Re: [Resend PATCH v3] psi: fix possible trigger missing in the window

[Suren Baghdasaryan]

On Mon, Jan 24, 2022 at 7:53 AM Johannes Weiner <hannes@xxxxxxxxxxx> wrote:
>
> On Thu, Jan 20, 2022 at 11:31:56AM +0800, Huangzhaoyang wrote:
> > From: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
> >
> > When a new threshold breaching stall happens after a psi event was
> > generated and within the window duration, the new event is not
> > generated because the events are rate-limited to one per window. If
> > after that no new stall is recorded then the event will not be
> > generated even after rate-limiting duration has passed. This is
> > happening because with no new stall, window_update will not be called
> > even though threshold was previously breached. To fix this, record
> > threshold breaching occurrence and generate the event once window
> > duration is passed.
> >
> > Suggested-by: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> > Signed-off-by: Zhaoyang Huang <zhaoyang.huang@xxxxxxxxxx>
>
> Good catch. The change makes sense to me.
>
> However, I had to re-read the discussion to understand *why*
> triggering once per window can be a practical problem. Could you
> please include the lkmd scenario you mentioned?
>
> Suren, even though it's your suggested code, can you please also add
> ack/review tags? Thanks!

Will do as soon as the new version is posted and your comments are
addressed. Thanks!

>
> Some minor inline comments:
>
> > diff --git a/include/linux/psi_types.h b/include/linux/psi_types.h
> > index 0a23300..87b694a 100644
> > --- a/include/linux/psi_types.h
> > +++ b/include/linux/psi_types.h
> > @@ -132,6 +132,8 @@ struct psi_trigger {
> >
> >       /* Refcounting to prevent premature destruction */
> >       struct kref refcount;
> > +
> > +     bool threshold_breach;
>
> Something like bool pending_event would be more descriptive, IMO.
>
> Also please remember to add a short comment like we have for the other
> struct members. For example:
>
>         /* Deferred event(s) from previous ratelimit window */
>
> > @@ -524,24 +524,29 @@ static u64 update_triggers(struct psi_group *group, u64 now)
> >        */
> >       list_for_each_entry(t, &group->triggers, node) {
> >               u64 growth;
> > +             bool trigger_stalled =
> > +                     group->polling_total[t->state] != total[t->state];
>
> Triggers don't stall, they trigger on stalls. How about this:
>
>                 bool new_stall;
>                 u64 growth;
>
>                 new_stall = group->polling_total[t->state] != total[t->state];
>
> (order local declarations by length, avoid line wraps where possible)
>
> > -             /* Check for stall activity */
> > -             if (group->polling_total[t->state] == total[t->state])
> > -                     continue;
> > -
> > -             /*
> > -              * Multiple triggers might be looking at the same state,
> > -              * remember to update group->polling_total[] once we've
> > -              * been through all of them. Also remember to extend the
> > -              * polling time if we see new stall activity.
> > -              */
> > -             new_stall = true;
> > -
> > -             /* Calculate growth since last update */
> > -             growth = window_update(&t->win, now, total[t->state]);
> > -             if (growth < t->threshold)
> > +             /* Check for stall activity or a previous threshold breach */
> > +             if (!trigger_stalled && !t->threshold_breach)
> >                       continue;
>
> This could use a bit more explanation imo:
>
>                 /*
>                  * Check for new stall activity, as well as deferred
>                  * events that occurred in the last window after the
>                  * trigger had already fired (we want to ratelimit
>                  * events without dropping any).
>                  */
>                 if (!new_stall && !t->pending_event)
>                         continue;
>
> > +             if (trigger_stalled) {
> > +                     /*
> > +                      * Multiple triggers might be looking at the same state,
> > +                      * remember to update group->polling_total[] once we've
> > +                      * been through all of them. Also remember to extend the
> > +                      * polling time if we see new stall activity.
> > +                      */
> > +                     new_stall = true;
>
> and then rename this flag `update_total'.
>
> > +                     /* Calculate growth since last update */
> > +                     growth = window_update(&t->win, now, total[t->state]);
> > +                     if (growth < t->threshold)
> > +                             continue;
> > +
> > +                     t->threshold_breach = true;
> > +             }
> >               /* Limit event signaling to once per window */
> >               if (now < t->last_event_time + t->win.size)
> >                       continue;
> > @@ -550,6 +555,8 @@ static u64 update_triggers(struct psi_group *group, u64 now)
> >               if (cmpxchg(&t->event, 0, 1) == 0)
> >                       wake_up_interruptible(&t->event_wait);
> >               t->last_event_time = now;
> > +             /* Reset threshold breach flag once event got generated */
> > +             t->threshold_breach = false;
> >       }
> >
> >       if (new_stall)
> > @@ -1152,6 +1159,7 @@ struct psi_trigger *psi_trigger_create(struct psi_group *group,
> >       t->last_event_time = 0;
> >       init_waitqueue_head(&t->event_wait);
> >       kref_init(&t->refcount);
> > +     t->threshold_breach = false;
> >
> >       mutex_lock(&group->trigger_lock);
>
> Thanks!