HardwarePWM-aware led driver

[NeonDaniel]

Description

Adds compatibility for operation with a dtoverlay controlling the PWM fan

Issues

Other Notes

[valorekhov]

Thank you for initiating the PR. Below is some additional information for future reference:

This PR currently is intended for Rev 10 boards and supports co-existence of the NeoPixel LED array connected to Pin 12 with the PWM Fan connected to Pin 13 of the board. The PWM Fan shall be controlled via a the pwm_fan kernel driver using hardware-generated PWM signal. The specific issue being resolved is:

• RPi4 has a single PWM timer used to generate signal over the PWM0 (Pin 12) and PWM1 (Pin13) domains
• Adafruit_CircuitPython_NeoPixel -> Adafruit_Blinka-> Adafruit_CircuitPython_Pixelbuf -> rpi_ws281x Lib dependency chain appears to have been written with a presumption of no coexistence with other connected PWM hardware.

This creates the following problems:

• The pwm_fan kernel driver starts up first, configuring the shared PWM clock divider and PWM1 range to values which the rpi_ws281x lib does not expect while performing its PWM signal generation with DMA.
• At the same time, Adafruit libs change PWM1 channel config prior to performing DMA writes (https://github.com/adafruit/Adafruit_Blinka/blob/b84de4800b2c31023f29a536d26209bd7517d504/src/adafruit_blinka/microcontroller/bcm283x/neopixel.py#L41), and also turn off the shared PWM clock at the end of a write op (https://github.com/adafruit/Adafruit_Blinka/blob/b84de4800b2c31023f29a536d26209bd7517d504/src/adafruit_blinka/microcontroller/bcm283x/neopixel.py#L122, https://github.com/jgarff/rpi_ws281x/blob/1f47b59ed603223d1376d36c788c89af67ae2fdc/ws2811.c#L1078)

The HwPwmAwareLed class attempts to fix these issues by:

• Configuring the PWM Clock divider to the expected value of 8 (240KHz) on init
• Setting PWM1 range to 32 prior to each buffer write, and
• Restoring Range, Clock Div and PWM Mode registers on exit for the PWM fan to continue to be operational

So far, this code has only been tested in the sj201 reset-led CLI scenario. Using it as a part of PHAL with no explicit process exit may require future adjustments.

[NeonDaniel]

Thanks for the detailed explanation. I will plan on implementing your kernel driver when I find time to revisit the Neon images; it looks like the "correct" solution to fan control for the Mark2 since that really should happen at a kernel level.