[openib-general] Usage of ib_mad_recv_wc.recv_buf?

Mon Oct 18 09:11:34 PDT 2004

    Hal> I believe that the intention is that the receive completion
    Hal> is not indicated to the consumer until the complete RMPP
    Hal> transaction has been received. The multiple receive
    Hal> completions (one per segment) are "coaelsced" into one and
    Hal> provided to the client when the final RMPP segment is
    Hal> received.

Right, and then the consumer can call

        void ib_coalesce_recv_mad(struct ib_mad_recv_wc *mad_recv_wc,
                                  void *buf);

to get the MAD packets' payloads copied into a single buffer.

What I was trying to point out was that we have:

        struct ib_mad_recv_buf {
                struct list_head        list;
                struct ib_grh           *grh;
                struct ib_mad           *mad;
        };

with "list" described as "Reference to next data buffer for a received
RMPP MAD," but:

        struct ib_mad_recv_wc {
                struct ib_wc            *wc;
                struct ib_mad_recv_buf  *recv_buf;
                int                     mad_len;
        };

with only a pointer to a single ib_mad_recv_buf (although the comment
describes recv_buf as "Specifies the location of the received data
buffer(s)").

It seems to me that if ib_mad_recv_wc is supposed to return a list of
MAD packets, then instead of the recv_buf member, it should have a
struct list_head member to serve as the head of the list of packets.

 - R.