On Tue, Mar 15, 2016 at 06:42:58PM +0300, Kirill A. Shutemov wrote: > On Tue, Mar 15, 2016 at 08:30:54AM -0700, Dave Hansen wrote: > > On 03/15/2016 07:55 AM, Kirill A. Shutemov wrote: > > > +Transparent huge page is represented by single entry on a lru list and > > > +therefore we can only make unevictable entire compound page, not > > > +individual subpages. > > > > A few grammar nits: > > > > A transparent huge page is represented by a single entry on an lru list. > > Therefore, we can only make unevictable an entire compound page, not > > individual subpages. > > Thanks. > > > > +We handle this by forbidding mlocking PTE-mapped huge pages. This way we > > > +keep the huge page accessible for vmscan. Under memory pressure the page > > > +will be split, subpages from VM_LOCKED VMAs moved to unevictable lru and > > > +the rest can be evicted. > > > > What do you mean by "mlocking" in this context? Do you mean that we > > actually return -ESOMETHING from mlock() on PTE-mapped huge pages? Or, > > do you just mean that we defer treating PTE-mapped huge pages as > > PageUnevictable() inside the kernel? > > The latter. > > > I think we should probably avoid saying "mlocking" when we really mean > > "kernel-internal mlocked page handling" aka. the unevictable list. > > What about > > "We handle this by keeping PTE-mapped huge pages on normal LRU lists." > > ? > > The updated patch is below. Err. That was the old patch. >From 2d805c15cf5f2ebee324f12646da6db3e095a4ba Mon Sep 17 00:00:00 2001 From: "Kirill A. Shutemov" <kirill.shutemov@xxxxxxxxxxxxxxx> Date: Tue, 15 Mar 2016 17:48:16 +0300 Subject: [PATCH] thp, mlock: update unevictable-lru.txt Add description of THP handling into unevictable-lru.txt. Signed-off-by: Kirill A. Shutemov <kirill.shutemov@xxxxxxxxxxxxxxx> --- Documentation/vm/unevictable-lru.txt | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/Documentation/vm/unevictable-lru.txt b/Documentation/vm/unevictable-lru.txt index fa3b527086fa..0026a8d33fc0 100644 --- a/Documentation/vm/unevictable-lru.txt +++ b/Documentation/vm/unevictable-lru.txt @@ -461,6 +461,27 @@ unevictable LRU is enabled, the work of compaction is mostly handled by the page migration code and the same work flow as described in MIGRATING MLOCKED PAGES will apply. +MLOCKING TRANSPARENT HUGE PAGES +------------------------------- + +A transparent huge page is represented by a single entry on an LRU list. +Therefore, we can only make unevictable an entire compound page, not +individual subpages. + +If a user tries to mlock() part of a huge page, we want the rest of the +page to be reclaimable. + +We cannot just split the page on partial mlock() as split_huge_page() can +fail and new intermittent failure mode for the syscall is undesirable. + +We handle this by keeping PTE-mapped huge pages on normal LRU lists: the +PMD on border of VM_LOCKED VMA will be split into PTE table. + +This way the huge page is accessible for vmscan. Under memory pressure the +page will be split, subpages which belong to VM_LOCKED VMAs will be moved +to unevictable LRU and the rest can be reclaimed. + +See also comment in follow_trans_huge_pmd(). mmap(MAP_LOCKED) SYSTEM CALL HANDLING ------------------------------------- -- Kirill A. Shutemov -- 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>