On Tue, Oct 20, 2020 at 02:09:23PM +0100, Tvrtko Ursulin wrote: > Not just the max, but the granularity as well. If byte granularity (well > less than PAGE_SIZE/4k) exists in some hw then I agree the API change makes > sense. scatter/gather lists are byte granular in HW, this is basically the norm. Page lists are something a little different. At least in RDMA we use scatterlist for both types of objects, but their treatement is quite different. For page lists the max_segment_size is just something that gets in the way, we really want the SGLs to be as large, and as highly aligned as possible. Fewer SGE's means less work and less memory everywhere that processes them. When it comes times to progam the HW the SGL is analyzed and the HW specific page size selected, then the SGL is broken up into a page list. Each SGE may be split into many HW DMA blocks. Again, this has nothing to do with PAGE_SIZE, the HW DMA block size selection is HW specific and handled when splitting the SGL. Jason
