Re: [PATCH V3 5/5] rtc: max77686: move initialisation of rtc regmap, irq chip locally

[Javier Martinez Canillas]

Hello Laxman,

On 02/04/2016 07:13 AM, Laxman Dewangan wrote:
> On Thursday 04 February 2016 02:38 PM, Krzysztof Kozlowski wrote:
> > On 04.02.2016 15:58, Krzysztof Kozlowski wrote:
> > > On 04.02.2016 11:03, Laxman Dewangan wrote:
> > > > 3. Can you try locally to not use devm_regmap_init_i2c() and just use
> > > > the regmap_init_i2c() and proper removal of this from error path and
> > > > remove callback?
> > > I'll try to find some time for that. Maybe tomorrow.
> > regmap_init_i2c does not help. However helps commenting out the:
> >     regmap_del_irq_chip(info->rtc_irq, info->rtc_irq_data);
> > from remove() callback.
>
> Thanks for debugging.  I do not see anything on regmap_del_irq_chip() which can be suspected. Is this because of sharing interrupt between mfd and rtc driver?

I don't think so, as long as all the callers pass the IRQF_SHARED flag,
there shouldn't be issues if different drivers call request / free irq.

AFAIK the only thing to take into account is that all callers should
pass a non-NULL cookie (the void * parameter in request and free irq)
so the kernel has a way to differentiate at request / free time.

And this is the case for regmap irq since the regmap_irq_chip_data *
that is passed to regmap_{add,del}_irq_chip() is used as the cookie.

> What is your suggestion here?  Should we avoid this inside RTC driver (creating this in mfd) and pass this information to the rtc driver using customized platform driver?

That wouldn't be the right approach IMHO, that would just mask the real
issue by making the call to regmap_del_irq_chip() to never happen since
the MFD device is not usually unbound due regulators being too critical.

I think we should find the real cause why this is happening and fix it.

Best regards,
--
Javier Martinez Canillas
Open Source Group
Samsung Research America