Re: [U]VM subsystem for L4 X.2 (and later)

[Farid Hajji]

Hi Ian,

> I think the first goal should be to just get something that works
> implemented. We could refine things after that. You mentioned before
> that the OSKIT had a lib that utilized the UVM. The first step should be
> to create a memserver that uses the simple VM in OSKIT, or if the author
> is ambitious, jump right to the OSKIT UVM. That would allow us to better
> understand what the mem server's interfaces need to look like. It would
> allow us to do some prototyping before building the real UVM memserver.

OSKIT 2002-03-17 finally shipped with uvm and uvm glue code. I didn't
have the time yet to examine the OSKIT UVM in detail, but it looks
very promising at a first glance.

Creating a memserver with OSKIT UVM would be an interesting exercise
indeed. IMO we should not was too much time doing a SLAB based memserver
if OSKIT UVM interfaces are cleanly cut. But, as said, I'm not yet
familiar with the OSKIT UVM variant at the moment (just with NetBSD
-CURRENT's variant).

> > As far as I understand the X.0 and new X.2 specs, a single thread can
> > be used for all other threads in the system. I see no problems with
> > this. The architecture would be:
> > 
> >     Hurd/L4 threads        NetBSD/L4 task
> >           ^                   ^
> >           |   .............../
> >           |  /
> >           v  v
> >        vmserver
> >           ^
> >           |
> >           v
> >         pmap/l4
> >          ^  ^
> >         /    \
> >        v      v
> >      DMphys  L4 X.2
> >       on      API
> >      X.0      on
> >             X.2/V4
> > 
> > Perhaps, perhaps not.
> 
> Nice art. I just wonder how the heuristics in the VM will be affected
> when it has two different OSes driving it rather than one. It will be
> interesting to see.

Yes, that is one of the main motivations: using UVM clients with
probably widely differing VM needs at the same time ;)

> Like I said, I think we should focus on building the simple thing first,
> and then work on refining it. I'm putting my effort into getting a set
> of device drivers working for the system, looking at the DDE stuff and
> possibly OSKIT as a first step. If someone else wanted to work on the
> memserver that would be great. If no one has picked it up when I finish
> the driver stuff, then I'll take a whack at it. It's just always nice to
> parallelize effort.

ACK.

Regards,

-Farid.

-- 
Farid Hajji -- Unix Systems and Network Admin | Phone: +49-2131-67-555
Broicherdorfstr. 83, D-41564 Kaarst, Germany  | address@hidden
- - - - - - - - - - - - - - - - - - - - - - - + - - - - - - - - - - - -
One OS To Rule Them All And In The Darkness Bind Them... --Bill Gates.