From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755383AbYAXNLi (ORCPT ); Thu, 24 Jan 2008 08:11:38 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1753282AbYAXNL1 (ORCPT ); Thu, 24 Jan 2008 08:11:27 -0500 Received: from E23SMTP01.au.ibm.com ([202.81.18.162]:35353 "EHLO e23smtp01.au.ibm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752922AbYAXNL0 (ORCPT ); Thu, 24 Jan 2008 08:11:26 -0500 Date: Thu, 24 Jan 2008 18:41:26 +0530 From: Kamalesh Babulal To: Ilpo =?iso-8859-1?Q?J=E4rvinen?= Cc: Dave Young , Kamalesh Babulal , Krishna Kumar2 , Denys Fedoryshchenko , David Miller , LKML , Netdev , Andrew Morton Subject: Re: 2.6.24-rc8-mm1 : net tcp_input.c warnings Message-ID: <20080124131126.GA698@linux.vnet.ibm.com> Reply-To: Kamalesh Babulal References:

<20080122.190137.206066966.davem@davemloft.net>

MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline Content-Transfer-Encoding: 8bit In-Reply-To: User-Agent: Mutt/1.5.13 (2006-08-11) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, Jan 24, 2008 at 11:54:18AM +0200, Ilpo Järvinen wrote: > On Thu, 24 Jan 2008, Dave Young wrote: > > Hi Dave (& others), > > > Thanks. > > Thanks a lot, I was first to ignore all these because they occurred > with newreno, but looked again... :-/ > > > New warning trigged with your debug patch: > > This was probably with the earlier one I sent to you because there's still > this case remaining which itself is valid: > > > P: 5 L: 0 vs 0 S: 0 vs 1 w: 2044790889-2044796616 (0) > > ...snip... this is still ok state (S+L <= P): > > > P: 5 L: 0 vs 0 S: 0 vs 3 w: 2044790889-2044796616 (0) > > TCP wq(s) < > > TCP wq(h) +++h+< > > l0 s3 f0 p5 seq: su2044790889 hs2044795029 sn2044796616 > > ------------[ cut here ]------------ > > WARNING: at net/ipv4/tcp_input.c:2169 tcp_mark_head_lost+0x122/0x150() > > Modules linked in: snd_seq_dummy snd_seq_oss snd_seq_midi_event > > snd_seq snd_seq_device snd_pcm_oss snd_mixer_oss eeprom e100 psmouse > > snd_hda_intel snd_pcm snd_timer btusb bluetooth serio_raw snd 3c59x sg > > evdev thermal soundcore rtc_cmos snd_page_alloc rtc_core rtc_lib > > i2c_i801 processor button intel_agp dcdbas pcspkr agpgart > > Pid: 0, comm: swapper Not tainted 2.6.24-rc8-mm1 #8 > > [] ? have_callable_console+0x20/0x30 > > [] warn_on_slowpath+0x54/0x80 > > [] ? tcp_print_queue+0x1a4/0x230 > > [] ? vprintk+0x308/0x320 . . . . > > ---[ end trace 14b601818e6903ac ]--- > > ...But this no longer is, and even more, L: 5 is not valid state at this > point all (should only happen if we went to RTO but it would reset S to > zero with newreno): > > > P: 5 L: 5 vs 5 S: 0 vs 3 w: 2044790889-2044796616 (0) > > TCP wq(s) LLLLl< > > TCP wq(h) +++h+< > > l5 s3 f0 p5 seq: su2044790889 hs2044795029 sn2044796616 > > Surprisingly, it was the first time the WARN_ON for left_out returned > correct location. This also explains why the patch I sent to Krishna > didn't print anything (it didn't end up into printing because I forgot > to add L+S>P check into to the state checking if). > > ...so please, could you (others than Denys) try this patch, it should > solve the issue. And Denys, could you confirm (and if necessary double > check) that the kernel you saw this similar problem with is the pure > Linus' mainline, i.e., without any net-2.6.25 or mm bits please, if so, > that problem persists. And anyway, there were some fackets_out related > problems reported as well and this doesn't help for that but I think I've > lost track of who was seeing it due to large number of reports :-), could > somebody refresh my memory because I currently don't have time to dig it > up from archives (at least on this week). > > > -- > i. > > -- > [PATCH] [TCP]: NewReno must count every skb while marking losses > > NewReno should add cnt per skb (as with FACK) instead of depending > on SACKED_ACKED bits which won't be set with it at all. > Effectively, NewReno should always exists after the first > iteration anyway (or immediately if there's already head in > lost_out. > > This was fixed earlier in net-2.6.25 but got reverted among other > stuff and I didn't notice that this is still necessary (actually > wasn't even considering this case while trying to figure out the > reports because I lived with different kind of code than it in > reality was). > > This should solve the WARN_ONs in TCP code that as a result of > this triggered multiple times in every place we check for this > invariant. > > Special thanks to Dave Young and > Krishna Kumar2 for trying with my debug > patches. Hi, Thanks, after applying the patch the warning is not seen. Tested-by: Kamalesh Babulal > Signed-off-by: Ilpo Järvinen > --- > net/ipv4/tcp_input.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/net/ipv4/tcp_input.c b/net/ipv4/tcp_input.c > index 295490e..aa409a5 100644 > --- a/net/ipv4/tcp_input.c > +++ b/net/ipv4/tcp_input.c > @@ -2156,7 +2156,7 @@ static void tcp_mark_head_lost(struct sock *sk, int packets, int fast_rexmit) > tp->lost_skb_hint = skb; > tp->lost_cnt_hint = cnt; > > - if (tcp_is_fack(tp) || > + if (tcp_is_fack(tp) || tcp_is_reno(tp) || > (TCP_SKB_CB(skb)->sacked & TCPCB_SACKED_ACKED)) > cnt += tcp_skb_pcount(skb); > > -- > 1.5.2.2 -- Thanks & Regards, Kamalesh Babulal, Linux Technology Center, IBM, ISTL.