Re: Panic on kern_event.c

On Fri, Nov 16, 2018 at 03:47:39PM +0100, Sylvain GALLIANO wrote:
> Le jeu. 15 nov. 2018 à 23:10, Mark Johnston <markj_at_freebsd.org> a écrit :
> 
> > On Thu, Nov 08, 2018 at 05:05:03PM +0100, Sylvain GALLIANO wrote:
> > > Hi,
> > >
> > > I replaced
> > > << printf("XXX knote %p already in tailq  status:%x kq_count:%d  [%p %p]
> > >
> > %u\n",kn,kn->kn_status,kq->kq_count,kn->kn_tqe.tqe_next,kn->kn_tqe.tqe_prev,__LINE__);
> > > by
> > > >> panic("XXX knote %p already in tailq  status:%x kq_count:%d  [%p %p]
> > >
> > %u\n",kn,kn->kn_status,kq->kq_count,kn->kn_tqe.tqe_next,kn->kn_tqe.tqe_prev,__LINE__);
> > >
> > > Here is the stack during panic:
> > > panic: XXX knote 0xfffff801e1c6ddc0 already in tailq  status:1 kq_count:2
> > > [0 0xfffff8000957a978]  2671
> > >
> > Could you please give the following patch a try?
> >
> > If possible, could you also ktrace one of the active syslog-ng processes
> > for some time, perhaps 15 seconds, and share the kdump?  I have been
> > trying to reproduce the problem without any luck.
> >
> Unfortunately patched kernel is not stable:
> - some processes run at 100% CPU (STOP state) and cannot be killed
> - sometime the system completely freeze (need a hard reboot)
> 
> I cannot reproduce the issue as soon as syslog-ng is under ktrace (even
> after 10GB of ktrace file)
> When I stop ktrace, issue come back after few minutes.

That's ok, I'd like to see part of the ktrace even if the problem
doesn't occur; this bug appears to be a race condition, so it's not
surprising that ktrace might hide it.