2019-12-19 11:11:57

by Bartosz Golaszewski

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

śr., 18 gru 2019 o 14:26 Jia-Ju Bai <[email protected]> napisał(a):
>
> The driver may sleep while holding a spinlock.
> The function call path (from bottom to top) in Linux 4.19 is:
>
> 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, gpiochip_remove() and irq_domain_remove() are called
> without holding the spinlock.
>
> These bugs are found by a static analysis tool STCheck written by myself.
>
> Signed-off-by: Jia-Ju Bai <[email protected]>
> ---
> drivers/gpio/gpio-grgpio.c | 5 ++++-
> sound/soc/sti/uniperif_player.c | 3 ++-
> 2 files changed, 6 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpio/gpio-grgpio.c b/drivers/gpio/gpio-grgpio.c
> index 08234e64993a..60a2871c5ba7 100644
> --- a/drivers/gpio/gpio-grgpio.c
> +++ b/drivers/gpio/gpio-grgpio.c
> @@ -448,13 +448,16 @@ static int grgpio_remove(struct platform_device *ofdev)
> }
> }
>
> + spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
> +
> gpiochip_remove(&priv->gc);
>
> if (priv->domain)
> irq_domain_remove(priv->domain);
>
> out:
> - spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
> + if (ret)
> + spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);

In general there is no need for locking in remove() callbacks. I guess
you can safely remove the spinlock here all together.

>
> return ret;
> }
> diff --git a/sound/soc/sti/uniperif_player.c b/sound/soc/sti/uniperif_player.c
> index 48ea915b24ba..62244e207679 100644
> --- a/sound/soc/sti/uniperif_player.c
> +++ b/sound/soc/sti/uniperif_player.c
> @@ -601,13 +601,14 @@ static int uni_player_ctl_iec958_put(struct snd_kcontrol *kcontrol,
> mutex_unlock(&player->ctrl_lock);
>
> spin_lock_irqsave(&player->irq_lock, flags);
> + spin_unlock_irqrestore(&player->irq_lock, flags);

Yeah I can tell this was generated automatically - what does this line
is expected to achieve?

Bart

> +
> if (player->substream && player->substream->runtime)
> uni_player_set_channel_status(player,
> player->substream->runtime);
> else
> uni_player_set_channel_status(player, NULL);
>
> - spin_unlock_irqrestore(&player->irq_lock, flags);
> return 0;
> }
>
> --
> 2.17.1
>


2019-12-19 12:37:19

by Jia-Ju Bai

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



On 2019/12/19 19:10, Bartosz Golaszewski wrote:
> śr., 18 gru 2019 o 14:26 Jia-Ju Bai <[email protected]> napisał(a):
>> The driver may sleep while holding a spinlock.
>> The function call path (from bottom to top) in Linux 4.19 is:
>>
>> 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, gpiochip_remove() and irq_domain_remove() are called
>> without holding the spinlock.
>>
>> These bugs are found by a static analysis tool STCheck written by myself.
>>
>> Signed-off-by: Jia-Ju Bai <[email protected]>
>> ---
>> drivers/gpio/gpio-grgpio.c | 5 ++++-
>> sound/soc/sti/uniperif_player.c | 3 ++-
>> 2 files changed, 6 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpio/gpio-grgpio.c b/drivers/gpio/gpio-grgpio.c
>> index 08234e64993a..60a2871c5ba7 100644
>> --- a/drivers/gpio/gpio-grgpio.c
>> +++ b/drivers/gpio/gpio-grgpio.c
>> @@ -448,13 +448,16 @@ static int grgpio_remove(struct platform_device *ofdev)
>> }
>> }
>>
>> + spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
>> +
>> gpiochip_remove(&priv->gc);
>>
>> if (priv->domain)
>> irq_domain_remove(priv->domain);
>>
>> out:
>> - spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
>> + if (ret)
>> + spin_unlock_irqrestore(&priv->gc.bgpio_lock, flags);
> In general there is no need for locking in remove() callbacks. I guess
> you can safely remove the spinlock here all together.

Okay, I will send a new patch.

>
>> return ret;
>> }
>> diff --git a/sound/soc/sti/uniperif_player.c b/sound/soc/sti/uniperif_player.c
>> index 48ea915b24ba..62244e207679 100644
>> --- a/sound/soc/sti/uniperif_player.c
>> +++ b/sound/soc/sti/uniperif_player.c
>> @@ -601,13 +601,14 @@ static int uni_player_ctl_iec958_put(struct snd_kcontrol *kcontrol,
>> mutex_unlock(&player->ctrl_lock);
>>
>> spin_lock_irqsave(&player->irq_lock, flags);
>> + spin_unlock_irqrestore(&player->irq_lock, flags);
> Yeah I can tell this was generated automatically - what does this line
> is expected to achieve?

Ah, sorry, this is my mistake.
I forgot to reset the kernel code before writing the patch...


Best wishes,
Jia-Ju Bai