[PATCH] iio: fix kobject_put warning in iio_device_register

[PATCH] iio: fix kobject_put warning in iio_device_register

[Zeng Heng]

There is warning reported by kobject lib in kobject_put():

kobject: '(null)' (00000000be81a546): is not initialized, yet kobject_put() is being called.
WARNING: CPU: 0 PID: 535 at lib/kobject.c:718 kobject_put+0x12c/0x180
Call Trace:
 cdev_device_add
 __iio_device_register
 __devm_iio_device_register
 tmp117_probe

If don't need to register chardev for most of IIO devices,
we just register them with device_add() only, and use device_del()
to unregister them.

Otherwise, when device_add() fails in internal and calls kobject_put()
in error handling path, it would report warning because the device
never be registered as chardev and there is no release function for it.

Fixes: 8ebaa3ff1e71 ("iio: core: register chardev only if needed")
Signed-off-by: Zeng Heng <zengheng4@huawei.com>
---
 drivers/iio/industrialio-core.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/iio/industrialio-core.c b/drivers/iio/industrialio-core.c
index 151ff3993354..f4f48bda07f7 100644
--- a/drivers/iio/industrialio-core.c
+++ b/drivers/iio/industrialio-core.c
@@ -1982,7 +1982,11 @@ int __iio_device_register(struct iio_dev *indio_dev, struct module *this_mod)
 	/* assign device groups now; they should be all registered now */
 	indio_dev->dev.groups = iio_dev_opaque->groups;
 
-	ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
+	if (iio_dev_opaque->attached_buffers_cnt || iio_dev_opaque->event_interface)
+		ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
+	else
+		ret = device_add(&indio_dev->dev);
+
 	if (ret < 0)
 		goto error_unreg_eventset;
 
@@ -2008,7 +2012,10 @@ void iio_device_unregister(struct iio_dev *indio_dev)
 {
 	struct iio_dev_opaque *iio_dev_opaque = to_iio_dev_opaque(indio_dev);
 
-	cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
+	if (iio_dev_opaque->chrdev.kobj.state_initialized)
+		cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
+	else
+		device_del(&indio_dev->dev);
 
 	mutex_lock(&iio_dev_opaque->info_exist_lock);
 
-- 
2.25.1

Re: [PATCH] iio: fix kobject_put warning in iio_device_register

[Jonathan Cameron]

On Thu, 10 Nov 2022 21:26:15 +0800
Zeng Heng <zengheng4@huawei.com> wrote:

> There is warning reported by kobject lib in kobject_put():
> 
> kobject: '(null)' (00000000be81a546): is not initialized, yet kobject_put() is being called.
> WARNING: CPU: 0 PID: 535 at lib/kobject.c:718 kobject_put+0x12c/0x180
> Call Trace:
>  cdev_device_add
>  __iio_device_register
>  __devm_iio_device_register
>  tmp117_probe
> 
> If don't need to register chardev for most of IIO devices,
> we just register them with device_add() only, and use device_del()
> to unregister them.



> 
> Otherwise, when device_add() fails in internal and calls kobject_put()
> in error handling path, it would report warning because the device
> never be registered as chardev and there is no release function for it.
> 
> Fixes: 8ebaa3ff1e71 ("iio: core: register chardev only if needed")
> Signed-off-by: Zeng Heng <zengheng4@huawei.com>

Interesting corner case. The cdev_device_add() call is fine with
!dev->devt which is what this code was taking advantage of. The exception
as you have highlighted is the error path of device_add().

So I think it should also cope with unwinding if device_add() fails
and not be calling cdev_del()  Note that cdev_device_del() has the
appropriate guards to be safe whether or not (dev->devt) is true.

