Re: Stop scheduler on panic

On Thursday, December 01, 2011 3:42:24 pm Andriy Gapon wrote:
> on 01/12/2011 20:49 John Baldwin said the following:
> > On Thursday, December 01, 2011 11:59:10 am Andriy Gapon wrote:
> >>
> >> [cc list trimmed]
> >>
> >> on 21/11/2011 18:32 John Baldwin said the following:
> >>> On Friday, November 18, 2011 4:59:32 pm Andriy Gapon wrote:
> >>>> on 17/11/2011 23:38 John Baldwin said the following:
> >>>>> On Thursday, November 17, 2011 4:35:07 pm John Baldwin wrote:
> >>>>>> Hmmm, you could also make critical_exit() not perform deferred preemptions
> >>>>>> if SCHEDULER_STOPPED?  That would fix the recursion and still let the
> >>>>>> preemption "work" when resuming from the debugger?
> >>
> >>
> >> Just to clarify, probably you are actually suggesting to not perform deferred
> >> preemptions if kdb_active == TRUE.  Because that's where we get the recursion (via
> >> kdb_switch).
> >>
> >> I think that if we get into the mi_switch in a state where !kdb_active &&
> >> SCHEDULER_STOPPED(), then we probably should just - I don't know - panic again?
> >>
> >> [the following is preserved for context]
> > 
> > Hmmm.  I'd be tempted to just ignore pending preemptions anytime
> > SCHEDULER_STOPPED() is true.  If it's stopped for a reason other than being
> > in the debugger (e.g. panic), I'd rather make a best effort at getting a dump
> > than panic again.
> 
> Yep, me too.  It's just that I assumed that ending up at mi_switch in the panic
> thread/context meant that something had gone very wrong already.  But I am not
> sure if this was a valid assumption.
> 
> Returning to critical_exit, what do you think about the following patch?
> I guess that it could be committed independently of / before the
> SCHEDULER_STOPPED thing.
> 
> commit ee3d1a04985e86911a68d854439ae8c5429b7bd5
> Author: Andriy Gapon <avg_at_icyb.net.ua>
> Date:   Thu Dec 1 18:53:36 2011 +0200
> 
>     critical_exit: ignore td_owepreempt if kdb_active
> 
>     calling mi_switch in such a context result in a recursion via
>     kdb_switch
> 
> diff --git a/sys/kern/kern_switch.c b/sys/kern/kern_switch.c
> index 93cbf7b..885dc22 100644
> --- a/sys/kern/kern_switch.c
> +++ b/sys/kern/kern_switch.c
> _at__at_ -200,7 +200,7 _at__at_ critical_exit(void)
> 
>  	if (td->td_critnest == 1) {
>  		td->td_critnest = 0;
> -		if (td->td_owepreempt) {
> +		if (td->td_owepreempt && !kdb_active) {
>  			td->td_critnest = 1;
>  			thread_lock(td);
>  			td->td_critnest--;

I think this is fine, but I'd probably change this to SCHEDULER_STOPPED()
in the SCHEDULER_STOPPED() patch.

> Would it make sense wrap kdb_active check with __predict_false?

I don't think so.

-- 
John Baldwin