Re: [avr-chat] wdt_disable bug found

[Michael Hennebry]

On Mon, 20 Dec 2010, KL wrote:

> .at least it looks like that.
>
> Foreword: ATmega16A, 4 MHz, nothing special peripheral devices, just stupid
> Nokia1100 display.
>
> In my device, Watchdog disabling routine were used frequently enough (every
> second). Sometimes device was reset. (Oh, that boring "sometimes".)
> I found why.
>
> wdt_disable routine (wdt.h) works like this:
> 1) Save SREG
> 2) cli
> 3) WD_CONTROL_REG = _BV(_WD_CHANGE_BIT) | _BV(WDE)
> 4) WD_CONTROL_REG = 0
> 5) Restore SREG
>
> Looks like ok, yeah?
> But look at (3). This line not only sets WD_CHANGE and WDE bits. In addition
> it resets to zero WDT prescaler bits. This makes WD work at maximum speed.
> If WDT_RESET was not performed recently enough, line (3) will reset device
> immediately.
> Even if I would write
>         wdt_reset();
>         wdt_disable();
> I can stiil catch an interrupt right between reset and disable.
> Sure, I can write ATOMIC_BLOCK, but all this becomes too bulky.
>
> Workaround:
> a) Add wdr between lines (2) and (3)
> Or
> b) Replace (3) to something that saves prescaler settings.

Preservation isn't necessary.
Setting the prescale bits to the slowest available would also do the trick.
It has the advantage of allowing a constant.

--
Michael   address@hidden
"Pessimist: The glass is half empty.
Optimist:   The glass is half full.
Engineer:   The glass is twice as big as it needs to be."