svn commit: r304426 - stable/11/sys/net

[Harry Schmalzbauer]

Bezüglich Alexander Motin's Nachricht vom 24.08.2016 11:02 (localtime):
> On 24.08.16 11:42, Harry Schmalzbauer wrote:
>>  Bezüglich Alexander Motin's Nachricht vom 18.08.2016 14:08 (localtime):
>>> Author: mav
>>> Date: Thu Aug 18 12:08:39 2016
>>> New Revision: 304426
>>> URL: https://svnweb.freebsd.org/changeset/base/304426
>>>
>>> Log:
>>>   MFC r303009: Negotiate/disable TXCSUM_IPV6 same as TXCSUM.
>>
>> Have you asked for RE-approval to get this into 11.0?
>> If I understand it correctly it's a essential fix for jail+epair+IPv6
>> useres, solving nasty hard to track oddities?
> 
> No, I haven't asked, because I haven't seen related bug reports and
> found this accidentally.  But if anybody wants it -- be my guest.

I haven't filed a PR because I ran out of time tracking my bridge(4)
oddities, since so many exotics are involved:
VT-d PCIe-passthrough of two igb(4) = lagg(4) - bridge(4) - epair(4).
At some point I clearly found bridge(4) misbehaving, and I did some(tm)
workarround and went on. This was with stable/10 between 10.0 and 10.1
(currently running 10.1 on that setup).
Can't tell if igb(4) in 10.1 is even TXCSUM_IPV6 capable (this was added
quiet lately in igb(4)'s live, afair), but the symptoms match very well
this missing TXCSUM_IPV6 disabling.

Even if not in my case, as far as I understand, everybody else using
11.0-RC2 with TXCSUM_IPV6-enabledNic(4)+bridge(4) would suffer from
network failures.

You found + fixed it, so 11.0 should get this :-)

Thanks,

-harry