sched: Optimize cgroup vs wakeup a bit

We don't need to call update_shares() for each domain we iterate,
just got the largets one.

However, we should call it before wake_affine() as well, so that
that can use up-to-date values too.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

3 files changed, 10 insertions, 22 deletions

diff --git a/kernel/sched.c b/kernel/sched.c
index af04ede6dd2f..5049d959bb26 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -376,13 +376,6 @@ static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
 
 #else
 
-#ifdef CONFIG_SMP
-static int root_task_group_empty(void)
-{
-        return 1;
-}
-#endif
-
 static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
 static inline struct task_group *task_group(struct task_struct *p)
 {
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index acf16a8d934b..722d392b0dac 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1348,7 +1348,7 @@ find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
  */
 static int select_task_rq_fair(struct task_struct *p, int sd_flag, int flags)
 {
-        struct sched_domain *tmp, *sd = NULL;
+        struct sched_domain *tmp, *shares = NULL, *sd = NULL;
         int cpu = smp_processor_id();
         int prev_cpu = task_cpu(p);
         int new_cpu = cpu;
@@ -1387,22 +1387,14 @@ static int select_task_rq_fair(struct task_struct *p, int sd_flag, int flags)
                                 break;
                 }
 
-                switch (sd_flag) {
-                case SD_BALANCE_WAKE:
-                        if (!sched_feat(LB_WAKEUP_UPDATE))
-                                break;
-                case SD_BALANCE_FORK:
-                case SD_BALANCE_EXEC:
-                        if (root_task_group_empty())
-                                break;
-                        update_shares(tmp);
-                default:
-                        break;
-                }
-
                 if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
                     cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
 
+                        if (sched_feat(LB_SHARES_UPDATE)) {
+                                update_shares(tmp);
+                                shares = tmp;
+                        }
+
                         if (wake_affine(tmp, p, sync)) {
                                 new_cpu = cpu;
                                 goto out;
@@ -1417,6 +1409,9 @@ static int select_task_rq_fair(struct task_struct *p, int sd_flag, int flags)
                 sd = tmp;
         }
 
+        if (sd && sd != shares && sched_feat(LB_SHARES_UPDATE))
+                update_shares(sd);
+
         while (sd) {
                 struct sched_group *group;
                 int weight;
diff --git a/kernel/sched_features.h b/kernel/sched_features.h
index fd375675f834..d5059fd761d9 100644
--- a/kernel/sched_features.h
+++ b/kernel/sched_features.h
@@ -107,7 +107,7 @@ SCHED_FEAT(ARCH_POWER, 0)
 SCHED_FEAT(HRTICK, 0)
 SCHED_FEAT(DOUBLE_TICK, 0)
 SCHED_FEAT(LB_BIAS, 1)
-SCHED_FEAT(LB_WAKEUP_UPDATE, 1)
+SCHED_FEAT(LB_SHARES_UPDATE, 1)
 SCHED_FEAT(ASYM_EFF_LOAD, 1)
 
 /*