Re: [PATCH v4 for-next 00/12] Add network namespace support in the RDMA-CM

On Tue, May 19, 2015 at 11:31 PM, Haggai Eran [off-list ref] wrote:

On Tuesday, May 19, 2015 7:39 PM, Parav Pandit [off-list ref] wrote:

quoted

This has come up during our first round of patch set for RoCEv2.
However due to last set of changes in this patch set, we decided to
move in for later submission.

Current high design approach to support this is:

A callback function registered to IP stack for IP address
assignment/removal. (it could be done possibly done for GID entry
update, instead of IP)
For each such namespace, for device supporting RoCEv2, one UDP socket
with RoCEv2 UDP port would be created and destroyed.

This needs to be done to take ownership of the RoCEv2 UDP port of course, but there is more to RoCE namespace support then that. You will need to de-multiplex rdma_cm_id requests from a shared ib_cm_id similarly to what is done in the InfiniBand patch-set here. Using the new RoCE GID cache should make it relatively easy. In addition, in order to support link-layer address resolution in the verbs layer as it is done today, we may need to add namespace support to the modify QP verb and the create AH verb.

Yes. We talked sometime back with Liran and Matan on same topic.

Haggai

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html