dependency for iommu series? Doesn't look like it to
me.
This patch is a replacement for patch 03/12 in the iommu series. I
think that patch 03/12 of the iommu series could be dropped and my
patch could be applied independently of the iommu series.
There is likely a conflict between my patch and patch 06/12 of the
iommu series because in my patch the first line of the patch chunk
of drivers/of/device.c includes a line that is changed in 06/12
of the iommu series. If this is the case then the iommu series
should take precedence over my patch (and I should subsequently
fixup my patch).