[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Fri, 4 Apr 2008 22:15:43 +0300 (EEST)
From: "Ilpo Järvinen" <ilpo.jarvinen@...sinki.fi>
To: Georgi Chorbadzhiyski <gf@...xsol.org>
cc: Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
Netdev <netdev@...r.kernel.org>
Subject: Re: Linux 2.6.25-rc6: WARNING: at net/ipv4/tcp_input.c:2510
On Fri, 4 Apr 2008, Georgi Chorbadzhiyski wrote:
> Around 04/03/08 13:17, Ilpo Järvinen scribbled:
> > On Thu, 27 Mar 2008, Georgi Chorbadzhiyski wrote:
> >
> >> A week ago I updated one of my servers to 2.6.25-rc6 and today looking
> >> through logs I've found these:
> >>
> >> [380219.823434] ------------[ cut here ]------------
> >> [380219.823450] WARNING: at net/ipv4/tcp_input.c:2510 tcp_ack+0xc7e/0x177d()
> >
> > Is this reproducable for you? If yes, is it possible for you to run
>
> Not in the moment.
>
> I'm not getting such errors in the log since my last report. Today I'll
> update to latest -git and will restart the server.
Yeah, the reason why these still happen is that they are quite hard to
hit.
> The machine that gave me this error is a http and ftp server handling not
> very busy distro mirror (if this can give you any clues).
>
> > processing expensive verification of TCP state (validates TCP write queue
> > invariants multiple times per ACK by scanning through the write queue) on
> > such machine?
>
> How to run such a thing, can you give me some pointers?
I sent the patch recently to netdev/lkml:
http://marc.info/?l=linux-netdev&m=120733529216457&w=2
But it's very expensive in processing wise so it might not be wise to put
it in if performance is important for you.
--
i.
Powered by blists - more mailing lists