[PATCH v6] irq: add irq_domain support to generic-chip

[Rob Herring]

On 02/09/2012 01:48 PM, Shawn Guo wrote:
> On Wed, Feb 08, 2012 at 04:55:22PM -0600, Rob Herring wrote:
>> From: Rob Herring <rob.herring at calxeda.com>
>>
>> Add irq domain support to irq generic-chip. This enables users of
>> generic-chip to support dynamic irq assignment needed for DT interrupt
>> binding.
>>
>> Thanks to Shawn Guo for fixes and testing.
>>
>> Signed-off-by: Rob Herring <rob.herring at calxeda.com>
>> Cc: Grant Likely <grant.likely at secretlab.ca>
>> Cc: Thomas Gleixner <tglx at linutronix.de>
>> ---
>> Here's the latest version. This has fixes from Shawn Guo, so should be
>> working. This version is also available here:
>>
> Rob, yes it's working.  But I think it should be working in a better
> way after I play it a little bit more.  See comment below.
> ...
>> +int irq_setup_generic_chip_domain(const char *name, struct device_node *node,
>> +				int num_ct, unsigned int irq_base,
>> +				void __iomem *reg_base,
>> +				irq_flow_handler_t handler, u32 hwirq_cnt,
>> +				enum irq_gc_flags flags, unsigned int clr,
>> +				unsigned int set,
>> +				void (*gc_init_cb)(struct irq_chip_generic *),
>> +				void *private)
>> +{
>> +	int ret, i;
>> +	struct irq_chip_generic **gc;
>> +	struct irq_domain *d;
>> +	int num_gc = ALIGN(hwirq_cnt, 32) / 32;
>> +
>> +	gc = kzalloc(num_gc * sizeof(struct irq_chip_generic *), GFP_KERNEL);
>> +	if (!gc)
>> +		return -ENOMEM;
>> +
>> +	for (i = 0; i < num_gc; i++) {
>> +		gc[i] = irq_alloc_generic_chip(name, num_ct, irq_base,
>> +			       reg_base, handler);
>> +		if (!gc[i]) {
>> +			ret = -ENOMEM;
>> +			goto err;
>> +		}
>> +		gc[i]->hwirq_base = i * 32;
>> +		gc[i]->private = private;
>> +
>> +		gc_init_cb(gc[i]);
>> +
>> +		irq_setup_generic_chip(gc[i], 0, flags, clr, set);
>> +	}
>> +
>> +	if (node)
>> +		d = irq_domain_add_linear(node, hwirq_cnt,
>> +					  &irq_gc_irq_domain_ops, gc);
>> +	else
>> +		d = irq_domain_add_legacy(node, hwirq_cnt, irq_base, 0,
>> +					  &irq_gc_irq_domain_ops, gc);
>> +
> I do not think it's good to make this decision based on whether it's
> dt or non-dt case.  That said, it's fairly valid and actually
> encouraged that non-dt users use linear domain too, and also the
> irqdomain core has no limit on dt users to use legacy domain.  Instead,
> it's much more reasonable to make this decision based on if irq_base,
> something like you did in your v3 patch.
> 
>  
> -       if (node)
> +       if ((int) irq_base < 0)
> 

Really, we want to discourage/limit the use of legacy domains. This is
only used for ISA irqs on powerpc. Unfortunately, there's not really a
way to use linear domains for non-DT ATM. But maybe we can fix that.

> Otherwise,
> 
> Acked-by: Shawn Guo <shawn.guo at linaro.org>
> 
> Tested on imx51 with tzic and gpio interrupt controller.

Thanks!

Rob
> 
> Regards,
> Shawn
> 
>> +	for (i = 0; i < num_gc; i++)
>> +		gc[i]->domain = d;
>> +
>> +	return 0;
>> +
>> + err:
>> +	for ( ; i >= 0; i--)
>> +		kfree(gc[i]);
>> +	kfree(gc);
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(irq_setup_generic_chip_domain);
>> +#endif
>> +