Here is a backtrace report about memory leak detected in
gpiochip_setup_dev():

unreferenced object 0xffff88810b406400 (size 512):
  comm "python3", pid 1682, jiffies 4295346908 (age 24.090s)
  backtrace:
    kmalloc_trace
    device_add		device_private_init at drivers/base/core.c:3361
			(inlined by) device_add at drivers/base/core.c:3411
    cdev_device_add
    gpiolib_cdev_register
    gpiochip_setup_dev
    gpiochip_add_data_with_key

gcdev_register() & gcdev_unregister() would call device_add() &
device_del() (no matter CONFIG_GPIO_CDEV is enabled or not) to
register/unregister device.

However, if device_add() succeeds, some resource (like
struct device_private allocated by device_private_init())
is not released by device_del().

Therefore, after device_add() succeeds by gcdev_register(), it
needs to call put_device() to release resource in the error handle
path.

Here we move forward the register of release function, and let it
release every piece of resource by put_device() instead of kfree().

While at it, fix another subtle issue, i.e. when gc->ngpio is equal
to 0, we still call kcalloc() and, in case of further error, kfree()
on the ZERO_PTR pointer, which is not NULL. It's not a bug per se,
but rather waste of the resources and potentially wrong expectation
about contents of the gdev->descs variable.

Fixes: 159f3cd9 ("gpiolib: Defer gpio device setup until after gpiolib initialization")
Signed-off-by: Zeng Heng <zengheng4@huawei.com>
Co-developed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>