C-func call in ISR seems to cause stack corruption

Currently functional code:

New code snippets:

;  Define instructions executed on a hardware timer interrupt.
HW_TIMER_CODE            MACRO
;  start of user code to update soft timers via C-func
        EXTRN   CODE (TIMER_Update)
        PUSH    ACC
        LCALL   TIMER_Update
        POP     ACC
        RETI
        ENDM

and the C-func:

void TIMER_Update( void ) using 2 
{   unsigned char data i;

    for( i=0; i<NUM_TIMERS; i++ )
    {   timers[i]--;
        if( timers[i] == 0 )
        {   timers[i] = reload[i];
            if( events[i] < 255 )
                events[i]++;
        }
    }
}

There still seems to be a discrepancy whether or not it would be better (more efficient) to wrap a full context switch in the assembly code (PUSH ACC, PSW, .. POP AR7, AR6, ...) around the call to the C_func rather than using a different register bank.

This is code that is running under RTS51 Tiny so this exercise will use Reg Bank 0 (in general) Reg Bank 1 (for the timer interrupt) and then extend the timer interrupt to use Reg Bank 2 just for this C-func.

You might need to also add USING 2 to the TIMER_Update( ) function, or omit the register bank part in the ISR vector routine.

Maybe I should have said "exclusive or" (XOR).

My point was to ensure that both code sections are identical. If you are going to use the "USING" command, do it in both places. Otherwise don't use the "USING" directive in either place (maybe).

Not that I think that this is your problem, it is just a risk area I saw in your code.

The Keil compiler does not know if one module has a different 'USING' value than another module. When you have a file "Timer.C" and another file "Interrupts.C" and one of them uses a non-standard "USING" directive, then you could have a problem.

Case in point:

/*
** Timer.C
*/
int Get_Timer1( void ) using 3
{
    return( TimerCounter ); // 'TimerCounter' maintained elsewhere.
}

And then your interrupt service vector table is in another location

/*
** ISR_Vectors.C
*/
void Ext1_ISR( void ) interrupt EXTERNAL1_ISR
{
    ...
    val = Get_Timer1( );
    if( val > LAUNCH_ICBM_DELAY_TIME )
    {
        Launch_Arsenal( ALL );
    }
    ...
}

When T1_ISR launches Timer1( ), there MAY be a mismatch in the register bank.

Again, I'm not saying that this is your problem. Just a risk area.

Are you sure that the uPSD3454 is configured properly? especially with respect to the memory mapping (PSD).

--Cpt. Vince Foster
2nd Cannon Place
Fort Marcy Park, VA

There is missing PUSH / POP PSW in your interrupt header, I think. And if your "C_funct" uses DPTR, you must add PUSH / POP DPL,DPH. And if your function uses dual DPTR - you must PUSH both DPL/DPH ... I prefer interrupt handler in "C" style. The compiler can decide what is necessary to PUSH to stack in this moment.

Good call Martin Persich, I totally agree. That has got to be it.

I just saw the who MACRO thing and thought there was some missing context around it. (I should have known. Especially since RETI is right there).

(I hate MACROs)

--Cpt. Vince Foster
2nd Cannon Place
Fort Marcy Park, VA

By using the declaration

void TIMER_Update( void ) using 2
{ ... }

C51 is intelligent enough to generate

    PUSH    PSW
    MOV     PSW,#010H
    ...
    POP     PSW
    RET

as a wrapper around the C-func.

RTX51 Tiny is sharing RegBank 0, but the RTOS timer interrupt is using RegBank 1. Therefore, the declaration in the C-func of using 2 was specified to let the compiler keep track of the used / unused registers of RegBank 2.

All appears to be working correctly now.
Thanks for the insights.

(I agree with both "I hate macros. All of them." and "I prefer interrupt handler in "C" style."!)