Re: [QUESTION] problems report: rcu_read_unlock_special() called in irq_exit() causes dead loop

[Frederic Weisbecker <frederic@kernel.org>]

Le Wed, Jul 02, 2025 at 06:59:38AM -0400, Joel Fernandes a écrit :
> 
> 
> On 7/2/2025 5:14 AM, Qi Xi wrote:
> > Hi Joel,
> > 
> > After applying the 2 patches, the problem still exists. Compared to the previous
> > fixes which did solve the problem, the difference is ct_in_irq() in the first
> > patch.
> > 
> > I am wondering why "nesting != CT_NESTING_IRQ_NONIDLE" is added?
> > 
> > 
> > (previous fix: problem is solved)
> > 
> > +bool ct_in_irq(void)
> > +{
> > +    return ct_nmi_nesting() != 0;
> > +}
> > 
> > (current fix: problem still exists)
> > 
> > +bool ct_in_irq(void)
> > +{
> > +    long nesting = ct_nmi_nesting();
> > +
> > +    return (nesting && nesting != CT_NESTING_IRQ_NONIDLE);
> > +}
> 
> Oh gosh, thanks for spotting that! Indeed,  I had changed it to != 0 in the last
> version but applied an older patch. I will fix it in the tree. Thank you again!
> 
> Neeraj, would you like this as a separate commit that you can then squash? Or
> could you fix it up in your tree?

nesting == CT_NESTING_IRQ_NONIDLE means that the rcu_is_watching() and we are
not in an interrupt. We still need to rule out that case, right?

Is there something else that makes it not working?

-- 
Frederic Weisbecker
SUSE Labs