[openib-general] Re: [PATCH] CMA and iWARP

[Roland Dreier]

    Caitlin> It is important to separate two issues here: L2-L3
    Caitlin> coordination and L4 coordination.

    Caitlin> The patches that Tom recently posted address the L2-L3
    Caitlin> coordination.  They ensure that the TCP/IP stack used by
    Caitlin> the RNIC is consistent in its L2-L3 configuration with
    Caitlin> the host stack. For example, it will route packets to the
    Caitlin> same destination IP address the same way as the host
    Caitlin> stack, use the same MAC addresses for first hops, etc.

But the patches don't provide full L2 coordination.  For example if I
install a netfilter rule that discards packets from a particular
source MAC (a completely, 100% L2 notion), the RNIC will still happily
accept connections from that MAC, right?

    Caitlin> Those will be compicated give and takes. There is no
    Caitlin> reason to hold up the L2-L3 work because of it. If it
    Caitlin> were, then SDP/IB should have been deferred until
    Caitlin> complete consistency was achieved. Updating one step at a
    Caitlin> time makes more sense.

SDP/IB is not in the upstream kernel.  And I agree that there are many
issues with respect to network stack integration to work out before
SDP is suitable for merging.

 - R.