Trellis M4 latest build causing issue with NeoPixel code

[kattni]

I updated CircuitPython on my Trellis M4 and any code involving the NeoPixels is failing to function properly.

The failure is:

1. All of the NeoPIxels will turn bright white (evidently (255, 255, 255)).
2. After a seemingly consistent period of time (a few seconds), CIRCUITPY will disconnect, all of the NeoPixels will become a dimmer white (as though it's more like (45, 45, 45) or something), and the status DotStar LED on the back will turn a slightly purple-tinged red (which is to say, you can see that there is a little blue to it, but it is primarily red).
3. If connected to the serial console, no errors show up, there is nothing there to indicate what happened.
4. Pressing buttons on the front does nothing at this point.
5. A single reset returns it to the bootloader and TRELM4BOOT mounts.
6. However, attempting to load the same or a different version of CircuitPython at this point does not change the state of the board - copy the build, TRELM4BOOT will disconnect, and then the board will restart with the same purple-red status LED on the back and no CIRCUITPY drive will mount.

The resolution is:

1. Unplug the board.
2. Plug it back in.
3. IMMEDIATELY double-tap reset to get into the bootloader.
4. Load an earlier version of CircuitPython onto the TRELM4BOOT drive. This returns it to a functional state.

Tested with the following examples:

• This demo starts with a rainbow already lit up, and lights up every LED the same color when you press a button. When the newer version of CircuitPython is loaded, on a button press, it flickers solid white as well as the correct color, it continues to do so for a short period of time, and then drops into the failure steps listed above.
  https://github.com/adafruit/Adafruit_CircuitPython_TrellisM4/blob/master/examples/neopixel_trellism4.simpletest.py
• This demo starts with all the LEDs off and you can toggle individual buttons on and off with different colors. When a newer version of CircuitPython is loaded, it starts up with all the LEDS white, button presses will toggle individual buttons colors while flashing all the pixels white as well, and then after a short period of time drop into the failure steps listed above.
  https://github.com/adafruit/Adafruit_CircuitPython_TrellisM4/blob/master/examples/trellism4_neopixel_toggle.py
• To verify that this isn't an issue with the Adafruit_CircuitPython_TrellisM4 library, I created a simple code.py that used neopixel and turned on the first NeoPixel. With the newer version of CircuitPython loaded, it turns on the first LED, flashes white for a bit, then drops into the failure steps listed above.

Issue reproduced with the following versions:
adafruit-circuitpython-trellis_m4_express-en_US-20181110-50f5d27.uf2
adafruit-circuitpython-trellis_m4_express-en_US-20181109-97bc951.uf2

The following versions do not have the issue:
adafruit-circuitpython-trellis_m4_express-en_US-20181107-d08747d.uf2
adafruit-circuitpython-trellis_m4_express-en_US-20181105-283e072.uf2

[jerryneedell]

Just a thought. I found that setting all of the pixels white causes an over current condition so some of this issue may be masked by that.
I have reproduced this issue on my trellis m4.

[tannewt]

Thank you for the detailed report. I was able to reproduce this on my end. I believe @jerryneedell is right that the bad failure mode is a result of overdrawing the usb line with all white pixels.

The question is, why the white pixels? Well, I haven't figured it out. It appears that the neopixel clocking is slightly slower than in d08747d. The zero high time is 0.5 us instead of 0.38 us which is enough to be a one. I get these numbers on a Metro M4 without neopixels attached so it isn't a fundamental power issue on the Trellis. I also tested the sleep duration between neopixel writes and its consistently 1.004 seconds between both commits. So, this makes me think we're hitting some sort of caching boundary or memory bus contention we weren't hitting before. I have no idea exactly what it is.

[dhalbert]

I wonder if the instruction/data cache is on at the right time? Maybe that got lost in some refactoring? It gets turned on in the beginning, and also gets turned off/on during some external flash operations.

[estiens]

I still have this issue. Currently my board is in a bricked state from accidentally turning all the pixels white. They turn white for a second, then the drive disappears and the board is stuck with all yellowish looking pixels

[dhalbert]

@estiens The problem is getting your board into bootloader mode. Then you can load the Trellis M4 .uf2 listed in the Trellis learn guide that does not have this problem. What I'd suggest trying is this:

1. Unplug the board from USB.
2. Hold down the reset button.
3. Continuing to hold down the reset button, plug the board into USB.
4. Release and quickly double-click the reset button, so that CircuitPython does not get a chance to start. I've tried this and gotten it into TRELM4BOOT without entering CircuitPython.

If this doesn't work it may be possible to pull down the input to the first NeoPixel to ground with, say a 1k resistor, to avoid illuminating them, and then double-click.

[tannewt]

@estiens If your code is the source of all white pixels we can give you a safe-mode only build to use to change it.

[dhalbert]

Here is a safe-mode build. I built this last night but then thought you probably had a clock issue. Unzip before using.
trellis_m4-SAFE-MODE.uf2.zip