[openib-general] GSI compromise

[Roland Dreier]

    Roland> - how does the MAD layer know how many elements are in the
    Roland> method_array parameter of ib_mad_reg_class()?

    Sean> I was assuming a fixed size of 127 elements.  (At least I
    Sean> think it's 127 methods per class.)  This could probably be
    Sean> an optional argument if the registration is for all methods
    Sean> of a given class.

How does the response bit get handled then?  (eg lots of consumers
will want to see get responses but not gets).

Also if it's just a bitmask of which method we want, we might as well
make it a bitmap (ie use DECLARE_BITMAP(method_mask, 256) or something
like that).

    Roland> - should we use a "struct list_head" instead of our own type
    Roland> for ib_mad_msg.next?

    Sean> I'm fine with this.  I followed what I had for the work
    Sean> requests.  Maybe those should change as well?

That's a good idea.  It adds one more pointer to the work request
(since struct list_head is doubly linked) but I think being able to
use macros like list_for_each(), list_del(), etc. is worth it.

    Sean> I'm good with redirection helper routines.  I'd just like to
    Sean> get a clean layering.  Btw, does anyone know if you can
    Sean> redirect in the midding of sending an RMPP message?

I don't think so -- my impression is that a redirect message counts as
a response to a request, so you can't send one until you get the whole request.

 - R.