[ofa-general] Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review

[Michael S. Tsirkin]

> Quoting Pradeep Satyanarayana <pradeeps at linux.vnet.ibm.com>:
> Subject: Re: IPOIB CM (NOSRQ)[PATCH V4] patch for review
> 
> If there are no other issues than the small restructure suggestion that
> Michael had, can this patch be merged into the for-2.6.22 tree?

I'm not sure.

I haven't the time, at the moment, to go over the patch again in depth.
Have the issues from this message been addressed?

http://www.mail-archive.com/general@lists.openfabrics.org/msg02056.html

Just a quick review, it seems that two most important issues have
apparently not been addressed yet:

1. Testing device SRQ capability twice on each RX packet is just too ugly,
   and it *should* be possible to structure the code
   by separating common functionality in separate
   functions instead of scattering if (srq) tests around.

2. Once the number of created connections exceeds
   the constant that you allow, all attempts to communicate
   with this node over IP over IB will fail.
   A way needs to be designed to switch to the datagram mode,
   and to retry going back to connected after some time.
   [We actually have this theoretical issue in SRQ
    as well - it is just much more severe in the nonSRQ case].

If connected mode works much worse than datagram in some setups,
people won't be able to enable it by default.

-- 
MST