sched: Add new wakeup preemption mode: WAKEUP_RUNNING

Create a new wakeup preemption mode, preempt towards tasks that run
shorter on avg. It sets next buddy to be sure we actually run the task
we preempted for.

Test results:

 root@twins:~# while :; do :; done &
 [1] 6537
 root@twins:~# while :; do :; done &
 [2] 6538
 root@twins:~# while :; do :; done &
 [3] 6539
 root@twins:~# while :; do :; done &
 [4] 6540

 root@twins:/home/peter# ./latt -c4 sleep 4
 Entries: 48 (clients=4)

 Averages:
 ------------------------------
        Max          4750 usec
        Avg           497 usec
        Stdev         737 usec

 root@twins:/home/peter# echo WAKEUP_RUNNING > /debug/sched_features

 root@twins:/home/peter# ./latt -c4 sleep 4
 Entries: 48 (clients=4)

 Averages:
 ------------------------------
        Max            14 usec
        Avg             5 usec
        Stdev           3 usec

Disabled by default - needs more testing.

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

4 files changed, 27 insertions, 10 deletions

diff --git a/kernel/sched.c b/kernel/sched.c
index 969dfaef2465..3bb4ea2ee6f0 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -2458,6 +2458,7 @@ static void __sched_fork(struct task_struct *p)
         p->se.avg_overlap               = 0;
         p->se.start_runtime             = 0;
         p->se.avg_wakeup                = sysctl_sched_wakeup_granularity;
+        p->se.avg_running               = 0;
 
 #ifdef CONFIG_SCHEDSTATS
         p->se.wait_start                        = 0;
@@ -5310,14 +5311,13 @@ static inline void schedule_debug(struct task_struct *prev)
 #endif
 }
 
-static void put_prev_task(struct rq *rq, struct task_struct *prev)
+static void put_prev_task(struct rq *rq, struct task_struct *p)
 {
-        if (prev->state == TASK_RUNNING) {
-                u64 runtime = prev->se.sum_exec_runtime;
+        u64 runtime = p->se.sum_exec_runtime - p->se.prev_sum_exec_runtime;
 
-                runtime -= prev->se.prev_sum_exec_runtime;
-                runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
+        update_avg(&p->se.avg_running, runtime);
 
+        if (p->state == TASK_RUNNING) {
                 /*
                  * In order to avoid avg_overlap growing stale when we are
                  * indeed overlapping and hence not getting put to sleep, grow
@@ -5327,9 +5327,12 @@ static void put_prev_task(struct rq *rq, struct task_struct *prev)
                  * correlates to the amount of cache footprint a task can
                  * build up.
                  */
-                update_avg(&prev->se.avg_overlap, runtime);
+                runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
+                update_avg(&p->se.avg_overlap, runtime);
+        } else {
+                update_avg(&p->se.avg_running, 0);
         }
-        prev->sched_class->put_prev_task(rq, prev);
+        p->sched_class->put_prev_task(rq, p);
 }
 
 /*
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index 5ddbd0891267..efb84409bc43 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -395,6 +395,7 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
         PN(se.sum_exec_runtime);
         PN(se.avg_overlap);
         PN(se.avg_wakeup);
+        PN(se.avg_running);
 
         nr_switches = p->nvcsw + p->nivcsw;
 
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index c741cd9d38de..3e6f78c66876 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1605,9 +1605,6 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_
                 return;
         }
 
-        if (!sched_feat(WAKEUP_PREEMPT))
-                return;
-
         if ((sched_feat(WAKEUP_SYNC) && sync) ||
             (sched_feat(WAKEUP_OVERLAP) &&
              (se->avg_overlap < sysctl_sched_migration_cost &&
@@ -1616,6 +1613,17 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int wake_
                 return;
         }
 
+        if (sched_feat(WAKEUP_RUNNING)) {
+                if (pse->avg_running < se->avg_running) {
+                        set_next_buddy(pse);
+                        resched_task(curr);
+                        return;
+                }
+        }
+
+        if (!sched_feat(WAKEUP_PREEMPT))
+                return;
+
         find_matching_se(&se, &pse);
 
         BUG_ON(!pse);
diff --git a/kernel/sched_features.h b/kernel/sched_features.h
index d5059fd761d9..0d94083582c7 100644
--- a/kernel/sched_features.h
+++ b/kernel/sched_features.h
@@ -54,6 +54,11 @@ SCHED_FEAT(WAKEUP_SYNC, 0)
 SCHED_FEAT(WAKEUP_OVERLAP, 0)
 
 /*
+ * Wakeup preemption towards tasks that run short
+ */
+SCHED_FEAT(WAKEUP_RUNNING, 0)
+
+/*
  * Use the SYNC wakeup hint, pipes and the likes use this to indicate
  * the remote end is likely to consume the data we just wrote, and
  * therefore has cache benefit from being placed on the same cpu, see