Michael S. Tsirkin wrote:
That's unfortunate, I think CMA needs this and it would be nice to develop CMA out of kernel against in-kernel CM. Is that right, Sean? Does CMA require private data matching to work?
The RDMA CM requires these changes. Yes, it would have been nice to sync the kernel tree, but only so I wouldn't have to generate those patches again. :)
Still, we really want to prevent userspace from using CMA/SDP service IDs, so I think we need at least this part. Sean, if you agree, please send Roland this patch.
The CMA/SDP service IDs are defined by the IB spec, so a userspace consumer shouldn't be able to use those ranges. Allowing this probably isn't a big issue until either are in the kernel. I.e. if the CMA isn't loaded, then it doesn't really matter if someone uses the CMA's service ID. And if the CMA is loaded, then this patch would have been applied locally.
- 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
