Roland Dreier wrote:
> So indeed the assumption in the patch is that mgids which translate to
> legal IP multicast addresses are inserted into the database either by
> ipoib or rdma-cm consumers who use IPOIB_PS for their ID's.
I guess since it's confiruable, it's OK. But I think that you miss
the fact that there might be other consumers of ib_sa creating
multicast groups, and there might be other rdma_cm consumers using
IPOIB_PS also.
I understand that there may be other ib_sa consumers that use multicast
and other rdma_cm consumers that use IPOIB_PS, however, the point I was
trying to make is that if there are consumers that join the --same--
multicast groups as ipoib they can actually avoid ipoib to join these
group if they provide different attributes for the group.
This is why Sean made the rdma_cm to use the --same-- attributes as
ipoib does for IPOIB_PS joins. For other joins the rdma-cm makes sure
that the MGID is different (rdma-cm signature instead of the ipv4/v6
one) and uses a different qkey.
Anyway, I understand that we are more or less on the same page regarding
this point, correct?
Or.
_______________________________________________
general mailing list
[email protected]
http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
