Date:   Fri, 6 Apr 2018 03:07:11 +0000
From:   Naoya Horiguchi <n-horiguchi@...jp.nec.com>
To:     Michal Hocko <mhocko@...nel.org>
CC:     "Kirill A. Shutemov" <kirill@...temov.name>,
        Zi Yan <zi.yan@...t.com>,
        "linux-mm@...ck.org" <linux-mm@...ck.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Vlastimil Babka <vbabka@...e.cz>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: [PATCH] mm: shmem: enable thp migration (Re: [PATCH v1] mm:
 consider non-anonymous thp as unmovable page)

Hi everyone,

On Thu, Apr 05, 2018 at 06:03:17PM +0200, Michal Hocko wrote:
> On Thu 05-04-18 18:55:51, Kirill A. Shutemov wrote:
> > On Thu, Apr 05, 2018 at 05:05:47PM +0200, Michal Hocko wrote:
> > > On Thu 05-04-18 16:40:45, Kirill A. Shutemov wrote:
> > > > On Thu, Apr 05, 2018 at 02:48:30PM +0200, Michal Hocko wrote:
> > > [...]
> > > > > RIght, I confused the two. What is the proper layer to fix that then?
> > > > > rmap_walk_file?
> > > > 
> > > > Maybe something like this? Totally untested.
> > > 
> > > This looks way too complex. Why cannot we simply split THP page cache
> > > during migration?
> > 
> > This way we unify the codepath for archictures that don't support THP
> > migration and shmem THP.
> 
> But why? There shouldn't be really nothing to prevent THP (anon or
> shemem) to be migratable. If we cannot migrate it at once we can always
> split it. So why should we add another thp specific handling all over
> the place?

If thp migration works fine for shmem, we can keep anon/shmem thp to
be migratable and we don't need any ad-hoc workaround.
So I wrote a patch to enable it.
This patch does not change any shmem specific code, so I think that
it works for file thp (not only shmem,) but I don't test it yet.

Thanks,
Naoya Horiguchi
-----
>From e31ec037701d1cc76b26226e4b66d8c783d40889 Mon Sep 17 00:00:00 2001
From: Naoya Horiguchi <n-horiguchi@...jp.nec.com>
Date: Fri, 6 Apr 2018 10:58:35 +0900
Subject: [PATCH] mm: enable thp migration for shmem thp

My testing for the latest kernel supporting thp migration showed an
infinite loop in offlining the memory block that is filled with shmem
thps.  We can get out of the loop with a signal, but kernel should
return with failure in this case.

What happens in the loop is that scan_movable_pages() repeats returning
the same pfn without any progress. That's because page migration always
fails for shmem thps.

In memory offline code, memory blocks containing unmovable pages should
be prevented from being offline targets by has_unmovable_pages() inside
start_isolate_page_range(). So it's possible to change migratability
for non-anonymous thps to avoid the issue, but it introduces more complex
and thp-specific handling in migration code, so it might not good.

So this patch is suggesting to fix the issue by enabling thp migration
for shmem thp. Both of anon/shmem thp are migratable so we don't need
precheck about the type of thps.

Fixes: commit 72b39cfc4d75 ("mm, memory_hotplug: do not fail offlining too early")
Signed-off-by: Naoya Horiguchi <n-horiguchi@...jp.nec.com>
Cc: stable@...r.kernel.org # v4.15+
---
 mm/huge_memory.c |  5 ++++-
 mm/migrate.c     | 19 ++++++++++++++++---
 mm/rmap.c        |  3 ---
 3 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 2aff58624886..933c1bbd3464 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2926,7 +2926,10 @@ void remove_migration_pmd(struct page_vma_mapped_walk *pvmw, struct page *new)
 		pmde = maybe_pmd_mkwrite(pmde, vma);
 
 	flush_cache_range(vma, mmun_start, mmun_start + HPAGE_PMD_SIZE);
-	page_add_anon_rmap(new, vma, mmun_start, true);
+	if (PageAnon(new))
+		page_add_anon_rmap(new, vma, mmun_start, true);
+	else
+		page_add_file_rmap(new, true);
 	set_pmd_at(mm, mmun_start, pvmw->pmd, pmde);
 	if (vma->vm_flags & VM_LOCKED)
 		mlock_vma_page(new);
diff --git a/mm/migrate.c b/mm/migrate.c
index bdef905b1737..f92dd9f50981 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -472,7 +472,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
 	pslot = radix_tree_lookup_slot(&mapping->i_pages,
  					page_index(page));
 
-	expected_count += 1 + page_has_private(page);
+	expected_count += hpage_nr_pages(page) + page_has_private(page);
 	if (page_count(page) != expected_count ||
 		radix_tree_deref_slot_protected(pslot,
 					&mapping->i_pages.xa_lock) != page) {
@@ -505,7 +505,7 @@ int migrate_page_move_mapping(struct address_space *mapping,
 	 */
 	newpage->index = page->index;
 	newpage->mapping = page->mapping;
-	get_page(newpage);	/* add cache reference */
+	page_ref_add(newpage, hpage_nr_pages(page)); /* add cache reference */
 	if (PageSwapBacked(page)) {
 		__SetPageSwapBacked(newpage);
 		if (PageSwapCache(page)) {
@@ -524,13 +524,26 @@ int migrate_page_move_mapping(struct address_space *mapping,
 	}
 
 	radix_tree_replace_slot(&mapping->i_pages, pslot, newpage);
+	if (PageTransHuge(page)) {
+		int i;
+		int index = page_index(page);
+
+		for (i = 0; i < HPAGE_PMD_NR; i++) {
+			pslot = radix_tree_lookup_slot(&mapping->i_pages,
+						       index + i);
+			radix_tree_replace_slot(&mapping->i_pages, pslot,
+						newpage + i);
+		}
+	} else {
+		radix_tree_replace_slot(&mapping->i_pages, pslot, newpage);
+	}
 
 	/*
 	 * Drop cache reference from old page by unfreezing
 	 * to one less reference.
 	 * We know this isn't the last reference.
 	 */
-	page_ref_unfreeze(page, expected_count - 1);
+	page_ref_unfreeze(page, expected_count - hpage_nr_pages(page));
 
 	xa_unlock(&mapping->i_pages);
 	/* Leave irq disabled to prevent preemption while updating stats */
diff --git a/mm/rmap.c b/mm/rmap.c
index f0dd4e4565bc..8d5337fed37b 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1374,9 +1374,6 @@ static bool try_to_unmap_one(struct page *page, struct vm_area_struct *vma,
 		if (!pvmw.pte && (flags & TTU_MIGRATION)) {
 			VM_BUG_ON_PAGE(PageHuge(page) || !PageTransCompound(page), page);
 
-			if (!PageAnon(page))
-				continue;
-
 			set_pmd_migration_entry(&pvmw, page);
 			continue;
 		}
-- 
2.7.4

Powered by blists - more mailing lists