06:48shelter: Is this a known error?
06:48shelter: [mån dec 22 07:43:32 2025] amdgpu 0000:03:00.0: amdgpu: MES(1) failed to respond to msg=INVALIDATE_TLBS
06:48shelter: Kernel 6.18.2
06:48shelter: not present on 6.17.11
09:59varsi: idk if this is place to ask but is firefox having issues with AMD webrendering?
10:31shelter: not that im aware of, but amd is a pretty wide statement considering the range of gpu's
10:51fililip: shelter: I saw it a few times but it seems to be completely harmless
10:51johnny0: Venemo: ahh, the erratic/low power reporting on Raphael looks to be smu metrics overflow
11:03varsi: i think i found the issue which was not mozilla but xfce4 compositor. dunno if it relates to AMD specifically but im on very modern gpu (9070xt)
12:37shelter: fililip: yeah... nothing seems to be affected by it
12:47fililip: I have however noticed random ring comp timeouts when trying to launch mpv (using vulkan), but this is probably a mesa issue
12:47fililip: suspend-resume fixes them
14:04Venemo: johnny0: yeah?
14:04Venemo: fililip: which gpu?
14:50fililip: RX 9070
14:50fililip: though this issue is very difficult to start reproducing
14:50fililip: once it happens, it keeps happening until suspend-resume, but I have no idea what triggers it in the first place
14:51fililip: it might be sensible to play a game for ~80-90minutes and then give it a try, I was playing Detroit and, after I closed the game, it started happening and went away after s3 resume
14:53fililip: Is there a memory allocation tester I could use to verify if this is due to lots of allocations causing some kind of leak?
14:53varsi: wait whats your issue?
14:53fililip: ring comp_(numbers) timeout when trying to open a Vulkan app
14:54fililip: it recovers just fine and I don't get a reset but it's weird that it happens in the first place
14:54fililip: it means I can't use mpv though
14:54varsi: well i have the exact same card and i just fought with some power saving issues
14:55fililip: power saving?
14:57Venemo: fililip: please open an issue on the drm/amd github, with exact steps to reproduce
14:59varsi: fililip: well im not sure of it but if i force my card out of powersaving mode i have no issues o_O
14:59fililip: Venemo: the issue is that I don't really know how to reproduce this properly, it may or may not happen with different things I've tried, that's why I've left it unreported for now
14:59Venemo: I see. That's unfortunate
15:00Venemo: worth a try to disable all power saving features and see what happens then
17:19fililip: hm, there's something interesting I've noticed as well
17:19fililip: when an app's pages are evicted into RAM from VRAM and then the app is closed, I get this:... (full message at <https://matrix.org/oftc/media/v1/media/download/Act_71-diCg6VswB5STmk-CDqA-nuYaMtgiCblVuar-IZQTf6Mxw3copHLz5pJYoymPLwEl7kuH0awlxDM6s_1xCebjywvngAG1hdHJpeC5vcmcvR1pNWm5oUFpTa0tpckpldVlMUWZmV1Vp>)
17:20fililip: easiest reproducer is to open mpv, run amdgpu_stress -b v 14G -b v 2G and close mpv
17:27fililip: (oh, I just noticed what someone meant by matrix as pastebin, I had no idea it was automatically linking to its own thing, does that link work?)
17:28dwfreed: The link works for a while; I do not know the exact lifetime
23:28johnny0: Venemo: yeah. the metrics table stores power as unscaled mW in a 16-bit field, not so good when the product ships with a 170W default tdp