The author discusses a persistent issue of Linux crashing upon sleep due to a driver bug, taking over a year to diagnose and solve. They share their journey of debugging, analyzing crash logs, experimenting with system changes, setting up a serial console for data retrieval during crashes, and proposing kernel patches to address the bug. Surprising content includes using Ghidra to decompile the amdgpu module and suggesting a workaround of copying NVIDIA’s userspace VRAM backup system. The author highlights challenges faced when implementing solutions and decides against upstreaming changes due to lack of complete resolution.
https://nyanpasu64.gitlab.io/blog/amdgpu-sleep-wake-hang/