Re: [PATCH 1/2] sched/fair: Use task_groups instead of leaf_cfs_rq_list to walk all cfs_rqs

[Peter Zijlstra]

On Tue, Apr 25, 2017 at 05:40:39PM -0700, Tejun Heo wrote:

> task_groups list is protected by its own lock.  While it allows RCU
> protected traversal and the order of operations guarantees that all
> online cfs_rqs will be visited, holding rq->lock won't protect against
> iterating an already unregistered cfs_rq.

So I think the below is sufficient.

Yes we can hit an (almost) dead cfs_rq, but poking the bandwidth
variables thereof is harmless.

unthrottle_cfs_rq() also stops doing anything much when it finds the
cfs_rq is empty, which must be the case if we're removing it.

I don't know Paul's opinion on RCU GPs happening while stop_machine(),
but just in case he feels that's fair game, I did add the
rcu_read_lock() thingies.

The lockdep assert is mostly documentation, to more easily see
it is indeed held when we get there.

I left print_cfs_stats using the leaf list, no point in printing stuff
that's empty.

This way we can avoid taking all RQ locks on cgroup destruction.

---
 kernel/sched/fair.c | 21 ++++++++++++++++-----
 1 file changed, 16 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index cd6c3f957d44..c897ad6d714c 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4644,22 +4644,32 @@ static void destroy_cfs_bandwidth(struct cfs_bandwidth 
*cfs_b)
 
 static void __maybe_unused update_runtime_enabled(struct rq *rq)
 {
-       struct cfs_rq *cfs_rq;
+       struct task_group *tg;
 
-       for_each_leaf_cfs_rq(rq, cfs_rq) {
-               struct cfs_bandwidth *cfs_b = &cfs_rq->tg->cfs_bandwidth;
+       lockdep_assert_held(&rq->lock);
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(tg, &task_groups, list) {
+               struct cfs_bandwidth *cfs_b = &tg->cfs_bandwidth;
+               struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
 
                raw_spin_lock(&cfs_b->lock);
                cfs_rq->runtime_enabled = cfs_b->quota != RUNTIME_INF;
                raw_spin_unlock(&cfs_b->lock);
        }
+       rcu_read_unlock();
 }
 
 static void __maybe_unused unthrottle_offline_cfs_rqs(struct rq *rq)
 {
-       struct cfs_rq *cfs_rq;
+       struct task_group *tg;
+
+       lockdep_assert_held(&rq->lock);
+
+       rcu_read_lock();
+       list_for_each_entry_rcu(tg, &task_groups, list) {
+               struct cfs_rq *cfs_rq = tg->cfs_rq[cpu_of(rq)];
 
-       for_each_leaf_cfs_rq(rq, cfs_rq) {
                if (!cfs_rq->runtime_enabled)
                        continue;
 
@@ -4677,6 +4687,7 @@ static void __maybe_unused 
unthrottle_offline_cfs_rqs(struct rq *rq)
                if (cfs_rq_throttled(cfs_rq))
                        unthrottle_cfs_rq(cfs_rq);
        }
+       rcu_read_unlock();
 }
 
 #else /* CONFIG_CFS_BANDWIDTH */