had a very similar issue on my (launch) ryzen 1700, tried some things from https://wiki.archlinux.org/title/Ryzen but in the end the only thing that solved it was to replace the cpu with something newer...
possibly, but hard to say with the data here. you could try things like disabling power state management (e.g. in the bios of with pcie_aspm=off on the kernel commandline)
sorry if my last post was badly worded, what i actually wanted you to...
Until there is a patch available, here is a documented mitigation:
https://github.com/V4bel/dirtyfrag/tree/master#mitigation
(EDIT: i know the link was in the first post already, but wanted to explicitly mention the mitigation section)
maybe a different pool/allocation policy scheme makes more sense for you? see e.g. the 'weekday' scheme in the docs:
https://pbs.proxmox.com/docs/tape-backup.html#weekday-scheme
yes i can confirm. the bug is that the cli parameter has the wrong name
on the api the parameter is named 'group-filter' and the cli only passes it's parameter through to there
i sent a patch to the list to correct the cli parameter name to...
it's just a general requirement that the pdm needs network access, otherwise it can't do anything. I'd not read too much into the 'dedicated' wording here. Of course it depends on the actual setup what kind of nic is needed. e.g. with many...
ok thanks for the logs.
while the ubsan line is something somebody should take a look at, i think it's also just a symptom of the underlying issue.
if the card is not in use, i doubt this is a software bug since the kernel does not just...
where did you find these sysfs paths?
with a b50 here, these files do not exist, and i dont see any other files that would give control over that i'm afraid...
it seems something wants to change the power state of the device and is unable to
do you passthrough any other device on that server? do you use the acs override kernel commandline?
more from the log would also be interesting (e.g. 'journalctl...