Q about PowerPC new exception processing.

Eric Valette valette at crf.canon.fr
Fri Sep 29 15:21:50 UTC 2000


Sergei Organov wrote:
> 
> Eric Valette <valette at crf.canon.fr> writes:
> [...]
> > > Do we need to switch back to usual method of stack allocation or fix comments
> > > in the header?
> >
> > I would suggest no to switch back as interrupt stack may need to be
> > initialized much more early than what is actually done in rtems...
> 
> I thought interrupts are enabled as a result of first context load. And stack
> allocation was made before that. What for do we need interrupt stack even
> earlier? Am I missing something?

No this is correct but also quite limitative. In thory, interrupt could
be needed very eraly in the process and this can be done provided the
code executed do not call the executive. Stack allocation is this way.
> 
> What actually bothers me is the fact of initialization of interrupt stack from
> __rtems_end variable. It seems to be more useful to have separate symbol,
> let's say, __rtems_isr_stack (and maybe __rtems_isr_stack_size), that are to
> be defined by linker. Having this I can put interrupt stack to whatever memory
> region I like (e.g., internal RAM of MPC505). Does it sound reasonable?

Perfect. Note that if you are going to use the new execption scheme for
other procs, I already have the code for mbx860. I'm just finishing the
implementation of the new interrupt model for the @#$%@#$% CPM/SIU
components...


> But the new code itself doesn't depend on the way SPRG1 is initialized,
> isn't it?

Unfortunately it does depend on SPRG1. See
c/src/lib/libbsp/powerpc/shared/irq/irq_asm.S. Note that it is just
related to performance...

-- 
   __                 
  /  `                   	Eric Valette - Canon CRF
 /--   __  o _.          	Canon Development Europe Team Leader
(___, / (_(_(__         	Rue de la touche lambert
				35517 Cesson-Sevigne  Cedex
				FRANCE
Tel: +33 (0)2 99 87 68 91	Fax: +33 (0)2 99 84 11 30
E-mail: valette at crf.canon.fr	http://www.crf.canon.fr



More information about the users mailing list