[PATCH] mfd: mt6397: irq domain should initialize before mfd_add_devices()

[Henry Chen]

On Thu, 2016-03-31 at 16:07 +0200, John Crispin wrote:
> 
> On 31/03/2016 15:41, Yingjoe Chen wrote:
> > On Thu, 2016-03-31 at 11:08 +0200, John Crispin wrote:
> >>
> >> On 31/03/2016 04:32, Yingjoe Chen wrote:
> >>> On Thu, 2016-03-31 at 09:40 +0800, Henry Chen wrote:
> >>>> On Wed, 2016-03-30 at 11:18 +0200, John Crispin wrote:
> >>>>> Hi,
> >>>>>
> >>>>> small nitpick inline
> >>>>>
> >>>>> On 30/03/2016 09:25, Henry Chen wrote:
> >>>>>> Some sub driver like RTC module need irq domain from parent to create
> >>>>>> irq mapping when driver initialize. so move mt6397_irq_init() before
> >>>>>> mfd_add_devices().
> >>>>>>
> >>>>>> Signed-off-by: Henry Chen <henryc.chen at mediatek.com>
> >>>>>> ---
> >>>>>> This patch fixed the below warning based on "Linux kernel v4.6-rc1"
> >>>>>> WARNING: CPU: 1 PID: 132 at kernel/mediatek/kernel/irq/irqdomain.c:471
> >>>>>> irq_create_mapping+0xc4/0xd0
> >>>>>> ---
> >>>>>>  drivers/mfd/mt6397-core.c | 37 ++++++++++++++++++++++---------------
> >>>>>>  1 file changed, 22 insertions(+), 15 deletions(-)
> >>>>>>
> >>>>>> diff --git a/drivers/mfd/mt6397-core.c b/drivers/mfd/mt6397-core.c
> >>>>>> index 8e8d932..a879223 100644
> >>>>>> --- a/drivers/mfd/mt6397-core.c
> >>>>>> +++ b/drivers/mfd/mt6397-core.c
> >>>>>> @@ -270,22 +270,36 @@ static int mt6397_probe(struct platform_device *pdev)
> >>>>>>  		goto fail_irq;
> >>>>>>  	}
> >>>>>>  
> >>>>>> +	pmic->irq = platform_get_irq(pdev, 0);
> >>>>>> +
> >>>>>>  	switch (id & 0xff) {
> >>>>>>  	case MT6323_CID_CODE:
> >>>>>> -		pmic->int_con[0] = MT6323_INT_CON0;
> >>>>>> -		pmic->int_con[1] = MT6323_INT_CON1;
> >>>>>> -		pmic->int_status[0] = MT6323_INT_STATUS0;
> >>>>>> -		pmic->int_status[1] = MT6323_INT_STATUS1;
> >>>>>> +		if (pmic->irq > 0) {
> >>>>>
> >>>>> should this not be
> >>>>>
> >>>>> 		if (pmic->irq >= 0) {
> >>>>>
> >>>>> i think the code before your patch was wrong as linux irqs start with 0.
> >>>>>
> >>>>> 	John
> >>>> Hi John,
> >>>>
> >>>> Thanks, I will modify this.
> >>>
> >>> Linux irq start from 1, 0 is invalid. I can't find the document saying
> >>> this now, but you could see this from irq_create_mapping() in
> >>> kernel/irq/irqdomain.c
> >>>
> >>> I think the code should have check return from platform_get_irq and
> >>> handle -EPROBE_DEFER, but maybe it should be another patch?
> >>>
> >>> BTW, in this function, it is possible that pmic->irq_domain will be NULL
> >>> in fail_irq error handling. We should check before calling
> >>> irq_domain_remove.
> >>>
> >>> Joe.C
> >>>
> >>
> >> Hi,
> >>
> >> looking at
> >> http://lxr.free-electrons.com/source/drivers/base/platform.c#L87 there
> >> is a check in line #100 ret >= 0
> >>
> >> checking the return value of pmic->irq = platform_get_irq(pdev, 0);
> >> should follow the same pattern i think .. unless i have a thinko and am
> >> reading the code wrong.
> > 
> > 
> > I'm not sure why platform_get_irq() check for 0, but I think the code
> > logic is differnet.
> > 
> > When platform_get_irq() return 0 to our code, it means we don't have
> > valid irq to use. In this case it doesn't make any sense to continue
> > init irq.
> > 
> > 
> > Joe.C
> > 
> 
> 
> --> http://lwn.net/Articles/470820/
> 
> indeed ARM has changed this is seems. was not aware of this change,
> sorry for the noise
> 
> 	John

Hi Lee/Joe/John,

Thanks for comment, that means we can keep this patch for 4.6-rc1
regression (RTC get NULL irq issue), right?

I will send another patch to fixed above error handling problem.

Thanks.
Henry