Bug fixes for rare glitches on WS2812B pixels (NeoPixels) and 030 errors on micro:bit V2 (1/3)

[kevinjwalters]

Issue Tickets

This is one of a trio of PRs which fixes issues with neopixel_send_buffer() used by MicroPython's:

• Rare glitches on WS2812B pixels using neopixel library on micro:bit V2 microbit-foundation/micropython-microbit-v2#227 & Occasional corruption of neopixel wire data causing visual glitching codal-microbit-v2#475 - glitches on WS2812B RGB LEDs (includes Adafruit's NeoPixels)
• 030 DEVICE_HEAP_ERROR codal error on long-running code using neopixel and display microbit-foundation/micropython-microbit-v2#228 - after 8 days of no low-level errors in testing I believe these were all caused by these bugs
• neopixels WS2812B PWM source reads one byte beyond the end of the data codal-microbit-v2#241
• periodUs incorrect due to int maths #58 - spotted whilst using debugger
• Bugs and improvements in WS2812B and streaming NRF52PWM #59 - concise summary ticket of bug fixes, enhancements and hygiene issues.
• Improve efficiency of NRF52TouchSensor interrupt handler #60 - noted while searching for inefficiencies in priority 3 interrupt handlers

@finneyj @martinwork @microbit-carlos may be interested in this.

Summary

The summary of changes for all three PRs (this one has the bulk of them in WS2812B and NRF52PWM classes) is:

• added volatile to variables in classes for better safety between main and interrupt handler code
• made WS2812B::play() fully blocking to prevent hazardous overlapping executions of neopixel_send_buffer() - this was implemented using the existing dataWanted() perhaps not quite how it's intended to be used?
• fixed the one extra pulse bug
• fixied PWM stopping using same technique as nrfx_pwm_stop() + sprinkling of event clears of EVENTS_SEQEND
• dramatically reduced chance of bogus pulses occuring after end of transmission by zeroing data buffer and using slightly enlarged emptyBuffer
• added an errorFlag and errorFlagCumulative (not yet exposed via API or MicroPython)
• #if disabled currently: added detail pwm statistics and error counters with history and anomalying DMESG printing (only intended for future debugging - this could be removed)
• safer conditionals with dataReady
• made interrupt handler in NRF52TouchSensor more efficient.
• more efficient transfer of from WS2812B to NRF52PWM by reusing existing buffers, removing data initialisation for ManagedBuffer and optimised code - this required a minor addition to the interface for DataSource
• after some testing and consideration increased the priority of NRF52PWM interrupts by changing PWM2_IRQn from 3 to 2.
• trivial changes: fixes for periodUs, remove of redundant setSampleRate(), change lock initalisation in WS28128B - these had no impact on glitching bugs.

I've highlighted the three things in bold italics which I think need the most reviewing.

The fixes were applied to a branch from master made on the 24th July 2025. This will clearly be different code to the production release where these bugs were originally observed:MicroPython v1.18 on 2023-10-30; micro:bit v2.1.2 with nRF52833 on a V2.2.1 micro;bit.

The diff looks a lot bigger than it really is, the bulk of the lines are the PWM statistics code which is #ifed out.

---

Testing

The bug fixes have been tested:

1. As they were developed in debugger and with logic analyzer, documented in comments in Occasional corruption of neopixel wire data causing visual glitching codal-microbit-v2#475
2. Recently for four days on three micro:bits with occasional visual observation running neopixel-thirty-bug-colourcycle-26.py
3. Four days on four micro:bits with occasional visual observation running neopixel-thirty-bug-timedetector-16.py & neopixel-thirty-bug-timedetector-23.py & neopixel-thirty-bug-colourcycle-26.py & zhhclock - video

---

Partner PRs:

• DataSource interface extension for Bug fixes for rare glitches on WS2812B pixels (NeoPixels) and 030 errors on micro:bit V2 (2/3) codal-core#184
• PWM2_IRQn (neopixel) priority increase for Bug fixes for rare glitches on WS2812B pixels (NeoPixels) and 030 errors on micro:bit V2 (3/3) codal-microbit-v2#505

---

Additional information (not required reading):

• Debugging Rare RGB LED Glitches in MicroPython on BBC micro:bit
• Everything You Always Wanted to Know About Debugging the Nordic nRF52 PWM Peripheral and Cortex ARM Interrupts* (*But were afraid to ask)

[kevinjwalters]

Unrelated to this I would be very grateful if you could discuss increasing the file system size from 24k to at least 32k with the creator (microbit-foundation/micropython-microbit-v2#226).