page fault while in kernel mode - after upgrade from 12.2 to 13.0

[Mark Johnston]

On Thu, May 06, 2021 at 06:00:05PM +0200, Michael Schmiedgen wrote:
> On 05.05.2021 20:38, Mark Johnston wrote:
> > On Wed, May 05, 2021 at 06:35:32PM +0200, Michael Schmiedgen wrote:
> >> On 04.05.2021 21:02, Mark Johnston wrote:
> >>> This looks like fairly random kernel memory corruption.  Are you able to
> >>> build an INVARIANTS kernel and test that?  Assuming you're using 13.0,
> >>> you'd grab the 13.0 sources, add "options INVARIANT_SUPPORT" and
> >>> "options INVARIANTS" to the GENERIC kernel configuration in
> >>> sys/amd64/conf, and do a "make buildkernel installkernel".
> >>
> >> Below some info with an INVARIANTS kernel. Please let me know if I can provide
> >> further information. Thank you!
> >
> > Thanks, this helped a lot.  I believe https://reviews.freebsd.org/D30129
> > will fix the problem.  That patch is against the main branch but applies
> > cleanly to 13.0.
> 
> I applied the patch and the server is running fine now for 8 hours with the
> INVARIANTS kernel, including the Samba jail and SIP VM. I just compiled my
> custom kernel and it looks like it is working too. Are there plans to get
> this MFCed or even as Errata?

Great, thanks.  Yes I think we will do an EN for this.

> BTW, we got 2 other systems, also with userland NAT but different workload.
> After an uncertain amount of time, mostly weeks, the natd starts to spin 100%
> CPU on these systems. Quick noobish workaround was restarting natd every night.
> I saw your recent commits that applied some more safety in that area, do you
> plan to MFC these as well? I can imagine that could help with my NAT problems.

I am skeptical that anything I did recently would fix this.  Did you try
attaching a debugger to natd to see where it's getting stuck?  Is it
also a regression from upgrading to 13.0?

> Anyway, many thanks for your investigation and your fix, much appreciated!
> 
>    Michael
>