RP2040 will lockup if a watchdog timeout occurs while lightsleep() is running
Port, board and/or hardware
Raspberry PI Pico
MicroPython version
MicroPython v1.25.0 on 2025-04-15; Raspberry Pi Pico with RP2040
Reproduction
Test file named wdt_lightsleep.py.
from machine import lightsleep, WDT
wdt = WDT(timeout=3000)
lightsleep(5000)
Do 'mpremote run wdt_lightsleep.py'.
>> mpremote run wdt_lightsleep.py
Traceback (most recent call last):
File "/home/picompute/.local/bin/mpremote", line 8, in <module>
sys.exit(main())
^^^^^^
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/mpremote/main.py", line 569, in main
handler_func(state, args)
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/mpremote/commands.py", line 463, in do_run
_do_execbuffer(state, buf, args.follow)
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/mpremote/commands.py", line 437, in _do_execbuffer
ret, ret_err = state.transport.follow(timeout=None, data_consumer=stdout_write_bytes)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/mpremote/transport_serial.py", line 184, in follow
data = self.read_until(1, b"\x04", timeout=timeout, data_consumer=data_consumer)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/mpremote/transport_serial.py", line 123, in read_until
elif self.serial.inWaiting() > 0:
^^^^^^^^^^^^^^^^^^^^^^^
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/serial/serialutil.py", line 594, in inWaiting
return self.in_waiting
^^^^^^^^^^^^^^^
File "/home/picompute/.local/pipx/venvs/mpremote/lib/python3.11/site-packages/serial/serialposix.py", line 549, in in_waiting
s = fcntl.ioctl(self.fd, TIOCINQ, TIOCM_zero_str)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
OSError: [Errno 5] Input/output error
>> mpremote
mpremote: no device found
Expected behaviour
Expect it to behave the same as if lightsleep were replaced with time.sleep_ms.
Call this test file wdt_sleep_ms.py
from machine import WDT
from time import sleep_ms
wdt = WDT(timeout=3000)
sleep_ms(5000)
In this case, after do 'mpremote run wdt_sleep_ms.py'.
After 3 seconds mpremote will report an OSError because the RP2040 does a soft reset.
Run mpremote again and it will connect successfully.
Observed behaviour
The RP2040 seems to get stuck. After running 'mpremote run wdt_lightsleep.py' attempts to connect running mpremote fail.
On my Raspberry Pi 4 I see an error message of 'mpremote: no device found'.
A power cycle is required to recover.
One more observation, I have a USB power meter connected between the PI 4 and RPI Pico.
Normally it reports about 19 milli-amps consumed, about 5 milli-amps while the lightsleep is running.
After the test completes it reports about 1.00 milli-amps.
Additional Information
I have changes to lightsleep in modmachine.c which fix this problem.
I will prepare a pull request with this fix.
The root cause is lightsleep turns off the ROSC.
When the watchdog timeout occurs the Power-On State Machine does not re-initialize it.
This occurs because the PICO-SDK function _watchdog_enable() zero's sm_hw->wdsel bits for the ROSC and XOSC.
The behavior reported here is seen on RP2040 only.
Code of Conduct
Yes, I agree
ports/rp2/modmachine.c: Fix lightsleep watchdog interactions.
Summary
Aside: I am a little hesitant to propose these changes. If the maintainers decide to reject this pull request I will understand.
Lightsleep on rp2 is gradually getting more complex and these changes would contribute to that complexity.
Fixes RP2040 issue #17228 and RP2350 issue #17229.
RP2040 issue is when watchdog timer expires while lightsleep is running the RP2040 will lock up during power-on startup.
This occurs because the power-on state machine is configured to not re-initialize the ROSC and lightsleep stopped the ROSC.
RP2350 issue is the watchdog timer does not decrement while lightsleep is running.
Once lightsleep returns the timer will start decrementing again. This occurs because lightsleep clears the bit
CLOCKS_SLEEP_EN1_CLK_SYS_WATCHDOG_BITS while sleeping.
Testing
I ran the reproduction methods from the two issues.
I also ran tests/port/rp2/rp2_lightsleep.py.
Trade-offs and Alternatives
General alternative. Document that lightsleep and watchdog don't play well together and leave the code as is.
Alternatives for RP2040 issue #17228:
- Raise issue for pico-sdk. Get them to change function _watchdog_enable() so the ROSC and XOSC do get reset on a watchdog timeout. The existing behavior in _watchdog_enable() has been there from day one. There might not be a good reason for it. Disadvantage: a possible long wait for a new pico-sdk release.
- Modify lightsleep to not stop the ROSC. I measure 4.96 milli-amps when lightsleep disables ROSC and 5.09 milli-amps when it does not. This is a small power increase.
Alternatives for the RP2350 issue #17229:
- I don't have any :-).