lkml.org 
[lkml]   [2022]   [Oct]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Date
SubjectRe: [External] [tip: sched/core] sched/psi: Fix avgs_work re-arm in psi_avgs_work()
On Thu, Oct 27, 2022 at 11:50 PM Chengming Zhou
<zhouchengming@bytedance.com> wrote:
>
> Hello,
>
> Thanks for picking this up. There is a newer version which has been acked:
> https://lore.kernel.org/all/20221014110551.22695-1-zhouchengming@bytedance.com/

Hmm. Indeed this seems to be an older version and not the one I asked
Peter to pick up in
https://lore.kernel.org/all/CAJuCfpHeJuZBbv-q+WXjgNHwt_caMomFPL3L9rxosXOrZz3fBw@mail.gmail.com/.
Not sure what went wrong. Peter, could you please replace this one
with https://lore.kernel.org/all/20221014110551.22695-1-zhouchengming@bytedance.com/?

Chengming, please do not top-post next time. Would be better if you
posted your note under the "Link:" field in this email.
Thanks!

>
> As well another PSI patch that has been acked by Johannes:
> https://lore.kernel.org/all/20220926081931.45420-1-zhouchengming@bytedance.com/
>
> Thanks!
>
>
> On 2022/10/28 14:42, tip-bot2 for Chengming Zhou wrote:
> > The following commit has been merged into the sched/core branch of tip:
> >
> > Commit-ID: 7d89d7bb921c5ae5a428df282e64ee5692e26fe0
> > Gitweb: https://git.kernel.org/tip/7d89d7bb921c5ae5a428df282e64ee5692e26fe0
> > Author: Chengming Zhou <zhouchengming@bytedance.com>
> > AuthorDate: Mon, 10 Oct 2022 18:42:06 +08:00
> > Committer: Peter Zijlstra <peterz@infradead.org>
> > CommitterDate: Thu, 27 Oct 2022 11:01:23 +02:00
> >
> > sched/psi: Fix avgs_work re-arm in psi_avgs_work()
> >
> > Pavan reported a problem that PSI avgs_work idle shutoff is not
> > working at all. Because PSI_NONIDLE condition would be observed in
> > psi_avgs_work()->collect_percpu_times()->get_recent_times() even if
> > only the kworker running avgs_work on the CPU.
> >
> > Although commit 1b69ac6b40eb ("psi: fix aggregation idle shut-off")
> > avoided the ping-pong wake problem when the worker sleep, psi_avgs_work()
> > still will always re-arm the avgs_work, so shutoff is not working.
> >
> > This patch changes to consider current CPU groupc as IDLE if the
> > kworker running avgs_work is the only task running and no IOWAIT
> > or MEMSTALL sleep tasks, in which case we will shut off the avgs_work
> > if other CPUs' groupc are also IDLE.
> >
> > One potential problem is that the brief period of non-idle time
> > incurred between the aggregation run and the kworker's dequeue will
> > be stranded in the per-cpu buckets until avgs_work run next time.
> > The buckets can hold 4s worth of time, and future activity will wake
> > the avgs_work with a 2s delay, giving us 2s worth of data we can leave
> > behind when shut off the avgs_work. If the kworker run other works after
> > avgs_work shut off and doesn't have any scheduler activities for 2s,
> > this maybe a problem.
> >
> > Reported-by: Pavan Kondeti <quic_pkondeti@quicinc.com>
> > Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > Link: https://lore.kernel.org/r/20221010104206.12184-1-zhouchengming@bytedance.com
> > ---
> > kernel/sched/psi.c | 15 +++++++++++++++
> > 1 file changed, 15 insertions(+)
> >
> > diff --git a/kernel/sched/psi.c b/kernel/sched/psi.c
> > index ee2ecc0..f4cdf6f 100644
> > --- a/kernel/sched/psi.c
> > +++ b/kernel/sched/psi.c
> > @@ -242,6 +242,8 @@ static void get_recent_times(struct psi_group *group, int cpu,
> > u32 *pchanged_states)
> > {
> > struct psi_group_cpu *groupc = per_cpu_ptr(group->pcpu, cpu);
> > + int current_cpu = raw_smp_processor_id();
> > + bool only_avgs_work = false;
> > u64 now, state_start;
> > enum psi_states s;
> > unsigned int seq;
> > @@ -256,6 +258,15 @@ static void get_recent_times(struct psi_group *group, int cpu,
> > memcpy(times, groupc->times, sizeof(groupc->times));
> > state_mask = groupc->state_mask;
> > state_start = groupc->state_start;
> > + /*
> > + * This CPU has only avgs_work kworker running, snapshot the
> > + * newest times then don't need to re-arm for this groupc.
> > + * Normally this kworker will sleep soon and won't wake
> > + * avgs_work back up in psi_group_change().
> > + */
> > + if (current_cpu == cpu && groupc->tasks[NR_RUNNING] == 1 &&
> > + !groupc->tasks[NR_IOWAIT] && !groupc->tasks[NR_MEMSTALL])
> > + only_avgs_work = true;
> > } while (read_seqcount_retry(&groupc->seq, seq));
> >
> > /* Calculate state time deltas against the previous snapshot */
> > @@ -280,6 +291,10 @@ static void get_recent_times(struct psi_group *group, int cpu,
> > if (delta)
> > *pchanged_states |= (1 << s);
> > }
> > +
> > + /* Clear PSI_NONIDLE so avgs_work won't be re-armed for this groupc */
> > + if (only_avgs_work)
> > + *pchanged_states &= ~(1 << PSI_NONIDLE);
> > }
> >
> > static void calc_avgs(unsigned long avg[3], int missed_periods,

\
 
 \ /
  Last update: 2022-10-28 18:02    [W:0.115 / U:25.100 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site