mm/vmscan.c: prevent useless kswapd loops

In production we have noticed hard lockups on large machines running
large jobs due to kswaps hoarding lru lock within isolate_lru_pages when
sc->reclaim_idx is 0 which is a small zone.  The lru was couple hundred
GiBs and the condition (page_zonenum(page) > sc->reclaim_idx) in
isolate_lru_pages() was basically skipping GiBs of pages while holding
the LRU spinlock with interrupt disabled.

On further inspection, it seems like there are two issues:

(1) If kswapd on the return from balance_pgdat() could not sleep (i.e.
    node is still unbalanced), the classzone_idx is unintentionally set
    to 0 and the whole reclaim cycle of kswapd will try to reclaim only
    the lowest and smallest zone while traversing the whole memory.

(2) Fundamentally isolate_lru_pages() is really bad when the
    allocation has woken kswapd for a smaller zone on a very large machine
    running very large jobs.  It can hoard the LRU spinlock while skipping
    over 100s of GiBs of pages.

This patch only fixes (1).  (2) needs a more fundamental solution.  To
fix (1), in the kswapd context, if pgdat->kswapd_classzone_idx is
invalid use the classzone_idx of the previous kswapd loop otherwise use
the one the waker has requested.

Link: http://lkml.kernel.org/r/[email protected]
Fixes: e716f2e ("mm, vmscan: prevent kswapd sleeping prematurely due to mismatched classzone_idx")
Signed-off-by: Shakeel Butt <[email protected]>
Reviewed-by: Yang Shi <[email protected]>
Acked-by: Mel Gorman <[email protected]>
Cc: Johannes Weiner <[email protected]>
Cc: Michal Hocko <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Hillf Danton <[email protected]>
Cc: Roman Gushchin <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

mm/vmscan.c

@@ -3644,19 +3644,18 @@ static int balance_pgdat(pg_data_t *pgdat, int order, int classzone_idx)
 }
 
 /*
- * pgdat->kswapd_classzone_idx is the highest zone index that a recent
- * allocation request woke kswapd for. When kswapd has not woken recently,
- * the value is MAX_NR_ZONES which is not a valid index. This compares a
- * given classzone and returns it or the highest classzone index kswapd
- * was recently woke for.
+ * The pgdat->kswapd_classzone_idx is used to pass the highest zone index to be
+ * reclaimed by kswapd from the waker. If the value is MAX_NR_ZONES which is not
+ * a valid index then either kswapd runs for first time or kswapd couldn't sleep
+ * after previous reclaim attempt (node is still unbalanced). In that case
+ * return the zone index of the previous kswapd reclaim cycle.
  */
 static enum zone_type kswapd_classzone_idx(pg_data_t *pgdat,
-					   enum zone_type classzone_idx)
+					   enum zone_type prev_classzone_idx)
 {
 	if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
-		return classzone_idx;
-
-	return max(pgdat->kswapd_classzone_idx, classzone_idx);
+		return prev_classzone_idx;
+	return pgdat->kswapd_classzone_idx;
 }
 
 static void kswapd_try_to_sleep(pg_data_t *pgdat, int alloc_order, int reclaim_order,
@@ -3797,7 +3796,7 @@ static int kswapd(void *p)
 
 		/* Read the new order and classzone_idx */
 		alloc_order = reclaim_order = pgdat->kswapd_order;
-		classzone_idx = kswapd_classzone_idx(pgdat, 0);
+		classzone_idx = kswapd_classzone_idx(pgdat, classzone_idx);
 		pgdat->kswapd_order = 0;
 		pgdat->kswapd_classzone_idx = MAX_NR_ZONES;
 
@@ -3851,8 +3850,12 @@ void wakeup_kswapd(struct zone *zone, gfp_t gfp_flags, int order,
 	if (!cpuset_zone_allowed(zone, gfp_flags))
 		return;
 	pgdat = zone->zone_pgdat;
-	pgdat->kswapd_classzone_idx = kswapd_classzone_idx(pgdat,
-							   classzone_idx);
+
+	if (pgdat->kswapd_classzone_idx == MAX_NR_ZONES)
+		pgdat->kswapd_classzone_idx = classzone_idx;
+	else
+		pgdat->kswapd_classzone_idx = max(pgdat->kswapd_classzone_idx,
+						  classzone_idx);
 	pgdat->kswapd_order = max(pgdat->kswapd_order, order);
 	if (!waitqueue_active(&pgdat->kswapd_wait))
 		return;