[ofa-general] [PATCH 2/4] ipoib: fix loss of connectivity after bonding failover on both sides

[Yossi Etigin]

I think it is safe in this case. The only interesting case
is if the correct value just randomly appears there, before
path record query completed successfully. This means that the
test can "pass" when it shouldn't only when neigh->ah is NULL.
This reverts to the situation before the patch, when neigh->ah != NULL
was also needed to perform the "neighbour refresh" stuff.
On the other hand, the patch intends to fix only a situation
when neigh->dgid is already initialized by a successful path query.

Roland Dreier wrote:
>  >  > also
>  >  > initialize neigh->dgid.raw to have value to compare with.
> 
>  > That was removed from the patch because Moni Shoua found it had
>  > increased the traffic renewal time in case of SM failover. I forgot to
>  > remove it from the
>  > changelog as well.
> 
> So does that mean the patch compares against a possibly uninitialized
> gid value?  Is that always safe?
> 
>  - R.

-- 
--Yossi