mm/vmscan: do not check compaction_ready on promoted zones

We abort direct reclaim if we find the zone is ready for compaction.
Sometimes the zone is just a promoted highmem zone to force a scan of
highmem, which is not the intended zone the caller want to allocate a
page from.  In this situation, setting aborted_reclaim to indicate the
caller turned back to retry the allocation is waste of time and could
cause a loop in __alloc_pages_slowpath().

This patch does not check compaction_ready() on promoted zones to avoid
the above situation.  Only set aborted_reclaim if the caller intended
zone is ready for compaction.

Signed-off-by: Weijie Yang <[email protected]>
Acked-by: Rik van Riel <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/vmscan.c

@@ -2318,6 +2318,7 @@ static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
 	struct shrink_control shrink = {
 		.gfp_mask = sc->gfp_mask,
 	};
+	enum zone_type requested_highidx = gfp_zone(sc->gfp_mask);
 
 	/*
 	 * If the number of buffer_heads in the machine exceeds the maximum
@@ -2358,7 +2359,8 @@ static bool shrink_zones(struct zonelist *zonelist, struct scan_control *sc)
 				 * noticeable problem, like transparent huge
 				 * page allocations.
 				 */
-				if (compaction_ready(zone, sc)) {
+				if ((zonelist_zone_idx(z) <= requested_highidx)
+				    && compaction_ready(zone, sc)) {
 					aborted_reclaim = true;
 					continue;
 				}