[openib-general] CM header file

Libor Michalek libor at topspin.com
Wed Dec 15 17:48:14 PST 2004


On Wed, Dec 15, 2004 at 04:25:23PM -0800, Sean Hefty wrote:
> Listed below is a start at an updated CM API.  Several details are 
> missing, but there should be enough there to evaluate the direction of 
> the CM.  Some design notes:
> Questions:
> 
> * Connection identifiers cannot be destroyed from within a CM callback. 
>   I think that we can support this by adding a flag (destroying within 
>   callback) to the destroy call.  Is this worth adding?

  The other option is to destroy the connection if the consumer returns
an error value from the callback. Along the same lines, will the 
consumer be allowed to call the corresponding response function from
a callback? (e.g. ib_send_cm_rep() from the REQ callback) If not then
the same behaviour could also be achieved with a callback return value.

> * Should listening clients call an "accept" routine to wait for a 
>   connection request?  Currently, the API operates asynchronously and 
>   inokves a CM event handler.

  I don't think this is necessary. Presumably the biggest reason to use
"accept" is to force the consumer to use its own thread to handle CM state
changes, thus avoiding CM or MAD thread deadlock if there is a buggy
consumer, or is there another reason to add this step?

   However, a nice to have feature which I've grown use to is the ability
to listen to an entire range of service IDs using a value/mask combo.

  Thanks for taking this on, once you have something minimal I'll be happy
to try it out, I've been waiting. :) 


-Libor










More information about the general mailing list