[v3,2/2] gpiolib: check the 'ngpios' property in core gpiolib code

Several drivers read the 'ngpios' device property on their own, but
since it's defined as a standard GPIO property in the device tree bindings
anyway, it's a good candidate for generalization. If the driver didn't
set its gc->ngpio, try to read the 'ngpios' property from the GPIO
device's firmware node before bailing out.

Signed-off-by: Bartosz Golaszewski <brgl@bgdev.pl>
Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
---
v1 -> v2:
- use device_property_read_u32() instead of fwnode_property_read_u32()
- reverse the error check logic

v2 -> v3:
- don't shadow errors other than -ENODATA in device_property_read_u32()

 drivers/gpio/gpiolib.c | 15 ++++++++++++---
 1 file changed, 12 insertions(+), 3 deletions(-)

On Wed, Nov 24, 2021 at 01:28:50PM +0100, Bartosz Golaszewski wrote:
> Several drivers read the 'ngpios' device property on their own, but
> since it's defined as a standard GPIO property in the device tree bindings
> anyway, it's a good candidate for generalization. If the driver didn't
> set its gc->ngpio, try to read the 'ngpios' property from the GPIO
> device's firmware node before bailing out.

> Signed-off-by: Bartosz Golaszewski <brgl@bgdev.pl>
> Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
> ---
> v1 -> v2:
> - use device_property_read_u32() instead of fwnode_property_read_u32()
> - reverse the error check logic
> 
> v2 -> v3:
> - don't shadow errors other than -ENODATA in device_property_read_u32()
> 
>  drivers/gpio/gpiolib.c | 15 ++++++++++++---
>  1 file changed, 12 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> index ede8b8a7aa18..f79fd2551cf7 100644
> --- a/drivers/gpio/gpiolib.c
> +++ b/drivers/gpio/gpiolib.c
> @@ -599,6 +599,7 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
>  	int base = gc->base;
>  	unsigned int i;
>  	int ret = 0;
> +	u32 ngpios;
>  
>  	/*
>  	 * First: allocate and populate the internal stat container, and
> @@ -647,9 +648,17 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
>  	}

>  	if (gc->ngpio == 0) {
> -		chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> -		ret = -EINVAL;
> -		goto err_free_descs;
> +		ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> +		if (ret) {
> +			if (ret == -ENODATA) {
> +				chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> +				ret = -EINVAL;
> +			}
> +
> +			goto err_free_descs;
> +		}

And if the property returns 0 in ngpios?

What about the modified suggestion from previous version:

	if (gc->ngpio == 0) {
		ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
		/*
		 * -ENODATA means that there is no property found and
		 * we want to issue the error message to the user. Besides
		 * that, we want to return different error code to state
		 * that supplied value is not valid.
		 */
		if (ret == -ENODATA)
			ngpios = 0;
		else if (ret)
			return ret;

		gc->ngpio = ngpios;
	}

	if (gc->ngpio == 0) {
		chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
		ret = -EINVAL;
		goto err_free_descs;
	}

?

> +		gc->ngpio = ngpios;
>  	}
>  
>  	if (gc->ngpio > FASTPATH_NGPIO)
> -- 
> 2.25.1
>

On Wed, Nov 24, 2021 at 04:46:58PM +0200, Andy Shevchenko wrote:
> On Wed, Nov 24, 2021 at 01:28:50PM +0100, Bartosz Golaszewski wrote:
> > Several drivers read the 'ngpios' device property on their own, but
> > since it's defined as a standard GPIO property in the device tree bindings
> > anyway, it's a good candidate for generalization. If the driver didn't
> > set its gc->ngpio, try to read the 'ngpios' property from the GPIO
> > device's firmware node before bailing out.

...

> What about the modified suggestion from previous version:
> 
> 	if (gc->ngpio == 0) {
> 		ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> 		/*
> 		 * -ENODATA means that there is no property found and
> 		 * we want to issue the error message to the user. Besides
> 		 * that, we want to return different error code to state
> 		 * that supplied value is not valid.
> 		 */
> 		if (ret == -ENODATA)
> 			ngpios = 0;
> 		else if (ret)
> 			return ret;
> 
> 		gc->ngpio = ngpios;
> 	}
> 
> 	if (gc->ngpio == 0) {
> 		chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> 		ret = -EINVAL;
> 		goto err_free_descs;
> 	}
> 
> ?

