On Wed, 2020-04-29 at 00:09 +0200, Peter Zijlstra wrote: > On Tue, Apr 28, 2020 at 10:37:18PM +0100, Valentin Schneider wrote: > > On 28/04/20 06:02, Scott Wood wrote: > > > Thus, newidle_balance() is entered with interrupts enabled, which > > > allows > > > (in the next patch) enabling interrupts when the lock is dropped. > > > > > > Signed-off-by: Scott Wood <swood@xxxxxxxxxx> > > > --- > > > kernel/sched/core.c | 7 ++++--- > > > kernel/sched/fair.c | 45 ++++++++++++++++--------------------------- > > > - > > > kernel/sched/sched.h | 6 ++---- > > > 3 files changed, 22 insertions(+), 36 deletions(-) > > > > > > diff --git a/kernel/sched/core.c b/kernel/sched/core.c > > > index 9a2fbf98fd6f..0294beb8d16c 100644 > > > --- a/kernel/sched/core.c > > > +++ b/kernel/sched/core.c > > > @@ -3241,6 +3241,10 @@ static struct rq *finish_task_switch(struct > > > task_struct *prev) > > > } > > > > > > tick_nohz_task_switch(); > > > + > > > + if (is_idle_task(current)) > > > + newidle_balance(); > > > + > > > > This means we must go through a switch_to(idle) before figuring out we > > could've switched to a CFS task, and do it then. I'm curious to see the > > performance impact of that. > > Also, if you move it this late, this is entirely the wrong place. If you > do it after the context switch either use the balance_callback or put it > in the idle path. > > But what Valentin said; this needs a fair bit of support, the whole > reason we've never done this is to avoid that double context switch... > balance_callback() enters with the rq lock held but BH not separately disabled, which interferes with the ability to enable interrupts but not BH. It also gets called from rt_mutex_setprio() and __sched_setscheduler(), and I didn't want the caller of those to be stuck with the latency. -Scott
