It's not completely obvious why we have to shuffle the complete zone -
introduced in commit e900a918b098 ("mm: shuffle initial free memory to
improve memory-side-cache utilization") - because some sort of shuffling is
already performed when onlining pages via __free_one_page(), placing
MAX_ORDER-1 pages either to the head or the tail of the freelist. Let's
document why we have to shuffle the complete zone when exposing larger,
contiguous physical memory areas to the buddy.
Acked-by: Dan Williams <dan.j.williams@xxxxxxxxx>
Acked-by: Michal Hocko <mhocko@xxxxxxxx>
Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Cc: Alexander Duyck <alexander.h.duyck@xxxxxxxxxxxxxxx>
Cc: Dan Williams <dan.j.williams@xxxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
Signed-off-by: David Hildenbrand <david@xxxxxxxxxx>
---
mm/memory_hotplug.c | 8 ++++++++
1 file changed, 8 insertions(+)
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index be3c62e3fb95c..ac6961abaa103 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -831,6 +831,14 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages,
zone->zone_pgdat->node_present_pages += onlined_pages;
pgdat_resize_unlock(zone->zone_pgdat, &flags);
+ /*
+ * When exposing larger, physically contiguous memory areas to the
+ * buddy, shuffling in the buddy (when freeing onlined pages, putting
+ * them either to the head or the tail of the freelist) is only helpful
+ * for maintaining the shuffle, but not for creating the initial
+ * shuffle. Shuffle the whole zone to make sure the just onlined pages
+ * are properly distributed across the whole freelist.
+ */
shuffle_zone(zone);
node_states_set_node(nid, &arg);
--
2.26.2
