[ofa-general] RE: [RFC v2 PATCH 3/5] rdma/cma: add high availability mode attribute to IDs

[Or Gerlitz]

Steve Wise wrote:
> Sean Hefty wrote:
>> I think we should just always report this event, and let users ignore 
>> it if they
>> want.  We don't seem to gain much by filtering the event at a lower 
>> level.  
> Um, doesn't that then change the ABI?  Some apps might hurl on a new 
> (unexpected) event.
Steve, I think ULPs should be designed/coded to live well with new 
events delivered by the rdma-cm as (A) the model is event based and (B) 
such events can be introduced while developing new features... So my 
suggestion is that a ULP which is limited in that sense would have to 
state in its package dependency requirements that they are dependent on 
librdmacm 1.0.7 or earlier. Ofcourse, the kernel is one package so I 
will make sure that the current intree rdma-cm consumers (iser, rnfs) 
live well with this event.

Sean, please let me know your preference (as it was somehow unclear from 
the thread) if you want the delivery of this event to be dependent on 
the ulp asking for it or no.


Or.