Re: RFC Proposal to enable MSI support in Linux kernel

[Zwane Mwaikambo]

On Wed, 28 May 2003, Jeff Garzik wrote:

> > +#ifdef CONFIG_PCI_USE_VECTOR
> > +	for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
> > +#else
> >  	for (i = 0; i < NR_IRQS; i++) {
> > +#endif
> 
> This ifdef isn't necessary.  ifdef in a header somewhere.  For example, 
> make NR_IRQS (or some other constant, if changing NR_IRQS definition 
> isn't ok) condition on the CONFIG_xxx options.

Actually we can simply use;

for (i = 0; i < (NR_VECTORS - FIRST_EXTERNAL_VECTOR); i++) {
	vector = ...
	if (i >= NR_IRQS)
		break;
	...
	set_intr_gate(vector, interrupt[i]);
}

> split into two functions, maybe?  (due to the ifdef)  Your call, it's 
> mainly a programmer preference.

I agree.

> A bigger question though:  if platform_legacy_irq() returns zero, will 
> the handler _ever_ be edge-triggered?

Good question, i wouldn't think so, that would collapse that section into 
two lines.

> As I see more and more of these ifdefs, I wonder if they couldn't be 
> hidden inside wrappers?

Yes some of them are a bit easy to botch up later on.. like the following 
excerpt.

> > +#ifdef CONFIG_PCI_USE_VECTOR
> > +			if (!platform_legacy_irq(irq))
> > +				entry->irq = IO_APIC_VECTOR(irq);
> > +			else
> > +#endif
> >  			entry->irq = irq;
> >  			continue;

Thanks,
	Zwane
-- 
function.linuxpower.ca