atmel-samd: Erratic I2C scanning behavior with nothing connected

[tdicola]

I noticed an interesting thing on a feather M0 express with nothing attached, if you scan for I2C devices you actually get random results. It's not necessarily a bug in CircuitPython but something worth noting that might trip folks up. Here's an example of scanning with nothing connected:

>>> import busio
>>> import board
>>> i2c = busio.I2C(board.SCL, board.SDA)
>>> while not i2c.try_lock():
...     pass
... 
>>> i2c.scan()
[65, 102]
>>> i2c.scan()
[33, 71]
>>> i2c.scan()
[]
>>> i2c.scan()
[65, 102]
>>> i2c.scan()
[]

Likely this is from the SCL and SDA pins not having internal pull-ups applied and floating to random high/low values. I wonder would it make sense to automatically turn these pull-ups on when I2C is initialized or even the board resets?

Impact-wise this isn't a huge issue. The only trouble it might cause is with drivers that validate the device exists by scanning for them first. In super rare cases they might get a false positive with nothing connected and interpret random noise as the expected I2C scan result. The driver will eventually fail when it tries to talk to the device, but it might be nice to catch it earlier if possible.

As far as I2C behavior goes it's all good once you plug in a device (which has its own pull-ups), I get a steady and expected I2C scan result:

>>> i2c.scan()
[116]
>>> i2c.scan()
[116]
>>> i2c.scan()
[116]

[tannewt]

Good point @tdicola. We could also fix it by scanning "found" addresses a second or third time before adding them to the list.

[deshipu]

The problem with turning the pull-ups on is that then it makes it impossible for the I²C bus to talk to devices that use different voltages than 3.3V. Normally the pins are open-drain to make that possible. There can also be some bad interactions with pull-up resistors already included on many modules and shields — they would all add up, at some point making it impossible to pull the pin down. Granted, that particular problem is also present without the internal pull-ups, but it's made worse with them.

On the other hand, communicating with devices that use different voltages seems like an advanced topic, and requires some careful design of the circuit anyways, so perhaps the pull-ups could be enabled by default, but have an option for disabling them in the I²C initialization?

[tdicola]

Some reference info, to compare with Arduino's I2C behavior it looks like there's no scanning built-in but if you use a scan sketch (like this one: https://playground.arduino.cc/Main/I2cScanner) it will require initializing the I2C library (Wire.begin() call) which turns on the AVR's pull-ups for SCL and SDA: https://github.com/arduino/Arduino/blob/master/hardware/arduino/avr/libraries/Wire/src/utility/twi.c#L75-L77 It seems reasonable to match Arduino's behavior here and temporarily turn on pull-ups if they aren't already on when doing a scan.

[tdicola]

And for the Pi it looks like they build in physical pull-ups (1.8K on the GPIO2 and GPIO3 lines for SCL, SDA): https://www.raspberrypi.org/documentation/hardware/raspberrypi/schematics/Raspberry-Pi-3B-V1.2-Schematics.pdf Seems reasonable to match the behavior and turn on pull-ups when doing an I2C action like scanning.

[deshipu]

The fact that Arduino turns on pull-ups automatically is a well-known problem in the Arduino community, and the reason why the advanced users use other libraries (or just ad-hoc code) for I²C when they need to interact with mixed-voltage devices.

[deshipu]

You also need to consider how the pull-ups will interact with the external resistors already included on the I²C lines. If high-speed communication is needed, strong pull-ups will be used, and that can be too strong for the pins to drive when combined with the internal pull-ups.

[dhalbert]

@deshipu Is it a problem electrically if we just turn on the pullups during scan() and then turn them off again?

[deshipu]

I agree pull-ups are a reasonable default, but I would also strongly advise to have an option to suppress them.

[deshipu]

@deshipu Is it a problem electrically if we just turn on the pullups during scan() and then turn them off again?

Yes, it's still a problem. If you have a, say, 5V device on your I²C bus, and consequently, external pull-ups to 5V, then enabling the internal pull-ups to 3.3V will result in a short — through two pairs of large resistors, so maybe not that catastrophic, but a short none the less.

[dhalbert]

If I understand correctly, Adafruit's philosophy is to have pullups on the I2C peripheral boards, not on the master board. So our core I2C implementations do not turn on internal pullups.

[deshipu]

Well, Arduino does.

[tannewt]

Why can't we fix this by double or triple probing addresses that are available? If its really just line noise from no pull-ups then it shouldn't be consistent over repeated probes. There is no need to turn on internal pull ups.