Intel igb 82576 (netmap mode) and RSS queue issuse with fragmented UDP packets.

[Adrian Chadd]

On 1 June 2015 at 07:54, Slawa Olhovchenkov <slw at zxy.spb.ru> wrote:
> On Mon, Jun 01, 2015 at 07:50:31AM -0700, Adrian Chadd wrote:
>
>> They all behave the same. You can fix the intel driver(s) by looking
>> for the TCP/UDP queue config and only enabling IPv4/IPv6 (not TCP/UDP)
>> hashing.
>
> Are you sure that enabling only IPv4/IPv6 hashing fix this?
> I see hashing in may case depends from flags fields.
> flags field is not part of TCP/UDP.

Yes. Just try doing ipv4/ipv6 only hashing. See if that fixes it.


-a

>
>> On 1 June 2015 at 01:25, Slawa Olhovchenkov <slw at zxy.spb.ru> wrote:
>> > I am use Intel 82576 (igb driver) in netmap mode.
>> > I see isssuse: flow routed in different queue depends of IP flags.
>> >
>> > For example:
>> >
>> > Total 4 queue.
>> > UDP flow.
>> > SRC: 91.214.70.167:12062 DST: 95.31.64.226:49939 Flags:DF routed to queue 0
>> > SRC: 91.214.70.167:12062 DST: 95.31.64.226:49939 Flags:More_Frags (offset 0, length 1396) routed to queue 3
>> >
>> > I understand that next fragment (IP (tos 0x0, ttl 62, id 36361, offset
>> > 1376, flags [none], proto UDP (17), length 102) 91.214.70.167 >
>> > 95.72.171.116: ip-proto-17) may be routed in different queue, but why
>> > fragment 0 routed to different queue? This is software (driver,
>> > netmap) issuse? Or this is hardware (silicon) isssuse?
>> >
>> > What about Intel 10G/40G cards? Chelsio?
>> >
>> > _______________________________________________
>> > freebsd-net at freebsd.org mailing list
>> > http://lists.freebsd.org/mailman/listinfo/freebsd-net
>> > To unsubscribe, send any mail to "freebsd-net-unsubscribe at freebsd.org"