Inside rmqueue() there is a dedicated WARN_ON_ONCE() to make sure
__GFP_NOFAIL is not combined with any order > 1 allocation.
However the following locations are doing the same check, but with
order > PAGE_ALLOC_COSTLY_ORDER as the threshold:
- VM_BUG_ON() inside pcp_allowed_order()
- WARN_ON_ONCE_GFP() inside __alloc_pages_may_oom()
So this looks like a behavior mismatch, and it's better to unify all the
__GFP_NOFAIL check using PAGE_ALLOC_COSTLY_ORDER for the order.
Signed-off-by: Qu Wenruo <wqu@xxxxxxxx>
---
mm/page_alloc.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 9ecf99190ea2..697e5cc7f3f1 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -3005,9 +3005,11 @@ struct page *rmqueue(struct zone *preferred_zone,
/*
* We most definitely don't want callers attempting to
- * allocate greater than order-1 page units with __GFP_NOFAIL.
+ * allocate greater than PAGE_ALLOC_COSTLY_ORDER page units
+ * with __GFP_NOFAIL.
*/
- WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
+ WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) &&
+ (order > PAGE_ALLOC_COSTLY_ORDER));
if (likely(pcp_allowed_order(order))) {
page = rmqueue_pcplist(preferred_zone, zone, order,
--
2.45.2
