Re: [PATCH] gpio: Document the 'valid_mask' being internal

[Matti Vaittinen]

On 28/02/2025 11:42, Matti Vaittinen wrote:
> On 28/02/2025 11:28, Matti Vaittinen wrote:
> > CC: Geert (because, I think he was asked about the Rcar GPIO check 
> > before).
> >
> > On 28/02/2025 10:23, Linus Walleij wrote:
> > > On Thu, Feb 27, 2025 at 9:24 AM Matti Vaittinen
> > > <mazziesaccount@xxxxxxxxx> wrote:
>
> > > The call graph should look like this:
> > >
> > > devm_gpiod_get_array()
> > >      gpiod_get_array()
> > >          gpiod_get_index(0...n)
> > >              gpiod_find_and_request()
> > >                  gpiod_request()
> > >                      gpiod_request_commit()
> >
> > Here in my setup the guard.gc->request == NULL. Thus the code never 
> > goes to the branch with the validation. And just before you ask me why 
> > the guard.gc->request is NULL - what do you call a blind bambi? :)
> >   - No idea.
>
> Oh, I suppose the 'guard.gc' is just the chip structure. So, these 
> validity checks are only applied if the gc provides the request 
> callback? As far as I understand, the request callback is optional, and 
> thus the validity check for GPIOs may be omitted.
>
> > >                          gpiochip_line_is_valid()

Would something like this be appropriate? It seems to work "on my 
machine" :) Do you see any unwanted side-effects?

+++ b/drivers/gpio/gpiolib.c
@@ -2315,6 +2315,10 @@ static int gpiod_request_commit(struct gpio_desc 
*desc, const char *label)
        if (!guard.gc)
                return -ENODEV;

+       offset = gpio_chip_hwgpio(desc);
+       if (!gpiochip_line_is_valid(guard.gc, offset))
+               return -EINVAL;
+
        if (test_and_set_bit(FLAG_REQUESTED, &desc->flags))
                return -EBUSY;

@@ -2323,11 +2327,7 @@ static int gpiod_request_commit(struct gpio_desc 
*desc, const char *label)
         */

        if (guard.gc->request) {
-               offset = gpio_chip_hwgpio(desc);
-               if (gpiochip_line_is_valid(guard.gc, offset))
-                       ret = guard.gc->request(guard.gc, offset);
-               else
-                       ret = -EINVAL;
+               ret = guard.gc->request(guard.gc, offset);
                if (ret)
                        goto out_clear_bit;
        }

I can craft a formal patch if this seems reasonable.

Yours,
	-- Matti