Re: [PATCH] pwm: Avoid double mutex lock on pwm_enable

From: Anand Moon
Date: Fri Dec 11 2015 - 00:57:37 EST


Hi Krzysztof

On 11 December 2015 at 09:53, Krzysztof Kozlowski
<k.kozlowski@xxxxxxxxxxx> wrote:
> On 11.12.2015 13:07, Anand Moon wrote:
>> Hi Krzysztof,
>>
>> On 22 November 2015 at 05:43, Krzysztof Kozlowski
>> <k.kozlowski@xxxxxxxxxxx> wrote:
>>> 2015-11-22 3:14 GMT+09:00 Anand Moon <linux.amoon@xxxxxxxxx>:
>>>> Hi Krzysztof,
>
> [...]
>
>>> Yes, now you pasted the same warning I did...
>>>
>>> This is still the same issue. I already wrote it:
>>>> 1. BUG: sleeping function called from invalid context
>>>> 2. DEBUG_LOCKS_WARN_ON(in_interrupt())
>>>
>>> We can repeat it many times but that won't change anything...
>>>
>>> Best regards,
>>> Krzysztof
>>
>> Would you consider below changes to fix the above issue.
>> I have tested this change by enabling CONFIG_DEBUG_ATOMIC_SLEEP
>> And I don't observed issue.
>>
>> 1. BUG: sleeping function called from invalid context
>> 2. DEBUG_LOCKS_WARN_ON(in_interrupt())
>>
>> Please share your thought on this changes.
>>
>> root@odroidxu4:/usr/src/odroidxu3-4.y-devel# git diff drivers/pwm/core.c
>> diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
>> index d24ca5f..f3f6cf9 100644
>> --- a/drivers/pwm/core.c
>> +++ b/drivers/pwm/core.c
>> @@ -506,6 +506,9 @@ int pwm_enable(struct pwm_device *pwm)
>> if (!pwm)
>> return -EINVAL;
>>
>> + if (!mutex_is_locked(&pwm->lock))
>> + return -EINVAL;
>> +
>> mutex_lock(&pwm->lock);
>>
>> if (!test_and_set_bit(PWMF_ENABLED, &pwm->flags)) {
>
> First of all, Thierry suggested way of fixing this:
> "Any objections to simply removing it and make all users use a workqueue
> or some such if they need to control a PWM as a result of an interrupt
> trigger?"

You are correct, current design need to be changes.

> what is wrong with his approach?
>

pwm->lock is locked but it never get unlocked, although the code looks correct.

> Second, you are writing something that looks like mutex-try-lock...
> which will fail the pwm_enable(). IMHO this *hides* the real issue and
> does not solve anything (except hiding also the warning).

Thanks for the inputs. I have realized my mistake.

>
> Best regards,
> Krzysztof

-Anand Moon
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/