Hi Jurij,
On Mon, May 02, 2005 at 11:03:55PM -0400, Jurij Smakov wrote:
> Hi,
>
> After some further investigation I can now pretty confidently state that
> the culprit for this problem is the ipsec patch, introduced into 2.4.27
> kernel by the previous kernel maintainer Herbert Xu. I have built two
> kernels, a pristine 2.4.27 from www.kernel.org, and another one, with the
> ipsec patch [0] applied. With the pristine kernel everything works fine,
> with the ipsec one I see bad tcp checksums in tcpdump log when connecting
> both to m-nas1 and esslingen (even though in the latter case the connection
> does not hang). And since the ipsec patch got accepted upstream somewhere
> in 2.5.x, it is not surprising that the failures persist with all later
> kernels, including 2.6.x.
>
> The ipsec patch is not the only condition for this failure though. In
> particular, I have completely failed to reproduce the problem on my own
> Ultra1, which is very similar (it is not SMP though). So, it might be that
> it is caused by a combination of ipsec and SMP (not likely, because tested
> kernels are all UP), or some other unidentified factors.
Do you have hme or lance on your ultra1? Remember that we only have
problems on Ultras IIs with on-board-hme. Does the ipsec-patch do
anything differently depending on the type of network-hardware? Or
does the lance-driver fix the bad checksums before sending the packets
over the wire (which would be a strange thing)?
> In summary, I do not have a clear idea of how we could fix this. Ripping
> out ipsec out of 2.6 kernels by now is out of the question. It you want a
> "quick and dirty" solution, you can build a 2.4.27 kernel without applying
> ipsec patch. I'll send a summary of this problem to debian-sparc and see
> if someone else can reproduce that using the kernels I've built.
Next week I will have access to an Ultra1 with hme and see if the
problem can be reproduced on that box.
P.S.: Sorry that my c-skills are too bad (would love to
do some kernel-hacking, maybe in my next life...)
Regads and thanks for the help!
Raoul
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
