[PATCH v2 11/19] sched/numa: Restrict migrating in parallel to the same node.

[Srikar Dronamraju]

Since task migration under numa balancing can happen in parallel, more
than one task might choose to move to the same node at the same time.
This can cause load imbalances at the node level.

The problem is more likely if there are more cores per node or more
nodes in system.

Use a per-node variable to indicate if task migration
to the node under numa balance is currently active.
This per-node variable will not track swapping of tasks.

Running SPECjbb2005 on a 4 node machine and comparing bops/JVM
JVMS  LAST_PATCH  WITH_PATCH  %CHANGE
16    25436.1     25657.9     0.87
1     74031       74435       0.54

Running SPECjbb2005 on a 16 node machine and comparing bops/JVM
JVMS  LAST_PATCH  WITH_PATCH  %CHANGE
8     110355      101748      -7.79
1     178401      170818      -4.25

(numbers from v1 based on v4.17-rc5)
Testcase       Time:         Min         Max         Avg      StdDev
numa01.sh      Real:      414.64      819.20      556.08      147.70
numa01.sh       Sys:       77.52      205.04      139.40       52.05
numa01.sh      User:    37043.24    61757.88    45517.48     9290.38
numa02.sh      Real:       60.80       63.32       61.63        0.88
numa02.sh       Sys:       17.35       39.37       25.71        7.33
numa02.sh      User:     5213.79     5374.73     5268.90       55.09
numa03.sh      Real:      780.09      948.64      831.43       63.02
numa03.sh       Sys:      104.96      136.92      116.31       11.34
numa03.sh      User:    60465.42    73339.78    64368.03     4700.14
numa04.sh      Real:      412.60      681.92      521.29       96.64
numa04.sh       Sys:      210.32      314.10      251.77       37.71
numa04.sh      User:    34026.38    45581.20    38534.49     4198.53
numa05.sh      Real:      394.79      439.63      411.35       16.87
numa05.sh       Sys:      238.32      330.09      292.31       38.32
numa05.sh      User:    33456.45    34876.07    34138.62      609.45

Testcase       Time:         Min         Max         Avg      StdDev 	 %Change
numa01.sh      Real:      434.84      676.90      550.53      106.24 	 1.008%
numa01.sh       Sys:      125.98      217.34      179.41       30.35 	 -22.3%
numa01.sh      User:    38318.48    53789.56    45864.17     6620.80 	 -0.75%
numa02.sh      Real:       60.06       61.27       60.59        0.45 	 1.716%
numa02.sh       Sys:       14.25       17.86       16.09        1.28 	 59.78%
numa02.sh      User:     5190.13     5225.67     5209.24       13.19 	 1.145%
numa03.sh      Real:      748.21      960.25      823.15       73.51 	 1.005%
numa03.sh       Sys:       96.68      122.10      110.42       11.29 	 5.334%
numa03.sh      User:    58222.16    72595.27    63552.22     5048.87 	 1.283%
numa04.sh      Real:      433.08      630.55      499.30       68.15 	 4.404%
numa04.sh       Sys:      245.22      386.75      306.09       63.32 	 -17.7%
numa04.sh      User:    35014.68    46151.72    38530.26     3924.65 	 0.010%
numa05.sh      Real:      394.77      410.07      401.41        5.99 	 2.476%
numa05.sh       Sys:      212.40      301.82      256.23       35.41 	 14.08%
numa05.sh      User:    33224.86    34201.40    33665.61      313.40 	 1.405%

Signed-off-by: Srikar Dronamraju <srikar@xxxxxxxxxxxxxxxxxx>
---
 include/linux/mmzone.h |  1 +
 kernel/sched/fair.c    | 14 ++++++++++++++
 mm/page_alloc.c        |  1 +
 3 files changed, 16 insertions(+)

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 32699b2..b0767703 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -677,6 +677,7 @@ struct zonelist {
 
 	/* Number of pages migrated during the rate limiting time interval */
 	unsigned long numabalancing_migrate_nr_pages;
+	int active_node_migrate;
 #endif
 	/*
 	 * This is a per-node reserve of pages that are not available
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 50c7727..87fb20e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1478,11 +1478,22 @@ struct task_numa_env {
 static void task_numa_assign(struct task_numa_env *env,
 			     struct task_struct *p, long imp)
 {
+	pg_data_t *pgdat = NODE_DATA(cpu_to_node(env->dst_cpu));
 	struct rq *rq = cpu_rq(env->dst_cpu);
 
 	if (xchg(&rq->numa_migrate_on, 1))
 		return;
 
+	if (!env->best_task && env->best_cpu != -1)
+		WRITE_ONCE(pgdat->active_node_migrate, 0);
+
+	if (!p) {
+		if (xchg(&pgdat->active_node_migrate, 1)) {
+			WRITE_ONCE(rq->numa_migrate_on, 0);
+			return;
+		}
+	}
+
 	if (env->best_cpu != -1) {
 		rq = cpu_rq(env->best_cpu);
 		WRITE_ONCE(rq->numa_migrate_on, 0);
@@ -1818,8 +1829,11 @@ static int task_numa_migrate(struct task_struct *p)
 
 	best_rq = cpu_rq(env.best_cpu);
 	if (env.best_task == NULL) {
+		pg_data_t *pgdat = NODE_DATA(cpu_to_node(env.dst_cpu));
+
 		ret = migrate_task_to(p, env.best_cpu);
 		WRITE_ONCE(best_rq->numa_migrate_on, 0);
+		WRITE_ONCE(pgdat->active_node_migrate, 0);
 		if (ret != 0)
 			trace_sched_stick_numa(p, env.src_cpu, env.best_cpu);
 		return ret;
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 22320ea27..8a522d2 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6209,6 +6209,7 @@ static void __paginginit free_area_init_core(struct pglist_data *pgdat)
 #ifdef CONFIG_NUMA_BALANCING
 	spin_lock_init(&pgdat->numabalancing_migrate_lock);
 	pgdat->numabalancing_migrate_nr_pages = 0;
+	pgdat->active_node_migrate = 0;
 	pgdat->numabalancing_migrate_next_window = jiffies;
 #endif
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-- 
1.8.3.1