[PATCH] gpiolib: Fix irq_disable() semantics

[PATCH] gpiolib: Fix irq_disable() semantics

[Linus Walleij]

The implementation if .irq_disable() which kicks in between
the gpiolib and the driver is not properly mimicking the
expected semantics of the irqchip core: the irqchip will
call .irq_disable() if that exists, else it will call
mask_irq() which first checks if .irq_mask() is defined
before calling it.

Since we are calling it unconditionally, we get this bug
from drivers/pinctrl/qcom/pinctrl-ssbi-gpio.c, as it only
defines .irq_mask_ack and not .irq_mask:

Unable to handle kernel NULL pointer dereference at virtual address 00000000
pgd = (ptrval)
(...)
PC is at 0x0
LR is at gpiochip_irq_disable+0x20/0x30

Fix this by only calling .irq_mask() if it exists.

Cc: Brian Masney <masneyb@onstation.org>
Cc: Hans Verkuil <hans.verkuil@cisco.com>
Cc: stable@vger.kernel.org
Fixes: 461c1a7d4733 ("gpiolib: override irq_enable/disable")
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
---
 drivers/gpio/gpiolib.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
index bdbc1649eafa..d0bb962f42d5 100644
--- a/drivers/gpio/gpiolib.c
+++ b/drivers/gpio/gpiolib.c
@@ -2169,9 +2169,16 @@ static void gpiochip_irq_disable(struct irq_data *d)
 {
 	struct gpio_chip *chip = irq_data_get_irq_chip_data(d);
 
+	/*
+	 * Since we override .irq_disable() we need to mimic the
+	 * behaviour of __irq_disable() in irq/chip.c.
+	 * First call .irq_disable() if it exists, else mimic the
+	 * behaviour of mask_irq() which calls .irq_mask() if
+	 * it exists.
+	 */
 	if (chip->irq.irq_disable)
 		chip->irq.irq_disable(d);
-	else
+	else if (chip->irq.chip->irq_mask)
 		chip->irq.chip->irq_mask(d);
 	gpiochip_disable_irq(chip, d->hwirq);
 }
-- 
2.24.1

Re: [PATCH] gpiolib: Fix irq_disable() semantics

[Bartosz Golaszewski]

pt., 6 mar 2020 o 14:23 Linus Walleij <linus.walleij@linaro.org> napisał(a):
>
> The implementation if .irq_disable() which kicks in between
> the gpiolib and the driver is not properly mimicking the
> expected semantics of the irqchip core: the irqchip will
> call .irq_disable() if that exists, else it will call
> mask_irq() which first checks if .irq_mask() is defined
> before calling it.
>
> Since we are calling it unconditionally, we get this bug
> from drivers/pinctrl/qcom/pinctrl-ssbi-gpio.c, as it only
> defines .irq_mask_ack and not .irq_mask:
>
> Unable to handle kernel NULL pointer dereference at virtual address 00000000
> pgd = (ptrval)
> (...)
> PC is at 0x0
> LR is at gpiochip_irq_disable+0x20/0x30
>
> Fix this by only calling .irq_mask() if it exists.
>
> Cc: Brian Masney <masneyb@onstation.org>
> Cc: Hans Verkuil <hans.verkuil@cisco.com>
> Cc: stable@vger.kernel.org
> Fixes: 461c1a7d4733 ("gpiolib: override irq_enable/disable")
> Signed-off-by: Linus Walleij <linus.walleij@linaro.org>
> ---
>  drivers/gpio/gpiolib.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpio/gpiolib.c b/drivers/gpio/gpiolib.c
> index bdbc1649eafa..d0bb962f42d5 100644
> --- a/drivers/gpio/gpiolib.c
> +++ b/drivers/gpio/gpiolib.c
> @@ -2169,9 +2169,16 @@ static void gpiochip_irq_disable(struct irq_data *d)
>  {
>         struct gpio_chip *chip = irq_data_get_irq_chip_data(d);
>
> +       /*
> +        * Since we override .irq_disable() we need to mimic the
> +        * behaviour of __irq_disable() in irq/chip.c.
> +        * First call .irq_disable() if it exists, else mimic the
> +        * behaviour of mask_irq() which calls .irq_mask() if
> +        * it exists.
> +        */
>         if (chip->irq.irq_disable)
>                 chip->irq.irq_disable(d);
> -       else
> +       else if (chip->irq.chip->irq_mask)
>                 chip->irq.chip->irq_mask(d);
>         gpiochip_disable_irq(chip, d->hwirq);
>  }
> --
> 2.24.1
>

Reviewed-by: Bartosz Golaszewski <bgolaszewski@baylibre.com>