Re: [linux-next][s390] Boot hang after merge c50f05bd3c4e (sched/hrtick)

[Heiko Carstens <hca@linux.ibm.com>]

On Thu, Mar 05, 2026 at 02:07:25PM +0100, Peter Zijlstra wrote:
> > > Moving that before irq_enter_rcu() means it doesn't see HARDIRQ_OFFSET
> > > in preempt_count(). As such, it might actually call into schedule() from
> > > hardirq context.
> > > 
> > > Or am I missing something?
> > 
> > N/m, it turns into __this_cpu_add() and that doesn't have
> > preempt_enable().
> > 
> > 00d8b035eb71 ("s390/idle: Slightly optimize idle time accounting")
> > 
> > Is actually a correctness fix afaict.

Yes. Those dependencies become quite subtle when calling early into C code.

> Another change is that you clear I and E in the PSW bit before
> irq_enter_rcu(), which, per:
> 
>   7e641e52cf5f ("softirq: Prepare for deferred hrtimer rearming")
> 
> can re-arm the timer.
> 
> So where previously it would re-arm and still have the I/E bits set, so
> the timer could fire, they are now disabled.
> 
> I really don't know if this is a problem; I'm clutching at s390 straws
> here that I really don't know much about.

That's the old PSW where the interrupt happened, not the one with which the
CPU is running with, and shouldn't have any effect.

...but reverting that commit actually does fix it for me. I don't see
immediately any code which checks if the _previous_ context had irqs enabled
or disabled, since that is the only difference.