[RFC tg_shares_up improvements - v1 03/12] sched: make tg_shares_up() walk on-demand

From: pjt
Date: Sat Oct 16 2010 - 00:53:55 EST


From: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>

Make tg_shares_up() use the active cgroup list, this means we cannot
do a strict bottom-up walk of the hierarchy, but assuming its a very
wide tree with a small number of active groups it should be a win.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
Signed-off-by: Paul Turner <pjt@xxxxxxxxxx>

---
kernel/sched.c | 67 ----------------------------------------------------
kernel/sched_fair.c | 58 +++++++++++++++++++++++++++++++++++++++++++++
2 files changed, 58 insertions(+), 67 deletions(-)

Index: kernel/sched.c
===================================================================
--- kernel/sched.c.orig
+++ kernel/sched.c
@@ -281,13 +281,6 @@ static DEFINE_SPINLOCK(task_group_lock);

#ifdef CONFIG_FAIR_GROUP_SCHED

-#ifdef CONFIG_SMP
-static int root_task_group_empty(void)
-{
- return list_empty(&root_task_group.children);
-}
-#endif
-
# define INIT_TASK_GROUP_LOAD NICE_0_LOAD

/*
@@ -1531,48 +1524,6 @@ static unsigned long cpu_avg_load_per_ta

#ifdef CONFIG_FAIR_GROUP_SCHED

-static void update_cfs_load(struct cfs_rq *cfs_rq, int lb);
-static void update_cfs_shares(struct cfs_rq *cfs_rq);
-
-/*
- * update tg->load_weight by folding this cpu's load_avg
- */
-static int tg_shares_up(struct task_group *tg, void *data)
-{
- long load_avg;
- struct cfs_rq *cfs_rq;
- unsigned long flags;
- int cpu = (long)data;
- struct rq *rq;
-
- if (!tg->se[cpu])
- return 0;
-
- rq = cpu_rq(cpu);
- cfs_rq = tg->cfs_rq[cpu];
-
- raw_spin_lock_irqsave(&rq->lock, flags);
-
- update_rq_clock(rq);
- update_cfs_load(cfs_rq, 1);
-
- load_avg = div64_u64(cfs_rq->load_avg, cfs_rq->load_period+1);
- load_avg -= cfs_rq->load_contribution;
-
- atomic_add(load_avg, &tg->load_weight);
- cfs_rq->load_contribution += load_avg;
-
- /*
- * We need to update shares after updating tg->load_weight in
- * order to adjust the weight of groups with long running tasks.
- */
- update_cfs_shares(cfs_rq);
-
- raw_spin_unlock_irqrestore(&rq->lock, flags);
-
- return 0;
-}
-
/*
* Compute the cpu's hierarchical load factor for each task group.
* This needs to be done in a top-down fashion because the load of a child
@@ -1596,29 +1547,11 @@ static int tg_load_down(struct task_grou
return 0;
}

-static void update_shares(long cpu)
-{
- if (root_task_group_empty())
- return;
-
- /*
- * XXX: replace with an on-demand list
- */
-
- walk_tg_tree(tg_nop, tg_shares_up, (void *)cpu);
-}
-
static void update_h_load(long cpu)
{
walk_tg_tree(tg_load_down, tg_nop, (void *)cpu);
}

-#else
-
-static inline void update_shares(int cpu)
-{
-}
-
#endif

#ifdef CONFIG_PREEMPT
Index: kernel/sched_fair.c
===================================================================
--- kernel/sched_fair.c.orig
+++ kernel/sched_fair.c
@@ -2000,6 +2000,60 @@ out:
}

#ifdef CONFIG_FAIR_GROUP_SCHED
+/*
+ * update tg->load_weight by folding this cpu's load_avg
+ */
+static int tg_shares_up(struct task_group *tg, int cpu)
+{
+ struct cfs_rq *cfs_rq;
+ unsigned long flags;
+ struct rq *rq;
+ long load_avg;
+
+ if (!tg->se[cpu])
+ return 0;
+
+ rq = cpu_rq(cpu);
+ cfs_rq = tg->cfs_rq[cpu];
+
+ raw_spin_lock_irqsave(&rq->lock, flags);
+
+ update_rq_clock(rq);
+ update_cfs_load(cfs_rq, 1);
+
+ load_avg = div64_u64(cfs_rq->load_avg, cfs_rq->load_period+1);
+ load_avg -= cfs_rq->load_contribution;
+ atomic_add(load_avg, &tg->load_weight);
+ cfs_rq->load_contribution += load_avg;
+
+ /*
+ * We need to update shares after updating tg->load_weight in
+ * order to adjust the weight of groups with long running tasks.
+ */
+ update_cfs_shares(cfs_rq);
+
+ raw_spin_unlock_irqrestore(&rq->lock, flags);
+
+ return 0;
+}
+
+static void update_shares(int cpu)
+{
+ struct cfs_rq *cfs_rq;
+ struct rq *rq = cpu_rq(cpu);
+
+ rcu_read_lock();
+ for_each_leaf_cfs_rq(rq, cfs_rq) {
+ struct task_group *tg = cfs_rq->tg;
+
+ do {
+ tg_shares_up(tg, cpu);
+ tg = tg->parent;
+ } while (tg);
+ }
+ rcu_read_unlock();
+}
+
static unsigned long
load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
unsigned long max_load_move,
@@ -2047,6 +2101,10 @@ load_balance_fair(struct rq *this_rq, in
return max_load_move - rem_load_move;
}
#else
+static inline void update_shares(int cpu)
+{
+}
+
static unsigned long
load_balance_fair(struct rq *this_rq, int this_cpu, struct rq *busiest,
unsigned long max_load_move,

--

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/