zsmalloc: partial page ordering within a fullness_list
authorSergey Senozhatsky <sergey.senozhatsky@gmail.com>
Tue, 8 Sep 2015 22:04:44 +0000 (15:04 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 8 Sep 2015 22:35:28 +0000 (15:35 -0700)
We want to see more ZS_FULL pages and less ZS_ALMOST_{FULL, EMPTY}
pages.  Put a page with higher ->inuse count first within its
->fullness_list, which will give us better chances to fill up this page
with new objects (find_get_zspage() return ->fullness_list head for new
object allocation), so some zspages will become ZS_ALMOST_FULL/ZS_FULL
quicker.

It performs a trivial and cheap ->inuse compare which does not slow down
zsmalloc and in the worst case keeps the list pages in no particular
order.

A more expensive solution could sort fullness_list by ->inuse count.

[minchan@kernel.org: code adjustments]
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Minchan Kim <minchan@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/zsmalloc.c

index 27b9661c8fa67fa8f434ecd9fc5e92d08621c64e..615b9b9b45ebfd1f1db017b4ac06b811c52a7d2e 100644 (file)
@@ -643,13 +643,22 @@ static void insert_zspage(struct page *page, struct size_class *class,
        if (fullness >= _ZS_NR_FULLNESS_GROUPS)
                return;
 
-       head = &class->fullness_list[fullness];
-       if (*head)
-               list_add_tail(&page->lru, &(*head)->lru);
-
-       *head = page;
        zs_stat_inc(class, fullness == ZS_ALMOST_EMPTY ?
                        CLASS_ALMOST_EMPTY : CLASS_ALMOST_FULL, 1);
+
+       head = &class->fullness_list[fullness];
+       if (!*head) {
+               *head = page;
+               return;
+       }
+
+       /*
+        * We want to see more ZS_FULL pages and less almost
+        * empty/full. Put pages with higher ->inuse first.
+        */
+       list_add_tail(&page->lru, &(*head)->lru);
+       if (page->inuse >= (*head)->inuse)
+               *head = page;
 }
 
 /*