Re: [RFC PATCH 1/3] lockdep/irq: Be more strict about IRQ-threadable code end

[Peter Zijlstra]

On Mon, Mar 23, 2020 at 04:30:40PM +0100, Frederic Weisbecker wrote:
> On Mon, Mar 23, 2020 at 02:53:20PM +0100, Peter Zijlstra wrote:
> > On Mon, Mar 23, 2020 at 04:32:05AM +0100, Frederic Weisbecker wrote:
> > > --- a/kernel/irq/handle.c
> > > +++ b/kernel/irq/handle.c
> > > @@ -144,18 +144,24 @@ irqreturn_t __handle_irq_event_percpu(struct irq_desc *desc, unsigned int *flags
> > >  
> > >  	for_each_action_of_desc(desc, action) {
> > >  		irqreturn_t res;
> > > +		bool threadable;
> > >  
> > >  		/*
> > >  		 * If this IRQ would be threaded under force_irqthreads, mark it so.
> > >  		 */
> > > -		if (irq_settings_can_thread(desc) &&
> > > -		    !(action->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT)))
> > > +		threadable = (irq_settings_can_thread(desc) &&
> > > +			      !(action->flags & (IRQF_NO_THREAD | IRQF_PERCPU | IRQF_ONESHOT)));
> > > +
> > > +		if (threadable)
> > >  			trace_hardirq_threaded();
> > >  
> > >  		trace_irq_handler_entry(irq, action);
> > >  		res = action->handler(irq, action->dev_id);
> > >  		trace_irq_handler_exit(irq, action, res);
> > >  
> > > +		if (threadable)
> > > +			trace_hardirq_unthreaded();
> > 
> > AFAICT this doesn't work for nested IRQ handlers.
> 
> So current->hardirq_threaded should be a counter perhaps?

Yeah, see how the old code used the hardirq_context counter for exactly
that. Also note how the old code was actually cheaper than this
(minimally, but still).