Messages in this thread | | | From | Valentin Schneider <> | Subject | Re: [PATCH] sched/fair: Fix nohz.next_balance update | Date | Mon, 04 May 2020 16:48:06 +0100 |
| |
On 04/05/20 16:17, Vincent Guittot wrote: > On Sun, 3 May 2020 at 10:34, Peng Liu <iwtbavbm@gmail.com> wrote: >> >> commit c5afb6a87f23 ("sched/fair: Fix nohz.next_balance update") >> During idle load balance, this_cpu(ilb) do load balance for the other >> idle CPUs, also gather the earliest (nohz.)next_balance. >> >> Since commit: >> 'b7031a02ec75 ("sched/fair: Add NOHZ_STATS_KICK")' >> >> We update nohz.next_balance like this: >> >> _nohz_idle_balance() { >> for_each_cpu(nohz.idle_cpus_mask) { >> rebalance_domains() { >> update nohz.next_balance <-- compare and update >> } >> } >> rebalance_domains(this_cpu) { >> update nohz.next_balance <-- compare and update >> } >> update nohz.next_balance <-- unconditionally update >> } >> >> For instance, nohz.idle_cpus_mask spans {cpu2,3,5,8}, and this_cpu is >> cpu5. After the above loop we could gather the earliest *next_balance* >> among {cpu2,3,8}, then rebalance_domains(this_cpu) update >> nohz.next_balance with this_rq->next_balance, but finally overwrite >> nohz.next_balance with the earliest *next_balance* among {cpu2,3,8}, >> we may end up with not getting the earliest next_balance. >> >> Since we can gather all the updated rq->next_balance, including this_cpu, >> in _nohz_idle_balance(), it's safe to remove the extra lines in >> rebalance_domains() which are originally intended for this_cpu. And >> finally the updating only happen in _nohz_idle_balance(). > > I'm not sure that's always true. Nothing prevents nohz_idle_balance() > to return false . Then run_rebalance_domains() calls > rebalance_domains(this_rq ,SCHED_IDLE) outside _nohz_idle_balance(). > In this case we must keep the code in rebalance_domains(). > > For example when the tick is not stopped when entering idle. Or when > need_resched() returns true. >
I had missed that, good points.
> So instead of removing the code from rebalance_domains, you should > move the one in _nohz_idle_balance() to make sure that the "if > (likely(update_next_balance)) ..." is called before calling > rebalance_domains for the local cpu >
Why not just get rid of the update in _nohz_idle_balance() entirely then? The nohz.next_balance update in rebalance_domains() will always happen if it is required (and we have idle == CPU_IDLE), so the extra update in _nohz_idle_balance() doesn't seem to be any useful.
| |