On Sun, Jan 23, 2022 at 10:51:57PM -0800, Hugh Dickins wrote: > I don't know of any problem that needs fixing with the zero page: > how do you suppose the zero page gets into a truncatable or hole-punchable > mapping? We use it for read faults in anonymous mappings. And I told the > story of how once-upon-a-time it could get inserted into any mapping by > reading from /dev/zero, but that odd case was dropped years ago. And I > am open to (even encouraging) a change to make use of zero page for read > faults of holes in shmem: but that's potential future work, which would > require some changes elsewhere (though perhaps none here: the zero page > could never be used for the result of a COW). > > Please explain the zero page problem you hope to fix here. After I tried to learn the old/new worlds of zero page somehow I thought there can be zero pfns installed, but it seems not at all.. Please ignore above, sorry for the noise. -- Peter Xu
