[openib-general] problem with user-verb WC's

[Roland Dreier]

    Kyle> According to a 'TODO' buried in the gen2
    Kyle> src/linux-kernel/infiniband/hw/ : "MW support: ib_mthca does
    Kyle> not support memory windows"

Yes, it's true that mthca doesn't support memory windows yet.

    Kyle> The opcodes that I receive for non-rdma requests are all
    Kyle> correct, however, when posting rdma requests, I'm
    Kyle> consistently getting work completions with opcodes of:
    Kyle> IBV_WC_BIND_MW

    Kyle> I'm not making any (known) calls or requests to bind to a
    Kyle> memory window, or for that matter to create a memory window.
    Kyle> So how does a completion event get generated with an opcode
    Kyle> indicating a currently unimplemented feature has just
    Kyle> finished?  And are there other reasons why I should/would be
    Kyle> getting this type of completion?

Looking at the code, I don't see anything obviously wrong.  What is
the status field of the completions that you see with the opcode
IBV_WC_BIND_MW?  If the status is not success, then the opcode field
is undefined (only the wrid is valid).

 - R.