15:02 soreau: With kernel 6.8 on RX580, `cat amdgpu_gpu_recover` seems to wedge the gpu with bands of corrupted/uninitted vram where the compositor used to be. Despite being able to ssh in, the only option to get the display working again is the reset button
15:03 soreau: I've also experienced 'natural' gpu resets with this exact behavior. IIRC it also happened with 6.5 and earlier.
15:04 MrCooper: that's an amdgpu kernel driver issue
15:04 soreau: well I'm kinda wondering if it's been addressed in more recent kernels or if there's any other info
15:07 agd5f: soreau, does it work again if you restart your desktop environment?
15:07 soreau: agd5f: I guess I didn't try restarting the compositor, no
15:08 soreau: I just verified it was running
15:08 soreau: do you think it's waiting on something for the compositor to do?
15:08 soreau: trying to handle context lost in the compositor automatically
15:08 agd5f: if your reset the entire GPU, all contexts using the GPU are lost including the compositor
15:08 Venemo: sounds like your compositor is not written to withstang a GPU reset
15:08 Venemo: withstand*
15:09 soreau: Venemo: I'm trying to implement it
15:09 soreau: it resets the allocator and renderer
15:09 Venemo: if you're using GL, look into the GL robustness extension
15:10 soreau: yes that is what signal I'm using to trigger it
15:10 soreau: context lost
15:11 soreau: Mesa: error: GL_CONTEXT_LOST in context lost
15:11 soreau: EE 17-01-25 07:51:42.145 - [GLES2] GL_CONTEXT_LOST in context lost
15:11 soreau: EE 17-01-25 07:51:42.145 - [src/core/opengl.cpp:48] gles2: function glDisable(0x0C11) in void OpenGL::render_end() line 322: UNKNOWN GL ERROR
15:11 soreau: hm
15:11 Venemo: which compositor is this, btw?
15:11 soreau: wayfire
15:12 soreau: I think I know what to try
15:16 Venemo: are there any errors in dmesg other than the reset itself?
15:19 soreau: let me try again
15:19 MrCooper: soreau: "the only option to get the display working again is the reset button" → "I guess I didn't try restarting the compositor" you skipped a few steps there :)
15:19 soreau: MrCooper: heh
15:19 soreau: ok brb
15:22 soreau: back
15:22 soreau: this time the code crashed and the reset survived (haven't rebooted)
15:23 soreau: but the crash appears to be the same one that kode54 gets and it's in wayfire, so I can probably work more from here
15:23 soreau: thanks guys, might have more questions later