Re: physmem, simple containers

[Niels Möller]

Marcus Brinkmann <address@hidden> writes:

> On Tue, Jan 27, 2004 at 02:14:02PM +0100, Niels Möller wrote:
> > This is under the assumption that it's
> > physmem, not wortel, that is responsible for starting the bootstrap
> > tasks.
> 
> This assumption is wrong.

Then you really have to spell out how you want to divide the work out.
I'm only getting more confused for each mail. *sigh* :-(

I first thought that wortel was responsible for starting task-server
and rootfs. Then wortel has to map the task-server image into an
address space and start a thread there. Seems reasonable, but as
wortel grants the memory image to physmem, a requirement for this is
that physmem gets the memory mapped back from physmem, so that it can
be mapped/granted further into the new address-space.

But you said that wortel will NOT EVER mmap any memory back from
physmem, which rules out the above scheme.

Second theory was that physmem is responsible for starting the tasks,
after it has got the images from wortel. That's also almost
reasonable. Coordination between tasks, like telling the rootfs where
to find the taskserver, should be fairly trivial to do for physmem. In
this case, all wortel ever needs to know about the tasks is the list
of who's allowed to talk to wortel. Main drawbach is that it seems
that physmem would need to have some knowledge about ELF-images, which
seems out-of-place.

And now you say physmem will not be responsible for starting the
bootstrap tasks.

So, how's it gonna be? Who should do what?

/Niels