Perhaps change cdev_device_add() to have

	rc = device_add(dev);
	if (rc && dev->devt)
		cdev_del(cdev);

	return rc;

Jonathan

> ---
>  drivers/iio/industrialio-core.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/iio/industrialio-core.c b/drivers/iio/industrialio-core.c
> index 151ff3993354..f4f48bda07f7 100644
> --- a/drivers/iio/industrialio-core.c
> +++ b/drivers/iio/industrialio-core.c
> @@ -1982,7 +1982,11 @@ int __iio_device_register(struct iio_dev *indio_dev, struct module *this_mod)
>  	/* assign device groups now; they should be all registered now */
>  	indio_dev->dev.groups = iio_dev_opaque->groups;
>  
> -	ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
> +	if (iio_dev_opaque->attached_buffers_cnt || iio_dev_opaque->event_interface)
> +		ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
> +	else
> +		ret = device_add(&indio_dev->dev);
> +
>  	if (ret < 0)
>  		goto error_unreg_eventset;
>  
> @@ -2008,7 +2012,10 @@ void iio_device_unregister(struct iio_dev *indio_dev)
>  {
>  	struct iio_dev_opaque *iio_dev_opaque = to_iio_dev_opaque(indio_dev);
>  
> -	cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
> +	if (iio_dev_opaque->chrdev.kobj.state_initialized)
> +		cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
> +	else
> +		device_del(&indio_dev->dev);
>  
>  	mutex_lock(&iio_dev_opaque->info_exist_lock);
>  

Re: [PATCH] iio: fix kobject_put warning in iio_device_register

[Zeng Heng]

On 2022/11/11 19:31, Jonathan Cameron wrote:
> On Thu, 10 Nov 2022 21:26:15 +0800
> Zeng Heng <zengheng4@huawei.com> wrote:
>
>> There is warning reported by kobject lib in kobject_put():
>>
>> kobject: '(null)' (00000000be81a546): is not initialized, yet kobject_put() is being called.
>> WARNING: CPU: 0 PID: 535 at lib/kobject.c:718 kobject_put+0x12c/0x180
>> Call Trace:
>>   cdev_device_add
>>   __iio_device_register
>>   __devm_iio_device_register
>>   tmp117_probe
>>
>> If don't need to register chardev for most of IIO devices,
>> we just register them with device_add() only, and use device_del()
>> to unregister them.
>
>
>> Otherwise, when device_add() fails in internal and calls kobject_put()
>> in error handling path, it would report warning because the device
>> never be registered as chardev and there is no release function for it.
>>
>> Fixes: 8ebaa3ff1e71 ("iio: core: register chardev only if needed")
>> Signed-off-by: Zeng Heng <zengheng4@huawei.com>
> Interesting corner case. The cdev_device_add() call is fine with
> !dev->devt which is what this code was taking advantage of. The exception
> as you have highlighted is the error path of device_add().
>
> So I think it should also cope with unwinding if device_add() fails
> and not be calling cdev_del()  Note that cdev_device_del() has the
> appropriate guards to be safe whether or not (dev->devt) is true.
>
> Perhaps change cdev_device_add() to have
>
> 	rc = device_add(dev);
> 	if (rc && dev->devt)
> 		cdev_del(cdev);
>
> 	return rc;

yes, I agree with your opinion about cdev_device_add(), which would be 
more flexible

for callers.

And I found there is existing patch trying to fix this out.

https://lore.kernel.org/lkml/Y1fmgCS7fuf%2FLQBc@kroah.com/

So just pass the patch.


Best Regards,

Zeng Heng

>> ---
>>   drivers/iio/industrialio-core.c | 11 +++++++++--
>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/iio/industrialio-core.c b/drivers/iio/industrialio-core.c
>> index 151ff3993354..f4f48bda07f7 100644
>> --- a/drivers/iio/industrialio-core.c
>> +++ b/drivers/iio/industrialio-core.c
>> @@ -1982,7 +1982,11 @@ int __iio_device_register(struct iio_dev *indio_dev, struct module *this_mod)
>>   	/* assign device groups now; they should be all registered now */
>>   	indio_dev->dev.groups = iio_dev_opaque->groups;
>>   
>> -	ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
>> +	if (iio_dev_opaque->attached_buffers_cnt || iio_dev_opaque->event_interface)
>> +		ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
>> +	else
>> +		ret = device_add(&indio_dev->dev);
>> +
>>   	if (ret < 0)
>>   		goto error_unreg_eventset;
>>   
>> @@ -2008,7 +2012,10 @@ void iio_device_unregister(struct iio_dev *indio_dev)
>>   {
>>   	struct iio_dev_opaque *iio_dev_opaque = to_iio_dev_opaque(indio_dev);
>>   
>> -	cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
>> +	if (iio_dev_opaque->chrdev.kobj.state_initialized)
>> +		cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
>> +	else
>> +		device_del(&indio_dev->dev);
>>   
>>   	mutex_lock(&iio_dev_opaque->info_exist_lock);
>>   

