On Mon, May 12, 2025 at 04:08:35PM +0200, Thomas Richard wrote:

On 5/9/25 11:07, Geert Uytterhoeven wrote:

quoted

On Tue, 6 May 2025 at 17:21, Thomas Richard [off-list ref] wrote:

...

quoted

quoted

+int gpio_fwd_set_config(struct gpio_chip *chip, unsigned int offset,
+                       unsigned long config);
+
+int gpio_fwd_to_irq(struct gpio_chip *chip, unsigned int offset);

I would expect all of these to take gpiochip_fwd pointers instead of
gpio_chip pointers.  What prevents you from passing a gpio_chip pointer
that does not correspond to a gpiochip_fwd object, causing a crash?

Indeed nothing prevents from passing gpio_chip pointer which does not
correspond to a gpiochip_fwd object.
And it is also a bit weird to pass a gpiochip_fwd pointer in some cases
(for example gpio_fwd_gpio_add()) and a gpio_chip in other cases.

I can keep GPIO operations as is, and create exported wrappers which
take a gpiochip_fwd pointer as parameter, for example:

int gpiochip_fwd_get_multiple(struct gpiochip_fwd *fwd,
			      unsigned long *mask,
			      unsigned long *bits)
{
	struct gpio_chip *gc = gpiochip_fwd_get_gpiochip(fwd);

	return gpio_fwd_get_multiple_locked(chip, mask, bits);
}
EXPORT_SYMBOL_NS_GPL(gpiochip_fwd_get_multiple, "GPIO_FORWARDER");

So exported functions are gpiochip_fwd_*().

Sounds good for me. Let's wait for Geert's opinoin on this.

-- 
With Best Regards,
Andy Shevchenko