Re: Hurd IPC

[Marcus Brinkmann]

On Sun, Oct 13, 2002 at 01:08:00PM +0200, Ludovic Courtès wrote:
> The Hurd on top of L4 will need to have its own "handle management server",
> and I guess that would be the same for other second-generation micro-kernels
> such as Fluke.

Correct. I think we can even provide lean wrappers for Mach, so we have
common server and client code.

> This means that all occurences of Mach ports and ports
> management functions in libports, glibc and the likes should be replaced by
> handles and RPCs to this server. Moreover, as L4 only supports synchronous
> message passing, it may be necessary to implement an asynchronous mechanism
> (with message queues and so on) in user-space.

We don't need asynchronous IPC for most if not all operations.  I am not
sure if that is good enough for notification messages from the server to the
client, we might consider buffering one of those per client.

> All these features are natively supported by Mach with the ports abstraction.
> However, do you think that each port to a new microkernel should lead to the
> creation of a new branch of the Hurd libs? Or do you think that once this
> server and the related mechanisms are implemented for L4 we should also try to
> implement them on top of Mach so that the Hurd remains the same, whatever the
> underlying kernel is?

Yes, there will be a library for objects and object handles.  The Hurd code
will be revised to only make use and provide with interfaces the minimal
features needed.

> I guess implementing such IPC mechanisms on top of Mach, i.e. using Mach
> ports, would be very unefficient

There is no reason for it to be much more inefficient than it is now. 
Except maybe in the I/O path, I am not sure about how strong the influence
of changes in the data buffer interface will be.

Thanks,
Marcus

-- 
`Rhubarb is no Egyptian god.' GNU      http://www.gnu.org    address@hidden
Marcus Brinkmann              The Hurd http://www.gnu.org/software/hurd/
address@hidden
http://www.marcus-brinkmann.de/