[openib-general] [PATCH][RFC] CMA automatic port number assignment

[James Lentini]

On Fri, 24 Feb 2006, Steve Wise wrote:

> I don't think this is going to work for iWARP providers.  From an 
> iWARP perspective, either the CMA needs to track the entire port 
> number space across all providers (ie replicate what the native 
> stack does), or it needs to track port spaces per provider, or it 
> needs to query the provider to get an ephemeral port.  The problem 
> is that the provider might already have that port in use from a 
> previous connection.

My patch was intended to fix a very specific problem on IB. Stepping 
back and looking at the big picture, there are several different 
guarantees the CMA could make.

When the CMA automatically generates a port number, does it
 1) coordinate with the native transport (TCP/UDP/SCTP) stack? or
 2) coordinate across RDMA transports (IB and iWARP)? or
 3) coordinate across individual devices within a transport
    type? (all IB devices share a port space, all iWARP devices share
    a port space, but iB and iWARP aren't coordinated)
 4) coordinate on a device by device basis (an mthca device's port
    space would be separate from other mthca devices, ipath devices,
    etc.)

For IB, the CMA can guarantee #3 relatively easily. If this is going 
to be an issue for iWARP, it is acceptable for the CMA to only 
guarantee #4.

Do you agree Sean?