[ofa-general] IB credit-based flow control

[Sean Hefty]

>I'm reading C9-162 and still not seeing why (according to the spec
>anyways) there should ever be RNR retries on a connection. I would think
>the receiving HCA would not credit its last WQE to the sender, and thus
>retries should never happen?
>
>The whole point of this feature is to eliminate RNR retries, no?

What I'm looking at for C9-162 is:

C9-162: When the requester encounters a WQE on its send queue for
which it has no available credits, that WQE is said to be limited.
If the limited request WQE is a SEND request, the send queue shall
transmit no more than a single packet of the request message before
it must stop transmission and wait for an acknowledge packet.

My assumption is that if no credits are available when the SEND request arrives,
then the receiver generates a RNR message, but I didn't read through the entire
section to verify this.

This is totally a guess, but there needs to be some sort of recovery mechanism
in place to handle a lost credit update message.  Allowing the requester to
issue a limited request in the absence of credits will force a credit update if
any are available.

Did you verify that the HCAs you're using implement e2e flow control?

- Sean