Low Level Init

Hi Jeff,

That may be the exact problem!  Actually I have changed microcontrollers and upgraded IAR from a while back,  On the initial design, I used:

void __low_level_init(void) {
WDTCTL = WDTPW + WDTHOLD; // Stop WDT (Watch Dog Timer)
}

As far as I remember, this worked fine in my original design.  However, when I tried to use this in my new design, the compiler gave me an error, saying the function should be: int __low_level_init(void)

I then changed the function to:

int __low_level_init(void) {
WDTCTL = WDTPW + WDTHOLD; // Stop WDT (Watch Dog Timer)
}

Notice how there is no return value.  I got no error from the compiler.  I believe it should really be: 

int __low_level_init(void) {
WDTCTL = WDTPW + WDTHOLD; // Stop WDT (Watch Dog Timer)
return 1;
}

I tried this, and there appears to be no problems (at least for now).  Would this error explain the incorrect variable initialization?  Why was it that it seemed to happen only sometimes?

Thanks for the advice!

Sincerely,

Mehdi

I think you need to find our the meaning of the returned integer. It might direct the c startup to do different things.

Mehdi Rahman said:

Notice how there is no return value.  I got no error from the compiler.  I believe it should really be: 

Jens-Michael Gross said:

It's no error to return nothing. But you should get a warning form the compiler.

Do you think this would cause the problem of improper variable initialization, or do you think the source of that problem is something else?

Mehdi Rahman said:

Would this error explain the incorrect variable initialization?  Why was it that it seemed to happen only sometimes?

#include "io430.h"

volatile static unsigned int temp = 164;

int __low_level_init(void) {
   WDTCTL = WDTPW + WDTHOLD; // Stop WDT (Watch Dog Timer)
}

 

int main( void )
{
  if (temp == 164)
  {
     // Turn on LED
     PJOUT |= BIT3;
     PJDIR |= BIT3;
  }
  temp++;
 
  return 0;
}

When the program was run with the debugger not attached, the LED wasn't lit, indicating the temp variable was no longer being initialized correctly.

Making the __low_level_init function return the value one then caused the temp variable to be initialized correctly both when the program was run in debugger or standalone.

i.e. the missing return from __low_level_init does explain why the variable initialization sometimes failed.

P.S. I noted the MSP430 FAQ had __low_level_init (IAR) and _system_pre_init (CCS) as a void return type. I have corrected the page.

Mehdi Rahman said:

It's no error to return nothing. But you should get a warning form the compiler.

Do you think this would cause the problem of improper variable initialization, or do you think the source of that problem is something else?[/quote]
Well, being no error from the ompiler side doesn't mean it is no error from teh application side. If the calling funciton expects a certain return code, returning a random value in the return value register isn't a good idea. And in this case, it seems a return value of '1' is expected to perform further initialization.

I've got a friend who ignored compiler warnigns about suspicious pinter conversions, saying that they were no errors so all should be fine. However, while the code was valid C code, some of his pointer conversions were simply wrong, leading to memory leaks that were difficult to track down. 'legal' doesn't always mean 'correct' :)

Hi Mehdi,

You have found the problem, which is that you must be sure to return 1 if you want cstartup to continue with variable initializations.

Also you should #include <intrinsics.h> since the proper function declaration for __low_level_init() is in that file.

Jeff

Hey everyone,

Thanks for the help!

Looks like the TI Software coding techniques application notes need an update as well (it uses the void declaration of the low level init): http://www.ti.com/lit/an/slaa294a/slaa294a.pdf

Mehdi

Mehdi Rahman said:

Looks like the TI Software coding techniques application notes need an update as well (it uses the void declaration of the low level init)

Jens-Michael Gross said:

Well, the low-level init funcitonality and how to use it is purely compiler/linker specific. And IAR is no TI compiler. So why should TI change their documents when a 3rd party product changes its proprietary nehaviour?

Good point.

I know the post is old but I stumbled upon it and so may people after me.

Mehdi Rahman said:

Hello,

I am in a situation where I have to initialize a lookup table that is large enough that under normal conditions, the watchdog timer expires before the table can finish loading. (...)

Supposing your lookup table is not a fixed one but changes during runtime, you could have used __no_init  instead, and initialized the table after stopping the watchdog, as in:

const int LookUpTableInit[] = {init values};  // Array in FLASH

__no_init int LookUpTable[sizeof(LookUpTableInit)/sizeof(LookUpTableInit[0])];// Uninitialized array in RAM

int main (void)

{

  WDTCTL = WDTPW | WDTHOLD;

  for (int i=0; i<sizeof(LookUpTableInit)/sizeof(LookUpTableInit[0]); i++)

    LookUpTable[i] = LookUpTableInit[i];// Initialize RAM array with FLASH array values

}

Regards,

Michael K.