> > > The RoCE Verbs interface references the HCA GID table in QPs and AHs, > > for all RoCE versions. > > The IBTA specifically does not define software interfaces. The concern here > is > the architecture and definition of the linux rdma software stack, not verbs, > despite the fact that the layer is named that. >
Whatever. The ib_* API is the Linux implementation for IBTA Verbs (both Infiniband and RoCE flavors) and iWARP Verbs. This interface uses a GID table. In this patch-set, there is no conceptual change in this regard. It adds GID types, as stated by the RoCE specification, which is a natural extension to the existing code. I am sure that we can consider additional changes in the future, but one patch-set at a time... > > As I said earlier, the network layer (e.g., IPv4, IPv6, or GRH) cannot > > be a port attribute because RoCE HCAs support all of them over the > > same physical port. > > IMO, the architecture needs to change to support multiple network and > transport layers over the same link layer. The current restriction of 1 > "transport" per port is too restrictive, and doesn't seem to be true even for > current devices, like usnic. > usnic uses UDP transport. Infiniband and RoCE use the IB transport (all the BTH+ header stuff as defined by IBTA). iWARP uses IETF defined protocols over TCP, namely RDMAP+DDP+MPA. I am not aware of any device that mixes these transports today. If we ever see such a need in the future, let's talk again. Currently, this is irrelevant. > > Maybe we should change these patches to encode the port "summary" as a > > bit mask, and provide convenient masks for queries? > > Switching to bit masks is the longer term objective. -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html