[openib-general] mthca_mad.c

[Roland Dreier]

    Hal> Shouldn't most of the functionality in mthca_mad.c be above
    Hal> the driver rather than part of the driver ? It seems to me
    Hal> that most of this is part of the access layer rather than the
    Hal> driver.

We had some discussion about how to implement the SMA earlier, and it
was suggested that more of the SMA should be handled in common code.
However my feeling is that the SMA is quite device-specific and the
low-level driver should have maximum flexibility about how to handle
things.  For example, the Tavor generates traps that need to be
explicitly forwarded to the SM.

    Hal> Also, a nit: In Sean's ib_verbs.h it is process_mad and
    Hal> whereas in Roland's version it is mad_process.

This is because I have not yet updated that entry point in my tree.
In my old API, I named functions like {OBJECT}_{METHOD} (so that eg
qp_create, qp_destroy, qp_modify etc all sort together), while Sean
uses a {METHOD}_{OBJECT} naming convention.  So this and all the other
naming differences will be fixed as I finish merging with Sean's
ib_verbs.h.

 - Roland