On Thu, Jun 09, 2005 at 08:22:11AM -0700, Tom Duffy wrote: > On Thu, 2005-06-09 at 11:47 +0200, Christoph Hellwig wrote: > > It is. B?ut I think it'll change even more later on - having a pointer > > to the method table in every object is an enormous waste of space, in > > the end the method should probably be only in a hca-level object, and > > all other objects should have public pointers to it. > > Can you please explain this a bit more? A 32 or 64 bit pointer is a > waste of space? Or are you saying that each provider should just define > a struct and have a global pointer to it?
No. What I mean is that we don't need a pointer to the method table in every single object. We don't have network driver methods in every sk_buff, struct sock and what other objects exist in the network stack. Instead the method table should be in some top-level per-HCA or per-port which most other object will have pointers to anyway (maybe through one or two indirections). This will need quite a bit of revamp of the dat codebase to move more members of the objects to the public part of it, but that should happen anyway. _______________________________________________ openib-general mailing list [email protected] http://openib.org/mailman/listinfo/openib-general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
