This may have been caused by the same anon-vma bug that is being debugged for mainline as it'd free anon-vma too early leading to memory corruption. split_huge_page is a super heavy user of anon-vmas, so that would explain all the issues, and also another crash I had in split_huge_page where the rmap chain found a number of pmd mapping the hugepage different than page_mapcount. For swap that means mlock, for split_huge_page not being able to update some huge pmd is fatal so there's plenty of BUG_ON to check all goes well and there's no risk of corruption later on. We need to test with Linus's fix for anon-vma bug, and I guess everything will go fine when that bug is solved. I added more commentary as well to cover an ordering issue in the anon_vma_prepare. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>