2019-12-19 13:17:36

by Jia-Ju Bai

Subject: [PATCH 1/2 v2] gpio: gpio-grgpio: fix possible sleep-in-atomic-context bugs in grgpio_remove()

drivers/gpio/gpiolib-sysfs.c, 796:
mutex_lock in gpiochip_sysfs_unregister
drivers/gpio/gpiolib.c, 1455:
gpiochip_sysfs_unregister in gpiochip_remove
drivers/gpio/gpio-grgpio.c, 460:
gpiochip_remove in grgpio_remove
drivers/gpio/gpio-grgpio.c, 449:
_raw_spin_lock_irqsave in grgpio_remove

kernel/irq/irqdomain.c, 243:
mutex_lock in irq_domain_remove
drivers/gpio/gpio-grgpio.c, 463:
irq_domain_remove in grgpio_remove
drivers/gpio/gpio-grgpio.c, 449:
_raw_spin_lock_irqsave in grgpio_remove

mutex_lock() can sleep at runtime.

To fix these bugs, the lock is dropped in grgpio_remove(), because there
is no need for locking in remove() callbacks.

These bugs are found by a static analysis tool STCheck written by
myself.

Signed-off-by: Jia-Ju Bai <[email protected]>
---
v2:
* Drop the lock instead of moving the calls to lock functions.
Thank Bartosz for good advice.

---
drivers/gpio/gpio-grgpio.c | 4 ----
1 file changed, 4 deletions(-)

diff --git a/drivers/gpio/gpio-grgpio.c b/drivers/gpio/gpio-grgpio.c
index 08234e64993a..a49f0711ca94 100644
--- a/drivers/gpio/gpio-grgpio.c
+++ b/drivers/gpio/gpio-grgpio.c
@@ -437,8 +437,6 @@ static int grgpio_remove(struct platform_device *ofdev)
int i;
int ret = 0;

- spin_lock_irqsave(&priv->gc.bgpio_lock, flags);
-
if (priv->domain) {
for (i = 0; i < GRGPIO_MAX_NGPIO; i++) {
if (priv->uirqs[i].refcnt != 0) {
@@ -454,8 +452,6 @@ static int grgpio_remove(struct platform_device *ofdev)
irq_domain_remove(priv->domain);

out:
- spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
-
return ret;
}

--
2.17.1

2020-01-07 09:02:15

by Linus Walleij

Subject: Re: [PATCH 1/2 v2] gpio: gpio-grgpio: fix possible sleep-in-atomic-context bugs in grgpio_remove()

On Thu, Dec 19, 2019 at 2:15 PM Jia-Ju Bai <[email protected]> wrote:

> drivers/gpio/gpiolib-sysfs.c, 796:
> mutex_lock in gpiochip_sysfs_unregister
> drivers/gpio/gpiolib.c, 1455:
> gpiochip_sysfs_unregister in gpiochip_remove
> drivers/gpio/gpio-grgpio.c, 460:
> gpiochip_remove in grgpio_remove
> drivers/gpio/gpio-grgpio.c, 449:
> _raw_spin_lock_irqsave in grgpio_remove
>
> kernel/irq/irqdomain.c, 243:
> mutex_lock in irq_domain_remove
> drivers/gpio/gpio-grgpio.c, 463:
> irq_domain_remove in grgpio_remove
> drivers/gpio/gpio-grgpio.c, 449:
> _raw_spin_lock_irqsave in grgpio_remove
>
> mutex_lock() can sleep at runtime.
>
> To fix these bugs, the lock is dropped in grgpio_remove(), because there
> is no need for locking in remove() callbacks.
>
> These bugs are found by a static analysis tool STCheck written by
> myself.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
> ---
> v2:
> * Drop the lock instead of moving the calls to lock functions.
> Thank Bartosz for good advice.

Looks good to me, patch applied!

Yours,
Linus Walleij