Message-ID: <1375941095.18565.19.camel@marge.simpson.net>
Subject: Re: [PATCH V3]hrtimer: Fix a performance regression by disable
 reprogramming in remove_hrtimer
From: Mike Galbraith <bitbucket@online.de>
To: "ethan.zhao" <ethan.kernel@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>, Ingo Molnar <mingo@kernel.org>,
        LKML <linux-kernel@vger.kernel.org>, johlstei@codeaurora.org,
        Yinghai Lu <yinghai@kernel.org>, Jin Feng <joe.jin@oracle.com>,
        Youquan Song <youquan.song@intel.com>, LenBrown <len.brown@intel.com>
Date: Thu, 08 Aug 2013 07:51:35 +0200
In-Reply-To: <1375939750.18565.11.camel@marge.simpson.net>
References: <1374955447-5051-1-git-send-email-ethan.kernel@gmail.com>
	 <alpine.DEB.2.02.1307291146060.4089@ionos.tec.linutronix.de>
	 <CABawtvP4oLuvHOS3prbbgPShXVziV_wTo7i6KCqJ9KkoVdz0ag@mail.gmail.com>
	 <20130730093519.GP3008@twins.programming.kicks-ass.net>
	 <1375774140.5412.9.camel@marge.simpson.net>
	 <1375775172.5412.14.camel@marge.simpson.net>
	 <BD6ACAD8-25D7-406B-A843-22696F9D8DCC@gmail.com>
	 <1375939750.18565.11.camel@marge.simpson.net>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Mime-Version: 1.0
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 1415
Lines: 36

On Thu, 2013-08-08 at 07:29 +0200, Mike Galbraith wrote:

> On Thu, 2013-08-08 at 12:31 +0800, ethan.zhao wrote:
> 
> > diff --git a/arch/x86/include/asm/smp.h b/arch/x86/include/asm/smp.h
> > index 4137890..c27f04f 100644
> > --- a/arch/x86/include/asm/smp.h
> > +++ b/arch/x86/include/asm/smp.h
> > @@ -137,7 +137,7 @@ static inline void play_dead(void)
> > 
> > static inline void smp_send_reschedule(int cpu)
> > {
> > -       smp_ops.smp_send_reschedule(cpu);
> > +       /* smp_ops.smp_send_reschedule(cpu); */
> > }
> 
> Hm.  You're much braver than I am.

BTW, according to my testing, Peter's patch should make your box a lot
happier.  It won't make reschedule_interrupt crawl back under its rock,
but should improve your pipe-test numbers quite a bit.

Also note that pipe-test is really kinda silly cross core, it's really
good for measuring fastpath weight gain when pinned.  When scheduled
cross core (we do that to cut latency for stuff that does real work), it
will always suck rocks, as it's 100% synchronous, there's nada to gain
by occupying two cores.  (fortunately, real programs tend to do more
than play ping-pong with a microscopic ball;)

-Mike

--
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/