diff mbox series

[v1] gpiolib: Don't WARN on gpiod_put() for optional GPIO

Message ID 20190326152114.40591-1-andriy.shevchenko@linux.intel.com
State New
Headers show
Series [v1] gpiolib: Don't WARN on gpiod_put() for optional GPIO | expand

Commit Message

Andy Shevchenko March 26, 2019, 3:21 p.m. UTC
In case of debug and optional GPIO requested, the gpiod_put() is not aware of
and will WARN, which is not the case.

Make gpiod_put() NULL-aware to keep silent for optional GPIOs.

Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
---
 drivers/gpio/gpiolib.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Comments

Bartosz Golaszewski March 26, 2019, 5:48 p.m. UTC | #1
wt., 26 mar 2019 o 16:21 Andy Shevchenko
<andriy.shevchenko@linux.intel.com> napisał(a):
>
> In case of debug and optional GPIO requested, the gpiod_put() is not aware of
> and will WARN, which is not the case.
>
> Make gpiod_put() NULL-aware to keep silent for optional GPIOs.
>
> Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> ---
>  drivers/gpio/gpiolib.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> index 144af0733581..36445e24ee89 100644
> --- a/drivers/gpio/gpiolib.c
> +++ b/drivers/gpio/gpiolib.c
> @@ -4616,7 +4616,8 @@ EXPORT_SYMBOL_GPL(gpiod_get_array_optional);
>   */
>  void gpiod_put(struct gpio_desc *desc)
>  {
> -       gpiod_free(desc);
> +       if (desc)
> +               gpiod_free(desc);
>  }
>  EXPORT_SYMBOL_GPL(gpiod_put);
>
> --
> 2.20.1
>

In that case I don't think we need to keep checking desc for NULL in
gpiod_free().

Bart
Andy Shevchenko March 26, 2019, 7:44 p.m. UTC | #2
On Tue, Mar 26, 2019 at 06:48:25PM +0100, Bartosz Golaszewski wrote:
> wt., 26 mar 2019 o 16:21 Andy Shevchenko
> <andriy.shevchenko@linux.intel.com> napisał(a):
> >
> > In case of debug and optional GPIO requested, the gpiod_put() is not aware of
> > and will WARN, which is not the case.
> >
> > Make gpiod_put() NULL-aware to keep silent for optional GPIOs.
> >
> > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > ---
> >  drivers/gpio/gpiolib.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> > index 144af0733581..36445e24ee89 100644
> > --- a/drivers/gpio/gpiolib.c
> > +++ b/drivers/gpio/gpiolib.c
> > @@ -4616,7 +4616,8 @@ EXPORT_SYMBOL_GPL(gpiod_get_array_optional);
> >   */
> >  void gpiod_put(struct gpio_desc *desc)
> >  {
> > -       gpiod_free(desc);
> > +       if (desc)
> > +               gpiod_free(desc);
> >  }
> >  EXPORT_SYMBOL_GPL(gpiod_put);
> >
> > --
> > 2.20.1
> >
> 
> In that case I don't think we need to keep checking desc for NULL in
> gpiod_free().

