Re: [PATCH v3 1/2] irq: add irq_domain support to generic-chip

[Rob Herring <robherring2@gmail.com>]

On 01/31/2012 09:06 AM, Shawn Guo wrote:
> On Tue, Jan 31, 2012 at 08:32:29AM -0600, Rob Herring wrote:
>> On 01/31/2012 08:13 AM, Shawn Guo wrote:
>>> On Mon, Jan 30, 2012 at 11:31:38AM -0600, Rob Herring wrote:
>>> ...
>>>> +#ifdef CONFIG_IRQ_DOMAIN
>>>> +static int irq_gc_irq_domain_match(struct irq_domain *d, struct device_node *np)
>>>> +{
>>>> +	struct irq_chip_generic *gc;
>>>> +
>>>> +	if (d->of_node != NULL && d->of_node == np) {
>>>> +		list_for_each_entry(gc, &gc_list, list) {
>>>> +			if ((gc == d->host_data) && (d == gc->domain))
>>>> +				return 1;
>>>> +		}
>>>> +	}
>>>
>>> IIRC, we talked about this a little bit, but I'm still unsure how this
>>> works for imx5 tzic case, where we have the same one tzic device_node
>>> for 4 irqdomains representing 128 irq lines.  It seems to me the match
>>> function here will always find the first irqdomain of the 4 for any
>>> of those 128 tzic irqs.
>>>
>>> The following is my code change against your branch for testing.  Am I
>>> missing anything?
>>
>> The irq domain code is a bit different now, so it's matching differently
>> than before. See the match function. The host_data ptr for a domain is
>> set to the gc ptr. I then check that the gc->domain matches the domain
>> passed in. So the fact that 4 domains point to 1 device_node doesn't matter.
>>
> I do not quite follow on that.  The gc->domain always matches the
> domain passed in anyway for those 4 pairs of domain/gc.  That said,
> the condition is all true for any of those 4 tzic domains.
> 
> 	if ((gc == d->host_data) && (d == gc->domain))
> 
> It does not help on identifying the correct domain from those 4 with
> given device_node, which is exactly same for those 4 domains.
> 

Crap. You're right. It worked in my head... ;)

Let me think about this some more.

>>>
>>> 8<---
>>> diff --git a/arch/arm/mach-imx/imx51-dt.c b/arch/arm/mach-imx/imx51-dt.c
>>> index e1b5edf..45abf11 100644
>>> --- a/arch/arm/mach-imx/imx51-dt.c
>>> +++ b/arch/arm/mach-imx/imx51-dt.c
>>> @@ -44,13 +44,6 @@ static const struct of_dev_auxdata
>>> imx51_auxdata_lookup[] __initconst = {
>>>         { /* sentinel */ }
>>>  };
>>>
>>> -static int __init imx51_tzic_add_irq_domain(struct device_node *np,
>>> -                               struct device_node *interrupt_parent)
>>> -{
>>> -       irq_domain_add_legacy(np, 32, 0, 0, &irq_domain_simple_ops, NULL);
>>> -       return 0;
>>> -}
>>> -
>>>  static int __init imx51_gpio_add_irq_domain(struct device_node *np,
>>>                                 struct device_node *interrupt_parent)
>>>  {
>>> @@ -63,7 +56,6 @@ static int __init imx51_gpio_add_irq_domain(struct
>>> device_node *np,
>>>  }
>>>
>>>  static const struct of_device_id imx51_irq_match[] __initconst = {
>>> -       { .compatible = "fsl,imx51-tzic", .data = imx51_tzic_add_irq_domain, },
>>>         { .compatible = "fsl,imx51-gpio", .data = imx51_gpio_add_irq_domain, },
>>>         { /* sentinel */ }
>>>  };
>>>
>>> diff --git a/arch/arm/plat-mxc/tzic.c b/arch/arm/plat-mxc/tzic.c
>>> index 98308ec..ffb615d 100644
>>> --- a/arch/arm/plat-mxc/tzic.c
>>> +++ b/arch/arm/plat-mxc/tzic.c
>>> @@ -122,7 +122,9 @@ static __init void tzic_init_gc(unsigned int irq_start)
>>>         ct->regs.disable = TZIC_ENCLEAR0(idx);
>>>         ct->regs.enable = TZIC_ENSET0(idx);
>>>
>>> -       irq_setup_generic_chip(gc, IRQ_MSK(32), 0, IRQ_NOREQUEST, 0);
>>> +       irq_setup_generic_chip_domain(gc,
>>> +               of_find_compatible_node(NULL, NULL, "fsl,tzic"),
>>> +               IRQ_MSK(32), 0, IRQ_NOREQUEST, 0);
>>
>> Looks right, but I wouldn't lookup the node ptr every time.
>>
> Yeah, will avoid it in the final patch.
> 
>>>  }
>>>
>>>  asmlinkage void __exception_irq_entry tzic_handle_irq(struct pt_regs *regs)
>>> --->8
>>>
>>>> +	return 0;
>>>> +}
>>>> +
>>> ...
>>>> +void irq_setup_generic_chip_domain(struct irq_chip_generic *gc,
>>>> +			    struct device_node *node, u32 msk,
>>>> +			    enum irq_gc_flags flags, unsigned int clr,
>>>> +			    unsigned int set)
>>>> +{
>>>> +	struct irq_chip_type *ct = gc->chip_types;
>>>> +
>>>> +	if (!node) {
>>>> +		irq_setup_generic_chip(gc, msk, flags, clr, set);
>>>> +		return;
>>>> +	}
>>>> +
>>>> +	raw_spin_lock(&gc_lock);
>>>> +	list_add_tail(&gc->list, &gc_list);
>>>> +	raw_spin_unlock(&gc_lock);
>>>> +
>>>> +	/* Init mask cache ? */
>>>> +	if (flags & IRQ_GC_INIT_MASK_CACHE)
>>>> +		gc->mask_cache = irq_reg_readl(gc->reg_base + ct->regs.mask);
>>>> +
>>>> +	gc->flags = flags;
>>>> +	gc->irq_clr = clr;
>>>> +	gc->irq_set = set;
>>>> +
>>>> +	/* Users of domains should not use irq_base */
>>>> +	if ((int)gc->irq_base > 0)
>>>> +		gc->domain = irq_domain_add_legacy(node, fls(msk),
>>>> +						   gc->irq_base, 0,
>>>> +						   &irq_gc_irq_domain_ops, gc);
>>>> +	else {
>>>> +		gc->irq_base = 0;
>>>> +		gc->domain = irq_domain_add_linear(node, fls(msk),
>>>> +						   &irq_gc_irq_domain_ops, gc);
>>>> +	}
>>>
>>> We have 4 generic_chips for tzic with irq_base as 0, 32, 64, 96.  In
>>> this case, we end up with having the first domain as the linear, and
>>> the other 3 as the legacy?
>>
>> Umm, yes. So it should be '>= 0' instead until we stop using IRQ0. I
>> could base it on NULL node ptr instead...
>>
>> For the DT case, you want irq_base to be -1.
>>
> Unless we have to, I would keep the same tzic code for dt and non-dt
> to save #ifdef CONFIG_OF.

You need to stop using Linux vIRQ0 as a valid vIRQ# and with DT, all
knowledge about irq_base should be removed. Whether that can be
accomplished with the same code is a separate issue.

Rob