[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20090317114557.GA32233@linux.vnet.ibm.com>
Date: Tue, 17 Mar 2009 17:15:58 +0530
From: Arun R Bharadwaj <arun@...ux.vnet.ibm.com>
To: Thomas Gleixner <tglx@...utronix.de>
Cc: linux-kernel@...r.kernel.org, linux-pm@...ts.linux-foundation.org,
a.p.zijlstra@...llo.nl, ego@...ibm.com, mingo@...e.hu,
andi@...stfloor.org, venkatesh.pallipadi@...el.com,
vatsa@...ux.vnet.ibm.com, arjan@...radead.org,
svaidy@...ux.vnet.ibm.com
Subject: Re: [v3 PATCH 4/4] timers: logic to move non pinned timers
* Thomas Gleixner <tglx@...utronix.de> [2009-03-17 11:22:24]:
> On Mon, 16 Mar 2009, Arun R Bharadwaj wrote:
> > @@ -627,6 +628,16 @@ __mod_timer(struct timer_list *timer, un
> >
> > new_base = __get_cpu_var(tvec_bases);
> >
> > + current_cpu = smp_processor_id();
> > + preferred_cpu = get_nohz_load_balancer();
> > + if (get_sysctl_timer_migration() && idle_cpu(current_cpu) &&
> > + !pinned && preferred_cpu != -1) {
> > + new_base = per_cpu(tvec_bases, preferred_cpu);
> > + timer_set_base(timer, new_base);
> > + timer->expires = expires;
> > + internal_add_timer(new_base, timer);
> > + goto out_unlock;
> > + }
>
> Err. This change breaks the timer->base logic. Why can't it just
> select the base and use the existing code ?
>
Sure, I'll take care of this.
> > @@ -198,8 +200,16 @@ switch_hrtimer_base(struct hrtimer *time
> > {
> > struct hrtimer_clock_base *new_base;
> > struct hrtimer_cpu_base *new_cpu_base;
> > + int current_cpu, preferred_cpu;
> > +
> > + current_cpu = smp_processor_id();
> > + preferred_cpu = get_nohz_load_balancer();
> > + if (get_sysctl_timer_migration() && !pinned && preferred_cpu != -1
> > + && idle_cpu(current_cpu))
> > + new_cpu_base = &per_cpu(hrtimer_bases, preferred_cpu);
> > + else
> > + new_cpu_base = &__get_cpu_var(hrtimer_bases);
> >
> > - new_cpu_base = &__get_cpu_var(hrtimer_bases);
> > new_base = &new_cpu_base->clock_base[base->index];
>
> Hmm. This can lead to high latencies when you enqueue the timer on
> the other CPU simply because we can not reprogram the timer hardware
> on the other CPU in the CONFIG_HIGH_RES=y case.
>
> Let's assume we are on CPU0 and try to enqueue the timer on CPU1,
> where the next timer expiry is 5ms away. The timer which we enqueue
> is due in 500us. So you introduce 4.5ms latency.
>
We are moving timers to the ilb which wakes up every jiffy.
So we can move the timer to the ilb only if it's expiry
time is greater than 1 jiffy. Else we can fire on the same CPU.
This would prevent any latency from being added, right?
--arun
> Thanks,
>
> tglx
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists