On Mon, Jul 02, 2018 at 06:29:15AM -0700, Christoph Hellwig wrote: > This commit is missing an explanation. The patch is for abiv1 & abiv2 CPU series' MMU support. - abiv1 CPU (CK610) is VIPT cache and it doesn't support highmem. - abiv2 CPUs are all PIPT cache and they could support highmem. We seperate abiv1 and abiv2 into two direcotries for coding convention. > For the dma-mapping code please use the generic kernel/dma/noncoherent.c > code instead of duplicating it. Thx for the tips and I think you mean the lib/dma-noncoherent.c in the lastest kernel source and not in linux-4.16.2. I'll rebase on the newest RC version of linux in next version patch and reuse the code in lib/dma-noncoherent.c. Current csky_dma_alloc implementation is not good, it use 512MB uncached area to mirror the Normal cachable-512MB area. In next version patch, we will increase the max-Normal memory zone to (1GB + 768MB). In csky_dma_alloc we will seperate the atomic_dma and non-atomic_dma and reserve the area in fixmap area. Here is my memory layout plan in next version patch: Fixmap : 0xffc02000 – 0xfffff000 (4 MB - 12KB) kmap_atomic, dma_atomic ... Pkmap : 0xff800000 – 0xffc00000 (4 MB) PTR_PER_PTE = 1024 Vmalloc : 0xf0200000 – 0xff000000 (238 MB) max: 238MB + 256MB + 1GB Lowmem : 0x80000000 – 0xf0000000 (1G + 768 MB) Guo Ren
