espressif: BLE prevents true deepsleep, even when not enabled (?)

[Sola85]

CircuitPython version

Adafruit CircuitPython 9.1.3 on 2024-08-30; ESP32-S3-DevKitM-1-N8 with ESP32S3
and latest

Code/REPL

import alarm
import microcontroller as uc
uc.cpu.frequency = 240_000_000 # change

alarm.exit_and_deep_sleep_until_alarms()

Behavior

The title is my current guess about what is going on. I am currently debugging a custom board with a ESP32-S3-Mini-1-N8 (8MB flash) which does not want to enter deep sleep and instead draws ~ 40mA, which is obviously way too high. (Note that USB is detached and the board reset before the current measurement.)

Description

• Clue 1: Sleep Current consumption depends on CPU frequency: ~15mA@40MHz, ~40mA@240MHz. This fact makes me conclude that its not an issue with the board but rather a software issue.
• Clue 2: Waveshare ESP32-S3-Zero with ESP32S3 (4MB flash) is not affected.
• Clue 3: If I change this line in main.c to if (true || awoke_from_true_deep_sleep || !supervisor_workflow_active()) { the board sleeps correctly with the expected sleep current of a few microamps.
• Clue 4: I added the line
  mp_printf(MP_PYTHON_PRINTER, "tud_ready(): %d, websocket_connected(): %d, ble_serial_connected(): %d", tud_ready(), websocket_connected(), ble_serial_connected()); to the top of supervisor_workflow_active(), which prints:
  "tud_ready(): 1, websocket_connected(): 0, ble_serial_connected(): 1"

Hence I arrive at the conclusion that BLE is preventing deepsleep. This also explains why the Waveshare ESP32-S3-Zero is not affected, since it doesn't have BLE enabled at build time.

Note that I have not enabled anything regarding BLE to reproduce this. settings.toml is empty and not even adafruit_ble is installed. The above code is the only file on the board after a fresh install of CircuitPython.

Additional information

No response

[tannewt]

Can you make a PR for Clue 3 please? That makes it easier to see changes.

Is the issue that ble_serial_connected is true when it shouldn't be?

Thanks for looking into this!

[Sola85]

Can you make a PR for Clue 3 please?

Yes, will do tomorrow. But like I said, that was just a temporary change for debugging, by making CP never fake deep sleep, even when it thinks that a workflow is currently active.

Is the issue that ble_serial_connected is true when it shouldn't be?

Yes that's my current guess. If I'm not mistaken this means that deepsleep is currently broken on all 8MB esp boards (specifically: all boards that have BLE enabled)

[bablokb]

Yes that's my current guess. If I'm not mistaken this means that deepsleep is currently broken on all 8MB esp boards (specifically: all boards that have BLE enabled)

There is #9464 that seems to address this.

[Sola85]

Can you make a PR for Clue 3 please?

Here is a PR to compare the changes: Sola85#2. Removing ble_serial_connected() from supervisor_workflow_active() makes deep sleep work again (For course this is not a proper fix, just a debugging step).

[dhalbert]

I have a fix - will be doing a PR soon.