Hi Christoph, On Thu, Nov 29, 2018 at 05:57:15PM +0100, Christoph Hellwig wrote: > > As for the buffer sharing: at least for the DMA API side I want to > move the current buffer sharing users away from dma_alloc_coherent > (and coherent dma_alloc_attrs users) and the remapping done in there > required for non-coherent architectures. Instead I'd like to allocate > plain old pages, and then just dma map them for each device separately, > with DMA_ATTR_SKIP_CPU_SYNC passed for all but the first user to map > or last user to unmap. On the iommu side it could probably work > similar. > > I have done some preliminary work on this, and want to get it into this > merge window, but there is a few other bits I need to sort out first. > This sounds very useful for ion, to avoid CPU cache maintenance as long as the buffer stays in device-land. One question though: How would you determine "the last user to unmap" to know when to do the final "make visible to CPU" step? Thanks, -Brian
