Comment # 11 on bug 1187264 from
------- Comment From geraldsc@de.ibm.com 2021-06-15 06:53 EDT-------
(In reply to comment #14)
> @R?diger: Where can I see whether THP is enabled on OBS worker nodes?
>
> for a while we had "transparent_hugepage=never" in the guest kernel
> commandline, you could see that in the buildlog. but since mid of December
> 2020 we dropped
> this again.
>
> I've created a dump of such a stuck machine during the nodejs16 build but
> now the question is where to place the 4G (sparse) tar file with
> buildlog/vm-dump/root/swap

Not sure if a dump would help much, because the system does not panic at the
point where it detects the bad page state. If you take a dump later, the struct
page will most likely have changed.

However, the kernel dmesg would help, and it would show in the dump. But you
could also attach it here. The kernel prints a lot more information than just
the "Bad page state" warning, including the reason why it thinks that it is
bad. Not sure why that does not show in your log, could you please attach the
dmesg output?

Which kernel version is this, and where can we find the corresponding source
code?

------- Comment From geraldsc@de.ibm.com 2021-06-15 06:58 EDT-------
(In reply to comment #14)
> @R?diger: Where can I see whether THP is enabled on OBS worker nodes?
>
> for a while we had "transparent_hugepage=never" in the guest kernel
> commandline, you could see that in the buildlog. but since mid of December
> 2020 we dropped
> this again.

That's interesting, and also a bit courageous, given the problems you saw in
LTC bug#190818 / SUSE bug#1180917. Could you try if disabling THP again would
help here? Is this easily reproducible at all?

BTW, does that mean that you cannot reproduce the issue from the other bugzilla
any more, even with THP enabled?


You are receiving this mail because: