thp: increase split_huge_page() success rate
authorKirill A. Shutemov <kirill.shutemov@linux.intel.com>
Sat, 16 Jan 2016 00:54:43 +0000 (16:54 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 16 Jan 2016 01:56:32 +0000 (17:56 -0800)
During freeze_page(), we remove the page from rmap.  It munlocks the
page if it was mlocked.  clear_page_mlock() uses thelru cache, which
temporary pins the page.

Let's drain the lru cache before checking page's count vs.  mapcount.
The change makes mlocked page split on first attempt, if it was not
pinned by somebody else.

Signed-off-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/huge_memory.c

index a0b910a0c2cbf3ba5f5fbc82bcf2be74370fc3c7..882b04449904a2f819fc51db609de9491dd209a3 100644 (file)
@@ -3201,6 +3201,7 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
        struct page *head = compound_head(page);
        struct anon_vma *anon_vma;
        int count, mapcount, ret;
+       bool mlocked;
 
        VM_BUG_ON_PAGE(is_huge_zero_page(page), page);
        VM_BUG_ON_PAGE(!PageAnon(page), page);
@@ -3231,9 +3232,14 @@ int split_huge_page_to_list(struct page *page, struct list_head *list)
                goto out_unlock;
        }
 
+       mlocked = PageMlocked(page);
        freeze_page(anon_vma, head);
        VM_BUG_ON_PAGE(compound_mapcount(head), head);
 
+       /* Make sure the page is not on per-CPU pagevec as it takes pin */
+       if (mlocked)
+               lru_add_drain();
+
        /* Prevent deferred_split_scan() touching ->_count */
        spin_lock(&split_queue_lock);
        count = page_count(head);