Re: scheduler syscall requirements

[Al ]
> Hi.
> I am interested in scheduling for L4 so I had a look at the source
> of Scheduler syscall in Pistachio and wondered how the scheduler and
> its scheduled thread necessarily should be in the same address
> space.

Wrong.  The scheduler thread of the destination and the current thread
must reside in the same addres space, i.e., the current thread must be
a thread in the "scheduler server" of the destination.

> pistachio/kernel/src/api/v4/schedule.cc

> 609     // are we in the same address space as the scheduler of the thread?
> 610     tcb_t * sched_tcb =
> get_current_space()->get_tcb(dest_tcb->get_scheduler());
> 611     if (sched_tcb->get_global_id() != dest_tcb->get_scheduler() ||
> 612          sched_tcb->get_space() != get_current_space())
> 613     {
> 614          get_current_tcb ()->set_error_code (ENO_PRIVILEGE);
> 615          return_schedule(0, 0);
> 616     }

> For me it restricts the opportunities for scheduling policy and the
> manageability for program systems on top of L4.

It does not restrict any oppoprtunities.  It just more easily allows
for schedulers to be multi-threaded; for example, in an MP system you
might have one scheduler thread per CPU.

> Also nothing is mentioned about these requirements in "L4 kernel
> reference manual".

Thought this had already been updated in the refman.  Have updated it
to be more clear on this point now.

> Is there any reasonable explanation for this?

> BTW, there is no checking that Scheduler syscall invoked by thread
> registered as scheduler for destination thread, and proposed
> priority less than current thread's. It seems easy to fix, so maybe
> it provided deviation?

Right.  The test seems to be missing from the implementation.

	eSk