Re: [PATCH v2 1/7] ARM: mmp: parse irq from DT

[Grant Likely <grant.likely@secretlab.ca>]

On Mon, Aug 1, 2011 at 3:42 PM, Haojian Zhuang <haojian.zhuang@gmail.com> wrote:
> On Mon, Aug 1, 2011 at 10:10 PM, Grant Likely <grant.likely@secretlab.ca> wrote:
>> On Mon, Aug 01, 2011 at 10:47:06AM +0800, Haojian Zhuang wrote:
>>> On Sat, Jul 30, 2011 at 12:36 AM, Grant Likely
>>> <grant.likely@secretlab.ca> wrote:
>>> >> diff --git a/arch/arm/mach-mmp/common.h b/arch/arm/mach-mmp/common.h
>>> >> index ec8d65d..1c563c2 100644
>>> >> --- a/arch/arm/mach-mmp/common.h
>>> >> +++ b/arch/arm/mach-mmp/common.h
>>> >> @@ -6,3 +6,4 @@ extern void timer_init(int irq);
>>> >>
>>> >>  extern void __init icu_init_irq(void);
>>> >>  extern void __init mmp_map_io(void);
>>> >> +extern void __init mmp_init_intc(void);
>>> >> diff --git a/arch/arm/mach-mmp/include/mach/irqs.h b/arch/arm/mach-mmp/include/mach/irqs.h
>>> >> index a09d328..65ec176 100644
>>> >> --- a/arch/arm/mach-mmp/include/mach/irqs.h
>>> >> +++ b/arch/arm/mach-mmp/include/mach/irqs.h
>>> >> @@ -224,6 +224,6 @@
>>> >>
>>> >>  #define IRQ_BOARD_START                      (IRQ_GPIO_START + IRQ_GPIO_NUM)
>>> >>
>>> >> -#define NR_IRQS                              (IRQ_BOARD_START)
>>> >> +#define NR_IRQS                              0
>>> >
>>> > Why is NR_IRQs getting changed?  I think this will break
>>> > !CONFIG_SPARSE_IRQS, and it shouldn't be necessary for this
>>> > conversion.
>>> >
>>> If CONFIG_SPARSE_IRQ is enabled, arch_probe_nr_irqs() returns NR_IRQS in arm.
>>> It results registering NR_IRQS in early_irq_init(). If NR_IRQS is 200,
>>> 200 irqs are
>>> registered in early_irq_init(). It's not my requirement. What I need
>>> is registering irq
>>> from DT.
>>>
>>> So I have to define NR_IRQS to 0. And CONFIG_SPARSE_IRQS is always enabled
>>> in ARCH_MMP. So it's selected in Kconfig by default.
>>
>> My point is, that applying this patch will break anyone depending on
>> non-DT mach-mmp support.  I'm completely fine with you doing so, but
>> you need to be extra careful that it is done in a bisectable way.  At
>> no point in the commit series should the kernel be unable to build a
>> working mmp image.
>>
>> g.
>>
>>
>
> Since .nr_irqs property is assigned in machine descriptor of brownstone.c or
> ttc_dkb.c, nr_irqs equals to machine_desc->nr_irqs (arch/arm/kernel/irq.c).
>
> Even NR_IRQS is defined as 0, machine_desc->nr_irqs can help us to
> pre-allocate irq numbers while CONFIG_OF isn't defined.
>
> So it's not an issue in ARCH_MMP. And I tested it that everything is well.

Okay, that answers my question.  Thanks.

g.