Re: [PATCHv2] pwm: avoid holding mutex in interrupt context

[Krzysztof Kozlowski]

On 18.01.2016 06:01, Anand Moon wrote:
> The introduction of the mutex in commit d1cd21427747 ("pwm: Set enable
> state properly on failed call to enable") effectively makes all PWM drivers
> potentially sleeping. That in turn makes the .can_sleep field obsolete
> since all drivers can now sleep.
> 
> Changes fix the below bug by using spinlocks instead of mutex
> 
> [   22.300239] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   22.307212] in_atomic(): 1, irqs_disabled(): 0, pid: 2257, name: sh
> [   22.313454] Preemption disabled at:[<  (null)>]   (null)
> [   23.655232] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   23.662174] in_atomic(): 1, irqs_disabled(): 0, pid: 2404, name: upowerd
> [   23.668932] Preemption disabled at:[<  (null)>]   (null)
> [   25.010207] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   25.017125] in_atomic(): 1, irqs_disabled(): 0, pid: 2262, name: indicator-keybo
> [   25.024491] Preemption disabled at:[<  (null)>]   (null)
> [   26.355237] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   26.362141] in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper/0
> [   26.368728] Preemption disabled at:[<  (null)>]   (null)
> [   27.680220] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   27.687119] in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper/0
> [   27.693698] Preemption disabled at:[<  (null)>]   (null)
> [   29.005199] BUG: sleeping function called from invalid context at kernel/locking/mutex.c:97
> [   29.012124] in_atomic(): 1, irqs_disabled(): 0, pid: 0, name: swapper/0
> 
> [thierry.reding@xxxxxxxxx: Fixed the commit message]
> Signed-off-by: Anand Moon <linux.amoon@xxxxxxxxx>
> ---
> Changes logs: droped my prevoius approch.
> ---
>  drivers/pwm/core.c  | 10 +++++-----
>  include/linux/pwm.h |  4 ++--
>  2 files changed, 7 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
> index d24ca5f..58e7091 100644
> --- a/drivers/pwm/core.c
> +++ b/drivers/pwm/core.c
> @@ -269,7 +269,7 @@ int pwmchip_add_with_polarity(struct pwm_chip *chip,
>  		pwm->pwm = chip->base + i;
>  		pwm->hwpwm = i;
>  		pwm->polarity = polarity;
> -		mutex_init(&pwm->lock);
> +		spin_lock_init(&pwm->lock);
>  
>  		radix_tree_insert(&pwm_tree, pwm->pwm, pwm);
>  	}
> @@ -474,7 +474,7 @@ int pwm_set_polarity(struct pwm_device *pwm, enum pwm_polarity polarity)
>  	if (!pwm->chip->ops->set_polarity)
>  		return -ENOSYS;
>  
> -	mutex_lock(&pwm->lock);
> +	spin_lock_irq(&pwm->lock);

Anand,

Thank you for the effort put into digging into this issue. Unfortunately
this approach is bad. You cannot fix one issue without looking at the
big picture of the given subsystem. This patch does exactly this - fixes
your warning but probably introduces bugs all over the place.

Although the set_polarity callback (called under the lock) is not
described as sleeping-allowed but some implementations do it in a
sleeping way. This is really easy to find, e.g.:
pwm_omap_dmtimer_set_polarity.

This means: no.

Best regards,
Krzysztof

>  
>  	if (pwm_is_enabled(pwm)) {
>  		err = -EBUSY;
> @@ -488,7 +488,7 @@ int pwm_set_polarity(struct pwm_device *pwm, enum pwm_polarity polarity)
>  	pwm->polarity = polarity;
>  
>  unlock:
> -	mutex_unlock(&pwm->lock);
> +	spin_unlock_irq(&pwm->lock);
>  	return err;
>  }
>  EXPORT_SYMBOL_GPL(pwm_set_polarity);
> @@ -506,7 +506,7 @@ int pwm_enable(struct pwm_device *pwm)
>  	if (!pwm)
>  		return -EINVAL;
>  
> -	mutex_lock(&pwm->lock);
> +	spin_lock_irq(&pwm->lock);
>  
>  	if (!test_and_set_bit(PWMF_ENABLED, &pwm->flags)) {
>  		err = pwm->chip->ops->enable(pwm->chip, pwm);
> @@ -514,7 +514,7 @@ int pwm_enable(struct pwm_device *pwm)
>  			clear_bit(PWMF_ENABLED, &pwm->flags);
>  	}
>  
> -	mutex_unlock(&pwm->lock);
> +	spin_unlock_irq(&pwm->lock);
>  
>  	return err;
>  }
> diff --git a/include/linux/pwm.h b/include/linux/pwm.h
> index cfc3ed4..86ad4c2 100644
> --- a/include/linux/pwm.h
> +++ b/include/linux/pwm.h
> @@ -2,7 +2,7 @@
>  #define __LINUX_PWM_H
>  
>  #include <linux/err.h>
> -#include <linux/mutex.h>
> +#include <linux/spinlock.h>
>  #include <linux/of.h>
>  
>  struct pwm_device;
> @@ -100,7 +100,7 @@ struct pwm_device {
>  	unsigned int pwm;
>  	struct pwm_chip *chip;
>  	void *chip_data;
> -	struct mutex lock;
> +	spinlock_t lock;
>  
>  	unsigned int period;
>  	unsigned int duty_cycle;
>