On 05/03/2020 18:29, Vincent Guittot wrote:
> When a cfs rq is throttled, the latter and its child are removed from the
> leaf list but their nr_running is not changed which includes staying higher
> than 1. When a task is enqueued in this throttled branch, the cfs rqs must
> be added back in order to ensure correct ordering in the list but this can
> only happens if nr_running == 1.
> When cfs bandwidth is used, we call unconditionnaly list_add_leaf_cfs_rq()
> when enqueuing an entity to make sure that the complete branch will be
> added.
>
> Reported-by: Christian Borntraeger <borntraeger@xxxxxxxxxx>
> Tested-by: Christian Borntraeger <borntraeger@xxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx #v5.1+
> Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> ---
> kernel/sched/fair.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index fcc968669aea..bdc5bb72ab31 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -4117,6 +4117,7 @@ static inline void check_schedstat_required(void)
> #endif
> }
>
> +static inline bool cfs_bandwidth_used(void);
>
> /*
> * MIGRATION
> @@ -4195,10 +4196,16 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
> __enqueue_entity(cfs_rq, se);
> se->on_rq = 1;
>
> - if (cfs_rq->nr_running == 1) {
> + /*
> + * When bandwidth control is enabled, cfs might have been removed because of
> + * a parent been throttled but cfs->nr_running > 1. Try to add it
> + * unconditionnally.
> + */
> + if (cfs_rq->nr_running == 1 || cfs_bandwidth_used())
> list_add_leaf_cfs_rq(cfs_rq);
> +
> + if (cfs_rq->nr_running == 1)
> check_enqueue_throttle(cfs_rq);
> - }
> }
>
> static void __clear_buddies_last(struct sched_entity *se)
I experimented with an rt-app based setup on Arm64 Juno (6 CPUs):
cgroupv1 hierarchy A/B/C, all CFS bw controlled (30,000/100,000)
I create A/B/C outside rt-app so I can have rt-app runs with an already
existing taskgroup hierarchy. There is a 4 secs gap between consecutive
rt-app runs.
The rt-app files contains 6 periodic CFS tasks (25,000/100,000) running
in /A/B/C, /A/B, /A (3 rt-app task phases).
I get w/ the patch (and the debug patch applied to unthrottle_cfs_rq()):
root@juno:~#
[ 409.236925] CPU1 path=/A/B on_list=1 nr_running=1 throttled=1
[ 409.242682] CPU1 path=/A on_list=0 nr_running=0 throttled=1
[ 409.248260] CPU1 path=/ on_list=1 nr_running=0 throttled=0
[ 409.253748] ------------[ cut here ]------------
[ 409.258365] rq->tmp_alone_branch != &rq->leaf_cfs_rq_list
[ 409.258382] WARNING: CPU: 1 PID: 0 at kernel/sched/fair.c:380
unthrottle_cfs_rq+0x21c/0x2a8
...
[ 409.275196] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 5.6.0-rc3-dirty #62
[ 409.281990] Hardware name: ARM Juno development board (r0) (DT)
...
[ 409.384644] Call trace:
[ 409.387089] unthrottle_cfs_rq+0x21c/0x2a8
[ 409.391188] distribute_cfs_runtime+0xf4/0x198
[ 409.395634] sched_cfs_period_timer+0x134/0x240
[ 409.400168] __hrtimer_run_queues+0x10c/0x3c0
[ 409.404527] hrtimer_interrupt+0xd4/0x250
[ 409.408539] tick_handle_oneshot_broadcast+0x17c/0x208
[ 409.413683] sp804_timer_interrupt+0x30/0x40
If I add the following snippet the issue goes away:
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index e9fd5379bb7e..5e03be046aba 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4627,11 +4627,17 @@ void unthrottle_cfs_rq(struct cfs_rq *cfs_rq)
break;
}
- assert_list_leaf_cfs_rq(rq);
-
if (!se)
add_nr_running(rq, task_delta);
+ for_each_sched_entity(se) {
+ cfs_rq = cfs_rq_of(se);
+
+ list_add_leaf_cfs_rq(cfs_rq);
+ }
+
+ assert_list_leaf_cfs_rq(rq);
+
/* Determine whether we need to wake up potentially idle CPU: */
if (rq->curr == rq->idle && rq->cfs.nr_running)
resched_curr(rq);
