On Mon, Jul 29, 2019 at 01:27:02PM +0200, Juri Lelli wrote: > On 29/07/19 13:15, Peter Zijlstra wrote: > > On Mon, Jul 29, 2019 at 11:25:19AM +0200, Juri Lelli wrote: > > > Hi, > > > > > > On 26/07/19 16:54, Peter Zijlstra wrote: > > > > Because pick_next_task() implies set_curr_task() and some of the > > > > details haven't matter too much, some of what _should_ be in > > > > set_curr_task() ended up in pick_next_task, correct this. > > > > > > > > This prepares the way for a pick_next_task() variant that does not > > > > affect the current state; allowing remote picking. > > > > > > > > Signed-off-by: Peter Zijlstra (Intel) <[email protected]> > > > > --- > > > > kernel/sched/deadline.c | 23 ++++++++++++----------- > > > > kernel/sched/rt.c | 27 ++++++++++++++------------- > > > > 2 files changed, 26 insertions(+), 24 deletions(-) > > > > > > > > --- a/kernel/sched/deadline.c > > > > +++ b/kernel/sched/deadline.c > > > > @@ -1694,12 +1694,21 @@ static void start_hrtick_dl(struct rq *r > > > > } > > > > #endif > > > > > > > > -static inline void set_next_task(struct rq *rq, struct task_struct *p) > > > > +static void set_next_task_dl(struct rq *rq, struct task_struct *p) > > > > { > > > > p->se.exec_start = rq_clock_task(rq); > > > > > > > > /* You can't push away the running task */ > > > > dequeue_pushable_dl_task(rq, p); > > > > + > > > > + if (hrtick_enabled(rq)) > > > > + start_hrtick_dl(rq, p); > > > > + > > > > + if (rq->curr->sched_class != &dl_sched_class) > > > > + update_dl_rq_load_avg(rq_clock_pelt(rq), rq, 0); > > > > + > > > > + if (rq->curr != p) > > > > + deadline_queue_push_tasks(rq); > > > > > > It's a minor thing, but I was wondering why you added the check on curr. > > > deadline_queue_push_tasks() already checks if are there pushable tasks, > > > plus curr can still be of a different class at this point? > > > > Hmm, so by moving that code into set_next_task() it is exposed to the: > > > > if (queued) > > deuque_task(); > > if (running) > > put_prev_task(); > > > > /* do stuff */ > > > > if (queued) > > enqueue_task(); > > if (running) > > set_next_task(); > > > > patter from core.c; and in that case nothing changes. That said; I > > might've gotten it wrong. > > Right. But, I was wondering about the __schedule()->pick_next_task() > case, where, say, prev (rq->curr) is RT/CFS and next (p) is DEADLINE.
So we do pick_next_task() first and then set rq->curr (obviously). So the first set_next_task() will see rq->curr != p and we'll do the push balance stuff. Then the above pattern will always see rq->curr == p and we'll not trigger push balancing. Now, looking at it, this also doesn't do push balancing when we re-select the same task, even though we really should be doing it. So I suppose not adding the condition, and always doing the push balance, while wasteful, is not wrong. Hmm?
