Re: [PATCH v3 1/6] irqchip: add support for Marvell Orion SoCs

[Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>]

On 06/11/13 15:45, Thomas Gleixner wrote:
> On Tue, 11 Jun 2013, Sebastian Hesselbarth wrote:
>> On 06/11/13 15:30, Thomas Gleixner wrote:
>>> On Tue, 11 Jun 2013, Thomas Gleixner wrote:
>>>> On Thu, 6 Jun 2013, Sebastian Hesselbarth wrote:
>>>>> This patch adds an irqchip driver for the main interrupt controller
>>>>> found
>>>>> on Marvell Orion SoCs (Kirkwood, Dove, Orion5x, Discovery Innovation).
>>>>> Corresponding device tree documentation is also added.
>>>>>
>>>>> Signed-off-by: Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>
>>>>
>>>> Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
>>>
>>> Second thoughts:
>>>
>>>> +static void orion_bridge_irq_handler(unsigned int irq, struct irq_desc
>>>> *desc)
>>>> +{
>>>> +	struct irq_domain *d = irq_get_handler_data(irq);
>>>> +	struct irq_chip_generic *gc = irq_get_domain_generic_chip(d, irq);
>>>> +	u32 stat = readl_relaxed(gc->reg_base + ORION_BRIDGE_IRQ_CAUSE) &
>>>> +		gc->mask_cache;
>>>
>>> In init you map the first irq of that chip and install the chain
>>> handler for it. Now if that first irq fires, isn't that set in the
>>> cause register as well? And what acks that first irq?
>>
>> It is "acked" by acking all unmasked bridge irqs.
>
> Ok. A comment would be nice.
>
> But what about the bit in of that first irq in the cause register? If
> it's set on entry you call generic_handle_irq() for that as well. So
> if it's set you need to mask it in stat. If not, then it wants a
> comment.

I am not sure I can follow. orion_bridge_irq_init() maps the first
parent irq, i.e. hwirq 0 of orion_irq. The parent irq controller
clears that irq cause when all corresponding chained irqs are
cleared. The chained (bridge) irqs are cleared by
orion_bridge_irq_handler above.

I can put a note to the parent irqchip in orion_irq_init() that we
don't need to clear its cause register, if it is that what you mean?

Sebastian