Re: [GIT pull] locking/urgent for v5.10-rc6

[Heiko Carstens]

On Wed, Dec 02, 2020 at 10:21:16AM +0100, Peter Zijlstra wrote:
> On Tue, Dec 01, 2020 at 08:18:56PM +0100, Heiko Carstens wrote:
> OK, so with a little help from s390/PoO and Sven, the code removed skips
> the TRACE_IRQS_OFF when IRQs were enabled in the old PSW (the previous
> context).
> 
> That sounds entirely the right thing. Irrespective of what the previous
> IRQ state was, the current state is off.
> 
> > diff --git a/arch/s390/kernel/idle.c b/arch/s390/kernel/idle.c
> > index 2b85096964f8..5bd8c1044d09 100644
> > --- a/arch/s390/kernel/idle.c
> > +++ b/arch/s390/kernel/idle.c
> > @@ -123,7 +123,6 @@ void arch_cpu_idle_enter(void)
> >  void arch_cpu_idle(void)
> >  {
> >  	enabled_wait();
> > -	raw_local_irq_enable();
> >  }
> 
> Currently arch_cpu_idle() is defined as to return with IRQs enabled,
> however, the very first thing we do when we return is
> raw_local_irq_disable(), so this change is harmless.
> 
> It is also the direction I've been arguing for elsewhere in this thread.
> So I'm certainly not complaining.

So I left that raw_local_irq_enable() in to be consistent with other
architectures. enabled_wait() now returns with irqs disabled, but with
a lockdep state that tells irqs are on...  See patch below.
Works and hopefully makes sense ;)

In addition (but not for rc7) I want to get rid of our complex udelay
implementation. I think we don't need that anymore.. so there would be
only the idle code left where we have to play tricks.