Re: [PATCH] pinctrl: qcom: Release pin ranges when gpiochip_irqchip_add fails

[Bjorn Andersson <bjorn.andersson@sonymobile.com>]

On Thu 28 Aug 00:13 PDT 2014, Pramod Gurav wrote:

> On Thursday 28 August 2014 02:54 AM, Bjorn Andersson wrote:
> > On Wed, Aug 27, 2014 at 3:57 AM, Pramod Gurav
> > <pramod.gurav@smartplayin.com> wrote:
> >> This patches adds a call to gpiochip_remove_pin_ranges when
> >> gpiochip_irqchip_add fails to release memory allocated for pin_ranges.
> >>
> >> diff --git a/drivers/pinctrl/qcom/pinctrl-msm.c b/drivers/pinctrl/qcom/pinctrl-msm.c
> >> @@ -845,6 +845,7 @@ static int msm_gpio_init(struct msm_pinctrl *pctrl)
> >>                                    IRQ_TYPE_NONE);
> >>         if (ret) {
> >>                 dev_err(pctrl->dev, "Failed to add irqchip to gpiochip\n");
> >> +               gpiochip_remove_pin_ranges(chip);
> >>                 return -ENOSYS;
> >>         }
> > 
> > Good catch, I guess this was lost in the introduction of gpiochip_irqchip...
> > 
> > 
> > Rather than just releasing the pin_ranges of the gpio_chip you should
> > probably add a gpiochip_remove() both here and in the case of
> > gpiochip_add_pin_range() failing.
> 
> Thanks for review. But if I see implementation of gpiochip_remove() it does:
> 	gpiochip_irqchip_remove(chip);
>         gpiochip_remove_pin_ranges(chip);
>         of_gpiochip_remove(chip);
> 
> In above failure case only gpiochip_add() and gpiochip_add_pin_range()
> have been successful hence I thought that would cause any problem to add
> gpiochip_remove(). If that is not a problem I think we can call
> gpiochip_remove() in fail case of gpiochip_add_pin_range() as well.
> Do I make sense?
> 

As soon as gpiochip_add() have returned successfully we will have a live
gpio_chip, upon returning unsuccessfully from probe devres will free the pctrl
node and the gpio core will continue to operate on freed memory.

Therefor we need to call gpio_remove() in the ccase of both
gpiochip_add_pin_range() and gpiochip_irqchip_add() failing.

The gpio_remove() does as you say remove those additional items, but handles
the case where they are not yet "allocated".

I hope this answers your conserns.

Regards,
Bjorn