[PATCH v2 2/2] sched/fair: Don't increase sd->balance_interval on newidle balance

From: Valentin Schneider
Date: Wed Sep 26 2018 - 11:13:44 EST


When load_balance() fails to move some load because of task affinity,
we end up increasing sd->balance_interval to delay the next periodic
balance in the hopes that next time we look, that annoying pinned
task(s) will be gone.

However, idle_balance() pays no attention to sd->balance_interval, yet
it will still lead to an increase in balance_interval in case of
pinned tasks.

If we're going through several newidle balances (e.g. we have a
periodic task), this can lead to a huge increase of the
balance_interval in a very small amount of time.

To prevent that, don't increase the balance interval when going
through a newidle balance.

This is a similar approach to what is done in commit 58b26c4c0257
("sched: Increment cache_nice_tries only on periodic lb"), where we
disregard newidle balance and rely on periodic balance for more stable
results.

Signed-off-by: Valentin Schneider <valentin.schneider@xxxxxxx>
---
kernel/sched/fair.c | 13 +++++++++++--
1 file changed, 11 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 9cf93ba..4c33283 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -8782,13 +8782,22 @@ static int load_balance(int this_cpu, struct rq *this_rq,
sd->nr_balance_failed = 0;

out_one_pinned:
+ ld_moved = 0;
+
+ /*
+ * idle_balance() disregards balance intervals, so we could repeatedly
+ * reach this code, which would lead to balance_interval skyrocketting
+ * in a short amount of time. Skip the balance_interval increase logic
+ * to avoid that.
+ */
+ if (env.idle == CPU_NEWLY_IDLE)
+ goto out;
+
/* tune up the balancing interval */
if ((env.flags & LBF_ALL_PINNED &&
sd->balance_interval < MAX_PINNED_INTERVAL) ||
sd->balance_interval < sd->max_interval)
sd->balance_interval *= 2;
-
- ld_moved = 0;
out:
return ld_moved;
}
--
2.7.4