Re: [PATCH] iio: fix kobject_put warning in iio_device_register

[Zeng Heng]

On 2022/11/11 20:05, Zeng Heng wrote:
>
> On 2022/11/11 19:31, Jonathan Cameron wrote:
>> On Thu, 10 Nov 2022 21:26:15 +0800
>> Zeng Heng <zengheng4@huawei.com> wrote:
>>
>>> There is warning reported by kobject lib in kobject_put():
>>>
>>> kobject: '(null)' (00000000be81a546): is not initialized, yet 
>>> kobject_put() is being called.
>>> WARNING: CPU: 0 PID: 535 at lib/kobject.c:718 kobject_put+0x12c/0x180
>>> Call Trace:
>>>   cdev_device_add
>>>   __iio_device_register
>>>   __devm_iio_device_register
>>>   tmp117_probe
>>>
>>> If don't need to register chardev for most of IIO devices,
>>> we just register them with device_add() only, and use device_del()
>>> to unregister them.
>>
>>
>>> Otherwise, when device_add() fails in internal and calls kobject_put()
>>> in error handling path, it would report warning because the device
>>> never be registered as chardev and there is no release function for it.
>>>
>>> Fixes: 8ebaa3ff1e71 ("iio: core: register chardev only if needed")
>>> Signed-off-by: Zeng Heng <zengheng4@huawei.com>
>> Interesting corner case. The cdev_device_add() call is fine with
>> !dev->devt which is what this code was taking advantage of. The 
>> exception
>> as you have highlighted is the error path of device_add().
>>
>> So I think it should also cope with unwinding if device_add() fails
>> and not be calling cdev_del()  Note that cdev_device_del() has the
>> appropriate guards to be safe whether or not (dev->devt) is true.
>>
>> Perhaps change cdev_device_add() to have
>>
>>     rc = device_add(dev);
>>     if (rc && dev->devt)
>>         cdev_del(cdev);
>>
>>     return rc;
>
yes, I agree with your opinion about cdev_device_add(), which would be 
more flexible

for callers.

And I *just find* there is existing patch trying to fix this out.

https://lore.kernel.org/lkml/Y1fmgCS7fuf%2FLQBc@kroah.com/

So just pass the patch.


Best Regards,

Zeng Heng

>>> ---
>>>   drivers/iio/industrialio-core.c | 11 +++++++++--
>>>   1 file changed, 9 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/iio/industrialio-core.c 
>>> b/drivers/iio/industrialio-core.c
>>> index 151ff3993354..f4f48bda07f7 100644
>>> --- a/drivers/iio/industrialio-core.c
>>> +++ b/drivers/iio/industrialio-core.c
>>> @@ -1982,7 +1982,11 @@ int __iio_device_register(struct iio_dev 
>>> *indio_dev, struct module *this_mod)
>>>       /* assign device groups now; they should be all registered now */
>>>       indio_dev->dev.groups = iio_dev_opaque->groups;
>>>   -    ret = cdev_device_add(&iio_dev_opaque->chrdev, &indio_dev->dev);
>>> +    if (iio_dev_opaque->attached_buffers_cnt || 
>>> iio_dev_opaque->event_interface)
>>> +        ret = cdev_device_add(&iio_dev_opaque->chrdev, 
>>> &indio_dev->dev);
>>> +    else
>>> +        ret = device_add(&indio_dev->dev);
>>> +
>>>       if (ret < 0)
>>>           goto error_unreg_eventset;
>>>   @@ -2008,7 +2012,10 @@ void iio_device_unregister(struct iio_dev 
>>> *indio_dev)
>>>   {
>>>       struct iio_dev_opaque *iio_dev_opaque = 
>>> to_iio_dev_opaque(indio_dev);
>>>   -    cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
>>> +    if (iio_dev_opaque->chrdev.kobj.state_initialized)
>>> +        cdev_device_del(&iio_dev_opaque->chrdev, &indio_dev->dev);
>>> +    else
>>> +        device_del(&indio_dev->dev);
>>>         mutex_lock(&iio_dev_opaque->info_exist_lock);