If you okay to go with this code,
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>

On Wed, Nov 24, 2021 at 3:47 PM Andy Shevchenko
<andriy.shevchenko@linux.intel.com> wrote:
>
> On Wed, Nov 24, 2021 at 01:28:50PM +0100, Bartosz Golaszewski wrote:
> > Several drivers read the 'ngpios' device property on their own, but
> > since it's defined as a standard GPIO property in the device tree bindings
> > anyway, it's a good candidate for generalization. If the driver didn't
> > set its gc->ngpio, try to read the 'ngpios' property from the GPIO
> > device's firmware node before bailing out.
>
> > Signed-off-by: Bartosz Golaszewski <brgl@bgdev.pl>
> > Suggested-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > Reviewed-by: Linus Walleij <linus.walleij@linaro.org>
> > ---
> > v1 -> v2:
> > - use device_property_read_u32() instead of fwnode_property_read_u32()
> > - reverse the error check logic
> >
> > v2 -> v3:
> > - don't shadow errors other than -ENODATA in device_property_read_u32()
> >
> >  drivers/gpio/gpiolib.c | 15 ++++++++++++---
> >  1 file changed, 12 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> > index ede8b8a7aa18..f79fd2551cf7 100644
> > --- a/drivers/gpio/gpiolib.c
> > +++ b/drivers/gpio/gpiolib.c
> > @@ -599,6 +599,7 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> >       int base = gc->base;
> >       unsigned int i;
> >       int ret = 0;
> > +     u32 ngpios;
> >
> >       /*
> >        * First: allocate and populate the internal stat container, and
> > @@ -647,9 +648,17 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
> >       }
>
> >       if (gc->ngpio == 0) {
> > -             chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> > -             ret = -EINVAL;
> > -             goto err_free_descs;
> > +             ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> > +             if (ret) {
> > +                     if (ret == -ENODATA) {
> > +                             chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> > +                             ret = -EINVAL;
> > +                     }
> > +
> > +                     goto err_free_descs;
> > +             }
>
> And if the property returns 0 in ngpios?
>
> What about the modified suggestion from previous version:
>
>         if (gc->ngpio == 0) {
>                 ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);

The comment is a good idea but other than that - it's overcomplicating things.

>                 if (ret == -ENODATA)
>                         ngpios = 0;
>                 else if (ret)
>                         return ret;

You still need to goto err_free_descs here.

>
>                 gc->ngpio = ngpios;
>         }
>
>         if (gc->ngpio == 0) {

Why check that again? We already know the driver set it to 0, we
checked it a couple lines before. If we can't get the setting from the
properties then it won't be non 0 here right?

>                 chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
>                 ret = -EINVAL;
>                 goto err_free_descs;
>         }
>
> ?
>
> > +             gc->ngpio = ngpios;
> >       }
> >
> >       if (gc->ngpio > FASTPATH_NGPIO)
> > --
> > 2.25.1
> >
>
> --
> With Best Regards,
> Andy Shevchenko
>
>

I suggest the following:

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index ede8b8a7aa18..08c1e8fc0dfa 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -599,6 +599,7 @@ int gpiochip_add_data_with_key(struct gpio_chip
*gc, void *data,
     int base = gc->base;
     unsigned int i;
     int ret = 0;
+    u32 ngpios;

     /*
      * First: allocate and populate the internal stat container, and
@@ -647,9 +648,23 @@ int gpiochip_add_data_with_key(struct gpio_chip
*gc, void *data,
     }

     if (gc->ngpio == 0) {
-        chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
-        ret = -EINVAL;
-        goto err_free_descs;
+        /*
+         * If the driver didn't specify the number of GPIOs, try to
+         * get this value from the device properties as a fall-back.
+         */
+        ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
+        if (ret) {
+            /*
+             * -ENODATA means that there is no property found and
+             * we want to issue the error message to the user.
+             * Besides that, we want to return different error code
+             * to state that supplied value is not valid.
+             * */
+            if (ret == -ENODATA) {
+                chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
+                ret = -EINVAL;
+            }
+
+            goto err_free_descs;
+        }
+
+        gc->ngpio = ngpios;
     }

     if (gc->ngpio > FASTPATH_NGPIO)

On Thu, Nov 25, 2021 at 11:10:08AM +0100, Bartosz Golaszewski wrote:
> On Wed, Nov 24, 2021 at 3:47 PM Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> wrote:
> > On Wed, Nov 24, 2021 at 01:28:50PM +0100, Bartosz Golaszewski wrote:

...

> > > +             ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> > > +             if (ret) {
> > > +                     if (ret == -ENODATA) {
> > > +                             chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> > > +                             ret = -EINVAL;
> > > +                     }
> > > +
> > > +                     goto err_free_descs;
> > > +             }
> >
> > And if the property returns 0 in ngpios?
> >
> > What about the modified suggestion from previous version:
> >
> >         if (gc->ngpio == 0) {
> >                 ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);

> The comment is a good idea but other than that - it's overcomplicating things.

