panic: Assertion td->td_lock == TDQ_LOCKPTR(tdq) failed at /usr/src/sys/kern/sched_ule.c:2137

[Konstantin Belousov]

On Sun, Jun 24, 2018 at 10:52:32PM +0200, Ronald Klop wrote:
> On Sat, 02 Jun 2018 23:04:50 +0200, Konstantin Belousov  
> <kostikbel at gmail.com> wrote:
> 
> > On Sat, Jun 02, 2018 at 02:39:00PM -0600, Warner Losh wrote:
> >> On Sat, Jun 2, 2018, 1:32 PM bob prohaska <fbsd at www.zefox.net> wrote:
> >>
> >> > On Sat, Jun 02, 2018 at 09:14:18AM -0600, Warner Losh wrote:
> >> > >
> >> > > Do you have r334508? It fixes the OOM falsely triggering.
> >> > >
> >> > At the moment sources are at 334456, world and kernel are at
> >> > 334276.
> >> >
> >> > Might it be feasible to stop the present buildworld, update
> >> > sources and then try to build a new kernel before building
> >> > world?
> >> >
> >>
> >> That's recent enough that you should be fine. There might be an lld vs
> >> binutils ld issue. There was on i386 recently, but I think it was i386
> >> specific. It's a good risk.
> >
> > This happens on arm64 ?  In fact, try this.
> > I did not even compiled the change.
> >
> > diff --git a/sys/arm64/arm64/swtch.S b/sys/arm64/arm64/swtch.S
> > index c9843303b1d..4c2c3aca583 100644
> > --- a/sys/arm64/arm64/swtch.S
> > +++ b/sys/arm64/arm64/swtch.S
> > @@ -165,10 +165,9 @@ ENTRY(cpu_switch)
> >  	mov	x0, x19
> > 	/*
> > -	 * Release the old thread. This doesn't need to be a store-release
> > -	 * as the above dsb instruction will provide release semantics.
> > +	 * Release the old thread.
> >  	 */
> > -	str	x2, [x0, #TD_LOCK]
> > +	stlr	x2, [x0, #TD_LOCK]
> >  #if defined(SCHED_ULE) && defined(SMP)
> >  	/* Spin if TD_LOCK points to a blocked_lock */
> >  	ldr	x2, =_C_LABEL(blocked_lock)
> 
> Hi, this patch was committed in r334851. Thanks for that.
> I just wanted to note that since I'm running a GENERIC-NODEBUG kernel of  
> 333449 (before the commit) I have not seen the TDQ_LOCKPTR panics anymore.  
> While the RPI3B+ has been compiling quite heavy.
> 
> So, my question, is it expected that the panic does not occur (or far  
> less) with a -NODEBUG kernel?

Kernel without INVARIANTS does not have asserts compiled in.