Re: Task destruction

[Niels Möller]

Marcus Brinkmann <address@hidden> writes:

> I don't see any problem, can you be specific about what problem you are
> thinking here?  I assume a proxy task server used for the child Hurd,
> where the user of the child Hurd has privilege to register parent-child
> relationships.

Using a proxy task server solves the problem, I think. Having a single
shared task server somehow appeals to me, and I don't think it has to
be overly complex, but perhaps you get a simpler system if you instead
have a several of them.

> Actually, I think that our disagreement is more about the nature of the
> task server we are talking about.  There is only one task server we must
> have, and that is the one sitting directly on L4.  And I am somewhat concerned
> about overloading it with semantics like Hurdish IPC.

To me, the requirements for an L4 task server are quite similar to
those for a Hurd task server. An L4 task server still needs to
associate some "master" with each task (which could be the Hurd proc
server or some l4-linux singleserver, for example), which is
inherited, it needs to send death notifications, and it needs some
magic integer that is inherited. That's not much different from having
<master, id> handle that is inherited. The main difference is that it
gets away with a single handle.

I think I'm also I'm assuming that ordinary hurd tasks can talk to the
task server and create new tasks (unless they hit a limit or
something), as if there were an unprivileged system call that creates
a task and does the needed bookkeeping. If that's true for the hurd
task server, but not for the L4 one, then maybe that gives some more
opportunities for simplifying the L4 task server.

Regards,
/Niels