Re: unaligned accesses
On Mon 29 Jul 2002, Thomas Evans wrote:
>
> I'm still using a 2.2 kernel on a UDB that I use as a NAT.
> It's also a seriously hacked RH installation from way
> back, before I saw the Debian light. I hack on the machine
> quite a bit, but I'm hesitant to make a major chane (like
> wiping it and starting with a Debian install).
>
> The networking code seems to be generating millions of
> kernel alignment fixups per day ( and during large file
> transfers, like an apt-get dist upgrade from a client behind
> the NAT, about a million per minute ). Is this normal
> or have I done something wrong?
>
> I am using a WOLK kernel on the UDB ( it has back ported EXT3
> and other "features" I wanted on that machine ), so that may be
> the issue - perhaps I can can EXT3 support in a different way.
Does it also have iptables? That's where I found a *lot* were generated
(in one place). Also the packet filter generates them (tcpdump uses
that).
When I was using 2.2 on my firewall alpha, I never got any kernel
unaligned accesses (and that was with plenty of ipchains rules). That's
why I looked into them when I saw them appear with the 2.4 kernel.
> The ipv4 netorking code is my main suspect based on info
> from /proc/cpuinfo and the kernel .map file.
> I've tried loading the symbols with gdb and correlating lines,
> but have had limited success.
Yeah, I tried that too.
I ended up looking at the assembler code, making a change to the source
where I suspected the problem was (adding a call to a function blabla()
or something stupid yet recognizable like that :-) and using gcc to
generate assembler code from that to see whether that was indeed the
right place. Up 10 days now with 0 unaligned traps :-)
Paul Slootman
--
To UNSUBSCRIBE, email to debian-alpha-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: