[1500] in linux-net channel archive
Re: UDP: bad checksum.
daemon@ATHENA.MIT.EDU (Dave Caswell)
Thu Dec 7 17:58:59 1995
Date: Wed, 6 Dec 1995 05:56:07 -0500
From: Dave Caswell <davec@helium.ftp.com>
To: dholland@hcs.harvard.edu
CC: STEFFENR@UWSTOUT.EDU, linux-kernel@vger.rutgers.edu,
linux-net@vger.rutgers.edu, davec@world.std.com
In-reply-to: <199511302339.SAA15361@hcs.harvard.edu> (message from David
Holland on Thu, 30 Nov 1995 18:39:45 -0500 (EST))
> > The messages I see are a little different:
> >
> > UDP: bad checksum. From 00000000:68 to FFFFFFFF:67 ulen 308
> >
>It is a report that a bad packet came in from someplace off the
>network. It is not a bug. Chances are there's a messed-up DOS machine
>someplace using bootp.
In my case, I had configured the cards (I have two in this machine.) like:
ether=15,0x280,0xD0000,0xD3FFF,eth0
ether=5,0x300,0xCA000,0xCBFFF,eth1
Which is one byte different from
ether=15,0x280,0xD0000,0xD4000,eth0
ether=5,0x300,0xCA000,0xCC000,eth1
This small difference was causing amazingly garbled packets... Not
just single dropped bytes.
I'd like to suggest that when it's possible, driver arguments taken
from the kernel arguments be sanity checked. In this case, the card
memory comes in 256 byte units. When the ASIC chip was programmed,
rounding caused it to not use the last block of memory on the card,
while the driver was trying to use 255 bytes. If the value had
generated a warning, or had rounded to the nearest correct value, or
both I wouldn't have had any problem.
>Perhaps this message shouldn't be unconditionally sent to the console
>though. Linus?
I think that for the development kernels, I like to see more error
messages, rather than less.
-davec
