]> err.no Git - linux-2.6/commitdiff
sched: wakeup granularity increase
authorIngo Molnar <mingo@elte.hu>
Mon, 15 Oct 2007 15:00:05 +0000 (17:00 +0200)
committerIngo Molnar <mingo@elte.hu>
Mon, 15 Oct 2007 15:00:05 +0000 (17:00 +0200)
increase wakeup granularity - we were overscheduling a bit.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Mike Galbraith <efault@gmx.de>
kernel/sched_fair.c

index 45c7493d8ca879e72ec2d23615cfee962615de1e..a60b1dac598a52693f9036ab9b2cdd4506b099b0 100644 (file)
@@ -74,7 +74,7 @@ const_debug unsigned int sysctl_sched_batch_wakeup_granularity = 25000000UL;
  * and reduces their over-scheduling. Synchronous workloads will still
  * have immediate wakeup/sleep latencies.
  */
-const_debug unsigned int sysctl_sched_wakeup_granularity = 1000000UL;
+const_debug unsigned int sysctl_sched_wakeup_granularity = 2000000UL;
 
 unsigned int sysctl_sched_runtime_limit __read_mostly;
 
@@ -582,7 +582,7 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int sleep)
  * Preempt the current task with a newly woken task if needed:
  */
 static void
-__check_preempt_curr_fair(struct cfs_rq *cfs_rq, struct sched_entity *curr)
+check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
 {
        unsigned long ideal_runtime, delta_exec;
 
@@ -646,8 +646,6 @@ static void put_prev_entity(struct cfs_rq *cfs_rq, struct sched_entity *prev)
 
 static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
 {
-       struct sched_entity *next;
-
        /*
         * Dequeue and enqueue the task to update its
         * position within the tree:
@@ -655,14 +653,8 @@ static void entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
        dequeue_entity(cfs_rq, curr, 0);
        enqueue_entity(cfs_rq, curr, 0);
 
-       /*
-        * Reschedule if another task tops the current one.
-        */
-       next = __pick_next_entity(cfs_rq);
-       if (next == curr)
-               return;
-
-       __check_preempt_curr_fair(cfs_rq, curr);
+       if (cfs_rq->nr_running > 1)
+               check_preempt_tick(cfs_rq, curr);
 }
 
 /**************************************************
@@ -852,7 +844,7 @@ static void yield_task_fair(struct rq *rq, struct task_struct *p)
 /*
  * Preempt the current task with a newly woken task if needed:
  */
-static void check_preempt_curr_fair(struct rq *rq, struct task_struct *p)
+static void check_preempt_wakeup(struct rq *rq, struct task_struct *p)
 {
        struct task_struct *curr = rq->curr;
        struct cfs_rq *cfs_rq = task_cfs_rq(curr);
@@ -863,9 +855,12 @@ static void check_preempt_curr_fair(struct rq *rq, struct task_struct *p)
                resched_task(curr);
                return;
        }
+       if (is_same_group(curr, p)) {
+               s64 delta = curr->se.vruntime - p->se.vruntime;
 
-       if (is_same_group(curr, p))
-               __check_preempt_curr_fair(cfs_rq, &curr->se);
+               if (delta > (s64)sysctl_sched_wakeup_granularity)
+                       resched_task(curr);
+       }
 }
 
 static struct task_struct *pick_next_task_fair(struct rq *rq)
@@ -1095,7 +1090,7 @@ struct sched_class fair_sched_class __read_mostly = {
        .dequeue_task           = dequeue_task_fair,
        .yield_task             = yield_task_fair,
 
-       .check_preempt_curr     = check_preempt_curr_fair,
+       .check_preempt_curr     = check_preempt_wakeup,
 
        .pick_next_task         = pick_next_task_fair,
        .put_prev_task          = put_prev_task_fair,