Return-Path: Received: by vger.rutgers.edu id <154518-28471>; Mon, 7 Sep 1998 09:15:29 -0400 Received: from styx.cs.kuleuven.ac.be ([134.58.40.3]:35323 "EHLO styx.cs.kuleuven.ac.be" ident: "TIMEDOUT2") by vger.rutgers.edu with ESMTP id <154761-28471>; Mon, 7 Sep 1998 08:21:17 -0400 Date: Mon, 7 Sep 1998 16:47:23 +0200 (CEST) From: Geert Uytterhoeven To: "David S. Miller" , Ulrik De Bie cc: Linux kernel , Linux/m68k Subject: Re: IPv4 kernel messages In-Reply-To: <199809070016.RAA09205@dm.cobaltmicro.com> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: owner-linux-kernel@vger.rutgers.edu Content-Length: 2479 Lines: 61 On Sun, 6 Sep 1998, David S. Miller wrote: > Date: Mon, 7 Sep 1998 01:12:16 +0200 (CEST) > From: Geert Uytterhoeven > > They're still there: > > | Socket destroy delayed (r=0 w=160) > | TCPv4 bad checksum from 10.0.24.8:0071 to 10.0.24.4:040c, len=20/20/40 > > Are these just warnings to be disabled in 2.2, or is there something wrong? > > 10.0.24.4 runs Linux/m68k 2.1.119. 10.0.24.8 is a PPC running vger-19980903. > I see them only rarely on my main workstation which talks a lot to the > rest of the world. On my internal Linux-2.1.x machines I never see > it. > > I would check out the csum routines on m68k and PPC to make sure they > are sane in all cases. > > Here is the lifesaver development utility I have been using for a long > time to verify all major changes made to Sparc and MIPS checksumming > routines. It won't work for you as-is, but you can figure out what it > is supposed to do and link the PPC and M68k csum routines into it to > perform verifications. Note there is an ugly piece of MIPS inline asm > in here which you'll need to remove too as the last thing I used this > for was the Cobalt kernels :-) Note it also performs performance > metrics on your code too, so it's useful for speed tuning as well as > verification. Thanks! The PPC checksumming code passed the test, but the m68k code seems to be broken. I got: verify_ip_fast_csum: buffer 0 gets bogus csum abe7! Ulrik De Bie wrote: > I think the following information is interesting here. I have the same with > Linux/i486 running recent-vger and PPC running recent-vger. > > So both intel and m68k are wrong, or ppc is wrong. Which would give a big > hint to the latter. Since PPC passed the test, this would indicate that ia32 is incorrect, too. But if I understand it correctly, the test isn't exhaustive, right? Greetings, Geert P.S. I put the code I used (i.e. Dave's with adaptations for ppc and m68k) on my webpage: http://www.cs.kuleuven.ac.be/~geert/bin/cksum_helper.c -- Geert Uytterhoeven Geert.Uytterhoeven@cs.kuleuven.ac.be Wavelets, Linux/{m68k~Amiga,PPC~CHRP} http://www.cs.kuleuven.ac.be/~geert/ Department of Computer Science -- Katholieke Universiteit Leuven -- Belgium - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/faq.html