> For scheduler activations, a way for the kernel to arrange to call a
> specific userlevel function is needed, as well as passing it
> arguments.

As Bill Sommerfeld suggests, this will be done by signal.

> I suspect that once the *context() stuff is done, we can use that.

I don't think so. *context() functions are developed for legacy
user/kernel mixed thread, and not appropriate to scheduler activations.

> The application would set up the context, and the kernel can clone it
> and restore it when it wants to run again.

Mmm, what does "the application would set up the context" mean?
On scheduler activations, userland don't have to setup a context,
it just can use system call.
--
soda