Re: [PATCH 1/3] genirq: Add support for priority-drop/deactivate interrupt controllers

[Thomas Gleixner]

On Thu, 30 Oct 2014, Marc Zyngier wrote:
> So I actually implemented this, and did hit another snag: per cpu interrupts.
> They don't use the startup/shutdown methods, and reproducing the above logic
> on a per-cpu basis is not very pretty.

Hmm. Have not looked at the percpu stuff yet.
 
>  /**
> + *	handle_spliteoi_irq - irq handler for 2-phase-eoi controllers
> + *	@irq:	the interrupt number
> + *	@desc:	the interrupt description structure for this irq
> + *
> + *	This relies on mask being a very cheap operation, and on
> + *	unmask performing both unmask+EOI. This avoids additional
> + *	operations for threaded interrupts (typically ARM's GICv2/v3).
> + */
> +void
> +handle_spliteoi_irq(unsigned int irq, struct irq_desc *desc)
> +{
> +	raw_spin_lock(&desc->lock);
> +
> +	if (!irq_may_run(desc))
> +		goto out;
> +
> +	desc->istate &= ~(IRQS_REPLAY | IRQS_WAITING);
> +	kstat_incr_irqs_this_cpu(irq, desc);
> +
> +	/* Mark the IRQ as in progress */
> +	mask_irq(desc);
> +
> +	/*
> +	 * If it's disabled or no action available
> +	 * then just get out of here:
> +	 */
> +	if (unlikely(!desc->action || irqd_irq_disabled(&desc->irq_data))) {
> +		desc->istate |= IRQS_PENDING;
> +		goto out_unmask;

If this handler is used with the lazy disable approach then this goto
causes an irq storm if the interrupt stays active (LEVEL).

So this relies on irq_disable() actually disabling the interrupt at
the hardware level. That really wants a big fat comment if we take
this approach.

Now there is another issue. Assume the following:

CPU 0	     	     	    	CPU 1
handle_spliteoi_irq()
  mask_irq();
     handle_event();
        wake_thread();
  return;

run_thread()
   call_handler();
				disable_irq()
				    irq_disable()
   finalize_oneshot()
     if (disabled)
     	return;

So that particular interrupt gets never acknowledged with a write to
DIR. 

What happens if you enable it again at the hardware level via
enable_irq()? Is it still in dropped priority mode and waits for the
write to DIR forever? That's what I tried to avoid with my approach.

Thanks,

	tglx
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/