AMDGPU: cp queue preemption timeout

[Kuba Pawlak]

Happened on Lenovo T14 AMD gen 2 5850U running Mantic 6.5.0-13-generic

Under load, GPU crashed resulting in a weird behavior. Screen froze, blinked a few times and the laptop suspended. After wakeup a normal login screen showed up but after logging in, it crashed to black screen with a message: " cp queue preemption timeout". Eventually I was able to log in to a clean session.

Log attached

X-HWE-Bug: Bug #2047389

[Mario Limonciello (superm1)]

This will be fixed by this patch:

https://<email address hidden>/

Until Canonical pulls it in for now you can use amdgpu.mcbp=0 on kernel command line to avoid it.

[Timo Aaltonen (tjaalton)]

@AMD, any idea about the above?

[Mario Limonciello (superm1)]

that patch I linked will go into the next 6.7-fixes PR. I suggest backporting it.

[Mario Limonciello (superm1)]

Here's the fix commit ID to backport to fix this issue.
https://github.com/torvalds/linux/commit/d6a57588666301acd9d42d3b00d74240964f07f6

[Timo Aaltonen (tjaalton)]

noble has 6.7 now, so this is essentially fixed.

Kuba, if you need this for mantic, open a task and submit it for inclusion, thanks.

[Timo Aaltonen (tjaalton)]

hah, should probably refresh old tabs before posting on them

[OEM Taipei Bot (oem-taipei-bot)]

linux-oem-22.04d 6.5.0.1014.16 in jammy-proposed contains the fix. https://changelogs.ubuntu.com/changelogs/pool/main/l/linux-oem-6.5/linux-oem-6.5_6.5.0-1014.15/changelog

[OEM Taipei Bot (oem-taipei-bot)]

linux-oem-22.04d 6.5.0.1013.15 in jammy-updates contains the fix. https://changelogs.ubuntu.com/changelogs/pool/main/l/linux-oem-6.5/linux-oem-6.5_6.5.0-1013.14/changelog