From mboxrd@z Thu Jan  1 00:00:00 1970
From: Eric Dumazet <eric.dumazet@gmail.com>
Subject: Re: [PATCH] net: tcp: Fix a PTO timing granularity issue
Date: Wed, 27 May 2015 06:41:17 -0700
Message-ID: <1432734077.4060.382.camel@edumazet-glaptop2.roam.corp.google.com>
References: <1432663992.4060.286.camel@edumazet-glaptop2.roam.corp.google.com>
	 <1432671437-19140-1-git-send-email-ido@wizery.com>
	 <063D6719AE5E284EB5DD2968C1650D6D1CB40A1A@AcuExch.aculab.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Cc: 'Ido Yariv' <ido@wizery.com>,
	"David S. Miller" <davem@davemloft.net>,
	Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>,
	James Morris <jmorris@namei.org>,
	Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
	Patrick McHardy <kaber@trash.net>,
	Nandita Dukkipati <nanditad@google.com>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	Ido Yariv <idox.yariv@intel.com>
To: David Laight <David.Laight@ACULAB.COM>
Return-path: <netdev-owner@vger.kernel.org>
Received: from mail-ig0-f169.google.com ([209.85.213.169]:37484 "EHLO
	mail-ig0-f169.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1751307AbbE0NlT (ORCPT
	<rfc822;netdev@vger.kernel.org>); Wed, 27 May 2015 09:41:19 -0400
In-Reply-To: <063D6719AE5E284EB5DD2968C1650D6D1CB40A1A@AcuExch.aculab.com>
Sender: netdev-owner@vger.kernel.org
List-ID: <netdev.vger.kernel.org>

On Wed, 2015-05-27 at 11:36 +0000, David Laight wrote:
> From: Of Ido Yariv
> > Sent: 26 May 2015 21:17
> > The Tail Loss Probe RFC specifies that the PTO value should be set to
> > max(2 * SRTT, 10ms), where SRTT is the smoothed round-trip time.
> > 
> > The PTO value is converted to jiffies, so the timer may expire
> > prematurely.
> > 
> > This is especially problematic on systems in which HZ <= 100, so work
> > around this by setting the timeout to at least 2 jiffies on such
> > systems.
> > 
> > The 10ms figure was originally selected based on tests performed with
> > the current implementation and HZ = 1000. Thus, leave the behavior on
> > systems with HZ > 100 unchanged.
> > 
> > Signed-off-by: Ido Yariv <idox.yariv@intel.com>
> > ---
> >  net/ipv4/tcp_output.c | 3 +++
> >  1 file changed, 3 insertions(+)
> > 
> > diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> > index 534e5fd..5321df8 100644
> > --- a/net/ipv4/tcp_output.c
> > +++ b/net/ipv4/tcp_output.c
> > @@ -2208,6 +2208,9 @@ bool tcp_schedule_loss_probe(struct sock *sk)
> >  		timeout = max_t(u32, timeout,
> >  				(rtt + (rtt >> 1) + TCP_DELACK_MAX));
> >  	timeout = max_t(u32, timeout, msecs_to_jiffies(10));
> > +#if HZ <= 100
> > +	timeout = max_t(u32, timeout, 2);
> > +#endif
> 
> Why not:
> 	timeout = max_t(u32, timeout, max_t(u32, 2, msecs_to_jiffies(10)));
> I think the RH max_t() is a compile time constant.
> 
> You need 2 jiffies to guarantee a non-zero timeout.
> Even if HZ=199 with a 'rounding down' msecs_to_jiffies() you get 1 jiffy
> and a possible immediate timeout.
> 

Have you followed previous discussions ?

I guess we can have a helper macro, but for the moment only one spot was
found.

Its kind of depressing having to deal with HZ=100 issues, with modern
NO_HZ configurations.

TCP rtts have now usec resolution, so HZ=100 is pushing TCP to very
imprecise behavior.