esp32/machine_rtc: Preserve RTC user memory over most reset causes. - #7298

Obviously RTC memory will survive a power cycle only if the RTC domain is supplied by a battery. And pulling the EN pin causes a power cycle.

@dpgeorge: This is now almost a year old.
It's a very small change and doesn't change anything in the absence of the #define above.
Maybe you find time for a review ?

This change crashes the microcontroller for me when MICROPY_HW_RTC_USER_MEM_NOINIT is defined and RTC().memory() is accessed. I noticed that init() is not called when RTC() object is created. To avoid the crash on uninitialized memory, RTC().init(datetime) must be called first. Maybe 'MAGIC' check should be moved to new()/ojbect creation instead?

This change crashes the microcontroller for me when MICROPY_HW_RTC_USER_MEM_NOINIT is defined and RTC().memory() is accessed. I noticed that init() is not called when RTC() object is created. To avoid the crash on uninitialized memory, RTC().init(datetime) must be called first. Maybe 'MAGIC' check should be moved to new()/ojbect creation instead?

You are right. Obviously this is a bug coming to light when the user memory is not initialized.
The magic number in the memory is also not set without MICROPY_HW_RTC_USER_MEM_NOINIT, but there it doesn't cause any harm as the memory gets cleaned at startup.

This is fixed by now.
My Program to test:

import machine
import time
from machine import RTC
from machine import WDT

text = {
    machine.DEEPSLEEP_RESET: "DEEP_SLEEP_RESET",
    machine.PWRON_RESET: "PWRON_RESET",
    machine.HARD_RESET: "HARD_RESET",
    machine.WDT_RESET: "WDT_RESET",
    machine.SOFT_RESET: "SOFT_RESET"
    }
print("Reset cause: {}".format(text.get(machine.reset_cause(), "???")) )

if machine.reset_cause() == machine.PWRON_RESET:
    RTC().memory('Hello, World')
    print("set RTC.memory(): {}".format(RTC().memory()) )
    machine.reset()
if machine.reset_cause() == machine.HARD_RESET:
    print("RTC.memory(): {}".format(RTC().memory()) )
    wdt = WDT(timeout=1000)  # enable it with a timeout of 1s
if machine.reset_cause() == machine.WDT_RESET:
    print("RTC.memory(): {}".format(RTC().memory()) )

gives me the output (edited for clarity)

rst:0x1 (POWERON_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
...
Reset cause: PWRON_RESET
set RTC.memory(): b'Hello, World'
...
rst:0xc (SW_CPU_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
...
Reset cause: HARD_RESET
RTC.memory(): b'Hello, World'
MicroPython v1.18 ... 
Type "help()" for more information.
>>> E (6611) task_wdt: Task watchdog got triggered. The following tasks did not reset the watchdog in time:
....
rst:0xc (SW_CPU_RESET),boot:0x13 (SPI_FAST_FLASH_BOOT)
...
Reset cause: WDT_RESET
RTC.memory(): b'Hello, World'
MicroPython v1.18 ...
Type "help()" for more information.

@jimmo, @dpgeorge: This is now almost two years old.
It's a very small change and doesn't change anything in the absence of the #define above.
Maybe you find time for a review ?

Sorry that this didn't get any attention for so long.

It looks good! But how about just enabling this new behaviour by default? Ie no need for the new config option MICROPY_HW_RTC_USER_MEM_NOINIT, just always use the RTC_NOINIT_ATTR attribute.

But how about just enabling this new behaviour by default?

As far as I remember was the idea to allow the (compiling) user to enable this feature when required, so all applications relying on an empty RTC.memory() regardless of the reset cause will work like before.

However, in the next 2-3 daysI I could add a MICROPY_HW_RTC_USER_MEM_BOOT_CLEAN option or similar (defaulting to false), so the user can at least disable this feature.

Now we have a new (optional) #define MICROPY_HW_RTC_MEM_INIT_ALWAYS to clean the RTC memory on boot (like before this PR).

Thanks for updating. I tested this and it works well.

Merged.