[ofa-general] Re: Re: [PATCH V3 0/7] net/bonding: ADD IPoIB support for the bonding driver

[Michael S. Tsirkin]

> Quoting Moni Shoua <monisonlists at gmail.com>:
> Subject: Re: Re: [PATCH V3 0/7] net/bonding: ADD IPoIB support for?the bonding driver
> 
> Roland Dreier wrote:
> >  > 1. When bonding enslaves an IPoIB device the bonding neighbor holds a 
> >  > reference to a cleanup function in the IPoIB drives. This makes it unsafe to 
> >  > unload the IPoIB module if there are bonding neighbors in the air. So, to 
> >  > avoid this race one must unload bonding before unloading IPoIB. 
> > 
> > I think we really want to resolve this somehow.  Getting an oops by
> > doing "modprobe -r ipoib" isn't that friendly.
> > 
> You are right and we want to resolve that.
> One way is to clean the neigh destructor function from all IPoIB neighs.
> The other way is to prevent ipoib unload if device is a slave or is referenced from 
> somewhere else.
> 
> I guess I would like an advice here.

I had this idea:


Maybe we could use hard_header_cache/header_cache_update methods instead of
neighbour cleanup calls.

To do this, it is possible that we'll have to switch from storing pointers
inside the neighbour to keeping an index there, but I expect the
performance impact to be minimal.

As a result, bonding would not have to copy pointers into ipoib module
and module removal would get fixed.

-- 
MST