gpiolib: don't check the retval of get_direction() when registering a chip

[PATCH] gpiolib: don't check the retval of get_direction() when registering a chip

Posted by Bartosz Golaszewski 9 months, 1 week ago

From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>

During chip registration we should neither check the return value of
gc->get_direction() nor hold the SRCU lock when calling it. The former
is because pin controllers may have pins set to alternate functions and
return errors from their get_direction() callbacks. That's alright - we
should default to the safe INPUT state and not bail-out. The latter is
not needed because we haven't registered the chip yet so there's nothing
to protect against dynamic removal. In fact: we currently hit a lockdep
splat. Revert to calling the gc->get_direction() callback directly and
*not* checking its value.

Fixes: 9d846b1aebbe ("gpiolib: check the return value of gpio_chip::get_direction()")
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Closes: https://lore.kernel.org/all/81f890fc-6688-42f0-9756-567efc8bb97a@samsung.com/
Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Link: https://lore.kernel.org/r/20250226-retval-fixes-v2-1-c8dc57182441@linaro.org
Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
---
This commit is in linux-next as 9becde08f1bc ("gpiolib: don't use
gpiochip_get_direction() when registering a chip") and was applied as
a fix to commits e623c4303ed1 ("gpiolib: sanitize the return value of
gpio_chip::get_direction()") and 9d846b1aebbe ("gpiolib: check the return
value of gpio_chip::get_direction()"). Becuase the former is queued for
v6.15-rc1, this fix was never applied to v6.14 and sent upstream.

However, the warning it addresses is now queued for v6.14. I've rebased
this commit on top of v6.14-rc6 and would like to send it upstream. Once
merged, I'll pull v6.14-rc7 back into my for-next branch and fix the
conflicts.

 drivers/gpio/gpiolib.c | 27 +++++++++++----------------
 1 file changed, 11 insertions(+), 16 deletions(-)

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index 8741600af7ef..de708d081858 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -1056,24 +1056,19 @@ int gpiochip_add_data_with_key(struct gpio_chip *gc, void *data,
 
 		desc->gdev = gdev;
 
-		if (gc->get_direction && gpiochip_line_is_valid(gc, desc_index)) {
-			ret = gc->get_direction(gc, desc_index);
-			if (ret < 0)
-				/*
-				 * FIXME: Bail-out here once all GPIO drivers
-				 * are updated to not return errors in
-				 * situations that can be considered normal
-				 * operation.
-				 */
-				dev_warn(&gdev->dev,
-					 "%s: get_direction failed: %d\n",
-					 __func__, ret);
-
-			assign_bit(FLAG_IS_OUT, &desc->flags, !ret);
-		} else {
+		/*
+		 * We would typically want to check the return value of
+		 * get_direction() here but we must not check the return value
+		 * and bail-out as pin controllers can have pins configured to
+		 * alternate functions and return -EINVAL. Also: there's no
+		 * need to take the SRCU lock here.
+		 */
+		if (gc->get_direction && gpiochip_line_is_valid(gc, desc_index))
+			assign_bit(FLAG_IS_OUT, &desc->flags,
+				   !gc->get_direction(gc, desc_index));
+		else
 			assign_bit(FLAG_IS_OUT,
 				   &desc->flags, !gc->direction_input);
-		}
 	}
 
 	ret = of_gpiochip_add(gc);
-- 
2.45.2

Re: [PATCH] gpiolib: don't check the retval of get_direction() when registering a chip

Posted by Bartosz Golaszewski 9 months, 1 week ago

From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>


On Tue, 11 Mar 2025 18:56:31 +0100, Bartosz Golaszewski wrote:
> During chip registration we should neither check the return value of
> gc->get_direction() nor hold the SRCU lock when calling it. The former
> is because pin controllers may have pins set to alternate functions and
> return errors from their get_direction() callbacks. That's alright - we
> should default to the safe INPUT state and not bail-out. The latter is
> not needed because we haven't registered the chip yet so there's nothing
> to protect against dynamic removal. In fact: we currently hit a lockdep
> splat. Revert to calling the gc->get_direction() callback directly and
> *not* checking its value.
> 
> [...]

