sched: Implement demand based update_cfs_load()

When the system is busy, dilation of rq->next_balance makes lb->update_shares()
insufficiently frequent for threads which don't sleep (no dequeue/enqueue
updates).  Adjust for this by making demand based updates based on the
accumulation of execution time sufficient to wrap our averaging window.

Signed-off-by: Paul Turner <pjt@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <20101115234938.291159744@google.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/kernel/sched.c b/kernel/sched.c
index dadab4d13875a529f37cd61f4a9fd87b041179e5..e914a716e1d44004005f6b63a541bd5380ac8f83 100644 (file)
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -353,9 +353,16 @@ struct cfs_rq {
         */
        unsigned long h_load;
 
+       /*
+        * Maintaining per-cpu shares distribution for group scheduling
+        *
+        * load_stamp is the last time we updated the load average
+        * load_last is the last time we updated the load average and saw load
+        * load_unacc_exec_time is currently unaccounted execution time
+        */
        u64 load_avg;
        u64 load_period;
-       u64 load_stamp, load_last;
+       u64 load_stamp, load_last, load_unacc_exec_time;
 
        unsigned long load_contribution;
 #endif

diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 33f941dcf88c2a3cad53515c734761b060a4cad5..e7e2f08e6d01376ba71daa7eee5f25f23a068f72 100644 (file)
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -539,6 +539,9 @@ static u64 sched_vslice(struct cfs_rq *cfs_rq, struct sched_entity *se)
        return calc_delta_fair(sched_slice(cfs_rq, se), se);
 }
 
+static void update_cfs_load(struct cfs_rq *cfs_rq);
+static void update_cfs_shares(struct cfs_rq *cfs_rq, long weight_delta);
+
 /*
  * Update the current task's runtime statistics. Skip current tasks that
  * are not in our scheduling class.
@@ -558,6 +561,14 @@ __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
 
        curr->vruntime += delta_exec_weighted;
        update_min_vruntime(cfs_rq);
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+       cfs_rq->load_unacc_exec_time += delta_exec;
+       if (cfs_rq->load_unacc_exec_time > sysctl_sched_shares_window) {
+               update_cfs_load(cfs_rq);
+               update_cfs_shares(cfs_rq, 0);
+       }
+#endif
 }
 
 static void update_curr(struct cfs_rq *cfs_rq)
@@ -713,6 +724,7 @@ static void update_cfs_load(struct cfs_rq *cfs_rq)
        }
 
        cfs_rq->load_stamp = now;
+       cfs_rq->load_unacc_exec_time = 0;
        cfs_rq->load_period += delta;
        if (load) {
                cfs_rq->load_last = now;