Re: Extending the problems we have with caps to endpoints

[Jonathan S. Shapiro]

On Tue, 2005-10-11 at 17:23 +0200, Marcus Brinkmann wrote:
> At Tue, 11 Oct 2005 14:39:12 +0200,
> ness <address@hidden> wrote:
> > 
> > As endpoints are mappable objects in upcoming L4s, we have the same 
> > problems like with capabilities, IMHO. Maybye it's not that important, 
> > as endpoints aren't exchanged that often, but I guess we needed an 
> > endpoint server, too?
> > One more point to hope for copy semantics or to emulate 'em?
> 
> Endpoints _are_ capabilities, and general purpose capabilities in L4s
> can only be implemented reasonably using endpoints.

A small correction:

The actual endpoint itself is really a kernel-implemented queue that is
the place where the sender and the receiver rendezvous to perform the
IPC. In the "first class endpoints" designs of L4sec and Coyotos, this
is a first class resource.

The endpoint is *named* by an endpoint capability.

> Note that from what I heard the L4 people do not have fundamental
> objections against a copy operation, on the condition that it is
> "necessary".  What is needed to convince them of necessity I don't
> know :)

For a while, I had them convinced that they wanted it. Then they looked
at the required interaction with the rest of the mapping database and
decided that it would be very difficult to add to their existing system.

I have been saying for years that the right solution is to get rid of
MAP/UNMAP. For some strange reason, they have been mildly reluctant to
adopt this proposal. :-)

shap