[PATCH kernel v3] powerpc/xive: Drop deregistered irqs

[Cédric Le Goater]

Hello Alexey,

>> --- a/arch/powerpc/kernel/irq.c
>> +++ b/arch/powerpc/kernel/irq.c
>> @@ -632,10 +632,13 @@ void __do_irq(struct pt_regs *regs)
>>       may_hard_irq_enable();
>>         /* And finally process it */
>> -    if (unlikely(!irq))
>> +    if (unlikely(!irq)) {
>>           __this_cpu_inc(irq_stat.spurious_irqs);
>> -    else
>> -        generic_handle_irq(irq);
>> +    } else if (generic_handle_irq(irq)) {
>> +        if (ppc_md.orphan_irq)
>> +            ppc_md.orphan_irq(irq);
>> +        __this_cpu_inc(irq_stat.spurious_irqs);
>> +    }

I think we still have an issue here. 

The fix is relying on the fact that generic_handle_irq() will return 
EINVAL if irq desc is NULL, and if this is the case the top interrupt 
handler will consider we have an orphan interrupt.

But, we could also be in the middle of irq_domain->map() and have a 
partially initialized descriptor if the same interrupt is being remapped 
on a CPU while another is trying to handle orphans. 

Calling can_request_irq() (which checks the IRQ_NOREQUEST flag) should 
close that window and it would be clearer than relying on the return value 
of generic_handle_irq().  

C.