I didn't remove the check there because it's called more than just from
gpiod_put(). I didn't check all code paths myself. So, if it seems good for you
to remove the check I can do it in next version.
Andy Shevchenko April 2, 2019, 10:49 a.m. UTC | #3
On Tue, Mar 26, 2019 at 09:44:20PM +0200, Andy Shevchenko wrote:
> On Tue, Mar 26, 2019 at 06:48:25PM +0100, Bartosz Golaszewski wrote:
> > wt., 26 mar 2019 o 16:21 Andy Shevchenko
> > <andriy.shevchenko@linux.intel.com> napisał(a):
> > >
> > > In case of debug and optional GPIO requested, the gpiod_put() is not aware of
> > > and will WARN, which is not the case.
> > >
> > > Make gpiod_put() NULL-aware to keep silent for optional GPIOs.
> > >
> > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > > ---
> > >  drivers/gpio/gpiolib.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> > > index 144af0733581..36445e24ee89 100644
> > > --- a/drivers/gpio/gpiolib.c
> > > +++ b/drivers/gpio/gpiolib.c
> > > @@ -4616,7 +4616,8 @@ EXPORT_SYMBOL_GPL(gpiod_get_array_optional);
> > >   */
> > >  void gpiod_put(struct gpio_desc *desc)
> > >  {
> > > -       gpiod_free(desc);
> > > +       if (desc)
> > > +               gpiod_free(desc);
> > >  }
> > >  EXPORT_SYMBOL_GPL(gpiod_put);
> > >
> > > --
> > > 2.20.1
> > >
> > 
> > In that case I don't think we need to keep checking desc for NULL in
> > gpiod_free().
> 
> I didn't remove the check there because it's called more than just from
> gpiod_put(). I didn't check all code paths myself. So, if it seems good for you
> to remove the check I can do it in next version.

Any additional comments on this? I have looked briefly at this and found that
gpiod_free() is used to support legacy gpio_free() which is used in a plenty of
drivers. There are more than dozen are using non static number for these calls.

I think the patch as is is the best approach we can do right now.
Bartosz Golaszewski April 4, 2019, 7:34 a.m. UTC | #4
wt., 2 kwi 2019 o 12:49 Andy Shevchenko
<andriy.shevchenko@linux.intel.com> napisał(a):
>
> On Tue, Mar 26, 2019 at 09:44:20PM +0200, Andy Shevchenko wrote:
> > On Tue, Mar 26, 2019 at 06:48:25PM +0100, Bartosz Golaszewski wrote:
> > > wt., 26 mar 2019 o 16:21 Andy Shevchenko
> > > <andriy.shevchenko@linux.intel.com> napisał(a):
> > > >
> > > > In case of debug and optional GPIO requested, the gpiod_put() is not aware of
> > > > and will WARN, which is not the case.
> > > >
> > > > Make gpiod_put() NULL-aware to keep silent for optional GPIOs.
> > > >
> > > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > > > ---
> > > >  drivers/gpio/gpiolib.c | 3 ++-
> > > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> > > > index 144af0733581..36445e24ee89 100644
> > > > --- a/drivers/gpio/gpiolib.c
> > > > +++ b/drivers/gpio/gpiolib.c
> > > > @@ -4616,7 +4616,8 @@ EXPORT_SYMBOL_GPL(gpiod_get_array_optional);
> > > >   */
> > > >  void gpiod_put(struct gpio_desc *desc)
> > > >  {
> > > > -       gpiod_free(desc);
> > > > +       if (desc)
> > > > +               gpiod_free(desc);
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(gpiod_put);
> > > >
> > > > --
> > > > 2.20.1
> > > >
> > >
> > > In that case I don't think we need to keep checking desc for NULL in
> > > gpiod_free().
> >
> > I didn't remove the check there because it's called more than just from
> > gpiod_put(). I didn't check all code paths myself. So, if it seems good for you
> > to remove the check I can do it in next version.
>
> Any additional comments on this? I have looked briefly at this and found that
> gpiod_free() is used to support legacy gpio_free() which is used in a plenty of
> drivers. There are more than dozen are using non static number for these calls.
>
> I think the patch as is is the best approach we can do right now.
>
> --
> With Best Regards,
> Andy Shevchenko
>
>

Now applied.

Bart
diff mbox series

Patch

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index 144af0733581..36445e24ee89 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -4616,7 +4616,8 @@  EXPORT_SYMBOL_GPL(gpiod_get_array_optional);
  */
 void gpiod_put(struct gpio_desc *desc)
 {
-	gpiod_free(desc);
+	if (desc)
+		gpiod_free(desc);
 }
 EXPORT_SYMBOL_GPL(gpiod_put);