Re: [PATCH 0/2] genirq: Make irqchip flags work with stacked irq domains

[Marc Zyngier <marc.zyngier@arm.com>]

On 13/01/15 10:41, Thomas Gleixner wrote:
> On Fri, 9 Jan 2015, Marc Zyngier wrote:
>> Hi Jiang,
>>
>> On 09/01/15 03:07, Jiang Liu wrote:
>>> On 2015/1/9 1:32, Marc Zyngier wrote:
>>>> With the landing of stacked irq domains in 3.19, we have ended up in a
>>>> situation where we have a stack of IRQ controllers, each with their
>>>> set of flags, but the core code is only able to look at the top-most,
>>>> which is not very helpful. This small series is trying to fix this.
>>>>
>>>> The first patch converts all access to desc->irq_data.chip->flags to
>>>> using an accessor, without changing anything else. The second patch
>>>> adds some logic to combine these flags as we allocate the interrupts,
>>>> ultimately storing the resulting set as part of the irq_desc
>>>> structure.
>>>>
>>>> We end-up with a configuration where the flags can either be located
>>>> in the irq_chip structure (non stacked case), or in the irq_desc
>>>> (stacked case). While this isn't really ideal, this gives at least the
>>>> right level of information to the rest of the IRQ framework.
>>> Hi Mark,
>>> 	By this way, we need to aggregate irq_chip flags for every
>>> irq. How about changing irq_desc_get_chip_flags(struct irq_desc *desc)
>>> to irq_desc_check_chip_flags(struct irq_desc *desc, unsigned int flags)
>>> which dynamically walks the stacked irqchips?
>>
>> That was the other option, but I felt that going through the list of
>> irqchips was not a very nice idea performance wise. This is used is some
>> moderately hot paths (cond_unmask_eoi_irq, for example), and I wondered
>> if we wanted to pay this price at runtime...
> 
> We rather do it at setup time.

That was my thought as well. I'll respin the series with Jiang's fix.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...