mm: Drain pages after onlining
When onlining, the onlined pages must be added to the kernel's
list of free pages using __free_page(). However, pages are not
immediately added but placed in a queue to be processed
when the queue size reaches a watermark. The last pages in
the queue may not be processed in time, and if you try to
offline that memory before it is processed, offlining will
always fail.
This fix calls drain_all_pages(), which will process every
free page in the queue. This ensures that all pages are
accounted for when onlining and nothing gets stuck in the queue.
Change-Id: I54dbc0749556702407090e51ce9246abc5db7d1c
Signed-off-by: Jack Cheung <jackc@codeaurora.org>
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
index 70ea0df..f8cf37d 100644
--- a/mm/memory_hotplug.c
+++ b/mm/memory_hotplug.c
@@ -466,6 +466,7 @@
zone->present_pages += onlined_pages;
zone->zone_pgdat->node_present_pages += onlined_pages;
+ drain_all_pages();
if (need_zonelists_rebuild)
build_all_zonelists(zone);
else