2023-09-26 14:26:43

by Chengfeng Ye

Subject: [PATCH] gpio: timberdale: Fix potential deadlock on &tgpio->lock

As timbgpio_irq_enable()/timbgpio_irq_disable() callback could be
executed under irq context, it could introduce double locks on
&tgpio->lock if it preempts other execution units requiring
the same locks.

timbgpio_gpio_set()
--> timbgpio_update_bit()
--> spin_lock(&tgpio->lock)
<interrupt>
--> timbgpio_irq_disable()
--> spin_lock_irqsave(&tgpio->lock)

This flaw was found by an experimental static analysis tool I am
developing for irq-related deadlock.

To prevent the potential deadlock, the patch uses spin_lock_irqsave()
on &tgpio->lock inside timbgpio_gpio_set() to prevent the possible
deadlock scenario.

Signed-off-by: Chengfeng Ye <[email protected]>
---
drivers/gpio/gpio-timberdale.c | 5 +++--
1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/gpio/gpio-timberdale.c b/drivers/gpio/gpio-timberdale.c
index bbd9e9191199..fad979797486 100644
--- a/drivers/gpio/gpio-timberdale.c
+++ b/drivers/gpio/gpio-timberdale.c
@@ -43,9 +43,10 @@ static int timbgpio_update_bit(struct gpio_chip *gpio, unsigned index,
unsigned offset, bool enabled)
{
struct timbgpio *tgpio = gpiochip_get_data(gpio);
+ unsigned long flags;
u32 reg;

- spin_lock(&tgpio->lock);
+ spin_lock_irqsave(&tgpio->lock, flags);
reg = ioread32(tgpio->membase + offset);

if (enabled)
@@ -54,7 +55,7 @@ static int timbgpio_update_bit(struct gpio_chip *gpio, unsigned index,
reg &= ~(1 << index);

iowrite32(reg, tgpio->membase + offset);
- spin_unlock(&tgpio->lock);
+ spin_unlock_irqrestore(&tgpio->lock, flags);

return 0;
}
--
2.17.1

2023-09-27 07:05:05

by Chengfeng Ye

Subject: Re: [PATCH] gpio: timberdale: Fix potential deadlock on &tgpio->lock

Thanks much for the review!

Best,
Chengfeng

2023-09-27 11:06:58

by Bartosz Golaszewski

Subject: Re: [PATCH] gpio: timberdale: Fix potential deadlock on &tgpio->lock

On Tue, Sep 26, 2023 at 12:29 PM Chengfeng Ye <[email protected]> wrote:
>
> As timbgpio_irq_enable()/timbgpio_irq_disable() callback could be
> executed under irq context, it could introduce double locks on
> &tgpio->lock if it preempts other execution units requiring
> the same locks.
>
> timbgpio_gpio_set()
> --> timbgpio_update_bit()
> --> spin_lock(&tgpio->lock)
> <interrupt>
> --> timbgpio_irq_disable()
> --> spin_lock_irqsave(&tgpio->lock)
>
> This flaw was found by an experimental static analysis tool I am
> developing for irq-related deadlock.
>
> To prevent the potential deadlock, the patch uses spin_lock_irqsave()
> on &tgpio->lock inside timbgpio_gpio_set() to prevent the possible
> deadlock scenario.
>
> Signed-off-by: Chengfeng Ye <[email protected]>
> ---
> drivers/gpio/gpio-timberdale.c | 5 +++--
> 1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpio/gpio-timberdale.c b/drivers/gpio/gpio-timberdale.c
> index bbd9e9191199..fad979797486 100644
> --- a/drivers/gpio/gpio-timberdale.c
> +++ b/drivers/gpio/gpio-timberdale.c
> @@ -43,9 +43,10 @@ static int timbgpio_update_bit(struct gpio_chip *gpio, unsigned index,
> unsigned offset, bool enabled)
> {
> struct timbgpio *tgpio = gpiochip_get_data(gpio);
> + unsigned long flags;
> u32 reg;
>
> - spin_lock(&tgpio->lock);
> + spin_lock_irqsave(&tgpio->lock, flags);
> reg = ioread32(tgpio->membase + offset);
>
> if (enabled)
> @@ -54,7 +55,7 @@ static int timbgpio_update_bit(struct gpio_chip *gpio, unsigned index,
> reg &= ~(1 << index);
>
> iowrite32(reg, tgpio->membase + offset);
> - spin_unlock(&tgpio->lock);
> + spin_unlock_irqrestore(&tgpio->lock, flags);
>
> return 0;
> }
> --
> 2.17.1
>

Applied, thanks!

Bart