Re: [avr-libc-dev] Interrupt vector redirection scheme

[Dmitry K.]

On Tuesday 04 September 2007 18:20, Michael Schulze wrote:
> Dmitry K. wrote:
> > On Monday 03 September 2007 19:59, Michael Schulze wrote:
> >> Hi,
> >>
> >>> Very interesting idea. Though there is a discrepancy: function
> >>> redir_func() keeps the SREG already after it is scratched (sbiw,
> >>> eor).
> >>
> >> I can't find the definition of that function. Where is it defined?
> >
> > Your file redirvec.S:
> >
> >    redir_func:
> >         sbiw ZL, 0                      ; test for zero
> >         ...
> >         eor r1, r1                      ; clear zero register, as assumed
> > to ...
> >         in r0, _SFR_IO_ADDR(SREG)       ; load status. SREG is a SFR
> > macro push r0                         ; save loaded status. Assumes irqs
> > are
>
> I can't see a discrepancy. I save the SREG and restore it at the end of
> the function, because of the called function could it change. However,
> the interrupt have to ensure that the state of all registers is the same
> like before.

Check, please, the interrupt trace:

      jmp TIMER1_OVR_vect

   TIMER1_OVR_vect:             ; naked fuction
      push r31
      push r30
      lds  r30, TIMER1_OVF_vect_REDIR
      lds  r31, TIMER1_OVF_vect_REDIR + 1
      jmp  redir_func

   redir_func:
      sbiw ZL, 0

So, the 'sbiw' instruction may change 5 bits of SREG before saving.
Yes?

Regards,
Dmitry.