On 01.06.24 16:09, Lance Yang wrote:
Completely agree with David's point[1]. I'm also not convinced that this is the right approach :) It seems like this patch won't handle all cases, as David mentioned[1] before. folio_remove_rmap_ptes() will immediately munlock a large folio (as large folios are not allowed to be batch-added to the LRU list) via munlock_vma_folio() when it is fully unmapped, so this patch does not work in this case. Even worse, if we encounter a COW mlocked folio, we would run into trouble (data corruption). Hi Andrew, I just noticed that this patch has become part of v6.10-rc1, but it has not been acked/reviewed yet. Is there any chance to revert it?
Thanks Lance, for paying attention. I think I spotted this on LWN and thought "I don't recall that we agreed this is the right approach" but didn't have time to follow up.
My opinion on this did not change. -- Cheers, David / dhildenb
