Re: [patch 03/15] sched: accumulate per-cfs_rq cpu usage

[Paul Turner]

On Tue, Apr 5, 2011 at 6:28 AM, Peter Zijlstra <a.p.zijlstra@xxxxxxxxx> wrote:
> On Tue, 2011-03-22 at 20:03 -0700, Paul Turner wrote:
>> +static void request_cfs_rq_quota(struct cfs_rq *cfs_rq)
>> +{
>> +       struct task_group *tg = cfs_rq->tg;
>> +       struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(tg);
>> +       u64 amount = 0, min_amount;
>> +
>> +       min_amount = sched_cfs_bandwidth_slice() + (-cfs_rq->quota_remaining);
>> +
>> +       if (cfs_b->runtime > 0 || cfs_b->quota == RUNTIME_INF) {
>> +               raw_spin_lock(&cfs_b->lock);
>> +               if (cfs_b->quota != RUNTIME_INF) {
>> +                       amount = min(cfs_b->runtime, min_amount);
>> +                       cfs_b->runtime -= amount;
>> +               } else {
>> +                       amount = min_amount;
>> +               }
>
> So why would quota be RUNTIME_INF and quota_enabled be true? If its due
> to a race when fiddling with the cgroup filesystem setting things up
> that else branch wants a comment, if its for another reason all together
> there's also a comment missing somewhere ;-)
>

It's the first -- it's possible for global quota to transition between
infinite/finite states, but since we own the lock we may not yet have
seen our cfs_rq->quota_enabled update yet.  In this case we don't want
to perturb the global state /out/ of RUNTIME_INF (by subtracting from
it).

While making this more readable the speculative check can also be
dropped since that only ever benefits us a single time when we're out
of quota (as the next operation in that case is to throttle).

Cleaning..


>> +               raw_spin_unlock(&cfs_b->lock);
>> +       }
>> +
>> +       cfs_rq->quota_remaining += amount;
>> +}
>> +
>> +static void account_cfs_rq_quota(struct cfs_rq *cfs_rq,
>> +               unsigned long delta_exec)
>> +{
>> +       if (!cfs_rq->quota_enabled)
>> +               return;
>> +
>> +       cfs_rq->quota_remaining -= delta_exec;
>> +
>> +       if (cfs_rq->quota_remaining > 0)
>> +               return;
>> +
>> +       request_cfs_rq_quota(cfs_rq);
>> +}
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/