On Mon, 22 Mar 2010, Andrea Arcangeli wrote: > Again: split_huge_page has nothing to do with the pte or pmd locking. But you are addinig sync points to the pte/pmd function... > Especially obvious in the case your proposed alternate design will > still use one form of split_huge_page but one that can fail if the > page is under gup (which would practically make it unusable anywhere > but swap and even in swap it would lead to potential livelocks in > unsolvable oom as it's not just slow-unfrequent-IO calling gup). It can fail and be retried. Breaking up a page is not a performance critical thing. As you have shown this occurs rarely. -- 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>