In test, some pages in swap-cache can't be migrated, as they aren't rmap.
unmap_and_move() ignores swap-cache page which is just read in and hasn't
rmap (see the comments in the code), but swap_aops provides .migratepage.
Better to migrate such pages instead of ignore them.
Signed-off-by: Shaohua Li <shaohua.li@intel.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Yakui Zhao <yakui.zhao@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
 static void remove_file_migration_ptes(struct page *old, struct page *new)
 {
        struct vm_area_struct *vma;
-       struct address_space *mapping = page_mapping(new);
+       struct address_space *mapping = new->mapping;
        struct prio_tree_iter iter;
        pgoff_t pgoff = new->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
 
                         *    needs to be effective.
                         */
                        try_to_free_buffers(page);
+                       goto rcu_unlock;
                }
-               goto rcu_unlock;
+               goto skip_unmap;
        }
 
        /* Establish migration ptes or remove ptes */
        try_to_unmap(page, 1);
 
+skip_unmap:
        if (!page_mapped(page))
                rc = move_to_new_page(newpage, page);