Re: [PATCH] sparse_irq aka dyn_irq v13

[Yinghai Lu <yinghai@kernel.org>]

Ingo Molnar wrote:

> it still looks ugly to me: couldnt we make it completely #ifdef-free, 
> by just adding the new API variants?
> 
> i.e. leave these present unconditionally:
> 
>>  extern void hpet_msi_unmask(unsigned int irq);
>>  extern void hpet_msi_mask(unsigned int irq);
> 
> and just _add_ these (unconditionally):
> 
>> +extern void hpet_msi_unmask_desc(unsigned int irq, struct irq_desc **descp);
>> +extern void hpet_msi_mask_desc(unsigned int irq, struct irq_desc **descp);
> 
> that gives us zero #ifdefs and much nicer to read patches. Am i 
> missing something why this isnt possible?

in io_apic.c

struct irq_chip hpet_msi_type = {
        .name = "HPET_MSI",
        .unmask = hpet_msi_unmask,
        .mask = hpet_msi_mask,
        .ack = ack_apic_edge,
#ifdef CONFIG_SMP
        .set_affinity = hpet_msi_set_affinity,
#endif
        .retrigger = ioapic_retrigger_irq,
};

so those ack, mask, unmak, eoi field will have different function prototype..when sparseirq is enabled or not.

if you want to remove those #ifdef, we need to go over all irq_chip definition to make all those field to take struct irq_desc *desc (or **descp) instead of unsigned int irq.

actually we only need to pass desc struct instead of irq, because we can get desc->irq in case.

YH