On Wed, Aug 14, 2019 at 04:20:34PM +0200, Christophe Leroy wrote: >> index 56a7c814160d..afe71b89dec3 100644 >> --- a/arch/powerpc/platforms/Kconfig.cputype >> +++ b/arch/powerpc/platforms/Kconfig.cputype >> @@ -450,8 +450,10 @@ config NOT_COHERENT_CACHE >> depends on 4xx || PPC_8xx || E200 || PPC_MPC512x || \ >> GAMECUBE_COMMON || AMIGAONE >> select ARCH_HAS_DMA_COHERENT_TO_PFN > > You drop arch_dma_coherent_to_pfn(), it's surprising to see > ARCH_HAS_DMA_COHERENT_TO_PFN remains. At first I thought I'd get a build > failure. > > After looking more, I see there is a arch_dma_coherent_to_pfn() > defined in kernel/dma/remap.c when DMA_DIRECT_REMAP is selected. > > I think the naming is not really consistant and should be fixed some how, > because that's misleading to have an arch_something() being common to all.
I actually have patches in the queue kill arch_dma_coherent_to_pfn off entirely, as we can always just get back to the physical address and thus pfn from the dma address using dma_to_phys for dma-direct. But there is a huge review backlog for my outstanding dma patcheset, so it might take a bit until it is posted.
