Hi Garrett, Thank you for your reply.
I have another question to ask. :) In current GLDv3 implementation, it is using mtops_header_info (mac type plugin function) to parse layer 2 packet header for both sending and receiving packets. However, IPoIB packets have different layer 2 header between sending packets and receiving packets. In other words, it can not use the same way to parse both sending and receiving packets. So it has to find way to differentiate sending, receiving and loopback packets in mtops_header_info function. What in my head is just add a new argument 'flag' to mtops_header_info. 'flag' will be used for indicating which path mtops_header_info is called. (sending path, receiving path or loopback). Actually, GLDv2 is using this way to deal with IPoIB packets, see function gld_interpret_ib. Should we make this change in GLDv3? Thanks, Lizhi Garrett D'Amore wrote: > On Thu, 2007-08-09 at 18:48 +0800, Lizhi Hou wrote: > >> In GLDv2 it has gldm_broadcast_addr in macinfo structure and it will >> be >> initialized during driver attaching. That means each driver instance >> will have its own broadcast address, so it works well with ibd >> driver. >> Should we keep this mechanism in GLDv3? >> > > I think if you need a per-instance broadcast address, then yes, having > it be in the instance structure is a good idea. > > -- Garrett > > _______________________________________________ > networking-discuss mailing list > [email protected] > > _______________________________________________ networking-discuss mailing list [email protected]
