Re: [PATCH] gpio: gpiolib: fix refcount imbalance in gpiochip_setup_dev()

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Thu, Dec 5, 2024 at 2:20 PM Andy Shevchenko
<andriy.shevchenko@xxxxxxxxx> wrote:
>
> On Wed, Dec 04, 2024 at 09:21:52PM +0900, Joe Hattori wrote:
> > In gpiochip_setup_dev(), the refcount incremented in device_initialize()
> > is not decremented in the error path. Fix it by calling put_device().
>
> First of all, we have gpio_put_device().
> Second, what the problem do you have in practice? Can you show any backtrace?
> Third, how had this change been tested?
>
> Looking at the current code I noticed the following:
> 1) gpiochip_add_data_with_key() has already that call;
> 2) gpiochip_setup_devs() misses that call.
>
> This effectively means that you inroduce a regression while fixing a bug.
>
> The GPIO device initialisation is non-trivial, please pay more attention to the
> code.
>
> Bart, can this be removed or reverted before it poisons stable?
>

Yes, sorry, should have paid more attention.

Bartosz