Re: [PATCH kernel v4 2/8] genirq/irqdomain: Clean legacy IRQ allocation

[Thomas Gleixner <tglx@xxxxxxxxxxxxx>]

Alexey,

On Tue, Nov 24 2020 at 17:17, Alexey Kardashevskiy wrote:
> There are 10 users of __irq_domain_alloc_irqs() and only one - IOAPIC -
> passes realloc==true. There is no obvious reason for handling this
> specific case in the generic code.

There is also no obvious reason for _NOT_ handling it at the core code.

> This splits out __irq_domain_alloc_irqs_data() to make it clear what
> IOAPIC does and makes __irq_domain_alloc_irqs() cleaner.

That's your interpretation of cleaner.

You need to expose __irq_domain_alloc_irqs_data() for that which is a
core only functionality, so it's not cleaner. It's exposing internals
which are not to be exposed.

The right thing to do is to get rid of the legacy allocation of x86
during early_irq_init() which is possible with the recent restructuring
of the interrupt initialization code in x86. That's a cleanup which will
actually remove code and not expose internals just because.

> This should cause no behavioral change.

Should not cause is a pretty weak statement.

You're missing a nasty detail here. Contrary to the normal irqdomain
rules virq 0 _IS_ valid on x86 for historical reasons and that's not
trivial to change.

Thanks,

        tglx