Re: physmem, simple containers

[Niels Möller]

Marcus Brinkmann <address@hidden> writes:

> No, it only has to map startup data/code into the new task.

Ok, that makes sense. It's just that the startup code is a part of
wortel that is very non-existant at the moment. Then one important
handle-passing thing is passing the container id for the image from
physmem to the startup code, via wortel. Should be doable.

A practical matter, is it possible to link an object file into wortel
and have it be automatically page aligned and put into it's own page?
Startup code can be a single page mapped read-only into all started
tasks. Startup data must be separate for each task, though. (Ideally,
the startup data page should somehow be returned to physmem when the
task is up and running).

> From then on the task is self-paged. Of course, the startup code may
> opt to map it all in at once, as doing paging in the task server
> might be insane.

For the time being, I'd say paging the task-server is insane. In the
longer term, I think data structures needed to keep track of a
particular task and the handles it owns should be kept on a separate
page. Then it might make sense to page out data that is associated
with particular tasks, while always keeping central task-server data
structures in core. This won't be important until we want to support a
huge number of tasks.

> Is that clearer?

Yeah, I think so. Thanks.

/Niels