[ofa-general] Further 2.6.23 merge plans...

[Or Gerlitz]

Roland Dreier wrote:
>  > IMHO, I still think that without some kind of SM/SA sourced
>  > invalidation mechanism all client side caching (including the ipoib
>  > stuff we have now) is a bad idea.
> 
> But for IPoIB at least doing a path lookup for every packet is
> obviously not feasible.  And ARP table aging gives a way to recover
> from stale cached data, eventually at least.

> In fact this may be a good argument in favor of local SA caching -- by
> analogy with IPoIB it makes sense to avoid going to the SA too often.

for each neighbour IPoIB-UD (*) keeps an IB UD Address Handle (AH), so 
the neighbouring subsystem GC mechanism which does unicast ARP probes 
etc actually --verifies-- that the cached AH is valid.

With the local SA, even though the network stack has invalidated the AH 
(neighbour), a new path query would not be initiated. If this is the 
case also with the current IPoIB code, it seems to me as a bug. Actually 
I never managed to under --why-- there's a need to keep the path (except 
for debugfs reasons) record in ipoib and not only the ah ?!

Or.

(*) for IPoIB-CM its the same idea, the neighbour points to IB 
connection and the probe is sent over the connection