From mboxrd@z Thu Jan 1 00:00:00 1970 From: Oleksandr Natalenko Subject: Re: [REGRESSION] tcp/ipv4: kernel panic because of (possible) division by zero Date: Sat, 09 Jan 2016 19:34:47 +0200 Message-ID: <2440018.ccxNiMU8Fd@spock> References: <26396443.iDTTxgChSj@spock> Mime-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: QUOTED-PRINTABLE Cc: netdev , "linux-kernel@vger.kernel.org" , Patrick McHardy , Hideaki YOSHIFUJI , James Morris , Alexey Kuznetsov , "David S. Miller" To: Yuchung Cheng Return-path: In-Reply-To: Sender: linux-kernel-owner@vger.kernel.org List-Id: netdev.vger.kernel.org Here is stacktrace got via netconsole with ECN enabled and first patch=20 applied: https://gist.github.com/57fd5e2795b86c40eb80 Now will recompile kernel with patch you've sent to upstream and test i= t with=20 ECN enabled again. On =D1=81=D0=B5=D1=80=D0=B5=D0=B4=D0=B0, 6 =D1=81=D1=96=D1=87=D0=BD=D1=8F= 2016 =D1=80. 10:43:45 EET Yuchung Cheng wrote: > On Wed, Jan 6, 2016 at 10:19 AM, Yuchung Cheng wr= ote: > > On Wed, Jan 6, 2016 at 8:50 AM, Oleksandr Natalenko > >=20 > > wrote: > >> Unfortunately, the patch didn't help -- I've got the same stacktra= ce with > >> slightly different offset (+3) within the function. > >>=20 > >> Now trying to get full stacktrace via netconsole. Need more time. > >>=20 > >> Meanwhile, any other ideas on what went wrong? > >=20 > > That's odd b/c the patch already checks and avoids div0. Can post m= e > > the stacktrace and kernel warnings if any ... > >=20 > > One possibility is that tcp_cwnd_reduction() may set a cwnd of 0, > > which then gets used to start another recovery phase. This may or m= ay > > not be the culprit of this div0 issue because I wasn't able to > > reproduce exactly your issue on our servers. But I will post the fi= x > > today and CC you. >=20 > Could you turn off ecn (sysctl net.ipv4.tcp_ecn=3D0) to see if this s= till > happen? > >> On December 22, 2015 4:10:32 AM EET, Yuchung Cheng =20 wrote: > >> >On Mon, Dec 21, 2015 at 12:25 PM, Oleksandr Natalenko > >> > > >> > wrote: > >> >> Commit 3759824da87b30ce7a35b4873b62b0ba38905ef5 (tcp: PRR uses = CRB > >> > > >> >mode by > >> > > >> >> default and SS mode conditionally) introduced changes to > >> > > >> >net/ipv4/tcp_input.c > >> > > >> >> tcp_cwnd_reduction() that, possibly, cause division by zero, an= d > >> > > >> >therefore, > >> > > >> >> kernel panic in interrupt handler [1]. > >> >>=20 > >> >> Reverting 3759824da87b30ce7a35b4873b62b0ba38905ef5 seems to fix= the > >> > > >> >issue. > >> > > >> >> I'm able to reproduce the issue on 4.3.0=E2=80=934.3.3 once per= several day > >> >> (occasionally). > >> >>=20 > >> >> What could be done to help in debugging this issue? > >> > > >> >Do you have ECN enabled (i.e. sysctl net.ipv4.tcp_ecn > 0)? > >> > > >> >If so I suspect an ACK carrying ECE during CA_Loss causes enterin= g CWR > >> >state w/o calling tcp_init_cwnd_reduct() to set tp->prior_cwnd. C= an > >> >you try this debug / quick-fix patch and send me the error messag= e if > >> >any? > >> > > >> >> Regards, > >> >>=20 > >> >> Oleksandr. > >> >>=20 > >> >> [1] http://i.piccy.info/ > >> > > >> >i9/6f5cb187c4ff282d189f78c63f95af43/1450729403/283985/951663/pani= c.jpg > >>=20 > >> -- > >> Sent from my Android device with K-9 Mail. Please excuse my brevit= y.