I don't think so. It is plain and self-explaining each step. See at the end of
the message how.

> >                 if (ret == -ENODATA)
> >                         ngpios = 0;
> >                 else if (ret)
> >                         return ret;

> You still need to goto err_free_descs here.

Right, this doesn't affect the main point / logic here.

> >                 gc->ngpio = ngpios;
> >         }
> >
> >         if (gc->ngpio == 0) {
> 
> Why check that again? We already know the driver set it to 0, we
> checked it a couple lines before. If we can't get the setting from the
> properties then it won't be non 0 here right?

No, it's not right. The check is needed to tell that properties supplied 0.

> >                 chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> >                 ret = -EINVAL;
> >                 goto err_free_descs;
> >         }
> >
> > ?
> >
> > > +             gc->ngpio = ngpios;
> > >       }
> > >
> > >       if (gc->ngpio > FASTPATH_NGPIO)

> I suggest the following:

It's buggy as submitted version (I actually haven't found any difference in
the code, but comments).

You see, I propose less changes and straight forward logic:

1. Check if the supplied ->ngpio equal to 0
2. If so, try device properties
2.1. If there is no property found, make sure we a) don't use uninitialized
     variable, b) we don't change ->ngpio, so it stays 0
2.2. If there is an error, return it as is to the caller
2.3. Assign ->ngpio by value from property (which very well may be 0!)
3. Check ->ngpio for 0 again, if so, issue a message and return -EINVAL to
the user.

We have three places where ->ngpio can be 0, all of them I covered.

On Thu, Nov 25, 2021 at 12:18 PM Andy Shevchenko
<andriy.shevchenko@linux.intel.com> wrote:
>
> On Thu, Nov 25, 2021 at 11:10:08AM +0100, Bartosz Golaszewski wrote:
> > On Wed, Nov 24, 2021 at 3:47 PM Andy Shevchenko
> > <andriy.shevchenko@linux.intel.com> wrote:
> > > On Wed, Nov 24, 2021 at 01:28:50PM +0100, Bartosz Golaszewski wrote:
>
> ...
>
> > > > +             ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
> > > > +             if (ret) {
> > > > +                     if (ret == -ENODATA) {
> > > > +                             chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> > > > +                             ret = -EINVAL;
> > > > +                     }
> > > > +
> > > > +                     goto err_free_descs;
> > > > +             }
> > >
> > > And if the property returns 0 in ngpios?
> > >
> > > What about the modified suggestion from previous version:
> > >
> > >         if (gc->ngpio == 0) {
> > >                 ret = device_property_read_u32(&gdev->dev, "ngpios", &ngpios);
>
> > The comment is a good idea but other than that - it's overcomplicating things.
>
> I don't think so. It is plain and self-explaining each step. See at the end of
> the message how.
>
> > >                 if (ret == -ENODATA)
> > >                         ngpios = 0;
> > >                 else if (ret)
> > >                         return ret;
>
> > You still need to goto err_free_descs here.
>
> Right, this doesn't affect the main point / logic here.
>
> > >                 gc->ngpio = ngpios;
> > >         }
> > >
> > >         if (gc->ngpio == 0) {
> >
> > Why check that again? We already know the driver set it to 0, we
> > checked it a couple lines before. If we can't get the setting from the
> > properties then it won't be non 0 here right?
>
> No, it's not right. The check is needed to tell that properties supplied 0.
>
> > >                 chip_err(gc, "tried to insert a GPIO chip with zero lines\n");
> > >                 ret = -EINVAL;
> > >                 goto err_free_descs;
> > >         }
> > >
> > > ?
> > >
> > > > +             gc->ngpio = ngpios;
> > > >       }
> > > >
> > > >       if (gc->ngpio > FASTPATH_NGPIO)
>
> > I suggest the following:
>
> It's buggy as submitted version (I actually haven't found any difference in
> the code, but comments).
>
> You see, I propose less changes and straight forward logic:
>
> 1. Check if the supplied ->ngpio equal to 0
> 2. If so, try device properties
> 2.1. If there is no property found, make sure we a) don't use uninitialized
>      variable, b) we don't change ->ngpio, so it stays 0
> 2.2. If there is an error, return it as is to the caller
> 2.3. Assign ->ngpio by value from property (which very well may be 0!)

Ok, this is the thing I didn't realize, the property can indeed be 0
so your approach is correct.

Bart

> 3. Check ->ngpio for 0 again, if so, issue a message and return -EINVAL to
> the user.
>
> We have three places where ->ngpio can be 0, all of them I covered.
>
> --
> With Best Regards,
> Andy Shevchenko
>
>