Re: [PATCH] pinctrl: qcom: spmi-gpio: make the irqchip immutable

[Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx>]

On Wed, Jul 13, 2022 at 02:33:32PM +0200, Robert Marko wrote:
> On Wed, 13 Jul 2022 at 13:47, Marc Zyngier <maz@xxxxxxxxxx> wrote:
> >
> > On 2022-07-13 12:08, Robert Marko wrote:
> > > On Tue, 12 Jul 2022 at 17:12, Marc Zyngier <maz@xxxxxxxxxx> wrote:
> > >>
> > >> On Tue, 12 Jul 2022 13:44:45 +0100,
> > >> Manivannan Sadhasivam <manivannan.sadhasivam@xxxxxxxxxx> wrote:
> > >> >
> > >> > On Tue, Jul 12, 2022 at 11:42:32AM +0100, Marc Zyngier wrote:
> > >> > > On Fri, 24 Jun 2022 20:51:12 +0100,
> > >> > > Robert Marko <robimarko@xxxxxxxxx> wrote:
> > >> > > >
> > >> > > > Commit 6c846d026d49 ("gpio: Don't fiddle with irqchips marked as
> > >> > > > immutable") added a warning to indicate if the gpiolib is altering the
> > >> > > > internals of irqchips.
> > >> > > >
> > >> > > > Following this change the following warning is now observed for the SPMI
> > >> > > > PMIC pinctrl driver:
> > >> > > > gpio gpiochip1: (200f000.spmi:pmic@0:gpio@c000): not an immutable chip, please consider fixing it!
> > >> > > >
> > >> > > > Fix this by making the irqchip in the SPMI PMIC pinctrl driver immutable.
> > >> > > >
> > >> > > > Signed-off-by: Robert Marko <robimarko@xxxxxxxxx>
> > >> > > > ---
> > >> > > >  drivers/pinctrl/qcom/pinctrl-spmi-gpio.c | 22 ++++++++++++----------
> > >> > > >  1 file changed, 12 insertions(+), 10 deletions(-)
> > >> > > >
> > >> > > > diff --git a/drivers/pinctrl/qcom/pinctrl-spmi-gpio.c b/drivers/pinctrl/qcom/pinctrl-spmi-gpio.c
> > >> > > > index c3255b0bece4..406ee0933d0b 100644
> > >> > > > --- a/drivers/pinctrl/qcom/pinctrl-spmi-gpio.c
> > >> > > > +++ b/drivers/pinctrl/qcom/pinctrl-spmi-gpio.c
> > >> > > > @@ -171,7 +171,6 @@ struct pmic_gpio_state {
> > >> > > >   struct regmap   *map;
> > >> > > >   struct pinctrl_dev *ctrl;
> > >> > > >   struct gpio_chip chip;
> > >> > > > - struct irq_chip irq;
> > >> > > >   u8 usid;
> > >> > > >   u8 pid_base;
> > >> > > >  };
> > >> > > > @@ -988,6 +987,17 @@ static void *pmic_gpio_populate_parent_fwspec(struct gpio_chip *chip,
> > >> > > >   return fwspec;
> > >> > > >  }
> > >> > > >
> > >> > > > +static const struct irq_chip spmi_gpio_irq_chip = {
> > >> > > > + .name           = "spmi-gpio",
> > >> > > > + .irq_ack        = irq_chip_ack_parent,
> > >> > > > + .irq_mask       = irq_chip_mask_parent,
> > >> > > > + .irq_unmask     = irq_chip_unmask_parent,
> > >> > >
> > >> > > No, this is wrong. Please look at the documentation to see how you
> > >> > > must now directly call into the gpiolib helpers for these two
> > >> > > callbacks.
> > >> > >
> > >> >
> > >> > IIUC, you are referring to gpiochip_disable_irq() and
> > >> > gpiochip_enable_irq() APIs.
> > >>
> > >> I am indeed.
> > >>
> > >> > These APIs are supposed to let the gpiolib know about that the IRQ
> > >> > usage of these GPIOs. But for the case of hierarchial IRQ domain,
> > >> > isn't the parent is going to do that?
> > >>
> > >> Why would it? The parent has no clue about what sits above it. In a
> > >> hierarchical configuration, each level is responsible for its own
> > >> level, and the GPIO layer should be responsible for its own
> > >> management.
> > >>
> > >> > Please correct me if I'm wrong.
> > >>
> > >> I'm afraid you are, and this patch is a fairly obvious change in
> > >> behaviour, as the callbacks you mention above are not called anymore,
> > >> while they were before.
> > >>
> > >> If they are not necessary (for reasons I can't fathom), then this
> > >> should be clearly explained.
> > >
> > > Hi Marc,
> > > I will look at IRQ GPIO docs, but in this case, then we have more
> > > conversions that
> > > are not correct.
> >
> > Then please point them out.
> 
> Oh, now I get the issue, I was misunderstanding it completely.
> gpiochip_enable_irq and gpiochip_disable_irq are not being called
> at all.
> 
> However, I dont see them being called before the conversion as well.
> I am not really familiar with the PMIC IRQ-s, looked like an easy conversion
> to get rid of the warning.
> 
> Manivannan can you shed some light on this?
> 

I hope you got the answer by now. When I looked into the conversion I saw that
there were missing calls to gpiochip_{enable/disable}_irq APIs. But at that
time I blindly assumed (yeah very bad of myself) that the parent irqchip will
handle that :(

Anyway, you should call these helpers from the mask/unmask callbacks as a part
of the conversion patch. Let me know if you are onto it or not!

Thanks,
Mani

> Regards,
> Robert
> 
> 
> 
> 
> 
> >
> >          M.
> > --
> > Jazz is not dead. It just smells funny...

-- 
மணிவண்ணன் சதாசிவம்