[ofw] patch 2/2 Add support for RDMAoEth to the low level driver

[Sean Hefty]

>> >+ attr.port_num           = (u8)port;
>> >  attr.qp_state   = IBV_QPS_RTR;
>> >  attr.path_mtu   = mtu;
>> >  attr.dest_qp_num  = dest->qpn;
>> >@@ -81,6 +86,15 @@
>> >  attr.ah_attr.src_path_bits = 0;
>> >  attr.ah_attr.port_num  = (uint8_t) port;
>>
>> Why isn't this sufficient to specify the port?
>>
>Can you please explain what you mean?

port is set twice: attr.port_num and attr.ah_attr.port_num.  Why isn't the
latter still sufficient?

>> >@@ -606,7 +624,7 @@
>> >  my_dest.psn = rand() & 0xffffff;
>> >  if (!my_dest.lid) {
>> >   fprintf(stderr, "Couldn't get local LID\n");
>> >-  return 1;
>> >+  //return 1;
>>
>> This is an unusual change.
>The real issue is that if you are running on native IB than you should
>have a lid. On the other hand if you are running on LLE, than you don't
>have a lid (and this is not an error). As a result I'll change the test
>to realy exit if this is running over ib only.
>I'm thinking of something like: "if I have a lid but you don't then
>exit".

Removing the return statement would have made more sense than simply commenting
it out.

Are these changes sufficient to use RDMAoE?  An out of tree driver is still
needed, correct?

- Sean