RE: unaligned accesses
>OK; UDB has built-in TULIP, are you using DE4X5 or TULIP driver? Both
>should be fine WRT not generating unaligned accesses, IIRC.
I'm using DE4X5 for the NIC.
>However, as you're doing NAT, you prolly have installed some other NIC
>in the PCI slot? What might it be, and what driver is it using?
It's a random D-Link card using via-rhine.
>You should be able to identify which driver (or perhaps non-driver) code
>the unaligned accesses are coming from, from the addresses in the
>messages in the log, I would think.
I have - the messages are from various places in the ipv4 network
code, tcp_ack, udp_recvmsg, tcp_recvmsg are the more recent places.
I don't know much about the Linux kernel internals, but perhaps these
folks have modified the kernel heap allocator to give out memory
on non-aligned boundaries? Or perhaps the via-rhine driver does
something similar. I have a few spare DE500s around, perhaps I
can switch to one of those?
>If you are using modules, it may be helpful to generate a monolithic
>kernel which might make the addresses in the messages more meaningful.
Okay - I'll do that too! :-)
>And as Paul noted, some of the network-related non-driver code would
>also be a good suspect, being used as heavily as it is, with (I think)
>some known problems (iptables?).
I couldn't find the exact correlation to 1 of the 2 places Paul
had found (the iptables stuff) in this source pool.
Is there a preferred method of bringing EXT3 support to the
2.2.x world? That's the only feature I would really need right
now to go back to the standard kernel dist (ant I really do not
*need* that for testing pruposes...).
To UNSUBSCRIBE, email to email@example.com
with a subject of "unsubscribe". Trouble? Contact firstname.lastname@example.org