[openib-general] [PATCH 00/12] ofed_1_2 - Neighbour update support

[Steve Wise]

On Fri, 2007-02-02 at 00:24 +0200, Michael S. Tsirkin wrote:
> > There's no easy way to tell who asked for notifications. And
> > particularly why they asked for notification.
> > 
> > I think we should leave it as-is.  If we have problems, we'll fix it.
> > 
> > Or you could put your arp snoop code back in addr.c and address
> > translation will not use netevents.  But still thing we should leave
> > it...
> 
> I think the issues need to be addressed in some way.
> 
> I think I see another issue with the destructor approach: ib_core could
> be unloaded while skb with destructor pointing to our code is still around.
> This will lead to nasty crashes without clear backtrace on screen if text
> segment memory gets over-written and the destructor gets called afterwards.
> 

Yes...hmm...  We could reference the module in the snoop function and
deref it in the destructor function.

> It currently seems that invoking the callback function directly rather than
> sticking it in skb->destructor is the lesser of evils at this point.
> But I'll think all this over, and I'd like to ask you to do this too,
> and post some suggestions.
> 

Ok.

> I can think of some more complicated approaches that might work better
> for iwarp. Off the top of my head, our netevents implementation could
> keep a reference on the skb, start a timer, check the users counter on skb and
> call the notifier chain when it drops to 1. Let's sleep on it.
> 

Ok.  I'll ponder it some more.  But we could solve the module unload
issue via module refs methinks.


Steve.