On Mon, 23 Aug 2021 16:05:46 +0200 Gerald Schaefer <gerald.schaefer@xxxxxxxxxxxxx> wrote: > On Fri, 20 Aug 2021 07:43:40 +0200 > Christoph Hellwig <hch@xxxxxx> wrote: > > > Hi all, > > > > looking at the recent ZONE_DEVICE related changes we still have a > > horrible maze of different code paths. I already suggested to > > depend on ARCH_HAS_PTE_SPECIAL for ZONE_DEVICE there, which all modern Oh, we do have PTE_SPECIAL, actually that took away the last free bit in the pte. So, if there is a chance that ZONE_DEVICE would depend on PTE_SPECIAL instead of PTE_DEVMAP, we might be back in the game and get rid of that CONFIG_FS_DAX_LIMITED. Or did you rather mean depend on ARCH_HAS_PTE_SPECIAL on top of ARCH_HAS_PTE_DEVMAP? > > architectures have anyway. But the other odd special case is > > CONFIG_FS_DAX_LIMITED which is just used for the xpram driver. Does > > this driver still see use? If so can we make it behave like the > > other DAX drivers and require a pgmap? I think the biggest missing > > part would be to implement ARCH_HAS_PTE_DEVMAP for s390. > > Puh, yes, that seems to be needed in order to enable ZONE_DEVICE, and > then we could use devm_memremap_pages(), at least that was my plan > some time ago. However, either the ARCH_HAS_PTE_DEVMAP dependency > is new, or I overlooked it before, but we do not have any free bits > in the pte left, so this is not going to work. > > Would it strictly be necessary to implement ZONE_DEVICE, or would > it be enough if we would use e.g. add_memory() instead of just > adding the DCSS memory directly to the kernel mapping via > vmem_add_mapping()? That way we might at least get the struct pages, > but somehow it doesn't feel completely right.
