05:56 <evx256> so delving into this gl pinephone saga a bit more, running glxgears, it appears to use a `format = PIPE_FORMAT_B8G8R8X8_UNORM` for the swapchain/framebuffer surface; the apps i've been testing attempt to use PIPE_FORMAT_R8G8B8A8_SRGB or PIPE_FORMAT_B8G8R8A8_UNORM (i think the first is just a gamma corrected texture and the latter is a swapchain, but i'm not sure). anyway, i'm not familiar with the X in the glxgears example, doe

05:56 <evx256> mean "dont care"?

05:58 <evx256> fwiw, i've just been setting a breakpoint in panfrost_create_surface, and printing `surf_tmp` or `pt`, assuming one of these calls has the actual swapchain surface type

11:59 <alyssa> evx256: the X means "don't care" or more precisely "unused"

11:59 <alyssa> ignored when reading, undefined contents when writing

14:21 <robmur01> macc24: Not a GPU problem, by the look of it - the rockchip-pm-domain error is the smoking gun; the subsequent SError is almost certainly just a symptom of touching something inside a power domain that was supposed to be powered on but isn't.

14:23 <robmur01> I'd guess you're probably looking for races or other badness between devfreq and runtime PM

17:29 <tomeu> jenneron: you can try playing with https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/panfrost/panfrost_job.c#L24 , but TBH, the values for DEFAULT_JS_HARD_STOP_TICKS_SS_8408 and friends look way higher than I would expect

17:29 <tomeu> scheduling in mainline and kbase is very different, so I'm not sure you can work that issue around by just changing that value

20:23 <jenneron> tomeu: i've already tried to increase this value. i think 8408 requires more, e.g. less amount of soft stopping/resetting, i tried also to delay in soft reset function, but it looks like it just delays it, but amount of resets and time between them are same.

20:24 <jenneron> here is description of 8408, https://elixir.bootlin.com/linux/latest/source/drivers/gpu/drm/panfrost/panfrost_issues.h#L43, it seems to happen for me on T604 and seems that T604 will be usable after fixing that

23:00 <alyssa> jenneron: That doesn't look like your bug

23:00 <alyssa> There's no cache flush job in the job chain, Panfrost doesn't produce cache flush jobs

23:00 <alyssa> so Panfrost should be unaffected by #8408