On 4 January 2017 at 18:20, Dietmar Eggemann [off-list ref] wrote:

On 21/12/16 15:50, Vincent Guittot wrote:

IMHO, the overall idea makes sense to me. Just a couple of small
questions ...

quoted

The update of the share of a cfs_rq is done when its load_avg is updated
but before the group_entity's load_avg has been updated for the past time
slot. This generates wrong load_avg accounting which can be significant
when small tasks are involved in the scheduling.

Why for small tasks? Is it because we use load =
scale_load_down(cfs_rq->load.weight) and not load = cfs_rq->avg.load_avg
in calc_cfs_shares()?

small tasks are significantly impacted because the tick has less
chance to fire while task is running and as a result less chance to
update group entity's load_avg of the past tick with the correct
weight and minimize the impact of this bug.

quoted

Let take the example of a task a that is dequeued of its task group A:
   root
  (cfs_rq)
    \
    (se)
     A
    (cfs_rq)
      \
      (se)
       a

Task "a" was the only task in task group A which becomes idle when a is
dequeued.

We have the sequence:

- dequeue_entity a->se
    - update_load_avg(a->se)
    - dequeue_entity_load_avg(A->cfs_rq, a->se)
    - update_cfs_shares(A->cfs_rq)
      A->cfs_rq->load.weight == 0

You mean A->cfs_rq->load.weight = 0 ?

no A->cfs_rq->load.weight == 0. Just to point out that
A->cfs_rq->load.weight equals 0 at that time.
A->cfs_rq->load.weight has been set to 0 in account_entity_dequeue()
which is not described here and happen between
dequeue_entity_load_avg() and update_cfs_shares()

quoted

        A->se->load.weight is updated with the new share (0 in this case)

Shouldn't this be MIN_SHARES (2) instead of 0?

yes you're right
I had in mind scale_load_down(se->load.weight) which is used when
calling  __update_load_avg() in update_load_avg() and which is 0 on
the 64bits platform that I used

quoted

- dequeue_entity A->se
    - update_load_avg(A->se) but its weight is now null so the last time
      slot (up to a tick) will be accounted with a weight of 0 instead of
      its real weight during the time slot. The last time slot will be
      accounted as an idle one whereas it was a running one.

If the running time of task a is short enough that no tick happens when it
runs, all running time of group entity A->se will be accounted as idle
time.

Instead, we should update the share of a cfs_rq (in fact the weight of its
group entity) only after having updated the load_avg of the group_entity.

This is because we use 'se->on_rq * scale_load_down(se->load.weight)' in
__update_load_avg() as weight parameter for PELT load_avg update?

yes

quoted

update_cfs_shares() now take the sched_entity as parameter instead of the
cfs_rq and  the weight of the group_entity is updated only once its load_avg
has been synced with current time.

[...]