Re: [Regression] "irqdomain: Don't set type when mapping an IRQ" breaks nexus7 gpio buttons

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

On 11/08/16 14:29, Jon Hunter wrote:
> 
> On 11/08/16 13:46, Marc Zyngier wrote:
>> On 11/08/16 10:47, Jon Hunter wrote:
>>>
>>> On 11/08/16 09:37, Marc Zyngier wrote:
>>>> On 08/08/16 22:48, Linus Walleij wrote:
>>>>> On Sat, Aug 6, 2016 at 1:45 AM, John Stultz <john.stultz@linaro.org> wrote:
>>>>>
>>>>>> @@ -614,7 +615,11 @@ unsigned int irq_create_fwspec_mapping(struct
>>>>>> irq_fwspec *fwspec)
>>>>>>                  * it now and return the interrupt number.
>>>>>>                  */
>>>>>>                 if (irq_get_trigger_type(virq) == IRQ_TYPE_NONE) {
>>>>>> -                       irq_set_irq_type(virq, type);
>>>>>> +                       irq_data = irq_get_irq_data(virq);
>>>>>> +                       if (!irq_data)
>>>>>> +                               return 0;
>>>>>> +
>>>>>> +                       irqd_set_trigger_type(irq_data, type);
>>>>>>                         return virq;
>>>>>>                 }
>>>>>>
>>>>>> If I revert just that, it works again.
>>>>>
>>>>> This makes my platform work too.
>>>>> Tested-by: Linus Walleij <linus.walleij@linaro.org>
>>>>
>>>> Hmmm. I'm now booting your kernel on the APQ8060, and reverting this
>>>> hunk doesn't fix it for me. I'm confused...
>>>>
>>>> The interesting part is this:
>>>> 109:     100000          0   msmgpio  88 Level     (null)
>>>
>>> 88 is the pm8058 parent interrupt and so I am surprised you would even
>>> see this in /proc/interrupts as it should be a chained interrupt, right?
>>>
>>> Are you seeing this with all the ethernet updates for the APQ8060 in
>>> Linus' branch? I am curious what you see with stock v4.8-rc1 and if
>>> interrupts work ok with the change I had proposed. Hard to tell if there
>>> is more than one issue here.
>>
>> Nailed the sucker:
> 
> Great!
> 
>> diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
>> index b4c1bc7..9d7284a 100644
>> --- a/kernel/irq/chip.c
>> +++ b/kernel/irq/chip.c
>> @@ -820,6 +820,18 @@ __irq_do_set_handler(struct irq_desc *desc, irq_flow_handler_t handle,
>>  	desc->name = name;
>>  
>>  	if (handle != handle_bad_irq && is_chained) {
>> +		int ret;
>> +
>> +		ret = __irq_set_trigger(desc,
>> +					irqd_get_trigger_type(&desc->irq_data));
>> +		WARN_ON(ret);
> 
> You could wrap the entire call in the WARN_ON(). I was not sure if there
> was a better way to handle that.

Actually, I've decided to drop it. We already have a message in
__irq_set_trigger(), and if we really want to scream, that's the one we
should consider upgrading to a WARN_ON().

> 
>> +		/*
>> +		 * This is beyond ugly: .set_type may have overridden
>> +		 * the flow, not not knowing that we're dealing with a
>> +		 * chained handler. Reset it here because we know
>> +		 * better.
>> +		 */
>> +		desc->handle_irq = handle;
> 
> Yes I see the call to irq_set_handler in the pinctrl-msm.c set_type.
> Good catch!

I can't believe it took me that much time to realize that. Guess I need
an extended weekend! ;-)

> 
> Apart from the above ...
> 
> Acked-by: Jon Hunter <jonathanh@nvidia.com>

Thanks a lot Jon,

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