diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index 75a717c03800..bd964e5b395f 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -2971,6 +2971,26 @@
 			firmware feature for updating multiple TCE entries
 			at a time.
 
+	kswapd_per_node=
+			kswapd_per_node allows you to control the number of kswapd threads
+			running on the system. This provides the ability to devote additional
+			CPU resources toward proactive page replacement with the goal of
+			reducing direct reclaims. When direct reclaims are prevented, the CPU
+			consumed by them is prevented as well. Depending on the workload, the
+			result can cause aggregate CPU usage on the system to go up, down or
+			stay the same.
+
+			More aggressive page replacement can reduce direct reclaims which
+			cause latency for tasks and decrease throughput when doing filesystem
+			IO through the pagecache. Direct reclaims are recorded using the
+			allocstall counter in /proc/vmstat.
+
+			The range of acceptible values are 1-16. Always start with lower
+			values in the 2-6 range. Higher values should be justified with
+			testing. If direct reclaims occur in spite of high values, the cost
+			of direct reclaims (in latency) that occur can be higher due to
+			increased lock contention.
+
 	onenand.bdry=	[HW,MTD] Flex-OneNAND Boundary Configuration
 
 			Format: [die0_boundary][,die0_lock][,die1_boundary][,die1_lock]
diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 1df06a7f10dd..1ccd1ad16015 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -38,6 +38,8 @@
  */
 #define PAGE_ALLOC_COSTLY_ORDER 3
 
+#define MAX_KSWAPD_THREADS 16
+
 enum migratetype {
 	MIGRATE_UNMOVABLE,
 	MIGRATE_MOVABLE,
@@ -769,6 +771,7 @@ typedef struct pglist_data {
 	wait_queue_head_t pfmemalloc_wait;
 	struct task_struct *kswapd;	/* Protected by
 					   mem_hotplug_begin/end() */
+	struct task_struct *mkswapd[MAX_KSWAPD_THREADS];
 	int kswapd_order;
 	enum zone_type kswapd_highest_zoneidx;
 
diff --git a/mm/vmscan.c b/mm/vmscan.c
index 4c5a9b2286bf..0facbf9930d3 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -171,6 +171,23 @@ struct scan_control {
  */
 int vm_swappiness = 60;
 
+#define DEF_KSWAPD_THREADS_PER_NODE 1
+int kswapd_threads = DEF_KSWAPD_THREADS_PER_NODE;
+static int __init kswapd_per_node_setup(char *str)
+{
+	int tmp;
+
+	if (kstrtoint(str, 0, &tmp) < 0)
+		return 0;
+
+	if (tmp > MAX_KSWAPD_THREADS || tmp <= 0)
+		return 0;
+
+	kswapd_threads = tmp;
+	return 1;
+}
+__setup("kswapd_per_node=", kswapd_per_node_setup);
+
 static void set_task_reclaim_state(struct task_struct *task,
 				   struct reclaim_state *rs)
 {
@@ -3935,6 +3952,46 @@ kswapd_try_sleep:
 	return 0;
 }
 
+static int kswapd_per_node_run(int nid)
+{
+	pg_data_t *pgdat = NODE_DATA(nid);
+	int hid;
+	int ret = 0;
+
+	for (hid = 0; hid < kswapd_threads; ++hid) {
+		pgdat->mkswapd[hid] = kthread_run(kswapd, pgdat, "kswapd%d:%d",
+								nid, hid);
+		if (IS_ERR(pgdat->mkswapd[hid])) {
+			/* failure at boot is fatal */
+			WARN_ON(system_state < SYSTEM_RUNNING);
+			pr_err("Failed to start kswapd%d on node %d\n",
+				hid, nid);
+			ret = PTR_ERR(pgdat->mkswapd[hid]);
+			pgdat->mkswapd[hid] = NULL;
+			continue;
+		}
+		if (!pgdat->kswapd)
+			pgdat->kswapd = pgdat->mkswapd[hid];
+	}
+
+	return ret;
+}
+
+static void kswapd_per_node_stop(int nid)
+{
+	int hid = 0;
+	struct task_struct *kswapd;
+
+	for (hid = 0; hid < kswapd_threads; hid++) {
+		kswapd = NODE_DATA(nid)->mkswapd[hid];
+		if (kswapd) {
+			kthread_stop(kswapd);
+			NODE_DATA(nid)->mkswapd[hid] = NULL;
+		}
+	}
+	NODE_DATA(nid)->kswapd = NULL;
+}
+
 /*
  * A zone is low on free memory or too fragmented for high-order memory.  If
  * kswapd should reclaim (direct reclaim is deferred), wake it up for the zone's
@@ -4038,6 +4095,9 @@ int kswapd_run(int nid)
 	if (pgdat->kswapd)
 		return 0;
 
+	if (kswapd_threads > 1)
+		return kswapd_per_node_run(nid);
+
 	pgdat->kswapd = kthread_run(kswapd, pgdat, "kswapd%d", nid);
 	if (IS_ERR(pgdat->kswapd)) {
 		/* failure at boot is fatal */
@@ -4057,6 +4117,11 @@ void kswapd_stop(int nid)
 {
 	struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
 
+	if (kswapd_threads > 1) {
+		kswapd_per_node_stop(nid);
+		return;
+	}
+
 	if (kswapd) {
 		kthread_stop(kswapd);
 		NODE_DATA(nid)->kswapd = NULL;