Reynolds, Terry (Contractor-SIMTECH) wrote:
Hi All,
I apologize for the long posting, but wanted to give sufficient detail for folks to see what's going on.
I have made a good bit of progress on the ppc64 port. I'm trying to get the RTAI example test latency to work.
Things fail in a way that looks like I've missed a function somewhere. I've added quite a few printk statements
in trying to track down the problem.
What I'm seeing is:
Root domain registered.
Pipeline started.
Start latency test:
RTAI[hal] domain registered:
switch from root to RTAI[hal].
calls suspend.
switches back to root.
Domain IShield registered.
root switches to IShield
IShield suspends & switches back to root.
RTAI[nucleus] RTAI/fusion v0.6.9 started
all initialization routines seem to run OK.
__fusion_skin_init processes all the way through and exits without error.
THEN
In Root:
sched.c's need_resched: calls -
adeos_schedule_tail, which calls __adeos_handle_event with
ADEOS_SCHEDULE_TAIL
__adeos_handle_event does a switch_to RTAI[hal] domain.
it resumes in suspend_domain, from which it returns to rthal_domain_entry.
At that point it is in a loop constantly calling suspend_domain.
So, RTAI[hal] instantly suspends again, switching back to root.
Root ends up back in need_resced, and this process repeats several hundred times,
until the latency test gives up and exits with a "failed to create display task,
code -38"
IShield is unregistered.
RTAI/fusion stops.
RTAI is unregistered.
RTAI[hal] is unloaded.
At least things seem to exit gracefully!
So, my questions are:
Did any of that description look totally wrong?
The behaviour described above looks sane.
What is suppose to cause RTAI[hal] to break out of the suspend_domain loop,
and process the event being passed in from root via the schedule_tail?
Should it be suspended at a different place?
adeos_suspend_domain() looks for events to process (traps & IRQs) before
returning to the caller. So the proper callbacks for the current work
cycle have already been fired when adeos_suspend_domain() returns back
to the calling statement in the RTHAL entry point. This is why an Adeos
work cycle usually looks like an endless loop calling
adeos_suspend_domain().
I'm not sure if this is an Adeos or an RTAI problem at this point.
Everything looks ok so far.
TIA!
Terry.
_______________________________________________
Adeos-main mailing list
[email protected]
https://mail.gna.org/listinfo/adeos-main
--
Philippe.
