On Fri, Mar 27, 2020 at 01:23:07PM -0400, Zi Yan wrote: > On 27 Mar 2020, at 13:03, Kirill A. Shutemov wrote: > > > Splitting PMD mapping huge zero page can be simplified a lot: we can > > just unmap it and fallback to PTE handling. > > So we will have an extra page fault for the first read to each subpage, > but nothing changes if the first access to a subpage is a write, right? > BTW, what is the motivation for this code simplification? Match what we do for file-THP. I found a problem with the patch. Ignore it for now. -- Kirill A. Shutemov
