Hello, On Tue, Sep 16, 2025 at 08:42:06PM +0200, Richard wrote: > I've got some bad news, that may help figuring out where the issue > lies though. I just went back to kernel 6.6.106 because of both this So you compiled your own 6.6.106 kernel, right? > issue and several amdgpu issues. I thought it couldn't happen here, > but it did. Wherever the source lies, it must have been backported to > 6.6. I first set up my FW16 with Debian 12 back in late April 2024, > but this issue didn't appear before kernel 6.9, maybe even with 6.10 > or 6.11. While I'm not sure when exactly, it most certainly didn't > appear before the second half of 2024, most likely only in Q4. Hmm, v6.6.36 was released on 2024-06-27 and v6.6.37 on 2024-07-05. So v6.6.36 is the latest commit where you are reasonably sure that it's still good, right. Unfortunately there are more than 11000 commits between v6.6.36 and 6.6.106. If the problem was easier to reproduce, I'd ask for bisection. But given that you can only diagnose that a kernel is bad but not that it's good, maybe do a bisect that I'd call a 90:10 bisection. That is, instead of picking the middle between known good and known bad, pick a rev that is much nearer to the bad commit to test. So in a first step maybe test 6.6.102 or so. (Of course running an old stable kernel for a time long enough to be able to trigger that problem is risky. So ideally do that in a known good network and with proper backups.) Best regards Uwe
Attachment:
signature.asc
Description: PGP signature