[PATCH v4 09/19] ARM: use common irqchip_init for GIC init

[robherring2@gmail.com (Rob Herring)]

On 02/12/2013 10:52 AM, Doug Anderson wrote:
> Rob,
> 
> 
> On Thu, Jan 3, 2013 at 2:04 PM, Rob Herring <robherring2@gmail.com> wrote:
>> On 01/03/2013 03:41 PM, Arnd Bergmann wrote:
>>> On Thursday 03 January 2013, Rob Herring wrote:
>>>> @@ -637,8 +638,6 @@ static int __init combiner_of_init(struct device_node *np,
>>>>  }
>>>>
>>>>  static const struct of_device_id exynos_dt_irq_match[] = {
>>>> -       { .compatible = "arm,cortex-a9-gic", .data = gic_of_init, },
>>>> -       { .compatible = "arm,cortex-a15-gic", .data = gic_of_init, },
>>>>         { .compatible = "samsung,exynos4210-combiner",
>>>>                         .data = combiner_of_init, },
>>>>         {},
>>>> @@ -654,8 +653,10 @@ void __init exynos4_init_irq(void)
>>>>         if (!of_have_populated_dt())
>>>>                 gic_init_bases(0, IRQ_PPI(0), S5P_VA_GIC_DIST, S5P_VA_GIC_CPU, gic_bank_offset, NULL);
>>>>  #ifdef CONFIG_OF
>>>> -       else
>>>> +       else {
>>>> +               irqchip_init();
>>>>                 of_irq_init(exynos_dt_irq_match);
>>>> +       }
>>>>  #endif
>>>
>>> Is this an equivalent change? I thought you are supposed to call of_irq_init
>>> only once so it can see the entire hierarchy. Or is it ok as long as the
>>> top-level controllers are probed first?
>>
>> Yes, it is equivalent as the GIC will be initialized first. Ideally,
>> this the exynos controller needs to be moved to drivers/irqchip as well
>> and this can be removed, but I can't have all the fun.
> 
> My ARM Chromebook that no longer boots on linux-next begs to differ.
> I tracked it down to this change.  If I hack it so that the exynos
> combiner is registered with irqchip or if I revert just the exynos
> part of this patch then things work again.
> 
> Is anyone from Samsung working to move the exynos-combiner into IRQ
> Chip?  Until that happens (or this change is partly reverted) any
> exynos boards won't boot.  ...and tracking the problem down to this
> change is a bit of a pain since there's no nice crash.

Actually, I was mistaken and of_irq_init can't really be called twice.
The parents need to be in the match list as well. I will fix this and
there's probably a few others broken as well.

Rob