Re: [PATCH v2 1/3] ARM: clps711x: Add CLPS711X irqchip driver

[Alexander Shiyan]

Hello.

> On Friday 03 January 2014 17:26:45 Alexander Shiyan wrote:
> > > > +   err = irq_alloc_descs(-1, 0, ARRAY_SIZE(clps711x_irqs), numa_node_id());
> > > > +   if (IS_ERR_VALUE(err))
> > > > +           goto out_iounmap;
> > > > +
> > > > +   clps711x_intc->ops.map = clps711x_intc_irq_map;
> > > > +   clps711x_intc->ops.xlate = irq_domain_xlate_onecell;
> > > > +   clps711x_intc->domain =
> > > > +           irq_domain_add_legacy(np, ARRAY_SIZE(clps711x_irqs),
> > > > +                                 0, 0, &clps711x_intc->ops, NULL);
> > > 
> > > Can't you use irq_domain_add_simple() here? When you are booting with DT
> > > and SPARSE_IRQ, you should not need to allocate the descriptors at boot time,
> > > but only as needed.
> > > 
> > > You will have to pass '0' as the first_irq argument for the DT case then,
> > > and '1' for the non-DT case to actually allocate virqs starting at number 1.
> > 
> > Using irq_domain_add_simple() cause double usage of irq_alloc_descs(),
> > so we cannot do it. At least now.
> 
> Well, the idea would be to remove the irq_alloc_descs() from your code,
> and have irq_domain_add_simple() call it for the ATAGS case but not
> at all for the DT case. That is exactly the purpose of '_simple', i.e.
> to let the irq_alloc_descs() part be handled automatically if needed.

OK. In any case this need to be tested on HW (this is not a fast thing :) ),
and such change can be provided later as a patch to current version.

> > PS: Arnd, can you merge this series for 3.14-rc within arm-soc?
> 
> I'd feel more comfortable if Thomas could have a look first and either
> merge it himself or provide an Ack. I'm not much of an irqchip expert
> and I don't want to step on his toes there.

OK, let's wait. I'll notify again about this series within 2 weeks or so.
Thanks.

---