Re: [patch] genirq/MSI: restore __do_IRQ() compat logic temporarily

[ebiederm@xmission.com (Eric W. Biederman)]

Ingo Molnar <mingo@elte.hu> writes:

> * Eric W. Biederman <ebiederm@xmission.com> wrote:
>
>> Ok.  Looking at it I almost certain the problem is that
>> we lost the hunk of code removed in: 266f0566761cf88906d634727b3d9fc2556f5cbd
>> i386: Fix stack switching in do_IRQ
>> 
>> -       if (!irq_desc[irq].handle_irq) {
>> -               __do_IRQ(irq, regs);
>> -               goto out_exit;
>> -       }
>> 
>> The msi code does not yet set desc->handle_irq.  So when we attempt to 
>> call it we get a NULL pointer dereference.
>
> indeed ... We thought the MSI cleanup went all the way with the irqchips 
> conversion, that's we suggested to Andrew to drop this chunk in -mm too.

Sorry.  At the time I was trying for a minimal fix to much more
fundamental brain damage.

>> Except for adding that hunk back in and breaking 4K stacks I don't 
>> have an immediate fix.
>
> i've attached a bandaid patch for -mm below. Brandon, does this solve 
> the crash you are seeing?
>
>> I do have a pending cleanup that should result in us setting 
>> handle_irq in all cases.  I will see if I can advance that shortly.
>
> yeah, that's the right solution.

The core problem at the moment is that the generic code in msi.c
still knows about apics.  So struct irq_chip needs to be pushed
to the individual architectures.  At which point it becomes easy to
ensure we have a proper handle_irq value as we aren't trying to be
impossibly generic.

Eric