[ofa-general] [PATCH] IB/core: handle race between elements in qork queues after event

[Moni Shoua]

Roland Dreier wrote:
>  > Can we please go on with this patch? We would like to see it in the next kernel.
> 
> I still don't get why this is important to you.  Is there a concrete
> example of a situation where this actually makes a measurable difference?
> 
> We need some justification for adding this locking complexity beyond "it
> doesn't hurt."  (And also of course we need it fixed so there aren't races)
> 
>  - R.
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
> 
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
> 
Hi,
OK. Here is an example that was viewed in our tests.
One IPoIB host (client) sends a stream of multicast packets to another IPoIB host (server).
SM takeover event takes place during traffic and as a result multicast info is flushed
and there is a need to rejoin by hosts. Without the patch there is a chance (which according to our experience
is a very big chance) that the request to rejoin will be to the old  SM and only after a retry join completes successfully.
This takes too long and the patch solves it.

I hope that this is convincing enough for you because for us it is important that a
recovery from a failure will be as quick as possible.

thanks

 MoniS