sched/numa: Favor placing a task on the preferred node
authorMel Gorman <mgorman@suse.de>
Mon, 7 Oct 2013 10:29:18 +0000 (11:29 +0100)
committerIngo Molnar <mingo@kernel.org>
Wed, 9 Oct 2013 12:47:41 +0000 (14:47 +0200)
A tasks preferred node is selected based on the number of faults
recorded for a node but the actual task_numa_migate() conducts a global
search regardless of the preferred nid. This patch checks if the
preferred nid has capacity and if so, searches for a CPU within that
node. This avoids a global search when the preferred node is not
overloaded.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reviewed-by: Rik van Riel <riel@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1381141781-10992-41-git-send-email-mgorman@suse.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c

index 1422765d4b86b950243179b228520be2779a768c..09aac90df89e86938b57f3af3af7b74f2caf12ae 100644 (file)
@@ -1052,6 +1052,20 @@ unlock:
        rcu_read_unlock();
 }
 
+static void task_numa_find_cpu(struct task_numa_env *env, long imp)
+{
+       int cpu;
+
+       for_each_cpu(cpu, cpumask_of_node(env->dst_nid)) {
+               /* Skip this CPU if the source task cannot migrate */
+               if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(env->p)))
+                       continue;
+
+               env->dst_cpu = cpu;
+               task_numa_compare(env, imp);
+       }
+}
+
 static int task_numa_migrate(struct task_struct *p)
 {
        struct task_numa_env env = {
@@ -1068,7 +1082,8 @@ static int task_numa_migrate(struct task_struct *p)
        };
        struct sched_domain *sd;
        unsigned long faults;
-       int nid, cpu, ret;
+       int nid, ret;
+       long imp;
 
        /*
         * Pick the lowest SD_NUMA domain, as that would have the smallest
@@ -1085,28 +1100,29 @@ static int task_numa_migrate(struct task_struct *p)
 
        faults = task_faults(p, env.src_nid);
        update_numa_stats(&env.src_stats, env.src_nid);
+       env.dst_nid = p->numa_preferred_nid;
+       imp = task_faults(env.p, env.dst_nid) - faults;
+       update_numa_stats(&env.dst_stats, env.dst_nid);
 
-       /* Find an alternative node with relatively better statistics */
-       for_each_online_node(nid) {
-               long imp;
-
-               if (nid == env.src_nid)
-                       continue;
-
-               /* Only consider nodes that recorded more faults */
-               imp = task_faults(p, nid) - faults;
-               if (imp < 0)
-                       continue;
+       /*
+        * If the preferred nid has capacity then use it. Otherwise find an
+        * alternative node with relatively better statistics.
+        */
+       if (env.dst_stats.has_capacity) {
+               task_numa_find_cpu(&env, imp);
+       } else {
+               for_each_online_node(nid) {
+                       if (nid == env.src_nid || nid == p->numa_preferred_nid)
+                               continue;
 
-               env.dst_nid = nid;
-               update_numa_stats(&env.dst_stats, env.dst_nid);
-               for_each_cpu(cpu, cpumask_of_node(nid)) {
-                       /* Skip this CPU if the source task cannot migrate */
-                       if (!cpumask_test_cpu(cpu, tsk_cpus_allowed(p)))
+                       /* Only consider nodes that recorded more faults */
+                       imp = task_faults(env.p, nid) - faults;
+                       if (imp < 0)
                                continue;
 
-                       env.dst_cpu = cpu;
-                       task_numa_compare(&env, imp);
+                       env.dst_nid = nid;
+                       update_numa_stats(&env.dst_stats, env.dst_nid);
+                       task_numa_find_cpu(&env, imp);
                }
        }