Re: TCP-MD5 checksum failure on x86_64 SMP

[Bhaskar Dutta <bhaskie@gmail.com>]

On Fri, May 7, 2010 at 3:07 PM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> Le vendredi 07 mai 2010 à 14:29 +0530, Bhaskar Dutta a écrit :
>
>> Eric,
>>
>> Thanks a lot! I will test it out and let you know.
>> BTW this patch seems to essentially do the same as the earlier fix you
>> had posted (where you just do bh disable/enable).
>> Am I missing something?
>>
>> With the earlier fix, I ran load tests with 80 TCP connections for
>> over 6 hrs and found 5 bad checksum packets.
>> So there is still a problem. Without the fix I see a bad packet every
>> minute or so.
>
> My second patch is cleaner, using only out of line code (inline was not
> necessary and made include file bigger than necessary). Inline is fine
> if we can avoid a function call, but it was not the case.
>
> If you notice another corruption, it may be because of another problem,
> yet to be discovered.
>
> To you have a userland suite to test/stress tcp md5 connections ?
>
>

We are still trying to find the other corruption.

I will send you the tarball of a userland client/server suite to
stress test the TCP-MD5 that we've been using to reproduce the issue.

Thanks,
Bhaskar