This is great! And a reasonably simple fix too. The 10GbE core will be
under the knife in the coming weeks. It'll be simplified to improve
timing because its fabric interface is currently limiting ROACH
compiles to <220MHz.
We'll hopefully role-in this patch at the same time. Another thing I'd
like to add to the 10GbE core is VLAN tagging, but that's a low
priority at the moment.
Thanks,
Jason
On 05 Feb 2009, at 13:49, Jan Wagner wrote:
Hi,
I added UDP multicast sending support to the 10G IP core.
Now all 224.0.0.0 through 239.255.255.255 are sent correctly with
the corresponding multicast MAC.
Earlier these UDP packets would have incorrectly been routed through
the gateway MAC (and e.g. our HP6400cl and Summit X450 did not
detect and convert the header multicast-IP target MAC from the wrong
into the correct multicast MAC upon switching the packet to another
10G port).
The "patch" is attached. It is a diff to "some" version of the 10G
core, not the SVN, but the patch is rather short and you can apply
it manually if it doesn't work directly.
Maybe you can commit the changes to the ibob/roach SVN?
By the way, there is one remaining TODO: I added provisions for TTL
('tx_ip_ttl'). Default TTL is 0xFF, but for multicast 0x01 is
preferrable. The magic value X"8412" in the tx_ip_checksum_fixed_0
sum needs to be changed to include 'tx_ip_ttl'. I couldn't make
sense of the X"8412" so the 0x01 TTL is not supported by the patch
yet. Maybe somebody can figure out the computation?
- Jan
--
****************************************************
Helsinki University of Technology
Dept. of Metsähovi Radio Observatory
http://www.metsahovi.fi/~jwagner/
Work +358-9-428320-36<patch.txt>
