lkml.org 
[lkml]   [2010]   [Oct]   [21]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    Date
    SubjectRe: [RFC tg_shares_up improvements - v1 05/12] sched: fix update_cfs_load synchronization
    On Thu, Oct 21, 2010 at 2:52 AM, Bharata B Rao
    <bharata@linux.vnet.ibm.com> wrote:
    > On Fri, Oct 15, 2010 at 09:43:54PM -0700, pjt@google.com wrote:
    >> Using cfs_rq->nr_running is not sufficient to synchronize update_cfs_load with
    >> the put path since nr_running accounting occurs at deactivation.
    >>
    >> It's also not safe to make the removal decision based on load_avg as this fails
    >> with both high periods and low shares.  Resolve this by clipping history at 8
    >> foldings.
    >>
    >
    > Did you mean 4 foldings (and not 8) above since I see you are truncating
    > the load history at 4 idle periods ?

    We end up folding every period/2 (after the first fold) since after
    that we start at period/2 and not 0.

    Thus, if we were issuing 'continuous' load updates triggering this
    condition would mean that we had folded idle periods into our load_avg
    8 times.

    Granted, it's possible for there to be less folds given sporadic load
    updates so I'll just update the comment to reference periods.

    Thanks

    >
    >
    >> @@ -685,9 +686,19 @@ static void update_cfs_load(struct cfs_r
    >>       now = rq_of(cfs_rq)->clock;
    >>       delta = now - cfs_rq->load_stamp;
    >>
    >> +     /* truncate load history at 4 idle periods */
    >> +     if (cfs_rq->load_stamp > cfs_rq->load_last &&
    >> +         now - cfs_rq->load_last > 4 * period) {
    >> +             cfs_rq->load_period = 0;
    >> +             cfs_rq->load_avg = 0;
    >> +     }
    >> +
    >>       cfs_rq->load_stamp = now;
    >>       cfs_rq->load_period += delta;
    >> -     cfs_rq->load_avg += delta * cfs_rq->load.weight;
    >> +     if (load) {
    >> +             cfs_rq->load_last = now;
    >> +             cfs_rq->load_avg += delta * load;
    >> +     }
    >>
    >>       while (cfs_rq->load_period > period) {
    >>               /*
    >> @@ -700,10 +711,8 @@ static void update_cfs_load(struct cfs_r
    >>               cfs_rq->load_avg /= 2;
    >>       }
    >>
    >> -     if (lb && !cfs_rq->nr_running) {
    >> -             if (cfs_rq->load_avg < (period / 8))
    >> -                     list_del_leaf_cfs_rq(cfs_rq);
    >> -     }
    >> +     if (!cfs_rq->curr && !cfs_rq->nr_running && !cfs_rq->load_avg)
    >> +             list_del_leaf_cfs_rq(cfs_rq);
    >>  }
    >>
    >
    --
    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to majordomo@vger.kernel.org
    More majordomo info at http://vger.kernel.org/majordomo-info.html
    Please read the FAQ at http://www.tux.org/lkml/

    \
     
     \ /
      Last update: 2010-10-21 20:27    [W:0.026 / U:30.392 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site