On 2/19/2021 8:57 AM, Jason Gunthorpe wrote:

On Fri, Feb 19, 2021 at 01:06:26PM +0000, Bernard Metzler wrote:

quoted

Actually, all this GID and GUID and friends for iWARP
CM looks more like squeezing things into InfiniBand terms,
where we could just rely on plain ARP and IP
(ARP resolve interface, see if there is an RDMA device
bound to, done)... or do I miss something?

I don't know how iWarp cM works very well, it would not be surprising
if the gid table code has gained general rocee behaviors that are not
applicable to iwarp modes.

iWarp doesn't really need a CM, it is capable of peer-to-peer without
any need to assign connection and queuepair ID's. The CM infrastructure
basically just implements a state machine to allow upper layers to have
a consistent connection API.

I'm with Bernard here, forcing iWarp to use CM is a fairly unnatural
act. Assigning a GID/GUID is unnecessary from a protocol perspective.

Tom.

With Steve gone I don't think there is really anyone left that even
really knows how the iWarp stuff works??

Jason