boards/common/atmega: initialize watchdog

[ZetaR60]

This initializes the watchdog on ATmega boards if a watchdog reset occurred, preventing reboot loops when using the reboot command.

Adapted from #6837

Fixes #6836

[ZetaR60]

Note: @mali reports that this PR does not fix #6836 on some boards, for unknown reasons.

[mali]

As per the datasheet, "the application software should always clear the WDRF and the WDE control bit in the initialization routine"and interrupt should be disabled while doing this. So I suggest 'WDTCSR |= 1 << WDCE | 1 << WDE;' and starting by irq_disable() and then irq_enable() before leaving

[ZetaR60]

If I am not mistaken, interrupts are enabled for the first time only after atmega_wdog_init is being run (at the end of board_init).

[mali]

@ZetaR60 , you're right about interrupts. Any thoughts about the WDE bit ?

[ZetaR60]

@mali WDE should be cleared when all of WDTCSR is set to 0.

I have an idea: What if some versions of the Arduino bootloader are clearing MCUSR / WDRF, but not WDTCSR ? The atmega_wdog_init function would do nothing (because it does not know the board was watchdog reset), but it would still have the watchdog set to reboot again. Could you try #9141 again, but with the condition removed (so that atmega_wdog_init always clears the watchdog no matter what)?

[mali]

Good idea but sorry, same result :-/
Honestly, I'd say ... nevermind the duemilanove, at least these PR doesn't break anything, so if it fix problems on other boards I'm ok with that.

[Josar]

As there is a problem with the watchdog beeing set to the shortest time after reset in some atmega MCU the WDT has to be reseted as fast as possible. Thus in the bootloader.

https://www.microchip.com/webdoc/AVRLibcReferenceManual/group__avr__watchdog.html

But the application is then not able to find out the reset cause. As for the atmegas the WDT is used for softreset this can not be distinguished. So we implemented a solution for that in the Jiminy bootloader.

https://github.com/Josar/arduino_stk500v2/blob/da56ca393e89ef59696b7de77a5d7c03c772af82/stk500boot.c#L560

And the atmega256rfr2
https://github.com/Josar/RIOT/blob/f1e5fccc728396c4d30a075a4a838af79f90b1c6/cpu/atmega256rfr2/cpu.c#L30

And the reboot command.
https://github.com/Josar/RIOT/blob/f1e5fccc728396c4d30a075a4a838af79f90b1c6/cpu/atmega_common/periph/pm.c#L33

My proposal would be to clear the WDT in __attribute__((section(".init0")));.
This could be adapted for all atmegas like follows, With no bootloader just save MCUSR and R3 to signlize softreset, With bootloader save R2 and R3. Use the values later to distinguish reset cause and softreset.

Related arduino bootloader code shows that MCUSR gets cleared

https://github.com/arduino/ArduinoCore-avr/blob/b7c607663fecc232e598f2c0acf419ceb0b7078c/bootloaders/atmega/ATmegaBOOT_168.c#L282

https://github.com/arduino/ArduinoCore-avr/blob/b7c607663fecc232e598f2c0acf419ceb0b7078c/bootloaders/stk500v2/stk500boot.c#L559

[ZetaR60]

@Josar I am not very familiar with the section attribute and how this is modifying the startup process. Could you submit an alternate PR so that we could test the method as you intend it to be written? I think that since there is no established API for getting the reset cause it should be a secondary priority (unless a specific board needs it, like Jiminy), and the main goal should be getting all CPUs to reset properly.

[ZetaR60]

ping @Josar

[Josar]

@ZetaR60 i will have a look into it. Maybe after #9130 was merged as this will place all CPU depended stuff in one place and then the init section can be in one place for all CPUs.

[ZetaR60]

Fixed by #9866. Closing.