[linux-yocto] [PATCH 22/28] sched/numa: Allow task switch if load imbalance improves

[Yang Shi]

From: Rik van Riel <riel at redhat.com>

commit e63da03639cc9e6e83b62e7ef8ffdbb92421416a upstream

Currently the NUMA balancing code only allows moving tasks between NUMA
nodes when the load on both nodes is in balance. This breaks down when
the load was imbalanced to begin with.

Allow tasks to be moved between NUMA nodes if the imbalance is small,
or if the new imbalance is be smaller than the original one.

Suggested-by: Peter Zijlstra <peterz at infradead.org>
Signed-off-by: Rik van Riel <riel at redhat.com>
Signed-off-by: Peter Zijlstra <peterz at infradead.org>
Cc: mgorman at suse.de
Cc: chegu_vinod at hp.com
Signed-off-by: Ingo Molnar <mingo at kernel.org>
Link: http://lkml.kernel.org/r/20140514132221.274b3463@annuminas.surriel.com
Signed-off-by: Yang Shi <yang.shi at windriver.com>
---
 kernel/sched/fair.c | 46 ++++++++++++++++++++++++++++++++++++----------
 1 file changed, 36 insertions(+), 10 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e6b287e..7f361d4 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1112,6 +1112,34 @@ static void task_numa_assign(struct task_numa_env *env,
 	env->best_cpu = env->dst_cpu;
 }
 
+static bool load_too_imbalanced(long orig_src_load, long orig_dst_load,
+				long src_load, long dst_load,
+				struct task_numa_env *env)
+{
+	long imb, old_imb;
+
+	/* We care about the slope of the imbalance, not the direction. */
+	if (dst_load < src_load)
+		swap(dst_load, src_load);
+
+	/* Is the difference below the threshold? */
+	imb = dst_load * 100 - src_load * env->imbalance_pct;
+	if (imb <= 0)
+		return false;
+
+	/*
+	 * The imbalance is above the allowed threshold.
+	 * Compare it with the old imbalance.
+	 */
+	if (orig_dst_load < orig_src_load)
+		swap(orig_dst_load, orig_src_load);
+
+	old_imb = orig_dst_load * 100 - orig_src_load * env->imbalance_pct;
+
+	/* Would this change make things worse? */
+	return (old_imb > imb);
+}
+
 /*
  * This checks if the overall compute and NUMA accesses of the system would
  * be improved if the source tasks was migrated to the target dst_cpu taking
@@ -1124,7 +1152,8 @@ static void task_numa_compare(struct task_numa_env *env,
 	struct rq *src_rq = cpu_rq(env->src_cpu);
 	struct rq *dst_rq = cpu_rq(env->dst_cpu);
 	struct task_struct *cur;
-	long dst_load, src_load;
+	long orig_src_load, src_load;
+	long orig_dst_load, dst_load;
 	long load;
 	long imp = (groupimp > 0) ? groupimp : taskimp;
 
@@ -1198,13 +1227,13 @@ static void task_numa_compare(struct task_numa_env *env,
 	 * In the overloaded case, try and keep the load balanced.
 	 */
 balance:
-	dst_load = env->dst_stats.load;
-	src_load = env->src_stats.load;
+	orig_dst_load = env->dst_stats.load;
+	orig_src_load = env->src_stats.load;
 
 	/* XXX missing power terms */
 	load = task_h_load(env->p);
-	dst_load += load;
-	src_load -= load;
+	dst_load = orig_dst_load + load;
+	src_load = orig_src_load - load;
 
 	if (cur) {
 		load = task_h_load(cur);
@@ -1212,11 +1241,8 @@ balance:
 		src_load += load;
 	}
 
-	/* make src_load the smaller */
-	if (dst_load < src_load)
-		swap(dst_load, src_load);
-
-	if (src_load * env->imbalance_pct < dst_load * 100)
+	if (load_too_imbalanced(orig_src_load, orig_dst_load,
+				src_load, dst_load, env))
 		goto unlock;
 
 assign:
-- 
2.0.2