On Wed, 19 Sep 2012, Hugh Dickins wrote: > Good catch, and the patch looks right to me, as far as it goes: > but does it go far enough? > > I hesitate because it looks as if the NR_MLOCK zone page state is > maintained (with incs and decs) in ignorance of THP; so although > you will be correcting the Unevictable kB with your mlock_vma_page(), > the Mlocked kB just above it in /proc/meminfo would still be wrong? > Indeed, NR_MLOCK is a separate problem with regard to thp and it's currently incremented once for every hugepage rather than HPAGE_PMD_NR. mlock_vma_page() needs to increment by hpage_nr_pages(page) like add_page_to_lru_list() does. > I suppose I'm not sure whether this is material for late-3.6: > surely it's not a fix for a recent regression? > Ok, sounds good. If there's no objection, I'd like to ask Andrew to apply this to -mm and remove the cc to stable@xxxxxxxxxxxxxxx since the mlock_vma_page() problem above is separate and doesn't conflict with this code, so I'll send a followup patch to address that. Thanks! -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>