Strange behaviour of incrementing variable in "noinit" section while desetting debugger

[HuEl_264296]

I am trying to use a variable in the "noinit" memory section. I am testing it with some very simple code:

#include "project.h"
CY_NOINIT int myVar;

int main(void)
{
    CY_NOP;
    myVar++;
    for(;;) {}
}

I then try to use that code in the debugger. However, I see a very strange effect when I click the Reset button:

Note that I never click the Play button. I just click Reset again and again.

When I first start the debugger, I see this:

There's the code, and the CY_NOP is highlighted in yellow, meaning that line hasn't been executed yet.

Then I don't click Play.

But I just click Reset again, and this is what I see:

Look at that, myVar has increased in value by 4! Each time I click Reset, the value increases by 4. This is strange for two reasons:

1.  No code should have been executed yet, as I haven't clicked Play.
2. There's nowhere in the code that adds 4 to the variable.

What is going on here?

Here's some more evidence:

• If I step through the code and execute the line myVar++; then I see the variable increase by 1.
• If I change the type of myVar to char, it still increases by 4.

[HuEl_264296]

After some more investigation, it looks like the Reset command of the debugger actually causes the firmware to execute 4 times, each time for about 2.7ms, before eventually getting it to stop back at the beginning of the code:

I tested some new firmware:

#include "project.h"

CY_NOINIT int myVar;

int main(void)
{
    CY_NOP;
    myVar++;
    
    while(1)
    {
        LED_Write(1);
        CyDelayUs(3);
        LED_Write(0);
        CyDelayUs(3);
    }
        
    for(;;) {}
}

Then I used my logic analyser to observe the GPIO pin. Here we can see 4 groups of pulse trains:

Looking more closely, it looks like each one lasts for about 2.7ms.

Is this the expected behaviour of the debugger?

[HuEl_264296]

I can see it's actually pulsing the reset line:

Is it supposed to do that?

BTW, I'm using the MiniProg4.

[DennisS_46 · Accepted solution]

The chip runs a lot of code in the boot process, this includes un-packing a bit-stream that defines the initial state of ALL of the registers and assigned memory locations (but not NOINIT section), exercising the Reset line and lots of other stuff. Who knows what memory locations can get over-written in the process? It's even messier in de-bug mode. What you are seeing is non un-expected behavior. If you need to save myvar so that it doesn't get randomly over-written at start, save it in Flash every time you use it so the next time you need it at start-up it will have your known value.
---- Dennis Seguine, PSoC Applications Engineer (with help from our software development team)

[HuEl_264296]

Hi Dennis,

Thanks for your reply. Two things:

1a. Whatever else happens in the boot process, the NOINIT memory shouldn't be overwritten. That is after all the whole point of the NOINIT memory. If you're telling me that maybe the NOINIT memory gets overwritten, and maybe it doesn't, "who knows?", then essentially the NOINIT memory is 100% useless.

1b. I'm pretty sure it's not 100% useless as the Cypress library code actually uses it (e.g. in CyLib.c).

2. If you read my replies, you'll see that I worked out what was happening. When I click "Reset" in the debugger, it releases the reset line 4 times, each for about 2.5ms. This allows the first 2.5ms of code to execute before the chip is held in reset again.

This is very surprising behaviour. I would not expect any code at all to be executed when I click Reset in the debugger. I would expect that code would only be executed when I clicked Play.

In my test, I was incrementing a NOINIT variable near the beginning of my code, and so I was seeing it incremented 4 times each time I clicked "Reset".

Kind regards

Hugo