Guy German wrote:
I believe that ib_at is still a valuable module even if ATS reverse ARP is broken, and I think we should discuss this.
Here's my thinking on this. ATS is broken as you mentioned for reverse lookups. However, if we want to keep ATS, I think that ATS registration/deregistration should be integrated with IPoIB. To keep it separate, we will need to patch net_device to provide an rdma_ptr as suggested by Roland.
For ARP, we can extract ARP code from ib_at and use it with the CMA. The CMA can then use the "ib_arp" module to map network addresses to GIDs, then issue path record queries using ib_sa.
Any caching should be done in other places. For example, ARP already does its own caching, and I believe that a generic SA caching module should be a part of ib_sa or exist separately.
This seems the best approach to getting a higher level RDMA connection interface into the kernel.
- Sean _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
