atmega328p/startup.c: Clear watchdog flag

[zqad]

When the atmega328p is restarted by the watchdog, the WDRF bit in
the MCUSR register will also set the WDE bit, enabling the
watchdog. Since we use the watchdog for reboot (see
cpu/atmega_common/periph/pm.c), this means that the CPU will just
loop-reboot when a reboot is issued.

From the Atmel documentation about WDE in WDTCSR:
"WDE is overridden by WDRF in MCUSR. This means that WDE is
always set when WDRF is set. To clear WDE, WDRF must be cleared
first. This feature ensures multiple resets during conditions
causing failure, and a safe start-up after the failure."

[mali]

I need to investigate more, because I can't reproduce the issue described in #6836.

[zqad]

That is strange, I'm still seeing this. And the documentation seems to indicate that the patch is needed. Might it be a revision thing?
For full disclosure: I'm using a 5V Arduino Mini Pro which is as similar to the Uno as they come, at least processor-wise.

[zqad]

..one thing just hit me. Are you running it with or without the Arduino bootloader? I'm running without it, so it might be that it clears the bit if the board was reset using the watchdog.
I'm not sure how RIOT works when running in "Arduino-mode", if it uses the bootloader or not..

[mali]

@zqad : Yes I'm using arduino bootloader, indeed, that should be related.
thank's for mentionning this.

[mali]

For now I'm unable to build Riot due to obsolete gcc toolchain. I'm waiting for the new debian stable which should be there soon.

[mali]

Ok, so here are the results of my tests.

• arduino uno : no problem found, with or without the patch.
• arduino duemilanove : problem is here but is not corrected by this patch, the board is bricked in the two cases and go in infernal loop-reboot. Using another board as ISP is needed to recover.

[mali]

Got something to track down, I've found this piece of code in arduino bootloader which can explain why the board go in loop-reboot

#ifdef WATCHDOG_MODS
ch = MCUSR;
MCUSR = 0;

WDTCSR |= _BV(WDCE) | _BV(WDE);
WDTCSR = 0;

// Check if the WDT was used to reset, in which case we dont bootload and skip
straight to the code. woot.

if (! (ch & _BV(EXTRF))) // if its a not an external reset...
app_start(); // skip bootloader
#else
asm volatile("nop\n\t");
#endif

But this said, I should have had hand on board while pressing reset button which was not the case ...

[kaspar030]

any news on this?

[ZetaR60]

The ATmega1284P has this reboot-loop issue as well. Tested on BOARD=mega-xplained. Placing the watchdog reset code at the beginning of board_init in boards/common/atmega/board.c fixes the reboot-loops.

[ZetaR60]

Other ATmegas have the reboot-loop issue. I suggest moving this to boards/common/atmega/board.c

[ZetaR60]

Ping @zqad

[ZetaR60]

Updated PR that fixes all ATmega boards: #9141

[PeterKietzmann]

@ZetaR60 can this one be closed in favor of #9141?

[ZetaR60]

@PeterKietzmann It should be okay to close. #9141 is not perfect (does not work on the Duemilanove, for unknown reasons), but it is a more comprehensive fix than this PR. Closing.