Since getting all these structures right is critical for the GPU to work and the firmware to not crash, I needed a way of quickly experimenting with them while I reverse engineered things. Thankfully, the Asahi Linux project already has a tool for this: The m1n1 Python framework! Since I was already writing a GPU tracer for the m1n1 hypervisor and filling out structure definitions in Python, I decided to just flip it on its head and start writing a Python GPU kernel driver, using the same structure definitions. Python is great for this, since it is very easy to iterate with! Even better, it can already talk the basic RTKit protocols and parse crash logs, and I improved the tools for that so I could see exactly what the firmware was doing when it crashes. This is all done by running scripts on a development machine which connects to the M1 machine via USB, so you can easily reboot it every time you want to test something and the test cycle is very fast!
Good for testing and iterating, but what about performance? Though I guess getting everything right is more important right now, translating it into another language will probably require less work that way
It has already been translated into rust. Python wasn’t ever intended to be used in the “real” driver, but I thought it was a fun anecdote none the less.
Fun fact!
The Asahi Linux drivers for the Apple M1 GPU were originally written in Python: https://asahilinux.org/2022/11/tales-of-the-m1-gpu/
Good for testing and iterating, but what about performance? Though I guess getting everything right is more important right now, translating it into another language will probably require less work that way
It has already been translated into rust. Python wasn’t ever intended to be used in the “real” driver, but I thought it was a fun anecdote none the less.