[openib-general] ib_gid_is_link_local

[Hal Rosenstock]

On Thu, 2007-01-04 at 13:42, Jason Gunthorpe wrote:
> On Thu, Jan 04, 2007 at 07:04:19AM -0500, Hal Rosenstock wrote:
> 
> > > I think it depends how you code it.. In IP-like terms you have two
> > > processes going on here, a route lookup and a 'neighbour lookup' (GID
> > > to LID/etc conversion) for the link address.
> > > 
> > > An non-existing link local GID should pass the routing lookup, but fail the
> > > neighbour lookup since there is no LID.
> > > 
> > > So typically you'd do the routing lookup and end up with an on-link
> > > GID. Then you'd do the neighbour lookup and get a link address (IB path).
> > 
> > I think you are talking here about the end node and router function
> > rather than SM/SA, right ?
> 
> The way I was hoping to start out is by putting this in the SA and the
> routers, not in the end nodes.

We can start there but this is a very fundamental question. I have heard
people weigh in on both sides...

> > And in terms of IB path lookup, is the path just local with a
> > destination of the neighbor (router) or global to the end point ?
> 
> With this kind of model the IB path lookup would return a LID/SL/etc

Map S/DGID and perhaps TClass to LID/SL/MTU ?

> so it is a local path but might terminate on a router and might
> require a GRH so it can be forwarded.

depending on whether the DGID is on or off link.

> > > As an initial start I'd like to see support for an entry like this:
> > > 
> > > default via fe80::17:77ff:feff:5
> > > 
> > > Where fe80::17:77ff:feff:5 is the GID of a router. A full table for a
> > > routing environment might look like:
> > >  
> > > fe80::/64 onlink
> > > fe80::/10 unreachable ; IBA requires link local scope addresses never
> > >                       ; to be routed
> > > ff00::/8 onlink       ; Multicast is 'on link'
> > 
> > Why is all multicast scoped onlink ? I guess this is a starting point
> > but not what IPoIB needs to run across IB subnets.
> 
> I'm thinking to start it would be OK to just force routers to be
> subscribed to all MLIDs 

all known MLIDs on that subnet rather than all possible MLIDs ? It's
really the MGIDs that are of interest rather than the MLIDs. The router
needs to subscribe to traps 66/67 multicast groups in and out of
existence. MLIDs on each side of the router may not be the same for a
non link local MGID.

> until we have a better idea how this needs to
> work. This would work for IPoIB and is why I put it as onlink.

By onlink, are you saying these wouldn't be forwarded ?

> If the SM is doing the inter-subnet spanning tree then it probably
> needs more routing entries.. 

Are you referring to running a spanning tree for multicast ? In any
case, I think it will be a while before the routing protocols come into
the picture and whether the SM is involved or not is another piece of
some of the fundamental routing questions/devisions to be made.

> If it is staying out of the picture then
> this is probably enough.

I think it will be out of the picture for a while.

-- Hal

> Jason