[ofa-general] recv with rdma immediate completion questions

[Ralph Campbell]

On Fri, 2008-06-06 at 12:46 -0500, Steve Wise wrote:
> Roland Dreier wrote:
> >  > Why is there a work completion opcode IB_WC_RECV_RDMA_WITH_IMM -and- a
> >  > work completion flag IB_WC_WITH_IMM indicating the immediate data is
> >  > present in the WC?
> >
> > Off the top of my head ... because RDMA operations usually generate no
> > completion on the target side, so you need some opcode for when one
> > comes in with immediate data.
> >
> > I'd have to check what the IB verbs spec says about this case (obviously
> > iWARP doesn't have to worry about immediate data)
> >   
> What I was thinking was that:
> 
> completion with IB_WC_RECV + IB_WC_WITH_IMM is the same information as
> completion wtih IB_WC_RECV_RDMA_WITH_IMM + IB_WC_WITH_IMMM.
> 
> And I was wondering if there was a good reason for this, and whether we 
> need both an opcode -and- a flag for recv completions that are due to an 
> ingress SEND_WITH_INV...
> 
> Seems like you only either need the opcode or the flag, but not both...
> 
> Steve.

The cases are: opcode == IB_WC_RECV, flags 0 or IB_WC_WITH_IMM,
opcode == IB_WC_RECV_RDMA_WITH_IMM, flags IB_WC_WITH_IMM.
There is no IB_WC_RECV_RDMA opcode because it doesn't create a CQE.