[ofa-general] [RFC v2 PATCH 4/5] rdma/cma: implement RDMA_ALIGN_WITH_NETDEVICE ha mode

[Steve Wise]

Or Gerlitz wrote:
> Sean Hefty wrote:
>>> So instead of adding a new function rdma_set_high_availability_mode, 
>>> you
>>> could just set an option saying WANT_NETDEV_CHANGE_EVENTS.  Maybe we
>>> need to add rdma_set_option() to the kernel RDMA-CM API?
>> I agree with this.  Having a generic mechanism to report rare events 
>> would be
>> useful.  Maybe the device removal notification can be adapted for 
>> this purpose?
> Sean, as suggested in the past (eg over the QoS discussion) 
> rdma_set_option can serve from more purposes similarly to setsockopt, 
> and I guess that down the road as the RDMA stack would get enhanced by 
> more features adding these rdma_set/get_opt calls would make sense. So 
> (Steve) in that respect, I don't see rdma_set_opt as a mechanism to 
> report rare events.
>
I don't understand your rationale above on why adding a new function is 
better than using an extensive "set this option" function.  Can you 
clarify? 

Function rdma_set_option() wouldn't be the mechanism to report the 
events.  It would be the mechanism to set an option indicating the cm_id 
wants NETDEV_CHANGE events.  Seems like the exact fit rather than a new 
API call to set some very specific mode or option.  I missed the QoS 
thread you're referencing, so please excuse me if I'm rehashing 
something that has been agreed-to in the past...


Steve.