[openib-general] ib_mad_recv_done_handler questions

[Sean Hefty]

Roland Dreier wrote:

>     Sean> * If the underlying driver provides a process_mad routine, a
>     Sean> response MAD is allocated every time a MAD is received on QP
>     Sean> 0 or 1.  Can we either push this allocation down into the
>     Sean> HCA driver, or find an alternative way of interacting
>     Sean> between the two drivers that doesn't require this allocation
>     Sean> unless a response will be generated?
> 
> How about if the MAD layer allocates a response MAD when a MAD is
> received, and if the process_mad call doesn't actually generate a
> response the MAD layer just stashed the response MAD away to use for
> the next receive?  This should keep the number of allocations within 1
> of the number of responses actually generated, but save us from
> tracking allocations between two layers.

That sounds reasonable, and I think avoiding allocations in the HCA 
driver is desirable given the current design.

- Sean