Messages in this thread | | | Subject | Re: [patch 13/15] sched: expire slack quota using generation counters | From | Peter Zijlstra <> | Date | Tue, 05 Apr 2011 15:28:09 +0200 |
| |
On Tue, 2011-03-22 at 20:03 -0700, Paul Turner wrote:
Argh, this patch is terrible for the reason that it changes the whole accounting just introduced and me having to re-open all the previous patches to look up hth stuff worked before.
> @@ -436,8 +438,10 @@ void init_cfs_bandwidth(struct cfs_bandw > raw_spin_lock_init(&cfs_b->lock); > cfs_b->quota = cfs_b->runtime = quota; > cfs_b->period = ns_to_ktime(period); > + cfs_b->quota_generation = 0; > INIT_LIST_HEAD(&cfs_b->throttled_cfs_rq); > > + > hrtimer_init(&cfs_b->period_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); > cfs_b->period_timer.function = sched_cfs_period_timer;
We're in desperate need of more whitespace there? :-)
> @@ -9333,6 +9337,8 @@ static int tg_set_cfs_bandwidth(struct t > raw_spin_lock_irq(&cfs_b->lock); > cfs_b->period = ns_to_ktime(period); > cfs_b->runtime = cfs_b->quota = quota; > + > + cfs_bump_quota_generation(cfs_b); > raw_spin_unlock_irq(&cfs_b->lock); > > for_each_possible_cpu(i) { > Index: tip/kernel/sched_fair.c > =================================================================== > --- tip.orig/kernel/sched_fair.c > +++ tip/kernel/sched_fair.c > @@ -1331,11 +1331,25 @@ static void check_cfs_rq_quota(struct cf > resched_task(rq_of(cfs_rq)->curr); > } > > +static void cfs_bump_quota_generation(struct cfs_bandwidth *cfs_b) > +{ > + cfs_b->quota_generation++; > + smp_mb(); > +}
Memory barriers come in pairs and with a comment, you fail on both counts.
> + > +static inline int cfs_rq_quota_current(struct cfs_rq *cfs_rq) > +{ > + struct cfs_bandwidth *cfs_b = tg_cfs_bandwidth(cfs_rq->tg); > + > + return cfs_rq->quota_generation == cfs_b->quota_generation; > +} > + > 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; > + int generation;
Not initialized,
> min_amount = sched_cfs_bandwidth_slice() + (-cfs_rq->quota_remaining); > > @@ -1347,10 +1361,18 @@ static void request_cfs_rq_quota(struct > } else { > amount = min_amount; > } > + generation = cfs_b->quota_generation; > raw_spin_unlock(&cfs_b->lock); > }
and since there's an if there, one can fail it, leaving generation uninitialized,
> > + /* a deficit should be carried forwards, surplus should be dropped */ > + > + if (generation != cfs_rq->quota_generation && > + cfs_rq->quota_remaining > 0) > + cfs_rq->quota_remaining = 0; > + > cfs_rq->quota_remaining += amount; > + cfs_rq->quota_generation = generation; > }
Resulting in uninitialized usage right there.
| |