Applied, thanks!

[1/1] gpiolib: don't check the retval of get_direction() when registering a chip
      commit: 0102fbf52b93e609fec0dab53b1fb4fe69113f5e

Best regards,
-- 
Bartosz Golaszewski <bartosz.golaszewski@linaro.org>

Re: [PATCH] gpiolib: don't check the retval of get_direction() when registering a chip

Posted by Genes Lists 9 months, 1 week ago

On Tue, 2025-03-11 at 18:56 +0100, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> 
> During chip registration we should neither check the return value of
> gc->get_direction() nor hold the SRCU lock when calling it. The
> former
> is because pin controllers may have pins set to alternate functions
> and
> return errors from their get_direction() callbacks. That's alright -
> we
> should default to the safe INPUT state and not bail-out. The latter
> is
> not needed because we haven't registered the chip yet so there's
> nothing
> to protect against dynamic removal. In fact: we currently hit a
> lockdep
> splat. Revert to calling the gc->get_direction() callback directly
> and
> *not* checking its value.


Tested-by: Gene C <arch@sapience.com>


> 
> Fixes: 9d846b1aebbe ("gpiolib: check the return value of
> gpio_chip::get_direction()")
> Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
> Closes:
> https://lore.kernel.org/all/81f890fc-6688-42f0-9756-567efc8bb97a@samsung.com/
> Reviewed-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
> Link:
> https://lore.kernel.org/r/20250226-retval-fixes-v2-1-c8dc57182441@linaro.org
> Signed-off-by: Bartosz Golaszewski <bartosz.golaszewski@linaro.org>
> ---
> This commit is in linux-next as 9becde08f1bc ("gpiolib: don't use
> gpiochip_get_direction() when registering a chip") and was applied as
> a fix to commits e623c4303ed1 ("gpiolib: sanitize the return value of
> gpio_chip::get_direction()") and 9d846b1aebbe ("gpiolib: check the
> return
> value of gpio_chip::get_direction()"). Becuase the former is queued
> for
> v6.15-rc1, this fix was never applied to v6.14 and sent upstream.
> 
> However, the warning it addresses is now queued for v6.14. I've
> rebased
> this commit on top of v6.14-rc6 and would like to send it upstream.
> Once
> merged, I'll pull v6.14-rc7 back into my for-next branch and fix the
> conflicts.
> 
>  drivers/gpio/gpiolib.c | 27 +++++++++++----------------
>  1 file changed, 11 insertions(+), 16 deletions(-)
> 
> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> index 8741600af7ef..de708d081858 100644
> --- a/drivers/gpio/gpiolib.c
> +++ b/drivers/gpio/gpiolib.c
> @@ -1056,24 +1056,19 @@ int gpiochip_add_data_with_key(struct
> gpio_chip *gc, void *data,
>  
>   desc->gdev = gdev;
>  
> - if (gc->get_direction && gpiochip_line_is_valid(gc,
> desc_index)) {
> - ret = gc->get_direction(gc, desc_index);
> - if (ret < 0)
> - /*
> - * FIXME: Bail-out here once all GPIO
> drivers
> - * are updated to not return errors in
> - * situations that can be considered
> normal
> - * operation.
> - */
> - dev_warn(&gdev->dev,
> - "%s: get_direction failed:
> %d\n",
> - __func__, ret);
> -
> - assign_bit(FLAG_IS_OUT, &desc->flags, !ret);
> - } else {
> + /*
> + * We would typically want to check the return value
> of
> + * get_direction() here but we must not check the
> return value
> + * and bail-out as pin controllers can have pins
> configured to
> + * alternate functions and return -EINVAL. Also:
> there's no
> + * need to take the SRCU lock here.
> + */
> + if (gc->get_direction && gpiochip_line_is_valid(gc,
> desc_index))
> + assign_bit(FLAG_IS_OUT, &desc->flags,
> +    !gc->get_direction(gc,
> desc_index));
> + else
>   assign_bit(FLAG_IS_OUT,
>      &desc->flags, !gc-
> >direction_input);
> - }
>   }
>  
>   ret = of_gpiochip_add(gc);

-- 
Gene